xref: /freebsd/usr.sbin/bhyve/pci_e82545.c (revision 069ac184)
1 /*
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2016 Alexander Motin <mav@FreeBSD.org>
5  * Copyright (c) 2015 Peter Grehan <grehan@freebsd.org>
6  * Copyright (c) 2013 Jeremiah Lott, Avere Systems
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer
14  *    in this position and unchanged.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/types.h>
33 #ifndef WITHOUT_CAPSICUM
34 #include <sys/capsicum.h>
35 #endif
36 #include <sys/limits.h>
37 #include <sys/ioctl.h>
38 #include <sys/uio.h>
39 #include <net/ethernet.h>
40 #include <netinet/in.h>
41 #include <netinet/tcp.h>
42 
43 #ifndef WITHOUT_CAPSICUM
44 #include <capsicum_helpers.h>
45 #endif
46 
47 #include <err.h>
48 #include <errno.h>
49 #include <fcntl.h>
50 #include <md5.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <sysexits.h>
55 #include <unistd.h>
56 #include <pthread.h>
57 #include <pthread_np.h>
58 
59 #include "e1000_regs.h"
60 #include "e1000_defines.h"
61 #include "mii.h"
62 
63 #include "bhyverun.h"
64 #include "config.h"
65 #include "debug.h"
66 #include "pci_emul.h"
67 #ifdef BHYVE_SNAPSHOT
68 #include "snapshot.h"
69 #endif
70 #include "mevent.h"
71 #include "net_utils.h"
72 #include "net_backends.h"
73 
74 /* Hardware/register definitions XXX: move some to common code. */
75 #define E82545_VENDOR_ID_INTEL			0x8086
76 #define E82545_DEV_ID_82545EM_COPPER		0x100F
77 #define E82545_SUBDEV_ID			0x1008
78 
79 #define E82545_REVISION_4			4
80 
81 #define E82545_MDIC_DATA_MASK			0x0000FFFF
82 #define E82545_MDIC_OP_MASK			0x0c000000
83 #define E82545_MDIC_IE				0x20000000
84 
85 #define E82545_EECD_FWE_DIS	0x00000010 /* Flash writes disabled */
86 #define E82545_EECD_FWE_EN	0x00000020 /* Flash writes enabled */
87 #define E82545_EECD_FWE_MASK	0x00000030 /* Flash writes mask */
88 
89 #define E82545_BAR_REGISTER			0
90 #define E82545_BAR_REGISTER_LEN			(128*1024)
91 #define E82545_BAR_FLASH			1
92 #define E82545_BAR_FLASH_LEN			(64*1024)
93 #define E82545_BAR_IO				2
94 #define E82545_BAR_IO_LEN			8
95 
96 #define E82545_IOADDR				0x00000000
97 #define E82545_IODATA				0x00000004
98 #define E82545_IO_REGISTER_MAX			0x0001FFFF
99 #define E82545_IO_FLASH_BASE			0x00080000
100 #define E82545_IO_FLASH_MAX			0x000FFFFF
101 
102 #define E82545_ARRAY_ENTRY(reg, offset)		(reg + (offset<<2))
103 #define E82545_RAR_MAX				15
104 #define E82545_MTA_MAX				127
105 #define E82545_VFTA_MAX				127
106 
107 /* Slightly modified from the driver versions, hardcoded for 3 opcode bits,
108  * followed by 6 address bits.
109  * TODO: make opcode bits and addr bits configurable?
110  * NVM Commands - Microwire */
111 #define E82545_NVM_OPCODE_BITS	3
112 #define E82545_NVM_ADDR_BITS	6
113 #define E82545_NVM_DATA_BITS	16
114 #define E82545_NVM_OPADDR_BITS	(E82545_NVM_OPCODE_BITS + E82545_NVM_ADDR_BITS)
115 #define E82545_NVM_ADDR_MASK	((1 << E82545_NVM_ADDR_BITS)-1)
116 #define E82545_NVM_OPCODE_MASK	\
117     (((1 << E82545_NVM_OPCODE_BITS) - 1) << E82545_NVM_ADDR_BITS)
118 #define E82545_NVM_OPCODE_READ	(0x6 << E82545_NVM_ADDR_BITS)	/* read */
119 #define E82545_NVM_OPCODE_WRITE	(0x5 << E82545_NVM_ADDR_BITS)	/* write */
120 #define E82545_NVM_OPCODE_ERASE	(0x7 << E82545_NVM_ADDR_BITS)	/* erase */
121 #define	E82545_NVM_OPCODE_EWEN	(0x4 << E82545_NVM_ADDR_BITS)	/* wr-enable */
122 
123 #define	E82545_NVM_EEPROM_SIZE	64 /* 64 * 16-bit values == 128K */
124 
125 #define E1000_ICR_SRPD		0x00010000
126 
127 /* This is an arbitrary number.  There is no hard limit on the chip. */
128 #define I82545_MAX_TXSEGS	64
129 
130 /* Legacy receive descriptor */
131 struct e1000_rx_desc {
132 	uint64_t buffer_addr;	/* Address of the descriptor's data buffer */
133 	uint16_t length;	/* Length of data DMAed into data buffer */
134 	uint16_t csum;		/* Packet checksum */
135 	uint8_t	 status;       	/* Descriptor status */
136 	uint8_t  errors;	/* Descriptor Errors */
137 	uint16_t special;
138 };
139 
140 /* Transmit descriptor types */
141 #define	E1000_TXD_MASK		(E1000_TXD_CMD_DEXT | 0x00F00000)
142 #define E1000_TXD_TYP_L		(0)
143 #define E1000_TXD_TYP_C		(E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_C)
144 #define E1000_TXD_TYP_D		(E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D)
145 
146 /* Legacy transmit descriptor */
147 struct e1000_tx_desc {
148 	uint64_t buffer_addr;   /* Address of the descriptor's data buffer */
149 	union {
150 		uint32_t data;
151 		struct {
152 			uint16_t length;  /* Data buffer length */
153 			uint8_t  cso;  /* Checksum offset */
154 			uint8_t  cmd;  /* Descriptor control */
155 		} flags;
156 	} lower;
157 	union {
158 		uint32_t data;
159 		struct {
160 			uint8_t status; /* Descriptor status */
161 			uint8_t css;  /* Checksum start */
162 			uint16_t special;
163 		} fields;
164 	} upper;
165 };
166 
167 /* Context descriptor */
168 struct e1000_context_desc {
169 	union {
170 		uint32_t ip_config;
171 		struct {
172 			uint8_t ipcss;  /* IP checksum start */
173 			uint8_t ipcso;  /* IP checksum offset */
174 			uint16_t ipcse;  /* IP checksum end */
175 		} ip_fields;
176 	} lower_setup;
177 	union {
178 		uint32_t tcp_config;
179 		struct {
180 			uint8_t tucss;  /* TCP checksum start */
181 			uint8_t tucso;  /* TCP checksum offset */
182 			uint16_t tucse;  /* TCP checksum end */
183 		} tcp_fields;
184 	} upper_setup;
185 	uint32_t cmd_and_length;
186 	union {
187 		uint32_t data;
188 		struct {
189 			uint8_t status;  /* Descriptor status */
190 			uint8_t hdr_len;  /* Header length */
191 			uint16_t mss;  /* Maximum segment size */
192 		} fields;
193 	} tcp_seg_setup;
194 };
195 
196 /* Data descriptor */
197 struct e1000_data_desc {
198 	uint64_t buffer_addr;  /* Address of the descriptor's buffer address */
199 	union {
200 		uint32_t data;
201 		struct {
202 			uint16_t length;  /* Data buffer length */
203 			uint8_t typ_len_ext;
204 			uint8_t cmd;
205 		} flags;
206 	} lower;
207 	union {
208 		uint32_t data;
209 		struct {
210 			uint8_t status;  /* Descriptor status */
211 			uint8_t popts;  /* Packet Options */
212 			uint16_t special;
213 		} fields;
214 	} upper;
215 };
216 
217 union e1000_tx_udesc {
218 	struct e1000_tx_desc td;
219 	struct e1000_context_desc cd;
220 	struct e1000_data_desc dd;
221 };
222 
223 /* Tx checksum info for a packet. */
224 struct ck_info {
225 	int	ck_valid;	/* ck_info is valid */
226 	uint8_t	ck_start;	/* start byte of cksum calcuation */
227 	uint8_t	ck_off;		/* offset of cksum insertion */
228 	uint16_t ck_len;	/* length of cksum calc: 0 is to packet-end */
229 };
230 
231 /*
232  * Debug printf
233  */
234 static int e82545_debug = 0;
235 #define WPRINTF(msg,params...) PRINTLN("e82545: " msg, ##params)
236 #define DPRINTF(msg,params...) if (e82545_debug) WPRINTF(msg, params)
237 
238 #define	MIN(a,b) (((a)<(b))?(a):(b))
239 #define	MAX(a,b) (((a)>(b))?(a):(b))
240 
241 /* s/w representation of the RAL/RAH regs */
242 struct  eth_uni {
243 	int		eu_valid;
244 	int		eu_addrsel;
245 	struct ether_addr eu_eth;
246 };
247 
248 
249 struct e82545_softc {
250 	struct pci_devinst *esc_pi;
251 	struct vmctx	*esc_ctx;
252 	struct mevent   *esc_mevpitr;
253 	pthread_mutex_t	esc_mtx;
254 	struct ether_addr esc_mac;
255 	net_backend_t	*esc_be;
256 
257 	/* General */
258 	uint32_t	esc_CTRL;	/* x0000 device ctl */
259 	uint32_t	esc_FCAL;	/* x0028 flow ctl addr lo */
260 	uint32_t	esc_FCAH;	/* x002C flow ctl addr hi */
261 	uint32_t	esc_FCT;	/* x0030 flow ctl type */
262 	uint32_t	esc_VET;	/* x0038 VLAN eth type */
263 	uint32_t	esc_FCTTV;	/* x0170 flow ctl tx timer */
264 	uint32_t	esc_LEDCTL;	/* x0E00 LED control */
265 	uint32_t	esc_PBA;	/* x1000 pkt buffer allocation */
266 
267 	/* Interrupt control */
268 	int		esc_irq_asserted;
269 	uint32_t	esc_ICR;	/* x00C0 cause read/clear */
270 	uint32_t	esc_ITR;	/* x00C4 intr throttling */
271 	uint32_t	esc_ICS;	/* x00C8 cause set */
272 	uint32_t	esc_IMS;	/* x00D0 mask set/read */
273 	uint32_t	esc_IMC;	/* x00D8 mask clear */
274 
275 	/* Transmit */
276 	union e1000_tx_udesc *esc_txdesc;
277 	struct e1000_context_desc esc_txctx;
278 	pthread_t	esc_tx_tid;
279 	pthread_cond_t	esc_tx_cond;
280 	int		esc_tx_enabled;
281 	int		esc_tx_active;
282 	uint32_t	esc_TXCW;	/* x0178 transmit config */
283 	uint32_t	esc_TCTL;	/* x0400 transmit ctl */
284 	uint32_t	esc_TIPG;	/* x0410 inter-packet gap */
285 	uint16_t	esc_AIT;	/* x0458 Adaptive Interframe Throttle */
286 	uint64_t	esc_tdba;      	/* verified 64-bit desc table addr */
287 	uint32_t	esc_TDBAL;	/* x3800 desc table addr, low bits */
288 	uint32_t	esc_TDBAH;	/* x3804 desc table addr, hi 32-bits */
289 	uint32_t	esc_TDLEN;	/* x3808 # descriptors in bytes */
290 	uint16_t	esc_TDH;	/* x3810 desc table head idx */
291 	uint16_t	esc_TDHr;	/* internal read version of TDH */
292 	uint16_t	esc_TDT;	/* x3818 desc table tail idx */
293 	uint32_t	esc_TIDV;	/* x3820 intr delay */
294 	uint32_t	esc_TXDCTL;	/* x3828 desc control */
295 	uint32_t	esc_TADV;	/* x382C intr absolute delay */
296 
297 	/* L2 frame acceptance */
298 	struct eth_uni	esc_uni[16];	/* 16 x unicast MAC addresses */
299 	uint32_t	esc_fmcast[128]; /* Multicast filter bit-match */
300 	uint32_t	esc_fvlan[128]; /* VLAN 4096-bit filter */
301 
302 	/* Receive */
303 	struct e1000_rx_desc *esc_rxdesc;
304 	pthread_cond_t	esc_rx_cond;
305 	int		esc_rx_enabled;
306 	int		esc_rx_active;
307 	int		esc_rx_loopback;
308 	uint32_t	esc_RCTL;	/* x0100 receive ctl */
309 	uint32_t	esc_FCRTL;	/* x2160 flow cntl thresh, low */
310 	uint32_t	esc_FCRTH;	/* x2168 flow cntl thresh, hi */
311 	uint64_t	esc_rdba;	/* verified 64-bit desc table addr */
312 	uint32_t	esc_RDBAL;	/* x2800 desc table addr, low bits */
313 	uint32_t	esc_RDBAH;	/* x2804 desc table addr, hi 32-bits*/
314 	uint32_t	esc_RDLEN;	/* x2808 #descriptors */
315 	uint16_t	esc_RDH;	/* x2810 desc table head idx */
316 	uint16_t	esc_RDT;	/* x2818 desc table tail idx */
317 	uint32_t	esc_RDTR;	/* x2820 intr delay */
318 	uint32_t	esc_RXDCTL;	/* x2828 desc control */
319 	uint32_t	esc_RADV;	/* x282C intr absolute delay */
320 	uint32_t	esc_RSRPD;	/* x2C00 recv small packet detect */
321 	uint32_t	esc_RXCSUM;     /* x5000 receive cksum ctl */
322 
323 	/* IO Port register access */
324 	uint32_t io_addr;
325 
326 	/* Shadow copy of MDIC */
327 	uint32_t mdi_control;
328 	/* Shadow copy of EECD */
329 	uint32_t eeprom_control;
330 	/* Latest NVM in/out */
331 	uint16_t nvm_data;
332 	uint16_t nvm_opaddr;
333 	/* stats */
334 	uint32_t missed_pkt_count; /* dropped for no room in rx queue */
335 	uint32_t pkt_rx_by_size[6];
336 	uint32_t pkt_tx_by_size[6];
337 	uint32_t good_pkt_rx_count;
338 	uint32_t bcast_pkt_rx_count;
339 	uint32_t mcast_pkt_rx_count;
340 	uint32_t good_pkt_tx_count;
341 	uint32_t bcast_pkt_tx_count;
342 	uint32_t mcast_pkt_tx_count;
343 	uint32_t oversize_rx_count;
344 	uint32_t tso_tx_count;
345 	uint64_t good_octets_rx;
346 	uint64_t good_octets_tx;
347 	uint64_t missed_octets; /* counts missed and oversized */
348 
349 	uint8_t nvm_bits:6; /* number of bits remaining in/out */
350 	uint8_t nvm_mode:2;
351 #define E82545_NVM_MODE_OPADDR  0x0
352 #define E82545_NVM_MODE_DATAIN  0x1
353 #define E82545_NVM_MODE_DATAOUT 0x2
354 	/* EEPROM data */
355 	uint16_t eeprom_data[E82545_NVM_EEPROM_SIZE];
356 };
357 
358 static void e82545_reset(struct e82545_softc *sc, int dev);
359 static void e82545_rx_enable(struct e82545_softc *sc);
360 static void e82545_rx_disable(struct e82545_softc *sc);
361 static void e82545_rx_callback(int fd, enum ev_type type, void *param);
362 static void e82545_tx_start(struct e82545_softc *sc);
363 static void e82545_tx_enable(struct e82545_softc *sc);
364 static void e82545_tx_disable(struct e82545_softc *sc);
365 
366 static inline int __unused
367 e82545_size_stat_index(uint32_t size)
368 {
369 	if (size <= 64) {
370 		return 0;
371 	} else if (size >= 1024) {
372 		return 5;
373 	} else {
374 		/* should be 1-4 */
375 		return (ffs(size) - 6);
376 	}
377 }
378 
379 static void
380 e82545_init_eeprom(struct e82545_softc *sc)
381 {
382 	uint16_t checksum, i;
383 
384         /* mac addr */
385 	sc->eeprom_data[NVM_MAC_ADDR] = ((uint16_t)sc->esc_mac.octet[0]) |
386 		(((uint16_t)sc->esc_mac.octet[1]) << 8);
387 	sc->eeprom_data[NVM_MAC_ADDR+1] = ((uint16_t)sc->esc_mac.octet[2]) |
388 		(((uint16_t)sc->esc_mac.octet[3]) << 8);
389 	sc->eeprom_data[NVM_MAC_ADDR+2] = ((uint16_t)sc->esc_mac.octet[4]) |
390 		(((uint16_t)sc->esc_mac.octet[5]) << 8);
391 
392 	/* pci ids */
393 	sc->eeprom_data[NVM_SUB_DEV_ID] = E82545_SUBDEV_ID;
394 	sc->eeprom_data[NVM_SUB_VEN_ID] = E82545_VENDOR_ID_INTEL;
395 	sc->eeprom_data[NVM_DEV_ID] = E82545_DEV_ID_82545EM_COPPER;
396 	sc->eeprom_data[NVM_VEN_ID] = E82545_VENDOR_ID_INTEL;
397 
398 	/* fill in the checksum */
399         checksum = 0;
400 	for (i = 0; i < NVM_CHECKSUM_REG; i++) {
401 		checksum += sc->eeprom_data[i];
402 	}
403 	checksum = NVM_SUM - checksum;
404 	sc->eeprom_data[NVM_CHECKSUM_REG] = checksum;
405 	DPRINTF("eeprom checksum: 0x%x", checksum);
406 }
407 
408 static void
409 e82545_write_mdi(struct e82545_softc *sc __unused, uint8_t reg_addr,
410     uint8_t phy_addr, uint32_t data)
411 {
412 	DPRINTF("Write mdi reg:0x%x phy:0x%x data: 0x%x", reg_addr, phy_addr, data);
413 }
414 
415 static uint32_t
416 e82545_read_mdi(struct e82545_softc *sc __unused, uint8_t reg_addr,
417     uint8_t phy_addr)
418 {
419 	//DPRINTF("Read mdi reg:0x%x phy:0x%x", reg_addr, phy_addr);
420 	switch (reg_addr) {
421 	case PHY_STATUS:
422 		return (MII_SR_LINK_STATUS | MII_SR_AUTONEG_CAPS |
423 			MII_SR_AUTONEG_COMPLETE);
424 	case PHY_AUTONEG_ADV:
425 		return NWAY_AR_SELECTOR_FIELD;
426 	case PHY_LP_ABILITY:
427 		return 0;
428 	case PHY_1000T_STATUS:
429 		return (SR_1000T_LP_FD_CAPS | SR_1000T_REMOTE_RX_STATUS |
430 			SR_1000T_LOCAL_RX_STATUS);
431 	case PHY_ID1:
432 		return (M88E1011_I_PHY_ID >> 16) & 0xFFFF;
433 	case PHY_ID2:
434 		return (M88E1011_I_PHY_ID | E82545_REVISION_4) & 0xFFFF;
435 	default:
436 		DPRINTF("Unknown mdi read reg:0x%x phy:0x%x", reg_addr, phy_addr);
437 		return 0;
438 	}
439 	/* not reached */
440 }
441 
442 static void
443 e82545_eecd_strobe(struct e82545_softc *sc)
444 {
445 	/* Microwire state machine */
446 	/*
447 	DPRINTF("eeprom state machine srtobe "
448 		"0x%x 0x%x 0x%x 0x%x",
449 		sc->nvm_mode, sc->nvm_bits,
450 		sc->nvm_opaddr, sc->nvm_data);*/
451 
452 	if (sc->nvm_bits == 0) {
453 		DPRINTF("eeprom state machine not expecting data! "
454 			"0x%x 0x%x 0x%x 0x%x",
455 			sc->nvm_mode, sc->nvm_bits,
456 			sc->nvm_opaddr, sc->nvm_data);
457 		return;
458 	}
459 	sc->nvm_bits--;
460 	if (sc->nvm_mode == E82545_NVM_MODE_DATAOUT) {
461 		/* shifting out */
462 		if (sc->nvm_data & 0x8000) {
463 			sc->eeprom_control |= E1000_EECD_DO;
464 		} else {
465 			sc->eeprom_control &= ~E1000_EECD_DO;
466 		}
467 		sc->nvm_data <<= 1;
468 		if (sc->nvm_bits == 0) {
469 			/* read done, back to opcode mode. */
470 			sc->nvm_opaddr = 0;
471 			sc->nvm_mode = E82545_NVM_MODE_OPADDR;
472 			sc->nvm_bits = E82545_NVM_OPADDR_BITS;
473 		}
474 	} else if (sc->nvm_mode == E82545_NVM_MODE_DATAIN) {
475 		/* shifting in */
476 		sc->nvm_data <<= 1;
477 		if (sc->eeprom_control & E1000_EECD_DI) {
478 			sc->nvm_data |= 1;
479 		}
480 		if (sc->nvm_bits == 0) {
481 			/* eeprom write */
482 			uint16_t op = sc->nvm_opaddr & E82545_NVM_OPCODE_MASK;
483 			uint16_t addr = sc->nvm_opaddr & E82545_NVM_ADDR_MASK;
484 			if (op != E82545_NVM_OPCODE_WRITE) {
485 				DPRINTF("Illegal eeprom write op 0x%x",
486 					sc->nvm_opaddr);
487 			} else if (addr >= E82545_NVM_EEPROM_SIZE) {
488 				DPRINTF("Illegal eeprom write addr 0x%x",
489 					sc->nvm_opaddr);
490 			} else {
491 				DPRINTF("eeprom write eeprom[0x%x] = 0x%x",
492 				addr, sc->nvm_data);
493 				sc->eeprom_data[addr] = sc->nvm_data;
494 			}
495 			/* back to opcode mode */
496 			sc->nvm_opaddr = 0;
497 			sc->nvm_mode = E82545_NVM_MODE_OPADDR;
498 			sc->nvm_bits = E82545_NVM_OPADDR_BITS;
499 		}
500 	} else if (sc->nvm_mode == E82545_NVM_MODE_OPADDR) {
501 		sc->nvm_opaddr <<= 1;
502 		if (sc->eeprom_control & E1000_EECD_DI) {
503 			sc->nvm_opaddr |= 1;
504 		}
505 		if (sc->nvm_bits == 0) {
506 			uint16_t op = sc->nvm_opaddr & E82545_NVM_OPCODE_MASK;
507 			switch (op) {
508 			case E82545_NVM_OPCODE_EWEN:
509 				DPRINTF("eeprom write enable: 0x%x",
510 					sc->nvm_opaddr);
511 				/* back to opcode mode */
512 				sc->nvm_opaddr = 0;
513 				sc->nvm_mode = E82545_NVM_MODE_OPADDR;
514 				sc->nvm_bits = E82545_NVM_OPADDR_BITS;
515 				break;
516 			case E82545_NVM_OPCODE_READ:
517 			{
518 				uint16_t addr = sc->nvm_opaddr &
519 					E82545_NVM_ADDR_MASK;
520 				sc->nvm_mode = E82545_NVM_MODE_DATAOUT;
521 				sc->nvm_bits = E82545_NVM_DATA_BITS;
522 				if (addr < E82545_NVM_EEPROM_SIZE) {
523 					sc->nvm_data = sc->eeprom_data[addr];
524 					DPRINTF("eeprom read: eeprom[0x%x] = 0x%x",
525 						addr, sc->nvm_data);
526 				} else {
527 					DPRINTF("eeprom illegal read: 0x%x",
528 						sc->nvm_opaddr);
529 					sc->nvm_data = 0;
530 				}
531 				break;
532 			}
533 			case E82545_NVM_OPCODE_WRITE:
534 				sc->nvm_mode = E82545_NVM_MODE_DATAIN;
535 				sc->nvm_bits = E82545_NVM_DATA_BITS;
536 				sc->nvm_data = 0;
537 				break;
538 			default:
539 				DPRINTF("eeprom unknown op: 0x%x",
540 					sc->nvm_opaddr);
541 				/* back to opcode mode */
542 				sc->nvm_opaddr = 0;
543 				sc->nvm_mode = E82545_NVM_MODE_OPADDR;
544 				sc->nvm_bits = E82545_NVM_OPADDR_BITS;
545 			}
546 		}
547 	} else {
548 		DPRINTF("eeprom state machine wrong state! "
549 			"0x%x 0x%x 0x%x 0x%x",
550 			sc->nvm_mode, sc->nvm_bits,
551 			sc->nvm_opaddr, sc->nvm_data);
552 	}
553 }
554 
555 static void
556 e82545_itr_callback(int fd __unused, enum ev_type type __unused, void *param)
557 {
558 	uint32_t new;
559 	struct e82545_softc *sc = param;
560 
561 	pthread_mutex_lock(&sc->esc_mtx);
562 	new = sc->esc_ICR & sc->esc_IMS;
563 	if (new && !sc->esc_irq_asserted) {
564 		DPRINTF("itr callback: lintr assert %x", new);
565 		sc->esc_irq_asserted = 1;
566 		pci_lintr_assert(sc->esc_pi);
567 	} else {
568 		mevent_delete(sc->esc_mevpitr);
569 		sc->esc_mevpitr = NULL;
570 	}
571 	pthread_mutex_unlock(&sc->esc_mtx);
572 }
573 
574 static void
575 e82545_icr_assert(struct e82545_softc *sc, uint32_t bits)
576 {
577 	uint32_t new;
578 
579 	DPRINTF("icr assert: 0x%x", bits);
580 
581 	/*
582 	 * An interrupt is only generated if bits are set that
583 	 * aren't already in the ICR, these bits are unmasked,
584 	 * and there isn't an interrupt already pending.
585 	 */
586 	new = bits & ~sc->esc_ICR & sc->esc_IMS;
587 	sc->esc_ICR |= bits;
588 
589 	if (new == 0) {
590 		DPRINTF("icr assert: masked %x, ims %x", new, sc->esc_IMS);
591 	} else if (sc->esc_mevpitr != NULL) {
592 		DPRINTF("icr assert: throttled %x, ims %x", new, sc->esc_IMS);
593 	} else if (!sc->esc_irq_asserted) {
594 		DPRINTF("icr assert: lintr assert %x", new);
595 		sc->esc_irq_asserted = 1;
596 		pci_lintr_assert(sc->esc_pi);
597 		if (sc->esc_ITR != 0) {
598 			sc->esc_mevpitr = mevent_add(
599 			    (sc->esc_ITR + 3905) / 3906,  /* 256ns -> 1ms */
600 			    EVF_TIMER, e82545_itr_callback, sc);
601 		}
602 	}
603 }
604 
605 static void
606 e82545_ims_change(struct e82545_softc *sc, uint32_t bits)
607 {
608 	uint32_t new;
609 
610 	/*
611 	 * Changing the mask may allow previously asserted
612 	 * but masked interrupt requests to generate an interrupt.
613 	 */
614 	new = bits & sc->esc_ICR & ~sc->esc_IMS;
615 	sc->esc_IMS |= bits;
616 
617 	if (new == 0) {
618 		DPRINTF("ims change: masked %x, ims %x", new, sc->esc_IMS);
619 	} else if (sc->esc_mevpitr != NULL) {
620 		DPRINTF("ims change: throttled %x, ims %x", new, sc->esc_IMS);
621 	} else if (!sc->esc_irq_asserted) {
622 		DPRINTF("ims change: lintr assert %x", new);
623 		sc->esc_irq_asserted = 1;
624 		pci_lintr_assert(sc->esc_pi);
625 		if (sc->esc_ITR != 0) {
626 			sc->esc_mevpitr = mevent_add(
627 			    (sc->esc_ITR + 3905) / 3906,  /* 256ns -> 1ms */
628 			    EVF_TIMER, e82545_itr_callback, sc);
629 		}
630 	}
631 }
632 
633 static void
634 e82545_icr_deassert(struct e82545_softc *sc, uint32_t bits)
635 {
636 
637 	DPRINTF("icr deassert: 0x%x", bits);
638 	sc->esc_ICR &= ~bits;
639 
640 	/*
641 	 * If there are no longer any interrupt sources and there
642 	 * was an asserted interrupt, clear it
643 	 */
644 	if (sc->esc_irq_asserted && !(sc->esc_ICR & sc->esc_IMS)) {
645 		DPRINTF("icr deassert: lintr deassert %x", bits);
646 		pci_lintr_deassert(sc->esc_pi);
647 		sc->esc_irq_asserted = 0;
648 	}
649 }
650 
651 static void
652 e82545_intr_write(struct e82545_softc *sc, uint32_t offset, uint32_t value)
653 {
654 
655 	DPRINTF("intr_write: off %x, val %x", offset, value);
656 
657 	switch (offset) {
658 	case E1000_ICR:
659 		e82545_icr_deassert(sc, value);
660 		break;
661 	case E1000_ITR:
662 		sc->esc_ITR = value;
663 		break;
664 	case E1000_ICS:
665 		sc->esc_ICS = value;	/* not used: store for debug */
666 		e82545_icr_assert(sc, value);
667 		break;
668 	case E1000_IMS:
669 		e82545_ims_change(sc, value);
670 		break;
671 	case E1000_IMC:
672 		sc->esc_IMC = value;	/* for debug */
673 		sc->esc_IMS &= ~value;
674 		// XXX clear interrupts if all ICR bits now masked
675 		// and interrupt was pending ?
676 		break;
677 	default:
678 		break;
679 	}
680 }
681 
682 static uint32_t
683 e82545_intr_read(struct e82545_softc *sc, uint32_t offset)
684 {
685 	uint32_t retval;
686 
687 	retval = 0;
688 
689 	DPRINTF("intr_read: off %x", offset);
690 
691 	switch (offset) {
692 	case E1000_ICR:
693 		retval = sc->esc_ICR;
694 		sc->esc_ICR = 0;
695 		e82545_icr_deassert(sc, ~0);
696 		break;
697 	case E1000_ITR:
698 		retval = sc->esc_ITR;
699 		break;
700 	case E1000_ICS:
701 		/* write-only register */
702 		break;
703 	case E1000_IMS:
704 		retval = sc->esc_IMS;
705 		break;
706 	case E1000_IMC:
707 		/* write-only register */
708 		break;
709 	default:
710 		break;
711 	}
712 
713 	return (retval);
714 }
715 
716 static void
717 e82545_devctl(struct e82545_softc *sc, uint32_t val)
718 {
719 
720 	sc->esc_CTRL = val & ~E1000_CTRL_RST;
721 
722 	if (val & E1000_CTRL_RST) {
723 		DPRINTF("e1k: s/w reset, ctl %x", val);
724 		e82545_reset(sc, 1);
725 	}
726 	/* XXX check for phy reset ? */
727 }
728 
729 static void
730 e82545_rx_update_rdba(struct e82545_softc *sc)
731 {
732 
733 	/* XXX verify desc base/len within phys mem range */
734 	sc->esc_rdba = (uint64_t)sc->esc_RDBAH << 32 |
735 	    sc->esc_RDBAL;
736 
737 	/* Cache host mapping of guest descriptor array */
738 	sc->esc_rxdesc = paddr_guest2host(sc->esc_ctx,
739 	    sc->esc_rdba, sc->esc_RDLEN);
740 }
741 
742 static void
743 e82545_rx_ctl(struct e82545_softc *sc, uint32_t val)
744 {
745 	int on;
746 
747 	on = ((val & E1000_RCTL_EN) == E1000_RCTL_EN);
748 
749 	/* Save RCTL after stripping reserved bits 31:27,24,21,14,11:10,0 */
750 	sc->esc_RCTL = val & ~0xF9204c01;
751 
752 	DPRINTF("rx_ctl - %s RCTL %x, val %x",
753 		on ? "on" : "off", sc->esc_RCTL, val);
754 
755 	/* state change requested */
756 	if (on != sc->esc_rx_enabled) {
757 		if (on) {
758 			/* Catch disallowed/unimplemented settings */
759 			//assert(!(val & E1000_RCTL_LBM_TCVR));
760 
761 			if (sc->esc_RCTL & E1000_RCTL_LBM_TCVR) {
762 				sc->esc_rx_loopback = 1;
763 			} else {
764 				sc->esc_rx_loopback = 0;
765 			}
766 
767 			e82545_rx_update_rdba(sc);
768 			e82545_rx_enable(sc);
769 		} else {
770 			e82545_rx_disable(sc);
771 			sc->esc_rx_loopback = 0;
772 			sc->esc_rdba = 0;
773 			sc->esc_rxdesc = NULL;
774 		}
775 	}
776 }
777 
778 static void
779 e82545_tx_update_tdba(struct e82545_softc *sc)
780 {
781 
782 	/* XXX verify desc base/len within phys mem range */
783 	sc->esc_tdba = (uint64_t)sc->esc_TDBAH << 32 | sc->esc_TDBAL;
784 
785 	/* Cache host mapping of guest descriptor array */
786 	sc->esc_txdesc = paddr_guest2host(sc->esc_ctx, sc->esc_tdba,
787             sc->esc_TDLEN);
788 }
789 
790 static void
791 e82545_tx_ctl(struct e82545_softc *sc, uint32_t val)
792 {
793 	int on;
794 
795 	on = ((val & E1000_TCTL_EN) == E1000_TCTL_EN);
796 
797 	/* ignore TCTL_EN settings that don't change state */
798 	if (on == sc->esc_tx_enabled)
799 		return;
800 
801 	if (on) {
802 		e82545_tx_update_tdba(sc);
803 		e82545_tx_enable(sc);
804 	} else {
805 		e82545_tx_disable(sc);
806 		sc->esc_tdba = 0;
807 		sc->esc_txdesc = NULL;
808 	}
809 
810 	/* Save TCTL value after stripping reserved bits 31:25,23,2,0 */
811 	sc->esc_TCTL = val & ~0xFE800005;
812 }
813 
814 static int
815 e82545_bufsz(uint32_t rctl)
816 {
817 
818 	switch (rctl & (E1000_RCTL_BSEX | E1000_RCTL_SZ_256)) {
819 	case (E1000_RCTL_SZ_2048): return (2048);
820 	case (E1000_RCTL_SZ_1024): return (1024);
821 	case (E1000_RCTL_SZ_512): return (512);
822 	case (E1000_RCTL_SZ_256): return (256);
823 	case (E1000_RCTL_BSEX|E1000_RCTL_SZ_16384): return (16384);
824 	case (E1000_RCTL_BSEX|E1000_RCTL_SZ_8192): return (8192);
825 	case (E1000_RCTL_BSEX|E1000_RCTL_SZ_4096): return (4096);
826 	}
827 	return (256);	/* Forbidden value. */
828 }
829 
830 /* XXX one packet at a time until this is debugged */
831 static void
832 e82545_rx_callback(int fd __unused, enum ev_type type __unused, void *param)
833 {
834 	struct e82545_softc *sc = param;
835 	struct e1000_rx_desc *rxd;
836 	struct iovec vec[64];
837 	ssize_t len;
838 	int left, lim, maxpktsz, maxpktdesc, bufsz, i, n, size;
839 	uint32_t cause = 0;
840 	uint16_t *tp, tag, head;
841 
842 	pthread_mutex_lock(&sc->esc_mtx);
843 	DPRINTF("rx_run: head %x, tail %x", sc->esc_RDH, sc->esc_RDT);
844 
845 	if (!sc->esc_rx_enabled || sc->esc_rx_loopback) {
846 		DPRINTF("rx disabled (!%d || %d) -- packet(s) dropped",
847 		    sc->esc_rx_enabled, sc->esc_rx_loopback);
848 		while (netbe_rx_discard(sc->esc_be) > 0) {
849 		}
850 		goto done1;
851 	}
852 	bufsz = e82545_bufsz(sc->esc_RCTL);
853 	maxpktsz = (sc->esc_RCTL & E1000_RCTL_LPE) ? 16384 : 1522;
854 	maxpktdesc = (maxpktsz + bufsz - 1) / bufsz;
855 	size = sc->esc_RDLEN / 16;
856 	head = sc->esc_RDH;
857 	left = (size + sc->esc_RDT - head) % size;
858 	if (left < maxpktdesc) {
859 		DPRINTF("rx overflow (%d < %d) -- packet(s) dropped",
860 		    left, maxpktdesc);
861 		while (netbe_rx_discard(sc->esc_be) > 0) {
862 		}
863 		goto done1;
864 	}
865 
866 	sc->esc_rx_active = 1;
867 	pthread_mutex_unlock(&sc->esc_mtx);
868 
869 	for (lim = size / 4; lim > 0 && left >= maxpktdesc; lim -= n) {
870 
871 		/* Grab rx descriptor pointed to by the head pointer */
872 		for (i = 0; i < maxpktdesc; i++) {
873 			rxd = &sc->esc_rxdesc[(head + i) % size];
874 			vec[i].iov_base = paddr_guest2host(sc->esc_ctx,
875 			    rxd->buffer_addr, bufsz);
876 			vec[i].iov_len = bufsz;
877 		}
878 		len = netbe_recv(sc->esc_be, vec, maxpktdesc);
879 		if (len <= 0) {
880 			DPRINTF("netbe_recv() returned %zd", len);
881 			goto done;
882 		}
883 
884 		/*
885 		 * Adjust the packet length based on whether the CRC needs
886 		 * to be stripped or if the packet is less than the minimum
887 		 * eth packet size.
888 		 */
889 		if (len < ETHER_MIN_LEN - ETHER_CRC_LEN)
890 			len = ETHER_MIN_LEN - ETHER_CRC_LEN;
891 		if (!(sc->esc_RCTL & E1000_RCTL_SECRC))
892 			len += ETHER_CRC_LEN;
893 		n = (len + bufsz - 1) / bufsz;
894 
895 		DPRINTF("packet read %zd bytes, %d segs, head %d",
896 		    len, n, head);
897 
898 		/* Apply VLAN filter. */
899 		tp = (uint16_t *)vec[0].iov_base + 6;
900 		if ((sc->esc_RCTL & E1000_RCTL_VFE) &&
901 		    (ntohs(tp[0]) == sc->esc_VET)) {
902 			tag = ntohs(tp[1]) & 0x0fff;
903 			if ((sc->esc_fvlan[tag >> 5] &
904 			    (1 << (tag & 0x1f))) != 0) {
905 				DPRINTF("known VLAN %d", tag);
906 			} else {
907 				DPRINTF("unknown VLAN %d", tag);
908 				n = 0;
909 				continue;
910 			}
911 		}
912 
913 		/* Update all consumed descriptors. */
914 		for (i = 0; i < n - 1; i++) {
915 			rxd = &sc->esc_rxdesc[(head + i) % size];
916 			rxd->length = bufsz;
917 			rxd->csum = 0;
918 			rxd->errors = 0;
919 			rxd->special = 0;
920 			rxd->status = E1000_RXD_STAT_DD;
921 		}
922 		rxd = &sc->esc_rxdesc[(head + i) % size];
923 		rxd->length = len % bufsz;
924 		rxd->csum = 0;
925 		rxd->errors = 0;
926 		rxd->special = 0;
927 		/* XXX signal no checksum for now */
928 		rxd->status = E1000_RXD_STAT_PIF | E1000_RXD_STAT_IXSM |
929 		    E1000_RXD_STAT_EOP | E1000_RXD_STAT_DD;
930 
931 		/* Schedule receive interrupts. */
932 		if ((uint32_t)len <= sc->esc_RSRPD) {
933 			cause |= E1000_ICR_SRPD | E1000_ICR_RXT0;
934 		} else {
935 			/* XXX: RDRT and RADV timers should be here. */
936 			cause |= E1000_ICR_RXT0;
937 		}
938 
939 		head = (head + n) % size;
940 		left -= n;
941 	}
942 
943 done:
944 	pthread_mutex_lock(&sc->esc_mtx);
945 	sc->esc_rx_active = 0;
946 	if (sc->esc_rx_enabled == 0)
947 		pthread_cond_signal(&sc->esc_rx_cond);
948 
949 	sc->esc_RDH = head;
950 	/* Respect E1000_RCTL_RDMTS */
951 	left = (size + sc->esc_RDT - head) % size;
952 	if (left < (size >> (((sc->esc_RCTL >> 8) & 3) + 1)))
953 		cause |= E1000_ICR_RXDMT0;
954 	/* Assert all accumulated interrupts. */
955 	if (cause != 0)
956 		e82545_icr_assert(sc, cause);
957 done1:
958 	DPRINTF("rx_run done: head %x, tail %x", sc->esc_RDH, sc->esc_RDT);
959 	pthread_mutex_unlock(&sc->esc_mtx);
960 }
961 
962 static uint16_t
963 e82545_carry(uint32_t sum)
964 {
965 
966 	sum = (sum & 0xFFFF) + (sum >> 16);
967 	if (sum > 0xFFFF)
968 		sum -= 0xFFFF;
969 	return (sum);
970 }
971 
972 static uint16_t
973 e82545_buf_checksum(uint8_t *buf, int len)
974 {
975 	int i;
976 	uint32_t sum = 0;
977 
978 	/* Checksum all the pairs of bytes first... */
979 	for (i = 0; i < (len & ~1); i += 2)
980 		sum += *((u_int16_t *)(buf + i));
981 
982 	/*
983 	 * If there's a single byte left over, checksum it, too.
984 	 * Network byte order is big-endian, so the remaining byte is
985 	 * the high byte.
986 	 */
987 	if (i < len)
988 		sum += htons(buf[i] << 8);
989 
990 	return (e82545_carry(sum));
991 }
992 
993 static uint16_t
994 e82545_iov_checksum(struct iovec *iov, int iovcnt, unsigned int off,
995     unsigned int len)
996 {
997 	unsigned int now, odd;
998 	uint32_t sum = 0, s;
999 
1000 	/* Skip completely unneeded vectors. */
1001 	while (iovcnt > 0 && iov->iov_len <= off && off > 0) {
1002 		off -= iov->iov_len;
1003 		iov++;
1004 		iovcnt--;
1005 	}
1006 
1007 	/* Calculate checksum of requested range. */
1008 	odd = 0;
1009 	while (len > 0 && iovcnt > 0) {
1010 		now = MIN(len, iov->iov_len - off);
1011 		s = e82545_buf_checksum((uint8_t *)iov->iov_base + off, now);
1012 		sum += odd ? (s << 8) : s;
1013 		odd ^= (now & 1);
1014 		len -= now;
1015 		off = 0;
1016 		iov++;
1017 		iovcnt--;
1018 	}
1019 
1020 	return (e82545_carry(sum));
1021 }
1022 
1023 /*
1024  * Return the transmit descriptor type.
1025  */
1026 static int
1027 e82545_txdesc_type(uint32_t lower)
1028 {
1029 	int type;
1030 
1031 	type = 0;
1032 
1033 	if (lower & E1000_TXD_CMD_DEXT)
1034 		type = lower & E1000_TXD_MASK;
1035 
1036 	return (type);
1037 }
1038 
1039 static void
1040 e82545_transmit_checksum(struct iovec *iov, int iovcnt, struct ck_info *ck)
1041 {
1042 	uint16_t cksum;
1043 	unsigned int cklen;
1044 
1045 	DPRINTF("tx cksum: iovcnt/s/off/len %d/%d/%d/%d",
1046 	    iovcnt, ck->ck_start, ck->ck_off, ck->ck_len);
1047 	cklen = ck->ck_len ? ck->ck_len - ck->ck_start + 1U : UINT_MAX;
1048 	cksum = e82545_iov_checksum(iov, iovcnt, ck->ck_start, cklen);
1049 	*(uint16_t *)((uint8_t *)iov[0].iov_base + ck->ck_off) = ~cksum;
1050 }
1051 
1052 static void
1053 e82545_transmit_backend(struct e82545_softc *sc, struct iovec *iov, int iovcnt)
1054 {
1055 
1056 	if (sc->esc_be == NULL)
1057 		return;
1058 
1059 	(void) netbe_send(sc->esc_be, iov, iovcnt);
1060 }
1061 
1062 static void
1063 e82545_transmit_done(struct e82545_softc *sc, uint16_t head, uint16_t tail,
1064     uint16_t dsize, int *tdwb)
1065 {
1066 	union e1000_tx_udesc *dsc;
1067 
1068 	for ( ; head != tail; head = (head + 1) % dsize) {
1069 		dsc = &sc->esc_txdesc[head];
1070 		if (dsc->td.lower.data & E1000_TXD_CMD_RS) {
1071 			dsc->td.upper.data |= E1000_TXD_STAT_DD;
1072 			*tdwb = 1;
1073 		}
1074 	}
1075 }
1076 
1077 static int
1078 e82545_transmit(struct e82545_softc *sc, uint16_t head, uint16_t tail,
1079     uint16_t dsize, uint16_t *rhead, int *tdwb)
1080 {
1081 	uint8_t *hdr, *hdrp;
1082 	struct iovec iovb[I82545_MAX_TXSEGS + 2];
1083 	struct iovec tiov[I82545_MAX_TXSEGS + 2];
1084 	struct e1000_context_desc *cd;
1085 	struct ck_info ckinfo[2];
1086 	struct iovec *iov;
1087 	union  e1000_tx_udesc *dsc;
1088 	int desc, dtype, ntype, iovcnt, tcp, tso, paylen, seg, tiovcnt, pv;
1089 	unsigned hdrlen, vlen, pktlen, len, left, mss, now, nnow, nleft, pvoff;
1090 	uint32_t tcpsum, tcpseq;
1091 	uint16_t ipcs, tcpcs, ipid, ohead;
1092 	bool invalid;
1093 
1094 	ckinfo[0].ck_valid = ckinfo[1].ck_valid = 0;
1095 	iovcnt = 0;
1096 	ntype = 0;
1097 	tso = 0;
1098 	pktlen = 0;
1099 	ohead = head;
1100 	invalid = false;
1101 
1102 	/* iovb[0/1] may be used for writable copy of headers. */
1103 	iov = &iovb[2];
1104 
1105 	for (desc = 0; ; desc++, head = (head + 1) % dsize) {
1106 		if (head == tail) {
1107 			*rhead = head;
1108 			return (0);
1109 		}
1110 		dsc = &sc->esc_txdesc[head];
1111 		dtype = e82545_txdesc_type(dsc->td.lower.data);
1112 
1113 		if (desc == 0) {
1114 			switch (dtype) {
1115 			case E1000_TXD_TYP_C:
1116 				DPRINTF("tx ctxt desc idx %d: %016jx "
1117 				    "%08x%08x",
1118 				    head, dsc->td.buffer_addr,
1119 				    dsc->td.upper.data, dsc->td.lower.data);
1120 				/* Save context and return */
1121 				sc->esc_txctx = dsc->cd;
1122 				goto done;
1123 			case E1000_TXD_TYP_L:
1124 				DPRINTF("tx legacy desc idx %d: %08x%08x",
1125 				    head, dsc->td.upper.data, dsc->td.lower.data);
1126 				/*
1127 				 * legacy cksum start valid in first descriptor
1128 				 */
1129 				ntype = dtype;
1130 				ckinfo[0].ck_start = dsc->td.upper.fields.css;
1131 				break;
1132 			case E1000_TXD_TYP_D:
1133 				DPRINTF("tx data desc idx %d: %08x%08x",
1134 				    head, dsc->td.upper.data, dsc->td.lower.data);
1135 				ntype = dtype;
1136 				break;
1137 			default:
1138 				break;
1139 			}
1140 		} else {
1141 			/* Descriptor type must be consistent */
1142 			assert(dtype == ntype);
1143 			DPRINTF("tx next desc idx %d: %08x%08x",
1144 			    head, dsc->td.upper.data, dsc->td.lower.data);
1145 		}
1146 
1147 		len = (dtype == E1000_TXD_TYP_L) ? dsc->td.lower.flags.length :
1148 		    dsc->dd.lower.data & 0xFFFFF;
1149 
1150 		/* Strip checksum supplied by guest. */
1151 		if ((dsc->td.lower.data & E1000_TXD_CMD_EOP) != 0 &&
1152 		    (dsc->td.lower.data & E1000_TXD_CMD_IFCS) == 0) {
1153 			if (len <= 2) {
1154 				WPRINTF("final descriptor too short (%d) -- dropped",
1155 				    len);
1156 				invalid = true;
1157 			} else
1158 				len -= 2;
1159 		}
1160 
1161 		if (len > 0 && iovcnt < I82545_MAX_TXSEGS) {
1162 			iov[iovcnt].iov_base = paddr_guest2host(sc->esc_ctx,
1163 			    dsc->td.buffer_addr, len);
1164 			iov[iovcnt].iov_len = len;
1165 			iovcnt++;
1166 			pktlen += len;
1167 		}
1168 
1169 		/*
1170 		 * Pull out info that is valid in the final descriptor
1171 		 * and exit descriptor loop.
1172 		 */
1173 		if (dsc->td.lower.data & E1000_TXD_CMD_EOP) {
1174 			if (dtype == E1000_TXD_TYP_L) {
1175 				if (dsc->td.lower.data & E1000_TXD_CMD_IC) {
1176 					ckinfo[0].ck_valid = 1;
1177 					ckinfo[0].ck_off =
1178 					    dsc->td.lower.flags.cso;
1179 					ckinfo[0].ck_len = 0;
1180 				}
1181 			} else {
1182 				cd = &sc->esc_txctx;
1183 				if (dsc->dd.lower.data & E1000_TXD_CMD_TSE)
1184 					tso = 1;
1185 				if (dsc->dd.upper.fields.popts &
1186 				    E1000_TXD_POPTS_IXSM)
1187 					ckinfo[0].ck_valid = 1;
1188 				if (dsc->dd.upper.fields.popts &
1189 				    E1000_TXD_POPTS_IXSM || tso) {
1190 					ckinfo[0].ck_start =
1191 					    cd->lower_setup.ip_fields.ipcss;
1192 					ckinfo[0].ck_off =
1193 					    cd->lower_setup.ip_fields.ipcso;
1194 					ckinfo[0].ck_len =
1195 					    cd->lower_setup.ip_fields.ipcse;
1196 				}
1197 				if (dsc->dd.upper.fields.popts &
1198 				    E1000_TXD_POPTS_TXSM)
1199 					ckinfo[1].ck_valid = 1;
1200 				if (dsc->dd.upper.fields.popts &
1201 				    E1000_TXD_POPTS_TXSM || tso) {
1202 					ckinfo[1].ck_start =
1203 					    cd->upper_setup.tcp_fields.tucss;
1204 					ckinfo[1].ck_off =
1205 					    cd->upper_setup.tcp_fields.tucso;
1206 					ckinfo[1].ck_len =
1207 					    cd->upper_setup.tcp_fields.tucse;
1208 				}
1209 			}
1210 			break;
1211 		}
1212 	}
1213 
1214 	if (invalid)
1215 		goto done;
1216 
1217 	if (iovcnt > I82545_MAX_TXSEGS) {
1218 		WPRINTF("tx too many descriptors (%d > %d) -- dropped",
1219 		    iovcnt, I82545_MAX_TXSEGS);
1220 		goto done;
1221 	}
1222 
1223 	hdrlen = vlen = 0;
1224 	/* Estimate writable space for VLAN header insertion. */
1225 	if ((sc->esc_CTRL & E1000_CTRL_VME) &&
1226 	    (dsc->td.lower.data & E1000_TXD_CMD_VLE)) {
1227 		hdrlen = ETHER_ADDR_LEN*2;
1228 		vlen = ETHER_VLAN_ENCAP_LEN;
1229 	}
1230 	if (!tso) {
1231 		/* Estimate required writable space for checksums. */
1232 		if (ckinfo[0].ck_valid)
1233 			hdrlen = MAX(hdrlen, ckinfo[0].ck_off + 2U);
1234 		if (ckinfo[1].ck_valid)
1235 			hdrlen = MAX(hdrlen, ckinfo[1].ck_off + 2U);
1236 		/* Round up writable space to the first vector. */
1237 		if (hdrlen != 0 && iov[0].iov_len > hdrlen &&
1238 		    iov[0].iov_len < hdrlen + 100)
1239 			hdrlen = iov[0].iov_len;
1240 	} else {
1241 		/* In case of TSO header length provided by software. */
1242 		hdrlen = sc->esc_txctx.tcp_seg_setup.fields.hdr_len;
1243 
1244 		/*
1245 		 * Cap the header length at 240 based on 7.2.4.5 of
1246 		 * the Intel 82576EB (Rev 2.63) datasheet.
1247 		 */
1248 		if (hdrlen > 240) {
1249 			WPRINTF("TSO hdrlen too large: %d", hdrlen);
1250 			goto done;
1251 		}
1252 
1253 		/*
1254 		 * If VLAN insertion is requested, ensure the header
1255 		 * at least holds the amount of data copied during
1256 		 * VLAN insertion below.
1257 		 *
1258 		 * XXX: Realistic packets will include a full Ethernet
1259 		 * header before the IP header at ckinfo[0].ck_start,
1260 		 * but this check is sufficient to prevent
1261 		 * out-of-bounds access below.
1262 		 */
1263 		if (vlen != 0 && hdrlen < ETHER_ADDR_LEN*2) {
1264 			WPRINTF("TSO hdrlen too small for vlan insertion "
1265 			    "(%d vs %d) -- dropped", hdrlen,
1266 			    ETHER_ADDR_LEN*2);
1267 			goto done;
1268 		}
1269 
1270 		/*
1271 		 * Ensure that the header length covers the used fields
1272 		 * in the IP and TCP headers as well as the IP and TCP
1273 		 * checksums.  The following fields are accessed below:
1274 		 *
1275 		 * Header | Field | Offset | Length
1276 		 * -------+-------+--------+-------
1277 		 * IPv4   | len   | 2      | 2
1278 		 * IPv4   | ID    | 4      | 2
1279 		 * IPv6   | len   | 4      | 2
1280 		 * TCP    | seq # | 4      | 4
1281 		 * TCP    | flags | 13     | 1
1282 		 * UDP    | len   | 4      | 4
1283 		 */
1284 		if (hdrlen < ckinfo[0].ck_start + 6U ||
1285 		    hdrlen < ckinfo[0].ck_off + 2U) {
1286 			WPRINTF("TSO hdrlen too small for IP fields (%d) "
1287 			    "-- dropped", hdrlen);
1288 			goto done;
1289 		}
1290 		if (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_TCP) {
1291 			if (hdrlen < ckinfo[1].ck_start + 14U) {
1292 				WPRINTF("TSO hdrlen too small for TCP fields "
1293 				    "(%d) -- dropped", hdrlen);
1294 				goto done;
1295 			}
1296 		} else {
1297 			if (hdrlen < ckinfo[1].ck_start + 8U) {
1298 				WPRINTF("TSO hdrlen too small for UDP fields "
1299 				    "(%d) -- dropped", hdrlen);
1300 				goto done;
1301 			}
1302 		}
1303 		if (ckinfo[1].ck_valid && hdrlen < ckinfo[1].ck_off + 2U) {
1304 			WPRINTF("TSO hdrlen too small for TCP/UDP fields "
1305 			    "(%d) -- dropped", hdrlen);
1306 			goto done;
1307 		}
1308 	}
1309 
1310 	if (pktlen < hdrlen + vlen) {
1311 		WPRINTF("packet too small for writable header");
1312 		goto done;
1313 	}
1314 
1315 	/* Allocate, fill and prepend writable header vector. */
1316 	if (hdrlen + vlen != 0) {
1317 		hdr = __builtin_alloca(hdrlen + vlen);
1318 		hdr += vlen;
1319 		for (left = hdrlen, hdrp = hdr; left > 0;
1320 		    left -= now, hdrp += now) {
1321 			now = MIN(left, iov->iov_len);
1322 			memcpy(hdrp, iov->iov_base, now);
1323 			iov->iov_base = (uint8_t *)iov->iov_base + now;
1324 			iov->iov_len -= now;
1325 			if (iov->iov_len == 0) {
1326 				iov++;
1327 				iovcnt--;
1328 			}
1329 		}
1330 		iov--;
1331 		iovcnt++;
1332 		iov->iov_base = hdr;
1333 		iov->iov_len = hdrlen;
1334 	} else
1335 		hdr = NULL;
1336 
1337 	/* Insert VLAN tag. */
1338 	if (vlen != 0) {
1339 		hdr -= ETHER_VLAN_ENCAP_LEN;
1340 		memmove(hdr, hdr + ETHER_VLAN_ENCAP_LEN, ETHER_ADDR_LEN*2);
1341 		hdrlen += ETHER_VLAN_ENCAP_LEN;
1342 		hdr[ETHER_ADDR_LEN*2 + 0] = sc->esc_VET >> 8;
1343 		hdr[ETHER_ADDR_LEN*2 + 1] = sc->esc_VET & 0xff;
1344 		hdr[ETHER_ADDR_LEN*2 + 2] = dsc->td.upper.fields.special >> 8;
1345 		hdr[ETHER_ADDR_LEN*2 + 3] = dsc->td.upper.fields.special & 0xff;
1346 		iov->iov_base = hdr;
1347 		iov->iov_len += ETHER_VLAN_ENCAP_LEN;
1348 		/* Correct checksum offsets after VLAN tag insertion. */
1349 		ckinfo[0].ck_start += ETHER_VLAN_ENCAP_LEN;
1350 		ckinfo[0].ck_off += ETHER_VLAN_ENCAP_LEN;
1351 		if (ckinfo[0].ck_len != 0)
1352 			ckinfo[0].ck_len += ETHER_VLAN_ENCAP_LEN;
1353 		ckinfo[1].ck_start += ETHER_VLAN_ENCAP_LEN;
1354 		ckinfo[1].ck_off += ETHER_VLAN_ENCAP_LEN;
1355 		if (ckinfo[1].ck_len != 0)
1356 			ckinfo[1].ck_len += ETHER_VLAN_ENCAP_LEN;
1357 	}
1358 
1359 	/* Simple non-TSO case. */
1360 	if (!tso) {
1361 		/* Calculate checksums and transmit. */
1362 		if (ckinfo[0].ck_valid)
1363 			e82545_transmit_checksum(iov, iovcnt, &ckinfo[0]);
1364 		if (ckinfo[1].ck_valid)
1365 			e82545_transmit_checksum(iov, iovcnt, &ckinfo[1]);
1366 		e82545_transmit_backend(sc, iov, iovcnt);
1367 		goto done;
1368 	}
1369 
1370 	/* Doing TSO. */
1371 	tcp = (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_TCP) != 0;
1372 	mss = sc->esc_txctx.tcp_seg_setup.fields.mss;
1373 	paylen = (sc->esc_txctx.cmd_and_length & 0x000fffff);
1374 	DPRINTF("tx %s segmentation offload %d+%d/%u bytes %d iovs",
1375 	    tcp ? "TCP" : "UDP", hdrlen, paylen, mss, iovcnt);
1376 	ipid = ntohs(*(uint16_t *)&hdr[ckinfo[0].ck_start + 4]);
1377 	tcpseq = 0;
1378 	if (tcp)
1379 		tcpseq = ntohl(*(uint32_t *)&hdr[ckinfo[1].ck_start + 4]);
1380 	ipcs = *(uint16_t *)&hdr[ckinfo[0].ck_off];
1381 	tcpcs = 0;
1382 	if (ckinfo[1].ck_valid)	/* Save partial pseudo-header checksum. */
1383 		tcpcs = *(uint16_t *)&hdr[ckinfo[1].ck_off];
1384 	pv = 1;
1385 	pvoff = 0;
1386 	for (seg = 0, left = paylen; left > 0; seg++, left -= now) {
1387 		now = MIN(left, mss);
1388 
1389 		/* Construct IOVs for the segment. */
1390 		/* Include whole original header. */
1391 		tiov[0].iov_base = hdr;
1392 		tiov[0].iov_len = hdrlen;
1393 		tiovcnt = 1;
1394 		/* Include respective part of payload IOV. */
1395 		for (nleft = now; pv < iovcnt && nleft > 0; nleft -= nnow) {
1396 			nnow = MIN(nleft, iov[pv].iov_len - pvoff);
1397 			tiov[tiovcnt].iov_base = (uint8_t *)iov[pv].iov_base +
1398 			    pvoff;
1399 			tiov[tiovcnt++].iov_len = nnow;
1400 			if (pvoff + nnow == iov[pv].iov_len) {
1401 				pv++;
1402 				pvoff = 0;
1403 			} else
1404 				pvoff += nnow;
1405 		}
1406 		DPRINTF("tx segment %d %d+%d bytes %d iovs",
1407 		    seg, hdrlen, now, tiovcnt);
1408 
1409 		/* Update IP header. */
1410 		if (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_IP) {
1411 			/* IPv4 -- set length and ID */
1412 			*(uint16_t *)&hdr[ckinfo[0].ck_start + 2] =
1413 			    htons(hdrlen - ckinfo[0].ck_start + now);
1414 			*(uint16_t *)&hdr[ckinfo[0].ck_start + 4] =
1415 			    htons(ipid + seg);
1416 		} else {
1417 			/* IPv6 -- set length */
1418 			*(uint16_t *)&hdr[ckinfo[0].ck_start + 4] =
1419 			    htons(hdrlen - ckinfo[0].ck_start - 40 +
1420 				  now);
1421 		}
1422 
1423 		/* Update pseudo-header checksum. */
1424 		tcpsum = tcpcs;
1425 		tcpsum += htons(hdrlen - ckinfo[1].ck_start + now);
1426 
1427 		/* Update TCP/UDP headers. */
1428 		if (tcp) {
1429 			/* Update sequence number and FIN/PUSH flags. */
1430 			*(uint32_t *)&hdr[ckinfo[1].ck_start + 4] =
1431 			    htonl(tcpseq + paylen - left);
1432 			if (now < left) {
1433 				hdr[ckinfo[1].ck_start + 13] &=
1434 				    ~(TH_FIN | TH_PUSH);
1435 			}
1436 		} else {
1437 			/* Update payload length. */
1438 			*(uint32_t *)&hdr[ckinfo[1].ck_start + 4] =
1439 			    hdrlen - ckinfo[1].ck_start + now;
1440 		}
1441 
1442 		/* Calculate checksums and transmit. */
1443 		if (ckinfo[0].ck_valid) {
1444 			*(uint16_t *)&hdr[ckinfo[0].ck_off] = ipcs;
1445 			e82545_transmit_checksum(tiov, tiovcnt, &ckinfo[0]);
1446 		}
1447 		if (ckinfo[1].ck_valid) {
1448 			*(uint16_t *)&hdr[ckinfo[1].ck_off] =
1449 			    e82545_carry(tcpsum);
1450 			e82545_transmit_checksum(tiov, tiovcnt, &ckinfo[1]);
1451 		}
1452 		e82545_transmit_backend(sc, tiov, tiovcnt);
1453 	}
1454 
1455 done:
1456 	head = (head + 1) % dsize;
1457 	e82545_transmit_done(sc, ohead, head, dsize, tdwb);
1458 
1459 	*rhead = head;
1460 	return (desc + 1);
1461 }
1462 
1463 static void
1464 e82545_tx_run(struct e82545_softc *sc)
1465 {
1466 	uint32_t cause;
1467 	uint16_t head, rhead, tail, size;
1468 	int lim, tdwb, sent;
1469 
1470 	size = sc->esc_TDLEN / 16;
1471 	if (size == 0)
1472 		return;
1473 
1474 	head = sc->esc_TDH % size;
1475 	tail = sc->esc_TDT % size;
1476 	DPRINTF("tx_run: head %x, rhead %x, tail %x",
1477 	    sc->esc_TDH, sc->esc_TDHr, sc->esc_TDT);
1478 
1479 	pthread_mutex_unlock(&sc->esc_mtx);
1480 	rhead = head;
1481 	tdwb = 0;
1482 	for (lim = size / 4; sc->esc_tx_enabled && lim > 0; lim -= sent) {
1483 		sent = e82545_transmit(sc, head, tail, size, &rhead, &tdwb);
1484 		if (sent == 0)
1485 			break;
1486 		head = rhead;
1487 	}
1488 	pthread_mutex_lock(&sc->esc_mtx);
1489 
1490 	sc->esc_TDH = head;
1491 	sc->esc_TDHr = rhead;
1492 	cause = 0;
1493 	if (tdwb)
1494 		cause |= E1000_ICR_TXDW;
1495 	if (lim != size / 4 && sc->esc_TDH == sc->esc_TDT)
1496 		cause |= E1000_ICR_TXQE;
1497 	if (cause)
1498 		e82545_icr_assert(sc, cause);
1499 
1500 	DPRINTF("tx_run done: head %x, rhead %x, tail %x",
1501 	    sc->esc_TDH, sc->esc_TDHr, sc->esc_TDT);
1502 }
1503 
1504 static _Noreturn void *
1505 e82545_tx_thread(void *param)
1506 {
1507 	struct e82545_softc *sc = param;
1508 
1509 	pthread_mutex_lock(&sc->esc_mtx);
1510 	for (;;) {
1511 		while (!sc->esc_tx_enabled || sc->esc_TDHr == sc->esc_TDT) {
1512 			if (sc->esc_tx_enabled && sc->esc_TDHr != sc->esc_TDT)
1513 				break;
1514 			sc->esc_tx_active = 0;
1515 			if (sc->esc_tx_enabled == 0)
1516 				pthread_cond_signal(&sc->esc_tx_cond);
1517 			pthread_cond_wait(&sc->esc_tx_cond, &sc->esc_mtx);
1518 		}
1519 		sc->esc_tx_active = 1;
1520 
1521 		/* Process some tx descriptors.  Lock dropped inside. */
1522 		e82545_tx_run(sc);
1523 	}
1524 }
1525 
1526 static void
1527 e82545_tx_start(struct e82545_softc *sc)
1528 {
1529 
1530 	if (sc->esc_tx_active == 0)
1531 		pthread_cond_signal(&sc->esc_tx_cond);
1532 }
1533 
1534 static void
1535 e82545_tx_enable(struct e82545_softc *sc)
1536 {
1537 
1538 	sc->esc_tx_enabled = 1;
1539 }
1540 
1541 static void
1542 e82545_tx_disable(struct e82545_softc *sc)
1543 {
1544 
1545 	sc->esc_tx_enabled = 0;
1546 	while (sc->esc_tx_active)
1547 		pthread_cond_wait(&sc->esc_tx_cond, &sc->esc_mtx);
1548 }
1549 
1550 static void
1551 e82545_rx_enable(struct e82545_softc *sc)
1552 {
1553 
1554 	sc->esc_rx_enabled = 1;
1555 }
1556 
1557 static void
1558 e82545_rx_disable(struct e82545_softc *sc)
1559 {
1560 
1561 	sc->esc_rx_enabled = 0;
1562 	while (sc->esc_rx_active)
1563 		pthread_cond_wait(&sc->esc_rx_cond, &sc->esc_mtx);
1564 }
1565 
1566 static void
1567 e82545_write_ra(struct e82545_softc *sc, int reg, uint32_t wval)
1568 {
1569 	struct eth_uni *eu;
1570 	int idx;
1571 
1572 	idx = reg >> 1;
1573 	assert(idx < 15);
1574 
1575 	eu = &sc->esc_uni[idx];
1576 
1577 	if (reg & 0x1) {
1578 		/* RAH */
1579 		eu->eu_valid = ((wval & E1000_RAH_AV) == E1000_RAH_AV);
1580 		eu->eu_addrsel = (wval >> 16) & 0x3;
1581 		eu->eu_eth.octet[5] = wval >> 8;
1582 		eu->eu_eth.octet[4] = wval;
1583 	} else {
1584 		/* RAL */
1585 		eu->eu_eth.octet[3] = wval >> 24;
1586 		eu->eu_eth.octet[2] = wval >> 16;
1587 		eu->eu_eth.octet[1] = wval >> 8;
1588 		eu->eu_eth.octet[0] = wval;
1589 	}
1590 }
1591 
1592 static uint32_t
1593 e82545_read_ra(struct e82545_softc *sc, int reg)
1594 {
1595 	struct eth_uni *eu;
1596 	uint32_t retval;
1597 	int idx;
1598 
1599 	idx = reg >> 1;
1600 	assert(idx < 15);
1601 
1602 	eu = &sc->esc_uni[idx];
1603 
1604 	if (reg & 0x1) {
1605 		/* RAH */
1606 		retval = (eu->eu_valid << 31) |
1607 			 (eu->eu_addrsel << 16) |
1608 			 (eu->eu_eth.octet[5] << 8) |
1609 			 eu->eu_eth.octet[4];
1610 	} else {
1611 		/* RAL */
1612 		retval = (eu->eu_eth.octet[3] << 24) |
1613 			 (eu->eu_eth.octet[2] << 16) |
1614 			 (eu->eu_eth.octet[1] << 8) |
1615 			 eu->eu_eth.octet[0];
1616 	}
1617 
1618 	return (retval);
1619 }
1620 
1621 static void
1622 e82545_write_register(struct e82545_softc *sc, uint32_t offset, uint32_t value)
1623 {
1624 	int ridx;
1625 
1626 	if (offset & 0x3) {
1627 		DPRINTF("Unaligned register write offset:0x%x value:0x%x", offset, value);
1628 		return;
1629 	}
1630 	DPRINTF("Register write: 0x%x value: 0x%x", offset, value);
1631 
1632 	switch (offset) {
1633 	case E1000_CTRL:
1634 	case E1000_CTRL_DUP:
1635 		e82545_devctl(sc, value);
1636 		break;
1637 	case E1000_FCAL:
1638 		sc->esc_FCAL = value;
1639 		break;
1640 	case E1000_FCAH:
1641 		sc->esc_FCAH = value & ~0xFFFF0000;
1642 		break;
1643 	case E1000_FCT:
1644 		sc->esc_FCT = value & ~0xFFFF0000;
1645 		break;
1646 	case E1000_VET:
1647 		sc->esc_VET = value & ~0xFFFF0000;
1648 		break;
1649 	case E1000_FCTTV:
1650 		sc->esc_FCTTV = value & ~0xFFFF0000;
1651 		break;
1652 	case E1000_LEDCTL:
1653 		sc->esc_LEDCTL = value & ~0x30303000;
1654 		break;
1655 	case E1000_PBA:
1656 		sc->esc_PBA = value & 0x0000FF80;
1657 		break;
1658 	case E1000_ICR:
1659 	case E1000_ITR:
1660 	case E1000_ICS:
1661 	case E1000_IMS:
1662 	case E1000_IMC:
1663 		e82545_intr_write(sc, offset, value);
1664 		break;
1665 	case E1000_RCTL:
1666 		e82545_rx_ctl(sc, value);
1667 		break;
1668 	case E1000_FCRTL:
1669 		sc->esc_FCRTL = value & ~0xFFFF0007;
1670 		break;
1671 	case E1000_FCRTH:
1672 		sc->esc_FCRTH = value & ~0xFFFF0007;
1673 		break;
1674 	case E1000_RDBAL(0):
1675 		sc->esc_RDBAL = value & ~0xF;
1676 		if (sc->esc_rx_enabled) {
1677 			/* Apparently legal: update cached address */
1678 			e82545_rx_update_rdba(sc);
1679 		}
1680 		break;
1681 	case E1000_RDBAH(0):
1682 		assert(!sc->esc_rx_enabled);
1683 		sc->esc_RDBAH = value;
1684 		break;
1685 	case E1000_RDLEN(0):
1686 		assert(!sc->esc_rx_enabled);
1687 		sc->esc_RDLEN = value & ~0xFFF0007F;
1688 		break;
1689 	case E1000_RDH(0):
1690 		/* XXX should only ever be zero ? Range check ? */
1691 		sc->esc_RDH = value;
1692 		break;
1693 	case E1000_RDT(0):
1694 		/* XXX if this opens up the rx ring, do something ? */
1695 		sc->esc_RDT = value;
1696 		break;
1697 	case E1000_RDTR:
1698 		/* ignore FPD bit 31 */
1699 		sc->esc_RDTR = value & ~0xFFFF0000;
1700 		break;
1701 	case E1000_RXDCTL(0):
1702 		sc->esc_RXDCTL = value & ~0xFEC0C0C0;
1703 		break;
1704 	case E1000_RADV:
1705 		sc->esc_RADV = value & ~0xFFFF0000;
1706 		break;
1707 	case E1000_RSRPD:
1708 		sc->esc_RSRPD = value & ~0xFFFFF000;
1709 		break;
1710 	case E1000_RXCSUM:
1711 		sc->esc_RXCSUM = value & ~0xFFFFF800;
1712 		break;
1713 	case E1000_TXCW:
1714 		sc->esc_TXCW = value & ~0x3FFF0000;
1715 		break;
1716 	case E1000_TCTL:
1717 		e82545_tx_ctl(sc, value);
1718 		break;
1719 	case E1000_TIPG:
1720 		sc->esc_TIPG = value;
1721 		break;
1722 	case E1000_AIT:
1723 		sc->esc_AIT = value;
1724 		break;
1725 	case E1000_TDBAL(0):
1726 		sc->esc_TDBAL = value & ~0xF;
1727 		if (sc->esc_tx_enabled)
1728 			e82545_tx_update_tdba(sc);
1729 		break;
1730 	case E1000_TDBAH(0):
1731 		sc->esc_TDBAH = value;
1732 		if (sc->esc_tx_enabled)
1733 			e82545_tx_update_tdba(sc);
1734 		break;
1735 	case E1000_TDLEN(0):
1736 		sc->esc_TDLEN = value & ~0xFFF0007F;
1737 		if (sc->esc_tx_enabled)
1738 			e82545_tx_update_tdba(sc);
1739 		break;
1740 	case E1000_TDH(0):
1741 		if (sc->esc_tx_enabled) {
1742 			WPRINTF("ignoring write to TDH while transmit enabled");
1743 			break;
1744 		}
1745 		if (value != 0) {
1746 			WPRINTF("ignoring non-zero value written to TDH");
1747 			break;
1748 		}
1749 		sc->esc_TDHr = sc->esc_TDH = value;
1750 		break;
1751 	case E1000_TDT(0):
1752 		sc->esc_TDT = value;
1753 		if (sc->esc_tx_enabled)
1754 			e82545_tx_start(sc);
1755 		break;
1756 	case E1000_TIDV:
1757 		sc->esc_TIDV = value & ~0xFFFF0000;
1758 		break;
1759 	case E1000_TXDCTL(0):
1760 		//assert(!sc->esc_tx_enabled);
1761 		sc->esc_TXDCTL = value & ~0xC0C0C0;
1762 		break;
1763 	case E1000_TADV:
1764 		sc->esc_TADV = value & ~0xFFFF0000;
1765 		break;
1766 	case E1000_RAL(0) ... E1000_RAH(15):
1767 		/* convert to u32 offset */
1768 		ridx = (offset - E1000_RAL(0)) >> 2;
1769 		e82545_write_ra(sc, ridx, value);
1770 		break;
1771 	case E1000_MTA ... (E1000_MTA + (127*4)):
1772 		sc->esc_fmcast[(offset - E1000_MTA) >> 2] = value;
1773 		break;
1774 	case E1000_VFTA ... (E1000_VFTA + (127*4)):
1775 		sc->esc_fvlan[(offset - E1000_VFTA) >> 2] = value;
1776 		break;
1777 	case E1000_EECD:
1778 	{
1779 		//DPRINTF("EECD write 0x%x -> 0x%x", sc->eeprom_control, value);
1780 		/* edge triggered low->high */
1781 		uint32_t eecd_strobe = ((sc->eeprom_control & E1000_EECD_SK) ?
1782 			0 : (value & E1000_EECD_SK));
1783 		uint32_t eecd_mask = (E1000_EECD_SK|E1000_EECD_CS|
1784 					E1000_EECD_DI|E1000_EECD_REQ);
1785 		sc->eeprom_control &= ~eecd_mask;
1786 		sc->eeprom_control |= (value & eecd_mask);
1787 		/* grant/revoke immediately */
1788 		if (value & E1000_EECD_REQ) {
1789 			sc->eeprom_control |= E1000_EECD_GNT;
1790 		} else {
1791                         sc->eeprom_control &= ~E1000_EECD_GNT;
1792 		}
1793 		if (eecd_strobe && (sc->eeprom_control & E1000_EECD_CS)) {
1794 			e82545_eecd_strobe(sc);
1795 		}
1796 		return;
1797 	}
1798 	case E1000_MDIC:
1799 	{
1800 		uint8_t reg_addr = (uint8_t)((value & E1000_MDIC_REG_MASK) >>
1801 						E1000_MDIC_REG_SHIFT);
1802 		uint8_t phy_addr = (uint8_t)((value & E1000_MDIC_PHY_MASK) >>
1803 						E1000_MDIC_PHY_SHIFT);
1804 		sc->mdi_control =
1805 			(value & ~(E1000_MDIC_ERROR|E1000_MDIC_DEST));
1806 		if ((value & E1000_MDIC_READY) != 0) {
1807 			DPRINTF("Incorrect MDIC ready bit: 0x%x", value);
1808 			return;
1809 		}
1810 		switch (value & E82545_MDIC_OP_MASK) {
1811 		case E1000_MDIC_OP_READ:
1812 			sc->mdi_control &= ~E82545_MDIC_DATA_MASK;
1813 			sc->mdi_control |= e82545_read_mdi(sc, reg_addr, phy_addr);
1814 			break;
1815 		case E1000_MDIC_OP_WRITE:
1816 			e82545_write_mdi(sc, reg_addr, phy_addr,
1817 				value & E82545_MDIC_DATA_MASK);
1818 			break;
1819 		default:
1820 			DPRINTF("Unknown MDIC op: 0x%x", value);
1821 			return;
1822 		}
1823 		/* TODO: barrier? */
1824 		sc->mdi_control |= E1000_MDIC_READY;
1825 		if (value & E82545_MDIC_IE) {
1826 			// TODO: generate interrupt
1827 		}
1828 		return;
1829 	}
1830 	case E1000_MANC:
1831 	case E1000_STATUS:
1832 		return;
1833 	default:
1834 		DPRINTF("Unknown write register: 0x%x value:%x", offset, value);
1835 		return;
1836 	}
1837 }
1838 
1839 static uint32_t
1840 e82545_read_register(struct e82545_softc *sc, uint32_t offset)
1841 {
1842 	uint32_t retval;
1843 	int ridx;
1844 
1845 	if (offset & 0x3) {
1846 		DPRINTF("Unaligned register read offset:0x%x", offset);
1847 		return 0;
1848 	}
1849 
1850 	DPRINTF("Register read: 0x%x", offset);
1851 
1852 	switch (offset) {
1853 	case E1000_CTRL:
1854 		retval = sc->esc_CTRL;
1855 		break;
1856 	case E1000_STATUS:
1857 		retval = E1000_STATUS_FD | E1000_STATUS_LU |
1858 		    E1000_STATUS_SPEED_1000;
1859 		break;
1860 	case E1000_FCAL:
1861 		retval = sc->esc_FCAL;
1862 		break;
1863 	case E1000_FCAH:
1864 		retval = sc->esc_FCAH;
1865 		break;
1866 	case E1000_FCT:
1867 		retval = sc->esc_FCT;
1868 		break;
1869 	case E1000_VET:
1870 		retval = sc->esc_VET;
1871 		break;
1872 	case E1000_FCTTV:
1873 		retval = sc->esc_FCTTV;
1874 		break;
1875 	case E1000_LEDCTL:
1876 		retval = sc->esc_LEDCTL;
1877 		break;
1878 	case E1000_PBA:
1879 		retval = sc->esc_PBA;
1880 		break;
1881 	case E1000_ICR:
1882 	case E1000_ITR:
1883 	case E1000_ICS:
1884 	case E1000_IMS:
1885 	case E1000_IMC:
1886 		retval = e82545_intr_read(sc, offset);
1887 		break;
1888 	case E1000_RCTL:
1889 		retval = sc->esc_RCTL;
1890 		break;
1891 	case E1000_FCRTL:
1892 		retval = sc->esc_FCRTL;
1893 		break;
1894 	case E1000_FCRTH:
1895 		retval = sc->esc_FCRTH;
1896 		break;
1897 	case E1000_RDBAL(0):
1898 		retval = sc->esc_RDBAL;
1899 		break;
1900 	case E1000_RDBAH(0):
1901 		retval = sc->esc_RDBAH;
1902 		break;
1903 	case E1000_RDLEN(0):
1904 		retval = sc->esc_RDLEN;
1905 		break;
1906 	case E1000_RDH(0):
1907 		retval = sc->esc_RDH;
1908 		break;
1909 	case E1000_RDT(0):
1910 		retval = sc->esc_RDT;
1911 		break;
1912 	case E1000_RDTR:
1913 		retval = sc->esc_RDTR;
1914 		break;
1915 	case E1000_RXDCTL(0):
1916 		retval = sc->esc_RXDCTL;
1917 		break;
1918 	case E1000_RADV:
1919 		retval = sc->esc_RADV;
1920 		break;
1921 	case E1000_RSRPD:
1922 		retval = sc->esc_RSRPD;
1923 		break;
1924 	case E1000_RXCSUM:
1925 		retval = sc->esc_RXCSUM;
1926 		break;
1927 	case E1000_TXCW:
1928 		retval = sc->esc_TXCW;
1929 		break;
1930 	case E1000_TCTL:
1931 		retval = sc->esc_TCTL;
1932 		break;
1933 	case E1000_TIPG:
1934 		retval = sc->esc_TIPG;
1935 		break;
1936 	case E1000_AIT:
1937 		retval = sc->esc_AIT;
1938 		break;
1939 	case E1000_TDBAL(0):
1940 		retval = sc->esc_TDBAL;
1941 		break;
1942 	case E1000_TDBAH(0):
1943 		retval = sc->esc_TDBAH;
1944 		break;
1945 	case E1000_TDLEN(0):
1946 		retval = sc->esc_TDLEN;
1947 		break;
1948 	case E1000_TDH(0):
1949 		retval = sc->esc_TDH;
1950 		break;
1951 	case E1000_TDT(0):
1952 		retval = sc->esc_TDT;
1953 		break;
1954 	case E1000_TIDV:
1955 		retval = sc->esc_TIDV;
1956 		break;
1957 	case E1000_TXDCTL(0):
1958 		retval = sc->esc_TXDCTL;
1959 		break;
1960 	case E1000_TADV:
1961 		retval = sc->esc_TADV;
1962 		break;
1963 	case E1000_RAL(0) ... E1000_RAH(15):
1964 		/* convert to u32 offset */
1965 		ridx = (offset - E1000_RAL(0)) >> 2;
1966 		retval = e82545_read_ra(sc, ridx);
1967 		break;
1968 	case E1000_MTA ... (E1000_MTA + (127*4)):
1969 		retval = sc->esc_fmcast[(offset - E1000_MTA) >> 2];
1970 		break;
1971 	case E1000_VFTA ... (E1000_VFTA + (127*4)):
1972 		retval = sc->esc_fvlan[(offset - E1000_VFTA) >> 2];
1973 		break;
1974 	case E1000_EECD:
1975 		//DPRINTF("EECD read %x", sc->eeprom_control);
1976 		retval = sc->eeprom_control;
1977 		break;
1978 	case E1000_MDIC:
1979 		retval = sc->mdi_control;
1980 		break;
1981 	case E1000_MANC:
1982 		retval = 0;
1983 		break;
1984 	/* stats that we emulate. */
1985 	case E1000_MPC:
1986 		retval = sc->missed_pkt_count;
1987 		break;
1988 	case E1000_PRC64:
1989 		retval = sc->pkt_rx_by_size[0];
1990 		break;
1991 	case E1000_PRC127:
1992 		retval = sc->pkt_rx_by_size[1];
1993 		break;
1994 	case E1000_PRC255:
1995 		retval = sc->pkt_rx_by_size[2];
1996 		break;
1997 	case E1000_PRC511:
1998 		retval = sc->pkt_rx_by_size[3];
1999 		break;
2000 	case E1000_PRC1023:
2001 		retval = sc->pkt_rx_by_size[4];
2002 		break;
2003 	case E1000_PRC1522:
2004 		retval = sc->pkt_rx_by_size[5];
2005 		break;
2006 	case E1000_GPRC:
2007 		retval = sc->good_pkt_rx_count;
2008 		break;
2009 	case E1000_BPRC:
2010 		retval = sc->bcast_pkt_rx_count;
2011 		break;
2012 	case E1000_MPRC:
2013 		retval = sc->mcast_pkt_rx_count;
2014 		break;
2015 	case E1000_GPTC:
2016 	case E1000_TPT:
2017 		retval = sc->good_pkt_tx_count;
2018 		break;
2019 	case E1000_GORCL:
2020 		retval = (uint32_t)sc->good_octets_rx;
2021 		break;
2022 	case E1000_GORCH:
2023 		retval = (uint32_t)(sc->good_octets_rx >> 32);
2024 		break;
2025 	case E1000_TOTL:
2026 	case E1000_GOTCL:
2027 		retval = (uint32_t)sc->good_octets_tx;
2028 		break;
2029 	case E1000_TOTH:
2030 	case E1000_GOTCH:
2031 		retval = (uint32_t)(sc->good_octets_tx >> 32);
2032 		break;
2033 	case E1000_ROC:
2034 		retval = sc->oversize_rx_count;
2035 		break;
2036 	case E1000_TORL:
2037 		retval = (uint32_t)(sc->good_octets_rx + sc->missed_octets);
2038 		break;
2039 	case E1000_TORH:
2040 		retval = (uint32_t)((sc->good_octets_rx +
2041 		    sc->missed_octets) >> 32);
2042 		break;
2043 	case E1000_TPR:
2044 		retval = sc->good_pkt_rx_count + sc->missed_pkt_count +
2045 		    sc->oversize_rx_count;
2046 		break;
2047 	case E1000_PTC64:
2048 		retval = sc->pkt_tx_by_size[0];
2049 		break;
2050 	case E1000_PTC127:
2051 		retval = sc->pkt_tx_by_size[1];
2052 		break;
2053 	case E1000_PTC255:
2054 		retval = sc->pkt_tx_by_size[2];
2055 		break;
2056 	case E1000_PTC511:
2057 		retval = sc->pkt_tx_by_size[3];
2058 		break;
2059 	case E1000_PTC1023:
2060 		retval = sc->pkt_tx_by_size[4];
2061 		break;
2062 	case E1000_PTC1522:
2063 		retval = sc->pkt_tx_by_size[5];
2064 		break;
2065 	case E1000_MPTC:
2066 		retval = sc->mcast_pkt_tx_count;
2067 		break;
2068 	case E1000_BPTC:
2069 		retval = sc->bcast_pkt_tx_count;
2070 		break;
2071 	case E1000_TSCTC:
2072 		retval = sc->tso_tx_count;
2073 		break;
2074 	/* stats that are always 0. */
2075 	case E1000_CRCERRS:
2076 	case E1000_ALGNERRC:
2077 	case E1000_SYMERRS:
2078 	case E1000_RXERRC:
2079 	case E1000_SCC:
2080 	case E1000_ECOL:
2081 	case E1000_MCC:
2082 	case E1000_LATECOL:
2083 	case E1000_COLC:
2084 	case E1000_DC:
2085 	case E1000_TNCRS:
2086 	case E1000_SEC:
2087 	case E1000_CEXTERR:
2088 	case E1000_RLEC:
2089 	case E1000_XONRXC:
2090 	case E1000_XONTXC:
2091 	case E1000_XOFFRXC:
2092 	case E1000_XOFFTXC:
2093 	case E1000_FCRUC:
2094 	case E1000_RNBC:
2095 	case E1000_RUC:
2096 	case E1000_RFC:
2097 	case E1000_RJC:
2098 	case E1000_MGTPRC:
2099 	case E1000_MGTPDC:
2100 	case E1000_MGTPTC:
2101 	case E1000_TSCTFC:
2102 		retval = 0;
2103 		break;
2104 	default:
2105 		DPRINTF("Unknown read register: 0x%x", offset);
2106 		retval = 0;
2107 		break;
2108 	}
2109 
2110 	return (retval);
2111 }
2112 
2113 static void
2114 e82545_write(struct pci_devinst *pi, int baridx, uint64_t offset, int size,
2115     uint64_t value)
2116 {
2117 	struct e82545_softc *sc;
2118 
2119 	//DPRINTF("Write bar:%d offset:0x%lx value:0x%lx size:%d", baridx, offset, value, size);
2120 
2121 	sc = pi->pi_arg;
2122 
2123 	pthread_mutex_lock(&sc->esc_mtx);
2124 
2125 	switch (baridx) {
2126 	case E82545_BAR_IO:
2127 		switch (offset) {
2128 		case E82545_IOADDR:
2129 			if (size != 4) {
2130 				DPRINTF("Wrong io addr write sz:%d value:0x%lx", size, value);
2131 			} else
2132 				sc->io_addr = (uint32_t)value;
2133 			break;
2134 		case E82545_IODATA:
2135 			if (size != 4) {
2136 				DPRINTF("Wrong io data write size:%d value:0x%lx", size, value);
2137 			} else if (sc->io_addr > E82545_IO_REGISTER_MAX) {
2138 				DPRINTF("Non-register io write addr:0x%x value:0x%lx", sc->io_addr, value);
2139 			} else
2140 				e82545_write_register(sc, sc->io_addr,
2141 						      (uint32_t)value);
2142 			break;
2143 		default:
2144 			DPRINTF("Unknown io bar write offset:0x%lx value:0x%lx size:%d", offset, value, size);
2145 			break;
2146 		}
2147 		break;
2148 	case E82545_BAR_REGISTER:
2149 		if (size != 4) {
2150 			DPRINTF("Wrong register write size:%d offset:0x%lx value:0x%lx", size, offset, value);
2151 		} else
2152 			e82545_write_register(sc, (uint32_t)offset,
2153 					      (uint32_t)value);
2154 		break;
2155 	default:
2156 		DPRINTF("Unknown write bar:%d off:0x%lx val:0x%lx size:%d",
2157 			baridx, offset, value, size);
2158 	}
2159 
2160 	pthread_mutex_unlock(&sc->esc_mtx);
2161 }
2162 
2163 static uint64_t
2164 e82545_read(struct pci_devinst *pi, int baridx, uint64_t offset, int size)
2165 {
2166 	struct e82545_softc *sc;
2167 	uint64_t retval;
2168 
2169 	//DPRINTF("Read  bar:%d offset:0x%lx size:%d", baridx, offset, size);
2170 	sc = pi->pi_arg;
2171 	retval = 0;
2172 
2173 	pthread_mutex_lock(&sc->esc_mtx);
2174 
2175 	switch (baridx) {
2176 	case E82545_BAR_IO:
2177 		switch (offset) {
2178 		case E82545_IOADDR:
2179 			if (size != 4) {
2180 				DPRINTF("Wrong io addr read sz:%d", size);
2181 			} else
2182 				retval = sc->io_addr;
2183 			break;
2184 		case E82545_IODATA:
2185 			if (size != 4) {
2186 				DPRINTF("Wrong io data read sz:%d", size);
2187 			}
2188 			if (sc->io_addr > E82545_IO_REGISTER_MAX) {
2189 				DPRINTF("Non-register io read addr:0x%x",
2190 					sc->io_addr);
2191 			} else
2192 				retval = e82545_read_register(sc, sc->io_addr);
2193 			break;
2194 		default:
2195 			DPRINTF("Unknown io bar read offset:0x%lx size:%d",
2196 				offset, size);
2197 			break;
2198 		}
2199 		break;
2200 	case E82545_BAR_REGISTER:
2201 		if (size != 4) {
2202 			DPRINTF("Wrong register read size:%d offset:0x%lx",
2203 				size, offset);
2204 		} else
2205 			retval = e82545_read_register(sc, (uint32_t)offset);
2206 		break;
2207 	default:
2208 		DPRINTF("Unknown read bar:%d offset:0x%lx size:%d",
2209 			baridx, offset, size);
2210 		break;
2211 	}
2212 
2213 	pthread_mutex_unlock(&sc->esc_mtx);
2214 
2215 	return (retval);
2216 }
2217 
2218 static void
2219 e82545_reset(struct e82545_softc *sc, int drvr)
2220 {
2221 	int i;
2222 
2223 	e82545_rx_disable(sc);
2224 	e82545_tx_disable(sc);
2225 
2226 	/* clear outstanding interrupts */
2227 	if (sc->esc_irq_asserted)
2228 		pci_lintr_deassert(sc->esc_pi);
2229 
2230 	/* misc */
2231 	if (!drvr) {
2232 		sc->esc_FCAL = 0;
2233 		sc->esc_FCAH = 0;
2234 		sc->esc_FCT = 0;
2235 		sc->esc_VET = 0;
2236 		sc->esc_FCTTV = 0;
2237 	}
2238 	sc->esc_LEDCTL = 0x07061302;
2239 	sc->esc_PBA = 0x00100030;
2240 
2241 	/* start nvm in opcode mode. */
2242 	sc->nvm_opaddr = 0;
2243 	sc->nvm_mode = E82545_NVM_MODE_OPADDR;
2244 	sc->nvm_bits = E82545_NVM_OPADDR_BITS;
2245 	sc->eeprom_control = E1000_EECD_PRES | E82545_EECD_FWE_EN;
2246 	e82545_init_eeprom(sc);
2247 
2248 	/* interrupt */
2249 	sc->esc_ICR = 0;
2250 	sc->esc_ITR = 250;
2251 	sc->esc_ICS = 0;
2252 	sc->esc_IMS = 0;
2253 	sc->esc_IMC = 0;
2254 
2255 	/* L2 filters */
2256 	if (!drvr) {
2257 		memset(sc->esc_fvlan, 0, sizeof(sc->esc_fvlan));
2258 		memset(sc->esc_fmcast, 0, sizeof(sc->esc_fmcast));
2259 		memset(sc->esc_uni, 0, sizeof(sc->esc_uni));
2260 
2261 		/* XXX not necessary on 82545 ?? */
2262 		sc->esc_uni[0].eu_valid = 1;
2263 		memcpy(sc->esc_uni[0].eu_eth.octet, sc->esc_mac.octet,
2264 		    ETHER_ADDR_LEN);
2265 	} else {
2266 		/* Clear RAH valid bits */
2267 		for (i = 0; i < 16; i++)
2268 			sc->esc_uni[i].eu_valid = 0;
2269 	}
2270 
2271 	/* receive */
2272 	if (!drvr) {
2273 		sc->esc_RDBAL = 0;
2274 		sc->esc_RDBAH = 0;
2275 	}
2276 	sc->esc_RCTL = 0;
2277 	sc->esc_FCRTL = 0;
2278 	sc->esc_FCRTH = 0;
2279 	sc->esc_RDLEN = 0;
2280 	sc->esc_RDH = 0;
2281 	sc->esc_RDT = 0;
2282 	sc->esc_RDTR = 0;
2283 	sc->esc_RXDCTL = (1 << 24) | (1 << 16); /* default GRAN/WTHRESH */
2284 	sc->esc_RADV = 0;
2285 	sc->esc_RXCSUM = 0;
2286 
2287 	/* transmit */
2288 	if (!drvr) {
2289 		sc->esc_TDBAL = 0;
2290 		sc->esc_TDBAH = 0;
2291 		sc->esc_TIPG = 0;
2292 		sc->esc_AIT = 0;
2293 		sc->esc_TIDV = 0;
2294 		sc->esc_TADV = 0;
2295 	}
2296 	sc->esc_tdba = 0;
2297 	sc->esc_txdesc = NULL;
2298 	sc->esc_TXCW = 0;
2299 	sc->esc_TCTL = 0;
2300 	sc->esc_TDLEN = 0;
2301 	sc->esc_TDT = 0;
2302 	sc->esc_TDHr = sc->esc_TDH = 0;
2303 	sc->esc_TXDCTL = 0;
2304 }
2305 
2306 static int
2307 e82545_init(struct pci_devinst *pi, nvlist_t *nvl)
2308 {
2309 	char nstr[80];
2310 	struct e82545_softc *sc;
2311 	const char *mac;
2312 	int err;
2313 
2314 	/* Setup our softc */
2315 	sc = calloc(1, sizeof(*sc));
2316 
2317 	pi->pi_arg = sc;
2318 	sc->esc_pi = pi;
2319 	sc->esc_ctx = pi->pi_vmctx;
2320 
2321 	pthread_mutex_init(&sc->esc_mtx, NULL);
2322 	pthread_cond_init(&sc->esc_rx_cond, NULL);
2323 	pthread_cond_init(&sc->esc_tx_cond, NULL);
2324 	pthread_create(&sc->esc_tx_tid, NULL, e82545_tx_thread, sc);
2325 	snprintf(nstr, sizeof(nstr), "e82545-%d:%d tx", pi->pi_slot,
2326 	    pi->pi_func);
2327         pthread_set_name_np(sc->esc_tx_tid, nstr);
2328 
2329 	pci_set_cfgdata16(pi, PCIR_DEVICE, E82545_DEV_ID_82545EM_COPPER);
2330 	pci_set_cfgdata16(pi, PCIR_VENDOR, E82545_VENDOR_ID_INTEL);
2331 	pci_set_cfgdata8(pi,  PCIR_CLASS, PCIC_NETWORK);
2332 	pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_NETWORK_ETHERNET);
2333 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, E82545_SUBDEV_ID);
2334 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, E82545_VENDOR_ID_INTEL);
2335 
2336 	pci_set_cfgdata8(pi,  PCIR_HDRTYPE, PCIM_HDRTYPE_NORMAL);
2337 	pci_set_cfgdata8(pi,  PCIR_INTPIN, 0x1);
2338 
2339 	/* TODO: this card also supports msi, but the freebsd driver for it
2340 	 * does not, so I have not implemented it. */
2341 	pci_lintr_request(pi);
2342 
2343 	pci_emul_alloc_bar(pi, E82545_BAR_REGISTER, PCIBAR_MEM32,
2344 		E82545_BAR_REGISTER_LEN);
2345 	pci_emul_alloc_bar(pi, E82545_BAR_FLASH, PCIBAR_MEM32,
2346 		E82545_BAR_FLASH_LEN);
2347 	pci_emul_alloc_bar(pi, E82545_BAR_IO, PCIBAR_IO,
2348 		E82545_BAR_IO_LEN);
2349 
2350 	mac = get_config_value_node(nvl, "mac");
2351 	if (mac != NULL) {
2352 		err = net_parsemac(mac, sc->esc_mac.octet);
2353 		if (err) {
2354 			free(sc);
2355 			return (err);
2356 		}
2357 	} else
2358 		net_genmac(pi, sc->esc_mac.octet);
2359 
2360 	err = netbe_init(&sc->esc_be, nvl, e82545_rx_callback, sc);
2361 	if (err) {
2362 		free(sc);
2363 		return (err);
2364 	}
2365 
2366 	netbe_rx_enable(sc->esc_be);
2367 
2368 	/* H/w initiated reset */
2369 	e82545_reset(sc, 0);
2370 
2371 	return (0);
2372 }
2373 
2374 #ifdef BHYVE_SNAPSHOT
2375 static int
2376 e82545_snapshot(struct vm_snapshot_meta *meta)
2377 {
2378 	int i;
2379 	int ret;
2380 	struct e82545_softc *sc;
2381 	struct pci_devinst *pi;
2382 	uint64_t bitmap_value;
2383 
2384 	pi = meta->dev_data;
2385 	sc = pi->pi_arg;
2386 
2387 	/* esc_mevp and esc_mevpitr should be reinitiated at init. */
2388 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_mac, meta, ret, done);
2389 
2390 	/* General */
2391 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_CTRL, meta, ret, done);
2392 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCAL, meta, ret, done);
2393 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCAH, meta, ret, done);
2394 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCT, meta, ret, done);
2395 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_VET, meta, ret, done);
2396 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCTTV, meta, ret, done);
2397 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_LEDCTL, meta, ret, done);
2398 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_PBA, meta, ret, done);
2399 
2400 	/* Interrupt control */
2401 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_irq_asserted, meta, ret, done);
2402 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_ICR, meta, ret, done);
2403 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_ITR, meta, ret, done);
2404 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_ICS, meta, ret, done);
2405 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_IMS, meta, ret, done);
2406 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_IMC, meta, ret, done);
2407 
2408 	/*
2409 	 * Transmit
2410 	 *
2411 	 * The fields in the unions are in superposition to access certain
2412 	 * bytes in the larger uint variables.
2413 	 * e.g., ip_config = [ipcss|ipcso|ipcse0|ipcse1]
2414 	 */
2415 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_txctx.lower_setup.ip_config, meta, ret, done);
2416 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_txctx.upper_setup.tcp_config, meta, ret, done);
2417 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_txctx.cmd_and_length, meta, ret, done);
2418 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_txctx.tcp_seg_setup.data, meta, ret, done);
2419 
2420 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_tx_enabled, meta, ret, done);
2421 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_tx_active, meta, ret, done);
2422 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TXCW, meta, ret, done);
2423 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TCTL, meta, ret, done);
2424 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TIPG, meta, ret, done);
2425 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_AIT, meta, ret, done);
2426 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_tdba, meta, ret, done);
2427 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDBAL, meta, ret, done);
2428 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDBAH, meta, ret, done);
2429 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDLEN, meta, ret, done);
2430 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDH, meta, ret, done);
2431 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDHr, meta, ret, done);
2432 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TDT, meta, ret, done);
2433 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TIDV, meta, ret, done);
2434 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TXDCTL, meta, ret, done);
2435 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_TADV, meta, ret, done);
2436 
2437 	/* Has dependency on esc_TDLEN; reoreder of fields from struct. */
2438 	SNAPSHOT_GUEST2HOST_ADDR_OR_LEAVE(pi->pi_vmctx, sc->esc_txdesc,
2439 	    sc->esc_TDLEN, true, meta, ret, done);
2440 
2441 	/* L2 frame acceptance */
2442 	for (i = 0; i < (int)nitems(sc->esc_uni); i++) {
2443 		SNAPSHOT_VAR_OR_LEAVE(sc->esc_uni[i].eu_valid, meta, ret, done);
2444 		SNAPSHOT_VAR_OR_LEAVE(sc->esc_uni[i].eu_addrsel, meta, ret, done);
2445 		SNAPSHOT_VAR_OR_LEAVE(sc->esc_uni[i].eu_eth, meta, ret, done);
2446 	}
2447 
2448 	SNAPSHOT_BUF_OR_LEAVE(sc->esc_fmcast, sizeof(sc->esc_fmcast),
2449 			      meta, ret, done);
2450 	SNAPSHOT_BUF_OR_LEAVE(sc->esc_fvlan, sizeof(sc->esc_fvlan),
2451 			      meta, ret, done);
2452 
2453 	/* Receive */
2454 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_rx_enabled, meta, ret, done);
2455 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_rx_active, meta, ret, done);
2456 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_rx_loopback, meta, ret, done);
2457 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RCTL, meta, ret, done);
2458 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCRTL, meta, ret, done);
2459 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_FCRTH, meta, ret, done);
2460 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_rdba, meta, ret, done);
2461 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDBAL, meta, ret, done);
2462 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDBAH, meta, ret, done);
2463 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDLEN, meta, ret, done);
2464 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDH, meta, ret, done);
2465 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDT, meta, ret, done);
2466 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RDTR, meta, ret, done);
2467 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RXDCTL, meta, ret, done);
2468 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RADV, meta, ret, done);
2469 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RSRPD, meta, ret, done);
2470 	SNAPSHOT_VAR_OR_LEAVE(sc->esc_RXCSUM, meta, ret, done);
2471 
2472 	/* Has dependency on esc_RDLEN; reoreder of fields from struct. */
2473 	SNAPSHOT_GUEST2HOST_ADDR_OR_LEAVE(pi->pi_vmctx, sc->esc_rxdesc,
2474 	    sc->esc_TDLEN, true, meta, ret, done);
2475 
2476 	/* IO Port register access */
2477 	SNAPSHOT_VAR_OR_LEAVE(sc->io_addr, meta, ret, done);
2478 
2479 	/* Shadow copy of MDIC */
2480 	SNAPSHOT_VAR_OR_LEAVE(sc->mdi_control, meta, ret, done);
2481 
2482 	/* Shadow copy of EECD */
2483 	SNAPSHOT_VAR_OR_LEAVE(sc->eeprom_control, meta, ret, done);
2484 
2485 	/* Latest NVM in/out */
2486 	SNAPSHOT_VAR_OR_LEAVE(sc->nvm_data, meta, ret, done);
2487 	SNAPSHOT_VAR_OR_LEAVE(sc->nvm_opaddr, meta, ret, done);
2488 
2489 	/* Stats */
2490 	SNAPSHOT_VAR_OR_LEAVE(sc->missed_pkt_count, meta, ret, done);
2491 	SNAPSHOT_BUF_OR_LEAVE(sc->pkt_rx_by_size, sizeof(sc->pkt_rx_by_size),
2492 			      meta, ret, done);
2493 	SNAPSHOT_BUF_OR_LEAVE(sc->pkt_tx_by_size, sizeof(sc->pkt_tx_by_size),
2494 			      meta, ret, done);
2495 	SNAPSHOT_VAR_OR_LEAVE(sc->good_pkt_rx_count, meta, ret, done);
2496 	SNAPSHOT_VAR_OR_LEAVE(sc->bcast_pkt_rx_count, meta, ret, done);
2497 	SNAPSHOT_VAR_OR_LEAVE(sc->mcast_pkt_rx_count, meta, ret, done);
2498 	SNAPSHOT_VAR_OR_LEAVE(sc->good_pkt_tx_count, meta, ret, done);
2499 	SNAPSHOT_VAR_OR_LEAVE(sc->bcast_pkt_tx_count, meta, ret, done);
2500 	SNAPSHOT_VAR_OR_LEAVE(sc->mcast_pkt_tx_count, meta, ret, done);
2501 	SNAPSHOT_VAR_OR_LEAVE(sc->oversize_rx_count, meta, ret, done);
2502 	SNAPSHOT_VAR_OR_LEAVE(sc->tso_tx_count, meta, ret, done);
2503 	SNAPSHOT_VAR_OR_LEAVE(sc->good_octets_rx, meta, ret, done);
2504 	SNAPSHOT_VAR_OR_LEAVE(sc->good_octets_tx, meta, ret, done);
2505 	SNAPSHOT_VAR_OR_LEAVE(sc->missed_octets, meta, ret, done);
2506 
2507 	if (meta->op == VM_SNAPSHOT_SAVE)
2508 		bitmap_value = sc->nvm_bits;
2509 	SNAPSHOT_VAR_OR_LEAVE(bitmap_value, meta, ret, done);
2510 	if (meta->op == VM_SNAPSHOT_RESTORE)
2511 		sc->nvm_bits = bitmap_value;
2512 
2513 	if (meta->op == VM_SNAPSHOT_SAVE)
2514 		bitmap_value = sc->nvm_bits;
2515 	SNAPSHOT_VAR_OR_LEAVE(bitmap_value, meta, ret, done);
2516 	if (meta->op == VM_SNAPSHOT_RESTORE)
2517 		sc->nvm_bits = bitmap_value;
2518 
2519 	/* EEPROM data */
2520 	SNAPSHOT_BUF_OR_LEAVE(sc->eeprom_data, sizeof(sc->eeprom_data),
2521 			      meta, ret, done);
2522 
2523 done:
2524 	return (ret);
2525 }
2526 #endif
2527 
2528 static const struct pci_devemu pci_de_e82545 = {
2529 	.pe_emu = 	"e1000",
2530 	.pe_init =	e82545_init,
2531 	.pe_legacy_config = netbe_legacy_config,
2532 	.pe_barwrite =	e82545_write,
2533 	.pe_barread =	e82545_read,
2534 #ifdef BHYVE_SNAPSHOT
2535 	.pe_snapshot =	e82545_snapshot,
2536 #endif
2537 };
2538 PCI_EMUL_SET(pci_de_e82545);
2539