ixv.c revision 1.133 1 /*$NetBSD: ixv.c,v 1.133 2019/09/12 11:48:44 msaitoh Exp $*/
2
3 /******************************************************************************
4
5 Copyright (c) 2001-2017, Intel Corporation
6 All rights reserved.
7
8 Redistribution and use in source and binary forms, with or without
9 modification, are permitted provided that the following conditions are met:
10
11 1. Redistributions of source code must retain the above copyright notice,
12 this list of conditions and the following disclaimer.
13
14 2. Redistributions in binary form must reproduce the above copyright
15 notice, this list of conditions and the following disclaimer in the
16 documentation and/or other materials provided with the distribution.
17
18 3. Neither the name of the Intel Corporation nor the names of its
19 contributors may be used to endorse or promote products derived from
20 this software without specific prior written permission.
21
22 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
23 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 POSSIBILITY OF SUCH DAMAGE.
33
34 ******************************************************************************/
35 /*$FreeBSD: head/sys/dev/ixgbe/if_ixv.c 331224 2018-03-19 20:55:05Z erj $*/
36
37 #ifdef _KERNEL_OPT
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_net_mpsafe.h"
41 #endif
42
43 #include "ixgbe.h"
44 #include "vlan.h"
45
46 /************************************************************************
47 * Driver version
48 ************************************************************************/
49 static const char ixv_driver_version[] = "2.0.1-k";
50 /* XXX NetBSD: + 1.5.17 */
51
52 /************************************************************************
53 * PCI Device ID Table
54 *
55 * Used by probe to select devices to load on
56 * Last field stores an index into ixv_strings
57 * Last entry must be all 0s
58 *
59 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
60 ************************************************************************/
61 static const ixgbe_vendor_info_t ixv_vendor_info_array[] =
62 {
63 {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_VF, 0, 0, 0},
64 {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540_VF, 0, 0, 0},
65 {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550_VF, 0, 0, 0},
66 {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_VF, 0, 0, 0},
67 {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_VF, 0, 0, 0},
68 /* required last entry */
69 {0, 0, 0, 0, 0}
70 };
71
72 /************************************************************************
73 * Table of branding strings
74 ************************************************************************/
75 static const char *ixv_strings[] = {
76 "Intel(R) PRO/10GbE Virtual Function Network Driver"
77 };
78
79 /*********************************************************************
80 * Function prototypes
81 *********************************************************************/
82 static int ixv_probe(device_t, cfdata_t, void *);
83 static void ixv_attach(device_t, device_t, void *);
84 static int ixv_detach(device_t, int);
85 #if 0
86 static int ixv_shutdown(device_t);
87 #endif
88 static int ixv_ifflags_cb(struct ethercom *);
89 static int ixv_ioctl(struct ifnet *, u_long, void *);
90 static int ixv_init(struct ifnet *);
91 static void ixv_init_locked(struct adapter *);
92 static void ixv_ifstop(struct ifnet *, int);
93 static void ixv_stop(void *);
94 static void ixv_init_device_features(struct adapter *);
95 static void ixv_media_status(struct ifnet *, struct ifmediareq *);
96 static int ixv_media_change(struct ifnet *);
97 static int ixv_allocate_pci_resources(struct adapter *,
98 const struct pci_attach_args *);
99 static int ixv_allocate_msix(struct adapter *,
100 const struct pci_attach_args *);
101 static int ixv_configure_interrupts(struct adapter *);
102 static void ixv_free_pci_resources(struct adapter *);
103 static void ixv_local_timer(void *);
104 static void ixv_local_timer_locked(void *);
105 static int ixv_setup_interface(device_t, struct adapter *);
106 static int ixv_negotiate_api(struct adapter *);
107
108 static void ixv_initialize_transmit_units(struct adapter *);
109 static void ixv_initialize_receive_units(struct adapter *);
110 static void ixv_initialize_rss_mapping(struct adapter *);
111 static s32 ixv_check_link(struct adapter *);
112
113 static void ixv_enable_intr(struct adapter *);
114 static void ixv_disable_intr(struct adapter *);
115 static void ixv_set_multi(struct adapter *);
116 static void ixv_update_link_status(struct adapter *);
117 static int ixv_sysctl_debug(SYSCTLFN_PROTO);
118 static void ixv_set_ivar(struct adapter *, u8, u8, s8);
119 static void ixv_configure_ivars(struct adapter *);
120 static u8 * ixv_mc_array_itr(struct ixgbe_hw *, u8 **, u32 *);
121 static void ixv_eitr_write(struct adapter *, uint32_t, uint32_t);
122
123 static void ixv_setup_vlan_tagging(struct adapter *);
124 static int ixv_setup_vlan_support(struct adapter *);
125 static int ixv_vlan_cb(struct ethercom *, uint16_t, bool);
126 static int ixv_register_vlan(struct adapter *, u16);
127 static int ixv_unregister_vlan(struct adapter *, u16);
128
129 static void ixv_add_device_sysctls(struct adapter *);
130 static void ixv_save_stats(struct adapter *);
131 static void ixv_init_stats(struct adapter *);
132 static void ixv_update_stats(struct adapter *);
133 static void ixv_add_stats_sysctls(struct adapter *);
134 static void ixv_clear_evcnt(struct adapter *);
135
136 /* Sysctl handlers */
137 static void ixv_set_sysctl_value(struct adapter *, const char *,
138 const char *, int *, int);
139 static int ixv_sysctl_interrupt_rate_handler(SYSCTLFN_PROTO);
140 static int ixv_sysctl_next_to_check_handler(SYSCTLFN_PROTO);
141 static int ixv_sysctl_rdh_handler(SYSCTLFN_PROTO);
142 static int ixv_sysctl_rdt_handler(SYSCTLFN_PROTO);
143 static int ixv_sysctl_tdt_handler(SYSCTLFN_PROTO);
144 static int ixv_sysctl_tdh_handler(SYSCTLFN_PROTO);
145
146 /* The MSI-X Interrupt handlers */
147 static int ixv_msix_que(void *);
148 static int ixv_msix_mbx(void *);
149
150 /* Deferred interrupt tasklets */
151 static void ixv_handle_que(void *);
152 static void ixv_handle_link(void *);
153
154 /* Workqueue handler for deferred work */
155 static void ixv_handle_que_work(struct work *, void *);
156
157 const struct sysctlnode *ixv_sysctl_instance(struct adapter *);
158 static const ixgbe_vendor_info_t *ixv_lookup(const struct pci_attach_args *);
159
160 /************************************************************************
161 * FreeBSD Device Interface Entry Points
162 ************************************************************************/
163 CFATTACH_DECL3_NEW(ixv, sizeof(struct adapter),
164 ixv_probe, ixv_attach, ixv_detach, NULL, NULL, NULL,
165 DVF_DETACH_SHUTDOWN);
166
167 #if 0
168 static driver_t ixv_driver = {
169 "ixv", ixv_methods, sizeof(struct adapter),
170 };
171
172 devclass_t ixv_devclass;
173 DRIVER_MODULE(ixv, pci, ixv_driver, ixv_devclass, 0, 0);
174 MODULE_DEPEND(ixv, pci, 1, 1, 1);
175 MODULE_DEPEND(ixv, ether, 1, 1, 1);
176 #endif
177
178 /*
179 * TUNEABLE PARAMETERS:
180 */
181
182 /* Number of Queues - do not exceed MSI-X vectors - 1 */
183 static int ixv_num_queues = 0;
184 #define TUNABLE_INT(__x, __y)
185 TUNABLE_INT("hw.ixv.num_queues", &ixv_num_queues);
186
187 /*
188 * AIM: Adaptive Interrupt Moderation
189 * which means that the interrupt rate
190 * is varied over time based on the
191 * traffic for that interrupt vector
192 */
193 static bool ixv_enable_aim = false;
194 TUNABLE_INT("hw.ixv.enable_aim", &ixv_enable_aim);
195
196 static int ixv_max_interrupt_rate = (4000000 / IXGBE_LOW_LATENCY);
197 TUNABLE_INT("hw.ixv.max_interrupt_rate", &ixv_max_interrupt_rate);
198
199 /* How many packets rxeof tries to clean at a time */
200 static int ixv_rx_process_limit = 256;
201 TUNABLE_INT("hw.ixv.rx_process_limit", &ixv_rx_process_limit);
202
203 /* How many packets txeof tries to clean at a time */
204 static int ixv_tx_process_limit = 256;
205 TUNABLE_INT("hw.ixv.tx_process_limit", &ixv_tx_process_limit);
206
207 /* Which packet processing uses workqueue or softint */
208 static bool ixv_txrx_workqueue = false;
209
210 /*
211 * Number of TX descriptors per ring,
212 * setting higher than RX as this seems
213 * the better performing choice.
214 */
215 static int ixv_txd = PERFORM_TXD;
216 TUNABLE_INT("hw.ixv.txd", &ixv_txd);
217
218 /* Number of RX descriptors per ring */
219 static int ixv_rxd = PERFORM_RXD;
220 TUNABLE_INT("hw.ixv.rxd", &ixv_rxd);
221
222 /* Legacy Transmit (single queue) */
223 static int ixv_enable_legacy_tx = 0;
224 TUNABLE_INT("hw.ixv.enable_legacy_tx", &ixv_enable_legacy_tx);
225
226 #ifdef NET_MPSAFE
227 #define IXGBE_MPSAFE 1
228 #define IXGBE_CALLOUT_FLAGS CALLOUT_MPSAFE
229 #define IXGBE_SOFTINFT_FLAGS SOFTINT_MPSAFE
230 #define IXGBE_WORKQUEUE_FLAGS WQ_PERCPU | WQ_MPSAFE
231 #else
232 #define IXGBE_CALLOUT_FLAGS 0
233 #define IXGBE_SOFTINFT_FLAGS 0
234 #define IXGBE_WORKQUEUE_FLAGS WQ_PERCPU
235 #endif
236 #define IXGBE_WORKQUEUE_PRI PRI_SOFTNET
237
238 #if 0
239 static int (*ixv_start_locked)(struct ifnet *, struct tx_ring *);
240 static int (*ixv_ring_empty)(struct ifnet *, struct buf_ring *);
241 #endif
242
243 /************************************************************************
244 * ixv_probe - Device identification routine
245 *
246 * Determines if the driver should be loaded on
247 * adapter based on its PCI vendor/device ID.
248 *
249 * return BUS_PROBE_DEFAULT on success, positive on failure
250 ************************************************************************/
251 static int
252 ixv_probe(device_t dev, cfdata_t cf, void *aux)
253 {
254 #ifdef __HAVE_PCI_MSI_MSIX
255 const struct pci_attach_args *pa = aux;
256
257 return (ixv_lookup(pa) != NULL) ? 1 : 0;
258 #else
259 return 0;
260 #endif
261 } /* ixv_probe */
262
263 static const ixgbe_vendor_info_t *
264 ixv_lookup(const struct pci_attach_args *pa)
265 {
266 const ixgbe_vendor_info_t *ent;
267 pcireg_t subid;
268
269 INIT_DEBUGOUT("ixv_lookup: begin");
270
271 if (PCI_VENDOR(pa->pa_id) != IXGBE_INTEL_VENDOR_ID)
272 return NULL;
273
274 subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
275
276 for (ent = ixv_vendor_info_array; ent->vendor_id != 0; ent++) {
277 if ((PCI_VENDOR(pa->pa_id) == ent->vendor_id) &&
278 (PCI_PRODUCT(pa->pa_id) == ent->device_id) &&
279 ((PCI_SUBSYS_VENDOR(subid) == ent->subvendor_id) ||
280 (ent->subvendor_id == 0)) &&
281 ((PCI_SUBSYS_ID(subid) == ent->subdevice_id) ||
282 (ent->subdevice_id == 0))) {
283 return ent;
284 }
285 }
286
287 return NULL;
288 }
289
290 /************************************************************************
291 * ixv_attach - Device initialization routine
292 *
293 * Called when the driver is being loaded.
294 * Identifies the type of hardware, allocates all resources
295 * and initializes the hardware.
296 *
297 * return 0 on success, positive on failure
298 ************************************************************************/
299 static void
300 ixv_attach(device_t parent, device_t dev, void *aux)
301 {
302 struct adapter *adapter;
303 struct ixgbe_hw *hw;
304 int error = 0;
305 pcireg_t id, subid;
306 const ixgbe_vendor_info_t *ent;
307 const struct pci_attach_args *pa = aux;
308 const char *apivstr;
309 const char *str;
310 char buf[256];
311
312 INIT_DEBUGOUT("ixv_attach: begin");
313
314 /*
315 * Make sure BUSMASTER is set, on a VM under
316 * KVM it may not be and will break things.
317 */
318 ixgbe_pci_enable_busmaster(pa->pa_pc, pa->pa_tag);
319
320 /* Allocate, clear, and link in our adapter structure */
321 adapter = device_private(dev);
322 adapter->dev = dev;
323 adapter->hw.back = adapter;
324 hw = &adapter->hw;
325
326 adapter->init_locked = ixv_init_locked;
327 adapter->stop_locked = ixv_stop;
328
329 adapter->osdep.pc = pa->pa_pc;
330 adapter->osdep.tag = pa->pa_tag;
331 if (pci_dma64_available(pa))
332 adapter->osdep.dmat = pa->pa_dmat64;
333 else
334 adapter->osdep.dmat = pa->pa_dmat;
335 adapter->osdep.attached = false;
336
337 ent = ixv_lookup(pa);
338
339 KASSERT(ent != NULL);
340
341 aprint_normal(": %s, Version - %s\n",
342 ixv_strings[ent->index], ixv_driver_version);
343
344 /* Core Lock Init*/
345 IXGBE_CORE_LOCK_INIT(adapter, device_xname(dev));
346
347 /* Do base PCI setup - map BAR0 */
348 if (ixv_allocate_pci_resources(adapter, pa)) {
349 aprint_error_dev(dev, "ixv_allocate_pci_resources() failed!\n");
350 error = ENXIO;
351 goto err_out;
352 }
353
354 /* SYSCTL APIs */
355 ixv_add_device_sysctls(adapter);
356
357 /* Set up the timer callout */
358 callout_init(&adapter->timer, IXGBE_CALLOUT_FLAGS);
359
360 /* Save off the information about this board */
361 id = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_ID_REG);
362 subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
363 hw->vendor_id = PCI_VENDOR(id);
364 hw->device_id = PCI_PRODUCT(id);
365 hw->revision_id =
366 PCI_REVISION(pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_CLASS_REG));
367 hw->subsystem_vendor_id = PCI_SUBSYS_VENDOR(subid);
368 hw->subsystem_device_id = PCI_SUBSYS_ID(subid);
369
370 /* A subset of set_mac_type */
371 switch (hw->device_id) {
372 case IXGBE_DEV_ID_82599_VF:
373 hw->mac.type = ixgbe_mac_82599_vf;
374 str = "82599 VF";
375 break;
376 case IXGBE_DEV_ID_X540_VF:
377 hw->mac.type = ixgbe_mac_X540_vf;
378 str = "X540 VF";
379 break;
380 case IXGBE_DEV_ID_X550_VF:
381 hw->mac.type = ixgbe_mac_X550_vf;
382 str = "X550 VF";
383 break;
384 case IXGBE_DEV_ID_X550EM_X_VF:
385 hw->mac.type = ixgbe_mac_X550EM_x_vf;
386 str = "X550EM X VF";
387 break;
388 case IXGBE_DEV_ID_X550EM_A_VF:
389 hw->mac.type = ixgbe_mac_X550EM_a_vf;
390 str = "X550EM A VF";
391 break;
392 default:
393 /* Shouldn't get here since probe succeeded */
394 aprint_error_dev(dev, "Unknown device ID!\n");
395 error = ENXIO;
396 goto err_out;
397 break;
398 }
399 aprint_normal_dev(dev, "device %s\n", str);
400
401 ixv_init_device_features(adapter);
402
403 /* Initialize the shared code */
404 error = ixgbe_init_ops_vf(hw);
405 if (error) {
406 aprint_error_dev(dev, "ixgbe_init_ops_vf() failed!\n");
407 error = EIO;
408 goto err_out;
409 }
410
411 /* Setup the mailbox */
412 ixgbe_init_mbx_params_vf(hw);
413
414 /* Set the right number of segments */
415 adapter->num_segs = IXGBE_82599_SCATTER;
416
417 /* Reset mbox api to 1.0 */
418 error = hw->mac.ops.reset_hw(hw);
419 if (error == IXGBE_ERR_RESET_FAILED)
420 aprint_error_dev(dev, "...reset_hw() failure: Reset Failed!\n");
421 else if (error)
422 aprint_error_dev(dev, "...reset_hw() failed with error %d\n",
423 error);
424 if (error) {
425 error = EIO;
426 goto err_out;
427 }
428
429 error = hw->mac.ops.init_hw(hw);
430 if (error) {
431 aprint_error_dev(dev, "...init_hw() failed!\n");
432 error = EIO;
433 goto err_out;
434 }
435
436 /* Negotiate mailbox API version */
437 error = ixv_negotiate_api(adapter);
438 if (error)
439 aprint_normal_dev(dev,
440 "MBX API negotiation failed during attach!\n");
441 switch (hw->api_version) {
442 case ixgbe_mbox_api_10:
443 apivstr = "1.0";
444 break;
445 case ixgbe_mbox_api_20:
446 apivstr = "2.0";
447 break;
448 case ixgbe_mbox_api_11:
449 apivstr = "1.1";
450 break;
451 case ixgbe_mbox_api_12:
452 apivstr = "1.2";
453 break;
454 case ixgbe_mbox_api_13:
455 apivstr = "1.3";
456 break;
457 default:
458 apivstr = "unknown";
459 break;
460 }
461 aprint_normal_dev(dev, "Mailbox API %s\n", apivstr);
462
463 /* If no mac address was assigned, make a random one */
464 if (!ixv_check_ether_addr(hw->mac.addr)) {
465 u8 addr[ETHER_ADDR_LEN];
466 uint64_t rndval = cprng_strong64();
467
468 memcpy(addr, &rndval, sizeof(addr));
469 addr[0] &= 0xFE;
470 addr[0] |= 0x02;
471 bcopy(addr, hw->mac.addr, sizeof(addr));
472 }
473
474 /* Register for VLAN events */
475 ether_set_vlan_cb(&adapter->osdep.ec, ixv_vlan_cb);
476
477 /* Sysctls for limiting the amount of work done in the taskqueues */
478 ixv_set_sysctl_value(adapter, "rx_processing_limit",
479 "max number of rx packets to process",
480 &adapter->rx_process_limit, ixv_rx_process_limit);
481
482 ixv_set_sysctl_value(adapter, "tx_processing_limit",
483 "max number of tx packets to process",
484 &adapter->tx_process_limit, ixv_tx_process_limit);
485
486 /* Do descriptor calc and sanity checks */
487 if (((ixv_txd * sizeof(union ixgbe_adv_tx_desc)) % DBA_ALIGN) != 0 ||
488 ixv_txd < MIN_TXD || ixv_txd > MAX_TXD) {
489 aprint_error_dev(dev, "TXD config issue, using default!\n");
490 adapter->num_tx_desc = DEFAULT_TXD;
491 } else
492 adapter->num_tx_desc = ixv_txd;
493
494 if (((ixv_rxd * sizeof(union ixgbe_adv_rx_desc)) % DBA_ALIGN) != 0 ||
495 ixv_rxd < MIN_RXD || ixv_rxd > MAX_RXD) {
496 aprint_error_dev(dev, "RXD config issue, using default!\n");
497 adapter->num_rx_desc = DEFAULT_RXD;
498 } else
499 adapter->num_rx_desc = ixv_rxd;
500
501 /* Setup MSI-X */
502 error = ixv_configure_interrupts(adapter);
503 if (error)
504 goto err_out;
505
506 /* Allocate our TX/RX Queues */
507 if (ixgbe_allocate_queues(adapter)) {
508 aprint_error_dev(dev, "ixgbe_allocate_queues() failed!\n");
509 error = ENOMEM;
510 goto err_out;
511 }
512
513 /* hw.ix defaults init */
514 adapter->enable_aim = ixv_enable_aim;
515
516 adapter->txrx_use_workqueue = ixv_txrx_workqueue;
517
518 error = ixv_allocate_msix(adapter, pa);
519 if (error) {
520 aprint_error_dev(dev, "ixv_allocate_msix() failed!\n");
521 goto err_late;
522 }
523
524 /* Setup OS specific network interface */
525 error = ixv_setup_interface(dev, adapter);
526 if (error != 0) {
527 aprint_error_dev(dev, "ixv_setup_interface() failed!\n");
528 goto err_late;
529 }
530
531 /* Do the stats setup */
532 ixv_save_stats(adapter);
533 ixv_init_stats(adapter);
534 ixv_add_stats_sysctls(adapter);
535
536 if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
537 ixgbe_netmap_attach(adapter);
538
539 snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_cap);
540 aprint_verbose_dev(dev, "feature cap %s\n", buf);
541 snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_en);
542 aprint_verbose_dev(dev, "feature ena %s\n", buf);
543
544 INIT_DEBUGOUT("ixv_attach: end");
545 adapter->osdep.attached = true;
546
547 return;
548
549 err_late:
550 ixgbe_free_transmit_structures(adapter);
551 ixgbe_free_receive_structures(adapter);
552 free(adapter->queues, M_DEVBUF);
553 err_out:
554 ixv_free_pci_resources(adapter);
555 IXGBE_CORE_LOCK_DESTROY(adapter);
556
557 return;
558 } /* ixv_attach */
559
560 /************************************************************************
561 * ixv_detach - Device removal routine
562 *
563 * Called when the driver is being removed.
564 * Stops the adapter and deallocates all the resources
565 * that were allocated for driver operation.
566 *
567 * return 0 on success, positive on failure
568 ************************************************************************/
569 static int
570 ixv_detach(device_t dev, int flags)
571 {
572 struct adapter *adapter = device_private(dev);
573 struct ixgbe_hw *hw = &adapter->hw;
574 struct ix_queue *que = adapter->queues;
575 struct tx_ring *txr = adapter->tx_rings;
576 struct rx_ring *rxr = adapter->rx_rings;
577 struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
578
579 INIT_DEBUGOUT("ixv_detach: begin");
580 if (adapter->osdep.attached == false)
581 return 0;
582
583 /* Stop the interface. Callouts are stopped in it. */
584 ixv_ifstop(adapter->ifp, 1);
585
586 #if NVLAN > 0
587 /* Make sure VLANs are not using driver */
588 if (!VLAN_ATTACHED(&adapter->osdep.ec))
589 ; /* nothing to do: no VLANs */
590 else if ((flags & (DETACH_SHUTDOWN | DETACH_FORCE)) != 0)
591 vlan_ifdetach(adapter->ifp);
592 else {
593 aprint_error_dev(dev, "VLANs in use, detach first\n");
594 return EBUSY;
595 }
596 #endif
597
598 for (int i = 0; i < adapter->num_queues; i++, que++, txr++) {
599 if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
600 softint_disestablish(txr->txr_si);
601 softint_disestablish(que->que_si);
602 }
603 if (adapter->txr_wq != NULL)
604 workqueue_destroy(adapter->txr_wq);
605 if (adapter->txr_wq_enqueued != NULL)
606 percpu_free(adapter->txr_wq_enqueued, sizeof(u_int));
607 if (adapter->que_wq != NULL)
608 workqueue_destroy(adapter->que_wq);
609
610 /* Drain the Mailbox(link) queue */
611 softint_disestablish(adapter->link_si);
612
613 ether_ifdetach(adapter->ifp);
614 callout_halt(&adapter->timer, NULL);
615
616 if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
617 netmap_detach(adapter->ifp);
618
619 ixv_free_pci_resources(adapter);
620 #if 0 /* XXX the NetBSD port is probably missing something here */
621 bus_generic_detach(dev);
622 #endif
623 if_detach(adapter->ifp);
624 if_percpuq_destroy(adapter->ipq);
625
626 sysctl_teardown(&adapter->sysctllog);
627 evcnt_detach(&adapter->efbig_tx_dma_setup);
628 evcnt_detach(&adapter->mbuf_defrag_failed);
629 evcnt_detach(&adapter->efbig2_tx_dma_setup);
630 evcnt_detach(&adapter->einval_tx_dma_setup);
631 evcnt_detach(&adapter->other_tx_dma_setup);
632 evcnt_detach(&adapter->eagain_tx_dma_setup);
633 evcnt_detach(&adapter->enomem_tx_dma_setup);
634 evcnt_detach(&adapter->watchdog_events);
635 evcnt_detach(&adapter->tso_err);
636 evcnt_detach(&adapter->link_irq);
637
638 txr = adapter->tx_rings;
639 for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
640 evcnt_detach(&adapter->queues[i].irqs);
641 evcnt_detach(&adapter->queues[i].handleq);
642 evcnt_detach(&adapter->queues[i].req);
643 evcnt_detach(&txr->no_desc_avail);
644 evcnt_detach(&txr->total_packets);
645 evcnt_detach(&txr->tso_tx);
646 #ifndef IXGBE_LEGACY_TX
647 evcnt_detach(&txr->pcq_drops);
648 #endif
649
650 evcnt_detach(&rxr->rx_packets);
651 evcnt_detach(&rxr->rx_bytes);
652 evcnt_detach(&rxr->rx_copies);
653 evcnt_detach(&rxr->no_jmbuf);
654 evcnt_detach(&rxr->rx_discarded);
655 }
656 evcnt_detach(&stats->ipcs);
657 evcnt_detach(&stats->l4cs);
658 evcnt_detach(&stats->ipcs_bad);
659 evcnt_detach(&stats->l4cs_bad);
660
661 /* Packet Reception Stats */
662 evcnt_detach(&stats->vfgorc);
663 evcnt_detach(&stats->vfgprc);
664 evcnt_detach(&stats->vfmprc);
665
666 /* Packet Transmission Stats */
667 evcnt_detach(&stats->vfgotc);
668 evcnt_detach(&stats->vfgptc);
669
670 /* Mailbox Stats */
671 evcnt_detach(&hw->mbx.stats.msgs_tx);
672 evcnt_detach(&hw->mbx.stats.msgs_rx);
673 evcnt_detach(&hw->mbx.stats.acks);
674 evcnt_detach(&hw->mbx.stats.reqs);
675 evcnt_detach(&hw->mbx.stats.rsts);
676
677 ixgbe_free_transmit_structures(adapter);
678 ixgbe_free_receive_structures(adapter);
679 for (int i = 0; i < adapter->num_queues; i++) {
680 struct ix_queue *lque = &adapter->queues[i];
681 mutex_destroy(&lque->dc_mtx);
682 }
683 free(adapter->queues, M_DEVBUF);
684
685 IXGBE_CORE_LOCK_DESTROY(adapter);
686
687 return (0);
688 } /* ixv_detach */
689
690 /************************************************************************
691 * ixv_init_locked - Init entry point
692 *
693 * Used in two ways: It is used by the stack as an init entry
694 * point in network interface structure. It is also used
695 * by the driver as a hw/sw initialization routine to get
696 * to a consistent state.
697 *
698 * return 0 on success, positive on failure
699 ************************************************************************/
700 static void
701 ixv_init_locked(struct adapter *adapter)
702 {
703 struct ifnet *ifp = adapter->ifp;
704 device_t dev = adapter->dev;
705 struct ixgbe_hw *hw = &adapter->hw;
706 struct ix_queue *que;
707 int error = 0;
708 uint32_t mask;
709 int i;
710
711 INIT_DEBUGOUT("ixv_init_locked: begin");
712 KASSERT(mutex_owned(&adapter->core_mtx));
713 hw->adapter_stopped = FALSE;
714 hw->mac.ops.stop_adapter(hw);
715 callout_stop(&adapter->timer);
716 for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
717 que->disabled_count = 0;
718
719 /* reprogram the RAR[0] in case user changed it. */
720 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
721
722 /* Get the latest mac address, User can use a LAA */
723 memcpy(hw->mac.addr, CLLADDR(ifp->if_sadl),
724 IXGBE_ETH_LENGTH_OF_ADDRESS);
725 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, 1);
726
727 /* Prepare transmit descriptors and buffers */
728 if (ixgbe_setup_transmit_structures(adapter)) {
729 aprint_error_dev(dev, "Could not setup transmit structures\n");
730 ixv_stop(adapter);
731 return;
732 }
733
734 /* Reset VF and renegotiate mailbox API version */
735 hw->mac.ops.reset_hw(hw);
736 hw->mac.ops.start_hw(hw);
737 error = ixv_negotiate_api(adapter);
738 if (error)
739 device_printf(dev,
740 "Mailbox API negotiation failed in init_locked!\n");
741
742 ixv_initialize_transmit_units(adapter);
743
744 /* Setup Multicast table */
745 ixv_set_multi(adapter);
746
747 /*
748 * Determine the correct mbuf pool
749 * for doing jumbo/headersplit
750 */
751 if (ifp->if_mtu > ETHERMTU)
752 adapter->rx_mbuf_sz = MJUMPAGESIZE;
753 else
754 adapter->rx_mbuf_sz = MCLBYTES;
755
756 /* Prepare receive descriptors and buffers */
757 if (ixgbe_setup_receive_structures(adapter)) {
758 device_printf(dev, "Could not setup receive structures\n");
759 ixv_stop(adapter);
760 return;
761 }
762
763 /* Configure RX settings */
764 ixv_initialize_receive_units(adapter);
765
766 #if 0 /* XXX isn't it required? -- msaitoh */
767 /* Set the various hardware offload abilities */
768 ifp->if_hwassist = 0;
769 if (ifp->if_capenable & IFCAP_TSO4)
770 ifp->if_hwassist |= CSUM_TSO;
771 if (ifp->if_capenable & IFCAP_TXCSUM) {
772 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP);
773 #if __FreeBSD_version >= 800000
774 ifp->if_hwassist |= CSUM_SCTP;
775 #endif
776 }
777 #endif
778
779 /* Set up VLAN offload and filter */
780 ixv_setup_vlan_support(adapter);
781
782 /* Set up MSI-X routing */
783 ixv_configure_ivars(adapter);
784
785 /* Set up auto-mask */
786 mask = (1 << adapter->vector);
787 for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
788 mask |= (1 << que->msix);
789 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, mask);
790
791 /* Set moderation on the Link interrupt */
792 ixv_eitr_write(adapter, adapter->vector, IXGBE_LINK_ITR);
793
794 /* Stats init */
795 ixv_init_stats(adapter);
796
797 /* Config/Enable Link */
798 hw->mac.get_link_status = TRUE;
799 hw->mac.ops.check_link(hw, &adapter->link_speed, &adapter->link_up,
800 FALSE);
801
802 /* Start watchdog */
803 callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
804
805 /* And now turn on interrupts */
806 ixv_enable_intr(adapter);
807
808 /* Update saved flags. See ixgbe_ifflags_cb() */
809 adapter->if_flags = ifp->if_flags;
810 adapter->ec_capenable = adapter->osdep.ec.ec_capenable;
811
812 /* Now inform the stack we're ready */
813 ifp->if_flags |= IFF_RUNNING;
814 ifp->if_flags &= ~IFF_OACTIVE;
815
816 return;
817 } /* ixv_init_locked */
818
819 /************************************************************************
820 * ixv_enable_queue
821 ************************************************************************/
822 static inline void
823 ixv_enable_queue(struct adapter *adapter, u32 vector)
824 {
825 struct ixgbe_hw *hw = &adapter->hw;
826 struct ix_queue *que = &adapter->queues[vector];
827 u32 queue = 1UL << vector;
828 u32 mask;
829
830 mutex_enter(&que->dc_mtx);
831 if (que->disabled_count > 0 && --que->disabled_count > 0)
832 goto out;
833
834 mask = (IXGBE_EIMS_RTX_QUEUE & queue);
835 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
836 out:
837 mutex_exit(&que->dc_mtx);
838 } /* ixv_enable_queue */
839
840 /************************************************************************
841 * ixv_disable_queue
842 ************************************************************************/
843 static inline void
844 ixv_disable_queue(struct adapter *adapter, u32 vector)
845 {
846 struct ixgbe_hw *hw = &adapter->hw;
847 struct ix_queue *que = &adapter->queues[vector];
848 u32 queue = 1UL << vector;
849 u32 mask;
850
851 mutex_enter(&que->dc_mtx);
852 if (que->disabled_count++ > 0)
853 goto out;
854
855 mask = (IXGBE_EIMS_RTX_QUEUE & queue);
856 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, mask);
857 out:
858 mutex_exit(&que->dc_mtx);
859 } /* ixv_disable_queue */
860
861 #if 0
862 static inline void
863 ixv_rearm_queues(struct adapter *adapter, u64 queues)
864 {
865 u32 mask = (IXGBE_EIMS_RTX_QUEUE & queues);
866 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEICS, mask);
867 } /* ixv_rearm_queues */
868 #endif
869
870
871 /************************************************************************
872 * ixv_msix_que - MSI-X Queue Interrupt Service routine
873 ************************************************************************/
874 static int
875 ixv_msix_que(void *arg)
876 {
877 struct ix_queue *que = arg;
878 struct adapter *adapter = que->adapter;
879 struct tx_ring *txr = que->txr;
880 struct rx_ring *rxr = que->rxr;
881 bool more;
882 u32 newitr = 0;
883
884 ixv_disable_queue(adapter, que->msix);
885 ++que->irqs.ev_count;
886
887 #ifdef __NetBSD__
888 /* Don't run ixgbe_rxeof in interrupt context */
889 more = true;
890 #else
891 more = ixgbe_rxeof(que);
892 #endif
893
894 IXGBE_TX_LOCK(txr);
895 ixgbe_txeof(txr);
896 IXGBE_TX_UNLOCK(txr);
897
898 /* Do AIM now? */
899
900 if (adapter->enable_aim == false)
901 goto no_calc;
902 /*
903 * Do Adaptive Interrupt Moderation:
904 * - Write out last calculated setting
905 * - Calculate based on average size over
906 * the last interval.
907 */
908 if (que->eitr_setting)
909 ixv_eitr_write(adapter, que->msix, que->eitr_setting);
910
911 que->eitr_setting = 0;
912
913 /* Idle, do nothing */
914 if ((txr->bytes == 0) && (rxr->bytes == 0))
915 goto no_calc;
916
917 if ((txr->bytes) && (txr->packets))
918 newitr = txr->bytes/txr->packets;
919 if ((rxr->bytes) && (rxr->packets))
920 newitr = uimax(newitr, (rxr->bytes / rxr->packets));
921 newitr += 24; /* account for hardware frame, crc */
922
923 /* set an upper boundary */
924 newitr = uimin(newitr, 3000);
925
926 /* Be nice to the mid range */
927 if ((newitr > 300) && (newitr < 1200))
928 newitr = (newitr / 3);
929 else
930 newitr = (newitr / 2);
931
932 /*
933 * When RSC is used, ITR interval must be larger than RSC_DELAY.
934 * Currently, we use 2us for RSC_DELAY. The minimum value is always
935 * greater than 2us on 100M (and 10M?(not documented)), but it's not
936 * on 1G and higher.
937 */
938 if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
939 && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
940 if (newitr < IXGBE_MIN_RSC_EITR_10G1G)
941 newitr = IXGBE_MIN_RSC_EITR_10G1G;
942 }
943
944 /* save for next interrupt */
945 que->eitr_setting = newitr;
946
947 /* Reset state */
948 txr->bytes = 0;
949 txr->packets = 0;
950 rxr->bytes = 0;
951 rxr->packets = 0;
952
953 no_calc:
954 if (more)
955 softint_schedule(que->que_si);
956 else /* Re-enable this interrupt */
957 ixv_enable_queue(adapter, que->msix);
958
959 return 1;
960 } /* ixv_msix_que */
961
962 /************************************************************************
963 * ixv_msix_mbx
964 ************************************************************************/
965 static int
966 ixv_msix_mbx(void *arg)
967 {
968 struct adapter *adapter = arg;
969 struct ixgbe_hw *hw = &adapter->hw;
970
971 ++adapter->link_irq.ev_count;
972 /* NetBSD: We use auto-clear, so it's not required to write VTEICR */
973
974 /* Link status change */
975 hw->mac.get_link_status = TRUE;
976 softint_schedule(adapter->link_si);
977
978 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
979
980 return 1;
981 } /* ixv_msix_mbx */
982
983 static void
984 ixv_eitr_write(struct adapter *adapter, uint32_t index, uint32_t itr)
985 {
986
987 /*
988 * Newer devices than 82598 have VF function, so this function is
989 * simple.
990 */
991 itr |= IXGBE_EITR_CNT_WDIS;
992
993 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEITR(index), itr);
994 }
995
996
997 /************************************************************************
998 * ixv_media_status - Media Ioctl callback
999 *
1000 * Called whenever the user queries the status of
1001 * the interface using ifconfig.
1002 ************************************************************************/
1003 static void
1004 ixv_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1005 {
1006 struct adapter *adapter = ifp->if_softc;
1007
1008 INIT_DEBUGOUT("ixv_media_status: begin");
1009 IXGBE_CORE_LOCK(adapter);
1010 ixv_update_link_status(adapter);
1011
1012 ifmr->ifm_status = IFM_AVALID;
1013 ifmr->ifm_active = IFM_ETHER;
1014
1015 if (adapter->link_active != LINK_STATE_UP) {
1016 ifmr->ifm_active |= IFM_NONE;
1017 IXGBE_CORE_UNLOCK(adapter);
1018 return;
1019 }
1020
1021 ifmr->ifm_status |= IFM_ACTIVE;
1022
1023 switch (adapter->link_speed) {
1024 case IXGBE_LINK_SPEED_10GB_FULL:
1025 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
1026 break;
1027 case IXGBE_LINK_SPEED_5GB_FULL:
1028 ifmr->ifm_active |= IFM_5000_T | IFM_FDX;
1029 break;
1030 case IXGBE_LINK_SPEED_2_5GB_FULL:
1031 ifmr->ifm_active |= IFM_2500_T | IFM_FDX;
1032 break;
1033 case IXGBE_LINK_SPEED_1GB_FULL:
1034 ifmr->ifm_active |= IFM_1000_T | IFM_FDX;
1035 break;
1036 case IXGBE_LINK_SPEED_100_FULL:
1037 ifmr->ifm_active |= IFM_100_TX | IFM_FDX;
1038 break;
1039 case IXGBE_LINK_SPEED_10_FULL:
1040 ifmr->ifm_active |= IFM_10_T | IFM_FDX;
1041 break;
1042 }
1043
1044 ifp->if_baudrate = ifmedia_baudrate(ifmr->ifm_active);
1045
1046 IXGBE_CORE_UNLOCK(adapter);
1047 } /* ixv_media_status */
1048
1049 /************************************************************************
1050 * ixv_media_change - Media Ioctl callback
1051 *
1052 * Called when the user changes speed/duplex using
1053 * media/mediopt option with ifconfig.
1054 ************************************************************************/
1055 static int
1056 ixv_media_change(struct ifnet *ifp)
1057 {
1058 struct adapter *adapter = ifp->if_softc;
1059 struct ifmedia *ifm = &adapter->media;
1060
1061 INIT_DEBUGOUT("ixv_media_change: begin");
1062
1063 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1064 return (EINVAL);
1065
1066 switch (IFM_SUBTYPE(ifm->ifm_media)) {
1067 case IFM_AUTO:
1068 break;
1069 default:
1070 device_printf(adapter->dev, "Only auto media type\n");
1071 return (EINVAL);
1072 }
1073
1074 return (0);
1075 } /* ixv_media_change */
1076
1077
1078 /************************************************************************
1079 * ixv_negotiate_api
1080 *
1081 * Negotiate the Mailbox API with the PF;
1082 * start with the most featured API first.
1083 ************************************************************************/
1084 static int
1085 ixv_negotiate_api(struct adapter *adapter)
1086 {
1087 struct ixgbe_hw *hw = &adapter->hw;
1088 int mbx_api[] = { ixgbe_mbox_api_11,
1089 ixgbe_mbox_api_10,
1090 ixgbe_mbox_api_unknown };
1091 int i = 0;
1092
1093 while (mbx_api[i] != ixgbe_mbox_api_unknown) {
1094 if (ixgbevf_negotiate_api_version(hw, mbx_api[i]) == 0)
1095 return (0);
1096 i++;
1097 }
1098
1099 return (EINVAL);
1100 } /* ixv_negotiate_api */
1101
1102
1103 /************************************************************************
1104 * ixv_set_multi - Multicast Update
1105 *
1106 * Called whenever multicast address list is updated.
1107 ************************************************************************/
1108 static void
1109 ixv_set_multi(struct adapter *adapter)
1110 {
1111 struct ether_multi *enm;
1112 struct ether_multistep step;
1113 struct ethercom *ec = &adapter->osdep.ec;
1114 u8 mta[IXGBE_MAX_VF_MC * IXGBE_ETH_LENGTH_OF_ADDRESS];
1115 u8 *update_ptr;
1116 int mcnt = 0;
1117
1118 KASSERT(mutex_owned(&adapter->core_mtx));
1119 IOCTL_DEBUGOUT("ixv_set_multi: begin");
1120
1121 ETHER_LOCK(ec);
1122 ETHER_FIRST_MULTI(step, ec, enm);
1123 while (enm != NULL) {
1124 bcopy(enm->enm_addrlo,
1125 &mta[mcnt * IXGBE_ETH_LENGTH_OF_ADDRESS],
1126 IXGBE_ETH_LENGTH_OF_ADDRESS);
1127 mcnt++;
1128 /* XXX This might be required --msaitoh */
1129 if (mcnt >= IXGBE_MAX_VF_MC)
1130 break;
1131 ETHER_NEXT_MULTI(step, enm);
1132 }
1133 ETHER_UNLOCK(ec);
1134
1135 update_ptr = mta;
1136
1137 adapter->hw.mac.ops.update_mc_addr_list(&adapter->hw, update_ptr, mcnt,
1138 ixv_mc_array_itr, TRUE);
1139 } /* ixv_set_multi */
1140
1141 /************************************************************************
1142 * ixv_mc_array_itr
1143 *
1144 * An iterator function needed by the multicast shared code.
1145 * It feeds the shared code routine the addresses in the
1146 * array of ixv_set_multi() one by one.
1147 ************************************************************************/
1148 static u8 *
1149 ixv_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq)
1150 {
1151 u8 *addr = *update_ptr;
1152 u8 *newptr;
1153
1154 *vmdq = 0;
1155
1156 newptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
1157 *update_ptr = newptr;
1158
1159 return addr;
1160 } /* ixv_mc_array_itr */
1161
1162 /************************************************************************
1163 * ixv_local_timer - Timer routine
1164 *
1165 * Checks for link status, updates statistics,
1166 * and runs the watchdog check.
1167 ************************************************************************/
1168 static void
1169 ixv_local_timer(void *arg)
1170 {
1171 struct adapter *adapter = arg;
1172
1173 IXGBE_CORE_LOCK(adapter);
1174 ixv_local_timer_locked(adapter);
1175 IXGBE_CORE_UNLOCK(adapter);
1176 }
1177
1178 static void
1179 ixv_local_timer_locked(void *arg)
1180 {
1181 struct adapter *adapter = arg;
1182 device_t dev = adapter->dev;
1183 struct ix_queue *que = adapter->queues;
1184 u64 queues = 0;
1185 u64 v0, v1, v2, v3, v4, v5, v6, v7;
1186 int hung = 0;
1187 int i;
1188
1189 KASSERT(mutex_owned(&adapter->core_mtx));
1190
1191 if (ixv_check_link(adapter)) {
1192 ixv_init_locked(adapter);
1193 return;
1194 }
1195
1196 /* Stats Update */
1197 ixv_update_stats(adapter);
1198
1199 /* Update some event counters */
1200 v0 = v1 = v2 = v3 = v4 = v5 = v6 = v7 = 0;
1201 que = adapter->queues;
1202 for (i = 0; i < adapter->num_queues; i++, que++) {
1203 struct tx_ring *txr = que->txr;
1204
1205 v0 += txr->q_efbig_tx_dma_setup;
1206 v1 += txr->q_mbuf_defrag_failed;
1207 v2 += txr->q_efbig2_tx_dma_setup;
1208 v3 += txr->q_einval_tx_dma_setup;
1209 v4 += txr->q_other_tx_dma_setup;
1210 v5 += txr->q_eagain_tx_dma_setup;
1211 v6 += txr->q_enomem_tx_dma_setup;
1212 v7 += txr->q_tso_err;
1213 }
1214 adapter->efbig_tx_dma_setup.ev_count = v0;
1215 adapter->mbuf_defrag_failed.ev_count = v1;
1216 adapter->efbig2_tx_dma_setup.ev_count = v2;
1217 adapter->einval_tx_dma_setup.ev_count = v3;
1218 adapter->other_tx_dma_setup.ev_count = v4;
1219 adapter->eagain_tx_dma_setup.ev_count = v5;
1220 adapter->enomem_tx_dma_setup.ev_count = v6;
1221 adapter->tso_err.ev_count = v7;
1222
1223 /*
1224 * Check the TX queues status
1225 * - mark hung queues so we don't schedule on them
1226 * - watchdog only if all queues show hung
1227 */
1228 que = adapter->queues;
1229 for (i = 0; i < adapter->num_queues; i++, que++) {
1230 /* Keep track of queues with work for soft irq */
1231 if (que->txr->busy)
1232 queues |= ((u64)1 << que->me);
1233 /*
1234 * Each time txeof runs without cleaning, but there
1235 * are uncleaned descriptors it increments busy. If
1236 * we get to the MAX we declare it hung.
1237 */
1238 if (que->busy == IXGBE_QUEUE_HUNG) {
1239 ++hung;
1240 /* Mark the queue as inactive */
1241 adapter->active_queues &= ~((u64)1 << que->me);
1242 continue;
1243 } else {
1244 /* Check if we've come back from hung */
1245 if ((adapter->active_queues & ((u64)1 << que->me)) == 0)
1246 adapter->active_queues |= ((u64)1 << que->me);
1247 }
1248 if (que->busy >= IXGBE_MAX_TX_BUSY) {
1249 device_printf(dev,
1250 "Warning queue %d appears to be hung!\n", i);
1251 que->txr->busy = IXGBE_QUEUE_HUNG;
1252 ++hung;
1253 }
1254 }
1255
1256 /* Only truly watchdog if all queues show hung */
1257 if (hung == adapter->num_queues)
1258 goto watchdog;
1259 #if 0
1260 else if (queues != 0) { /* Force an IRQ on queues with work */
1261 ixv_rearm_queues(adapter, queues);
1262 }
1263 #endif
1264
1265 callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
1266
1267 return;
1268
1269 watchdog:
1270
1271 device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
1272 adapter->ifp->if_flags &= ~IFF_RUNNING;
1273 adapter->watchdog_events.ev_count++;
1274 ixv_init_locked(adapter);
1275 } /* ixv_local_timer */
1276
1277 /************************************************************************
1278 * ixv_update_link_status - Update OS on link state
1279 *
1280 * Note: Only updates the OS on the cached link state.
1281 * The real check of the hardware only happens with
1282 * a link interrupt.
1283 ************************************************************************/
1284 static void
1285 ixv_update_link_status(struct adapter *adapter)
1286 {
1287 struct ifnet *ifp = adapter->ifp;
1288 device_t dev = adapter->dev;
1289
1290 KASSERT(mutex_owned(&adapter->core_mtx));
1291
1292 if (adapter->link_up) {
1293 if (adapter->link_active != LINK_STATE_UP) {
1294 if (bootverbose) {
1295 const char *bpsmsg;
1296
1297 switch (adapter->link_speed) {
1298 case IXGBE_LINK_SPEED_10GB_FULL:
1299 bpsmsg = "10 Gbps";
1300 break;
1301 case IXGBE_LINK_SPEED_5GB_FULL:
1302 bpsmsg = "5 Gbps";
1303 break;
1304 case IXGBE_LINK_SPEED_2_5GB_FULL:
1305 bpsmsg = "2.5 Gbps";
1306 break;
1307 case IXGBE_LINK_SPEED_1GB_FULL:
1308 bpsmsg = "1 Gbps";
1309 break;
1310 case IXGBE_LINK_SPEED_100_FULL:
1311 bpsmsg = "100 Mbps";
1312 break;
1313 case IXGBE_LINK_SPEED_10_FULL:
1314 bpsmsg = "10 Mbps";
1315 break;
1316 default:
1317 bpsmsg = "unknown speed";
1318 break;
1319 }
1320 device_printf(dev, "Link is up %s %s \n",
1321 bpsmsg, "Full Duplex");
1322 }
1323 adapter->link_active = LINK_STATE_UP;
1324 if_link_state_change(ifp, LINK_STATE_UP);
1325 }
1326 } else {
1327 /*
1328 * Do it when link active changes to DOWN. i.e.
1329 * a) LINK_STATE_UNKNOWN -> LINK_STATE_DOWN
1330 * b) LINK_STATE_UP -> LINK_STATE_DOWN
1331 */
1332 if (adapter->link_active != LINK_STATE_DOWN) {
1333 if (bootverbose)
1334 device_printf(dev, "Link is Down\n");
1335 if_link_state_change(ifp, LINK_STATE_DOWN);
1336 adapter->link_active = LINK_STATE_DOWN;
1337 }
1338 }
1339 } /* ixv_update_link_status */
1340
1341
1342 /************************************************************************
1343 * ixv_stop - Stop the hardware
1344 *
1345 * Disables all traffic on the adapter by issuing a
1346 * global reset on the MAC and deallocates TX/RX buffers.
1347 ************************************************************************/
1348 static void
1349 ixv_ifstop(struct ifnet *ifp, int disable)
1350 {
1351 struct adapter *adapter = ifp->if_softc;
1352
1353 IXGBE_CORE_LOCK(adapter);
1354 ixv_stop(adapter);
1355 IXGBE_CORE_UNLOCK(adapter);
1356 }
1357
1358 static void
1359 ixv_stop(void *arg)
1360 {
1361 struct ifnet *ifp;
1362 struct adapter *adapter = arg;
1363 struct ixgbe_hw *hw = &adapter->hw;
1364
1365 ifp = adapter->ifp;
1366
1367 KASSERT(mutex_owned(&adapter->core_mtx));
1368
1369 INIT_DEBUGOUT("ixv_stop: begin\n");
1370 ixv_disable_intr(adapter);
1371
1372 /* Tell the stack that the interface is no longer active */
1373 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1374
1375 hw->mac.ops.reset_hw(hw);
1376 adapter->hw.adapter_stopped = FALSE;
1377 hw->mac.ops.stop_adapter(hw);
1378 callout_stop(&adapter->timer);
1379
1380 /* reprogram the RAR[0] in case user changed it. */
1381 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
1382
1383 return;
1384 } /* ixv_stop */
1385
1386
1387 /************************************************************************
1388 * ixv_allocate_pci_resources
1389 ************************************************************************/
1390 static int
1391 ixv_allocate_pci_resources(struct adapter *adapter,
1392 const struct pci_attach_args *pa)
1393 {
1394 pcireg_t memtype, csr;
1395 device_t dev = adapter->dev;
1396 bus_addr_t addr;
1397 int flags;
1398
1399 memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, PCI_BAR(0));
1400 switch (memtype) {
1401 case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
1402 case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
1403 adapter->osdep.mem_bus_space_tag = pa->pa_memt;
1404 if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, PCI_BAR(0),
1405 memtype, &addr, &adapter->osdep.mem_size, &flags) != 0)
1406 goto map_err;
1407 if ((flags & BUS_SPACE_MAP_PREFETCHABLE) != 0) {
1408 aprint_normal_dev(dev, "clearing prefetchable bit\n");
1409 flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
1410 }
1411 if (bus_space_map(adapter->osdep.mem_bus_space_tag, addr,
1412 adapter->osdep.mem_size, flags,
1413 &adapter->osdep.mem_bus_space_handle) != 0) {
1414 map_err:
1415 adapter->osdep.mem_size = 0;
1416 aprint_error_dev(dev, "unable to map BAR0\n");
1417 return ENXIO;
1418 }
1419 /*
1420 * Enable address decoding for memory range in case it's not
1421 * set.
1422 */
1423 csr = pci_conf_read(pa->pa_pc, pa->pa_tag,
1424 PCI_COMMAND_STATUS_REG);
1425 csr |= PCI_COMMAND_MEM_ENABLE;
1426 pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
1427 csr);
1428 break;
1429 default:
1430 aprint_error_dev(dev, "unexpected type on BAR0\n");
1431 return ENXIO;
1432 }
1433
1434 /* Pick up the tuneable queues */
1435 adapter->num_queues = ixv_num_queues;
1436
1437 return (0);
1438 } /* ixv_allocate_pci_resources */
1439
1440 /************************************************************************
1441 * ixv_free_pci_resources
1442 ************************************************************************/
1443 static void
1444 ixv_free_pci_resources(struct adapter * adapter)
1445 {
1446 struct ix_queue *que = adapter->queues;
1447 int rid;
1448
1449 /*
1450 * Release all msix queue resources:
1451 */
1452 for (int i = 0; i < adapter->num_queues; i++, que++) {
1453 if (que->res != NULL)
1454 pci_intr_disestablish(adapter->osdep.pc,
1455 adapter->osdep.ihs[i]);
1456 }
1457
1458
1459 /* Clean the Mailbox interrupt last */
1460 rid = adapter->vector;
1461
1462 if (adapter->osdep.ihs[rid] != NULL) {
1463 pci_intr_disestablish(adapter->osdep.pc,
1464 adapter->osdep.ihs[rid]);
1465 adapter->osdep.ihs[rid] = NULL;
1466 }
1467
1468 pci_intr_release(adapter->osdep.pc, adapter->osdep.intrs,
1469 adapter->osdep.nintrs);
1470
1471 if (adapter->osdep.mem_size != 0) {
1472 bus_space_unmap(adapter->osdep.mem_bus_space_tag,
1473 adapter->osdep.mem_bus_space_handle,
1474 adapter->osdep.mem_size);
1475 }
1476
1477 return;
1478 } /* ixv_free_pci_resources */
1479
1480 /************************************************************************
1481 * ixv_setup_interface
1482 *
1483 * Setup networking device structure and register an interface.
1484 ************************************************************************/
1485 static int
1486 ixv_setup_interface(device_t dev, struct adapter *adapter)
1487 {
1488 struct ethercom *ec = &adapter->osdep.ec;
1489 struct ifnet *ifp;
1490 int rv;
1491
1492 INIT_DEBUGOUT("ixv_setup_interface: begin");
1493
1494 ifp = adapter->ifp = &ec->ec_if;
1495 strlcpy(ifp->if_xname, device_xname(dev), IFNAMSIZ);
1496 ifp->if_baudrate = IF_Gbps(10);
1497 ifp->if_init = ixv_init;
1498 ifp->if_stop = ixv_ifstop;
1499 ifp->if_softc = adapter;
1500 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1501 #ifdef IXGBE_MPSAFE
1502 ifp->if_extflags = IFEF_MPSAFE;
1503 #endif
1504 ifp->if_ioctl = ixv_ioctl;
1505 if (adapter->feat_en & IXGBE_FEATURE_LEGACY_TX) {
1506 #if 0
1507 ixv_start_locked = ixgbe_legacy_start_locked;
1508 #endif
1509 } else {
1510 ifp->if_transmit = ixgbe_mq_start;
1511 #if 0
1512 ixv_start_locked = ixgbe_mq_start_locked;
1513 #endif
1514 }
1515 ifp->if_start = ixgbe_legacy_start;
1516 IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 2);
1517 IFQ_SET_READY(&ifp->if_snd);
1518
1519 rv = if_initialize(ifp);
1520 if (rv != 0) {
1521 aprint_error_dev(dev, "if_initialize failed(%d)\n", rv);
1522 return rv;
1523 }
1524 adapter->ipq = if_percpuq_create(&adapter->osdep.ec.ec_if);
1525 ether_ifattach(ifp, adapter->hw.mac.addr);
1526 /*
1527 * We use per TX queue softint, so if_deferred_start_init() isn't
1528 * used.
1529 */
1530 ether_set_ifflags_cb(ec, ixv_ifflags_cb);
1531
1532 adapter->max_frame_size = ifp->if_mtu + IXGBE_MTU_HDR;
1533
1534 /*
1535 * Tell the upper layer(s) we support long frames.
1536 */
1537 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1538
1539 /* Set capability flags */
1540 ifp->if_capabilities |= IFCAP_HWCSUM
1541 | IFCAP_TSOv4
1542 | IFCAP_TSOv6;
1543 ifp->if_capenable = 0;
1544
1545 ec->ec_capabilities |= ETHERCAP_VLAN_HWFILTER
1546 | ETHERCAP_VLAN_HWTAGGING
1547 | ETHERCAP_VLAN_HWCSUM
1548 | ETHERCAP_JUMBO_MTU
1549 | ETHERCAP_VLAN_MTU;
1550
1551 /* Enable the above capabilities by default */
1552 ec->ec_capenable = ec->ec_capabilities;
1553
1554 /* Don't enable LRO by default */
1555 #if 0
1556 /* NetBSD doesn't support LRO yet */
1557 ifp->if_capabilities |= IFCAP_LRO;
1558 #endif
1559
1560 /*
1561 * Specify the media types supported by this adapter and register
1562 * callbacks to update media and link information
1563 */
1564 ec->ec_ifmedia = &adapter->media;
1565 ifmedia_init(&adapter->media, IFM_IMASK, ixv_media_change,
1566 ixv_media_status);
1567 ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1568 ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
1569
1570 if_register(ifp);
1571
1572 return 0;
1573 } /* ixv_setup_interface */
1574
1575
1576 /************************************************************************
1577 * ixv_initialize_transmit_units - Enable transmit unit.
1578 ************************************************************************/
1579 static void
1580 ixv_initialize_transmit_units(struct adapter *adapter)
1581 {
1582 struct tx_ring *txr = adapter->tx_rings;
1583 struct ixgbe_hw *hw = &adapter->hw;
1584 int i;
1585
1586 for (i = 0; i < adapter->num_queues; i++, txr++) {
1587 u64 tdba = txr->txdma.dma_paddr;
1588 u32 txctrl, txdctl;
1589 int j = txr->me;
1590
1591 /* Set WTHRESH to 8, burst writeback */
1592 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1593 txdctl |= (8 << 16);
1594 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1595
1596 /* Set the HW Tx Head and Tail indices */
1597 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDH(j), 0);
1598 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDT(j), 0);
1599
1600 /* Set Tx Tail register */
1601 txr->tail = IXGBE_VFTDT(j);
1602
1603 txr->txr_no_space = false;
1604
1605 /* Set Ring parameters */
1606 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1607 (tdba & 0x00000000ffffffffULL));
1608 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1609 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j),
1610 adapter->num_tx_desc * sizeof(struct ixgbe_legacy_tx_desc));
1611 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1612 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
1613 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1614
1615 /* Now enable */
1616 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1617 txdctl |= IXGBE_TXDCTL_ENABLE;
1618 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1619 }
1620
1621 return;
1622 } /* ixv_initialize_transmit_units */
1623
1624
1625 /************************************************************************
1626 * ixv_initialize_rss_mapping
1627 ************************************************************************/
1628 static void
1629 ixv_initialize_rss_mapping(struct adapter *adapter)
1630 {
1631 struct ixgbe_hw *hw = &adapter->hw;
1632 u32 reta = 0, mrqc, rss_key[10];
1633 int queue_id;
1634 int i, j;
1635 u32 rss_hash_config;
1636
1637 /* force use default RSS key. */
1638 #ifdef __NetBSD__
1639 rss_getkey((uint8_t *) &rss_key);
1640 #else
1641 if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1642 /* Fetch the configured RSS key */
1643 rss_getkey((uint8_t *)&rss_key);
1644 } else {
1645 /* set up random bits */
1646 cprng_fast(&rss_key, sizeof(rss_key));
1647 }
1648 #endif
1649
1650 /* Now fill out hash function seeds */
1651 for (i = 0; i < 10; i++)
1652 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1653
1654 /* Set up the redirection table */
1655 for (i = 0, j = 0; i < 64; i++, j++) {
1656 if (j == adapter->num_queues)
1657 j = 0;
1658
1659 if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1660 /*
1661 * Fetch the RSS bucket id for the given indirection
1662 * entry. Cap it at the number of configured buckets
1663 * (which is num_queues.)
1664 */
1665 queue_id = rss_get_indirection_to_bucket(i);
1666 queue_id = queue_id % adapter->num_queues;
1667 } else
1668 queue_id = j;
1669
1670 /*
1671 * The low 8 bits are for hash value (n+0);
1672 * The next 8 bits are for hash value (n+1), etc.
1673 */
1674 reta >>= 8;
1675 reta |= ((uint32_t)queue_id) << 24;
1676 if ((i & 3) == 3) {
1677 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), reta);
1678 reta = 0;
1679 }
1680 }
1681
1682 /* Perform hash on these packet types */
1683 if (adapter->feat_en & IXGBE_FEATURE_RSS)
1684 rss_hash_config = rss_gethashconfig();
1685 else {
1686 /*
1687 * Disable UDP - IP fragments aren't currently being handled
1688 * and so we end up with a mix of 2-tuple and 4-tuple
1689 * traffic.
1690 */
1691 rss_hash_config = RSS_HASHTYPE_RSS_IPV4
1692 | RSS_HASHTYPE_RSS_TCP_IPV4
1693 | RSS_HASHTYPE_RSS_IPV6
1694 | RSS_HASHTYPE_RSS_TCP_IPV6;
1695 }
1696
1697 mrqc = IXGBE_MRQC_RSSEN;
1698 if (rss_hash_config & RSS_HASHTYPE_RSS_IPV4)
1699 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
1700 if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV4)
1701 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
1702 if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6)
1703 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
1704 if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6)
1705 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
1706 if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6_EX)
1707 device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_IPV6_EX defined, but not supported\n",
1708 __func__);
1709 if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6_EX)
1710 device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_TCP_IPV6_EX defined, but not supported\n",
1711 __func__);
1712 if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV4)
1713 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
1714 if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6)
1715 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1716 if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6_EX)
1717 device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_UDP_IPV6_EX defined, but not supported\n",
1718 __func__);
1719 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, mrqc);
1720 } /* ixv_initialize_rss_mapping */
1721
1722
1723 /************************************************************************
1724 * ixv_initialize_receive_units - Setup receive registers and features.
1725 ************************************************************************/
1726 static void
1727 ixv_initialize_receive_units(struct adapter *adapter)
1728 {
1729 struct rx_ring *rxr = adapter->rx_rings;
1730 struct ixgbe_hw *hw = &adapter->hw;
1731 struct ifnet *ifp = adapter->ifp;
1732 u32 bufsz, psrtype;
1733
1734 if (ifp->if_mtu > ETHERMTU)
1735 bufsz = 4096 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1736 else
1737 bufsz = 2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1738
1739 psrtype = IXGBE_PSRTYPE_TCPHDR
1740 | IXGBE_PSRTYPE_UDPHDR
1741 | IXGBE_PSRTYPE_IPV4HDR
1742 | IXGBE_PSRTYPE_IPV6HDR
1743 | IXGBE_PSRTYPE_L2HDR;
1744
1745 if (adapter->num_queues > 1)
1746 psrtype |= 1 << 29;
1747
1748 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1749
1750 /* Tell PF our max_frame size */
1751 if (ixgbevf_rlpml_set_vf(hw, adapter->max_frame_size) != 0) {
1752 device_printf(adapter->dev, "There is a problem with the PF setup. It is likely the receive unit for this VF will not function correctly.\n");
1753 }
1754
1755 for (int i = 0; i < adapter->num_queues; i++, rxr++) {
1756 u64 rdba = rxr->rxdma.dma_paddr;
1757 u32 reg, rxdctl;
1758 int j = rxr->me;
1759
1760 /* Disable the queue */
1761 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1762 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1763 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1764 for (int k = 0; k < 10; k++) {
1765 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1766 IXGBE_RXDCTL_ENABLE)
1767 msec_delay(1);
1768 else
1769 break;
1770 }
1771 wmb();
1772 /* Setup the Base and Length of the Rx Descriptor Ring */
1773 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1774 (rdba & 0x00000000ffffffffULL));
1775 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1776 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j),
1777 adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc));
1778
1779 /* Reset the ring indices */
1780 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
1781 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), 0);
1782
1783 /* Set up the SRRCTL register */
1784 reg = IXGBE_READ_REG(hw, IXGBE_VFSRRCTL(j));
1785 reg &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1786 reg &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1787 reg |= bufsz;
1788 reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1789 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(j), reg);
1790
1791 /* Capture Rx Tail index */
1792 rxr->tail = IXGBE_VFRDT(rxr->me);
1793
1794 /* Do the queue enabling last */
1795 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1796 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1797 for (int k = 0; k < 10; k++) {
1798 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1799 IXGBE_RXDCTL_ENABLE)
1800 break;
1801 msec_delay(1);
1802 }
1803 wmb();
1804
1805 /* Set the Tail Pointer */
1806 #ifdef DEV_NETMAP
1807 /*
1808 * In netmap mode, we must preserve the buffers made
1809 * available to userspace before the if_init()
1810 * (this is true by default on the TX side, because
1811 * init makes all buffers available to userspace).
1812 *
1813 * netmap_reset() and the device specific routines
1814 * (e.g. ixgbe_setup_receive_rings()) map these
1815 * buffers at the end of the NIC ring, so here we
1816 * must set the RDT (tail) register to make sure
1817 * they are not overwritten.
1818 *
1819 * In this driver the NIC ring starts at RDH = 0,
1820 * RDT points to the last slot available for reception (?),
1821 * so RDT = num_rx_desc - 1 means the whole ring is available.
1822 */
1823 if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) &&
1824 (ifp->if_capenable & IFCAP_NETMAP)) {
1825 struct netmap_adapter *na = NA(adapter->ifp);
1826 struct netmap_kring *kring = na->rx_rings[i];
1827 int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring);
1828
1829 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), t);
1830 } else
1831 #endif /* DEV_NETMAP */
1832 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
1833 adapter->num_rx_desc - 1);
1834 }
1835
1836 ixv_initialize_rss_mapping(adapter);
1837 } /* ixv_initialize_receive_units */
1838
1839 /************************************************************************
1840 * ixv_sysctl_tdh_handler - Transmit Descriptor Head handler function
1841 *
1842 * Retrieves the TDH value from the hardware
1843 ************************************************************************/
1844 static int
1845 ixv_sysctl_tdh_handler(SYSCTLFN_ARGS)
1846 {
1847 struct sysctlnode node = *rnode;
1848 struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1849 uint32_t val;
1850
1851 if (!txr)
1852 return (0);
1853
1854 val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDH(txr->me));
1855 node.sysctl_data = &val;
1856 return sysctl_lookup(SYSCTLFN_CALL(&node));
1857 } /* ixv_sysctl_tdh_handler */
1858
1859 /************************************************************************
1860 * ixgbe_sysctl_tdt_handler - Transmit Descriptor Tail handler function
1861 *
1862 * Retrieves the TDT value from the hardware
1863 ************************************************************************/
1864 static int
1865 ixv_sysctl_tdt_handler(SYSCTLFN_ARGS)
1866 {
1867 struct sysctlnode node = *rnode;
1868 struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1869 uint32_t val;
1870
1871 if (!txr)
1872 return (0);
1873
1874 val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDT(txr->me));
1875 node.sysctl_data = &val;
1876 return sysctl_lookup(SYSCTLFN_CALL(&node));
1877 } /* ixv_sysctl_tdt_handler */
1878
1879 /************************************************************************
1880 * ixv_sysctl_next_to_check_handler - Receive Descriptor next to check
1881 * handler function
1882 *
1883 * Retrieves the next_to_check value
1884 ************************************************************************/
1885 static int
1886 ixv_sysctl_next_to_check_handler(SYSCTLFN_ARGS)
1887 {
1888 struct sysctlnode node = *rnode;
1889 struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1890 uint32_t val;
1891
1892 if (!rxr)
1893 return (0);
1894
1895 val = rxr->next_to_check;
1896 node.sysctl_data = &val;
1897 return sysctl_lookup(SYSCTLFN_CALL(&node));
1898 } /* ixv_sysctl_next_to_check_handler */
1899
1900 /************************************************************************
1901 * ixv_sysctl_rdh_handler - Receive Descriptor Head handler function
1902 *
1903 * Retrieves the RDH value from the hardware
1904 ************************************************************************/
1905 static int
1906 ixv_sysctl_rdh_handler(SYSCTLFN_ARGS)
1907 {
1908 struct sysctlnode node = *rnode;
1909 struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1910 uint32_t val;
1911
1912 if (!rxr)
1913 return (0);
1914
1915 val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDH(rxr->me));
1916 node.sysctl_data = &val;
1917 return sysctl_lookup(SYSCTLFN_CALL(&node));
1918 } /* ixv_sysctl_rdh_handler */
1919
1920 /************************************************************************
1921 * ixv_sysctl_rdt_handler - Receive Descriptor Tail handler function
1922 *
1923 * Retrieves the RDT value from the hardware
1924 ************************************************************************/
1925 static int
1926 ixv_sysctl_rdt_handler(SYSCTLFN_ARGS)
1927 {
1928 struct sysctlnode node = *rnode;
1929 struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1930 uint32_t val;
1931
1932 if (!rxr)
1933 return (0);
1934
1935 val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDT(rxr->me));
1936 node.sysctl_data = &val;
1937 return sysctl_lookup(SYSCTLFN_CALL(&node));
1938 } /* ixv_sysctl_rdt_handler */
1939
1940 static void
1941 ixv_setup_vlan_tagging(struct adapter *adapter)
1942 {
1943 struct ethercom *ec = &adapter->osdep.ec;
1944 struct ixgbe_hw *hw = &adapter->hw;
1945 struct rx_ring *rxr;
1946 u32 ctrl;
1947 int i;
1948 bool hwtagging;
1949
1950 /* Enable HW tagging only if any vlan is attached */
1951 hwtagging = (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING)
1952 && VLAN_ATTACHED(ec);
1953
1954 /* Enable the queues */
1955 for (i = 0; i < adapter->num_queues; i++) {
1956 rxr = &adapter->rx_rings[i];
1957 ctrl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(rxr->me));
1958 if (hwtagging)
1959 ctrl |= IXGBE_RXDCTL_VME;
1960 else
1961 ctrl &= ~IXGBE_RXDCTL_VME;
1962 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(rxr->me), ctrl);
1963 /*
1964 * Let Rx path know that it needs to store VLAN tag
1965 * as part of extra mbuf info.
1966 */
1967 rxr->vtag_strip = hwtagging ? TRUE : FALSE;
1968 }
1969 } /* ixv_setup_vlan_tagging */
1970
1971 /************************************************************************
1972 * ixv_setup_vlan_support
1973 ************************************************************************/
1974 static int
1975 ixv_setup_vlan_support(struct adapter *adapter)
1976 {
1977 struct ethercom *ec = &adapter->osdep.ec;
1978 struct ixgbe_hw *hw = &adapter->hw;
1979 u32 vid, vfta, retry;
1980 struct vlanid_list *vlanidp;
1981 int rv, error = 0;
1982
1983 /*
1984 * This function is called from both if_init and ifflags_cb()
1985 * on NetBSD.
1986 */
1987
1988 /*
1989 * Part 1:
1990 * Setup VLAN HW tagging
1991 */
1992 ixv_setup_vlan_tagging(adapter);
1993
1994 if (!VLAN_ATTACHED(ec))
1995 return 0;
1996
1997 /*
1998 * Part 2:
1999 * Setup VLAN HW filter
2000 */
2001 /* Cleanup shadow_vfta */
2002 for (int i = 0; i < IXGBE_VFTA_SIZE; i++)
2003 adapter->shadow_vfta[i] = 0;
2004 /* Generate shadow_vfta from ec_vids */
2005 ETHER_LOCK(ec);
2006 SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
2007 uint32_t idx;
2008
2009 idx = vlanidp->vid / 32;
2010 KASSERT(idx < IXGBE_VFTA_SIZE);
2011 adapter->shadow_vfta[idx] |= (u32)1 << (vlanidp->vid % 32);
2012 }
2013 ETHER_UNLOCK(ec);
2014
2015 /*
2016 * A soft reset zero's out the VFTA, so
2017 * we need to repopulate it now.
2018 */
2019 for (int i = 0; i < IXGBE_VFTA_SIZE; i++) {
2020 if (adapter->shadow_vfta[i] == 0)
2021 continue;
2022 vfta = adapter->shadow_vfta[i];
2023 /*
2024 * Reconstruct the vlan id's
2025 * based on the bits set in each
2026 * of the array ints.
2027 */
2028 for (int j = 0; j < 32; j++) {
2029 retry = 0;
2030 if ((vfta & ((u32)1 << j)) == 0)
2031 continue;
2032 vid = (i * 32) + j;
2033
2034 /* Call the shared code mailbox routine */
2035 while ((rv = hw->mac.ops.set_vfta(hw, vid, 0, TRUE,
2036 FALSE)) != 0) {
2037 if (++retry > 5) {
2038 device_printf(adapter->dev,
2039 "%s: max retry exceeded\n",
2040 __func__);
2041 break;
2042 }
2043 }
2044 if (rv != 0) {
2045 device_printf(adapter->dev,
2046 "failed to set vlan %d\n", vid);
2047 error = EACCES;
2048 }
2049 }
2050 }
2051 return error;
2052 } /* ixv_setup_vlan_support */
2053
2054 static int
2055 ixv_vlan_cb(struct ethercom *ec, uint16_t vid, bool set)
2056 {
2057 struct ifnet *ifp = &ec->ec_if;
2058 struct adapter *adapter = ifp->if_softc;
2059 int rv;
2060
2061 if (set)
2062 rv = ixv_register_vlan(adapter, vid);
2063 else
2064 rv = ixv_unregister_vlan(adapter, vid);
2065
2066 if (rv != 0)
2067 return rv;
2068
2069 /*
2070 * Control VLAN HW tagging when ec_nvlan is changed from 1 to 0
2071 * or 0 to 1.
2072 */
2073 if ((set && (ec->ec_nvlans == 1)) || (!set && (ec->ec_nvlans == 0)))
2074 ixv_setup_vlan_tagging(adapter);
2075
2076 return rv;
2077 }
2078
2079 /************************************************************************
2080 * ixv_register_vlan
2081 *
2082 * Run via a vlan config EVENT, it enables us to use the
2083 * HW Filter table since we can get the vlan id. This just
2084 * creates the entry in the soft version of the VFTA, init
2085 * will repopulate the real table.
2086 ************************************************************************/
2087 static int
2088 ixv_register_vlan(struct adapter *adapter, u16 vtag)
2089 {
2090 struct ixgbe_hw *hw = &adapter->hw;
2091 u16 index, bit;
2092 int error;
2093
2094 if ((vtag == 0) || (vtag > 4095)) /* Invalid */
2095 return EINVAL;
2096 IXGBE_CORE_LOCK(adapter);
2097 index = (vtag >> 5) & 0x7F;
2098 bit = vtag & 0x1F;
2099 adapter->shadow_vfta[index] |= ((u32)1 << bit);
2100 error = hw->mac.ops.set_vfta(hw, vtag, 0, true, false);
2101 IXGBE_CORE_UNLOCK(adapter);
2102
2103 if (error != 0) {
2104 device_printf(adapter->dev, "failed to register vlan %hu\n",
2105 vtag);
2106 error = EACCES;
2107 }
2108 return error;
2109 } /* ixv_register_vlan */
2110
2111 /************************************************************************
2112 * ixv_unregister_vlan
2113 *
2114 * Run via a vlan unconfig EVENT, remove our entry
2115 * in the soft vfta.
2116 ************************************************************************/
2117 static int
2118 ixv_unregister_vlan(struct adapter *adapter, u16 vtag)
2119 {
2120 struct ixgbe_hw *hw = &adapter->hw;
2121 u16 index, bit;
2122 int error;
2123
2124 if ((vtag == 0) || (vtag > 4095)) /* Invalid */
2125 return EINVAL;
2126
2127 IXGBE_CORE_LOCK(adapter);
2128 index = (vtag >> 5) & 0x7F;
2129 bit = vtag & 0x1F;
2130 adapter->shadow_vfta[index] &= ~((u32)1 << bit);
2131 error = hw->mac.ops.set_vfta(hw, vtag, 0, false, false);
2132 IXGBE_CORE_UNLOCK(adapter);
2133
2134 if (error != 0) {
2135 device_printf(adapter->dev, "failed to unregister vlan %hu\n",
2136 vtag);
2137 error = EIO;
2138 }
2139 return error;
2140 } /* ixv_unregister_vlan */
2141
2142 /************************************************************************
2143 * ixv_enable_intr
2144 ************************************************************************/
2145 static void
2146 ixv_enable_intr(struct adapter *adapter)
2147 {
2148 struct ixgbe_hw *hw = &adapter->hw;
2149 struct ix_queue *que = adapter->queues;
2150 u32 mask;
2151 int i;
2152
2153 /* For VTEIAC */
2154 mask = (1 << adapter->vector);
2155 for (i = 0; i < adapter->num_queues; i++, que++)
2156 mask |= (1 << que->msix);
2157 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
2158
2159 /* For VTEIMS */
2160 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
2161 que = adapter->queues;
2162 for (i = 0; i < adapter->num_queues; i++, que++)
2163 ixv_enable_queue(adapter, que->msix);
2164
2165 IXGBE_WRITE_FLUSH(hw);
2166 } /* ixv_enable_intr */
2167
2168 /************************************************************************
2169 * ixv_disable_intr
2170 ************************************************************************/
2171 static void
2172 ixv_disable_intr(struct adapter *adapter)
2173 {
2174 struct ix_queue *que = adapter->queues;
2175
2176 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIAC, 0);
2177
2178 /* disable interrupts other than queues */
2179 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIMC, adapter->vector);
2180
2181 for (int i = 0; i < adapter->num_queues; i++, que++)
2182 ixv_disable_queue(adapter, que->msix);
2183
2184 IXGBE_WRITE_FLUSH(&adapter->hw);
2185 } /* ixv_disable_intr */
2186
2187 /************************************************************************
2188 * ixv_set_ivar
2189 *
2190 * Setup the correct IVAR register for a particular MSI-X interrupt
2191 * - entry is the register array entry
2192 * - vector is the MSI-X vector for this queue
2193 * - type is RX/TX/MISC
2194 ************************************************************************/
2195 static void
2196 ixv_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type)
2197 {
2198 struct ixgbe_hw *hw = &adapter->hw;
2199 u32 ivar, index;
2200
2201 vector |= IXGBE_IVAR_ALLOC_VAL;
2202
2203 if (type == -1) { /* MISC IVAR */
2204 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
2205 ivar &= ~0xFF;
2206 ivar |= vector;
2207 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
2208 } else { /* RX/TX IVARS */
2209 index = (16 * (entry & 1)) + (8 * type);
2210 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(entry >> 1));
2211 ivar &= ~(0xffUL << index);
2212 ivar |= ((u32)vector << index);
2213 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(entry >> 1), ivar);
2214 }
2215 } /* ixv_set_ivar */
2216
2217 /************************************************************************
2218 * ixv_configure_ivars
2219 ************************************************************************/
2220 static void
2221 ixv_configure_ivars(struct adapter *adapter)
2222 {
2223 struct ix_queue *que = adapter->queues;
2224
2225 /* XXX We should sync EITR value calculation with ixgbe.c? */
2226
2227 for (int i = 0; i < adapter->num_queues; i++, que++) {
2228 /* First the RX queue entry */
2229 ixv_set_ivar(adapter, i, que->msix, 0);
2230 /* ... and the TX */
2231 ixv_set_ivar(adapter, i, que->msix, 1);
2232 /* Set an initial value in EITR */
2233 ixv_eitr_write(adapter, que->msix, IXGBE_EITR_DEFAULT);
2234 }
2235
2236 /* For the mailbox interrupt */
2237 ixv_set_ivar(adapter, 1, adapter->vector, -1);
2238 } /* ixv_configure_ivars */
2239
2240
2241 /************************************************************************
2242 * ixv_save_stats
2243 *
2244 * The VF stats registers never have a truly virgin
2245 * starting point, so this routine tries to make an
2246 * artificial one, marking ground zero on attach as
2247 * it were.
2248 ************************************************************************/
2249 static void
2250 ixv_save_stats(struct adapter *adapter)
2251 {
2252 struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2253
2254 if (stats->vfgprc.ev_count || stats->vfgptc.ev_count) {
2255 stats->saved_reset_vfgprc +=
2256 stats->vfgprc.ev_count - stats->base_vfgprc;
2257 stats->saved_reset_vfgptc +=
2258 stats->vfgptc.ev_count - stats->base_vfgptc;
2259 stats->saved_reset_vfgorc +=
2260 stats->vfgorc.ev_count - stats->base_vfgorc;
2261 stats->saved_reset_vfgotc +=
2262 stats->vfgotc.ev_count - stats->base_vfgotc;
2263 stats->saved_reset_vfmprc +=
2264 stats->vfmprc.ev_count - stats->base_vfmprc;
2265 }
2266 } /* ixv_save_stats */
2267
2268 /************************************************************************
2269 * ixv_init_stats
2270 ************************************************************************/
2271 static void
2272 ixv_init_stats(struct adapter *adapter)
2273 {
2274 struct ixgbe_hw *hw = &adapter->hw;
2275
2276 adapter->stats.vf.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2277 adapter->stats.vf.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2278 adapter->stats.vf.last_vfgorc |=
2279 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2280
2281 adapter->stats.vf.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2282 adapter->stats.vf.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2283 adapter->stats.vf.last_vfgotc |=
2284 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2285
2286 adapter->stats.vf.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2287
2288 adapter->stats.vf.base_vfgprc = adapter->stats.vf.last_vfgprc;
2289 adapter->stats.vf.base_vfgorc = adapter->stats.vf.last_vfgorc;
2290 adapter->stats.vf.base_vfgptc = adapter->stats.vf.last_vfgptc;
2291 adapter->stats.vf.base_vfgotc = adapter->stats.vf.last_vfgotc;
2292 adapter->stats.vf.base_vfmprc = adapter->stats.vf.last_vfmprc;
2293 } /* ixv_init_stats */
2294
2295 #define UPDATE_STAT_32(reg, last, count) \
2296 { \
2297 u32 current = IXGBE_READ_REG(hw, (reg)); \
2298 if (current < (last)) \
2299 count.ev_count += 0x100000000LL; \
2300 (last) = current; \
2301 count.ev_count &= 0xFFFFFFFF00000000LL; \
2302 count.ev_count |= current; \
2303 }
2304
2305 #define UPDATE_STAT_36(lsb, msb, last, count) \
2306 { \
2307 u64 cur_lsb = IXGBE_READ_REG(hw, (lsb)); \
2308 u64 cur_msb = IXGBE_READ_REG(hw, (msb)); \
2309 u64 current = ((cur_msb << 32) | cur_lsb); \
2310 if (current < (last)) \
2311 count.ev_count += 0x1000000000LL; \
2312 (last) = current; \
2313 count.ev_count &= 0xFFFFFFF000000000LL; \
2314 count.ev_count |= current; \
2315 }
2316
2317 /************************************************************************
2318 * ixv_update_stats - Update the board statistics counters.
2319 ************************************************************************/
2320 void
2321 ixv_update_stats(struct adapter *adapter)
2322 {
2323 struct ixgbe_hw *hw = &adapter->hw;
2324 struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2325
2326 UPDATE_STAT_32(IXGBE_VFGPRC, stats->last_vfgprc, stats->vfgprc);
2327 UPDATE_STAT_32(IXGBE_VFGPTC, stats->last_vfgptc, stats->vfgptc);
2328 UPDATE_STAT_36(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB, stats->last_vfgorc,
2329 stats->vfgorc);
2330 UPDATE_STAT_36(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB, stats->last_vfgotc,
2331 stats->vfgotc);
2332 UPDATE_STAT_32(IXGBE_VFMPRC, stats->last_vfmprc, stats->vfmprc);
2333
2334 /* Fill out the OS statistics structure */
2335 /*
2336 * NetBSD: Don't override if_{i|o}{packets|bytes|mcasts} with
2337 * adapter->stats counters. It's required to make ifconfig -z
2338 * (SOICZIFDATA) work.
2339 */
2340 } /* ixv_update_stats */
2341
2342 /************************************************************************
2343 * ixv_sysctl_interrupt_rate_handler
2344 ************************************************************************/
2345 static int
2346 ixv_sysctl_interrupt_rate_handler(SYSCTLFN_ARGS)
2347 {
2348 struct sysctlnode node = *rnode;
2349 struct ix_queue *que = (struct ix_queue *)node.sysctl_data;
2350 struct adapter *adapter = que->adapter;
2351 uint32_t reg, usec, rate;
2352 int error;
2353
2354 if (que == NULL)
2355 return 0;
2356 reg = IXGBE_READ_REG(&que->adapter->hw, IXGBE_VTEITR(que->msix));
2357 usec = ((reg & 0x0FF8) >> 3);
2358 if (usec > 0)
2359 rate = 500000 / usec;
2360 else
2361 rate = 0;
2362 node.sysctl_data = &rate;
2363 error = sysctl_lookup(SYSCTLFN_CALL(&node));
2364 if (error || newp == NULL)
2365 return error;
2366 reg &= ~0xfff; /* default, no limitation */
2367 if (rate > 0 && rate < 500000) {
2368 if (rate < 1000)
2369 rate = 1000;
2370 reg |= ((4000000/rate) & 0xff8);
2371 /*
2372 * When RSC is used, ITR interval must be larger than
2373 * RSC_DELAY. Currently, we use 2us for RSC_DELAY.
2374 * The minimum value is always greater than 2us on 100M
2375 * (and 10M?(not documented)), but it's not on 1G and higher.
2376 */
2377 if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
2378 && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
2379 if ((adapter->num_queues > 1)
2380 && (reg < IXGBE_MIN_RSC_EITR_10G1G))
2381 return EINVAL;
2382 }
2383 ixv_max_interrupt_rate = rate;
2384 } else
2385 ixv_max_interrupt_rate = 0;
2386 ixv_eitr_write(adapter, que->msix, reg);
2387
2388 return (0);
2389 } /* ixv_sysctl_interrupt_rate_handler */
2390
2391 const struct sysctlnode *
2392 ixv_sysctl_instance(struct adapter *adapter)
2393 {
2394 const char *dvname;
2395 struct sysctllog **log;
2396 int rc;
2397 const struct sysctlnode *rnode;
2398
2399 log = &adapter->sysctllog;
2400 dvname = device_xname(adapter->dev);
2401
2402 if ((rc = sysctl_createv(log, 0, NULL, &rnode,
2403 0, CTLTYPE_NODE, dvname,
2404 SYSCTL_DESCR("ixv information and settings"),
2405 NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0)
2406 goto err;
2407
2408 return rnode;
2409 err:
2410 device_printf(adapter->dev,
2411 "%s: sysctl_createv failed, rc = %d\n", __func__, rc);
2412 return NULL;
2413 }
2414
2415 static void
2416 ixv_add_device_sysctls(struct adapter *adapter)
2417 {
2418 struct sysctllog **log;
2419 const struct sysctlnode *rnode, *cnode;
2420 device_t dev;
2421
2422 dev = adapter->dev;
2423 log = &adapter->sysctllog;
2424
2425 if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2426 aprint_error_dev(dev, "could not create sysctl root\n");
2427 return;
2428 }
2429
2430 if (sysctl_createv(log, 0, &rnode, &cnode,
2431 CTLFLAG_READWRITE, CTLTYPE_INT,
2432 "debug", SYSCTL_DESCR("Debug Info"),
2433 ixv_sysctl_debug, 0, (void *)adapter, 0, CTL_CREATE, CTL_EOL) != 0)
2434 aprint_error_dev(dev, "could not create sysctl\n");
2435
2436 if (sysctl_createv(log, 0, &rnode, &cnode,
2437 CTLFLAG_READWRITE, CTLTYPE_BOOL,
2438 "enable_aim", SYSCTL_DESCR("Interrupt Moderation"),
2439 NULL, 0, &adapter->enable_aim, 0, CTL_CREATE, CTL_EOL) != 0)
2440 aprint_error_dev(dev, "could not create sysctl\n");
2441
2442 if (sysctl_createv(log, 0, &rnode, &cnode,
2443 CTLFLAG_READWRITE, CTLTYPE_BOOL,
2444 "txrx_workqueue", SYSCTL_DESCR("Use workqueue for packet processing"),
2445 NULL, 0, &adapter->txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL) != 0)
2446 aprint_error_dev(dev, "could not create sysctl\n");
2447 }
2448
2449 /************************************************************************
2450 * ixv_add_stats_sysctls - Add statistic sysctls for the VF.
2451 ************************************************************************/
2452 static void
2453 ixv_add_stats_sysctls(struct adapter *adapter)
2454 {
2455 device_t dev = adapter->dev;
2456 struct tx_ring *txr = adapter->tx_rings;
2457 struct rx_ring *rxr = adapter->rx_rings;
2458 struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2459 struct ixgbe_hw *hw = &adapter->hw;
2460 const struct sysctlnode *rnode, *cnode;
2461 struct sysctllog **log = &adapter->sysctllog;
2462 const char *xname = device_xname(dev);
2463
2464 /* Driver Statistics */
2465 evcnt_attach_dynamic(&adapter->efbig_tx_dma_setup, EVCNT_TYPE_MISC,
2466 NULL, xname, "Driver tx dma soft fail EFBIG");
2467 evcnt_attach_dynamic(&adapter->mbuf_defrag_failed, EVCNT_TYPE_MISC,
2468 NULL, xname, "m_defrag() failed");
2469 evcnt_attach_dynamic(&adapter->efbig2_tx_dma_setup, EVCNT_TYPE_MISC,
2470 NULL, xname, "Driver tx dma hard fail EFBIG");
2471 evcnt_attach_dynamic(&adapter->einval_tx_dma_setup, EVCNT_TYPE_MISC,
2472 NULL, xname, "Driver tx dma hard fail EINVAL");
2473 evcnt_attach_dynamic(&adapter->other_tx_dma_setup, EVCNT_TYPE_MISC,
2474 NULL, xname, "Driver tx dma hard fail other");
2475 evcnt_attach_dynamic(&adapter->eagain_tx_dma_setup, EVCNT_TYPE_MISC,
2476 NULL, xname, "Driver tx dma soft fail EAGAIN");
2477 evcnt_attach_dynamic(&adapter->enomem_tx_dma_setup, EVCNT_TYPE_MISC,
2478 NULL, xname, "Driver tx dma soft fail ENOMEM");
2479 evcnt_attach_dynamic(&adapter->watchdog_events, EVCNT_TYPE_MISC,
2480 NULL, xname, "Watchdog timeouts");
2481 evcnt_attach_dynamic(&adapter->tso_err, EVCNT_TYPE_MISC,
2482 NULL, xname, "TSO errors");
2483 evcnt_attach_dynamic(&adapter->link_irq, EVCNT_TYPE_INTR,
2484 NULL, xname, "Link MSI-X IRQ Handled");
2485
2486 for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2487 snprintf(adapter->queues[i].evnamebuf,
2488 sizeof(adapter->queues[i].evnamebuf), "%s q%d",
2489 xname, i);
2490 snprintf(adapter->queues[i].namebuf,
2491 sizeof(adapter->queues[i].namebuf), "q%d", i);
2492
2493 if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2494 aprint_error_dev(dev, "could not create sysctl root\n");
2495 break;
2496 }
2497
2498 if (sysctl_createv(log, 0, &rnode, &rnode,
2499 0, CTLTYPE_NODE,
2500 adapter->queues[i].namebuf, SYSCTL_DESCR("Queue Name"),
2501 NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL) != 0)
2502 break;
2503
2504 if (sysctl_createv(log, 0, &rnode, &cnode,
2505 CTLFLAG_READWRITE, CTLTYPE_INT,
2506 "interrupt_rate", SYSCTL_DESCR("Interrupt Rate"),
2507 ixv_sysctl_interrupt_rate_handler, 0,
2508 (void *)&adapter->queues[i], 0, CTL_CREATE, CTL_EOL) != 0)
2509 break;
2510
2511 if (sysctl_createv(log, 0, &rnode, &cnode,
2512 CTLFLAG_READONLY, CTLTYPE_INT,
2513 "txd_head", SYSCTL_DESCR("Transmit Descriptor Head"),
2514 ixv_sysctl_tdh_handler, 0, (void *)txr,
2515 0, CTL_CREATE, CTL_EOL) != 0)
2516 break;
2517
2518 if (sysctl_createv(log, 0, &rnode, &cnode,
2519 CTLFLAG_READONLY, CTLTYPE_INT,
2520 "txd_tail", SYSCTL_DESCR("Transmit Descriptor Tail"),
2521 ixv_sysctl_tdt_handler, 0, (void *)txr,
2522 0, CTL_CREATE, CTL_EOL) != 0)
2523 break;
2524
2525 evcnt_attach_dynamic(&adapter->queues[i].irqs, EVCNT_TYPE_INTR,
2526 NULL, adapter->queues[i].evnamebuf, "IRQs on queue");
2527 evcnt_attach_dynamic(&adapter->queues[i].handleq,
2528 EVCNT_TYPE_MISC, NULL, adapter->queues[i].evnamebuf,
2529 "Handled queue in softint");
2530 evcnt_attach_dynamic(&adapter->queues[i].req, EVCNT_TYPE_MISC,
2531 NULL, adapter->queues[i].evnamebuf, "Requeued in softint");
2532 evcnt_attach_dynamic(&txr->tso_tx, EVCNT_TYPE_MISC,
2533 NULL, adapter->queues[i].evnamebuf, "TSO");
2534 evcnt_attach_dynamic(&txr->no_desc_avail, EVCNT_TYPE_MISC,
2535 NULL, adapter->queues[i].evnamebuf,
2536 "Queue No Descriptor Available");
2537 evcnt_attach_dynamic(&txr->total_packets, EVCNT_TYPE_MISC,
2538 NULL, adapter->queues[i].evnamebuf,
2539 "Queue Packets Transmitted");
2540 #ifndef IXGBE_LEGACY_TX
2541 evcnt_attach_dynamic(&txr->pcq_drops, EVCNT_TYPE_MISC,
2542 NULL, adapter->queues[i].evnamebuf,
2543 "Packets dropped in pcq");
2544 #endif
2545
2546 #ifdef LRO
2547 struct lro_ctrl *lro = &rxr->lro;
2548 #endif /* LRO */
2549
2550 if (sysctl_createv(log, 0, &rnode, &cnode,
2551 CTLFLAG_READONLY,
2552 CTLTYPE_INT,
2553 "rxd_nxck", SYSCTL_DESCR("Receive Descriptor next to check"),
2554 ixv_sysctl_next_to_check_handler, 0, (void *)rxr, 0,
2555 CTL_CREATE, CTL_EOL) != 0)
2556 break;
2557
2558 if (sysctl_createv(log, 0, &rnode, &cnode,
2559 CTLFLAG_READONLY,
2560 CTLTYPE_INT,
2561 "rxd_head", SYSCTL_DESCR("Receive Descriptor Head"),
2562 ixv_sysctl_rdh_handler, 0, (void *)rxr, 0,
2563 CTL_CREATE, CTL_EOL) != 0)
2564 break;
2565
2566 if (sysctl_createv(log, 0, &rnode, &cnode,
2567 CTLFLAG_READONLY,
2568 CTLTYPE_INT,
2569 "rxd_tail", SYSCTL_DESCR("Receive Descriptor Tail"),
2570 ixv_sysctl_rdt_handler, 0, (void *)rxr, 0,
2571 CTL_CREATE, CTL_EOL) != 0)
2572 break;
2573
2574 evcnt_attach_dynamic(&rxr->rx_packets, EVCNT_TYPE_MISC,
2575 NULL, adapter->queues[i].evnamebuf, "Queue Packets Received");
2576 evcnt_attach_dynamic(&rxr->rx_bytes, EVCNT_TYPE_MISC,
2577 NULL, adapter->queues[i].evnamebuf, "Queue Bytes Received");
2578 evcnt_attach_dynamic(&rxr->rx_copies, EVCNT_TYPE_MISC,
2579 NULL, adapter->queues[i].evnamebuf, "Copied RX Frames");
2580 evcnt_attach_dynamic(&rxr->no_jmbuf, EVCNT_TYPE_MISC,
2581 NULL, adapter->queues[i].evnamebuf, "Rx no jumbo mbuf");
2582 evcnt_attach_dynamic(&rxr->rx_discarded, EVCNT_TYPE_MISC,
2583 NULL, adapter->queues[i].evnamebuf, "Rx discarded");
2584 #ifdef LRO
2585 SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_queued",
2586 CTLFLAG_RD, &lro->lro_queued, 0,
2587 "LRO Queued");
2588 SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_flushed",
2589 CTLFLAG_RD, &lro->lro_flushed, 0,
2590 "LRO Flushed");
2591 #endif /* LRO */
2592 }
2593
2594 /* MAC stats get their own sub node */
2595
2596 snprintf(stats->namebuf,
2597 sizeof(stats->namebuf), "%s MAC Statistics", xname);
2598
2599 evcnt_attach_dynamic(&stats->ipcs, EVCNT_TYPE_MISC, NULL,
2600 stats->namebuf, "rx csum offload - IP");
2601 evcnt_attach_dynamic(&stats->l4cs, EVCNT_TYPE_MISC, NULL,
2602 stats->namebuf, "rx csum offload - L4");
2603 evcnt_attach_dynamic(&stats->ipcs_bad, EVCNT_TYPE_MISC, NULL,
2604 stats->namebuf, "rx csum offload - IP bad");
2605 evcnt_attach_dynamic(&stats->l4cs_bad, EVCNT_TYPE_MISC, NULL,
2606 stats->namebuf, "rx csum offload - L4 bad");
2607
2608 /* Packet Reception Stats */
2609 evcnt_attach_dynamic(&stats->vfgprc, EVCNT_TYPE_MISC, NULL,
2610 xname, "Good Packets Received");
2611 evcnt_attach_dynamic(&stats->vfgorc, EVCNT_TYPE_MISC, NULL,
2612 xname, "Good Octets Received");
2613 evcnt_attach_dynamic(&stats->vfmprc, EVCNT_TYPE_MISC, NULL,
2614 xname, "Multicast Packets Received");
2615 evcnt_attach_dynamic(&stats->vfgptc, EVCNT_TYPE_MISC, NULL,
2616 xname, "Good Packets Transmitted");
2617 evcnt_attach_dynamic(&stats->vfgotc, EVCNT_TYPE_MISC, NULL,
2618 xname, "Good Octets Transmitted");
2619
2620 /* Mailbox Stats */
2621 evcnt_attach_dynamic(&hw->mbx.stats.msgs_tx, EVCNT_TYPE_MISC, NULL,
2622 xname, "message TXs");
2623 evcnt_attach_dynamic(&hw->mbx.stats.msgs_rx, EVCNT_TYPE_MISC, NULL,
2624 xname, "message RXs");
2625 evcnt_attach_dynamic(&hw->mbx.stats.acks, EVCNT_TYPE_MISC, NULL,
2626 xname, "ACKs");
2627 evcnt_attach_dynamic(&hw->mbx.stats.reqs, EVCNT_TYPE_MISC, NULL,
2628 xname, "REQs");
2629 evcnt_attach_dynamic(&hw->mbx.stats.rsts, EVCNT_TYPE_MISC, NULL,
2630 xname, "RSTs");
2631
2632 } /* ixv_add_stats_sysctls */
2633
2634 static void
2635 ixv_clear_evcnt(struct adapter *adapter)
2636 {
2637 struct tx_ring *txr = adapter->tx_rings;
2638 struct rx_ring *rxr = adapter->rx_rings;
2639 struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2640 struct ixgbe_hw *hw = &adapter->hw;
2641 int i;
2642
2643 /* Driver Statistics */
2644 adapter->efbig_tx_dma_setup.ev_count = 0;
2645 adapter->mbuf_defrag_failed.ev_count = 0;
2646 adapter->efbig2_tx_dma_setup.ev_count = 0;
2647 adapter->einval_tx_dma_setup.ev_count = 0;
2648 adapter->other_tx_dma_setup.ev_count = 0;
2649 adapter->eagain_tx_dma_setup.ev_count = 0;
2650 adapter->enomem_tx_dma_setup.ev_count = 0;
2651 adapter->watchdog_events.ev_count = 0;
2652 adapter->tso_err.ev_count = 0;
2653 adapter->link_irq.ev_count = 0;
2654
2655 for (i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2656 adapter->queues[i].irqs.ev_count = 0;
2657 adapter->queues[i].handleq.ev_count = 0;
2658 adapter->queues[i].req.ev_count = 0;
2659 txr->tso_tx.ev_count = 0;
2660 txr->no_desc_avail.ev_count = 0;
2661 txr->total_packets.ev_count = 0;
2662 #ifndef IXGBE_LEGACY_TX
2663 txr->pcq_drops.ev_count = 0;
2664 #endif
2665 txr->q_efbig_tx_dma_setup = 0;
2666 txr->q_mbuf_defrag_failed = 0;
2667 txr->q_efbig2_tx_dma_setup = 0;
2668 txr->q_einval_tx_dma_setup = 0;
2669 txr->q_other_tx_dma_setup = 0;
2670 txr->q_eagain_tx_dma_setup = 0;
2671 txr->q_enomem_tx_dma_setup = 0;
2672 txr->q_tso_err = 0;
2673
2674 rxr->rx_packets.ev_count = 0;
2675 rxr->rx_bytes.ev_count = 0;
2676 rxr->rx_copies.ev_count = 0;
2677 rxr->no_jmbuf.ev_count = 0;
2678 rxr->rx_discarded.ev_count = 0;
2679 }
2680
2681 /* MAC stats get their own sub node */
2682
2683 stats->ipcs.ev_count = 0;
2684 stats->l4cs.ev_count = 0;
2685 stats->ipcs_bad.ev_count = 0;
2686 stats->l4cs_bad.ev_count = 0;
2687
2688 /* Packet Reception Stats */
2689 stats->vfgprc.ev_count = 0;
2690 stats->vfgorc.ev_count = 0;
2691 stats->vfmprc.ev_count = 0;
2692 stats->vfgptc.ev_count = 0;
2693 stats->vfgotc.ev_count = 0;
2694
2695 /* Mailbox Stats */
2696 hw->mbx.stats.msgs_tx.ev_count = 0;
2697 hw->mbx.stats.msgs_rx.ev_count = 0;
2698 hw->mbx.stats.acks.ev_count = 0;
2699 hw->mbx.stats.reqs.ev_count = 0;
2700 hw->mbx.stats.rsts.ev_count = 0;
2701
2702 } /* ixv_clear_evcnt */
2703
2704 /************************************************************************
2705 * ixv_set_sysctl_value
2706 ************************************************************************/
2707 static void
2708 ixv_set_sysctl_value(struct adapter *adapter, const char *name,
2709 const char *description, int *limit, int value)
2710 {
2711 device_t dev = adapter->dev;
2712 struct sysctllog **log;
2713 const struct sysctlnode *rnode, *cnode;
2714
2715 log = &adapter->sysctllog;
2716 if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2717 aprint_error_dev(dev, "could not create sysctl root\n");
2718 return;
2719 }
2720 if (sysctl_createv(log, 0, &rnode, &cnode,
2721 CTLFLAG_READWRITE, CTLTYPE_INT,
2722 name, SYSCTL_DESCR(description),
2723 NULL, 0, limit, 0, CTL_CREATE, CTL_EOL) != 0)
2724 aprint_error_dev(dev, "could not create sysctl\n");
2725 *limit = value;
2726 } /* ixv_set_sysctl_value */
2727
2728 /************************************************************************
2729 * ixv_print_debug_info
2730 *
2731 * Called only when em_display_debug_stats is enabled.
2732 * Provides a way to take a look at important statistics
2733 * maintained by the driver and hardware.
2734 ************************************************************************/
2735 static void
2736 ixv_print_debug_info(struct adapter *adapter)
2737 {
2738 device_t dev = adapter->dev;
2739 struct ix_queue *que = adapter->queues;
2740 struct rx_ring *rxr;
2741 struct tx_ring *txr;
2742 #ifdef LRO
2743 struct lro_ctrl *lro;
2744 #endif /* LRO */
2745
2746 for (int i = 0; i < adapter->num_queues; i++, que++) {
2747 txr = que->txr;
2748 rxr = que->rxr;
2749 #ifdef LRO
2750 lro = &rxr->lro;
2751 #endif /* LRO */
2752 device_printf(dev, "QUE(%d) IRQs Handled: %lu\n",
2753 que->msix, (long)que->irqs.ev_count);
2754 device_printf(dev, "RX(%d) Packets Received: %lld\n",
2755 rxr->me, (long long)rxr->rx_packets.ev_count);
2756 device_printf(dev, "RX(%d) Bytes Received: %lu\n",
2757 rxr->me, (long)rxr->rx_bytes.ev_count);
2758 #ifdef LRO
2759 device_printf(dev, "RX(%d) LRO Queued= %ju\n",
2760 rxr->me, (uintmax_t)lro->lro_queued);
2761 device_printf(dev, "RX(%d) LRO Flushed= %ju\n",
2762 rxr->me, (uintmax_t)lro->lro_flushed);
2763 #endif /* LRO */
2764 device_printf(dev, "TX(%d) Packets Sent: %lu\n",
2765 txr->me, (long)txr->total_packets.ev_count);
2766 device_printf(dev, "TX(%d) NO Desc Avail: %lu\n",
2767 txr->me, (long)txr->no_desc_avail.ev_count);
2768 }
2769
2770 device_printf(dev, "MBX IRQ Handled: %lu\n",
2771 (long)adapter->link_irq.ev_count);
2772 } /* ixv_print_debug_info */
2773
2774 /************************************************************************
2775 * ixv_sysctl_debug
2776 ************************************************************************/
2777 static int
2778 ixv_sysctl_debug(SYSCTLFN_ARGS)
2779 {
2780 struct sysctlnode node = *rnode;
2781 struct adapter *adapter = (struct adapter *)node.sysctl_data;
2782 int error, result;
2783
2784 node.sysctl_data = &result;
2785 error = sysctl_lookup(SYSCTLFN_CALL(&node));
2786
2787 if (error || newp == NULL)
2788 return error;
2789
2790 if (result == 1)
2791 ixv_print_debug_info(adapter);
2792
2793 return 0;
2794 } /* ixv_sysctl_debug */
2795
2796 /************************************************************************
2797 * ixv_init_device_features
2798 ************************************************************************/
2799 static void
2800 ixv_init_device_features(struct adapter *adapter)
2801 {
2802 adapter->feat_cap = IXGBE_FEATURE_NETMAP
2803 | IXGBE_FEATURE_VF
2804 | IXGBE_FEATURE_RSS
2805 | IXGBE_FEATURE_LEGACY_TX;
2806
2807 /* A tad short on feature flags for VFs, atm. */
2808 switch (adapter->hw.mac.type) {
2809 case ixgbe_mac_82599_vf:
2810 break;
2811 case ixgbe_mac_X540_vf:
2812 break;
2813 case ixgbe_mac_X550_vf:
2814 case ixgbe_mac_X550EM_x_vf:
2815 case ixgbe_mac_X550EM_a_vf:
2816 adapter->feat_cap |= IXGBE_FEATURE_NEEDS_CTXD;
2817 break;
2818 default:
2819 break;
2820 }
2821
2822 /* Enabled by default... */
2823 /* Is a virtual function (VF) */
2824 if (adapter->feat_cap & IXGBE_FEATURE_VF)
2825 adapter->feat_en |= IXGBE_FEATURE_VF;
2826 /* Netmap */
2827 if (adapter->feat_cap & IXGBE_FEATURE_NETMAP)
2828 adapter->feat_en |= IXGBE_FEATURE_NETMAP;
2829 /* Receive-Side Scaling (RSS) */
2830 if (adapter->feat_cap & IXGBE_FEATURE_RSS)
2831 adapter->feat_en |= IXGBE_FEATURE_RSS;
2832 /* Needs advanced context descriptor regardless of offloads req'd */
2833 if (adapter->feat_cap & IXGBE_FEATURE_NEEDS_CTXD)
2834 adapter->feat_en |= IXGBE_FEATURE_NEEDS_CTXD;
2835
2836 /* Enabled via sysctl... */
2837 /* Legacy (single queue) transmit */
2838 if ((adapter->feat_cap & IXGBE_FEATURE_LEGACY_TX) &&
2839 ixv_enable_legacy_tx)
2840 adapter->feat_en |= IXGBE_FEATURE_LEGACY_TX;
2841 } /* ixv_init_device_features */
2842
2843 /************************************************************************
2844 * ixv_shutdown - Shutdown entry point
2845 ************************************************************************/
2846 #if 0 /* XXX NetBSD ought to register something like this through pmf(9) */
2847 static int
2848 ixv_shutdown(device_t dev)
2849 {
2850 struct adapter *adapter = device_private(dev);
2851 IXGBE_CORE_LOCK(adapter);
2852 ixv_stop(adapter);
2853 IXGBE_CORE_UNLOCK(adapter);
2854
2855 return (0);
2856 } /* ixv_shutdown */
2857 #endif
2858
2859 static int
2860 ixv_ifflags_cb(struct ethercom *ec)
2861 {
2862 struct ifnet *ifp = &ec->ec_if;
2863 struct adapter *adapter = ifp->if_softc;
2864 int change, rv = 0;
2865
2866 IXGBE_CORE_LOCK(adapter);
2867
2868 change = ifp->if_flags ^ adapter->if_flags;
2869 if (change != 0)
2870 adapter->if_flags = ifp->if_flags;
2871
2872 if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
2873 rv = ENETRESET;
2874 goto out;
2875 }
2876
2877 /* Check for ec_capenable. */
2878 change = ec->ec_capenable ^ adapter->ec_capenable;
2879 adapter->ec_capenable = ec->ec_capenable;
2880 if ((change & ~(ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING
2881 | ETHERCAP_VLAN_HWFILTER)) != 0) {
2882 rv = ENETRESET;
2883 goto out;
2884 }
2885
2886 /*
2887 * Special handling is not required for ETHERCAP_VLAN_MTU.
2888 * PF's MAXFRS(MHADD) does not include the 4bytes of the VLAN header.
2889 */
2890
2891 /* Set up VLAN support and filter */
2892 if ((change & (ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_HWFILTER)) != 0)
2893 rv = ixv_setup_vlan_support(adapter);
2894
2895 out:
2896 IXGBE_CORE_UNLOCK(adapter);
2897
2898 return rv;
2899 }
2900
2901
2902 /************************************************************************
2903 * ixv_ioctl - Ioctl entry point
2904 *
2905 * Called when the user wants to configure the interface.
2906 *
2907 * return 0 on success, positive on failure
2908 ************************************************************************/
2909 static int
2910 ixv_ioctl(struct ifnet *ifp, u_long command, void *data)
2911 {
2912 struct adapter *adapter = ifp->if_softc;
2913 struct ifcapreq *ifcr = data;
2914 int error = 0;
2915 int l4csum_en;
2916 const int l4csum = IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx |
2917 IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_UDPv6_Rx;
2918
2919 switch (command) {
2920 case SIOCSIFFLAGS:
2921 IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)");
2922 break;
2923 case SIOCADDMULTI: {
2924 struct ether_multi *enm;
2925 struct ether_multistep step;
2926 struct ethercom *ec = &adapter->osdep.ec;
2927 int mcnt = 0;
2928
2929 /*
2930 * Check the number of multicast address. If it exceeds,
2931 * return ENOSPC.
2932 * Update this code when we support API 1.3.
2933 */
2934 ETHER_LOCK(ec);
2935 ETHER_FIRST_MULTI(step, ec, enm);
2936 while (enm != NULL) {
2937 mcnt++;
2938
2939 /*
2940 * This code is before adding, so one room is required
2941 * at least.
2942 */
2943 if (mcnt > (IXGBE_MAX_VF_MC - 1)) {
2944 device_printf(adapter->dev,
2945 "number of Ethernet multicast addresses "
2946 "exceeds the limit (%d)\n",
2947 IXGBE_MAX_VF_MC);
2948 error = ENOSPC;
2949 break;
2950 }
2951 ETHER_NEXT_MULTI(step, enm);
2952 }
2953 ETHER_UNLOCK(ec);
2954 if (error)
2955 return error;
2956 }
2957 /*FALLTHROUGH*/
2958 case SIOCDELMULTI:
2959 IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI");
2960 break;
2961 case SIOCSIFMEDIA:
2962 case SIOCGIFMEDIA:
2963 IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)");
2964 break;
2965 case SIOCSIFCAP:
2966 IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)");
2967 break;
2968 case SIOCSIFMTU:
2969 IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)");
2970 break;
2971 case SIOCZIFDATA:
2972 IOCTL_DEBUGOUT("ioctl: SIOCZIFDATA (Zero counter)");
2973 ixv_update_stats(adapter);
2974 ixv_clear_evcnt(adapter);
2975 break;
2976 default:
2977 IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)", (int)command);
2978 break;
2979 }
2980
2981 switch (command) {
2982 case SIOCSIFCAP:
2983 /* Layer-4 Rx checksum offload has to be turned on and
2984 * off as a unit.
2985 */
2986 l4csum_en = ifcr->ifcr_capenable & l4csum;
2987 if (l4csum_en != l4csum && l4csum_en != 0)
2988 return EINVAL;
2989 /*FALLTHROUGH*/
2990 case SIOCADDMULTI:
2991 case SIOCDELMULTI:
2992 case SIOCSIFFLAGS:
2993 case SIOCSIFMTU:
2994 default:
2995 if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
2996 return error;
2997 if ((ifp->if_flags & IFF_RUNNING) == 0)
2998 ;
2999 else if (command == SIOCSIFCAP || command == SIOCSIFMTU) {
3000 IXGBE_CORE_LOCK(adapter);
3001 ixv_init_locked(adapter);
3002 IXGBE_CORE_UNLOCK(adapter);
3003 } else if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
3004 /*
3005 * Multicast list has changed; set the hardware filter
3006 * accordingly.
3007 */
3008 IXGBE_CORE_LOCK(adapter);
3009 ixv_disable_intr(adapter);
3010 ixv_set_multi(adapter);
3011 ixv_enable_intr(adapter);
3012 IXGBE_CORE_UNLOCK(adapter);
3013 }
3014 return 0;
3015 }
3016 } /* ixv_ioctl */
3017
3018 /************************************************************************
3019 * ixv_init
3020 ************************************************************************/
3021 static int
3022 ixv_init(struct ifnet *ifp)
3023 {
3024 struct adapter *adapter = ifp->if_softc;
3025
3026 IXGBE_CORE_LOCK(adapter);
3027 ixv_init_locked(adapter);
3028 IXGBE_CORE_UNLOCK(adapter);
3029
3030 return 0;
3031 } /* ixv_init */
3032
3033 /************************************************************************
3034 * ixv_handle_que
3035 ************************************************************************/
3036 static void
3037 ixv_handle_que(void *context)
3038 {
3039 struct ix_queue *que = context;
3040 struct adapter *adapter = que->adapter;
3041 struct tx_ring *txr = que->txr;
3042 struct ifnet *ifp = adapter->ifp;
3043 bool more;
3044
3045 que->handleq.ev_count++;
3046
3047 if (ifp->if_flags & IFF_RUNNING) {
3048 more = ixgbe_rxeof(que);
3049 IXGBE_TX_LOCK(txr);
3050 more |= ixgbe_txeof(txr);
3051 if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
3052 if (!ixgbe_mq_ring_empty(ifp, txr->txr_interq))
3053 ixgbe_mq_start_locked(ifp, txr);
3054 /* Only for queue 0 */
3055 /* NetBSD still needs this for CBQ */
3056 if ((&adapter->queues[0] == que)
3057 && (!ixgbe_legacy_ring_empty(ifp, NULL)))
3058 ixgbe_legacy_start_locked(ifp, txr);
3059 IXGBE_TX_UNLOCK(txr);
3060 if (more) {
3061 que->req.ev_count++;
3062 if (adapter->txrx_use_workqueue) {
3063 /*
3064 * "enqueued flag" is not required here
3065 * the same as ixg(4). See ixgbe_msix_que().
3066 */
3067 workqueue_enqueue(adapter->que_wq,
3068 &que->wq_cookie, curcpu());
3069 } else
3070 softint_schedule(que->que_si);
3071 return;
3072 }
3073 }
3074
3075 /* Re-enable this interrupt */
3076 ixv_enable_queue(adapter, que->msix);
3077
3078 return;
3079 } /* ixv_handle_que */
3080
3081 /************************************************************************
3082 * ixv_handle_que_work
3083 ************************************************************************/
3084 static void
3085 ixv_handle_que_work(struct work *wk, void *context)
3086 {
3087 struct ix_queue *que = container_of(wk, struct ix_queue, wq_cookie);
3088
3089 /*
3090 * "enqueued flag" is not required here the same as ixg(4).
3091 * See ixgbe_msix_que().
3092 */
3093 ixv_handle_que(que);
3094 }
3095
3096 /************************************************************************
3097 * ixv_allocate_msix - Setup MSI-X Interrupt resources and handlers
3098 ************************************************************************/
3099 static int
3100 ixv_allocate_msix(struct adapter *adapter, const struct pci_attach_args *pa)
3101 {
3102 device_t dev = adapter->dev;
3103 struct ix_queue *que = adapter->queues;
3104 struct tx_ring *txr = adapter->tx_rings;
3105 int error, msix_ctrl, rid, vector = 0;
3106 pci_chipset_tag_t pc;
3107 pcitag_t tag;
3108 char intrbuf[PCI_INTRSTR_LEN];
3109 char wqname[MAXCOMLEN];
3110 char intr_xname[32];
3111 const char *intrstr = NULL;
3112 kcpuset_t *affinity;
3113 int cpu_id = 0;
3114
3115 pc = adapter->osdep.pc;
3116 tag = adapter->osdep.tag;
3117
3118 adapter->osdep.nintrs = adapter->num_queues + 1;
3119 if (pci_msix_alloc_exact(pa, &adapter->osdep.intrs,
3120 adapter->osdep.nintrs) != 0) {
3121 aprint_error_dev(dev,
3122 "failed to allocate MSI-X interrupt\n");
3123 return (ENXIO);
3124 }
3125
3126 kcpuset_create(&affinity, false);
3127 for (int i = 0; i < adapter->num_queues; i++, vector++, que++, txr++) {
3128 snprintf(intr_xname, sizeof(intr_xname), "%s TXRX%d",
3129 device_xname(dev), i);
3130 intrstr = pci_intr_string(pc, adapter->osdep.intrs[i], intrbuf,
3131 sizeof(intrbuf));
3132 #ifdef IXGBE_MPSAFE
3133 pci_intr_setattr(pc, &adapter->osdep.intrs[i], PCI_INTR_MPSAFE,
3134 true);
3135 #endif
3136 /* Set the handler function */
3137 que->res = adapter->osdep.ihs[i] = pci_intr_establish_xname(pc,
3138 adapter->osdep.intrs[i], IPL_NET, ixv_msix_que, que,
3139 intr_xname);
3140 if (que->res == NULL) {
3141 pci_intr_release(pc, adapter->osdep.intrs,
3142 adapter->osdep.nintrs);
3143 aprint_error_dev(dev,
3144 "Failed to register QUE handler\n");
3145 kcpuset_destroy(affinity);
3146 return (ENXIO);
3147 }
3148 que->msix = vector;
3149 adapter->active_queues |= (u64)(1 << que->msix);
3150
3151 cpu_id = i;
3152 /* Round-robin affinity */
3153 kcpuset_zero(affinity);
3154 kcpuset_set(affinity, cpu_id % ncpu);
3155 error = interrupt_distribute(adapter->osdep.ihs[i], affinity,
3156 NULL);
3157 aprint_normal_dev(dev, "for TX/RX, interrupting at %s",
3158 intrstr);
3159 if (error == 0)
3160 aprint_normal(", bound queue %d to cpu %d\n",
3161 i, cpu_id % ncpu);
3162 else
3163 aprint_normal("\n");
3164
3165 #ifndef IXGBE_LEGACY_TX
3166 txr->txr_si
3167 = softint_establish(SOFTINT_NET | IXGBE_SOFTINFT_FLAGS,
3168 ixgbe_deferred_mq_start, txr);
3169 #endif
3170 que->que_si
3171 = softint_establish(SOFTINT_NET | IXGBE_SOFTINFT_FLAGS,
3172 ixv_handle_que, que);
3173 if (que->que_si == NULL) {
3174 aprint_error_dev(dev,
3175 "could not establish software interrupt\n");
3176 }
3177 }
3178 snprintf(wqname, sizeof(wqname), "%sdeferTx", device_xname(dev));
3179 error = workqueue_create(&adapter->txr_wq, wqname,
3180 ixgbe_deferred_mq_start_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3181 IXGBE_WORKQUEUE_FLAGS);
3182 if (error) {
3183 aprint_error_dev(dev, "couldn't create workqueue for deferred Tx\n");
3184 }
3185 adapter->txr_wq_enqueued = percpu_alloc(sizeof(u_int));
3186
3187 snprintf(wqname, sizeof(wqname), "%sTxRx", device_xname(dev));
3188 error = workqueue_create(&adapter->que_wq, wqname,
3189 ixv_handle_que_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3190 IXGBE_WORKQUEUE_FLAGS);
3191 if (error) {
3192 aprint_error_dev(dev,
3193 "couldn't create workqueue\n");
3194 }
3195
3196 /* and Mailbox */
3197 cpu_id++;
3198 snprintf(intr_xname, sizeof(intr_xname), "%s link", device_xname(dev));
3199 adapter->vector = vector;
3200 intrstr = pci_intr_string(pc, adapter->osdep.intrs[vector], intrbuf,
3201 sizeof(intrbuf));
3202 #ifdef IXGBE_MPSAFE
3203 pci_intr_setattr(pc, &adapter->osdep.intrs[vector], PCI_INTR_MPSAFE,
3204 true);
3205 #endif
3206 /* Set the mbx handler function */
3207 adapter->osdep.ihs[vector] = pci_intr_establish_xname(pc,
3208 adapter->osdep.intrs[vector], IPL_NET, ixv_msix_mbx, adapter,
3209 intr_xname);
3210 if (adapter->osdep.ihs[vector] == NULL) {
3211 aprint_error_dev(dev, "Failed to register LINK handler\n");
3212 kcpuset_destroy(affinity);
3213 return (ENXIO);
3214 }
3215 /* Round-robin affinity */
3216 kcpuset_zero(affinity);
3217 kcpuset_set(affinity, cpu_id % ncpu);
3218 error = interrupt_distribute(adapter->osdep.ihs[vector],
3219 affinity, NULL);
3220
3221 aprint_normal_dev(dev,
3222 "for link, interrupting at %s", intrstr);
3223 if (error == 0)
3224 aprint_normal(", affinity to cpu %d\n", cpu_id % ncpu);
3225 else
3226 aprint_normal("\n");
3227
3228 /* Tasklets for Mailbox */
3229 adapter->link_si = softint_establish(SOFTINT_NET |IXGBE_SOFTINFT_FLAGS,
3230 ixv_handle_link, adapter);
3231 /*
3232 * Due to a broken design QEMU will fail to properly
3233 * enable the guest for MSI-X unless the vectors in
3234 * the table are all set up, so we must rewrite the
3235 * ENABLE in the MSI-X control register again at this
3236 * point to cause it to successfully initialize us.
3237 */
3238 if (adapter->hw.mac.type == ixgbe_mac_82599_vf) {
3239 pci_get_capability(pc, tag, PCI_CAP_MSIX, &rid, NULL);
3240 rid += PCI_MSIX_CTL;
3241 msix_ctrl = pci_conf_read(pc, tag, rid);
3242 msix_ctrl |= PCI_MSIX_CTL_ENABLE;
3243 pci_conf_write(pc, tag, rid, msix_ctrl);
3244 }
3245
3246 kcpuset_destroy(affinity);
3247 return (0);
3248 } /* ixv_allocate_msix */
3249
3250 /************************************************************************
3251 * ixv_configure_interrupts - Setup MSI-X resources
3252 *
3253 * Note: The VF device MUST use MSI-X, there is no fallback.
3254 ************************************************************************/
3255 static int
3256 ixv_configure_interrupts(struct adapter *adapter)
3257 {
3258 device_t dev = adapter->dev;
3259 int want, queues, msgs;
3260
3261 /* Must have at least 2 MSI-X vectors */
3262 msgs = pci_msix_count(adapter->osdep.pc, adapter->osdep.tag);
3263 if (msgs < 2) {
3264 aprint_error_dev(dev, "MSIX config error\n");
3265 return (ENXIO);
3266 }
3267 msgs = MIN(msgs, IXG_MAX_NINTR);
3268
3269 /* Figure out a reasonable auto config value */
3270 queues = (ncpu > (msgs - 1)) ? (msgs - 1) : ncpu;
3271
3272 if (ixv_num_queues != 0)
3273 queues = ixv_num_queues;
3274 else if ((ixv_num_queues == 0) && (queues > IXGBE_VF_MAX_TX_QUEUES))
3275 queues = IXGBE_VF_MAX_TX_QUEUES;
3276
3277 /*
3278 * Want vectors for the queues,
3279 * plus an additional for mailbox.
3280 */
3281 want = queues + 1;
3282 if (msgs >= want)
3283 msgs = want;
3284 else {
3285 aprint_error_dev(dev,
3286 "MSI-X Configuration Problem, "
3287 "%d vectors but %d queues wanted!\n",
3288 msgs, want);
3289 return -1;
3290 }
3291
3292 adapter->msix_mem = (void *)1; /* XXX */
3293 aprint_normal_dev(dev,
3294 "Using MSI-X interrupts with %d vectors\n", msgs);
3295 adapter->num_queues = queues;
3296
3297 return (0);
3298 } /* ixv_configure_interrupts */
3299
3300
3301 /************************************************************************
3302 * ixv_handle_link - Tasklet handler for MSI-X MBX interrupts
3303 *
3304 * Done outside of interrupt context since the driver might sleep
3305 ************************************************************************/
3306 static void
3307 ixv_handle_link(void *context)
3308 {
3309 struct adapter *adapter = context;
3310
3311 IXGBE_CORE_LOCK(adapter);
3312
3313 adapter->hw.mac.ops.check_link(&adapter->hw, &adapter->link_speed,
3314 &adapter->link_up, FALSE);
3315 ixv_update_link_status(adapter);
3316
3317 IXGBE_CORE_UNLOCK(adapter);
3318 } /* ixv_handle_link */
3319
3320 /************************************************************************
3321 * ixv_check_link - Used in the local timer to poll for link changes
3322 ************************************************************************/
3323 static s32
3324 ixv_check_link(struct adapter *adapter)
3325 {
3326 s32 error;
3327
3328 KASSERT(mutex_owned(&adapter->core_mtx));
3329
3330 adapter->hw.mac.get_link_status = TRUE;
3331
3332 error = adapter->hw.mac.ops.check_link(&adapter->hw,
3333 &adapter->link_speed, &adapter->link_up, FALSE);
3334 ixv_update_link_status(adapter);
3335
3336 return error;
3337 } /* ixv_check_link */
3338