2 * Copyright (c) 2004 Joerg Sonnenberger <joerg@bec.de>. All rights reserved.
4 * Copyright (c) 2001-2008, Intel Corporation
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * 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.
17 * 3. Neither the name of the Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived from
19 * this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
25 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
34 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
36 * This code is derived from software contributed to The DragonFly Project
37 * by Matthew Dillon <dillon@backplane.com>
39 * Redistribution and use in source and binary forms, with or without
40 * modification, are permitted provided that the following conditions
43 * 1. Redistributions of source code must retain the above copyright
44 * notice, this list of conditions and the following disclaimer.
45 * 2. Redistributions in binary form must reproduce the above copyright
46 * notice, this list of conditions and the following disclaimer in
47 * the documentation and/or other materials provided with the
49 * 3. Neither the name of The DragonFly Project nor the names of its
50 * contributors may be used to endorse or promote products derived
51 * from this software without specific, prior written permission.
53 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
54 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
55 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
56 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
57 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
58 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
59 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
60 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
61 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
62 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
63 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 #include "opt_ifpoll.h"
70 #include <sys/param.h>
72 #include <sys/endian.h>
73 #include <sys/interrupt.h>
74 #include <sys/kernel.h>
76 #include <sys/malloc.h>
80 #include <sys/serialize.h>
81 #include <sys/serialize2.h>
82 #include <sys/socket.h>
83 #include <sys/sockio.h>
84 #include <sys/sysctl.h>
85 #include <sys/systm.h>
88 #include <net/ethernet.h>
90 #include <net/if_arp.h>
91 #include <net/if_dl.h>
92 #include <net/if_media.h>
93 #include <net/ifq_var.h>
94 #include <net/toeplitz.h>
95 #include <net/toeplitz2.h>
96 #include <net/vlan/if_vlan_var.h>
97 #include <net/vlan/if_vlan_ether.h>
98 #include <net/if_poll.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/in.h>
102 #include <netinet/ip.h>
103 #include <netinet/tcp.h>
104 #include <netinet/udp.h>
106 #include <bus/pci/pcivar.h>
107 #include <bus/pci/pcireg.h>
109 #include <dev/netif/ig_hal/e1000_api.h>
110 #include <dev/netif/ig_hal/e1000_82571.h>
111 #include <dev/netif/emx/if_emx.h>
114 #define EMX_RSS_DPRINTF(sc, lvl, fmt, ...) \
116 if (sc->rss_debug >= lvl) \
117 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
119 #else /* !EMX_RSS_DEBUG */
120 #define EMX_RSS_DPRINTF(sc, lvl, fmt, ...) ((void)0)
121 #endif /* EMX_RSS_DEBUG */
123 #define EMX_TX_SERIALIZE 1
124 #define EMX_RX_SERIALIZE 2
126 #define EMX_NAME "Intel(R) PRO/1000 "
128 #define EMX_DEVICE(id) \
129 { EMX_VENDOR_ID, E1000_DEV_ID_##id, EMX_NAME #id }
130 #define EMX_DEVICE_NULL { 0, 0, NULL }
132 static const struct emx_device {
137 EMX_DEVICE(82571EB_COPPER),
138 EMX_DEVICE(82571EB_FIBER),
139 EMX_DEVICE(82571EB_SERDES),
140 EMX_DEVICE(82571EB_SERDES_DUAL),
141 EMX_DEVICE(82571EB_SERDES_QUAD),
142 EMX_DEVICE(82571EB_QUAD_COPPER),
143 EMX_DEVICE(82571EB_QUAD_COPPER_BP),
144 EMX_DEVICE(82571EB_QUAD_COPPER_LP),
145 EMX_DEVICE(82571EB_QUAD_FIBER),
146 EMX_DEVICE(82571PT_QUAD_COPPER),
148 EMX_DEVICE(82572EI_COPPER),
149 EMX_DEVICE(82572EI_FIBER),
150 EMX_DEVICE(82572EI_SERDES),
154 EMX_DEVICE(82573E_IAMT),
157 EMX_DEVICE(80003ES2LAN_COPPER_SPT),
158 EMX_DEVICE(80003ES2LAN_SERDES_SPT),
159 EMX_DEVICE(80003ES2LAN_COPPER_DPT),
160 EMX_DEVICE(80003ES2LAN_SERDES_DPT),
165 /* required last entry */
169 static int emx_probe(device_t);
170 static int emx_attach(device_t);
171 static int emx_detach(device_t);
172 static int emx_shutdown(device_t);
173 static int emx_suspend(device_t);
174 static int emx_resume(device_t);
176 static void emx_init(void *);
177 static void emx_stop(struct emx_softc *);
178 static int emx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
179 static void emx_start(struct ifnet *);
181 static void emx_npoll(struct ifnet *, struct ifpoll_info *);
182 static void emx_npoll_status(struct ifnet *);
183 static void emx_npoll_tx(struct ifnet *, void *, int);
184 static void emx_npoll_rx(struct ifnet *, void *, int);
186 static void emx_watchdog(struct ifnet *);
187 static void emx_media_status(struct ifnet *, struct ifmediareq *);
188 static int emx_media_change(struct ifnet *);
189 static void emx_timer(void *);
190 static void emx_serialize(struct ifnet *, enum ifnet_serialize);
191 static void emx_deserialize(struct ifnet *, enum ifnet_serialize);
192 static int emx_tryserialize(struct ifnet *, enum ifnet_serialize);
194 static void emx_serialize_assert(struct ifnet *, enum ifnet_serialize,
198 static void emx_intr(void *);
199 static void emx_intr_mask(void *);
200 static void emx_intr_body(struct emx_softc *, boolean_t);
201 static void emx_rxeof(struct emx_softc *, int, int);
202 static void emx_txeof(struct emx_softc *);
203 static void emx_tx_collect(struct emx_softc *);
204 static void emx_tx_purge(struct emx_softc *);
205 static void emx_enable_intr(struct emx_softc *);
206 static void emx_disable_intr(struct emx_softc *);
208 static int emx_dma_alloc(struct emx_softc *);
209 static void emx_dma_free(struct emx_softc *);
210 static void emx_init_tx_ring(struct emx_softc *);
211 static int emx_init_rx_ring(struct emx_softc *, struct emx_rxdata *);
212 static void emx_free_rx_ring(struct emx_softc *, struct emx_rxdata *);
213 static int emx_create_tx_ring(struct emx_softc *);
214 static int emx_create_rx_ring(struct emx_softc *, struct emx_rxdata *);
215 static void emx_destroy_tx_ring(struct emx_softc *, int);
216 static void emx_destroy_rx_ring(struct emx_softc *,
217 struct emx_rxdata *, int);
218 static int emx_newbuf(struct emx_softc *, struct emx_rxdata *, int, int);
219 static int emx_encap(struct emx_softc *, struct mbuf **);
220 static int emx_txcsum(struct emx_softc *, struct mbuf *,
221 uint32_t *, uint32_t *);
222 static int emx_tso_pullup(struct emx_softc *, struct mbuf **);
223 static int emx_tso_setup(struct emx_softc *, struct mbuf *,
224 uint32_t *, uint32_t *);
226 static int emx_is_valid_eaddr(const uint8_t *);
227 static int emx_reset(struct emx_softc *);
228 static void emx_setup_ifp(struct emx_softc *);
229 static void emx_init_tx_unit(struct emx_softc *);
230 static void emx_init_rx_unit(struct emx_softc *);
231 static void emx_update_stats(struct emx_softc *);
232 static void emx_set_promisc(struct emx_softc *);
233 static void emx_disable_promisc(struct emx_softc *);
234 static void emx_set_multi(struct emx_softc *);
235 static void emx_update_link_status(struct emx_softc *);
236 static void emx_smartspeed(struct emx_softc *);
237 static void emx_set_itr(struct emx_softc *, uint32_t);
238 static void emx_disable_aspm(struct emx_softc *);
240 static void emx_print_debug_info(struct emx_softc *);
241 static void emx_print_nvm_info(struct emx_softc *);
242 static void emx_print_hw_stats(struct emx_softc *);
244 static int emx_sysctl_stats(SYSCTL_HANDLER_ARGS);
245 static int emx_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
246 static int emx_sysctl_int_throttle(SYSCTL_HANDLER_ARGS);
247 static int emx_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS);
249 static int emx_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS);
250 static int emx_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS);
252 static void emx_add_sysctl(struct emx_softc *);
254 static void emx_serialize_skipmain(struct emx_softc *);
255 static void emx_deserialize_skipmain(struct emx_softc *);
257 /* Management and WOL Support */
258 static void emx_get_mgmt(struct emx_softc *);
259 static void emx_rel_mgmt(struct emx_softc *);
260 static void emx_get_hw_control(struct emx_softc *);
261 static void emx_rel_hw_control(struct emx_softc *);
262 static void emx_enable_wol(device_t);
264 static device_method_t emx_methods[] = {
265 /* Device interface */
266 DEVMETHOD(device_probe, emx_probe),
267 DEVMETHOD(device_attach, emx_attach),
268 DEVMETHOD(device_detach, emx_detach),
269 DEVMETHOD(device_shutdown, emx_shutdown),
270 DEVMETHOD(device_suspend, emx_suspend),
271 DEVMETHOD(device_resume, emx_resume),
275 static driver_t emx_driver = {
278 sizeof(struct emx_softc),
281 static devclass_t emx_devclass;
283 DECLARE_DUMMY_MODULE(if_emx);
284 MODULE_DEPEND(emx, ig_hal, 1, 1, 1);
285 DRIVER_MODULE(if_emx, pci, emx_driver, emx_devclass, NULL, NULL);
290 static int emx_int_throttle_ceil = EMX_DEFAULT_ITR;
291 static int emx_rxd = EMX_DEFAULT_RXD;
292 static int emx_txd = EMX_DEFAULT_TXD;
293 static int emx_smart_pwr_down = 0;
294 static int emx_rxr = 0;
296 /* Controls whether promiscuous also shows bad packets */
297 static int emx_debug_sbp = 0;
299 static int emx_82573_workaround = 1;
300 static int emx_msi_enable = 1;
302 TUNABLE_INT("hw.emx.int_throttle_ceil", &emx_int_throttle_ceil);
303 TUNABLE_INT("hw.emx.rxd", &emx_rxd);
304 TUNABLE_INT("hw.emx.rxr", &emx_rxr);
305 TUNABLE_INT("hw.emx.txd", &emx_txd);
306 TUNABLE_INT("hw.emx.smart_pwr_down", &emx_smart_pwr_down);
307 TUNABLE_INT("hw.emx.sbp", &emx_debug_sbp);
308 TUNABLE_INT("hw.emx.82573_workaround", &emx_82573_workaround);
309 TUNABLE_INT("hw.emx.msi.enable", &emx_msi_enable);
311 /* Global used in WOL setup with multiport cards */
312 static int emx_global_quad_port_a = 0;
314 /* Set this to one to display debug statistics */
315 static int emx_display_debug_stats = 0;
317 #if !defined(KTR_IF_EMX)
318 #define KTR_IF_EMX KTR_ALL
320 KTR_INFO_MASTER(if_emx);
321 KTR_INFO(KTR_IF_EMX, if_emx, intr_beg, 0, "intr begin");
322 KTR_INFO(KTR_IF_EMX, if_emx, intr_end, 1, "intr end");
323 KTR_INFO(KTR_IF_EMX, if_emx, pkt_receive, 4, "rx packet");
324 KTR_INFO(KTR_IF_EMX, if_emx, pkt_txqueue, 5, "tx packet");
325 KTR_INFO(KTR_IF_EMX, if_emx, pkt_txclean, 6, "tx clean");
326 #define logif(name) KTR_LOG(if_emx_ ## name)
329 emx_setup_rxdesc(emx_rxdesc_t *rxd, const struct emx_rxbuf *rxbuf)
331 rxd->rxd_bufaddr = htole64(rxbuf->paddr);
332 /* DD bit must be cleared */
333 rxd->rxd_staterr = 0;
337 emx_rxcsum(uint32_t staterr, struct mbuf *mp)
339 /* Ignore Checksum bit is set */
340 if (staterr & E1000_RXD_STAT_IXSM)
343 if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
345 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
347 if ((staterr & (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
348 E1000_RXD_STAT_TCPCS) {
349 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
351 CSUM_FRAG_NOT_CHECKED;
352 mp->m_pkthdr.csum_data = htons(0xffff);
356 static __inline struct pktinfo *
357 emx_rssinfo(struct mbuf *m, struct pktinfo *pi,
358 uint32_t mrq, uint32_t hash, uint32_t staterr)
360 switch (mrq & EMX_RXDMRQ_RSSTYPE_MASK) {
361 case EMX_RXDMRQ_IPV4_TCP:
362 pi->pi_netisr = NETISR_IP;
364 pi->pi_l3proto = IPPROTO_TCP;
367 case EMX_RXDMRQ_IPV6_TCP:
368 pi->pi_netisr = NETISR_IPV6;
370 pi->pi_l3proto = IPPROTO_TCP;
373 case EMX_RXDMRQ_IPV4:
374 if (staterr & E1000_RXD_STAT_IXSM)
378 (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
379 E1000_RXD_STAT_TCPCS) {
380 pi->pi_netisr = NETISR_IP;
382 pi->pi_l3proto = IPPROTO_UDP;
390 m->m_flags |= M_HASH;
391 m->m_pkthdr.hash = toeplitz_hash(hash);
396 emx_probe(device_t dev)
398 const struct emx_device *d;
401 vid = pci_get_vendor(dev);
402 did = pci_get_device(dev);
404 for (d = emx_devices; d->desc != NULL; ++d) {
405 if (vid == d->vid && did == d->did) {
406 device_set_desc(dev, d->desc);
407 device_set_async_attach(dev, TRUE);
415 emx_attach(device_t dev)
417 struct emx_softc *sc = device_get_softc(dev);
418 struct ifnet *ifp = &sc->arpcom.ac_if;
419 int error = 0, i, throttle, msi_enable;
421 uint16_t eeprom_data, device_id, apme_mask;
422 driver_intr_t *intr_func;
424 int offset, offset_def;
427 lwkt_serialize_init(&sc->main_serialize);
428 lwkt_serialize_init(&sc->tx_serialize);
429 for (i = 0; i < EMX_NRX_RING; ++i)
430 lwkt_serialize_init(&sc->rx_data[i].rx_serialize);
433 sc->serializes[i++] = &sc->main_serialize;
434 sc->serializes[i++] = &sc->tx_serialize;
435 sc->serializes[i++] = &sc->rx_data[0].rx_serialize;
436 sc->serializes[i++] = &sc->rx_data[1].rx_serialize;
437 KKASSERT(i == EMX_NSERIALIZE);
439 callout_init_mp(&sc->timer);
441 sc->dev = sc->osdep.dev = dev;
444 * Determine hardware and mac type
446 sc->hw.vendor_id = pci_get_vendor(dev);
447 sc->hw.device_id = pci_get_device(dev);
448 sc->hw.revision_id = pci_get_revid(dev);
449 sc->hw.subsystem_vendor_id = pci_get_subvendor(dev);
450 sc->hw.subsystem_device_id = pci_get_subdevice(dev);
452 if (e1000_set_mac_type(&sc->hw))
456 * Pullup extra 4bytes into the first data segment, see:
457 * 82571/82572 specification update errata #7
460 * 4bytes instead of 2bytes, which are mentioned in the errata,
461 * are pulled; mainly to keep rest of the data properly aligned.
463 if (sc->hw.mac.type == e1000_82571 || sc->hw.mac.type == e1000_82572)
464 sc->flags |= EMX_FLAG_TSO_PULLEX;
466 /* Enable bus mastering */
467 pci_enable_busmaster(dev);
472 sc->memory_rid = EMX_BAR_MEM;
473 sc->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
474 &sc->memory_rid, RF_ACTIVE);
475 if (sc->memory == NULL) {
476 device_printf(dev, "Unable to allocate bus resource: memory\n");
480 sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->memory);
481 sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->memory);
483 /* XXX This is quite goofy, it is not actually used */
484 sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
487 * Don't enable MSI-X on 82574, see:
488 * 82574 specification update errata #15
490 * Don't enable MSI on 82571/82572, see:
491 * 82571/82572 specification update errata #63
493 msi_enable = emx_msi_enable;
495 (sc->hw.mac.type == e1000_82571 ||
496 sc->hw.mac.type == e1000_82572))
502 sc->intr_type = pci_alloc_1intr(dev, msi_enable,
503 &sc->intr_rid, &intr_flags);
505 if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
508 unshared = device_getenv_int(dev, "irq.unshared", 0);
510 sc->flags |= EMX_FLAG_SHARED_INTR;
512 device_printf(dev, "IRQ shared\n");
514 intr_flags &= ~RF_SHAREABLE;
516 device_printf(dev, "IRQ unshared\n");
520 sc->intr_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->intr_rid,
522 if (sc->intr_res == NULL) {
523 device_printf(dev, "Unable to allocate bus resource: "
529 /* Save PCI command register for Shared Code */
530 sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
531 sc->hw.back = &sc->osdep;
533 /* Do Shared Code initialization */
534 if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
535 device_printf(dev, "Setup of Shared code failed\n");
539 e1000_get_bus_info(&sc->hw);
541 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
542 sc->hw.phy.autoneg_wait_to_complete = FALSE;
543 sc->hw.phy.autoneg_advertised = EMX_AUTONEG_ADV_DEFAULT;
546 * Interrupt throttle rate
548 throttle = device_getenv_int(dev, "int_throttle_ceil",
549 emx_int_throttle_ceil);
551 sc->int_throttle_ceil = 0;
554 throttle = EMX_DEFAULT_ITR;
556 /* Recalculate the tunable value to get the exact frequency. */
557 throttle = 1000000000 / 256 / throttle;
559 /* Upper 16bits of ITR is reserved and should be zero */
560 if (throttle & 0xffff0000)
561 throttle = 1000000000 / 256 / EMX_DEFAULT_ITR;
563 sc->int_throttle_ceil = 1000000000 / 256 / throttle;
566 e1000_init_script_state_82541(&sc->hw, TRUE);
567 e1000_set_tbi_compatibility_82543(&sc->hw, TRUE);
570 if (sc->hw.phy.media_type == e1000_media_type_copper) {
571 sc->hw.phy.mdix = EMX_AUTO_ALL_MODES;
572 sc->hw.phy.disable_polarity_correction = FALSE;
573 sc->hw.phy.ms_type = EMX_MASTER_SLAVE;
576 /* Set the frame limits assuming standard ethernet sized frames. */
577 sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
578 sc->min_frame_size = ETHER_MIN_LEN;
580 /* This controls when hardware reports transmit completion status. */
581 sc->hw.mac.report_tx_early = 1;
583 /* Calculate # of RX rings */
584 sc->rx_ring_cnt = device_getenv_int(dev, "rxr", emx_rxr);
585 sc->rx_ring_cnt = if_ring_count2(sc->rx_ring_cnt, EMX_NRX_RING);
587 /* Allocate RX/TX rings' busdma(9) stuffs */
588 error = emx_dma_alloc(sc);
592 /* Allocate multicast array memory. */
593 sc->mta = kmalloc(ETH_ADDR_LEN * EMX_MCAST_ADDR_MAX,
596 /* Indicate SOL/IDER usage */
597 if (e1000_check_reset_block(&sc->hw)) {
599 "PHY reset is blocked due to SOL/IDER session.\n");
603 * Start from a known state, this is important in reading the
604 * nvm and mac from that.
606 e1000_reset_hw(&sc->hw);
608 /* Make sure we have a good EEPROM before we read from it */
609 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
611 * Some PCI-E parts fail the first check due to
612 * the link being in sleep state, call it again,
613 * if it fails a second time its a real issue.
615 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
617 "The EEPROM Checksum Is Not Valid\n");
623 /* Copy the permanent MAC address out of the EEPROM */
624 if (e1000_read_mac_addr(&sc->hw) < 0) {
625 device_printf(dev, "EEPROM read error while reading MAC"
630 if (!emx_is_valid_eaddr(sc->hw.mac.addr)) {
631 device_printf(dev, "Invalid MAC address\n");
636 /* Determine if we have to control management hardware */
637 if (e1000_enable_mng_pass_thru(&sc->hw))
638 sc->flags |= EMX_FLAG_HAS_MGMT;
643 apme_mask = EMX_EEPROM_APME;
645 switch (sc->hw.mac.type) {
647 sc->flags |= EMX_FLAG_HAS_AMT;
652 case e1000_80003es2lan:
653 if (sc->hw.bus.func == 1) {
654 e1000_read_nvm(&sc->hw,
655 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
657 e1000_read_nvm(&sc->hw,
658 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
663 e1000_read_nvm(&sc->hw,
664 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
667 if (eeprom_data & apme_mask)
668 sc->wol = E1000_WUFC_MAG | E1000_WUFC_MC;
671 * We have the eeprom settings, now apply the special cases
672 * where the eeprom may be wrong or the board won't support
673 * wake on lan on a particular port
675 device_id = pci_get_device(dev);
677 case E1000_DEV_ID_82571EB_FIBER:
679 * Wake events only supported on port A for dual fiber
680 * regardless of eeprom setting
682 if (E1000_READ_REG(&sc->hw, E1000_STATUS) &
687 case E1000_DEV_ID_82571EB_QUAD_COPPER:
688 case E1000_DEV_ID_82571EB_QUAD_FIBER:
689 case E1000_DEV_ID_82571EB_QUAD_COPPER_LP:
690 /* if quad port sc, disable WoL on all but port A */
691 if (emx_global_quad_port_a != 0)
693 /* Reset for multiple quad port adapters */
694 if (++emx_global_quad_port_a == 4)
695 emx_global_quad_port_a = 0;
699 /* XXX disable wol */
704 * NPOLLING RX CPU offset
706 if (sc->rx_ring_cnt == ncpus2) {
709 offset_def = (sc->rx_ring_cnt * device_get_unit(dev)) % ncpus2;
710 offset = device_getenv_int(dev, "npoll.rxoff", offset_def);
711 if (offset >= ncpus2 ||
712 offset % sc->rx_ring_cnt != 0) {
713 device_printf(dev, "invalid npoll.rxoff %d, use %d\n",
718 sc->rx_npoll_off = offset;
721 * NPOLLING TX CPU offset
723 offset_def = sc->rx_npoll_off;
724 offset = device_getenv_int(dev, "npoll.txoff", offset_def);
725 if (offset >= ncpus2) {
726 device_printf(dev, "invalid npoll.txoff %d, use %d\n",
730 sc->tx_npoll_off = offset;
733 /* Setup OS specific network interface */
736 /* Add sysctl tree, must after em_setup_ifp() */
739 /* Reset the hardware */
740 error = emx_reset(sc);
742 device_printf(dev, "Unable to reset the hardware\n");
746 /* Initialize statistics */
747 emx_update_stats(sc);
749 sc->hw.mac.get_link_status = 1;
750 emx_update_link_status(sc);
752 sc->spare_tx_desc = EMX_TX_SPARE;
755 * Keep following relationship between spare_tx_desc, oact_tx_desc
757 * (spare_tx_desc + EMX_TX_RESERVED) <=
758 * oact_tx_desc <= EMX_TX_OACTIVE_MAX <= tx_int_nsegs
760 sc->oact_tx_desc = sc->num_tx_desc / 8;
761 if (sc->oact_tx_desc > EMX_TX_OACTIVE_MAX)
762 sc->oact_tx_desc = EMX_TX_OACTIVE_MAX;
763 if (sc->oact_tx_desc < sc->spare_tx_desc + EMX_TX_RESERVED)
764 sc->oact_tx_desc = sc->spare_tx_desc + EMX_TX_RESERVED;
766 sc->tx_int_nsegs = sc->num_tx_desc / 16;
767 if (sc->tx_int_nsegs < sc->oact_tx_desc)
768 sc->tx_int_nsegs = sc->oact_tx_desc;
770 /* Non-AMT based hardware can now take control from firmware */
771 if ((sc->flags & (EMX_FLAG_HAS_MGMT | EMX_FLAG_HAS_AMT)) ==
773 emx_get_hw_control(sc);
776 * Missing Interrupt Following ICR read:
778 * 82571/82572 specification update errata #76
779 * 82573 specification update errata #31
780 * 82574 specification update errata #12
782 intr_func = emx_intr;
783 if ((sc->flags & EMX_FLAG_SHARED_INTR) &&
784 (sc->hw.mac.type == e1000_82571 ||
785 sc->hw.mac.type == e1000_82572 ||
786 sc->hw.mac.type == e1000_82573 ||
787 sc->hw.mac.type == e1000_82574))
788 intr_func = emx_intr_mask;
790 error = bus_setup_intr(dev, sc->intr_res, INTR_MPSAFE, intr_func, sc,
791 &sc->intr_tag, &sc->main_serialize);
793 device_printf(dev, "Failed to register interrupt handler");
794 ether_ifdetach(&sc->arpcom.ac_if);
798 ifp->if_cpuid = rman_get_cpuid(sc->intr_res);
799 KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
807 emx_detach(device_t dev)
809 struct emx_softc *sc = device_get_softc(dev);
811 if (device_is_attached(dev)) {
812 struct ifnet *ifp = &sc->arpcom.ac_if;
814 ifnet_serialize_all(ifp);
818 e1000_phy_hw_reset(&sc->hw);
821 emx_rel_hw_control(sc);
824 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
825 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
829 bus_teardown_intr(dev, sc->intr_res, sc->intr_tag);
831 ifnet_deserialize_all(ifp);
834 } else if (sc->memory != NULL) {
835 emx_rel_hw_control(sc);
837 bus_generic_detach(dev);
839 if (sc->intr_res != NULL) {
840 bus_release_resource(dev, SYS_RES_IRQ, sc->intr_rid,
844 if (sc->intr_type == PCI_INTR_TYPE_MSI)
845 pci_release_msi(dev);
847 if (sc->memory != NULL) {
848 bus_release_resource(dev, SYS_RES_MEMORY, sc->memory_rid,
854 /* Free sysctl tree */
855 if (sc->sysctl_tree != NULL)
856 sysctl_ctx_free(&sc->sysctl_ctx);
859 kfree(sc->mta, M_DEVBUF);
865 emx_shutdown(device_t dev)
867 return emx_suspend(dev);
871 emx_suspend(device_t dev)
873 struct emx_softc *sc = device_get_softc(dev);
874 struct ifnet *ifp = &sc->arpcom.ac_if;
876 ifnet_serialize_all(ifp);
881 emx_rel_hw_control(sc);
884 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
885 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
889 ifnet_deserialize_all(ifp);
891 return bus_generic_suspend(dev);
895 emx_resume(device_t dev)
897 struct emx_softc *sc = device_get_softc(dev);
898 struct ifnet *ifp = &sc->arpcom.ac_if;
900 ifnet_serialize_all(ifp);
906 ifnet_deserialize_all(ifp);
908 return bus_generic_resume(dev);
912 emx_start(struct ifnet *ifp)
914 struct emx_softc *sc = ifp->if_softc;
917 ASSERT_SERIALIZED(&sc->tx_serialize);
919 if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
922 if (!sc->link_active) {
923 ifq_purge(&ifp->if_snd);
927 while (!ifq_is_empty(&ifp->if_snd)) {
928 /* Now do we at least have a minimal? */
929 if (EMX_IS_OACTIVE(sc)) {
931 if (EMX_IS_OACTIVE(sc)) {
932 ifp->if_flags |= IFF_OACTIVE;
933 sc->no_tx_desc_avail1++;
939 m_head = ifq_dequeue(&ifp->if_snd, NULL);
943 if (emx_encap(sc, &m_head)) {
949 /* Send a copy of the frame to the BPF listener */
950 ETHER_BPF_MTAP(ifp, m_head);
952 /* Set timeout in case hardware has problems transmitting. */
953 ifp->if_timer = EMX_TX_TIMEOUT;
958 emx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
960 struct emx_softc *sc = ifp->if_softc;
961 struct ifreq *ifr = (struct ifreq *)data;
962 uint16_t eeprom_data = 0;
963 int max_frame_size, mask, reinit;
966 ASSERT_IFNET_SERIALIZED_ALL(ifp);
970 switch (sc->hw.mac.type) {
973 * 82573 only supports jumbo frames
974 * if ASPM is disabled.
976 e1000_read_nvm(&sc->hw, NVM_INIT_3GIO_3, 1,
978 if (eeprom_data & NVM_WORD1A_ASPM_MASK) {
979 max_frame_size = ETHER_MAX_LEN;
984 /* Limit Jumbo Frame size */
988 case e1000_80003es2lan:
989 max_frame_size = 9234;
993 max_frame_size = MAX_JUMBO_FRAME_SIZE;
996 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
1002 ifp->if_mtu = ifr->ifr_mtu;
1003 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
1006 if (ifp->if_flags & IFF_RUNNING)
1011 if (ifp->if_flags & IFF_UP) {
1012 if ((ifp->if_flags & IFF_RUNNING)) {
1013 if ((ifp->if_flags ^ sc->if_flags) &
1014 (IFF_PROMISC | IFF_ALLMULTI)) {
1015 emx_disable_promisc(sc);
1016 emx_set_promisc(sc);
1021 } else if (ifp->if_flags & IFF_RUNNING) {
1024 sc->if_flags = ifp->if_flags;
1029 if (ifp->if_flags & IFF_RUNNING) {
1030 emx_disable_intr(sc);
1032 #ifdef IFPOLL_ENABLE
1033 if (!(ifp->if_flags & IFF_NPOLLING))
1035 emx_enable_intr(sc);
1040 /* Check SOL/IDER usage */
1041 if (e1000_check_reset_block(&sc->hw)) {
1042 device_printf(sc->dev, "Media change is"
1043 " blocked due to SOL/IDER session.\n");
1049 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
1054 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1055 if (mask & IFCAP_RXCSUM) {
1056 ifp->if_capenable ^= IFCAP_RXCSUM;
1059 if (mask & IFCAP_VLAN_HWTAGGING) {
1060 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1063 if (mask & IFCAP_TXCSUM) {
1064 ifp->if_capenable ^= IFCAP_TXCSUM;
1065 if (ifp->if_capenable & IFCAP_TXCSUM)
1066 ifp->if_hwassist |= EMX_CSUM_FEATURES;
1068 ifp->if_hwassist &= ~EMX_CSUM_FEATURES;
1070 if (mask & IFCAP_TSO) {
1071 ifp->if_capenable ^= IFCAP_TSO;
1072 if (ifp->if_capenable & IFCAP_TSO)
1073 ifp->if_hwassist |= CSUM_TSO;
1075 ifp->if_hwassist &= ~CSUM_TSO;
1077 if (mask & IFCAP_RSS)
1078 ifp->if_capenable ^= IFCAP_RSS;
1079 if (reinit && (ifp->if_flags & IFF_RUNNING))
1084 error = ether_ioctl(ifp, command, data);
1091 emx_watchdog(struct ifnet *ifp)
1093 struct emx_softc *sc = ifp->if_softc;
1095 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1098 * The timer is set to 5 every time start queues a packet.
1099 * Then txeof keeps resetting it as long as it cleans at
1100 * least one descriptor.
1101 * Finally, anytime all descriptors are clean the timer is
1105 if (E1000_READ_REG(&sc->hw, E1000_TDT(0)) ==
1106 E1000_READ_REG(&sc->hw, E1000_TDH(0))) {
1108 * If we reach here, all TX jobs are completed and
1109 * the TX engine should have been idled for some time.
1110 * We don't need to call if_devstart() here.
1112 ifp->if_flags &= ~IFF_OACTIVE;
1118 * If we are in this routine because of pause frames, then
1119 * don't reset the hardware.
1121 if (E1000_READ_REG(&sc->hw, E1000_STATUS) & E1000_STATUS_TXOFF) {
1122 ifp->if_timer = EMX_TX_TIMEOUT;
1126 if (e1000_check_for_link(&sc->hw) == 0)
1127 if_printf(ifp, "watchdog timeout -- resetting\n");
1130 sc->watchdog_events++;
1134 if (!ifq_is_empty(&ifp->if_snd))
1141 struct emx_softc *sc = xsc;
1142 struct ifnet *ifp = &sc->arpcom.ac_if;
1143 device_t dev = sc->dev;
1147 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1152 * Packet Buffer Allocation (PBA)
1153 * Writing PBA sets the receive portion of the buffer
1154 * the remainder is used for the transmit buffer.
1156 switch (sc->hw.mac.type) {
1157 /* Total Packet Buffer on these is 48K */
1160 case e1000_80003es2lan:
1161 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1164 case e1000_82573: /* 82573: Total Packet Buffer is 32K */
1165 pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
1169 pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
1173 /* Devices before 82547 had a Packet Buffer of 64K. */
1174 if (sc->max_frame_size > 8192)
1175 pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
1177 pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
1179 E1000_WRITE_REG(&sc->hw, E1000_PBA, pba);
1181 /* Get the latest mac address, User can use a LAA */
1182 bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
1184 /* Put the address into the Receive Address Array */
1185 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
1188 * With the 82571 sc, RAR[0] may be overwritten
1189 * when the other port is reset, we make a duplicate
1190 * in RAR[14] for that eventuality, this assures
1191 * the interface continues to function.
1193 if (sc->hw.mac.type == e1000_82571) {
1194 e1000_set_laa_state_82571(&sc->hw, TRUE);
1195 e1000_rar_set(&sc->hw, sc->hw.mac.addr,
1196 E1000_RAR_ENTRIES - 1);
1199 /* Initialize the hardware */
1200 if (emx_reset(sc)) {
1201 device_printf(dev, "Unable to reset the hardware\n");
1202 /* XXX emx_stop()? */
1205 emx_update_link_status(sc);
1207 /* Setup VLAN support, basic and offload if available */
1208 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1210 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1213 ctrl = E1000_READ_REG(&sc->hw, E1000_CTRL);
1214 ctrl |= E1000_CTRL_VME;
1215 E1000_WRITE_REG(&sc->hw, E1000_CTRL, ctrl);
1218 /* Configure for OS presence */
1221 /* Prepare transmit descriptors and buffers */
1222 emx_init_tx_ring(sc);
1223 emx_init_tx_unit(sc);
1225 /* Setup Multicast table */
1228 /* Prepare receive descriptors and buffers */
1229 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1230 if (emx_init_rx_ring(sc, &sc->rx_data[i])) {
1232 "Could not setup receive structures\n");
1237 emx_init_rx_unit(sc);
1239 /* Don't lose promiscuous settings */
1240 emx_set_promisc(sc);
1242 ifp->if_flags |= IFF_RUNNING;
1243 ifp->if_flags &= ~IFF_OACTIVE;
1245 callout_reset(&sc->timer, hz, emx_timer, sc);
1246 e1000_clear_hw_cntrs_base_generic(&sc->hw);
1248 /* MSI/X configuration for 82574 */
1249 if (sc->hw.mac.type == e1000_82574) {
1252 tmp = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
1253 tmp |= E1000_CTRL_EXT_PBA_CLR;
1254 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT, tmp);
1257 * Set the IVAR - interrupt vector routing.
1258 * Each nibble represents a vector, high bit
1259 * is enable, other 3 bits are the MSIX table
1260 * entry, we map RXQ0 to 0, TXQ0 to 1, and
1261 * Link (other) to 2, hence the magic number.
1263 E1000_WRITE_REG(&sc->hw, E1000_IVAR, 0x800A0908);
1266 #ifdef IFPOLL_ENABLE
1268 * Only enable interrupts if we are not polling, make sure
1269 * they are off otherwise.
1271 if (ifp->if_flags & IFF_NPOLLING)
1272 emx_disable_intr(sc);
1274 #endif /* IFPOLL_ENABLE */
1275 emx_enable_intr(sc);
1277 /* AMT based hardware can now take control from firmware */
1278 if ((sc->flags & (EMX_FLAG_HAS_MGMT | EMX_FLAG_HAS_AMT)) ==
1279 (EMX_FLAG_HAS_MGMT | EMX_FLAG_HAS_AMT))
1280 emx_get_hw_control(sc);
1282 /* Don't reset the phy next time init gets called */
1283 sc->hw.phy.reset_disable = TRUE;
1289 emx_intr_body(xsc, TRUE);
1293 emx_intr_body(struct emx_softc *sc, boolean_t chk_asserted)
1295 struct ifnet *ifp = &sc->arpcom.ac_if;
1299 ASSERT_SERIALIZED(&sc->main_serialize);
1301 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
1303 if (chk_asserted && (reg_icr & E1000_ICR_INT_ASSERTED) == 0) {
1309 * XXX: some laptops trigger several spurious interrupts
1310 * on emx(4) when in the resume cycle. The ICR register
1311 * reports all-ones value in this case. Processing such
1312 * interrupts would lead to a freeze. I don't know why.
1314 if (reg_icr == 0xffffffff) {
1319 if (ifp->if_flags & IFF_RUNNING) {
1321 (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
1324 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1325 lwkt_serialize_enter(
1326 &sc->rx_data[i].rx_serialize);
1327 emx_rxeof(sc, i, -1);
1328 lwkt_serialize_exit(
1329 &sc->rx_data[i].rx_serialize);
1332 if (reg_icr & E1000_ICR_TXDW) {
1333 lwkt_serialize_enter(&sc->tx_serialize);
1335 if (!ifq_is_empty(&ifp->if_snd))
1337 lwkt_serialize_exit(&sc->tx_serialize);
1341 /* Link status change */
1342 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1343 emx_serialize_skipmain(sc);
1345 callout_stop(&sc->timer);
1346 sc->hw.mac.get_link_status = 1;
1347 emx_update_link_status(sc);
1349 /* Deal with TX cruft when link lost */
1352 callout_reset(&sc->timer, hz, emx_timer, sc);
1354 emx_deserialize_skipmain(sc);
1357 if (reg_icr & E1000_ICR_RXO)
1364 emx_intr_mask(void *xsc)
1366 struct emx_softc *sc = xsc;
1368 E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
1371 * ICR.INT_ASSERTED bit will never be set if IMS is 0,
1372 * so don't check it.
1374 emx_intr_body(sc, FALSE);
1375 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
1379 emx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1381 struct emx_softc *sc = ifp->if_softc;
1383 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1385 emx_update_link_status(sc);
1387 ifmr->ifm_status = IFM_AVALID;
1388 ifmr->ifm_active = IFM_ETHER;
1390 if (!sc->link_active)
1393 ifmr->ifm_status |= IFM_ACTIVE;
1395 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1396 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1397 ifmr->ifm_active |= IFM_1000_SX | IFM_FDX;
1399 switch (sc->link_speed) {
1401 ifmr->ifm_active |= IFM_10_T;
1404 ifmr->ifm_active |= IFM_100_TX;
1408 ifmr->ifm_active |= IFM_1000_T;
1411 if (sc->link_duplex == FULL_DUPLEX)
1412 ifmr->ifm_active |= IFM_FDX;
1414 ifmr->ifm_active |= IFM_HDX;
1419 emx_media_change(struct ifnet *ifp)
1421 struct emx_softc *sc = ifp->if_softc;
1422 struct ifmedia *ifm = &sc->media;
1424 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1426 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1429 switch (IFM_SUBTYPE(ifm->ifm_media)) {
1431 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
1432 sc->hw.phy.autoneg_advertised = EMX_AUTONEG_ADV_DEFAULT;
1438 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
1439 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1443 sc->hw.mac.autoneg = FALSE;
1444 sc->hw.phy.autoneg_advertised = 0;
1445 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1446 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1448 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1452 sc->hw.mac.autoneg = FALSE;
1453 sc->hw.phy.autoneg_advertised = 0;
1454 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1455 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1457 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1461 if_printf(ifp, "Unsupported media type\n");
1466 * As the speed/duplex settings my have changed we need to
1469 sc->hw.phy.reset_disable = FALSE;
1477 emx_encap(struct emx_softc *sc, struct mbuf **m_headp)
1479 bus_dma_segment_t segs[EMX_MAX_SCATTER];
1481 struct emx_txbuf *tx_buffer, *tx_buffer_mapped;
1482 struct e1000_tx_desc *ctxd = NULL;
1483 struct mbuf *m_head = *m_headp;
1484 uint32_t txd_upper, txd_lower, cmd = 0;
1485 int maxsegs, nsegs, i, j, first, last = 0, error;
1487 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
1488 error = emx_tso_pullup(sc, m_headp);
1494 txd_upper = txd_lower = 0;
1497 * Capture the first descriptor index, this descriptor
1498 * will have the index of the EOP which is the only one
1499 * that now gets a DONE bit writeback.
1501 first = sc->next_avail_tx_desc;
1502 tx_buffer = &sc->tx_buf[first];
1503 tx_buffer_mapped = tx_buffer;
1504 map = tx_buffer->map;
1506 maxsegs = sc->num_tx_desc_avail - EMX_TX_RESERVED;
1507 KASSERT(maxsegs >= sc->spare_tx_desc, ("not enough spare TX desc"));
1508 if (maxsegs > EMX_MAX_SCATTER)
1509 maxsegs = EMX_MAX_SCATTER;
1511 error = bus_dmamap_load_mbuf_defrag(sc->txtag, map, m_headp,
1512 segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1514 if (error == ENOBUFS)
1515 sc->mbuf_alloc_failed++;
1517 sc->no_tx_dma_setup++;
1523 bus_dmamap_sync(sc->txtag, map, BUS_DMASYNC_PREWRITE);
1526 sc->tx_nsegs += nsegs;
1528 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
1529 /* TSO will consume one TX desc */
1530 sc->tx_nsegs += emx_tso_setup(sc, m_head,
1531 &txd_upper, &txd_lower);
1532 } else if (m_head->m_pkthdr.csum_flags & EMX_CSUM_FEATURES) {
1533 /* TX csum offloading will consume one TX desc */
1534 sc->tx_nsegs += emx_txcsum(sc, m_head, &txd_upper, &txd_lower);
1536 i = sc->next_avail_tx_desc;
1538 /* Set up our transmit descriptors */
1539 for (j = 0; j < nsegs; j++) {
1540 tx_buffer = &sc->tx_buf[i];
1541 ctxd = &sc->tx_desc_base[i];
1543 ctxd->buffer_addr = htole64(segs[j].ds_addr);
1544 ctxd->lower.data = htole32(E1000_TXD_CMD_IFCS |
1545 txd_lower | segs[j].ds_len);
1546 ctxd->upper.data = htole32(txd_upper);
1549 if (++i == sc->num_tx_desc)
1553 sc->next_avail_tx_desc = i;
1555 KKASSERT(sc->num_tx_desc_avail > nsegs);
1556 sc->num_tx_desc_avail -= nsegs;
1558 /* Handle VLAN tag */
1559 if (m_head->m_flags & M_VLANTAG) {
1560 /* Set the vlan id. */
1561 ctxd->upper.fields.special =
1562 htole16(m_head->m_pkthdr.ether_vlantag);
1564 /* Tell hardware to add tag */
1565 ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE);
1568 tx_buffer->m_head = m_head;
1569 tx_buffer_mapped->map = tx_buffer->map;
1570 tx_buffer->map = map;
1572 if (sc->tx_nsegs >= sc->tx_int_nsegs) {
1576 * Report Status (RS) is turned on
1577 * every tx_int_nsegs descriptors.
1579 cmd = E1000_TXD_CMD_RS;
1582 * Keep track of the descriptor, which will
1583 * be written back by hardware.
1585 sc->tx_dd[sc->tx_dd_tail] = last;
1586 EMX_INC_TXDD_IDX(sc->tx_dd_tail);
1587 KKASSERT(sc->tx_dd_tail != sc->tx_dd_head);
1591 * Last Descriptor of Packet needs End Of Packet (EOP)
1593 ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | cmd);
1596 * Advance the Transmit Descriptor Tail (TDT), this tells
1597 * the E1000 that this frame is available to transmit.
1599 E1000_WRITE_REG(&sc->hw, E1000_TDT(0), i);
1605 emx_set_promisc(struct emx_softc *sc)
1607 struct ifnet *ifp = &sc->arpcom.ac_if;
1610 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1612 if (ifp->if_flags & IFF_PROMISC) {
1613 reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1614 /* Turn this on if you want to see bad packets */
1616 reg_rctl |= E1000_RCTL_SBP;
1617 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1618 } else if (ifp->if_flags & IFF_ALLMULTI) {
1619 reg_rctl |= E1000_RCTL_MPE;
1620 reg_rctl &= ~E1000_RCTL_UPE;
1621 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1626 emx_disable_promisc(struct emx_softc *sc)
1630 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1632 reg_rctl &= ~E1000_RCTL_UPE;
1633 reg_rctl &= ~E1000_RCTL_MPE;
1634 reg_rctl &= ~E1000_RCTL_SBP;
1635 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1639 emx_set_multi(struct emx_softc *sc)
1641 struct ifnet *ifp = &sc->arpcom.ac_if;
1642 struct ifmultiaddr *ifma;
1643 uint32_t reg_rctl = 0;
1648 bzero(mta, ETH_ADDR_LEN * EMX_MCAST_ADDR_MAX);
1650 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1651 if (ifma->ifma_addr->sa_family != AF_LINK)
1654 if (mcnt == EMX_MCAST_ADDR_MAX)
1657 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1658 &mta[mcnt * ETHER_ADDR_LEN], ETHER_ADDR_LEN);
1662 if (mcnt >= EMX_MCAST_ADDR_MAX) {
1663 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1664 reg_rctl |= E1000_RCTL_MPE;
1665 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1667 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1672 * This routine checks for link status and updates statistics.
1675 emx_timer(void *xsc)
1677 struct emx_softc *sc = xsc;
1678 struct ifnet *ifp = &sc->arpcom.ac_if;
1680 lwkt_serialize_enter(&sc->main_serialize);
1682 emx_update_link_status(sc);
1683 emx_update_stats(sc);
1685 /* Reset LAA into RAR[0] on 82571 */
1686 if (e1000_get_laa_state_82571(&sc->hw) == TRUE)
1687 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
1689 if (emx_display_debug_stats && (ifp->if_flags & IFF_RUNNING))
1690 emx_print_hw_stats(sc);
1694 callout_reset(&sc->timer, hz, emx_timer, sc);
1696 lwkt_serialize_exit(&sc->main_serialize);
1700 emx_update_link_status(struct emx_softc *sc)
1702 struct e1000_hw *hw = &sc->hw;
1703 struct ifnet *ifp = &sc->arpcom.ac_if;
1704 device_t dev = sc->dev;
1705 uint32_t link_check = 0;
1707 /* Get the cached link value or read phy for real */
1708 switch (hw->phy.media_type) {
1709 case e1000_media_type_copper:
1710 if (hw->mac.get_link_status) {
1711 /* Do the work to read phy */
1712 e1000_check_for_link(hw);
1713 link_check = !hw->mac.get_link_status;
1714 if (link_check) /* ESB2 fix */
1715 e1000_cfg_on_link_up(hw);
1721 case e1000_media_type_fiber:
1722 e1000_check_for_link(hw);
1723 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1726 case e1000_media_type_internal_serdes:
1727 e1000_check_for_link(hw);
1728 link_check = sc->hw.mac.serdes_has_link;
1731 case e1000_media_type_unknown:
1736 /* Now check for a transition */
1737 if (link_check && sc->link_active == 0) {
1738 e1000_get_speed_and_duplex(hw, &sc->link_speed,
1742 * Check if we should enable/disable SPEED_MODE bit on
1745 if (sc->link_speed != SPEED_1000 &&
1746 (hw->mac.type == e1000_82571 ||
1747 hw->mac.type == e1000_82572)) {
1750 tarc0 = E1000_READ_REG(hw, E1000_TARC(0));
1751 tarc0 &= ~EMX_TARC_SPEED_MODE;
1752 E1000_WRITE_REG(hw, E1000_TARC(0), tarc0);
1755 device_printf(dev, "Link is up %d Mbps %s\n",
1757 ((sc->link_duplex == FULL_DUPLEX) ?
1758 "Full Duplex" : "Half Duplex"));
1760 sc->link_active = 1;
1762 ifp->if_baudrate = sc->link_speed * 1000000;
1763 ifp->if_link_state = LINK_STATE_UP;
1764 if_link_state_change(ifp);
1765 } else if (!link_check && sc->link_active == 1) {
1766 ifp->if_baudrate = sc->link_speed = 0;
1767 sc->link_duplex = 0;
1769 device_printf(dev, "Link is Down\n");
1770 sc->link_active = 0;
1772 /* Link down, disable watchdog */
1775 ifp->if_link_state = LINK_STATE_DOWN;
1776 if_link_state_change(ifp);
1781 emx_stop(struct emx_softc *sc)
1783 struct ifnet *ifp = &sc->arpcom.ac_if;
1786 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1788 emx_disable_intr(sc);
1790 callout_stop(&sc->timer);
1792 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1796 * Disable multiple receive queues.
1799 * We should disable multiple receive queues before
1800 * resetting the hardware.
1802 E1000_WRITE_REG(&sc->hw, E1000_MRQC, 0);
1804 e1000_reset_hw(&sc->hw);
1805 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1807 for (i = 0; i < sc->num_tx_desc; i++) {
1808 struct emx_txbuf *tx_buffer = &sc->tx_buf[i];
1810 if (tx_buffer->m_head != NULL) {
1811 bus_dmamap_unload(sc->txtag, tx_buffer->map);
1812 m_freem(tx_buffer->m_head);
1813 tx_buffer->m_head = NULL;
1817 for (i = 0; i < sc->rx_ring_cnt; ++i)
1818 emx_free_rx_ring(sc, &sc->rx_data[i]);
1822 sc->csum_iphlen = 0;
1825 sc->csum_pktlen = 0;
1833 emx_reset(struct emx_softc *sc)
1835 device_t dev = sc->dev;
1836 uint16_t rx_buffer_size;
1838 /* Set up smart power down as default off on newer adapters. */
1839 if (!emx_smart_pwr_down &&
1840 (sc->hw.mac.type == e1000_82571 ||
1841 sc->hw.mac.type == e1000_82572)) {
1842 uint16_t phy_tmp = 0;
1844 /* Speed up time to link by disabling smart power down. */
1845 e1000_read_phy_reg(&sc->hw,
1846 IGP02E1000_PHY_POWER_MGMT, &phy_tmp);
1847 phy_tmp &= ~IGP02E1000_PM_SPD;
1848 e1000_write_phy_reg(&sc->hw,
1849 IGP02E1000_PHY_POWER_MGMT, phy_tmp);
1853 * These parameters control the automatic generation (Tx) and
1854 * response (Rx) to Ethernet PAUSE frames.
1855 * - High water mark should allow for at least two frames to be
1856 * received after sending an XOFF.
1857 * - Low water mark works best when it is very near the high water mark.
1858 * This allows the receiver to restart by sending XON when it has
1859 * drained a bit. Here we use an arbitary value of 1500 which will
1860 * restart after one full frame is pulled from the buffer. There
1861 * could be several smaller frames in the buffer and if so they will
1862 * not trigger the XON until their total number reduces the buffer
1864 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1866 rx_buffer_size = (E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff) << 10;
1868 sc->hw.fc.high_water = rx_buffer_size -
1869 roundup2(sc->max_frame_size, 1024);
1870 sc->hw.fc.low_water = sc->hw.fc.high_water - 1500;
1872 if (sc->hw.mac.type == e1000_80003es2lan)
1873 sc->hw.fc.pause_time = 0xFFFF;
1875 sc->hw.fc.pause_time = EMX_FC_PAUSE_TIME;
1876 sc->hw.fc.send_xon = TRUE;
1877 sc->hw.fc.requested_mode = e1000_fc_full;
1879 /* Issue a global reset */
1880 e1000_reset_hw(&sc->hw);
1881 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1882 emx_disable_aspm(sc);
1884 if (e1000_init_hw(&sc->hw) < 0) {
1885 device_printf(dev, "Hardware Initialization Failed\n");
1889 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1890 e1000_get_phy_info(&sc->hw);
1891 e1000_check_for_link(&sc->hw);
1897 emx_setup_ifp(struct emx_softc *sc)
1899 struct ifnet *ifp = &sc->arpcom.ac_if;
1901 if_initname(ifp, device_get_name(sc->dev),
1902 device_get_unit(sc->dev));
1904 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1905 ifp->if_init = emx_init;
1906 ifp->if_ioctl = emx_ioctl;
1907 ifp->if_start = emx_start;
1908 #ifdef IFPOLL_ENABLE
1909 ifp->if_npoll = emx_npoll;
1911 ifp->if_watchdog = emx_watchdog;
1912 ifp->if_serialize = emx_serialize;
1913 ifp->if_deserialize = emx_deserialize;
1914 ifp->if_tryserialize = emx_tryserialize;
1916 ifp->if_serialize_assert = emx_serialize_assert;
1918 ifq_set_maxlen(&ifp->if_snd, sc->num_tx_desc - 1);
1919 ifq_set_ready(&ifp->if_snd);
1921 ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1923 ifp->if_capabilities = IFCAP_HWCSUM |
1924 IFCAP_VLAN_HWTAGGING |
1927 if (sc->rx_ring_cnt > 1)
1928 ifp->if_capabilities |= IFCAP_RSS;
1929 ifp->if_capenable = ifp->if_capabilities;
1930 ifp->if_hwassist = EMX_CSUM_FEATURES | CSUM_TSO;
1933 * Tell the upper layer(s) we support long frames.
1935 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1938 * Specify the media types supported by this sc and register
1939 * callbacks to update media and link information
1941 ifmedia_init(&sc->media, IFM_IMASK,
1942 emx_media_change, emx_media_status);
1943 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1944 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1945 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1947 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1949 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1950 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1952 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1953 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1955 if (sc->hw.phy.type != e1000_phy_ife) {
1956 ifmedia_add(&sc->media,
1957 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1958 ifmedia_add(&sc->media,
1959 IFM_ETHER | IFM_1000_T, 0, NULL);
1962 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1963 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1967 * Workaround for SmartSpeed on 82541 and 82547 controllers
1970 emx_smartspeed(struct emx_softc *sc)
1974 if (sc->link_active || sc->hw.phy.type != e1000_phy_igp ||
1975 sc->hw.mac.autoneg == 0 ||
1976 (sc->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0)
1979 if (sc->smartspeed == 0) {
1981 * If Master/Slave config fault is asserted twice,
1982 * we assume back-to-back
1984 e1000_read_phy_reg(&sc->hw, PHY_1000T_STATUS, &phy_tmp);
1985 if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
1987 e1000_read_phy_reg(&sc->hw, PHY_1000T_STATUS, &phy_tmp);
1988 if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
1989 e1000_read_phy_reg(&sc->hw,
1990 PHY_1000T_CTRL, &phy_tmp);
1991 if (phy_tmp & CR_1000T_MS_ENABLE) {
1992 phy_tmp &= ~CR_1000T_MS_ENABLE;
1993 e1000_write_phy_reg(&sc->hw,
1994 PHY_1000T_CTRL, phy_tmp);
1996 if (sc->hw.mac.autoneg &&
1997 !e1000_phy_setup_autoneg(&sc->hw) &&
1998 !e1000_read_phy_reg(&sc->hw,
1999 PHY_CONTROL, &phy_tmp)) {
2000 phy_tmp |= MII_CR_AUTO_NEG_EN |
2001 MII_CR_RESTART_AUTO_NEG;
2002 e1000_write_phy_reg(&sc->hw,
2003 PHY_CONTROL, phy_tmp);
2008 } else if (sc->smartspeed == EMX_SMARTSPEED_DOWNSHIFT) {
2009 /* If still no link, perhaps using 2/3 pair cable */
2010 e1000_read_phy_reg(&sc->hw, PHY_1000T_CTRL, &phy_tmp);
2011 phy_tmp |= CR_1000T_MS_ENABLE;
2012 e1000_write_phy_reg(&sc->hw, PHY_1000T_CTRL, phy_tmp);
2013 if (sc->hw.mac.autoneg &&
2014 !e1000_phy_setup_autoneg(&sc->hw) &&
2015 !e1000_read_phy_reg(&sc->hw, PHY_CONTROL, &phy_tmp)) {
2016 phy_tmp |= MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG;
2017 e1000_write_phy_reg(&sc->hw, PHY_CONTROL, phy_tmp);
2021 /* Restart process after EMX_SMARTSPEED_MAX iterations */
2022 if (sc->smartspeed++ == EMX_SMARTSPEED_MAX)
2027 emx_create_tx_ring(struct emx_softc *sc)
2029 device_t dev = sc->dev;
2030 struct emx_txbuf *tx_buffer;
2031 int error, i, tsize, ntxd;
2034 * Validate number of transmit descriptors. It must not exceed
2035 * hardware maximum, and must be multiple of E1000_DBA_ALIGN.
2037 ntxd = device_getenv_int(dev, "txd", emx_txd);
2038 if ((ntxd * sizeof(struct e1000_tx_desc)) % EMX_DBA_ALIGN != 0 ||
2039 ntxd > EMX_MAX_TXD || ntxd < EMX_MIN_TXD) {
2040 device_printf(dev, "Using %d TX descriptors instead of %d!\n",
2041 EMX_DEFAULT_TXD, ntxd);
2042 sc->num_tx_desc = EMX_DEFAULT_TXD;
2044 sc->num_tx_desc = ntxd;
2048 * Allocate Transmit Descriptor ring
2050 tsize = roundup2(sc->num_tx_desc * sizeof(struct e1000_tx_desc),
2052 sc->tx_desc_base = bus_dmamem_coherent_any(sc->parent_dtag,
2053 EMX_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
2054 &sc->tx_desc_dtag, &sc->tx_desc_dmap,
2055 &sc->tx_desc_paddr);
2056 if (sc->tx_desc_base == NULL) {
2057 device_printf(dev, "Unable to allocate tx_desc memory\n");
2061 sc->tx_buf = kmalloc(sizeof(struct emx_txbuf) * sc->num_tx_desc,
2062 M_DEVBUF, M_WAITOK | M_ZERO);
2065 * Create DMA tags for tx buffers
2067 error = bus_dma_tag_create(sc->parent_dtag, /* parent */
2068 1, 0, /* alignment, bounds */
2069 BUS_SPACE_MAXADDR, /* lowaddr */
2070 BUS_SPACE_MAXADDR, /* highaddr */
2071 NULL, NULL, /* filter, filterarg */
2072 EMX_TSO_SIZE, /* maxsize */
2073 EMX_MAX_SCATTER, /* nsegments */
2074 EMX_MAX_SEGSIZE, /* maxsegsize */
2075 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
2076 BUS_DMA_ONEBPAGE, /* flags */
2079 device_printf(dev, "Unable to allocate TX DMA tag\n");
2080 kfree(sc->tx_buf, M_DEVBUF);
2086 * Create DMA maps for tx buffers
2088 for (i = 0; i < sc->num_tx_desc; i++) {
2089 tx_buffer = &sc->tx_buf[i];
2091 error = bus_dmamap_create(sc->txtag,
2092 BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
2095 device_printf(dev, "Unable to create TX DMA map\n");
2096 emx_destroy_tx_ring(sc, i);
2104 emx_init_tx_ring(struct emx_softc *sc)
2106 /* Clear the old ring contents */
2107 bzero(sc->tx_desc_base,
2108 sizeof(struct e1000_tx_desc) * sc->num_tx_desc);
2111 sc->next_avail_tx_desc = 0;
2112 sc->next_tx_to_clean = 0;
2113 sc->num_tx_desc_avail = sc->num_tx_desc;
2117 emx_init_tx_unit(struct emx_softc *sc)
2119 uint32_t tctl, tarc, tipg = 0;
2122 /* Setup the Base and Length of the Tx Descriptor Ring */
2123 bus_addr = sc->tx_desc_paddr;
2124 E1000_WRITE_REG(&sc->hw, E1000_TDLEN(0),
2125 sc->num_tx_desc * sizeof(struct e1000_tx_desc));
2126 E1000_WRITE_REG(&sc->hw, E1000_TDBAH(0),
2127 (uint32_t)(bus_addr >> 32));
2128 E1000_WRITE_REG(&sc->hw, E1000_TDBAL(0),
2129 (uint32_t)bus_addr);
2130 /* Setup the HW Tx Head and Tail descriptor pointers */
2131 E1000_WRITE_REG(&sc->hw, E1000_TDT(0), 0);
2132 E1000_WRITE_REG(&sc->hw, E1000_TDH(0), 0);
2134 /* Set the default values for the Tx Inter Packet Gap timer */
2135 switch (sc->hw.mac.type) {
2136 case e1000_80003es2lan:
2137 tipg = DEFAULT_82543_TIPG_IPGR1;
2138 tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 <<
2139 E1000_TIPG_IPGR2_SHIFT;
2143 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
2144 sc->hw.phy.media_type == e1000_media_type_internal_serdes)
2145 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
2147 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
2148 tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2149 tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2153 E1000_WRITE_REG(&sc->hw, E1000_TIPG, tipg);
2155 /* NOTE: 0 is not allowed for TIDV */
2156 E1000_WRITE_REG(&sc->hw, E1000_TIDV, 1);
2157 E1000_WRITE_REG(&sc->hw, E1000_TADV, 0);
2159 if (sc->hw.mac.type == e1000_82571 ||
2160 sc->hw.mac.type == e1000_82572) {
2161 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(0));
2162 tarc |= EMX_TARC_SPEED_MODE;
2163 E1000_WRITE_REG(&sc->hw, E1000_TARC(0), tarc);
2164 } else if (sc->hw.mac.type == e1000_80003es2lan) {
2165 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(0));
2167 E1000_WRITE_REG(&sc->hw, E1000_TARC(0), tarc);
2168 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(1));
2170 E1000_WRITE_REG(&sc->hw, E1000_TARC(1), tarc);
2173 /* Program the Transmit Control Register */
2174 tctl = E1000_READ_REG(&sc->hw, E1000_TCTL);
2175 tctl &= ~E1000_TCTL_CT;
2176 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
2177 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2178 tctl |= E1000_TCTL_MULR;
2180 /* This write will effectively turn on the transmit unit. */
2181 E1000_WRITE_REG(&sc->hw, E1000_TCTL, tctl);
2185 emx_destroy_tx_ring(struct emx_softc *sc, int ndesc)
2187 struct emx_txbuf *tx_buffer;
2190 /* Free Transmit Descriptor ring */
2191 if (sc->tx_desc_base) {
2192 bus_dmamap_unload(sc->tx_desc_dtag, sc->tx_desc_dmap);
2193 bus_dmamem_free(sc->tx_desc_dtag, sc->tx_desc_base,
2195 bus_dma_tag_destroy(sc->tx_desc_dtag);
2197 sc->tx_desc_base = NULL;
2200 if (sc->tx_buf == NULL)
2203 for (i = 0; i < ndesc; i++) {
2204 tx_buffer = &sc->tx_buf[i];
2206 KKASSERT(tx_buffer->m_head == NULL);
2207 bus_dmamap_destroy(sc->txtag, tx_buffer->map);
2209 bus_dma_tag_destroy(sc->txtag);
2211 kfree(sc->tx_buf, M_DEVBUF);
2216 * The offload context needs to be set when we transfer the first
2217 * packet of a particular protocol (TCP/UDP). This routine has been
2218 * enhanced to deal with inserted VLAN headers.
2220 * If the new packet's ether header length, ip header length and
2221 * csum offloading type are same as the previous packet, we should
2222 * avoid allocating a new csum context descriptor; mainly to take
2223 * advantage of the pipeline effect of the TX data read request.
2225 * This function returns number of TX descrptors allocated for
2229 emx_txcsum(struct emx_softc *sc, struct mbuf *mp,
2230 uint32_t *txd_upper, uint32_t *txd_lower)
2232 struct e1000_context_desc *TXD;
2233 int curr_txd, ehdrlen, csum_flags;
2234 uint32_t cmd, hdr_len, ip_hlen;
2236 csum_flags = mp->m_pkthdr.csum_flags & EMX_CSUM_FEATURES;
2237 ip_hlen = mp->m_pkthdr.csum_iphlen;
2238 ehdrlen = mp->m_pkthdr.csum_lhlen;
2240 if (sc->csum_lhlen == ehdrlen && sc->csum_iphlen == ip_hlen &&
2241 sc->csum_flags == csum_flags) {
2243 * Same csum offload context as the previous packets;
2246 *txd_upper = sc->csum_txd_upper;
2247 *txd_lower = sc->csum_txd_lower;
2252 * Setup a new csum offload context.
2255 curr_txd = sc->next_avail_tx_desc;
2256 TXD = (struct e1000_context_desc *)&sc->tx_desc_base[curr_txd];
2260 /* Setup of IP header checksum. */
2261 if (csum_flags & CSUM_IP) {
2263 * Start offset for header checksum calculation.
2264 * End offset for header checksum calculation.
2265 * Offset of place to put the checksum.
2267 TXD->lower_setup.ip_fields.ipcss = ehdrlen;
2268 TXD->lower_setup.ip_fields.ipcse =
2269 htole16(ehdrlen + ip_hlen - 1);
2270 TXD->lower_setup.ip_fields.ipcso =
2271 ehdrlen + offsetof(struct ip, ip_sum);
2272 cmd |= E1000_TXD_CMD_IP;
2273 *txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2275 hdr_len = ehdrlen + ip_hlen;
2277 if (csum_flags & CSUM_TCP) {
2279 * Start offset for payload checksum calculation.
2280 * End offset for payload checksum calculation.
2281 * Offset of place to put the checksum.
2283 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2284 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2285 TXD->upper_setup.tcp_fields.tucso =
2286 hdr_len + offsetof(struct tcphdr, th_sum);
2287 cmd |= E1000_TXD_CMD_TCP;
2288 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2289 } else if (csum_flags & CSUM_UDP) {
2291 * Start offset for header checksum calculation.
2292 * End offset for header checksum calculation.
2293 * Offset of place to put the checksum.
2295 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2296 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2297 TXD->upper_setup.tcp_fields.tucso =
2298 hdr_len + offsetof(struct udphdr, uh_sum);
2299 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2302 *txd_lower = E1000_TXD_CMD_DEXT | /* Extended descr type */
2303 E1000_TXD_DTYP_D; /* Data descr */
2305 /* Save the information for this csum offloading context */
2306 sc->csum_lhlen = ehdrlen;
2307 sc->csum_iphlen = ip_hlen;
2308 sc->csum_flags = csum_flags;
2309 sc->csum_txd_upper = *txd_upper;
2310 sc->csum_txd_lower = *txd_lower;
2312 TXD->tcp_seg_setup.data = htole32(0);
2313 TXD->cmd_and_length =
2314 htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
2316 if (++curr_txd == sc->num_tx_desc)
2319 KKASSERT(sc->num_tx_desc_avail > 0);
2320 sc->num_tx_desc_avail--;
2322 sc->next_avail_tx_desc = curr_txd;
2327 emx_txeof(struct emx_softc *sc)
2329 struct ifnet *ifp = &sc->arpcom.ac_if;
2330 struct emx_txbuf *tx_buffer;
2331 int first, num_avail;
2333 if (sc->tx_dd_head == sc->tx_dd_tail)
2336 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2339 num_avail = sc->num_tx_desc_avail;
2340 first = sc->next_tx_to_clean;
2342 while (sc->tx_dd_head != sc->tx_dd_tail) {
2343 int dd_idx = sc->tx_dd[sc->tx_dd_head];
2344 struct e1000_tx_desc *tx_desc;
2346 tx_desc = &sc->tx_desc_base[dd_idx];
2347 if (tx_desc->upper.fields.status & E1000_TXD_STAT_DD) {
2348 EMX_INC_TXDD_IDX(sc->tx_dd_head);
2350 if (++dd_idx == sc->num_tx_desc)
2353 while (first != dd_idx) {
2358 tx_buffer = &sc->tx_buf[first];
2359 if (tx_buffer->m_head) {
2361 bus_dmamap_unload(sc->txtag,
2363 m_freem(tx_buffer->m_head);
2364 tx_buffer->m_head = NULL;
2367 if (++first == sc->num_tx_desc)
2374 sc->next_tx_to_clean = first;
2375 sc->num_tx_desc_avail = num_avail;
2377 if (sc->tx_dd_head == sc->tx_dd_tail) {
2382 if (!EMX_IS_OACTIVE(sc)) {
2383 ifp->if_flags &= ~IFF_OACTIVE;
2385 /* All clean, turn off the timer */
2386 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2392 emx_tx_collect(struct emx_softc *sc)
2394 struct ifnet *ifp = &sc->arpcom.ac_if;
2395 struct emx_txbuf *tx_buffer;
2396 int tdh, first, num_avail, dd_idx = -1;
2398 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2401 tdh = E1000_READ_REG(&sc->hw, E1000_TDH(0));
2402 if (tdh == sc->next_tx_to_clean)
2405 if (sc->tx_dd_head != sc->tx_dd_tail)
2406 dd_idx = sc->tx_dd[sc->tx_dd_head];
2408 num_avail = sc->num_tx_desc_avail;
2409 first = sc->next_tx_to_clean;
2411 while (first != tdh) {
2416 tx_buffer = &sc->tx_buf[first];
2417 if (tx_buffer->m_head) {
2419 bus_dmamap_unload(sc->txtag,
2421 m_freem(tx_buffer->m_head);
2422 tx_buffer->m_head = NULL;
2425 if (first == dd_idx) {
2426 EMX_INC_TXDD_IDX(sc->tx_dd_head);
2427 if (sc->tx_dd_head == sc->tx_dd_tail) {
2432 dd_idx = sc->tx_dd[sc->tx_dd_head];
2436 if (++first == sc->num_tx_desc)
2439 sc->next_tx_to_clean = first;
2440 sc->num_tx_desc_avail = num_avail;
2442 if (!EMX_IS_OACTIVE(sc)) {
2443 ifp->if_flags &= ~IFF_OACTIVE;
2445 /* All clean, turn off the timer */
2446 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2452 * When Link is lost sometimes there is work still in the TX ring
2453 * which will result in a watchdog, rather than allow that do an
2454 * attempted cleanup and then reinit here. Note that this has been
2455 * seens mostly with fiber adapters.
2458 emx_tx_purge(struct emx_softc *sc)
2460 struct ifnet *ifp = &sc->arpcom.ac_if;
2462 if (!sc->link_active && ifp->if_timer) {
2464 if (ifp->if_timer) {
2465 if_printf(ifp, "Link lost, TX pending, reinit\n");
2473 emx_newbuf(struct emx_softc *sc, struct emx_rxdata *rdata, int i, int init)
2476 bus_dma_segment_t seg;
2478 struct emx_rxbuf *rx_buffer;
2481 m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2483 rdata->mbuf_cluster_failed++;
2485 if_printf(&sc->arpcom.ac_if,
2486 "Unable to allocate RX mbuf\n");
2490 m->m_len = m->m_pkthdr.len = MCLBYTES;
2492 if (sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2493 m_adj(m, ETHER_ALIGN);
2495 error = bus_dmamap_load_mbuf_segment(rdata->rxtag,
2496 rdata->rx_sparemap, m,
2497 &seg, 1, &nseg, BUS_DMA_NOWAIT);
2501 if_printf(&sc->arpcom.ac_if,
2502 "Unable to load RX mbuf\n");
2507 rx_buffer = &rdata->rx_buf[i];
2508 if (rx_buffer->m_head != NULL)
2509 bus_dmamap_unload(rdata->rxtag, rx_buffer->map);
2511 map = rx_buffer->map;
2512 rx_buffer->map = rdata->rx_sparemap;
2513 rdata->rx_sparemap = map;
2515 rx_buffer->m_head = m;
2516 rx_buffer->paddr = seg.ds_addr;
2518 emx_setup_rxdesc(&rdata->rx_desc[i], rx_buffer);
2523 emx_create_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
2525 device_t dev = sc->dev;
2526 struct emx_rxbuf *rx_buffer;
2527 int i, error, rsize, nrxd;
2530 * Validate number of receive descriptors. It must not exceed
2531 * hardware maximum, and must be multiple of E1000_DBA_ALIGN.
2533 nrxd = device_getenv_int(dev, "rxd", emx_rxd);
2534 if ((nrxd * sizeof(emx_rxdesc_t)) % EMX_DBA_ALIGN != 0 ||
2535 nrxd > EMX_MAX_RXD || nrxd < EMX_MIN_RXD) {
2536 device_printf(dev, "Using %d RX descriptors instead of %d!\n",
2537 EMX_DEFAULT_RXD, nrxd);
2538 rdata->num_rx_desc = EMX_DEFAULT_RXD;
2540 rdata->num_rx_desc = nrxd;
2544 * Allocate Receive Descriptor ring
2546 rsize = roundup2(rdata->num_rx_desc * sizeof(emx_rxdesc_t),
2548 rdata->rx_desc = bus_dmamem_coherent_any(sc->parent_dtag,
2549 EMX_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
2550 &rdata->rx_desc_dtag, &rdata->rx_desc_dmap,
2551 &rdata->rx_desc_paddr);
2552 if (rdata->rx_desc == NULL) {
2553 device_printf(dev, "Unable to allocate rx_desc memory\n");
2557 rdata->rx_buf = kmalloc(sizeof(struct emx_rxbuf) * rdata->num_rx_desc,
2558 M_DEVBUF, M_WAITOK | M_ZERO);
2561 * Create DMA tag for rx buffers
2563 error = bus_dma_tag_create(sc->parent_dtag, /* parent */
2564 1, 0, /* alignment, bounds */
2565 BUS_SPACE_MAXADDR, /* lowaddr */
2566 BUS_SPACE_MAXADDR, /* highaddr */
2567 NULL, NULL, /* filter, filterarg */
2568 MCLBYTES, /* maxsize */
2570 MCLBYTES, /* maxsegsize */
2571 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2574 device_printf(dev, "Unable to allocate RX DMA tag\n");
2575 kfree(rdata->rx_buf, M_DEVBUF);
2576 rdata->rx_buf = NULL;
2581 * Create spare DMA map for rx buffers
2583 error = bus_dmamap_create(rdata->rxtag, BUS_DMA_WAITOK,
2584 &rdata->rx_sparemap);
2586 device_printf(dev, "Unable to create spare RX DMA map\n");
2587 bus_dma_tag_destroy(rdata->rxtag);
2588 kfree(rdata->rx_buf, M_DEVBUF);
2589 rdata->rx_buf = NULL;
2594 * Create DMA maps for rx buffers
2596 for (i = 0; i < rdata->num_rx_desc; i++) {
2597 rx_buffer = &rdata->rx_buf[i];
2599 error = bus_dmamap_create(rdata->rxtag, BUS_DMA_WAITOK,
2602 device_printf(dev, "Unable to create RX DMA map\n");
2603 emx_destroy_rx_ring(sc, rdata, i);
2611 emx_free_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
2615 for (i = 0; i < rdata->num_rx_desc; i++) {
2616 struct emx_rxbuf *rx_buffer = &rdata->rx_buf[i];
2618 if (rx_buffer->m_head != NULL) {
2619 bus_dmamap_unload(rdata->rxtag, rx_buffer->map);
2620 m_freem(rx_buffer->m_head);
2621 rx_buffer->m_head = NULL;
2625 if (rdata->fmp != NULL)
2626 m_freem(rdata->fmp);
2632 emx_init_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
2636 /* Reset descriptor ring */
2637 bzero(rdata->rx_desc, sizeof(emx_rxdesc_t) * rdata->num_rx_desc);
2639 /* Allocate new ones. */
2640 for (i = 0; i < rdata->num_rx_desc; i++) {
2641 error = emx_newbuf(sc, rdata, i, 1);
2646 /* Setup our descriptor pointers */
2647 rdata->next_rx_desc_to_check = 0;
2653 emx_init_rx_unit(struct emx_softc *sc)
2655 struct ifnet *ifp = &sc->arpcom.ac_if;
2657 uint32_t rctl, itr, rfctl;
2661 * Make sure receives are disabled while setting
2662 * up the descriptor ring
2664 rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
2665 E1000_WRITE_REG(&sc->hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2668 * Set the interrupt throttling rate. Value is calculated
2669 * as ITR = 1 / (INT_THROTTLE_CEIL * 256ns)
2671 if (sc->int_throttle_ceil)
2672 itr = 1000000000 / 256 / sc->int_throttle_ceil;
2675 emx_set_itr(sc, itr);
2677 /* Use extended RX descriptor */
2678 rfctl = E1000_RFCTL_EXTEN;
2680 /* Disable accelerated ackknowledge */
2681 if (sc->hw.mac.type == e1000_82574)
2682 rfctl |= E1000_RFCTL_ACK_DIS;
2684 E1000_WRITE_REG(&sc->hw, E1000_RFCTL, rfctl);
2687 * Receive Checksum Offload for TCP and UDP
2689 * Checksum offloading is also enabled if multiple receive
2690 * queue is to be supported, since we need it to figure out
2693 if ((ifp->if_capenable & IFCAP_RXCSUM) ||
2694 sc->rx_ring_cnt > 1) {
2697 rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2701 * PCSD must be enabled to enable multiple
2704 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2706 E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2710 * Configure multiple receive queue (RSS)
2712 if (sc->rx_ring_cnt > 1) {
2713 uint8_t key[EMX_NRSSRK * EMX_RSSRK_SIZE];
2716 KASSERT(sc->rx_ring_cnt == EMX_NRX_RING,
2717 ("invalid number of RX ring (%d)", sc->rx_ring_cnt));
2721 * When we reach here, RSS has already been disabled
2722 * in emx_stop(), so we could safely configure RSS key
2723 * and redirect table.
2729 toeplitz_get_key(key, sizeof(key));
2730 for (i = 0; i < EMX_NRSSRK; ++i) {
2733 rssrk = EMX_RSSRK_VAL(key, i);
2734 EMX_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2736 E1000_WRITE_REG(&sc->hw, E1000_RSSRK(i), rssrk);
2740 * Configure RSS redirect table in following fashion:
2741 * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
2744 for (i = 0; i < EMX_RETA_SIZE; ++i) {
2747 q = (i % sc->rx_ring_cnt) << EMX_RETA_RINGIDX_SHIFT;
2748 reta |= q << (8 * i);
2750 EMX_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2752 for (i = 0; i < EMX_NRETA; ++i)
2753 E1000_WRITE_REG(&sc->hw, E1000_RETA(i), reta);
2756 * Enable multiple receive queues.
2757 * Enable IPv4 RSS standard hash functions.
2758 * Disable RSS interrupt.
2760 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2761 E1000_MRQC_ENABLE_RSS_2Q |
2762 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2763 E1000_MRQC_RSS_FIELD_IPV4);
2767 * XXX TEMPORARY WORKAROUND: on some systems with 82573
2768 * long latencies are observed, like Lenovo X60. This
2769 * change eliminates the problem, but since having positive
2770 * values in RDTR is a known source of problems on other
2771 * platforms another solution is being sought.
2773 if (emx_82573_workaround && sc->hw.mac.type == e1000_82573) {
2774 E1000_WRITE_REG(&sc->hw, E1000_RADV, EMX_RADV_82573);
2775 E1000_WRITE_REG(&sc->hw, E1000_RDTR, EMX_RDTR_82573);
2778 for (i = 0; i < sc->rx_ring_cnt; ++i) {
2779 struct emx_rxdata *rdata = &sc->rx_data[i];
2782 * Setup the Base and Length of the Rx Descriptor Ring
2784 bus_addr = rdata->rx_desc_paddr;
2785 E1000_WRITE_REG(&sc->hw, E1000_RDLEN(i),
2786 rdata->num_rx_desc * sizeof(emx_rxdesc_t));
2787 E1000_WRITE_REG(&sc->hw, E1000_RDBAH(i),
2788 (uint32_t)(bus_addr >> 32));
2789 E1000_WRITE_REG(&sc->hw, E1000_RDBAL(i),
2790 (uint32_t)bus_addr);
2793 * Setup the HW Rx Head and Tail Descriptor Pointers
2795 E1000_WRITE_REG(&sc->hw, E1000_RDH(i), 0);
2796 E1000_WRITE_REG(&sc->hw, E1000_RDT(i),
2797 sc->rx_data[i].num_rx_desc - 1);
2800 /* Setup the Receive Control Register */
2801 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2802 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2803 E1000_RCTL_RDMTS_HALF | E1000_RCTL_SECRC |
2804 (sc->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2806 /* Make sure VLAN Filters are off */
2807 rctl &= ~E1000_RCTL_VFE;
2809 /* Don't store bad paket */
2810 rctl &= ~E1000_RCTL_SBP;
2813 rctl |= E1000_RCTL_SZ_2048;
2815 if (ifp->if_mtu > ETHERMTU)
2816 rctl |= E1000_RCTL_LPE;
2818 rctl &= ~E1000_RCTL_LPE;
2820 /* Enable Receives */
2821 E1000_WRITE_REG(&sc->hw, E1000_RCTL, rctl);
2825 emx_destroy_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata, int ndesc)
2827 struct emx_rxbuf *rx_buffer;
2830 /* Free Receive Descriptor ring */
2831 if (rdata->rx_desc) {
2832 bus_dmamap_unload(rdata->rx_desc_dtag, rdata->rx_desc_dmap);
2833 bus_dmamem_free(rdata->rx_desc_dtag, rdata->rx_desc,
2834 rdata->rx_desc_dmap);
2835 bus_dma_tag_destroy(rdata->rx_desc_dtag);
2837 rdata->rx_desc = NULL;
2840 if (rdata->rx_buf == NULL)
2843 for (i = 0; i < ndesc; i++) {
2844 rx_buffer = &rdata->rx_buf[i];
2846 KKASSERT(rx_buffer->m_head == NULL);
2847 bus_dmamap_destroy(rdata->rxtag, rx_buffer->map);
2849 bus_dmamap_destroy(rdata->rxtag, rdata->rx_sparemap);
2850 bus_dma_tag_destroy(rdata->rxtag);
2852 kfree(rdata->rx_buf, M_DEVBUF);
2853 rdata->rx_buf = NULL;
2857 emx_rxeof(struct emx_softc *sc, int ring_idx, int count)
2859 struct emx_rxdata *rdata = &sc->rx_data[ring_idx];
2860 struct ifnet *ifp = &sc->arpcom.ac_if;
2862 emx_rxdesc_t *current_desc;
2866 i = rdata->next_rx_desc_to_check;
2867 current_desc = &rdata->rx_desc[i];
2868 staterr = le32toh(current_desc->rxd_staterr);
2870 if (!(staterr & E1000_RXD_STAT_DD))
2873 while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
2874 struct pktinfo *pi = NULL, pi0;
2875 struct emx_rxbuf *rx_buf = &rdata->rx_buf[i];
2876 struct mbuf *m = NULL;
2881 mp = rx_buf->m_head;
2884 * Can't defer bus_dmamap_sync(9) because TBI_ACCEPT
2885 * needs to access the last received byte in the mbuf.
2887 bus_dmamap_sync(rdata->rxtag, rx_buf->map,
2888 BUS_DMASYNC_POSTREAD);
2890 len = le16toh(current_desc->rxd_length);
2891 if (staterr & E1000_RXD_STAT_EOP) {
2898 if (!(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
2900 uint32_t mrq, rss_hash;
2903 * Save several necessary information,
2904 * before emx_newbuf() destroy it.
2906 if ((staterr & E1000_RXD_STAT_VP) && eop)
2907 vlan = le16toh(current_desc->rxd_vlan);
2909 mrq = le32toh(current_desc->rxd_mrq);
2910 rss_hash = le32toh(current_desc->rxd_rss);
2912 EMX_RSS_DPRINTF(sc, 10,
2913 "ring%d, mrq 0x%08x, rss_hash 0x%08x\n",
2914 ring_idx, mrq, rss_hash);
2916 if (emx_newbuf(sc, rdata, i, 0) != 0) {
2921 /* Assign correct length to the current fragment */
2924 if (rdata->fmp == NULL) {
2925 mp->m_pkthdr.len = len;
2926 rdata->fmp = mp; /* Store the first mbuf */
2930 * Chain mbuf's together
2932 rdata->lmp->m_next = mp;
2933 rdata->lmp = rdata->lmp->m_next;
2934 rdata->fmp->m_pkthdr.len += len;
2938 rdata->fmp->m_pkthdr.rcvif = ifp;
2941 if (ifp->if_capenable & IFCAP_RXCSUM)
2942 emx_rxcsum(staterr, rdata->fmp);
2944 if (staterr & E1000_RXD_STAT_VP) {
2945 rdata->fmp->m_pkthdr.ether_vlantag =
2947 rdata->fmp->m_flags |= M_VLANTAG;
2953 if (ifp->if_capenable & IFCAP_RSS) {
2954 pi = emx_rssinfo(m, &pi0, mrq,
2957 #ifdef EMX_RSS_DEBUG
2964 emx_setup_rxdesc(current_desc, rx_buf);
2965 if (rdata->fmp != NULL) {
2966 m_freem(rdata->fmp);
2974 ether_input_pkt(ifp, m, pi);
2976 /* Advance our pointers to the next descriptor. */
2977 if (++i == rdata->num_rx_desc)
2980 current_desc = &rdata->rx_desc[i];
2981 staterr = le32toh(current_desc->rxd_staterr);
2983 rdata->next_rx_desc_to_check = i;
2985 /* Advance the E1000's Receive Queue "Tail Pointer". */
2987 i = rdata->num_rx_desc - 1;
2988 E1000_WRITE_REG(&sc->hw, E1000_RDT(ring_idx), i);
2992 emx_enable_intr(struct emx_softc *sc)
2994 uint32_t ims_mask = IMS_ENABLE_MASK;
2996 lwkt_serialize_handler_enable(&sc->main_serialize);
2999 if (sc->hw.mac.type == e1000_82574) {
3000 E1000_WRITE_REG(hw, EMX_EIAC, EM_MSIX_MASK);
3001 ims_mask |= EM_MSIX_MASK;
3004 E1000_WRITE_REG(&sc->hw, E1000_IMS, ims_mask);
3008 emx_disable_intr(struct emx_softc *sc)
3010 if (sc->hw.mac.type == e1000_82574)
3011 E1000_WRITE_REG(&sc->hw, EMX_EIAC, 0);
3012 E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
3014 lwkt_serialize_handler_disable(&sc->main_serialize);
3018 * Bit of a misnomer, what this really means is
3019 * to enable OS management of the system... aka
3020 * to disable special hardware management features
3023 emx_get_mgmt(struct emx_softc *sc)
3025 /* A shared code workaround */
3026 if (sc->flags & EMX_FLAG_HAS_MGMT) {
3027 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
3028 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
3030 /* disable hardware interception of ARP */
3031 manc &= ~(E1000_MANC_ARP_EN);
3033 /* enable receiving management packets to the host */
3034 manc |= E1000_MANC_EN_MNG2HOST;
3035 #define E1000_MNG2HOST_PORT_623 (1 << 5)
3036 #define E1000_MNG2HOST_PORT_664 (1 << 6)
3037 manc2h |= E1000_MNG2HOST_PORT_623;
3038 manc2h |= E1000_MNG2HOST_PORT_664;
3039 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
3041 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
3046 * Give control back to hardware management
3047 * controller if there is one.
3050 emx_rel_mgmt(struct emx_softc *sc)
3052 if (sc->flags & EMX_FLAG_HAS_MGMT) {
3053 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
3055 /* re-enable hardware interception of ARP */
3056 manc |= E1000_MANC_ARP_EN;
3057 manc &= ~E1000_MANC_EN_MNG2HOST;
3059 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
3064 * emx_get_hw_control() sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3065 * For ASF and Pass Through versions of f/w this means that
3066 * the driver is loaded. For AMT version (only with 82573)
3067 * of the f/w this means that the network i/f is open.
3070 emx_get_hw_control(struct emx_softc *sc)
3072 /* Let firmware know the driver has taken over */
3073 if (sc->hw.mac.type == e1000_82573) {
3076 swsm = E1000_READ_REG(&sc->hw, E1000_SWSM);
3077 E1000_WRITE_REG(&sc->hw, E1000_SWSM,
3078 swsm | E1000_SWSM_DRV_LOAD);
3082 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3083 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
3084 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
3086 sc->flags |= EMX_FLAG_HW_CTRL;
3090 * emx_rel_hw_control() resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3091 * For ASF and Pass Through versions of f/w this means that the
3092 * driver is no longer loaded. For AMT version (only with 82573)
3093 * of the f/w this means that the network i/f is closed.
3096 emx_rel_hw_control(struct emx_softc *sc)
3098 if ((sc->flags & EMX_FLAG_HW_CTRL) == 0)
3100 sc->flags &= ~EMX_FLAG_HW_CTRL;
3102 /* Let firmware taken over control of h/w */
3103 if (sc->hw.mac.type == e1000_82573) {
3106 swsm = E1000_READ_REG(&sc->hw, E1000_SWSM);
3107 E1000_WRITE_REG(&sc->hw, E1000_SWSM,
3108 swsm & ~E1000_SWSM_DRV_LOAD);
3112 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3113 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
3114 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
3119 emx_is_valid_eaddr(const uint8_t *addr)
3121 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
3123 if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3130 * Enable PCI Wake On Lan capability
3133 emx_enable_wol(device_t dev)
3135 uint16_t cap, status;
3138 /* First find the capabilities pointer*/
3139 cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
3141 /* Read the PM Capabilities */
3142 id = pci_read_config(dev, cap, 1);
3143 if (id != PCIY_PMG) /* Something wrong */
3147 * OK, we have the power capabilities,
3148 * so now get the status register
3150 cap += PCIR_POWER_STATUS;
3151 status = pci_read_config(dev, cap, 2);
3152 status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3153 pci_write_config(dev, cap, status, 2);
3157 emx_update_stats(struct emx_softc *sc)
3159 struct ifnet *ifp = &sc->arpcom.ac_if;
3161 if (sc->hw.phy.media_type == e1000_media_type_copper ||
3162 (E1000_READ_REG(&sc->hw, E1000_STATUS) & E1000_STATUS_LU)) {
3163 sc->stats.symerrs += E1000_READ_REG(&sc->hw, E1000_SYMERRS);
3164 sc->stats.sec += E1000_READ_REG(&sc->hw, E1000_SEC);
3166 sc->stats.crcerrs += E1000_READ_REG(&sc->hw, E1000_CRCERRS);
3167 sc->stats.mpc += E1000_READ_REG(&sc->hw, E1000_MPC);
3168 sc->stats.scc += E1000_READ_REG(&sc->hw, E1000_SCC);
3169 sc->stats.ecol += E1000_READ_REG(&sc->hw, E1000_ECOL);
3171 sc->stats.mcc += E1000_READ_REG(&sc->hw, E1000_MCC);
3172 sc->stats.latecol += E1000_READ_REG(&sc->hw, E1000_LATECOL);
3173 sc->stats.colc += E1000_READ_REG(&sc->hw, E1000_COLC);
3174 sc->stats.dc += E1000_READ_REG(&sc->hw, E1000_DC);
3175 sc->stats.rlec += E1000_READ_REG(&sc->hw, E1000_RLEC);
3176 sc->stats.xonrxc += E1000_READ_REG(&sc->hw, E1000_XONRXC);
3177 sc->stats.xontxc += E1000_READ_REG(&sc->hw, E1000_XONTXC);
3178 sc->stats.xoffrxc += E1000_READ_REG(&sc->hw, E1000_XOFFRXC);
3179 sc->stats.xofftxc += E1000_READ_REG(&sc->hw, E1000_XOFFTXC);
3180 sc->stats.fcruc += E1000_READ_REG(&sc->hw, E1000_FCRUC);
3181 sc->stats.prc64 += E1000_READ_REG(&sc->hw, E1000_PRC64);
3182 sc->stats.prc127 += E1000_READ_REG(&sc->hw, E1000_PRC127);
3183 sc->stats.prc255 += E1000_READ_REG(&sc->hw, E1000_PRC255);
3184 sc->stats.prc511 += E1000_READ_REG(&sc->hw, E1000_PRC511);
3185 sc->stats.prc1023 += E1000_READ_REG(&sc->hw, E1000_PRC1023);
3186 sc->stats.prc1522 += E1000_READ_REG(&sc->hw, E1000_PRC1522);
3187 sc->stats.gprc += E1000_READ_REG(&sc->hw, E1000_GPRC);
3188 sc->stats.bprc += E1000_READ_REG(&sc->hw, E1000_BPRC);
3189 sc->stats.mprc += E1000_READ_REG(&sc->hw, E1000_MPRC);
3190 sc->stats.gptc += E1000_READ_REG(&sc->hw, E1000_GPTC);
3192 /* For the 64-bit byte counters the low dword must be read first. */
3193 /* Both registers clear on the read of the high dword */
3195 sc->stats.gorc += E1000_READ_REG(&sc->hw, E1000_GORCH);
3196 sc->stats.gotc += E1000_READ_REG(&sc->hw, E1000_GOTCH);
3198 sc->stats.rnbc += E1000_READ_REG(&sc->hw, E1000_RNBC);
3199 sc->stats.ruc += E1000_READ_REG(&sc->hw, E1000_RUC);
3200 sc->stats.rfc += E1000_READ_REG(&sc->hw, E1000_RFC);
3201 sc->stats.roc += E1000_READ_REG(&sc->hw, E1000_ROC);
3202 sc->stats.rjc += E1000_READ_REG(&sc->hw, E1000_RJC);
3204 sc->stats.tor += E1000_READ_REG(&sc->hw, E1000_TORH);
3205 sc->stats.tot += E1000_READ_REG(&sc->hw, E1000_TOTH);
3207 sc->stats.tpr += E1000_READ_REG(&sc->hw, E1000_TPR);
3208 sc->stats.tpt += E1000_READ_REG(&sc->hw, E1000_TPT);
3209 sc->stats.ptc64 += E1000_READ_REG(&sc->hw, E1000_PTC64);
3210 sc->stats.ptc127 += E1000_READ_REG(&sc->hw, E1000_PTC127);
3211 sc->stats.ptc255 += E1000_READ_REG(&sc->hw, E1000_PTC255);
3212 sc->stats.ptc511 += E1000_READ_REG(&sc->hw, E1000_PTC511);
3213 sc->stats.ptc1023 += E1000_READ_REG(&sc->hw, E1000_PTC1023);
3214 sc->stats.ptc1522 += E1000_READ_REG(&sc->hw, E1000_PTC1522);
3215 sc->stats.mptc += E1000_READ_REG(&sc->hw, E1000_MPTC);
3216 sc->stats.bptc += E1000_READ_REG(&sc->hw, E1000_BPTC);
3218 sc->stats.algnerrc += E1000_READ_REG(&sc->hw, E1000_ALGNERRC);
3219 sc->stats.rxerrc += E1000_READ_REG(&sc->hw, E1000_RXERRC);
3220 sc->stats.tncrs += E1000_READ_REG(&sc->hw, E1000_TNCRS);
3221 sc->stats.cexterr += E1000_READ_REG(&sc->hw, E1000_CEXTERR);
3222 sc->stats.tsctc += E1000_READ_REG(&sc->hw, E1000_TSCTC);
3223 sc->stats.tsctfc += E1000_READ_REG(&sc->hw, E1000_TSCTFC);
3225 ifp->if_collisions = sc->stats.colc;
3228 ifp->if_ierrors = sc->dropped_pkts + sc->stats.rxerrc +
3229 sc->stats.crcerrs + sc->stats.algnerrc +
3230 sc->stats.ruc + sc->stats.roc +
3231 sc->stats.mpc + sc->stats.cexterr;
3234 ifp->if_oerrors = sc->stats.ecol + sc->stats.latecol +
3235 sc->watchdog_events;
3239 emx_print_debug_info(struct emx_softc *sc)
3241 device_t dev = sc->dev;
3242 uint8_t *hw_addr = sc->hw.hw_addr;
3244 device_printf(dev, "Adapter hardware address = %p \n", hw_addr);
3245 device_printf(dev, "CTRL = 0x%x RCTL = 0x%x \n",
3246 E1000_READ_REG(&sc->hw, E1000_CTRL),
3247 E1000_READ_REG(&sc->hw, E1000_RCTL));
3248 device_printf(dev, "Packet buffer = Tx=%dk Rx=%dk \n",
3249 ((E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff0000) >> 16),\
3250 (E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff) );
3251 device_printf(dev, "Flow control watermarks high = %d low = %d\n",
3252 sc->hw.fc.high_water, sc->hw.fc.low_water);
3253 device_printf(dev, "tx_int_delay = %d, tx_abs_int_delay = %d\n",
3254 E1000_READ_REG(&sc->hw, E1000_TIDV),
3255 E1000_READ_REG(&sc->hw, E1000_TADV));
3256 device_printf(dev, "rx_int_delay = %d, rx_abs_int_delay = %d\n",
3257 E1000_READ_REG(&sc->hw, E1000_RDTR),
3258 E1000_READ_REG(&sc->hw, E1000_RADV));
3259 device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
3260 E1000_READ_REG(&sc->hw, E1000_TDH(0)),
3261 E1000_READ_REG(&sc->hw, E1000_TDT(0)));
3262 device_printf(dev, "hw rdh = %d, hw rdt = %d\n",
3263 E1000_READ_REG(&sc->hw, E1000_RDH(0)),
3264 E1000_READ_REG(&sc->hw, E1000_RDT(0)));
3265 device_printf(dev, "Num Tx descriptors avail = %d\n",
3266 sc->num_tx_desc_avail);
3267 device_printf(dev, "Tx Descriptors not avail1 = %ld\n",
3268 sc->no_tx_desc_avail1);
3269 device_printf(dev, "Tx Descriptors not avail2 = %ld\n",
3270 sc->no_tx_desc_avail2);
3271 device_printf(dev, "Std mbuf failed = %ld\n",
3272 sc->mbuf_alloc_failed);
3273 device_printf(dev, "Std mbuf cluster failed = %ld\n",
3274 sc->rx_data[0].mbuf_cluster_failed);
3275 device_printf(dev, "Driver dropped packets = %ld\n",
3277 device_printf(dev, "Driver tx dma failure in encap = %ld\n",
3278 sc->no_tx_dma_setup);
3280 device_printf(dev, "TSO segments %lu\n", sc->tso_segments);
3281 device_printf(dev, "TSO ctx reused %lu\n", sc->tso_ctx_reused);
3285 emx_print_hw_stats(struct emx_softc *sc)
3287 device_t dev = sc->dev;
3289 device_printf(dev, "Excessive collisions = %lld\n",
3290 (long long)sc->stats.ecol);
3291 #if (DEBUG_HW > 0) /* Dont output these errors normally */
3292 device_printf(dev, "Symbol errors = %lld\n",
3293 (long long)sc->stats.symerrs);
3295 device_printf(dev, "Sequence errors = %lld\n",
3296 (long long)sc->stats.sec);
3297 device_printf(dev, "Defer count = %lld\n",
3298 (long long)sc->stats.dc);
3299 device_printf(dev, "Missed Packets = %lld\n",
3300 (long long)sc->stats.mpc);
3301 device_printf(dev, "Receive No Buffers = %lld\n",
3302 (long long)sc->stats.rnbc);
3303 /* RLEC is inaccurate on some hardware, calculate our own. */
3304 device_printf(dev, "Receive Length Errors = %lld\n",
3305 ((long long)sc->stats.roc + (long long)sc->stats.ruc));
3306 device_printf(dev, "Receive errors = %lld\n",
3307 (long long)sc->stats.rxerrc);
3308 device_printf(dev, "Crc errors = %lld\n",
3309 (long long)sc->stats.crcerrs);
3310 device_printf(dev, "Alignment errors = %lld\n",
3311 (long long)sc->stats.algnerrc);
3312 device_printf(dev, "Collision/Carrier extension errors = %lld\n",
3313 (long long)sc->stats.cexterr);
3314 device_printf(dev, "RX overruns = %ld\n", sc->rx_overruns);
3315 device_printf(dev, "watchdog timeouts = %ld\n",
3316 sc->watchdog_events);
3317 device_printf(dev, "XON Rcvd = %lld\n",
3318 (long long)sc->stats.xonrxc);
3319 device_printf(dev, "XON Xmtd = %lld\n",
3320 (long long)sc->stats.xontxc);
3321 device_printf(dev, "XOFF Rcvd = %lld\n",
3322 (long long)sc->stats.xoffrxc);
3323 device_printf(dev, "XOFF Xmtd = %lld\n",
3324 (long long)sc->stats.xofftxc);
3325 device_printf(dev, "Good Packets Rcvd = %lld\n",
3326 (long long)sc->stats.gprc);
3327 device_printf(dev, "Good Packets Xmtd = %lld\n",
3328 (long long)sc->stats.gptc);
3332 emx_print_nvm_info(struct emx_softc *sc)
3334 uint16_t eeprom_data;
3337 /* Its a bit crude, but it gets the job done */
3338 kprintf("\nInterface EEPROM Dump:\n");
3339 kprintf("Offset\n0x0000 ");
3340 for (i = 0, j = 0; i < 32; i++, j++) {
3341 if (j == 8) { /* Make the offset block */
3343 kprintf("\n0x00%x0 ",row);
3345 e1000_read_nvm(&sc->hw, i, 1, &eeprom_data);
3346 kprintf("%04x ", eeprom_data);
3352 emx_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
3354 struct emx_softc *sc;
3359 error = sysctl_handle_int(oidp, &result, 0, req);
3360 if (error || !req->newptr)
3363 sc = (struct emx_softc *)arg1;
3364 ifp = &sc->arpcom.ac_if;
3366 ifnet_serialize_all(ifp);
3369 emx_print_debug_info(sc);
3372 * This value will cause a hex dump of the
3373 * first 32 16-bit words of the EEPROM to
3377 emx_print_nvm_info(sc);
3379 ifnet_deserialize_all(ifp);
3385 emx_sysctl_stats(SYSCTL_HANDLER_ARGS)
3390 error = sysctl_handle_int(oidp, &result, 0, req);
3391 if (error || !req->newptr)
3395 struct emx_softc *sc = (struct emx_softc *)arg1;
3396 struct ifnet *ifp = &sc->arpcom.ac_if;
3398 ifnet_serialize_all(ifp);
3399 emx_print_hw_stats(sc);
3400 ifnet_deserialize_all(ifp);
3406 emx_add_sysctl(struct emx_softc *sc)
3408 #ifdef EMX_RSS_DEBUG
3413 sysctl_ctx_init(&sc->sysctl_ctx);
3414 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
3415 SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
3416 device_get_nameunit(sc->dev),
3418 if (sc->sysctl_tree == NULL) {
3419 device_printf(sc->dev, "can't add sysctl node\n");
3423 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3424 OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, sc, 0,
3425 emx_sysctl_debug_info, "I", "Debug Information");
3427 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3428 OID_AUTO, "stats", CTLTYPE_INT|CTLFLAG_RW, sc, 0,
3429 emx_sysctl_stats, "I", "Statistics");
3431 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3432 OID_AUTO, "rxd", CTLFLAG_RD,
3433 &sc->rx_data[0].num_rx_desc, 0, NULL);
3434 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3435 OID_AUTO, "txd", CTLFLAG_RD, &sc->num_tx_desc, 0, NULL);
3437 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3438 OID_AUTO, "int_throttle_ceil", CTLTYPE_INT|CTLFLAG_RW,
3439 sc, 0, emx_sysctl_int_throttle, "I",
3440 "interrupt throttling rate");
3441 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3442 OID_AUTO, "int_tx_nsegs", CTLTYPE_INT|CTLFLAG_RW,
3443 sc, 0, emx_sysctl_int_tx_nsegs, "I",
3444 "# segments per TX interrupt");
3446 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3447 OID_AUTO, "rx_ring_cnt", CTLFLAG_RD,
3448 &sc->rx_ring_cnt, 0, "RX ring count");
3450 #ifdef IFPOLL_ENABLE
3451 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3452 OID_AUTO, "npoll_rxoff", CTLTYPE_INT|CTLFLAG_RW,
3453 sc, 0, emx_sysctl_npoll_rxoff, "I",
3454 "NPOLLING RX cpu offset");
3455 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3456 OID_AUTO, "npoll_txoff", CTLTYPE_INT|CTLFLAG_RW,
3457 sc, 0, emx_sysctl_npoll_txoff, "I",
3458 "NPOLLING TX cpu offset");
3461 #ifdef EMX_RSS_DEBUG
3462 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3463 OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug,
3464 0, "RSS debug level");
3465 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3466 ksnprintf(rx_pkt, sizeof(rx_pkt), "rx%d_pkt", i);
3467 SYSCTL_ADD_UINT(&sc->sysctl_ctx,
3468 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO,
3470 &sc->rx_data[i].rx_pkts, 0, "RXed packets");
3476 emx_sysctl_int_throttle(SYSCTL_HANDLER_ARGS)
3478 struct emx_softc *sc = (void *)arg1;
3479 struct ifnet *ifp = &sc->arpcom.ac_if;
3480 int error, throttle;
3482 throttle = sc->int_throttle_ceil;
3483 error = sysctl_handle_int(oidp, &throttle, 0, req);
3484 if (error || req->newptr == NULL)
3486 if (throttle < 0 || throttle > 1000000000 / 256)
3491 * Set the interrupt throttling rate in 256ns increments,
3492 * recalculate sysctl value assignment to get exact frequency.
3494 throttle = 1000000000 / 256 / throttle;
3496 /* Upper 16bits of ITR is reserved and should be zero */
3497 if (throttle & 0xffff0000)
3501 ifnet_serialize_all(ifp);
3504 sc->int_throttle_ceil = 1000000000 / 256 / throttle;
3506 sc->int_throttle_ceil = 0;
3508 if (ifp->if_flags & IFF_RUNNING)
3509 emx_set_itr(sc, throttle);
3511 ifnet_deserialize_all(ifp);
3514 if_printf(ifp, "Interrupt moderation set to %d/sec\n",
3515 sc->int_throttle_ceil);
3521 emx_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS)
3523 struct emx_softc *sc = (void *)arg1;
3524 struct ifnet *ifp = &sc->arpcom.ac_if;
3527 segs = sc->tx_int_nsegs;
3528 error = sysctl_handle_int(oidp, &segs, 0, req);
3529 if (error || req->newptr == NULL)
3534 ifnet_serialize_all(ifp);
3537 * Don't allow int_tx_nsegs to become:
3538 * o Less the oact_tx_desc
3539 * o Too large that no TX desc will cause TX interrupt to
3540 * be generated (OACTIVE will never recover)
3541 * o Too small that will cause tx_dd[] overflow
3543 if (segs < sc->oact_tx_desc ||
3544 segs >= sc->num_tx_desc - sc->oact_tx_desc ||
3545 segs < sc->num_tx_desc / EMX_TXDD_SAFE) {
3549 sc->tx_int_nsegs = segs;
3552 ifnet_deserialize_all(ifp);
3557 #ifdef IFPOLL_ENABLE
3560 emx_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS)
3562 struct emx_softc *sc = (void *)arg1;
3563 struct ifnet *ifp = &sc->arpcom.ac_if;
3566 off = sc->rx_npoll_off;
3567 error = sysctl_handle_int(oidp, &off, 0, req);
3568 if (error || req->newptr == NULL)
3573 ifnet_serialize_all(ifp);
3574 if (off >= ncpus2 || off % sc->rx_ring_cnt != 0) {
3578 sc->rx_npoll_off = off;
3580 ifnet_deserialize_all(ifp);
3586 emx_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS)
3588 struct emx_softc *sc = (void *)arg1;
3589 struct ifnet *ifp = &sc->arpcom.ac_if;
3592 off = sc->tx_npoll_off;
3593 error = sysctl_handle_int(oidp, &off, 0, req);
3594 if (error || req->newptr == NULL)
3599 ifnet_serialize_all(ifp);
3600 if (off >= ncpus2) {
3604 sc->tx_npoll_off = off;
3606 ifnet_deserialize_all(ifp);
3611 #endif /* IFPOLL_ENABLE */
3614 emx_dma_alloc(struct emx_softc *sc)
3619 * Create top level busdma tag
3621 error = bus_dma_tag_create(NULL, 1, 0,
3622 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3624 BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
3625 0, &sc->parent_dtag);
3627 device_printf(sc->dev, "could not create top level DMA tag\n");
3632 * Allocate transmit descriptors ring and buffers
3634 error = emx_create_tx_ring(sc);
3636 device_printf(sc->dev, "Could not setup transmit structures\n");
3641 * Allocate receive descriptors ring and buffers
3643 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3644 error = emx_create_rx_ring(sc, &sc->rx_data[i]);
3646 device_printf(sc->dev,
3647 "Could not setup receive structures\n");
3655 emx_dma_free(struct emx_softc *sc)
3659 emx_destroy_tx_ring(sc, sc->num_tx_desc);
3661 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3662 emx_destroy_rx_ring(sc, &sc->rx_data[i],
3663 sc->rx_data[i].num_rx_desc);
3666 /* Free top level busdma tag */
3667 if (sc->parent_dtag != NULL)
3668 bus_dma_tag_destroy(sc->parent_dtag);
3672 emx_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3674 struct emx_softc *sc = ifp->if_softc;
3676 ifnet_serialize_array_enter(sc->serializes, EMX_NSERIALIZE,
3677 EMX_TX_SERIALIZE, EMX_RX_SERIALIZE, slz);
3681 emx_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3683 struct emx_softc *sc = ifp->if_softc;
3685 ifnet_serialize_array_exit(sc->serializes, EMX_NSERIALIZE,
3686 EMX_TX_SERIALIZE, EMX_RX_SERIALIZE, slz);
3690 emx_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3692 struct emx_softc *sc = ifp->if_softc;
3694 return ifnet_serialize_array_try(sc->serializes, EMX_NSERIALIZE,
3695 EMX_TX_SERIALIZE, EMX_RX_SERIALIZE, slz);
3699 emx_serialize_skipmain(struct emx_softc *sc)
3701 lwkt_serialize_array_enter(sc->serializes, EMX_NSERIALIZE, 1);
3705 emx_deserialize_skipmain(struct emx_softc *sc)
3707 lwkt_serialize_array_exit(sc->serializes, EMX_NSERIALIZE, 1);
3713 emx_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3714 boolean_t serialized)
3716 struct emx_softc *sc = ifp->if_softc;
3718 ifnet_serialize_array_assert(sc->serializes, EMX_NSERIALIZE,
3719 EMX_TX_SERIALIZE, EMX_RX_SERIALIZE, slz, serialized);
3722 #endif /* INVARIANTS */
3724 #ifdef IFPOLL_ENABLE
3727 emx_npoll_status(struct ifnet *ifp)
3729 struct emx_softc *sc = ifp->if_softc;
3732 ASSERT_SERIALIZED(&sc->main_serialize);
3734 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3735 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3736 callout_stop(&sc->timer);
3737 sc->hw.mac.get_link_status = 1;
3738 emx_update_link_status(sc);
3739 callout_reset(&sc->timer, hz, emx_timer, sc);
3744 emx_npoll_tx(struct ifnet *ifp, void *arg __unused, int cycle __unused)
3746 struct emx_softc *sc = ifp->if_softc;
3748 ASSERT_SERIALIZED(&sc->tx_serialize);
3751 if (!ifq_is_empty(&ifp->if_snd))
3756 emx_npoll_rx(struct ifnet *ifp, void *arg, int cycle)
3758 struct emx_softc *sc = ifp->if_softc;
3759 struct emx_rxdata *rdata = arg;
3761 ASSERT_SERIALIZED(&rdata->rx_serialize);
3763 emx_rxeof(sc, rdata - sc->rx_data, cycle);
3767 emx_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3769 struct emx_softc *sc = ifp->if_softc;
3771 ASSERT_IFNET_SERIALIZED_ALL(ifp);
3776 info->ifpi_status.status_func = emx_npoll_status;
3777 info->ifpi_status.serializer = &sc->main_serialize;
3779 off = sc->tx_npoll_off;
3780 KKASSERT(off < ncpus2);
3781 info->ifpi_tx[off].poll_func = emx_npoll_tx;
3782 info->ifpi_tx[off].arg = NULL;
3783 info->ifpi_tx[off].serializer = &sc->tx_serialize;
3785 off = sc->rx_npoll_off;
3786 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3787 struct emx_rxdata *rdata = &sc->rx_data[i];
3790 KKASSERT(idx < ncpus2);
3791 info->ifpi_rx[idx].poll_func = emx_npoll_rx;
3792 info->ifpi_rx[idx].arg = rdata;
3793 info->ifpi_rx[idx].serializer = &rdata->rx_serialize;
3796 if (ifp->if_flags & IFF_RUNNING)
3797 emx_disable_intr(sc);
3798 ifp->if_npoll_cpuid = sc->tx_npoll_off;
3800 if (ifp->if_flags & IFF_RUNNING)
3801 emx_enable_intr(sc);
3802 ifp->if_npoll_cpuid = -1;
3806 #endif /* IFPOLL_ENABLE */
3809 emx_set_itr(struct emx_softc *sc, uint32_t itr)
3811 E1000_WRITE_REG(&sc->hw, E1000_ITR, itr);
3812 if (sc->hw.mac.type == e1000_82574) {
3816 * When using MSIX interrupts we need to
3817 * throttle using the EITR register
3819 for (i = 0; i < 4; ++i)
3820 E1000_WRITE_REG(&sc->hw, E1000_EITR_82574(i), itr);
3825 * Disable the L0s, 82574L Errata #20
3828 emx_disable_aspm(struct emx_softc *sc)
3830 uint16_t link_cap, link_ctrl, disable;
3831 uint8_t pcie_ptr, reg;
3832 device_t dev = sc->dev;
3834 switch (sc->hw.mac.type) {
3839 * 82573 specification update
3840 * errata #8 disable L0s
3841 * errata #41 disable L1
3843 * 82571/82572 specification update
3844 # errata #13 disable L1
3845 * errata #68 disable L0s
3847 disable = PCIEM_LNKCTL_ASPM_L0S | PCIEM_LNKCTL_ASPM_L1;
3852 * 82574 specification update errata #20
3854 * There is no need to disable L1
3856 disable = PCIEM_LNKCTL_ASPM_L0S;
3863 pcie_ptr = pci_get_pciecap_ptr(dev);
3867 link_cap = pci_read_config(dev, pcie_ptr + PCIER_LINKCAP, 2);
3868 if ((link_cap & PCIEM_LNKCAP_ASPM_MASK) == 0)
3872 if_printf(&sc->arpcom.ac_if, "disable ASPM %#02x\n", disable);
3874 reg = pcie_ptr + PCIER_LINKCTRL;
3875 link_ctrl = pci_read_config(dev, reg, 2);
3876 link_ctrl &= ~disable;
3877 pci_write_config(dev, reg, link_ctrl, 2);
3881 emx_tso_pullup(struct emx_softc *sc, struct mbuf **mp)
3883 int iphlen, hoff, thoff, ex = 0;
3888 KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
3890 iphlen = m->m_pkthdr.csum_iphlen;
3891 thoff = m->m_pkthdr.csum_thlen;
3892 hoff = m->m_pkthdr.csum_lhlen;
3894 KASSERT(iphlen > 0, ("invalid ip hlen"));
3895 KASSERT(thoff > 0, ("invalid tcp hlen"));
3896 KASSERT(hoff > 0, ("invalid ether hlen"));
3898 if (sc->flags & EMX_FLAG_TSO_PULLEX)
3901 if (m->m_len < hoff + iphlen + thoff + ex) {
3902 m = m_pullup(m, hoff + iphlen + thoff + ex);
3909 ip = mtodoff(m, struct ip *, hoff);
3916 emx_tso_setup(struct emx_softc *sc, struct mbuf *mp,
3917 uint32_t *txd_upper, uint32_t *txd_lower)
3919 struct e1000_context_desc *TXD;
3920 int hoff, iphlen, thoff, hlen;
3921 int mss, pktlen, curr_txd;
3923 #ifdef EMX_TSO_DEBUG
3927 iphlen = mp->m_pkthdr.csum_iphlen;
3928 thoff = mp->m_pkthdr.csum_thlen;
3929 hoff = mp->m_pkthdr.csum_lhlen;
3930 mss = mp->m_pkthdr.tso_segsz;
3931 pktlen = mp->m_pkthdr.len;
3933 if (sc->csum_flags == CSUM_TSO &&
3934 sc->csum_iphlen == iphlen &&
3935 sc->csum_lhlen == hoff &&
3936 sc->csum_thlen == thoff &&
3937 sc->csum_mss == mss &&
3938 sc->csum_pktlen == pktlen) {
3939 *txd_upper = sc->csum_txd_upper;
3940 *txd_lower = sc->csum_txd_lower;
3941 #ifdef EMX_TSO_DEBUG
3942 sc->tso_ctx_reused++;
3946 hlen = hoff + iphlen + thoff;
3949 * Setup a new TSO context.
3952 curr_txd = sc->next_avail_tx_desc;
3953 TXD = (struct e1000_context_desc *)&sc->tx_desc_base[curr_txd];
3955 *txd_lower = E1000_TXD_CMD_DEXT | /* Extended descr type */
3956 E1000_TXD_DTYP_D | /* Data descr type */
3957 E1000_TXD_CMD_TSE; /* Do TSE on this packet */
3959 /* IP and/or TCP header checksum calculation and insertion. */
3960 *txd_upper = (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8;
3963 * Start offset for header checksum calculation.
3964 * End offset for header checksum calculation.
3965 * Offset of place put the checksum.
3967 TXD->lower_setup.ip_fields.ipcss = hoff;
3968 TXD->lower_setup.ip_fields.ipcse = htole16(hoff + iphlen - 1);
3969 TXD->lower_setup.ip_fields.ipcso = hoff + offsetof(struct ip, ip_sum);
3972 * Start offset for payload checksum calculation.
3973 * End offset for payload checksum calculation.
3974 * Offset of place to put the checksum.
3976 TXD->upper_setup.tcp_fields.tucss = hoff + iphlen;
3977 TXD->upper_setup.tcp_fields.tucse = 0;
3978 TXD->upper_setup.tcp_fields.tucso =
3979 hoff + iphlen + offsetof(struct tcphdr, th_sum);
3982 * Payload size per packet w/o any headers.
3983 * Length of all headers up to payload.
3985 TXD->tcp_seg_setup.fields.mss = htole16(mss);
3986 TXD->tcp_seg_setup.fields.hdr_len = hlen;
3987 TXD->cmd_and_length = htole32(E1000_TXD_CMD_IFCS |
3988 E1000_TXD_CMD_DEXT | /* Extended descr */
3989 E1000_TXD_CMD_TSE | /* TSE context */
3990 E1000_TXD_CMD_IP | /* Do IP csum */
3991 E1000_TXD_CMD_TCP | /* Do TCP checksum */
3992 (pktlen - hlen)); /* Total len */
3994 /* Save the information for this TSO context */
3995 sc->csum_flags = CSUM_TSO;
3996 sc->csum_lhlen = hoff;
3997 sc->csum_iphlen = iphlen;
3998 sc->csum_thlen = thoff;
4000 sc->csum_pktlen = pktlen;
4001 sc->csum_txd_upper = *txd_upper;
4002 sc->csum_txd_lower = *txd_lower;
4004 if (++curr_txd == sc->num_tx_desc)
4007 KKASSERT(sc->num_tx_desc_avail > 0);
4008 sc->num_tx_desc_avail--;
4010 sc->next_avail_tx_desc = curr_txd;