1/*-
2 * SPDX-License-Identifier: BSD-4-Clause
3 *
4 * Copyright (c) 1997, 1998-2003
5 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 *    must display the following acknowledgement:
17 *	This product includes software developed by Bill Paul.
18 * 4. Neither the name of the author nor the names of any co-contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
23 * 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 Bill Paul OR THE VOICES IN HIS HEAD
26 * BE 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
32 * THE POSSIBILITY OF SUCH DAMAGE.
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD$");
37
38/*
39 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
40 *
41 * Written by Bill Paul <wpaul@windriver.com>
42 * Senior Networking Software Engineer
43 * Wind River Systems
44 */
45
46/*
47 * This driver is designed to support RealTek's next generation of
48 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
49 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
50 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
51 *
52 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
53 * with the older 8139 family, however it also supports a special
54 * C+ mode of operation that provides several new performance enhancing
55 * features. These include:
56 *
57 *	o Descriptor based DMA mechanism. Each descriptor represents
58 *	  a single packet fragment. Data buffers may be aligned on
59 *	  any byte boundary.
60 *
61 *	o 64-bit DMA
62 *
63 *	o TCP/IP checksum offload for both RX and TX
64 *
65 *	o High and normal priority transmit DMA rings
66 *
67 *	o VLAN tag insertion and extraction
68 *
69 *	o TCP large send (segmentation offload)
70 *
71 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
72 * programming API is fairly straightforward. The RX filtering, EEPROM
73 * access and PHY access is the same as it is on the older 8139 series
74 * chips.
75 *
76 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
77 * same programming API and feature set as the 8139C+ with the following
78 * differences and additions:
79 *
80 *	o 1000Mbps mode
81 *
82 *	o Jumbo frames
83 *
84 *	o GMII and TBI ports/registers for interfacing with copper
85 *	  or fiber PHYs
86 *
87 *	o RX and TX DMA rings can have up to 1024 descriptors
88 *	  (the 8139C+ allows a maximum of 64)
89 *
90 *	o Slight differences in register layout from the 8139C+
91 *
92 * The TX start and timer interrupt registers are at different locations
93 * on the 8169 than they are on the 8139C+. Also, the status word in the
94 * RX descriptor has a slightly different bit layout. The 8169 does not
95 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
96 * copper gigE PHY.
97 *
98 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
99 * (the 'S' stands for 'single-chip'). These devices have the same
100 * programming API as the older 8169, but also have some vendor-specific
101 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
102 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
103 *
104 * This driver takes advantage of the RX and TX checksum offload and
105 * VLAN tag insertion/extraction features. It also implements TX
106 * interrupt moderation using the timer interrupt registers, which
107 * significantly reduces TX interrupt load. There is also support
108 * for jumbo frames, however the 8169/8169S/8110S can not transmit
109 * jumbo frames larger than 7440, so the max MTU possible with this
110 * driver is 7422 bytes.
111 */
112
113#ifdef HAVE_KERNEL_OPTION_HEADERS
114#include "opt_device_polling.h"
115#endif
116
117#include <sys/param.h>
118#include <sys/endian.h>
119#include <sys/systm.h>
120#include <sys/sockio.h>
121#include <sys/mbuf.h>
122#include <sys/malloc.h>
123#include <sys/module.h>
124#include <sys/kernel.h>
125#include <sys/socket.h>
126#include <sys/lock.h>
127#include <sys/mutex.h>
128#include <sys/sysctl.h>
129#include <sys/taskqueue.h>
130
131#include <net/debugnet.h>
132#include <net/if.h>
133#include <net/if_var.h>
134#include <net/if_arp.h>
135#include <net/ethernet.h>
136#include <net/if_dl.h>
137#include <net/if_media.h>
138#include <net/if_types.h>
139#include <net/if_vlan_var.h>
140
141#include <net/bpf.h>
142
143#include <machine/bus.h>
144#include <machine/resource.h>
145#include <sys/bus.h>
146#include <sys/rman.h>
147
148#include <dev/mii/mii.h>
149#include <dev/mii/miivar.h>
150
151#include <dev/pci/pcireg.h>
152#include <dev/pci/pcivar.h>
153
154#include <dev/rl/if_rlreg.h>
155
156MODULE_DEPEND(re, pci, 1, 1, 1);
157MODULE_DEPEND(re, ether, 1, 1, 1);
158MODULE_DEPEND(re, miibus, 1, 1, 1);
159
160/* "device miibus" required.  See GENERIC if you get errors here. */
161#include "miibus_if.h"
162
163/* Tunables. */
164static int intr_filter = 0;
165TUNABLE_INT("hw.re.intr_filter", &intr_filter);
166static int msi_disable = 0;
167TUNABLE_INT("hw.re.msi_disable", &msi_disable);
168static int msix_disable = 0;
169TUNABLE_INT("hw.re.msix_disable", &msix_disable);
170static int prefer_iomap = 0;
171TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
172
173#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
174
175/*
176 * Various supported device vendors/types and their names.
177 */
178static const struct rl_type re_devs[] = {
179	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
180	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
181	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
182	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
183	{ RT_VENDORID, RT_DEVICEID_8139, 0,
184	    "RealTek 8139C+ 10/100BaseTX" },
185	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
186	    "RealTek 810xE PCIe 10/100baseTX" },
187	{ RT_VENDORID, RT_DEVICEID_8168, 0,
188	    "RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
189	{ RT_VENDORID, RT_DEVICEID_8161, 0,
190	    "RealTek 8168 Gigabit Ethernet" },
191	{ NCUBE_VENDORID, RT_DEVICEID_8168, 0,
192	    "TP-Link TG-3468 v2 (RTL8168) Gigabit Ethernet" },
193	{ RT_VENDORID, RT_DEVICEID_8169, 0,
194	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
195	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
196	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
197	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
198	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
199	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
200	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
201	{ USR_VENDORID, USR_DEVICEID_997902, 0,
202	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
203};
204
205static const struct rl_hwrev re_hwrevs[] = {
206	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
207	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
208	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
209	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
210	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
211	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
212	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
213	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
214	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
215	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
216	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
217	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
218	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
219	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
220	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
221	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
222	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
223	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
224	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
225	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
226	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
227	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
228	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
229	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
230	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
231	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
232	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
233	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
234	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
235	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
236	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
237	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
238	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
239	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
240	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
241	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
242	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
243	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
244	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
245	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
246	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
247	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
248	{ RL_HWREV_8168H, RL_8169, "8168H/8111H", RL_JUMBO_MTU_9K},
249	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
250	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
251	{ 0, 0, NULL, 0 }
252};
253
254static int re_probe		(device_t);
255static int re_attach		(device_t);
256static int re_detach		(device_t);
257
258static int re_encap		(struct rl_softc *, struct mbuf **);
259
260static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
261static int re_allocmem		(device_t, struct rl_softc *);
262static __inline void re_discard_rxbuf
263				(struct rl_softc *, int);
264static int re_newbuf		(struct rl_softc *, int);
265static int re_jumbo_newbuf	(struct rl_softc *, int);
266static int re_rx_list_init	(struct rl_softc *);
267static int re_jrx_list_init	(struct rl_softc *);
268static int re_tx_list_init	(struct rl_softc *);
269#ifdef RE_FIXUP_RX
270static __inline void re_fixup_rx
271				(struct mbuf *);
272#endif
273static int re_rxeof		(struct rl_softc *, int *);
274static void re_txeof		(struct rl_softc *);
275#ifdef DEVICE_POLLING
276static int re_poll		(struct ifnet *, enum poll_cmd, int);
277static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
278#endif
279static int re_intr		(void *);
280static void re_intr_msi		(void *);
281static void re_tick		(void *);
282static void re_int_task		(void *, int);
283static void re_start		(struct ifnet *);
284static void re_start_locked	(struct ifnet *);
285static void re_start_tx		(struct rl_softc *);
286static int re_ioctl		(struct ifnet *, u_long, caddr_t);
287static void re_init		(void *);
288static void re_init_locked	(struct rl_softc *);
289static void re_stop		(struct rl_softc *);
290static void re_watchdog		(struct rl_softc *);
291static int re_suspend		(device_t);
292static int re_resume		(device_t);
293static int re_shutdown		(device_t);
294static int re_ifmedia_upd	(struct ifnet *);
295static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
296
297static void re_eeprom_putbyte	(struct rl_softc *, int);
298static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
299static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
300static int re_gmii_readreg	(device_t, int, int);
301static int re_gmii_writereg	(device_t, int, int, int);
302
303static int re_miibus_readreg	(device_t, int, int);
304static int re_miibus_writereg	(device_t, int, int, int);
305static void re_miibus_statchg	(device_t);
306
307static void re_set_jumbo	(struct rl_softc *, int);
308static void re_set_rxmode		(struct rl_softc *);
309static void re_reset		(struct rl_softc *);
310static void re_setwol		(struct rl_softc *);
311static void re_clrwol		(struct rl_softc *);
312static void re_set_linkspeed	(struct rl_softc *);
313
314DEBUGNET_DEFINE(re);
315
316#ifdef DEV_NETMAP	/* see ixgbe.c for details */
317#include <dev/netmap/if_re_netmap.h>
318MODULE_DEPEND(re, netmap, 1, 1, 1);
319#endif /* !DEV_NETMAP */
320
321#ifdef RE_DIAG
322static int re_diag		(struct rl_softc *);
323#endif
324
325static void re_add_sysctls	(struct rl_softc *);
326static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
327static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
328static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
329
330static device_method_t re_methods[] = {
331	/* Device interface */
332	DEVMETHOD(device_probe,		re_probe),
333	DEVMETHOD(device_attach,	re_attach),
334	DEVMETHOD(device_detach,	re_detach),
335	DEVMETHOD(device_suspend,	re_suspend),
336	DEVMETHOD(device_resume,	re_resume),
337	DEVMETHOD(device_shutdown,	re_shutdown),
338
339	/* MII interface */
340	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
341	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
342	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
343
344	DEVMETHOD_END
345};
346
347static driver_t re_driver = {
348	"re",
349	re_methods,
350	sizeof(struct rl_softc)
351};
352
353static devclass_t re_devclass;
354
355DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
356DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
357
358#define EE_SET(x)					\
359	CSR_WRITE_1(sc, RL_EECMD,			\
360		CSR_READ_1(sc, RL_EECMD) | x)
361
362#define EE_CLR(x)					\
363	CSR_WRITE_1(sc, RL_EECMD,			\
364		CSR_READ_1(sc, RL_EECMD) & ~x)
365
366/*
367 * Send a read command and address to the EEPROM, check for ACK.
368 */
369static void
370re_eeprom_putbyte(struct rl_softc *sc, int addr)
371{
372	int			d, i;
373
374	d = addr | (RL_9346_READ << sc->rl_eewidth);
375
376	/*
377	 * Feed in each bit and strobe the clock.
378	 */
379
380	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
381		if (d & i) {
382			EE_SET(RL_EE_DATAIN);
383		} else {
384			EE_CLR(RL_EE_DATAIN);
385		}
386		DELAY(100);
387		EE_SET(RL_EE_CLK);
388		DELAY(150);
389		EE_CLR(RL_EE_CLK);
390		DELAY(100);
391	}
392}
393
394/*
395 * Read a word of data stored in the EEPROM at address 'addr.'
396 */
397static void
398re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
399{
400	int			i;
401	u_int16_t		word = 0;
402
403	/*
404	 * Send address of word we want to read.
405	 */
406	re_eeprom_putbyte(sc, addr);
407
408	/*
409	 * Start reading bits from EEPROM.
410	 */
411	for (i = 0x8000; i; i >>= 1) {
412		EE_SET(RL_EE_CLK);
413		DELAY(100);
414		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
415			word |= i;
416		EE_CLR(RL_EE_CLK);
417		DELAY(100);
418	}
419
420	*dest = word;
421}
422
423/*
424 * Read a sequence of words from the EEPROM.
425 */
426static void
427re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
428{
429	int			i;
430	u_int16_t		word = 0, *ptr;
431
432	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
433
434        DELAY(100);
435
436	for (i = 0; i < cnt; i++) {
437		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
438		re_eeprom_getword(sc, off + i, &word);
439		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
440		ptr = (u_int16_t *)(dest + (i * 2));
441                *ptr = word;
442	}
443
444	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
445}
446
447static int
448re_gmii_readreg(device_t dev, int phy, int reg)
449{
450	struct rl_softc		*sc;
451	u_int32_t		rval;
452	int			i;
453
454	sc = device_get_softc(dev);
455
456	/* Let the rgephy driver read the GMEDIASTAT register */
457
458	if (reg == RL_GMEDIASTAT) {
459		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
460		return (rval);
461	}
462
463	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
464
465	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
466		rval = CSR_READ_4(sc, RL_PHYAR);
467		if (rval & RL_PHYAR_BUSY)
468			break;
469		DELAY(25);
470	}
471
472	if (i == RL_PHY_TIMEOUT) {
473		device_printf(sc->rl_dev, "PHY read failed\n");
474		return (0);
475	}
476
477	/*
478	 * Controller requires a 20us delay to process next MDIO request.
479	 */
480	DELAY(20);
481
482	return (rval & RL_PHYAR_PHYDATA);
483}
484
485static int
486re_gmii_writereg(device_t dev, int phy, int reg, int data)
487{
488	struct rl_softc		*sc;
489	u_int32_t		rval;
490	int			i;
491
492	sc = device_get_softc(dev);
493
494	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
495	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
496
497	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
498		rval = CSR_READ_4(sc, RL_PHYAR);
499		if (!(rval & RL_PHYAR_BUSY))
500			break;
501		DELAY(25);
502	}
503
504	if (i == RL_PHY_TIMEOUT) {
505		device_printf(sc->rl_dev, "PHY write failed\n");
506		return (0);
507	}
508
509	/*
510	 * Controller requires a 20us delay to process next MDIO request.
511	 */
512	DELAY(20);
513
514	return (0);
515}
516
517static int
518re_miibus_readreg(device_t dev, int phy, int reg)
519{
520	struct rl_softc		*sc;
521	u_int16_t		rval = 0;
522	u_int16_t		re8139_reg = 0;
523
524	sc = device_get_softc(dev);
525
526	if (sc->rl_type == RL_8169) {
527		rval = re_gmii_readreg(dev, phy, reg);
528		return (rval);
529	}
530
531	switch (reg) {
532	case MII_BMCR:
533		re8139_reg = RL_BMCR;
534		break;
535	case MII_BMSR:
536		re8139_reg = RL_BMSR;
537		break;
538	case MII_ANAR:
539		re8139_reg = RL_ANAR;
540		break;
541	case MII_ANER:
542		re8139_reg = RL_ANER;
543		break;
544	case MII_ANLPAR:
545		re8139_reg = RL_LPAR;
546		break;
547	case MII_PHYIDR1:
548	case MII_PHYIDR2:
549		return (0);
550	/*
551	 * Allow the rlphy driver to read the media status
552	 * register. If we have a link partner which does not
553	 * support NWAY, this is the register which will tell
554	 * us the results of parallel detection.
555	 */
556	case RL_MEDIASTAT:
557		rval = CSR_READ_1(sc, RL_MEDIASTAT);
558		return (rval);
559	default:
560		device_printf(sc->rl_dev, "bad phy register\n");
561		return (0);
562	}
563	rval = CSR_READ_2(sc, re8139_reg);
564	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
565		/* 8139C+ has different bit layout. */
566		rval &= ~(BMCR_LOOP | BMCR_ISO);
567	}
568	return (rval);
569}
570
571static int
572re_miibus_writereg(device_t dev, int phy, int reg, int data)
573{
574	struct rl_softc		*sc;
575	u_int16_t		re8139_reg = 0;
576	int			rval = 0;
577
578	sc = device_get_softc(dev);
579
580	if (sc->rl_type == RL_8169) {
581		rval = re_gmii_writereg(dev, phy, reg, data);
582		return (rval);
583	}
584
585	switch (reg) {
586	case MII_BMCR:
587		re8139_reg = RL_BMCR;
588		if (sc->rl_type == RL_8139CPLUS) {
589			/* 8139C+ has different bit layout. */
590			data &= ~(BMCR_LOOP | BMCR_ISO);
591		}
592		break;
593	case MII_BMSR:
594		re8139_reg = RL_BMSR;
595		break;
596	case MII_ANAR:
597		re8139_reg = RL_ANAR;
598		break;
599	case MII_ANER:
600		re8139_reg = RL_ANER;
601		break;
602	case MII_ANLPAR:
603		re8139_reg = RL_LPAR;
604		break;
605	case MII_PHYIDR1:
606	case MII_PHYIDR2:
607		return (0);
608		break;
609	default:
610		device_printf(sc->rl_dev, "bad phy register\n");
611		return (0);
612	}
613	CSR_WRITE_2(sc, re8139_reg, data);
614	return (0);
615}
616
617static void
618re_miibus_statchg(device_t dev)
619{
620	struct rl_softc		*sc;
621	struct ifnet		*ifp;
622	struct mii_data		*mii;
623
624	sc = device_get_softc(dev);
625	mii = device_get_softc(sc->rl_miibus);
626	ifp = sc->rl_ifp;
627	if (mii == NULL || ifp == NULL ||
628	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
629		return;
630
631	sc->rl_flags &= ~RL_FLAG_LINK;
632	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
633	    (IFM_ACTIVE | IFM_AVALID)) {
634		switch (IFM_SUBTYPE(mii->mii_media_active)) {
635		case IFM_10_T:
636		case IFM_100_TX:
637			sc->rl_flags |= RL_FLAG_LINK;
638			break;
639		case IFM_1000_T:
640			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
641				break;
642			sc->rl_flags |= RL_FLAG_LINK;
643			break;
644		default:
645			break;
646		}
647	}
648	/*
649	 * RealTek controllers do not provide any interface to the RX/TX
650	 * MACs for resolved speed, duplex and flow-control parameters.
651	 */
652}
653
654static u_int
655re_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
656{
657	uint32_t h, *hashes = arg;
658
659	h = ether_crc32_be(LLADDR(sdl), ETHER_ADDR_LEN) >> 26;
660	if (h < 32)
661		hashes[0] |= (1 << h);
662	else
663		hashes[1] |= (1 << (h - 32));
664
665	return (1);
666}
667
668/*
669 * Set the RX configuration and 64-bit multicast hash filter.
670 */
671static void
672re_set_rxmode(struct rl_softc *sc)
673{
674	struct ifnet		*ifp;
675	uint32_t		h, hashes[2] = { 0, 0 };
676	uint32_t		rxfilt;
677
678	RL_LOCK_ASSERT(sc);
679
680	ifp = sc->rl_ifp;
681
682	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
683	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
684		rxfilt |= RL_RXCFG_EARLYOFF;
685	else if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
686		rxfilt |= RL_RXCFG_EARLYOFFV2;
687
688	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
689		if (ifp->if_flags & IFF_PROMISC)
690			rxfilt |= RL_RXCFG_RX_ALLPHYS;
691		/*
692		 * Unlike other hardwares, we have to explicitly set
693		 * RL_RXCFG_RX_MULTI to receive multicast frames in
694		 * promiscuous mode.
695		 */
696		rxfilt |= RL_RXCFG_RX_MULTI;
697		hashes[0] = hashes[1] = 0xffffffff;
698		goto done;
699	}
700
701	if_foreach_llmaddr(ifp, re_hash_maddr, hashes);
702
703	if (hashes[0] != 0 || hashes[1] != 0) {
704		/*
705		 * For some unfathomable reason, RealTek decided to
706		 * reverse the order of the multicast hash registers
707		 * in the PCI Express parts.  This means we have to
708		 * write the hash pattern in reverse order for those
709		 * devices.
710		 */
711		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
712			h = bswap32(hashes[0]);
713			hashes[0] = bswap32(hashes[1]);
714			hashes[1] = h;
715		}
716		rxfilt |= RL_RXCFG_RX_MULTI;
717	}
718
719	if  (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
720		/* Disable multicast filtering due to silicon bug. */
721		hashes[0] = 0xffffffff;
722		hashes[1] = 0xffffffff;
723	}
724
725done:
726	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
727	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
728	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
729}
730
731static void
732re_reset(struct rl_softc *sc)
733{
734	int			i;
735
736	RL_LOCK_ASSERT(sc);
737
738	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
739
740	for (i = 0; i < RL_TIMEOUT; i++) {
741		DELAY(10);
742		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
743			break;
744	}
745	if (i == RL_TIMEOUT)
746		device_printf(sc->rl_dev, "reset never completed!\n");
747
748	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
749		CSR_WRITE_1(sc, 0x82, 1);
750	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
751		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
752}
753
754#ifdef RE_DIAG
755
756/*
757 * The following routine is designed to test for a defect on some
758 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
759 * lines connected to the bus, however for a 32-bit only card, they
760 * should be pulled high. The result of this defect is that the
761 * NIC will not work right if you plug it into a 64-bit slot: DMA
762 * operations will be done with 64-bit transfers, which will fail
763 * because the 64-bit data lines aren't connected.
764 *
765 * There's no way to work around this (short of talking a soldering
766 * iron to the board), however we can detect it. The method we use
767 * here is to put the NIC into digital loopback mode, set the receiver
768 * to promiscuous mode, and then try to send a frame. We then compare
769 * the frame data we sent to what was received. If the data matches,
770 * then the NIC is working correctly, otherwise we know the user has
771 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
772 * slot. In the latter case, there's no way the NIC can work correctly,
773 * so we print out a message on the console and abort the device attach.
774 */
775
776static int
777re_diag(struct rl_softc *sc)
778{
779	struct ifnet		*ifp = sc->rl_ifp;
780	struct mbuf		*m0;
781	struct ether_header	*eh;
782	struct rl_desc		*cur_rx;
783	u_int16_t		status;
784	u_int32_t		rxstat;
785	int			total_len, i, error = 0, phyaddr;
786	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
787	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
788
789	/* Allocate a single mbuf */
790	MGETHDR(m0, M_NOWAIT, MT_DATA);
791	if (m0 == NULL)
792		return (ENOBUFS);
793
794	RL_LOCK(sc);
795
796	/*
797	 * Initialize the NIC in test mode. This sets the chip up
798	 * so that it can send and receive frames, but performs the
799	 * following special functions:
800	 * - Puts receiver in promiscuous mode
801	 * - Enables digital loopback mode
802	 * - Leaves interrupts turned off
803	 */
804
805	ifp->if_flags |= IFF_PROMISC;
806	sc->rl_testmode = 1;
807	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
808	re_init_locked(sc);
809	sc->rl_flags |= RL_FLAG_LINK;
810	if (sc->rl_type == RL_8169)
811		phyaddr = 1;
812	else
813		phyaddr = 0;
814
815	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
816	for (i = 0; i < RL_TIMEOUT; i++) {
817		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
818		if (!(status & BMCR_RESET))
819			break;
820	}
821
822	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
823	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
824
825	DELAY(100000);
826
827	/* Put some data in the mbuf */
828
829	eh = mtod(m0, struct ether_header *);
830	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
831	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
832	eh->ether_type = htons(ETHERTYPE_IP);
833	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
834
835	/*
836	 * Queue the packet, start transmission.
837	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
838	 */
839
840	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
841	RL_UNLOCK(sc);
842	/* XXX: re_diag must not be called when in ALTQ mode */
843	IF_HANDOFF(&ifp->if_snd, m0, ifp);
844	RL_LOCK(sc);
845	m0 = NULL;
846
847	/* Wait for it to propagate through the chip */
848
849	DELAY(100000);
850	for (i = 0; i < RL_TIMEOUT; i++) {
851		status = CSR_READ_2(sc, RL_ISR);
852		CSR_WRITE_2(sc, RL_ISR, status);
853		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
854		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
855			break;
856		DELAY(10);
857	}
858
859	if (i == RL_TIMEOUT) {
860		device_printf(sc->rl_dev,
861		    "diagnostic failed, failed to receive packet in"
862		    " loopback mode\n");
863		error = EIO;
864		goto done;
865	}
866
867	/*
868	 * The packet should have been dumped into the first
869	 * entry in the RX DMA ring. Grab it from there.
870	 */
871
872	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
873	    sc->rl_ldata.rl_rx_list_map,
874	    BUS_DMASYNC_POSTREAD);
875	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
876	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
877	    BUS_DMASYNC_POSTREAD);
878	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
879	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
880
881	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
882	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
883	eh = mtod(m0, struct ether_header *);
884
885	cur_rx = &sc->rl_ldata.rl_rx_list[0];
886	total_len = RL_RXBYTES(cur_rx);
887	rxstat = le32toh(cur_rx->rl_cmdstat);
888
889	if (total_len != ETHER_MIN_LEN) {
890		device_printf(sc->rl_dev,
891		    "diagnostic failed, received short packet\n");
892		error = EIO;
893		goto done;
894	}
895
896	/* Test that the received packet data matches what we sent. */
897
898	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
899	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
900	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
901		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
902		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
903		    dst, ":", src, ":", ETHERTYPE_IP);
904		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
905		    eh->ether_dhost, ":", eh->ether_shost, ":",
906		    ntohs(eh->ether_type));
907		device_printf(sc->rl_dev, "You may have a defective 32-bit "
908		    "NIC plugged into a 64-bit PCI slot.\n");
909		device_printf(sc->rl_dev, "Please re-install the NIC in a "
910		    "32-bit slot for proper operation.\n");
911		device_printf(sc->rl_dev, "Read the re(4) man page for more "
912		    "details.\n");
913		error = EIO;
914	}
915
916done:
917	/* Turn interface off, release resources */
918
919	sc->rl_testmode = 0;
920	sc->rl_flags &= ~RL_FLAG_LINK;
921	ifp->if_flags &= ~IFF_PROMISC;
922	re_stop(sc);
923	if (m0 != NULL)
924		m_freem(m0);
925
926	RL_UNLOCK(sc);
927
928	return (error);
929}
930
931#endif
932
933/*
934 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
935 * IDs against our list and return a device name if we find a match.
936 */
937static int
938re_probe(device_t dev)
939{
940	const struct rl_type	*t;
941	uint16_t		devid, vendor;
942	uint16_t		revid, sdevid;
943	int			i;
944
945	vendor = pci_get_vendor(dev);
946	devid = pci_get_device(dev);
947	revid = pci_get_revid(dev);
948	sdevid = pci_get_subdevice(dev);
949
950	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
951		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
952			/*
953			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
954			 * Rev. 2 is supported by sk(4).
955			 */
956			return (ENXIO);
957		}
958	}
959
960	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
961		if (revid != 0x20) {
962			/* 8139, let rl(4) take care of this device. */
963			return (ENXIO);
964		}
965	}
966
967	t = re_devs;
968	for (i = 0; i < nitems(re_devs); i++, t++) {
969		if (vendor == t->rl_vid && devid == t->rl_did) {
970			device_set_desc(dev, t->rl_name);
971			return (BUS_PROBE_DEFAULT);
972		}
973	}
974
975	return (ENXIO);
976}
977
978/*
979 * Map a single buffer address.
980 */
981
982static void
983re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
984{
985	bus_addr_t		*addr;
986
987	if (error)
988		return;
989
990	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
991	addr = arg;
992	*addr = segs->ds_addr;
993}
994
995static int
996re_allocmem(device_t dev, struct rl_softc *sc)
997{
998	bus_addr_t		lowaddr;
999	bus_size_t		rx_list_size, tx_list_size;
1000	int			error;
1001	int			i;
1002
1003	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
1004	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
1005
1006	/*
1007	 * Allocate the parent bus DMA tag appropriate for PCI.
1008	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
1009	 * register should be set. However some RealTek chips are known
1010	 * to be buggy on DAC handling, therefore disable DAC by limiting
1011	 * DMA address space to 32bit. PCIe variants of RealTek chips
1012	 * may not have the limitation.
1013	 */
1014	lowaddr = BUS_SPACE_MAXADDR;
1015	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
1016		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1017	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
1018	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
1019	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1020	    NULL, NULL, &sc->rl_parent_tag);
1021	if (error) {
1022		device_printf(dev, "could not allocate parent DMA tag\n");
1023		return (error);
1024	}
1025
1026	/*
1027	 * Allocate map for TX mbufs.
1028	 */
1029	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1030	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1031	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1032	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1033	if (error) {
1034		device_printf(dev, "could not allocate TX DMA tag\n");
1035		return (error);
1036	}
1037
1038	/*
1039	 * Allocate map for RX mbufs.
1040	 */
1041
1042	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1043		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1044		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1045		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1046		    &sc->rl_ldata.rl_jrx_mtag);
1047		if (error) {
1048			device_printf(dev,
1049			    "could not allocate jumbo RX DMA tag\n");
1050			return (error);
1051		}
1052	}
1053	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1054	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1055	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1056	if (error) {
1057		device_printf(dev, "could not allocate RX DMA tag\n");
1058		return (error);
1059	}
1060
1061	/*
1062	 * Allocate map for TX descriptor list.
1063	 */
1064	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1065	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1066	    NULL, tx_list_size, 1, tx_list_size, 0,
1067	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1068	if (error) {
1069		device_printf(dev, "could not allocate TX DMA ring tag\n");
1070		return (error);
1071	}
1072
1073	/* Allocate DMA'able memory for the TX ring */
1074
1075	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1076	    (void **)&sc->rl_ldata.rl_tx_list,
1077	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1078	    &sc->rl_ldata.rl_tx_list_map);
1079	if (error) {
1080		device_printf(dev, "could not allocate TX DMA ring\n");
1081		return (error);
1082	}
1083
1084	/* Load the map for the TX ring. */
1085
1086	sc->rl_ldata.rl_tx_list_addr = 0;
1087	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1088	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1089	     tx_list_size, re_dma_map_addr,
1090	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1091	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1092		device_printf(dev, "could not load TX DMA ring\n");
1093		return (ENOMEM);
1094	}
1095
1096	/* Create DMA maps for TX buffers */
1097
1098	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1099		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1100		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1101		if (error) {
1102			device_printf(dev, "could not create DMA map for TX\n");
1103			return (error);
1104		}
1105	}
1106
1107	/*
1108	 * Allocate map for RX descriptor list.
1109	 */
1110	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1111	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1112	    NULL, rx_list_size, 1, rx_list_size, 0,
1113	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1114	if (error) {
1115		device_printf(dev, "could not create RX DMA ring tag\n");
1116		return (error);
1117	}
1118
1119	/* Allocate DMA'able memory for the RX ring */
1120
1121	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1122	    (void **)&sc->rl_ldata.rl_rx_list,
1123	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1124	    &sc->rl_ldata.rl_rx_list_map);
1125	if (error) {
1126		device_printf(dev, "could not allocate RX DMA ring\n");
1127		return (error);
1128	}
1129
1130	/* Load the map for the RX ring. */
1131
1132	sc->rl_ldata.rl_rx_list_addr = 0;
1133	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1134	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1135	     rx_list_size, re_dma_map_addr,
1136	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1137	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1138		device_printf(dev, "could not load RX DMA ring\n");
1139		return (ENOMEM);
1140	}
1141
1142	/* Create DMA maps for RX buffers */
1143
1144	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1145		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1146		    &sc->rl_ldata.rl_jrx_sparemap);
1147		if (error) {
1148			device_printf(dev,
1149			    "could not create spare DMA map for jumbo RX\n");
1150			return (error);
1151		}
1152		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1153			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1154			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1155			if (error) {
1156				device_printf(dev,
1157				    "could not create DMA map for jumbo RX\n");
1158				return (error);
1159			}
1160		}
1161	}
1162	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1163	    &sc->rl_ldata.rl_rx_sparemap);
1164	if (error) {
1165		device_printf(dev, "could not create spare DMA map for RX\n");
1166		return (error);
1167	}
1168	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1169		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1170		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1171		if (error) {
1172			device_printf(dev, "could not create DMA map for RX\n");
1173			return (error);
1174		}
1175	}
1176
1177	/* Create DMA map for statistics. */
1178	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1179	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1180	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1181	    &sc->rl_ldata.rl_stag);
1182	if (error) {
1183		device_printf(dev, "could not create statistics DMA tag\n");
1184		return (error);
1185	}
1186	/* Allocate DMA'able memory for statistics. */
1187	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1188	    (void **)&sc->rl_ldata.rl_stats,
1189	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1190	    &sc->rl_ldata.rl_smap);
1191	if (error) {
1192		device_printf(dev,
1193		    "could not allocate statistics DMA memory\n");
1194		return (error);
1195	}
1196	/* Load the map for statistics. */
1197	sc->rl_ldata.rl_stats_addr = 0;
1198	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1199	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1200	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1201	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1202		device_printf(dev, "could not load statistics DMA memory\n");
1203		return (ENOMEM);
1204	}
1205
1206	return (0);
1207}
1208
1209/*
1210 * Attach the interface. Allocate softc structures, do ifmedia
1211 * setup and ethernet/BPF attach.
1212 */
1213static int
1214re_attach(device_t dev)
1215{
1216	u_char			eaddr[ETHER_ADDR_LEN];
1217	u_int16_t		as[ETHER_ADDR_LEN / 2];
1218	struct rl_softc		*sc;
1219	struct ifnet		*ifp;
1220	const struct rl_hwrev	*hw_rev;
1221	int			capmask, error = 0, hwrev, i, msic, msixc,
1222				phy, reg, rid;
1223	u_int32_t		cap, ctl;
1224	u_int16_t		devid, re_did = 0;
1225	uint8_t			cfg;
1226
1227	sc = device_get_softc(dev);
1228	sc->rl_dev = dev;
1229
1230	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1231	    MTX_DEF);
1232	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1233
1234	/*
1235	 * Map control/status registers.
1236	 */
1237	pci_enable_busmaster(dev);
1238
1239	devid = pci_get_device(dev);
1240	/*
1241	 * Prefer memory space register mapping over IO space.
1242	 * Because RTL8169SC does not seem to work when memory mapping
1243	 * is used always activate io mapping.
1244	 */
1245	if (devid == RT_DEVICEID_8169SC)
1246		prefer_iomap = 1;
1247	if (prefer_iomap == 0) {
1248		sc->rl_res_id = PCIR_BAR(1);
1249		sc->rl_res_type = SYS_RES_MEMORY;
1250		/* RTL8168/8101E seems to use different BARs. */
1251		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1252			sc->rl_res_id = PCIR_BAR(2);
1253	} else {
1254		sc->rl_res_id = PCIR_BAR(0);
1255		sc->rl_res_type = SYS_RES_IOPORT;
1256	}
1257	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1258	    &sc->rl_res_id, RF_ACTIVE);
1259	if (sc->rl_res == NULL && prefer_iomap == 0) {
1260		sc->rl_res_id = PCIR_BAR(0);
1261		sc->rl_res_type = SYS_RES_IOPORT;
1262		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1263		    &sc->rl_res_id, RF_ACTIVE);
1264	}
1265	if (sc->rl_res == NULL) {
1266		device_printf(dev, "couldn't map ports/memory\n");
1267		error = ENXIO;
1268		goto fail;
1269	}
1270
1271	sc->rl_btag = rman_get_bustag(sc->rl_res);
1272	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1273
1274	msic = pci_msi_count(dev);
1275	msixc = pci_msix_count(dev);
1276	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1277		sc->rl_flags |= RL_FLAG_PCIE;
1278		sc->rl_expcap = reg;
1279	}
1280	if (bootverbose) {
1281		device_printf(dev, "MSI count : %d\n", msic);
1282		device_printf(dev, "MSI-X count : %d\n", msixc);
1283	}
1284	if (msix_disable > 0)
1285		msixc = 0;
1286	if (msi_disable > 0)
1287		msic = 0;
1288	/* Prefer MSI-X to MSI. */
1289	if (msixc > 0) {
1290		msixc = RL_MSI_MESSAGES;
1291		rid = PCIR_BAR(4);
1292		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1293		    &rid, RF_ACTIVE);
1294		if (sc->rl_res_pba == NULL) {
1295			device_printf(sc->rl_dev,
1296			    "could not allocate MSI-X PBA resource\n");
1297		}
1298		if (sc->rl_res_pba != NULL &&
1299		    pci_alloc_msix(dev, &msixc) == 0) {
1300			if (msixc == RL_MSI_MESSAGES) {
1301				device_printf(dev, "Using %d MSI-X message\n",
1302				    msixc);
1303				sc->rl_flags |= RL_FLAG_MSIX;
1304			} else
1305				pci_release_msi(dev);
1306		}
1307		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1308			if (sc->rl_res_pba != NULL)
1309				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1310				    sc->rl_res_pba);
1311			sc->rl_res_pba = NULL;
1312			msixc = 0;
1313		}
1314	}
1315	/* Prefer MSI to INTx. */
1316	if (msixc == 0 && msic > 0) {
1317		msic = RL_MSI_MESSAGES;
1318		if (pci_alloc_msi(dev, &msic) == 0) {
1319			if (msic == RL_MSI_MESSAGES) {
1320				device_printf(dev, "Using %d MSI message\n",
1321				    msic);
1322				sc->rl_flags |= RL_FLAG_MSI;
1323				/* Explicitly set MSI enable bit. */
1324				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1325				cfg = CSR_READ_1(sc, RL_CFG2);
1326				cfg |= RL_CFG2_MSI;
1327				CSR_WRITE_1(sc, RL_CFG2, cfg);
1328				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1329			} else
1330				pci_release_msi(dev);
1331		}
1332		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1333			msic = 0;
1334	}
1335
1336	/* Allocate interrupt */
1337	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1338		rid = 0;
1339		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1340		    RF_SHAREABLE | RF_ACTIVE);
1341		if (sc->rl_irq[0] == NULL) {
1342			device_printf(dev, "couldn't allocate IRQ resources\n");
1343			error = ENXIO;
1344			goto fail;
1345		}
1346	} else {
1347		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1348			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1349			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1350			if (sc->rl_irq[i] == NULL) {
1351				device_printf(dev,
1352				    "couldn't allocate IRQ resources for "
1353				    "message %d\n", rid);
1354				error = ENXIO;
1355				goto fail;
1356			}
1357		}
1358	}
1359
1360	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1361		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1362		cfg = CSR_READ_1(sc, RL_CFG2);
1363		if ((cfg & RL_CFG2_MSI) != 0) {
1364			device_printf(dev, "turning off MSI enable bit.\n");
1365			cfg &= ~RL_CFG2_MSI;
1366			CSR_WRITE_1(sc, RL_CFG2, cfg);
1367		}
1368		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1369	}
1370
1371	/* Disable ASPM L0S/L1 and CLKREQ. */
1372	if (sc->rl_expcap != 0) {
1373		cap = pci_read_config(dev, sc->rl_expcap +
1374		    PCIER_LINK_CAP, 2);
1375		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1376			ctl = pci_read_config(dev, sc->rl_expcap +
1377			    PCIER_LINK_CTL, 2);
1378			if ((ctl & (PCIEM_LINK_CTL_ECPM |
1379			    PCIEM_LINK_CTL_ASPMC))!= 0) {
1380				ctl &= ~(PCIEM_LINK_CTL_ECPM |
1381				    PCIEM_LINK_CTL_ASPMC);
1382				pci_write_config(dev, sc->rl_expcap +
1383				    PCIER_LINK_CTL, ctl, 2);
1384				device_printf(dev, "ASPM disabled\n");
1385			}
1386		} else
1387			device_printf(dev, "no ASPM capability\n");
1388	}
1389
1390	hw_rev = re_hwrevs;
1391	hwrev = CSR_READ_4(sc, RL_TXCFG);
1392	switch (hwrev & 0x70000000) {
1393	case 0x00000000:
1394	case 0x10000000:
1395		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1396		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1397		break;
1398	default:
1399		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1400		sc->rl_macrev = hwrev & 0x00700000;
1401		hwrev &= RL_TXCFG_HWREV;
1402		break;
1403	}
1404	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1405	while (hw_rev->rl_desc != NULL) {
1406		if (hw_rev->rl_rev == hwrev) {
1407			sc->rl_type = hw_rev->rl_type;
1408			sc->rl_hwrev = hw_rev;
1409			break;
1410		}
1411		hw_rev++;
1412	}
1413	if (hw_rev->rl_desc == NULL) {
1414		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1415		error = ENXIO;
1416		goto fail;
1417	}
1418
1419	switch (hw_rev->rl_rev) {
1420	case RL_HWREV_8139CPLUS:
1421		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1422		break;
1423	case RL_HWREV_8100E:
1424	case RL_HWREV_8101E:
1425		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1426		break;
1427	case RL_HWREV_8102E:
1428	case RL_HWREV_8102EL:
1429	case RL_HWREV_8102EL_SPIN1:
1430		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1431		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1432		    RL_FLAG_AUTOPAD;
1433		break;
1434	case RL_HWREV_8103E:
1435		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1436		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1437		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1438		break;
1439	case RL_HWREV_8401E:
1440	case RL_HWREV_8105E:
1441	case RL_HWREV_8105E_SPIN1:
1442	case RL_HWREV_8106E:
1443		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1444		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1445		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1446		break;
1447	case RL_HWREV_8402:
1448		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1449		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1450		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1451		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1452		break;
1453	case RL_HWREV_8168B_SPIN1:
1454	case RL_HWREV_8168B_SPIN2:
1455		sc->rl_flags |= RL_FLAG_WOLRXENB;
1456		/* FALLTHROUGH */
1457	case RL_HWREV_8168B_SPIN3:
1458		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1459		break;
1460	case RL_HWREV_8168C_SPIN2:
1461		sc->rl_flags |= RL_FLAG_MACSLEEP;
1462		/* FALLTHROUGH */
1463	case RL_HWREV_8168C:
1464		if (sc->rl_macrev == 0x00200000)
1465			sc->rl_flags |= RL_FLAG_MACSLEEP;
1466		/* FALLTHROUGH */
1467	case RL_HWREV_8168CP:
1468		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1469		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1470		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1471		break;
1472	case RL_HWREV_8168D:
1473		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1474		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1475		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1476		    RL_FLAG_WOL_MANLINK;
1477		break;
1478	case RL_HWREV_8168DP:
1479		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1480		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1481		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1482		break;
1483	case RL_HWREV_8168E:
1484		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1485		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1486		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1487		    RL_FLAG_WOL_MANLINK;
1488		break;
1489	case RL_HWREV_8168E_VL:
1490	case RL_HWREV_8168F:
1491		sc->rl_flags |= RL_FLAG_EARLYOFF;
1492		/* FALLTHROUGH */
1493	case RL_HWREV_8411:
1494		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1495		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1496		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1497		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1498		break;
1499	case RL_HWREV_8168EP:
1500	case RL_HWREV_8168G:
1501	case RL_HWREV_8411B:
1502		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1503		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1504		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1505		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
1506		    RL_FLAG_8168G_PLUS;
1507		break;
1508	case RL_HWREV_8168GU:
1509	case RL_HWREV_8168H:
1510		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1511			/* RTL8106E(US), RTL8107E */
1512			sc->rl_flags |= RL_FLAG_FASTETHER;
1513		} else
1514			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1515
1516		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1517		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1518		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
1519		    RL_FLAG_8168G_PLUS;
1520		break;
1521	case RL_HWREV_8169_8110SB:
1522	case RL_HWREV_8169_8110SBL:
1523	case RL_HWREV_8169_8110SC:
1524	case RL_HWREV_8169_8110SCE:
1525		sc->rl_flags |= RL_FLAG_PHYWAKE;
1526		/* FALLTHROUGH */
1527	case RL_HWREV_8169:
1528	case RL_HWREV_8169S:
1529	case RL_HWREV_8110S:
1530		sc->rl_flags |= RL_FLAG_MACRESET;
1531		break;
1532	default:
1533		break;
1534	}
1535
1536	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1537		sc->rl_cfg0 = RL_8139_CFG0;
1538		sc->rl_cfg1 = RL_8139_CFG1;
1539		sc->rl_cfg2 = 0;
1540		sc->rl_cfg3 = RL_8139_CFG3;
1541		sc->rl_cfg4 = RL_8139_CFG4;
1542		sc->rl_cfg5 = RL_8139_CFG5;
1543	} else {
1544		sc->rl_cfg0 = RL_CFG0;
1545		sc->rl_cfg1 = RL_CFG1;
1546		sc->rl_cfg2 = RL_CFG2;
1547		sc->rl_cfg3 = RL_CFG3;
1548		sc->rl_cfg4 = RL_CFG4;
1549		sc->rl_cfg5 = RL_CFG5;
1550	}
1551
1552	/* Reset the adapter. */
1553	RL_LOCK(sc);
1554	re_reset(sc);
1555	RL_UNLOCK(sc);
1556
1557	/* Enable PME. */
1558	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1559	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1560	cfg |= RL_CFG1_PME;
1561	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1562	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1563	cfg &= RL_CFG5_PME_STS;
1564	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1565	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1566
1567	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1568		/*
1569		 * XXX Should have a better way to extract station
1570		 * address from EEPROM.
1571		 */
1572		for (i = 0; i < ETHER_ADDR_LEN; i++)
1573			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1574	} else {
1575		sc->rl_eewidth = RL_9356_ADDR_LEN;
1576		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1577		if (re_did != 0x8129)
1578			sc->rl_eewidth = RL_9346_ADDR_LEN;
1579
1580		/*
1581		 * Get station address from the EEPROM.
1582		 */
1583		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1584		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1585			as[i] = le16toh(as[i]);
1586		bcopy(as, eaddr, ETHER_ADDR_LEN);
1587	}
1588
1589	if (sc->rl_type == RL_8169) {
1590		/* Set RX length mask and number of descriptors. */
1591		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1592		sc->rl_txstart = RL_GTXSTART;
1593		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1594		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1595	} else {
1596		/* Set RX length mask and number of descriptors. */
1597		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1598		sc->rl_txstart = RL_TXSTART;
1599		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1600		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1601	}
1602
1603	error = re_allocmem(dev, sc);
1604	if (error)
1605		goto fail;
1606	re_add_sysctls(sc);
1607
1608	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1609	if (ifp == NULL) {
1610		device_printf(dev, "can not if_alloc()\n");
1611		error = ENOSPC;
1612		goto fail;
1613	}
1614
1615	/* Take controller out of deep sleep mode. */
1616	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1617		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1618			CSR_WRITE_1(sc, RL_GPIO,
1619			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1620		else
1621			CSR_WRITE_1(sc, RL_GPIO,
1622			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1623	}
1624
1625	/* Take PHY out of power down mode. */
1626	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1627		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1628		if (hw_rev->rl_rev == RL_HWREV_8401E)
1629			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1630	}
1631	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1632		re_gmii_writereg(dev, 1, 0x1f, 0);
1633		re_gmii_writereg(dev, 1, 0x0e, 0);
1634	}
1635
1636	ifp->if_softc = sc;
1637	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1638	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1639	ifp->if_ioctl = re_ioctl;
1640	ifp->if_start = re_start;
1641	/*
1642	 * RTL8168/8111C generates wrong IP checksummed frame if the
1643	 * packet has IP options so disable TX checksum offloading.
1644	 */
1645	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1646	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1647	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1648		ifp->if_hwassist = 0;
1649		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1650	} else {
1651		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1652		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1653	}
1654	ifp->if_hwassist |= CSUM_TSO;
1655	ifp->if_capenable = ifp->if_capabilities;
1656	ifp->if_init = re_init;
1657	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1658	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1659	IFQ_SET_READY(&ifp->if_snd);
1660
1661	NET_TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1662
1663#define	RE_PHYAD_INTERNAL	 0
1664
1665	/* Do MII setup. */
1666	phy = RE_PHYAD_INTERNAL;
1667	if (sc->rl_type == RL_8169)
1668		phy = 1;
1669	capmask = BMSR_DEFCAPMASK;
1670	if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
1671		 capmask &= ~BMSR_EXTSTAT;
1672	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1673	    re_ifmedia_sts, capmask, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1674	if (error != 0) {
1675		device_printf(dev, "attaching PHYs failed\n");
1676		goto fail;
1677	}
1678
1679	/*
1680	 * Call MI attach routine.
1681	 */
1682	ether_ifattach(ifp, eaddr);
1683
1684	/* VLAN capability setup */
1685	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1686	if (ifp->if_capabilities & IFCAP_HWCSUM)
1687		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1688	/* Enable WOL if PM is supported. */
1689	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1690		ifp->if_capabilities |= IFCAP_WOL;
1691	ifp->if_capenable = ifp->if_capabilities;
1692	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1693	/*
1694	 * Don't enable TSO by default.  It is known to generate
1695	 * corrupted TCP segments(bad TCP options) under certain
1696	 * circumstances.
1697	 */
1698	ifp->if_hwassist &= ~CSUM_TSO;
1699	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1700#ifdef DEVICE_POLLING
1701	ifp->if_capabilities |= IFCAP_POLLING;
1702#endif
1703	/*
1704	 * Tell the upper layer(s) we support long frames.
1705	 * Must appear after the call to ether_ifattach() because
1706	 * ether_ifattach() sets ifi_hdrlen to the default value.
1707	 */
1708	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1709
1710#ifdef DEV_NETMAP
1711	re_netmap_attach(sc);
1712#endif /* DEV_NETMAP */
1713
1714#ifdef RE_DIAG
1715	/*
1716	 * Perform hardware diagnostic on the original RTL8169.
1717	 * Some 32-bit cards were incorrectly wired and would
1718	 * malfunction if plugged into a 64-bit slot.
1719	 */
1720	if (hwrev == RL_HWREV_8169) {
1721		error = re_diag(sc);
1722		if (error) {
1723			device_printf(dev,
1724		    	"attach aborted due to hardware diag failure\n");
1725			ether_ifdetach(ifp);
1726			goto fail;
1727		}
1728	}
1729#endif
1730
1731#ifdef RE_TX_MODERATION
1732	intr_filter = 1;
1733#endif
1734	/* Hook interrupt last to avoid having to lock softc */
1735	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1736	    intr_filter == 0) {
1737		error = bus_setup_intr(dev, sc->rl_irq[0],
1738		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1739		    &sc->rl_intrhand[0]);
1740	} else {
1741		error = bus_setup_intr(dev, sc->rl_irq[0],
1742		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1743		    &sc->rl_intrhand[0]);
1744	}
1745	if (error) {
1746		device_printf(dev, "couldn't set up irq\n");
1747		ether_ifdetach(ifp);
1748		goto fail;
1749	}
1750
1751	DEBUGNET_SET(ifp, re);
1752
1753fail:
1754	if (error)
1755		re_detach(dev);
1756
1757	return (error);
1758}
1759
1760/*
1761 * Shutdown hardware and free up resources. This can be called any
1762 * time after the mutex has been initialized. It is called in both
1763 * the error case in attach and the normal detach case so it needs
1764 * to be careful about only freeing resources that have actually been
1765 * allocated.
1766 */
1767static int
1768re_detach(device_t dev)
1769{
1770	struct rl_softc		*sc;
1771	struct ifnet		*ifp;
1772	int			i, rid;
1773
1774	sc = device_get_softc(dev);
1775	ifp = sc->rl_ifp;
1776	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1777
1778	/* These should only be active if attach succeeded */
1779	if (device_is_attached(dev)) {
1780#ifdef DEVICE_POLLING
1781		if (ifp->if_capenable & IFCAP_POLLING)
1782			ether_poll_deregister(ifp);
1783#endif
1784		RL_LOCK(sc);
1785#if 0
1786		sc->suspended = 1;
1787#endif
1788		re_stop(sc);
1789		RL_UNLOCK(sc);
1790		callout_drain(&sc->rl_stat_callout);
1791		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1792		/*
1793		 * Force off the IFF_UP flag here, in case someone
1794		 * still had a BPF descriptor attached to this
1795		 * interface. If they do, ether_ifdetach() will cause
1796		 * the BPF code to try and clear the promisc mode
1797		 * flag, which will bubble down to re_ioctl(),
1798		 * which will try to call re_init() again. This will
1799		 * turn the NIC back on and restart the MII ticker,
1800		 * which will panic the system when the kernel tries
1801		 * to invoke the re_tick() function that isn't there
1802		 * anymore.
1803		 */
1804		ifp->if_flags &= ~IFF_UP;
1805		ether_ifdetach(ifp);
1806	}
1807	if (sc->rl_miibus)
1808		device_delete_child(dev, sc->rl_miibus);
1809	bus_generic_detach(dev);
1810
1811	/*
1812	 * The rest is resource deallocation, so we should already be
1813	 * stopped here.
1814	 */
1815
1816	if (sc->rl_intrhand[0] != NULL) {
1817		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1818		sc->rl_intrhand[0] = NULL;
1819	}
1820	if (ifp != NULL) {
1821#ifdef DEV_NETMAP
1822		netmap_detach(ifp);
1823#endif /* DEV_NETMAP */
1824		if_free(ifp);
1825	}
1826	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1827		rid = 0;
1828	else
1829		rid = 1;
1830	if (sc->rl_irq[0] != NULL) {
1831		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1832		sc->rl_irq[0] = NULL;
1833	}
1834	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1835		pci_release_msi(dev);
1836	if (sc->rl_res_pba) {
1837		rid = PCIR_BAR(4);
1838		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1839	}
1840	if (sc->rl_res)
1841		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1842		    sc->rl_res);
1843
1844	/* Unload and free the RX DMA ring memory and map */
1845
1846	if (sc->rl_ldata.rl_rx_list_tag) {
1847		if (sc->rl_ldata.rl_rx_list_addr)
1848			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1849			    sc->rl_ldata.rl_rx_list_map);
1850		if (sc->rl_ldata.rl_rx_list)
1851			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1852			    sc->rl_ldata.rl_rx_list,
1853			    sc->rl_ldata.rl_rx_list_map);
1854		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1855	}
1856
1857	/* Unload and free the TX DMA ring memory and map */
1858
1859	if (sc->rl_ldata.rl_tx_list_tag) {
1860		if (sc->rl_ldata.rl_tx_list_addr)
1861			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1862			    sc->rl_ldata.rl_tx_list_map);
1863		if (sc->rl_ldata.rl_tx_list)
1864			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1865			    sc->rl_ldata.rl_tx_list,
1866			    sc->rl_ldata.rl_tx_list_map);
1867		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1868	}
1869
1870	/* Destroy all the RX and TX buffer maps */
1871
1872	if (sc->rl_ldata.rl_tx_mtag) {
1873		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1874			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1875				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1876				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1877		}
1878		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1879	}
1880	if (sc->rl_ldata.rl_rx_mtag) {
1881		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1882			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1883				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1884				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1885		}
1886		if (sc->rl_ldata.rl_rx_sparemap)
1887			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1888			    sc->rl_ldata.rl_rx_sparemap);
1889		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1890	}
1891	if (sc->rl_ldata.rl_jrx_mtag) {
1892		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1893			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1894				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1895				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1896		}
1897		if (sc->rl_ldata.rl_jrx_sparemap)
1898			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1899			    sc->rl_ldata.rl_jrx_sparemap);
1900		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1901	}
1902	/* Unload and free the stats buffer and map */
1903
1904	if (sc->rl_ldata.rl_stag) {
1905		if (sc->rl_ldata.rl_stats_addr)
1906			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1907			    sc->rl_ldata.rl_smap);
1908		if (sc->rl_ldata.rl_stats)
1909			bus_dmamem_free(sc->rl_ldata.rl_stag,
1910			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1911		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1912	}
1913
1914	if (sc->rl_parent_tag)
1915		bus_dma_tag_destroy(sc->rl_parent_tag);
1916
1917	mtx_destroy(&sc->rl_mtx);
1918
1919	return (0);
1920}
1921
1922static __inline void
1923re_discard_rxbuf(struct rl_softc *sc, int idx)
1924{
1925	struct rl_desc		*desc;
1926	struct rl_rxdesc	*rxd;
1927	uint32_t		cmdstat;
1928
1929	if (sc->rl_ifp->if_mtu > RL_MTU &&
1930	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1931		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1932	else
1933		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1934	desc = &sc->rl_ldata.rl_rx_list[idx];
1935	desc->rl_vlanctl = 0;
1936	cmdstat = rxd->rx_size;
1937	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1938		cmdstat |= RL_RDESC_CMD_EOR;
1939	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1940}
1941
1942static int
1943re_newbuf(struct rl_softc *sc, int idx)
1944{
1945	struct mbuf		*m;
1946	struct rl_rxdesc	*rxd;
1947	bus_dma_segment_t	segs[1];
1948	bus_dmamap_t		map;
1949	struct rl_desc		*desc;
1950	uint32_t		cmdstat;
1951	int			error, nsegs;
1952
1953	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1954	if (m == NULL)
1955		return (ENOBUFS);
1956
1957	m->m_len = m->m_pkthdr.len = MCLBYTES;
1958#ifdef RE_FIXUP_RX
1959	/*
1960	 * This is part of an evil trick to deal with non-x86 platforms.
1961	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1962	 * boundaries, but that will hose non-x86 machines. To get around
1963	 * this, we leave some empty space at the start of each buffer
1964	 * and for non-x86 hosts, we copy the buffer back six bytes
1965	 * to achieve word alignment. This is slightly more efficient
1966	 * than allocating a new buffer, copying the contents, and
1967	 * discarding the old buffer.
1968	 */
1969	m_adj(m, RE_ETHER_ALIGN);
1970#endif
1971	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1972	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1973	if (error != 0) {
1974		m_freem(m);
1975		return (ENOBUFS);
1976	}
1977	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1978
1979	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1980	if (rxd->rx_m != NULL) {
1981		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1982		    BUS_DMASYNC_POSTREAD);
1983		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1984	}
1985
1986	rxd->rx_m = m;
1987	map = rxd->rx_dmamap;
1988	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1989	rxd->rx_size = segs[0].ds_len;
1990	sc->rl_ldata.rl_rx_sparemap = map;
1991	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1992	    BUS_DMASYNC_PREREAD);
1993
1994	desc = &sc->rl_ldata.rl_rx_list[idx];
1995	desc->rl_vlanctl = 0;
1996	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1997	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1998	cmdstat = segs[0].ds_len;
1999	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2000		cmdstat |= RL_RDESC_CMD_EOR;
2001	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2002
2003	return (0);
2004}
2005
2006static int
2007re_jumbo_newbuf(struct rl_softc *sc, int idx)
2008{
2009	struct mbuf		*m;
2010	struct rl_rxdesc	*rxd;
2011	bus_dma_segment_t	segs[1];
2012	bus_dmamap_t		map;
2013	struct rl_desc		*desc;
2014	uint32_t		cmdstat;
2015	int			error, nsegs;
2016
2017	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
2018	if (m == NULL)
2019		return (ENOBUFS);
2020	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
2021#ifdef RE_FIXUP_RX
2022	m_adj(m, RE_ETHER_ALIGN);
2023#endif
2024	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
2025	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
2026	if (error != 0) {
2027		m_freem(m);
2028		return (ENOBUFS);
2029	}
2030	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2031
2032	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2033	if (rxd->rx_m != NULL) {
2034		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2035		    BUS_DMASYNC_POSTREAD);
2036		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2037	}
2038
2039	rxd->rx_m = m;
2040	map = rxd->rx_dmamap;
2041	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2042	rxd->rx_size = segs[0].ds_len;
2043	sc->rl_ldata.rl_jrx_sparemap = map;
2044	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2045	    BUS_DMASYNC_PREREAD);
2046
2047	desc = &sc->rl_ldata.rl_rx_list[idx];
2048	desc->rl_vlanctl = 0;
2049	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2050	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2051	cmdstat = segs[0].ds_len;
2052	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2053		cmdstat |= RL_RDESC_CMD_EOR;
2054	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2055
2056	return (0);
2057}
2058
2059#ifdef RE_FIXUP_RX
2060static __inline void
2061re_fixup_rx(struct mbuf *m)
2062{
2063	int                     i;
2064	uint16_t                *src, *dst;
2065
2066	src = mtod(m, uint16_t *);
2067	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2068
2069	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2070		*dst++ = *src++;
2071
2072	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2073}
2074#endif
2075
2076static int
2077re_tx_list_init(struct rl_softc *sc)
2078{
2079	struct rl_desc		*desc;
2080	int			i;
2081
2082	RL_LOCK_ASSERT(sc);
2083
2084	bzero(sc->rl_ldata.rl_tx_list,
2085	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2086	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2087		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2088#ifdef DEV_NETMAP
2089	re_netmap_tx_init(sc);
2090#endif /* DEV_NETMAP */
2091	/* Set EOR. */
2092	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2093	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2094
2095	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2096	    sc->rl_ldata.rl_tx_list_map,
2097	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2098
2099	sc->rl_ldata.rl_tx_prodidx = 0;
2100	sc->rl_ldata.rl_tx_considx = 0;
2101	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2102
2103	return (0);
2104}
2105
2106static int
2107re_rx_list_init(struct rl_softc *sc)
2108{
2109	int			error, i;
2110
2111	bzero(sc->rl_ldata.rl_rx_list,
2112	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2113	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2114		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2115		if ((error = re_newbuf(sc, i)) != 0)
2116			return (error);
2117	}
2118#ifdef DEV_NETMAP
2119	re_netmap_rx_init(sc);
2120#endif /* DEV_NETMAP */
2121
2122	/* Flush the RX descriptors */
2123
2124	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2125	    sc->rl_ldata.rl_rx_list_map,
2126	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2127
2128	sc->rl_ldata.rl_rx_prodidx = 0;
2129	sc->rl_head = sc->rl_tail = NULL;
2130	sc->rl_int_rx_act = 0;
2131
2132	return (0);
2133}
2134
2135static int
2136re_jrx_list_init(struct rl_softc *sc)
2137{
2138	int			error, i;
2139
2140	bzero(sc->rl_ldata.rl_rx_list,
2141	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2142	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2143		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2144		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2145			return (error);
2146	}
2147
2148	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2149	    sc->rl_ldata.rl_rx_list_map,
2150	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2151
2152	sc->rl_ldata.rl_rx_prodidx = 0;
2153	sc->rl_head = sc->rl_tail = NULL;
2154	sc->rl_int_rx_act = 0;
2155
2156	return (0);
2157}
2158
2159/*
2160 * RX handler for C+ and 8169. For the gigE chips, we support
2161 * the reception of jumbo frames that have been fragmented
2162 * across multiple 2K mbuf cluster buffers.
2163 */
2164static int
2165re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2166{
2167	struct mbuf		*m;
2168	struct ifnet		*ifp;
2169	int			i, rxerr, total_len;
2170	struct rl_desc		*cur_rx;
2171	u_int32_t		rxstat, rxvlan;
2172	int			jumbo, maxpkt = 16, rx_npkts = 0;
2173
2174	RL_LOCK_ASSERT(sc);
2175
2176	ifp = sc->rl_ifp;
2177#ifdef DEV_NETMAP
2178	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2179		return 0;
2180#endif /* DEV_NETMAP */
2181	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2182		jumbo = 1;
2183	else
2184		jumbo = 0;
2185
2186	/* Invalidate the descriptor memory */
2187
2188	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2189	    sc->rl_ldata.rl_rx_list_map,
2190	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2191
2192	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2193	    i = RL_RX_DESC_NXT(sc, i)) {
2194		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2195			break;
2196		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2197		rxstat = le32toh(cur_rx->rl_cmdstat);
2198		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2199			break;
2200		total_len = rxstat & sc->rl_rxlenmask;
2201		rxvlan = le32toh(cur_rx->rl_vlanctl);
2202		if (jumbo != 0)
2203			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2204		else
2205			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2206
2207		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2208		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2209		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2210			/*
2211			 * RTL8168C or later controllers do not
2212			 * support multi-fragment packet.
2213			 */
2214			re_discard_rxbuf(sc, i);
2215			continue;
2216		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2217			if (re_newbuf(sc, i) != 0) {
2218				/*
2219				 * If this is part of a multi-fragment packet,
2220				 * discard all the pieces.
2221				 */
2222				if (sc->rl_head != NULL) {
2223					m_freem(sc->rl_head);
2224					sc->rl_head = sc->rl_tail = NULL;
2225				}
2226				re_discard_rxbuf(sc, i);
2227				continue;
2228			}
2229			m->m_len = RE_RX_DESC_BUFLEN;
2230			if (sc->rl_head == NULL)
2231				sc->rl_head = sc->rl_tail = m;
2232			else {
2233				m->m_flags &= ~M_PKTHDR;
2234				sc->rl_tail->m_next = m;
2235				sc->rl_tail = m;
2236			}
2237			continue;
2238		}
2239
2240		/*
2241		 * NOTE: for the 8139C+, the frame length field
2242		 * is always 12 bits in size, but for the gigE chips,
2243		 * it is 13 bits (since the max RX frame length is 16K).
2244		 * Unfortunately, all 32 bits in the status word
2245		 * were already used, so to make room for the extra
2246		 * length bit, RealTek took out the 'frame alignment
2247		 * error' bit and shifted the other status bits
2248		 * over one slot. The OWN, EOR, FS and LS bits are
2249		 * still in the same places. We have already extracted
2250		 * the frame length and checked the OWN bit, so rather
2251		 * than using an alternate bit mapping, we shift the
2252		 * status bits one space to the right so we can evaluate
2253		 * them using the 8169 status as though it was in the
2254		 * same format as that of the 8139C+.
2255		 */
2256		if (sc->rl_type == RL_8169)
2257			rxstat >>= 1;
2258
2259		/*
2260		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2261		 * set, but if CRC is clear, it will still be a valid frame.
2262		 */
2263		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2264			rxerr = 1;
2265			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2266			    total_len > 8191 &&
2267			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2268				rxerr = 0;
2269			if (rxerr != 0) {
2270				if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2271				/*
2272				 * If this is part of a multi-fragment packet,
2273				 * discard all the pieces.
2274				 */
2275				if (sc->rl_head != NULL) {
2276					m_freem(sc->rl_head);
2277					sc->rl_head = sc->rl_tail = NULL;
2278				}
2279				re_discard_rxbuf(sc, i);
2280				continue;
2281			}
2282		}
2283
2284		/*
2285		 * If allocating a replacement mbuf fails,
2286		 * reload the current one.
2287		 */
2288		if (jumbo != 0)
2289			rxerr = re_jumbo_newbuf(sc, i);
2290		else
2291			rxerr = re_newbuf(sc, i);
2292		if (rxerr != 0) {
2293			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2294			if (sc->rl_head != NULL) {
2295				m_freem(sc->rl_head);
2296				sc->rl_head = sc->rl_tail = NULL;
2297			}
2298			re_discard_rxbuf(sc, i);
2299			continue;
2300		}
2301
2302		if (sc->rl_head != NULL) {
2303			if (jumbo != 0)
2304				m->m_len = total_len;
2305			else {
2306				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2307				if (m->m_len == 0)
2308					m->m_len = RE_RX_DESC_BUFLEN;
2309			}
2310			/*
2311			 * Special case: if there's 4 bytes or less
2312			 * in this buffer, the mbuf can be discarded:
2313			 * the last 4 bytes is the CRC, which we don't
2314			 * care about anyway.
2315			 */
2316			if (m->m_len <= ETHER_CRC_LEN) {
2317				sc->rl_tail->m_len -=
2318				    (ETHER_CRC_LEN - m->m_len);
2319				m_freem(m);
2320			} else {
2321				m->m_len -= ETHER_CRC_LEN;
2322				m->m_flags &= ~M_PKTHDR;
2323				sc->rl_tail->m_next = m;
2324			}
2325			m = sc->rl_head;
2326			sc->rl_head = sc->rl_tail = NULL;
2327			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2328		} else
2329			m->m_pkthdr.len = m->m_len =
2330			    (total_len - ETHER_CRC_LEN);
2331
2332#ifdef RE_FIXUP_RX
2333		re_fixup_rx(m);
2334#endif
2335		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2336		m->m_pkthdr.rcvif = ifp;
2337
2338		/* Do RX checksumming if enabled */
2339
2340		if (ifp->if_capenable & IFCAP_RXCSUM) {
2341			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2342				/* Check IP header checksum */
2343				if (rxstat & RL_RDESC_STAT_PROTOID)
2344					m->m_pkthdr.csum_flags |=
2345					    CSUM_IP_CHECKED;
2346				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2347					m->m_pkthdr.csum_flags |=
2348					    CSUM_IP_VALID;
2349
2350				/* Check TCP/UDP checksum */
2351				if ((RL_TCPPKT(rxstat) &&
2352				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2353				    (RL_UDPPKT(rxstat) &&
2354				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2355					m->m_pkthdr.csum_flags |=
2356						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2357					m->m_pkthdr.csum_data = 0xffff;
2358				}
2359			} else {
2360				/*
2361				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2362				 */
2363				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2364				    (rxvlan & RL_RDESC_IPV4))
2365					m->m_pkthdr.csum_flags |=
2366					    CSUM_IP_CHECKED;
2367				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2368				    (rxvlan & RL_RDESC_IPV4))
2369					m->m_pkthdr.csum_flags |=
2370					    CSUM_IP_VALID;
2371				if (((rxstat & RL_RDESC_STAT_TCP) &&
2372				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2373				    ((rxstat & RL_RDESC_STAT_UDP) &&
2374				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2375					m->m_pkthdr.csum_flags |=
2376						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2377					m->m_pkthdr.csum_data = 0xffff;
2378				}
2379			}
2380		}
2381		maxpkt--;
2382		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2383			m->m_pkthdr.ether_vtag =
2384			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2385			m->m_flags |= M_VLANTAG;
2386		}
2387		RL_UNLOCK(sc);
2388		(*ifp->if_input)(ifp, m);
2389		RL_LOCK(sc);
2390		rx_npkts++;
2391	}
2392
2393	/* Flush the RX DMA ring */
2394
2395	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2396	    sc->rl_ldata.rl_rx_list_map,
2397	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2398
2399	sc->rl_ldata.rl_rx_prodidx = i;
2400
2401	if (rx_npktsp != NULL)
2402		*rx_npktsp = rx_npkts;
2403	if (maxpkt)
2404		return (EAGAIN);
2405
2406	return (0);
2407}
2408
2409static void
2410re_txeof(struct rl_softc *sc)
2411{
2412	struct ifnet		*ifp;
2413	struct rl_txdesc	*txd;
2414	u_int32_t		txstat;
2415	int			cons;
2416
2417	cons = sc->rl_ldata.rl_tx_considx;
2418	if (cons == sc->rl_ldata.rl_tx_prodidx)
2419		return;
2420
2421	ifp = sc->rl_ifp;
2422#ifdef DEV_NETMAP
2423	if (netmap_tx_irq(ifp, 0))
2424		return;
2425#endif /* DEV_NETMAP */
2426	/* Invalidate the TX descriptor list */
2427	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2428	    sc->rl_ldata.rl_tx_list_map,
2429	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2430
2431	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2432	    cons = RL_TX_DESC_NXT(sc, cons)) {
2433		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2434		if (txstat & RL_TDESC_STAT_OWN)
2435			break;
2436		/*
2437		 * We only stash mbufs in the last descriptor
2438		 * in a fragment chain, which also happens to
2439		 * be the only place where the TX status bits
2440		 * are valid.
2441		 */
2442		if (txstat & RL_TDESC_CMD_EOF) {
2443			txd = &sc->rl_ldata.rl_tx_desc[cons];
2444			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2445			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2446			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2447			    txd->tx_dmamap);
2448			KASSERT(txd->tx_m != NULL,
2449			    ("%s: freeing NULL mbufs!", __func__));
2450			m_freem(txd->tx_m);
2451			txd->tx_m = NULL;
2452			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2453			    RL_TDESC_STAT_COLCNT))
2454				if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
2455			if (txstat & RL_TDESC_STAT_TXERRSUM)
2456				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2457			else
2458				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
2459		}
2460		sc->rl_ldata.rl_tx_free++;
2461		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2462	}
2463	sc->rl_ldata.rl_tx_considx = cons;
2464
2465	/* No changes made to the TX ring, so no flush needed */
2466
2467	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2468#ifdef RE_TX_MODERATION
2469		/*
2470		 * If not all descriptors have been reaped yet, reload
2471		 * the timer so that we will eventually get another
2472		 * interrupt that will cause us to re-enter this routine.
2473		 * This is done in case the transmitter has gone idle.
2474		 */
2475		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2476#endif
2477	} else
2478		sc->rl_watchdog_timer = 0;
2479}
2480
2481static void
2482re_tick(void *xsc)
2483{
2484	struct rl_softc		*sc;
2485	struct mii_data		*mii;
2486
2487	sc = xsc;
2488
2489	RL_LOCK_ASSERT(sc);
2490
2491	mii = device_get_softc(sc->rl_miibus);
2492	mii_tick(mii);
2493	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2494		re_miibus_statchg(sc->rl_dev);
2495	/*
2496	 * Reclaim transmitted frames here. Technically it is not
2497	 * necessary to do here but it ensures periodic reclamation
2498	 * regardless of Tx completion interrupt which seems to be
2499	 * lost on PCIe based controllers under certain situations.
2500	 */
2501	re_txeof(sc);
2502	re_watchdog(sc);
2503	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2504}
2505
2506#ifdef DEVICE_POLLING
2507static int
2508re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2509{
2510	struct rl_softc *sc = ifp->if_softc;
2511	int rx_npkts = 0;
2512
2513	RL_LOCK(sc);
2514	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2515		rx_npkts = re_poll_locked(ifp, cmd, count);
2516	RL_UNLOCK(sc);
2517	return (rx_npkts);
2518}
2519
2520static int
2521re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2522{
2523	struct rl_softc *sc = ifp->if_softc;
2524	int rx_npkts;
2525
2526	RL_LOCK_ASSERT(sc);
2527
2528	sc->rxcycles = count;
2529	re_rxeof(sc, &rx_npkts);
2530	re_txeof(sc);
2531
2532	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2533		re_start_locked(ifp);
2534
2535	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2536		u_int16_t       status;
2537
2538		status = CSR_READ_2(sc, RL_ISR);
2539		if (status == 0xffff)
2540			return (rx_npkts);
2541		if (status)
2542			CSR_WRITE_2(sc, RL_ISR, status);
2543		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2544		    (sc->rl_flags & RL_FLAG_PCIE))
2545			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2546
2547		/*
2548		 * XXX check behaviour on receiver stalls.
2549		 */
2550
2551		if (status & RL_ISR_SYSTEM_ERR) {
2552			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2553			re_init_locked(sc);
2554		}
2555	}
2556	return (rx_npkts);
2557}
2558#endif /* DEVICE_POLLING */
2559
2560static int
2561re_intr(void *arg)
2562{
2563	struct rl_softc		*sc;
2564	uint16_t		status;
2565
2566	sc = arg;
2567
2568	status = CSR_READ_2(sc, RL_ISR);
2569	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2570                return (FILTER_STRAY);
2571	CSR_WRITE_2(sc, RL_IMR, 0);
2572
2573	taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2574
2575	return (FILTER_HANDLED);
2576}
2577
2578static void
2579re_int_task(void *arg, int npending)
2580{
2581	struct rl_softc		*sc;
2582	struct ifnet		*ifp;
2583	u_int16_t		status;
2584	int			rval = 0;
2585
2586	sc = arg;
2587	ifp = sc->rl_ifp;
2588
2589	RL_LOCK(sc);
2590
2591	status = CSR_READ_2(sc, RL_ISR);
2592        CSR_WRITE_2(sc, RL_ISR, status);
2593
2594	if (sc->suspended ||
2595	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2596		RL_UNLOCK(sc);
2597		return;
2598	}
2599
2600#ifdef DEVICE_POLLING
2601	if  (ifp->if_capenable & IFCAP_POLLING) {
2602		RL_UNLOCK(sc);
2603		return;
2604	}
2605#endif
2606
2607	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2608		rval = re_rxeof(sc, NULL);
2609
2610	/*
2611	 * Some chips will ignore a second TX request issued
2612	 * while an existing transmission is in progress. If
2613	 * the transmitter goes idle but there are still
2614	 * packets waiting to be sent, we need to restart the
2615	 * channel here to flush them out. This only seems to
2616	 * be required with the PCIe devices.
2617	 */
2618	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2619	    (sc->rl_flags & RL_FLAG_PCIE))
2620		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2621	if (status & (
2622#ifdef RE_TX_MODERATION
2623	    RL_ISR_TIMEOUT_EXPIRED|
2624#else
2625	    RL_ISR_TX_OK|
2626#endif
2627	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2628		re_txeof(sc);
2629
2630	if (status & RL_ISR_SYSTEM_ERR) {
2631		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2632		re_init_locked(sc);
2633	}
2634
2635	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2636		re_start_locked(ifp);
2637
2638	RL_UNLOCK(sc);
2639
2640        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2641		taskqueue_enqueue(taskqueue_fast, &sc->rl_inttask);
2642		return;
2643	}
2644
2645	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2646}
2647
2648static void
2649re_intr_msi(void *xsc)
2650{
2651	struct rl_softc		*sc;
2652	struct ifnet		*ifp;
2653	uint16_t		intrs, status;
2654
2655	sc = xsc;
2656	RL_LOCK(sc);
2657
2658	ifp = sc->rl_ifp;
2659#ifdef DEVICE_POLLING
2660	if (ifp->if_capenable & IFCAP_POLLING) {
2661		RL_UNLOCK(sc);
2662		return;
2663	}
2664#endif
2665	/* Disable interrupts. */
2666	CSR_WRITE_2(sc, RL_IMR, 0);
2667	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2668		RL_UNLOCK(sc);
2669		return;
2670	}
2671
2672	intrs = RL_INTRS_CPLUS;
2673	status = CSR_READ_2(sc, RL_ISR);
2674        CSR_WRITE_2(sc, RL_ISR, status);
2675	if (sc->rl_int_rx_act > 0) {
2676		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2677		    RL_ISR_RX_OVERRUN);
2678		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2679		    RL_ISR_RX_OVERRUN);
2680	}
2681
2682	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2683	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2684		re_rxeof(sc, NULL);
2685		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2686			if (sc->rl_int_rx_mod != 0 &&
2687			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2688			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2689				/* Rearm one-shot timer. */
2690				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2691				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2692				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2693				sc->rl_int_rx_act = 1;
2694			} else {
2695				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2696				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2697				sc->rl_int_rx_act = 0;
2698			}
2699		}
2700	}
2701
2702	/*
2703	 * Some chips will ignore a second TX request issued
2704	 * while an existing transmission is in progress. If
2705	 * the transmitter goes idle but there are still
2706	 * packets waiting to be sent, we need to restart the
2707	 * channel here to flush them out. This only seems to
2708	 * be required with the PCIe devices.
2709	 */
2710	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2711	    (sc->rl_flags & RL_FLAG_PCIE))
2712		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2713	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2714		re_txeof(sc);
2715
2716	if (status & RL_ISR_SYSTEM_ERR) {
2717		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2718		re_init_locked(sc);
2719	}
2720
2721	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2722		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2723			re_start_locked(ifp);
2724		CSR_WRITE_2(sc, RL_IMR, intrs);
2725	}
2726	RL_UNLOCK(sc);
2727}
2728
2729static int
2730re_encap(struct rl_softc *sc, struct mbuf **m_head)
2731{
2732	struct rl_txdesc	*txd, *txd_last;
2733	bus_dma_segment_t	segs[RL_NTXSEGS];
2734	bus_dmamap_t		map;
2735	struct mbuf		*m_new;
2736	struct rl_desc		*desc;
2737	int			nsegs, prod;
2738	int			i, error, ei, si;
2739	int			padlen;
2740	uint32_t		cmdstat, csum_flags, vlanctl;
2741
2742	RL_LOCK_ASSERT(sc);
2743	M_ASSERTPKTHDR((*m_head));
2744
2745	/*
2746	 * With some of the RealTek chips, using the checksum offload
2747	 * support in conjunction with the autopadding feature results
2748	 * in the transmission of corrupt frames. For example, if we
2749	 * need to send a really small IP fragment that's less than 60
2750	 * bytes in size, and IP header checksumming is enabled, the
2751	 * resulting ethernet frame that appears on the wire will
2752	 * have garbled payload. To work around this, if TX IP checksum
2753	 * offload is enabled, we always manually pad short frames out
2754	 * to the minimum ethernet frame size.
2755	 */
2756	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2757	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2758	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2759		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2760		if (M_WRITABLE(*m_head) == 0) {
2761			/* Get a writable copy. */
2762			m_new = m_dup(*m_head, M_NOWAIT);
2763			m_freem(*m_head);
2764			if (m_new == NULL) {
2765				*m_head = NULL;
2766				return (ENOBUFS);
2767			}
2768			*m_head = m_new;
2769		}
2770		if ((*m_head)->m_next != NULL ||
2771		    M_TRAILINGSPACE(*m_head) < padlen) {
2772			m_new = m_defrag(*m_head, M_NOWAIT);
2773			if (m_new == NULL) {
2774				m_freem(*m_head);
2775				*m_head = NULL;
2776				return (ENOBUFS);
2777			}
2778		} else
2779			m_new = *m_head;
2780
2781		/*
2782		 * Manually pad short frames, and zero the pad space
2783		 * to avoid leaking data.
2784		 */
2785		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2786		m_new->m_pkthdr.len += padlen;
2787		m_new->m_len = m_new->m_pkthdr.len;
2788		*m_head = m_new;
2789	}
2790
2791	prod = sc->rl_ldata.rl_tx_prodidx;
2792	txd = &sc->rl_ldata.rl_tx_desc[prod];
2793	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2794	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2795	if (error == EFBIG) {
2796		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2797		if (m_new == NULL) {
2798			m_freem(*m_head);
2799			*m_head = NULL;
2800			return (ENOBUFS);
2801		}
2802		*m_head = m_new;
2803		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2804		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2805		if (error != 0) {
2806			m_freem(*m_head);
2807			*m_head = NULL;
2808			return (error);
2809		}
2810	} else if (error != 0)
2811		return (error);
2812	if (nsegs == 0) {
2813		m_freem(*m_head);
2814		*m_head = NULL;
2815		return (EIO);
2816	}
2817
2818	/* Check for number of available descriptors. */
2819	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2820		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2821		return (ENOBUFS);
2822	}
2823
2824	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2825	    BUS_DMASYNC_PREWRITE);
2826
2827	/*
2828	 * Set up checksum offload. Note: checksum offload bits must
2829	 * appear in all descriptors of a multi-descriptor transmit
2830	 * attempt. This is according to testing done with an 8169
2831	 * chip. This is a requirement.
2832	 */
2833	vlanctl = 0;
2834	csum_flags = 0;
2835	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2836		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2837			csum_flags |= RL_TDESC_CMD_LGSEND;
2838			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2839			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2840		} else {
2841			csum_flags |= RL_TDESC_CMD_LGSEND |
2842			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2843			    RL_TDESC_CMD_MSSVAL_SHIFT);
2844		}
2845	} else {
2846		/*
2847		 * Unconditionally enable IP checksum if TCP or UDP
2848		 * checksum is required. Otherwise, TCP/UDP checksum
2849		 * doesn't make effects.
2850		 */
2851		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2852			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2853				csum_flags |= RL_TDESC_CMD_IPCSUM;
2854				if (((*m_head)->m_pkthdr.csum_flags &
2855				    CSUM_TCP) != 0)
2856					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2857				if (((*m_head)->m_pkthdr.csum_flags &
2858				    CSUM_UDP) != 0)
2859					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2860			} else {
2861				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2862				if (((*m_head)->m_pkthdr.csum_flags &
2863				    CSUM_TCP) != 0)
2864					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2865				if (((*m_head)->m_pkthdr.csum_flags &
2866				    CSUM_UDP) != 0)
2867					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2868			}
2869		}
2870	}
2871
2872	/*
2873	 * Set up hardware VLAN tagging. Note: vlan tag info must
2874	 * appear in all descriptors of a multi-descriptor
2875	 * transmission attempt.
2876	 */
2877	if ((*m_head)->m_flags & M_VLANTAG)
2878		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2879		    RL_TDESC_VLANCTL_TAG;
2880
2881	si = prod;
2882	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2883		desc = &sc->rl_ldata.rl_tx_list[prod];
2884		desc->rl_vlanctl = htole32(vlanctl);
2885		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2886		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2887		cmdstat = segs[i].ds_len;
2888		if (i != 0)
2889			cmdstat |= RL_TDESC_CMD_OWN;
2890		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2891			cmdstat |= RL_TDESC_CMD_EOR;
2892		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2893		sc->rl_ldata.rl_tx_free--;
2894	}
2895	/* Update producer index. */
2896	sc->rl_ldata.rl_tx_prodidx = prod;
2897
2898	/* Set EOF on the last descriptor. */
2899	ei = RL_TX_DESC_PRV(sc, prod);
2900	desc = &sc->rl_ldata.rl_tx_list[ei];
2901	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2902
2903	desc = &sc->rl_ldata.rl_tx_list[si];
2904	/* Set SOF and transfer ownership of packet to the chip. */
2905	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2906
2907	/*
2908	 * Insure that the map for this transmission
2909	 * is placed at the array index of the last descriptor
2910	 * in this chain.  (Swap last and first dmamaps.)
2911	 */
2912	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2913	map = txd->tx_dmamap;
2914	txd->tx_dmamap = txd_last->tx_dmamap;
2915	txd_last->tx_dmamap = map;
2916	txd_last->tx_m = *m_head;
2917
2918	return (0);
2919}
2920
2921static void
2922re_start(struct ifnet *ifp)
2923{
2924	struct rl_softc		*sc;
2925
2926	sc = ifp->if_softc;
2927	RL_LOCK(sc);
2928	re_start_locked(ifp);
2929	RL_UNLOCK(sc);
2930}
2931
2932/*
2933 * Main transmit routine for C+ and gigE NICs.
2934 */
2935static void
2936re_start_locked(struct ifnet *ifp)
2937{
2938	struct rl_softc		*sc;
2939	struct mbuf		*m_head;
2940	int			queued;
2941
2942	sc = ifp->if_softc;
2943
2944#ifdef DEV_NETMAP
2945	/* XXX is this necessary ? */
2946	if (ifp->if_capenable & IFCAP_NETMAP) {
2947		struct netmap_kring *kring = NA(ifp)->tx_rings[0];
2948		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2949			/* kick the tx unit */
2950			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2951#ifdef RE_TX_MODERATION
2952			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2953#endif
2954			sc->rl_watchdog_timer = 5;
2955		}
2956		return;
2957	}
2958#endif /* DEV_NETMAP */
2959
2960	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2961	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2962		return;
2963
2964	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2965	    sc->rl_ldata.rl_tx_free > 1;) {
2966		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2967		if (m_head == NULL)
2968			break;
2969
2970		if (re_encap(sc, &m_head) != 0) {
2971			if (m_head == NULL)
2972				break;
2973			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2974			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2975			break;
2976		}
2977
2978		/*
2979		 * If there's a BPF listener, bounce a copy of this frame
2980		 * to him.
2981		 */
2982		ETHER_BPF_MTAP(ifp, m_head);
2983
2984		queued++;
2985	}
2986
2987	if (queued == 0) {
2988#ifdef RE_TX_MODERATION
2989		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2990			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2991#endif
2992		return;
2993	}
2994
2995	re_start_tx(sc);
2996}
2997
2998static void
2999re_start_tx(struct rl_softc *sc)
3000{
3001
3002	/* Flush the TX descriptors */
3003	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
3004	    sc->rl_ldata.rl_tx_list_map,
3005	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
3006
3007	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
3008
3009#ifdef RE_TX_MODERATION
3010	/*
3011	 * Use the countdown timer for interrupt moderation.
3012	 * 'TX done' interrupts are disabled. Instead, we reset the
3013	 * countdown timer, which will begin counting until it hits
3014	 * the value in the TIMERINT register, and then trigger an
3015	 * interrupt. Each time we write to the TIMERCNT register,
3016	 * the timer count is reset to 0.
3017	 */
3018	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
3019#endif
3020
3021	/*
3022	 * Set a timeout in case the chip goes out to lunch.
3023	 */
3024	sc->rl_watchdog_timer = 5;
3025}
3026
3027static void
3028re_set_jumbo(struct rl_softc *sc, int jumbo)
3029{
3030
3031	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
3032		pci_set_max_read_req(sc->rl_dev, 4096);
3033		return;
3034	}
3035
3036	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3037	if (jumbo != 0) {
3038		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3039		    RL_CFG3_JUMBO_EN0);
3040		switch (sc->rl_hwrev->rl_rev) {
3041		case RL_HWREV_8168DP:
3042			break;
3043		case RL_HWREV_8168E:
3044			CSR_WRITE_1(sc, sc->rl_cfg4,
3045			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3046			break;
3047		default:
3048			CSR_WRITE_1(sc, sc->rl_cfg4,
3049			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3050		}
3051	} else {
3052		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3053		    ~RL_CFG3_JUMBO_EN0);
3054		switch (sc->rl_hwrev->rl_rev) {
3055		case RL_HWREV_8168DP:
3056			break;
3057		case RL_HWREV_8168E:
3058			CSR_WRITE_1(sc, sc->rl_cfg4,
3059			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3060			break;
3061		default:
3062			CSR_WRITE_1(sc, sc->rl_cfg4,
3063			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3064		}
3065	}
3066	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3067
3068	switch (sc->rl_hwrev->rl_rev) {
3069	case RL_HWREV_8168DP:
3070		pci_set_max_read_req(sc->rl_dev, 4096);
3071		break;
3072	default:
3073		if (jumbo != 0)
3074			pci_set_max_read_req(sc->rl_dev, 512);
3075		else
3076			pci_set_max_read_req(sc->rl_dev, 4096);
3077	}
3078}
3079
3080static void
3081re_init(void *xsc)
3082{
3083	struct rl_softc		*sc = xsc;
3084
3085	RL_LOCK(sc);
3086	re_init_locked(sc);
3087	RL_UNLOCK(sc);
3088}
3089
3090static void
3091re_init_locked(struct rl_softc *sc)
3092{
3093	struct ifnet		*ifp = sc->rl_ifp;
3094	struct mii_data		*mii;
3095	uint32_t		reg;
3096	uint16_t		cfg;
3097	union {
3098		uint32_t align_dummy;
3099		u_char eaddr[ETHER_ADDR_LEN];
3100        } eaddr;
3101
3102	RL_LOCK_ASSERT(sc);
3103
3104	mii = device_get_softc(sc->rl_miibus);
3105
3106	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3107		return;
3108
3109	/*
3110	 * Cancel pending I/O and free all RX/TX buffers.
3111	 */
3112	re_stop(sc);
3113
3114	/* Put controller into known state. */
3115	re_reset(sc);
3116
3117	/*
3118	 * For C+ mode, initialize the RX descriptors and mbufs.
3119	 */
3120	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3121		if (ifp->if_mtu > RL_MTU) {
3122			if (re_jrx_list_init(sc) != 0) {
3123				device_printf(sc->rl_dev,
3124				    "no memory for jumbo RX buffers\n");
3125				re_stop(sc);
3126				return;
3127			}
3128			/* Disable checksum offloading for jumbo frames. */
3129			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3130			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3131		} else {
3132			if (re_rx_list_init(sc) != 0) {
3133				device_printf(sc->rl_dev,
3134				    "no memory for RX buffers\n");
3135				re_stop(sc);
3136				return;
3137			}
3138		}
3139		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3140	} else {
3141		if (re_rx_list_init(sc) != 0) {
3142			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3143			re_stop(sc);
3144			return;
3145		}
3146		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3147		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3148			if (ifp->if_mtu > RL_MTU)
3149				pci_set_max_read_req(sc->rl_dev, 512);
3150			else
3151				pci_set_max_read_req(sc->rl_dev, 4096);
3152		}
3153	}
3154	re_tx_list_init(sc);
3155
3156	/*
3157	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3158	 * RX checksum offload. We must configure the C+ register
3159	 * before all others.
3160	 */
3161	cfg = RL_CPLUSCMD_PCI_MRW;
3162	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3163		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3164	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3165		cfg |= RL_CPLUSCMD_VLANSTRIP;
3166	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3167		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3168		/* XXX magic. */
3169		cfg |= 0x0001;
3170	} else
3171		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3172	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3173	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3174	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3175		reg = 0x000fff00;
3176		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3177			reg |= 0x000000ff;
3178		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3179			reg |= 0x00f00000;
3180		CSR_WRITE_4(sc, 0x7c, reg);
3181		/* Disable interrupt mitigation. */
3182		CSR_WRITE_2(sc, 0xe2, 0);
3183	}
3184	/*
3185	 * Disable TSO if interface MTU size is greater than MSS
3186	 * allowed in controller.
3187	 */
3188	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3189		ifp->if_capenable &= ~IFCAP_TSO4;
3190		ifp->if_hwassist &= ~CSUM_TSO;
3191	}
3192
3193	/*
3194	 * Init our MAC address.  Even though the chipset
3195	 * documentation doesn't mention it, we need to enter "Config
3196	 * register write enable" mode to modify the ID registers.
3197	 */
3198	/* Copy MAC address on stack to align. */
3199	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3200	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3201	CSR_WRITE_4(sc, RL_IDR0,
3202	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3203	CSR_WRITE_4(sc, RL_IDR4,
3204	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3205	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3206
3207	/*
3208	 * Load the addresses of the RX and TX lists into the chip.
3209	 */
3210
3211	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3212	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3213	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3214	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3215
3216	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3217	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3218	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3219	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3220
3221	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3222		/* Disable RXDV gate. */
3223		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3224		    ~0x00080000);
3225	}
3226
3227	/*
3228	 * Enable transmit and receive for pre-RTL8168G controllers.
3229	 * RX/TX MACs should be enabled before RX/TX configuration.
3230	 */
3231	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) == 0)
3232		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3233
3234	/*
3235	 * Set the initial TX configuration.
3236	 */
3237	if (sc->rl_testmode) {
3238		if (sc->rl_type == RL_8169)
3239			CSR_WRITE_4(sc, RL_TXCFG,
3240			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3241		else
3242			CSR_WRITE_4(sc, RL_TXCFG,
3243			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3244	} else
3245		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3246
3247	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3248
3249	/*
3250	 * Set the initial RX configuration.
3251	 */
3252	re_set_rxmode(sc);
3253
3254	/* Configure interrupt moderation. */
3255	if (sc->rl_type == RL_8169) {
3256		/* Magic from vendor. */
3257		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3258	}
3259
3260	/*
3261	 * Enable transmit and receive for RTL8168G and later controllers.
3262	 * RX/TX MACs should be enabled after RX/TX configuration.
3263	 */
3264	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0)
3265		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
3266
3267#ifdef DEVICE_POLLING
3268	/*
3269	 * Disable interrupts if we are polling.
3270	 */
3271	if (ifp->if_capenable & IFCAP_POLLING)
3272		CSR_WRITE_2(sc, RL_IMR, 0);
3273	else	/* otherwise ... */
3274#endif
3275
3276	/*
3277	 * Enable interrupts.
3278	 */
3279	if (sc->rl_testmode)
3280		CSR_WRITE_2(sc, RL_IMR, 0);
3281	else
3282		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3283	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3284
3285	/* Set initial TX threshold */
3286	sc->rl_txthresh = RL_TX_THRESH_INIT;
3287
3288	/* Start RX/TX process. */
3289	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3290
3291	/*
3292	 * Initialize the timer interrupt register so that
3293	 * a timer interrupt will be generated once the timer
3294	 * reaches a certain number of ticks. The timer is
3295	 * reloaded on each transmit.
3296	 */
3297#ifdef RE_TX_MODERATION
3298	/*
3299	 * Use timer interrupt register to moderate TX interrupt
3300	 * moderation, which dramatically improves TX frame rate.
3301	 */
3302	if (sc->rl_type == RL_8169)
3303		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3304	else
3305		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3306#else
3307	/*
3308	 * Use timer interrupt register to moderate RX interrupt
3309	 * moderation.
3310	 */
3311	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3312	    intr_filter == 0) {
3313		if (sc->rl_type == RL_8169)
3314			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3315			    RL_USECS(sc->rl_int_rx_mod));
3316	} else {
3317		if (sc->rl_type == RL_8169)
3318			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3319	}
3320#endif
3321
3322	/*
3323	 * For 8169 gigE NICs, set the max allowed RX packet
3324	 * size so we can receive jumbo frames.
3325	 */
3326	if (sc->rl_type == RL_8169) {
3327		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3328			/*
3329			 * For controllers that use new jumbo frame scheme,
3330			 * set maximum size of jumbo frame depending on
3331			 * controller revisions.
3332			 */
3333			if (ifp->if_mtu > RL_MTU)
3334				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3335				    sc->rl_hwrev->rl_max_mtu +
3336				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3337				    ETHER_CRC_LEN);
3338			else
3339				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3340				    RE_RX_DESC_BUFLEN);
3341		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3342		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3343			/* RTL810x has no jumbo frame support. */
3344			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3345		} else
3346			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3347	}
3348
3349	if (sc->rl_testmode)
3350		return;
3351
3352	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3353	    RL_CFG1_DRVLOAD);
3354
3355	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3356	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3357
3358	sc->rl_flags &= ~RL_FLAG_LINK;
3359	mii_mediachg(mii);
3360
3361	sc->rl_watchdog_timer = 0;
3362	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3363
3364#ifdef DEV_NETMAP
3365	netmap_enable_all_rings(ifp);
3366#endif /* DEV_NETMAP */
3367}
3368
3369/*
3370 * Set media options.
3371 */
3372static int
3373re_ifmedia_upd(struct ifnet *ifp)
3374{
3375	struct rl_softc		*sc;
3376	struct mii_data		*mii;
3377	int			error;
3378
3379	sc = ifp->if_softc;
3380	mii = device_get_softc(sc->rl_miibus);
3381	RL_LOCK(sc);
3382	error = mii_mediachg(mii);
3383	RL_UNLOCK(sc);
3384
3385	return (error);
3386}
3387
3388/*
3389 * Report current media status.
3390 */
3391static void
3392re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3393{
3394	struct rl_softc		*sc;
3395	struct mii_data		*mii;
3396
3397	sc = ifp->if_softc;
3398	mii = device_get_softc(sc->rl_miibus);
3399
3400	RL_LOCK(sc);
3401	mii_pollstat(mii);
3402	ifmr->ifm_active = mii->mii_media_active;
3403	ifmr->ifm_status = mii->mii_media_status;
3404	RL_UNLOCK(sc);
3405}
3406
3407static int
3408re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3409{
3410	struct rl_softc		*sc = ifp->if_softc;
3411	struct ifreq		*ifr = (struct ifreq *) data;
3412	struct mii_data		*mii;
3413	int			error = 0;
3414
3415	switch (command) {
3416	case SIOCSIFMTU:
3417		if (ifr->ifr_mtu < ETHERMIN ||
3418		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3419		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3420		    ifr->ifr_mtu > RL_MTU)) {
3421			error = EINVAL;
3422			break;
3423		}
3424		RL_LOCK(sc);
3425		if (ifp->if_mtu != ifr->ifr_mtu) {
3426			ifp->if_mtu = ifr->ifr_mtu;
3427			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3428			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3429				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3430				re_init_locked(sc);
3431			}
3432			if (ifp->if_mtu > RL_TSO_MTU &&
3433			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3434				ifp->if_capenable &= ~(IFCAP_TSO4 |
3435				    IFCAP_VLAN_HWTSO);
3436				ifp->if_hwassist &= ~CSUM_TSO;
3437			}
3438			VLAN_CAPABILITIES(ifp);
3439		}
3440		RL_UNLOCK(sc);
3441		break;
3442	case SIOCSIFFLAGS:
3443		RL_LOCK(sc);
3444		if ((ifp->if_flags & IFF_UP) != 0) {
3445			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3446				if (((ifp->if_flags ^ sc->rl_if_flags)
3447				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3448					re_set_rxmode(sc);
3449			} else
3450				re_init_locked(sc);
3451		} else {
3452			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3453				re_stop(sc);
3454		}
3455		sc->rl_if_flags = ifp->if_flags;
3456		RL_UNLOCK(sc);
3457		break;
3458	case SIOCADDMULTI:
3459	case SIOCDELMULTI:
3460		RL_LOCK(sc);
3461		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3462			re_set_rxmode(sc);
3463		RL_UNLOCK(sc);
3464		break;
3465	case SIOCGIFMEDIA:
3466	case SIOCSIFMEDIA:
3467		mii = device_get_softc(sc->rl_miibus);
3468		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3469		break;
3470	case SIOCSIFCAP:
3471	    {
3472		int mask, reinit;
3473
3474		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3475		reinit = 0;
3476#ifdef DEVICE_POLLING
3477		if (mask & IFCAP_POLLING) {
3478			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3479				error = ether_poll_register(re_poll, ifp);
3480				if (error)
3481					return (error);
3482				RL_LOCK(sc);
3483				/* Disable interrupts */
3484				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3485				ifp->if_capenable |= IFCAP_POLLING;
3486				RL_UNLOCK(sc);
3487			} else {
3488				error = ether_poll_deregister(ifp);
3489				/* Enable interrupts. */
3490				RL_LOCK(sc);
3491				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3492				ifp->if_capenable &= ~IFCAP_POLLING;
3493				RL_UNLOCK(sc);
3494			}
3495		}
3496#endif /* DEVICE_POLLING */
3497		RL_LOCK(sc);
3498		if ((mask & IFCAP_TXCSUM) != 0 &&
3499		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3500			ifp->if_capenable ^= IFCAP_TXCSUM;
3501			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3502				ifp->if_hwassist |= RE_CSUM_FEATURES;
3503			else
3504				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3505			reinit = 1;
3506		}
3507		if ((mask & IFCAP_RXCSUM) != 0 &&
3508		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3509			ifp->if_capenable ^= IFCAP_RXCSUM;
3510			reinit = 1;
3511		}
3512		if ((mask & IFCAP_TSO4) != 0 &&
3513		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3514			ifp->if_capenable ^= IFCAP_TSO4;
3515			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3516				ifp->if_hwassist |= CSUM_TSO;
3517			else
3518				ifp->if_hwassist &= ~CSUM_TSO;
3519			if (ifp->if_mtu > RL_TSO_MTU &&
3520			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3521				ifp->if_capenable &= ~IFCAP_TSO4;
3522				ifp->if_hwassist &= ~CSUM_TSO;
3523			}
3524		}
3525		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3526		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3527			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3528		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3529		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3530			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3531			/* TSO over VLAN requires VLAN hardware tagging. */
3532			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3533				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3534			reinit = 1;
3535		}
3536		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3537		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3538		    IFCAP_VLAN_HWTSO)) != 0)
3539				reinit = 1;
3540		if ((mask & IFCAP_WOL) != 0 &&
3541		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3542			if ((mask & IFCAP_WOL_UCAST) != 0)
3543				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3544			if ((mask & IFCAP_WOL_MCAST) != 0)
3545				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3546			if ((mask & IFCAP_WOL_MAGIC) != 0)
3547				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3548		}
3549		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3550			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3551			re_init_locked(sc);
3552		}
3553		RL_UNLOCK(sc);
3554		VLAN_CAPABILITIES(ifp);
3555	    }
3556		break;
3557	default:
3558		error = ether_ioctl(ifp, command, data);
3559		break;
3560	}
3561
3562	return (error);
3563}
3564
3565static void
3566re_watchdog(struct rl_softc *sc)
3567{
3568	struct ifnet		*ifp;
3569
3570	RL_LOCK_ASSERT(sc);
3571
3572	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3573		return;
3574
3575	ifp = sc->rl_ifp;
3576	re_txeof(sc);
3577	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3578		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3579		    "-- recovering\n");
3580		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3581			re_start_locked(ifp);
3582		return;
3583	}
3584
3585	if_printf(ifp, "watchdog timeout\n");
3586	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
3587
3588	re_rxeof(sc, NULL);
3589	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3590	re_init_locked(sc);
3591	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3592		re_start_locked(ifp);
3593}
3594
3595/*
3596 * Stop the adapter and free any mbufs allocated to the
3597 * RX and TX lists.
3598 */
3599static void
3600re_stop(struct rl_softc *sc)
3601{
3602	int			i;
3603	struct ifnet		*ifp;
3604	struct rl_txdesc	*txd;
3605	struct rl_rxdesc	*rxd;
3606
3607	RL_LOCK_ASSERT(sc);
3608
3609	ifp = sc->rl_ifp;
3610
3611	sc->rl_watchdog_timer = 0;
3612	callout_stop(&sc->rl_stat_callout);
3613	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3614
3615#ifdef DEV_NETMAP
3616	netmap_disable_all_rings(ifp);
3617#endif /* DEV_NETMAP */
3618
3619	/*
3620	 * Disable accepting frames to put RX MAC into idle state.
3621	 * Otherwise it's possible to get frames while stop command
3622	 * execution is in progress and controller can DMA the frame
3623	 * to already freed RX buffer during that period.
3624	 */
3625	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3626	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3627	    RL_RXCFG_RX_BROAD));
3628
3629	if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3630		/* Enable RXDV gate. */
3631		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) |
3632		    0x00080000);
3633	}
3634
3635	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3636		for (i = RL_TIMEOUT; i > 0; i--) {
3637			if ((CSR_READ_1(sc, sc->rl_txstart) &
3638			    RL_TXSTART_START) == 0)
3639				break;
3640			DELAY(20);
3641		}
3642		if (i == 0)
3643			device_printf(sc->rl_dev,
3644			    "stopping TX poll timed out!\n");
3645		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3646	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3647		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3648		    RL_CMD_RX_ENB);
3649		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3650			for (i = RL_TIMEOUT; i > 0; i--) {
3651				if ((CSR_READ_4(sc, RL_TXCFG) &
3652				    RL_TXCFG_QUEUE_EMPTY) != 0)
3653					break;
3654				DELAY(100);
3655			}
3656			if (i == 0)
3657				device_printf(sc->rl_dev,
3658				   "stopping TXQ timed out!\n");
3659		}
3660	} else
3661		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3662	DELAY(1000);
3663	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3664	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3665
3666	if (sc->rl_head != NULL) {
3667		m_freem(sc->rl_head);
3668		sc->rl_head = sc->rl_tail = NULL;
3669	}
3670
3671	/* Free the TX list buffers. */
3672	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3673		txd = &sc->rl_ldata.rl_tx_desc[i];
3674		if (txd->tx_m != NULL) {
3675			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3676			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3677			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3678			    txd->tx_dmamap);
3679			m_freem(txd->tx_m);
3680			txd->tx_m = NULL;
3681		}
3682	}
3683
3684	/* Free the RX list buffers. */
3685	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3686		rxd = &sc->rl_ldata.rl_rx_desc[i];
3687		if (rxd->rx_m != NULL) {
3688			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3689			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3690			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3691			    rxd->rx_dmamap);
3692			m_freem(rxd->rx_m);
3693			rxd->rx_m = NULL;
3694		}
3695	}
3696
3697	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3698		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3699			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3700			if (rxd->rx_m != NULL) {
3701				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3702				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3703				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3704				    rxd->rx_dmamap);
3705				m_freem(rxd->rx_m);
3706				rxd->rx_m = NULL;
3707			}
3708		}
3709	}
3710}
3711
3712/*
3713 * Device suspend routine.  Stop the interface and save some PCI
3714 * settings in case the BIOS doesn't restore them properly on
3715 * resume.
3716 */
3717static int
3718re_suspend(device_t dev)
3719{
3720	struct rl_softc		*sc;
3721
3722	sc = device_get_softc(dev);
3723
3724	RL_LOCK(sc);
3725	re_stop(sc);
3726	re_setwol(sc);
3727	sc->suspended = 1;
3728	RL_UNLOCK(sc);
3729
3730	return (0);
3731}
3732
3733/*
3734 * Device resume routine.  Restore some PCI settings in case the BIOS
3735 * doesn't, re-enable busmastering, and restart the interface if
3736 * appropriate.
3737 */
3738static int
3739re_resume(device_t dev)
3740{
3741	struct rl_softc		*sc;
3742	struct ifnet		*ifp;
3743
3744	sc = device_get_softc(dev);
3745
3746	RL_LOCK(sc);
3747
3748	ifp = sc->rl_ifp;
3749	/* Take controller out of sleep mode. */
3750	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3751		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3752			CSR_WRITE_1(sc, RL_GPIO,
3753			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3754	}
3755
3756	/*
3757	 * Clear WOL matching such that normal Rx filtering
3758	 * wouldn't interfere with WOL patterns.
3759	 */
3760	re_clrwol(sc);
3761
3762	/* reinitialize interface if necessary */
3763	if (ifp->if_flags & IFF_UP)
3764		re_init_locked(sc);
3765
3766	sc->suspended = 0;
3767	RL_UNLOCK(sc);
3768
3769	return (0);
3770}
3771
3772/*
3773 * Stop all chip I/O so that the kernel's probe routines don't
3774 * get confused by errant DMAs when rebooting.
3775 */
3776static int
3777re_shutdown(device_t dev)
3778{
3779	struct rl_softc		*sc;
3780
3781	sc = device_get_softc(dev);
3782
3783	RL_LOCK(sc);
3784	re_stop(sc);
3785	/*
3786	 * Mark interface as down since otherwise we will panic if
3787	 * interrupt comes in later on, which can happen in some
3788	 * cases.
3789	 */
3790	sc->rl_ifp->if_flags &= ~IFF_UP;
3791	re_setwol(sc);
3792	RL_UNLOCK(sc);
3793
3794	return (0);
3795}
3796
3797static void
3798re_set_linkspeed(struct rl_softc *sc)
3799{
3800	struct mii_softc *miisc;
3801	struct mii_data *mii;
3802	int aneg, i, phyno;
3803
3804	RL_LOCK_ASSERT(sc);
3805
3806	mii = device_get_softc(sc->rl_miibus);
3807	mii_pollstat(mii);
3808	aneg = 0;
3809	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3810	    (IFM_ACTIVE | IFM_AVALID)) {
3811		switch IFM_SUBTYPE(mii->mii_media_active) {
3812		case IFM_10_T:
3813		case IFM_100_TX:
3814			return;
3815		case IFM_1000_T:
3816			aneg++;
3817			break;
3818		default:
3819			break;
3820		}
3821	}
3822	miisc = LIST_FIRST(&mii->mii_phys);
3823	phyno = miisc->mii_phy;
3824	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3825		PHY_RESET(miisc);
3826	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3827	re_miibus_writereg(sc->rl_dev, phyno,
3828	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3829	re_miibus_writereg(sc->rl_dev, phyno,
3830	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3831	DELAY(1000);
3832	if (aneg != 0) {
3833		/*
3834		 * Poll link state until re(4) get a 10/100Mbps link.
3835		 */
3836		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3837			mii_pollstat(mii);
3838			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3839			    == (IFM_ACTIVE | IFM_AVALID)) {
3840				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3841				case IFM_10_T:
3842				case IFM_100_TX:
3843					return;
3844				default:
3845					break;
3846				}
3847			}
3848			RL_UNLOCK(sc);
3849			pause("relnk", hz);
3850			RL_LOCK(sc);
3851		}
3852		if (i == MII_ANEGTICKS_GIGE)
3853			device_printf(sc->rl_dev,
3854			    "establishing a link failed, WOL may not work!");
3855	}
3856	/*
3857	 * No link, force MAC to have 100Mbps, full-duplex link.
3858	 * MAC does not require reprogramming on resolved speed/duplex,
3859	 * so this is just for completeness.
3860	 */
3861	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3862	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3863}
3864
3865static void
3866re_setwol(struct rl_softc *sc)
3867{
3868	struct ifnet		*ifp;
3869	int			pmc;
3870	uint16_t		pmstat;
3871	uint8_t			v;
3872
3873	RL_LOCK_ASSERT(sc);
3874
3875	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3876		return;
3877
3878	ifp = sc->rl_ifp;
3879	/* Put controller into sleep mode. */
3880	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3881		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3882			CSR_WRITE_1(sc, RL_GPIO,
3883			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3884	}
3885	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3886		if ((sc->rl_flags & RL_FLAG_8168G_PLUS) != 0) {
3887			/* Disable RXDV gate. */
3888			CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3889			    ~0x00080000);
3890		}
3891		re_set_rxmode(sc);
3892		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3893			re_set_linkspeed(sc);
3894		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3895			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3896	}
3897	/* Enable config register write. */
3898	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3899
3900	/* Enable PME. */
3901	v = CSR_READ_1(sc, sc->rl_cfg1);
3902	v &= ~RL_CFG1_PME;
3903	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3904		v |= RL_CFG1_PME;
3905	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3906
3907	v = CSR_READ_1(sc, sc->rl_cfg3);
3908	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3909	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3910		v |= RL_CFG3_WOL_MAGIC;
3911	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3912
3913	v = CSR_READ_1(sc, sc->rl_cfg5);
3914	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3915	    RL_CFG5_WOL_LANWAKE);
3916	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3917		v |= RL_CFG5_WOL_UCAST;
3918	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3919		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3920	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3921		v |= RL_CFG5_WOL_LANWAKE;
3922	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3923
3924	/* Config register write done. */
3925	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3926
3927	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3928	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3929		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3930	/*
3931	 * It seems that hardware resets its link speed to 100Mbps in
3932	 * power down mode so switching to 100Mbps in driver is not
3933	 * needed.
3934	 */
3935
3936	/* Request PME if WOL is requested. */
3937	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3938	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3939	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3940		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3941	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3942}
3943
3944static void
3945re_clrwol(struct rl_softc *sc)
3946{
3947	int			pmc;
3948	uint8_t			v;
3949
3950	RL_LOCK_ASSERT(sc);
3951
3952	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3953		return;
3954
3955	/* Enable config register write. */
3956	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3957
3958	v = CSR_READ_1(sc, sc->rl_cfg3);
3959	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3960	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3961
3962	/* Config register write done. */
3963	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3964
3965	v = CSR_READ_1(sc, sc->rl_cfg5);
3966	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3967	v &= ~RL_CFG5_WOL_LANWAKE;
3968	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3969}
3970
3971static void
3972re_add_sysctls(struct rl_softc *sc)
3973{
3974	struct sysctl_ctx_list	*ctx;
3975	struct sysctl_oid_list	*children;
3976	int			error;
3977
3978	ctx = device_get_sysctl_ctx(sc->rl_dev);
3979	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3980
3981	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3982	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
3983	    re_sysctl_stats, "I", "Statistics Information");
3984	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3985		return;
3986
3987	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3988	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
3989	    &sc->rl_int_rx_mod, 0, sysctl_hw_re_int_mod, "I",
3990	    "re RX interrupt moderation");
3991	/* Pull in device tunables. */
3992	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3993	error = resource_int_value(device_get_name(sc->rl_dev),
3994	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3995	if (error == 0) {
3996		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3997		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3998			device_printf(sc->rl_dev, "int_rx_mod value out of "
3999			    "range; using default: %d\n",
4000			    RL_TIMER_DEFAULT);
4001			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
4002		}
4003	}
4004}
4005
4006static int
4007re_sysctl_stats(SYSCTL_HANDLER_ARGS)
4008{
4009	struct rl_softc		*sc;
4010	struct rl_stats		*stats;
4011	int			error, i, result;
4012
4013	result = -1;
4014	error = sysctl_handle_int(oidp, &result, 0, req);
4015	if (error || req->newptr == NULL)
4016		return (error);
4017
4018	if (result == 1) {
4019		sc = (struct rl_softc *)arg1;
4020		RL_LOCK(sc);
4021		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
4022			RL_UNLOCK(sc);
4023			goto done;
4024		}
4025		bus_dmamap_sync(sc->rl_ldata.rl_stag,
4026		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
4027		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
4028		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
4029		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
4030		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
4031		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
4032		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
4033		    RL_DUMPSTATS_START));
4034		for (i = RL_TIMEOUT; i > 0; i--) {
4035			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
4036			    RL_DUMPSTATS_START) == 0)
4037				break;
4038			DELAY(1000);
4039		}
4040		bus_dmamap_sync(sc->rl_ldata.rl_stag,
4041		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
4042		RL_UNLOCK(sc);
4043		if (i == 0) {
4044			device_printf(sc->rl_dev,
4045			    "DUMP statistics request timed out\n");
4046			return (ETIMEDOUT);
4047		}
4048done:
4049		stats = sc->rl_ldata.rl_stats;
4050		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
4051		printf("Tx frames : %ju\n",
4052		    (uintmax_t)le64toh(stats->rl_tx_pkts));
4053		printf("Rx frames : %ju\n",
4054		    (uintmax_t)le64toh(stats->rl_rx_pkts));
4055		printf("Tx errors : %ju\n",
4056		    (uintmax_t)le64toh(stats->rl_tx_errs));
4057		printf("Rx errors : %u\n",
4058		    le32toh(stats->rl_rx_errs));
4059		printf("Rx missed frames : %u\n",
4060		    (uint32_t)le16toh(stats->rl_missed_pkts));
4061		printf("Rx frame alignment errs : %u\n",
4062		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4063		printf("Tx single collisions : %u\n",
4064		    le32toh(stats->rl_tx_onecoll));
4065		printf("Tx multiple collisions : %u\n",
4066		    le32toh(stats->rl_tx_multicolls));
4067		printf("Rx unicast frames : %ju\n",
4068		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4069		printf("Rx broadcast frames : %ju\n",
4070		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4071		printf("Rx multicast frames : %u\n",
4072		    le32toh(stats->rl_rx_mcasts));
4073		printf("Tx aborts : %u\n",
4074		    (uint32_t)le16toh(stats->rl_tx_aborts));
4075		printf("Tx underruns : %u\n",
4076		    (uint32_t)le16toh(stats->rl_rx_underruns));
4077	}
4078
4079	return (error);
4080}
4081
4082static int
4083sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4084{
4085	int error, value;
4086
4087	if (arg1 == NULL)
4088		return (EINVAL);
4089	value = *(int *)arg1;
4090	error = sysctl_handle_int(oidp, &value, 0, req);
4091	if (error || req->newptr == NULL)
4092		return (error);
4093	if (value < low || value > high)
4094		return (EINVAL);
4095	*(int *)arg1 = value;
4096
4097	return (0);
4098}
4099
4100static int
4101sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4102{
4103
4104	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4105	    RL_TIMER_MAX));
4106}
4107
4108#ifdef DEBUGNET
4109static void
4110re_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize)
4111{
4112	struct rl_softc *sc;
4113
4114	sc = if_getsoftc(ifp);
4115	RL_LOCK(sc);
4116	*nrxr = sc->rl_ldata.rl_rx_desc_cnt;
4117	*ncl = DEBUGNET_MAX_IN_FLIGHT;
4118	*clsize = (ifp->if_mtu > RL_MTU &&
4119	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0) ? MJUM9BYTES : MCLBYTES;
4120	RL_UNLOCK(sc);
4121}
4122
4123static void
4124re_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused)
4125{
4126}
4127
4128static int
4129re_debugnet_transmit(struct ifnet *ifp, struct mbuf *m)
4130{
4131	struct rl_softc *sc;
4132	int error;
4133
4134	sc = if_getsoftc(ifp);
4135	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
4136	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
4137		return (EBUSY);
4138
4139	error = re_encap(sc, &m);
4140	if (error == 0)
4141		re_start_tx(sc);
4142	return (error);
4143}
4144
4145static int
4146re_debugnet_poll(struct ifnet *ifp, int count)
4147{
4148	struct rl_softc *sc;
4149	int error;
4150
4151	sc = if_getsoftc(ifp);
4152	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 ||
4153	    (sc->rl_flags & RL_FLAG_LINK) == 0)
4154		return (EBUSY);
4155
4156	re_txeof(sc);
4157	error = re_rxeof(sc, NULL);
4158	if (error != 0 && error != EAGAIN)
4159		return (error);
4160	return (0);
4161}
4162#endif /* DEBUGNET */
4163