projects / dragonfly.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | history | HEAD
Modify ktr(4) to be typesafe
[dragonfly.git] / sys / dev / netif / emx / if_emx.c
CommitLineData
5330213c
SZ		 1/*
2 * Copyright (c) 2004 Joerg Sonnenberger <joerg@bec.de>. All rights reserved.
3 *
4 * Copyright (c) 2001-2008, Intel Corporation
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
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.
20 *
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.
32 *
33 *
34 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
35 *
36 * This code is derived from software contributed to The DragonFly Project
37 * by Matthew Dillon <dillon@backplane.com>
38 *
39 * Redistribution and use in source and binary forms, with or without
40 * modification, are permitted provided that the following conditions
41 * are met:
42 *
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
48 * distribution.
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.
52 *
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
64 * SUCH DAMAGE.
65 */
66
b3a7093f 67#include "opt_ifpoll.h"
8434a83b 68#include "opt_rss.h"
e6cde6e6 69#include "opt_emx.h"
5330213c
SZ		 70
71#include <sys/param.h>
72#include <sys/bus.h>
73#include <sys/endian.h>
74#include <sys/interrupt.h>
75#include <sys/kernel.h>
76#include <sys/ktr.h>
77#include <sys/malloc.h>
78#include <sys/mbuf.h>
79#include <sys/proc.h>
80#include <sys/rman.h>
81#include <sys/serialize.h>
bc197380 82#include <sys/serialize2.h>
5330213c
SZ		 83#include <sys/socket.h>
84#include <sys/sockio.h>
85#include <sys/sysctl.h>
86#include <sys/systm.h>
87
88#include <net/bpf.h>
89#include <net/ethernet.h>
90#include <net/if.h>
91#include <net/if_arp.h>
92#include <net/if_dl.h>
93#include <net/if_media.h>
94#include <net/ifq_var.h>
89d8e73d 95#include <net/toeplitz.h>
9cc86e17 96#include <net/toeplitz2.h>
5330213c
SZ		 97#include <net/vlan/if_vlan_var.h>
98#include <net/vlan/if_vlan_ether.h>
b3a7093f 99#include <net/if_poll.h>
5330213c
SZ		 100
101#include <netinet/in_systm.h>
102#include <netinet/in.h>
103#include <netinet/ip.h>
104#include <netinet/tcp.h>
105#include <netinet/udp.h>
106
107#include <bus/pci/pcivar.h>
108#include <bus/pci/pcireg.h>
109
110#include <dev/netif/ig_hal/e1000_api.h>
111#include <dev/netif/ig_hal/e1000_82571.h>
112#include <dev/netif/emx/if_emx.h>
113
3f939c23
SZ		 114#ifdef EMX_RSS_DEBUG
115#define EMX_RSS_DPRINTF(sc, lvl, fmt, ...) \
116do { \
89d8e73d 117 if (sc->rss_debug >= lvl) \
3f939c23
SZ		 118 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
119} while (0)
120#else /* !EMX_RSS_DEBUG */
121#define EMX_RSS_DPRINTF(sc, lvl, fmt, ...) ((void)0)
122#endif /* EMX_RSS_DEBUG */
123
5330213c
SZ		 124#define EMX_NAME "Intel(R) PRO/1000 "
125
126#define EMX_DEVICE(id) \
127 { EMX_VENDOR_ID, E1000_DEV_ID_##id, EMX_NAME #id }
128#define EMX_DEVICE_NULL { 0, 0, NULL }
129
130static const struct emx_device {
131 uint16_t vid;
132 uint16_t did;
133 const char *desc;
134} emx_devices[] = {
135 EMX_DEVICE(82571EB_COPPER),
136 EMX_DEVICE(82571EB_FIBER),
137 EMX_DEVICE(82571EB_SERDES),
138 EMX_DEVICE(82571EB_SERDES_DUAL),
139 EMX_DEVICE(82571EB_SERDES_QUAD),
140 EMX_DEVICE(82571EB_QUAD_COPPER),
75a5634e 141 EMX_DEVICE(82571EB_QUAD_COPPER_BP),
5330213c
SZ		 142 EMX_DEVICE(82571EB_QUAD_COPPER_LP),
143 EMX_DEVICE(82571EB_QUAD_FIBER),
144 EMX_DEVICE(82571PT_QUAD_COPPER),
145
146 EMX_DEVICE(82572EI_COPPER),
147 EMX_DEVICE(82572EI_FIBER),
148 EMX_DEVICE(82572EI_SERDES),
149 EMX_DEVICE(82572EI),
150
151 EMX_DEVICE(82573E),
152 EMX_DEVICE(82573E_IAMT),
153 EMX_DEVICE(82573L),
154
155 EMX_DEVICE(80003ES2LAN_COPPER_SPT),
156 EMX_DEVICE(80003ES2LAN_SERDES_SPT),
157 EMX_DEVICE(80003ES2LAN_COPPER_DPT),
158 EMX_DEVICE(80003ES2LAN_SERDES_DPT),
159
160 EMX_DEVICE(82574L),
2d0e5700 161 EMX_DEVICE(82574LA),
5330213c
SZ		 162
163 /* required last entry */
164 EMX_DEVICE_NULL
165};
166
167static int emx_probe(device_t);
168static int emx_attach(device_t);
169static int emx_detach(device_t);
170static int emx_shutdown(device_t);
171static int emx_suspend(device_t);
172static int emx_resume(device_t);
173
174static void emx_init(void *);
175static void emx_stop(struct emx_softc *);
176static int emx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
177static void emx_start(struct ifnet *);
b3a7093f
SZ		 178#ifdef IFPOLL_ENABLE
179static void emx_qpoll(struct ifnet *, struct ifpoll_info *);
5330213c
SZ		 180#endif
181static void emx_watchdog(struct ifnet *);
182static void emx_media_status(struct ifnet *, struct ifmediareq *);
183static int emx_media_change(struct ifnet *);
184static void emx_timer(void *);
6d435846
SZ		 185static void emx_serialize(struct ifnet *, enum ifnet_serialize);
186static void emx_deserialize(struct ifnet *, enum ifnet_serialize);
187static int emx_tryserialize(struct ifnet *, enum ifnet_serialize);
2c9effcf
SZ		 188#ifdef INVARIANTS
189static void emx_serialize_assert(struct ifnet *, enum ifnet_serialize,
190 boolean_t);
191#endif
5330213c
SZ		 192
193static void emx_intr(void *);
c39e3a1f 194static void emx_rxeof(struct emx_softc *, int, int);
5330213c
SZ		 195static void emx_txeof(struct emx_softc *);
196static void emx_tx_collect(struct emx_softc *);
197static void emx_tx_purge(struct emx_softc *);
198static void emx_enable_intr(struct emx_softc *);
199static void emx_disable_intr(struct emx_softc *);
200
071699f8
SZ		 201static int emx_dma_alloc(struct emx_softc *);
202static void emx_dma_free(struct emx_softc *);
5330213c 203static void emx_init_tx_ring(struct emx_softc *);
c39e3a1f
SZ		 204static int emx_init_rx_ring(struct emx_softc *, struct emx_rxdata *);
205static void emx_free_rx_ring(struct emx_softc *, struct emx_rxdata *);
5330213c 206static int emx_create_tx_ring(struct emx_softc *);
c39e3a1f 207static int emx_create_rx_ring(struct emx_softc *, struct emx_rxdata *);
5330213c 208static void emx_destroy_tx_ring(struct emx_softc *, int);
c39e3a1f
SZ		 209static void emx_destroy_rx_ring(struct emx_softc *,
210 struct emx_rxdata *, int);
211static int emx_newbuf(struct emx_softc *, struct emx_rxdata *, int, int);
5330213c 212static int emx_encap(struct emx_softc *, struct mbuf **);
5330213c
SZ		 213static int emx_txcsum_pullup(struct emx_softc *, struct mbuf **);
214static int emx_txcsum(struct emx_softc *, struct mbuf *,
215 uint32_t *, uint32_t *);
216
217static int emx_is_valid_eaddr(const uint8_t *);
2d0e5700 218static int emx_reset(struct emx_softc *);
5330213c
SZ		 219static void emx_setup_ifp(struct emx_softc *);
220static void emx_init_tx_unit(struct emx_softc *);
221static void emx_init_rx_unit(struct emx_softc *);
222static void emx_update_stats(struct emx_softc *);
223static void emx_set_promisc(struct emx_softc *);
224static void emx_disable_promisc(struct emx_softc *);
225static void emx_set_multi(struct emx_softc *);
226static void emx_update_link_status(struct emx_softc *);
227static void emx_smartspeed(struct emx_softc *);
2d0e5700 228static void emx_set_itr(struct emx_softc *, uint32_t);
5330213c
SZ		 229
230static void emx_print_debug_info(struct emx_softc *);
231static void emx_print_nvm_info(struct emx_softc *);
232static void emx_print_hw_stats(struct emx_softc *);
233
234static int emx_sysctl_stats(SYSCTL_HANDLER_ARGS);
235static int emx_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
236static int emx_sysctl_int_throttle(SYSCTL_HANDLER_ARGS);
237static int emx_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS);
238static void emx_add_sysctl(struct emx_softc *);
239
bca7c435
SZ		 240static void emx_serialize_skipmain(struct emx_softc *);
241static void emx_deserialize_skipmain(struct emx_softc *);
242
5330213c
SZ		 243/* Management and WOL Support */
244static void emx_get_mgmt(struct emx_softc *);
245static void emx_rel_mgmt(struct emx_softc *);
246static void emx_get_hw_control(struct emx_softc *);
247static void emx_rel_hw_control(struct emx_softc *);
248static void emx_enable_wol(device_t);
249
250static device_method_t emx_methods[] = {
251 /* Device interface */
252 DEVMETHOD(device_probe, emx_probe),
253 DEVMETHOD(device_attach, emx_attach),
254 DEVMETHOD(device_detach, emx_detach),
255 DEVMETHOD(device_shutdown, emx_shutdown),
256 DEVMETHOD(device_suspend, emx_suspend),
257 DEVMETHOD(device_resume, emx_resume),
258 { 0, 0 }
259};
260
261static driver_t emx_driver = {
262 "emx",
263 emx_methods,
264 sizeof(struct emx_softc),
265};
266
267static devclass_t emx_devclass;
268
269DECLARE_DUMMY_MODULE(if_emx);
270MODULE_DEPEND(emx, ig_hal, 1, 1, 1);
aa2b9d05 271DRIVER_MODULE(if_emx, pci, emx_driver, emx_devclass, NULL, NULL);
5330213c
SZ		 272
273/*
274 * Tunables
275 */
276static int emx_int_throttle_ceil = EMX_DEFAULT_ITR;
277static int emx_rxd = EMX_DEFAULT_RXD;
278static int emx_txd = EMX_DEFAULT_TXD;
704b6287 279static int emx_smart_pwr_down = 0;
5330213c
SZ		 280
281/* Controls whether promiscuous also shows bad packets */
282static int emx_debug_sbp = FALSE;
283
704b6287
SZ		 284static int emx_82573_workaround = 1;
285static int emx_msi_enable = 1;
5330213c
SZ		 286
287TUNABLE_INT("hw.emx.int_throttle_ceil", &emx_int_throttle_ceil);
288TUNABLE_INT("hw.emx.rxd", &emx_rxd);
289TUNABLE_INT("hw.emx.txd", &emx_txd);
290TUNABLE_INT("hw.emx.smart_pwr_down", &emx_smart_pwr_down);
291TUNABLE_INT("hw.emx.sbp", &emx_debug_sbp);
292TUNABLE_INT("hw.emx.82573_workaround", &emx_82573_workaround);
704b6287 293TUNABLE_INT("hw.emx.msi.enable", &emx_msi_enable);
5330213c
SZ		 294
295/* Global used in WOL setup with multiport cards */
296static int emx_global_quad_port_a = 0;
297
298/* Set this to one to display debug statistics */
299static int emx_display_debug_stats = 0;
300
301#if !defined(KTR_IF_EMX)
302#define KTR_IF_EMX KTR_ALL
303#endif
304KTR_INFO_MASTER(if_emx);
5bf48697
AE		 305KTR_INFO(KTR_IF_EMX, if_emx, intr_beg, 0, "intr begin");
306KTR_INFO(KTR_IF_EMX, if_emx, intr_end, 1, "intr end");
307KTR_INFO(KTR_IF_EMX, if_emx, pkt_receive, 4, "rx packet");
308KTR_INFO(KTR_IF_EMX, if_emx, pkt_txqueue, 5, "tx packet");
309KTR_INFO(KTR_IF_EMX, if_emx, pkt_txclean, 6, "tx clean");
5330213c
SZ		 310#define logif(name) KTR_LOG(if_emx_ ## name)
311
235b9d30
SZ		 312static __inline void
313emx_setup_rxdesc(emx_rxdesc_t *rxd, const struct emx_rxbuf *rxbuf)
314{
315 rxd->rxd_bufaddr = htole64(rxbuf->paddr);
3f939c23 316 /* DD bit must be cleared */
235b9d30
SZ		 317 rxd->rxd_staterr = 0;
318}
319
320static __inline void
321emx_rxcsum(uint32_t staterr, struct mbuf *mp)
322{
323 /* Ignore Checksum bit is set */
324 if (staterr & E1000_RXD_STAT_IXSM)
325 return;
326
327 if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
328 E1000_RXD_STAT_IPCS)
329 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
330
331 if ((staterr & (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
332 E1000_RXD_STAT_TCPCS) {
333 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
334 CSUM_PSEUDO_HDR |
335 CSUM_FRAG_NOT_CHECKED;
336 mp->m_pkthdr.csum_data = htons(0xffff);
337 }
338}
339
9cc86e17
SZ		 340static __inline struct pktinfo *
341emx_rssinfo(struct mbuf *m, struct pktinfo *pi,
342 uint32_t mrq, uint32_t hash, uint32_t staterr)
343{
344 switch (mrq & EMX_RXDMRQ_RSSTYPE_MASK) {
345 case EMX_RXDMRQ_IPV4_TCP:
346 pi->pi_netisr = NETISR_IP;
347 pi->pi_flags = 0;
348 pi->pi_l3proto = IPPROTO_TCP;
349 break;
350
351 case EMX_RXDMRQ_IPV6_TCP:
352 pi->pi_netisr = NETISR_IPV6;
353 pi->pi_flags = 0;
354 pi->pi_l3proto = IPPROTO_TCP;
355 break;
356
357 case EMX_RXDMRQ_IPV4:
358 if (staterr & E1000_RXD_STAT_IXSM)
359 return NULL;
360
361 if ((staterr &
362 (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
363 E1000_RXD_STAT_TCPCS) {
364 pi->pi_netisr = NETISR_IP;
365 pi->pi_flags = 0;
366 pi->pi_l3proto = IPPROTO_UDP;
367 break;
368 }
369 /* FALL THROUGH */
370 default:
371 return NULL;
372 }
373
374 m->m_flags |= M_HASH;
375 m->m_pkthdr.hash = toeplitz_hash(hash);
376 return pi;
377}
378
5330213c
SZ		 379static int
380emx_probe(device_t dev)
381{
382 const struct emx_device *d;
383 uint16_t vid, did;
384
385 vid = pci_get_vendor(dev);
386 did = pci_get_device(dev);
387
388 for (d = emx_devices; d->desc != NULL; ++d) {
389 if (vid == d->vid && did == d->did) {
390 device_set_desc(dev, d->desc);
391 device_set_async_attach(dev, TRUE);
392 return 0;
393 }
394 }
395 return ENXIO;
396}
397
398static int
399emx_attach(device_t dev)
400{
401 struct emx_softc *sc = device_get_softc(dev);
402 struct ifnet *ifp = &sc->arpcom.ac_if;
7fb43956 403 int error = 0, i;
704b6287 404 u_int intr_flags;
2d0e5700 405 uint16_t eeprom_data, device_id, apme_mask;
5330213c 406
6d435846
SZ		 407 lwkt_serialize_init(&sc->main_serialize);
408 lwkt_serialize_init(&sc->tx_serialize);
409 for (i = 0; i < EMX_NRX_RING; ++i)
410 lwkt_serialize_init(&sc->rx_data[i].rx_serialize);
411
6d435846
SZ		 412 i = 0;
413 sc->serializes[i++] = &sc->main_serialize;
414 sc->serializes[i++] = &sc->tx_serialize;
415 sc->serializes[i++] = &sc->rx_data[0].rx_serialize;
416 sc->serializes[i++] = &sc->rx_data[1].rx_serialize;
417 KKASSERT(i == EMX_NSERIALIZE);
418
c2022416 419 callout_init_mp(&sc->timer);
5330213c
SZ		 420
421 sc->dev = sc->osdep.dev = dev;
422
423 /*
424 * Determine hardware and mac type
425 */
426 sc->hw.vendor_id = pci_get_vendor(dev);
427 sc->hw.device_id = pci_get_device(dev);
428 sc->hw.revision_id = pci_get_revid(dev);
429 sc->hw.subsystem_vendor_id = pci_get_subvendor(dev);
430 sc->hw.subsystem_device_id = pci_get_subdevice(dev);
431
432 if (e1000_set_mac_type(&sc->hw))
433 return ENXIO;
434
435 /* Enable bus mastering */
436 pci_enable_busmaster(dev);
437
438 /*
439 * Allocate IO memory
440 */
441 sc->memory_rid = EMX_BAR_MEM;
442 sc->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
443 &sc->memory_rid, RF_ACTIVE);
444 if (sc->memory == NULL) {
445 device_printf(dev, "Unable to allocate bus resource: memory\n");
446 error = ENXIO;
447 goto fail;
448 }
449 sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->memory);
450 sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->memory);
451
452 /* XXX This is quite goofy, it is not actually used */
453 sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
454
455 /*
456 * Allocate interrupt
457 */
7fb43956
SZ		 458 sc->intr_type = pci_alloc_1intr(dev, emx_msi_enable,
459 &sc->intr_rid, &intr_flags);
704b6287 460
5330213c 461 sc->intr_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->intr_rid,
704b6287 462 intr_flags);
5330213c
SZ		 463 if (sc->intr_res == NULL) {
464 device_printf(dev, "Unable to allocate bus resource: "
465 "interrupt\n");
466 error = ENXIO;
467 goto fail;
468 }
469
470 /* Save PCI command register for Shared Code */
471 sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
472 sc->hw.back = &sc->osdep;
473
474 /* Do Shared Code initialization */
475 if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
476 device_printf(dev, "Setup of Shared code failed\n");
477 error = ENXIO;
478 goto fail;
479 }
480 e1000_get_bus_info(&sc->hw);
481
5330213c
SZ		 482 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
483 sc->hw.phy.autoneg_wait_to_complete = FALSE;
484 sc->hw.phy.autoneg_advertised = EMX_AUTONEG_ADV_DEFAULT;
5330213c
SZ		 485
486 /*
487 * Interrupt throttle rate
488 */
489 if (emx_int_throttle_ceil == 0) {
490 sc->int_throttle_ceil = 0;
491 } else {
492 int throttle = emx_int_throttle_ceil;
493
494 if (throttle < 0)
495 throttle = EMX_DEFAULT_ITR;
496
497 /* Recalculate the tunable value to get the exact frequency. */
498 throttle = 1000000000 / 256 / throttle;
499
500 /* Upper 16bits of ITR is reserved and should be zero */
501 if (throttle & 0xffff0000)
502 throttle = 1000000000 / 256 / EMX_DEFAULT_ITR;
503
504 sc->int_throttle_ceil = 1000000000 / 256 / throttle;
505 }
506
507 e1000_init_script_state_82541(&sc->hw, TRUE);
508 e1000_set_tbi_compatibility_82543(&sc->hw, TRUE);
509
510 /* Copper options */
511 if (sc->hw.phy.media_type == e1000_media_type_copper) {
512 sc->hw.phy.mdix = EMX_AUTO_ALL_MODES;
513 sc->hw.phy.disable_polarity_correction = FALSE;
514 sc->hw.phy.ms_type = EMX_MASTER_SLAVE;
515 }
516
517 /* Set the frame limits assuming standard ethernet sized frames. */
518 sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
519 sc->min_frame_size = ETHER_MIN_LEN;
520
521 /* This controls when hardware reports transmit completion status. */
522 sc->hw.mac.report_tx_early = 1;
523
65c7a6af 524 /* Calculate # of RX rings */
8434a83b 525 if (ncpus > 1)
65c7a6af
SZ		 526 sc->rx_ring_cnt = EMX_NRX_RING;
527 else
528 sc->rx_ring_cnt = 1;
8434a83b 529 sc->rx_ring_inuse = sc->rx_ring_cnt;
65c7a6af 530
071699f8
SZ		 531 /* Allocate RX/TX rings' busdma(9) stuffs */
532 error = emx_dma_alloc(sc);
533 if (error)
e5b3bcc4 534 goto fail;
e5b3bcc4 535
2d0e5700
SZ		 536 /* Allocate multicast array memory. */
537 sc->mta = kmalloc(ETH_ADDR_LEN * EMX_MCAST_ADDR_MAX,
538 M_DEVBUF, M_WAITOK);
539
540 /* Indicate SOL/IDER usage */
541 if (e1000_check_reset_block(&sc->hw)) {
542 device_printf(dev,
543 "PHY reset is blocked due to SOL/IDER session.\n");
544 }
545
546 /*
547 * Start from a known state, this is important in reading the
548 * nvm and mac from that.
549 */
550 e1000_reset_hw(&sc->hw);
551
5330213c
SZ		 552 /* Make sure we have a good EEPROM before we read from it */
553 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
554 /*
555 * Some PCI-E parts fail the first check due to
556 * the link being in sleep state, call it again,
557 * if it fails a second time its a real issue.
558 */
559 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
560 device_printf(dev,
561 "The EEPROM Checksum Is Not Valid\n");
562 error = EIO;
563 goto fail;
564 }
565 }
566
5330213c
SZ		 567 /* Copy the permanent MAC address out of the EEPROM */
568 if (e1000_read_mac_addr(&sc->hw) < 0) {
569 device_printf(dev, "EEPROM read error while reading MAC"
570 " address\n");
571 error = EIO;
572 goto fail;
573 }
574 if (!emx_is_valid_eaddr(sc->hw.mac.addr)) {
575 device_printf(dev, "Invalid MAC address\n");
576 error = EIO;
577 goto fail;
578 }
579
5330213c
SZ		 580 /* Determine if we have to control management hardware */
581 sc->has_manage = e1000_enable_mng_pass_thru(&sc->hw);
582
583 /*
584 * Setup Wake-on-Lan
585 */
2d0e5700
SZ		 586 apme_mask = EMX_EEPROM_APME;
587 eeprom_data = 0;
5330213c 588 switch (sc->hw.mac.type) {
2d0e5700
SZ		 589 case e1000_82573:
590 sc->has_amt = 1;
591 /* FALL THROUGH */
592
5330213c 593 case e1000_82571:
2d0e5700 594 case e1000_82572:
5330213c
SZ		 595 case e1000_80003es2lan:
596 if (sc->hw.bus.func == 1) {
597 e1000_read_nvm(&sc->hw,
598 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
599 } else {
600 e1000_read_nvm(&sc->hw,
601 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
602 }
5330213c
SZ		 603 break;
604
605 default:
2d0e5700
SZ		 606 e1000_read_nvm(&sc->hw,
607 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5330213c
SZ		 608 break;
609 }
2d0e5700
SZ		 610 if (eeprom_data & apme_mask)
611 sc->wol = E1000_WUFC_MAG | E1000_WUFC_MC;
612
5330213c
SZ		 613 /*
614 * We have the eeprom settings, now apply the special cases
615 * where the eeprom may be wrong or the board won't support
616 * wake on lan on a particular port
617 */
618 device_id = pci_get_device(dev);
619 switch (device_id) {
620 case E1000_DEV_ID_82571EB_FIBER:
621 /*
622 * Wake events only supported on port A for dual fiber
623 * regardless of eeprom setting
624 */
625 if (E1000_READ_REG(&sc->hw, E1000_STATUS) &
626 E1000_STATUS_FUNC_1)
627 sc->wol = 0;
628 break;
629
630 case E1000_DEV_ID_82571EB_QUAD_COPPER:
631 case E1000_DEV_ID_82571EB_QUAD_FIBER:
632 case E1000_DEV_ID_82571EB_QUAD_COPPER_LP:
633 /* if quad port sc, disable WoL on all but port A */
634 if (emx_global_quad_port_a != 0)
635 sc->wol = 0;
636 /* Reset for multiple quad port adapters */
637 if (++emx_global_quad_port_a == 4)
638 emx_global_quad_port_a = 0;
639 break;
640 }
641
642 /* XXX disable wol */
643 sc->wol = 0;
644
2d0e5700
SZ		 645 /* Setup OS specific network interface */
646 emx_setup_ifp(sc);
647
648 /* Add sysctl tree, must after em_setup_ifp() */
649 emx_add_sysctl(sc);
650
651 /* Reset the hardware */
652 error = emx_reset(sc);
653 if (error) {
654 device_printf(dev, "Unable to reset the hardware\n");
655 goto fail;
656 }
657
658 /* Initialize statistics */
659 emx_update_stats(sc);
660
661 sc->hw.mac.get_link_status = 1;
662 emx_update_link_status(sc);
663
5330213c
SZ		 664 sc->spare_tx_desc = EMX_TX_SPARE;
665
666 /*
667 * Keep following relationship between spare_tx_desc, oact_tx_desc
668 * and tx_int_nsegs:
669 * (spare_tx_desc + EMX_TX_RESERVED) <=
670 * oact_tx_desc <= EMX_TX_OACTIVE_MAX <= tx_int_nsegs
671 */
672 sc->oact_tx_desc = sc->num_tx_desc / 8;
673 if (sc->oact_tx_desc > EMX_TX_OACTIVE_MAX)
674 sc->oact_tx_desc = EMX_TX_OACTIVE_MAX;
675 if (sc->oact_tx_desc < sc->spare_tx_desc + EMX_TX_RESERVED)
676 sc->oact_tx_desc = sc->spare_tx_desc + EMX_TX_RESERVED;
677
678 sc->tx_int_nsegs = sc->num_tx_desc / 16;
679 if (sc->tx_int_nsegs < sc->oact_tx_desc)
680 sc->tx_int_nsegs = sc->oact_tx_desc;
681
2d0e5700
SZ		 682 /* Non-AMT based hardware can now take control from firmware */
683 if (sc->has_manage && !sc->has_amt)
684 emx_get_hw_control(sc);
685
5330213c 686 error = bus_setup_intr(dev, sc->intr_res, INTR_MPSAFE, emx_intr, sc,
6d435846 687 &sc->intr_tag, &sc->main_serialize);
5330213c
SZ		 688 if (error) {
689 device_printf(dev, "Failed to register interrupt handler");
690 ether_ifdetach(&sc->arpcom.ac_if);
691 goto fail;
692 }
693
704b6287 694 ifp->if_cpuid = rman_get_cpuid(sc->intr_res);
5330213c
SZ		 695 KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
696 return (0);
697fail:
698 emx_detach(dev);
699 return (error);
700}
701
702static int
703emx_detach(device_t dev)
704{
705 struct emx_softc *sc = device_get_softc(dev);
706
707 if (device_is_attached(dev)) {
708 struct ifnet *ifp = &sc->arpcom.ac_if;
709
6d435846 710 ifnet_serialize_all(ifp);
5330213c
SZ		 711
712 emx_stop(sc);
713
714 e1000_phy_hw_reset(&sc->hw);
715
716 emx_rel_mgmt(sc);
2d0e5700 717 emx_rel_hw_control(sc);
5330213c
SZ		 718
719 if (sc->wol) {
720 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
721 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
722 emx_enable_wol(dev);
723 }
724
725 bus_teardown_intr(dev, sc->intr_res, sc->intr_tag);
726
6d435846 727 ifnet_deserialize_all(ifp);
5330213c
SZ		 728
729 ether_ifdetach(ifp);
2d0e5700
SZ		 730 } else {
731 emx_rel_hw_control(sc);
5330213c
SZ		 732 }
733 bus_generic_detach(dev);
734
735 if (sc->intr_res != NULL) {
736 bus_release_resource(dev, SYS_RES_IRQ, sc->intr_rid,
737 sc->intr_res);
738 }
739
7fb43956 740 if (sc->intr_type == PCI_INTR_TYPE_MSI)
704b6287
SZ		 741 pci_release_msi(dev);
742
5330213c
SZ		 743 if (sc->memory != NULL) {
744 bus_release_resource(dev, SYS_RES_MEMORY, sc->memory_rid,
745 sc->memory);
746 }
747
071699f8 748 emx_dma_free(sc);
5330213c
SZ		 749
750 /* Free sysctl tree */
751 if (sc->sysctl_tree != NULL)
752 sysctl_ctx_free(&sc->sysctl_ctx);
753
754 return (0);
755}
756
757static int
758emx_shutdown(device_t dev)
759{
760 return emx_suspend(dev);
761}
762
763static int
764emx_suspend(device_t dev)
765{
766 struct emx_softc *sc = device_get_softc(dev);
767 struct ifnet *ifp = &sc->arpcom.ac_if;
768
6d435846 769 ifnet_serialize_all(ifp);
5330213c
SZ		 770
771 emx_stop(sc);
772
773 emx_rel_mgmt(sc);
2d0e5700 774 emx_rel_hw_control(sc);
5330213c 775
2d0e5700 776 if (sc->wol) {
5330213c
SZ		 777 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
778 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
779 emx_enable_wol(dev);
2d0e5700 780 }
5330213c 781
6d435846 782 ifnet_deserialize_all(ifp);
5330213c
SZ		 783
784 return bus_generic_suspend(dev);
785}
786
787static int
788emx_resume(device_t dev)
789{
790 struct emx_softc *sc = device_get_softc(dev);
791 struct ifnet *ifp = &sc->arpcom.ac_if;
792
6d435846 793 ifnet_serialize_all(ifp);
5330213c
SZ		 794
795 emx_init(sc);
796 emx_get_mgmt(sc);
797 if_devstart(ifp);
798
6d435846 799 ifnet_deserialize_all(ifp);
5330213c
SZ		 800
801 return bus_generic_resume(dev);
802}
803
804static void
805emx_start(struct ifnet *ifp)
806{
807 struct emx_softc *sc = ifp->if_softc;
808 struct mbuf *m_head;
809
6d435846 810 ASSERT_SERIALIZED(&sc->tx_serialize);
5330213c
SZ		 811
812 if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
813 return;
814
815 if (!sc->link_active) {
816 ifq_purge(&ifp->if_snd);
817 return;
818 }
819
820 while (!ifq_is_empty(&ifp->if_snd)) {
821 /* Now do we at least have a minimal? */
822 if (EMX_IS_OACTIVE(sc)) {
823 emx_tx_collect(sc);
824 if (EMX_IS_OACTIVE(sc)) {
825 ifp->if_flags |= IFF_OACTIVE;
826 sc->no_tx_desc_avail1++;
827 break;
828 }
829 }
830
831 logif(pkt_txqueue);
832 m_head = ifq_dequeue(&ifp->if_snd, NULL);
833 if (m_head == NULL)
834 break;
835
836 if (emx_encap(sc, &m_head)) {
837 ifp->if_oerrors++;
838 emx_tx_collect(sc);
839 continue;
840 }
841
842 /* Send a copy of the frame to the BPF listener */
843 ETHER_BPF_MTAP(ifp, m_head);
844
845 /* Set timeout in case hardware has problems transmitting. */
846 ifp->if_timer = EMX_TX_TIMEOUT;
847 }
848}
849
850static int
851emx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
852{
853 struct emx_softc *sc = ifp->if_softc;
854 struct ifreq *ifr = (struct ifreq *)data;
855 uint16_t eeprom_data = 0;
856 int max_frame_size, mask, reinit;
857 int error = 0;
858
2c9effcf 859 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 860
861 switch (command) {
862 case SIOCSIFMTU:
863 switch (sc->hw.mac.type) {
864 case e1000_82573:
865 /*
866 * 82573 only supports jumbo frames
867 * if ASPM is disabled.
868 */
869 e1000_read_nvm(&sc->hw, NVM_INIT_3GIO_3, 1,
870 &eeprom_data);
871 if (eeprom_data & NVM_WORD1A_ASPM_MASK) {
872 max_frame_size = ETHER_MAX_LEN;
873 break;
874 }
875 /* FALL THROUGH */
876
877 /* Limit Jumbo Frame size */
878 case e1000_82571:
879 case e1000_82572:
880 case e1000_82574:
881 case e1000_80003es2lan:
882 max_frame_size = 9234;
883 break;
884
885 default:
886 max_frame_size = MAX_JUMBO_FRAME_SIZE;
887 break;
888 }
889 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
890 ETHER_CRC_LEN) {
891 error = EINVAL;
892 break;
893 }
894
895 ifp->if_mtu = ifr->ifr_mtu;
896 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
897 ETHER_CRC_LEN;
898
899 if (ifp->if_flags & IFF_RUNNING)
900 emx_init(sc);
901 break;
902
903 case SIOCSIFFLAGS:
904 if (ifp->if_flags & IFF_UP) {
905 if ((ifp->if_flags & IFF_RUNNING)) {
906 if ((ifp->if_flags ^ sc->if_flags) &
907 (IFF_PROMISC | IFF_ALLMULTI)) {
908 emx_disable_promisc(sc);
909 emx_set_promisc(sc);
910 }
911 } else {
912 emx_init(sc);
913 }
914 } else if (ifp->if_flags & IFF_RUNNING) {
915 emx_stop(sc);
916 }
917 sc->if_flags = ifp->if_flags;
918 break;
919
920 case SIOCADDMULTI:
921 case SIOCDELMULTI:
922 if (ifp->if_flags & IFF_RUNNING) {
923 emx_disable_intr(sc);
924 emx_set_multi(sc);
b3a7093f
SZ		 925#ifdef IFPOLL_ENABLE
926 if (!(ifp->if_flags & IFF_NPOLLING))
5330213c
SZ		 927#endif
928 emx_enable_intr(sc);
929 }
930 break;
931
932 case SIOCSIFMEDIA:
933 /* Check SOL/IDER usage */
934 if (e1000_check_reset_block(&sc->hw)) {
935 device_printf(sc->dev, "Media change is"
936 " blocked due to SOL/IDER session.\n");
937 break;
938 }
939 /* FALL THROUGH */
940
941 case SIOCGIFMEDIA:
942 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
943 break;
944
945 case SIOCSIFCAP:
946 reinit = 0;
947 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
948 if (mask & IFCAP_HWCSUM) {
949 ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
950 reinit = 1;
951 }
952 if (mask & IFCAP_VLAN_HWTAGGING) {
953 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
954 reinit = 1;
955 }
8434a83b
SZ		 956 if (mask & IFCAP_RSS) {
957 ifp->if_capenable ^= IFCAP_RSS;
958 reinit = 1;
959 }
5330213c
SZ		 960 if (reinit && (ifp->if_flags & IFF_RUNNING))
961 emx_init(sc);
962 break;
963
964 default:
965 error = ether_ioctl(ifp, command, data);
966 break;
967 }
968 return (error);
969}
970
971static void
972emx_watchdog(struct ifnet *ifp)
973{
974 struct emx_softc *sc = ifp->if_softc;
975
2c9effcf 976 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 977
978 /*
979 * The timer is set to 5 every time start queues a packet.
980 * Then txeof keeps resetting it as long as it cleans at
981 * least one descriptor.
982 * Finally, anytime all descriptors are clean the timer is
983 * set to 0.
984 */
985
986 if (E1000_READ_REG(&sc->hw, E1000_TDT(0)) ==
987 E1000_READ_REG(&sc->hw, E1000_TDH(0))) {
988 /*
989 * If we reach here, all TX jobs are completed and
990 * the TX engine should have been idled for some time.
991 * We don't need to call if_devstart() here.
992 */
993 ifp->if_flags &= ~IFF_OACTIVE;
994 ifp->if_timer = 0;
995 return;
996 }
997
998 /*
999 * If we are in this routine because of pause frames, then
1000 * don't reset the hardware.
1001 */
1002 if (E1000_READ_REG(&sc->hw, E1000_STATUS) & E1000_STATUS_TXOFF) {
1003 ifp->if_timer = EMX_TX_TIMEOUT;
1004 return;
1005 }
1006
1007 if (e1000_check_for_link(&sc->hw) == 0)
1008 if_printf(ifp, "watchdog timeout -- resetting\n");
1009
1010 ifp->if_oerrors++;
1011 sc->watchdog_events++;
1012
1013 emx_init(sc);
1014
1015 if (!ifq_is_empty(&ifp->if_snd))
1016 if_devstart(ifp);
1017}
1018
1019static void
1020emx_init(void *xsc)
1021{
1022 struct emx_softc *sc = xsc;
1023 struct ifnet *ifp = &sc->arpcom.ac_if;
1024 device_t dev = sc->dev;
1025 uint32_t pba;
3f939c23 1026 int i;
5330213c 1027
2c9effcf 1028 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 1029
1030 emx_stop(sc);
1031
1032 /*
1033 * Packet Buffer Allocation (PBA)
1034 * Writing PBA sets the receive portion of the buffer
1035 * the remainder is used for the transmit buffer.
1036 */
1037 switch (sc->hw.mac.type) {
1038 /* Total Packet Buffer on these is 48K */
1039 case e1000_82571:
1040 case e1000_82572:
1041 case e1000_80003es2lan:
1042 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1043 break;
1044
1045 case e1000_82573: /* 82573: Total Packet Buffer is 32K */
1046 pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
1047 break;
1048
1049 case e1000_82574:
1050 pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
1051 break;
1052
1053 default:
1054 /* Devices before 82547 had a Packet Buffer of 64K. */
1055 if (sc->max_frame_size > 8192)
1056 pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
1057 else
1058 pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
1059 }
1060 E1000_WRITE_REG(&sc->hw, E1000_PBA, pba);
1061
1062 /* Get the latest mac address, User can use a LAA */
1063 bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
1064
1065 /* Put the address into the Receive Address Array */
1066 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
1067
1068 /*
1069 * With the 82571 sc, RAR[0] may be overwritten
1070 * when the other port is reset, we make a duplicate
1071 * in RAR[14] for that eventuality, this assures
1072 * the interface continues to function.
1073 */
1074 if (sc->hw.mac.type == e1000_82571) {
1075 e1000_set_laa_state_82571(&sc->hw, TRUE);
1076 e1000_rar_set(&sc->hw, sc->hw.mac.addr,
1077 E1000_RAR_ENTRIES - 1);
1078 }
1079
1080 /* Initialize the hardware */
2d0e5700
SZ		 1081 if (emx_reset(sc)) {
1082 device_printf(dev, "Unable to reset the hardware\n");
5330213c
SZ		 1083 /* XXX emx_stop()? */
1084 return;
1085 }
1086 emx_update_link_status(sc);
1087
1088 /* Setup VLAN support, basic and offload if available */
1089 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1090
1091 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1092 uint32_t ctrl;
1093
1094 ctrl = E1000_READ_REG(&sc->hw, E1000_CTRL);
1095 ctrl |= E1000_CTRL_VME;
1096 E1000_WRITE_REG(&sc->hw, E1000_CTRL, ctrl);
1097 }
1098
1099 /* Set hardware offload abilities */
1100 if (ifp->if_capenable & IFCAP_TXCSUM)
1101 ifp->if_hwassist = EMX_CSUM_FEATURES;
1102 else
1103 ifp->if_hwassist = 0;
1104
1105 /* Configure for OS presence */
1106 emx_get_mgmt(sc);
1107
1108 /* Prepare transmit descriptors and buffers */
1109 emx_init_tx_ring(sc);
1110 emx_init_tx_unit(sc);
1111
1112 /* Setup Multicast table */
1113 emx_set_multi(sc);
1114
8434a83b
SZ		 1115 /*
1116 * Adjust # of RX ring to be used based on IFCAP_RSS
1117 */
1118 if (ifp->if_capenable & IFCAP_RSS)
1119 sc->rx_ring_inuse = sc->rx_ring_cnt;
1120 else
1121 sc->rx_ring_inuse = 1;
1122
5330213c 1123 /* Prepare receive descriptors and buffers */
8434a83b 1124 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3f939c23
SZ		 1125 if (emx_init_rx_ring(sc, &sc->rx_data[i])) {
1126 device_printf(dev,
1127 "Could not setup receive structures\n");
1128 emx_stop(sc);
1129 return;
1130 }
5330213c
SZ		 1131 }
1132 emx_init_rx_unit(sc);
1133
1134 /* Don't lose promiscuous settings */
1135 emx_set_promisc(sc);
1136
1137 ifp->if_flags |= IFF_RUNNING;
1138 ifp->if_flags &= ~IFF_OACTIVE;
1139
1140 callout_reset(&sc->timer, hz, emx_timer, sc);
1141 e1000_clear_hw_cntrs_base_generic(&sc->hw);
1142
1143 /* MSI/X configuration for 82574 */
1144 if (sc->hw.mac.type == e1000_82574) {
1145 int tmp;
1146
1147 tmp = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
1148 tmp |= E1000_CTRL_EXT_PBA_CLR;
1149 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT, tmp);
1150 /*
2d0e5700 1151 * XXX MSIX
5330213c
SZ		 1152 * Set the IVAR - interrupt vector routing.
1153 * Each nibble represents a vector, high bit
1154 * is enable, other 3 bits are the MSIX table
1155 * entry, we map RXQ0 to 0, TXQ0 to 1, and
1156 * Link (other) to 2, hence the magic number.
1157 */
1158 E1000_WRITE_REG(&sc->hw, E1000_IVAR, 0x800A0908);
1159 }
1160
b3a7093f 1161#ifdef IFPOLL_ENABLE
5330213c
SZ		 1162 /*
1163 * Only enable interrupts if we are not polling, make sure
1164 * they are off otherwise.
1165 */
b3a7093f 1166 if (ifp->if_flags & IFF_NPOLLING)
5330213c
SZ		 1167 emx_disable_intr(sc);
1168 else
b3a7093f 1169#endif /* IFPOLL_ENABLE */
5330213c
SZ		 1170 emx_enable_intr(sc);
1171
2d0e5700
SZ		 1172 /* AMT based hardware can now take control from firmware */
1173 if (sc->has_manage && sc->has_amt)
1174 emx_get_hw_control(sc);
1175
5330213c
SZ		 1176 /* Don't reset the phy next time init gets called */
1177 sc->hw.phy.reset_disable = TRUE;
1178}
1179
5330213c
SZ		 1180static void
1181emx_intr(void *xsc)
1182{
1183 struct emx_softc *sc = xsc;
1184 struct ifnet *ifp = &sc->arpcom.ac_if;
1185 uint32_t reg_icr;
1186
1187 logif(intr_beg);
6d435846 1188 ASSERT_SERIALIZED(&sc->main_serialize);
5330213c
SZ		 1189
1190 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
1191
1192 if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0) {
1193 logif(intr_end);
1194 return;
1195 }
1196
1197 /*
1198 * XXX: some laptops trigger several spurious interrupts
df50f778 1199 * on emx(4) when in the resume cycle. The ICR register
5330213c
SZ		 1200 * reports all-ones value in this case. Processing such
1201 * interrupts would lead to a freeze. I don't know why.
1202 */
1203 if (reg_icr == 0xffffffff) {
1204 logif(intr_end);
1205 return;
1206 }
1207
1208 if (ifp->if_flags & IFF_RUNNING) {
1209 if (reg_icr &
3f939c23
SZ		 1210 (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
1211 int i;
1212
6d435846
SZ		 1213 for (i = 0; i < sc->rx_ring_inuse; ++i) {
1214 lwkt_serialize_enter(
1215 &sc->rx_data[i].rx_serialize);
3f939c23 1216 emx_rxeof(sc, i, -1);
6d435846
SZ		 1217 lwkt_serialize_exit(
1218 &sc->rx_data[i].rx_serialize);
1219 }
3f939c23 1220 }
6446af7b 1221 if (reg_icr & E1000_ICR_TXDW) {
6d435846 1222 lwkt_serialize_enter(&sc->tx_serialize);
5330213c
SZ		 1223 emx_txeof(sc);
1224 if (!ifq_is_empty(&ifp->if_snd))
1225 if_devstart(ifp);
6d435846 1226 lwkt_serialize_exit(&sc->tx_serialize);
5330213c
SZ		 1227 }
1228 }
1229
1230 /* Link status change */
1231 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
bca7c435 1232 emx_serialize_skipmain(sc);
6d435846 1233
5330213c
SZ		 1234 callout_stop(&sc->timer);
1235 sc->hw.mac.get_link_status = 1;
1236 emx_update_link_status(sc);
1237
1238 /* Deal with TX cruft when link lost */
1239 emx_tx_purge(sc);
1240
1241 callout_reset(&sc->timer, hz, emx_timer, sc);
6d435846 1242
bca7c435 1243 emx_deserialize_skipmain(sc);
5330213c
SZ		 1244 }
1245
1246 if (reg_icr & E1000_ICR_RXO)
1247 sc->rx_overruns++;
1248
1249 logif(intr_end);
1250}
1251
1252static void
1253emx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1254{
1255 struct emx_softc *sc = ifp->if_softc;
1256
2c9effcf 1257 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 1258
1259 emx_update_link_status(sc);
1260
1261 ifmr->ifm_status = IFM_AVALID;
1262 ifmr->ifm_active = IFM_ETHER;
1263
1264 if (!sc->link_active)
1265 return;
1266
1267 ifmr->ifm_status |= IFM_ACTIVE;
1268
1269 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1270 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1271 ifmr->ifm_active |= IFM_1000_SX | IFM_FDX;
1272 } else {
1273 switch (sc->link_speed) {
1274 case 10:
1275 ifmr->ifm_active |= IFM_10_T;
1276 break;
1277 case 100:
1278 ifmr->ifm_active |= IFM_100_TX;
1279 break;
1280
1281 case 1000:
1282 ifmr->ifm_active |= IFM_1000_T;
1283 break;
1284 }
1285 if (sc->link_duplex == FULL_DUPLEX)
1286 ifmr->ifm_active |= IFM_FDX;
1287 else
1288 ifmr->ifm_active |= IFM_HDX;
1289 }
1290}
1291
1292static int
1293emx_media_change(struct ifnet *ifp)
1294{
1295 struct emx_softc *sc = ifp->if_softc;
1296 struct ifmedia *ifm = &sc->media;
1297
2c9effcf 1298 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 1299
1300 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1301 return (EINVAL);
1302
1303 switch (IFM_SUBTYPE(ifm->ifm_media)) {
1304 case IFM_AUTO:
1305 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
1306 sc->hw.phy.autoneg_advertised = EMX_AUTONEG_ADV_DEFAULT;
1307 break;
1308
1309 case IFM_1000_LX:
1310 case IFM_1000_SX:
1311 case IFM_1000_T:
1312 sc->hw.mac.autoneg = EMX_DO_AUTO_NEG;
1313 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1314 break;
1315
1316 case IFM_100_TX:
1317 sc->hw.mac.autoneg = FALSE;
1318 sc->hw.phy.autoneg_advertised = 0;
1319 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1320 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1321 else
1322 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1323 break;
1324
1325 case IFM_10_T:
1326 sc->hw.mac.autoneg = FALSE;
1327 sc->hw.phy.autoneg_advertised = 0;
1328 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1329 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1330 else
1331 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1332 break;
1333
1334 default:
1335 if_printf(ifp, "Unsupported media type\n");
1336 break;
1337 }
1338
1339 /*
1340 * As the speed/duplex settings my have changed we need to
1341 * reset the PHY.
1342 */
1343 sc->hw.phy.reset_disable = FALSE;
1344
1345 emx_init(sc);
1346
1347 return (0);
1348}
1349
1350static int
1351emx_encap(struct emx_softc *sc, struct mbuf **m_headp)
1352{
1353 bus_dma_segment_t segs[EMX_MAX_SCATTER];
1354 bus_dmamap_t map;
323e5ecd 1355 struct emx_txbuf *tx_buffer, *tx_buffer_mapped;
5330213c
SZ		 1356 struct e1000_tx_desc *ctxd = NULL;
1357 struct mbuf *m_head = *m_headp;
1358 uint32_t txd_upper, txd_lower, cmd = 0;
1359 int maxsegs, nsegs, i, j, first, last = 0, error;
1360
3752657e 1361 if (m_head->m_len < EMX_TXCSUM_MINHL &&
5330213c
SZ		 1362 (m_head->m_flags & EMX_CSUM_FEATURES)) {
1363 /*
1364 * Make sure that ethernet header and ip.ip_hl are in
1365 * contiguous memory, since if TXCSUM is enabled, later
1366 * TX context descriptor's setup need to access ip.ip_hl.
1367 */
1368 error = emx_txcsum_pullup(sc, m_headp);
1369 if (error) {
1370 KKASSERT(*m_headp == NULL);
1371 return error;
1372 }
1373 m_head = *m_headp;
1374 }
1375
1376 txd_upper = txd_lower = 0;
1377
1378 /*
1379 * Capture the first descriptor index, this descriptor
1380 * will have the index of the EOP which is the only one
1381 * that now gets a DONE bit writeback.
1382 */
1383 first = sc->next_avail_tx_desc;
323e5ecd 1384 tx_buffer = &sc->tx_buf[first];
5330213c
SZ		 1385 tx_buffer_mapped = tx_buffer;
1386 map = tx_buffer->map;
1387
1388 maxsegs = sc->num_tx_desc_avail - EMX_TX_RESERVED;
1389 KASSERT(maxsegs >= sc->spare_tx_desc, ("not enough spare TX desc\n"));
1390 if (maxsegs > EMX_MAX_SCATTER)
1391 maxsegs = EMX_MAX_SCATTER;
1392
1393 error = bus_dmamap_load_mbuf_defrag(sc->txtag, map, m_headp,
1394 segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1395 if (error) {
1396 if (error == ENOBUFS)
1397 sc->mbuf_alloc_failed++;
1398 else
1399 sc->no_tx_dma_setup++;
1400
1401 m_freem(*m_headp);
1402 *m_headp = NULL;
1403 return error;
1404 }
1405 bus_dmamap_sync(sc->txtag, map, BUS_DMASYNC_PREWRITE);
1406
1407 m_head = *m_headp;
1408 sc->tx_nsegs += nsegs;
1409
1410 if (m_head->m_pkthdr.csum_flags & EMX_CSUM_FEATURES) {
1411 /* TX csum offloading will consume one TX desc */
1412 sc->tx_nsegs += emx_txcsum(sc, m_head, &txd_upper, &txd_lower);
1413 }
1414 i = sc->next_avail_tx_desc;
1415
1416 /* Set up our transmit descriptors */
1417 for (j = 0; j < nsegs; j++) {
323e5ecd 1418 tx_buffer = &sc->tx_buf[i];
5330213c
SZ		 1419 ctxd = &sc->tx_desc_base[i];
1420
1421 ctxd->buffer_addr = htole64(segs[j].ds_addr);
1422 ctxd->lower.data = htole32(E1000_TXD_CMD_IFCS |
1423 txd_lower | segs[j].ds_len);
1424 ctxd->upper.data = htole32(txd_upper);
1425
1426 last = i;
1427 if (++i == sc->num_tx_desc)
1428 i = 0;
5330213c
SZ		 1429 }
1430
1431 sc->next_avail_tx_desc = i;
1432
1433 KKASSERT(sc->num_tx_desc_avail > nsegs);
1434 sc->num_tx_desc_avail -= nsegs;
1435
1436 /* Handle VLAN tag */
1437 if (m_head->m_flags & M_VLANTAG) {
1438 /* Set the vlan id. */
1439 ctxd->upper.fields.special =
1440 htole16(m_head->m_pkthdr.ether_vlantag);
1441
1442 /* Tell hardware to add tag */
1443 ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE);
1444 }
1445
1446 tx_buffer->m_head = m_head;
1447 tx_buffer_mapped->map = tx_buffer->map;
1448 tx_buffer->map = map;
1449
1450 if (sc->tx_nsegs >= sc->tx_int_nsegs) {
1451 sc->tx_nsegs = 0;
4e4e8481
SZ		 1452
1453 /*
1454 * Report Status (RS) is turned on
1455 * every tx_int_nsegs descriptors.
1456 */
5330213c
SZ		 1457 cmd = E1000_TXD_CMD_RS;
1458
b4b0a2b4
SZ		 1459 /*
1460 * Keep track of the descriptor, which will
1461 * be written back by hardware.
1462 */
5330213c
SZ		 1463 sc->tx_dd[sc->tx_dd_tail] = last;
1464 EMX_INC_TXDD_IDX(sc->tx_dd_tail);
1465 KKASSERT(sc->tx_dd_tail != sc->tx_dd_head);
1466 }
1467
1468 /*
1469 * Last Descriptor of Packet needs End Of Packet (EOP)
5330213c
SZ		 1470 */
1471 ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | cmd);
1472
1473 /*
5330213c
SZ		 1474 * Advance the Transmit Descriptor Tail (TDT), this tells
1475 * the E1000 that this frame is available to transmit.
1476 */
1477 E1000_WRITE_REG(&sc->hw, E1000_TDT(0), i);
1478
1479 return (0);
1480}
1481
1482static void
1483emx_set_promisc(struct emx_softc *sc)
1484{
1485 struct ifnet *ifp = &sc->arpcom.ac_if;
1486 uint32_t reg_rctl;
1487
1488 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1489
1490 if (ifp->if_flags & IFF_PROMISC) {
1491 reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1492 /* Turn this on if you want to see bad packets */
1493 if (emx_debug_sbp)
1494 reg_rctl |= E1000_RCTL_SBP;
1495 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1496 } else if (ifp->if_flags & IFF_ALLMULTI) {
1497 reg_rctl |= E1000_RCTL_MPE;
1498 reg_rctl &= ~E1000_RCTL_UPE;
1499 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1500 }
1501}
1502
1503static void
1504emx_disable_promisc(struct emx_softc *sc)
1505{
1506 uint32_t reg_rctl;
1507
1508 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1509
1510 reg_rctl &= ~E1000_RCTL_UPE;
1511 reg_rctl &= ~E1000_RCTL_MPE;
1512 reg_rctl &= ~E1000_RCTL_SBP;
1513 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1514}
1515
1516static void
1517emx_set_multi(struct emx_softc *sc)
1518{
1519 struct ifnet *ifp = &sc->arpcom.ac_if;
1520 struct ifmultiaddr *ifma;
1521 uint32_t reg_rctl = 0;
2d0e5700 1522 uint8_t *mta;
5330213c
SZ		 1523 int mcnt = 0;
1524
2d0e5700
SZ		 1525 mta = sc->mta;
1526 bzero(mta, ETH_ADDR_LEN * EMX_MCAST_ADDR_MAX);
1527
441d34b2 1528 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
5330213c
SZ		 1529 if (ifma->ifma_addr->sa_family != AF_LINK)
1530 continue;
1531
1532 if (mcnt == EMX_MCAST_ADDR_MAX)
1533 break;
1534
1535 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1536 &mta[mcnt * ETHER_ADDR_LEN], ETHER_ADDR_LEN);
1537 mcnt++;
1538 }
1539
1540 if (mcnt >= EMX_MCAST_ADDR_MAX) {
1541 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1542 reg_rctl |= E1000_RCTL_MPE;
1543 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1544 } else {
6a5a645e 1545 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
5330213c
SZ		 1546 }
1547}
1548
1549/*
1550 * This routine checks for link status and updates statistics.
1551 */
1552static void
1553emx_timer(void *xsc)
1554{
1555 struct emx_softc *sc = xsc;
1556 struct ifnet *ifp = &sc->arpcom.ac_if;
1557
6d435846 1558 ifnet_serialize_all(ifp);
5330213c
SZ		 1559
1560 emx_update_link_status(sc);
1561 emx_update_stats(sc);
1562
1563 /* Reset LAA into RAR[0] on 82571 */
1564 if (e1000_get_laa_state_82571(&sc->hw) == TRUE)
1565 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
1566
1567 if (emx_display_debug_stats && (ifp->if_flags & IFF_RUNNING))
1568 emx_print_hw_stats(sc);
1569
1570 emx_smartspeed(sc);
1571
1572 callout_reset(&sc->timer, hz, emx_timer, sc);
1573
6d435846 1574 ifnet_deserialize_all(ifp);
5330213c
SZ		 1575}
1576
1577static void
1578emx_update_link_status(struct emx_softc *sc)
1579{
1580 struct e1000_hw *hw = &sc->hw;
1581 struct ifnet *ifp = &sc->arpcom.ac_if;
1582 device_t dev = sc->dev;
1583 uint32_t link_check = 0;
1584
1585 /* Get the cached link value or read phy for real */
1586 switch (hw->phy.media_type) {
1587 case e1000_media_type_copper:
1588 if (hw->mac.get_link_status) {
1589 /* Do the work to read phy */
1590 e1000_check_for_link(hw);
1591 link_check = !hw->mac.get_link_status;
1592 if (link_check) /* ESB2 fix */
1593 e1000_cfg_on_link_up(hw);
1594 } else {
1595 link_check = TRUE;
1596 }
1597 break;
1598
1599 case e1000_media_type_fiber:
1600 e1000_check_for_link(hw);
1601 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1602 break;
1603
1604 case e1000_media_type_internal_serdes:
1605 e1000_check_for_link(hw);
1606 link_check = sc->hw.mac.serdes_has_link;
1607 break;
1608
1609 case e1000_media_type_unknown:
1610 default:
1611 break;
1612 }
1613
1614 /* Now check for a transition */
1615 if (link_check && sc->link_active == 0) {
1616 e1000_get_speed_and_duplex(hw, &sc->link_speed,
1617 &sc->link_duplex);
1618
1619 /*
1620 * Check if we should enable/disable SPEED_MODE bit on
1621 * 82571EB/82572EI
1622 */
2d0e5700
SZ		 1623 if (sc->link_speed != SPEED_1000 &&
1624 (hw->mac.type == e1000_82571 ||
1625 hw->mac.type == e1000_82572)) {
5330213c
SZ		 1626 int tarc0;
1627
1628 tarc0 = E1000_READ_REG(hw, E1000_TARC(0));
2d0e5700 1629 tarc0 &= ~EMX_TARC_SPEED_MODE;
5330213c
SZ		 1630 E1000_WRITE_REG(hw, E1000_TARC(0), tarc0);
1631 }
1632 if (bootverbose) {
1633 device_printf(dev, "Link is up %d Mbps %s\n",
1634 sc->link_speed,
1635 ((sc->link_duplex == FULL_DUPLEX) ?
1636 "Full Duplex" : "Half Duplex"));
1637 }
1638 sc->link_active = 1;
1639 sc->smartspeed = 0;
1640 ifp->if_baudrate = sc->link_speed * 1000000;
1641 ifp->if_link_state = LINK_STATE_UP;
1642 if_link_state_change(ifp);
1643 } else if (!link_check && sc->link_active == 1) {
1644 ifp->if_baudrate = sc->link_speed = 0;
1645 sc->link_duplex = 0;
1646 if (bootverbose)
1647 device_printf(dev, "Link is Down\n");
1648 sc->link_active = 0;
1649#if 0
1650 /* Link down, disable watchdog */
1651 if->if_timer = 0;
1652#endif
1653 ifp->if_link_state = LINK_STATE_DOWN;
1654 if_link_state_change(ifp);
1655 }
1656}
1657
1658static void
1659emx_stop(struct emx_softc *sc)
1660{
1661 struct ifnet *ifp = &sc->arpcom.ac_if;
1662 int i;
1663
2c9effcf 1664 ASSERT_IFNET_SERIALIZED_ALL(ifp);
5330213c
SZ		 1665
1666 emx_disable_intr(sc);
1667
1668 callout_stop(&sc->timer);
1669
1670 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1671 ifp->if_timer = 0;
1672
3f939c23
SZ		 1673 /*
1674 * Disable multiple receive queues.
1675 *
1676 * NOTE:
1677 * We should disable multiple receive queues before
1678 * resetting the hardware.
1679 */
1680 E1000_WRITE_REG(&sc->hw, E1000_MRQC, 0);
1681
5330213c
SZ		 1682 e1000_reset_hw(&sc->hw);
1683 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1684
1685 for (i = 0; i < sc->num_tx_desc; i++) {
323e5ecd 1686 struct emx_txbuf *tx_buffer = &sc->tx_buf[i];
5330213c
SZ		 1687
1688 if (tx_buffer->m_head != NULL) {
1689 bus_dmamap_unload(sc->txtag, tx_buffer->map);
1690 m_freem(tx_buffer->m_head);
1691 tx_buffer->m_head = NULL;
1692 }
1693 }
1694
8434a83b 1695 for (i = 0; i < sc->rx_ring_inuse; ++i)
3f939c23 1696 emx_free_rx_ring(sc, &sc->rx_data[i]);
5330213c
SZ		 1697
1698 sc->csum_flags = 0;
1699 sc->csum_ehlen = 0;
1700 sc->csum_iphlen = 0;
1701
1702 sc->tx_dd_head = 0;
1703 sc->tx_dd_tail = 0;
1704 sc->tx_nsegs = 0;
1705}
1706
1707static int
2d0e5700 1708emx_reset(struct emx_softc *sc)
5330213c
SZ		 1709{
1710 device_t dev = sc->dev;
1711 uint16_t rx_buffer_size;
1712
5330213c
SZ		 1713 /* Set up smart power down as default off on newer adapters. */
1714 if (!emx_smart_pwr_down &&
1715 (sc->hw.mac.type == e1000_82571 ||
1716 sc->hw.mac.type == e1000_82572)) {
1717 uint16_t phy_tmp = 0;
1718
1719 /* Speed up time to link by disabling smart power down. */
1720 e1000_read_phy_reg(&sc->hw,
1721 IGP02E1000_PHY_POWER_MGMT, &phy_tmp);
1722 phy_tmp &= ~IGP02E1000_PM_SPD;
1723 e1000_write_phy_reg(&sc->hw,
1724 IGP02E1000_PHY_POWER_MGMT, phy_tmp);
1725 }
1726
1727 /*
1728 * These parameters control the automatic generation (Tx) and
1729 * response (Rx) to Ethernet PAUSE frames.
1730 * - High water mark should allow for at least two frames to be
1731 * received after sending an XOFF.
1732 * - Low water mark works best when it is very near the high water mark.
1733 * This allows the receiver to restart by sending XON when it has
1734 * drained a bit. Here we use an arbitary value of 1500 which will
1735 * restart after one full frame is pulled from the buffer. There
1736 * could be several smaller frames in the buffer and if so they will
1737 * not trigger the XON until their total number reduces the buffer
1738 * by 1500.
1739 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1740 */
1741 rx_buffer_size = (E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff) << 10;
1742
1743 sc->hw.fc.high_water = rx_buffer_size -
1744 roundup2(sc->max_frame_size, 1024);
1745 sc->hw.fc.low_water = sc->hw.fc.high_water - 1500;
1746
1747 if (sc->hw.mac.type == e1000_80003es2lan)
1748 sc->hw.fc.pause_time = 0xFFFF;
1749 else
1750 sc->hw.fc.pause_time = EMX_FC_PAUSE_TIME;
1751 sc->hw.fc.send_xon = TRUE;
1752 sc->hw.fc.requested_mode = e1000_fc_full;
1753
2d0e5700
SZ		 1754 /* Issue a global reset */
1755 e1000_reset_hw(&sc->hw);
1756 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1757
5330213c
SZ		 1758 if (e1000_init_hw(&sc->hw) < 0) {
1759 device_printf(dev, "Hardware Initialization Failed\n");
1760 return (EIO);
1761 }
1762
2d0e5700
SZ		 1763 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1764 e1000_get_phy_info(&sc->hw);
5330213c
SZ		 1765 e1000_check_for_link(&sc->hw);
1766
1767 return (0);
1768}
1769
1770static void
1771emx_setup_ifp(struct emx_softc *sc)
1772{
1773 struct ifnet *ifp = &sc->arpcom.ac_if;
1774
1775 if_initname(ifp, device_get_name(sc->dev),
1776 device_get_unit(sc->dev));
1777 ifp->if_softc = sc;
1778 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1779 ifp->if_init = emx_init;
1780 ifp->if_ioctl = emx_ioctl;
1781 ifp->if_start = emx_start;
b3a7093f
SZ		 1782#ifdef IFPOLL_ENABLE
1783 ifp->if_qpoll = emx_qpoll;
5330213c
SZ		 1784#endif
1785 ifp->if_watchdog = emx_watchdog;
6d435846
SZ		 1786 ifp->if_serialize = emx_serialize;
1787 ifp->if_deserialize = emx_deserialize;
1788 ifp->if_tryserialize = emx_tryserialize;
2c9effcf
SZ		 1789#ifdef INVARIANTS
1790 ifp->if_serialize_assert = emx_serialize_assert;
1791#endif
5330213c
SZ		 1792 ifq_set_maxlen(&ifp->if_snd, sc->num_tx_desc - 1);
1793 ifq_set_ready(&ifp->if_snd);
1794
ae474cfa 1795 ether_ifattach(ifp, sc->hw.mac.addr, NULL);
5330213c
SZ		 1796
1797 ifp->if_capabilities = IFCAP_HWCSUM |
1798 IFCAP_VLAN_HWTAGGING |
1799 IFCAP_VLAN_MTU;
8434a83b
SZ		 1800 if (sc->rx_ring_cnt > 1)
1801 ifp->if_capabilities |= IFCAP_RSS;
5330213c
SZ		 1802 ifp->if_capenable = ifp->if_capabilities;
1803 ifp->if_hwassist = EMX_CSUM_FEATURES;
1804
1805 /*
1806 * Tell the upper layer(s) we support long frames.
1807 */
1808 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1809
1810 /*
1811 * Specify the media types supported by this sc and register
1812 * callbacks to update media and link information
1813 */
1814 ifmedia_init(&sc->media, IFM_IMASK,
1815 emx_media_change, emx_media_status);
1816 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1817 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1818 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1819 0, NULL);
1820 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1821 } else {
1822 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1823 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1824 0, NULL);
1825 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1826 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1827 0, NULL);
1828 if (sc->hw.phy.type != e1000_phy_ife) {
1829 ifmedia_add(&sc->media,
1830 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1831 ifmedia_add(&sc->media,
1832 IFM_ETHER | IFM_1000_T, 0, NULL);
1833 }
1834 }
1835 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1836 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1837}
1838
1839/*
1840 * Workaround for SmartSpeed on 82541 and 82547 controllers
1841 */
1842static void
1843emx_smartspeed(struct emx_softc *sc)
1844{
1845 uint16_t phy_tmp;
1846
1847 if (sc->link_active || sc->hw.phy.type != e1000_phy_igp ||
1848 sc->hw.mac.autoneg == 0 ||
1849 (sc->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0)
1850 return;
1851
1852 if (sc->smartspeed == 0) {
1853 /*
1854 * If Master/Slave config fault is asserted twice,
1855 * we assume back-to-back
1856 */
1857 e1000_read_phy_reg(&sc->hw, PHY_1000T_STATUS, &phy_tmp);
1858 if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
1859 return;
1860 e1000_read_phy_reg(&sc->hw, PHY_1000T_STATUS, &phy_tmp);
1861 if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
1862 e1000_read_phy_reg(&sc->hw,
1863 PHY_1000T_CTRL, &phy_tmp);
1864 if (phy_tmp & CR_1000T_MS_ENABLE) {
1865 phy_tmp &= ~CR_1000T_MS_ENABLE;
1866 e1000_write_phy_reg(&sc->hw,
1867 PHY_1000T_CTRL, phy_tmp);
1868 sc->smartspeed++;
1869 if (sc->hw.mac.autoneg &&
1870 !e1000_phy_setup_autoneg(&sc->hw) &&
1871 !e1000_read_phy_reg(&sc->hw,
1872 PHY_CONTROL, &phy_tmp)) {
1873 phy_tmp |= MII_CR_AUTO_NEG_EN |
1874 MII_CR_RESTART_AUTO_NEG;
1875 e1000_write_phy_reg(&sc->hw,
1876 PHY_CONTROL, phy_tmp);
1877 }
1878 }
1879 }
1880 return;
1881 } else if (sc->smartspeed == EMX_SMARTSPEED_DOWNSHIFT) {
1882 /* If still no link, perhaps using 2/3 pair cable */
1883 e1000_read_phy_reg(&sc->hw, PHY_1000T_CTRL, &phy_tmp);
1884 phy_tmp |= CR_1000T_MS_ENABLE;
1885 e1000_write_phy_reg(&sc->hw, PHY_1000T_CTRL, phy_tmp);
1886 if (sc->hw.mac.autoneg &&
1887 !e1000_phy_setup_autoneg(&sc->hw) &&
1888 !e1000_read_phy_reg(&sc->hw, PHY_CONTROL, &phy_tmp)) {
1889 phy_tmp |= MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG;
1890 e1000_write_phy_reg(&sc->hw, PHY_CONTROL, phy_tmp);
1891 }
1892 }
1893
1894 /* Restart process after EMX_SMARTSPEED_MAX iterations */
1895 if (sc->smartspeed++ == EMX_SMARTSPEED_MAX)
1896 sc->smartspeed = 0;
1897}
1898
1899static int
5330213c
SZ		 1900emx_create_tx_ring(struct emx_softc *sc)
1901{
1902 device_t dev = sc->dev;
323e5ecd 1903 struct emx_txbuf *tx_buffer;
bdca134f
SZ		 1904 int error, i, tsize;
1905
1906 /*
1907 * Validate number of transmit descriptors. It must not exceed
1908 * hardware maximum, and must be multiple of E1000_DBA_ALIGN.
1909 */
1910 if ((emx_txd * sizeof(struct e1000_tx_desc)) % EMX_DBA_ALIGN != 0 ||
1911 emx_txd > EMX_MAX_TXD || emx_txd < EMX_MIN_TXD) {
1912 device_printf(dev, "Using %d TX descriptors instead of %d!\n",
1913 EMX_DEFAULT_TXD, emx_txd);
1914 sc->num_tx_desc = EMX_DEFAULT_TXD;
1915 } else {
1916 sc->num_tx_desc = emx_txd;
1917 }
1918
1919 /*
1920 * Allocate Transmit Descriptor ring
1921 */
1922 tsize = roundup2(sc->num_tx_desc * sizeof(struct e1000_tx_desc),
1923 EMX_DBA_ALIGN);
a596084c
SZ		 1924 sc->tx_desc_base = bus_dmamem_coherent_any(sc->parent_dtag,
1925 EMX_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1926 &sc->tx_desc_dtag, &sc->tx_desc_dmap,
1927 &sc->tx_desc_paddr);
1928 if (sc->tx_desc_base == NULL) {
bdca134f 1929 device_printf(dev, "Unable to allocate tx_desc memory\n");
a596084c 1930 return ENOMEM;
bdca134f 1931 }
5330213c 1932
323e5ecd
SZ		 1933 sc->tx_buf = kmalloc(sizeof(struct emx_txbuf) * sc->num_tx_desc,
1934 M_DEVBUF, M_WAITOK | M_ZERO);
5330213c
SZ		 1935
1936 /*
1937 * Create DMA tags for tx buffers
1938 */
1939 error = bus_dma_tag_create(sc->parent_dtag, /* parent */
1940 1, 0, /* alignment, bounds */
1941 BUS_SPACE_MAXADDR, /* lowaddr */
1942 BUS_SPACE_MAXADDR, /* highaddr */
1943 NULL, NULL, /* filter, filterarg */
1944 EMX_TSO_SIZE, /* maxsize */
1945 EMX_MAX_SCATTER, /* nsegments */
1946 EMX_MAX_SEGSIZE, /* maxsegsize */
1947 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1948 BUS_DMA_ONEBPAGE, /* flags */
1949 &sc->txtag);
1950 if (error) {
1951 device_printf(dev, "Unable to allocate TX DMA tag\n");
323e5ecd
SZ		 1952 kfree(sc->tx_buf, M_DEVBUF);
1953 sc->tx_buf = NULL;
5330213c
SZ		 1954 return error;
1955 }
1956
1957 /*
1958 * Create DMA maps for tx buffers
1959 */
1960 for (i = 0; i < sc->num_tx_desc; i++) {
323e5ecd 1961 tx_buffer = &sc->tx_buf[i];
5330213c
SZ		 1962
1963 error = bus_dmamap_create(sc->txtag,
1964 BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1965 &tx_buffer->map);
1966 if (error) {
1967 device_printf(dev, "Unable to create TX DMA map\n");
1968 emx_destroy_tx_ring(sc, i);
1969 return error;
1970 }
1971 }
1972 return (0);
1973}
1974
1975static void
1976emx_init_tx_ring(struct emx_softc *sc)
1977{
1978 /* Clear the old ring contents */
1979 bzero(sc->tx_desc_base,
1980 sizeof(struct e1000_tx_desc) * sc->num_tx_desc);
1981
1982 /* Reset state */
1983 sc->next_avail_tx_desc = 0;
1984 sc->next_tx_to_clean = 0;
1985 sc->num_tx_desc_avail = sc->num_tx_desc;
1986}
1987
1988static void
1989emx_init_tx_unit(struct emx_softc *sc)
1990{
1991 uint32_t tctl, tarc, tipg = 0;
1992 uint64_t bus_addr;
1993
1994 /* Setup the Base and Length of the Tx Descriptor Ring */
a596084c 1995 bus_addr = sc->tx_desc_paddr;
5330213c
SZ		 1996 E1000_WRITE_REG(&sc->hw, E1000_TDLEN(0),
1997 sc->num_tx_desc * sizeof(struct e1000_tx_desc));
1998 E1000_WRITE_REG(&sc->hw, E1000_TDBAH(0),
1999 (uint32_t)(bus_addr >> 32));
2000 E1000_WRITE_REG(&sc->hw, E1000_TDBAL(0),
2001 (uint32_t)bus_addr);
2002 /* Setup the HW Tx Head and Tail descriptor pointers */
2003 E1000_WRITE_REG(&sc->hw, E1000_TDT(0), 0);
2004 E1000_WRITE_REG(&sc->hw, E1000_TDH(0), 0);
2005
2006 /* Set the default values for the Tx Inter Packet Gap timer */
2007 switch (sc->hw.mac.type) {
2008 case e1000_80003es2lan:
2009 tipg = DEFAULT_82543_TIPG_IPGR1;
2010 tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 <<
2011 E1000_TIPG_IPGR2_SHIFT;
2012 break;
2013
2014 default:
2015 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
2016 sc->hw.phy.media_type == e1000_media_type_internal_serdes)
2017 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
2018 else
2019 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
2020 tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2021 tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2022 break;
2023 }
2024
2025 E1000_WRITE_REG(&sc->hw, E1000_TIPG, tipg);
2026
2027 /* NOTE: 0 is not allowed for TIDV */
2028 E1000_WRITE_REG(&sc->hw, E1000_TIDV, 1);
2029 E1000_WRITE_REG(&sc->hw, E1000_TADV, 0);
2030
2031 if (sc->hw.mac.type == e1000_82571 ||
2032 sc->hw.mac.type == e1000_82572) {
2033 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(0));
2034 tarc |= EMX_TARC_SPEED_MODE;
2035 E1000_WRITE_REG(&sc->hw, E1000_TARC(0), tarc);
2036 } else if (sc->hw.mac.type == e1000_80003es2lan) {
2037 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(0));
2038 tarc |= 1;
2039 E1000_WRITE_REG(&sc->hw, E1000_TARC(0), tarc);
2040 tarc = E1000_READ_REG(&sc->hw, E1000_TARC(1));
2041 tarc |= 1;
2042 E1000_WRITE_REG(&sc->hw, E1000_TARC(1), tarc);
2043 }
2044
2045 /* Program the Transmit Control Register */
2046 tctl = E1000_READ_REG(&sc->hw, E1000_TCTL);
2047 tctl &= ~E1000_TCTL_CT;
2048 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
2049 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2050 tctl |= E1000_TCTL_MULR;
2051
2052 /* This write will effectively turn on the transmit unit. */
2053 E1000_WRITE_REG(&sc->hw, E1000_TCTL, tctl);
2054}
2055
2056static void
2057emx_destroy_tx_ring(struct emx_softc *sc, int ndesc)
2058{
323e5ecd 2059 struct emx_txbuf *tx_buffer;
5330213c
SZ		 2060 int i;
2061
bdca134f 2062 /* Free Transmit Descriptor ring */
a596084c
SZ		 2063 if (sc->tx_desc_base) {
2064 bus_dmamap_unload(sc->tx_desc_dtag, sc->tx_desc_dmap);
2065 bus_dmamem_free(sc->tx_desc_dtag, sc->tx_desc_base,
2066 sc->tx_desc_dmap);
2067 bus_dma_tag_destroy(sc->tx_desc_dtag);
2068
2069 sc->tx_desc_base = NULL;
2070 }
bdca134f 2071
323e5ecd 2072 if (sc->tx_buf == NULL)
5330213c
SZ		 2073 return;
2074
2075 for (i = 0; i < ndesc; i++) {
323e5ecd 2076 tx_buffer = &sc->tx_buf[i];
5330213c
SZ		 2077
2078 KKASSERT(tx_buffer->m_head == NULL);
2079 bus_dmamap_destroy(sc->txtag, tx_buffer->map);
2080 }
2081 bus_dma_tag_destroy(sc->txtag);
2082
323e5ecd
SZ		 2083 kfree(sc->tx_buf, M_DEVBUF);
2084 sc->tx_buf = NULL;
5330213c
SZ		 2085}
2086
2087/*
2088 * The offload context needs to be set when we transfer the first
2089 * packet of a particular protocol (TCP/UDP). This routine has been
2090 * enhanced to deal with inserted VLAN headers.
2091 *
2092 * If the new packet's ether header length, ip header length and
2093 * csum offloading type are same as the previous packet, we should
2094 * avoid allocating a new csum context descriptor; mainly to take
2095 * advantage of the pipeline effect of the TX data read request.
2096 *
2097 * This function returns number of TX descrptors allocated for
2098 * csum context.
2099 */
2100static int
2101emx_txcsum(struct emx_softc *sc, struct mbuf *mp,
2102 uint32_t *txd_upper, uint32_t *txd_lower)
2103{
2104 struct e1000_context_desc *TXD;
323e5ecd 2105 struct emx_txbuf *tx_buffer;
5330213c
SZ		 2106 struct ether_vlan_header *eh;
2107 struct ip *ip;
2108 int curr_txd, ehdrlen, csum_flags;
2109 uint32_t cmd, hdr_len, ip_hlen;
2110 uint16_t etype;
2111
2112 /*
2113 * Determine where frame payload starts.
2114 * Jump over vlan headers if already present,
2115 * helpful for QinQ too.
2116 */
2117 KASSERT(mp->m_len >= ETHER_HDR_LEN,
2118 ("emx_txcsum_pullup is not called (eh)?\n"));
2119 eh = mtod(mp, struct ether_vlan_header *);
2120 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2121 KASSERT(mp->m_len >= ETHER_HDR_LEN + EVL_ENCAPLEN,
2122 ("emx_txcsum_pullup is not called (evh)?\n"));
2123 etype = ntohs(eh->evl_proto);
2124 ehdrlen = ETHER_HDR_LEN + EVL_ENCAPLEN;
2125 } else {
2126 etype = ntohs(eh->evl_encap_proto);
2127 ehdrlen = ETHER_HDR_LEN;
2128 }
2129
2130 /*
2131 * We only support TCP/UDP for IPv4 for the moment.
2132 * TODO: Support SCTP too when it hits the tree.
2133 */
2134 if (etype != ETHERTYPE_IP)
2135 return 0;
2136
2137 KASSERT(mp->m_len >= ehdrlen + EMX_IPVHL_SIZE,
2138 ("emx_txcsum_pullup is not called (eh+ip_vhl)?\n"));
2139
2140 /* NOTE: We could only safely access ip.ip_vhl part */
2141 ip = (struct ip *)(mp->m_data + ehdrlen);
2142 ip_hlen = ip->ip_hl << 2;
2143
2144 csum_flags = mp->m_pkthdr.csum_flags & EMX_CSUM_FEATURES;
2145
2146 if (sc->csum_ehlen == ehdrlen && sc->csum_iphlen == ip_hlen &&
2147 sc->csum_flags == csum_flags) {
2148 /*
2149 * Same csum offload context as the previous packets;
2150 * just return.
2151 */
2152 *txd_upper = sc->csum_txd_upper;
2153 *txd_lower = sc->csum_txd_lower;
2154 return 0;
2155 }
2156
2157 /*
2158 * Setup a new csum offload context.
2159 */
2160
2161 curr_txd = sc->next_avail_tx_desc;
323e5ecd 2162 tx_buffer = &sc->tx_buf[curr_txd];
5330213c
SZ		 2163 TXD = (struct e1000_context_desc *)&sc->tx_desc_base[curr_txd];
2164
2165 cmd = 0;
2166
2167 /* Setup of IP header checksum. */
2168 if (csum_flags & CSUM_IP) {
2169 /*
2170 * Start offset for header checksum calculation.
2171 * End offset for header checksum calculation.
2172 * Offset of place to put the checksum.
2173 */
2174 TXD->lower_setup.ip_fields.ipcss = ehdrlen;
2175 TXD->lower_setup.ip_fields.ipcse =
2176 htole16(ehdrlen + ip_hlen - 1);
2177 TXD->lower_setup.ip_fields.ipcso =
2178 ehdrlen + offsetof(struct ip, ip_sum);
2179 cmd |= E1000_TXD_CMD_IP;
2180 *txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2181 }
2182 hdr_len = ehdrlen + ip_hlen;
2183
2184 if (csum_flags & CSUM_TCP) {
2185 /*
2186 * Start offset for payload checksum calculation.
2187 * End offset for payload checksum calculation.
2188 * Offset of place to put the checksum.
2189 */
2190 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2191 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2192 TXD->upper_setup.tcp_fields.tucso =
2193 hdr_len + offsetof(struct tcphdr, th_sum);
2194 cmd |= E1000_TXD_CMD_TCP;
2195 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2196 } else if (csum_flags & CSUM_UDP) {
2197 /*
2198 * Start offset for header checksum calculation.
2199 * End offset for header checksum calculation.
2200 * Offset of place to put the checksum.
2201 */
2202 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2203 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2204 TXD->upper_setup.tcp_fields.tucso =
2205 hdr_len + offsetof(struct udphdr, uh_sum);
2206 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2207 }
2208
2209 *txd_lower = E1000_TXD_CMD_DEXT | /* Extended descr type */
2210 E1000_TXD_DTYP_D; /* Data descr */
2211
2212 /* Save the information for this csum offloading context */
2213 sc->csum_ehlen = ehdrlen;
2214 sc->csum_iphlen = ip_hlen;
2215 sc->csum_flags = csum_flags;
2216 sc->csum_txd_upper = *txd_upper;
2217 sc->csum_txd_lower = *txd_lower;
2218
2219 TXD->tcp_seg_setup.data = htole32(0);
2220 TXD->cmd_and_length =
2221 htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
5330213c
SZ		 2222
2223 if (++curr_txd == sc->num_tx_desc)
2224 curr_txd = 0;
2225
2226 KKASSERT(sc->num_tx_desc_avail > 0);
2227 sc->num_tx_desc_avail--;
2228
2229 sc->next_avail_tx_desc = curr_txd;
2230 return 1;
2231}
2232
2233static int
2234emx_txcsum_pullup(struct emx_softc *sc, struct mbuf **m0)
2235{
2236 struct mbuf *m = *m0;
2237 struct ether_header *eh;
2238 int len;
2239
2240 sc->tx_csum_try_pullup++;
2241
2242 len = ETHER_HDR_LEN + EMX_IPVHL_SIZE;
2243
2244 if (__predict_false(!M_WRITABLE(m))) {
2245 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2246 sc->tx_csum_drop1++;
2247 m_freem(m);
2248 *m0 = NULL;
2249 return ENOBUFS;
2250 }
2251 eh = mtod(m, struct ether_header *);
2252
2253 if (eh->ether_type == htons(ETHERTYPE_VLAN))
2254 len += EVL_ENCAPLEN;
2255
3752657e 2256 if (m->m_len < len) {
5330213c
SZ		 2257 sc->tx_csum_drop2++;
2258 m_freem(m);
2259 *m0 = NULL;
2260 return ENOBUFS;
2261 }
2262 return 0;
2263 }
2264
2265 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2266 sc->tx_csum_pullup1++;
2267 m = m_pullup(m, ETHER_HDR_LEN);
2268 if (m == NULL) {
2269 sc->tx_csum_pullup1_failed++;
2270 *m0 = NULL;
2271 return ENOBUFS;
2272 }
2273 *m0 = m;
2274 }
2275 eh = mtod(m, struct ether_header *);
2276
2277 if (eh->ether_type == htons(ETHERTYPE_VLAN))
2278 len += EVL_ENCAPLEN;
2279
3752657e 2280 if (m->m_len < len) {
5330213c
SZ		 2281 sc->tx_csum_pullup2++;
2282 m = m_pullup(m, len);
2283 if (m == NULL) {
2284 sc->tx_csum_pullup2_failed++;
2285 *m0 = NULL;
2286 return ENOBUFS;
2287 }
2288 *m0 = m;
2289 }
2290 return 0;
2291}
2292
2293static void
2294emx_txeof(struct emx_softc *sc)
2295{
2296 struct ifnet *ifp = &sc->arpcom.ac_if;
323e5ecd 2297 struct emx_txbuf *tx_buffer;
5330213c
SZ		 2298 int first, num_avail;
2299
2300 if (sc->tx_dd_head == sc->tx_dd_tail)
2301 return;
2302
2303 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2304 return;
2305
2306 num_avail = sc->num_tx_desc_avail;
2307 first = sc->next_tx_to_clean;
2308
2309 while (sc->tx_dd_head != sc->tx_dd_tail) {
2310 int dd_idx = sc->tx_dd[sc->tx_dd_head];
70172a73 2311 struct e1000_tx_desc *tx_desc;
5330213c
SZ		 2312
2313 tx_desc = &sc->tx_desc_base[dd_idx];
5330213c
SZ		 2314 if (tx_desc->upper.fields.status & E1000_TXD_STAT_DD) {
2315 EMX_INC_TXDD_IDX(sc->tx_dd_head);
2316
2317 if (++dd_idx == sc->num_tx_desc)
2318 dd_idx = 0;
2319
2320 while (first != dd_idx) {
2321 logif(pkt_txclean);
2322
5330213c
SZ		 2323 num_avail++;
2324
323e5ecd 2325 tx_buffer = &sc->tx_buf[first];
5330213c
SZ		 2326 if (tx_buffer->m_head) {
2327 ifp->if_opackets++;
2328 bus_dmamap_unload(sc->txtag,
2329 tx_buffer->map);
2330 m_freem(tx_buffer->m_head);
2331 tx_buffer->m_head = NULL;
2332 }
2333
2334 if (++first == sc->num_tx_desc)
2335 first = 0;
2336 }
2337 } else {
2338 break;
2339 }
2340 }
2341 sc->next_tx_to_clean = first;
2342 sc->num_tx_desc_avail = num_avail;
2343
2344 if (sc->tx_dd_head == sc->tx_dd_tail) {
2345 sc->tx_dd_head = 0;
2346 sc->tx_dd_tail = 0;
2347 }
2348
2349 if (!EMX_IS_OACTIVE(sc)) {
2350 ifp->if_flags &= ~IFF_OACTIVE;
2351
2352 /* All clean, turn off the timer */
2353 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2354 ifp->if_timer = 0;
2355 }
2356}
2357
2358static void
2359emx_tx_collect(struct emx_softc *sc)
2360{
2361 struct ifnet *ifp = &sc->arpcom.ac_if;
323e5ecd 2362 struct emx_txbuf *tx_buffer;
5330213c
SZ		 2363 int tdh, first, num_avail, dd_idx = -1;
2364
2365 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2366 return;
2367
2368 tdh = E1000_READ_REG(&sc->hw, E1000_TDH(0));
2369 if (tdh == sc->next_tx_to_clean)
2370 return;
2371
2372 if (sc->tx_dd_head != sc->tx_dd_tail)
2373 dd_idx = sc->tx_dd[sc->tx_dd_head];
2374
2375 num_avail = sc->num_tx_desc_avail;
2376 first = sc->next_tx_to_clean;
2377
2378 while (first != tdh) {
2379 logif(pkt_txclean);
2380
5330213c
SZ		 2381 num_avail++;
2382
323e5ecd 2383 tx_buffer = &sc->tx_buf[first];
5330213c
SZ		 2384 if (tx_buffer->m_head) {
2385 ifp->if_opackets++;
2386 bus_dmamap_unload(sc->txtag,
2387 tx_buffer->map);
2388 m_freem(tx_buffer->m_head);
2389 tx_buffer->m_head = NULL;
2390 }
2391
2392 if (first == dd_idx) {
2393 EMX_INC_TXDD_IDX(sc->tx_dd_head);
2394 if (sc->tx_dd_head == sc->tx_dd_tail) {
2395 sc->tx_dd_head = 0;
2396 sc->tx_dd_tail = 0;
2397 dd_idx = -1;
2398 } else {
2399 dd_idx = sc->tx_dd[sc->tx_dd_head];
2400 }
2401 }
2402
2403 if (++first == sc->num_tx_desc)
2404 first = 0;
2405 }
2406 sc->next_tx_to_clean = first;
2407 sc->num_tx_desc_avail = num_avail;
2408
2409 if (!EMX_IS_OACTIVE(sc)) {
2410 ifp->if_flags &= ~IFF_OACTIVE;
2411
2412 /* All clean, turn off the timer */
2413 if (sc->num_tx_desc_avail == sc->num_tx_desc)
2414 ifp->if_timer = 0;
2415 }
2416}
2417
2418/*
2419 * When Link is lost sometimes there is work still in the TX ring
2420 * which will result in a watchdog, rather than allow that do an
2421 * attempted cleanup and then reinit here. Note that this has been
2422 * seens mostly with fiber adapters.
2423 */
2424static void
2425emx_tx_purge(struct emx_softc *sc)
2426{
2427 struct ifnet *ifp = &sc->arpcom.ac_if;
2428
2429 if (!sc->link_active && ifp->if_timer) {
2430 emx_tx_collect(sc);
2431 if (ifp->if_timer) {
2432 if_printf(ifp, "Link lost, TX pending, reinit\n");
2433 ifp->if_timer = 0;
2434 emx_init(sc);
2435 }
2436 }
2437}
2438
2439static int
c39e3a1f 2440emx_newbuf(struct emx_softc *sc, struct emx_rxdata *rdata, int i, int init)
5330213c
SZ		 2441{
2442 struct mbuf *m;
2443 bus_dma_segment_t seg;
2444 bus_dmamap_t map;
323e5ecd 2445 struct emx_rxbuf *rx_buffer;
5330213c
SZ		 2446 int error, nseg;
2447
2448 m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2449 if (m == NULL) {
c39e3a1f 2450 rdata->mbuf_cluster_failed++;
5330213c
SZ		 2451 if (init) {
2452 if_printf(&sc->arpcom.ac_if,
2453 "Unable to allocate RX mbuf\n");
2454 }
2455 return (ENOBUFS);
2456 }
2457 m->m_len = m->m_pkthdr.len = MCLBYTES;
2458
2459 if (sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2460 m_adj(m, ETHER_ALIGN);
2461
c39e3a1f
SZ		 2462 error = bus_dmamap_load_mbuf_segment(rdata->rxtag,
2463 rdata->rx_sparemap, m,
5330213c
SZ		 2464 &seg, 1, &nseg, BUS_DMA_NOWAIT);
2465 if (error) {
2466 m_freem(m);
2467 if (init) {
2468 if_printf(&sc->arpcom.ac_if,
2469 "Unable to load RX mbuf\n");
2470 }
2471 return (error);
2472 }
2473
323e5ecd 2474 rx_buffer = &rdata->rx_buf[i];
5330213c 2475 if (rx_buffer->m_head != NULL)
c39e3a1f 2476 bus_dmamap_unload(rdata->rxtag, rx_buffer->map);
5330213c
SZ		 2477
2478 map = rx_buffer->map;
c39e3a1f
SZ		 2479 rx_buffer->map = rdata->rx_sparemap;
2480 rdata->rx_sparemap = map;
5330213c
SZ		 2481
2482 rx_buffer->m_head = m;
235b9d30 2483 rx_buffer->paddr = seg.ds_addr;
5330213c 2484
235b9d30 2485 emx_setup_rxdesc(&rdata->rx_desc[i], rx_buffer);
5330213c
SZ		 2486 return (0);
2487}
2488
2489static int
c39e3a1f 2490emx_create_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
5330213c
SZ		 2491{
2492 device_t dev = sc->dev;
323e5ecd 2493 struct emx_rxbuf *rx_buffer;
bdca134f
SZ		 2494 int i, error, rsize;
2495
2496 /*
2497 * Validate number of receive descriptors. It must not exceed
2498 * hardware maximum, and must be multiple of E1000_DBA_ALIGN.
2499 */
3f939c23 2500 if ((emx_rxd * sizeof(emx_rxdesc_t)) % EMX_DBA_ALIGN != 0 ||
bdca134f
SZ		 2501 emx_rxd > EMX_MAX_RXD || emx_rxd < EMX_MIN_RXD) {
2502 device_printf(dev, "Using %d RX descriptors instead of %d!\n",
2503 EMX_DEFAULT_RXD, emx_rxd);
c39e3a1f 2504 rdata->num_rx_desc = EMX_DEFAULT_RXD;
bdca134f 2505 } else {
c39e3a1f 2506 rdata->num_rx_desc = emx_rxd;
bdca134f
SZ		 2507 }
2508
2509 /*
2510 * Allocate Receive Descriptor ring
2511 */
235b9d30 2512 rsize = roundup2(rdata->num_rx_desc * sizeof(emx_rxdesc_t),
bdca134f 2513 EMX_DBA_ALIGN);
235b9d30 2514 rdata->rx_desc = bus_dmamem_coherent_any(sc->parent_dtag,
a596084c 2515 EMX_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
c39e3a1f
SZ		 2516 &rdata->rx_desc_dtag, &rdata->rx_desc_dmap,
2517 &rdata->rx_desc_paddr);
235b9d30 2518 if (rdata->rx_desc == NULL) {
bdca134f 2519 device_printf(dev, "Unable to allocate rx_desc memory\n");
a596084c 2520 return ENOMEM;
bdca134f 2521 }
5330213c 2522
323e5ecd
SZ		 2523 rdata->rx_buf = kmalloc(sizeof(struct emx_rxbuf) * rdata->num_rx_desc,
2524 M_DEVBUF, M_WAITOK | M_ZERO);
5330213c
SZ		 2525
2526 /*
2527 * Create DMA tag for rx buffers
2528 */
2529 error = bus_dma_tag_create(sc->parent_dtag, /* parent */
2530 1, 0, /* alignment, bounds */
2531 BUS_SPACE_MAXADDR, /* lowaddr */
2532 BUS_SPACE_MAXADDR, /* highaddr */
2533 NULL, NULL, /* filter, filterarg */
2534 MCLBYTES, /* maxsize */
2535 1, /* nsegments */
2536 MCLBYTES, /* maxsegsize */
2537 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
c39e3a1f 2538 &rdata->rxtag);
5330213c
SZ		 2539 if (error) {
2540 device_printf(dev, "Unable to allocate RX DMA tag\n");
323e5ecd
SZ		 2541 kfree(rdata->rx_buf, M_DEVBUF);
2542 rdata->rx_buf = NULL;
5330213c
SZ		 2543 return error;
2544 }
2545
2546 /*
2547 * Create spare DMA map for rx buffers
2548 */
c39e3a1f
SZ		 2549 error = bus_dmamap_create(rdata->rxtag, BUS_DMA_WAITOK,
2550 &rdata->rx_sparemap);
5330213c
SZ		 2551 if (error) {
2552 device_printf(dev, "Unable to create spare RX DMA map\n");
c39e3a1f 2553 bus_dma_tag_destroy(rdata->rxtag);
323e5ecd
SZ		 2554 kfree(rdata->rx_buf, M_DEVBUF);
2555 rdata->rx_buf = NULL;
5330213c
SZ		 2556 return error;
2557 }
2558
2559 /*
2560 * Create DMA maps for rx buffers
2561 */
c39e3a1f 2562 for (i = 0; i < rdata->num_rx_desc; i++) {
323e5ecd 2563 rx_buffer = &rdata->rx_buf[i];
5330213c 2564
c39e3a1f 2565 error = bus_dmamap_create(rdata->rxtag, BUS_DMA_WAITOK,
5330213c
SZ		 2566 &rx_buffer->map);
2567 if (error) {
2568 device_printf(dev, "Unable to create RX DMA map\n");
c39e3a1f 2569 emx_destroy_rx_ring(sc, rdata, i);
5330213c
SZ		 2570 return error;
2571 }
2572 }
2573 return (0);
2574}
2575
c39e3a1f
SZ		 2576static void
2577emx_free_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
2578{
2579 int i;
2580
2581 for (i = 0; i < rdata->num_rx_desc; i++) {
323e5ecd 2582 struct emx_rxbuf *rx_buffer = &rdata->rx_buf[i];
c39e3a1f
SZ		 2583
2584 if (rx_buffer->m_head != NULL) {
2585 bus_dmamap_unload(rdata->rxtag, rx_buffer->map);
2586 m_freem(rx_buffer->m_head);
2587 rx_buffer->m_head = NULL;
2588 }
2589 }
2590
2591 if (rdata->fmp != NULL)
2592 m_freem(rdata->fmp);
2593 rdata->fmp = NULL;
2594 rdata->lmp = NULL;
2595}
2596
5330213c 2597static int
c39e3a1f 2598emx_init_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata)
5330213c
SZ		 2599{
2600 int i, error;
2601
2602 /* Reset descriptor ring */
235b9d30 2603 bzero(rdata->rx_desc, sizeof(emx_rxdesc_t) * rdata->num_rx_desc);
5330213c
SZ		 2604
2605 /* Allocate new ones. */
c39e3a1f
SZ		 2606 for (i = 0; i < rdata->num_rx_desc; i++) {
2607 error = emx_newbuf(sc, rdata, i, 1);
5330213c
SZ		 2608 if (error)
2609 return (error);
2610 }
2611
2612 /* Setup our descriptor pointers */
c39e3a1f 2613 rdata->next_rx_desc_to_check = 0;
5330213c
SZ		 2614
2615 return (0);
2616}
2617
2618static void
2619emx_init_rx_unit(struct emx_softc *sc)
2620{
2621 struct ifnet *ifp = &sc->arpcom.ac_if;
2622 uint64_t bus_addr;
2d0e5700 2623 uint32_t rctl, itr, rfctl;
3f939c23 2624 int i;
5330213c
SZ		 2625
2626 /*
2627 * Make sure receives are disabled while setting
2628 * up the descriptor ring
2629 */
2630 rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
2631 E1000_WRITE_REG(&sc->hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2632
2633 /*
2634 * Set the interrupt throttling rate. Value is calculated
2635 * as ITR = 1 / (INT_THROTTLE_CEIL * 256ns)
2636 */
2d0e5700
SZ		 2637 if (sc->int_throttle_ceil)
2638 itr = 1000000000 / 256 / sc->int_throttle_ceil;
2639 else
2640 itr = 0;
2641 emx_set_itr(sc, itr);
5330213c 2642
235b9d30
SZ		 2643 /* Use extended RX descriptor */
2644 rfctl = E1000_RFCTL_EXTEN;
2645
5330213c 2646 /* Disable accelerated ackknowledge */
235b9d30
SZ		 2647 if (sc->hw.mac.type == e1000_82574)
2648 rfctl |= E1000_RFCTL_ACK_DIS;
2649
2650 E1000_WRITE_REG(&sc->hw, E1000_RFCTL, rfctl);
5330213c 2651
65c7a6af
SZ		 2652 /*
2653 * Receive Checksum Offload for TCP and UDP
2654 *
2655 * Checksum offloading is also enabled if multiple receive
2656 * queue is to be supported, since we need it to figure out
2657 * packet type.
2658 */
8434a83b 2659 if (ifp->if_capenable & (IFCAP_RSS | IFCAP_RXCSUM)) {
2d0e5700
SZ		 2660 uint32_t rxcsum;
2661
5330213c 2662 rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
3f939c23
SZ		 2663
2664 /*
2665 * NOTE:
2666 * PCSD must be enabled to enable multiple
2667 * receive queues.
2668 */
2669 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2670 E1000_RXCSUM_PCSD;
5330213c
SZ		 2671 E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2672 }
2673
2674 /*
65c7a6af 2675 * Configure multiple receive queue (RSS)
3f939c23 2676 */
8434a83b 2677 if (ifp->if_capenable & IFCAP_RSS) {
89d8e73d
SZ		 2678 uint8_t key[EMX_NRSSRK * EMX_RSSRK_SIZE];
2679 uint32_t reta;
2680
2681 KASSERT(sc->rx_ring_inuse == EMX_NRX_RING,
2682 ("invalid number of RX ring (%d)",
2683 sc->rx_ring_inuse));
2684
65c7a6af
SZ		 2685 /*
2686 * NOTE:
2687 * When we reach here, RSS has already been disabled
2688 * in emx_stop(), so we could safely configure RSS key
2689 * and redirect table.
2690 */
3f939c23 2691
65c7a6af
SZ		 2692 /*
2693 * Configure RSS key
2694 */
89d8e73d
SZ		 2695 toeplitz_get_key(key, sizeof(key));
2696 for (i = 0; i < EMX_NRSSRK; ++i) {
2697 uint32_t rssrk;
2698
2699 rssrk = EMX_RSSRK_VAL(key, i);
2700 EMX_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2701
2702 E1000_WRITE_REG(&sc->hw, E1000_RSSRK(i), rssrk);
2703 }
3f939c23 2704
65c7a6af 2705 /*
89d8e73d
SZ		 2706 * Configure RSS redirect table in following fashion:
2707 * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
65c7a6af 2708 */
89d8e73d
SZ		 2709 reta = 0;
2710 for (i = 0; i < EMX_RETA_SIZE; ++i) {
2711 uint32_t q;
2712
2713 q = (i % sc->rx_ring_inuse) << EMX_RETA_RINGIDX_SHIFT;
2714 reta |= q << (8 * i);
2715 }
2716 EMX_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2717
65c7a6af
SZ		 2718 for (i = 0; i < EMX_NRETA; ++i)
2719 E1000_WRITE_REG(&sc->hw, E1000_RETA(i), reta);
3f939c23 2720
65c7a6af
SZ		 2721 /*
2722 * Enable multiple receive queues.
2723 * Enable IPv4 RSS standard hash functions.
2724 * Disable RSS interrupt.
2725 */
2726 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2727 E1000_MRQC_ENABLE_RSS_2Q |
2728 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2729 E1000_MRQC_RSS_FIELD_IPV4);
2730 }
3f939c23
SZ		 2731
2732 /*
5330213c
SZ		 2733 * XXX TEMPORARY WORKAROUND: on some systems with 82573
2734 * long latencies are observed, like Lenovo X60. This
2735 * change eliminates the problem, but since having positive
2736 * values in RDTR is a known source of problems on other
2737 * platforms another solution is being sought.
2738 */
2739 if (emx_82573_workaround && sc->hw.mac.type == e1000_82573) {
2740 E1000_WRITE_REG(&sc->hw, E1000_RADV, EMX_RADV_82573);
2741 E1000_WRITE_REG(&sc->hw, E1000_RDTR, EMX_RDTR_82573);
2742 }
2743
8434a83b 2744 for (i = 0; i < sc->rx_ring_inuse; ++i) {
2d0e5700
SZ		 2745 struct emx_rxdata *rdata = &sc->rx_data[i];
2746
2747 /*
2748 * Setup the Base and Length of the Rx Descriptor Ring
2749 */
2750 bus_addr = rdata->rx_desc_paddr;
2751 E1000_WRITE_REG(&sc->hw, E1000_RDLEN(i),
2752 rdata->num_rx_desc * sizeof(emx_rxdesc_t));
2753 E1000_WRITE_REG(&sc->hw, E1000_RDBAH(i),
2754 (uint32_t)(bus_addr >> 32));
2755 E1000_WRITE_REG(&sc->hw, E1000_RDBAL(i),
2756 (uint32_t)bus_addr);
2757
2758 /*
2759 * Setup the HW Rx Head and Tail Descriptor Pointers
2760 */
3f939c23
SZ		 2761 E1000_WRITE_REG(&sc->hw, E1000_RDH(i), 0);
2762 E1000_WRITE_REG(&sc->hw, E1000_RDT(i),
2763 sc->rx_data[i].num_rx_desc - 1);
2764 }
2765
2d0e5700
SZ		 2766 /* Setup the Receive Control Register */
2767 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2768 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2769 E1000_RCTL_RDMTS_HALF | E1000_RCTL_SECRC |
2770 (sc->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2771
2772 /* Make sure VLAN Filters are off */
2773 rctl &= ~E1000_RCTL_VFE;
2774
2775 /* Don't store bad paket */
2776 rctl &= ~E1000_RCTL_SBP;
2777
2778 /* MCLBYTES */
2779 rctl |= E1000_RCTL_SZ_2048;
2780
2781 if (ifp->if_mtu > ETHERMTU)
2782 rctl |= E1000_RCTL_LPE;
2783 else
2784 rctl &= ~E1000_RCTL_LPE;
2785
3f939c23
SZ		 2786 /* Enable Receives */
2787 E1000_WRITE_REG(&sc->hw, E1000_RCTL, rctl);
5330213c
SZ		 2788}
2789
2790static void
c39e3a1f 2791emx_destroy_rx_ring(struct emx_softc *sc, struct emx_rxdata *rdata, int ndesc)
5330213c 2792{
323e5ecd 2793 struct emx_rxbuf *rx_buffer;
5330213c
SZ		 2794 int i;
2795
bdca134f 2796 /* Free Receive Descriptor ring */
235b9d30 2797 if (rdata->rx_desc) {
c39e3a1f 2798 bus_dmamap_unload(rdata->rx_desc_dtag, rdata->rx_desc_dmap);
235b9d30 2799 bus_dmamem_free(rdata->rx_desc_dtag, rdata->rx_desc,
c39e3a1f
SZ		 2800 rdata->rx_desc_dmap);
2801 bus_dma_tag_destroy(rdata->rx_desc_dtag);
a596084c 2802
235b9d30 2803 rdata->rx_desc = NULL;
a596084c 2804 }
bdca134f 2805
323e5ecd 2806 if (rdata->rx_buf == NULL)
5330213c
SZ		 2807 return;
2808
2809 for (i = 0; i < ndesc; i++) {
323e5ecd 2810 rx_buffer = &rdata->rx_buf[i];
5330213c
SZ		 2811
2812 KKASSERT(rx_buffer->m_head == NULL);
c39e3a1f 2813 bus_dmamap_destroy(rdata->rxtag, rx_buffer->map);
5330213c 2814 }
c39e3a1f
SZ		 2815 bus_dmamap_destroy(rdata->rxtag, rdata->rx_sparemap);
2816 bus_dma_tag_destroy(rdata->rxtag);
5330213c 2817
323e5ecd
SZ		 2818 kfree(rdata->rx_buf, M_DEVBUF);
2819 rdata->rx_buf = NULL;
5330213c
SZ		 2820}
2821
2822static void
c39e3a1f 2823emx_rxeof(struct emx_softc *sc, int ring_idx, int count)
5330213c 2824{
c39e3a1f 2825 struct emx_rxdata *rdata = &sc->rx_data[ring_idx];
5330213c 2826 struct ifnet *ifp = &sc->arpcom.ac_if;
235b9d30 2827 uint32_t staterr;
235b9d30 2828 emx_rxdesc_t *current_desc;
5330213c
SZ		 2829 struct mbuf *mp;
2830 int i;
2831 struct mbuf_chain chain[MAXCPU];
2832
c39e3a1f 2833 i = rdata->next_rx_desc_to_check;
235b9d30
SZ		 2834 current_desc = &rdata->rx_desc[i];
2835 staterr = le32toh(current_desc->rxd_staterr);
5330213c 2836
235b9d30 2837 if (!(staterr & E1000_RXD_STAT_DD))
5330213c
SZ		 2838 return;
2839
2840 ether_input_chain_init(chain);
2841
235b9d30 2842 while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
9cc86e17 2843 struct pktinfo *pi = NULL, pi0;
235b9d30 2844 struct emx_rxbuf *rx_buf = &rdata->rx_buf[i];
5330213c 2845 struct mbuf *m = NULL;
0acc29d6 2846 int eop, len;
5330213c
SZ		 2847
2848 logif(pkt_receive);
2849
235b9d30 2850 mp = rx_buf->m_head;
5330213c
SZ		 2851
2852 /*
2853 * Can't defer bus_dmamap_sync(9) because TBI_ACCEPT
2854 * needs to access the last received byte in the mbuf.
2855 */
235b9d30 2856 bus_dmamap_sync(rdata->rxtag, rx_buf->map,
5330213c
SZ		 2857 BUS_DMASYNC_POSTREAD);
2858
0acc29d6 2859 len = le16toh(current_desc->rxd_length);
235b9d30 2860 if (staterr & E1000_RXD_STAT_EOP) {
5330213c
SZ		 2861 count--;
2862 eop = 1;
5330213c
SZ		 2863 } else {
2864 eop = 0;
5330213c
SZ		 2865 }
2866
235b9d30
SZ		 2867 if (!(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
2868 uint16_t vlan = 0;
3f939c23 2869 uint32_t mrq, rss_hash;
235b9d30
SZ		 2870
2871 /*
2872 * Save several necessary information,
2873 * before emx_newbuf() destroy it.
2874 */
2875 if ((staterr & E1000_RXD_STAT_VP) && eop)
2876 vlan = le16toh(current_desc->rxd_vlan);
5330213c 2877
3f939c23
SZ		 2878 mrq = le32toh(current_desc->rxd_mrq);
2879 rss_hash = le32toh(current_desc->rxd_rss);
2880
2881 EMX_RSS_DPRINTF(sc, 10,
2882 "ring%d, mrq 0x%08x, rss_hash 0x%08x\n",
2883 ring_idx, mrq, rss_hash);
2884
c39e3a1f 2885 if (emx_newbuf(sc, rdata, i, 0) != 0) {
5330213c
SZ		 2886 ifp->if_iqdrops++;
2887 goto discard;
2888 }
2889
2890 /* Assign correct length to the current fragment */
2891 mp->m_len = len;
2892
c39e3a1f 2893 if (rdata->fmp == NULL) {
5330213c 2894 mp->m_pkthdr.len = len;
c39e3a1f
SZ		 2895 rdata->fmp = mp; /* Store the first mbuf */
2896 rdata->lmp = mp;
5330213c
SZ		 2897 } else {
2898 /*
2899 * Chain mbuf's together
2900 */
c39e3a1f
SZ		 2901 rdata->lmp->m_next = mp;
2902 rdata->lmp = rdata->lmp->m_next;
2903 rdata->fmp->m_pkthdr.len += len;
5330213c
SZ		 2904 }
2905
2906 if (eop) {
c39e3a1f 2907 rdata->fmp->m_pkthdr.rcvif = ifp;
5330213c
SZ		 2908 ifp->if_ipackets++;
2909
235b9d30
SZ		 2910 if (ifp->if_capenable & IFCAP_RXCSUM)
2911 emx_rxcsum(staterr, rdata->fmp);
5330213c 2912
235b9d30 2913 if (staterr & E1000_RXD_STAT_VP) {
c39e3a1f 2914 rdata->fmp->m_pkthdr.ether_vlantag =
235b9d30 2915 vlan;
c39e3a1f 2916 rdata->fmp->m_flags |= M_VLANTAG;
5330213c 2917 }
c39e3a1f
SZ		 2918 m = rdata->fmp;
2919 rdata->fmp = NULL;
2920 rdata->lmp = NULL;
3f939c23 2921
9cc86e17
SZ		 2922 if (ifp->if_capenable & IFCAP_RSS) {
2923 pi = emx_rssinfo(m, &pi0, mrq,
2924 rss_hash, staterr);
2925 }
3f939c23
SZ		 2926#ifdef EMX_RSS_DEBUG
2927 rdata->rx_pkts++;
2928#endif
5330213c
SZ		 2929 }
2930 } else {
2931 ifp->if_ierrors++;
2932discard:
235b9d30 2933 emx_setup_rxdesc(current_desc, rx_buf);
c39e3a1f
SZ		 2934 if (rdata->fmp != NULL) {
2935 m_freem(rdata->fmp);
2936 rdata->fmp = NULL;
2937 rdata->lmp = NULL;
5330213c
SZ		 2938 }
2939 m = NULL;
2940 }
2941
5330213c 2942 if (m != NULL)
9cc86e17 2943 ether_input_chain(ifp, m, pi, chain);
5330213c
SZ		 2944
2945 /* Advance our pointers to the next descriptor. */
c39e3a1f 2946 if (++i == rdata->num_rx_desc)
5330213c 2947 i = 0;
235b9d30
SZ		 2948
2949 current_desc = &rdata->rx_desc[i];
2950 staterr = le32toh(current_desc->rxd_staterr);
5330213c 2951 }
c39e3a1f 2952 rdata->next_rx_desc_to_check = i;
5330213c
SZ		 2953
2954 ether_input_dispatch(chain);
2955
3f939c23 2956 /* Advance the E1000's Receive Queue "Tail Pointer". */
5330213c 2957 if (--i < 0)
c39e3a1f 2958 i = rdata->num_rx_desc - 1;
3f939c23 2959 E1000_WRITE_REG(&sc->hw, E1000_RDT(ring_idx), i);
5330213c
SZ		 2960}
2961
2962static void
5330213c
SZ		 2963emx_enable_intr(struct emx_softc *sc)
2964{
2d0e5700
SZ		 2965 uint32_t ims_mask = IMS_ENABLE_MASK;
2966
6d435846 2967 lwkt_serialize_handler_enable(&sc->main_serialize);
2d0e5700
SZ		 2968
2969#if 0
2970 if (sc->hw.mac.type == e1000_82574) {
2971 E1000_WRITE_REG(hw, EMX_EIAC, EM_MSIX_MASK);
2972 ims_mask |= EM_MSIX_MASK;
2973 }
2974#endif
2975 E1000_WRITE_REG(&sc->hw, E1000_IMS, ims_mask);
5330213c
SZ		 2976}
2977
2978static void
2979emx_disable_intr(struct emx_softc *sc)
2980{
2d0e5700
SZ		 2981 if (sc->hw.mac.type == e1000_82574)
2982 E1000_WRITE_REG(&sc->hw, EMX_EIAC, 0);
5330213c 2983 E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2d0e5700 2984
6d435846 2985 lwkt_serialize_handler_disable(&sc->main_serialize);
5330213c
SZ		 2986}
2987
2988/*
2989 * Bit of a misnomer, what this really means is
2990 * to enable OS management of the system... aka
2991 * to disable special hardware management features
2992 */
2993static void
2994emx_get_mgmt(struct emx_softc *sc)
2995{
2996 /* A shared code workaround */
2997 if (sc->has_manage) {
2998 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2999 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
3000
3001 /* disable hardware interception of ARP */
3002 manc &= ~(E1000_MANC_ARP_EN);
3003
3004 /* enable receiving management packets to the host */
3005 manc |= E1000_MANC_EN_MNG2HOST;
3006#define E1000_MNG2HOST_PORT_623 (1 << 5)
3007#define E1000_MNG2HOST_PORT_664 (1 << 6)
3008 manc2h |= E1000_MNG2HOST_PORT_623;
3009 manc2h |= E1000_MNG2HOST_PORT_664;
3010 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
3011
3012 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
3013 }
3014}
3015
3016/*
3017 * Give control back to hardware management
3018 * controller if there is one.
3019 */
3020static void
3021emx_rel_mgmt(struct emx_softc *sc)
3022{
3023 if (sc->has_manage) {
3024 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
3025
3026 /* re-enable hardware interception of ARP */
3027 manc |= E1000_MANC_ARP_EN;
3028 manc &= ~E1000_MANC_EN_MNG2HOST;
3029
3030 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
3031 }
3032}
3033
3034/*
3035 * emx_get_hw_control() sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3036 * For ASF and Pass Through versions of f/w this means that
3037 * the driver is loaded. For AMT version (only with 82573)
3038 * of the f/w this means that the network i/f is open.
3039 */
3040static void
3041emx_get_hw_control(struct emx_softc *sc)
3042{
5330213c 3043 /* Let firmware know the driver has taken over */
2d0e5700
SZ		 3044 if (sc->hw.mac.type == e1000_82573) {
3045 uint32_t swsm;
3046
5330213c
SZ		 3047 swsm = E1000_READ_REG(&sc->hw, E1000_SWSM);
3048 E1000_WRITE_REG(&sc->hw, E1000_SWSM,
3049 swsm | E1000_SWSM_DRV_LOAD);
2d0e5700
SZ		 3050 } else {
3051 uint32_t ctrl_ext;
5330213c 3052
5330213c
SZ		 3053 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3054 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
3055 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
5330213c 3056 }
2d0e5700 3057 sc->control_hw = 1;
5330213c
SZ		 3058}
3059
3060/*
3061 * emx_rel_hw_control() resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3062 * For ASF and Pass Through versions of f/w this means that the
3063 * driver is no longer loaded. For AMT version (only with 82573)
3064 * of the f/w this means that the network i/f is closed.
3065 */
3066static void
3067emx_rel_hw_control(struct emx_softc *sc)
3068{
2d0e5700
SZ		 3069 if (!sc->control_hw)
3070 return;
3071 sc->control_hw = 0;
5330213c
SZ		 3072
3073 /* Let firmware taken over control of h/w */
2d0e5700
SZ		 3074 if (sc->hw.mac.type == e1000_82573) {
3075 uint32_t swsm;
3076
5330213c
SZ		 3077 swsm = E1000_READ_REG(&sc->hw, E1000_SWSM);
3078 E1000_WRITE_REG(&sc->hw, E1000_SWSM,
3079 swsm & ~E1000_SWSM_DRV_LOAD);
2d0e5700
SZ		 3080 } else {
3081 uint32_t ctrl_ext;
5330213c 3082
5330213c
SZ		 3083 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3084 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
3085 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
5330213c
SZ		 3086 }
3087}
3088
3089static int
3090emx_is_valid_eaddr(const uint8_t *addr)
3091{
3092 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
3093
3094 if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3095 return (FALSE);
3096
3097 return (TRUE);
3098}
3099
3100/*
3101 * Enable PCI Wake On Lan capability
3102 */
3103void
3104emx_enable_wol(device_t dev)
3105{
3106 uint16_t cap, status;
3107 uint8_t id;
3108
3109 /* First find the capabilities pointer*/
3110 cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
3111
3112 /* Read the PM Capabilities */
3113 id = pci_read_config(dev, cap, 1);
3114 if (id != PCIY_PMG) /* Something wrong */
3115 return;
3116
3117 /*
3118 * OK, we have the power capabilities,
3119 * so now get the status register
3120 */
3121 cap += PCIR_POWER_STATUS;
3122 status = pci_read_config(dev, cap, 2);
3123 status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3124 pci_write_config(dev, cap, status, 2);
3125}
3126
3127static void
3128emx_update_stats(struct emx_softc *sc)
3129{
3130 struct ifnet *ifp = &sc->arpcom.ac_if;
3131
3132 if (sc->hw.phy.media_type == e1000_media_type_copper ||
3133 (E1000_READ_REG(&sc->hw, E1000_STATUS) & E1000_STATUS_LU)) {
3134 sc->stats.symerrs += E1000_READ_REG(&sc->hw, E1000_SYMERRS);
3135 sc->stats.sec += E1000_READ_REG(&sc->hw, E1000_SEC);
3136 }
3137 sc->stats.crcerrs += E1000_READ_REG(&sc->hw, E1000_CRCERRS);
3138 sc->stats.mpc += E1000_READ_REG(&sc->hw, E1000_MPC);
3139 sc->stats.scc += E1000_READ_REG(&sc->hw, E1000_SCC);
3140 sc->stats.ecol += E1000_READ_REG(&sc->hw, E1000_ECOL);
3141
3142 sc->stats.mcc += E1000_READ_REG(&sc->hw, E1000_MCC);
3143 sc->stats.latecol += E1000_READ_REG(&sc->hw, E1000_LATECOL);
3144 sc->stats.colc += E1000_READ_REG(&sc->hw, E1000_COLC);
3145 sc->stats.dc += E1000_READ_REG(&sc->hw, E1000_DC);
3146 sc->stats.rlec += E1000_READ_REG(&sc->hw, E1000_RLEC);
3147 sc->stats.xonrxc += E1000_READ_REG(&sc->hw, E1000_XONRXC);
3148 sc->stats.xontxc += E1000_READ_REG(&sc->hw, E1000_XONTXC);
3149 sc->stats.xoffrxc += E1000_READ_REG(&sc->hw, E1000_XOFFRXC);
3150 sc->stats.xofftxc += E1000_READ_REG(&sc->hw, E1000_XOFFTXC);
3151 sc->stats.fcruc += E1000_READ_REG(&sc->hw, E1000_FCRUC);
3152 sc->stats.prc64 += E1000_READ_REG(&sc->hw, E1000_PRC64);
3153 sc->stats.prc127 += E1000_READ_REG(&sc->hw, E1000_PRC127);
3154 sc->stats.prc255 += E1000_READ_REG(&sc->hw, E1000_PRC255);
3155 sc->stats.prc511 += E1000_READ_REG(&sc->hw, E1000_PRC511);
3156 sc->stats.prc1023 += E1000_READ_REG(&sc->hw, E1000_PRC1023);
3157 sc->stats.prc1522 += E1000_READ_REG(&sc->hw, E1000_PRC1522);
3158 sc->stats.gprc += E1000_READ_REG(&sc->hw, E1000_GPRC);
3159 sc->stats.bprc += E1000_READ_REG(&sc->hw, E1000_BPRC);
3160 sc->stats.mprc += E1000_READ_REG(&sc->hw, E1000_MPRC);
3161 sc->stats.gptc += E1000_READ_REG(&sc->hw, E1000_GPTC);
3162
3163 /* For the 64-bit byte counters the low dword must be read first. */
3164 /* Both registers clear on the read of the high dword */
3165
3166 sc->stats.gorc += E1000_READ_REG(&sc->hw, E1000_GORCH);
3167 sc->stats.gotc += E1000_READ_REG(&sc->hw, E1000_GOTCH);
3168
3169 sc->stats.rnbc += E1000_READ_REG(&sc->hw, E1000_RNBC);
3170 sc->stats.ruc += E1000_READ_REG(&sc->hw, E1000_RUC);
3171 sc->stats.rfc += E1000_READ_REG(&sc->hw, E1000_RFC);
3172 sc->stats.roc += E1000_READ_REG(&sc->hw, E1000_ROC);
3173 sc->stats.rjc += E1000_READ_REG(&sc->hw, E1000_RJC);
3174
3175 sc->stats.tor += E1000_READ_REG(&sc->hw, E1000_TORH);
3176 sc->stats.tot += E1000_READ_REG(&sc->hw, E1000_TOTH);
3177
3178 sc->stats.tpr += E1000_READ_REG(&sc->hw, E1000_TPR);
3179 sc->stats.tpt += E1000_READ_REG(&sc->hw, E1000_TPT);
3180 sc->stats.ptc64 += E1000_READ_REG(&sc->hw, E1000_PTC64);
3181 sc->stats.ptc127 += E1000_READ_REG(&sc->hw, E1000_PTC127);
3182 sc->stats.ptc255 += E1000_READ_REG(&sc->hw, E1000_PTC255);
3183 sc->stats.ptc511 += E1000_READ_REG(&sc->hw, E1000_PTC511);
3184 sc->stats.ptc1023 += E1000_READ_REG(&sc->hw, E1000_PTC1023);
3185 sc->stats.ptc1522 += E1000_READ_REG(&sc->hw, E1000_PTC1522);
3186 sc->stats.mptc += E1000_READ_REG(&sc->hw, E1000_MPTC);
3187 sc->stats.bptc += E1000_READ_REG(&sc->hw, E1000_BPTC);
3188
3189 sc->stats.algnerrc += E1000_READ_REG(&sc->hw, E1000_ALGNERRC);
3190 sc->stats.rxerrc += E1000_READ_REG(&sc->hw, E1000_RXERRC);
3191 sc->stats.tncrs += E1000_READ_REG(&sc->hw, E1000_TNCRS);
3192 sc->stats.cexterr += E1000_READ_REG(&sc->hw, E1000_CEXTERR);
3193 sc->stats.tsctc += E1000_READ_REG(&sc->hw, E1000_TSCTC);
3194 sc->stats.tsctfc += E1000_READ_REG(&sc->hw, E1000_TSCTFC);
3195
3196 ifp->if_collisions = sc->stats.colc;
3197
3198 /* Rx Errors */
3199 ifp->if_ierrors = sc->dropped_pkts + sc->stats.rxerrc +
3200 sc->stats.crcerrs + sc->stats.algnerrc +
3201 sc->stats.ruc + sc->stats.roc +
3202 sc->stats.mpc + sc->stats.cexterr;
3203
3204 /* Tx Errors */
3205 ifp->if_oerrors = sc->stats.ecol + sc->stats.latecol +
3206 sc->watchdog_events;
3207}
3208
3209static void
3210emx_print_debug_info(struct emx_softc *sc)
3211{
3212 device_t dev = sc->dev;
3213 uint8_t *hw_addr = sc->hw.hw_addr;
3214
3215 device_printf(dev, "Adapter hardware address = %p \n", hw_addr);
3216 device_printf(dev, "CTRL = 0x%x RCTL = 0x%x \n",
3217 E1000_READ_REG(&sc->hw, E1000_CTRL),
3218 E1000_READ_REG(&sc->hw, E1000_RCTL));
3219 device_printf(dev, "Packet buffer = Tx=%dk Rx=%dk \n",
3220 ((E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff0000) >> 16),\
3221 (E1000_READ_REG(&sc->hw, E1000_PBA) & 0xffff) );
3222 device_printf(dev, "Flow control watermarks high = %d low = %d\n",
3223 sc->hw.fc.high_water, sc->hw.fc.low_water);
3224 device_printf(dev, "tx_int_delay = %d, tx_abs_int_delay = %d\n",
3225 E1000_READ_REG(&sc->hw, E1000_TIDV),
3226 E1000_READ_REG(&sc->hw, E1000_TADV));
3227 device_printf(dev, "rx_int_delay = %d, rx_abs_int_delay = %d\n",
3228 E1000_READ_REG(&sc->hw, E1000_RDTR),
3229 E1000_READ_REG(&sc->hw, E1000_RADV));
3230 device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
3231 E1000_READ_REG(&sc->hw, E1000_TDH(0)),
3232 E1000_READ_REG(&sc->hw, E1000_TDT(0)));
3233 device_printf(dev, "hw rdh = %d, hw rdt = %d\n",
3234 E1000_READ_REG(&sc->hw, E1000_RDH(0)),
3235 E1000_READ_REG(&sc->hw, E1000_RDT(0)));
3236 device_printf(dev, "Num Tx descriptors avail = %d\n",
3237 sc->num_tx_desc_avail);
3238 device_printf(dev, "Tx Descriptors not avail1 = %ld\n",
3239 sc->no_tx_desc_avail1);
3240 device_printf(dev, "Tx Descriptors not avail2 = %ld\n",
3241 sc->no_tx_desc_avail2);
3242 device_printf(dev, "Std mbuf failed = %ld\n",
3243 sc->mbuf_alloc_failed);
3244 device_printf(dev, "Std mbuf cluster failed = %ld\n",
c39e3a1f 3245 sc->rx_data[0].mbuf_cluster_failed);
5330213c
SZ		 3246 device_printf(dev, "Driver dropped packets = %ld\n",
3247 sc->dropped_pkts);
3248 device_printf(dev, "Driver tx dma failure in encap = %ld\n",
3249 sc->no_tx_dma_setup);
3250
3251 device_printf(dev, "TXCSUM try pullup = %lu\n",
3252 sc->tx_csum_try_pullup);
3253 device_printf(dev, "TXCSUM m_pullup(eh) called = %lu\n",
3254 sc->tx_csum_pullup1);
3255 device_printf(dev, "TXCSUM m_pullup(eh) failed = %lu\n",
3256 sc->tx_csum_pullup1_failed);
3257 device_printf(dev, "TXCSUM m_pullup(eh+ip) called = %lu\n",
3258 sc->tx_csum_pullup2);
3259 device_printf(dev, "TXCSUM m_pullup(eh+ip) failed = %lu\n",
3260 sc->tx_csum_pullup2_failed);
3261 device_printf(dev, "TXCSUM non-writable(eh) droped = %lu\n",
3262 sc->tx_csum_drop1);
3263 device_printf(dev, "TXCSUM non-writable(eh+ip) droped = %lu\n",
3264 sc->tx_csum_drop2);
3265}
3266
3267static void
3268emx_print_hw_stats(struct emx_softc *sc)
3269{
3270 device_t dev = sc->dev;
3271
3272 device_printf(dev, "Excessive collisions = %lld\n",
3273 (long long)sc->stats.ecol);
3274#if (DEBUG_HW > 0) /* Dont output these errors normally */
3275 device_printf(dev, "Symbol errors = %lld\n",
3276 (long long)sc->stats.symerrs);
3277#endif
3278 device_printf(dev, "Sequence errors = %lld\n",
3279 (long long)sc->stats.sec);
3280 device_printf(dev, "Defer count = %lld\n",
3281 (long long)sc->stats.dc);
3282 device_printf(dev, "Missed Packets = %lld\n",
3283 (long long)sc->stats.mpc);
3284 device_printf(dev, "Receive No Buffers = %lld\n",
3285 (long long)sc->stats.rnbc);
3286 /* RLEC is inaccurate on some hardware, calculate our own. */
3287 device_printf(dev, "Receive Length Errors = %lld\n",
3288 ((long long)sc->stats.roc + (long long)sc->stats.ruc));
3289 device_printf(dev, "Receive errors = %lld\n",
3290 (long long)sc->stats.rxerrc);
3291 device_printf(dev, "Crc errors = %lld\n",
3292 (long long)sc->stats.crcerrs);
3293 device_printf(dev, "Alignment errors = %lld\n",
3294 (long long)sc->stats.algnerrc);
3295 device_printf(dev, "Collision/Carrier extension errors = %lld\n",
3296 (long long)sc->stats.cexterr);
3297 device_printf(dev, "RX overruns = %ld\n", sc->rx_overruns);
3298 device_printf(dev, "watchdog timeouts = %ld\n",
3299 sc->watchdog_events);
3300 device_printf(dev, "XON Rcvd = %lld\n",
3301 (long long)sc->stats.xonrxc);
3302 device_printf(dev, "XON Xmtd = %lld\n",
3303 (long long)sc->stats.xontxc);
3304 device_printf(dev, "XOFF Rcvd = %lld\n",
3305 (long long)sc->stats.xoffrxc);
3306 device_printf(dev, "XOFF Xmtd = %lld\n",
3307 (long long)sc->stats.xofftxc);
3308 device_printf(dev, "Good Packets Rcvd = %lld\n",
3309 (long long)sc->stats.gprc);
3310 device_printf(dev, "Good Packets Xmtd = %lld\n",
3311 (long long)sc->stats.gptc);
3312}
3313
3314static void
3315emx_print_nvm_info(struct emx_softc *sc)
3316{
3317 uint16_t eeprom_data;
3318 int i, j, row = 0;
3319
3320 /* Its a bit crude, but it gets the job done */
3321 kprintf("\nInterface EEPROM Dump:\n");
3322 kprintf("Offset\n0x0000 ");
3323 for (i = 0, j = 0; i < 32; i++, j++) {
3324 if (j == 8) { /* Make the offset block */
3325 j = 0; ++row;
3326 kprintf("\n0x00%x0 ",row);
3327 }
3328 e1000_read_nvm(&sc->hw, i, 1, &eeprom_data);
3329 kprintf("%04x ", eeprom_data);
3330 }
3331 kprintf("\n");
3332}
3333
3334static int
3335emx_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
3336{
3337 struct emx_softc *sc;
3338 struct ifnet *ifp;
3339 int error, result;
3340
3341 result = -1;
3342 error = sysctl_handle_int(oidp, &result, 0, req);
3343 if (error || !req->newptr)
3344 return (error);
3345
3346 sc = (struct emx_softc *)arg1;
3347 ifp = &sc->arpcom.ac_if;
3348
6d435846 3349 ifnet_serialize_all(ifp);
5330213c
SZ		 3350
3351 if (result == 1)
3352 emx_print_debug_info(sc);
3353
3354 /*
3355 * This value will cause a hex dump of the
3356 * first 32 16-bit words of the EEPROM to
3357 * the screen.
3358 */
3359 if (result == 2)
3360 emx_print_nvm_info(sc);
3361
6d435846 3362 ifnet_deserialize_all(ifp);
5330213c
SZ		 3363
3364 return (error);
3365}
3366
3367static int
3368emx_sysctl_stats(SYSCTL_HANDLER_ARGS)
3369{
3370 int error, result;
3371
3372 result = -1;
3373 error = sysctl_handle_int(oidp, &result, 0, req);
3374 if (error || !req->newptr)
3375 return (error);
3376
3377 if (result == 1) {
3378 struct emx_softc *sc = (struct emx_softc *)arg1;
3379 struct ifnet *ifp = &sc->arpcom.ac_if;
3380
6d435846 3381 ifnet_serialize_all(ifp);
5330213c 3382 emx_print_hw_stats(sc);
6d435846 3383 ifnet_deserialize_all(ifp);
5330213c
SZ		 3384 }
3385 return (error);
3386}
3387
3388static void
3389emx_add_sysctl(struct emx_softc *sc)
3390{
3f939c23
SZ		 3391#ifdef EMX_RSS_DEBUG
3392 char rx_pkt[32];
3393 int i;
3394#endif
5330213c
SZ		 3395
3396 sysctl_ctx_init(&sc->sysctl_ctx);
3397 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
3398 SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
3399 device_get_nameunit(sc->dev),
3400 CTLFLAG_RD, 0, "");
3401 if (sc->sysctl_tree == NULL) {
3402 device_printf(sc->dev, "can't add sysctl node\n");
3403 return;
3404 }
3405
3406 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3407 OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, sc, 0,
3408 emx_sysctl_debug_info, "I", "Debug Information");
3409
3410 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3411 OID_AUTO, "stats", CTLTYPE_INT|CTLFLAG_RW, sc, 0,
3412 emx_sysctl_stats, "I", "Statistics");
3413
3414 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
c39e3a1f
SZ		 3415 OID_AUTO, "rxd", CTLFLAG_RD,
3416 &sc->rx_data[0].num_rx_desc, 0, NULL);
5330213c 3417 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
c39e3a1f 3418 OID_AUTO, "txd", CTLFLAG_RD, &sc->num_tx_desc, 0, NULL);
5330213c 3419
5330213c
SZ		 3420 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3421 OID_AUTO, "int_throttle_ceil", CTLTYPE_INT|CTLFLAG_RW,
3422 sc, 0, emx_sysctl_int_throttle, "I",
3423 "interrupt throttling rate");
3424 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3425 OID_AUTO, "int_tx_nsegs", CTLTYPE_INT|CTLFLAG_RW,
3426 sc, 0, emx_sysctl_int_tx_nsegs, "I",
3427 "# segments per TX interrupt");
3f939c23 3428
8434a83b
SZ		 3429 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3430 OID_AUTO, "rx_ring_inuse", CTLFLAG_RD,
3431 &sc->rx_ring_inuse, 0, "RX ring in use");
3432
3f939c23
SZ		 3433#ifdef EMX_RSS_DEBUG
3434 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
3435 OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug,
3436 0, "RSS debug level");
65c7a6af 3437 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3f939c23
SZ		 3438 ksnprintf(rx_pkt, sizeof(rx_pkt), "rx%d_pkt", i);
3439 SYSCTL_ADD_UINT(&sc->sysctl_ctx,
3440 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO,
89d8e73d 3441 rx_pkt, CTLFLAG_RW,
3f939c23
SZ		 3442 &sc->rx_data[i].rx_pkts, 0, "RXed packets");
3443 }
3444#endif
5330213c
SZ		 3445}
3446
3447static int
3448emx_sysctl_int_throttle(SYSCTL_HANDLER_ARGS)
3449{
3450 struct emx_softc *sc = (void *)arg1;
3451 struct ifnet *ifp = &sc->arpcom.ac_if;
3452 int error, throttle;
3453
3454 throttle = sc->int_throttle_ceil;
3455 error = sysctl_handle_int(oidp, &throttle, 0, req);
3456 if (error || req->newptr == NULL)
3457 return error;
3458 if (throttle < 0 || throttle > 1000000000 / 256)
3459 return EINVAL;
3460
3461 if (throttle) {
3462 /*
3463 * Set the interrupt throttling rate in 256ns increments,
3464 * recalculate sysctl value assignment to get exact frequency.
3465 */
3466 throttle = 1000000000 / 256 / throttle;
3467
3468 /* Upper 16bits of ITR is reserved and should be zero */
3469 if (throttle & 0xffff0000)
3470 return EINVAL;
3471 }
3472
6d435846 3473 ifnet_serialize_all(ifp);
5330213c
SZ		 3474
3475 if (throttle)
3476 sc->int_throttle_ceil = 1000000000 / 256 / throttle;
3477 else
3478 sc->int_throttle_ceil = 0;
3479
3480 if (ifp->if_flags & IFF_RUNNING)
2d0e5700 3481 emx_set_itr(sc, throttle);
5330213c 3482
6d435846 3483 ifnet_deserialize_all(ifp);
5330213c
SZ		 3484
3485 if (bootverbose) {
3486 if_printf(ifp, "Interrupt moderation set to %d/sec\n",
3487 sc->int_throttle_ceil);
3488 }
3489 return 0;
3490}
3491
3492static int
3493emx_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS)
3494{
3495 struct emx_softc *sc = (void *)arg1;
3496 struct ifnet *ifp = &sc->arpcom.ac_if;
3497 int error, segs;
3498
3499 segs = sc->tx_int_nsegs;
3500 error = sysctl_handle_int(oidp, &segs, 0, req);
3501 if (error || req->newptr == NULL)
3502 return error;
3503 if (segs <= 0)
3504 return EINVAL;
3505
6d435846 3506 ifnet_serialize_all(ifp);
5330213c
SZ		 3507
3508 /*
3509 * Don't allow int_tx_nsegs to become:
3510 * o Less the oact_tx_desc
3511 * o Too large that no TX desc will cause TX interrupt to
3512 * be generated (OACTIVE will never recover)
3513 * o Too small that will cause tx_dd[] overflow
3514 */
3515 if (segs < sc->oact_tx_desc ||
3516 segs >= sc->num_tx_desc - sc->oact_tx_desc ||
3517 segs < sc->num_tx_desc / EMX_TXDD_SAFE) {
3518 error = EINVAL;
3519 } else {
3520 error = 0;
3521 sc->tx_int_nsegs = segs;
3522 }
3523
6d435846 3524 ifnet_deserialize_all(ifp);
5330213c
SZ		 3525
3526 return error;
3527}
071699f8
SZ		 3528
3529static int
3530emx_dma_alloc(struct emx_softc *sc)
3531{
3f939c23 3532 int error, i;
071699f8
SZ		 3533
3534 /*
3535 * Create top level busdma tag
3536 */
3537 error = bus_dma_tag_create(NULL, 1, 0,
3538 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3539 NULL, NULL,
3540 BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
3541 0, &sc->parent_dtag);
3542 if (error) {
3543 device_printf(sc->dev, "could not create top level DMA tag\n");
3544 return error;
3545 }
3546
3547 /*
3548 * Allocate transmit descriptors ring and buffers
3549 */
3550 error = emx_create_tx_ring(sc);
3551 if (error) {
3552 device_printf(sc->dev, "Could not setup transmit structures\n");
3553 return error;
3554 }
3555
3556 /*
3557 * Allocate receive descriptors ring and buffers
3558 */
65c7a6af 3559 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3f939c23
SZ		 3560 error = emx_create_rx_ring(sc, &sc->rx_data[i]);
3561 if (error) {
3562 device_printf(sc->dev,
3563 "Could not setup receive structures\n");
3564 return error;
3565 }
071699f8
SZ		 3566 }
3567 return 0;
3568}
3569
3570static void
3571emx_dma_free(struct emx_softc *sc)
3572{
3f939c23
SZ		 3573 int i;
3574
071699f8 3575 emx_destroy_tx_ring(sc, sc->num_tx_desc);
3f939c23 3576
65c7a6af 3577 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3f939c23
SZ		 3578 emx_destroy_rx_ring(sc, &sc->rx_data[i],
3579 sc->rx_data[i].num_rx_desc);
3580 }
071699f8
SZ		 3581
3582 /* Free top level busdma tag */
3583 if (sc->parent_dtag != NULL)
3584 bus_dma_tag_destroy(sc->parent_dtag);
3585}
6d435846
SZ		 3586
3587static void
3588emx_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3589{
3590 struct emx_softc *sc = ifp->if_softc;
6d435846
SZ		 3591
3592 switch (slz) {
3593 case IFNET_SERIALIZE_ALL:
f61533ad 3594 lwkt_serialize_array_enter(sc->serializes, EMX_NSERIALIZE, 0);
6d435846
SZ		 3595 break;
3596
aabfe6fb
SZ		 3597 case IFNET_SERIALIZE_MAIN:
3598 lwkt_serialize_enter(&sc->main_serialize);
3599 break;
3600
6d435846
SZ		 3601 case IFNET_SERIALIZE_TX:
3602 lwkt_serialize_enter(&sc->tx_serialize);
3603 break;
3604
067b3d6b 3605 case IFNET_SERIALIZE_RX(0):
6d435846
SZ		 3606 lwkt_serialize_enter(&sc->rx_data[0].rx_serialize);
3607 break;
3608
067b3d6b 3609 case IFNET_SERIALIZE_RX(1):
6d435846
SZ		 3610 lwkt_serialize_enter(&sc->rx_data[1].rx_serialize);
3611 break;
3612
3613 default:
3614 panic("%s unsupported serialize type\n", ifp->if_xname);
3615 }
3616}
3617
3618static void
3619emx_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3620{
3621 struct emx_softc *sc = ifp->if_softc;
6d435846
SZ		 3622
3623 switch (slz) {
3624 case IFNET_SERIALIZE_ALL:
f61533ad 3625 lwkt_serialize_array_exit(sc->serializes, EMX_NSERIALIZE, 0);
6d435846
SZ		 3626 break;
3627
aabfe6fb
SZ		 3628 case IFNET_SERIALIZE_MAIN:
3629 lwkt_serialize_exit(&sc->main_serialize);
3630 break;
3631
6d435846
SZ		 3632 case IFNET_SERIALIZE_TX:
3633 lwkt_serialize_exit(&sc->tx_serialize);
3634 break;
3635
067b3d6b 3636 case IFNET_SERIALIZE_RX(0):
6d435846
SZ		 3637 lwkt_serialize_exit(&sc->rx_data[0].rx_serialize);
3638 break;
3639
067b3d6b 3640 case IFNET_SERIALIZE_RX(1):
6d435846
SZ		 3641 lwkt_serialize_exit(&sc->rx_data[1].rx_serialize);
3642 break;
3643
3644 default:
3645 panic("%s unsupported serialize type\n", ifp->if_xname);
3646 }
3647}
3648
3649static int
3650emx_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3651{
3652 struct emx_softc *sc = ifp->if_softc;
6d435846
SZ		 3653
3654 switch (slz) {
3655 case IFNET_SERIALIZE_ALL:
f61533ad
SZ		 3656 return lwkt_serialize_array_try(sc->serializes,
3657 EMX_NSERIALIZE, 0);
6d435846 3658
aabfe6fb
SZ		 3659 case IFNET_SERIALIZE_MAIN:
3660 return lwkt_serialize_try(&sc->main_serialize);
3661
6d435846
SZ		 3662 case IFNET_SERIALIZE_TX:
3663 return lwkt_serialize_try(&sc->tx_serialize);
3664
067b3d6b 3665 case IFNET_SERIALIZE_RX(0):
6d435846
SZ		 3666 return lwkt_serialize_try(&sc->rx_data[0].rx_serialize);
3667
067b3d6b 3668 case IFNET_SERIALIZE_RX(1):
6d435846
SZ		 3669 return lwkt_serialize_try(&sc->rx_data[1].rx_serialize);
3670
3671 default:
3672 panic("%s unsupported serialize type\n", ifp->if_xname);
3673 }
3674}
bca7c435
SZ		 3675
3676static void
3677emx_serialize_skipmain(struct emx_softc *sc)
3678{
3679 lwkt_serialize_array_enter(sc->serializes, EMX_NSERIALIZE, 1);
3680}
3681
3682static void
3683emx_deserialize_skipmain(struct emx_softc *sc)
3684{
3685 lwkt_serialize_array_exit(sc->serializes, EMX_NSERIALIZE, 1);
3686}
2c9effcf
SZ		 3687
3688#ifdef INVARIANTS
3689
3690static void
3691emx_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3692 boolean_t serialized)
3693{
3694 struct emx_softc *sc = ifp->if_softc;
3695 int i;
3696
3697 switch (slz) {
3698 case IFNET_SERIALIZE_ALL:
3699 if (serialized) {
3700 for (i = 0; i < EMX_NSERIALIZE; ++i)
3701 ASSERT_SERIALIZED(sc->serializes[i]);
3702 } else {
3703 for (i = 0; i < EMX_NSERIALIZE; ++i)
3704 ASSERT_NOT_SERIALIZED(sc->serializes[i]);
3705 }
3706 break;
3707
aabfe6fb
SZ		 3708 case IFNET_SERIALIZE_MAIN:
3709 if (serialized)
3710 ASSERT_SERIALIZED(&sc->main_serialize);
3711 else
3712 ASSERT_NOT_SERIALIZED(&sc->main_serialize);
3713 break;
3714
2c9effcf
SZ		 3715 case IFNET_SERIALIZE_TX:
3716 if (serialized)
3717 ASSERT_SERIALIZED(&sc->tx_serialize);
3718 else
3719 ASSERT_NOT_SERIALIZED(&sc->tx_serialize);
3720 break;
3721
067b3d6b 3722 case IFNET_SERIALIZE_RX(0):
2c9effcf
SZ		 3723 if (serialized)
3724 ASSERT_SERIALIZED(&sc->rx_data[0].rx_serialize);
3725 else
3726 ASSERT_NOT_SERIALIZED(&sc->rx_data[0].rx_serialize);
3727 break;
3728
067b3d6b 3729 case IFNET_SERIALIZE_RX(1):
2c9effcf
SZ		 3730 if (serialized)
3731 ASSERT_SERIALIZED(&sc->rx_data[1].rx_serialize);
3732 else
3733 ASSERT_NOT_SERIALIZED(&sc->rx_data[1].rx_serialize);
3734 break;
3735
3736 default:
3737 panic("%s unsupported serialize type\n", ifp->if_xname);
3738 }
3739}
3740
3741#endif /* INVARIANTS */