igb: Change TXCSUM does not require reinitialize hardware
[dragonfly.git] / sys / dev / netif / igb / if_igb.c
1 /*
2  * Copyright (c) 2001-2011, Intel Corporation 
3  * All rights reserved.
4  * 
5  * Redistribution and use in source and binary forms, with or without 
6  * modification, are permitted provided that the following conditions are met:
7  * 
8  *  1. Redistributions of source code must retain the above copyright notice, 
9  *     this list of conditions and the following disclaimer.
10  * 
11  *  2. Redistributions in binary form must reproduce the above copyright 
12  *     notice, this list of conditions and the following disclaimer in the 
13  *     documentation and/or other materials provided with the distribution.
14  * 
15  *  3. Neither the name of the Intel Corporation nor the names of its 
16  *     contributors may be used to endorse or promote products derived from 
17  *     this software without specific prior written permission.
18  * 
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 
23  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31
32 #include "opt_polling.h"
33 #include "opt_igb.h"
34
35 #include <sys/param.h>
36 #include <sys/bus.h>
37 #include <sys/endian.h>
38 #include <sys/interrupt.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/proc.h>
43 #include <sys/rman.h>
44 #include <sys/serialize.h>
45 #include <sys/serialize2.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50
51 #include <net/bpf.h>
52 #include <net/ethernet.h>
53 #include <net/if.h>
54 #include <net/if_arp.h>
55 #include <net/if_dl.h>
56 #include <net/if_media.h>
57 #include <net/ifq_var.h>
58 #include <net/toeplitz.h>
59 #include <net/toeplitz2.h>
60 #include <net/vlan/if_vlan_var.h>
61 #include <net/vlan/if_vlan_ether.h>
62 #include <net/if_poll.h>
63
64 #include <netinet/in_systm.h>
65 #include <netinet/in.h>
66 #include <netinet/ip.h>
67 #include <netinet/tcp.h>
68 #include <netinet/udp.h>
69
70 #include <bus/pci/pcivar.h>
71 #include <bus/pci/pcireg.h>
72
73 #include <dev/netif/ig_hal/e1000_api.h>
74 #include <dev/netif/ig_hal/e1000_82575.h>
75 #include <dev/netif/igb/if_igb.h>
76
77 #ifdef IGB_RSS_DEBUG
78 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) \
79 do { \
80         if (sc->rss_debug >= lvl) \
81                 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
82 } while (0)
83 #else   /* !IGB_RSS_DEBUG */
84 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...)      ((void)0)
85 #endif  /* IGB_RSS_DEBUG */
86
87 #define IGB_NAME        "Intel(R) PRO/1000 "
88 #define IGB_DEVICE(id)  \
89         { IGB_VENDOR_ID, E1000_DEV_ID_##id, IGB_NAME #id }
90 #define IGB_DEVICE_NULL { 0, 0, NULL }
91
92 static struct igb_device {
93         uint16_t        vid;
94         uint16_t        did;
95         const char      *desc;
96 } igb_devices[] = {
97         IGB_DEVICE(82575EB_COPPER),
98         IGB_DEVICE(82575EB_FIBER_SERDES),
99         IGB_DEVICE(82575GB_QUAD_COPPER),
100         IGB_DEVICE(82576),
101         IGB_DEVICE(82576_NS),
102         IGB_DEVICE(82576_NS_SERDES),
103         IGB_DEVICE(82576_FIBER),
104         IGB_DEVICE(82576_SERDES),
105         IGB_DEVICE(82576_SERDES_QUAD),
106         IGB_DEVICE(82576_QUAD_COPPER),
107         IGB_DEVICE(82576_QUAD_COPPER_ET2),
108         IGB_DEVICE(82576_VF),
109         IGB_DEVICE(82580_COPPER),
110         IGB_DEVICE(82580_FIBER),
111         IGB_DEVICE(82580_SERDES),
112         IGB_DEVICE(82580_SGMII),
113         IGB_DEVICE(82580_COPPER_DUAL),
114         IGB_DEVICE(82580_QUAD_FIBER),
115         IGB_DEVICE(DH89XXCC_SERDES),
116         IGB_DEVICE(DH89XXCC_SGMII),
117         IGB_DEVICE(DH89XXCC_SFP),
118         IGB_DEVICE(DH89XXCC_BACKPLANE),
119         IGB_DEVICE(I350_COPPER),
120         IGB_DEVICE(I350_FIBER),
121         IGB_DEVICE(I350_SERDES),
122         IGB_DEVICE(I350_SGMII),
123         IGB_DEVICE(I350_VF),
124
125         /* required last entry */
126         IGB_DEVICE_NULL
127 };
128
129 static int      igb_probe(device_t);
130 static int      igb_attach(device_t);
131 static int      igb_detach(device_t);
132 static int      igb_shutdown(device_t);
133 static int      igb_suspend(device_t);
134 static int      igb_resume(device_t);
135
136 static boolean_t igb_is_valid_ether_addr(const uint8_t *);
137 static void     igb_setup_ifp(struct igb_softc *);
138 static boolean_t igb_txcsum_ctx(struct igb_tx_ring *, struct mbuf *);
139 static void     igb_add_sysctl(struct igb_softc *);
140 static int      igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS);
141 static int      igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS);
142 static int      igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS);
143 static void     igb_set_ring_inuse(struct igb_softc *, boolean_t);
144
145 static void     igb_vf_init_stats(struct igb_softc *);
146 static void     igb_reset(struct igb_softc *);
147 static void     igb_update_stats_counters(struct igb_softc *);
148 static void     igb_update_vf_stats_counters(struct igb_softc *);
149 static void     igb_update_link_status(struct igb_softc *);
150 static void     igb_init_tx_unit(struct igb_softc *);
151 static void     igb_init_rx_unit(struct igb_softc *);
152
153 static void     igb_set_vlan(struct igb_softc *);
154 static void     igb_set_multi(struct igb_softc *);
155 static void     igb_set_promisc(struct igb_softc *);
156 static void     igb_disable_promisc(struct igb_softc *);
157
158 static int      igb_alloc_rings(struct igb_softc *);
159 static void     igb_free_rings(struct igb_softc *);
160 static int      igb_create_tx_ring(struct igb_tx_ring *);
161 static int      igb_create_rx_ring(struct igb_rx_ring *);
162 static void     igb_free_tx_ring(struct igb_tx_ring *);
163 static void     igb_free_rx_ring(struct igb_rx_ring *);
164 static void     igb_destroy_tx_ring(struct igb_tx_ring *, int);
165 static void     igb_destroy_rx_ring(struct igb_rx_ring *, int);
166 static void     igb_init_tx_ring(struct igb_tx_ring *);
167 static int      igb_init_rx_ring(struct igb_rx_ring *);
168 static int      igb_newbuf(struct igb_rx_ring *, int, boolean_t);
169 static int      igb_encap(struct igb_tx_ring *, struct mbuf **);
170
171 static void     igb_stop(struct igb_softc *);
172 static void     igb_init(void *);
173 static int      igb_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
174 static void     igb_media_status(struct ifnet *, struct ifmediareq *);
175 static int      igb_media_change(struct ifnet *);
176 static void     igb_timer(void *);
177 static void     igb_watchdog(struct ifnet *);
178 static void     igb_start(struct ifnet *);
179 #ifdef DEVICE_POLLING
180 static void     igb_poll(struct ifnet *, enum poll_cmd, int);
181 #endif
182 static void     igb_serialize(struct ifnet *, enum ifnet_serialize);
183 static void     igb_deserialize(struct ifnet *, enum ifnet_serialize);
184 static int      igb_tryserialize(struct ifnet *, enum ifnet_serialize);
185 #ifdef INVARIANTS
186 static void     igb_serialize_assert(struct ifnet *, enum ifnet_serialize,
187                     boolean_t);
188 #endif
189
190 static void     igb_intr(void *);
191 static void     igb_intr_shared(void *);
192 static void     igb_rxeof(struct igb_rx_ring *, int);
193 static void     igb_txeof(struct igb_tx_ring *);
194 static void     igb_set_eitr(struct igb_softc *, int, int);
195 static void     igb_enable_intr(struct igb_softc *);
196 static void     igb_disable_intr(struct igb_softc *);
197 static void     igb_init_unshared_intr(struct igb_softc *);
198 static void     igb_init_intr(struct igb_softc *);
199 static int      igb_setup_intr(struct igb_softc *);
200 static void     igb_set_txintr_mask(struct igb_tx_ring *, int *, int);
201 static void     igb_set_rxintr_mask(struct igb_rx_ring *, int *, int);
202 static void     igb_set_intr_mask(struct igb_softc *);
203 static int      igb_alloc_intr(struct igb_softc *);
204 static void     igb_free_intr(struct igb_softc *);
205 static void     igb_teardown_intr(struct igb_softc *);
206 static void     igb_msix_try_alloc(struct igb_softc *);
207 static void     igb_msix_free(struct igb_softc *, boolean_t);
208 static int      igb_msix_setup(struct igb_softc *);
209 static void     igb_msix_teardown(struct igb_softc *, int);
210 static void     igb_msix_rx(void *);
211 static void     igb_msix_tx(void *);
212 static void     igb_msix_status(void *);
213
214 /* Management and WOL Support */
215 static void     igb_get_mgmt(struct igb_softc *);
216 static void     igb_rel_mgmt(struct igb_softc *);
217 static void     igb_get_hw_control(struct igb_softc *);
218 static void     igb_rel_hw_control(struct igb_softc *);
219 static void     igb_enable_wol(device_t);
220
221 static device_method_t igb_methods[] = {
222         /* Device interface */
223         DEVMETHOD(device_probe,         igb_probe),
224         DEVMETHOD(device_attach,        igb_attach),
225         DEVMETHOD(device_detach,        igb_detach),
226         DEVMETHOD(device_shutdown,      igb_shutdown),
227         DEVMETHOD(device_suspend,       igb_suspend),
228         DEVMETHOD(device_resume,        igb_resume),
229         { 0, 0 }
230 };
231
232 static driver_t igb_driver = {
233         "igb",
234         igb_methods,
235         sizeof(struct igb_softc),
236 };
237
238 static devclass_t igb_devclass;
239
240 DECLARE_DUMMY_MODULE(if_igb);
241 MODULE_DEPEND(igb, ig_hal, 1, 1, 1);
242 DRIVER_MODULE(if_igb, pci, igb_driver, igb_devclass, NULL, NULL);
243
244 static int      igb_rxd = IGB_DEFAULT_RXD;
245 static int      igb_txd = IGB_DEFAULT_TXD;
246 static int      igb_rxr = 0;
247 static int      igb_msi_enable = 1;
248 static int      igb_msix_enable = 1;
249 static int      igb_eee_disabled = 1;   /* Energy Efficient Ethernet */
250 static int      igb_fc_setting = e1000_fc_full;
251
252 /*
253  * DMA Coalescing, only for i350 - default to off,
254  * this feature is for power savings
255  */
256 static int      igb_dma_coalesce = 0;
257
258 TUNABLE_INT("hw.igb.rxd", &igb_rxd);
259 TUNABLE_INT("hw.igb.txd", &igb_txd);
260 TUNABLE_INT("hw.igb.rxr", &igb_rxr);
261 TUNABLE_INT("hw.igb.msi.enable", &igb_msi_enable);
262 TUNABLE_INT("hw.igb.msix.enable", &igb_msix_enable);
263 TUNABLE_INT("hw.igb.fc_setting", &igb_fc_setting);
264
265 /* i350 specific */
266 TUNABLE_INT("hw.igb.eee_disabled", &igb_eee_disabled);
267 TUNABLE_INT("hw.igb.dma_coalesce", &igb_dma_coalesce);
268
269 static __inline void
270 igb_rxcsum(uint32_t staterr, struct mbuf *mp)
271 {
272         /* Ignore Checksum bit is set */
273         if (staterr & E1000_RXD_STAT_IXSM)
274                 return;
275
276         if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
277             E1000_RXD_STAT_IPCS)
278                 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
279
280         if (staterr & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
281                 if ((staterr & E1000_RXDEXT_STATERR_TCPE) == 0) {
282                         mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
283                             CSUM_PSEUDO_HDR | CSUM_FRAG_NOT_CHECKED;
284                         mp->m_pkthdr.csum_data = htons(0xffff);
285                 }
286         }
287 }
288
289 static __inline struct pktinfo *
290 igb_rssinfo(struct mbuf *m, struct pktinfo *pi,
291     uint32_t hash, uint32_t hashtype, uint32_t staterr)
292 {
293         switch (hashtype) {
294         case E1000_RXDADV_RSSTYPE_IPV4_TCP:
295                 pi->pi_netisr = NETISR_IP;
296                 pi->pi_flags = 0;
297                 pi->pi_l3proto = IPPROTO_TCP;
298                 break;
299
300         case E1000_RXDADV_RSSTYPE_IPV4:
301                 if (staterr & E1000_RXD_STAT_IXSM)
302                         return NULL;
303
304                 if ((staterr &
305                      (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
306                     E1000_RXD_STAT_TCPCS) {
307                         pi->pi_netisr = NETISR_IP;
308                         pi->pi_flags = 0;
309                         pi->pi_l3proto = IPPROTO_UDP;
310                         break;
311                 }
312                 /* FALL THROUGH */
313         default:
314                 return NULL;
315         }
316
317         m->m_flags |= M_HASH;
318         m->m_pkthdr.hash = toeplitz_hash(hash);
319         return pi;
320 }
321
322 static int
323 igb_probe(device_t dev)
324 {
325         const struct igb_device *d;
326         uint16_t vid, did;
327
328         vid = pci_get_vendor(dev);
329         did = pci_get_device(dev);
330
331         for (d = igb_devices; d->desc != NULL; ++d) {
332                 if (vid == d->vid && did == d->did) {
333                         device_set_desc(dev, d->desc);
334                         return 0;
335                 }
336         }
337         return ENXIO;
338 }
339
340 static int
341 igb_attach(device_t dev)
342 {
343         struct igb_softc *sc = device_get_softc(dev);
344         uint16_t eeprom_data;
345         int error = 0, i, j, ring_max;
346
347 #ifdef notyet
348         /* SYSCTL stuff */
349         SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
350             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
351             OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
352             igb_sysctl_nvm_info, "I", "NVM Information");
353
354         SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
355             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
356             OID_AUTO, "enable_aim", CTLTYPE_INT|CTLFLAG_RW,
357             &igb_enable_aim, 1, "Interrupt Moderation");
358
359         SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
360             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
361             OID_AUTO, "flow_control", CTLTYPE_INT|CTLFLAG_RW,
362             adapter, 0, igb_set_flowcntl, "I", "Flow Control");
363 #endif
364
365         callout_init_mp(&sc->timer);
366         lwkt_serialize_init(&sc->main_serialize);
367
368         sc->dev = sc->osdep.dev = dev;
369
370         /*
371          * Determine hardware and mac type
372          */
373         sc->hw.vendor_id = pci_get_vendor(dev);
374         sc->hw.device_id = pci_get_device(dev);
375         sc->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
376         sc->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2);
377         sc->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2);
378
379         if (e1000_set_mac_type(&sc->hw))
380                 return ENXIO;
381
382         /* Are we a VF device? */
383         if (sc->hw.mac.type == e1000_vfadapt ||
384             sc->hw.mac.type == e1000_vfadapt_i350)
385                 sc->vf_ifp = 1;
386         else
387                 sc->vf_ifp = 0;
388
389         /*
390          * Configure total supported RX/TX ring count
391          */
392         switch (sc->hw.mac.type) {
393         case e1000_82575:
394                 ring_max = IGB_MAX_RING_82575;
395                 break;
396         case e1000_82580:
397                 ring_max = IGB_MAX_RING_82580;
398                 break;
399         case e1000_i350:
400                 ring_max = IGB_MAX_RING_I350;
401                 break;
402         case e1000_82576:
403                 ring_max = IGB_MAX_RING_82576;
404                 break;
405         default:
406                 ring_max = IGB_MIN_RING;
407                 break;
408         }
409         sc->rx_ring_cnt = device_getenv_int(dev, "rxr", igb_rxr);
410         sc->rx_ring_cnt = if_ring_count2(sc->rx_ring_cnt, ring_max);
411 #ifdef IGB_RSS_DEBUG
412         sc->rx_ring_cnt = device_getenv_int(dev, "rxr_debug", sc->rx_ring_cnt);
413 #endif
414         sc->rx_ring_inuse = sc->rx_ring_cnt;
415         sc->tx_ring_cnt = 1; /* XXX */
416
417         /* Enable bus mastering */
418         pci_enable_busmaster(dev);
419
420         /*
421          * Allocate IO memory
422          */
423         sc->mem_rid = PCIR_BAR(0);
424         sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
425             RF_ACTIVE);
426         if (sc->mem_res == NULL) {
427                 device_printf(dev, "Unable to allocate bus resource: memory\n");
428                 error = ENXIO;
429                 goto failed;
430         }
431         sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->mem_res);
432         sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->mem_res);
433
434         sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
435
436         /* Save PCI command register for Shared Code */
437         sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
438         sc->hw.back = &sc->osdep;
439
440         /* Do Shared Code initialization */
441         if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
442                 device_printf(dev, "Setup of Shared code failed\n");
443                 error = ENXIO;
444                 goto failed;
445         }
446
447         e1000_get_bus_info(&sc->hw);
448
449         sc->hw.mac.autoneg = DO_AUTO_NEG;
450         sc->hw.phy.autoneg_wait_to_complete = FALSE;
451         sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
452
453         /* Copper options */
454         if (sc->hw.phy.media_type == e1000_media_type_copper) {
455                 sc->hw.phy.mdix = AUTO_ALL_MODES;
456                 sc->hw.phy.disable_polarity_correction = FALSE;
457                 sc->hw.phy.ms_type = IGB_MASTER_SLAVE;
458         }
459
460         /* Set the frame limits assuming  standard ethernet sized frames. */
461         sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
462
463         /* Allocate RX/TX rings */
464         error = igb_alloc_rings(sc);
465         if (error)
466                 goto failed;
467
468         /* Allocate interrupt */
469         error = igb_alloc_intr(sc);
470         if (error)
471                 goto failed;
472
473         /*
474          * Setup serializers
475          */
476         i = 0;
477         sc->serializes[i++] = &sc->main_serialize;
478
479         sc->tx_serialize = i;
480         for (j = 0; j < sc->tx_ring_cnt; ++j)
481                 sc->serializes[i++] = &sc->tx_rings[j].tx_serialize;
482
483         sc->rx_serialize = i;
484         for (j = 0; j < sc->rx_ring_cnt; ++j)
485                 sc->serializes[i++] = &sc->rx_rings[j].rx_serialize;
486
487         sc->serialize_cnt = i;
488         KKASSERT(sc->serialize_cnt <= IGB_NSERIALIZE);
489
490         /* Allocate the appropriate stats memory */
491         if (sc->vf_ifp) {
492                 sc->stats = kmalloc(sizeof(struct e1000_vf_stats), M_DEVBUF,
493                     M_WAITOK | M_ZERO);
494                 igb_vf_init_stats(sc);
495         } else {
496                 sc->stats = kmalloc(sizeof(struct e1000_hw_stats), M_DEVBUF,
497                     M_WAITOK | M_ZERO);
498         }
499
500         /* Allocate multicast array memory. */
501         sc->mta = kmalloc(ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES,
502             M_DEVBUF, M_WAITOK);
503
504         /* Some adapter-specific advanced features */
505         if (sc->hw.mac.type >= e1000_i350) {
506 #ifdef notyet
507                 igb_set_sysctl_value(adapter, "dma_coalesce",
508                     "configure dma coalesce",
509                     &adapter->dma_coalesce, igb_dma_coalesce);
510                 igb_set_sysctl_value(adapter, "eee_disabled",
511                     "enable Energy Efficient Ethernet",
512                     &adapter->hw.dev_spec._82575.eee_disable,
513                     igb_eee_disabled);
514 #else
515                 sc->dma_coalesce = igb_dma_coalesce;
516                 sc->hw.dev_spec._82575.eee_disable = igb_eee_disabled;
517 #endif
518                 e1000_set_eee_i350(&sc->hw);
519         }
520
521         /*
522          * Start from a known state, this is important in reading the nvm and
523          * mac from that.
524          */
525         e1000_reset_hw(&sc->hw);
526
527         /* Make sure we have a good EEPROM before we read from it */
528         if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
529                 /*
530                  * Some PCI-E parts fail the first check due to
531                  * the link being in sleep state, call it again,
532                  * if it fails a second time its a real issue.
533                  */
534                 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
535                         device_printf(dev,
536                             "The EEPROM Checksum Is Not Valid\n");
537                         error = EIO;
538                         goto failed;
539                 }
540         }
541
542         /* Copy the permanent MAC address out of the EEPROM */
543         if (e1000_read_mac_addr(&sc->hw) < 0) {
544                 device_printf(dev, "EEPROM read error while reading MAC"
545                     " address\n");
546                 error = EIO;
547                 goto failed;
548         }
549         if (!igb_is_valid_ether_addr(sc->hw.mac.addr)) {
550                 device_printf(dev, "Invalid MAC address\n");
551                 error = EIO;
552                 goto failed;
553         }
554
555 #ifdef notyet
556         /* 
557         ** Configure Interrupts
558         */
559         if ((adapter->msix > 1) && (igb_enable_msix))
560                 error = igb_allocate_msix(adapter);
561         else /* MSI or Legacy */
562                 error = igb_allocate_legacy(adapter);
563         if (error)
564                 goto err_late;
565 #endif
566
567         /* Setup OS specific network interface */
568         igb_setup_ifp(sc);
569
570         /* Add sysctl tree, must after igb_setup_ifp() */
571         igb_add_sysctl(sc);
572
573         /* Now get a good starting state */
574         igb_reset(sc);
575
576         /* Initialize statistics */
577         igb_update_stats_counters(sc);
578
579         sc->hw.mac.get_link_status = 1;
580         igb_update_link_status(sc);
581
582         /* Indicate SOL/IDER usage */
583         if (e1000_check_reset_block(&sc->hw)) {
584                 device_printf(dev,
585                     "PHY reset is blocked due to SOL/IDER session.\n");
586         }
587
588         /* Determine if we have to control management hardware */
589         if (e1000_enable_mng_pass_thru(&sc->hw))
590                 sc->flags |= IGB_FLAG_HAS_MGMT;
591
592         /*
593          * Setup Wake-on-Lan
594          */
595         /* APME bit in EEPROM is mapped to WUC.APME */
596         eeprom_data = E1000_READ_REG(&sc->hw, E1000_WUC) & E1000_WUC_APME;
597         if (eeprom_data)
598                 sc->wol = E1000_WUFC_MAG;
599         /* XXX disable WOL */
600         sc->wol = 0; 
601
602 #ifdef notyet
603         /* Register for VLAN events */
604         adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
605              igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
606         adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
607              igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
608 #endif
609
610 #ifdef notyet
611         igb_add_hw_stats(adapter);
612 #endif
613
614         error = igb_setup_intr(sc);
615         if (error) {
616                 ether_ifdetach(&sc->arpcom.ac_if);
617                 goto failed;
618         }
619         return 0;
620
621 failed:
622         igb_detach(dev);
623         return error;
624 }
625
626 static int
627 igb_detach(device_t dev)
628 {
629         struct igb_softc *sc = device_get_softc(dev);
630
631         if (device_is_attached(dev)) {
632                 struct ifnet *ifp = &sc->arpcom.ac_if;
633
634                 ifnet_serialize_all(ifp);
635
636                 igb_stop(sc);
637
638                 e1000_phy_hw_reset(&sc->hw);
639
640                 /* Give control back to firmware */
641                 igb_rel_mgmt(sc);
642                 igb_rel_hw_control(sc);
643
644                 if (sc->wol) {
645                         E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
646                         E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
647                         igb_enable_wol(dev);
648                 }
649
650                 igb_teardown_intr(sc);
651
652                 ifnet_deserialize_all(ifp);
653
654                 ether_ifdetach(ifp);
655         } else if (sc->mem_res != NULL) {
656                 igb_rel_hw_control(sc);
657         }
658         bus_generic_detach(dev);
659
660         if (sc->sysctl_tree != NULL)
661                 sysctl_ctx_free(&sc->sysctl_ctx);
662
663         igb_free_intr(sc);
664
665         if (sc->msix_mem_res != NULL) {
666                 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_mem_rid,
667                     sc->msix_mem_res);
668         }
669         if (sc->mem_res != NULL) {
670                 bus_release_resource(dev, SYS_RES_MEMORY, sc->mem_rid,
671                     sc->mem_res);
672         }
673
674         igb_free_rings(sc);
675
676         if (sc->mta != NULL)
677                 kfree(sc->mta, M_DEVBUF);
678         if (sc->stats != NULL)
679                 kfree(sc->stats, M_DEVBUF);
680
681         return 0;
682 }
683
684 static int
685 igb_shutdown(device_t dev)
686 {
687         return igb_suspend(dev);
688 }
689
690 static int
691 igb_suspend(device_t dev)
692 {
693         struct igb_softc *sc = device_get_softc(dev);
694         struct ifnet *ifp = &sc->arpcom.ac_if;
695
696         ifnet_serialize_all(ifp);
697
698         igb_stop(sc);
699
700         igb_rel_mgmt(sc);
701         igb_rel_hw_control(sc);
702
703         if (sc->wol) {
704                 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
705                 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
706                 igb_enable_wol(dev);
707         }
708
709         ifnet_deserialize_all(ifp);
710
711         return bus_generic_suspend(dev);
712 }
713
714 static int
715 igb_resume(device_t dev)
716 {
717         struct igb_softc *sc = device_get_softc(dev);
718         struct ifnet *ifp = &sc->arpcom.ac_if;
719
720         ifnet_serialize_all(ifp);
721
722         igb_init(sc);
723         igb_get_mgmt(sc);
724
725         if_devstart(ifp);
726
727         ifnet_deserialize_all(ifp);
728
729         return bus_generic_resume(dev);
730 }
731
732 static int
733 igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
734 {
735         struct igb_softc *sc = ifp->if_softc;
736         struct ifreq *ifr = (struct ifreq *)data;
737         int max_frame_size, mask, reinit;
738         int error = 0;
739
740         ASSERT_IFNET_SERIALIZED_ALL(ifp);
741
742         switch (command) {
743         case SIOCSIFMTU:
744                 max_frame_size = 9234;
745                 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
746                     ETHER_CRC_LEN) {
747                         error = EINVAL;
748                         break;
749                 }
750
751                 ifp->if_mtu = ifr->ifr_mtu;
752                 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
753                     ETHER_CRC_LEN;
754
755                 if (ifp->if_flags & IFF_RUNNING)
756                         igb_init(sc);
757                 break;
758
759         case SIOCSIFFLAGS:
760                 if (ifp->if_flags & IFF_UP) {
761                         if (ifp->if_flags & IFF_RUNNING) {
762                                 if ((ifp->if_flags ^ sc->if_flags) &
763                                     (IFF_PROMISC | IFF_ALLMULTI)) {
764                                         igb_disable_promisc(sc);
765                                         igb_set_promisc(sc);
766                                 }
767                         } else {
768                                 igb_init(sc);
769                         }
770                 } else if (ifp->if_flags & IFF_RUNNING) {
771                         igb_stop(sc);
772                 }
773                 sc->if_flags = ifp->if_flags;
774                 break;
775
776         case SIOCADDMULTI:
777         case SIOCDELMULTI:
778                 if (ifp->if_flags & IFF_RUNNING) {
779                         igb_disable_intr(sc);
780                         igb_set_multi(sc);
781 #ifdef DEVICE_POLLING
782                         if (!(ifp->if_flags & IFF_POLLING))
783 #endif
784                                 igb_enable_intr(sc);
785                 }
786                 break;
787
788         case SIOCSIFMEDIA:
789                 /*
790                  * As the speed/duplex settings are being
791                  * changed, we need toreset the PHY.
792                  */
793                 sc->hw.phy.reset_disable = FALSE;
794
795                 /* Check SOL/IDER usage */
796                 if (e1000_check_reset_block(&sc->hw)) {
797                         if_printf(ifp, "Media change is "
798                             "blocked due to SOL/IDER session.\n");
799                         break;
800                 }
801                 /* FALL THROUGH */
802
803         case SIOCGIFMEDIA:
804                 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
805                 break;
806
807         case SIOCSIFCAP:
808                 reinit = 0;
809                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
810                 if (mask & IFCAP_RXCSUM) {
811                         ifp->if_capenable ^= IFCAP_RXCSUM;
812                         reinit = 1;
813                 }
814                 if (mask & IFCAP_VLAN_HWTAGGING) {
815                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
816                         reinit = 1;
817                 }
818                 if (mask & IFCAP_TXCSUM) {
819                         ifp->if_capenable ^= IFCAP_TXCSUM;
820                         if (ifp->if_capenable & IFCAP_TXCSUM)
821                                 ifp->if_hwassist |= IGB_CSUM_FEATURES;
822                         else
823                                 ifp->if_hwassist &= ~IGB_CSUM_FEATURES;
824                 }
825                 if (mask & IFCAP_RSS)
826                         ifp->if_capenable ^= IFCAP_RSS;
827                 if (reinit && (ifp->if_flags & IFF_RUNNING))
828                         igb_init(sc);
829                 break;
830
831         default:
832                 error = ether_ioctl(ifp, command, data);
833                 break;
834         }
835         return error;
836 }
837
838 static void
839 igb_init(void *xsc)
840 {
841         struct igb_softc *sc = xsc;
842         struct ifnet *ifp = &sc->arpcom.ac_if;
843         boolean_t polling;
844         int i;
845
846         ASSERT_IFNET_SERIALIZED_ALL(ifp);
847
848         igb_stop(sc);
849
850         /* Get the latest mac address, User can use a LAA */
851         bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
852
853         /* Put the address into the Receive Address Array */
854         e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
855
856         igb_reset(sc);
857         igb_update_link_status(sc);
858
859         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
860
861         /* Configure for OS presence */
862         igb_get_mgmt(sc);
863
864         polling = FALSE;
865 #ifdef DEVICE_POLLING
866         if (ifp->if_flags & IFF_POLLING)
867                 polling = TRUE;
868 #endif
869
870         /* Configured used RX/TX rings */
871         igb_set_ring_inuse(sc, polling);
872
873         /* Initialize interrupt */
874         igb_init_intr(sc);
875
876         /* Prepare transmit descriptors and buffers */
877         for (i = 0; i < sc->tx_ring_cnt; ++i)
878                 igb_init_tx_ring(&sc->tx_rings[i]);
879         igb_init_tx_unit(sc);
880
881         /* Setup Multicast table */
882         igb_set_multi(sc);
883
884 #if 0
885         /*
886          * Figure out the desired mbuf pool
887          * for doing jumbo/packetsplit
888          */
889         if (adapter->max_frame_size <= 2048)
890                 adapter->rx_mbuf_sz = MCLBYTES;
891         else if (adapter->max_frame_size <= 4096)
892                 adapter->rx_mbuf_sz = MJUMPAGESIZE;
893         else
894                 adapter->rx_mbuf_sz = MJUM9BYTES;
895 #endif
896
897         /* Prepare receive descriptors and buffers */
898         for (i = 0; i < sc->rx_ring_inuse; ++i) {
899                 int error;
900
901                 error = igb_init_rx_ring(&sc->rx_rings[i]);
902                 if (error) {
903                         if_printf(ifp, "Could not setup receive structures\n");
904                         igb_stop(sc);
905                         return;
906                 }
907         }
908         igb_init_rx_unit(sc);
909
910         /* Enable VLAN support */
911         if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
912                 igb_set_vlan(sc);
913
914         /* Don't lose promiscuous settings */
915         igb_set_promisc(sc);
916
917         ifp->if_flags |= IFF_RUNNING;
918         ifp->if_flags &= ~IFF_OACTIVE;
919
920         callout_reset(&sc->timer, hz, igb_timer, sc);
921         e1000_clear_hw_cntrs_base_generic(&sc->hw);
922
923 #if 0
924         if (adapter->msix > 1) /* Set up queue routing */
925                 igb_configure_queues(adapter);
926 #endif
927
928         /* This clears any pending interrupts */
929         E1000_READ_REG(&sc->hw, E1000_ICR);
930
931         /*
932          * Only enable interrupts if we are not polling, make sure
933          * they are off otherwise.
934          */
935         if (polling) {
936                 igb_disable_intr(sc);
937         } else {
938                 igb_enable_intr(sc);
939                 E1000_WRITE_REG(&sc->hw, E1000_ICS, E1000_ICS_LSC);
940         }
941
942         /* Set Energy Efficient Ethernet */
943         e1000_set_eee_i350(&sc->hw);
944
945         /* Don't reset the phy next time init gets called */
946         sc->hw.phy.reset_disable = TRUE;
947 }
948
949 static void
950 igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
951 {
952         struct igb_softc *sc = ifp->if_softc;
953         u_char fiber_type = IFM_1000_SX;
954
955         ASSERT_IFNET_SERIALIZED_ALL(ifp);
956
957         igb_update_link_status(sc);
958
959         ifmr->ifm_status = IFM_AVALID;
960         ifmr->ifm_active = IFM_ETHER;
961
962         if (!sc->link_active)
963                 return;
964
965         ifmr->ifm_status |= IFM_ACTIVE;
966
967         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
968             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
969                 ifmr->ifm_active |= fiber_type | IFM_FDX;
970         } else {
971                 switch (sc->link_speed) {
972                 case 10:
973                         ifmr->ifm_active |= IFM_10_T;
974                         break;
975
976                 case 100:
977                         ifmr->ifm_active |= IFM_100_TX;
978                         break;
979
980                 case 1000:
981                         ifmr->ifm_active |= IFM_1000_T;
982                         break;
983                 }
984                 if (sc->link_duplex == FULL_DUPLEX)
985                         ifmr->ifm_active |= IFM_FDX;
986                 else
987                         ifmr->ifm_active |= IFM_HDX;
988         }
989 }
990
991 static int
992 igb_media_change(struct ifnet *ifp)
993 {
994         struct igb_softc *sc = ifp->if_softc;
995         struct ifmedia *ifm = &sc->media;
996
997         ASSERT_IFNET_SERIALIZED_ALL(ifp);
998
999         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1000                 return EINVAL;
1001
1002         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1003         case IFM_AUTO:
1004                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1005                 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1006                 break;
1007
1008         case IFM_1000_LX:
1009         case IFM_1000_SX:
1010         case IFM_1000_T:
1011                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1012                 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1013                 break;
1014
1015         case IFM_100_TX:
1016                 sc->hw.mac.autoneg = FALSE;
1017                 sc->hw.phy.autoneg_advertised = 0;
1018                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1019                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1020                 else
1021                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1022                 break;
1023
1024         case IFM_10_T:
1025                 sc->hw.mac.autoneg = FALSE;
1026                 sc->hw.phy.autoneg_advertised = 0;
1027                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1028                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1029                 else
1030                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1031                 break;
1032
1033         default:
1034                 if_printf(ifp, "Unsupported media type\n");
1035                 break;
1036         }
1037
1038         igb_init(sc);
1039
1040         return 0;
1041 }
1042
1043 static void
1044 igb_set_promisc(struct igb_softc *sc)
1045 {
1046         struct ifnet *ifp = &sc->arpcom.ac_if;
1047         struct e1000_hw *hw = &sc->hw;
1048         uint32_t reg;
1049
1050         if (sc->vf_ifp) {
1051                 e1000_promisc_set_vf(hw, e1000_promisc_enabled);
1052                 return;
1053         }
1054
1055         reg = E1000_READ_REG(hw, E1000_RCTL);
1056         if (ifp->if_flags & IFF_PROMISC) {
1057                 reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1058                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1059         } else if (ifp->if_flags & IFF_ALLMULTI) {
1060                 reg |= E1000_RCTL_MPE;
1061                 reg &= ~E1000_RCTL_UPE;
1062                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1063         }
1064 }
1065
1066 static void
1067 igb_disable_promisc(struct igb_softc *sc)
1068 {
1069         struct e1000_hw *hw = &sc->hw;
1070         uint32_t reg;
1071
1072         if (sc->vf_ifp) {
1073                 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
1074                 return;
1075         }
1076         reg = E1000_READ_REG(hw, E1000_RCTL);
1077         reg &= ~E1000_RCTL_UPE;
1078         reg &= ~E1000_RCTL_MPE;
1079         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1080 }
1081
1082 static void
1083 igb_set_multi(struct igb_softc *sc)
1084 {
1085         struct ifnet *ifp = &sc->arpcom.ac_if;
1086         struct ifmultiaddr *ifma;
1087         uint32_t reg_rctl = 0;
1088         uint8_t *mta;
1089         int mcnt = 0;
1090
1091         mta = sc->mta;
1092         bzero(mta, ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
1093
1094         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1095                 if (ifma->ifma_addr->sa_family != AF_LINK)
1096                         continue;
1097
1098                 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1099                         break;
1100
1101                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1102                     &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
1103                 mcnt++;
1104         }
1105
1106         if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1107                 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1108                 reg_rctl |= E1000_RCTL_MPE;
1109                 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1110         } else {
1111                 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1112         }
1113 }
1114
1115 static void
1116 igb_timer(void *xsc)
1117 {
1118         struct igb_softc *sc = xsc;
1119
1120         lwkt_serialize_enter(&sc->main_serialize);
1121
1122         igb_update_link_status(sc);
1123         igb_update_stats_counters(sc);
1124
1125         callout_reset(&sc->timer, hz, igb_timer, sc);
1126
1127         lwkt_serialize_exit(&sc->main_serialize);
1128 }
1129
1130 static void
1131 igb_update_link_status(struct igb_softc *sc)
1132 {
1133         struct ifnet *ifp = &sc->arpcom.ac_if;
1134         struct e1000_hw *hw = &sc->hw;
1135         uint32_t link_check, thstat, ctrl;
1136
1137         link_check = thstat = ctrl = 0;
1138
1139         /* Get the cached link value or read for real */
1140         switch (hw->phy.media_type) {
1141         case e1000_media_type_copper:
1142                 if (hw->mac.get_link_status) {
1143                         /* Do the work to read phy */
1144                         e1000_check_for_link(hw);
1145                         link_check = !hw->mac.get_link_status;
1146                 } else {
1147                         link_check = TRUE;
1148                 }
1149                 break;
1150
1151         case e1000_media_type_fiber:
1152                 e1000_check_for_link(hw);
1153                 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1154                 break;
1155
1156         case e1000_media_type_internal_serdes:
1157                 e1000_check_for_link(hw);
1158                 link_check = hw->mac.serdes_has_link;
1159                 break;
1160
1161         /* VF device is type_unknown */
1162         case e1000_media_type_unknown:
1163                 e1000_check_for_link(hw);
1164                 link_check = !hw->mac.get_link_status;
1165                 /* Fall thru */
1166         default:
1167                 break;
1168         }
1169
1170         /* Check for thermal downshift or shutdown */
1171         if (hw->mac.type == e1000_i350) {
1172                 thstat = E1000_READ_REG(hw, E1000_THSTAT);
1173                 ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
1174         }
1175
1176         /* Now we check if a transition has happened */
1177         if (link_check && sc->link_active == 0) {
1178                 e1000_get_speed_and_duplex(hw, 
1179                     &sc->link_speed, &sc->link_duplex);
1180                 if (bootverbose) {
1181                         if_printf(ifp, "Link is up %d Mbps %s\n",
1182                             sc->link_speed,
1183                             sc->link_duplex == FULL_DUPLEX ?
1184                             "Full Duplex" : "Half Duplex");
1185                 }
1186                 sc->link_active = 1;
1187
1188                 ifp->if_baudrate = sc->link_speed * 1000000;
1189                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1190                     (thstat & E1000_THSTAT_LINK_THROTTLE))
1191                         if_printf(ifp, "Link: thermal downshift\n");
1192                 /* This can sleep */
1193                 ifp->if_link_state = LINK_STATE_UP;
1194                 if_link_state_change(ifp);
1195         } else if (!link_check && sc->link_active == 1) {
1196                 ifp->if_baudrate = sc->link_speed = 0;
1197                 sc->link_duplex = 0;
1198                 if (bootverbose)
1199                         if_printf(ifp, "Link is Down\n");
1200                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1201                     (thstat & E1000_THSTAT_PWR_DOWN))
1202                         if_printf(ifp, "Link: thermal shutdown\n");
1203                 sc->link_active = 0;
1204                 /* This can sleep */
1205                 ifp->if_link_state = LINK_STATE_DOWN;
1206                 if_link_state_change(ifp);
1207         }
1208 }
1209
1210 static void
1211 igb_stop(struct igb_softc *sc)
1212 {
1213         struct ifnet *ifp = &sc->arpcom.ac_if;
1214         int i;
1215
1216         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1217
1218         igb_disable_intr(sc);
1219
1220         callout_stop(&sc->timer);
1221
1222         ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1223         ifp->if_timer = 0;
1224
1225         e1000_reset_hw(&sc->hw);
1226         E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1227
1228         e1000_led_off(&sc->hw);
1229         e1000_cleanup_led(&sc->hw);
1230
1231         for (i = 0; i < sc->tx_ring_cnt; ++i)
1232                 igb_free_tx_ring(&sc->tx_rings[i]);
1233         for (i = 0; i < sc->rx_ring_cnt; ++i)
1234                 igb_free_rx_ring(&sc->rx_rings[i]);
1235 }
1236
1237 static void
1238 igb_reset(struct igb_softc *sc)
1239 {
1240         struct ifnet *ifp = &sc->arpcom.ac_if;
1241         struct e1000_hw *hw = &sc->hw;
1242         struct e1000_fc_info *fc = &hw->fc;
1243         uint32_t pba = 0;
1244         uint16_t hwm;
1245
1246         /* Let the firmware know the OS is in control */
1247         igb_get_hw_control(sc);
1248
1249         /*
1250          * Packet Buffer Allocation (PBA)
1251          * Writing PBA sets the receive portion of the buffer
1252          * the remainder is used for the transmit buffer.
1253          */
1254         switch (hw->mac.type) {
1255         case e1000_82575:
1256                 pba = E1000_PBA_32K;
1257                 break;
1258
1259         case e1000_82576:
1260         case e1000_vfadapt:
1261                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1262                 pba &= E1000_RXPBS_SIZE_MASK_82576;
1263                 break;
1264
1265         case e1000_82580:
1266         case e1000_i350:
1267         case e1000_vfadapt_i350:
1268                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1269                 pba = e1000_rxpbs_adjust_82580(pba);
1270                 break;
1271                 /* XXX pba = E1000_PBA_35K; */
1272
1273         default:
1274                 break;
1275         }
1276
1277         /* Special needs in case of Jumbo frames */
1278         if (hw->mac.type == e1000_82575 && ifp->if_mtu > ETHERMTU) {
1279                 uint32_t tx_space, min_tx, min_rx;
1280
1281                 pba = E1000_READ_REG(hw, E1000_PBA);
1282                 tx_space = pba >> 16;
1283                 pba &= 0xffff;
1284
1285                 min_tx = (sc->max_frame_size +
1286                     sizeof(struct e1000_tx_desc) - ETHER_CRC_LEN) * 2;
1287                 min_tx = roundup2(min_tx, 1024);
1288                 min_tx >>= 10;
1289                 min_rx = sc->max_frame_size;
1290                 min_rx = roundup2(min_rx, 1024);
1291                 min_rx >>= 10;
1292                 if (tx_space < min_tx && (min_tx - tx_space) < pba) {
1293                         pba = pba - (min_tx - tx_space);
1294                         /*
1295                          * if short on rx space, rx wins
1296                          * and must trump tx adjustment
1297                          */
1298                         if (pba < min_rx)
1299                                 pba = min_rx;
1300                 }
1301                 E1000_WRITE_REG(hw, E1000_PBA, pba);
1302         }
1303
1304         /*
1305          * These parameters control the automatic generation (Tx) and
1306          * response (Rx) to Ethernet PAUSE frames.
1307          * - High water mark should allow for at least two frames to be
1308          *   received after sending an XOFF.
1309          * - Low water mark works best when it is very near the high water mark.
1310          *   This allows the receiver to restart by sending XON when it has
1311          *   drained a bit.
1312          */
1313         hwm = min(((pba << 10) * 9 / 10),
1314             ((pba << 10) - 2 * sc->max_frame_size));
1315
1316         if (hw->mac.type < e1000_82576) {
1317                 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1318                 fc->low_water = fc->high_water - 8;
1319         } else {
1320                 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1321                 fc->low_water = fc->high_water - 16;
1322         }
1323         fc->pause_time = IGB_FC_PAUSE_TIME;
1324         fc->send_xon = TRUE;
1325
1326         /* Issue a global reset */
1327         e1000_reset_hw(hw);
1328         E1000_WRITE_REG(hw, E1000_WUC, 0);
1329
1330         if (e1000_init_hw(hw) < 0)
1331                 if_printf(ifp, "Hardware Initialization Failed\n");
1332
1333         /* Setup DMA Coalescing */
1334         if (hw->mac.type == e1000_i350 && sc->dma_coalesce) {
1335                 uint32_t reg;
1336
1337                 hwm = (pba - 4) << 10;
1338                 reg = ((pba - 6) << E1000_DMACR_DMACTHR_SHIFT)
1339                     & E1000_DMACR_DMACTHR_MASK;
1340
1341                 /* transition to L0x or L1 if available..*/
1342                 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
1343
1344                 /* timer = +-1000 usec in 32usec intervals */
1345                 reg |= (1000 >> 5);
1346                 E1000_WRITE_REG(hw, E1000_DMACR, reg);
1347
1348                 /* No lower threshold */
1349                 E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
1350
1351                 /* set hwm to PBA -  2 * max frame size */
1352                 E1000_WRITE_REG(hw, E1000_FCRTC, hwm);
1353
1354                 /* Set the interval before transition */
1355                 reg = E1000_READ_REG(hw, E1000_DMCTLX);
1356                 reg |= 0x800000FF; /* 255 usec */
1357                 E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
1358
1359                 /* free space in tx packet buffer to wake from DMA coal */
1360                 E1000_WRITE_REG(hw, E1000_DMCTXTH,
1361                     (20480 - (2 * sc->max_frame_size)) >> 6);
1362
1363                 /* make low power state decision controlled by DMA coal */
1364                 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
1365                 E1000_WRITE_REG(hw, E1000_PCIEMISC,
1366                     reg | E1000_PCIEMISC_LX_DECISION);
1367                 if_printf(ifp, "DMA Coalescing enabled\n");
1368         }
1369
1370         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1371         e1000_get_phy_info(hw);
1372         e1000_check_for_link(hw);
1373 }
1374
1375 static void
1376 igb_setup_ifp(struct igb_softc *sc)
1377 {
1378         struct ifnet *ifp = &sc->arpcom.ac_if;
1379
1380         if_initname(ifp, device_get_name(sc->dev), device_get_unit(sc->dev));
1381         ifp->if_softc = sc;
1382         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1383         ifp->if_init = igb_init;
1384         ifp->if_ioctl = igb_ioctl;
1385         ifp->if_start = igb_start;
1386         ifp->if_serialize = igb_serialize;
1387         ifp->if_deserialize = igb_deserialize;
1388         ifp->if_tryserialize = igb_tryserialize;
1389 #ifdef INVARIANTS
1390         ifp->if_serialize_assert = igb_serialize_assert;
1391 #endif
1392 #ifdef DEVICE_POLLING
1393         ifp->if_poll = igb_poll;
1394 #endif
1395         ifp->if_watchdog = igb_watchdog;
1396
1397         ifq_set_maxlen(&ifp->if_snd, sc->tx_rings[0].num_tx_desc - 1);
1398         ifq_set_ready(&ifp->if_snd);
1399
1400         ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1401
1402         ifp->if_capabilities =
1403             IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
1404         if (IGB_ENABLE_HWRSS(sc))
1405                 ifp->if_capabilities |= IFCAP_RSS;
1406         ifp->if_capenable = ifp->if_capabilities;
1407         ifp->if_hwassist = IGB_CSUM_FEATURES;
1408
1409         /*
1410          * Tell the upper layer(s) we support long frames
1411          */
1412         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1413
1414         /*
1415          * Specify the media types supported by this adapter and register
1416          * callbacks to update media and link information
1417          */
1418         ifmedia_init(&sc->media, IFM_IMASK, igb_media_change, igb_media_status);
1419         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1420             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1421                 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1422                     0, NULL);
1423                 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1424         } else {
1425                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1426                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1427                     0, NULL);
1428                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1429                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1430                     0, NULL);
1431                 if (sc->hw.phy.type != e1000_phy_ife) {
1432                         ifmedia_add(&sc->media,
1433                             IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1434                         ifmedia_add(&sc->media,
1435                             IFM_ETHER | IFM_1000_T, 0, NULL);
1436                 }
1437         }
1438         ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1439         ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1440 }
1441
1442 static void
1443 igb_add_sysctl(struct igb_softc *sc)
1444 {
1445         char node[32];
1446         int i;
1447
1448         sysctl_ctx_init(&sc->sysctl_ctx);
1449         sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
1450             SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
1451             device_get_nameunit(sc->dev), CTLFLAG_RD, 0, "");
1452         if (sc->sysctl_tree == NULL) {
1453                 device_printf(sc->dev, "can't add sysctl node\n");
1454                 return;
1455         }
1456
1457         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1458             OID_AUTO, "rxr", CTLFLAG_RD, &sc->rx_ring_cnt, 0, "# of RX rings");
1459         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1460             OID_AUTO, "rxr_inuse", CTLFLAG_RD, &sc->rx_ring_inuse, 0,
1461             "# of RX rings used");
1462         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1463             OID_AUTO, "rxd", CTLFLAG_RD, &sc->rx_rings[0].num_rx_desc, 0,
1464             "# of RX descs");
1465         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1466             OID_AUTO, "txd", CTLFLAG_RD, &sc->tx_rings[0].num_tx_desc, 0,
1467             "# of TX descs");
1468
1469         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
1470                 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1471                     SYSCTL_CHILDREN(sc->sysctl_tree),
1472                     OID_AUTO, "intr_rate", CTLTYPE_INT | CTLFLAG_RW,
1473                     sc, 0, igb_sysctl_intr_rate, "I", "interrupt rate");
1474         } else {
1475                 for (i = 0; i < sc->msix_cnt; ++i) {
1476                         struct igb_msix_data *msix = &sc->msix_data[i];
1477
1478                         ksnprintf(node, sizeof(node), "msix%d_rate", i);
1479                         SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1480                             SYSCTL_CHILDREN(sc->sysctl_tree),
1481                             OID_AUTO, node, CTLTYPE_INT | CTLFLAG_RW,
1482                             msix, 0, igb_sysctl_msix_rate, "I",
1483                             msix->msix_rate_desc);
1484                 }
1485         }
1486
1487         SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1488             OID_AUTO, "tx_intr_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1489             sc, 0, igb_sysctl_tx_intr_nsegs, "I",
1490             "# of segments per TX interrupt");
1491
1492 #ifdef IGB_RSS_DEBUG
1493         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1494             OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug, 0,
1495             "RSS debug level");
1496         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1497                 ksnprintf(node, sizeof(node), "rx%d_pkt", i);
1498                 SYSCTL_ADD_ULONG(&sc->sysctl_ctx,
1499                     SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, node,
1500                     CTLFLAG_RW, &sc->rx_rings[i].rx_packets, "RXed packets");
1501         }
1502 #endif
1503 }
1504
1505 static int
1506 igb_alloc_rings(struct igb_softc *sc)
1507 {
1508         int error, i;
1509
1510         /*
1511          * Create top level busdma tag
1512          */
1513         error = bus_dma_tag_create(NULL, 1, 0,
1514             BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1515             BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1516             &sc->parent_tag);
1517         if (error) {
1518                 device_printf(sc->dev, "could not create top level DMA tag\n");
1519                 return error;
1520         }
1521
1522         /*
1523          * Allocate TX descriptor rings and buffers
1524          */
1525         sc->tx_rings = kmalloc(sizeof(struct igb_tx_ring) * sc->tx_ring_cnt,
1526             M_DEVBUF, M_WAITOK | M_ZERO);
1527         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1528                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1529
1530                 /* Set up some basics */
1531                 txr->sc = sc;
1532                 txr->me = i;
1533                 lwkt_serialize_init(&txr->tx_serialize);
1534
1535                 error = igb_create_tx_ring(txr);
1536                 if (error)
1537                         return error;
1538         }
1539
1540         /*
1541          * Allocate RX descriptor rings and buffers
1542          */ 
1543         sc->rx_rings = kmalloc(sizeof(struct igb_rx_ring) * sc->rx_ring_cnt,
1544             M_DEVBUF, M_WAITOK | M_ZERO);
1545         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1546                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1547
1548                 /* Set up some basics */
1549                 rxr->sc = sc;
1550                 rxr->me = i;
1551                 lwkt_serialize_init(&rxr->rx_serialize);
1552
1553                 error = igb_create_rx_ring(rxr);
1554                 if (error)
1555                         return error;
1556         }
1557
1558         return 0;
1559 }
1560
1561 static void
1562 igb_free_rings(struct igb_softc *sc)
1563 {
1564         int i;
1565
1566         if (sc->tx_rings != NULL) {
1567                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1568                         struct igb_tx_ring *txr = &sc->tx_rings[i];
1569
1570                         igb_destroy_tx_ring(txr, txr->num_tx_desc);
1571                 }
1572                 kfree(sc->tx_rings, M_DEVBUF);
1573         }
1574
1575         if (sc->rx_rings != NULL) {
1576                 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1577                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
1578
1579                         igb_destroy_rx_ring(rxr, rxr->num_rx_desc);
1580                 }
1581                 kfree(sc->rx_rings, M_DEVBUF);
1582         }
1583 }
1584
1585 static int
1586 igb_create_tx_ring(struct igb_tx_ring *txr)
1587 {
1588         int tsize, error, i;
1589
1590         /*
1591          * Validate number of transmit descriptors. It must not exceed
1592          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1593          */
1594         if (((igb_txd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN) != 0 ||
1595             (igb_txd > IGB_MAX_TXD) || (igb_txd < IGB_MIN_TXD)) {
1596                 device_printf(txr->sc->dev,
1597                     "Using %d TX descriptors instead of %d!\n",
1598                     IGB_DEFAULT_TXD, igb_txd);
1599                 txr->num_tx_desc = IGB_DEFAULT_TXD;
1600         } else {
1601                 txr->num_tx_desc = igb_txd;
1602         }
1603
1604         /*
1605          * Allocate TX descriptor ring
1606          */
1607         tsize = roundup2(txr->num_tx_desc * sizeof(union e1000_adv_tx_desc),
1608             IGB_DBA_ALIGN);
1609         txr->txdma.dma_vaddr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1610             IGB_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1611             &txr->txdma.dma_tag, &txr->txdma.dma_map, &txr->txdma.dma_paddr);
1612         if (txr->txdma.dma_vaddr == NULL) {
1613                 device_printf(txr->sc->dev,
1614                     "Unable to allocate TX Descriptor memory\n");
1615                 return ENOMEM;
1616         }
1617         txr->tx_base = txr->txdma.dma_vaddr;
1618         bzero(txr->tx_base, tsize);
1619
1620         txr->tx_buf = kmalloc(sizeof(struct igb_tx_buf) * txr->num_tx_desc,
1621             M_DEVBUF, M_WAITOK | M_ZERO);
1622
1623         /*
1624          * Allocate TX head write-back buffer
1625          */
1626         txr->tx_hdr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1627             __VM_CACHELINE_SIZE, __VM_CACHELINE_SIZE, BUS_DMA_WAITOK,
1628             &txr->tx_hdr_dtag, &txr->tx_hdr_dmap, &txr->tx_hdr_paddr);
1629         if (txr->tx_hdr == NULL) {
1630                 device_printf(txr->sc->dev,
1631                     "Unable to allocate TX head write-back buffer\n");
1632                 return ENOMEM;
1633         }
1634
1635         /*
1636          * Create DMA tag for TX buffers
1637          */
1638         error = bus_dma_tag_create(txr->sc->parent_tag,
1639             1, 0,               /* alignment, bounds */
1640             BUS_SPACE_MAXADDR,  /* lowaddr */
1641             BUS_SPACE_MAXADDR,  /* highaddr */
1642             NULL, NULL,         /* filter, filterarg */
1643             IGB_TSO_SIZE,       /* maxsize */
1644             IGB_MAX_SCATTER,    /* nsegments */
1645             PAGE_SIZE,          /* maxsegsize */
1646             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1647             BUS_DMA_ONEBPAGE,   /* flags */
1648             &txr->tx_tag);
1649         if (error) {
1650                 device_printf(txr->sc->dev, "Unable to allocate TX DMA tag\n");
1651                 kfree(txr->tx_buf, M_DEVBUF);
1652                 txr->tx_buf = NULL;
1653                 return error;
1654         }
1655
1656         /*
1657          * Create DMA maps for TX buffers
1658          */
1659         for (i = 0; i < txr->num_tx_desc; ++i) {
1660                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1661
1662                 error = bus_dmamap_create(txr->tx_tag,
1663                     BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE, &txbuf->map);
1664                 if (error) {
1665                         device_printf(txr->sc->dev,
1666                             "Unable to create TX DMA map\n");
1667                         igb_destroy_tx_ring(txr, i);
1668                         return error;
1669                 }
1670         }
1671
1672         /*
1673          * Initialize various watermark
1674          */
1675         txr->spare_desc = IGB_TX_SPARE;
1676         txr->intr_nsegs = txr->num_tx_desc / 16;
1677         txr->oact_hi_desc = txr->num_tx_desc / 2;
1678         txr->oact_lo_desc = txr->num_tx_desc / 8;
1679         if (txr->oact_lo_desc > IGB_TX_OACTIVE_MAX)
1680                 txr->oact_lo_desc = IGB_TX_OACTIVE_MAX;
1681         if (txr->oact_lo_desc < txr->spare_desc + IGB_TX_RESERVED)
1682                 txr->oact_lo_desc = txr->spare_desc + IGB_TX_RESERVED;
1683
1684         return 0;
1685 }
1686
1687 static void
1688 igb_free_tx_ring(struct igb_tx_ring *txr)
1689 {
1690         int i;
1691
1692         for (i = 0; i < txr->num_tx_desc; ++i) {
1693                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1694
1695                 if (txbuf->m_head != NULL) {
1696                         bus_dmamap_unload(txr->tx_tag, txbuf->map);
1697                         m_freem(txbuf->m_head);
1698                         txbuf->m_head = NULL;
1699                 }
1700         }
1701 }
1702
1703 static void
1704 igb_destroy_tx_ring(struct igb_tx_ring *txr, int ndesc)
1705 {
1706         int i;
1707
1708         if (txr->txdma.dma_vaddr != NULL) {
1709                 bus_dmamap_unload(txr->txdma.dma_tag, txr->txdma.dma_map);
1710                 bus_dmamem_free(txr->txdma.dma_tag, txr->txdma.dma_vaddr,
1711                     txr->txdma.dma_map);
1712                 bus_dma_tag_destroy(txr->txdma.dma_tag);
1713                 txr->txdma.dma_vaddr = NULL;
1714         }
1715
1716         if (txr->tx_hdr != NULL) {
1717                 bus_dmamap_unload(txr->tx_hdr_dtag, txr->tx_hdr_dmap);
1718                 bus_dmamem_free(txr->tx_hdr_dtag, txr->tx_hdr,
1719                     txr->tx_hdr_dmap);
1720                 bus_dma_tag_destroy(txr->tx_hdr_dtag);
1721                 txr->tx_hdr = NULL;
1722         }
1723
1724         if (txr->tx_buf == NULL)
1725                 return;
1726
1727         for (i = 0; i < ndesc; ++i) {
1728                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1729
1730                 KKASSERT(txbuf->m_head == NULL);
1731                 bus_dmamap_destroy(txr->tx_tag, txbuf->map);
1732         }
1733         bus_dma_tag_destroy(txr->tx_tag);
1734
1735         kfree(txr->tx_buf, M_DEVBUF);
1736         txr->tx_buf = NULL;
1737 }
1738
1739 static void
1740 igb_init_tx_ring(struct igb_tx_ring *txr)
1741 {
1742         /* Clear the old descriptor contents */
1743         bzero(txr->tx_base,
1744             sizeof(union e1000_adv_tx_desc) * txr->num_tx_desc);
1745
1746         /* Clear TX head write-back buffer */
1747         *(txr->tx_hdr) = 0;
1748
1749         /* Reset indices */
1750         txr->next_avail_desc = 0;
1751         txr->next_to_clean = 0;
1752         txr->tx_nsegs = 0;
1753
1754         /* Set number of descriptors available */
1755         txr->tx_avail = txr->num_tx_desc;
1756 }
1757
1758 static void
1759 igb_init_tx_unit(struct igb_softc *sc)
1760 {
1761         struct e1000_hw *hw = &sc->hw;
1762         uint32_t tctl;
1763         int i;
1764
1765         /* Setup the Tx Descriptor Rings */
1766         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1767                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1768                 uint64_t bus_addr = txr->txdma.dma_paddr;
1769                 uint64_t hdr_paddr = txr->tx_hdr_paddr;
1770                 uint32_t txdctl = 0;
1771                 uint32_t dca_txctrl;
1772
1773                 E1000_WRITE_REG(hw, E1000_TDLEN(i),
1774                     txr->num_tx_desc * sizeof(struct e1000_tx_desc));
1775                 E1000_WRITE_REG(hw, E1000_TDBAH(i),
1776                     (uint32_t)(bus_addr >> 32));
1777                 E1000_WRITE_REG(hw, E1000_TDBAL(i),
1778                     (uint32_t)bus_addr);
1779
1780                 /* Setup the HW Tx Head and Tail descriptor pointers */
1781                 E1000_WRITE_REG(hw, E1000_TDT(i), 0);
1782                 E1000_WRITE_REG(hw, E1000_TDH(i), 0);
1783
1784                 /*
1785                  * WTHRESH is ignored by the hardware, since header
1786                  * write back mode is used.
1787                  */
1788                 txdctl |= IGB_TX_PTHRESH;
1789                 txdctl |= IGB_TX_HTHRESH << 8;
1790                 txdctl |= IGB_TX_WTHRESH << 16;
1791                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1792                 E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
1793
1794                 dca_txctrl = E1000_READ_REG(hw, E1000_DCA_TXCTRL(i));
1795                 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1796                 E1000_WRITE_REG(hw, E1000_DCA_TXCTRL(i), dca_txctrl);
1797
1798                 /*
1799                  * Don't set WB_on_EITR:
1800                  * - 82575 does not have it
1801                  * - It almost has no effect on 82576, see:
1802                  *   82576 specification update errata #26
1803                  * - It causes unnecessary bus traffic
1804                  */
1805                 E1000_WRITE_REG(hw, E1000_TDWBAH(i),
1806                     (uint32_t)(hdr_paddr >> 32));
1807                 E1000_WRITE_REG(hw, E1000_TDWBAL(i),
1808                     ((uint32_t)hdr_paddr) | E1000_TX_HEAD_WB_ENABLE);
1809         }
1810
1811         if (sc->vf_ifp)
1812                 return;
1813
1814         e1000_config_collision_dist(hw);
1815
1816         /* Program the Transmit Control Register */
1817         tctl = E1000_READ_REG(hw, E1000_TCTL);
1818         tctl &= ~E1000_TCTL_CT;
1819         tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
1820             (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
1821
1822         /* This write will effectively turn on the transmit unit. */
1823         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1824 }
1825
1826 static boolean_t
1827 igb_txcsum_ctx(struct igb_tx_ring *txr, struct mbuf *mp)
1828 {
1829         struct e1000_adv_tx_context_desc *TXD;
1830         uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
1831         int ehdrlen, ctxd, ip_hlen = 0;
1832         uint16_t vlantag = 0;
1833         boolean_t offload = TRUE;
1834
1835         if ((mp->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) == 0)
1836                 offload = FALSE;
1837
1838         vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
1839
1840         ctxd = txr->next_avail_desc;
1841         TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
1842
1843         /*
1844          * In advanced descriptors the vlan tag must 
1845          * be placed into the context descriptor, thus
1846          * we need to be here just for that setup.
1847          */
1848         if (mp->m_flags & M_VLANTAG) {
1849                 vlantag = htole16(mp->m_pkthdr.ether_vlantag);
1850                 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
1851         } else if (!offload) {
1852                 return FALSE;
1853         }
1854
1855         ehdrlen = mp->m_pkthdr.csum_lhlen;
1856         KASSERT(ehdrlen > 0, ("invalid ether hlen"));
1857
1858         /* Set the ether header length */
1859         vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
1860
1861         if (mp->m_pkthdr.csum_flags & CSUM_IP) {
1862                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
1863                 ip_hlen = mp->m_pkthdr.csum_iphlen;
1864                 KASSERT(ip_hlen > 0, ("invalid ip hlen"));
1865         }
1866
1867         vlan_macip_lens |= ip_hlen;
1868         type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1869
1870         if (mp->m_pkthdr.csum_flags & CSUM_TCP)
1871                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
1872         else if (mp->m_pkthdr.csum_flags & CSUM_UDP)
1873                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
1874
1875         /* 82575 needs the queue index added */
1876         if (txr->sc->hw.mac.type == e1000_82575)
1877                 mss_l4len_idx = txr->me << 4;
1878
1879         /* Now copy bits into descriptor */
1880         TXD->vlan_macip_lens = htole32(vlan_macip_lens);
1881         TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
1882         TXD->seqnum_seed = htole32(0);
1883         TXD->mss_l4len_idx = htole32(mss_l4len_idx);
1884
1885         /* We've consumed the first desc, adjust counters */
1886         if (++ctxd == txr->num_tx_desc)
1887                 ctxd = 0;
1888         txr->next_avail_desc = ctxd;
1889         --txr->tx_avail;
1890
1891         return offload;
1892 }
1893
1894 static void
1895 igb_txeof(struct igb_tx_ring *txr)
1896 {
1897         struct ifnet *ifp = &txr->sc->arpcom.ac_if;
1898         int first, hdr, avail;
1899
1900         if (txr->tx_avail == txr->num_tx_desc)
1901                 return;
1902
1903         first = txr->next_to_clean;
1904         hdr = *(txr->tx_hdr);
1905
1906         if (first == hdr)
1907                 return;
1908
1909         avail = txr->tx_avail;
1910         while (first != hdr) {
1911                 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
1912
1913                 ++avail;
1914                 if (txbuf->m_head) {
1915                         bus_dmamap_unload(txr->tx_tag, txbuf->map);
1916                         m_freem(txbuf->m_head);
1917                         txbuf->m_head = NULL;
1918                         ++ifp->if_opackets;
1919                 }
1920                 if (++first == txr->num_tx_desc)
1921                         first = 0;
1922         }
1923         txr->next_to_clean = first;
1924         txr->tx_avail = avail;
1925
1926         /*
1927          * If we have a minimum free, clear IFF_OACTIVE
1928          * to tell the stack that it is OK to send packets.
1929          */
1930         if (IGB_IS_NOT_OACTIVE(txr)) {
1931                 ifp->if_flags &= ~IFF_OACTIVE;
1932
1933                 /*
1934                  * We have enough TX descriptors, turn off
1935                  * the watchdog.  We allow small amount of
1936                  * packets (roughly intr_nsegs) pending on
1937                  * the transmit ring.
1938                  */
1939                 ifp->if_timer = 0;
1940         }
1941 }
1942
1943 static int
1944 igb_create_rx_ring(struct igb_rx_ring *rxr)
1945 {
1946         int rsize, i, error;
1947
1948         /*
1949          * Validate number of receive descriptors. It must not exceed
1950          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1951          */
1952         if (((igb_rxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN) != 0 ||
1953             (igb_rxd > IGB_MAX_RXD) || (igb_rxd < IGB_MIN_RXD)) {
1954                 device_printf(rxr->sc->dev,
1955                     "Using %d RX descriptors instead of %d!\n",
1956                     IGB_DEFAULT_RXD, igb_rxd);
1957                 rxr->num_rx_desc = IGB_DEFAULT_RXD;
1958         } else {
1959                 rxr->num_rx_desc = igb_rxd;
1960         }
1961
1962         /*
1963          * Allocate RX descriptor ring
1964          */
1965         rsize = roundup2(rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc),
1966             IGB_DBA_ALIGN);
1967         rxr->rxdma.dma_vaddr = bus_dmamem_coherent_any(rxr->sc->parent_tag,
1968             IGB_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
1969             &rxr->rxdma.dma_tag, &rxr->rxdma.dma_map,
1970             &rxr->rxdma.dma_paddr);
1971         if (rxr->rxdma.dma_vaddr == NULL) {
1972                 device_printf(rxr->sc->dev,
1973                     "Unable to allocate RxDescriptor memory\n");
1974                 return ENOMEM;
1975         }
1976         rxr->rx_base = rxr->rxdma.dma_vaddr;
1977         bzero(rxr->rx_base, rsize);
1978
1979         rxr->rx_buf = kmalloc(sizeof(struct igb_rx_buf) * rxr->num_rx_desc,
1980             M_DEVBUF, M_WAITOK | M_ZERO);
1981
1982         /*
1983          * Create DMA tag for RX buffers
1984          */
1985         error = bus_dma_tag_create(rxr->sc->parent_tag,
1986             1, 0,               /* alignment, bounds */
1987             BUS_SPACE_MAXADDR,  /* lowaddr */
1988             BUS_SPACE_MAXADDR,  /* highaddr */
1989             NULL, NULL,         /* filter, filterarg */
1990             MCLBYTES,           /* maxsize */
1991             1,                  /* nsegments */
1992             MCLBYTES,           /* maxsegsize */
1993             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
1994             &rxr->rx_tag);
1995         if (error) {
1996                 device_printf(rxr->sc->dev,
1997                     "Unable to create RX payload DMA tag\n");
1998                 kfree(rxr->rx_buf, M_DEVBUF);
1999                 rxr->rx_buf = NULL;
2000                 return error;
2001         }
2002
2003         /*
2004          * Create spare DMA map for RX buffers
2005          */
2006         error = bus_dmamap_create(rxr->rx_tag, BUS_DMA_WAITOK,
2007             &rxr->rx_sparemap);
2008         if (error) {
2009                 device_printf(rxr->sc->dev,
2010                     "Unable to create spare RX DMA maps\n");
2011                 bus_dma_tag_destroy(rxr->rx_tag);
2012                 kfree(rxr->rx_buf, M_DEVBUF);
2013                 rxr->rx_buf = NULL;
2014                 return error;
2015         }
2016
2017         /*
2018          * Create DMA maps for RX buffers
2019          */
2020         for (i = 0; i < rxr->num_rx_desc; i++) {
2021                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2022
2023                 error = bus_dmamap_create(rxr->rx_tag,
2024                     BUS_DMA_WAITOK, &rxbuf->map);
2025                 if (error) {
2026                         device_printf(rxr->sc->dev,
2027                             "Unable to create RX DMA maps\n");
2028                         igb_destroy_rx_ring(rxr, i);
2029                         return error;
2030                 }
2031         }
2032         return 0;
2033 }
2034
2035 static void
2036 igb_free_rx_ring(struct igb_rx_ring *rxr)
2037 {
2038         int i;
2039
2040         for (i = 0; i < rxr->num_rx_desc; ++i) {
2041                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2042
2043                 if (rxbuf->m_head != NULL) {
2044                         bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2045                         m_freem(rxbuf->m_head);
2046                         rxbuf->m_head = NULL;
2047                 }
2048         }
2049
2050         if (rxr->fmp != NULL)
2051                 m_freem(rxr->fmp);
2052         rxr->fmp = NULL;
2053         rxr->lmp = NULL;
2054 }
2055
2056 static void
2057 igb_destroy_rx_ring(struct igb_rx_ring *rxr, int ndesc)
2058 {
2059         int i;
2060
2061         if (rxr->rxdma.dma_vaddr != NULL) {
2062                 bus_dmamap_unload(rxr->rxdma.dma_tag, rxr->rxdma.dma_map);
2063                 bus_dmamem_free(rxr->rxdma.dma_tag, rxr->rxdma.dma_vaddr,
2064                     rxr->rxdma.dma_map);
2065                 bus_dma_tag_destroy(rxr->rxdma.dma_tag);
2066                 rxr->rxdma.dma_vaddr = NULL;
2067         }
2068
2069         if (rxr->rx_buf == NULL)
2070                 return;
2071
2072         for (i = 0; i < ndesc; ++i) {
2073                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2074
2075                 KKASSERT(rxbuf->m_head == NULL);
2076                 bus_dmamap_destroy(rxr->rx_tag, rxbuf->map);
2077         }
2078         bus_dmamap_destroy(rxr->rx_tag, rxr->rx_sparemap);
2079         bus_dma_tag_destroy(rxr->rx_tag);
2080
2081         kfree(rxr->rx_buf, M_DEVBUF);
2082         rxr->rx_buf = NULL;
2083 }
2084
2085 static void
2086 igb_setup_rxdesc(union e1000_adv_rx_desc *rxd, const struct igb_rx_buf *rxbuf)
2087 {
2088         rxd->read.pkt_addr = htole64(rxbuf->paddr);
2089         rxd->wb.upper.status_error = 0;
2090 }
2091
2092 static int
2093 igb_newbuf(struct igb_rx_ring *rxr, int i, boolean_t wait)
2094 {
2095         struct mbuf *m;
2096         bus_dma_segment_t seg;
2097         bus_dmamap_t map;
2098         struct igb_rx_buf *rxbuf;
2099         int error, nseg;
2100
2101         m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2102         if (m == NULL) {
2103                 if (wait) {
2104                         if_printf(&rxr->sc->arpcom.ac_if,
2105                             "Unable to allocate RX mbuf\n");
2106                 }
2107                 return ENOBUFS;
2108         }
2109         m->m_len = m->m_pkthdr.len = MCLBYTES;
2110
2111         if (rxr->sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2112                 m_adj(m, ETHER_ALIGN);
2113
2114         error = bus_dmamap_load_mbuf_segment(rxr->rx_tag,
2115             rxr->rx_sparemap, m, &seg, 1, &nseg, BUS_DMA_NOWAIT);
2116         if (error) {
2117                 m_freem(m);
2118                 if (wait) {
2119                         if_printf(&rxr->sc->arpcom.ac_if,
2120                             "Unable to load RX mbuf\n");
2121                 }
2122                 return error;
2123         }
2124
2125         rxbuf = &rxr->rx_buf[i];
2126         if (rxbuf->m_head != NULL)
2127                 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2128
2129         map = rxbuf->map;
2130         rxbuf->map = rxr->rx_sparemap;
2131         rxr->rx_sparemap = map;
2132
2133         rxbuf->m_head = m;
2134         rxbuf->paddr = seg.ds_addr;
2135
2136         igb_setup_rxdesc(&rxr->rx_base[i], rxbuf);
2137         return 0;
2138 }
2139
2140 static int
2141 igb_init_rx_ring(struct igb_rx_ring *rxr)
2142 {
2143         int i;
2144
2145         /* Clear the ring contents */
2146         bzero(rxr->rx_base,
2147             rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc));
2148
2149         /* Now replenish the ring mbufs */
2150         for (i = 0; i < rxr->num_rx_desc; ++i) {
2151                 int error;
2152
2153                 error = igb_newbuf(rxr, i, TRUE);
2154                 if (error)
2155                         return error;
2156         }
2157
2158         /* Setup our descriptor indices */
2159         rxr->next_to_check = 0;
2160
2161         rxr->fmp = NULL;
2162         rxr->lmp = NULL;
2163         rxr->discard = FALSE;
2164
2165         return 0;
2166 }
2167
2168 static void
2169 igb_init_rx_unit(struct igb_softc *sc)
2170 {
2171         struct ifnet *ifp = &sc->arpcom.ac_if;
2172         struct e1000_hw *hw = &sc->hw;
2173         uint32_t rctl, rxcsum, srrctl = 0;
2174         int i;
2175
2176         /*
2177          * Make sure receives are disabled while setting
2178          * up the descriptor ring
2179          */
2180         rctl = E1000_READ_REG(hw, E1000_RCTL);
2181         E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2182
2183 #if 0
2184         /*
2185         ** Set up for header split
2186         */
2187         if (igb_header_split) {
2188                 /* Use a standard mbuf for the header */
2189                 srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2190                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2191         } else
2192 #endif
2193                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2194
2195         /*
2196         ** Set up for jumbo frames
2197         */
2198         if (ifp->if_mtu > ETHERMTU) {
2199                 rctl |= E1000_RCTL_LPE;
2200 #if 0
2201                 if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
2202                         srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2203                         rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
2204                 } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
2205                         srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2206                         rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
2207                 }
2208                 /* Set maximum packet len */
2209                 psize = adapter->max_frame_size;
2210                 /* are we on a vlan? */
2211                 if (adapter->ifp->if_vlantrunk != NULL)
2212                         psize += VLAN_TAG_SIZE;
2213                 E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
2214 #else
2215                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2216                 rctl |= E1000_RCTL_SZ_2048;
2217 #endif
2218         } else {
2219                 rctl &= ~E1000_RCTL_LPE;
2220                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2221                 rctl |= E1000_RCTL_SZ_2048;
2222         }
2223
2224         /* Setup the Base and Length of the Rx Descriptor Rings */
2225         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2226                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2227                 uint64_t bus_addr = rxr->rxdma.dma_paddr;
2228                 uint32_t rxdctl;
2229
2230                 E1000_WRITE_REG(hw, E1000_RDLEN(i),
2231                     rxr->num_rx_desc * sizeof(struct e1000_rx_desc));
2232                 E1000_WRITE_REG(hw, E1000_RDBAH(i),
2233                     (uint32_t)(bus_addr >> 32));
2234                 E1000_WRITE_REG(hw, E1000_RDBAL(i),
2235                     (uint32_t)bus_addr);
2236                 E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
2237                 /* Enable this Queue */
2238                 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
2239                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2240                 rxdctl &= 0xFFF00000;
2241                 rxdctl |= IGB_RX_PTHRESH;
2242                 rxdctl |= IGB_RX_HTHRESH << 8;
2243                 /*
2244                  * Don't set WTHRESH to a value above 1 on 82576, see:
2245                  * 82576 specification update errata #26
2246                  */
2247                 rxdctl |= IGB_RX_WTHRESH << 16;
2248                 E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
2249         }
2250
2251         rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2252         rxcsum &= ~(E1000_RXCSUM_PCSS_MASK | E1000_RXCSUM_IPPCSE);
2253
2254         /*
2255          * Receive Checksum Offload for TCP and UDP
2256          *
2257          * Checksum offloading is also enabled if multiple receive
2258          * queue is to be supported, since we need it to figure out
2259          * fragments.
2260          */
2261         if ((ifp->if_capenable & IFCAP_RXCSUM) || IGB_ENABLE_HWRSS(sc)) {
2262                 /*
2263                  * NOTE:
2264                  * PCSD must be enabled to enable multiple
2265                  * receive queues.
2266                  */
2267                 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2268                     E1000_RXCSUM_PCSD;
2269         } else {
2270                 rxcsum &= ~(E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2271                     E1000_RXCSUM_PCSD);
2272         }
2273         E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2274
2275         if (IGB_ENABLE_HWRSS(sc)) {
2276                 uint8_t key[IGB_NRSSRK * IGB_RSSRK_SIZE];
2277                 uint32_t reta_shift;
2278                 int j, r;
2279
2280                 /*
2281                  * NOTE:
2282                  * When we reach here, RSS has already been disabled
2283                  * in igb_stop(), so we could safely configure RSS key
2284                  * and redirect table.
2285                  */
2286
2287                 /*
2288                  * Configure RSS key
2289                  */
2290                 toeplitz_get_key(key, sizeof(key));
2291                 for (i = 0; i < IGB_NRSSRK; ++i) {
2292                         uint32_t rssrk;
2293
2294                         rssrk = IGB_RSSRK_VAL(key, i);
2295                         IGB_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2296
2297                         E1000_WRITE_REG(hw, E1000_RSSRK(i), rssrk);
2298                 }
2299
2300                 /*
2301                  * Configure RSS redirect table in following fashion:
2302                  * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
2303                  */
2304                 reta_shift = IGB_RETA_SHIFT;
2305                 if (hw->mac.type == e1000_82575)
2306                         reta_shift = IGB_RETA_SHIFT_82575;
2307
2308                 r = 0;
2309                 for (j = 0; j < IGB_NRETA; ++j) {
2310                         uint32_t reta = 0;
2311
2312                         for (i = 0; i < IGB_RETA_SIZE; ++i) {
2313                                 uint32_t q;
2314
2315                                 q = (r % sc->rx_ring_inuse) << reta_shift;
2316                                 reta |= q << (8 * i);
2317                                 ++r;
2318                         }
2319                         IGB_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2320                         E1000_WRITE_REG(hw, E1000_RETA(j), reta);
2321                 }
2322
2323                 /*
2324                  * Enable multiple receive queues.
2325                  * Enable IPv4 RSS standard hash functions.
2326                  * Disable RSS interrupt on 82575
2327                  */
2328                 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2329                                 E1000_MRQC_ENABLE_RSS_4Q |
2330                                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2331                                 E1000_MRQC_RSS_FIELD_IPV4);
2332         }
2333
2334         /* Setup the Receive Control Register */
2335         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2336         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2337             E1000_RCTL_RDMTS_HALF |
2338             (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2339         /* Strip CRC bytes. */
2340         rctl |= E1000_RCTL_SECRC;
2341         /* Make sure VLAN Filters are off */
2342         rctl &= ~E1000_RCTL_VFE;
2343         /* Don't store bad packets */
2344         rctl &= ~E1000_RCTL_SBP;
2345
2346         /* Enable Receives */
2347         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2348
2349         /*
2350          * Setup the HW Rx Head and Tail Descriptor Pointers
2351          *   - needs to be after enable
2352          */
2353         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2354                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2355
2356                 E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
2357                 E1000_WRITE_REG(hw, E1000_RDT(i), rxr->num_rx_desc - 1);
2358         }
2359 }
2360
2361 static void
2362 igb_rxeof(struct igb_rx_ring *rxr, int count)
2363 {
2364         struct ifnet *ifp = &rxr->sc->arpcom.ac_if;
2365         union e1000_adv_rx_desc *cur;
2366         uint32_t staterr;
2367         int i;
2368
2369         i = rxr->next_to_check;
2370         cur = &rxr->rx_base[i];
2371         staterr = le32toh(cur->wb.upper.status_error);
2372
2373         if ((staterr & E1000_RXD_STAT_DD) == 0)
2374                 return;
2375
2376         while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
2377                 struct pktinfo *pi = NULL, pi0;
2378                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2379                 struct mbuf *m = NULL;
2380                 boolean_t eop;
2381
2382                 eop = (staterr & E1000_RXD_STAT_EOP) ? TRUE : FALSE;
2383                 if (eop)
2384                         --count;
2385
2386                 if ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) == 0 &&
2387                     !rxr->discard) {
2388                         struct mbuf *mp = rxbuf->m_head;
2389                         uint32_t hash, hashtype;
2390                         uint16_t vlan;
2391                         int len;
2392
2393                         len = le16toh(cur->wb.upper.length);
2394                         if (rxr->sc->hw.mac.type == e1000_i350 &&
2395                             (staterr & E1000_RXDEXT_STATERR_LB))
2396                                 vlan = be16toh(cur->wb.upper.vlan);
2397                         else
2398                                 vlan = le16toh(cur->wb.upper.vlan);
2399
2400                         hash = le32toh(cur->wb.lower.hi_dword.rss);
2401                         hashtype = le32toh(cur->wb.lower.lo_dword.data) &
2402                             E1000_RXDADV_RSSTYPE_MASK;
2403
2404                         IGB_RSS_DPRINTF(rxr->sc, 10,
2405                             "ring%d, hash 0x%08x, hashtype %u\n",
2406                             rxr->me, hash, hashtype);
2407
2408                         bus_dmamap_sync(rxr->rx_tag, rxbuf->map,
2409                             BUS_DMASYNC_POSTREAD);
2410
2411                         if (igb_newbuf(rxr, i, FALSE) != 0) {
2412                                 ifp->if_iqdrops++;
2413                                 goto discard;
2414                         }
2415
2416                         mp->m_len = len;
2417                         if (rxr->fmp == NULL) {
2418                                 mp->m_pkthdr.len = len;
2419                                 rxr->fmp = mp;
2420                                 rxr->lmp = mp;
2421                         } else {
2422                                 rxr->lmp->m_next = mp;
2423                                 rxr->lmp = rxr->lmp->m_next;
2424                                 rxr->fmp->m_pkthdr.len += len;
2425                         }
2426
2427                         if (eop) {
2428                                 m = rxr->fmp;
2429                                 rxr->fmp = NULL;
2430                                 rxr->lmp = NULL;
2431
2432                                 m->m_pkthdr.rcvif = ifp;
2433                                 ifp->if_ipackets++;
2434
2435                                 if (ifp->if_capenable & IFCAP_RXCSUM)
2436                                         igb_rxcsum(staterr, m);
2437
2438                                 if (staterr & E1000_RXD_STAT_VP) {
2439                                         m->m_pkthdr.ether_vlantag = vlan;
2440                                         m->m_flags |= M_VLANTAG;
2441                                 }
2442
2443                                 if (ifp->if_capenable & IFCAP_RSS) {
2444                                         pi = igb_rssinfo(m, &pi0,
2445                                             hash, hashtype, staterr);
2446                                 }
2447 #ifdef IGB_RSS_DEBUG
2448                                 rxr->rx_packets++;
2449 #endif
2450                         }
2451                 } else {
2452                         ifp->if_ierrors++;
2453 discard:
2454                         igb_setup_rxdesc(cur, rxbuf);
2455                         if (!eop)
2456                                 rxr->discard = TRUE;
2457                         else
2458                                 rxr->discard = FALSE;
2459                         if (rxr->fmp != NULL) {
2460                                 m_freem(rxr->fmp);
2461                                 rxr->fmp = NULL;
2462                                 rxr->lmp = NULL;
2463                         }
2464                         m = NULL;
2465                 }
2466
2467                 if (m != NULL)
2468                         ether_input_pkt(ifp, m, pi);
2469
2470                 /* Advance our pointers to the next descriptor. */
2471                 if (++i == rxr->num_rx_desc)
2472                         i = 0;
2473
2474                 cur = &rxr->rx_base[i];
2475                 staterr = le32toh(cur->wb.upper.status_error);
2476         }
2477         rxr->next_to_check = i;
2478
2479         if (--i < 0)
2480                 i = rxr->num_rx_desc - 1;
2481         E1000_WRITE_REG(&rxr->sc->hw, E1000_RDT(rxr->me), i);
2482 }
2483
2484
2485 static void
2486 igb_set_vlan(struct igb_softc *sc)
2487 {
2488         struct e1000_hw *hw = &sc->hw;
2489         uint32_t reg;
2490 #if 0
2491         struct ifnet *ifp = sc->arpcom.ac_if;
2492 #endif
2493
2494         if (sc->vf_ifp) {
2495                 e1000_rlpml_set_vf(hw, sc->max_frame_size + VLAN_TAG_SIZE);
2496                 return;
2497         }
2498
2499         reg = E1000_READ_REG(hw, E1000_CTRL);
2500         reg |= E1000_CTRL_VME;
2501         E1000_WRITE_REG(hw, E1000_CTRL, reg);
2502
2503 #if 0
2504         /* Enable the Filter Table */
2505         if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2506                 reg = E1000_READ_REG(hw, E1000_RCTL);
2507                 reg &= ~E1000_RCTL_CFIEN;
2508                 reg |= E1000_RCTL_VFE;
2509                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
2510         }
2511 #endif
2512
2513         /* Update the frame size */
2514         E1000_WRITE_REG(&sc->hw, E1000_RLPML,
2515             sc->max_frame_size + VLAN_TAG_SIZE);
2516
2517 #if 0
2518         /* Don't bother with table if no vlans */
2519         if ((adapter->num_vlans == 0) ||
2520             ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
2521                 return;
2522         /*
2523         ** A soft reset zero's out the VFTA, so
2524         ** we need to repopulate it now.
2525         */
2526         for (int i = 0; i < IGB_VFTA_SIZE; i++)
2527                 if (adapter->shadow_vfta[i] != 0) {
2528                         if (adapter->vf_ifp)
2529                                 e1000_vfta_set_vf(hw,
2530                                     adapter->shadow_vfta[i], TRUE);
2531                         else
2532                                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
2533                                  i, adapter->shadow_vfta[i]);
2534                 }
2535 #endif
2536 }
2537
2538 static void
2539 igb_enable_intr(struct igb_softc *sc)
2540 {
2541         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2542                 lwkt_serialize_handler_enable(&sc->main_serialize);
2543         } else {
2544                 int i;
2545
2546                 for (i = 0; i < sc->msix_cnt; ++i) {
2547                         lwkt_serialize_handler_enable(
2548                             sc->msix_data[i].msix_serialize);
2549                 }
2550         }
2551
2552         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2553                 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
2554                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, sc->intr_mask);
2555                 else
2556                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2557                 E1000_WRITE_REG(&sc->hw, E1000_EIAM, sc->intr_mask);
2558                 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2559                 E1000_WRITE_REG(&sc->hw, E1000_IMS, E1000_IMS_LSC);
2560         } else {
2561                 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
2562         }
2563         E1000_WRITE_FLUSH(&sc->hw);
2564 }
2565
2566 static void
2567 igb_disable_intr(struct igb_softc *sc)
2568 {
2569         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2570                 E1000_WRITE_REG(&sc->hw, E1000_EIMC, 0xffffffff);
2571                 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2572         }
2573         E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2574         E1000_WRITE_FLUSH(&sc->hw);
2575
2576         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2577                 lwkt_serialize_handler_disable(&sc->main_serialize);
2578         } else {
2579                 int i;
2580
2581                 for (i = 0; i < sc->msix_cnt; ++i) {
2582                         lwkt_serialize_handler_disable(
2583                             sc->msix_data[i].msix_serialize);
2584                 }
2585         }
2586 }
2587
2588 /*
2589  * Bit of a misnomer, what this really means is
2590  * to enable OS management of the system... aka
2591  * to disable special hardware management features 
2592  */
2593 static void
2594 igb_get_mgmt(struct igb_softc *sc)
2595 {
2596         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2597                 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2598                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2599
2600                 /* disable hardware interception of ARP */
2601                 manc &= ~E1000_MANC_ARP_EN;
2602
2603                 /* enable receiving management packets to the host */
2604                 manc |= E1000_MANC_EN_MNG2HOST;
2605                 manc2h |= 1 << 5; /* Mng Port 623 */
2606                 manc2h |= 1 << 6; /* Mng Port 664 */
2607                 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
2608                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2609         }
2610 }
2611
2612 /*
2613  * Give control back to hardware management controller
2614  * if there is one.
2615  */
2616 static void
2617 igb_rel_mgmt(struct igb_softc *sc)
2618 {
2619         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2620                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2621
2622                 /* Re-enable hardware interception of ARP */
2623                 manc |= E1000_MANC_ARP_EN;
2624                 manc &= ~E1000_MANC_EN_MNG2HOST;
2625
2626                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2627         }
2628 }
2629
2630 /*
2631  * Sets CTRL_EXT:DRV_LOAD bit.
2632  *
2633  * For ASF and Pass Through versions of f/w this means that
2634  * the driver is loaded. 
2635  */
2636 static void
2637 igb_get_hw_control(struct igb_softc *sc)
2638 {
2639         uint32_t ctrl_ext;
2640
2641         if (sc->vf_ifp)
2642                 return;
2643
2644         /* Let firmware know the driver has taken over */
2645         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2646         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2647             ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2648 }
2649
2650 /*
2651  * Resets CTRL_EXT:DRV_LOAD bit.
2652  *
2653  * For ASF and Pass Through versions of f/w this means that the
2654  * driver is no longer loaded.
2655  */
2656 static void
2657 igb_rel_hw_control(struct igb_softc *sc)
2658 {
2659         uint32_t ctrl_ext;
2660
2661         if (sc->vf_ifp)
2662                 return;
2663
2664         /* Let firmware taken over control of h/w */
2665         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2666         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2667             ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2668 }
2669
2670 static int
2671 igb_is_valid_ether_addr(const uint8_t *addr)
2672 {
2673         uint8_t zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
2674
2675         if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
2676                 return FALSE;
2677         return TRUE;
2678 }
2679
2680 /*
2681  * Enable PCI Wake On Lan capability
2682  */
2683 static void
2684 igb_enable_wol(device_t dev)
2685 {
2686         uint16_t cap, status;
2687         uint8_t id;
2688
2689         /* First find the capabilities pointer*/
2690         cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
2691
2692         /* Read the PM Capabilities */
2693         id = pci_read_config(dev, cap, 1);
2694         if (id != PCIY_PMG)     /* Something wrong */
2695                 return;
2696
2697         /*
2698          * OK, we have the power capabilities,
2699          * so now get the status register
2700          */
2701         cap += PCIR_POWER_STATUS;
2702         status = pci_read_config(dev, cap, 2);
2703         status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2704         pci_write_config(dev, cap, status, 2);
2705 }
2706
2707 static void
2708 igb_update_stats_counters(struct igb_softc *sc)
2709 {
2710         struct e1000_hw *hw = &sc->hw;
2711         struct e1000_hw_stats *stats;
2712         struct ifnet *ifp = &sc->arpcom.ac_if;
2713
2714         /* 
2715          * The virtual function adapter has only a
2716          * small controlled set of stats, do only 
2717          * those and return.
2718          */
2719         if (sc->vf_ifp) {
2720                 igb_update_vf_stats_counters(sc);
2721                 return;
2722         }
2723         stats = sc->stats;
2724
2725         if (sc->hw.phy.media_type == e1000_media_type_copper ||
2726             (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
2727                 stats->symerrs +=
2728                     E1000_READ_REG(hw,E1000_SYMERRS);
2729                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
2730         }
2731
2732         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
2733         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
2734         stats->scc += E1000_READ_REG(hw, E1000_SCC);
2735         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
2736
2737         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
2738         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
2739         stats->colc += E1000_READ_REG(hw, E1000_COLC);
2740         stats->dc += E1000_READ_REG(hw, E1000_DC);
2741         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
2742         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
2743         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
2744
2745         /*
2746          * For watchdog management we need to know if we have been
2747          * paused during the last interval, so capture that here.
2748          */ 
2749         sc->pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
2750         stats->xoffrxc += sc->pause_frames;
2751         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
2752         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
2753         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
2754         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
2755         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
2756         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
2757         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
2758         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
2759         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
2760         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
2761         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
2762         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
2763
2764         /* For the 64-bit byte counters the low dword must be read first. */
2765         /* Both registers clear on the read of the high dword */
2766
2767         stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
2768             ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
2769         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
2770             ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
2771
2772         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
2773         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
2774         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
2775         stats->roc += E1000_READ_REG(hw, E1000_ROC);
2776         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
2777
2778         stats->tor += E1000_READ_REG(hw, E1000_TORH);
2779         stats->tot += E1000_READ_REG(hw, E1000_TOTH);
2780
2781         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
2782         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
2783         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
2784         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
2785         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
2786         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
2787         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
2788         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
2789         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
2790         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
2791
2792         /* Interrupt Counts */
2793
2794         stats->iac += E1000_READ_REG(hw, E1000_IAC);
2795         stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
2796         stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
2797         stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
2798         stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
2799         stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
2800         stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
2801         stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
2802         stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
2803
2804         /* Host to Card Statistics */
2805
2806         stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
2807         stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
2808         stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
2809         stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
2810         stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
2811         stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
2812         stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
2813         stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
2814             ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32));
2815         stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
2816             ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
2817         stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
2818         stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
2819         stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
2820
2821         stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
2822         stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
2823         stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
2824         stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
2825         stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
2826         stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
2827
2828         ifp->if_collisions = stats->colc;
2829
2830         /* Rx Errors */
2831         ifp->if_ierrors = stats->rxerrc + stats->crcerrs + stats->algnerrc +
2832             stats->ruc + stats->roc + stats->mpc + stats->cexterr;
2833
2834         /* Tx Errors */
2835         ifp->if_oerrors = stats->ecol + stats->latecol + sc->watchdog_events;
2836
2837         /* Driver specific counters */
2838         sc->device_control = E1000_READ_REG(hw, E1000_CTRL);
2839         sc->rx_control = E1000_READ_REG(hw, E1000_RCTL);
2840         sc->int_mask = E1000_READ_REG(hw, E1000_IMS);
2841         sc->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
2842         sc->packet_buf_alloc_tx =
2843             ((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
2844         sc->packet_buf_alloc_rx =
2845             (E1000_READ_REG(hw, E1000_PBA) & 0xffff);
2846 }
2847
2848 static void
2849 igb_vf_init_stats(struct igb_softc *sc)
2850 {
2851         struct e1000_hw *hw = &sc->hw;
2852         struct e1000_vf_stats *stats;
2853
2854         stats = sc->stats;
2855         stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
2856         stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
2857         stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
2858         stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
2859         stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
2860 }
2861  
2862 static void
2863 igb_update_vf_stats_counters(struct igb_softc *sc)
2864 {
2865         struct e1000_hw *hw = &sc->hw;
2866         struct e1000_vf_stats *stats;
2867
2868         if (sc->link_speed == 0)
2869                 return;
2870
2871         stats = sc->stats;
2872         UPDATE_VF_REG(E1000_VFGPRC, stats->last_gprc, stats->gprc);
2873         UPDATE_VF_REG(E1000_VFGORC, stats->last_gorc, stats->gorc);
2874         UPDATE_VF_REG(E1000_VFGPTC, stats->last_gptc, stats->gptc);
2875         UPDATE_VF_REG(E1000_VFGOTC, stats->last_gotc, stats->gotc);
2876         UPDATE_VF_REG(E1000_VFMPRC, stats->last_mprc, stats->mprc);
2877 }
2878
2879 #ifdef DEVICE_POLLING
2880
2881 static void
2882 igb_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2883 {
2884         struct igb_softc *sc = ifp->if_softc;
2885         uint32_t reg_icr;
2886
2887         switch (cmd) {
2888         case POLL_REGISTER:
2889         case POLL_DEREGISTER:
2890                 ASSERT_IFNET_SERIALIZED_ALL(ifp);
2891                 igb_init(sc);
2892                 break;
2893
2894         case POLL_AND_CHECK_STATUS:
2895                 ASSERT_SERIALIZED(&sc->main_serialize);
2896                 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2897                 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
2898                         sc->hw.mac.get_link_status = 1;
2899                         igb_update_link_status(sc);
2900                 }
2901                 /* FALL THROUGH */
2902         case POLL_ONLY:
2903                 ASSERT_SERIALIZED(&sc->main_serialize);
2904                 if (ifp->if_flags & IFF_RUNNING) {
2905                         struct igb_tx_ring *txr;
2906                         int i;
2907
2908                         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2909                                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2910
2911                                 lwkt_serialize_enter(&rxr->rx_serialize);
2912                                 igb_rxeof(rxr, count);
2913                                 lwkt_serialize_exit(&rxr->rx_serialize);
2914                         }
2915
2916                         txr = &sc->tx_rings[0];
2917                         lwkt_serialize_enter(&txr->tx_serialize);
2918                         igb_txeof(txr);
2919                         if (!ifq_is_empty(&ifp->if_snd))
2920                                 if_devstart(ifp);
2921                         lwkt_serialize_exit(&txr->tx_serialize);
2922                 }
2923                 break;
2924         }
2925 }
2926
2927 #endif /* DEVICE_POLLING */
2928
2929 static void
2930 igb_intr(void *xsc)
2931 {
2932         struct igb_softc *sc = xsc;
2933         struct ifnet *ifp = &sc->arpcom.ac_if;
2934         uint32_t eicr;
2935
2936         ASSERT_SERIALIZED(&sc->main_serialize);
2937
2938         eicr = E1000_READ_REG(&sc->hw, E1000_EICR);
2939
2940         if (eicr == 0)
2941                 return;
2942
2943         if (ifp->if_flags & IFF_RUNNING) {
2944                 struct igb_tx_ring *txr;
2945                 int i;
2946
2947                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
2948                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
2949
2950                         if (eicr & rxr->rx_intr_mask) {
2951                                 lwkt_serialize_enter(&rxr->rx_serialize);
2952                                 igb_rxeof(rxr, -1);
2953                                 lwkt_serialize_exit(&rxr->rx_serialize);
2954                         }
2955                 }
2956
2957                 txr = &sc->tx_rings[0];
2958                 if (eicr & txr->tx_intr_mask) {
2959                         lwkt_serialize_enter(&txr->tx_serialize);
2960                         igb_txeof(txr);
2961                         if (!ifq_is_empty(&ifp->if_snd))
2962                                 if_devstart(ifp);
2963                         lwkt_serialize_exit(&txr->tx_serialize);
2964                 }
2965         }
2966
2967         if (eicr & E1000_EICR_OTHER) {
2968                 uint32_t icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2969
2970                 /* Link status change */
2971                 if (icr & E1000_ICR_LSC) {
2972                         sc->hw.mac.get_link_status = 1;
2973                         igb_update_link_status(sc);
2974                 }
2975         }
2976
2977         /*
2978          * Reading EICR has the side effect to clear interrupt mask,
2979          * so all interrupts need to be enabled here.
2980          */
2981         E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2982 }
2983
2984 static void
2985 igb_intr_shared(void *xsc)
2986 {
2987         struct igb_softc *sc = xsc;
2988         struct ifnet *ifp = &sc->arpcom.ac_if;
2989         uint32_t reg_icr;
2990
2991         ASSERT_SERIALIZED(&sc->main_serialize);
2992
2993         reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2994
2995         /* Hot eject?  */
2996         if (reg_icr == 0xffffffff)
2997                 return;
2998
2999         /* Definitely not our interrupt.  */
3000         if (reg_icr == 0x0)
3001                 return;
3002
3003         if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
3004                 return;
3005
3006         if (ifp->if_flags & IFF_RUNNING) {
3007                 if (reg_icr &
3008                     (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
3009                         int i;
3010
3011                         for (i = 0; i < sc->rx_ring_inuse; ++i) {
3012                                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3013
3014                                 lwkt_serialize_enter(&rxr->rx_serialize);
3015                                 igb_rxeof(rxr, -1);
3016                                 lwkt_serialize_exit(&rxr->rx_serialize);
3017                         }
3018                 }
3019
3020                 if (reg_icr & E1000_ICR_TXDW) {
3021                         struct igb_tx_ring *txr = &sc->tx_rings[0];
3022
3023                         lwkt_serialize_enter(&txr->tx_serialize);
3024                         igb_txeof(txr);
3025                         if (!ifq_is_empty(&ifp->if_snd))
3026                                 if_devstart(ifp);
3027                         lwkt_serialize_exit(&txr->tx_serialize);
3028                 }
3029         }
3030
3031         /* Link status change */
3032         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3033                 sc->hw.mac.get_link_status = 1;
3034                 igb_update_link_status(sc);
3035         }
3036
3037         if (reg_icr & E1000_ICR_RXO)
3038                 sc->rx_overruns++;
3039 }
3040
3041 static int
3042 igb_encap(struct igb_tx_ring *txr, struct mbuf **m_headp)
3043 {
3044         bus_dma_segment_t segs[IGB_MAX_SCATTER];
3045         bus_dmamap_t map;
3046         struct igb_tx_buf *tx_buf, *tx_buf_mapped;
3047         union e1000_adv_tx_desc *txd = NULL;
3048         struct mbuf *m_head = *m_headp;
3049         uint32_t olinfo_status = 0, cmd_type_len = 0, cmd_rs = 0;
3050         int maxsegs, nsegs, i, j, error, last = 0;
3051         uint32_t hdrlen = 0;
3052
3053         /* Set basic descriptor constants */
3054         cmd_type_len |= E1000_ADVTXD_DTYP_DATA;
3055         cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
3056         if (m_head->m_flags & M_VLANTAG)
3057                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3058
3059         /*
3060          * Map the packet for DMA.
3061          */
3062         tx_buf = &txr->tx_buf[txr->next_avail_desc];
3063         tx_buf_mapped = tx_buf;
3064         map = tx_buf->map;
3065
3066         maxsegs = txr->tx_avail - IGB_TX_RESERVED;
3067         KASSERT(maxsegs >= txr->spare_desc, ("not enough spare TX desc\n"));
3068         if (maxsegs > IGB_MAX_SCATTER)
3069                 maxsegs = IGB_MAX_SCATTER;
3070
3071         error = bus_dmamap_load_mbuf_defrag(txr->tx_tag, map, m_headp,
3072             segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3073         if (error) {
3074                 if (error == ENOBUFS)
3075                         txr->sc->mbuf_defrag_failed++;
3076                 else
3077                         txr->sc->no_tx_dma_setup++;
3078
3079                 m_freem(*m_headp);
3080                 *m_headp = NULL;
3081                 return error;
3082         }
3083         bus_dmamap_sync(txr->tx_tag, map, BUS_DMASYNC_PREWRITE);
3084
3085         m_head = *m_headp;
3086
3087 #if 0
3088         /*
3089          * Set up the context descriptor:
3090          * used when any hardware offload is done.
3091          * This includes CSUM, VLAN, and TSO. It
3092          * will use the first descriptor.
3093          */
3094         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3095                 if (igb_tso_setup(txr, m_head, &hdrlen)) {
3096                         cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3097                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3098                         olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3099                 } else
3100                         return (ENXIO); 
3101         } else if (igb_tx_ctx_setup(txr, m_head))
3102                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3103 #else
3104         if (igb_txcsum_ctx(txr, m_head)) {
3105                 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3106                         olinfo_status |= (E1000_TXD_POPTS_IXSM << 8);
3107                 if (m_head->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_TCP))
3108                         olinfo_status |= (E1000_TXD_POPTS_TXSM << 8);
3109                 txr->tx_nsegs++;
3110         }
3111 #endif
3112
3113         txr->tx_nsegs += nsegs;
3114         if (txr->tx_nsegs >= txr->intr_nsegs) {
3115                 /*
3116                  * Report Status (RS) is turned on every intr_nsegs
3117                  * descriptors (roughly).
3118                  */
3119                 txr->tx_nsegs = 0;
3120                 cmd_rs = E1000_ADVTXD_DCMD_RS;
3121         }
3122
3123         /* Calculate payload length */
3124         olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
3125             << E1000_ADVTXD_PAYLEN_SHIFT);
3126
3127         /* 82575 needs the queue index added */
3128         if (txr->sc->hw.mac.type == e1000_82575)
3129                 olinfo_status |= txr->me << 4;
3130
3131         /* Set up our transmit descriptors */
3132         i = txr->next_avail_desc;
3133         for (j = 0; j < nsegs; j++) {
3134                 bus_size_t seg_len;
3135                 bus_addr_t seg_addr;
3136
3137                 tx_buf = &txr->tx_buf[i];
3138                 txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
3139                 seg_addr = segs[j].ds_addr;
3140                 seg_len = segs[j].ds_len;
3141
3142                 txd->read.buffer_addr = htole64(seg_addr);
3143                 txd->read.cmd_type_len = htole32(cmd_type_len | seg_len);
3144                 txd->read.olinfo_status = htole32(olinfo_status);
3145                 last = i;
3146                 if (++i == txr->num_tx_desc)
3147                         i = 0;
3148                 tx_buf->m_head = NULL;
3149         }
3150
3151         KASSERT(txr->tx_avail > nsegs, ("invalid avail TX desc\n"));
3152         txr->next_avail_desc = i;
3153         txr->tx_avail -= nsegs;
3154
3155         tx_buf->m_head = m_head;
3156         tx_buf_mapped->map = tx_buf->map;
3157         tx_buf->map = map;
3158
3159         /*
3160          * Last Descriptor of Packet needs End Of Packet (EOP)
3161          */
3162         txd->read.cmd_type_len |= htole32(E1000_ADVTXD_DCMD_EOP | cmd_rs);
3163
3164         /*
3165          * Advance the Transmit Descriptor Tail (TDT), this tells the E1000
3166          * that this frame is available to transmit.
3167          */
3168         E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), i);
3169         ++txr->tx_packets;
3170
3171         return 0;
3172 }
3173
3174 static void
3175 igb_start(struct ifnet *ifp)
3176 {
3177         struct igb_softc *sc = ifp->if_softc;
3178         struct igb_tx_ring *txr = &sc->tx_rings[0];
3179         struct mbuf *m_head;
3180
3181         ASSERT_SERIALIZED(&txr->tx_serialize);
3182
3183         if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
3184                 return;
3185
3186         if (!sc->link_active) {
3187                 ifq_purge(&ifp->if_snd);
3188                 return;
3189         }
3190
3191         if (!IGB_IS_NOT_OACTIVE(txr))
3192                 igb_txeof(txr);
3193
3194         while (!ifq_is_empty(&ifp->if_snd)) {
3195                 if (IGB_IS_OACTIVE(txr)) {
3196                         ifp->if_flags |= IFF_OACTIVE;
3197                         /* Set watchdog on */
3198                         ifp->if_timer = 5;
3199                         break;
3200                 }
3201
3202                 m_head = ifq_dequeue(&ifp->if_snd, NULL);
3203                 if (m_head == NULL)
3204                         break;
3205
3206                 if (igb_encap(txr, &m_head)) {
3207                         ifp->if_oerrors++;
3208                         continue;
3209                 }
3210
3211                 /* Send a copy of the frame to the BPF listener */
3212                 ETHER_BPF_MTAP(ifp, m_head);
3213         }
3214 }
3215
3216 static void
3217 igb_watchdog(struct ifnet *ifp)
3218 {
3219         struct igb_softc *sc = ifp->if_softc;
3220         struct igb_tx_ring *txr = &sc->tx_rings[0];
3221
3222         ASSERT_IFNET_SERIALIZED_ALL(ifp);
3223
3224         /* 
3225          * If flow control has paused us since last checking
3226          * it invalidates the watchdog timing, so dont run it.
3227          */
3228         if (sc->pause_frames) {
3229                 sc->pause_frames = 0;
3230                 ifp->if_timer = 5;
3231                 return;
3232         }
3233
3234         if_printf(ifp, "Watchdog timeout -- resetting\n");
3235         if_printf(ifp, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
3236             E1000_READ_REG(&sc->hw, E1000_TDH(txr->me)),
3237             E1000_READ_REG(&sc->hw, E1000_TDT(txr->me)));
3238         if_printf(ifp, "TX(%d) desc avail = %d, "
3239             "Next TX to Clean = %d\n",
3240             txr->me, txr->tx_avail, txr->next_to_clean);
3241
3242         ifp->if_oerrors++;
3243         sc->watchdog_events++;
3244
3245         igb_init(sc);
3246         if (!ifq_is_empty(&ifp->if_snd))
3247                 if_devstart(ifp);
3248 }
3249
3250 static void
3251 igb_set_eitr(struct igb_softc *sc, int idx, int rate)
3252 {
3253         uint32_t eitr = 0;
3254
3255         if (rate > 0) {
3256                 if (sc->hw.mac.type == e1000_82575) {
3257                         eitr = 1000000000 / 256 / rate;
3258                         /*
3259                          * NOTE:
3260                          * Document is wrong on the 2 bits left shift
3261                          */
3262                 } else {
3263                         eitr = 1000000 / rate;
3264                         eitr <<= IGB_EITR_INTVL_SHIFT;
3265                 }
3266
3267                 if (eitr == 0) {
3268                         /* Don't disable it */
3269                         eitr = 1 << IGB_EITR_INTVL_SHIFT;
3270                 } else if (eitr > IGB_EITR_INTVL_MASK) {
3271                         /* Don't allow it to be too large */
3272                         eitr = IGB_EITR_INTVL_MASK;
3273                 }
3274         }
3275         if (sc->hw.mac.type == e1000_82575)
3276                 eitr |= eitr << 16;
3277         else
3278                 eitr |= E1000_EITR_CNT_IGNR;
3279         E1000_WRITE_REG(&sc->hw, E1000_EITR(idx), eitr);
3280 }
3281
3282 static int
3283 igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS)
3284 {
3285         struct igb_softc *sc = (void *)arg1;
3286         struct ifnet *ifp = &sc->arpcom.ac_if;
3287         int error, intr_rate;
3288
3289         intr_rate = sc->intr_rate;
3290         error = sysctl_handle_int(oidp, &intr_rate, 0, req);
3291         if (error || req->newptr == NULL)
3292                 return error;
3293         if (intr_rate < 0)
3294                 return EINVAL;
3295
3296         ifnet_serialize_all(ifp);
3297
3298         sc->intr_rate = intr_rate;
3299         if (ifp->if_flags & IFF_RUNNING)
3300                 igb_set_eitr(sc, 0, sc->intr_rate);
3301
3302         if (bootverbose)
3303                 if_printf(ifp, "interrupt rate set to %d/sec\n", sc->intr_rate);
3304
3305         ifnet_deserialize_all(ifp);
3306
3307         return 0;
3308 }
3309
3310 static int
3311 igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS)
3312 {
3313         struct igb_msix_data *msix = (void *)arg1;
3314         struct igb_softc *sc = msix->msix_sc;
3315         struct ifnet *ifp = &sc->arpcom.ac_if;
3316         int error, msix_rate;
3317
3318         msix_rate = msix->msix_rate;
3319         error = sysctl_handle_int(oidp, &msix_rate, 0, req);
3320         if (error || req->newptr == NULL)
3321                 return error;
3322         if (msix_rate < 0)
3323                 return EINVAL;
3324
3325         lwkt_serialize_enter(msix->msix_serialize);
3326
3327         msix->msix_rate = msix_rate;
3328         if (ifp->if_flags & IFF_RUNNING)
3329                 igb_set_eitr(sc, msix->msix_vector, msix->msix_rate);
3330
3331         if (bootverbose) {
3332                 if_printf(ifp, "%s set to %d/sec\n", msix->msix_rate_desc,
3333                     msix->msix_rate);
3334         }
3335
3336         lwkt_serialize_exit(msix->msix_serialize);
3337
3338         return 0;
3339 }
3340
3341 static int
3342 igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS)
3343 {
3344         struct igb_softc *sc = (void *)arg1;
3345         struct ifnet *ifp = &sc->arpcom.ac_if;
3346         struct igb_tx_ring *txr = &sc->tx_rings[0];
3347         int error, nsegs;
3348
3349         nsegs = txr->intr_nsegs;
3350         error = sysctl_handle_int(oidp, &nsegs, 0, req);
3351         if (error || req->newptr == NULL)
3352                 return error;
3353         if (nsegs <= 0)
3354                 return EINVAL;
3355
3356         ifnet_serialize_all(ifp);
3357
3358         if (nsegs >= txr->num_tx_desc - txr->oact_lo_desc ||
3359             nsegs >= txr->oact_hi_desc - IGB_MAX_SCATTER) {
3360                 error = EINVAL;
3361         } else {
3362                 error = 0;
3363                 txr->intr_nsegs = nsegs;
3364         }
3365
3366         ifnet_deserialize_all(ifp);
3367
3368         return error;
3369 }
3370
3371 static void
3372 igb_init_intr(struct igb_softc *sc)
3373 {
3374         igb_set_intr_mask(sc);
3375
3376         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0)
3377                 igb_init_unshared_intr(sc);
3378
3379         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
3380                 igb_set_eitr(sc, 0, sc->intr_rate);
3381         } else {
3382                 int i;
3383
3384                 for (i = 0; i < sc->msix_cnt; ++i)
3385                         igb_set_eitr(sc, i, sc->msix_data[i].msix_rate);
3386         }
3387 }
3388
3389 static void
3390 igb_init_unshared_intr(struct igb_softc *sc)
3391 {
3392         struct e1000_hw *hw = &sc->hw;
3393         const struct igb_rx_ring *rxr;
3394         const struct igb_tx_ring *txr;
3395         uint32_t ivar, index;
3396         int i;
3397
3398         /*
3399          * Enable extended mode
3400          */
3401         if (sc->hw.mac.type != e1000_82575) {
3402                 uint32_t gpie;
3403                 int ivar_max;
3404
3405                 gpie = E1000_GPIE_NSICR;
3406                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3407                         gpie |= E1000_GPIE_MSIX_MODE |
3408                             E1000_GPIE_EIAME |
3409                             E1000_GPIE_PBA;
3410                 }
3411                 E1000_WRITE_REG(hw, E1000_GPIE, gpie);
3412
3413                 /*
3414                  * Clear IVARs
3415                  */
3416                 switch (sc->hw.mac.type) {
3417                 case e1000_82580:
3418                         ivar_max = IGB_MAX_IVAR_82580;
3419                         break;
3420
3421                 case e1000_i350:
3422                         ivar_max = IGB_MAX_IVAR_I350;
3423                         break;
3424
3425                 case e1000_vfadapt:
3426                 case e1000_vfadapt_i350:
3427                         ivar_max = IGB_MAX_IVAR_VF;
3428                         break;
3429
3430                 case e1000_82576:
3431                         ivar_max = IGB_MAX_IVAR_82576;
3432                         break;
3433
3434                 default:
3435                         panic("unknown mac type %d\n", sc->hw.mac.type);
3436                 }
3437                 for (i = 0; i < ivar_max; ++i)
3438                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, i, 0);
3439                 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
3440         } else {
3441                 uint32_t tmp;
3442
3443                 KASSERT(sc->intr_type != PCI_INTR_TYPE_MSIX,
3444                     ("82575 w/ MSI-X"));
3445                 tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
3446                 tmp |= E1000_CTRL_EXT_IRCA;
3447                 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
3448         }
3449
3450         /*
3451          * Map TX/RX interrupts to EICR
3452          */
3453         switch (sc->hw.mac.type) {
3454         case e1000_82580:
3455         case e1000_i350:
3456         case e1000_vfadapt:
3457         case e1000_vfadapt_i350:
3458                 /* RX entries */
3459                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3460                         rxr = &sc->rx_rings[i];
3461
3462                         index = i >> 1;
3463                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3464
3465                         if (i & 1) {
3466                                 ivar &= 0xff00ffff;
3467                                 ivar |=
3468                                 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3469                         } else {
3470                                 ivar &= 0xffffff00;
3471                                 ivar |=
3472                                 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3473                         }
3474                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3475                 }
3476                 /* TX entries */
3477                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3478                         txr = &sc->tx_rings[i];
3479
3480                         index = i >> 1;
3481                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3482
3483                         if (i & 1) {
3484                                 ivar &= 0x00ffffff;
3485                                 ivar |=
3486                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3487                         } else {
3488                                 ivar &= 0xffff00ff;
3489                                 ivar |=
3490                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3491                         }
3492                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3493                 }
3494                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3495                         ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3496                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3497                 }
3498                 break;
3499
3500         case e1000_82576:
3501                 /* RX entries */
3502                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3503                         rxr = &sc->rx_rings[i];
3504
3505                         index = i & 0x7; /* Each IVAR has two entries */
3506                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3507
3508                         if (i < 8) {
3509                                 ivar &= 0xffffff00;
3510                                 ivar |=
3511                                 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3512                         } else {
3513                                 ivar &= 0xff00ffff;
3514                                 ivar |=
3515                                 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3516                         }
3517                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3518                 }
3519                 /* TX entries */
3520                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3521                         txr = &sc->tx_rings[i];
3522
3523                         index = i & 0x7; /* Each IVAR has two entries */
3524                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3525
3526                         if (i < 8) {
3527                                 ivar &= 0xffff00ff;
3528                                 ivar |=
3529                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3530                         } else {
3531                                 ivar &= 0x00ffffff;
3532                                 ivar |=
3533                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3534                         }
3535                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3536                 }
3537                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3538                         ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3539                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3540                 }
3541                 break;
3542
3543         case e1000_82575:
3544                 /*
3545                  * Enable necessary interrupt bits.
3546                  *
3547                  * The name of the register is confusing; in addition to
3548                  * configuring the first vector of MSI-X, it also configures
3549                  * which bits of EICR could be set by the hardware even when
3550                  * MSI or line interrupt is used; it thus controls interrupt
3551                  * generation.  It MUST be configured explicitly; the default
3552                  * value mentioned in the datasheet is wrong: RX queue0 and
3553                  * TX queue0 are NOT enabled by default.
3554                  */
3555                 E1000_WRITE_REG(&sc->hw, E1000_MSIXBM(0), sc->intr_mask);
3556                 break;
3557
3558         default:
3559                 panic("unknown mac type %d\n", sc->hw.mac.type);
3560         }
3561 }
3562
3563 static int
3564 igb_setup_intr(struct igb_softc *sc)
3565 {
3566         struct ifnet *ifp = &sc->arpcom.ac_if;
3567         int error;
3568
3569         if (sc->intr_type == PCI_INTR_TYPE_MSIX)
3570                 return igb_msix_setup(sc);
3571
3572         error = bus_setup_intr(sc->dev, sc->intr_res, INTR_MPSAFE,
3573             (sc->flags & IGB_FLAG_SHARED_INTR) ? igb_intr_shared : igb_intr,
3574             sc, &sc->intr_tag, &sc->main_serialize);
3575         if (error) {
3576                 device_printf(sc->dev, "Failed to register interrupt handler");
3577                 return error;
3578         }
3579
3580         ifp->if_cpuid = rman_get_cpuid(sc->intr_res);
3581         KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
3582
3583         return 0;
3584 }
3585
3586 static void
3587 igb_set_txintr_mask(struct igb_tx_ring *txr, int *intr_bit0, int intr_bitmax)
3588 {
3589         if (txr->sc->hw.mac.type == e1000_82575) {
3590                 txr->tx_intr_bit = 0;   /* unused */
3591                 switch (txr->me) {
3592                 case 0:
3593                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE0;
3594                         break;
3595                 case 1:
3596                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE1;
3597                         break;
3598                 case 2:
3599                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE2;
3600                         break;
3601                 case 3:
3602                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE3;
3603                         break;
3604                 default:
3605                         panic("unsupported # of TX ring, %d\n", txr->me);
3606                 }
3607         } else {
3608                 int intr_bit = *intr_bit0;
3609
3610                 txr->tx_intr_bit = intr_bit % intr_bitmax;
3611                 txr->tx_intr_mask = 1 << txr->tx_intr_bit;
3612
3613                 *intr_bit0 = intr_bit + 1;
3614         }
3615 }
3616
3617 static void
3618 igb_set_rxintr_mask(struct igb_rx_ring *rxr, int *intr_bit0, int intr_bitmax)
3619 {
3620         if (rxr->sc->hw.mac.type == e1000_82575) {
3621                 rxr->rx_intr_bit = 0;   /* unused */
3622                 switch (rxr->me) {
3623                 case 0:
3624                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE0;
3625                         break;
3626                 case 1:
3627                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE1;
3628                         break;
3629                 case 2:
3630                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE2;
3631                         break;
3632                 case 3:
3633                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE3;
3634                         break;
3635                 default:
3636                         panic("unsupported # of RX ring, %d\n", rxr->me);
3637                 }
3638         } else {
3639                 int intr_bit = *intr_bit0;
3640
3641                 rxr->rx_intr_bit = intr_bit % intr_bitmax;
3642                 rxr->rx_intr_mask = 1 << rxr->rx_intr_bit;
3643
3644                 *intr_bit0 = intr_bit + 1;
3645         }
3646 }
3647
3648 static void
3649 igb_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3650 {
3651         struct igb_softc *sc = ifp->if_softc;
3652
3653         ifnet_serialize_array_enter(sc->serializes, sc->serialize_cnt,
3654             sc->tx_serialize, sc->rx_serialize, slz);
3655 }
3656
3657 static void
3658 igb_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3659 {
3660         struct igb_softc *sc = ifp->if_softc;
3661
3662         ifnet_serialize_array_exit(sc->serializes, sc->serialize_cnt,
3663             sc->tx_serialize, sc->rx_serialize, slz);
3664 }
3665
3666 static int
3667 igb_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3668 {
3669         struct igb_softc *sc = ifp->if_softc;
3670
3671         return ifnet_serialize_array_try(sc->serializes, sc->serialize_cnt,
3672             sc->tx_serialize, sc->rx_serialize, slz);
3673 }
3674
3675 #ifdef INVARIANTS
3676
3677 static void
3678 igb_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3679     boolean_t serialized)
3680 {
3681         struct igb_softc *sc = ifp->if_softc;
3682
3683         ifnet_serialize_array_assert(sc->serializes, sc->serialize_cnt,
3684             sc->tx_serialize, sc->rx_serialize, slz, serialized);
3685 }
3686
3687 #endif  /* INVARIANTS */
3688
3689 static void
3690 igb_set_intr_mask(struct igb_softc *sc)
3691 {
3692         int i;
3693
3694         sc->intr_mask = sc->sts_intr_mask;
3695         for (i = 0; i < sc->rx_ring_inuse; ++i)
3696                 sc->intr_mask |= sc->rx_rings[i].rx_intr_mask;
3697         for (i = 0; i < sc->tx_ring_cnt; ++i)
3698                 sc->intr_mask |= sc->tx_rings[i].tx_intr_mask;
3699         if (bootverbose)
3700                 device_printf(sc->dev, "intr mask 0x%08x\n", sc->intr_mask);
3701 }
3702
3703 static int
3704 igb_alloc_intr(struct igb_softc *sc)
3705 {
3706         int i, intr_bit, intr_bitmax;
3707         u_int intr_flags;
3708
3709         igb_msix_try_alloc(sc);
3710         if (sc->intr_type == PCI_INTR_TYPE_MSIX)
3711                 goto done;
3712
3713         /*
3714          * Allocate MSI/legacy interrupt resource
3715          */
3716         sc->intr_type = pci_alloc_1intr(sc->dev, igb_msi_enable,
3717             &sc->intr_rid, &intr_flags);
3718
3719         if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
3720                 int unshared;
3721
3722                 unshared = device_getenv_int(sc->dev, "irq.unshared", 0);
3723                 if (!unshared) {
3724                         sc->flags |= IGB_FLAG_SHARED_INTR;
3725                         if (bootverbose)
3726                                 device_printf(sc->dev, "IRQ shared\n");
3727                 } else {
3728                         intr_flags &= ~RF_SHAREABLE;
3729                         if (bootverbose)
3730                                 device_printf(sc->dev, "IRQ unshared\n");
3731                 }
3732         }
3733
3734         sc->intr_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
3735             &sc->intr_rid, intr_flags);
3736         if (sc->intr_res == NULL) {
3737                 device_printf(sc->dev, "Unable to allocate bus resource: "
3738                     "interrupt\n");
3739                 return ENXIO;
3740         }
3741
3742         /*
3743          * Setup MSI/legacy interrupt mask
3744          */
3745         switch (sc->hw.mac.type) {
3746         case e1000_82575:
3747                 intr_bitmax = IGB_MAX_TXRXINT_82575;
3748                 break;
3749         case e1000_82580:
3750                 intr_bitmax = IGB_MAX_TXRXINT_82580;
3751                 break;
3752         case e1000_i350:
3753                 intr_bitmax = IGB_MAX_TXRXINT_I350;
3754                 break;
3755         case e1000_82576:
3756                 intr_bitmax = IGB_MAX_TXRXINT_82576;
3757                 break;
3758         default:
3759                 intr_bitmax = IGB_MIN_TXRXINT;
3760                 break;
3761         }
3762         intr_bit = 0;
3763         for (i = 0; i < sc->tx_ring_cnt; ++i)
3764                 igb_set_txintr_mask(&sc->tx_rings[i], &intr_bit, intr_bitmax);
3765         for (i = 0; i < sc->rx_ring_cnt; ++i)
3766                 igb_set_rxintr_mask(&sc->rx_rings[i], &intr_bit, intr_bitmax);
3767         sc->sts_intr_bit = 0;
3768         sc->sts_intr_mask = E1000_EICR_OTHER;
3769
3770         /* Initialize interrupt rate */
3771         sc->intr_rate = IGB_INTR_RATE;
3772 done:
3773         igb_set_ring_inuse(sc, FALSE);
3774         igb_set_intr_mask(sc);
3775         return 0;
3776 }
3777
3778 static void
3779 igb_free_intr(struct igb_softc *sc)
3780 {
3781         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
3782                 if (sc->intr_res != NULL) {
3783                         bus_release_resource(sc->dev, SYS_RES_IRQ, sc->intr_rid,
3784                             sc->intr_res);
3785                 }
3786                 if (sc->intr_type == PCI_INTR_TYPE_MSI)
3787                         pci_release_msi(sc->dev);
3788         } else {
3789                 igb_msix_free(sc, TRUE);
3790         }
3791 }
3792
3793 static void
3794 igb_teardown_intr(struct igb_softc *sc)
3795 {
3796         if (sc->intr_type != PCI_INTR_TYPE_MSIX)
3797                 bus_teardown_intr(sc->dev, sc->intr_res, sc->intr_tag);
3798         else
3799                 igb_msix_teardown(sc, sc->msix_cnt);
3800 }
3801
3802 static void
3803 igb_msix_try_alloc(struct igb_softc *sc)
3804 {
3805         int msix_enable, msix_cnt, msix_cnt2, alloc_cnt;
3806         int i, x, error;
3807         struct igb_msix_data *msix;
3808         boolean_t aggregate, setup = FALSE;
3809
3810         /*
3811          * Don't enable MSI-X on 82575, see:
3812          * 82575 specification update errata #25
3813          */
3814         if (sc->hw.mac.type == e1000_82575)
3815                 return;
3816
3817         /* Don't enable MSI-X on VF */
3818         if (sc->vf_ifp)
3819                 return;
3820
3821         msix_enable = device_getenv_int(sc->dev, "msix.enable",
3822             igb_msix_enable);
3823         if (!msix_enable)
3824                 return;
3825
3826         msix_cnt = pci_msix_count(sc->dev);
3827 #ifdef IGB_MSIX_DEBUG
3828         msix_cnt = device_getenv_int(sc->dev, "msix.count", msix_cnt);
3829 #endif
3830         if (msix_cnt <= 1) {
3831                 /* One MSI-X model does not make sense */
3832                 return;
3833         }
3834
3835         i = 0;
3836         while ((1 << (i + 1)) <= msix_cnt)
3837                 ++i;
3838         msix_cnt2 = 1 << i;
3839
3840         if (bootverbose) {
3841                 device_printf(sc->dev, "MSI-X count %d/%d\n",
3842                     msix_cnt2, msix_cnt);
3843         }
3844
3845         KKASSERT(msix_cnt2 <= msix_cnt);
3846         if (msix_cnt == msix_cnt2) {
3847                 /* We need at least one MSI-X for link status */
3848                 msix_cnt2 >>= 1;
3849                 if (msix_cnt2 <= 1) {
3850                         /* One MSI-X for RX/TX does not make sense */
3851                         device_printf(sc->dev, "not enough MSI-X for TX/RX, "
3852                             "MSI-X count %d/%d\n", msix_cnt2, msix_cnt);
3853                         return;
3854                 }
3855                 KKASSERT(msix_cnt > msix_cnt2);
3856
3857                 if (bootverbose) {
3858                         device_printf(sc->dev, "MSI-X count fixup %d/%d\n",
3859                             msix_cnt2, msix_cnt);
3860                 }
3861         }
3862
3863         sc->rx_ring_msix = sc->rx_ring_cnt;
3864         if (sc->rx_ring_msix > msix_cnt2)
3865                 sc->rx_ring_msix = msix_cnt2;
3866
3867         if (msix_cnt >= sc->tx_ring_cnt + sc->rx_ring_msix + 1) {
3868                 /*
3869                  * Independent TX/RX MSI-X
3870                  */
3871                 aggregate = FALSE;
3872                 if (bootverbose)
3873                         device_printf(sc->dev, "independent TX/RX MSI-X\n");
3874                 alloc_cnt = sc->tx_ring_cnt + sc->rx_ring_msix;
3875         } else {
3876                 /*
3877                  * Aggregate TX/RX MSI-X
3878                  */
3879                 aggregate = TRUE;
3880                 if (bootverbose)
3881                         device_printf(sc->dev, "aggregate TX/RX MSI-X\n");
3882                 alloc_cnt = msix_cnt2;
3883                 if (alloc_cnt > ncpus2)
3884                         alloc_cnt = ncpus2;
3885                 if (sc->rx_ring_msix > alloc_cnt)
3886                         sc->rx_ring_msix = alloc_cnt;
3887         }
3888         ++alloc_cnt;    /* For link status */
3889
3890         if (bootverbose) {
3891                 device_printf(sc->dev, "MSI-X alloc %d, RX ring %d\n",
3892                     alloc_cnt, sc->rx_ring_msix);
3893         }
3894
3895         sc->msix_mem_rid = PCIR_BAR(IGB_MSIX_BAR);
3896         sc->msix_mem_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
3897             &sc->msix_mem_rid, RF_ACTIVE);
3898         if (sc->msix_mem_res == NULL) {
3899                 device_printf(sc->dev, "Unable to map MSI-X table\n");
3900                 return;
3901         }
3902
3903         sc->msix_cnt = alloc_cnt;
3904         sc->msix_data = kmalloc(sizeof(struct igb_msix_data) * sc->msix_cnt,
3905             M_DEVBUF, M_WAITOK | M_ZERO);
3906         for (x = 0; x < sc->msix_cnt; ++x) {
3907                 msix = &sc->msix_data[x];
3908
3909                 lwkt_serialize_init(&msix->msix_serialize0);
3910                 msix->msix_sc = sc;
3911                 msix->msix_rid = -1;
3912                 msix->msix_vector = x;
3913                 msix->msix_mask = 1 << msix->msix_vector;
3914                 msix->msix_rate = IGB_INTR_RATE;
3915         }
3916
3917         x = 0;
3918         if (!aggregate) {
3919                 int offset, offset_def;
3920
3921                 if (sc->rx_ring_msix == ncpus2) {
3922                         offset = 0;
3923                 } else {
3924                         offset_def = (sc->rx_ring_msix *
3925                             device_get_unit(sc->dev)) % ncpus2;
3926
3927                         offset = device_getenv_int(sc->dev,
3928                             "msix.rxoff", offset_def);
3929                         if (offset >= ncpus2 ||
3930                             offset % sc->rx_ring_msix != 0) {
3931                                 device_printf(sc->dev,
3932                                     "invalid msix.rxoff %d, use %d\n",
3933                                     offset, offset_def);
3934                                 offset = offset_def;
3935                         }
3936                 }
3937
3938                 /* RX rings */
3939                 for (i = 0; i < sc->rx_ring_msix; ++i) {
3940                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
3941
3942                         KKASSERT(x < sc->msix_cnt);
3943                         msix = &sc->msix_data[x++];
3944                         rxr->rx_intr_bit = msix->msix_vector;
3945                         rxr->rx_intr_mask = msix->msix_mask;
3946
3947                         msix->msix_serialize = &rxr->rx_serialize;
3948                         msix->msix_func = igb_msix_rx;
3949                         msix->msix_arg = rxr;
3950                         msix->msix_cpuid = i + offset;
3951                         KKASSERT(msix->msix_cpuid < ncpus2);
3952                         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
3953                             "%s rx%d", device_get_nameunit(sc->dev), i);
3954                         msix->msix_rate = IGB_MSIX_RX_RATE;
3955                         ksnprintf(msix->msix_rate_desc,
3956                             sizeof(msix->msix_rate_desc),
3957                             "RX%d interrupt rate", i);
3958                 }
3959
3960                 offset_def = device_get_unit(sc->dev) % ncpus2;
3961                 offset = device_getenv_int(sc->dev, "msix.txoff", offset_def);
3962                 if (offset >= ncpus2) {
3963                         device_printf(sc->dev, "invalid msix.txoff %d, "
3964                             "use %d\n", offset, offset_def);
3965                         offset = offset_def;
3966                 }
3967
3968                 /* TX rings */
3969                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3970                         struct igb_tx_ring *txr = &sc->tx_rings[i];
3971
3972                         KKASSERT(x < sc->msix_cnt);
3973                         msix = &sc->msix_data[x++];
3974                         txr->tx_intr_bit = msix->msix_vector;
3975                         txr->tx_intr_mask = msix->msix_mask;
3976
3977                         msix->msix_serialize = &txr->tx_serialize;
3978                         msix->msix_func = igb_msix_tx;
3979                         msix->msix_arg = txr;
3980                         msix->msix_cpuid = i + offset;
3981                         sc->msix_tx_cpuid = msix->msix_cpuid; /* XXX */
3982                         KKASSERT(msix->msix_cpuid < ncpus2);
3983                         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
3984                             "%s tx%d", device_get_nameunit(sc->dev), i);
3985                         msix->msix_rate = IGB_MSIX_TX_RATE;
3986                         ksnprintf(msix->msix_rate_desc,
3987                             sizeof(msix->msix_rate_desc),
3988                             "TX%d interrupt rate", i);
3989                 }
3990         } else {
3991                 /* TODO */
3992                 error = EOPNOTSUPP;
3993                 goto back;
3994         }
3995
3996         /*
3997          * Link status
3998          */
3999         KKASSERT(x < sc->msix_cnt);
4000         msix = &sc->msix_data[x++];
4001         sc->sts_intr_bit = msix->msix_vector;
4002         sc->sts_intr_mask = msix->msix_mask;
4003
4004         msix->msix_serialize = &sc->main_serialize;
4005         msix->msix_func = igb_msix_status;
4006         msix->msix_arg = sc;
4007         msix->msix_cpuid = 0; /* TODO tunable */
4008         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc), "%s sts",
4009             device_get_nameunit(sc->dev));
4010         ksnprintf(msix->msix_rate_desc, sizeof(msix->msix_rate_desc),
4011             "status interrupt rate");
4012
4013         KKASSERT(x == sc->msix_cnt);
4014
4015         error = pci_setup_msix(sc->dev);
4016         if (error) {
4017                 device_printf(sc->dev, "Setup MSI-X failed\n");
4018                 goto back;
4019         }
4020         setup = TRUE;
4021
4022         for (i = 0; i < sc->msix_cnt; ++i) {
4023                 msix = &sc->msix_data[i];
4024
4025                 error = pci_alloc_msix_vector(sc->dev, msix->msix_vector,
4026                     &msix->msix_rid, msix->msix_cpuid);
4027                 if (error) {
4028                         device_printf(sc->dev,
4029                             "Unable to allocate MSI-X %d on cpu%d\n",
4030                             msix->msix_vector, msix->msix_cpuid);
4031                         goto back;
4032                 }
4033
4034                 msix->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4035                     &msix->msix_rid, RF_ACTIVE);
4036                 if (msix->msix_res == NULL) {
4037                         device_printf(sc->dev,
4038                             "Unable to allocate MSI-X %d resource\n",
4039                             msix->msix_vector);
4040                         error = ENOMEM;
4041                         goto back;
4042                 }
4043         }
4044
4045         pci_enable_msix(sc->dev);
4046         sc->intr_type = PCI_INTR_TYPE_MSIX;
4047 back:
4048         if (error)
4049    &