xref: /netbsd-src/sys/dev/pci/ixgbe/ixv.c (revision 33881f779a77dce6440bdc44610d94de75bebefe)
1 /*$NetBSD: ixv.c,v 1.147 2020/03/15 23:04:50 thorpej 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 int	ixv_set_rxfilter(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_queues(adapter);
551 err_out:
552 	ixv_free_pci_resources(adapter);
553 	IXGBE_CORE_LOCK_DESTROY(adapter);
554 
555 	return;
556 } /* ixv_attach */
557 
558 /************************************************************************
559  * ixv_detach - Device removal routine
560  *
561  *   Called when the driver is being removed.
562  *   Stops the adapter and deallocates all the resources
563  *   that were allocated for driver operation.
564  *
565  *   return 0 on success, positive on failure
566  ************************************************************************/
567 static int
568 ixv_detach(device_t dev, int flags)
569 {
570 	struct adapter	*adapter = device_private(dev);
571 	struct ixgbe_hw *hw = &adapter->hw;
572 	struct ix_queue *que = adapter->queues;
573 	struct tx_ring *txr = adapter->tx_rings;
574 	struct rx_ring *rxr = adapter->rx_rings;
575 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
576 
577 	INIT_DEBUGOUT("ixv_detach: begin");
578 	if (adapter->osdep.attached == false)
579 		return 0;
580 
581 	/* Stop the interface. Callouts are stopped in it. */
582 	ixv_ifstop(adapter->ifp, 1);
583 
584 #if NVLAN > 0
585 	/* Make sure VLANs are not using driver */
586 	if (!VLAN_ATTACHED(&adapter->osdep.ec))
587 		;	/* nothing to do: no VLANs */
588 	else if ((flags & (DETACH_SHUTDOWN | DETACH_FORCE)) != 0)
589 		vlan_ifdetach(adapter->ifp);
590 	else {
591 		aprint_error_dev(dev, "VLANs in use, detach first\n");
592 		return EBUSY;
593 	}
594 #endif
595 
596 	for (int i = 0; i < adapter->num_queues; i++, que++, txr++) {
597 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
598 			softint_disestablish(txr->txr_si);
599 		softint_disestablish(que->que_si);
600 	}
601 	if (adapter->txr_wq != NULL)
602 		workqueue_destroy(adapter->txr_wq);
603 	if (adapter->txr_wq_enqueued != NULL)
604 		percpu_free(adapter->txr_wq_enqueued, sizeof(u_int));
605 	if (adapter->que_wq != NULL)
606 		workqueue_destroy(adapter->que_wq);
607 
608 	/* Drain the Mailbox(link) queue */
609 	softint_disestablish(adapter->link_si);
610 
611 	ether_ifdetach(adapter->ifp);
612 	callout_halt(&adapter->timer, NULL);
613 
614 	if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
615 		netmap_detach(adapter->ifp);
616 
617 	ixv_free_pci_resources(adapter);
618 #if 0 /* XXX the NetBSD port is probably missing something here */
619 	bus_generic_detach(dev);
620 #endif
621 	if_detach(adapter->ifp);
622 	ifmedia_fini(&adapter->media);
623 	if_percpuq_destroy(adapter->ipq);
624 
625 	sysctl_teardown(&adapter->sysctllog);
626 	evcnt_detach(&adapter->efbig_tx_dma_setup);
627 	evcnt_detach(&adapter->mbuf_defrag_failed);
628 	evcnt_detach(&adapter->efbig2_tx_dma_setup);
629 	evcnt_detach(&adapter->einval_tx_dma_setup);
630 	evcnt_detach(&adapter->other_tx_dma_setup);
631 	evcnt_detach(&adapter->eagain_tx_dma_setup);
632 	evcnt_detach(&adapter->enomem_tx_dma_setup);
633 	evcnt_detach(&adapter->watchdog_events);
634 	evcnt_detach(&adapter->tso_err);
635 	evcnt_detach(&adapter->link_irq);
636 
637 	txr = adapter->tx_rings;
638 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
639 		evcnt_detach(&adapter->queues[i].irqs);
640 		evcnt_detach(&adapter->queues[i].handleq);
641 		evcnt_detach(&adapter->queues[i].req);
642 		evcnt_detach(&txr->no_desc_avail);
643 		evcnt_detach(&txr->total_packets);
644 		evcnt_detach(&txr->tso_tx);
645 #ifndef IXGBE_LEGACY_TX
646 		evcnt_detach(&txr->pcq_drops);
647 #endif
648 
649 		evcnt_detach(&rxr->rx_packets);
650 		evcnt_detach(&rxr->rx_bytes);
651 		evcnt_detach(&rxr->rx_copies);
652 		evcnt_detach(&rxr->no_jmbuf);
653 		evcnt_detach(&rxr->rx_discarded);
654 	}
655 	evcnt_detach(&stats->ipcs);
656 	evcnt_detach(&stats->l4cs);
657 	evcnt_detach(&stats->ipcs_bad);
658 	evcnt_detach(&stats->l4cs_bad);
659 
660 	/* Packet Reception Stats */
661 	evcnt_detach(&stats->vfgorc);
662 	evcnt_detach(&stats->vfgprc);
663 	evcnt_detach(&stats->vfmprc);
664 
665 	/* Packet Transmission Stats */
666 	evcnt_detach(&stats->vfgotc);
667 	evcnt_detach(&stats->vfgptc);
668 
669 	/* Mailbox Stats */
670 	evcnt_detach(&hw->mbx.stats.msgs_tx);
671 	evcnt_detach(&hw->mbx.stats.msgs_rx);
672 	evcnt_detach(&hw->mbx.stats.acks);
673 	evcnt_detach(&hw->mbx.stats.reqs);
674 	evcnt_detach(&hw->mbx.stats.rsts);
675 
676 	ixgbe_free_queues(adapter);
677 
678 	IXGBE_CORE_LOCK_DESTROY(adapter);
679 
680 	return (0);
681 } /* ixv_detach */
682 
683 /************************************************************************
684  * ixv_init_locked - Init entry point
685  *
686  *   Used in two ways: It is used by the stack as an init entry
687  *   point in network interface structure. It is also used
688  *   by the driver as a hw/sw initialization routine to get
689  *   to a consistent state.
690  *
691  *   return 0 on success, positive on failure
692  ************************************************************************/
693 static void
694 ixv_init_locked(struct adapter *adapter)
695 {
696 	struct ifnet	*ifp = adapter->ifp;
697 	device_t	dev = adapter->dev;
698 	struct ixgbe_hw *hw = &adapter->hw;
699 	struct ix_queue	*que;
700 	int		error = 0;
701 	uint32_t mask;
702 	int i;
703 
704 	INIT_DEBUGOUT("ixv_init_locked: begin");
705 	KASSERT(mutex_owned(&adapter->core_mtx));
706 	hw->adapter_stopped = FALSE;
707 	hw->mac.ops.stop_adapter(hw);
708 	callout_stop(&adapter->timer);
709 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
710 		que->disabled_count = 0;
711 
712 	adapter->max_frame_size =
713 	    ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
714 
715 	/* reprogram the RAR[0] in case user changed it. */
716 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
717 
718 	/* Get the latest mac address, User can use a LAA */
719 	memcpy(hw->mac.addr, CLLADDR(ifp->if_sadl),
720 	     IXGBE_ETH_LENGTH_OF_ADDRESS);
721 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, 1);
722 
723 	/* Prepare transmit descriptors and buffers */
724 	if (ixgbe_setup_transmit_structures(adapter)) {
725 		aprint_error_dev(dev, "Could not setup transmit structures\n");
726 		ixv_stop(adapter);
727 		return;
728 	}
729 
730 	/* Reset VF and renegotiate mailbox API version */
731 	hw->mac.ops.reset_hw(hw);
732 	hw->mac.ops.start_hw(hw);
733 	error = ixv_negotiate_api(adapter);
734 	if (error)
735 		device_printf(dev,
736 		    "Mailbox API negotiation failed in init_locked!\n");
737 
738 	ixv_initialize_transmit_units(adapter);
739 
740 	/* Setup Multicast table */
741 	ixv_set_rxfilter(adapter);
742 
743 	/*
744 	 * Determine the correct mbuf pool
745 	 * for doing jumbo/headersplit
746 	 */
747 	if (adapter->max_frame_size <= MCLBYTES)
748 		adapter->rx_mbuf_sz = MCLBYTES;
749 	else
750 		adapter->rx_mbuf_sz = MJUMPAGESIZE;
751 
752 	/* Prepare receive descriptors and buffers */
753 	if (ixgbe_setup_receive_structures(adapter)) {
754 		device_printf(dev, "Could not setup receive structures\n");
755 		ixv_stop(adapter);
756 		return;
757 	}
758 
759 	/* Configure RX settings */
760 	ixv_initialize_receive_units(adapter);
761 
762 	/* Set up VLAN offload and filter */
763 	ixv_setup_vlan_support(adapter);
764 
765 	/* Set up MSI-X routing */
766 	ixv_configure_ivars(adapter);
767 
768 	/* Set up auto-mask */
769 	mask = (1 << adapter->vector);
770 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
771 		mask |= (1 << que->msix);
772 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, mask);
773 
774 	/* Set moderation on the Link interrupt */
775 	ixv_eitr_write(adapter, adapter->vector, IXGBE_LINK_ITR);
776 
777 	/* Stats init */
778 	ixv_init_stats(adapter);
779 
780 	/* Config/Enable Link */
781 	hw->mac.get_link_status = TRUE;
782 	hw->mac.ops.check_link(hw, &adapter->link_speed, &adapter->link_up,
783 	    FALSE);
784 
785 	/* Start watchdog */
786 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
787 
788 	/* And now turn on interrupts */
789 	ixv_enable_intr(adapter);
790 
791 	/* Update saved flags. See ixgbe_ifflags_cb() */
792 	adapter->if_flags = ifp->if_flags;
793 	adapter->ec_capenable = adapter->osdep.ec.ec_capenable;
794 
795 	/* Now inform the stack we're ready */
796 	ifp->if_flags |= IFF_RUNNING;
797 	ifp->if_flags &= ~IFF_OACTIVE;
798 
799 	return;
800 } /* ixv_init_locked */
801 
802 /************************************************************************
803  * ixv_enable_queue
804  ************************************************************************/
805 static inline void
806 ixv_enable_queue(struct adapter *adapter, u32 vector)
807 {
808 	struct ixgbe_hw *hw = &adapter->hw;
809 	struct ix_queue *que = &adapter->queues[vector];
810 	u32		queue = 1UL << vector;
811 	u32		mask;
812 
813 	mutex_enter(&que->dc_mtx);
814 	if (que->disabled_count > 0 && --que->disabled_count > 0)
815 		goto out;
816 
817 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
818 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
819 out:
820 	mutex_exit(&que->dc_mtx);
821 } /* ixv_enable_queue */
822 
823 /************************************************************************
824  * ixv_disable_queue
825  ************************************************************************/
826 static inline void
827 ixv_disable_queue(struct adapter *adapter, u32 vector)
828 {
829 	struct ixgbe_hw *hw = &adapter->hw;
830 	struct ix_queue *que = &adapter->queues[vector];
831 	u32		queue = 1UL << vector;
832 	u32		mask;
833 
834 	mutex_enter(&que->dc_mtx);
835 	if (que->disabled_count++ > 0)
836 		goto  out;
837 
838 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
839 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, mask);
840 out:
841 	mutex_exit(&que->dc_mtx);
842 } /* ixv_disable_queue */
843 
844 #if 0
845 static inline void
846 ixv_rearm_queues(struct adapter *adapter, u64 queues)
847 {
848 	u32 mask = (IXGBE_EIMS_RTX_QUEUE & queues);
849 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEICS, mask);
850 } /* ixv_rearm_queues */
851 #endif
852 
853 
854 /************************************************************************
855  * ixv_msix_que - MSI-X Queue Interrupt Service routine
856  ************************************************************************/
857 static int
858 ixv_msix_que(void *arg)
859 {
860 	struct ix_queue	*que = arg;
861 	struct adapter	*adapter = que->adapter;
862 	struct tx_ring	*txr = que->txr;
863 	struct rx_ring	*rxr = que->rxr;
864 	bool		more;
865 	u32		newitr = 0;
866 
867 	ixv_disable_queue(adapter, que->msix);
868 	++que->irqs.ev_count;
869 
870 #ifdef __NetBSD__
871 	/* Don't run ixgbe_rxeof in interrupt context */
872 	more = true;
873 #else
874 	more = ixgbe_rxeof(que);
875 #endif
876 
877 	IXGBE_TX_LOCK(txr);
878 	ixgbe_txeof(txr);
879 	IXGBE_TX_UNLOCK(txr);
880 
881 	/* Do AIM now? */
882 
883 	if (adapter->enable_aim == false)
884 		goto no_calc;
885 	/*
886 	 * Do Adaptive Interrupt Moderation:
887 	 *  - Write out last calculated setting
888 	 *  - Calculate based on average size over
889 	 *    the last interval.
890 	 */
891 	if (que->eitr_setting)
892 		ixv_eitr_write(adapter, que->msix, que->eitr_setting);
893 
894 	que->eitr_setting = 0;
895 
896 	/* Idle, do nothing */
897 	if ((txr->bytes == 0) && (rxr->bytes == 0))
898 		goto no_calc;
899 
900 	if ((txr->bytes) && (txr->packets))
901 		newitr = txr->bytes/txr->packets;
902 	if ((rxr->bytes) && (rxr->packets))
903 		newitr = uimax(newitr, (rxr->bytes / rxr->packets));
904 	newitr += 24; /* account for hardware frame, crc */
905 
906 	/* set an upper boundary */
907 	newitr = uimin(newitr, 3000);
908 
909 	/* Be nice to the mid range */
910 	if ((newitr > 300) && (newitr < 1200))
911 		newitr = (newitr / 3);
912 	else
913 		newitr = (newitr / 2);
914 
915 	/*
916 	 * When RSC is used, ITR interval must be larger than RSC_DELAY.
917 	 * Currently, we use 2us for RSC_DELAY. The minimum value is always
918 	 * greater than 2us on 100M (and 10M?(not documented)), but it's not
919 	 * on 1G and higher.
920 	 */
921 	if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
922 	    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
923 		if (newitr < IXGBE_MIN_RSC_EITR_10G1G)
924 			newitr = IXGBE_MIN_RSC_EITR_10G1G;
925 	}
926 
927 	/* save for next interrupt */
928 	que->eitr_setting = newitr;
929 
930 	/* Reset state */
931 	txr->bytes = 0;
932 	txr->packets = 0;
933 	rxr->bytes = 0;
934 	rxr->packets = 0;
935 
936 no_calc:
937 	if (more)
938 		softint_schedule(que->que_si);
939 	else /* Re-enable this interrupt */
940 		ixv_enable_queue(adapter, que->msix);
941 
942 	return 1;
943 } /* ixv_msix_que */
944 
945 /************************************************************************
946  * ixv_msix_mbx
947  ************************************************************************/
948 static int
949 ixv_msix_mbx(void *arg)
950 {
951 	struct adapter	*adapter = arg;
952 	struct ixgbe_hw *hw = &adapter->hw;
953 
954 	++adapter->link_irq.ev_count;
955 	/* NetBSD: We use auto-clear, so it's not required to write VTEICR */
956 
957 	/* Link status change */
958 	hw->mac.get_link_status = TRUE;
959 	softint_schedule(adapter->link_si);
960 
961 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
962 
963 	return 1;
964 } /* ixv_msix_mbx */
965 
966 static void
967 ixv_eitr_write(struct adapter *adapter, uint32_t index, uint32_t itr)
968 {
969 
970 	/*
971 	 * Newer devices than 82598 have VF function, so this function is
972 	 * simple.
973 	 */
974 	itr |= IXGBE_EITR_CNT_WDIS;
975 
976 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEITR(index), itr);
977 }
978 
979 
980 /************************************************************************
981  * ixv_media_status - Media Ioctl callback
982  *
983  *   Called whenever the user queries the status of
984  *   the interface using ifconfig.
985  ************************************************************************/
986 static void
987 ixv_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
988 {
989 	struct adapter *adapter = ifp->if_softc;
990 
991 	INIT_DEBUGOUT("ixv_media_status: begin");
992 	ixv_update_link_status(adapter);
993 
994 	ifmr->ifm_status = IFM_AVALID;
995 	ifmr->ifm_active = IFM_ETHER;
996 
997 	if (adapter->link_active != LINK_STATE_UP) {
998 		ifmr->ifm_active |= IFM_NONE;
999 		return;
1000 	}
1001 
1002 	ifmr->ifm_status |= IFM_ACTIVE;
1003 
1004 	switch (adapter->link_speed) {
1005 		case IXGBE_LINK_SPEED_10GB_FULL:
1006 			ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
1007 			break;
1008 		case IXGBE_LINK_SPEED_5GB_FULL:
1009 			ifmr->ifm_active |= IFM_5000_T | IFM_FDX;
1010 			break;
1011 		case IXGBE_LINK_SPEED_2_5GB_FULL:
1012 			ifmr->ifm_active |= IFM_2500_T | IFM_FDX;
1013 			break;
1014 		case IXGBE_LINK_SPEED_1GB_FULL:
1015 			ifmr->ifm_active |= IFM_1000_T | IFM_FDX;
1016 			break;
1017 		case IXGBE_LINK_SPEED_100_FULL:
1018 			ifmr->ifm_active |= IFM_100_TX | IFM_FDX;
1019 			break;
1020 		case IXGBE_LINK_SPEED_10_FULL:
1021 			ifmr->ifm_active |= IFM_10_T | IFM_FDX;
1022 			break;
1023 	}
1024 
1025 	ifp->if_baudrate = ifmedia_baudrate(ifmr->ifm_active);
1026 } /* ixv_media_status */
1027 
1028 /************************************************************************
1029  * ixv_media_change - Media Ioctl callback
1030  *
1031  *   Called when the user changes speed/duplex using
1032  *   media/mediopt option with ifconfig.
1033  ************************************************************************/
1034 static int
1035 ixv_media_change(struct ifnet *ifp)
1036 {
1037 	struct adapter *adapter = ifp->if_softc;
1038 	struct ifmedia *ifm = &adapter->media;
1039 
1040 	INIT_DEBUGOUT("ixv_media_change: begin");
1041 
1042 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1043 		return (EINVAL);
1044 
1045 	switch (IFM_SUBTYPE(ifm->ifm_media)) {
1046 	case IFM_AUTO:
1047 		break;
1048 	default:
1049 		device_printf(adapter->dev, "Only auto media type\n");
1050 		return (EINVAL);
1051 	}
1052 
1053 	return (0);
1054 } /* ixv_media_change */
1055 
1056 /************************************************************************
1057  * ixv_negotiate_api
1058  *
1059  *   Negotiate the Mailbox API with the PF;
1060  *   start with the most featured API first.
1061  ************************************************************************/
1062 static int
1063 ixv_negotiate_api(struct adapter *adapter)
1064 {
1065 	struct ixgbe_hw *hw = &adapter->hw;
1066 	int		mbx_api[] = { ixgbe_mbox_api_13,
1067 				      ixgbe_mbox_api_12,
1068 				      ixgbe_mbox_api_11,
1069 				      ixgbe_mbox_api_10,
1070 				      ixgbe_mbox_api_unknown };
1071 	int		i = 0;
1072 
1073 	while (mbx_api[i] != ixgbe_mbox_api_unknown) {
1074 		if (ixgbevf_negotiate_api_version(hw, mbx_api[i]) == 0)
1075 			return (0);
1076 		i++;
1077 	}
1078 
1079 	return (EINVAL);
1080 } /* ixv_negotiate_api */
1081 
1082 
1083 /************************************************************************
1084  * ixv_set_multi - Multicast Update
1085  *
1086  *   Called whenever multicast address list is updated.
1087  ************************************************************************/
1088 static int
1089 ixv_set_rxfilter(struct adapter *adapter)
1090 {
1091 	u8	mta[IXGBE_MAX_VF_MC * IXGBE_ETH_LENGTH_OF_ADDRESS];
1092 	struct ifnet		*ifp = adapter->ifp;
1093 	struct ixgbe_hw		*hw = &adapter->hw;
1094 	u8			*update_ptr;
1095 	int			mcnt = 0;
1096 	struct ethercom		*ec = &adapter->osdep.ec;
1097 	struct ether_multi	*enm;
1098 	struct ether_multistep	step;
1099 	bool			overflow = false;
1100 	int			error, rc = 0;
1101 
1102 	KASSERT(mutex_owned(&adapter->core_mtx));
1103 	IOCTL_DEBUGOUT("ixv_set_rxfilter: begin");
1104 
1105 	/* 1: For PROMISC */
1106 	if (ifp->if_flags & IFF_PROMISC) {
1107 		error = hw->mac.ops.update_xcast_mode(hw,
1108 		    IXGBEVF_XCAST_MODE_PROMISC);
1109 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1110 			device_printf(adapter->dev,
1111 			    "this interface is not trusted\n");
1112 			error = EPERM;
1113 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1114 			device_printf(adapter->dev,
1115 			    "the PF doesn't support promisc mode\n");
1116 			error = EOPNOTSUPP;
1117 		} else if (error == IXGBE_ERR_NOT_IN_PROMISC) {
1118 			device_printf(adapter->dev,
1119 			    "the PF may not in promisc mode\n");
1120 			error = EINVAL;
1121 		} else if (error) {
1122 			device_printf(adapter->dev,
1123 			    "failed to set promisc mode. error = %d\n",
1124 			    error);
1125 			error = EIO;
1126 		} else
1127 			return 0;
1128 		rc = error;
1129 	}
1130 
1131 	/* 2: For ALLMULTI or normal */
1132 	ETHER_LOCK(ec);
1133 	ETHER_FIRST_MULTI(step, ec, enm);
1134 	while (enm != NULL) {
1135 		if ((mcnt >= IXGBE_MAX_VF_MC) ||
1136 		    (memcmp(enm->enm_addrlo, enm->enm_addrhi,
1137 			ETHER_ADDR_LEN) != 0)) {
1138 			overflow = true;
1139 			break;
1140 		}
1141 		bcopy(enm->enm_addrlo,
1142 		    &mta[mcnt * IXGBE_ETH_LENGTH_OF_ADDRESS],
1143 		    IXGBE_ETH_LENGTH_OF_ADDRESS);
1144 		mcnt++;
1145 		ETHER_NEXT_MULTI(step, enm);
1146 	}
1147 	ETHER_UNLOCK(ec);
1148 
1149 	/* 3: For ALLMULTI */
1150 	if (overflow) {
1151 		error = hw->mac.ops.update_xcast_mode(hw,
1152 		    IXGBEVF_XCAST_MODE_ALLMULTI);
1153 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1154 			device_printf(adapter->dev,
1155 			    "this interface is not trusted\n");
1156 			error = EPERM;
1157 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1158 			device_printf(adapter->dev,
1159 			    "the PF doesn't support allmulti mode\n");
1160 			error = EOPNOTSUPP;
1161 		} else if (error) {
1162 			device_printf(adapter->dev,
1163 			    "number of Ethernet multicast addresses "
1164 			    "exceeds the limit (%d). error = %d\n",
1165 			    IXGBE_MAX_VF_MC, error);
1166 			error = ENOSPC;
1167 		} else {
1168 			ETHER_LOCK(ec);
1169 			ec->ec_flags |= ETHER_F_ALLMULTI;
1170 			ETHER_UNLOCK(ec);
1171 			return rc; /* Promisc might failed */
1172 		}
1173 
1174 		if (rc == 0)
1175 			rc = error;
1176 
1177 		/* Continue to update the multicast table as many as we can */
1178 	}
1179 
1180 	/* 4: For normal operation */
1181 	error = hw->mac.ops.update_xcast_mode(hw, IXGBEVF_XCAST_MODE_MULTI);
1182 	if ((error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) || (error == 0)) {
1183 		/* Normal operation */
1184 		ETHER_LOCK(ec);
1185 		ec->ec_flags &= ~ETHER_F_ALLMULTI;
1186 		ETHER_UNLOCK(ec);
1187 		error = 0;
1188 	} else if (error) {
1189 		device_printf(adapter->dev,
1190 		    "failed to set Ethernet multicast address "
1191 		    "operation to normal. error = %d\n", error);
1192 	}
1193 
1194 	update_ptr = mta;
1195 
1196 	error = adapter->hw.mac.ops.update_mc_addr_list(&adapter->hw,
1197 	    update_ptr, mcnt, ixv_mc_array_itr, TRUE);
1198 	if (rc == 0)
1199 		rc = error;
1200 
1201 	return rc;
1202 } /* ixv_set_rxfilter */
1203 
1204 /************************************************************************
1205  * ixv_mc_array_itr
1206  *
1207  *   An iterator function needed by the multicast shared code.
1208  *   It feeds the shared code routine the addresses in the
1209  *   array of ixv_set_rxfilter() one by one.
1210  ************************************************************************/
1211 static u8 *
1212 ixv_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq)
1213 {
1214 	u8 *addr = *update_ptr;
1215 	u8 *newptr;
1216 
1217 	*vmdq = 0;
1218 
1219 	newptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
1220 	*update_ptr = newptr;
1221 
1222 	return addr;
1223 } /* ixv_mc_array_itr */
1224 
1225 /************************************************************************
1226  * ixv_local_timer - Timer routine
1227  *
1228  *   Checks for link status, updates statistics,
1229  *   and runs the watchdog check.
1230  ************************************************************************/
1231 static void
1232 ixv_local_timer(void *arg)
1233 {
1234 	struct adapter *adapter = arg;
1235 
1236 	IXGBE_CORE_LOCK(adapter);
1237 	ixv_local_timer_locked(adapter);
1238 	IXGBE_CORE_UNLOCK(adapter);
1239 }
1240 
1241 static void
1242 ixv_local_timer_locked(void *arg)
1243 {
1244 	struct adapter	*adapter = arg;
1245 	device_t	dev = adapter->dev;
1246 	struct ix_queue	*que = adapter->queues;
1247 	u64		queues = 0;
1248 	u64		v0, v1, v2, v3, v4, v5, v6, v7;
1249 	int		hung = 0;
1250 	int		i;
1251 
1252 	KASSERT(mutex_owned(&adapter->core_mtx));
1253 
1254 	if (ixv_check_link(adapter)) {
1255 		ixv_init_locked(adapter);
1256 		return;
1257 	}
1258 
1259 	/* Stats Update */
1260 	ixv_update_stats(adapter);
1261 
1262 	/* Update some event counters */
1263 	v0 = v1 = v2 = v3 = v4 = v5 = v6 = v7 = 0;
1264 	que = adapter->queues;
1265 	for (i = 0; i < adapter->num_queues; i++, que++) {
1266 		struct tx_ring	*txr = que->txr;
1267 
1268 		v0 += txr->q_efbig_tx_dma_setup;
1269 		v1 += txr->q_mbuf_defrag_failed;
1270 		v2 += txr->q_efbig2_tx_dma_setup;
1271 		v3 += txr->q_einval_tx_dma_setup;
1272 		v4 += txr->q_other_tx_dma_setup;
1273 		v5 += txr->q_eagain_tx_dma_setup;
1274 		v6 += txr->q_enomem_tx_dma_setup;
1275 		v7 += txr->q_tso_err;
1276 	}
1277 	adapter->efbig_tx_dma_setup.ev_count = v0;
1278 	adapter->mbuf_defrag_failed.ev_count = v1;
1279 	adapter->efbig2_tx_dma_setup.ev_count = v2;
1280 	adapter->einval_tx_dma_setup.ev_count = v3;
1281 	adapter->other_tx_dma_setup.ev_count = v4;
1282 	adapter->eagain_tx_dma_setup.ev_count = v5;
1283 	adapter->enomem_tx_dma_setup.ev_count = v6;
1284 	adapter->tso_err.ev_count = v7;
1285 
1286 	/*
1287 	 * Check the TX queues status
1288 	 *	- mark hung queues so we don't schedule on them
1289 	 *	- watchdog only if all queues show hung
1290 	 */
1291 	que = adapter->queues;
1292 	for (i = 0; i < adapter->num_queues; i++, que++) {
1293 		/* Keep track of queues with work for soft irq */
1294 		if (que->txr->busy)
1295 			queues |= ((u64)1 << que->me);
1296 		/*
1297 		 * Each time txeof runs without cleaning, but there
1298 		 * are uncleaned descriptors it increments busy. If
1299 		 * we get to the MAX we declare it hung.
1300 		 */
1301 		if (que->busy == IXGBE_QUEUE_HUNG) {
1302 			++hung;
1303 			/* Mark the queue as inactive */
1304 			adapter->active_queues &= ~((u64)1 << que->me);
1305 			continue;
1306 		} else {
1307 			/* Check if we've come back from hung */
1308 			if ((adapter->active_queues & ((u64)1 << que->me)) == 0)
1309 				adapter->active_queues |= ((u64)1 << que->me);
1310 		}
1311 		if (que->busy >= IXGBE_MAX_TX_BUSY) {
1312 			device_printf(dev,
1313 			    "Warning queue %d appears to be hung!\n", i);
1314 			que->txr->busy = IXGBE_QUEUE_HUNG;
1315 			++hung;
1316 		}
1317 	}
1318 
1319 	/* Only truly watchdog if all queues show hung */
1320 	if (hung == adapter->num_queues)
1321 		goto watchdog;
1322 #if 0
1323 	else if (queues != 0) { /* Force an IRQ on queues with work */
1324 		ixv_rearm_queues(adapter, queues);
1325 	}
1326 #endif
1327 
1328 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
1329 
1330 	return;
1331 
1332 watchdog:
1333 
1334 	device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
1335 	adapter->ifp->if_flags &= ~IFF_RUNNING;
1336 	adapter->watchdog_events.ev_count++;
1337 	ixv_init_locked(adapter);
1338 } /* ixv_local_timer */
1339 
1340 /************************************************************************
1341  * ixv_update_link_status - Update OS on link state
1342  *
1343  * Note: Only updates the OS on the cached link state.
1344  *	 The real check of the hardware only happens with
1345  *	 a link interrupt.
1346  ************************************************************************/
1347 static void
1348 ixv_update_link_status(struct adapter *adapter)
1349 {
1350 	struct ifnet *ifp = adapter->ifp;
1351 	device_t     dev = adapter->dev;
1352 
1353 	KASSERT(mutex_owned(&adapter->core_mtx));
1354 
1355 	if (adapter->link_up) {
1356 		if (adapter->link_active != LINK_STATE_UP) {
1357 			if (bootverbose) {
1358 				const char *bpsmsg;
1359 
1360 				switch (adapter->link_speed) {
1361 				case IXGBE_LINK_SPEED_10GB_FULL:
1362 					bpsmsg = "10 Gbps";
1363 					break;
1364 				case IXGBE_LINK_SPEED_5GB_FULL:
1365 					bpsmsg = "5 Gbps";
1366 					break;
1367 				case IXGBE_LINK_SPEED_2_5GB_FULL:
1368 					bpsmsg = "2.5 Gbps";
1369 					break;
1370 				case IXGBE_LINK_SPEED_1GB_FULL:
1371 					bpsmsg = "1 Gbps";
1372 					break;
1373 				case IXGBE_LINK_SPEED_100_FULL:
1374 					bpsmsg = "100 Mbps";
1375 					break;
1376 				case IXGBE_LINK_SPEED_10_FULL:
1377 					bpsmsg = "10 Mbps";
1378 					break;
1379 				default:
1380 					bpsmsg = "unknown speed";
1381 					break;
1382 				}
1383 				device_printf(dev, "Link is up %s %s \n",
1384 				    bpsmsg, "Full Duplex");
1385 			}
1386 			adapter->link_active = LINK_STATE_UP;
1387 			if_link_state_change(ifp, LINK_STATE_UP);
1388 		}
1389 	} else {
1390 		/*
1391 		 * Do it when link active changes to DOWN. i.e.
1392 		 * a) LINK_STATE_UNKNOWN -> LINK_STATE_DOWN
1393 		 * b) LINK_STATE_UP	 -> LINK_STATE_DOWN
1394 		 */
1395 		if (adapter->link_active != LINK_STATE_DOWN) {
1396 			if (bootverbose)
1397 				device_printf(dev, "Link is Down\n");
1398 			if_link_state_change(ifp, LINK_STATE_DOWN);
1399 			adapter->link_active = LINK_STATE_DOWN;
1400 		}
1401 	}
1402 } /* ixv_update_link_status */
1403 
1404 
1405 /************************************************************************
1406  * ixv_stop - Stop the hardware
1407  *
1408  *   Disables all traffic on the adapter by issuing a
1409  *   global reset on the MAC and deallocates TX/RX buffers.
1410  ************************************************************************/
1411 static void
1412 ixv_ifstop(struct ifnet *ifp, int disable)
1413 {
1414 	struct adapter *adapter = ifp->if_softc;
1415 
1416 	IXGBE_CORE_LOCK(adapter);
1417 	ixv_stop(adapter);
1418 	IXGBE_CORE_UNLOCK(adapter);
1419 }
1420 
1421 static void
1422 ixv_stop(void *arg)
1423 {
1424 	struct ifnet	*ifp;
1425 	struct adapter	*adapter = arg;
1426 	struct ixgbe_hw *hw = &adapter->hw;
1427 
1428 	ifp = adapter->ifp;
1429 
1430 	KASSERT(mutex_owned(&adapter->core_mtx));
1431 
1432 	INIT_DEBUGOUT("ixv_stop: begin\n");
1433 	ixv_disable_intr(adapter);
1434 
1435 	/* Tell the stack that the interface is no longer active */
1436 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1437 
1438 	hw->mac.ops.reset_hw(hw);
1439 	adapter->hw.adapter_stopped = FALSE;
1440 	hw->mac.ops.stop_adapter(hw);
1441 	callout_stop(&adapter->timer);
1442 
1443 	/* reprogram the RAR[0] in case user changed it. */
1444 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
1445 
1446 	return;
1447 } /* ixv_stop */
1448 
1449 
1450 /************************************************************************
1451  * ixv_allocate_pci_resources
1452  ************************************************************************/
1453 static int
1454 ixv_allocate_pci_resources(struct adapter *adapter,
1455     const struct pci_attach_args *pa)
1456 {
1457 	pcireg_t	memtype, csr;
1458 	device_t	dev = adapter->dev;
1459 	bus_addr_t addr;
1460 	int flags;
1461 
1462 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, PCI_BAR(0));
1463 	switch (memtype) {
1464 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
1465 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
1466 		adapter->osdep.mem_bus_space_tag = pa->pa_memt;
1467 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, PCI_BAR(0),
1468 		      memtype, &addr, &adapter->osdep.mem_size, &flags) != 0)
1469 			goto map_err;
1470 		if ((flags & BUS_SPACE_MAP_PREFETCHABLE) != 0) {
1471 			aprint_normal_dev(dev, "clearing prefetchable bit\n");
1472 			flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
1473 		}
1474 		if (bus_space_map(adapter->osdep.mem_bus_space_tag, addr,
1475 		     adapter->osdep.mem_size, flags,
1476 		     &adapter->osdep.mem_bus_space_handle) != 0) {
1477 map_err:
1478 			adapter->osdep.mem_size = 0;
1479 			aprint_error_dev(dev, "unable to map BAR0\n");
1480 			return ENXIO;
1481 		}
1482 		/*
1483 		 * Enable address decoding for memory range in case it's not
1484 		 * set.
1485 		 */
1486 		csr = pci_conf_read(pa->pa_pc, pa->pa_tag,
1487 		    PCI_COMMAND_STATUS_REG);
1488 		csr |= PCI_COMMAND_MEM_ENABLE;
1489 		pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
1490 		    csr);
1491 		break;
1492 	default:
1493 		aprint_error_dev(dev, "unexpected type on BAR0\n");
1494 		return ENXIO;
1495 	}
1496 
1497 	/* Pick up the tuneable queues */
1498 	adapter->num_queues = ixv_num_queues;
1499 
1500 	return (0);
1501 } /* ixv_allocate_pci_resources */
1502 
1503 /************************************************************************
1504  * ixv_free_pci_resources
1505  ************************************************************************/
1506 static void
1507 ixv_free_pci_resources(struct adapter * adapter)
1508 {
1509 	struct		ix_queue *que = adapter->queues;
1510 	int		rid;
1511 
1512 	/*
1513 	 *  Release all msix queue resources:
1514 	 */
1515 	for (int i = 0; i < adapter->num_queues; i++, que++) {
1516 		if (que->res != NULL)
1517 			pci_intr_disestablish(adapter->osdep.pc,
1518 			    adapter->osdep.ihs[i]);
1519 	}
1520 
1521 
1522 	/* Clean the Mailbox interrupt last */
1523 	rid = adapter->vector;
1524 
1525 	if (adapter->osdep.ihs[rid] != NULL) {
1526 		pci_intr_disestablish(adapter->osdep.pc,
1527 		    adapter->osdep.ihs[rid]);
1528 		adapter->osdep.ihs[rid] = NULL;
1529 	}
1530 
1531 	pci_intr_release(adapter->osdep.pc, adapter->osdep.intrs,
1532 	    adapter->osdep.nintrs);
1533 
1534 	if (adapter->osdep.mem_size != 0) {
1535 		bus_space_unmap(adapter->osdep.mem_bus_space_tag,
1536 		    adapter->osdep.mem_bus_space_handle,
1537 		    adapter->osdep.mem_size);
1538 	}
1539 
1540 	return;
1541 } /* ixv_free_pci_resources */
1542 
1543 /************************************************************************
1544  * ixv_setup_interface
1545  *
1546  *   Setup networking device structure and register an interface.
1547  ************************************************************************/
1548 static int
1549 ixv_setup_interface(device_t dev, struct adapter *adapter)
1550 {
1551 	struct ethercom *ec = &adapter->osdep.ec;
1552 	struct ifnet   *ifp;
1553 	int rv;
1554 
1555 	INIT_DEBUGOUT("ixv_setup_interface: begin");
1556 
1557 	ifp = adapter->ifp = &ec->ec_if;
1558 	strlcpy(ifp->if_xname, device_xname(dev), IFNAMSIZ);
1559 	ifp->if_baudrate = IF_Gbps(10);
1560 	ifp->if_init = ixv_init;
1561 	ifp->if_stop = ixv_ifstop;
1562 	ifp->if_softc = adapter;
1563 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1564 #ifdef IXGBE_MPSAFE
1565 	ifp->if_extflags = IFEF_MPSAFE;
1566 #endif
1567 	ifp->if_ioctl = ixv_ioctl;
1568 	if (adapter->feat_en & IXGBE_FEATURE_LEGACY_TX) {
1569 #if 0
1570 		ixv_start_locked = ixgbe_legacy_start_locked;
1571 #endif
1572 	} else {
1573 		ifp->if_transmit = ixgbe_mq_start;
1574 #if 0
1575 		ixv_start_locked = ixgbe_mq_start_locked;
1576 #endif
1577 	}
1578 	ifp->if_start = ixgbe_legacy_start;
1579 	IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 2);
1580 	IFQ_SET_READY(&ifp->if_snd);
1581 
1582 	rv = if_initialize(ifp);
1583 	if (rv != 0) {
1584 		aprint_error_dev(dev, "if_initialize failed(%d)\n", rv);
1585 		return rv;
1586 	}
1587 	adapter->ipq = if_percpuq_create(&adapter->osdep.ec.ec_if);
1588 	ether_ifattach(ifp, adapter->hw.mac.addr);
1589 	aprint_normal_dev(dev, "Ethernet address %s\n",
1590 	    ether_sprintf(adapter->hw.mac.addr));
1591 	/*
1592 	 * We use per TX queue softint, so if_deferred_start_init() isn't
1593 	 * used.
1594 	 */
1595 	ether_set_ifflags_cb(ec, ixv_ifflags_cb);
1596 
1597 	adapter->max_frame_size = ifp->if_mtu + IXGBE_MTU_HDR;
1598 
1599 	/*
1600 	 * Tell the upper layer(s) we support long frames.
1601 	 */
1602 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1603 
1604 	/* Set capability flags */
1605 	ifp->if_capabilities |= IFCAP_HWCSUM
1606 			     |	IFCAP_TSOv4
1607 			     |	IFCAP_TSOv6;
1608 	ifp->if_capenable = 0;
1609 
1610 	ec->ec_capabilities |= ETHERCAP_VLAN_HWFILTER
1611 			    |  ETHERCAP_VLAN_HWTAGGING
1612 			    |  ETHERCAP_VLAN_HWCSUM
1613 			    |  ETHERCAP_JUMBO_MTU
1614 			    |  ETHERCAP_VLAN_MTU;
1615 
1616 	/* Enable the above capabilities by default */
1617 	ec->ec_capenable = ec->ec_capabilities;
1618 
1619 	/* Don't enable LRO by default */
1620 #if 0
1621 	/* NetBSD doesn't support LRO yet */
1622 	ifp->if_capabilities |= IFCAP_LRO;
1623 #endif
1624 
1625 	/*
1626 	 * Specify the media types supported by this adapter and register
1627 	 * callbacks to update media and link information
1628 	 */
1629 	ec->ec_ifmedia = &adapter->media;
1630 	ifmedia_init_with_lock(&adapter->media, IFM_IMASK, ixv_media_change,
1631 	    ixv_media_status, &adapter->core_mtx);
1632 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1633 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
1634 
1635 	if_register(ifp);
1636 
1637 	return 0;
1638 } /* ixv_setup_interface */
1639 
1640 
1641 /************************************************************************
1642  * ixv_initialize_transmit_units - Enable transmit unit.
1643  ************************************************************************/
1644 static void
1645 ixv_initialize_transmit_units(struct adapter *adapter)
1646 {
1647 	struct tx_ring	*txr = adapter->tx_rings;
1648 	struct ixgbe_hw	*hw = &adapter->hw;
1649 	int i;
1650 
1651 	for (i = 0; i < adapter->num_queues; i++, txr++) {
1652 		u64 tdba = txr->txdma.dma_paddr;
1653 		u32 txctrl, txdctl;
1654 		int j = txr->me;
1655 
1656 		/* Set WTHRESH to 8, burst writeback */
1657 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1658 		txdctl |= (8 << 16);
1659 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1660 
1661 		/* Set the HW Tx Head and Tail indices */
1662 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDH(j), 0);
1663 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDT(j), 0);
1664 
1665 		/* Set Tx Tail register */
1666 		txr->tail = IXGBE_VFTDT(j);
1667 
1668 		txr->txr_no_space = false;
1669 
1670 		/* Set Ring parameters */
1671 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1672 		    (tdba & 0x00000000ffffffffULL));
1673 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1674 		IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j),
1675 		    adapter->num_tx_desc * sizeof(struct ixgbe_legacy_tx_desc));
1676 		txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1677 		txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
1678 		IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1679 
1680 		/* Now enable */
1681 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1682 		txdctl |= IXGBE_TXDCTL_ENABLE;
1683 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1684 	}
1685 
1686 	return;
1687 } /* ixv_initialize_transmit_units */
1688 
1689 
1690 /************************************************************************
1691  * ixv_initialize_rss_mapping
1692  ************************************************************************/
1693 static void
1694 ixv_initialize_rss_mapping(struct adapter *adapter)
1695 {
1696 	struct ixgbe_hw *hw = &adapter->hw;
1697 	u32		reta = 0, mrqc, rss_key[10];
1698 	int		queue_id;
1699 	int		i, j;
1700 	u32		rss_hash_config;
1701 
1702 	/* force use default RSS key. */
1703 #ifdef __NetBSD__
1704 	rss_getkey((uint8_t *) &rss_key);
1705 #else
1706 	if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1707 		/* Fetch the configured RSS key */
1708 		rss_getkey((uint8_t *)&rss_key);
1709 	} else {
1710 		/* set up random bits */
1711 		cprng_fast(&rss_key, sizeof(rss_key));
1712 	}
1713 #endif
1714 
1715 	/* Now fill out hash function seeds */
1716 	for (i = 0; i < 10; i++)
1717 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1718 
1719 	/* Set up the redirection table */
1720 	for (i = 0, j = 0; i < 64; i++, j++) {
1721 		if (j == adapter->num_queues)
1722 			j = 0;
1723 
1724 		if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1725 			/*
1726 			 * Fetch the RSS bucket id for the given indirection
1727 			 * entry. Cap it at the number of configured buckets
1728 			 * (which is num_queues.)
1729 			 */
1730 			queue_id = rss_get_indirection_to_bucket(i);
1731 			queue_id = queue_id % adapter->num_queues;
1732 		} else
1733 			queue_id = j;
1734 
1735 		/*
1736 		 * The low 8 bits are for hash value (n+0);
1737 		 * The next 8 bits are for hash value (n+1), etc.
1738 		 */
1739 		reta >>= 8;
1740 		reta |= ((uint32_t)queue_id) << 24;
1741 		if ((i & 3) == 3) {
1742 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), reta);
1743 			reta = 0;
1744 		}
1745 	}
1746 
1747 	/* Perform hash on these packet types */
1748 	if (adapter->feat_en & IXGBE_FEATURE_RSS)
1749 		rss_hash_config = rss_gethashconfig();
1750 	else {
1751 		/*
1752 		 * Disable UDP - IP fragments aren't currently being handled
1753 		 * and so we end up with a mix of 2-tuple and 4-tuple
1754 		 * traffic.
1755 		 */
1756 		rss_hash_config = RSS_HASHTYPE_RSS_IPV4
1757 				| RSS_HASHTYPE_RSS_TCP_IPV4
1758 				| RSS_HASHTYPE_RSS_IPV6
1759 				| RSS_HASHTYPE_RSS_TCP_IPV6;
1760 	}
1761 
1762 	mrqc = IXGBE_MRQC_RSSEN;
1763 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV4)
1764 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
1765 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV4)
1766 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
1767 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6)
1768 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
1769 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6)
1770 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
1771 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6_EX)
1772 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_IPV6_EX defined, but not supported\n",
1773 		    __func__);
1774 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6_EX)
1775 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_TCP_IPV6_EX defined, but not supported\n",
1776 		    __func__);
1777 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV4)
1778 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
1779 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6)
1780 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1781 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6_EX)
1782 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_UDP_IPV6_EX defined, but not supported\n",
1783 		    __func__);
1784 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, mrqc);
1785 } /* ixv_initialize_rss_mapping */
1786 
1787 
1788 /************************************************************************
1789  * ixv_initialize_receive_units - Setup receive registers and features.
1790  ************************************************************************/
1791 static void
1792 ixv_initialize_receive_units(struct adapter *adapter)
1793 {
1794 	struct	rx_ring	*rxr = adapter->rx_rings;
1795 	struct ixgbe_hw	*hw = &adapter->hw;
1796 	struct ifnet	*ifp = adapter->ifp;
1797 	u32		bufsz, psrtype;
1798 
1799 	if (ifp->if_mtu > ETHERMTU)
1800 		bufsz = 4096 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1801 	else
1802 		bufsz = 2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1803 
1804 	psrtype = IXGBE_PSRTYPE_TCPHDR
1805 		| IXGBE_PSRTYPE_UDPHDR
1806 		| IXGBE_PSRTYPE_IPV4HDR
1807 		| IXGBE_PSRTYPE_IPV6HDR
1808 		| IXGBE_PSRTYPE_L2HDR;
1809 
1810 	if (adapter->num_queues > 1)
1811 		psrtype |= 1 << 29;
1812 
1813 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1814 
1815 	/* Tell PF our max_frame size */
1816 	if (ixgbevf_rlpml_set_vf(hw, adapter->max_frame_size) != 0) {
1817 		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");
1818 	}
1819 
1820 	for (int i = 0; i < adapter->num_queues; i++, rxr++) {
1821 		u64 rdba = rxr->rxdma.dma_paddr;
1822 		u32 reg, rxdctl;
1823 		int j = rxr->me;
1824 
1825 		/* Disable the queue */
1826 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1827 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1828 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1829 		for (int k = 0; k < 10; k++) {
1830 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1831 			    IXGBE_RXDCTL_ENABLE)
1832 				msec_delay(1);
1833 			else
1834 				break;
1835 		}
1836 		IXGBE_WRITE_BARRIER(hw);
1837 		/* Setup the Base and Length of the Rx Descriptor Ring */
1838 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1839 		    (rdba & 0x00000000ffffffffULL));
1840 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1841 		IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j),
1842 		    adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc));
1843 
1844 		/* Reset the ring indices */
1845 		IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
1846 		IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), 0);
1847 
1848 		/* Set up the SRRCTL register */
1849 		reg = IXGBE_READ_REG(hw, IXGBE_VFSRRCTL(j));
1850 		reg &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1851 		reg &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1852 		reg |= bufsz;
1853 		reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1854 		IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(j), reg);
1855 
1856 		/* Capture Rx Tail index */
1857 		rxr->tail = IXGBE_VFRDT(rxr->me);
1858 
1859 		/* Do the queue enabling last */
1860 		rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1861 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1862 		for (int k = 0; k < 10; k++) {
1863 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1864 			    IXGBE_RXDCTL_ENABLE)
1865 				break;
1866 			msec_delay(1);
1867 		}
1868 		IXGBE_WRITE_BARRIER(hw);
1869 
1870 		/* Set the Tail Pointer */
1871 #ifdef DEV_NETMAP
1872 		/*
1873 		 * In netmap mode, we must preserve the buffers made
1874 		 * available to userspace before the if_init()
1875 		 * (this is true by default on the TX side, because
1876 		 * init makes all buffers available to userspace).
1877 		 *
1878 		 * netmap_reset() and the device specific routines
1879 		 * (e.g. ixgbe_setup_receive_rings()) map these
1880 		 * buffers at the end of the NIC ring, so here we
1881 		 * must set the RDT (tail) register to make sure
1882 		 * they are not overwritten.
1883 		 *
1884 		 * In this driver the NIC ring starts at RDH = 0,
1885 		 * RDT points to the last slot available for reception (?),
1886 		 * so RDT = num_rx_desc - 1 means the whole ring is available.
1887 		 */
1888 		if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) &&
1889 		    (ifp->if_capenable & IFCAP_NETMAP)) {
1890 			struct netmap_adapter *na = NA(adapter->ifp);
1891 			struct netmap_kring *kring = na->rx_rings[i];
1892 			int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring);
1893 
1894 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), t);
1895 		} else
1896 #endif /* DEV_NETMAP */
1897 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
1898 			    adapter->num_rx_desc - 1);
1899 	}
1900 
1901 	if (adapter->hw.mac.type >= ixgbe_mac_X550_vf)
1902 		ixv_initialize_rss_mapping(adapter);
1903 } /* ixv_initialize_receive_units */
1904 
1905 /************************************************************************
1906  * ixv_sysctl_tdh_handler - Transmit Descriptor Head handler function
1907  *
1908  *   Retrieves the TDH value from the hardware
1909  ************************************************************************/
1910 static int
1911 ixv_sysctl_tdh_handler(SYSCTLFN_ARGS)
1912 {
1913 	struct sysctlnode node = *rnode;
1914 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1915 	uint32_t val;
1916 
1917 	if (!txr)
1918 		return (0);
1919 
1920 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDH(txr->me));
1921 	node.sysctl_data = &val;
1922 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1923 } /* ixv_sysctl_tdh_handler */
1924 
1925 /************************************************************************
1926  * ixgbe_sysctl_tdt_handler - Transmit Descriptor Tail handler function
1927  *
1928  *   Retrieves the TDT value from the hardware
1929  ************************************************************************/
1930 static int
1931 ixv_sysctl_tdt_handler(SYSCTLFN_ARGS)
1932 {
1933 	struct sysctlnode node = *rnode;
1934 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1935 	uint32_t val;
1936 
1937 	if (!txr)
1938 		return (0);
1939 
1940 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDT(txr->me));
1941 	node.sysctl_data = &val;
1942 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1943 } /* ixv_sysctl_tdt_handler */
1944 
1945 /************************************************************************
1946  * ixv_sysctl_next_to_check_handler - Receive Descriptor next to check
1947  * handler function
1948  *
1949  *   Retrieves the next_to_check value
1950  ************************************************************************/
1951 static int
1952 ixv_sysctl_next_to_check_handler(SYSCTLFN_ARGS)
1953 {
1954 	struct sysctlnode node = *rnode;
1955 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1956 	uint32_t val;
1957 
1958 	if (!rxr)
1959 		return (0);
1960 
1961 	val = rxr->next_to_check;
1962 	node.sysctl_data = &val;
1963 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1964 } /* ixv_sysctl_next_to_check_handler */
1965 
1966 /************************************************************************
1967  * ixv_sysctl_rdh_handler - Receive Descriptor Head handler function
1968  *
1969  *   Retrieves the RDH value from the hardware
1970  ************************************************************************/
1971 static int
1972 ixv_sysctl_rdh_handler(SYSCTLFN_ARGS)
1973 {
1974 	struct sysctlnode node = *rnode;
1975 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1976 	uint32_t val;
1977 
1978 	if (!rxr)
1979 		return (0);
1980 
1981 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDH(rxr->me));
1982 	node.sysctl_data = &val;
1983 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1984 } /* ixv_sysctl_rdh_handler */
1985 
1986 /************************************************************************
1987  * ixv_sysctl_rdt_handler - Receive Descriptor Tail handler function
1988  *
1989  *   Retrieves the RDT value from the hardware
1990  ************************************************************************/
1991 static int
1992 ixv_sysctl_rdt_handler(SYSCTLFN_ARGS)
1993 {
1994 	struct sysctlnode node = *rnode;
1995 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
1996 	uint32_t val;
1997 
1998 	if (!rxr)
1999 		return (0);
2000 
2001 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDT(rxr->me));
2002 	node.sysctl_data = &val;
2003 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2004 } /* ixv_sysctl_rdt_handler */
2005 
2006 static void
2007 ixv_setup_vlan_tagging(struct adapter *adapter)
2008 {
2009 	struct ethercom *ec = &adapter->osdep.ec;
2010 	struct ixgbe_hw *hw = &adapter->hw;
2011 	struct rx_ring	*rxr;
2012 	u32		ctrl;
2013 	int		i;
2014 	bool		hwtagging;
2015 
2016 	/* Enable HW tagging only if any vlan is attached */
2017 	hwtagging = (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING)
2018 	    && VLAN_ATTACHED(ec);
2019 
2020 	/* Enable the queues */
2021 	for (i = 0; i < adapter->num_queues; i++) {
2022 		rxr = &adapter->rx_rings[i];
2023 		ctrl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(rxr->me));
2024 		if (hwtagging)
2025 			ctrl |= IXGBE_RXDCTL_VME;
2026 		else
2027 			ctrl &= ~IXGBE_RXDCTL_VME;
2028 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(rxr->me), ctrl);
2029 		/*
2030 		 * Let Rx path know that it needs to store VLAN tag
2031 		 * as part of extra mbuf info.
2032 		 */
2033 		rxr->vtag_strip = hwtagging ? TRUE : FALSE;
2034 	}
2035 } /* ixv_setup_vlan_tagging */
2036 
2037 /************************************************************************
2038  * ixv_setup_vlan_support
2039  ************************************************************************/
2040 static int
2041 ixv_setup_vlan_support(struct adapter *adapter)
2042 {
2043 	struct ethercom *ec = &adapter->osdep.ec;
2044 	struct ixgbe_hw *hw = &adapter->hw;
2045 	u32		vid, vfta, retry;
2046 	struct vlanid_list *vlanidp;
2047 	int rv, error = 0;
2048 
2049 	/*
2050 	 *  This function is called from both if_init and ifflags_cb()
2051 	 * on NetBSD.
2052 	 */
2053 
2054 	/*
2055 	 * Part 1:
2056 	 * Setup VLAN HW tagging
2057 	 */
2058 	ixv_setup_vlan_tagging(adapter);
2059 
2060 	if (!VLAN_ATTACHED(ec))
2061 		return 0;
2062 
2063 	/*
2064 	 * Part 2:
2065 	 * Setup VLAN HW filter
2066 	 */
2067 	/* Cleanup shadow_vfta */
2068 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++)
2069 		adapter->shadow_vfta[i] = 0;
2070 	/* Generate shadow_vfta from ec_vids */
2071 	ETHER_LOCK(ec);
2072 	SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
2073 		uint32_t idx;
2074 
2075 		idx = vlanidp->vid / 32;
2076 		KASSERT(idx < IXGBE_VFTA_SIZE);
2077 		adapter->shadow_vfta[idx] |= (u32)1 << (vlanidp->vid % 32);
2078 	}
2079 	ETHER_UNLOCK(ec);
2080 
2081 	/*
2082 	 * A soft reset zero's out the VFTA, so
2083 	 * we need to repopulate it now.
2084 	 */
2085 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++) {
2086 		if (adapter->shadow_vfta[i] == 0)
2087 			continue;
2088 		vfta = adapter->shadow_vfta[i];
2089 		/*
2090 		 * Reconstruct the vlan id's
2091 		 * based on the bits set in each
2092 		 * of the array ints.
2093 		 */
2094 		for (int j = 0; j < 32; j++) {
2095 			retry = 0;
2096 			if ((vfta & ((u32)1 << j)) == 0)
2097 				continue;
2098 			vid = (i * 32) + j;
2099 
2100 			/* Call the shared code mailbox routine */
2101 			while ((rv = hw->mac.ops.set_vfta(hw, vid, 0, TRUE,
2102 			    FALSE)) != 0) {
2103 				if (++retry > 5) {
2104 					device_printf(adapter->dev,
2105 					    "%s: max retry exceeded\n",
2106 						__func__);
2107 					break;
2108 				}
2109 			}
2110 			if (rv != 0) {
2111 				device_printf(adapter->dev,
2112 				    "failed to set vlan %d\n", vid);
2113 				error = EACCES;
2114 			}
2115 		}
2116 	}
2117 	return error;
2118 } /* ixv_setup_vlan_support */
2119 
2120 static int
2121 ixv_vlan_cb(struct ethercom *ec, uint16_t vid, bool set)
2122 {
2123 	struct ifnet *ifp = &ec->ec_if;
2124 	struct adapter *adapter = ifp->if_softc;
2125 	int rv;
2126 
2127 	if (set)
2128 		rv = ixv_register_vlan(adapter, vid);
2129 	else
2130 		rv = ixv_unregister_vlan(adapter, vid);
2131 
2132 	if (rv != 0)
2133 		return rv;
2134 
2135 	/*
2136 	 * Control VLAN HW tagging when ec_nvlan is changed from 1 to 0
2137 	 * or 0 to 1.
2138 	 */
2139 	if ((set && (ec->ec_nvlans == 1)) || (!set && (ec->ec_nvlans == 0)))
2140 		ixv_setup_vlan_tagging(adapter);
2141 
2142 	return rv;
2143 }
2144 
2145 /************************************************************************
2146  * ixv_register_vlan
2147  *
2148  *   Run via a vlan config EVENT, it enables us to use the
2149  *   HW Filter table since we can get the vlan id. This just
2150  *   creates the entry in the soft version of the VFTA, init
2151  *   will repopulate the real table.
2152  ************************************************************************/
2153 static int
2154 ixv_register_vlan(struct adapter *adapter, u16 vtag)
2155 {
2156 	struct ixgbe_hw *hw = &adapter->hw;
2157 	u16		index, bit;
2158 	int error;
2159 
2160 	if ((vtag == 0) || (vtag > 4095)) /* Invalid */
2161 		return EINVAL;
2162 	IXGBE_CORE_LOCK(adapter);
2163 	index = (vtag >> 5) & 0x7F;
2164 	bit = vtag & 0x1F;
2165 	adapter->shadow_vfta[index] |= ((u32)1 << bit);
2166 	error = hw->mac.ops.set_vfta(hw, vtag, 0, true, false);
2167 	IXGBE_CORE_UNLOCK(adapter);
2168 
2169 	if (error != 0) {
2170 		device_printf(adapter->dev, "failed to register vlan %hu\n",
2171 		    vtag);
2172 		error = EACCES;
2173 	}
2174 	return error;
2175 } /* ixv_register_vlan */
2176 
2177 /************************************************************************
2178  * ixv_unregister_vlan
2179  *
2180  *   Run via a vlan unconfig EVENT, remove our entry
2181  *   in the soft vfta.
2182  ************************************************************************/
2183 static int
2184 ixv_unregister_vlan(struct adapter *adapter, u16 vtag)
2185 {
2186 	struct ixgbe_hw *hw = &adapter->hw;
2187 	u16		index, bit;
2188 	int		error;
2189 
2190 	if ((vtag == 0) || (vtag > 4095))  /* Invalid */
2191 		return EINVAL;
2192 
2193 	IXGBE_CORE_LOCK(adapter);
2194 	index = (vtag >> 5) & 0x7F;
2195 	bit = vtag & 0x1F;
2196 	adapter->shadow_vfta[index] &= ~((u32)1 << bit);
2197 	error = hw->mac.ops.set_vfta(hw, vtag, 0, false, false);
2198 	IXGBE_CORE_UNLOCK(adapter);
2199 
2200 	if (error != 0) {
2201 		device_printf(adapter->dev, "failed to unregister vlan %hu\n",
2202 		    vtag);
2203 		error = EIO;
2204 	}
2205 	return error;
2206 } /* ixv_unregister_vlan */
2207 
2208 /************************************************************************
2209  * ixv_enable_intr
2210  ************************************************************************/
2211 static void
2212 ixv_enable_intr(struct adapter *adapter)
2213 {
2214 	struct ixgbe_hw *hw = &adapter->hw;
2215 	struct ix_queue *que = adapter->queues;
2216 	u32		mask;
2217 	int i;
2218 
2219 	/* For VTEIAC */
2220 	mask = (1 << adapter->vector);
2221 	for (i = 0; i < adapter->num_queues; i++, que++)
2222 		mask |= (1 << que->msix);
2223 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
2224 
2225 	/* For VTEIMS */
2226 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
2227 	que = adapter->queues;
2228 	for (i = 0; i < adapter->num_queues; i++, que++)
2229 		ixv_enable_queue(adapter, que->msix);
2230 
2231 	IXGBE_WRITE_FLUSH(hw);
2232 } /* ixv_enable_intr */
2233 
2234 /************************************************************************
2235  * ixv_disable_intr
2236  ************************************************************************/
2237 static void
2238 ixv_disable_intr(struct adapter *adapter)
2239 {
2240 	struct ix_queue	*que = adapter->queues;
2241 
2242 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIAC, 0);
2243 
2244 	/* disable interrupts other than queues */
2245 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIMC, adapter->vector);
2246 
2247 	for (int i = 0; i < adapter->num_queues; i++, que++)
2248 		ixv_disable_queue(adapter, que->msix);
2249 
2250 	IXGBE_WRITE_FLUSH(&adapter->hw);
2251 } /* ixv_disable_intr */
2252 
2253 /************************************************************************
2254  * ixv_set_ivar
2255  *
2256  *   Setup the correct IVAR register for a particular MSI-X interrupt
2257  *    - entry is the register array entry
2258  *    - vector is the MSI-X vector for this queue
2259  *    - type is RX/TX/MISC
2260  ************************************************************************/
2261 static void
2262 ixv_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type)
2263 {
2264 	struct ixgbe_hw *hw = &adapter->hw;
2265 	u32		ivar, index;
2266 
2267 	vector |= IXGBE_IVAR_ALLOC_VAL;
2268 
2269 	if (type == -1) { /* MISC IVAR */
2270 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
2271 		ivar &= ~0xFF;
2272 		ivar |= vector;
2273 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
2274 	} else {	  /* RX/TX IVARS */
2275 		index = (16 * (entry & 1)) + (8 * type);
2276 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(entry >> 1));
2277 		ivar &= ~(0xffUL << index);
2278 		ivar |= ((u32)vector << index);
2279 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(entry >> 1), ivar);
2280 	}
2281 } /* ixv_set_ivar */
2282 
2283 /************************************************************************
2284  * ixv_configure_ivars
2285  ************************************************************************/
2286 static void
2287 ixv_configure_ivars(struct adapter *adapter)
2288 {
2289 	struct ix_queue *que = adapter->queues;
2290 
2291 	/* XXX We should sync EITR value calculation with ixgbe.c? */
2292 
2293 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2294 		/* First the RX queue entry */
2295 		ixv_set_ivar(adapter, i, que->msix, 0);
2296 		/* ... and the TX */
2297 		ixv_set_ivar(adapter, i, que->msix, 1);
2298 		/* Set an initial value in EITR */
2299 		ixv_eitr_write(adapter, que->msix, IXGBE_EITR_DEFAULT);
2300 	}
2301 
2302 	/* For the mailbox interrupt */
2303 	ixv_set_ivar(adapter, 1, adapter->vector, -1);
2304 } /* ixv_configure_ivars */
2305 
2306 
2307 /************************************************************************
2308  * ixv_save_stats
2309  *
2310  *   The VF stats registers never have a truly virgin
2311  *   starting point, so this routine tries to make an
2312  *   artificial one, marking ground zero on attach as
2313  *   it were.
2314  ************************************************************************/
2315 static void
2316 ixv_save_stats(struct adapter *adapter)
2317 {
2318 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2319 
2320 	if (stats->vfgprc.ev_count || stats->vfgptc.ev_count) {
2321 		stats->saved_reset_vfgprc +=
2322 		    stats->vfgprc.ev_count - stats->base_vfgprc;
2323 		stats->saved_reset_vfgptc +=
2324 		    stats->vfgptc.ev_count - stats->base_vfgptc;
2325 		stats->saved_reset_vfgorc +=
2326 		    stats->vfgorc.ev_count - stats->base_vfgorc;
2327 		stats->saved_reset_vfgotc +=
2328 		    stats->vfgotc.ev_count - stats->base_vfgotc;
2329 		stats->saved_reset_vfmprc +=
2330 		    stats->vfmprc.ev_count - stats->base_vfmprc;
2331 	}
2332 } /* ixv_save_stats */
2333 
2334 /************************************************************************
2335  * ixv_init_stats
2336  ************************************************************************/
2337 static void
2338 ixv_init_stats(struct adapter *adapter)
2339 {
2340 	struct ixgbe_hw *hw = &adapter->hw;
2341 
2342 	adapter->stats.vf.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2343 	adapter->stats.vf.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2344 	adapter->stats.vf.last_vfgorc |=
2345 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2346 
2347 	adapter->stats.vf.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2348 	adapter->stats.vf.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2349 	adapter->stats.vf.last_vfgotc |=
2350 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2351 
2352 	adapter->stats.vf.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2353 
2354 	adapter->stats.vf.base_vfgprc = adapter->stats.vf.last_vfgprc;
2355 	adapter->stats.vf.base_vfgorc = adapter->stats.vf.last_vfgorc;
2356 	adapter->stats.vf.base_vfgptc = adapter->stats.vf.last_vfgptc;
2357 	adapter->stats.vf.base_vfgotc = adapter->stats.vf.last_vfgotc;
2358 	adapter->stats.vf.base_vfmprc = adapter->stats.vf.last_vfmprc;
2359 } /* ixv_init_stats */
2360 
2361 #define UPDATE_STAT_32(reg, last, count)		\
2362 {							\
2363 	u32 current = IXGBE_READ_REG(hw, (reg));	\
2364 	if (current < (last))				\
2365 		count.ev_count += 0x100000000LL;	\
2366 	(last) = current;				\
2367 	count.ev_count &= 0xFFFFFFFF00000000LL;		\
2368 	count.ev_count |= current;			\
2369 }
2370 
2371 #define UPDATE_STAT_36(lsb, msb, last, count)		\
2372 {							\
2373 	u64 cur_lsb = IXGBE_READ_REG(hw, (lsb));	\
2374 	u64 cur_msb = IXGBE_READ_REG(hw, (msb));	\
2375 	u64 current = ((cur_msb << 32) | cur_lsb);	\
2376 	if (current < (last))				\
2377 		count.ev_count += 0x1000000000LL;	\
2378 	(last) = current;				\
2379 	count.ev_count &= 0xFFFFFFF000000000LL;		\
2380 	count.ev_count |= current;			\
2381 }
2382 
2383 /************************************************************************
2384  * ixv_update_stats - Update the board statistics counters.
2385  ************************************************************************/
2386 void
2387 ixv_update_stats(struct adapter *adapter)
2388 {
2389 	struct ixgbe_hw *hw = &adapter->hw;
2390 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2391 
2392 	UPDATE_STAT_32(IXGBE_VFGPRC, stats->last_vfgprc, stats->vfgprc);
2393 	UPDATE_STAT_32(IXGBE_VFGPTC, stats->last_vfgptc, stats->vfgptc);
2394 	UPDATE_STAT_36(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB, stats->last_vfgorc,
2395 	    stats->vfgorc);
2396 	UPDATE_STAT_36(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB, stats->last_vfgotc,
2397 	    stats->vfgotc);
2398 	UPDATE_STAT_32(IXGBE_VFMPRC, stats->last_vfmprc, stats->vfmprc);
2399 
2400 	/* VF doesn't count errors by hardware */
2401 
2402 } /* ixv_update_stats */
2403 
2404 /************************************************************************
2405  * ixv_sysctl_interrupt_rate_handler
2406  ************************************************************************/
2407 static int
2408 ixv_sysctl_interrupt_rate_handler(SYSCTLFN_ARGS)
2409 {
2410 	struct sysctlnode node = *rnode;
2411 	struct ix_queue *que = (struct ix_queue *)node.sysctl_data;
2412 	struct adapter	*adapter = que->adapter;
2413 	uint32_t reg, usec, rate;
2414 	int error;
2415 
2416 	if (que == NULL)
2417 		return 0;
2418 	reg = IXGBE_READ_REG(&que->adapter->hw, IXGBE_VTEITR(que->msix));
2419 	usec = ((reg & 0x0FF8) >> 3);
2420 	if (usec > 0)
2421 		rate = 500000 / usec;
2422 	else
2423 		rate = 0;
2424 	node.sysctl_data = &rate;
2425 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2426 	if (error || newp == NULL)
2427 		return error;
2428 	reg &= ~0xfff; /* default, no limitation */
2429 	if (rate > 0 && rate < 500000) {
2430 		if (rate < 1000)
2431 			rate = 1000;
2432 		reg |= ((4000000/rate) & 0xff8);
2433 		/*
2434 		 * When RSC is used, ITR interval must be larger than
2435 		 * RSC_DELAY. Currently, we use 2us for RSC_DELAY.
2436 		 * The minimum value is always greater than 2us on 100M
2437 		 * (and 10M?(not documented)), but it's not on 1G and higher.
2438 		 */
2439 		if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
2440 		    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
2441 			if ((adapter->num_queues > 1)
2442 			    && (reg < IXGBE_MIN_RSC_EITR_10G1G))
2443 				return EINVAL;
2444 		}
2445 		ixv_max_interrupt_rate = rate;
2446 	} else
2447 		ixv_max_interrupt_rate = 0;
2448 	ixv_eitr_write(adapter, que->msix, reg);
2449 
2450 	return (0);
2451 } /* ixv_sysctl_interrupt_rate_handler */
2452 
2453 const struct sysctlnode *
2454 ixv_sysctl_instance(struct adapter *adapter)
2455 {
2456 	const char *dvname;
2457 	struct sysctllog **log;
2458 	int rc;
2459 	const struct sysctlnode *rnode;
2460 
2461 	log = &adapter->sysctllog;
2462 	dvname = device_xname(adapter->dev);
2463 
2464 	if ((rc = sysctl_createv(log, 0, NULL, &rnode,
2465 	    0, CTLTYPE_NODE, dvname,
2466 	    SYSCTL_DESCR("ixv information and settings"),
2467 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0)
2468 		goto err;
2469 
2470 	return rnode;
2471 err:
2472 	device_printf(adapter->dev,
2473 	    "%s: sysctl_createv failed, rc = %d\n", __func__, rc);
2474 	return NULL;
2475 }
2476 
2477 static void
2478 ixv_add_device_sysctls(struct adapter *adapter)
2479 {
2480 	struct sysctllog **log;
2481 	const struct sysctlnode *rnode, *cnode;
2482 	device_t dev;
2483 
2484 	dev = adapter->dev;
2485 	log = &adapter->sysctllog;
2486 
2487 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2488 		aprint_error_dev(dev, "could not create sysctl root\n");
2489 		return;
2490 	}
2491 
2492 	if (sysctl_createv(log, 0, &rnode, &cnode,
2493 	    CTLFLAG_READWRITE, CTLTYPE_INT,
2494 	    "debug", SYSCTL_DESCR("Debug Info"),
2495 	    ixv_sysctl_debug, 0, (void *)adapter, 0, CTL_CREATE, CTL_EOL) != 0)
2496 		aprint_error_dev(dev, "could not create sysctl\n");
2497 
2498 	if (sysctl_createv(log, 0, &rnode, &cnode,
2499 	    CTLFLAG_READWRITE, CTLTYPE_BOOL,
2500 	    "enable_aim", SYSCTL_DESCR("Interrupt Moderation"),
2501 	    NULL, 0, &adapter->enable_aim, 0, CTL_CREATE, CTL_EOL) != 0)
2502 		aprint_error_dev(dev, "could not create sysctl\n");
2503 
2504 	if (sysctl_createv(log, 0, &rnode, &cnode,
2505 	    CTLFLAG_READWRITE, CTLTYPE_BOOL,
2506 	    "txrx_workqueue", SYSCTL_DESCR("Use workqueue for packet processing"),
2507 		NULL, 0, &adapter->txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL) != 0)
2508 		aprint_error_dev(dev, "could not create sysctl\n");
2509 }
2510 
2511 /************************************************************************
2512  * ixv_add_stats_sysctls - Add statistic sysctls for the VF.
2513  ************************************************************************/
2514 static void
2515 ixv_add_stats_sysctls(struct adapter *adapter)
2516 {
2517 	device_t		dev = adapter->dev;
2518 	struct tx_ring		*txr = adapter->tx_rings;
2519 	struct rx_ring		*rxr = adapter->rx_rings;
2520 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2521 	struct ixgbe_hw *hw = &adapter->hw;
2522 	const struct sysctlnode *rnode, *cnode;
2523 	struct sysctllog **log = &adapter->sysctllog;
2524 	const char *xname = device_xname(dev);
2525 
2526 	/* Driver Statistics */
2527 	evcnt_attach_dynamic(&adapter->efbig_tx_dma_setup, EVCNT_TYPE_MISC,
2528 	    NULL, xname, "Driver tx dma soft fail EFBIG");
2529 	evcnt_attach_dynamic(&adapter->mbuf_defrag_failed, EVCNT_TYPE_MISC,
2530 	    NULL, xname, "m_defrag() failed");
2531 	evcnt_attach_dynamic(&adapter->efbig2_tx_dma_setup, EVCNT_TYPE_MISC,
2532 	    NULL, xname, "Driver tx dma hard fail EFBIG");
2533 	evcnt_attach_dynamic(&adapter->einval_tx_dma_setup, EVCNT_TYPE_MISC,
2534 	    NULL, xname, "Driver tx dma hard fail EINVAL");
2535 	evcnt_attach_dynamic(&adapter->other_tx_dma_setup, EVCNT_TYPE_MISC,
2536 	    NULL, xname, "Driver tx dma hard fail other");
2537 	evcnt_attach_dynamic(&adapter->eagain_tx_dma_setup, EVCNT_TYPE_MISC,
2538 	    NULL, xname, "Driver tx dma soft fail EAGAIN");
2539 	evcnt_attach_dynamic(&adapter->enomem_tx_dma_setup, EVCNT_TYPE_MISC,
2540 	    NULL, xname, "Driver tx dma soft fail ENOMEM");
2541 	evcnt_attach_dynamic(&adapter->watchdog_events, EVCNT_TYPE_MISC,
2542 	    NULL, xname, "Watchdog timeouts");
2543 	evcnt_attach_dynamic(&adapter->tso_err, EVCNT_TYPE_MISC,
2544 	    NULL, xname, "TSO errors");
2545 	evcnt_attach_dynamic(&adapter->link_irq, EVCNT_TYPE_INTR,
2546 	    NULL, xname, "Link MSI-X IRQ Handled");
2547 
2548 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2549 		snprintf(adapter->queues[i].evnamebuf,
2550 		    sizeof(adapter->queues[i].evnamebuf), "%s q%d",
2551 		    xname, i);
2552 		snprintf(adapter->queues[i].namebuf,
2553 		    sizeof(adapter->queues[i].namebuf), "q%d", i);
2554 
2555 		if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2556 			aprint_error_dev(dev, "could not create sysctl root\n");
2557 			break;
2558 		}
2559 
2560 		if (sysctl_createv(log, 0, &rnode, &rnode,
2561 		    0, CTLTYPE_NODE,
2562 		    adapter->queues[i].namebuf, SYSCTL_DESCR("Queue Name"),
2563 		    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL) != 0)
2564 			break;
2565 
2566 		if (sysctl_createv(log, 0, &rnode, &cnode,
2567 		    CTLFLAG_READWRITE, CTLTYPE_INT,
2568 		    "interrupt_rate", SYSCTL_DESCR("Interrupt Rate"),
2569 		    ixv_sysctl_interrupt_rate_handler, 0,
2570 		    (void *)&adapter->queues[i], 0, CTL_CREATE, CTL_EOL) != 0)
2571 			break;
2572 
2573 		if (sysctl_createv(log, 0, &rnode, &cnode,
2574 		    CTLFLAG_READONLY, CTLTYPE_INT,
2575 		    "txd_head", SYSCTL_DESCR("Transmit Descriptor Head"),
2576 		    ixv_sysctl_tdh_handler, 0, (void *)txr,
2577 		    0, CTL_CREATE, CTL_EOL) != 0)
2578 			break;
2579 
2580 		if (sysctl_createv(log, 0, &rnode, &cnode,
2581 		    CTLFLAG_READONLY, CTLTYPE_INT,
2582 		    "txd_tail", SYSCTL_DESCR("Transmit Descriptor Tail"),
2583 		    ixv_sysctl_tdt_handler, 0, (void *)txr,
2584 		    0, CTL_CREATE, CTL_EOL) != 0)
2585 			break;
2586 
2587 		evcnt_attach_dynamic(&adapter->queues[i].irqs, EVCNT_TYPE_INTR,
2588 		    NULL, adapter->queues[i].evnamebuf, "IRQs on queue");
2589 		evcnt_attach_dynamic(&adapter->queues[i].handleq,
2590 		    EVCNT_TYPE_MISC, NULL, adapter->queues[i].evnamebuf,
2591 		    "Handled queue in softint");
2592 		evcnt_attach_dynamic(&adapter->queues[i].req, EVCNT_TYPE_MISC,
2593 		    NULL, adapter->queues[i].evnamebuf, "Requeued in softint");
2594 		evcnt_attach_dynamic(&txr->tso_tx, EVCNT_TYPE_MISC,
2595 		    NULL, adapter->queues[i].evnamebuf, "TSO");
2596 		evcnt_attach_dynamic(&txr->no_desc_avail, EVCNT_TYPE_MISC,
2597 		    NULL, adapter->queues[i].evnamebuf,
2598 		    "Queue No Descriptor Available");
2599 		evcnt_attach_dynamic(&txr->total_packets, EVCNT_TYPE_MISC,
2600 		    NULL, adapter->queues[i].evnamebuf,
2601 		    "Queue Packets Transmitted");
2602 #ifndef IXGBE_LEGACY_TX
2603 		evcnt_attach_dynamic(&txr->pcq_drops, EVCNT_TYPE_MISC,
2604 		    NULL, adapter->queues[i].evnamebuf,
2605 		    "Packets dropped in pcq");
2606 #endif
2607 
2608 #ifdef LRO
2609 		struct lro_ctrl *lro = &rxr->lro;
2610 #endif /* LRO */
2611 
2612 		if (sysctl_createv(log, 0, &rnode, &cnode,
2613 		    CTLFLAG_READONLY,
2614 		    CTLTYPE_INT,
2615 		    "rxd_nxck", SYSCTL_DESCR("Receive Descriptor next to check"),
2616 			ixv_sysctl_next_to_check_handler, 0, (void *)rxr, 0,
2617 		    CTL_CREATE, CTL_EOL) != 0)
2618 			break;
2619 
2620 		if (sysctl_createv(log, 0, &rnode, &cnode,
2621 		    CTLFLAG_READONLY,
2622 		    CTLTYPE_INT,
2623 		    "rxd_head", SYSCTL_DESCR("Receive Descriptor Head"),
2624 		    ixv_sysctl_rdh_handler, 0, (void *)rxr, 0,
2625 		    CTL_CREATE, CTL_EOL) != 0)
2626 			break;
2627 
2628 		if (sysctl_createv(log, 0, &rnode, &cnode,
2629 		    CTLFLAG_READONLY,
2630 		    CTLTYPE_INT,
2631 		    "rxd_tail", SYSCTL_DESCR("Receive Descriptor Tail"),
2632 		    ixv_sysctl_rdt_handler, 0, (void *)rxr, 0,
2633 		    CTL_CREATE, CTL_EOL) != 0)
2634 			break;
2635 
2636 		evcnt_attach_dynamic(&rxr->rx_packets, EVCNT_TYPE_MISC,
2637 		    NULL, adapter->queues[i].evnamebuf, "Queue Packets Received");
2638 		evcnt_attach_dynamic(&rxr->rx_bytes, EVCNT_TYPE_MISC,
2639 		    NULL, adapter->queues[i].evnamebuf, "Queue Bytes Received");
2640 		evcnt_attach_dynamic(&rxr->rx_copies, EVCNT_TYPE_MISC,
2641 		    NULL, adapter->queues[i].evnamebuf, "Copied RX Frames");
2642 		evcnt_attach_dynamic(&rxr->no_jmbuf, EVCNT_TYPE_MISC,
2643 		    NULL, adapter->queues[i].evnamebuf, "Rx no jumbo mbuf");
2644 		evcnt_attach_dynamic(&rxr->rx_discarded, EVCNT_TYPE_MISC,
2645 		    NULL, adapter->queues[i].evnamebuf, "Rx discarded");
2646 #ifdef LRO
2647 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_queued",
2648 				CTLFLAG_RD, &lro->lro_queued, 0,
2649 				"LRO Queued");
2650 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_flushed",
2651 				CTLFLAG_RD, &lro->lro_flushed, 0,
2652 				"LRO Flushed");
2653 #endif /* LRO */
2654 	}
2655 
2656 	/* MAC stats get their own sub node */
2657 
2658 	snprintf(stats->namebuf,
2659 	    sizeof(stats->namebuf), "%s MAC Statistics", xname);
2660 
2661 	evcnt_attach_dynamic(&stats->ipcs, EVCNT_TYPE_MISC, NULL,
2662 	    stats->namebuf, "rx csum offload - IP");
2663 	evcnt_attach_dynamic(&stats->l4cs, EVCNT_TYPE_MISC, NULL,
2664 	    stats->namebuf, "rx csum offload - L4");
2665 	evcnt_attach_dynamic(&stats->ipcs_bad, EVCNT_TYPE_MISC, NULL,
2666 	    stats->namebuf, "rx csum offload - IP bad");
2667 	evcnt_attach_dynamic(&stats->l4cs_bad, EVCNT_TYPE_MISC, NULL,
2668 	    stats->namebuf, "rx csum offload - L4 bad");
2669 
2670 	/* Packet Reception Stats */
2671 	evcnt_attach_dynamic(&stats->vfgprc, EVCNT_TYPE_MISC, NULL,
2672 	    xname, "Good Packets Received");
2673 	evcnt_attach_dynamic(&stats->vfgorc, EVCNT_TYPE_MISC, NULL,
2674 	    xname, "Good Octets Received");
2675 	evcnt_attach_dynamic(&stats->vfmprc, EVCNT_TYPE_MISC, NULL,
2676 	    xname, "Multicast Packets Received");
2677 	evcnt_attach_dynamic(&stats->vfgptc, EVCNT_TYPE_MISC, NULL,
2678 	    xname, "Good Packets Transmitted");
2679 	evcnt_attach_dynamic(&stats->vfgotc, EVCNT_TYPE_MISC, NULL,
2680 	    xname, "Good Octets Transmitted");
2681 
2682 	/* Mailbox Stats */
2683 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_tx, EVCNT_TYPE_MISC, NULL,
2684 	    xname, "message TXs");
2685 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_rx, EVCNT_TYPE_MISC, NULL,
2686 	    xname, "message RXs");
2687 	evcnt_attach_dynamic(&hw->mbx.stats.acks, EVCNT_TYPE_MISC, NULL,
2688 	    xname, "ACKs");
2689 	evcnt_attach_dynamic(&hw->mbx.stats.reqs, EVCNT_TYPE_MISC, NULL,
2690 	    xname, "REQs");
2691 	evcnt_attach_dynamic(&hw->mbx.stats.rsts, EVCNT_TYPE_MISC, NULL,
2692 	    xname, "RSTs");
2693 
2694 } /* ixv_add_stats_sysctls */
2695 
2696 static void
2697 ixv_clear_evcnt(struct adapter *adapter)
2698 {
2699 	struct tx_ring		*txr = adapter->tx_rings;
2700 	struct rx_ring		*rxr = adapter->rx_rings;
2701 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2702 	struct ixgbe_hw *hw = &adapter->hw;
2703 	int i;
2704 
2705 	/* Driver Statistics */
2706 	adapter->efbig_tx_dma_setup.ev_count = 0;
2707 	adapter->mbuf_defrag_failed.ev_count = 0;
2708 	adapter->efbig2_tx_dma_setup.ev_count = 0;
2709 	adapter->einval_tx_dma_setup.ev_count = 0;
2710 	adapter->other_tx_dma_setup.ev_count = 0;
2711 	adapter->eagain_tx_dma_setup.ev_count = 0;
2712 	adapter->enomem_tx_dma_setup.ev_count = 0;
2713 	adapter->watchdog_events.ev_count = 0;
2714 	adapter->tso_err.ev_count = 0;
2715 	adapter->link_irq.ev_count = 0;
2716 
2717 	for (i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2718 		adapter->queues[i].irqs.ev_count = 0;
2719 		adapter->queues[i].handleq.ev_count = 0;
2720 		adapter->queues[i].req.ev_count = 0;
2721 		txr->tso_tx.ev_count = 0;
2722 		txr->no_desc_avail.ev_count = 0;
2723 		txr->total_packets.ev_count = 0;
2724 #ifndef IXGBE_LEGACY_TX
2725 		txr->pcq_drops.ev_count = 0;
2726 #endif
2727 		txr->q_efbig_tx_dma_setup = 0;
2728 		txr->q_mbuf_defrag_failed = 0;
2729 		txr->q_efbig2_tx_dma_setup = 0;
2730 		txr->q_einval_tx_dma_setup = 0;
2731 		txr->q_other_tx_dma_setup = 0;
2732 		txr->q_eagain_tx_dma_setup = 0;
2733 		txr->q_enomem_tx_dma_setup = 0;
2734 		txr->q_tso_err = 0;
2735 
2736 		rxr->rx_packets.ev_count = 0;
2737 		rxr->rx_bytes.ev_count = 0;
2738 		rxr->rx_copies.ev_count = 0;
2739 		rxr->no_jmbuf.ev_count = 0;
2740 		rxr->rx_discarded.ev_count = 0;
2741 	}
2742 
2743 	/* MAC stats get their own sub node */
2744 
2745 	stats->ipcs.ev_count = 0;
2746 	stats->l4cs.ev_count = 0;
2747 	stats->ipcs_bad.ev_count = 0;
2748 	stats->l4cs_bad.ev_count = 0;
2749 
2750 	/* Packet Reception Stats */
2751 	stats->vfgprc.ev_count = 0;
2752 	stats->vfgorc.ev_count = 0;
2753 	stats->vfmprc.ev_count = 0;
2754 	stats->vfgptc.ev_count = 0;
2755 	stats->vfgotc.ev_count = 0;
2756 
2757 	/* Mailbox Stats */
2758 	hw->mbx.stats.msgs_tx.ev_count = 0;
2759 	hw->mbx.stats.msgs_rx.ev_count = 0;
2760 	hw->mbx.stats.acks.ev_count = 0;
2761 	hw->mbx.stats.reqs.ev_count = 0;
2762 	hw->mbx.stats.rsts.ev_count = 0;
2763 
2764 } /* ixv_clear_evcnt */
2765 
2766 /************************************************************************
2767  * ixv_set_sysctl_value
2768  ************************************************************************/
2769 static void
2770 ixv_set_sysctl_value(struct adapter *adapter, const char *name,
2771 	const char *description, int *limit, int value)
2772 {
2773 	device_t dev =	adapter->dev;
2774 	struct sysctllog **log;
2775 	const struct sysctlnode *rnode, *cnode;
2776 
2777 	log = &adapter->sysctllog;
2778 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2779 		aprint_error_dev(dev, "could not create sysctl root\n");
2780 		return;
2781 	}
2782 	if (sysctl_createv(log, 0, &rnode, &cnode,
2783 	    CTLFLAG_READWRITE, CTLTYPE_INT,
2784 	    name, SYSCTL_DESCR(description),
2785 	    NULL, 0, limit, 0, CTL_CREATE, CTL_EOL) != 0)
2786 		aprint_error_dev(dev, "could not create sysctl\n");
2787 	*limit = value;
2788 } /* ixv_set_sysctl_value */
2789 
2790 /************************************************************************
2791  * ixv_print_debug_info
2792  *
2793  *   Called only when em_display_debug_stats is enabled.
2794  *   Provides a way to take a look at important statistics
2795  *   maintained by the driver and hardware.
2796  ************************************************************************/
2797 static void
2798 ixv_print_debug_info(struct adapter *adapter)
2799 {
2800 	device_t	dev = adapter->dev;
2801 	struct ix_queue *que = adapter->queues;
2802 	struct rx_ring	*rxr;
2803 	struct tx_ring	*txr;
2804 #ifdef LRO
2805 	struct lro_ctrl *lro;
2806 #endif /* LRO */
2807 
2808 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2809 		txr = que->txr;
2810 		rxr = que->rxr;
2811 #ifdef LRO
2812 		lro = &rxr->lro;
2813 #endif /* LRO */
2814 		device_printf(dev, "QUE(%d) IRQs Handled: %lu\n",
2815 		    que->msix, (long)que->irqs.ev_count);
2816 		device_printf(dev, "RX(%d) Packets Received: %lld\n",
2817 		    rxr->me, (long long)rxr->rx_packets.ev_count);
2818 		device_printf(dev, "RX(%d) Bytes Received: %lu\n",
2819 		    rxr->me, (long)rxr->rx_bytes.ev_count);
2820 #ifdef LRO
2821 		device_printf(dev, "RX(%d) LRO Queued= %ju\n",
2822 		    rxr->me, (uintmax_t)lro->lro_queued);
2823 		device_printf(dev, "RX(%d) LRO Flushed= %ju\n",
2824 		    rxr->me, (uintmax_t)lro->lro_flushed);
2825 #endif /* LRO */
2826 		device_printf(dev, "TX(%d) Packets Sent: %lu\n",
2827 		    txr->me, (long)txr->total_packets.ev_count);
2828 		device_printf(dev, "TX(%d) NO Desc Avail: %lu\n",
2829 		    txr->me, (long)txr->no_desc_avail.ev_count);
2830 	}
2831 
2832 	device_printf(dev, "MBX IRQ Handled: %lu\n",
2833 	    (long)adapter->link_irq.ev_count);
2834 } /* ixv_print_debug_info */
2835 
2836 /************************************************************************
2837  * ixv_sysctl_debug
2838  ************************************************************************/
2839 static int
2840 ixv_sysctl_debug(SYSCTLFN_ARGS)
2841 {
2842 	struct sysctlnode node = *rnode;
2843 	struct adapter *adapter = (struct adapter *)node.sysctl_data;
2844 	int	       error, result;
2845 
2846 	node.sysctl_data = &result;
2847 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2848 
2849 	if (error || newp == NULL)
2850 		return error;
2851 
2852 	if (result == 1)
2853 		ixv_print_debug_info(adapter);
2854 
2855 	return 0;
2856 } /* ixv_sysctl_debug */
2857 
2858 /************************************************************************
2859  * ixv_init_device_features
2860  ************************************************************************/
2861 static void
2862 ixv_init_device_features(struct adapter *adapter)
2863 {
2864 	adapter->feat_cap = IXGBE_FEATURE_NETMAP
2865 			  | IXGBE_FEATURE_VF
2866 			  | IXGBE_FEATURE_RSS
2867 			  | IXGBE_FEATURE_LEGACY_TX;
2868 
2869 	/* A tad short on feature flags for VFs, atm. */
2870 	switch (adapter->hw.mac.type) {
2871 	case ixgbe_mac_82599_vf:
2872 		break;
2873 	case ixgbe_mac_X540_vf:
2874 		break;
2875 	case ixgbe_mac_X550_vf:
2876 	case ixgbe_mac_X550EM_x_vf:
2877 	case ixgbe_mac_X550EM_a_vf:
2878 		adapter->feat_cap |= IXGBE_FEATURE_NEEDS_CTXD;
2879 		break;
2880 	default:
2881 		break;
2882 	}
2883 
2884 	/* Enabled by default... */
2885 	/* Is a virtual function (VF) */
2886 	if (adapter->feat_cap & IXGBE_FEATURE_VF)
2887 		adapter->feat_en |= IXGBE_FEATURE_VF;
2888 	/* Netmap */
2889 	if (adapter->feat_cap & IXGBE_FEATURE_NETMAP)
2890 		adapter->feat_en |= IXGBE_FEATURE_NETMAP;
2891 	/* Receive-Side Scaling (RSS) */
2892 	if (adapter->feat_cap & IXGBE_FEATURE_RSS)
2893 		adapter->feat_en |= IXGBE_FEATURE_RSS;
2894 	/* Needs advanced context descriptor regardless of offloads req'd */
2895 	if (adapter->feat_cap & IXGBE_FEATURE_NEEDS_CTXD)
2896 		adapter->feat_en |= IXGBE_FEATURE_NEEDS_CTXD;
2897 
2898 	/* Enabled via sysctl... */
2899 	/* Legacy (single queue) transmit */
2900 	if ((adapter->feat_cap & IXGBE_FEATURE_LEGACY_TX) &&
2901 	    ixv_enable_legacy_tx)
2902 		adapter->feat_en |= IXGBE_FEATURE_LEGACY_TX;
2903 } /* ixv_init_device_features */
2904 
2905 /************************************************************************
2906  * ixv_shutdown - Shutdown entry point
2907  ************************************************************************/
2908 #if 0 /* XXX NetBSD ought to register something like this through pmf(9) */
2909 static int
2910 ixv_shutdown(device_t dev)
2911 {
2912 	struct adapter *adapter = device_private(dev);
2913 	IXGBE_CORE_LOCK(adapter);
2914 	ixv_stop(adapter);
2915 	IXGBE_CORE_UNLOCK(adapter);
2916 
2917 	return (0);
2918 } /* ixv_shutdown */
2919 #endif
2920 
2921 static int
2922 ixv_ifflags_cb(struct ethercom *ec)
2923 {
2924 	struct ifnet *ifp = &ec->ec_if;
2925 	struct adapter *adapter = ifp->if_softc;
2926 	u_short saved_flags;
2927 	u_short change;
2928 	int rv = 0;
2929 
2930 	IXGBE_CORE_LOCK(adapter);
2931 
2932 	saved_flags = adapter->if_flags;
2933 	change = ifp->if_flags ^ adapter->if_flags;
2934 	if (change != 0)
2935 		adapter->if_flags = ifp->if_flags;
2936 
2937 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
2938 		rv = ENETRESET;
2939 		goto out;
2940 	} else if ((change & IFF_PROMISC) != 0) {
2941 		rv = ixv_set_rxfilter(adapter);
2942 		if (rv != 0) {
2943 			/* Restore previous */
2944 			adapter->if_flags = saved_flags;
2945 			goto out;
2946 		}
2947 	}
2948 
2949 	/* Check for ec_capenable. */
2950 	change = ec->ec_capenable ^ adapter->ec_capenable;
2951 	adapter->ec_capenable = ec->ec_capenable;
2952 	if ((change & ~(ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING
2953 	    | ETHERCAP_VLAN_HWFILTER)) != 0) {
2954 		rv = ENETRESET;
2955 		goto out;
2956 	}
2957 
2958 	/*
2959 	 * Special handling is not required for ETHERCAP_VLAN_MTU.
2960 	 * PF's MAXFRS(MHADD) does not include the 4bytes of the VLAN header.
2961 	 */
2962 
2963 	/* Set up VLAN support and filter */
2964 	if ((change & (ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_HWFILTER)) != 0)
2965 		rv = ixv_setup_vlan_support(adapter);
2966 
2967 out:
2968 	IXGBE_CORE_UNLOCK(adapter);
2969 
2970 	return rv;
2971 }
2972 
2973 
2974 /************************************************************************
2975  * ixv_ioctl - Ioctl entry point
2976  *
2977  *   Called when the user wants to configure the interface.
2978  *
2979  *   return 0 on success, positive on failure
2980  ************************************************************************/
2981 static int
2982 ixv_ioctl(struct ifnet *ifp, u_long command, void *data)
2983 {
2984 	struct adapter	*adapter = ifp->if_softc;
2985 	struct ixgbe_hw *hw = &adapter->hw;
2986 	struct ifcapreq *ifcr = data;
2987 	int		error;
2988 	int l4csum_en;
2989 	const int l4csum = IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx |
2990 	     IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_UDPv6_Rx;
2991 
2992 	switch (command) {
2993 	case SIOCSIFFLAGS:
2994 		IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)");
2995 		break;
2996 	case SIOCADDMULTI: {
2997 		struct ether_multi *enm;
2998 		struct ether_multistep step;
2999 		struct ethercom *ec = &adapter->osdep.ec;
3000 		bool overflow = false;
3001 		int mcnt = 0;
3002 
3003 		/*
3004 		 * Check the number of multicast address. If it exceeds,
3005 		 * return ENOSPC.
3006 		 * Update this code when we support API 1.3.
3007 		 */
3008 		ETHER_LOCK(ec);
3009 		ETHER_FIRST_MULTI(step, ec, enm);
3010 		while (enm != NULL) {
3011 			mcnt++;
3012 
3013 			/*
3014 			 * This code is before adding, so one room is required
3015 			 * at least.
3016 			 */
3017 			if (mcnt > (IXGBE_MAX_VF_MC - 1)) {
3018 				overflow = true;
3019 				break;
3020 			}
3021 			ETHER_NEXT_MULTI(step, enm);
3022 		}
3023 		ETHER_UNLOCK(ec);
3024 		error = 0;
3025 		if (overflow && ((ec->ec_flags & ETHER_F_ALLMULTI) == 0)) {
3026 			error = hw->mac.ops.update_xcast_mode(hw,
3027 			    IXGBEVF_XCAST_MODE_ALLMULTI);
3028 			if (error == IXGBE_ERR_NOT_TRUSTED) {
3029 				device_printf(adapter->dev,
3030 				    "this interface is not trusted\n");
3031 				error = EPERM;
3032 			} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
3033 				device_printf(adapter->dev,
3034 				    "the PF doesn't support allmulti mode\n");
3035 				error = EOPNOTSUPP;
3036 			} else if (error) {
3037 				device_printf(adapter->dev,
3038 				    "number of Ethernet multicast addresses "
3039 				    "exceeds the limit (%d). error = %d\n",
3040 				    IXGBE_MAX_VF_MC, error);
3041 				error = ENOSPC;
3042 			} else
3043 				ec->ec_flags |= ETHER_F_ALLMULTI;
3044 		}
3045 		if (error)
3046 			return error;
3047 	}
3048 		/*FALLTHROUGH*/
3049 	case SIOCDELMULTI:
3050 		IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI");
3051 		break;
3052 	case SIOCSIFMEDIA:
3053 	case SIOCGIFMEDIA:
3054 		IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)");
3055 		break;
3056 	case SIOCSIFCAP:
3057 		IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)");
3058 		break;
3059 	case SIOCSIFMTU:
3060 		IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)");
3061 		break;
3062 	case SIOCZIFDATA:
3063 		IOCTL_DEBUGOUT("ioctl: SIOCZIFDATA (Zero counter)");
3064 		ixv_update_stats(adapter);
3065 		ixv_clear_evcnt(adapter);
3066 		break;
3067 	default:
3068 		IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)", (int)command);
3069 		break;
3070 	}
3071 
3072 	switch (command) {
3073 	case SIOCSIFCAP:
3074 		/* Layer-4 Rx checksum offload has to be turned on and
3075 		 * off as a unit.
3076 		 */
3077 		l4csum_en = ifcr->ifcr_capenable & l4csum;
3078 		if (l4csum_en != l4csum && l4csum_en != 0)
3079 			return EINVAL;
3080 		/*FALLTHROUGH*/
3081 	case SIOCADDMULTI:
3082 	case SIOCDELMULTI:
3083 	case SIOCSIFFLAGS:
3084 	case SIOCSIFMTU:
3085 	default:
3086 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
3087 			return error;
3088 		if ((ifp->if_flags & IFF_RUNNING) == 0)
3089 			;
3090 		else if (command == SIOCSIFCAP || command == SIOCSIFMTU) {
3091 			IXGBE_CORE_LOCK(adapter);
3092 			ixv_init_locked(adapter);
3093 			IXGBE_CORE_UNLOCK(adapter);
3094 		} else if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
3095 			/*
3096 			 * Multicast list has changed; set the hardware filter
3097 			 * accordingly.
3098 			 */
3099 			IXGBE_CORE_LOCK(adapter);
3100 			ixv_disable_intr(adapter);
3101 			ixv_set_rxfilter(adapter);
3102 			ixv_enable_intr(adapter);
3103 			IXGBE_CORE_UNLOCK(adapter);
3104 		}
3105 		return 0;
3106 	}
3107 } /* ixv_ioctl */
3108 
3109 /************************************************************************
3110  * ixv_init
3111  ************************************************************************/
3112 static int
3113 ixv_init(struct ifnet *ifp)
3114 {
3115 	struct adapter *adapter = ifp->if_softc;
3116 
3117 	IXGBE_CORE_LOCK(adapter);
3118 	ixv_init_locked(adapter);
3119 	IXGBE_CORE_UNLOCK(adapter);
3120 
3121 	return 0;
3122 } /* ixv_init */
3123 
3124 /************************************************************************
3125  * ixv_handle_que
3126  ************************************************************************/
3127 static void
3128 ixv_handle_que(void *context)
3129 {
3130 	struct ix_queue *que = context;
3131 	struct adapter	*adapter = que->adapter;
3132 	struct tx_ring	*txr = que->txr;
3133 	struct ifnet	*ifp = adapter->ifp;
3134 	bool		more;
3135 
3136 	que->handleq.ev_count++;
3137 
3138 	if (ifp->if_flags & IFF_RUNNING) {
3139 		more = ixgbe_rxeof(que);
3140 		IXGBE_TX_LOCK(txr);
3141 		more |= ixgbe_txeof(txr);
3142 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
3143 			if (!ixgbe_mq_ring_empty(ifp, txr->txr_interq))
3144 				ixgbe_mq_start_locked(ifp, txr);
3145 		/* Only for queue 0 */
3146 		/* NetBSD still needs this for CBQ */
3147 		if ((&adapter->queues[0] == que)
3148 		    && (!ixgbe_legacy_ring_empty(ifp, NULL)))
3149 			ixgbe_legacy_start_locked(ifp, txr);
3150 		IXGBE_TX_UNLOCK(txr);
3151 		if (more) {
3152 			que->req.ev_count++;
3153 			if (adapter->txrx_use_workqueue) {
3154 				/*
3155 				 * "enqueued flag" is not required here
3156 				 * the same as ixg(4). See ixgbe_msix_que().
3157 				 */
3158 				workqueue_enqueue(adapter->que_wq,
3159 				    &que->wq_cookie, curcpu());
3160 			} else
3161 				  softint_schedule(que->que_si);
3162 			return;
3163 		}
3164 	}
3165 
3166 	/* Re-enable this interrupt */
3167 	ixv_enable_queue(adapter, que->msix);
3168 
3169 	return;
3170 } /* ixv_handle_que */
3171 
3172 /************************************************************************
3173  * ixv_handle_que_work
3174  ************************************************************************/
3175 static void
3176 ixv_handle_que_work(struct work *wk, void *context)
3177 {
3178 	struct ix_queue *que = container_of(wk, struct ix_queue, wq_cookie);
3179 
3180 	/*
3181 	 * "enqueued flag" is not required here the same as ixg(4).
3182 	 * See ixgbe_msix_que().
3183 	 */
3184 	ixv_handle_que(que);
3185 }
3186 
3187 /************************************************************************
3188  * ixv_allocate_msix - Setup MSI-X Interrupt resources and handlers
3189  ************************************************************************/
3190 static int
3191 ixv_allocate_msix(struct adapter *adapter, const struct pci_attach_args *pa)
3192 {
3193 	device_t	dev = adapter->dev;
3194 	struct ix_queue *que = adapter->queues;
3195 	struct		tx_ring *txr = adapter->tx_rings;
3196 	int		error, msix_ctrl, rid, vector = 0;
3197 	pci_chipset_tag_t pc;
3198 	pcitag_t	tag;
3199 	char		intrbuf[PCI_INTRSTR_LEN];
3200 	char		wqname[MAXCOMLEN];
3201 	char		intr_xname[32];
3202 	const char	*intrstr = NULL;
3203 	kcpuset_t	*affinity;
3204 	int		cpu_id = 0;
3205 
3206 	pc = adapter->osdep.pc;
3207 	tag = adapter->osdep.tag;
3208 
3209 	adapter->osdep.nintrs = adapter->num_queues + 1;
3210 	if (pci_msix_alloc_exact(pa, &adapter->osdep.intrs,
3211 	    adapter->osdep.nintrs) != 0) {
3212 		aprint_error_dev(dev,
3213 		    "failed to allocate MSI-X interrupt\n");
3214 		return (ENXIO);
3215 	}
3216 
3217 	kcpuset_create(&affinity, false);
3218 	for (int i = 0; i < adapter->num_queues; i++, vector++, que++, txr++) {
3219 		snprintf(intr_xname, sizeof(intr_xname), "%s TXRX%d",
3220 		    device_xname(dev), i);
3221 		intrstr = pci_intr_string(pc, adapter->osdep.intrs[i], intrbuf,
3222 		    sizeof(intrbuf));
3223 #ifdef IXGBE_MPSAFE
3224 		pci_intr_setattr(pc, &adapter->osdep.intrs[i], PCI_INTR_MPSAFE,
3225 		    true);
3226 #endif
3227 		/* Set the handler function */
3228 		que->res = adapter->osdep.ihs[i] = pci_intr_establish_xname(pc,
3229 		    adapter->osdep.intrs[i], IPL_NET, ixv_msix_que, que,
3230 		    intr_xname);
3231 		if (que->res == NULL) {
3232 			pci_intr_release(pc, adapter->osdep.intrs,
3233 			    adapter->osdep.nintrs);
3234 			aprint_error_dev(dev,
3235 			    "Failed to register QUE handler\n");
3236 			kcpuset_destroy(affinity);
3237 			return (ENXIO);
3238 		}
3239 		que->msix = vector;
3240 		adapter->active_queues |= (u64)(1 << que->msix);
3241 
3242 		cpu_id = i;
3243 		/* Round-robin affinity */
3244 		kcpuset_zero(affinity);
3245 		kcpuset_set(affinity, cpu_id % ncpu);
3246 		error = interrupt_distribute(adapter->osdep.ihs[i], affinity,
3247 		    NULL);
3248 		aprint_normal_dev(dev, "for TX/RX, interrupting at %s",
3249 		    intrstr);
3250 		if (error == 0)
3251 			aprint_normal(", bound queue %d to cpu %d\n",
3252 			    i, cpu_id % ncpu);
3253 		else
3254 			aprint_normal("\n");
3255 
3256 #ifndef IXGBE_LEGACY_TX
3257 		txr->txr_si
3258 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINFT_FLAGS,
3259 			ixgbe_deferred_mq_start, txr);
3260 #endif
3261 		que->que_si
3262 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINFT_FLAGS,
3263 			ixv_handle_que, que);
3264 		if (que->que_si == NULL) {
3265 			aprint_error_dev(dev,
3266 			    "could not establish software interrupt\n");
3267 		}
3268 	}
3269 	snprintf(wqname, sizeof(wqname), "%sdeferTx", device_xname(dev));
3270 	error = workqueue_create(&adapter->txr_wq, wqname,
3271 	    ixgbe_deferred_mq_start_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3272 	    IXGBE_WORKQUEUE_FLAGS);
3273 	if (error) {
3274 		aprint_error_dev(dev, "couldn't create workqueue for deferred Tx\n");
3275 	}
3276 	adapter->txr_wq_enqueued = percpu_alloc(sizeof(u_int));
3277 
3278 	snprintf(wqname, sizeof(wqname), "%sTxRx", device_xname(dev));
3279 	error = workqueue_create(&adapter->que_wq, wqname,
3280 	    ixv_handle_que_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3281 	    IXGBE_WORKQUEUE_FLAGS);
3282 	if (error) {
3283 		aprint_error_dev(dev,
3284 		    "couldn't create workqueue\n");
3285 	}
3286 
3287 	/* and Mailbox */
3288 	cpu_id++;
3289 	snprintf(intr_xname, sizeof(intr_xname), "%s link", device_xname(dev));
3290 	adapter->vector = vector;
3291 	intrstr = pci_intr_string(pc, adapter->osdep.intrs[vector], intrbuf,
3292 	    sizeof(intrbuf));
3293 #ifdef IXGBE_MPSAFE
3294 	pci_intr_setattr(pc, &adapter->osdep.intrs[vector], PCI_INTR_MPSAFE,
3295 	    true);
3296 #endif
3297 	/* Set the mbx handler function */
3298 	adapter->osdep.ihs[vector] = pci_intr_establish_xname(pc,
3299 	    adapter->osdep.intrs[vector], IPL_NET, ixv_msix_mbx, adapter,
3300 	    intr_xname);
3301 	if (adapter->osdep.ihs[vector] == NULL) {
3302 		aprint_error_dev(dev, "Failed to register LINK handler\n");
3303 		kcpuset_destroy(affinity);
3304 		return (ENXIO);
3305 	}
3306 	/* Round-robin affinity */
3307 	kcpuset_zero(affinity);
3308 	kcpuset_set(affinity, cpu_id % ncpu);
3309 	error = interrupt_distribute(adapter->osdep.ihs[vector],
3310 	    affinity, NULL);
3311 
3312 	aprint_normal_dev(dev,
3313 	    "for link, interrupting at %s", intrstr);
3314 	if (error == 0)
3315 		aprint_normal(", affinity to cpu %d\n", cpu_id % ncpu);
3316 	else
3317 		aprint_normal("\n");
3318 
3319 	/* Tasklets for Mailbox */
3320 	adapter->link_si = softint_establish(SOFTINT_NET |IXGBE_SOFTINFT_FLAGS,
3321 	    ixv_handle_link, adapter);
3322 	/*
3323 	 * Due to a broken design QEMU will fail to properly
3324 	 * enable the guest for MSI-X unless the vectors in
3325 	 * the table are all set up, so we must rewrite the
3326 	 * ENABLE in the MSI-X control register again at this
3327 	 * point to cause it to successfully initialize us.
3328 	 */
3329 	if (adapter->hw.mac.type == ixgbe_mac_82599_vf) {
3330 		pci_get_capability(pc, tag, PCI_CAP_MSIX, &rid, NULL);
3331 		rid += PCI_MSIX_CTL;
3332 		msix_ctrl = pci_conf_read(pc, tag, rid);
3333 		msix_ctrl |= PCI_MSIX_CTL_ENABLE;
3334 		pci_conf_write(pc, tag, rid, msix_ctrl);
3335 	}
3336 
3337 	kcpuset_destroy(affinity);
3338 	return (0);
3339 } /* ixv_allocate_msix */
3340 
3341 /************************************************************************
3342  * ixv_configure_interrupts - Setup MSI-X resources
3343  *
3344  *   Note: The VF device MUST use MSI-X, there is no fallback.
3345  ************************************************************************/
3346 static int
3347 ixv_configure_interrupts(struct adapter *adapter)
3348 {
3349 	device_t dev = adapter->dev;
3350 	int want, queues, msgs;
3351 
3352 	/* Must have at least 2 MSI-X vectors */
3353 	msgs = pci_msix_count(adapter->osdep.pc, adapter->osdep.tag);
3354 	if (msgs < 2) {
3355 		aprint_error_dev(dev, "MSIX config error\n");
3356 		return (ENXIO);
3357 	}
3358 	msgs = MIN(msgs, IXG_MAX_NINTR);
3359 
3360 	/* Figure out a reasonable auto config value */
3361 	queues = (ncpu > (msgs - 1)) ? (msgs - 1) : ncpu;
3362 
3363 	if (ixv_num_queues != 0)
3364 		queues = ixv_num_queues;
3365 	else if ((ixv_num_queues == 0) && (queues > IXGBE_VF_MAX_TX_QUEUES))
3366 		queues = IXGBE_VF_MAX_TX_QUEUES;
3367 
3368 	/*
3369 	 * Want vectors for the queues,
3370 	 * plus an additional for mailbox.
3371 	 */
3372 	want = queues + 1;
3373 	if (msgs >= want)
3374 		msgs = want;
3375 	else {
3376 		aprint_error_dev(dev,
3377 		    "MSI-X Configuration Problem, "
3378 		    "%d vectors but %d queues wanted!\n",
3379 		    msgs, want);
3380 		return -1;
3381 	}
3382 
3383 	adapter->msix_mem = (void *)1; /* XXX */
3384 	aprint_normal_dev(dev,
3385 	    "Using MSI-X interrupts with %d vectors\n", msgs);
3386 	adapter->num_queues = queues;
3387 
3388 	return (0);
3389 } /* ixv_configure_interrupts */
3390 
3391 
3392 /************************************************************************
3393  * ixv_handle_link - Tasklet handler for MSI-X MBX interrupts
3394  *
3395  *   Done outside of interrupt context since the driver might sleep
3396  ************************************************************************/
3397 static void
3398 ixv_handle_link(void *context)
3399 {
3400 	struct adapter *adapter = context;
3401 
3402 	IXGBE_CORE_LOCK(adapter);
3403 
3404 	adapter->hw.mac.ops.check_link(&adapter->hw, &adapter->link_speed,
3405 	    &adapter->link_up, FALSE);
3406 	ixv_update_link_status(adapter);
3407 
3408 	IXGBE_CORE_UNLOCK(adapter);
3409 } /* ixv_handle_link */
3410 
3411 /************************************************************************
3412  * ixv_check_link - Used in the local timer to poll for link changes
3413  ************************************************************************/
3414 static s32
3415 ixv_check_link(struct adapter *adapter)
3416 {
3417 	s32 error;
3418 
3419 	KASSERT(mutex_owned(&adapter->core_mtx));
3420 
3421 	adapter->hw.mac.get_link_status = TRUE;
3422 
3423 	error = adapter->hw.mac.ops.check_link(&adapter->hw,
3424 	    &adapter->link_speed, &adapter->link_up, FALSE);
3425 	ixv_update_link_status(adapter);
3426 
3427 	return error;
3428 } /* ixv_check_link */
3429