1/*-
2 * Copyright (c) 1997, 1998-2003
3 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by Bill Paul.
16 * 4. Neither the name of the author nor the names of any co-contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33#include <sys/cdefs.h>
34__FBSDID("$FreeBSD$");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38 *
39 * Written by Bill Paul <wpaul@windriver.com>
40 * Senior Networking Software Engineer
41 * Wind River Systems
42 */
43
44/*
45 * This driver is designed to support RealTek's next generation of
46 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49 *
50 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51 * with the older 8139 family, however it also supports a special
52 * C+ mode of operation that provides several new performance enhancing
53 * features. These include:
54 *
55 *	o Descriptor based DMA mechanism. Each descriptor represents
56 *	  a single packet fragment. Data buffers may be aligned on
57 *	  any byte boundary.
58 *
59 *	o 64-bit DMA
60 *
61 *	o TCP/IP checksum offload for both RX and TX
62 *
63 *	o High and normal priority transmit DMA rings
64 *
65 *	o VLAN tag insertion and extraction
66 *
67 *	o TCP large send (segmentation offload)
68 *
69 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70 * programming API is fairly straightforward. The RX filtering, EEPROM
71 * access and PHY access is the same as it is on the older 8139 series
72 * chips.
73 *
74 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75 * same programming API and feature set as the 8139C+ with the following
76 * differences and additions:
77 *
78 *	o 1000Mbps mode
79 *
80 *	o Jumbo frames
81 *
82 *	o GMII and TBI ports/registers for interfacing with copper
83 *	  or fiber PHYs
84 *
85 *	o RX and TX DMA rings can have up to 1024 descriptors
86 *	  (the 8139C+ allows a maximum of 64)
87 *
88 *	o Slight differences in register layout from the 8139C+
89 *
90 * The TX start and timer interrupt registers are at different locations
91 * on the 8169 than they are on the 8139C+. Also, the status word in the
92 * RX descriptor has a slightly different bit layout. The 8169 does not
93 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94 * copper gigE PHY.
95 *
96 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97 * (the 'S' stands for 'single-chip'). These devices have the same
98 * programming API as the older 8169, but also have some vendor-specific
99 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101 *
102 * This driver takes advantage of the RX and TX checksum offload and
103 * VLAN tag insertion/extraction features. It also implements TX
104 * interrupt moderation using the timer interrupt registers, which
105 * significantly reduces TX interrupt load. There is also support
106 * for jumbo frames, however the 8169/8169S/8110S can not transmit
107 * jumbo frames larger than 7440, so the max MTU possible with this
108 * driver is 7422 bytes.
109 */
110
111#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/sysctl.h>
127#include <sys/taskqueue.h>
128
129#include <net/if.h>
130#include <net/if_arp.h>
131#include <net/ethernet.h>
132#include <net/if_dl.h>
133#include <net/if_media.h>
134#include <net/if_types.h>
135#include <net/if_vlan_var.h>
136
137#include <net/bpf.h>
138
139#include <machine/bus.h>
140#include <machine/resource.h>
141#include <sys/bus.h>
142#include <sys/rman.h>
143
144#include <dev/mii/mii.h>
145#include <dev/mii/miivar.h>
146
147#include <dev/pci/pcireg.h>
148#include <dev/pci/pcivar.h>
149
150#include <pci/if_rlreg.h>
151
152MODULE_DEPEND(re, pci, 1, 1, 1);
153MODULE_DEPEND(re, ether, 1, 1, 1);
154MODULE_DEPEND(re, miibus, 1, 1, 1);
155
156/* "device miibus" required.  See GENERIC if you get errors here. */
157#include "miibus_if.h"
158
159/* Tunables. */
160static int intr_filter = 0;
161TUNABLE_INT("hw.re.intr_filter", &intr_filter);
162static int msi_disable = 0;
163TUNABLE_INT("hw.re.msi_disable", &msi_disable);
164static int msix_disable = 0;
165TUNABLE_INT("hw.re.msix_disable", &msix_disable);
166static int prefer_iomap = 0;
167TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
168
169#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
170
171/*
172 * Various supported device vendors/types and their names.
173 */
174static const struct rl_type re_devs[] = {
175	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
178	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
179	{ RT_VENDORID, RT_DEVICEID_8139, 0,
180	    "RealTek 8139C+ 10/100BaseTX" },
181	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
182	    "RealTek 810xE PCIe 10/100baseTX" },
183	{ RT_VENDORID, RT_DEVICEID_8168, 0,
184	    "RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
185	{ RT_VENDORID, RT_DEVICEID_8169, 0,
186	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
187	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
188	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
189	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
190	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
192	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193	{ USR_VENDORID, USR_DEVICEID_997902, 0,
194	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
195};
196
197static const struct rl_hwrev re_hwrevs[] = {
198	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
199	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
200	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
201	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
202	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
203	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
204	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
205	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
206	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
207	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
208	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
209	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
210	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
211	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
213	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
214	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
215	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
216	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
217	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
218	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
219	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
220	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
221	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
222	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
223	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
224	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
225	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
226	{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
227	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
228	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
229	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
230	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
231	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
232	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
233	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
234	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
235	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
236	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
237	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
238	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
239	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
240	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
241	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
242	{ 0, 0, NULL, 0 }
243};
244
245static int re_probe		(device_t);
246static int re_attach		(device_t);
247static int re_detach		(device_t);
248
249static int re_encap		(struct rl_softc *, struct mbuf **);
250
251static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
252static int re_allocmem		(device_t, struct rl_softc *);
253static __inline void re_discard_rxbuf
254				(struct rl_softc *, int);
255static int re_newbuf		(struct rl_softc *, int);
256static int re_jumbo_newbuf	(struct rl_softc *, int);
257static int re_rx_list_init	(struct rl_softc *);
258static int re_jrx_list_init	(struct rl_softc *);
259static int re_tx_list_init	(struct rl_softc *);
260#ifdef RE_FIXUP_RX
261static __inline void re_fixup_rx
262				(struct mbuf *);
263#endif
264static int re_rxeof		(struct rl_softc *, int *);
265static void re_txeof		(struct rl_softc *);
266#ifdef DEVICE_POLLING
267static int re_poll		(struct ifnet *, enum poll_cmd, int);
268static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
269#endif
270static int re_intr		(void *);
271static void re_intr_msi		(void *);
272static void re_tick		(void *);
273static void re_int_task		(void *, int);
274static void re_start		(struct ifnet *);
275static void re_start_locked	(struct ifnet *);
276static int re_ioctl		(struct ifnet *, u_long, caddr_t);
277static void re_init		(void *);
278static void re_init_locked	(struct rl_softc *);
279static void re_stop		(struct rl_softc *);
280static void re_watchdog		(struct rl_softc *);
281static int re_suspend		(device_t);
282static int re_resume		(device_t);
283static int re_shutdown		(device_t);
284static int re_ifmedia_upd	(struct ifnet *);
285static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
286
287static void re_eeprom_putbyte	(struct rl_softc *, int);
288static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
289static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
290static int re_gmii_readreg	(device_t, int, int);
291static int re_gmii_writereg	(device_t, int, int, int);
292
293static int re_miibus_readreg	(device_t, int, int);
294static int re_miibus_writereg	(device_t, int, int, int);
295static void re_miibus_statchg	(device_t);
296
297static void re_set_jumbo	(struct rl_softc *, int);
298static void re_set_rxmode		(struct rl_softc *);
299static void re_reset		(struct rl_softc *);
300static void re_setwol		(struct rl_softc *);
301static void re_clrwol		(struct rl_softc *);
302static void re_set_linkspeed	(struct rl_softc *);
303
304#ifdef DEV_NETMAP	/* see ixgbe.c for details */
305#include <dev/netmap/if_re_netmap.h>
306#endif /* !DEV_NETMAP */
307
308#ifdef RE_DIAG
309static int re_diag		(struct rl_softc *);
310#endif
311
312static void re_add_sysctls	(struct rl_softc *);
313static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
314static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
315static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
316
317static device_method_t re_methods[] = {
318	/* Device interface */
319	DEVMETHOD(device_probe,		re_probe),
320	DEVMETHOD(device_attach,	re_attach),
321	DEVMETHOD(device_detach,	re_detach),
322	DEVMETHOD(device_suspend,	re_suspend),
323	DEVMETHOD(device_resume,	re_resume),
324	DEVMETHOD(device_shutdown,	re_shutdown),
325
326	/* MII interface */
327	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
328	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
329	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
330
331	DEVMETHOD_END
332};
333
334static driver_t re_driver = {
335	"re",
336	re_methods,
337	sizeof(struct rl_softc)
338};
339
340static devclass_t re_devclass;
341
342DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
343DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
344
345#define EE_SET(x)					\
346	CSR_WRITE_1(sc, RL_EECMD,			\
347		CSR_READ_1(sc, RL_EECMD) | x)
348
349#define EE_CLR(x)					\
350	CSR_WRITE_1(sc, RL_EECMD,			\
351		CSR_READ_1(sc, RL_EECMD) & ~x)
352
353/*
354 * Send a read command and address to the EEPROM, check for ACK.
355 */
356static void
357re_eeprom_putbyte(struct rl_softc *sc, int addr)
358{
359	int			d, i;
360
361	d = addr | (RL_9346_READ << sc->rl_eewidth);
362
363	/*
364	 * Feed in each bit and strobe the clock.
365	 */
366
367	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
368		if (d & i) {
369			EE_SET(RL_EE_DATAIN);
370		} else {
371			EE_CLR(RL_EE_DATAIN);
372		}
373		DELAY(100);
374		EE_SET(RL_EE_CLK);
375		DELAY(150);
376		EE_CLR(RL_EE_CLK);
377		DELAY(100);
378	}
379}
380
381/*
382 * Read a word of data stored in the EEPROM at address 'addr.'
383 */
384static void
385re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
386{
387	int			i;
388	u_int16_t		word = 0;
389
390	/*
391	 * Send address of word we want to read.
392	 */
393	re_eeprom_putbyte(sc, addr);
394
395	/*
396	 * Start reading bits from EEPROM.
397	 */
398	for (i = 0x8000; i; i >>= 1) {
399		EE_SET(RL_EE_CLK);
400		DELAY(100);
401		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
402			word |= i;
403		EE_CLR(RL_EE_CLK);
404		DELAY(100);
405	}
406
407	*dest = word;
408}
409
410/*
411 * Read a sequence of words from the EEPROM.
412 */
413static void
414re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
415{
416	int			i;
417	u_int16_t		word = 0, *ptr;
418
419	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
420
421        DELAY(100);
422
423	for (i = 0; i < cnt; i++) {
424		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
425		re_eeprom_getword(sc, off + i, &word);
426		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
427		ptr = (u_int16_t *)(dest + (i * 2));
428                *ptr = word;
429	}
430
431	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
432}
433
434static int
435re_gmii_readreg(device_t dev, int phy, int reg)
436{
437	struct rl_softc		*sc;
438	u_int32_t		rval;
439	int			i;
440
441	sc = device_get_softc(dev);
442
443	/* Let the rgephy driver read the GMEDIASTAT register */
444
445	if (reg == RL_GMEDIASTAT) {
446		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
447		return (rval);
448	}
449
450	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
451
452	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
453		rval = CSR_READ_4(sc, RL_PHYAR);
454		if (rval & RL_PHYAR_BUSY)
455			break;
456		DELAY(25);
457	}
458
459	if (i == RL_PHY_TIMEOUT) {
460		device_printf(sc->rl_dev, "PHY read failed\n");
461		return (0);
462	}
463
464	/*
465	 * Controller requires a 20us delay to process next MDIO request.
466	 */
467	DELAY(20);
468
469	return (rval & RL_PHYAR_PHYDATA);
470}
471
472static int
473re_gmii_writereg(device_t dev, int phy, int reg, int data)
474{
475	struct rl_softc		*sc;
476	u_int32_t		rval;
477	int			i;
478
479	sc = device_get_softc(dev);
480
481	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
482	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
483
484	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
485		rval = CSR_READ_4(sc, RL_PHYAR);
486		if (!(rval & RL_PHYAR_BUSY))
487			break;
488		DELAY(25);
489	}
490
491	if (i == RL_PHY_TIMEOUT) {
492		device_printf(sc->rl_dev, "PHY write failed\n");
493		return (0);
494	}
495
496	/*
497	 * Controller requires a 20us delay to process next MDIO request.
498	 */
499	DELAY(20);
500
501	return (0);
502}
503
504static int
505re_miibus_readreg(device_t dev, int phy, int reg)
506{
507	struct rl_softc		*sc;
508	u_int16_t		rval = 0;
509	u_int16_t		re8139_reg = 0;
510
511	sc = device_get_softc(dev);
512
513	if (sc->rl_type == RL_8169) {
514		rval = re_gmii_readreg(dev, phy, reg);
515		return (rval);
516	}
517
518	switch (reg) {
519	case MII_BMCR:
520		re8139_reg = RL_BMCR;
521		break;
522	case MII_BMSR:
523		re8139_reg = RL_BMSR;
524		break;
525	case MII_ANAR:
526		re8139_reg = RL_ANAR;
527		break;
528	case MII_ANER:
529		re8139_reg = RL_ANER;
530		break;
531	case MII_ANLPAR:
532		re8139_reg = RL_LPAR;
533		break;
534	case MII_PHYIDR1:
535	case MII_PHYIDR2:
536		return (0);
537	/*
538	 * Allow the rlphy driver to read the media status
539	 * register. If we have a link partner which does not
540	 * support NWAY, this is the register which will tell
541	 * us the results of parallel detection.
542	 */
543	case RL_MEDIASTAT:
544		rval = CSR_READ_1(sc, RL_MEDIASTAT);
545		return (rval);
546	default:
547		device_printf(sc->rl_dev, "bad phy register\n");
548		return (0);
549	}
550	rval = CSR_READ_2(sc, re8139_reg);
551	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
552		/* 8139C+ has different bit layout. */
553		rval &= ~(BMCR_LOOP | BMCR_ISO);
554	}
555	return (rval);
556}
557
558static int
559re_miibus_writereg(device_t dev, int phy, int reg, int data)
560{
561	struct rl_softc		*sc;
562	u_int16_t		re8139_reg = 0;
563	int			rval = 0;
564
565	sc = device_get_softc(dev);
566
567	if (sc->rl_type == RL_8169) {
568		rval = re_gmii_writereg(dev, phy, reg, data);
569		return (rval);
570	}
571
572	switch (reg) {
573	case MII_BMCR:
574		re8139_reg = RL_BMCR;
575		if (sc->rl_type == RL_8139CPLUS) {
576			/* 8139C+ has different bit layout. */
577			data &= ~(BMCR_LOOP | BMCR_ISO);
578		}
579		break;
580	case MII_BMSR:
581		re8139_reg = RL_BMSR;
582		break;
583	case MII_ANAR:
584		re8139_reg = RL_ANAR;
585		break;
586	case MII_ANER:
587		re8139_reg = RL_ANER;
588		break;
589	case MII_ANLPAR:
590		re8139_reg = RL_LPAR;
591		break;
592	case MII_PHYIDR1:
593	case MII_PHYIDR2:
594		return (0);
595		break;
596	default:
597		device_printf(sc->rl_dev, "bad phy register\n");
598		return (0);
599	}
600	CSR_WRITE_2(sc, re8139_reg, data);
601	return (0);
602}
603
604static void
605re_miibus_statchg(device_t dev)
606{
607	struct rl_softc		*sc;
608	struct ifnet		*ifp;
609	struct mii_data		*mii;
610
611	sc = device_get_softc(dev);
612	mii = device_get_softc(sc->rl_miibus);
613	ifp = sc->rl_ifp;
614	if (mii == NULL || ifp == NULL ||
615	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
616		return;
617
618	sc->rl_flags &= ~RL_FLAG_LINK;
619	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
620	    (IFM_ACTIVE | IFM_AVALID)) {
621		switch (IFM_SUBTYPE(mii->mii_media_active)) {
622		case IFM_10_T:
623		case IFM_100_TX:
624			sc->rl_flags |= RL_FLAG_LINK;
625			break;
626		case IFM_1000_T:
627			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
628				break;
629			sc->rl_flags |= RL_FLAG_LINK;
630			break;
631		default:
632			break;
633		}
634	}
635	/*
636	 * RealTek controllers does not provide any interface to
637	 * Tx/Rx MACs for resolved speed, duplex and flow-control
638	 * parameters.
639	 */
640}
641
642/*
643 * Set the RX configuration and 64-bit multicast hash filter.
644 */
645static void
646re_set_rxmode(struct rl_softc *sc)
647{
648	struct ifnet		*ifp;
649	struct ifmultiaddr	*ifma;
650	uint32_t		hashes[2] = { 0, 0 };
651	uint32_t		h, rxfilt;
652
653	RL_LOCK_ASSERT(sc);
654
655	ifp = sc->rl_ifp;
656
657	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
658	if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
659		rxfilt |= RL_RXCFG_EARLYOFF;
660	else if ((sc->rl_flags & RL_FLAG_EARLYOFFV2) != 0)
661		rxfilt |= RL_RXCFG_EARLYOFFV2;
662
663	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
664		if (ifp->if_flags & IFF_PROMISC)
665			rxfilt |= RL_RXCFG_RX_ALLPHYS;
666		/*
667		 * Unlike other hardwares, we have to explicitly set
668		 * RL_RXCFG_RX_MULTI to receive multicast frames in
669		 * promiscuous mode.
670		 */
671		rxfilt |= RL_RXCFG_RX_MULTI;
672		hashes[0] = hashes[1] = 0xffffffff;
673		goto done;
674	}
675
676	if_maddr_rlock(ifp);
677	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
678		if (ifma->ifma_addr->sa_family != AF_LINK)
679			continue;
680		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
681		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
682		if (h < 32)
683			hashes[0] |= (1 << h);
684		else
685			hashes[1] |= (1 << (h - 32));
686	}
687	if_maddr_runlock(ifp);
688
689	if (hashes[0] != 0 || hashes[1] != 0) {
690		/*
691		 * For some unfathomable reason, RealTek decided to
692		 * reverse the order of the multicast hash registers
693		 * in the PCI Express parts.  This means we have to
694		 * write the hash pattern in reverse order for those
695		 * devices.
696		 */
697		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
698			h = bswap32(hashes[0]);
699			hashes[0] = bswap32(hashes[1]);
700			hashes[1] = h;
701		}
702		rxfilt |= RL_RXCFG_RX_MULTI;
703	}
704
705done:
706	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
707	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
708	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
709}
710
711static void
712re_reset(struct rl_softc *sc)
713{
714	int			i;
715
716	RL_LOCK_ASSERT(sc);
717
718	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
719
720	for (i = 0; i < RL_TIMEOUT; i++) {
721		DELAY(10);
722		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
723			break;
724	}
725	if (i == RL_TIMEOUT)
726		device_printf(sc->rl_dev, "reset never completed!\n");
727
728	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
729		CSR_WRITE_1(sc, 0x82, 1);
730	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
731		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
732}
733
734#ifdef RE_DIAG
735
736/*
737 * The following routine is designed to test for a defect on some
738 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
739 * lines connected to the bus, however for a 32-bit only card, they
740 * should be pulled high. The result of this defect is that the
741 * NIC will not work right if you plug it into a 64-bit slot: DMA
742 * operations will be done with 64-bit transfers, which will fail
743 * because the 64-bit data lines aren't connected.
744 *
745 * There's no way to work around this (short of talking a soldering
746 * iron to the board), however we can detect it. The method we use
747 * here is to put the NIC into digital loopback mode, set the receiver
748 * to promiscuous mode, and then try to send a frame. We then compare
749 * the frame data we sent to what was received. If the data matches,
750 * then the NIC is working correctly, otherwise we know the user has
751 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
752 * slot. In the latter case, there's no way the NIC can work correctly,
753 * so we print out a message on the console and abort the device attach.
754 */
755
756static int
757re_diag(struct rl_softc *sc)
758{
759	struct ifnet		*ifp = sc->rl_ifp;
760	struct mbuf		*m0;
761	struct ether_header	*eh;
762	struct rl_desc		*cur_rx;
763	u_int16_t		status;
764	u_int32_t		rxstat;
765	int			total_len, i, error = 0, phyaddr;
766	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
767	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
768
769	/* Allocate a single mbuf */
770	MGETHDR(m0, M_NOWAIT, MT_DATA);
771	if (m0 == NULL)
772		return (ENOBUFS);
773
774	RL_LOCK(sc);
775
776	/*
777	 * Initialize the NIC in test mode. This sets the chip up
778	 * so that it can send and receive frames, but performs the
779	 * following special functions:
780	 * - Puts receiver in promiscuous mode
781	 * - Enables digital loopback mode
782	 * - Leaves interrupts turned off
783	 */
784
785	ifp->if_flags |= IFF_PROMISC;
786	sc->rl_testmode = 1;
787	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
788	re_init_locked(sc);
789	sc->rl_flags |= RL_FLAG_LINK;
790	if (sc->rl_type == RL_8169)
791		phyaddr = 1;
792	else
793		phyaddr = 0;
794
795	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
796	for (i = 0; i < RL_TIMEOUT; i++) {
797		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
798		if (!(status & BMCR_RESET))
799			break;
800	}
801
802	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
803	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
804
805	DELAY(100000);
806
807	/* Put some data in the mbuf */
808
809	eh = mtod(m0, struct ether_header *);
810	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
811	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
812	eh->ether_type = htons(ETHERTYPE_IP);
813	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
814
815	/*
816	 * Queue the packet, start transmission.
817	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
818	 */
819
820	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
821	RL_UNLOCK(sc);
822	/* XXX: re_diag must not be called when in ALTQ mode */
823	IF_HANDOFF(&ifp->if_snd, m0, ifp);
824	RL_LOCK(sc);
825	m0 = NULL;
826
827	/* Wait for it to propagate through the chip */
828
829	DELAY(100000);
830	for (i = 0; i < RL_TIMEOUT; i++) {
831		status = CSR_READ_2(sc, RL_ISR);
832		CSR_WRITE_2(sc, RL_ISR, status);
833		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
834		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
835			break;
836		DELAY(10);
837	}
838
839	if (i == RL_TIMEOUT) {
840		device_printf(sc->rl_dev,
841		    "diagnostic failed, failed to receive packet in"
842		    " loopback mode\n");
843		error = EIO;
844		goto done;
845	}
846
847	/*
848	 * The packet should have been dumped into the first
849	 * entry in the RX DMA ring. Grab it from there.
850	 */
851
852	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
853	    sc->rl_ldata.rl_rx_list_map,
854	    BUS_DMASYNC_POSTREAD);
855	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
856	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
857	    BUS_DMASYNC_POSTREAD);
858	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
859	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
860
861	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
862	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
863	eh = mtod(m0, struct ether_header *);
864
865	cur_rx = &sc->rl_ldata.rl_rx_list[0];
866	total_len = RL_RXBYTES(cur_rx);
867	rxstat = le32toh(cur_rx->rl_cmdstat);
868
869	if (total_len != ETHER_MIN_LEN) {
870		device_printf(sc->rl_dev,
871		    "diagnostic failed, received short packet\n");
872		error = EIO;
873		goto done;
874	}
875
876	/* Test that the received packet data matches what we sent. */
877
878	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
879	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
880	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
881		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
882		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
883		    dst, ":", src, ":", ETHERTYPE_IP);
884		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
885		    eh->ether_dhost, ":", eh->ether_shost, ":",
886		    ntohs(eh->ether_type));
887		device_printf(sc->rl_dev, "You may have a defective 32-bit "
888		    "NIC plugged into a 64-bit PCI slot.\n");
889		device_printf(sc->rl_dev, "Please re-install the NIC in a "
890		    "32-bit slot for proper operation.\n");
891		device_printf(sc->rl_dev, "Read the re(4) man page for more "
892		    "details.\n");
893		error = EIO;
894	}
895
896done:
897	/* Turn interface off, release resources */
898
899	sc->rl_testmode = 0;
900	sc->rl_flags &= ~RL_FLAG_LINK;
901	ifp->if_flags &= ~IFF_PROMISC;
902	re_stop(sc);
903	if (m0 != NULL)
904		m_freem(m0);
905
906	RL_UNLOCK(sc);
907
908	return (error);
909}
910
911#endif
912
913/*
914 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
915 * IDs against our list and return a device name if we find a match.
916 */
917static int
918re_probe(device_t dev)
919{
920	const struct rl_type	*t;
921	uint16_t		devid, vendor;
922	uint16_t		revid, sdevid;
923	int			i;
924
925	vendor = pci_get_vendor(dev);
926	devid = pci_get_device(dev);
927	revid = pci_get_revid(dev);
928	sdevid = pci_get_subdevice(dev);
929
930	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
931		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
932			/*
933			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
934			 * Rev. 2 is supported by sk(4).
935			 */
936			return (ENXIO);
937		}
938	}
939
940	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
941		if (revid != 0x20) {
942			/* 8139, let rl(4) take care of this device. */
943			return (ENXIO);
944		}
945	}
946
947	t = re_devs;
948	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
949		if (vendor == t->rl_vid && devid == t->rl_did) {
950			device_set_desc(dev, t->rl_name);
951			return (BUS_PROBE_DEFAULT);
952		}
953	}
954
955	return (ENXIO);
956}
957
958/*
959 * Map a single buffer address.
960 */
961
962static void
963re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
964{
965	bus_addr_t		*addr;
966
967	if (error)
968		return;
969
970	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
971	addr = arg;
972	*addr = segs->ds_addr;
973}
974
975static int
976re_allocmem(device_t dev, struct rl_softc *sc)
977{
978	bus_addr_t		lowaddr;
979	bus_size_t		rx_list_size, tx_list_size;
980	int			error;
981	int			i;
982
983	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
984	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
985
986	/*
987	 * Allocate the parent bus DMA tag appropriate for PCI.
988	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
989	 * register should be set. However some RealTek chips are known
990	 * to be buggy on DAC handling, therefore disable DAC by limiting
991	 * DMA address space to 32bit. PCIe variants of RealTek chips
992	 * may not have the limitation.
993	 */
994	lowaddr = BUS_SPACE_MAXADDR;
995	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
996		lowaddr = BUS_SPACE_MAXADDR_32BIT;
997	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
998	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
999	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1000	    NULL, NULL, &sc->rl_parent_tag);
1001	if (error) {
1002		device_printf(dev, "could not allocate parent DMA tag\n");
1003		return (error);
1004	}
1005
1006	/*
1007	 * Allocate map for TX mbufs.
1008	 */
1009	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1010	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1011	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1012	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1013	if (error) {
1014		device_printf(dev, "could not allocate TX DMA tag\n");
1015		return (error);
1016	}
1017
1018	/*
1019	 * Allocate map for RX mbufs.
1020	 */
1021
1022	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1023		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1024		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1025		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1026		    &sc->rl_ldata.rl_jrx_mtag);
1027		if (error) {
1028			device_printf(dev,
1029			    "could not allocate jumbo RX DMA tag\n");
1030			return (error);
1031		}
1032	}
1033	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1034	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1035	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1036	if (error) {
1037		device_printf(dev, "could not allocate RX DMA tag\n");
1038		return (error);
1039	}
1040
1041	/*
1042	 * Allocate map for TX descriptor list.
1043	 */
1044	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1045	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1046	    NULL, tx_list_size, 1, tx_list_size, 0,
1047	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1048	if (error) {
1049		device_printf(dev, "could not allocate TX DMA ring tag\n");
1050		return (error);
1051	}
1052
1053	/* Allocate DMA'able memory for the TX ring */
1054
1055	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1056	    (void **)&sc->rl_ldata.rl_tx_list,
1057	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1058	    &sc->rl_ldata.rl_tx_list_map);
1059	if (error) {
1060		device_printf(dev, "could not allocate TX DMA ring\n");
1061		return (error);
1062	}
1063
1064	/* Load the map for the TX ring. */
1065
1066	sc->rl_ldata.rl_tx_list_addr = 0;
1067	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1068	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1069	     tx_list_size, re_dma_map_addr,
1070	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1071	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1072		device_printf(dev, "could not load TX DMA ring\n");
1073		return (ENOMEM);
1074	}
1075
1076	/* Create DMA maps for TX buffers */
1077
1078	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1079		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1080		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1081		if (error) {
1082			device_printf(dev, "could not create DMA map for TX\n");
1083			return (error);
1084		}
1085	}
1086
1087	/*
1088	 * Allocate map for RX descriptor list.
1089	 */
1090	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1091	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1092	    NULL, rx_list_size, 1, rx_list_size, 0,
1093	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1094	if (error) {
1095		device_printf(dev, "could not create RX DMA ring tag\n");
1096		return (error);
1097	}
1098
1099	/* Allocate DMA'able memory for the RX ring */
1100
1101	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1102	    (void **)&sc->rl_ldata.rl_rx_list,
1103	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1104	    &sc->rl_ldata.rl_rx_list_map);
1105	if (error) {
1106		device_printf(dev, "could not allocate RX DMA ring\n");
1107		return (error);
1108	}
1109
1110	/* Load the map for the RX ring. */
1111
1112	sc->rl_ldata.rl_rx_list_addr = 0;
1113	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1114	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1115	     rx_list_size, re_dma_map_addr,
1116	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1117	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1118		device_printf(dev, "could not load RX DMA ring\n");
1119		return (ENOMEM);
1120	}
1121
1122	/* Create DMA maps for RX buffers */
1123
1124	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1125		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1126		    &sc->rl_ldata.rl_jrx_sparemap);
1127		if (error) {
1128			device_printf(dev,
1129			    "could not create spare DMA map for jumbo RX\n");
1130			return (error);
1131		}
1132		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1133			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1134			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1135			if (error) {
1136				device_printf(dev,
1137				    "could not create DMA map for jumbo RX\n");
1138				return (error);
1139			}
1140		}
1141	}
1142	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1143	    &sc->rl_ldata.rl_rx_sparemap);
1144	if (error) {
1145		device_printf(dev, "could not create spare DMA map for RX\n");
1146		return (error);
1147	}
1148	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1149		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1150		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1151		if (error) {
1152			device_printf(dev, "could not create DMA map for RX\n");
1153			return (error);
1154		}
1155	}
1156
1157	/* Create DMA map for statistics. */
1158	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1159	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1160	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1161	    &sc->rl_ldata.rl_stag);
1162	if (error) {
1163		device_printf(dev, "could not create statistics DMA tag\n");
1164		return (error);
1165	}
1166	/* Allocate DMA'able memory for statistics. */
1167	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1168	    (void **)&sc->rl_ldata.rl_stats,
1169	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1170	    &sc->rl_ldata.rl_smap);
1171	if (error) {
1172		device_printf(dev,
1173		    "could not allocate statistics DMA memory\n");
1174		return (error);
1175	}
1176	/* Load the map for statistics. */
1177	sc->rl_ldata.rl_stats_addr = 0;
1178	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1179	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1180	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1181	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1182		device_printf(dev, "could not load statistics DMA memory\n");
1183		return (ENOMEM);
1184	}
1185
1186	return (0);
1187}
1188
1189/*
1190 * Attach the interface. Allocate softc structures, do ifmedia
1191 * setup and ethernet/BPF attach.
1192 */
1193static int
1194re_attach(device_t dev)
1195{
1196	u_char			eaddr[ETHER_ADDR_LEN];
1197	u_int16_t		as[ETHER_ADDR_LEN / 2];
1198	struct rl_softc		*sc;
1199	struct ifnet		*ifp;
1200	const struct rl_hwrev	*hw_rev;
1201	u_int32_t		cap, ctl;
1202	int			hwrev;
1203	u_int16_t		devid, re_did = 0;
1204	int			error = 0, i, phy, rid;
1205	int			msic, msixc, reg;
1206	uint8_t			cfg;
1207
1208	sc = device_get_softc(dev);
1209	sc->rl_dev = dev;
1210
1211	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1212	    MTX_DEF);
1213	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1214
1215	/*
1216	 * Map control/status registers.
1217	 */
1218	pci_enable_busmaster(dev);
1219
1220	devid = pci_get_device(dev);
1221	/*
1222	 * Prefer memory space register mapping over IO space.
1223	 * Because RTL8169SC does not seem to work when memory mapping
1224	 * is used always activate io mapping.
1225	 */
1226	if (devid == RT_DEVICEID_8169SC)
1227		prefer_iomap = 1;
1228	if (prefer_iomap == 0) {
1229		sc->rl_res_id = PCIR_BAR(1);
1230		sc->rl_res_type = SYS_RES_MEMORY;
1231		/* RTL8168/8101E seems to use different BARs. */
1232		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1233			sc->rl_res_id = PCIR_BAR(2);
1234	} else {
1235		sc->rl_res_id = PCIR_BAR(0);
1236		sc->rl_res_type = SYS_RES_IOPORT;
1237	}
1238	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1239	    &sc->rl_res_id, RF_ACTIVE);
1240	if (sc->rl_res == NULL && prefer_iomap == 0) {
1241		sc->rl_res_id = PCIR_BAR(0);
1242		sc->rl_res_type = SYS_RES_IOPORT;
1243		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1244		    &sc->rl_res_id, RF_ACTIVE);
1245	}
1246	if (sc->rl_res == NULL) {
1247		device_printf(dev, "couldn't map ports/memory\n");
1248		error = ENXIO;
1249		goto fail;
1250	}
1251
1252	sc->rl_btag = rman_get_bustag(sc->rl_res);
1253	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1254
1255	msic = pci_msi_count(dev);
1256	msixc = pci_msix_count(dev);
1257	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1258		sc->rl_flags |= RL_FLAG_PCIE;
1259		sc->rl_expcap = reg;
1260	}
1261	if (bootverbose) {
1262		device_printf(dev, "MSI count : %d\n", msic);
1263		device_printf(dev, "MSI-X count : %d\n", msixc);
1264	}
1265	if (msix_disable > 0)
1266		msixc = 0;
1267	if (msi_disable > 0)
1268		msic = 0;
1269	/* Prefer MSI-X to MSI. */
1270	if (msixc > 0) {
1271		msixc = RL_MSI_MESSAGES;
1272		rid = PCIR_BAR(4);
1273		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1274		    &rid, RF_ACTIVE);
1275		if (sc->rl_res_pba == NULL) {
1276			device_printf(sc->rl_dev,
1277			    "could not allocate MSI-X PBA resource\n");
1278		}
1279		if (sc->rl_res_pba != NULL &&
1280		    pci_alloc_msix(dev, &msixc) == 0) {
1281			if (msixc == RL_MSI_MESSAGES) {
1282				device_printf(dev, "Using %d MSI-X message\n",
1283				    msixc);
1284				sc->rl_flags |= RL_FLAG_MSIX;
1285			} else
1286				pci_release_msi(dev);
1287		}
1288		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1289			if (sc->rl_res_pba != NULL)
1290				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1291				    sc->rl_res_pba);
1292			sc->rl_res_pba = NULL;
1293			msixc = 0;
1294		}
1295	}
1296	/* Prefer MSI to INTx. */
1297	if (msixc == 0 && msic > 0) {
1298		msic = RL_MSI_MESSAGES;
1299		if (pci_alloc_msi(dev, &msic) == 0) {
1300			if (msic == RL_MSI_MESSAGES) {
1301				device_printf(dev, "Using %d MSI message\n",
1302				    msic);
1303				sc->rl_flags |= RL_FLAG_MSI;
1304				/* Explicitly set MSI enable bit. */
1305				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1306				cfg = CSR_READ_1(sc, RL_CFG2);
1307				cfg |= RL_CFG2_MSI;
1308				CSR_WRITE_1(sc, RL_CFG2, cfg);
1309				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1310			} else
1311				pci_release_msi(dev);
1312		}
1313		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1314			msic = 0;
1315	}
1316
1317	/* Allocate interrupt */
1318	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1319		rid = 0;
1320		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1321		    RF_SHAREABLE | RF_ACTIVE);
1322		if (sc->rl_irq[0] == NULL) {
1323			device_printf(dev, "couldn't allocate IRQ resources\n");
1324			error = ENXIO;
1325			goto fail;
1326		}
1327	} else {
1328		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1329			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1330			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1331			if (sc->rl_irq[i] == NULL) {
1332				device_printf(dev,
1333				    "couldn't llocate IRQ resources for "
1334				    "message %d\n", rid);
1335				error = ENXIO;
1336				goto fail;
1337			}
1338		}
1339	}
1340
1341	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1342		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1343		cfg = CSR_READ_1(sc, RL_CFG2);
1344		if ((cfg & RL_CFG2_MSI) != 0) {
1345			device_printf(dev, "turning off MSI enable bit.\n");
1346			cfg &= ~RL_CFG2_MSI;
1347			CSR_WRITE_1(sc, RL_CFG2, cfg);
1348		}
1349		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1350	}
1351
1352	/* Disable ASPM L0S/L1. */
1353	if (sc->rl_expcap != 0) {
1354		cap = pci_read_config(dev, sc->rl_expcap +
1355		    PCIER_LINK_CAP, 2);
1356		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1357			ctl = pci_read_config(dev, sc->rl_expcap +
1358			    PCIER_LINK_CTL, 2);
1359			if ((ctl & 0x0003) != 0) {
1360				ctl &= ~0x0003;
1361				pci_write_config(dev, sc->rl_expcap +
1362				    PCIER_LINK_CTL, ctl, 2);
1363				device_printf(dev, "ASPM disabled\n");
1364			}
1365		} else
1366			device_printf(dev, "no ASPM capability\n");
1367	}
1368
1369	hw_rev = re_hwrevs;
1370	hwrev = CSR_READ_4(sc, RL_TXCFG);
1371	switch (hwrev & 0x70000000) {
1372	case 0x00000000:
1373	case 0x10000000:
1374		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1375		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1376		break;
1377	default:
1378		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1379		sc->rl_macrev = hwrev & 0x00700000;
1380		hwrev &= RL_TXCFG_HWREV;
1381		break;
1382	}
1383	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1384	while (hw_rev->rl_desc != NULL) {
1385		if (hw_rev->rl_rev == hwrev) {
1386			sc->rl_type = hw_rev->rl_type;
1387			sc->rl_hwrev = hw_rev;
1388			break;
1389		}
1390		hw_rev++;
1391	}
1392	if (hw_rev->rl_desc == NULL) {
1393		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1394		error = ENXIO;
1395		goto fail;
1396	}
1397
1398	switch (hw_rev->rl_rev) {
1399	case RL_HWREV_8139CPLUS:
1400		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1401		break;
1402	case RL_HWREV_8100E:
1403	case RL_HWREV_8101E:
1404		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1405		break;
1406	case RL_HWREV_8102E:
1407	case RL_HWREV_8102EL:
1408	case RL_HWREV_8102EL_SPIN1:
1409		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1410		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1411		    RL_FLAG_AUTOPAD;
1412		break;
1413	case RL_HWREV_8103E:
1414		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1415		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1416		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1417		break;
1418	case RL_HWREV_8401E:
1419	case RL_HWREV_8105E:
1420	case RL_HWREV_8105E_SPIN1:
1421	case RL_HWREV_8106E:
1422		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1423		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1424		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1425		break;
1426	case RL_HWREV_8402:
1427		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1428		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1429		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1430		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1431		break;
1432	case RL_HWREV_8168B_SPIN1:
1433	case RL_HWREV_8168B_SPIN2:
1434		sc->rl_flags |= RL_FLAG_WOLRXENB;
1435		/* FALLTHROUGH */
1436	case RL_HWREV_8168B_SPIN3:
1437		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1438		break;
1439	case RL_HWREV_8168C_SPIN2:
1440		sc->rl_flags |= RL_FLAG_MACSLEEP;
1441		/* FALLTHROUGH */
1442	case RL_HWREV_8168C:
1443		if (sc->rl_macrev == 0x00200000)
1444			sc->rl_flags |= RL_FLAG_MACSLEEP;
1445		/* FALLTHROUGH */
1446	case RL_HWREV_8168CP:
1447		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1448		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1449		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1450		break;
1451	case RL_HWREV_8168D:
1452		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1453		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1454		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1455		    RL_FLAG_WOL_MANLINK;
1456		break;
1457	case RL_HWREV_8168DP:
1458		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1459		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1460		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1461		break;
1462	case RL_HWREV_8168E:
1463		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1464		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1465		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1466		    RL_FLAG_WOL_MANLINK;
1467		break;
1468	case RL_HWREV_8168E_VL:
1469	case RL_HWREV_8168F:
1470		sc->rl_flags |= RL_FLAG_EARLYOFF;
1471		/* FALLTHROUGH */
1472	case RL_HWREV_8411:
1473		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1474		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1475		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1476		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1477		break;
1478	case RL_HWREV_8168EP:
1479	case RL_HWREV_8168G:
1480	case RL_HWREV_8411B:
1481		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1482		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1483		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1484		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
1485		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1486		break;
1487	case RL_HWREV_8168GU:
1488		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1489			/* RTL8106EUS */
1490			sc->rl_flags |= RL_FLAG_FASTETHER;
1491		} else
1492			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1493
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_CMDSTOP_WAIT_TXQ |
1497		    RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
1498		break;
1499	case RL_HWREV_8169_8110SB:
1500	case RL_HWREV_8169_8110SBL:
1501	case RL_HWREV_8169_8110SC:
1502	case RL_HWREV_8169_8110SCE:
1503		sc->rl_flags |= RL_FLAG_PHYWAKE;
1504		/* FALLTHROUGH */
1505	case RL_HWREV_8169:
1506	case RL_HWREV_8169S:
1507	case RL_HWREV_8110S:
1508		sc->rl_flags |= RL_FLAG_MACRESET;
1509		break;
1510	default:
1511		break;
1512	}
1513
1514	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1515		sc->rl_cfg0 = RL_8139_CFG0;
1516		sc->rl_cfg1 = RL_8139_CFG1;
1517		sc->rl_cfg2 = 0;
1518		sc->rl_cfg3 = RL_8139_CFG3;
1519		sc->rl_cfg4 = RL_8139_CFG4;
1520		sc->rl_cfg5 = RL_8139_CFG5;
1521	} else {
1522		sc->rl_cfg0 = RL_CFG0;
1523		sc->rl_cfg1 = RL_CFG1;
1524		sc->rl_cfg2 = RL_CFG2;
1525		sc->rl_cfg3 = RL_CFG3;
1526		sc->rl_cfg4 = RL_CFG4;
1527		sc->rl_cfg5 = RL_CFG5;
1528	}
1529
1530	/* Reset the adapter. */
1531	RL_LOCK(sc);
1532	re_reset(sc);
1533	RL_UNLOCK(sc);
1534
1535	/* Enable PME. */
1536	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1537	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1538	cfg |= RL_CFG1_PME;
1539	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1540	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1541	cfg &= RL_CFG5_PME_STS;
1542	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1543	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1544
1545	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1546		/*
1547		 * XXX Should have a better way to extract station
1548		 * address from EEPROM.
1549		 */
1550		for (i = 0; i < ETHER_ADDR_LEN; i++)
1551			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1552	} else {
1553		sc->rl_eewidth = RL_9356_ADDR_LEN;
1554		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1555		if (re_did != 0x8129)
1556			sc->rl_eewidth = RL_9346_ADDR_LEN;
1557
1558		/*
1559		 * Get station address from the EEPROM.
1560		 */
1561		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1562		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1563			as[i] = le16toh(as[i]);
1564		bcopy(as, eaddr, ETHER_ADDR_LEN);
1565	}
1566
1567	if (sc->rl_type == RL_8169) {
1568		/* Set RX length mask and number of descriptors. */
1569		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1570		sc->rl_txstart = RL_GTXSTART;
1571		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1572		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1573	} else {
1574		/* Set RX length mask and number of descriptors. */
1575		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1576		sc->rl_txstart = RL_TXSTART;
1577		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1578		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1579	}
1580
1581	error = re_allocmem(dev, sc);
1582	if (error)
1583		goto fail;
1584	re_add_sysctls(sc);
1585
1586	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1587	if (ifp == NULL) {
1588		device_printf(dev, "can not if_alloc()\n");
1589		error = ENOSPC;
1590		goto fail;
1591	}
1592
1593	/* Take controller out of deep sleep mode. */
1594	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1595		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1596			CSR_WRITE_1(sc, RL_GPIO,
1597			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1598		else
1599			CSR_WRITE_1(sc, RL_GPIO,
1600			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1601	}
1602
1603	/* Take PHY out of power down mode. */
1604	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1605		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1606		if (hw_rev->rl_rev == RL_HWREV_8401E)
1607			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1608	}
1609	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1610		re_gmii_writereg(dev, 1, 0x1f, 0);
1611		re_gmii_writereg(dev, 1, 0x0e, 0);
1612	}
1613
1614	ifp->if_softc = sc;
1615	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1616	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1617	ifp->if_ioctl = re_ioctl;
1618	ifp->if_start = re_start;
1619	/*
1620	 * RTL8168/8111C generates wrong IP checksummed frame if the
1621	 * packet has IP options so disable TX checksum offloading.
1622	 */
1623	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1624	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1625	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
1626		ifp->if_hwassist = 0;
1627		ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
1628	} else {
1629		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1630		ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1631	}
1632	ifp->if_hwassist |= CSUM_TSO;
1633	ifp->if_capenable = ifp->if_capabilities;
1634	ifp->if_init = re_init;
1635	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1636	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1637	IFQ_SET_READY(&ifp->if_snd);
1638
1639	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1640
1641#define	RE_PHYAD_INTERNAL	 0
1642
1643	/* Do MII setup. */
1644	phy = RE_PHYAD_INTERNAL;
1645	if (sc->rl_type == RL_8169)
1646		phy = 1;
1647	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1648	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1649	if (error != 0) {
1650		device_printf(dev, "attaching PHYs failed\n");
1651		goto fail;
1652	}
1653
1654	/*
1655	 * Call MI attach routine.
1656	 */
1657	ether_ifattach(ifp, eaddr);
1658
1659	/* VLAN capability setup */
1660	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1661	if (ifp->if_capabilities & IFCAP_HWCSUM)
1662		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1663	/* Enable WOL if PM is supported. */
1664	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1665		ifp->if_capabilities |= IFCAP_WOL;
1666	ifp->if_capenable = ifp->if_capabilities;
1667	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1668	/*
1669	 * Don't enable TSO by default.  It is known to generate
1670	 * corrupted TCP segments(bad TCP options) under certain
1671	 * circumtances.
1672	 */
1673	ifp->if_hwassist &= ~CSUM_TSO;
1674	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1675#ifdef DEVICE_POLLING
1676	ifp->if_capabilities |= IFCAP_POLLING;
1677#endif
1678	/*
1679	 * Tell the upper layer(s) we support long frames.
1680	 * Must appear after the call to ether_ifattach() because
1681	 * ether_ifattach() sets ifi_hdrlen to the default value.
1682	 */
1683	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1684
1685#ifdef DEV_NETMAP
1686	re_netmap_attach(sc);
1687#endif /* DEV_NETMAP */
1688#ifdef RE_DIAG
1689	/*
1690	 * Perform hardware diagnostic on the original RTL8169.
1691	 * Some 32-bit cards were incorrectly wired and would
1692	 * malfunction if plugged into a 64-bit slot.
1693	 */
1694
1695	if (hwrev == RL_HWREV_8169) {
1696		error = re_diag(sc);
1697		if (error) {
1698			device_printf(dev,
1699		    	"attach aborted due to hardware diag failure\n");
1700			ether_ifdetach(ifp);
1701			goto fail;
1702		}
1703	}
1704#endif
1705
1706#ifdef RE_TX_MODERATION
1707	intr_filter = 1;
1708#endif
1709	/* Hook interrupt last to avoid having to lock softc */
1710	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1711	    intr_filter == 0) {
1712		error = bus_setup_intr(dev, sc->rl_irq[0],
1713		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1714		    &sc->rl_intrhand[0]);
1715	} else {
1716		error = bus_setup_intr(dev, sc->rl_irq[0],
1717		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1718		    &sc->rl_intrhand[0]);
1719	}
1720	if (error) {
1721		device_printf(dev, "couldn't set up irq\n");
1722		ether_ifdetach(ifp);
1723	}
1724
1725fail:
1726
1727	if (error)
1728		re_detach(dev);
1729
1730	return (error);
1731}
1732
1733/*
1734 * Shutdown hardware and free up resources. This can be called any
1735 * time after the mutex has been initialized. It is called in both
1736 * the error case in attach and the normal detach case so it needs
1737 * to be careful about only freeing resources that have actually been
1738 * allocated.
1739 */
1740static int
1741re_detach(device_t dev)
1742{
1743	struct rl_softc		*sc;
1744	struct ifnet		*ifp;
1745	int			i, rid;
1746
1747	sc = device_get_softc(dev);
1748	ifp = sc->rl_ifp;
1749	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1750
1751	/* These should only be active if attach succeeded */
1752	if (device_is_attached(dev)) {
1753#ifdef DEVICE_POLLING
1754		if (ifp->if_capenable & IFCAP_POLLING)
1755			ether_poll_deregister(ifp);
1756#endif
1757		RL_LOCK(sc);
1758#if 0
1759		sc->suspended = 1;
1760#endif
1761		re_stop(sc);
1762		RL_UNLOCK(sc);
1763		callout_drain(&sc->rl_stat_callout);
1764		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1765		/*
1766		 * Force off the IFF_UP flag here, in case someone
1767		 * still had a BPF descriptor attached to this
1768		 * interface. If they do, ether_ifdetach() will cause
1769		 * the BPF code to try and clear the promisc mode
1770		 * flag, which will bubble down to re_ioctl(),
1771		 * which will try to call re_init() again. This will
1772		 * turn the NIC back on and restart the MII ticker,
1773		 * which will panic the system when the kernel tries
1774		 * to invoke the re_tick() function that isn't there
1775		 * anymore.
1776		 */
1777		ifp->if_flags &= ~IFF_UP;
1778		ether_ifdetach(ifp);
1779	}
1780	if (sc->rl_miibus)
1781		device_delete_child(dev, sc->rl_miibus);
1782	bus_generic_detach(dev);
1783
1784	/*
1785	 * The rest is resource deallocation, so we should already be
1786	 * stopped here.
1787	 */
1788
1789	if (sc->rl_intrhand[0] != NULL) {
1790		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1791		sc->rl_intrhand[0] = NULL;
1792	}
1793	if (ifp != NULL) {
1794#ifdef DEV_NETMAP
1795		netmap_detach(ifp);
1796#endif /* DEV_NETMAP */
1797		if_free(ifp);
1798	}
1799	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1800		rid = 0;
1801	else
1802		rid = 1;
1803	if (sc->rl_irq[0] != NULL) {
1804		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1805		sc->rl_irq[0] = NULL;
1806	}
1807	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1808		pci_release_msi(dev);
1809	if (sc->rl_res_pba) {
1810		rid = PCIR_BAR(4);
1811		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1812	}
1813	if (sc->rl_res)
1814		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1815		    sc->rl_res);
1816
1817	/* Unload and free the RX DMA ring memory and map */
1818
1819	if (sc->rl_ldata.rl_rx_list_tag) {
1820		if (sc->rl_ldata.rl_rx_list_map)
1821			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1822			    sc->rl_ldata.rl_rx_list_map);
1823		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1824			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1825			    sc->rl_ldata.rl_rx_list,
1826			    sc->rl_ldata.rl_rx_list_map);
1827		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1828	}
1829
1830	/* Unload and free the TX DMA ring memory and map */
1831
1832	if (sc->rl_ldata.rl_tx_list_tag) {
1833		if (sc->rl_ldata.rl_tx_list_map)
1834			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1835			    sc->rl_ldata.rl_tx_list_map);
1836		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1837			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1838			    sc->rl_ldata.rl_tx_list,
1839			    sc->rl_ldata.rl_tx_list_map);
1840		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1841	}
1842
1843	/* Destroy all the RX and TX buffer maps */
1844
1845	if (sc->rl_ldata.rl_tx_mtag) {
1846		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1847			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1848				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1849				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1850		}
1851		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1852	}
1853	if (sc->rl_ldata.rl_rx_mtag) {
1854		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1855			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1856				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1857				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1858		}
1859		if (sc->rl_ldata.rl_rx_sparemap)
1860			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1861			    sc->rl_ldata.rl_rx_sparemap);
1862		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1863	}
1864	if (sc->rl_ldata.rl_jrx_mtag) {
1865		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1866			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1867				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1868				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1869		}
1870		if (sc->rl_ldata.rl_jrx_sparemap)
1871			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1872			    sc->rl_ldata.rl_jrx_sparemap);
1873		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1874	}
1875	/* Unload and free the stats buffer and map */
1876
1877	if (sc->rl_ldata.rl_stag) {
1878		if (sc->rl_ldata.rl_smap)
1879			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1880			    sc->rl_ldata.rl_smap);
1881		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1882			bus_dmamem_free(sc->rl_ldata.rl_stag,
1883			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1884		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1885	}
1886
1887	if (sc->rl_parent_tag)
1888		bus_dma_tag_destroy(sc->rl_parent_tag);
1889
1890	mtx_destroy(&sc->rl_mtx);
1891
1892	return (0);
1893}
1894
1895static __inline void
1896re_discard_rxbuf(struct rl_softc *sc, int idx)
1897{
1898	struct rl_desc		*desc;
1899	struct rl_rxdesc	*rxd;
1900	uint32_t		cmdstat;
1901
1902	if (sc->rl_ifp->if_mtu > RL_MTU &&
1903	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1904		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1905	else
1906		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1907	desc = &sc->rl_ldata.rl_rx_list[idx];
1908	desc->rl_vlanctl = 0;
1909	cmdstat = rxd->rx_size;
1910	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1911		cmdstat |= RL_RDESC_CMD_EOR;
1912	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1913}
1914
1915static int
1916re_newbuf(struct rl_softc *sc, int idx)
1917{
1918	struct mbuf		*m;
1919	struct rl_rxdesc	*rxd;
1920	bus_dma_segment_t	segs[1];
1921	bus_dmamap_t		map;
1922	struct rl_desc		*desc;
1923	uint32_t		cmdstat;
1924	int			error, nsegs;
1925
1926	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1927	if (m == NULL)
1928		return (ENOBUFS);
1929
1930	m->m_len = m->m_pkthdr.len = MCLBYTES;
1931#ifdef RE_FIXUP_RX
1932	/*
1933	 * This is part of an evil trick to deal with non-x86 platforms.
1934	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1935	 * boundaries, but that will hose non-x86 machines. To get around
1936	 * this, we leave some empty space at the start of each buffer
1937	 * and for non-x86 hosts, we copy the buffer back six bytes
1938	 * to achieve word alignment. This is slightly more efficient
1939	 * than allocating a new buffer, copying the contents, and
1940	 * discarding the old buffer.
1941	 */
1942	m_adj(m, RE_ETHER_ALIGN);
1943#endif
1944	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1945	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1946	if (error != 0) {
1947		m_freem(m);
1948		return (ENOBUFS);
1949	}
1950	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1951
1952	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1953	if (rxd->rx_m != NULL) {
1954		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1955		    BUS_DMASYNC_POSTREAD);
1956		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1957	}
1958
1959	rxd->rx_m = m;
1960	map = rxd->rx_dmamap;
1961	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1962	rxd->rx_size = segs[0].ds_len;
1963	sc->rl_ldata.rl_rx_sparemap = map;
1964	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1965	    BUS_DMASYNC_PREREAD);
1966
1967	desc = &sc->rl_ldata.rl_rx_list[idx];
1968	desc->rl_vlanctl = 0;
1969	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1970	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1971	cmdstat = segs[0].ds_len;
1972	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1973		cmdstat |= RL_RDESC_CMD_EOR;
1974	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1975
1976	return (0);
1977}
1978
1979static int
1980re_jumbo_newbuf(struct rl_softc *sc, int idx)
1981{
1982	struct mbuf		*m;
1983	struct rl_rxdesc	*rxd;
1984	bus_dma_segment_t	segs[1];
1985	bus_dmamap_t		map;
1986	struct rl_desc		*desc;
1987	uint32_t		cmdstat;
1988	int			error, nsegs;
1989
1990	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1991	if (m == NULL)
1992		return (ENOBUFS);
1993	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1994#ifdef RE_FIXUP_RX
1995	m_adj(m, RE_ETHER_ALIGN);
1996#endif
1997	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1998	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1999	if (error != 0) {
2000		m_freem(m);
2001		return (ENOBUFS);
2002	}
2003	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
2004
2005	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
2006	if (rxd->rx_m != NULL) {
2007		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2008		    BUS_DMASYNC_POSTREAD);
2009		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
2010	}
2011
2012	rxd->rx_m = m;
2013	map = rxd->rx_dmamap;
2014	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2015	rxd->rx_size = segs[0].ds_len;
2016	sc->rl_ldata.rl_jrx_sparemap = map;
2017	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2018	    BUS_DMASYNC_PREREAD);
2019
2020	desc = &sc->rl_ldata.rl_rx_list[idx];
2021	desc->rl_vlanctl = 0;
2022	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2023	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2024	cmdstat = segs[0].ds_len;
2025	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2026		cmdstat |= RL_RDESC_CMD_EOR;
2027	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2028
2029	return (0);
2030}
2031
2032#ifdef RE_FIXUP_RX
2033static __inline void
2034re_fixup_rx(struct mbuf *m)
2035{
2036	int                     i;
2037	uint16_t                *src, *dst;
2038
2039	src = mtod(m, uint16_t *);
2040	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2041
2042	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2043		*dst++ = *src++;
2044
2045	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2046}
2047#endif
2048
2049static int
2050re_tx_list_init(struct rl_softc *sc)
2051{
2052	struct rl_desc		*desc;
2053	int			i;
2054
2055	RL_LOCK_ASSERT(sc);
2056
2057	bzero(sc->rl_ldata.rl_tx_list,
2058	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2059	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2060		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2061#ifdef DEV_NETMAP
2062	re_netmap_tx_init(sc);
2063#endif /* DEV_NETMAP */
2064	/* Set EOR. */
2065	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2066	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2067
2068	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2069	    sc->rl_ldata.rl_tx_list_map,
2070	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2071
2072	sc->rl_ldata.rl_tx_prodidx = 0;
2073	sc->rl_ldata.rl_tx_considx = 0;
2074	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2075
2076	return (0);
2077}
2078
2079static int
2080re_rx_list_init(struct rl_softc *sc)
2081{
2082	int			error, i;
2083
2084	bzero(sc->rl_ldata.rl_rx_list,
2085	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2086	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2087		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2088		if ((error = re_newbuf(sc, i)) != 0)
2089			return (error);
2090	}
2091#ifdef DEV_NETMAP
2092	re_netmap_rx_init(sc);
2093#endif /* DEV_NETMAP */
2094
2095	/* Flush the RX descriptors */
2096
2097	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2098	    sc->rl_ldata.rl_rx_list_map,
2099	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2100
2101	sc->rl_ldata.rl_rx_prodidx = 0;
2102	sc->rl_head = sc->rl_tail = NULL;
2103	sc->rl_int_rx_act = 0;
2104
2105	return (0);
2106}
2107
2108static int
2109re_jrx_list_init(struct rl_softc *sc)
2110{
2111	int			error, i;
2112
2113	bzero(sc->rl_ldata.rl_rx_list,
2114	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2115	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2116		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2117		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2118			return (error);
2119	}
2120
2121	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2122	    sc->rl_ldata.rl_rx_list_map,
2123	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2124
2125	sc->rl_ldata.rl_rx_prodidx = 0;
2126	sc->rl_head = sc->rl_tail = NULL;
2127	sc->rl_int_rx_act = 0;
2128
2129	return (0);
2130}
2131
2132/*
2133 * RX handler for C+ and 8169. For the gigE chips, we support
2134 * the reception of jumbo frames that have been fragmented
2135 * across multiple 2K mbuf cluster buffers.
2136 */
2137static int
2138re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2139{
2140	struct mbuf		*m;
2141	struct ifnet		*ifp;
2142	int			i, rxerr, total_len;
2143	struct rl_desc		*cur_rx;
2144	u_int32_t		rxstat, rxvlan;
2145	int			jumbo, maxpkt = 16, rx_npkts = 0;
2146
2147	RL_LOCK_ASSERT(sc);
2148
2149	ifp = sc->rl_ifp;
2150#ifdef DEV_NETMAP
2151	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2152		return 0;
2153#endif /* DEV_NETMAP */
2154	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2155		jumbo = 1;
2156	else
2157		jumbo = 0;
2158
2159	/* Invalidate the descriptor memory */
2160
2161	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2162	    sc->rl_ldata.rl_rx_list_map,
2163	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2164
2165	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2166	    i = RL_RX_DESC_NXT(sc, i)) {
2167		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2168			break;
2169		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2170		rxstat = le32toh(cur_rx->rl_cmdstat);
2171		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2172			break;
2173		total_len = rxstat & sc->rl_rxlenmask;
2174		rxvlan = le32toh(cur_rx->rl_vlanctl);
2175		if (jumbo != 0)
2176			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2177		else
2178			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2179
2180		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2181		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2182		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2183			/*
2184			 * RTL8168C or later controllers do not
2185			 * support multi-fragment packet.
2186			 */
2187			re_discard_rxbuf(sc, i);
2188			continue;
2189		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2190			if (re_newbuf(sc, i) != 0) {
2191				/*
2192				 * If this is part of a multi-fragment packet,
2193				 * discard all the pieces.
2194				 */
2195				if (sc->rl_head != NULL) {
2196					m_freem(sc->rl_head);
2197					sc->rl_head = sc->rl_tail = NULL;
2198				}
2199				re_discard_rxbuf(sc, i);
2200				continue;
2201			}
2202			m->m_len = RE_RX_DESC_BUFLEN;
2203			if (sc->rl_head == NULL)
2204				sc->rl_head = sc->rl_tail = m;
2205			else {
2206				m->m_flags &= ~M_PKTHDR;
2207				sc->rl_tail->m_next = m;
2208				sc->rl_tail = m;
2209			}
2210			continue;
2211		}
2212
2213		/*
2214		 * NOTE: for the 8139C+, the frame length field
2215		 * is always 12 bits in size, but for the gigE chips,
2216		 * it is 13 bits (since the max RX frame length is 16K).
2217		 * Unfortunately, all 32 bits in the status word
2218		 * were already used, so to make room for the extra
2219		 * length bit, RealTek took out the 'frame alignment
2220		 * error' bit and shifted the other status bits
2221		 * over one slot. The OWN, EOR, FS and LS bits are
2222		 * still in the same places. We have already extracted
2223		 * the frame length and checked the OWN bit, so rather
2224		 * than using an alternate bit mapping, we shift the
2225		 * status bits one space to the right so we can evaluate
2226		 * them using the 8169 status as though it was in the
2227		 * same format as that of the 8139C+.
2228		 */
2229		if (sc->rl_type == RL_8169)
2230			rxstat >>= 1;
2231
2232		/*
2233		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2234		 * set, but if CRC is clear, it will still be a valid frame.
2235		 */
2236		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2237			rxerr = 1;
2238			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2239			    total_len > 8191 &&
2240			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2241				rxerr = 0;
2242			if (rxerr != 0) {
2243				ifp->if_ierrors++;
2244				/*
2245				 * If this is part of a multi-fragment packet,
2246				 * discard all the pieces.
2247				 */
2248				if (sc->rl_head != NULL) {
2249					m_freem(sc->rl_head);
2250					sc->rl_head = sc->rl_tail = NULL;
2251				}
2252				re_discard_rxbuf(sc, i);
2253				continue;
2254			}
2255		}
2256
2257		/*
2258		 * If allocating a replacement mbuf fails,
2259		 * reload the current one.
2260		 */
2261		if (jumbo != 0)
2262			rxerr = re_jumbo_newbuf(sc, i);
2263		else
2264			rxerr = re_newbuf(sc, i);
2265		if (rxerr != 0) {
2266			ifp->if_iqdrops++;
2267			if (sc->rl_head != NULL) {
2268				m_freem(sc->rl_head);
2269				sc->rl_head = sc->rl_tail = NULL;
2270			}
2271			re_discard_rxbuf(sc, i);
2272			continue;
2273		}
2274
2275		if (sc->rl_head != NULL) {
2276			if (jumbo != 0)
2277				m->m_len = total_len;
2278			else {
2279				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2280				if (m->m_len == 0)
2281					m->m_len = RE_RX_DESC_BUFLEN;
2282			}
2283			/*
2284			 * Special case: if there's 4 bytes or less
2285			 * in this buffer, the mbuf can be discarded:
2286			 * the last 4 bytes is the CRC, which we don't
2287			 * care about anyway.
2288			 */
2289			if (m->m_len <= ETHER_CRC_LEN) {
2290				sc->rl_tail->m_len -=
2291				    (ETHER_CRC_LEN - m->m_len);
2292				m_freem(m);
2293			} else {
2294				m->m_len -= ETHER_CRC_LEN;
2295				m->m_flags &= ~M_PKTHDR;
2296				sc->rl_tail->m_next = m;
2297			}
2298			m = sc->rl_head;
2299			sc->rl_head = sc->rl_tail = NULL;
2300			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2301		} else
2302			m->m_pkthdr.len = m->m_len =
2303			    (total_len - ETHER_CRC_LEN);
2304
2305#ifdef RE_FIXUP_RX
2306		re_fixup_rx(m);
2307#endif
2308		ifp->if_ipackets++;
2309		m->m_pkthdr.rcvif = ifp;
2310
2311		/* Do RX checksumming if enabled */
2312
2313		if (ifp->if_capenable & IFCAP_RXCSUM) {
2314			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2315				/* Check IP header checksum */
2316				if (rxstat & RL_RDESC_STAT_PROTOID)
2317					m->m_pkthdr.csum_flags |=
2318					    CSUM_IP_CHECKED;
2319				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2320					m->m_pkthdr.csum_flags |=
2321					    CSUM_IP_VALID;
2322
2323				/* Check TCP/UDP checksum */
2324				if ((RL_TCPPKT(rxstat) &&
2325				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2326				    (RL_UDPPKT(rxstat) &&
2327				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2328					m->m_pkthdr.csum_flags |=
2329						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2330					m->m_pkthdr.csum_data = 0xffff;
2331				}
2332			} else {
2333				/*
2334				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2335				 */
2336				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2337				    (rxvlan & RL_RDESC_IPV4))
2338					m->m_pkthdr.csum_flags |=
2339					    CSUM_IP_CHECKED;
2340				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2341				    (rxvlan & RL_RDESC_IPV4))
2342					m->m_pkthdr.csum_flags |=
2343					    CSUM_IP_VALID;
2344				if (((rxstat & RL_RDESC_STAT_TCP) &&
2345				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2346				    ((rxstat & RL_RDESC_STAT_UDP) &&
2347				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2348					m->m_pkthdr.csum_flags |=
2349						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2350					m->m_pkthdr.csum_data = 0xffff;
2351				}
2352			}
2353		}
2354		maxpkt--;
2355		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2356			m->m_pkthdr.ether_vtag =
2357			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2358			m->m_flags |= M_VLANTAG;
2359		}
2360		RL_UNLOCK(sc);
2361		(*ifp->if_input)(ifp, m);
2362		RL_LOCK(sc);
2363		rx_npkts++;
2364	}
2365
2366	/* Flush the RX DMA ring */
2367
2368	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2369	    sc->rl_ldata.rl_rx_list_map,
2370	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2371
2372	sc->rl_ldata.rl_rx_prodidx = i;
2373
2374	if (rx_npktsp != NULL)
2375		*rx_npktsp = rx_npkts;
2376	if (maxpkt)
2377		return (EAGAIN);
2378
2379	return (0);
2380}
2381
2382static void
2383re_txeof(struct rl_softc *sc)
2384{
2385	struct ifnet		*ifp;
2386	struct rl_txdesc	*txd;
2387	u_int32_t		txstat;
2388	int			cons;
2389
2390	cons = sc->rl_ldata.rl_tx_considx;
2391	if (cons == sc->rl_ldata.rl_tx_prodidx)
2392		return;
2393
2394	ifp = sc->rl_ifp;
2395#ifdef DEV_NETMAP
2396	if (netmap_tx_irq(ifp, 0))
2397		return;
2398#endif /* DEV_NETMAP */
2399	/* Invalidate the TX descriptor list */
2400	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2401	    sc->rl_ldata.rl_tx_list_map,
2402	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2403
2404	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2405	    cons = RL_TX_DESC_NXT(sc, cons)) {
2406		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2407		if (txstat & RL_TDESC_STAT_OWN)
2408			break;
2409		/*
2410		 * We only stash mbufs in the last descriptor
2411		 * in a fragment chain, which also happens to
2412		 * be the only place where the TX status bits
2413		 * are valid.
2414		 */
2415		if (txstat & RL_TDESC_CMD_EOF) {
2416			txd = &sc->rl_ldata.rl_tx_desc[cons];
2417			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2418			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2419			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2420			    txd->tx_dmamap);
2421			KASSERT(txd->tx_m != NULL,
2422			    ("%s: freeing NULL mbufs!", __func__));
2423			m_freem(txd->tx_m);
2424			txd->tx_m = NULL;
2425			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2426			    RL_TDESC_STAT_COLCNT))
2427				ifp->if_collisions++;
2428			if (txstat & RL_TDESC_STAT_TXERRSUM)
2429				ifp->if_oerrors++;
2430			else
2431				ifp->if_opackets++;
2432		}
2433		sc->rl_ldata.rl_tx_free++;
2434		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2435	}
2436	sc->rl_ldata.rl_tx_considx = cons;
2437
2438	/* No changes made to the TX ring, so no flush needed */
2439
2440	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2441#ifdef RE_TX_MODERATION
2442		/*
2443		 * If not all descriptors have been reaped yet, reload
2444		 * the timer so that we will eventually get another
2445		 * interrupt that will cause us to re-enter this routine.
2446		 * This is done in case the transmitter has gone idle.
2447		 */
2448		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2449#endif
2450	} else
2451		sc->rl_watchdog_timer = 0;
2452}
2453
2454static void
2455re_tick(void *xsc)
2456{
2457	struct rl_softc		*sc;
2458	struct mii_data		*mii;
2459
2460	sc = xsc;
2461
2462	RL_LOCK_ASSERT(sc);
2463
2464	mii = device_get_softc(sc->rl_miibus);
2465	mii_tick(mii);
2466	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2467		re_miibus_statchg(sc->rl_dev);
2468	/*
2469	 * Reclaim transmitted frames here. Technically it is not
2470	 * necessary to do here but it ensures periodic reclamation
2471	 * regardless of Tx completion interrupt which seems to be
2472	 * lost on PCIe based controllers under certain situations.
2473	 */
2474	re_txeof(sc);
2475	re_watchdog(sc);
2476	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2477}
2478
2479#ifdef DEVICE_POLLING
2480static int
2481re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2482{
2483	struct rl_softc *sc = ifp->if_softc;
2484	int rx_npkts = 0;
2485
2486	RL_LOCK(sc);
2487	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2488		rx_npkts = re_poll_locked(ifp, cmd, count);
2489	RL_UNLOCK(sc);
2490	return (rx_npkts);
2491}
2492
2493static int
2494re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2495{
2496	struct rl_softc *sc = ifp->if_softc;
2497	int rx_npkts;
2498
2499	RL_LOCK_ASSERT(sc);
2500
2501	sc->rxcycles = count;
2502	re_rxeof(sc, &rx_npkts);
2503	re_txeof(sc);
2504
2505	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2506		re_start_locked(ifp);
2507
2508	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2509		u_int16_t       status;
2510
2511		status = CSR_READ_2(sc, RL_ISR);
2512		if (status == 0xffff)
2513			return (rx_npkts);
2514		if (status)
2515			CSR_WRITE_2(sc, RL_ISR, status);
2516		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2517		    (sc->rl_flags & RL_FLAG_PCIE))
2518			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2519
2520		/*
2521		 * XXX check behaviour on receiver stalls.
2522		 */
2523
2524		if (status & RL_ISR_SYSTEM_ERR) {
2525			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2526			re_init_locked(sc);
2527		}
2528	}
2529	return (rx_npkts);
2530}
2531#endif /* DEVICE_POLLING */
2532
2533static int
2534re_intr(void *arg)
2535{
2536	struct rl_softc		*sc;
2537	uint16_t		status;
2538
2539	sc = arg;
2540
2541	status = CSR_READ_2(sc, RL_ISR);
2542	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2543                return (FILTER_STRAY);
2544	CSR_WRITE_2(sc, RL_IMR, 0);
2545
2546	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2547
2548	return (FILTER_HANDLED);
2549}
2550
2551static void
2552re_int_task(void *arg, int npending)
2553{
2554	struct rl_softc		*sc;
2555	struct ifnet		*ifp;
2556	u_int16_t		status;
2557	int			rval = 0;
2558
2559	sc = arg;
2560	ifp = sc->rl_ifp;
2561
2562	RL_LOCK(sc);
2563
2564	status = CSR_READ_2(sc, RL_ISR);
2565        CSR_WRITE_2(sc, RL_ISR, status);
2566
2567	if (sc->suspended ||
2568	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2569		RL_UNLOCK(sc);
2570		return;
2571	}
2572
2573#ifdef DEVICE_POLLING
2574	if  (ifp->if_capenable & IFCAP_POLLING) {
2575		RL_UNLOCK(sc);
2576		return;
2577	}
2578#endif
2579
2580	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2581		rval = re_rxeof(sc, NULL);
2582
2583	/*
2584	 * Some chips will ignore a second TX request issued
2585	 * while an existing transmission is in progress. If
2586	 * the transmitter goes idle but there are still
2587	 * packets waiting to be sent, we need to restart the
2588	 * channel here to flush them out. This only seems to
2589	 * be required with the PCIe devices.
2590	 */
2591	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2592	    (sc->rl_flags & RL_FLAG_PCIE))
2593		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2594	if (status & (
2595#ifdef RE_TX_MODERATION
2596	    RL_ISR_TIMEOUT_EXPIRED|
2597#else
2598	    RL_ISR_TX_OK|
2599#endif
2600	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2601		re_txeof(sc);
2602
2603	if (status & RL_ISR_SYSTEM_ERR) {
2604		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2605		re_init_locked(sc);
2606	}
2607
2608	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2609		re_start_locked(ifp);
2610
2611	RL_UNLOCK(sc);
2612
2613        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2614		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2615		return;
2616	}
2617
2618	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2619}
2620
2621static void
2622re_intr_msi(void *xsc)
2623{
2624	struct rl_softc		*sc;
2625	struct ifnet		*ifp;
2626	uint16_t		intrs, status;
2627
2628	sc = xsc;
2629	RL_LOCK(sc);
2630
2631	ifp = sc->rl_ifp;
2632#ifdef DEVICE_POLLING
2633	if (ifp->if_capenable & IFCAP_POLLING) {
2634		RL_UNLOCK(sc);
2635		return;
2636	}
2637#endif
2638	/* Disable interrupts. */
2639	CSR_WRITE_2(sc, RL_IMR, 0);
2640	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2641		RL_UNLOCK(sc);
2642		return;
2643	}
2644
2645	intrs = RL_INTRS_CPLUS;
2646	status = CSR_READ_2(sc, RL_ISR);
2647        CSR_WRITE_2(sc, RL_ISR, status);
2648	if (sc->rl_int_rx_act > 0) {
2649		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2650		    RL_ISR_RX_OVERRUN);
2651		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2652		    RL_ISR_RX_OVERRUN);
2653	}
2654
2655	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2656	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2657		re_rxeof(sc, NULL);
2658		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2659			if (sc->rl_int_rx_mod != 0 &&
2660			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2661			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2662				/* Rearm one-shot timer. */
2663				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2664				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2665				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2666				sc->rl_int_rx_act = 1;
2667			} else {
2668				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2669				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2670				sc->rl_int_rx_act = 0;
2671			}
2672		}
2673	}
2674
2675	/*
2676	 * Some chips will ignore a second TX request issued
2677	 * while an existing transmission is in progress. If
2678	 * the transmitter goes idle but there are still
2679	 * packets waiting to be sent, we need to restart the
2680	 * channel here to flush them out. This only seems to
2681	 * be required with the PCIe devices.
2682	 */
2683	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2684	    (sc->rl_flags & RL_FLAG_PCIE))
2685		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2686	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2687		re_txeof(sc);
2688
2689	if (status & RL_ISR_SYSTEM_ERR) {
2690		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2691		re_init_locked(sc);
2692	}
2693
2694	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2695		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2696			re_start_locked(ifp);
2697		CSR_WRITE_2(sc, RL_IMR, intrs);
2698	}
2699	RL_UNLOCK(sc);
2700}
2701
2702static int
2703re_encap(struct rl_softc *sc, struct mbuf **m_head)
2704{
2705	struct rl_txdesc	*txd, *txd_last;
2706	bus_dma_segment_t	segs[RL_NTXSEGS];
2707	bus_dmamap_t		map;
2708	struct mbuf		*m_new;
2709	struct rl_desc		*desc;
2710	int			nsegs, prod;
2711	int			i, error, ei, si;
2712	int			padlen;
2713	uint32_t		cmdstat, csum_flags, vlanctl;
2714
2715	RL_LOCK_ASSERT(sc);
2716	M_ASSERTPKTHDR((*m_head));
2717
2718	/*
2719	 * With some of the RealTek chips, using the checksum offload
2720	 * support in conjunction with the autopadding feature results
2721	 * in the transmission of corrupt frames. For example, if we
2722	 * need to send a really small IP fragment that's less than 60
2723	 * bytes in size, and IP header checksumming is enabled, the
2724	 * resulting ethernet frame that appears on the wire will
2725	 * have garbled payload. To work around this, if TX IP checksum
2726	 * offload is enabled, we always manually pad short frames out
2727	 * to the minimum ethernet frame size.
2728	 */
2729	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2730	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2731	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2732		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2733		if (M_WRITABLE(*m_head) == 0) {
2734			/* Get a writable copy. */
2735			m_new = m_dup(*m_head, M_NOWAIT);
2736			m_freem(*m_head);
2737			if (m_new == NULL) {
2738				*m_head = NULL;
2739				return (ENOBUFS);
2740			}
2741			*m_head = m_new;
2742		}
2743		if ((*m_head)->m_next != NULL ||
2744		    M_TRAILINGSPACE(*m_head) < padlen) {
2745			m_new = m_defrag(*m_head, M_NOWAIT);
2746			if (m_new == NULL) {
2747				m_freem(*m_head);
2748				*m_head = NULL;
2749				return (ENOBUFS);
2750			}
2751		} else
2752			m_new = *m_head;
2753
2754		/*
2755		 * Manually pad short frames, and zero the pad space
2756		 * to avoid leaking data.
2757		 */
2758		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2759		m_new->m_pkthdr.len += padlen;
2760		m_new->m_len = m_new->m_pkthdr.len;
2761		*m_head = m_new;
2762	}
2763
2764	prod = sc->rl_ldata.rl_tx_prodidx;
2765	txd = &sc->rl_ldata.rl_tx_desc[prod];
2766	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2767	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2768	if (error == EFBIG) {
2769		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2770		if (m_new == NULL) {
2771			m_freem(*m_head);
2772			*m_head = NULL;
2773			return (ENOBUFS);
2774		}
2775		*m_head = m_new;
2776		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2777		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2778		if (error != 0) {
2779			m_freem(*m_head);
2780			*m_head = NULL;
2781			return (error);
2782		}
2783	} else if (error != 0)
2784		return (error);
2785	if (nsegs == 0) {
2786		m_freem(*m_head);
2787		*m_head = NULL;
2788		return (EIO);
2789	}
2790
2791	/* Check for number of available descriptors. */
2792	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2793		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2794		return (ENOBUFS);
2795	}
2796
2797	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2798	    BUS_DMASYNC_PREWRITE);
2799
2800	/*
2801	 * Set up checksum offload. Note: checksum offload bits must
2802	 * appear in all descriptors of a multi-descriptor transmit
2803	 * attempt. This is according to testing done with an 8169
2804	 * chip. This is a requirement.
2805	 */
2806	vlanctl = 0;
2807	csum_flags = 0;
2808	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2809		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2810			csum_flags |= RL_TDESC_CMD_LGSEND;
2811			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2812			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2813		} else {
2814			csum_flags |= RL_TDESC_CMD_LGSEND |
2815			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2816			    RL_TDESC_CMD_MSSVAL_SHIFT);
2817		}
2818	} else {
2819		/*
2820		 * Unconditionally enable IP checksum if TCP or UDP
2821		 * checksum is required. Otherwise, TCP/UDP checksum
2822		 * does't make effects.
2823		 */
2824		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2825			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2826				csum_flags |= RL_TDESC_CMD_IPCSUM;
2827				if (((*m_head)->m_pkthdr.csum_flags &
2828				    CSUM_TCP) != 0)
2829					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2830				if (((*m_head)->m_pkthdr.csum_flags &
2831				    CSUM_UDP) != 0)
2832					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2833			} else {
2834				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2835				if (((*m_head)->m_pkthdr.csum_flags &
2836				    CSUM_TCP) != 0)
2837					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2838				if (((*m_head)->m_pkthdr.csum_flags &
2839				    CSUM_UDP) != 0)
2840					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2841			}
2842		}
2843	}
2844
2845	/*
2846	 * Set up hardware VLAN tagging. Note: vlan tag info must
2847	 * appear in all descriptors of a multi-descriptor
2848	 * transmission attempt.
2849	 */
2850	if ((*m_head)->m_flags & M_VLANTAG)
2851		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2852		    RL_TDESC_VLANCTL_TAG;
2853
2854	si = prod;
2855	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2856		desc = &sc->rl_ldata.rl_tx_list[prod];
2857		desc->rl_vlanctl = htole32(vlanctl);
2858		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2859		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2860		cmdstat = segs[i].ds_len;
2861		if (i != 0)
2862			cmdstat |= RL_TDESC_CMD_OWN;
2863		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2864			cmdstat |= RL_TDESC_CMD_EOR;
2865		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2866		sc->rl_ldata.rl_tx_free--;
2867	}
2868	/* Update producer index. */
2869	sc->rl_ldata.rl_tx_prodidx = prod;
2870
2871	/* Set EOF on the last descriptor. */
2872	ei = RL_TX_DESC_PRV(sc, prod);
2873	desc = &sc->rl_ldata.rl_tx_list[ei];
2874	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2875
2876	desc = &sc->rl_ldata.rl_tx_list[si];
2877	/* Set SOF and transfer ownership of packet to the chip. */
2878	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2879
2880	/*
2881	 * Insure that the map for this transmission
2882	 * is placed at the array index of the last descriptor
2883	 * in this chain.  (Swap last and first dmamaps.)
2884	 */
2885	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2886	map = txd->tx_dmamap;
2887	txd->tx_dmamap = txd_last->tx_dmamap;
2888	txd_last->tx_dmamap = map;
2889	txd_last->tx_m = *m_head;
2890
2891	return (0);
2892}
2893
2894static void
2895re_start(struct ifnet *ifp)
2896{
2897	struct rl_softc		*sc;
2898
2899	sc = ifp->if_softc;
2900	RL_LOCK(sc);
2901	re_start_locked(ifp);
2902	RL_UNLOCK(sc);
2903}
2904
2905/*
2906 * Main transmit routine for C+ and gigE NICs.
2907 */
2908static void
2909re_start_locked(struct ifnet *ifp)
2910{
2911	struct rl_softc		*sc;
2912	struct mbuf		*m_head;
2913	int			queued;
2914
2915	sc = ifp->if_softc;
2916
2917#ifdef DEV_NETMAP
2918	/* XXX is this necessary ? */
2919	if (ifp->if_capenable & IFCAP_NETMAP) {
2920		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2921		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2922			/* kick the tx unit */
2923			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2924#ifdef RE_TX_MODERATION
2925			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2926#endif
2927			sc->rl_watchdog_timer = 5;
2928		}
2929		return;
2930	}
2931#endif /* DEV_NETMAP */
2932	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2933	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2934		return;
2935
2936	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2937	    sc->rl_ldata.rl_tx_free > 1;) {
2938		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2939		if (m_head == NULL)
2940			break;
2941
2942		if (re_encap(sc, &m_head) != 0) {
2943			if (m_head == NULL)
2944				break;
2945			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2946			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2947			break;
2948		}
2949
2950		/*
2951		 * If there's a BPF listener, bounce a copy of this frame
2952		 * to him.
2953		 */
2954		ETHER_BPF_MTAP(ifp, m_head);
2955
2956		queued++;
2957	}
2958
2959	if (queued == 0) {
2960#ifdef RE_TX_MODERATION
2961		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2962			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2963#endif
2964		return;
2965	}
2966
2967	/* Flush the TX descriptors */
2968
2969	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2970	    sc->rl_ldata.rl_tx_list_map,
2971	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2972
2973	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2974
2975#ifdef RE_TX_MODERATION
2976	/*
2977	 * Use the countdown timer for interrupt moderation.
2978	 * 'TX done' interrupts are disabled. Instead, we reset the
2979	 * countdown timer, which will begin counting until it hits
2980	 * the value in the TIMERINT register, and then trigger an
2981	 * interrupt. Each time we write to the TIMERCNT register,
2982	 * the timer count is reset to 0.
2983	 */
2984	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2985#endif
2986
2987	/*
2988	 * Set a timeout in case the chip goes out to lunch.
2989	 */
2990	sc->rl_watchdog_timer = 5;
2991}
2992
2993static void
2994re_set_jumbo(struct rl_softc *sc, int jumbo)
2995{
2996
2997	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2998		pci_set_max_read_req(sc->rl_dev, 4096);
2999		return;
3000	}
3001
3002	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3003	if (jumbo != 0) {
3004		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
3005		    RL_CFG3_JUMBO_EN0);
3006		switch (sc->rl_hwrev->rl_rev) {
3007		case RL_HWREV_8168DP:
3008			break;
3009		case RL_HWREV_8168E:
3010			CSR_WRITE_1(sc, sc->rl_cfg4,
3011			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
3012			break;
3013		default:
3014			CSR_WRITE_1(sc, sc->rl_cfg4,
3015			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3016		}
3017	} else {
3018		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3019		    ~RL_CFG3_JUMBO_EN0);
3020		switch (sc->rl_hwrev->rl_rev) {
3021		case RL_HWREV_8168DP:
3022			break;
3023		case RL_HWREV_8168E:
3024			CSR_WRITE_1(sc, sc->rl_cfg4,
3025			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3026			break;
3027		default:
3028			CSR_WRITE_1(sc, sc->rl_cfg4,
3029			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3030		}
3031	}
3032	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3033
3034	switch (sc->rl_hwrev->rl_rev) {
3035	case RL_HWREV_8168DP:
3036		pci_set_max_read_req(sc->rl_dev, 4096);
3037		break;
3038	default:
3039		if (jumbo != 0)
3040			pci_set_max_read_req(sc->rl_dev, 512);
3041		else
3042			pci_set_max_read_req(sc->rl_dev, 4096);
3043	}
3044}
3045
3046static void
3047re_init(void *xsc)
3048{
3049	struct rl_softc		*sc = xsc;
3050
3051	RL_LOCK(sc);
3052	re_init_locked(sc);
3053	RL_UNLOCK(sc);
3054}
3055
3056static void
3057re_init_locked(struct rl_softc *sc)
3058{
3059	struct ifnet		*ifp = sc->rl_ifp;
3060	struct mii_data		*mii;
3061	uint32_t		reg;
3062	uint16_t		cfg;
3063	union {
3064		uint32_t align_dummy;
3065		u_char eaddr[ETHER_ADDR_LEN];
3066        } eaddr;
3067
3068	RL_LOCK_ASSERT(sc);
3069
3070	mii = device_get_softc(sc->rl_miibus);
3071
3072	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3073		return;
3074
3075	/*
3076	 * Cancel pending I/O and free all RX/TX buffers.
3077	 */
3078	re_stop(sc);
3079
3080	/* Put controller into known state. */
3081	re_reset(sc);
3082
3083	/*
3084	 * For C+ mode, initialize the RX descriptors and mbufs.
3085	 */
3086	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3087		if (ifp->if_mtu > RL_MTU) {
3088			if (re_jrx_list_init(sc) != 0) {
3089				device_printf(sc->rl_dev,
3090				    "no memory for jumbo RX buffers\n");
3091				re_stop(sc);
3092				return;
3093			}
3094			/* Disable checksum offloading for jumbo frames. */
3095			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3096			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3097		} else {
3098			if (re_rx_list_init(sc) != 0) {
3099				device_printf(sc->rl_dev,
3100				    "no memory for RX buffers\n");
3101				re_stop(sc);
3102				return;
3103			}
3104		}
3105		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3106	} else {
3107		if (re_rx_list_init(sc) != 0) {
3108			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3109			re_stop(sc);
3110			return;
3111		}
3112		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3113		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3114			if (ifp->if_mtu > RL_MTU)
3115				pci_set_max_read_req(sc->rl_dev, 512);
3116			else
3117				pci_set_max_read_req(sc->rl_dev, 4096);
3118		}
3119	}
3120	re_tx_list_init(sc);
3121
3122	/*
3123	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3124	 * RX checksum offload. We must configure the C+ register
3125	 * before all others.
3126	 */
3127	cfg = RL_CPLUSCMD_PCI_MRW;
3128	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3129		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3130	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3131		cfg |= RL_CPLUSCMD_VLANSTRIP;
3132	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3133		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3134		/* XXX magic. */
3135		cfg |= 0x0001;
3136	} else
3137		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3138	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3139	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3140	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3141		reg = 0x000fff00;
3142		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3143			reg |= 0x000000ff;
3144		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3145			reg |= 0x00f00000;
3146		CSR_WRITE_4(sc, 0x7c, reg);
3147		/* Disable interrupt mitigation. */
3148		CSR_WRITE_2(sc, 0xe2, 0);
3149	}
3150	/*
3151	 * Disable TSO if interface MTU size is greater than MSS
3152	 * allowed in controller.
3153	 */
3154	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3155		ifp->if_capenable &= ~IFCAP_TSO4;
3156		ifp->if_hwassist &= ~CSUM_TSO;
3157	}
3158
3159	/*
3160	 * Init our MAC address.  Even though the chipset
3161	 * documentation doesn't mention it, we need to enter "Config
3162	 * register write enable" mode to modify the ID registers.
3163	 */
3164	/* Copy MAC address on stack to align. */
3165	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3166	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3167	CSR_WRITE_4(sc, RL_IDR0,
3168	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3169	CSR_WRITE_4(sc, RL_IDR4,
3170	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3171	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3172
3173	/*
3174	 * Load the addresses of the RX and TX lists into the chip.
3175	 */
3176
3177	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3178	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3179	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3180	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3181
3182	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3183	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3184	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3185	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3186
3187	if ((sc->rl_flags & RL_FLAG_RXDV_GATED) != 0)
3188		CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
3189		    ~0x00080000);
3190
3191	/*
3192	 * Enable transmit and receive.
3193	 */
3194	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3195
3196	/*
3197	 * Set the initial TX configuration.
3198	 */
3199	if (sc->rl_testmode) {
3200		if (sc->rl_type == RL_8169)
3201			CSR_WRITE_4(sc, RL_TXCFG,
3202			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3203		else
3204			CSR_WRITE_4(sc, RL_TXCFG,
3205			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3206	} else
3207		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3208
3209	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3210
3211	/*
3212	 * Set the initial RX configuration.
3213	 */
3214	re_set_rxmode(sc);
3215
3216	/* Configure interrupt moderation. */
3217	if (sc->rl_type == RL_8169) {
3218		/* Magic from vendor. */
3219		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3220	}
3221
3222#ifdef DEVICE_POLLING
3223	/*
3224	 * Disable interrupts if we are polling.
3225	 */
3226	if (ifp->if_capenable & IFCAP_POLLING)
3227		CSR_WRITE_2(sc, RL_IMR, 0);
3228	else	/* otherwise ... */
3229#endif
3230
3231	/*
3232	 * Enable interrupts.
3233	 */
3234	if (sc->rl_testmode)
3235		CSR_WRITE_2(sc, RL_IMR, 0);
3236	else
3237		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3238	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3239
3240	/* Set initial TX threshold */
3241	sc->rl_txthresh = RL_TX_THRESH_INIT;
3242
3243	/* Start RX/TX process. */
3244	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3245#ifdef notdef
3246	/* Enable receiver and transmitter. */
3247	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3248#endif
3249
3250	/*
3251	 * Initialize the timer interrupt register so that
3252	 * a timer interrupt will be generated once the timer
3253	 * reaches a certain number of ticks. The timer is
3254	 * reloaded on each transmit.
3255	 */
3256#ifdef RE_TX_MODERATION
3257	/*
3258	 * Use timer interrupt register to moderate TX interrupt
3259	 * moderation, which dramatically improves TX frame rate.
3260	 */
3261	if (sc->rl_type == RL_8169)
3262		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3263	else
3264		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3265#else
3266	/*
3267	 * Use timer interrupt register to moderate RX interrupt
3268	 * moderation.
3269	 */
3270	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3271	    intr_filter == 0) {
3272		if (sc->rl_type == RL_8169)
3273			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3274			    RL_USECS(sc->rl_int_rx_mod));
3275	} else {
3276		if (sc->rl_type == RL_8169)
3277			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3278	}
3279#endif
3280
3281	/*
3282	 * For 8169 gigE NICs, set the max allowed RX packet
3283	 * size so we can receive jumbo frames.
3284	 */
3285	if (sc->rl_type == RL_8169) {
3286		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3287			/*
3288			 * For controllers that use new jumbo frame scheme,
3289			 * set maximum size of jumbo frame depedning on
3290			 * controller revisions.
3291			 */
3292			if (ifp->if_mtu > RL_MTU)
3293				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3294				    sc->rl_hwrev->rl_max_mtu +
3295				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3296				    ETHER_CRC_LEN);
3297			else
3298				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3299				    RE_RX_DESC_BUFLEN);
3300		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3301		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3302			/* RTL810x has no jumbo frame support. */
3303			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3304		} else
3305			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3306	}
3307
3308	if (sc->rl_testmode)
3309		return;
3310
3311	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3312	    RL_CFG1_DRVLOAD);
3313
3314	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3315	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3316
3317	sc->rl_flags &= ~RL_FLAG_LINK;
3318	mii_mediachg(mii);
3319
3320	sc->rl_watchdog_timer = 0;
3321	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3322}
3323
3324/*
3325 * Set media options.
3326 */
3327static int
3328re_ifmedia_upd(struct ifnet *ifp)
3329{
3330	struct rl_softc		*sc;
3331	struct mii_data		*mii;
3332	int			error;
3333
3334	sc = ifp->if_softc;
3335	mii = device_get_softc(sc->rl_miibus);
3336	RL_LOCK(sc);
3337	error = mii_mediachg(mii);
3338	RL_UNLOCK(sc);
3339
3340	return (error);
3341}
3342
3343/*
3344 * Report current media status.
3345 */
3346static void
3347re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3348{
3349	struct rl_softc		*sc;
3350	struct mii_data		*mii;
3351
3352	sc = ifp->if_softc;
3353	mii = device_get_softc(sc->rl_miibus);
3354
3355	RL_LOCK(sc);
3356	mii_pollstat(mii);
3357	ifmr->ifm_active = mii->mii_media_active;
3358	ifmr->ifm_status = mii->mii_media_status;
3359	RL_UNLOCK(sc);
3360}
3361
3362static int
3363re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3364{
3365	struct rl_softc		*sc = ifp->if_softc;
3366	struct ifreq		*ifr = (struct ifreq *) data;
3367	struct mii_data		*mii;
3368	int			error = 0;
3369
3370	switch (command) {
3371	case SIOCSIFMTU:
3372		if (ifr->ifr_mtu < ETHERMIN ||
3373		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3374		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3375		    ifr->ifr_mtu > RL_MTU)) {
3376			error = EINVAL;
3377			break;
3378		}
3379		RL_LOCK(sc);
3380		if (ifp->if_mtu != ifr->ifr_mtu) {
3381			ifp->if_mtu = ifr->ifr_mtu;
3382			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3383			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3384				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3385				re_init_locked(sc);
3386			}
3387			if (ifp->if_mtu > RL_TSO_MTU &&
3388			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3389				ifp->if_capenable &= ~(IFCAP_TSO4 |
3390				    IFCAP_VLAN_HWTSO);
3391				ifp->if_hwassist &= ~CSUM_TSO;
3392			}
3393			VLAN_CAPABILITIES(ifp);
3394		}
3395		RL_UNLOCK(sc);
3396		break;
3397	case SIOCSIFFLAGS:
3398		RL_LOCK(sc);
3399		if ((ifp->if_flags & IFF_UP) != 0) {
3400			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3401				if (((ifp->if_flags ^ sc->rl_if_flags)
3402				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3403					re_set_rxmode(sc);
3404			} else
3405				re_init_locked(sc);
3406		} else {
3407			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3408				re_stop(sc);
3409		}
3410		sc->rl_if_flags = ifp->if_flags;
3411		RL_UNLOCK(sc);
3412		break;
3413	case SIOCADDMULTI:
3414	case SIOCDELMULTI:
3415		RL_LOCK(sc);
3416		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3417			re_set_rxmode(sc);
3418		RL_UNLOCK(sc);
3419		break;
3420	case SIOCGIFMEDIA:
3421	case SIOCSIFMEDIA:
3422		mii = device_get_softc(sc->rl_miibus);
3423		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3424		break;
3425	case SIOCSIFCAP:
3426	    {
3427		int mask, reinit;
3428
3429		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3430		reinit = 0;
3431#ifdef DEVICE_POLLING
3432		if (mask & IFCAP_POLLING) {
3433			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3434				error = ether_poll_register(re_poll, ifp);
3435				if (error)
3436					return (error);
3437				RL_LOCK(sc);
3438				/* Disable interrupts */
3439				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3440				ifp->if_capenable |= IFCAP_POLLING;
3441				RL_UNLOCK(sc);
3442			} else {
3443				error = ether_poll_deregister(ifp);
3444				/* Enable interrupts. */
3445				RL_LOCK(sc);
3446				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3447				ifp->if_capenable &= ~IFCAP_POLLING;
3448				RL_UNLOCK(sc);
3449			}
3450		}
3451#endif /* DEVICE_POLLING */
3452		RL_LOCK(sc);
3453		if ((mask & IFCAP_TXCSUM) != 0 &&
3454		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3455			ifp->if_capenable ^= IFCAP_TXCSUM;
3456			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
3457				ifp->if_hwassist |= RE_CSUM_FEATURES;
3458			else
3459				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3460			reinit = 1;
3461		}
3462		if ((mask & IFCAP_RXCSUM) != 0 &&
3463		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3464			ifp->if_capenable ^= IFCAP_RXCSUM;
3465			reinit = 1;
3466		}
3467		if ((mask & IFCAP_TSO4) != 0 &&
3468		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3469			ifp->if_capenable ^= IFCAP_TSO4;
3470			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3471				ifp->if_hwassist |= CSUM_TSO;
3472			else
3473				ifp->if_hwassist &= ~CSUM_TSO;
3474			if (ifp->if_mtu > RL_TSO_MTU &&
3475			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3476				ifp->if_capenable &= ~IFCAP_TSO4;
3477				ifp->if_hwassist &= ~CSUM_TSO;
3478			}
3479		}
3480		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3481		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3482			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3483		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3484		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3485			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3486			/* TSO over VLAN requires VLAN hardware tagging. */
3487			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3488				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3489			reinit = 1;
3490		}
3491		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3492		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3493		    IFCAP_VLAN_HWTSO)) != 0)
3494				reinit = 1;
3495		if ((mask & IFCAP_WOL) != 0 &&
3496		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3497			if ((mask & IFCAP_WOL_UCAST) != 0)
3498				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3499			if ((mask & IFCAP_WOL_MCAST) != 0)
3500				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3501			if ((mask & IFCAP_WOL_MAGIC) != 0)
3502				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3503		}
3504		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3505			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3506			re_init_locked(sc);
3507		}
3508		RL_UNLOCK(sc);
3509		VLAN_CAPABILITIES(ifp);
3510	    }
3511		break;
3512	default:
3513		error = ether_ioctl(ifp, command, data);
3514		break;
3515	}
3516
3517	return (error);
3518}
3519
3520static void
3521re_watchdog(struct rl_softc *sc)
3522{
3523	struct ifnet		*ifp;
3524
3525	RL_LOCK_ASSERT(sc);
3526
3527	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3528		return;
3529
3530	ifp = sc->rl_ifp;
3531	re_txeof(sc);
3532	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3533		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3534		    "-- recovering\n");
3535		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3536			re_start_locked(ifp);
3537		return;
3538	}
3539
3540	if_printf(ifp, "watchdog timeout\n");
3541	ifp->if_oerrors++;
3542
3543	re_rxeof(sc, NULL);
3544	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3545	re_init_locked(sc);
3546	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3547		re_start_locked(ifp);
3548}
3549
3550/*
3551 * Stop the adapter and free any mbufs allocated to the
3552 * RX and TX lists.
3553 */
3554static void
3555re_stop(struct rl_softc *sc)
3556{
3557	int			i;
3558	struct ifnet		*ifp;
3559	struct rl_txdesc	*txd;
3560	struct rl_rxdesc	*rxd;
3561
3562	RL_LOCK_ASSERT(sc);
3563
3564	ifp = sc->rl_ifp;
3565
3566	sc->rl_watchdog_timer = 0;
3567	callout_stop(&sc->rl_stat_callout);
3568	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3569
3570	/*
3571	 * Disable accepting frames to put RX MAC into idle state.
3572	 * Otherwise it's possible to get frames while stop command
3573	 * execution is in progress and controller can DMA the frame
3574	 * to already freed RX buffer during that period.
3575	 */
3576	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3577	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3578	    RL_RXCFG_RX_BROAD));
3579
3580	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3581		for (i = RL_TIMEOUT; i > 0; i--) {
3582			if ((CSR_READ_1(sc, sc->rl_txstart) &
3583			    RL_TXSTART_START) == 0)
3584				break;
3585			DELAY(20);
3586		}
3587		if (i == 0)
3588			device_printf(sc->rl_dev,
3589			    "stopping TX poll timed out!\n");
3590		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3591	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3592		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3593		    RL_CMD_RX_ENB);
3594		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3595			for (i = RL_TIMEOUT; i > 0; i--) {
3596				if ((CSR_READ_4(sc, RL_TXCFG) &
3597				    RL_TXCFG_QUEUE_EMPTY) != 0)
3598					break;
3599				DELAY(100);
3600			}
3601			if (i == 0)
3602				device_printf(sc->rl_dev,
3603				   "stopping TXQ timed out!\n");
3604		}
3605	} else
3606		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3607	DELAY(1000);
3608	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3609	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3610
3611	if (sc->rl_head != NULL) {
3612		m_freem(sc->rl_head);
3613		sc->rl_head = sc->rl_tail = NULL;
3614	}
3615
3616	/* Free the TX list buffers. */
3617	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3618		txd = &sc->rl_ldata.rl_tx_desc[i];
3619		if (txd->tx_m != NULL) {
3620			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3621			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3622			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3623			    txd->tx_dmamap);
3624			m_freem(txd->tx_m);
3625			txd->tx_m = NULL;
3626		}
3627	}
3628
3629	/* Free the RX list buffers. */
3630	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3631		rxd = &sc->rl_ldata.rl_rx_desc[i];
3632		if (rxd->rx_m != NULL) {
3633			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3634			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3635			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3636			    rxd->rx_dmamap);
3637			m_freem(rxd->rx_m);
3638			rxd->rx_m = NULL;
3639		}
3640	}
3641
3642	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3643		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3644			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3645			if (rxd->rx_m != NULL) {
3646				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3647				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3648				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3649				    rxd->rx_dmamap);
3650				m_freem(rxd->rx_m);
3651				rxd->rx_m = NULL;
3652			}
3653		}
3654	}
3655}
3656
3657/*
3658 * Device suspend routine.  Stop the interface and save some PCI
3659 * settings in case the BIOS doesn't restore them properly on
3660 * resume.
3661 */
3662static int
3663re_suspend(device_t dev)
3664{
3665	struct rl_softc		*sc;
3666
3667	sc = device_get_softc(dev);
3668
3669	RL_LOCK(sc);
3670	re_stop(sc);
3671	re_setwol(sc);
3672	sc->suspended = 1;
3673	RL_UNLOCK(sc);
3674
3675	return (0);
3676}
3677
3678/*
3679 * Device resume routine.  Restore some PCI settings in case the BIOS
3680 * doesn't, re-enable busmastering, and restart the interface if
3681 * appropriate.
3682 */
3683static int
3684re_resume(device_t dev)
3685{
3686	struct rl_softc		*sc;
3687	struct ifnet		*ifp;
3688
3689	sc = device_get_softc(dev);
3690
3691	RL_LOCK(sc);
3692
3693	ifp = sc->rl_ifp;
3694	/* Take controller out of sleep mode. */
3695	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3696		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3697			CSR_WRITE_1(sc, RL_GPIO,
3698			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3699	}
3700
3701	/*
3702	 * Clear WOL matching such that normal Rx filtering
3703	 * wouldn't interfere with WOL patterns.
3704	 */
3705	re_clrwol(sc);
3706
3707	/* reinitialize interface if necessary */
3708	if (ifp->if_flags & IFF_UP)
3709		re_init_locked(sc);
3710
3711	sc->suspended = 0;
3712	RL_UNLOCK(sc);
3713
3714	return (0);
3715}
3716
3717/*
3718 * Stop all chip I/O so that the kernel's probe routines don't
3719 * get confused by errant DMAs when rebooting.
3720 */
3721static int
3722re_shutdown(device_t dev)
3723{
3724	struct rl_softc		*sc;
3725
3726	sc = device_get_softc(dev);
3727
3728	RL_LOCK(sc);
3729	re_stop(sc);
3730	/*
3731	 * Mark interface as down since otherwise we will panic if
3732	 * interrupt comes in later on, which can happen in some
3733	 * cases.
3734	 */
3735	sc->rl_ifp->if_flags &= ~IFF_UP;
3736	re_setwol(sc);
3737	RL_UNLOCK(sc);
3738
3739	return (0);
3740}
3741
3742static void
3743re_set_linkspeed(struct rl_softc *sc)
3744{
3745	struct mii_softc *miisc;
3746	struct mii_data *mii;
3747	int aneg, i, phyno;
3748
3749	RL_LOCK_ASSERT(sc);
3750
3751	mii = device_get_softc(sc->rl_miibus);
3752	mii_pollstat(mii);
3753	aneg = 0;
3754	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3755	    (IFM_ACTIVE | IFM_AVALID)) {
3756		switch IFM_SUBTYPE(mii->mii_media_active) {
3757		case IFM_10_T:
3758		case IFM_100_TX:
3759			return;
3760		case IFM_1000_T:
3761			aneg++;
3762			break;
3763		default:
3764			break;
3765		}
3766	}
3767	miisc = LIST_FIRST(&mii->mii_phys);
3768	phyno = miisc->mii_phy;
3769	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3770		PHY_RESET(miisc);
3771	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3772	re_miibus_writereg(sc->rl_dev, phyno,
3773	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3774	re_miibus_writereg(sc->rl_dev, phyno,
3775	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3776	DELAY(1000);
3777	if (aneg != 0) {
3778		/*
3779		 * Poll link state until re(4) get a 10/100Mbps link.
3780		 */
3781		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3782			mii_pollstat(mii);
3783			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3784			    == (IFM_ACTIVE | IFM_AVALID)) {
3785				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3786				case IFM_10_T:
3787				case IFM_100_TX:
3788					return;
3789				default:
3790					break;
3791				}
3792			}
3793			RL_UNLOCK(sc);
3794			pause("relnk", hz);
3795			RL_LOCK(sc);
3796		}
3797		if (i == MII_ANEGTICKS_GIGE)
3798			device_printf(sc->rl_dev,
3799			    "establishing a link failed, WOL may not work!");
3800	}
3801	/*
3802	 * No link, force MAC to have 100Mbps, full-duplex link.
3803	 * MAC does not require reprogramming on resolved speed/duplex,
3804	 * so this is just for completeness.
3805	 */
3806	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3807	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3808}
3809
3810static void
3811re_setwol(struct rl_softc *sc)
3812{
3813	struct ifnet		*ifp;
3814	int			pmc;
3815	uint16_t		pmstat;
3816	uint8_t			v;
3817
3818	RL_LOCK_ASSERT(sc);
3819
3820	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3821		return;
3822
3823	ifp = sc->rl_ifp;
3824	/* Put controller into sleep mode. */
3825	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3826		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3827			CSR_WRITE_1(sc, RL_GPIO,
3828			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3829	}
3830	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3831		re_set_rxmode(sc);
3832		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3833			re_set_linkspeed(sc);
3834		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3835			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3836	}
3837	/* Enable config register write. */
3838	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3839
3840	/* Enable PME. */
3841	v = CSR_READ_1(sc, sc->rl_cfg1);
3842	v &= ~RL_CFG1_PME;
3843	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3844		v |= RL_CFG1_PME;
3845	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3846
3847	v = CSR_READ_1(sc, sc->rl_cfg3);
3848	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3849	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3850		v |= RL_CFG3_WOL_MAGIC;
3851	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3852
3853	v = CSR_READ_1(sc, sc->rl_cfg5);
3854	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3855	    RL_CFG5_WOL_LANWAKE);
3856	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3857		v |= RL_CFG5_WOL_UCAST;
3858	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3859		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3860	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3861		v |= RL_CFG5_WOL_LANWAKE;
3862	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3863
3864	/* Config register write done. */
3865	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3866
3867	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3868	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3869		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3870	/*
3871	 * It seems that hardware resets its link speed to 100Mbps in
3872	 * power down mode so switching to 100Mbps in driver is not
3873	 * needed.
3874	 */
3875
3876	/* Request PME if WOL is requested. */
3877	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3878	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3879	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3880		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3881	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3882}
3883
3884static void
3885re_clrwol(struct rl_softc *sc)
3886{
3887	int			pmc;
3888	uint8_t			v;
3889
3890	RL_LOCK_ASSERT(sc);
3891
3892	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3893		return;
3894
3895	/* Enable config register write. */
3896	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3897
3898	v = CSR_READ_1(sc, sc->rl_cfg3);
3899	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3900	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3901
3902	/* Config register write done. */
3903	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3904
3905	v = CSR_READ_1(sc, sc->rl_cfg5);
3906	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3907	v &= ~RL_CFG5_WOL_LANWAKE;
3908	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3909}
3910
3911static void
3912re_add_sysctls(struct rl_softc *sc)
3913{
3914	struct sysctl_ctx_list	*ctx;
3915	struct sysctl_oid_list	*children;
3916	int			error;
3917
3918	ctx = device_get_sysctl_ctx(sc->rl_dev);
3919	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3920
3921	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3922	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3923	    "Statistics Information");
3924	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3925		return;
3926
3927	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3928	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3929	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3930	/* Pull in device tunables. */
3931	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3932	error = resource_int_value(device_get_name(sc->rl_dev),
3933	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3934	if (error == 0) {
3935		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3936		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3937			device_printf(sc->rl_dev, "int_rx_mod value out of "
3938			    "range; using default: %d\n",
3939			    RL_TIMER_DEFAULT);
3940			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3941		}
3942	}
3943
3944}
3945
3946static int
3947re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3948{
3949	struct rl_softc		*sc;
3950	struct rl_stats		*stats;
3951	int			error, i, result;
3952
3953	result = -1;
3954	error = sysctl_handle_int(oidp, &result, 0, req);
3955	if (error || req->newptr == NULL)
3956		return (error);
3957
3958	if (result == 1) {
3959		sc = (struct rl_softc *)arg1;
3960		RL_LOCK(sc);
3961		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3962			RL_UNLOCK(sc);
3963			goto done;
3964		}
3965		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3966		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3967		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3968		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3969		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3970		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3971		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3972		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3973		    RL_DUMPSTATS_START));
3974		for (i = RL_TIMEOUT; i > 0; i--) {
3975			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3976			    RL_DUMPSTATS_START) == 0)
3977				break;
3978			DELAY(1000);
3979		}
3980		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3981		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3982		RL_UNLOCK(sc);
3983		if (i == 0) {
3984			device_printf(sc->rl_dev,
3985			    "DUMP statistics request timedout\n");
3986			return (ETIMEDOUT);
3987		}
3988done:
3989		stats = sc->rl_ldata.rl_stats;
3990		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3991		printf("Tx frames : %ju\n",
3992		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3993		printf("Rx frames : %ju\n",
3994		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3995		printf("Tx errors : %ju\n",
3996		    (uintmax_t)le64toh(stats->rl_tx_errs));
3997		printf("Rx errors : %u\n",
3998		    le32toh(stats->rl_rx_errs));
3999		printf("Rx missed frames : %u\n",
4000		    (uint32_t)le16toh(stats->rl_missed_pkts));
4001		printf("Rx frame alignment errs : %u\n",
4002		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
4003		printf("Tx single collisions : %u\n",
4004		    le32toh(stats->rl_tx_onecoll));
4005		printf("Tx multiple collisions : %u\n",
4006		    le32toh(stats->rl_tx_multicolls));
4007		printf("Rx unicast frames : %ju\n",
4008		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
4009		printf("Rx broadcast frames : %ju\n",
4010		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
4011		printf("Rx multicast frames : %u\n",
4012		    le32toh(stats->rl_rx_mcasts));
4013		printf("Tx aborts : %u\n",
4014		    (uint32_t)le16toh(stats->rl_tx_aborts));
4015		printf("Tx underruns : %u\n",
4016		    (uint32_t)le16toh(stats->rl_rx_underruns));
4017	}
4018
4019	return (error);
4020}
4021
4022static int
4023sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4024{
4025	int error, value;
4026
4027	if (arg1 == NULL)
4028		return (EINVAL);
4029	value = *(int *)arg1;
4030	error = sysctl_handle_int(oidp, &value, 0, req);
4031	if (error || req->newptr == NULL)
4032		return (error);
4033	if (value < low || value > high)
4034		return (EINVAL);
4035	*(int *)arg1 = value;
4036
4037	return (0);
4038}
4039
4040static int
4041sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4042{
4043
4044	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4045	    RL_TIMER_MAX));
4046}
4047