3 * Joerg Sonnenberger <joerg@bec.de>. All rights reserved.
5 * Copyright (c) 1997, 1998-2003
6 * Bill Paul <wpaul@windriver.com>. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by Bill Paul.
19 * 4. Neither the name of the author nor the names of any co-contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
27 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33 * THE POSSIBILITY OF SUCH DAMAGE.
35 * $FreeBSD: src/sys/dev/re/if_re.c,v 1.25 2004/06/09 14:34:01 naddy Exp $
36 * $DragonFly: src/sys/dev/netif/re/if_re.c,v 1.97 2008/10/21 12:31:00 sephe Exp $
40 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
42 * Written by Bill Paul <wpaul@windriver.com>
43 * Senior Networking Software Engineer
48 * This driver is designed to support RealTek's next generation of
49 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
50 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
51 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
53 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54 * with the older 8139 family, however it also supports a special
55 * C+ mode of operation that provides several new performance enhancing
56 * features. These include:
58 * o Descriptor based DMA mechanism. Each descriptor represents
59 * a single packet fragment. Data buffers may be aligned on
64 * o TCP/IP checksum offload for both RX and TX
66 * o High and normal priority transmit DMA rings
68 * o VLAN tag insertion and extraction
70 * o TCP large send (segmentation offload)
72 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73 * programming API is fairly straightforward. The RX filtering, EEPROM
74 * access and PHY access is the same as it is on the older 8139 series
77 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78 * same programming API and feature set as the 8139C+ with the following
79 * differences and additions:
85 * o GMII and TBI ports/registers for interfacing with copper
88 * o RX and TX DMA rings can have up to 1024 descriptors
89 * (the 8139C+ allows a maximum of 64)
91 * o Slight differences in register layout from the 8139C+
93 * The TX start and timer interrupt registers are at different locations
94 * on the 8169 than they are on the 8139C+. Also, the status word in the
95 * RX descriptor has a slightly different bit layout. The 8169 does not
96 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
99 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100 * (the 'S' stands for 'single-chip'). These devices have the same
101 * programming API as the older 8169, but also have some vendor-specific
102 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
105 * This driver takes advantage of the RX and TX checksum offload and
106 * VLAN tag insertion/extraction features. It also implements TX
107 * interrupt moderation using the timer interrupt registers, which
108 * significantly reduces TX interrupt load. There is also support
109 * for jumbo frames, however the 8169/8169S/8110S can not transmit
110 * jumbo frames larger than 7440, so the max MTU possible with this
111 * driver is 7422 bytes.
116 #include "opt_polling.h"
118 #include <sys/param.h>
120 #include <sys/endian.h>
121 #include <sys/kernel.h>
122 #include <sys/in_cksum.h>
123 #include <sys/interrupt.h>
124 #include <sys/malloc.h>
125 #include <sys/mbuf.h>
126 #include <sys/rman.h>
127 #include <sys/serialize.h>
128 #include <sys/socket.h>
129 #include <sys/sockio.h>
130 #include <sys/sysctl.h>
133 #include <net/ethernet.h>
135 #include <net/ifq_var.h>
136 #include <net/if_arp.h>
137 #include <net/if_dl.h>
138 #include <net/if_media.h>
139 #include <net/if_types.h>
140 #include <net/vlan/if_vlan_var.h>
141 #include <net/vlan/if_vlan_ether.h>
143 #include <netinet/ip.h>
145 #include <dev/netif/mii_layer/mii.h>
146 #include <dev/netif/mii_layer/miivar.h>
148 #include <bus/pci/pcidevs.h>
149 #include <bus/pci/pcireg.h>
150 #include <bus/pci/pcivar.h>
152 /* "device miibus" required. See GENERIC if you get errors here. */
153 #include "miibus_if.h"
155 #include <dev/netif/re/if_rereg.h>
156 #include <dev/netif/re/if_revar.h>
158 #define RE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
161 * Various supported device vendors/types and their names.
163 static const struct re_type {
168 { PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T,
169 "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
171 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139,
172 "RealTek 8139C+ 10/100BaseTX" },
174 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E,
175 "RealTek 810x PCIe 10/100baseTX" },
177 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168,
178 "RealTek 8111/8168 PCIe Gigabit Ethernet" },
180 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169,
181 "RealTek 8110/8169 Gigabit Ethernet" },
183 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC,
184 "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
186 { PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT,
187 "Corega CG-LAPCIGT Gigabit Ethernet" },
189 { PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032,
190 "Linksys EG1032 Gigabit Ethernet" },
192 { PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902,
193 "US Robotics 997902 Gigabit Ethernet" },
198 static const struct re_hwrev re_hwrevs[] = {
199 { RE_HWREV_8139CPLUS, RE_MACVER_UNKN, ETHERMTU,
200 RE_C_HWCSUM | RE_C_8139CP },
202 { RE_HWREV_8169, RE_MACVER_UNKN, RE_MTU_6K,
203 RE_C_HWCSUM | RE_C_8169 },
205 { RE_HWREV_8110S, RE_MACVER_03, RE_MTU_6K,
206 RE_C_HWCSUM | RE_C_8169 },
208 { RE_HWREV_8169S, RE_MACVER_03, RE_MTU_6K,
209 RE_C_HWCSUM | RE_C_8169 },
211 { RE_HWREV_8169SB, RE_MACVER_04, RE_MTU_6K,
212 RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
214 { RE_HWREV_8169SC1, RE_MACVER_05, RE_MTU_6K,
215 RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
217 { RE_HWREV_8169SC2, RE_MACVER_06, RE_MTU_6K,
218 RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
220 { RE_HWREV_8168B1, RE_MACVER_21, RE_MTU_6K,
221 RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT },
223 { RE_HWREV_8168B2, RE_MACVER_23, RE_MTU_6K,
224 RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
226 { RE_HWREV_8168B3, RE_MACVER_23, RE_MTU_6K,
227 RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
229 { RE_HWREV_8168C, RE_MACVER_29, RE_MTU_6K,
230 RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
231 RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
233 { RE_HWREV_8168CP, RE_MACVER_2B, RE_MTU_6K,
234 RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
235 RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
237 { RE_HWREV_8168D, RE_MACVER_2A, RE_MTU_9K,
238 RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
239 RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
241 { RE_HWREV_8100E, RE_MACVER_UNKN, ETHERMTU,
244 { RE_HWREV_8101E1, RE_MACVER_16, ETHERMTU,
247 { RE_HWREV_8101E2, RE_MACVER_16, ETHERMTU,
250 { RE_HWREV_8102E, RE_MACVER_15, ETHERMTU,
251 RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX },
253 { RE_HWREV_8102EL, RE_MACVER_15, ETHERMTU,
254 RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX },
256 { RE_HWREV_NULL, 0, 0, 0 }
259 static int re_probe(device_t);
260 static int re_attach(device_t);
261 static int re_detach(device_t);
262 static int re_suspend(device_t);
263 static int re_resume(device_t);
264 static void re_shutdown(device_t);
266 static void re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
267 static void re_dma_map_desc(void *, bus_dma_segment_t *, int,
269 static int re_allocmem(device_t);
270 static void re_freemem(device_t);
271 static void re_freebufmem(struct re_softc *, int, int);
272 static int re_encap(struct re_softc *, struct mbuf **, int *);
273 static int re_newbuf_std(struct re_softc *, int, int);
274 static int re_newbuf_jumbo(struct re_softc *, int, int);
275 static void re_setup_rxdesc(struct re_softc *, int);
276 static int re_rx_list_init(struct re_softc *);
277 static int re_tx_list_init(struct re_softc *);
278 static int re_rxeof(struct re_softc *);
279 static int re_txeof(struct re_softc *);
280 static int re_tx_collect(struct re_softc *);
281 static void re_intr(void *);
282 static void re_tick(void *);
283 static void re_tick_serialized(void *);
285 static void re_start(struct ifnet *);
286 static int re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
287 static void re_init(void *);
288 static void re_stop(struct re_softc *);
289 static void re_watchdog(struct ifnet *);
290 static int re_ifmedia_upd(struct ifnet *);
291 static void re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
293 static void re_eeprom_putbyte(struct re_softc *, int);
294 static void re_eeprom_getword(struct re_softc *, int, u_int16_t *);
295 static void re_read_eeprom(struct re_softc *, caddr_t, int, int);
296 static void re_get_eewidth(struct re_softc *);
298 static int re_gmii_readreg(device_t, int, int);
299 static int re_gmii_writereg(device_t, int, int, int);
301 static int re_miibus_readreg(device_t, int, int);
302 static int re_miibus_writereg(device_t, int, int, int);
303 static void re_miibus_statchg(device_t);
305 static void re_setmulti(struct re_softc *);
306 static void re_reset(struct re_softc *, int);
307 static void re_get_eaddr(struct re_softc *, uint8_t *);
309 static void re_setup_hw_im(struct re_softc *);
310 static void re_setup_sim_im(struct re_softc *);
311 static void re_disable_hw_im(struct re_softc *);
312 static void re_disable_sim_im(struct re_softc *);
313 static void re_config_imtype(struct re_softc *, int);
314 static void re_setup_intr(struct re_softc *, int, int);
316 static int re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *);
317 static int re_sysctl_rxtime(SYSCTL_HANDLER_ARGS);
318 static int re_sysctl_txtime(SYSCTL_HANDLER_ARGS);
319 static int re_sysctl_simtime(SYSCTL_HANDLER_ARGS);
320 static int re_sysctl_imtype(SYSCTL_HANDLER_ARGS);
322 static int re_jpool_alloc(struct re_softc *);
323 static void re_jpool_free(struct re_softc *);
324 static struct re_jbuf *re_jbuf_alloc(struct re_softc *);
325 static void re_jbuf_free(void *);
326 static void re_jbuf_ref(void *);
329 static int re_diag(struct re_softc *);
332 #ifdef DEVICE_POLLING
333 static void re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
336 static device_method_t re_methods[] = {
337 /* Device interface */
338 DEVMETHOD(device_probe, re_probe),
339 DEVMETHOD(device_attach, re_attach),
340 DEVMETHOD(device_detach, re_detach),
341 DEVMETHOD(device_suspend, re_suspend),
342 DEVMETHOD(device_resume, re_resume),
343 DEVMETHOD(device_shutdown, re_shutdown),
346 DEVMETHOD(bus_print_child, bus_generic_print_child),
347 DEVMETHOD(bus_driver_added, bus_generic_driver_added),
350 DEVMETHOD(miibus_readreg, re_miibus_readreg),
351 DEVMETHOD(miibus_writereg, re_miibus_writereg),
352 DEVMETHOD(miibus_statchg, re_miibus_statchg),
357 static driver_t re_driver = {
360 sizeof(struct re_softc)
363 static devclass_t re_devclass;
365 DECLARE_DUMMY_MODULE(if_re);
366 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
367 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
368 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
370 static int re_rx_desc_count = RE_RX_DESC_CNT_DEF;
371 static int re_tx_desc_count = RE_TX_DESC_CNT_DEF;
373 TUNABLE_INT("hw.re.rx_desc_count", &re_rx_desc_count);
374 TUNABLE_INT("hw.re.tx_desc_count", &re_tx_desc_count);
377 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
380 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
383 re_free_rxchain(struct re_softc *sc)
385 if (sc->re_head != NULL) {
386 m_freem(sc->re_head);
387 sc->re_head = sc->re_tail = NULL;
392 * Send a read command and address to the EEPROM, check for ACK.
395 re_eeprom_putbyte(struct re_softc *sc, int addr)
399 d = addr | (RE_9346_READ << sc->re_eewidth);
402 * Feed in each bit and strobe the clock.
404 for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
406 EE_SET(RE_EE_DATAIN);
408 EE_CLR(RE_EE_DATAIN);
418 * Read a word of data stored in the EEPROM at address 'addr.'
421 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
427 * Send address of word we want to read.
429 re_eeprom_putbyte(sc, addr);
432 * Start reading bits from EEPROM.
434 for (i = 0x8000; i != 0; i >>= 1) {
437 if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
447 * Read a sequence of words from the EEPROM.
450 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
453 uint16_t word = 0, *ptr;
455 CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
458 for (i = 0; i < cnt; i++) {
459 CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
460 re_eeprom_getword(sc, off + i, &word);
461 CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
462 ptr = (uint16_t *)(dest + (i * 2));
466 CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
470 re_get_eewidth(struct re_softc *sc)
475 re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
476 if (re_did != 0x8129)
481 re_gmii_readreg(device_t dev, int phy, int reg)
483 struct re_softc *sc = device_get_softc(dev);
490 /* Let the rgephy driver read the GMEDIASTAT register */
492 if (reg == RE_GMEDIASTAT)
493 return(CSR_READ_1(sc, RE_GMEDIASTAT));
495 CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
498 for (i = 0; i < RE_TIMEOUT; i++) {
499 rval = CSR_READ_4(sc, RE_PHYAR);
500 if (rval & RE_PHYAR_BUSY)
505 if (i == RE_TIMEOUT) {
506 device_printf(dev, "PHY read failed\n");
510 return(rval & RE_PHYAR_PHYDATA);
514 re_gmii_writereg(device_t dev, int phy, int reg, int data)
516 struct re_softc *sc = device_get_softc(dev);
520 CSR_WRITE_4(sc, RE_PHYAR,
521 (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
524 for (i = 0; i < RE_TIMEOUT; i++) {
525 rval = CSR_READ_4(sc, RE_PHYAR);
526 if ((rval & RE_PHYAR_BUSY) == 0)
532 device_printf(dev, "PHY write failed\n");
538 re_miibus_readreg(device_t dev, int phy, int reg)
540 struct re_softc *sc = device_get_softc(dev);
542 uint16_t re8139_reg = 0;
544 if (!RE_IS_8139CP(sc)) {
545 rval = re_gmii_readreg(dev, phy, reg);
549 /* Pretend the internal PHY is only at address 0 */
555 re8139_reg = RE_BMCR;
558 re8139_reg = RE_BMSR;
561 re8139_reg = RE_ANAR;
564 re8139_reg = RE_ANER;
567 re8139_reg = RE_LPAR;
573 * Allow the rlphy driver to read the media status
574 * register. If we have a link partner which does not
575 * support NWAY, this is the register which will tell
576 * us the results of parallel detection.
579 return(CSR_READ_1(sc, RE_MEDIASTAT));
581 device_printf(dev, "bad phy register\n");
584 rval = CSR_READ_2(sc, re8139_reg);
585 if (re8139_reg == RE_BMCR) {
586 /* 8139C+ has different bit layout. */
587 rval &= ~(BMCR_LOOP | BMCR_ISO);
593 re_miibus_writereg(device_t dev, int phy, int reg, int data)
595 struct re_softc *sc= device_get_softc(dev);
596 u_int16_t re8139_reg = 0;
598 if (!RE_IS_8139CP(sc))
599 return(re_gmii_writereg(dev, phy, reg, data));
601 /* Pretend the internal PHY is only at address 0 */
607 re8139_reg = RE_BMCR;
608 /* 8139C+ has different bit layout. */
609 data &= ~(BMCR_LOOP | BMCR_ISO);
612 re8139_reg = RE_BMSR;
615 re8139_reg = RE_ANAR;
618 re8139_reg = RE_ANER;
621 re8139_reg = RE_LPAR;
627 device_printf(dev, "bad phy register\n");
630 CSR_WRITE_2(sc, re8139_reg, data);
635 re_miibus_statchg(device_t dev)
640 * Program the 64-bit multicast hash filter.
643 re_setmulti(struct re_softc *sc)
645 struct ifnet *ifp = &sc->arpcom.ac_if;
647 uint32_t hashes[2] = { 0, 0 };
648 struct ifmultiaddr *ifma;
652 rxfilt = CSR_READ_4(sc, RE_RXCFG);
654 /* Set the individual bit to receive frames for this host only. */
655 rxfilt |= RE_RXCFG_RX_INDIV;
656 /* Set capture broadcast bit to capture broadcast frames. */
657 rxfilt |= RE_RXCFG_RX_BROAD;
659 rxfilt &= ~(RE_RXCFG_RX_ALLPHYS | RE_RXCFG_RX_MULTI);
660 if ((ifp->if_flags & IFF_ALLMULTI) || (ifp->if_flags & IFF_PROMISC)) {
661 rxfilt |= RE_RXCFG_RX_MULTI;
663 /* If we want promiscuous mode, set the allframes bit. */
664 if (ifp->if_flags & IFF_PROMISC)
665 rxfilt |= RE_RXCFG_RX_ALLPHYS;
667 CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
668 CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
669 CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
673 /* first, zot all the existing hash bits */
674 CSR_WRITE_4(sc, RE_MAR0, 0);
675 CSR_WRITE_4(sc, RE_MAR4, 0);
677 /* now program new ones */
678 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
679 if (ifma->ifma_addr->sa_family != AF_LINK)
681 h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
682 ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
684 hashes[0] |= (1 << h);
686 hashes[1] |= (1 << (h - 32));
691 rxfilt |= RE_RXCFG_RX_MULTI;
693 rxfilt &= ~RE_RXCFG_RX_MULTI;
695 CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
698 * For some unfathomable reason, RealTek decided to reverse
699 * the order of the multicast hash registers in the PCI Express
700 * parts. This means we have to write the hash pattern in reverse
701 * order for those devices.
703 if (sc->re_caps & RE_C_PCIE) {
704 CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
705 CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
707 CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
708 CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
713 re_reset(struct re_softc *sc, int running)
717 if ((sc->re_caps & RE_C_STOP_RXTX) && running) {
718 CSR_WRITE_1(sc, RE_COMMAND,
719 RE_CMD_STOPREQ | RE_CMD_TX_ENB | RE_CMD_RX_ENB);
723 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
725 for (i = 0; i < RE_TIMEOUT; i++) {
727 if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
731 if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
736 * The following routine is designed to test for a defect on some
737 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
738 * lines connected to the bus, however for a 32-bit only card, they
739 * should be pulled high. The result of this defect is that the
740 * NIC will not work right if you plug it into a 64-bit slot: DMA
741 * operations will be done with 64-bit transfers, which will fail
742 * because the 64-bit data lines aren't connected.
744 * There's no way to work around this (short of talking a soldering
745 * iron to the board), however we can detect it. The method we use
746 * here is to put the NIC into digital loopback mode, set the receiver
747 * to promiscuous mode, and then try to send a frame. We then compare
748 * the frame data we sent to what was received. If the data matches,
749 * then the NIC is working correctly, otherwise we know the user has
750 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
751 * slot. In the latter case, there's no way the NIC can work correctly,
752 * so we print out a message on the console and abort the device attach.
756 re_diag(struct re_softc *sc)
758 struct ifnet *ifp = &sc->arpcom.ac_if;
760 struct ether_header *eh;
761 struct re_desc *cur_rx;
764 int total_len, i, error = 0, phyaddr;
765 uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
766 uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
768 /* Allocate a single mbuf */
770 MGETHDR(m0, MB_DONTWAIT, MT_DATA);
775 * Initialize the NIC in test mode. This sets the chip up
776 * so that it can send and receive frames, but performs the
777 * following special functions:
778 * - Puts receiver in promiscuous mode
779 * - Enables digital loopback mode
780 * - Leaves interrupts turned off
783 ifp->if_flags |= IFF_PROMISC;
784 sc->re_flags |= RE_F_TESTMODE;
786 sc->re_flags |= RE_F_LINKED;
787 if (!RE_IS_8139CP(sc))
792 re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
793 for (i = 0; i < RE_TIMEOUT; i++) {
794 status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
795 if (!(status & BMCR_RESET))
799 re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
800 CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
804 /* Put some data in the mbuf */
806 eh = mtod(m0, struct ether_header *);
807 bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
808 bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
809 eh->ether_type = htons(ETHERTYPE_IP);
810 m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
813 * Queue the packet, start transmission.
814 * Note: ifq_handoff() ultimately calls re_start() for us.
817 CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
818 error = ifq_handoff(ifp, m0, NULL);
825 /* Wait for it to propagate through the chip */
828 for (i = 0; i < RE_TIMEOUT; i++) {
829 status = CSR_READ_2(sc, RE_ISR);
830 CSR_WRITE_2(sc, RE_ISR, status);
831 if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
832 (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
837 if (i == RE_TIMEOUT) {
838 if_printf(ifp, "diagnostic failed to receive packet "
839 "in loopback mode\n");
845 * The packet should have been dumped into the first
846 * entry in the RX DMA ring. Grab it from there.
849 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
850 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
851 bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
852 BUS_DMASYNC_POSTWRITE);
853 bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
855 m0 = sc->re_ldata.re_rx_mbuf[0];
856 sc->re_ldata.re_rx_mbuf[0] = NULL;
857 eh = mtod(m0, struct ether_header *);
859 cur_rx = &sc->re_ldata.re_rx_list[0];
860 total_len = RE_RXBYTES(cur_rx);
861 rxstat = le32toh(cur_rx->re_cmdstat);
863 if (total_len != ETHER_MIN_LEN) {
864 if_printf(ifp, "diagnostic failed, received short packet\n");
869 /* Test that the received packet data matches what we sent. */
871 if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
872 bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
873 be16toh(eh->ether_type) != ETHERTYPE_IP) {
874 if_printf(ifp, "WARNING, DMA FAILURE!\n");
875 if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
876 dst, ":", src, ":", ETHERTYPE_IP);
877 if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
878 eh->ether_dhost, ":", eh->ether_shost, ":",
879 ntohs(eh->ether_type));
880 if_printf(ifp, "You may have a defective 32-bit NIC plugged "
881 "into a 64-bit PCI slot.\n");
882 if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
883 "for proper operation.\n");
884 if_printf(ifp, "Read the re(4) man page for more details.\n");
889 /* Turn interface off, release resources */
891 sc->re_flags &= ~(RE_F_LINKED | RE_F_TESTMODE);
892 ifp->if_flags &= ~IFF_PROMISC;
902 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
903 * IDs against our list and return a device name if we find a match.
906 re_probe(device_t dev)
908 const struct re_type *t;
909 const struct re_hwrev *hw_rev;
912 uint32_t hwrev, macmode, txcfg;
913 uint16_t vendor, product;
915 vendor = pci_get_vendor(dev);
916 product = pci_get_device(dev);
919 * Only attach to rev.3 of the Linksys EG1032 adapter.
920 * Rev.2 is supported by sk(4).
922 if (vendor == PCI_VENDOR_LINKSYS &&
923 product == PCI_PRODUCT_LINKSYS_EG1032 &&
924 pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
927 for (t = re_devs; t->re_name != NULL; t++) {
928 if (product == t->re_did && vendor == t->re_vid)
933 * Check if we found a RealTek device.
935 if (t->re_name == NULL)
939 * Temporarily map the I/O space so we can read the chip ID register.
941 sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
943 sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
945 if (sc->re_res == NULL) {
946 device_printf(dev, "couldn't map ports/memory\n");
951 sc->re_btag = rman_get_bustag(sc->re_res);
952 sc->re_bhandle = rman_get_bushandle(sc->re_res);
954 txcfg = CSR_READ_4(sc, RE_TXCFG);
955 hwrev = txcfg & RE_TXCFG_HWREV;
956 macmode = txcfg & RE_TXCFG_MACMODE;
957 bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
961 * and continue matching for the specific chip...
963 for (hw_rev = re_hwrevs; hw_rev->re_hwrev != RE_HWREV_NULL; hw_rev++) {
964 if (hw_rev->re_hwrev == hwrev) {
965 sc = device_get_softc(dev);
967 sc->re_hwrev = hw_rev->re_hwrev;
968 sc->re_macver = hw_rev->re_macver;
969 sc->re_caps = hw_rev->re_caps;
970 sc->re_maxmtu = hw_rev->re_maxmtu;
971 sc->re_swcsum_lim = RE_SWCSUM_UNLIMITED;
974 * Apply chip property fixup
976 switch (sc->re_hwrev) {
978 sc->re_swcsum_lim = RE_SWCSUM_LIM_8169;
980 case RE_HWREV_8101E1:
981 case RE_HWREV_8101E2:
983 sc->re_macver = RE_MACVER_11;
984 else if (macmode == 0x200000)
985 sc->re_macver = RE_MACVER_12;
988 case RE_HWREV_8102EL:
990 sc->re_macver = RE_MACVER_13;
991 else if (macmode == 0x100000)
992 sc->re_macver = RE_MACVER_14;
994 case RE_HWREV_8168B2:
995 case RE_HWREV_8168B3:
997 sc->re_macver = RE_MACVER_22;
1001 sc->re_macver = RE_MACVER_24;
1002 else if (macmode == 0x200000)
1003 sc->re_macver = RE_MACVER_25;
1004 else if (macmode == 0x300000)
1005 sc->re_macver = RE_MACVER_27;
1007 case RE_HWREV_8168CP:
1009 sc->re_macver = RE_MACVER_26;
1010 else if (macmode == 0x100000)
1011 sc->re_macver = RE_MACVER_28;
1014 if (pci_get_pciecap_ptr(dev) != 0)
1015 sc->re_caps |= RE_C_PCIE;
1017 device_set_desc(dev, t->re_name);
1021 device_printf(dev, "unknown hwrev 0x%08x, macmode 0x%08x\n",
1027 re_dma_map_desc(void *xarg, bus_dma_segment_t *segs, int nsegs,
1028 bus_size_t mapsize, int error)
1030 struct re_dmaload_arg *arg = xarg;
1036 if (nsegs > arg->re_nsegs) {
1041 arg->re_nsegs = nsegs;
1042 for (i = 0; i < nsegs; ++i)
1043 arg->re_segs[i] = segs[i];
1047 * Map a single buffer address.
1051 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1058 KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1060 *addr = segs->ds_addr;
1064 re_allocmem(device_t dev)
1066 struct re_softc *sc = device_get_softc(dev);
1070 * Allocate list data
1072 sc->re_ldata.re_tx_mbuf =
1073 kmalloc(sc->re_tx_desc_cnt * sizeof(struct mbuf *),
1074 M_DEVBUF, M_ZERO | M_WAITOK);
1076 sc->re_ldata.re_rx_mbuf =
1077 kmalloc(sc->re_rx_desc_cnt * sizeof(struct mbuf *),
1078 M_DEVBUF, M_ZERO | M_WAITOK);
1080 sc->re_ldata.re_rx_paddr =
1081 kmalloc(sc->re_rx_desc_cnt * sizeof(bus_addr_t),
1082 M_DEVBUF, M_ZERO | M_WAITOK);
1084 sc->re_ldata.re_tx_dmamap =
1085 kmalloc(sc->re_tx_desc_cnt * sizeof(bus_dmamap_t),
1086 M_DEVBUF, M_ZERO | M_WAITOK);
1088 sc->re_ldata.re_rx_dmamap =
1089 kmalloc(sc->re_rx_desc_cnt * sizeof(bus_dmamap_t),
1090 M_DEVBUF, M_ZERO | M_WAITOK);
1093 * Allocate the parent bus DMA tag appropriate for PCI.
1095 error = bus_dma_tag_create(NULL, /* parent */
1096 1, 0, /* alignment, boundary */
1097 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1098 BUS_SPACE_MAXADDR, /* highaddr */
1099 NULL, NULL, /* filter, filterarg */
1100 MAXBSIZE, RE_MAXSEGS, /* maxsize, nsegments */
1101 BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1102 BUS_DMA_ALLOCNOW, /* flags */
1103 &sc->re_parent_tag);
1105 device_printf(dev, "could not allocate parent dma tag\n");
1109 /* Allocate tag for TX descriptor list. */
1110 error = bus_dma_tag_create(sc->re_parent_tag,
1112 BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1114 RE_TX_LIST_SZ(sc), 1, RE_TX_LIST_SZ(sc),
1116 &sc->re_ldata.re_tx_list_tag);
1118 device_printf(dev, "could not allocate TX ring dma tag\n");
1122 /* Allocate DMA'able memory for the TX ring */
1123 error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
1124 (void **)&sc->re_ldata.re_tx_list,
1125 BUS_DMA_WAITOK | BUS_DMA_ZERO,
1126 &sc->re_ldata.re_tx_list_map);
1128 device_printf(dev, "could not allocate TX ring\n");
1129 bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1130 sc->re_ldata.re_tx_list_tag = NULL;
1134 /* Load the map for the TX ring. */
1135 error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
1136 sc->re_ldata.re_tx_list_map,
1137 sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc),
1138 re_dma_map_addr, &sc->re_ldata.re_tx_list_addr,
1141 device_printf(dev, "could not get address of TX ring\n");
1142 bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1143 sc->re_ldata.re_tx_list,
1144 sc->re_ldata.re_tx_list_map);
1145 bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1146 sc->re_ldata.re_tx_list_tag = NULL;
1150 /* Allocate tag for RX descriptor list. */
1151 error = bus_dma_tag_create(sc->re_parent_tag,
1153 BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1155 RE_RX_LIST_SZ(sc), 1, RE_RX_LIST_SZ(sc),
1157 &sc->re_ldata.re_rx_list_tag);
1159 device_printf(dev, "could not allocate RX ring dma tag\n");
1163 /* Allocate DMA'able memory for the RX ring */
1164 error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1165 (void **)&sc->re_ldata.re_rx_list,
1166 BUS_DMA_WAITOK | BUS_DMA_ZERO,
1167 &sc->re_ldata.re_rx_list_map);
1169 device_printf(dev, "could not allocate RX ring\n");
1170 bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1171 sc->re_ldata.re_rx_list_tag = NULL;
1175 /* Load the map for the RX ring. */
1176 error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1177 sc->re_ldata.re_rx_list_map,
1178 sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc),
1179 re_dma_map_addr, &sc->re_ldata.re_rx_list_addr,
1182 device_printf(dev, "could not get address of RX ring\n");
1183 bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1184 sc->re_ldata.re_rx_list,
1185 sc->re_ldata.re_rx_list_map);
1186 bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1187 sc->re_ldata.re_rx_list_tag = NULL;
1191 /* Allocate map for RX/TX mbufs. */
1192 error = bus_dma_tag_create(sc->re_parent_tag,
1194 BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1196 RE_FRAMELEN_MAX, RE_MAXSEGS, MCLBYTES,
1198 &sc->re_ldata.re_mtag);
1200 device_printf(dev, "could not allocate buf dma tag\n");
1204 /* Create spare DMA map for RX */
1205 error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1206 &sc->re_ldata.re_rx_spare);
1208 device_printf(dev, "can't create spare DMA map for RX\n");
1209 bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1210 sc->re_ldata.re_mtag = NULL;
1214 /* Create DMA maps for TX buffers */
1215 for (i = 0; i < sc->re_tx_desc_cnt; i++) {
1216 error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1217 &sc->re_ldata.re_tx_dmamap[i]);
1219 device_printf(dev, "can't create DMA map for TX buf\n");
1220 re_freebufmem(sc, i, 0);
1225 /* Create DMA maps for RX buffers */
1226 for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1227 error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1228 &sc->re_ldata.re_rx_dmamap[i]);
1230 device_printf(dev, "can't create DMA map for RX buf\n");
1231 re_freebufmem(sc, sc->re_tx_desc_cnt, i);
1236 /* Create jumbo buffer pool for RX if required */
1237 if (sc->re_caps & RE_C_CONTIGRX) {
1238 error = re_jpool_alloc(sc);
1241 /* Disable jumbo frame support */
1242 sc->re_maxmtu = ETHERMTU;
1249 re_freebufmem(struct re_softc *sc, int tx_cnt, int rx_cnt)
1253 /* Destroy all the RX and TX buffer maps */
1254 if (sc->re_ldata.re_mtag) {
1255 for (i = 0; i < tx_cnt; i++) {
1256 bus_dmamap_destroy(sc->re_ldata.re_mtag,
1257 sc->re_ldata.re_tx_dmamap[i]);
1259 for (i = 0; i < rx_cnt; i++) {
1260 bus_dmamap_destroy(sc->re_ldata.re_mtag,
1261 sc->re_ldata.re_rx_dmamap[i]);
1263 bus_dmamap_destroy(sc->re_ldata.re_mtag,
1264 sc->re_ldata.re_rx_spare);
1265 bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1266 sc->re_ldata.re_mtag = NULL;
1271 re_freemem(device_t dev)
1273 struct re_softc *sc = device_get_softc(dev);
1275 /* Unload and free the RX DMA ring memory and map */
1276 if (sc->re_ldata.re_rx_list_tag) {
1277 bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1278 sc->re_ldata.re_rx_list_map);
1279 bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1280 sc->re_ldata.re_rx_list,
1281 sc->re_ldata.re_rx_list_map);
1282 bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1285 /* Unload and free the TX DMA ring memory and map */
1286 if (sc->re_ldata.re_tx_list_tag) {
1287 bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1288 sc->re_ldata.re_tx_list_map);
1289 bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1290 sc->re_ldata.re_tx_list,
1291 sc->re_ldata.re_tx_list_map);
1292 bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1295 /* Free RX/TX buf DMA stuffs */
1296 re_freebufmem(sc, sc->re_tx_desc_cnt, sc->re_rx_desc_cnt);
1298 /* Unload and free the stats buffer and map */
1299 if (sc->re_ldata.re_stag) {
1300 bus_dmamap_unload(sc->re_ldata.re_stag,
1301 sc->re_ldata.re_rx_list_map);
1302 bus_dmamem_free(sc->re_ldata.re_stag,
1303 sc->re_ldata.re_stats,
1304 sc->re_ldata.re_smap);
1305 bus_dma_tag_destroy(sc->re_ldata.re_stag);
1308 if (sc->re_caps & RE_C_CONTIGRX)
1311 if (sc->re_parent_tag)
1312 bus_dma_tag_destroy(sc->re_parent_tag);
1314 if (sc->re_ldata.re_tx_mbuf != NULL)
1315 kfree(sc->re_ldata.re_tx_mbuf, M_DEVBUF);
1316 if (sc->re_ldata.re_rx_mbuf != NULL)
1317 kfree(sc->re_ldata.re_rx_mbuf, M_DEVBUF);
1318 if (sc->re_ldata.re_rx_paddr != NULL)
1319 kfree(sc->re_ldata.re_rx_paddr, M_DEVBUF);
1320 if (sc->re_ldata.re_tx_dmamap != NULL)
1321 kfree(sc->re_ldata.re_tx_dmamap, M_DEVBUF);
1322 if (sc->re_ldata.re_rx_dmamap != NULL)
1323 kfree(sc->re_ldata.re_rx_dmamap, M_DEVBUF);
1327 * Attach the interface. Allocate softc structures, do ifmedia
1328 * setup and ethernet/BPF attach.
1331 re_attach(device_t dev)
1333 struct re_softc *sc = device_get_softc(dev);
1335 uint8_t eaddr[ETHER_ADDR_LEN];
1336 int error = 0, rid, qlen;
1338 callout_init(&sc->re_timer);
1341 if (RE_IS_8139CP(sc)) {
1342 sc->re_rx_desc_cnt = RE_RX_DESC_CNT_8139CP;
1343 sc->re_tx_desc_cnt = RE_TX_DESC_CNT_8139CP;
1345 sc->re_rx_desc_cnt = re_rx_desc_count;
1346 if (sc->re_rx_desc_cnt > RE_RX_DESC_CNT_MAX)
1347 sc->re_rx_desc_cnt = RE_RX_DESC_CNT_MAX;
1349 sc->re_tx_desc_cnt = re_tx_desc_count;
1350 if (sc->re_tx_desc_cnt > RE_TX_DESC_CNT_MAX)
1351 sc->re_tx_desc_cnt = RE_TX_DESC_CNT_MAX;
1354 qlen = RE_IFQ_MAXLEN;
1355 if (sc->re_tx_desc_cnt > qlen)
1356 qlen = sc->re_tx_desc_cnt;
1358 sc->re_rxbuf_size = MCLBYTES;
1359 sc->re_newbuf = re_newbuf_std;
1361 sc->re_tx_time = 5; /* 125us */
1362 sc->re_rx_time = 2; /* 50us */
1363 if (sc->re_caps & RE_C_PCIE)
1364 sc->re_sim_time = 75; /* 75us */
1366 sc->re_sim_time = 125; /* 125us */
1367 sc->re_imtype = RE_IMTYPE_SIM; /* simulated interrupt moderation */
1368 re_config_imtype(sc, sc->re_imtype);
1370 sysctl_ctx_init(&sc->re_sysctl_ctx);
1371 sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1372 SYSCTL_STATIC_CHILDREN(_hw),
1374 device_get_nameunit(dev),
1376 if (sc->re_sysctl_tree == NULL) {
1377 device_printf(dev, "can't add sysctl node\n");
1381 SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1382 SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1383 "rx_desc_count", CTLFLAG_RD, &sc->re_rx_desc_cnt,
1384 0, "RX desc count");
1385 SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1386 SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1387 "tx_desc_count", CTLFLAG_RD, &sc->re_tx_desc_cnt,
1388 0, "TX desc count");
1389 SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1390 SYSCTL_CHILDREN(sc->re_sysctl_tree),
1391 OID_AUTO, "sim_time",
1392 CTLTYPE_INT | CTLFLAG_RW,
1393 sc, 0, re_sysctl_simtime, "I",
1394 "Simulated interrupt moderation time (usec).");
1395 SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1396 SYSCTL_CHILDREN(sc->re_sysctl_tree),
1398 CTLTYPE_INT | CTLFLAG_RW,
1399 sc, 0, re_sysctl_imtype, "I",
1400 "Interrupt moderation type -- "
1401 "0:disable, 1:simulated, "
1402 "2:hardware(if supported)");
1403 if (sc->re_caps & RE_C_HWIM) {
1404 SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1405 SYSCTL_CHILDREN(sc->re_sysctl_tree),
1406 OID_AUTO, "hw_rxtime",
1407 CTLTYPE_INT | CTLFLAG_RW,
1408 sc, 0, re_sysctl_rxtime, "I",
1409 "Hardware interrupt moderation time "
1411 SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1412 SYSCTL_CHILDREN(sc->re_sysctl_tree),
1413 OID_AUTO, "hw_txtime",
1414 CTLTYPE_INT | CTLFLAG_RW,
1415 sc, 0, re_sysctl_txtime, "I",
1416 "Hardware interrupt moderation time "
1420 #ifndef BURN_BRIDGES
1422 * Handle power management nonsense.
1425 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1426 uint32_t membase, irq;
1428 /* Save important PCI config data. */
1429 membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1430 irq = pci_read_config(dev, PCIR_INTLINE, 4);
1432 /* Reset the power state. */
1433 device_printf(dev, "chip is in D%d power mode "
1434 "-- setting to D0\n", pci_get_powerstate(dev));
1436 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1438 /* Restore PCI config data. */
1439 pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1440 pci_write_config(dev, PCIR_INTLINE, irq, 4);
1444 * Map control/status registers.
1446 pci_enable_busmaster(dev);
1449 sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1452 if (sc->re_res == NULL) {
1453 device_printf(dev, "couldn't map ports\n");
1458 sc->re_btag = rman_get_bustag(sc->re_res);
1459 sc->re_bhandle = rman_get_bushandle(sc->re_res);
1461 /* Allocate interrupt */
1463 sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1464 RF_SHAREABLE | RF_ACTIVE);
1466 if (sc->re_irq == NULL) {
1467 device_printf(dev, "couldn't map interrupt\n");
1472 /* Reset the adapter. */
1475 if (RE_IS_8139CP(sc)) {
1476 sc->re_bus_speed = 33; /* XXX */
1477 } else if (sc->re_caps & RE_C_PCIE) {
1478 sc->re_bus_speed = 125;
1482 cfg2 = CSR_READ_1(sc, RE_CFG2);
1483 switch (cfg2 & RE_CFG2_PCICLK_MASK) {
1484 case RE_CFG2_PCICLK_33MHZ:
1485 sc->re_bus_speed = 33;
1487 case RE_CFG2_PCICLK_66MHZ:
1488 sc->re_bus_speed = 66;
1491 device_printf(dev, "unknown bus speed, assume 33MHz\n");
1492 sc->re_bus_speed = 33;
1495 if (cfg2 & RE_CFG2_PCI64)
1496 sc->re_caps |= RE_C_PCI64;
1498 device_printf(dev, "Hardware rev. 0x%08x; MAC ver. 0x%02x; "
1500 sc->re_hwrev, sc->re_macver,
1501 (sc->re_caps & RE_C_PCIE) ?
1502 "-E" : ((sc->re_caps & RE_C_PCI64) ? "64" : "32"),
1507 * DO NOT try to adjust config1 and config5 which was spotted in
1508 * Realtek's Linux drivers. It will _permanently_ damage certain
1509 * cards EEPROM, e.g. one of my 8168B (0x38000000) card ...
1512 re_get_eaddr(sc, eaddr);
1514 if (!RE_IS_8139CP(sc)) {
1515 /* Set RX length mask */
1516 sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1517 sc->re_txstart = RE_GTXSTART;
1519 /* Set RX length mask */
1520 sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1521 sc->re_txstart = RE_TXSTART;
1524 /* Allocate DMA stuffs */
1525 error = re_allocmem(dev);
1530 * Apply some magic PCI settings from Realtek ...
1532 if (RE_IS_8169(sc)) {
1533 CSR_WRITE_1(sc, 0x82, 1);
1534 pci_write_config(dev, PCIR_CACHELNSZ, 0x8, 1);
1536 pci_write_config(dev, PCIR_LATTIMER, 0x40, 1);
1538 if (sc->re_caps & RE_C_MAC2) {
1540 * Following part is extracted from Realtek BSD driver v176.
1541 * However, this does _not_ make much/any sense:
1542 * 8168C's PCI Express device control is located at 0x78,
1543 * so the reading from 0x79 (higher part of 0x78) and setting
1544 * the 4~6bits intend to enlarge the "max read request size"
1545 * (we will do it). The content of the rest part of this
1546 * register is not meaningful to other PCI registers, so
1547 * writing the value to 0x54 could be completely wrong.
1548 * 0x80 is the lower part of PCI Express device status, non-
1549 * reserved bits are RW1C, writing 0 to them will not have
1550 * any effect at all.
1555 val = pci_read_config(dev, 0x79, 1);
1556 val = (val & ~0x70) | 0x50;
1557 pci_write_config(dev, 0x54, val, 1);
1558 pci_write_config(dev, 0x80, 0, 1);
1563 * Apply some PHY fixup from Realtek ...
1565 if (sc->re_hwrev == RE_HWREV_8110S) {
1566 CSR_WRITE_1(sc, 0x82, 1);
1567 re_miibus_writereg(dev, 1, 0xb, 0);
1569 if (sc->re_caps & RE_C_PHYPMGT) {
1571 re_miibus_writereg(dev, 1, 0x1f, 0);
1572 re_miibus_writereg(dev, 1, 0xe, 0);
1576 if (mii_phy_probe(dev, &sc->re_miibus,
1577 re_ifmedia_upd, re_ifmedia_sts)) {
1578 device_printf(dev, "MII without any phy!\n");
1583 ifp = &sc->arpcom.ac_if;
1585 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1586 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1587 ifp->if_ioctl = re_ioctl;
1588 ifp->if_start = re_start;
1589 #ifdef DEVICE_POLLING
1590 ifp->if_poll = re_poll;
1592 ifp->if_watchdog = re_watchdog;
1593 ifp->if_init = re_init;
1594 if (!RE_IS_8139CP(sc)) /* XXX */
1595 ifp->if_baudrate = 1000000000;
1597 ifp->if_baudrate = 100000000;
1598 ifq_set_maxlen(&ifp->if_snd, qlen);
1599 ifq_set_ready(&ifp->if_snd);
1601 ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1602 if (sc->re_caps & RE_C_HWCSUM)
1603 ifp->if_capabilities |= IFCAP_HWCSUM;
1605 ifp->if_capenable = ifp->if_capabilities;
1606 if (ifp->if_capabilities & IFCAP_HWCSUM)
1607 ifp->if_hwassist = RE_CSUM_FEATURES;
1609 ifp->if_hwassist = 0;
1612 * Call MI attach routine.
1614 ether_ifattach(ifp, eaddr, NULL);
1618 * Perform hardware diagnostic on the original RTL8169.
1619 * Some 32-bit cards were incorrectly wired and would
1620 * malfunction if plugged into a 64-bit slot.
1622 if (sc->re_hwrev == RE_HWREV_8169) {
1623 lwkt_serialize_enter(ifp->if_serializer);
1624 error = re_diag(sc);
1625 lwkt_serialize_exit(ifp->if_serializer);
1628 device_printf(dev, "hardware diagnostic failure\n");
1629 ether_ifdetach(ifp);
1633 #endif /* RE_DIAG */
1635 /* Hook interrupt last to avoid having to lock softc */
1636 error = bus_setup_intr(dev, sc->re_irq, INTR_MPSAFE, re_intr, sc,
1637 &sc->re_intrhand, ifp->if_serializer);
1640 device_printf(dev, "couldn't set up irq\n");
1641 ether_ifdetach(ifp);
1645 ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1646 KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1656 * Shutdown hardware and free up resources. This can be called any
1657 * time after the mutex has been initialized. It is called in both
1658 * the error case in attach and the normal detach case so it needs
1659 * to be careful about only freeing resources that have actually been
1663 re_detach(device_t dev)
1665 struct re_softc *sc = device_get_softc(dev);
1666 struct ifnet *ifp = &sc->arpcom.ac_if;
1668 /* These should only be active if attach succeeded */
1669 if (device_is_attached(dev)) {
1670 lwkt_serialize_enter(ifp->if_serializer);
1672 bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1673 lwkt_serialize_exit(ifp->if_serializer);
1675 ether_ifdetach(ifp);
1678 device_delete_child(dev, sc->re_miibus);
1679 bus_generic_detach(dev);
1681 if (sc->re_sysctl_tree != NULL)
1682 sysctl_ctx_free(&sc->re_sysctl_ctx);
1685 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1687 bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1691 /* Free DMA stuffs */
1698 re_setup_rxdesc(struct re_softc *sc, int idx)
1704 paddr = sc->re_ldata.re_rx_paddr[idx];
1705 d = &sc->re_ldata.re_rx_list[idx];
1707 d->re_bufaddr_lo = htole32(RE_ADDR_LO(paddr));
1708 d->re_bufaddr_hi = htole32(RE_ADDR_HI(paddr));
1710 cmdstat = sc->re_rxbuf_size | RE_RDESC_CMD_OWN;
1711 if (idx == (sc->re_rx_desc_cnt - 1))
1712 cmdstat |= RE_RDESC_CMD_EOR;
1713 d->re_cmdstat = htole32(cmdstat);
1717 re_newbuf_std(struct re_softc *sc, int idx, int init)
1719 struct re_dmaload_arg arg;
1720 bus_dma_segment_t seg;
1725 m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
1730 if_printf(&sc->arpcom.ac_if, "m_getcl failed\n");
1736 m->m_len = m->m_pkthdr.len = MCLBYTES;
1740 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1741 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1746 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1747 sc->re_ldata.re_rx_spare, m,
1748 re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1749 if (error || arg.re_nsegs == 0) {
1751 if_printf(&sc->arpcom.ac_if, "too many segments?!\n");
1752 bus_dmamap_unload(sc->re_ldata.re_mtag,
1753 sc->re_ldata.re_rx_spare);
1759 if_printf(&sc->arpcom.ac_if, "can't load RX mbuf\n");
1767 bus_dmamap_sync(sc->re_ldata.re_mtag,
1768 sc->re_ldata.re_rx_dmamap[idx],
1769 BUS_DMASYNC_POSTREAD);
1770 bus_dmamap_unload(sc->re_ldata.re_mtag,
1771 sc->re_ldata.re_rx_dmamap[idx]);
1773 sc->re_ldata.re_rx_mbuf[idx] = m;
1774 sc->re_ldata.re_rx_paddr[idx] = seg.ds_addr;
1776 map = sc->re_ldata.re_rx_dmamap[idx];
1777 sc->re_ldata.re_rx_dmamap[idx] = sc->re_ldata.re_rx_spare;
1778 sc->re_ldata.re_rx_spare = map;
1780 re_setup_rxdesc(sc, idx);
1785 re_newbuf_jumbo(struct re_softc *sc, int idx, int init)
1788 struct re_jbuf *jbuf;
1791 MGETHDR(m, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1795 if_printf(&sc->arpcom.ac_if, "MGETHDR failed\n");
1802 jbuf = re_jbuf_alloc(sc);
1808 if_printf(&sc->arpcom.ac_if, "jpool is empty\n");
1815 m->m_ext.ext_arg = jbuf;
1816 m->m_ext.ext_buf = jbuf->re_buf;
1817 m->m_ext.ext_free = re_jbuf_free;
1818 m->m_ext.ext_ref = re_jbuf_ref;
1819 m->m_ext.ext_size = sc->re_rxbuf_size;
1821 m->m_data = m->m_ext.ext_buf;
1822 m->m_flags |= M_EXT;
1823 m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
1827 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1828 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1831 sc->re_ldata.re_rx_mbuf[idx] = m;
1832 sc->re_ldata.re_rx_paddr[idx] = jbuf->re_paddr;
1834 re_setup_rxdesc(sc, idx);
1839 re_tx_list_init(struct re_softc *sc)
1841 bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
1843 /* Flush the TX descriptors */
1844 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1845 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1847 sc->re_ldata.re_tx_prodidx = 0;
1848 sc->re_ldata.re_tx_considx = 0;
1849 sc->re_ldata.re_tx_free = sc->re_tx_desc_cnt;
1855 re_rx_list_init(struct re_softc *sc)
1859 bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc));
1861 for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1862 error = sc->re_newbuf(sc, i, 1);
1867 /* Flush the RX descriptors */
1868 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1869 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1871 sc->re_ldata.re_rx_prodidx = 0;
1872 sc->re_head = sc->re_tail = NULL;
1877 #define RE_IP4_PACKET 0x1
1878 #define RE_TCP_PACKET 0x2
1879 #define RE_UDP_PACKET 0x4
1881 static __inline uint8_t
1882 re_packet_type(struct re_softc *sc, uint32_t rxstat, uint32_t rxctrl)
1884 uint8_t packet_type = 0;
1886 if (sc->re_caps & RE_C_MAC2) {
1887 if (rxctrl & RE_RDESC_CTL_PROTOIP4)
1888 packet_type |= RE_IP4_PACKET;
1890 if (rxstat & RE_RDESC_STAT_PROTOID)
1891 packet_type |= RE_IP4_PACKET;
1893 if (RE_TCPPKT(rxstat))
1894 packet_type |= RE_TCP_PACKET;
1895 else if (RE_UDPPKT(rxstat))
1896 packet_type |= RE_UDP_PACKET;
1901 * RX handler for C+ and 8169. For the gigE chips, we support
1902 * the reception of jumbo frames that have been fragmented
1903 * across multiple 2K mbuf cluster buffers.
1906 re_rxeof(struct re_softc *sc)
1908 struct ifnet *ifp = &sc->arpcom.ac_if;
1910 struct re_desc *cur_rx;
1911 uint32_t rxstat, rxctrl;
1912 int i, total_len, rx = 0;
1913 struct mbuf_chain chain[MAXCPU];
1915 /* Invalidate the descriptor memory */
1917 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1918 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1920 ether_input_chain_init(chain);
1922 for (i = sc->re_ldata.re_rx_prodidx;
1923 RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0; RE_RXDESC_INC(sc, i)) {
1924 cur_rx = &sc->re_ldata.re_rx_list[i];
1925 m = sc->re_ldata.re_rx_mbuf[i];
1926 total_len = RE_RXBYTES(cur_rx);
1927 rxstat = le32toh(cur_rx->re_cmdstat);
1928 rxctrl = le32toh(cur_rx->re_control);
1933 if (sc->re_flags & RE_F_USE_JPOOL)
1934 KKASSERT(rxstat & RE_RDESC_STAT_EOF);
1937 if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1938 if (sc->re_flags & RE_F_DROP_RXFRAG) {
1939 re_setup_rxdesc(sc, i);
1943 if (sc->re_newbuf(sc, i, 0)) {
1944 /* Drop upcoming fragments */
1945 sc->re_flags |= RE_F_DROP_RXFRAG;
1949 m->m_len = MCLBYTES;
1950 if (sc->re_head == NULL) {
1951 sc->re_head = sc->re_tail = m;
1953 sc->re_tail->m_next = m;
1957 } else if (sc->re_flags & RE_F_DROP_RXFRAG) {
1959 * Last fragment of a multi-fragment packet.
1961 * Since error already happened, this fragment
1962 * must be dropped as well as the fragment chain.
1964 re_setup_rxdesc(sc, i);
1965 re_free_rxchain(sc);
1966 sc->re_flags &= ~RE_F_DROP_RXFRAG;
1971 * NOTE: for the 8139C+, the frame length field
1972 * is always 12 bits in size, but for the gigE chips,
1973 * it is 13 bits (since the max RX frame length is 16K).
1974 * Unfortunately, all 32 bits in the status word
1975 * were already used, so to make room for the extra
1976 * length bit, RealTek took out the 'frame alignment
1977 * error' bit and shifted the other status bits
1978 * over one slot. The OWN, EOR, FS and LS bits are
1979 * still in the same places. We have already extracted
1980 * the frame length and checked the OWN bit, so rather
1981 * than using an alternate bit mapping, we shift the
1982 * status bits one space to the right so we can evaluate
1983 * them using the 8169 status as though it was in the
1984 * same format as that of the 8139C+.
1986 if (!RE_IS_8139CP(sc))
1989 if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1992 * If this is part of a multi-fragment packet,
1993 * discard all the pieces.
1995 re_free_rxchain(sc);
1996 re_setup_rxdesc(sc, i);
2001 * If allocating a replacement mbuf fails,
2002 * reload the current one.
2005 if (sc->re_newbuf(sc, i, 0)) {
2010 if (sc->re_head != NULL) {
2011 m->m_len = total_len % MCLBYTES;
2013 * Special case: if there's 4 bytes or less
2014 * in this buffer, the mbuf can be discarded:
2015 * the last 4 bytes is the CRC, which we don't
2016 * care about anyway.
2018 if (m->m_len <= ETHER_CRC_LEN) {
2019 sc->re_tail->m_len -=
2020 (ETHER_CRC_LEN - m->m_len);
2023 m->m_len -= ETHER_CRC_LEN;
2024 sc->re_tail->m_next = m;
2027 sc->re_head = sc->re_tail = NULL;
2028 m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2030 m->m_pkthdr.len = m->m_len =
2031 (total_len - ETHER_CRC_LEN);
2035 m->m_pkthdr.rcvif = ifp;
2037 /* Do RX checksumming if enabled */
2039 if (ifp->if_capenable & IFCAP_RXCSUM) {
2040 uint8_t packet_type;
2042 packet_type = re_packet_type(sc, rxstat, rxctrl);
2044 /* Check IP header checksum */
2045 if (packet_type & RE_IP4_PACKET) {
2046 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2047 if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
2048 m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2051 /* Check TCP/UDP checksum */
2052 if (((packet_type & RE_TCP_PACKET) &&
2053 (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
2054 ((packet_type & RE_UDP_PACKET) &&
2055 (rxstat & RE_RDESC_STAT_UDPSUMBAD) == 0)) {
2056 m->m_pkthdr.csum_flags |=
2057 CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
2058 CSUM_FRAG_NOT_CHECKED;
2059 m->m_pkthdr.csum_data = 0xffff;
2063 if (rxctrl & RE_RDESC_CTL_HASTAG) {
2064 m->m_flags |= M_VLANTAG;
2065 m->m_pkthdr.ether_vlantag =
2066 be16toh((rxctrl & RE_RDESC_CTL_TAGDATA));
2068 ether_input_chain(ifp, m, chain);
2071 ether_input_dispatch(chain);
2073 /* Flush the RX DMA ring */
2075 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
2076 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
2078 sc->re_ldata.re_rx_prodidx = i;
2083 #undef RE_IP4_PACKET
2084 #undef RE_TCP_PACKET
2085 #undef RE_UDP_PACKET
2088 re_tx_collect(struct re_softc *sc)
2090 struct ifnet *ifp = &sc->arpcom.ac_if;
2094 /* Invalidate the TX descriptor list */
2095 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2096 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
2098 for (idx = sc->re_ldata.re_tx_considx;
2099 sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt;
2100 RE_TXDESC_INC(sc, idx)) {
2101 txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
2102 if (txstat & RE_TDESC_CMD_OWN)
2107 sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
2110 * We only stash mbufs in the last descriptor
2111 * in a fragment chain, which also happens to
2112 * be the only place where the TX status bits
2115 if (txstat & RE_TDESC_CMD_EOF) {
2116 m_freem(sc->re_ldata.re_tx_mbuf[idx]);
2117 sc->re_ldata.re_tx_mbuf[idx] = NULL;
2118 bus_dmamap_unload(sc->re_ldata.re_mtag,
2119 sc->re_ldata.re_tx_dmamap[idx]);
2120 if (txstat & (RE_TDESC_STAT_EXCESSCOL|
2121 RE_TDESC_STAT_COLCNT))
2122 ifp->if_collisions++;
2123 if (txstat & RE_TDESC_STAT_TXERRSUM)
2128 sc->re_ldata.re_tx_free++;
2130 sc->re_ldata.re_tx_considx = idx;
2136 re_txeof(struct re_softc *sc)
2138 struct ifnet *ifp = &sc->arpcom.ac_if;
2141 tx = re_tx_collect(sc);
2143 /* There is enough free TX descs */
2144 if (sc->re_ldata.re_tx_free > RE_TXDESC_SPARE)
2145 ifp->if_flags &= ~IFF_OACTIVE;
2148 * Some chips will ignore a second TX request issued while an
2149 * existing transmission is in progress. If the transmitter goes
2150 * idle but there are still packets waiting to be sent, we need
2151 * to restart the channel here to flush them out. This only seems
2152 * to be required with the PCIe devices.
2154 if (sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt)
2155 CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2165 struct re_softc *sc = xsc;
2167 lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
2168 re_tick_serialized(xsc);
2169 lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
2173 re_tick_serialized(void *xsc)
2175 struct re_softc *sc = xsc;
2176 struct ifnet *ifp = &sc->arpcom.ac_if;
2177 struct mii_data *mii;
2179 ASSERT_SERIALIZED(ifp->if_serializer);
2181 mii = device_get_softc(sc->re_miibus);
2183 if (sc->re_flags & RE_F_LINKED) {
2184 if (!(mii->mii_media_status & IFM_ACTIVE))
2185 sc->re_flags &= ~RE_F_LINKED;
2187 if (mii->mii_media_status & IFM_ACTIVE &&
2188 IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2189 sc->re_flags |= RE_F_LINKED;
2190 if (!ifq_is_empty(&ifp->if_snd))
2195 callout_reset(&sc->re_timer, hz, re_tick, sc);
2198 #ifdef DEVICE_POLLING
2201 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2203 struct re_softc *sc = ifp->if_softc;
2205 ASSERT_SERIALIZED(ifp->if_serializer);
2209 /* disable interrupts */
2210 re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2213 case POLL_DEREGISTER:
2214 /* enable interrupts */
2215 re_setup_intr(sc, 1, sc->re_imtype);
2219 sc->rxcycles = count;
2223 if (!ifq_is_empty(&ifp->if_snd))
2226 if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2229 status = CSR_READ_2(sc, RE_ISR);
2230 if (status == 0xffff)
2233 CSR_WRITE_2(sc, RE_ISR, status);
2236 * XXX check behaviour on receiver stalls.
2239 if (status & RE_ISR_SYSTEM_ERR)
2245 #endif /* DEVICE_POLLING */
2250 struct re_softc *sc = arg;
2251 struct ifnet *ifp = &sc->arpcom.ac_if;
2255 ASSERT_SERIALIZED(ifp->if_serializer);
2257 if ((sc->re_flags & RE_F_SUSPENDED) ||
2258 (ifp->if_flags & IFF_RUNNING) == 0)
2263 status = CSR_READ_2(sc, RE_ISR);
2264 /* If the card has gone away the read returns 0xffff. */
2265 if (status == 0xffff)
2268 CSR_WRITE_2(sc, RE_ISR, status);
2270 if ((status & sc->re_intrs) == 0)
2273 if (status & (sc->re_rx_ack | RE_ISR_RX_ERR))
2276 if (status & (sc->re_tx_ack | RE_ISR_TX_ERR))
2279 if (status & RE_ISR_SYSTEM_ERR)
2282 if (status & RE_ISR_LINKCHG) {
2283 callout_stop(&sc->re_timer);
2284 re_tick_serialized(sc);
2288 if (sc->re_imtype == RE_IMTYPE_SIM) {
2289 if ((sc->re_flags & RE_F_TIMER_INTR)) {
2291 re_setup_intr(sc, 1, RE_IMTYPE_NONE);
2293 CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
2294 } else if (tx | rx) {
2295 re_setup_intr(sc, 1, RE_IMTYPE_SIM);
2299 if (tx && !ifq_is_empty(&ifp->if_snd))
2304 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx0)
2306 struct ifnet *ifp = &sc->arpcom.ac_if;
2308 struct re_dmaload_arg arg;
2309 bus_dma_segment_t segs[RE_MAXSEGS];
2311 int error, maxsegs, idx, i;
2312 struct re_desc *d, *tx_ring;
2313 uint32_t cmd_csum, ctl_csum, vlantag;
2315 KASSERT(sc->re_ldata.re_tx_free > RE_TXDESC_SPARE,
2316 ("not enough free TX desc\n"));
2319 map = sc->re_ldata.re_tx_dmamap[*idx0];
2322 * Set up checksum offload. Note: checksum offload bits must
2323 * appear in all descriptors of a multi-descriptor transmit
2324 * attempt. (This is according to testing done with an 8169
2325 * chip. I'm not sure if this is a requirement or a bug.)
2327 cmd_csum = ctl_csum = 0;
2328 if (m->m_pkthdr.csum_flags & CSUM_IP) {
2329 cmd_csum |= RE_TDESC_CMD_IPCSUM;
2330 ctl_csum |= RE_TDESC_CTL_IPCSUM;
2332 if (m->m_pkthdr.csum_flags & CSUM_TCP) {
2333 cmd_csum |= RE_TDESC_CMD_TCPCSUM;
2334 ctl_csum |= RE_TDESC_CTL_TCPCSUM;
2336 if (m->m_pkthdr.csum_flags & CSUM_UDP) {
2337 cmd_csum |= RE_TDESC_CMD_UDPCSUM;
2338 ctl_csum |= RE_TDESC_CTL_UDPCSUM;
2341 /* For MAC2 chips, csum flags are set on re_control */
2342 if (sc->re_caps & RE_C_MAC2)
2347 if (m->m_pkthdr.len > sc->re_swcsum_lim &&
2348 (m->m_pkthdr.csum_flags & (CSUM_DELAY_IP | CSUM_DELAY_DATA))) {
2349 struct ether_header *eh;
2353 m = m_pullup(m, sizeof(struct ether_header *));
2358 eh = mtod(m, struct ether_header *);
2361 if (eh->ether_type == ETHERTYPE_VLAN)
2362 offset = sizeof(struct ether_vlan_header);
2364 offset = sizeof(struct ether_header);
2366 m = m_pullup(m, offset + sizeof(struct ip *));
2371 ip = (struct ip *)(mtod(m, uint8_t *) + offset);
2373 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2376 offset += IP_VHL_HL(ip->ip_vhl) << 2;
2377 csum = in_cksum_skip(m, ntohs(ip->ip_len), offset);
2378 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
2380 offset += m->m_pkthdr.csum_data; /* checksum offset */
2381 *(u_short *)(m->m_data + offset) = csum;
2383 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2385 if (m->m_pkthdr.csum_flags & CSUM_DELAY_IP) {
2387 if (ip->ip_vhl == IP_VHL_BORING) {
2388 ip->ip_sum = in_cksum_hdr(ip);
2391 in_cksum(m, IP_VHL_HL(ip->ip_vhl) << 2);
2393 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
2395 *m_head = m; /* 'm' may be changed by above two m_pullup() */
2397 /* Clear hardware CSUM flags */
2398 cmd_csum = ctl_csum = 0;
2401 if ((sc->re_caps & RE_C_AUTOPAD) == 0) {
2403 * With some of the RealTek chips, using the checksum offload
2404 * support in conjunction with the autopadding feature results
2405 * in the transmission of corrupt frames. For example, if we
2406 * need to send a really small IP fragment that's less than 60
2407 * bytes in size, and IP header checksumming is enabled, the
2408 * resulting ethernet frame that appears on the wire will
2409 * have garbled payload. To work around this, if TX checksum
2410 * offload is enabled, we always manually pad short frames out
2411 * to the minimum ethernet frame size.
2413 * Note: this appears unnecessary for TCP, and doing it for TCP
2414 * with PCIe adapters seems to result in bad checksums.
2416 if ((m->m_pkthdr.csum_flags &
2417 (CSUM_DELAY_IP | CSUM_DELAY_DATA)) &&
2418 (m->m_pkthdr.csum_flags & CSUM_TCP) == 0 &&
2419 m->m_pkthdr.len < RE_MIN_FRAMELEN) {
2420 error = m_devpad(m, RE_MIN_FRAMELEN);
2427 if (m->m_flags & M_VLANTAG) {
2428 vlantag = htobe16(m->m_pkthdr.ether_vlantag) |
2429 RE_TDESC_CTL_INSTAG;
2432 maxsegs = sc->re_ldata.re_tx_free;
2433 if (maxsegs > RE_MAXSEGS)
2434 maxsegs = RE_MAXSEGS;
2436 arg.re_nsegs = maxsegs;
2438 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2439 re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2440 if (error && error != EFBIG) {
2441 if_printf(ifp, "can't map mbuf (error %d)\n", error);
2446 * Too many segments to map, coalesce into a single mbuf
2448 if (!error && arg.re_nsegs == 0) {
2449 bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2455 m_new = m_defrag(m, MB_DONTWAIT);
2456 if (m_new == NULL) {
2457 if_printf(ifp, "can't defrag TX mbuf\n");
2461 *m_head = m = m_new;
2464 arg.re_nsegs = maxsegs;
2466 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2467 re_dma_map_desc, &arg,
2469 if (error || arg.re_nsegs == 0) {
2471 bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2474 if_printf(ifp, "can't map mbuf (error %d)\n", error);
2478 bus_dmamap_sync(sc->re_ldata.re_mtag, map, BUS_DMASYNC_PREWRITE);
2481 * Map the segment array into descriptors. We also keep track
2482 * of the end of the ring and set the end-of-ring bits as needed,
2483 * and we set the ownership bits in all except the very first
2484 * descriptor, whose ownership bits will be turned on later.
2486 tx_ring = sc->re_ldata.re_tx_list;
2494 cmdstat = segs[i].ds_len;
2495 d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
2496 d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
2498 cmdstat |= RE_TDESC_CMD_SOF;
2500 cmdstat |= RE_TDESC_CMD_OWN;
2501 if (idx == (sc->re_tx_desc_cnt - 1))
2502 cmdstat |= RE_TDESC_CMD_EOR;
2503 d->re_cmdstat = htole32(cmdstat | cmd_csum);
2504 d->re_control = htole32(ctl_csum | vlantag);
2507 if (i == arg.re_nsegs)
2509 RE_TXDESC_INC(sc, idx);
2511 d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
2513 /* Transfer ownership of packet to the chip. */
2514 d->re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2516 tx_ring[*idx0].re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2519 * Insure that the map for this transmission
2520 * is placed at the array index of the last descriptor
2523 sc->re_ldata.re_tx_dmamap[*idx0] = sc->re_ldata.re_tx_dmamap[idx];
2524 sc->re_ldata.re_tx_dmamap[idx] = map;
2526 sc->re_ldata.re_tx_mbuf[idx] = m;
2527 sc->re_ldata.re_tx_free -= arg.re_nsegs;
2529 RE_TXDESC_INC(sc, idx);
2540 * Main transmit routine for C+ and gigE NICs.
2544 re_start(struct ifnet *ifp)
2546 struct re_softc *sc = ifp->if_softc;
2547 struct mbuf *m_head;
2548 int idx, need_trans, oactive, error;
2550 ASSERT_SERIALIZED(ifp->if_serializer);
2552 if ((sc->re_flags & RE_F_LINKED) == 0) {
2553 ifq_purge(&ifp->if_snd);
2557 if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2560 idx = sc->re_ldata.re_tx_prodidx;
2564 while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2565 if (sc->re_ldata.re_tx_free <= RE_TXDESC_SPARE) {
2567 if (re_tx_collect(sc)) {
2572 ifp->if_flags |= IFF_OACTIVE;
2576 m_head = ifq_dequeue(&ifp->if_snd, NULL);
2580 error = re_encap(sc, &m_head, &idx);
2582 /* m_head is freed by re_encap(), if we reach here */
2585 if (error == EFBIG && !oactive) {
2586 if (re_tx_collect(sc)) {
2591 ifp->if_flags |= IFF_OACTIVE;
2599 * If there's a BPF listener, bounce a copy of this frame
2602 ETHER_BPF_MTAP(ifp, m_head);
2608 /* Flush the TX descriptors */
2609 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2610 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2612 sc->re_ldata.re_tx_prodidx = idx;
2615 * RealTek put the TX poll request register in a different
2616 * location on the 8169 gigE chip. I don't know why.
2618 CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2621 * Set a timeout in case the chip goes out to lunch.
2629 struct re_softc *sc = xsc;
2630 struct ifnet *ifp = &sc->arpcom.ac_if;
2631 struct mii_data *mii;
2632 int error, framelen;
2634 ASSERT_SERIALIZED(ifp->if_serializer);
2636 mii = device_get_softc(sc->re_miibus);
2639 * Cancel pending I/O and free all RX/TX buffers.
2643 if (sc->re_caps & RE_C_CONTIGRX) {
2644 if (ifp->if_mtu > ETHERMTU) {
2645 KKASSERT(sc->re_ldata.re_jbuf != NULL);
2646 sc->re_flags |= RE_F_USE_JPOOL;
2647 sc->re_rxbuf_size = RE_FRAMELEN_MAX;
2648 sc->re_newbuf = re_newbuf_jumbo;
2650 sc->re_flags &= ~RE_F_USE_JPOOL;
2651 sc->re_rxbuf_size = MCLBYTES;
2652 sc->re_newbuf = re_newbuf_std;
2657 * Adjust max read request size according to MTU.
2658 * Mainly to improve TX performance for common case (ETHERMTU).
2660 if (sc->re_caps & RE_C_PCIE) {
2661 if (ifp->if_mtu > ETHERMTU) {
2663 * 512 seems to be the only value that works
2664 * reliably with jumbo frame
2666 pcie_set_max_readrq(sc->re_dev,
2667 PCIEM_DEVCTL_MAX_READRQ_512);
2669 pcie_set_max_readrq(sc->re_dev,
2670 PCIEM_DEVCTL_MAX_READRQ_4096);
2675 * Enable C+ RX and TX mode, as well as VLAN stripping and
2676 * RX checksum offload. We must configure the C+ register
2677 * before all others.
2679 CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2680 RE_CPLUSCMD_PCI_MRW |
2681 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING ?
2682 RE_CPLUSCMD_VLANSTRIP : 0) |
2683 (ifp->if_capenable & IFCAP_RXCSUM ?
2684 RE_CPLUSCMD_RXCSUM_ENB : 0));
2687 * Init our MAC address. Even though the chipset
2688 * documentation doesn't mention it, we need to enter "Config
2689 * register write enable" mode to modify the ID registers.
2691 CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2692 CSR_WRITE_4(sc, RE_IDR0,
2693 htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2694 CSR_WRITE_2(sc, RE_IDR4,
2695 htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2696 CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2699 * For C+ mode, initialize the RX descriptors and mbufs.
2701 error = re_rx_list_init(sc);
2706 error = re_tx_list_init(sc);
2713 * Load the addresses of the RX and TX lists into the chip.
2715 CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2716 RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2717 CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2718 RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2720 CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2721 RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2722 CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2723 RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2726 * Enable transmit and receive.
2728 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2731 * Set the initial TX and RX configuration.
2733 if (sc->re_flags & RE_F_TESTMODE) {
2734 if (!RE_IS_8139CP(sc))
2735 CSR_WRITE_4(sc, RE_TXCFG,
2736 RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2738 CSR_WRITE_4(sc, RE_TXCFG,
2739 RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2741 CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2743 framelen = RE_FRAMELEN(ifp->if_mtu);
2744 if (framelen < MCLBYTES)
2745 CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(MCLBYTES, 128));
2747 CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(framelen, 128));
2749 CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2752 * Program the multicast filter, if necessary.
2756 #ifdef DEVICE_POLLING
2758 * Disable interrupts if we are polling.
2760 if (ifp->if_flags & IFF_POLLING)
2761 re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2762 else /* otherwise ... */
2763 #endif /* DEVICE_POLLING */
2765 * Enable interrupts.
2767 if (sc->re_flags & RE_F_TESTMODE)
2768 CSR_WRITE_2(sc, RE_IMR, 0);
2770 re_setup_intr(sc, 1, sc->re_imtype);
2771 CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2773 /* Start RX/TX process. */
2774 CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2777 /* Enable receiver and transmitter. */
2778 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2782 * For 8169 gigE NICs, set the max allowed RX packet
2783 * size so we can receive jumbo frames.
2785 if (!RE_IS_8139CP(sc)) {
2786 if (sc->re_caps & RE_C_CONTIGRX)
2787 CSR_WRITE_2(sc, RE_MAXRXPKTLEN, sc->re_rxbuf_size);
2789 CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2792 if (sc->re_flags & RE_F_TESTMODE)
2797 CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2799 ifp->if_flags |= IFF_RUNNING;
2800 ifp->if_flags &= ~IFF_OACTIVE;
2802 callout_reset(&sc->re_timer, hz, re_tick, sc);
2806 * Set media options.
2809 re_ifmedia_upd(struct ifnet *ifp)
2811 struct re_softc *sc = ifp->if_softc;
2812 struct mii_data *mii;
2814 ASSERT_SERIALIZED(ifp->if_serializer);
2816 mii = device_get_softc(sc->re_miibus);
2823 * Report current media status.
2826 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2828 struct re_softc *sc = ifp->if_softc;
2829 struct mii_data *mii;
2831 ASSERT_SERIALIZED(ifp->if_serializer);
2833 mii = device_get_softc(sc->re_miibus);
2836 ifmr->ifm_active = mii->mii_media_active;
2837 ifmr->ifm_status = mii->mii_media_status;
2841 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2843 struct re_softc *sc = ifp->if_softc;
2844 struct ifreq *ifr = (struct ifreq *) data;
2845 struct mii_data *mii;
2846 int error = 0, mask;
2848 ASSERT_SERIALIZED(ifp->if_serializer);
2852 if (ifr->ifr_mtu > sc->re_maxmtu) {
2854 } else if (ifp->if_mtu != ifr->ifr_mtu) {
2855 ifp->if_mtu = ifr->ifr_mtu;
2856 if (ifp->if_flags & IFF_RUNNING)
2862 if (ifp->if_flags & IFF_UP) {
2863 if (ifp->if_flags & IFF_RUNNING) {
2864 if ((ifp->if_flags ^ sc->re_if_flags) &
2865 (IFF_PROMISC | IFF_ALLMULTI))
2870 } else if (ifp->if_flags & IFF_RUNNING) {
2873 sc->re_if_flags = ifp->if_flags;
2883 mii = device_get_softc(sc->re_miibus);
2884 error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2888 mask = (ifr->ifr_reqcap ^ ifp->if_capenable) &
2889 ifp->if_capabilities;
2890 ifp->if_capenable ^= mask;
2892 if (mask & IFCAP_HWCSUM) {
2893 if (ifp->if_capenable & IFCAP_TXCSUM)
2894 ifp->if_hwassist = RE_CSUM_FEATURES;
2896 ifp->if_hwassist = 0;
2898 if (mask && (ifp->if_flags & IFF_RUNNING))
2903 error = ether_ioctl(ifp, command, data);
2910 re_watchdog(struct ifnet *ifp)
2912 struct re_softc *sc = ifp->if_softc;
2914 ASSERT_SERIALIZED(ifp->if_serializer);
2916 if_printf(ifp, "watchdog timeout\n");
2925 if (!ifq_is_empty(&ifp->if_snd))
2930 * Stop the adapter and free any mbufs allocated to the
2934 re_stop(struct re_softc *sc)
2936 struct ifnet *ifp = &sc->arpcom.ac_if;
2939 ASSERT_SERIALIZED(ifp->if_serializer);
2941 /* Reset the adapter. */
2942 re_reset(sc, ifp->if_flags & IFF_RUNNING);
2945 callout_stop(&sc->re_timer);
2947 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2948 sc->re_flags &= ~(RE_F_TIMER_INTR | RE_F_DROP_RXFRAG | RE_F_LINKED);
2950 CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2951 CSR_WRITE_2(sc, RE_IMR, 0x0000);
2952 CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2954 re_free_rxchain(sc);
2956 /* Free the TX list buffers. */
2957 for (i = 0; i < sc->re_tx_desc_cnt; i++) {
2958 if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2959 bus_dmamap_unload(sc->re_ldata.re_mtag,
2960 sc->re_ldata.re_tx_dmamap[i]);
2961 m_freem(sc->re_ldata.re_tx_mbuf[i]);
2962 sc->re_ldata.re_tx_mbuf[i] = NULL;
2966 /* Free the RX list buffers. */
2967 for (i = 0; i < sc->re_rx_desc_cnt; i++) {
2968 if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2969 if ((sc->re_flags & RE_F_USE_JPOOL) == 0) {
2970 bus_dmamap_unload(sc->re_ldata.re_mtag,
2971 sc->re_ldata.re_rx_dmamap[i]);
2973 m_freem(sc->re_ldata.re_rx_mbuf[i]);
2974 sc->re_ldata.re_rx_mbuf[i] = NULL;
2980 * Device suspend routine. Stop the interface and save some PCI
2981 * settings in case the BIOS doesn't restore them properly on
2985 re_suspend(device_t dev)
2987 #ifndef BURN_BRIDGES
2990 struct re_softc *sc = device_get_softc(dev);
2991 struct ifnet *ifp = &sc->arpcom.ac_if;
2993 lwkt_serialize_enter(ifp->if_serializer);
2997 #ifndef BURN_BRIDGES
2998 for (i = 0; i < 5; i++)
2999 sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
3000 sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
3001 sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
3002 sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3003 sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3006 sc->re_flags |= RE_F_SUSPENDED;
3008 lwkt_serialize_exit(ifp->if_serializer);
3014 * Device resume routine. Restore some PCI settings in case the BIOS
3015 * doesn't, re-enable busmastering, and restart the interface if
3019 re_resume(device_t dev)
3021 struct re_softc *sc = device_get_softc(dev);
3022 struct ifnet *ifp = &sc->arpcom.ac_if;
3023 #ifndef BURN_BRIDGES
3027 lwkt_serialize_enter(ifp->if_serializer);
3029 #ifndef BURN_BRIDGES
3030 /* better way to do this? */
3031 for (i = 0; i < 5; i++)
3032 pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
3033 pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
3034 pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
3035 pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
3036 pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
3038 /* reenable busmastering */
3039 pci_enable_busmaster(dev);
3040 pci_enable_io(dev, SYS_RES_IOPORT);
3043 /* reinitialize interface if necessary */
3044 if (ifp->if_flags & IFF_UP)
3047 sc->re_flags &= ~RE_F_SUSPENDED;
3049 lwkt_serialize_exit(ifp->if_serializer);
3055 * Stop all chip I/O so that the kernel's probe routines don't
3056 * get confused by errant DMAs when rebooting.
3059 re_shutdown(device_t dev)
3061 struct re_softc *sc = device_get_softc(dev);
3062 struct ifnet *ifp = &sc->arpcom.ac_if;
3064 lwkt_serialize_enter(ifp->if_serializer);
3066 lwkt_serialize_exit(ifp->if_serializer);
3070 re_sysctl_rxtime(SYSCTL_HANDLER_ARGS)
3072 struct re_softc *sc = arg1;
3074 return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_rx_time);
3078 re_sysctl_txtime(SYSCTL_HANDLER_ARGS)
3080 struct re_softc *sc = arg1;
3082 return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_tx_time);
3086 re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *hwtime)
3088 struct re_softc *sc = arg1;
3089 struct ifnet *ifp = &sc->arpcom.ac_if;
3092 lwkt_serialize_enter(ifp->if_serializer);
3095 error = sysctl_handle_int(oidp, &v, 0, req);
3096 if (error || req->newptr == NULL)
3107 if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3108 IFF_RUNNING && sc->re_imtype == RE_IMTYPE_HW)
3112 lwkt_serialize_exit(ifp->if_serializer);
3117 re_sysctl_simtime(SYSCTL_HANDLER_ARGS)
3119 struct re_softc *sc = arg1;
3120 struct ifnet *ifp = &sc->arpcom.ac_if;
3123 lwkt_serialize_enter(ifp->if_serializer);
3125 v = sc->re_sim_time;
3126 error = sysctl_handle_int(oidp, &v, 0, req);
3127 if (error || req->newptr == NULL)
3135 if (v != sc->re_sim_time) {
3136 sc->re_sim_time = v;
3138 if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3139 IFF_RUNNING && sc->re_imtype == RE_IMTYPE_SIM) {
3144 * Following code causes various strange
3145 * performance problems. Hmm ...
3147 CSR_WRITE_2(sc, RE_IMR, 0);
3148 if (!RE_IS_8139CP(sc))
3149 reg = RE_TIMERINT_8169;
3152 CSR_WRITE_4(sc, reg, 0);
3153 CSR_READ_4(sc, reg); /* flush */
3155 CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3156 re_setup_sim_im(sc);
3158 re_setup_intr(sc, 0, RE_IMTYPE_NONE);
3160 re_setup_intr(sc, 1, RE_IMTYPE_SIM);
3165 lwkt_serialize_exit(ifp->if_serializer);
3170 re_sysctl_imtype(SYSCTL_HANDLER_ARGS)
3172 struct re_softc *sc = arg1;
3173 struct ifnet *ifp = &sc->arpcom.ac_if;
3176 lwkt_serialize_enter(ifp->if_serializer);
3179 error = sysctl_handle_int(oidp, &v, 0, req);
3180 if (error || req->newptr == NULL)
3183 if (v != RE_IMTYPE_HW && v != RE_IMTYPE_SIM && v != RE_IMTYPE_NONE) {
3187 if (v == RE_IMTYPE_HW && (sc->re_caps & RE_C_HWIM) == 0) {
3188 /* Can't do hardware interrupt moderation */
3193 if (v != sc->re_imtype) {
3195 if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3197 re_setup_intr(sc, 1, sc->re_imtype);
3200 lwkt_serialize_exit(ifp->if_serializer);
3205 re_setup_hw_im(struct re_softc *sc)
3207 KKASSERT(sc->re_caps & RE_C_HWIM);
3210 * Interrupt moderation
3213 * A - unknown (maybe TX related)
3214 * B - TX timer (unit: 25us)
3215 * C - unknown (maybe RX related)
3216 * D - RX timer (unit: 25us)
3219 * re(4)'s interrupt moderation is actually controlled by
3220 * two variables, like most other NICs (bge, bce etc.)
3222 * o number of packets [P]
3224 * The logic relationship between these two variables is
3225 * similar to other NICs too:
3226 * if (timer expire || packets > [P])
3227 * Interrupt is delivered
3229 * Currently we only know how to set 'timer', but not
3230 * 'number of packets', which should be ~30, as far as I
3231 * tested (sink ~900Kpps, interrupt rate is 30KHz)
3233 CSR_WRITE_2(sc, RE_IM,
3234 RE_IM_RXTIME(sc->re_rx_time) |
3235 RE_IM_TXTIME(sc->re_tx_time) |
3240 re_disable_hw_im(struct re_softc *sc)
3242 if (sc->re_caps & RE_C_HWIM)
3243 CSR_WRITE_2(sc, RE_IM, 0);
3247 re_setup_sim_im(struct re_softc *sc)
3249 if (!RE_IS_8139CP(sc)) {
3253 * Datasheet says tick decreases at bus speed,
3254 * but it seems the clock runs a little bit
3255 * faster, so we do some compensation here.
3257 ticks = (sc->re_sim_time * sc->re_bus_speed * 8) / 5;
3258 CSR_WRITE_4(sc, RE_TIMERINT_8169, ticks);
3260 CSR_WRITE_4(sc, RE_TIMERINT, 0x400); /* XXX */
3262 CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
3263 sc->re_flags |= RE_F_TIMER_INTR;
3267 re_disable_sim_im(struct re_softc *sc)
3269 if (!RE_IS_8139CP(sc))
3270 CSR_WRITE_4(sc, RE_TIMERINT_8169, 0);
3272 CSR_WRITE_4(sc, RE_TIMERINT, 0);
3273 sc->re_flags &= ~RE_F_TIMER_INTR;
3277 re_config_imtype(struct re_softc *sc, int imtype)
3281 KKASSERT(sc->re_caps & RE_C_HWIM);
3283 case RE_IMTYPE_NONE:
3284 sc->re_intrs = RE_INTRS;
3285 sc->re_rx_ack = RE_ISR_RX_OK | RE_ISR_FIFO_OFLOW |
3287 sc->re_tx_ack = RE_ISR_TX_OK;
3291 sc->re_intrs = RE_INTRS_TIMER;
3292 sc->re_rx_ack = RE_ISR_TIMEOUT_EXPIRED;
3293 sc->re_tx_ack = RE_ISR_TIMEOUT_EXPIRED;
3297 panic("%s: unknown imtype %d\n",
3298 sc->arpcom.ac_if.if_xname, imtype);
3303 re_setup_intr(struct re_softc *sc, int enable_intrs, int imtype)
3305 re_config_imtype(sc, imtype);
3308 CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3310 CSR_WRITE_2(sc, RE_IMR, 0);
3313 case RE_IMTYPE_NONE:
3314 re_disable_sim_im(sc);
3315 re_disable_hw_im(sc);
3319 KKASSERT(sc->re_caps & RE_C_HWIM);
3320 re_disable_sim_im(sc);
3325 re_disable_hw_im(sc);
3326 re_setup_sim_im(sc);
3330 panic("%s: unknown imtype %d\n",
3331 sc->arpcom.ac_if.if_xname, imtype);
3336 re_get_eaddr(struct re_softc *sc, uint8_t *eaddr)
3340 if (sc->re_macver == RE_MACVER_11 || sc->re_macver == RE_MACVER_12) {
3344 re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
3345 if (re_did == 0x8128) {
3346 uint16_t as[ETHER_ADDR_LEN / 2];
3349 * Get station address from the EEPROM.
3351 re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
3352 for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
3353 as[i] = le16toh(as[i]);
3354 bcopy(as, eaddr, sizeof(eaddr));
3360 * Get station address from IDRx.
3362 for (i = 0; i < ETHER_ADDR_LEN; ++i)
3363 eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
3367 re_jpool_alloc(struct re_softc *sc)
3369 struct re_list_data *ldata = &sc->re_ldata;
3370 struct re_jbuf *jbuf;
3372 bus_size_t jpool_size;
3376 lwkt_serialize_init(&ldata->re_jbuf_serializer);
3378 ldata->re_jbuf = kmalloc(sizeof(struct re_jbuf) * RE_JBUF_COUNT(sc),
3379 M_DEVBUF, M_WAITOK | M_ZERO);
3381 jpool_size = RE_JBUF_COUNT(sc) * RE_JBUF_SIZE;
3383 error = bus_dma_tag_create(sc->re_parent_tag,
3384 RE_BUF_ALIGN, 0, /* alignment, boundary */
3385 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
3386 BUS_SPACE_MAXADDR, /* highaddr */
3387 NULL, NULL, /* filter, filterarg */
3388 jpool_size, 1, /* nsegments, maxsize */
3389 BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
3390 BUS_DMA_ALLOCNOW, /* flags */
3391 &ldata->re_jpool_tag);
3393 device_printf(sc->re_dev, "could not allocate jumbo dma tag\n");
3397 error = bus_dmamem_alloc(ldata->re_jpool_tag, (void **)&ldata->re_jpool,
3398 BUS_DMA_WAITOK, &ldata->re_jpool_map);
3400 device_printf(sc->re_dev,
3401 "could not allocate jumbo dma memory\n");
3402 bus_dma_tag_destroy(ldata->re_jpool_tag);
3403 ldata->re_jpool_tag = NULL;
3407 error = bus_dmamap_load(ldata->re_jpool_tag, ldata->re_jpool_map,
3408 ldata->re_jpool, jpool_size,
3409 re_dma_map_addr, &paddr, BUS_DMA_WAITOK);
3411 device_printf(sc->re_dev, "could not load jumbo dma map\n");
3412 bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3413 ldata->re_jpool_map);
3414 bus_dma_tag_destroy(ldata->re_jpool_tag);
3415 ldata->re_jpool_tag = NULL;
3419 /* ..and split it into 9KB chunks */
3420 SLIST_INIT(&ldata->re_jbuf_free);
3422 buf = ldata->re_jpool;
3423 for (i = 0; i < RE_JBUF_COUNT(sc); i++) {
3424 jbuf = &ldata->re_jbuf[i];
3430 jbuf->re_paddr = paddr;
3432 SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3434 buf += RE_JBUF_SIZE;
3435 paddr += RE_JBUF_SIZE;
3441 re_jpool_free(struct re_softc *sc)
3443 struct re_list_data *ldata = &sc->re_ldata;
3445 if (ldata->re_jpool_tag != NULL) {
3446 bus_dmamap_unload(ldata->re_jpool_tag, ldata->re_jpool_map);
3447 bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3448 ldata->re_jpool_map);
3449 bus_dma_tag_destroy(ldata->re_jpool_tag);
3450 ldata->re_jpool_tag = NULL;
3453 if (ldata->re_jbuf != NULL) {
3454 kfree(ldata->re_jbuf, M_DEVBUF);
3455 ldata->re_jbuf = NULL;
3459 static struct re_jbuf *
3460 re_jbuf_alloc(struct re_softc *sc)
3462 struct re_list_data *ldata = &sc->re_ldata;
3463 struct re_jbuf *jbuf;
3465 lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3467 jbuf = SLIST_FIRST(&ldata->re_jbuf_free);
3469 SLIST_REMOVE_HEAD(&ldata->re_jbuf_free, re_link);
3473 lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3479 re_jbuf_free(void *arg)
3481 struct re_jbuf *jbuf = arg;
3482 struct re_softc *sc = jbuf->re_sc;
3483 struct re_list_data *ldata = &sc->re_ldata;
3485 if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3486 panic("%s: free wrong jumbo buffer\n",
3487 sc->arpcom.ac_if.if_xname);
3488 } else if (jbuf->re_inuse == 0) {
3489 panic("%s: jumbo buffer already freed\n",
3490 sc->arpcom.ac_if.if_xname);
3493 lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3494 atomic_subtract_int(&jbuf->re_inuse, 1);
3495 if (jbuf->re_inuse == 0)
3496 SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3497 lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3501 re_jbuf_ref(void *arg)
3503 struct re_jbuf *jbuf = arg;
3504 struct re_softc *sc = jbuf->re_sc;
3505 struct re_list_data *ldata = &sc->re_ldata;
3507 if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3508 panic("%s: ref wrong jumbo buffer\n",
3509 sc->arpcom.ac_if.if_xname);
3510 } else if (jbuf->re_inuse == 0) {
3511 panic("%s: jumbo buffer already freed\n",
3512 sc->arpcom.ac_if.if_xname);
3514 atomic_add_int(&jbuf->re_inuse, 1);