xref: /netbsd-src/sys/dev/pci/ixgbe/ixv.c (revision 9fb66d812c00ebfb445c0b47dea128f32aa6fe96)
1 /*$NetBSD: ixv.c,v 1.158 2021/03/31 07:53:53 msaitoh Exp $*/
2 
3 /******************************************************************************
4 
5   Copyright (c) 2001-2017, Intel Corporation
6   All rights reserved.
7 
8   Redistribution and use in source and binary forms, with or without
9   modification, are permitted provided that the following conditions are met:
10 
11    1. Redistributions of source code must retain the above copyright notice,
12       this list of conditions and the following disclaimer.
13 
14    2. Redistributions in binary form must reproduce the above copyright
15       notice, this list of conditions and the following disclaimer in the
16       documentation and/or other materials provided with the distribution.
17 
18    3. Neither the name of the Intel Corporation nor the names of its
19       contributors may be used to endorse or promote products derived from
20       this software without specific prior written permission.
21 
22   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
23   AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24   IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25   ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26   LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27   CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28   SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29   INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30   CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31   ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32   POSSIBILITY OF SUCH DAMAGE.
33 
34 ******************************************************************************/
35 /*$FreeBSD: head/sys/dev/ixgbe/if_ixv.c 331224 2018-03-19 20:55:05Z erj $*/
36 
37 #ifdef _KERNEL_OPT
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_net_mpsafe.h"
41 #include "opt_ixgbe.h"
42 #endif
43 
44 #include "ixgbe.h"
45 #include "vlan.h"
46 
47 /************************************************************************
48  * Driver version
49  ************************************************************************/
50 static const char ixv_driver_version[] = "2.0.1-k";
51 /* XXX NetBSD: + 1.5.17 */
52 
53 /************************************************************************
54  * PCI Device ID Table
55  *
56  *   Used by probe to select devices to load on
57  *   Last field stores an index into ixv_strings
58  *   Last entry must be all 0s
59  *
60  *   { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
61  ************************************************************************/
62 static const ixgbe_vendor_info_t ixv_vendor_info_array[] =
63 {
64 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_VF, 0, 0, 0},
65 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540_VF, 0, 0, 0},
66 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550_VF, 0, 0, 0},
67 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_VF, 0, 0, 0},
68 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_VF, 0, 0, 0},
69 	/* required last entry */
70 	{0, 0, 0, 0, 0}
71 };
72 
73 /************************************************************************
74  * Table of branding strings
75  ************************************************************************/
76 static const char *ixv_strings[] = {
77 	"Intel(R) PRO/10GbE Virtual Function Network Driver"
78 };
79 
80 /*********************************************************************
81  *  Function prototypes
82  *********************************************************************/
83 static int	ixv_probe(device_t, cfdata_t, void *);
84 static void	ixv_attach(device_t, device_t, void *);
85 static int	ixv_detach(device_t, int);
86 #if 0
87 static int	ixv_shutdown(device_t);
88 #endif
89 static int	ixv_ifflags_cb(struct ethercom *);
90 static int	ixv_ioctl(struct ifnet *, u_long, void *);
91 static int	ixv_init(struct ifnet *);
92 static void	ixv_init_locked(struct adapter *);
93 static void	ixv_ifstop(struct ifnet *, int);
94 static void	ixv_stop_locked(void *);
95 static void	ixv_init_device_features(struct adapter *);
96 static void	ixv_media_status(struct ifnet *, struct ifmediareq *);
97 static int	ixv_media_change(struct ifnet *);
98 static int	ixv_allocate_pci_resources(struct adapter *,
99 		    const struct pci_attach_args *);
100 static void	ixv_free_deferred_handlers(struct adapter *);
101 static int	ixv_allocate_msix(struct adapter *,
102 		    const struct pci_attach_args *);
103 static int	ixv_configure_interrupts(struct adapter *);
104 static void	ixv_free_pci_resources(struct adapter *);
105 static void	ixv_local_timer(void *);
106 static void	ixv_handle_timer(struct work *, void *);
107 static int	ixv_setup_interface(device_t, struct adapter *);
108 static void	ixv_schedule_admin_tasklet(struct adapter *);
109 static int	ixv_negotiate_api(struct adapter *);
110 
111 static void	ixv_initialize_transmit_units(struct adapter *);
112 static void	ixv_initialize_receive_units(struct adapter *);
113 static void	ixv_initialize_rss_mapping(struct adapter *);
114 static s32	ixv_check_link(struct adapter *);
115 
116 static void	ixv_enable_intr(struct adapter *);
117 static void	ixv_disable_intr(struct adapter *);
118 static int	ixv_set_rxfilter(struct adapter *);
119 static void	ixv_update_link_status(struct adapter *);
120 static int	ixv_sysctl_debug(SYSCTLFN_PROTO);
121 static void	ixv_set_ivar(struct adapter *, u8, u8, s8);
122 static void	ixv_configure_ivars(struct adapter *);
123 static u8 *	ixv_mc_array_itr(struct ixgbe_hw *, u8 **, u32 *);
124 static void	ixv_eitr_write(struct adapter *, uint32_t, uint32_t);
125 
126 static void	ixv_setup_vlan_tagging(struct adapter *);
127 static int	ixv_setup_vlan_support(struct adapter *);
128 static int	ixv_vlan_cb(struct ethercom *, uint16_t, bool);
129 static int	ixv_register_vlan(struct adapter *, u16);
130 static int	ixv_unregister_vlan(struct adapter *, u16);
131 
132 static void	ixv_add_device_sysctls(struct adapter *);
133 static void	ixv_save_stats(struct adapter *);
134 static void	ixv_init_stats(struct adapter *);
135 static void	ixv_update_stats(struct adapter *);
136 static void	ixv_add_stats_sysctls(struct adapter *);
137 static void	ixv_clear_evcnt(struct adapter *);
138 
139 /* Sysctl handlers */
140 static void	ixv_set_sysctl_value(struct adapter *, const char *,
141 		    const char *, int *, int);
142 static int	ixv_sysctl_interrupt_rate_handler(SYSCTLFN_PROTO);
143 static int	ixv_sysctl_next_to_check_handler(SYSCTLFN_PROTO);
144 static int	ixv_sysctl_rdh_handler(SYSCTLFN_PROTO);
145 static int	ixv_sysctl_rdt_handler(SYSCTLFN_PROTO);
146 static int	ixv_sysctl_tdt_handler(SYSCTLFN_PROTO);
147 static int	ixv_sysctl_tdh_handler(SYSCTLFN_PROTO);
148 
149 /* The MSI-X Interrupt handlers */
150 static int	ixv_msix_que(void *);
151 static int	ixv_msix_mbx(void *);
152 
153 /* Event handlers running on workqueue */
154 static void	ixv_handle_que(void *);
155 
156 /* Deferred workqueue handlers */
157 static void	ixv_handle_admin(struct work *, void *);
158 static void	ixv_handle_que_work(struct work *, void *);
159 
160 const struct sysctlnode *ixv_sysctl_instance(struct adapter *);
161 static const ixgbe_vendor_info_t *ixv_lookup(const struct pci_attach_args *);
162 
163 /************************************************************************
164  * NetBSD Device Interface Entry Points
165  ************************************************************************/
166 CFATTACH_DECL3_NEW(ixv, sizeof(struct adapter),
167     ixv_probe, ixv_attach, ixv_detach, NULL, NULL, NULL,
168     DVF_DETACH_SHUTDOWN);
169 
170 #if 0
171 static driver_t ixv_driver = {
172 	"ixv", ixv_methods, sizeof(struct adapter),
173 };
174 
175 devclass_t ixv_devclass;
176 DRIVER_MODULE(ixv, pci, ixv_driver, ixv_devclass, 0, 0);
177 MODULE_DEPEND(ixv, pci, 1, 1, 1);
178 MODULE_DEPEND(ixv, ether, 1, 1, 1);
179 #endif
180 
181 /*
182  * TUNEABLE PARAMETERS:
183  */
184 
185 /* Number of Queues - do not exceed MSI-X vectors - 1 */
186 static int ixv_num_queues = 0;
187 #define	TUNABLE_INT(__x, __y)
188 TUNABLE_INT("hw.ixv.num_queues", &ixv_num_queues);
189 
190 /*
191  * AIM: Adaptive Interrupt Moderation
192  * which means that the interrupt rate
193  * is varied over time based on the
194  * traffic for that interrupt vector
195  */
196 static bool ixv_enable_aim = false;
197 TUNABLE_INT("hw.ixv.enable_aim", &ixv_enable_aim);
198 
199 static int ixv_max_interrupt_rate = (4000000 / IXGBE_LOW_LATENCY);
200 TUNABLE_INT("hw.ixv.max_interrupt_rate", &ixv_max_interrupt_rate);
201 
202 /* How many packets rxeof tries to clean at a time */
203 static int ixv_rx_process_limit = 256;
204 TUNABLE_INT("hw.ixv.rx_process_limit", &ixv_rx_process_limit);
205 
206 /* How many packets txeof tries to clean at a time */
207 static int ixv_tx_process_limit = 256;
208 TUNABLE_INT("hw.ixv.tx_process_limit", &ixv_tx_process_limit);
209 
210 /* Which packet processing uses workqueue or softint */
211 static bool ixv_txrx_workqueue = false;
212 
213 /*
214  * Number of TX descriptors per ring,
215  * setting higher than RX as this seems
216  * the better performing choice.
217  */
218 static int ixv_txd = PERFORM_TXD;
219 TUNABLE_INT("hw.ixv.txd", &ixv_txd);
220 
221 /* Number of RX descriptors per ring */
222 static int ixv_rxd = PERFORM_RXD;
223 TUNABLE_INT("hw.ixv.rxd", &ixv_rxd);
224 
225 /* Legacy Transmit (single queue) */
226 static int ixv_enable_legacy_tx = 0;
227 TUNABLE_INT("hw.ixv.enable_legacy_tx", &ixv_enable_legacy_tx);
228 
229 #ifdef NET_MPSAFE
230 #define IXGBE_MPSAFE		1
231 #define IXGBE_CALLOUT_FLAGS	CALLOUT_MPSAFE
232 #define IXGBE_SOFTINT_FLAGS	SOFTINT_MPSAFE
233 #define IXGBE_WORKQUEUE_FLAGS	WQ_PERCPU | WQ_MPSAFE
234 #define IXGBE_TASKLET_WQ_FLAGS	WQ_MPSAFE
235 #else
236 #define IXGBE_CALLOUT_FLAGS	0
237 #define IXGBE_SOFTINT_FLAGS	0
238 #define IXGBE_WORKQUEUE_FLAGS	WQ_PERCPU
239 #define IXGBE_TASKLET_WQ_FLAGS	0
240 #endif
241 #define IXGBE_WORKQUEUE_PRI PRI_SOFTNET
242 
243 #if 0
244 static int (*ixv_start_locked)(struct ifnet *, struct tx_ring *);
245 static int (*ixv_ring_empty)(struct ifnet *, struct buf_ring *);
246 #endif
247 
248 /************************************************************************
249  * ixv_probe - Device identification routine
250  *
251  *   Determines if the driver should be loaded on
252  *   adapter based on its PCI vendor/device ID.
253  *
254  *   return BUS_PROBE_DEFAULT on success, positive on failure
255  ************************************************************************/
256 static int
257 ixv_probe(device_t dev, cfdata_t cf, void *aux)
258 {
259 #ifdef __HAVE_PCI_MSI_MSIX
260 	const struct pci_attach_args *pa = aux;
261 
262 	return (ixv_lookup(pa) != NULL) ? 1 : 0;
263 #else
264 	return 0;
265 #endif
266 } /* ixv_probe */
267 
268 static const ixgbe_vendor_info_t *
269 ixv_lookup(const struct pci_attach_args *pa)
270 {
271 	const ixgbe_vendor_info_t *ent;
272 	pcireg_t subid;
273 
274 	INIT_DEBUGOUT("ixv_lookup: begin");
275 
276 	if (PCI_VENDOR(pa->pa_id) != IXGBE_INTEL_VENDOR_ID)
277 		return NULL;
278 
279 	subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
280 
281 	for (ent = ixv_vendor_info_array; ent->vendor_id != 0; ent++) {
282 		if ((PCI_VENDOR(pa->pa_id) == ent->vendor_id) &&
283 		    (PCI_PRODUCT(pa->pa_id) == ent->device_id) &&
284 		    ((PCI_SUBSYS_VENDOR(subid) == ent->subvendor_id) ||
285 		     (ent->subvendor_id == 0)) &&
286 		    ((PCI_SUBSYS_ID(subid) == ent->subdevice_id) ||
287 		     (ent->subdevice_id == 0))) {
288 			return ent;
289 		}
290 	}
291 
292 	return NULL;
293 }
294 
295 /************************************************************************
296  * ixv_attach - Device initialization routine
297  *
298  *   Called when the driver is being loaded.
299  *   Identifies the type of hardware, allocates all resources
300  *   and initializes the hardware.
301  *
302  *   return 0 on success, positive on failure
303  ************************************************************************/
304 static void
305 ixv_attach(device_t parent, device_t dev, void *aux)
306 {
307 	struct adapter *adapter;
308 	struct ixgbe_hw *hw;
309 	int		error = 0;
310 	pcireg_t	id, subid;
311 	const ixgbe_vendor_info_t *ent;
312 	const struct pci_attach_args *pa = aux;
313 	const char *apivstr;
314 	const char *str;
315 	char wqname[MAXCOMLEN];
316 	char buf[256];
317 
318 	INIT_DEBUGOUT("ixv_attach: begin");
319 
320 	/*
321 	 * Make sure BUSMASTER is set, on a VM under
322 	 * KVM it may not be and will break things.
323 	 */
324 	ixgbe_pci_enable_busmaster(pa->pa_pc, pa->pa_tag);
325 
326 	/* Allocate, clear, and link in our adapter structure */
327 	adapter = device_private(dev);
328 	adapter->hw.back = adapter;
329 	adapter->dev = dev;
330 	hw = &adapter->hw;
331 
332 	adapter->init_locked = ixv_init_locked;
333 	adapter->stop_locked = ixv_stop_locked;
334 
335 	adapter->osdep.pc = pa->pa_pc;
336 	adapter->osdep.tag = pa->pa_tag;
337 	if (pci_dma64_available(pa))
338 		adapter->osdep.dmat = pa->pa_dmat64;
339 	else
340 		adapter->osdep.dmat = pa->pa_dmat;
341 	adapter->osdep.attached = false;
342 
343 	ent = ixv_lookup(pa);
344 
345 	KASSERT(ent != NULL);
346 
347 	aprint_normal(": %s, Version - %s\n",
348 	    ixv_strings[ent->index], ixv_driver_version);
349 
350 	/* Core Lock Init */
351 	IXGBE_CORE_LOCK_INIT(adapter, device_xname(dev));
352 
353 	/* Do base PCI setup - map BAR0 */
354 	if (ixv_allocate_pci_resources(adapter, pa)) {
355 		aprint_error_dev(dev, "ixv_allocate_pci_resources() failed!\n");
356 		error = ENXIO;
357 		goto err_out;
358 	}
359 
360 	/* SYSCTL APIs */
361 	ixv_add_device_sysctls(adapter);
362 
363 	/* Set up the timer callout and workqueue */
364 	callout_init(&adapter->timer, IXGBE_CALLOUT_FLAGS);
365 	snprintf(wqname, sizeof(wqname), "%s-timer", device_xname(dev));
366 	error = workqueue_create(&adapter->timer_wq, wqname,
367 	    ixv_handle_timer, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
368 	    IXGBE_TASKLET_WQ_FLAGS);
369 	if (error) {
370 		aprint_error_dev(dev,
371 		    "could not create timer workqueue (%d)\n", error);
372 		goto err_out;
373 	}
374 
375 	/* Save off the information about this board */
376 	id = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_ID_REG);
377 	subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
378 	hw->vendor_id = PCI_VENDOR(id);
379 	hw->device_id = PCI_PRODUCT(id);
380 	hw->revision_id =
381 	    PCI_REVISION(pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_CLASS_REG));
382 	hw->subsystem_vendor_id = PCI_SUBSYS_VENDOR(subid);
383 	hw->subsystem_device_id = PCI_SUBSYS_ID(subid);
384 
385 	/* A subset of set_mac_type */
386 	switch (hw->device_id) {
387 	case IXGBE_DEV_ID_82599_VF:
388 		hw->mac.type = ixgbe_mac_82599_vf;
389 		str = "82599 VF";
390 		break;
391 	case IXGBE_DEV_ID_X540_VF:
392 		hw->mac.type = ixgbe_mac_X540_vf;
393 		str = "X540 VF";
394 		break;
395 	case IXGBE_DEV_ID_X550_VF:
396 		hw->mac.type = ixgbe_mac_X550_vf;
397 		str = "X550 VF";
398 		break;
399 	case IXGBE_DEV_ID_X550EM_X_VF:
400 		hw->mac.type = ixgbe_mac_X550EM_x_vf;
401 		str = "X550EM X VF";
402 		break;
403 	case IXGBE_DEV_ID_X550EM_A_VF:
404 		hw->mac.type = ixgbe_mac_X550EM_a_vf;
405 		str = "X550EM A VF";
406 		break;
407 	default:
408 		/* Shouldn't get here since probe succeeded */
409 		aprint_error_dev(dev, "Unknown device ID!\n");
410 		error = ENXIO;
411 		goto err_out;
412 		break;
413 	}
414 	aprint_normal_dev(dev, "device %s\n", str);
415 
416 	ixv_init_device_features(adapter);
417 
418 	/* Initialize the shared code */
419 	error = ixgbe_init_ops_vf(hw);
420 	if (error) {
421 		aprint_error_dev(dev, "ixgbe_init_ops_vf() failed!\n");
422 		error = EIO;
423 		goto err_out;
424 	}
425 
426 	/* Setup the mailbox */
427 	ixgbe_init_mbx_params_vf(hw);
428 
429 	/* Set the right number of segments */
430 	adapter->num_segs = IXGBE_82599_SCATTER;
431 
432 	/* Reset mbox api to 1.0 */
433 	error = hw->mac.ops.reset_hw(hw);
434 	if (error == IXGBE_ERR_RESET_FAILED)
435 		aprint_error_dev(dev, "...reset_hw() failure: Reset Failed!\n");
436 	else if (error)
437 		aprint_error_dev(dev, "...reset_hw() failed with error %d\n",
438 		    error);
439 	if (error) {
440 		error = EIO;
441 		goto err_out;
442 	}
443 
444 	error = hw->mac.ops.init_hw(hw);
445 	if (error) {
446 		aprint_error_dev(dev, "...init_hw() failed!\n");
447 		error = EIO;
448 		goto err_out;
449 	}
450 
451 	/* Negotiate mailbox API version */
452 	error = ixv_negotiate_api(adapter);
453 	if (error)
454 		aprint_normal_dev(dev,
455 		    "MBX API negotiation failed during attach!\n");
456 	switch (hw->api_version) {
457 	case ixgbe_mbox_api_10:
458 		apivstr = "1.0";
459 		break;
460 	case ixgbe_mbox_api_20:
461 		apivstr = "2.0";
462 		break;
463 	case ixgbe_mbox_api_11:
464 		apivstr = "1.1";
465 		break;
466 	case ixgbe_mbox_api_12:
467 		apivstr = "1.2";
468 		break;
469 	case ixgbe_mbox_api_13:
470 		apivstr = "1.3";
471 		break;
472 	default:
473 		apivstr = "unknown";
474 		break;
475 	}
476 	aprint_normal_dev(dev, "Mailbox API %s\n", apivstr);
477 
478 	/* If no mac address was assigned, make a random one */
479 	if (!ixv_check_ether_addr(hw->mac.addr)) {
480 		u8 addr[ETHER_ADDR_LEN];
481 		uint64_t rndval = cprng_strong64();
482 
483 		memcpy(addr, &rndval, sizeof(addr));
484 		addr[0] &= 0xFE;
485 		addr[0] |= 0x02;
486 		bcopy(addr, hw->mac.addr, sizeof(addr));
487 	}
488 
489 	/* Register for VLAN events */
490 	ether_set_vlan_cb(&adapter->osdep.ec, ixv_vlan_cb);
491 
492 	/* Sysctls for limiting the amount of work done in the taskqueues */
493 	ixv_set_sysctl_value(adapter, "rx_processing_limit",
494 	    "max number of rx packets to process",
495 	    &adapter->rx_process_limit, ixv_rx_process_limit);
496 
497 	ixv_set_sysctl_value(adapter, "tx_processing_limit",
498 	    "max number of tx packets to process",
499 	    &adapter->tx_process_limit, ixv_tx_process_limit);
500 
501 	/* Do descriptor calc and sanity checks */
502 	if (((ixv_txd * sizeof(union ixgbe_adv_tx_desc)) % DBA_ALIGN) != 0 ||
503 	    ixv_txd < MIN_TXD || ixv_txd > MAX_TXD) {
504 		aprint_error_dev(dev, "TXD config issue, using default!\n");
505 		adapter->num_tx_desc = DEFAULT_TXD;
506 	} else
507 		adapter->num_tx_desc = ixv_txd;
508 
509 	if (((ixv_rxd * sizeof(union ixgbe_adv_rx_desc)) % DBA_ALIGN) != 0 ||
510 	    ixv_rxd < MIN_RXD || ixv_rxd > MAX_RXD) {
511 		aprint_error_dev(dev, "RXD config issue, using default!\n");
512 		adapter->num_rx_desc = DEFAULT_RXD;
513 	} else
514 		adapter->num_rx_desc = ixv_rxd;
515 
516 	adapter->num_jcl = adapter->num_rx_desc * IXGBE_JCLNUM_MULTI;
517 
518 	/* Setup MSI-X */
519 	error = ixv_configure_interrupts(adapter);
520 	if (error)
521 		goto err_out;
522 
523 	/* Allocate our TX/RX Queues */
524 	if (ixgbe_allocate_queues(adapter)) {
525 		aprint_error_dev(dev, "ixgbe_allocate_queues() failed!\n");
526 		error = ENOMEM;
527 		goto err_out;
528 	}
529 
530 	/* hw.ix defaults init */
531 	adapter->enable_aim = ixv_enable_aim;
532 
533 	adapter->txrx_use_workqueue = ixv_txrx_workqueue;
534 
535 	error = ixv_allocate_msix(adapter, pa);
536 	if (error) {
537 		aprint_error_dev(dev, "ixv_allocate_msix() failed!\n");
538 		goto err_late;
539 	}
540 
541 	/* Setup OS specific network interface */
542 	error = ixv_setup_interface(dev, adapter);
543 	if (error != 0) {
544 		aprint_error_dev(dev, "ixv_setup_interface() failed!\n");
545 		goto err_late;
546 	}
547 
548 	/* Do the stats setup */
549 	ixv_save_stats(adapter);
550 	ixv_init_stats(adapter);
551 	ixv_add_stats_sysctls(adapter);
552 
553 	if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
554 		ixgbe_netmap_attach(adapter);
555 
556 	snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_cap);
557 	aprint_verbose_dev(dev, "feature cap %s\n", buf);
558 	snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_en);
559 	aprint_verbose_dev(dev, "feature ena %s\n", buf);
560 
561 	INIT_DEBUGOUT("ixv_attach: end");
562 	adapter->osdep.attached = true;
563 
564 	return;
565 
566 err_late:
567 	ixgbe_free_queues(adapter);
568 err_out:
569 	ixv_free_pci_resources(adapter);
570 	IXGBE_CORE_LOCK_DESTROY(adapter);
571 
572 	return;
573 } /* ixv_attach */
574 
575 /************************************************************************
576  * ixv_detach - Device removal routine
577  *
578  *   Called when the driver is being removed.
579  *   Stops the adapter and deallocates all the resources
580  *   that were allocated for driver operation.
581  *
582  *   return 0 on success, positive on failure
583  ************************************************************************/
584 static int
585 ixv_detach(device_t dev, int flags)
586 {
587 	struct adapter	*adapter = device_private(dev);
588 	struct ixgbe_hw *hw = &adapter->hw;
589 	struct tx_ring *txr = adapter->tx_rings;
590 	struct rx_ring *rxr = adapter->rx_rings;
591 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
592 
593 	INIT_DEBUGOUT("ixv_detach: begin");
594 	if (adapter->osdep.attached == false)
595 		return 0;
596 
597 	/* Stop the interface. Callouts are stopped in it. */
598 	ixv_ifstop(adapter->ifp, 1);
599 
600 #if NVLAN > 0
601 	/* Make sure VLANs are not using driver */
602 	if (!VLAN_ATTACHED(&adapter->osdep.ec))
603 		;	/* nothing to do: no VLANs */
604 	else if ((flags & (DETACH_SHUTDOWN | DETACH_FORCE)) != 0)
605 		vlan_ifdetach(adapter->ifp);
606 	else {
607 		aprint_error_dev(dev, "VLANs in use, detach first\n");
608 		return EBUSY;
609 	}
610 #endif
611 
612 	ether_ifdetach(adapter->ifp);
613 	callout_halt(&adapter->timer, NULL);
614 	ixv_free_deferred_handlers(adapter);
615 
616 	if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
617 		netmap_detach(adapter->ifp);
618 
619 	ixv_free_pci_resources(adapter);
620 #if 0 /* XXX the NetBSD port is probably missing something here */
621 	bus_generic_detach(dev);
622 #endif
623 	if_detach(adapter->ifp);
624 	ifmedia_fini(&adapter->media);
625 	if_percpuq_destroy(adapter->ipq);
626 
627 	sysctl_teardown(&adapter->sysctllog);
628 	evcnt_detach(&adapter->efbig_tx_dma_setup);
629 	evcnt_detach(&adapter->mbuf_defrag_failed);
630 	evcnt_detach(&adapter->efbig2_tx_dma_setup);
631 	evcnt_detach(&adapter->einval_tx_dma_setup);
632 	evcnt_detach(&adapter->other_tx_dma_setup);
633 	evcnt_detach(&adapter->eagain_tx_dma_setup);
634 	evcnt_detach(&adapter->enomem_tx_dma_setup);
635 	evcnt_detach(&adapter->watchdog_events);
636 	evcnt_detach(&adapter->tso_err);
637 	evcnt_detach(&adapter->admin_irqev);
638 	evcnt_detach(&adapter->link_workev);
639 
640 	txr = adapter->tx_rings;
641 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
642 		evcnt_detach(&adapter->queues[i].irqs);
643 		evcnt_detach(&adapter->queues[i].handleq);
644 		evcnt_detach(&adapter->queues[i].req);
645 		evcnt_detach(&txr->no_desc_avail);
646 		evcnt_detach(&txr->total_packets);
647 		evcnt_detach(&txr->tso_tx);
648 #ifndef IXGBE_LEGACY_TX
649 		evcnt_detach(&txr->pcq_drops);
650 #endif
651 
652 		evcnt_detach(&rxr->rx_packets);
653 		evcnt_detach(&rxr->rx_bytes);
654 		evcnt_detach(&rxr->rx_copies);
655 		evcnt_detach(&rxr->no_jmbuf);
656 		evcnt_detach(&rxr->rx_discarded);
657 	}
658 	evcnt_detach(&stats->ipcs);
659 	evcnt_detach(&stats->l4cs);
660 	evcnt_detach(&stats->ipcs_bad);
661 	evcnt_detach(&stats->l4cs_bad);
662 
663 	/* Packet Reception Stats */
664 	evcnt_detach(&stats->vfgorc);
665 	evcnt_detach(&stats->vfgprc);
666 	evcnt_detach(&stats->vfmprc);
667 
668 	/* Packet Transmission Stats */
669 	evcnt_detach(&stats->vfgotc);
670 	evcnt_detach(&stats->vfgptc);
671 
672 	/* Mailbox Stats */
673 	evcnt_detach(&hw->mbx.stats.msgs_tx);
674 	evcnt_detach(&hw->mbx.stats.msgs_rx);
675 	evcnt_detach(&hw->mbx.stats.acks);
676 	evcnt_detach(&hw->mbx.stats.reqs);
677 	evcnt_detach(&hw->mbx.stats.rsts);
678 
679 	ixgbe_free_queues(adapter);
680 
681 	IXGBE_CORE_LOCK_DESTROY(adapter);
682 
683 	return (0);
684 } /* ixv_detach */
685 
686 /************************************************************************
687  * ixv_init_locked - Init entry point
688  *
689  *   Used in two ways: It is used by the stack as an init entry
690  *   point in network interface structure. It is also used
691  *   by the driver as a hw/sw initialization routine to get
692  *   to a consistent state.
693  *
694  *   return 0 on success, positive on failure
695  ************************************************************************/
696 static void
697 ixv_init_locked(struct adapter *adapter)
698 {
699 	struct ifnet	*ifp = adapter->ifp;
700 	device_t	dev = adapter->dev;
701 	struct ixgbe_hw *hw = &adapter->hw;
702 	struct ix_queue	*que;
703 	int		error = 0;
704 	uint32_t mask;
705 	int i;
706 
707 	INIT_DEBUGOUT("ixv_init_locked: begin");
708 	KASSERT(mutex_owned(&adapter->core_mtx));
709 	hw->adapter_stopped = FALSE;
710 	hw->mac.ops.stop_adapter(hw);
711 	callout_stop(&adapter->timer);
712 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
713 		que->disabled_count = 0;
714 
715 	adapter->max_frame_size =
716 	    ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
717 
718 	/* reprogram the RAR[0] in case user changed it. */
719 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
720 
721 	/* Get the latest mac address, User can use a LAA */
722 	memcpy(hw->mac.addr, CLLADDR(ifp->if_sadl),
723 	     IXGBE_ETH_LENGTH_OF_ADDRESS);
724 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, 1);
725 
726 	/* Prepare transmit descriptors and buffers */
727 	if (ixgbe_setup_transmit_structures(adapter)) {
728 		aprint_error_dev(dev, "Could not setup transmit structures\n");
729 		ixv_stop_locked(adapter);
730 		return;
731 	}
732 
733 	/* Reset VF and renegotiate mailbox API version */
734 	hw->mac.ops.reset_hw(hw);
735 	hw->mac.ops.start_hw(hw);
736 	error = ixv_negotiate_api(adapter);
737 	if (error)
738 		device_printf(dev,
739 		    "Mailbox API negotiation failed in init_locked!\n");
740 
741 	ixv_initialize_transmit_units(adapter);
742 
743 	/* Setup Multicast table */
744 	ixv_set_rxfilter(adapter);
745 
746 	/*
747 	 * Determine the correct mbuf pool
748 	 * for doing jumbo/headersplit
749 	 */
750 	if (adapter->max_frame_size <= MCLBYTES)
751 		adapter->rx_mbuf_sz = MCLBYTES;
752 	else
753 		adapter->rx_mbuf_sz = MJUMPAGESIZE;
754 
755 	/* Prepare receive descriptors and buffers */
756 	if (ixgbe_setup_receive_structures(adapter)) {
757 		device_printf(dev, "Could not setup receive structures\n");
758 		ixv_stop_locked(adapter);
759 		return;
760 	}
761 
762 	/* Configure RX settings */
763 	ixv_initialize_receive_units(adapter);
764 
765 	/* Initialize variable holding task enqueue requests interrupts */
766 	adapter->task_requests = 0;
767 
768 	/* Set up VLAN offload and filter */
769 	ixv_setup_vlan_support(adapter);
770 
771 	/* Set up MSI-X routing */
772 	ixv_configure_ivars(adapter);
773 
774 	/* Set up auto-mask */
775 	mask = (1 << adapter->vector);
776 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
777 		mask |= (1 << que->msix);
778 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, mask);
779 
780 	/* Set moderation on the Link interrupt */
781 	ixv_eitr_write(adapter, adapter->vector, IXGBE_LINK_ITR);
782 
783 	/* Stats init */
784 	ixv_init_stats(adapter);
785 
786 	/* Config/Enable Link */
787 	hw->mac.get_link_status = TRUE;
788 	hw->mac.ops.check_link(hw, &adapter->link_speed, &adapter->link_up,
789 	    FALSE);
790 
791 	/* Start watchdog */
792 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
793 	atomic_store_relaxed(&adapter->timer_pending, 0);
794 
795 	/* OK to schedule workqueues. */
796 	adapter->schedule_wqs_ok = true;
797 
798 	/* And now turn on interrupts */
799 	ixv_enable_intr(adapter);
800 
801 	/* Update saved flags. See ixgbe_ifflags_cb() */
802 	adapter->if_flags = ifp->if_flags;
803 	adapter->ec_capenable = adapter->osdep.ec.ec_capenable;
804 
805 	/* Now inform the stack we're ready */
806 	ifp->if_flags |= IFF_RUNNING;
807 	ifp->if_flags &= ~IFF_OACTIVE;
808 
809 	return;
810 } /* ixv_init_locked */
811 
812 /************************************************************************
813  * ixv_enable_queue
814  ************************************************************************/
815 static inline void
816 ixv_enable_queue(struct adapter *adapter, u32 vector)
817 {
818 	struct ixgbe_hw *hw = &adapter->hw;
819 	struct ix_queue *que = &adapter->queues[vector];
820 	u32		queue = 1UL << vector;
821 	u32		mask;
822 
823 	mutex_enter(&que->dc_mtx);
824 	if (que->disabled_count > 0 && --que->disabled_count > 0)
825 		goto out;
826 
827 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
828 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
829 out:
830 	mutex_exit(&que->dc_mtx);
831 } /* ixv_enable_queue */
832 
833 /************************************************************************
834  * ixv_disable_queue
835  ************************************************************************/
836 static inline void
837 ixv_disable_queue(struct adapter *adapter, u32 vector)
838 {
839 	struct ixgbe_hw *hw = &adapter->hw;
840 	struct ix_queue *que = &adapter->queues[vector];
841 	u32		queue = 1UL << vector;
842 	u32		mask;
843 
844 	mutex_enter(&que->dc_mtx);
845 	if (que->disabled_count++ > 0)
846 		goto  out;
847 
848 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
849 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, mask);
850 out:
851 	mutex_exit(&que->dc_mtx);
852 } /* ixv_disable_queue */
853 
854 #if 0
855 static inline void
856 ixv_rearm_queues(struct adapter *adapter, u64 queues)
857 {
858 	u32 mask = (IXGBE_EIMS_RTX_QUEUE & queues);
859 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEICS, mask);
860 } /* ixv_rearm_queues */
861 #endif
862 
863 
864 /************************************************************************
865  * ixv_msix_que - MSI-X Queue Interrupt Service routine
866  ************************************************************************/
867 static int
868 ixv_msix_que(void *arg)
869 {
870 	struct ix_queue	*que = arg;
871 	struct adapter	*adapter = que->adapter;
872 	struct tx_ring	*txr = que->txr;
873 	struct rx_ring	*rxr = que->rxr;
874 	bool		more;
875 	u32		newitr = 0;
876 
877 	ixv_disable_queue(adapter, que->msix);
878 	++que->irqs.ev_count;
879 
880 #ifdef __NetBSD__
881 	/* Don't run ixgbe_rxeof in interrupt context */
882 	more = true;
883 #else
884 	more = ixgbe_rxeof(que);
885 #endif
886 
887 	IXGBE_TX_LOCK(txr);
888 	ixgbe_txeof(txr);
889 	IXGBE_TX_UNLOCK(txr);
890 
891 	/* Do AIM now? */
892 
893 	if (adapter->enable_aim == false)
894 		goto no_calc;
895 	/*
896 	 * Do Adaptive Interrupt Moderation:
897 	 *  - Write out last calculated setting
898 	 *  - Calculate based on average size over
899 	 *    the last interval.
900 	 */
901 	if (que->eitr_setting)
902 		ixv_eitr_write(adapter, que->msix, que->eitr_setting);
903 
904 	que->eitr_setting = 0;
905 
906 	/* Idle, do nothing */
907 	if ((txr->bytes == 0) && (rxr->bytes == 0))
908 		goto no_calc;
909 
910 	if ((txr->bytes) && (txr->packets))
911 		newitr = txr->bytes/txr->packets;
912 	if ((rxr->bytes) && (rxr->packets))
913 		newitr = uimax(newitr, (rxr->bytes / rxr->packets));
914 	newitr += 24; /* account for hardware frame, crc */
915 
916 	/* set an upper boundary */
917 	newitr = uimin(newitr, 3000);
918 
919 	/* Be nice to the mid range */
920 	if ((newitr > 300) && (newitr < 1200))
921 		newitr = (newitr / 3);
922 	else
923 		newitr = (newitr / 2);
924 
925 	/*
926 	 * When RSC is used, ITR interval must be larger than RSC_DELAY.
927 	 * Currently, we use 2us for RSC_DELAY. The minimum value is always
928 	 * greater than 2us on 100M (and 10M?(not documented)), but it's not
929 	 * on 1G and higher.
930 	 */
931 	if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
932 	    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
933 		if (newitr < IXGBE_MIN_RSC_EITR_10G1G)
934 			newitr = IXGBE_MIN_RSC_EITR_10G1G;
935 	}
936 
937 	/* save for next interrupt */
938 	que->eitr_setting = newitr;
939 
940 	/* Reset state */
941 	txr->bytes = 0;
942 	txr->packets = 0;
943 	rxr->bytes = 0;
944 	rxr->packets = 0;
945 
946 no_calc:
947 	if (more)
948 		softint_schedule(que->que_si);
949 	else /* Re-enable this interrupt */
950 		ixv_enable_queue(adapter, que->msix);
951 
952 	return 1;
953 } /* ixv_msix_que */
954 
955 /************************************************************************
956  * ixv_msix_mbx
957  ************************************************************************/
958 static int
959 ixv_msix_mbx(void *arg)
960 {
961 	struct adapter	*adapter = arg;
962 	struct ixgbe_hw *hw = &adapter->hw;
963 
964 	++adapter->admin_irqev.ev_count;
965 	/* NetBSD: We use auto-clear, so it's not required to write VTEICR */
966 
967 	/* Link status change */
968 	hw->mac.get_link_status = TRUE;
969 	atomic_or_32(&adapter->task_requests, IXGBE_REQUEST_TASK_MBX);
970 	ixv_schedule_admin_tasklet(adapter);
971 
972 	return 1;
973 } /* ixv_msix_mbx */
974 
975 static void
976 ixv_eitr_write(struct adapter *adapter, uint32_t index, uint32_t itr)
977 {
978 
979 	/*
980 	 * Newer devices than 82598 have VF function, so this function is
981 	 * simple.
982 	 */
983 	itr |= IXGBE_EITR_CNT_WDIS;
984 
985 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEITR(index), itr);
986 }
987 
988 
989 /************************************************************************
990  * ixv_media_status - Media Ioctl callback
991  *
992  *   Called whenever the user queries the status of
993  *   the interface using ifconfig.
994  ************************************************************************/
995 static void
996 ixv_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
997 {
998 	struct adapter *adapter = ifp->if_softc;
999 
1000 	INIT_DEBUGOUT("ixv_media_status: begin");
1001 	ixv_update_link_status(adapter);
1002 
1003 	ifmr->ifm_status = IFM_AVALID;
1004 	ifmr->ifm_active = IFM_ETHER;
1005 
1006 	if (adapter->link_active != LINK_STATE_UP) {
1007 		ifmr->ifm_active |= IFM_NONE;
1008 		return;
1009 	}
1010 
1011 	ifmr->ifm_status |= IFM_ACTIVE;
1012 
1013 	switch (adapter->link_speed) {
1014 		case IXGBE_LINK_SPEED_10GB_FULL:
1015 			ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
1016 			break;
1017 		case IXGBE_LINK_SPEED_5GB_FULL:
1018 			ifmr->ifm_active |= IFM_5000_T | IFM_FDX;
1019 			break;
1020 		case IXGBE_LINK_SPEED_2_5GB_FULL:
1021 			ifmr->ifm_active |= IFM_2500_T | IFM_FDX;
1022 			break;
1023 		case IXGBE_LINK_SPEED_1GB_FULL:
1024 			ifmr->ifm_active |= IFM_1000_T | IFM_FDX;
1025 			break;
1026 		case IXGBE_LINK_SPEED_100_FULL:
1027 			ifmr->ifm_active |= IFM_100_TX | IFM_FDX;
1028 			break;
1029 		case IXGBE_LINK_SPEED_10_FULL:
1030 			ifmr->ifm_active |= IFM_10_T | IFM_FDX;
1031 			break;
1032 	}
1033 
1034 	ifp->if_baudrate = ifmedia_baudrate(ifmr->ifm_active);
1035 } /* ixv_media_status */
1036 
1037 /************************************************************************
1038  * ixv_media_change - Media Ioctl callback
1039  *
1040  *   Called when the user changes speed/duplex using
1041  *   media/mediopt option with ifconfig.
1042  ************************************************************************/
1043 static int
1044 ixv_media_change(struct ifnet *ifp)
1045 {
1046 	struct adapter *adapter = ifp->if_softc;
1047 	struct ifmedia *ifm = &adapter->media;
1048 
1049 	INIT_DEBUGOUT("ixv_media_change: begin");
1050 
1051 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1052 		return (EINVAL);
1053 
1054 	switch (IFM_SUBTYPE(ifm->ifm_media)) {
1055 	case IFM_AUTO:
1056 		break;
1057 	default:
1058 		device_printf(adapter->dev, "Only auto media type\n");
1059 		return (EINVAL);
1060 	}
1061 
1062 	return (0);
1063 } /* ixv_media_change */
1064 
1065 static void
1066 ixv_schedule_admin_tasklet(struct adapter *adapter)
1067 {
1068 	if (adapter->schedule_wqs_ok) {
1069 		if (atomic_cas_uint(&adapter->admin_pending, 0, 1) == 0)
1070 			workqueue_enqueue(adapter->admin_wq,
1071 			    &adapter->admin_wc, NULL);
1072 	}
1073 }
1074 
1075 /************************************************************************
1076  * ixv_negotiate_api
1077  *
1078  *   Negotiate the Mailbox API with the PF;
1079  *   start with the most featured API first.
1080  ************************************************************************/
1081 static int
1082 ixv_negotiate_api(struct adapter *adapter)
1083 {
1084 	struct ixgbe_hw *hw = &adapter->hw;
1085 	int		mbx_api[] = { ixgbe_mbox_api_13,
1086 				      ixgbe_mbox_api_12,
1087 				      ixgbe_mbox_api_11,
1088 				      ixgbe_mbox_api_10,
1089 				      ixgbe_mbox_api_unknown };
1090 	int		i = 0;
1091 
1092 	while (mbx_api[i] != ixgbe_mbox_api_unknown) {
1093 		if (ixgbevf_negotiate_api_version(hw, mbx_api[i]) == 0)
1094 			return (0);
1095 		i++;
1096 	}
1097 
1098 	return (EINVAL);
1099 } /* ixv_negotiate_api */
1100 
1101 
1102 /************************************************************************
1103  * ixv_set_rxfilter - Multicast Update
1104  *
1105  *   Called whenever multicast address list is updated.
1106  ************************************************************************/
1107 static int
1108 ixv_set_rxfilter(struct adapter *adapter)
1109 {
1110 	u8	mta[IXGBE_MAX_VF_MC * IXGBE_ETH_LENGTH_OF_ADDRESS];
1111 	struct ifnet		*ifp = adapter->ifp;
1112 	struct ixgbe_hw		*hw = &adapter->hw;
1113 	u8			*update_ptr;
1114 	int			mcnt = 0;
1115 	struct ethercom		*ec = &adapter->osdep.ec;
1116 	struct ether_multi	*enm;
1117 	struct ether_multistep	step;
1118 	bool			overflow = false;
1119 	int			error, rc = 0;
1120 
1121 	KASSERT(mutex_owned(&adapter->core_mtx));
1122 	IOCTL_DEBUGOUT("ixv_set_rxfilter: begin");
1123 
1124 	/* 1: For PROMISC */
1125 	if (ifp->if_flags & IFF_PROMISC) {
1126 		error = hw->mac.ops.update_xcast_mode(hw,
1127 		    IXGBEVF_XCAST_MODE_PROMISC);
1128 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1129 			device_printf(adapter->dev,
1130 			    "this interface is not trusted\n");
1131 			error = EPERM;
1132 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1133 			device_printf(adapter->dev,
1134 			    "the PF doesn't support promisc mode\n");
1135 			error = EOPNOTSUPP;
1136 		} else if (error == IXGBE_ERR_NOT_IN_PROMISC) {
1137 			device_printf(adapter->dev,
1138 			    "the PF may not in promisc mode\n");
1139 			error = EINVAL;
1140 		} else if (error) {
1141 			device_printf(adapter->dev,
1142 			    "failed to set promisc mode. error = %d\n",
1143 			    error);
1144 			error = EIO;
1145 		} else
1146 			return 0;
1147 		rc = error;
1148 	}
1149 
1150 	/* 2: For ALLMULTI or normal */
1151 	ETHER_LOCK(ec);
1152 	ETHER_FIRST_MULTI(step, ec, enm);
1153 	while (enm != NULL) {
1154 		if ((mcnt >= IXGBE_MAX_VF_MC) ||
1155 		    (memcmp(enm->enm_addrlo, enm->enm_addrhi,
1156 			ETHER_ADDR_LEN) != 0)) {
1157 			overflow = true;
1158 			break;
1159 		}
1160 		bcopy(enm->enm_addrlo,
1161 		    &mta[mcnt * IXGBE_ETH_LENGTH_OF_ADDRESS],
1162 		    IXGBE_ETH_LENGTH_OF_ADDRESS);
1163 		mcnt++;
1164 		ETHER_NEXT_MULTI(step, enm);
1165 	}
1166 	ETHER_UNLOCK(ec);
1167 
1168 	/* 3: For ALLMULTI */
1169 	if (overflow) {
1170 		error = hw->mac.ops.update_xcast_mode(hw,
1171 		    IXGBEVF_XCAST_MODE_ALLMULTI);
1172 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1173 			device_printf(adapter->dev,
1174 			    "this interface is not trusted\n");
1175 			error = EPERM;
1176 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1177 			device_printf(adapter->dev,
1178 			    "the PF doesn't support allmulti mode\n");
1179 			error = EOPNOTSUPP;
1180 		} else if (error) {
1181 			device_printf(adapter->dev,
1182 			    "number of Ethernet multicast addresses "
1183 			    "exceeds the limit (%d). error = %d\n",
1184 			    IXGBE_MAX_VF_MC, error);
1185 			error = ENOSPC;
1186 		} else {
1187 			ETHER_LOCK(ec);
1188 			ec->ec_flags |= ETHER_F_ALLMULTI;
1189 			ETHER_UNLOCK(ec);
1190 			return rc; /* Promisc might have failed */
1191 		}
1192 
1193 		if (rc == 0)
1194 			rc = error;
1195 
1196 		/* Continue to update the multicast table as many as we can */
1197 	}
1198 
1199 	/* 4: For normal operation */
1200 	error = hw->mac.ops.update_xcast_mode(hw, IXGBEVF_XCAST_MODE_MULTI);
1201 	if ((error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) || (error == 0)) {
1202 		/* Normal operation */
1203 		ETHER_LOCK(ec);
1204 		ec->ec_flags &= ~ETHER_F_ALLMULTI;
1205 		ETHER_UNLOCK(ec);
1206 		error = 0;
1207 	} else if (error) {
1208 		device_printf(adapter->dev,
1209 		    "failed to set Ethernet multicast address "
1210 		    "operation to normal. error = %d\n", error);
1211 	}
1212 
1213 	update_ptr = mta;
1214 
1215 	error = adapter->hw.mac.ops.update_mc_addr_list(&adapter->hw,
1216 	    update_ptr, mcnt, ixv_mc_array_itr, TRUE);
1217 	if (rc == 0)
1218 		rc = error;
1219 
1220 	return rc;
1221 } /* ixv_set_rxfilter */
1222 
1223 /************************************************************************
1224  * ixv_mc_array_itr
1225  *
1226  *   An iterator function needed by the multicast shared code.
1227  *   It feeds the shared code routine the addresses in the
1228  *   array of ixv_set_rxfilter() one by one.
1229  ************************************************************************/
1230 static u8 *
1231 ixv_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq)
1232 {
1233 	u8 *addr = *update_ptr;
1234 	u8 *newptr;
1235 
1236 	*vmdq = 0;
1237 
1238 	newptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
1239 	*update_ptr = newptr;
1240 
1241 	return addr;
1242 } /* ixv_mc_array_itr */
1243 
1244 /************************************************************************
1245  * ixv_local_timer - Timer routine
1246  *
1247  *   Checks for link status, updates statistics,
1248  *   and runs the watchdog check.
1249  ************************************************************************/
1250 static void
1251 ixv_local_timer(void *arg)
1252 {
1253 	struct adapter *adapter = arg;
1254 
1255 	if (adapter->schedule_wqs_ok) {
1256 		if (atomic_cas_uint(&adapter->timer_pending, 0, 1) == 0)
1257 			workqueue_enqueue(adapter->timer_wq,
1258 			    &adapter->timer_wc, NULL);
1259 	}
1260 }
1261 
1262 static void
1263 ixv_handle_timer(struct work *wk, void *context)
1264 {
1265 	struct adapter	*adapter = context;
1266 	device_t	dev = adapter->dev;
1267 	struct ix_queue	*que = adapter->queues;
1268 	u64		queues = 0;
1269 	u64		v0, v1, v2, v3, v4, v5, v6, v7;
1270 	int		hung = 0;
1271 	int		i;
1272 
1273 	IXGBE_CORE_LOCK(adapter);
1274 
1275 	if (ixv_check_link(adapter)) {
1276 		ixv_init_locked(adapter);
1277 		IXGBE_CORE_UNLOCK(adapter);
1278 		return;
1279 	}
1280 
1281 	/* Stats Update */
1282 	ixv_update_stats(adapter);
1283 
1284 	/* Update some event counters */
1285 	v0 = v1 = v2 = v3 = v4 = v5 = v6 = v7 = 0;
1286 	que = adapter->queues;
1287 	for (i = 0; i < adapter->num_queues; i++, que++) {
1288 		struct tx_ring	*txr = que->txr;
1289 
1290 		v0 += txr->q_efbig_tx_dma_setup;
1291 		v1 += txr->q_mbuf_defrag_failed;
1292 		v2 += txr->q_efbig2_tx_dma_setup;
1293 		v3 += txr->q_einval_tx_dma_setup;
1294 		v4 += txr->q_other_tx_dma_setup;
1295 		v5 += txr->q_eagain_tx_dma_setup;
1296 		v6 += txr->q_enomem_tx_dma_setup;
1297 		v7 += txr->q_tso_err;
1298 	}
1299 	adapter->efbig_tx_dma_setup.ev_count = v0;
1300 	adapter->mbuf_defrag_failed.ev_count = v1;
1301 	adapter->efbig2_tx_dma_setup.ev_count = v2;
1302 	adapter->einval_tx_dma_setup.ev_count = v3;
1303 	adapter->other_tx_dma_setup.ev_count = v4;
1304 	adapter->eagain_tx_dma_setup.ev_count = v5;
1305 	adapter->enomem_tx_dma_setup.ev_count = v6;
1306 	adapter->tso_err.ev_count = v7;
1307 
1308 	/*
1309 	 * Check the TX queues status
1310 	 *	- mark hung queues so we don't schedule on them
1311 	 *	- watchdog only if all queues show hung
1312 	 */
1313 	que = adapter->queues;
1314 	for (i = 0; i < adapter->num_queues; i++, que++) {
1315 		/* Keep track of queues with work for soft irq */
1316 		if (que->txr->busy)
1317 			queues |= ((u64)1 << que->me);
1318 		/*
1319 		 * Each time txeof runs without cleaning, but there
1320 		 * are uncleaned descriptors it increments busy. If
1321 		 * we get to the MAX we declare it hung.
1322 		 */
1323 		if (que->busy == IXGBE_QUEUE_HUNG) {
1324 			++hung;
1325 			/* Mark the queue as inactive */
1326 			adapter->active_queues &= ~((u64)1 << que->me);
1327 			continue;
1328 		} else {
1329 			/* Check if we've come back from hung */
1330 			if ((adapter->active_queues & ((u64)1 << que->me)) == 0)
1331 				adapter->active_queues |= ((u64)1 << que->me);
1332 		}
1333 		if (que->busy >= IXGBE_MAX_TX_BUSY) {
1334 			device_printf(dev,
1335 			    "Warning queue %d appears to be hung!\n", i);
1336 			que->txr->busy = IXGBE_QUEUE_HUNG;
1337 			++hung;
1338 		}
1339 	}
1340 
1341 	/* Only truly watchdog if all queues show hung */
1342 	if (hung == adapter->num_queues)
1343 		goto watchdog;
1344 #if 0
1345 	else if (queues != 0) { /* Force an IRQ on queues with work */
1346 		ixv_rearm_queues(adapter, queues);
1347 	}
1348 #endif
1349 
1350 	atomic_store_relaxed(&adapter->timer_pending, 0);
1351 	IXGBE_CORE_UNLOCK(adapter);
1352 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
1353 
1354 	return;
1355 
1356 watchdog:
1357 	device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
1358 	adapter->ifp->if_flags &= ~IFF_RUNNING;
1359 	adapter->watchdog_events.ev_count++;
1360 	ixv_init_locked(adapter);
1361 	IXGBE_CORE_UNLOCK(adapter);
1362 } /* ixv_handle_timer */
1363 
1364 /************************************************************************
1365  * ixv_update_link_status - Update OS on link state
1366  *
1367  * Note: Only updates the OS on the cached link state.
1368  *	 The real check of the hardware only happens with
1369  *	 a link interrupt.
1370  ************************************************************************/
1371 static void
1372 ixv_update_link_status(struct adapter *adapter)
1373 {
1374 	struct ifnet *ifp = adapter->ifp;
1375 	device_t     dev = adapter->dev;
1376 
1377 	KASSERT(mutex_owned(&adapter->core_mtx));
1378 
1379 	if (adapter->link_up) {
1380 		if (adapter->link_active != LINK_STATE_UP) {
1381 			if (bootverbose) {
1382 				const char *bpsmsg;
1383 
1384 				switch (adapter->link_speed) {
1385 				case IXGBE_LINK_SPEED_10GB_FULL:
1386 					bpsmsg = "10 Gbps";
1387 					break;
1388 				case IXGBE_LINK_SPEED_5GB_FULL:
1389 					bpsmsg = "5 Gbps";
1390 					break;
1391 				case IXGBE_LINK_SPEED_2_5GB_FULL:
1392 					bpsmsg = "2.5 Gbps";
1393 					break;
1394 				case IXGBE_LINK_SPEED_1GB_FULL:
1395 					bpsmsg = "1 Gbps";
1396 					break;
1397 				case IXGBE_LINK_SPEED_100_FULL:
1398 					bpsmsg = "100 Mbps";
1399 					break;
1400 				case IXGBE_LINK_SPEED_10_FULL:
1401 					bpsmsg = "10 Mbps";
1402 					break;
1403 				default:
1404 					bpsmsg = "unknown speed";
1405 					break;
1406 				}
1407 				device_printf(dev, "Link is up %s %s \n",
1408 				    bpsmsg, "Full Duplex");
1409 			}
1410 			adapter->link_active = LINK_STATE_UP;
1411 			if_link_state_change(ifp, LINK_STATE_UP);
1412 		}
1413 	} else {
1414 		/*
1415 		 * Do it when link active changes to DOWN. i.e.
1416 		 * a) LINK_STATE_UNKNOWN -> LINK_STATE_DOWN
1417 		 * b) LINK_STATE_UP	 -> LINK_STATE_DOWN
1418 		 */
1419 		if (adapter->link_active != LINK_STATE_DOWN) {
1420 			if (bootverbose)
1421 				device_printf(dev, "Link is Down\n");
1422 			if_link_state_change(ifp, LINK_STATE_DOWN);
1423 			adapter->link_active = LINK_STATE_DOWN;
1424 		}
1425 	}
1426 } /* ixv_update_link_status */
1427 
1428 
1429 /************************************************************************
1430  * ixv_stop - Stop the hardware
1431  *
1432  *   Disables all traffic on the adapter by issuing a
1433  *   global reset on the MAC and deallocates TX/RX buffers.
1434  ************************************************************************/
1435 static void
1436 ixv_ifstop(struct ifnet *ifp, int disable)
1437 {
1438 	struct adapter *adapter = ifp->if_softc;
1439 
1440 	IXGBE_CORE_LOCK(adapter);
1441 	ixv_stop_locked(adapter);
1442 	IXGBE_CORE_UNLOCK(adapter);
1443 
1444 	workqueue_wait(adapter->admin_wq, &adapter->admin_wc);
1445 	atomic_store_relaxed(&adapter->admin_pending, 0);
1446 	workqueue_wait(adapter->timer_wq, &adapter->timer_wc);
1447 	atomic_store_relaxed(&adapter->timer_pending, 0);
1448 }
1449 
1450 static void
1451 ixv_stop_locked(void *arg)
1452 {
1453 	struct ifnet	*ifp;
1454 	struct adapter	*adapter = arg;
1455 	struct ixgbe_hw *hw = &adapter->hw;
1456 
1457 	ifp = adapter->ifp;
1458 
1459 	KASSERT(mutex_owned(&adapter->core_mtx));
1460 
1461 	INIT_DEBUGOUT("ixv_stop_locked: begin\n");
1462 	ixv_disable_intr(adapter);
1463 
1464 	/* Tell the stack that the interface is no longer active */
1465 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1466 
1467 	hw->mac.ops.reset_hw(hw);
1468 	adapter->hw.adapter_stopped = FALSE;
1469 	hw->mac.ops.stop_adapter(hw);
1470 	callout_stop(&adapter->timer);
1471 
1472 	/* Don't schedule workqueues. */
1473 	adapter->schedule_wqs_ok = false;
1474 
1475 	/* reprogram the RAR[0] in case user changed it. */
1476 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
1477 
1478 	return;
1479 } /* ixv_stop_locked */
1480 
1481 
1482 /************************************************************************
1483  * ixv_allocate_pci_resources
1484  ************************************************************************/
1485 static int
1486 ixv_allocate_pci_resources(struct adapter *adapter,
1487     const struct pci_attach_args *pa)
1488 {
1489 	pcireg_t	memtype, csr;
1490 	device_t	dev = adapter->dev;
1491 	bus_addr_t addr;
1492 	int flags;
1493 
1494 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, PCI_BAR(0));
1495 	switch (memtype) {
1496 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
1497 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
1498 		adapter->osdep.mem_bus_space_tag = pa->pa_memt;
1499 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, PCI_BAR(0),
1500 		      memtype, &addr, &adapter->osdep.mem_size, &flags) != 0)
1501 			goto map_err;
1502 		if ((flags & BUS_SPACE_MAP_PREFETCHABLE) != 0) {
1503 			aprint_normal_dev(dev, "clearing prefetchable bit\n");
1504 			flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
1505 		}
1506 		if (bus_space_map(adapter->osdep.mem_bus_space_tag, addr,
1507 		     adapter->osdep.mem_size, flags,
1508 		     &adapter->osdep.mem_bus_space_handle) != 0) {
1509 map_err:
1510 			adapter->osdep.mem_size = 0;
1511 			aprint_error_dev(dev, "unable to map BAR0\n");
1512 			return ENXIO;
1513 		}
1514 		/*
1515 		 * Enable address decoding for memory range in case it's not
1516 		 * set.
1517 		 */
1518 		csr = pci_conf_read(pa->pa_pc, pa->pa_tag,
1519 		    PCI_COMMAND_STATUS_REG);
1520 		csr |= PCI_COMMAND_MEM_ENABLE;
1521 		pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
1522 		    csr);
1523 		break;
1524 	default:
1525 		aprint_error_dev(dev, "unexpected type on BAR0\n");
1526 		return ENXIO;
1527 	}
1528 
1529 	/* Pick up the tuneable queues */
1530 	adapter->num_queues = ixv_num_queues;
1531 
1532 	return (0);
1533 } /* ixv_allocate_pci_resources */
1534 
1535 static void
1536 ixv_free_deferred_handlers(struct adapter *adapter)
1537 {
1538 	struct ix_queue *que = adapter->queues;
1539 	struct tx_ring *txr = adapter->tx_rings;
1540 	int i;
1541 
1542 	for (i = 0; i < adapter->num_queues; i++, que++, txr++) {
1543 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX)) {
1544 			if (txr->txr_si != NULL)
1545 				softint_disestablish(txr->txr_si);
1546 		}
1547 		if (que->que_si != NULL)
1548 			softint_disestablish(que->que_si);
1549 	}
1550 	if (adapter->txr_wq != NULL)
1551 		workqueue_destroy(adapter->txr_wq);
1552 	if (adapter->txr_wq_enqueued != NULL)
1553 		percpu_free(adapter->txr_wq_enqueued, sizeof(u_int));
1554 	if (adapter->que_wq != NULL)
1555 		workqueue_destroy(adapter->que_wq);
1556 
1557 	/* Drain the Mailbox(link) queue */
1558 	if (adapter->admin_wq != NULL) {
1559 		workqueue_destroy(adapter->admin_wq);
1560 		adapter->admin_wq = NULL;
1561 	}
1562 	if (adapter->timer_wq != NULL) {
1563 		workqueue_destroy(adapter->timer_wq);
1564 		adapter->timer_wq = NULL;
1565 	}
1566 } /* ixv_free_deferred_handlers */
1567 
1568 /************************************************************************
1569  * ixv_free_pci_resources
1570  ************************************************************************/
1571 static void
1572 ixv_free_pci_resources(struct adapter * adapter)
1573 {
1574 	struct		ix_queue *que = adapter->queues;
1575 	int		rid;
1576 
1577 	/*
1578 	 *  Release all msix queue resources:
1579 	 */
1580 	for (int i = 0; i < adapter->num_queues; i++, que++) {
1581 		if (que->res != NULL)
1582 			pci_intr_disestablish(adapter->osdep.pc,
1583 			    adapter->osdep.ihs[i]);
1584 	}
1585 
1586 
1587 	/* Clean the Mailbox interrupt last */
1588 	rid = adapter->vector;
1589 
1590 	if (adapter->osdep.ihs[rid] != NULL) {
1591 		pci_intr_disestablish(adapter->osdep.pc,
1592 		    adapter->osdep.ihs[rid]);
1593 		adapter->osdep.ihs[rid] = NULL;
1594 	}
1595 
1596 	pci_intr_release(adapter->osdep.pc, adapter->osdep.intrs,
1597 	    adapter->osdep.nintrs);
1598 
1599 	if (adapter->osdep.mem_size != 0) {
1600 		bus_space_unmap(adapter->osdep.mem_bus_space_tag,
1601 		    adapter->osdep.mem_bus_space_handle,
1602 		    adapter->osdep.mem_size);
1603 	}
1604 
1605 	return;
1606 } /* ixv_free_pci_resources */
1607 
1608 /************************************************************************
1609  * ixv_setup_interface
1610  *
1611  *   Setup networking device structure and register an interface.
1612  ************************************************************************/
1613 static int
1614 ixv_setup_interface(device_t dev, struct adapter *adapter)
1615 {
1616 	struct ethercom *ec = &adapter->osdep.ec;
1617 	struct ifnet   *ifp;
1618 	int rv;
1619 
1620 	INIT_DEBUGOUT("ixv_setup_interface: begin");
1621 
1622 	ifp = adapter->ifp = &ec->ec_if;
1623 	strlcpy(ifp->if_xname, device_xname(dev), IFNAMSIZ);
1624 	ifp->if_baudrate = IF_Gbps(10);
1625 	ifp->if_init = ixv_init;
1626 	ifp->if_stop = ixv_ifstop;
1627 	ifp->if_softc = adapter;
1628 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1629 #ifdef IXGBE_MPSAFE
1630 	ifp->if_extflags = IFEF_MPSAFE;
1631 #endif
1632 	ifp->if_ioctl = ixv_ioctl;
1633 	if (adapter->feat_en & IXGBE_FEATURE_LEGACY_TX) {
1634 #if 0
1635 		ixv_start_locked = ixgbe_legacy_start_locked;
1636 #endif
1637 	} else {
1638 		ifp->if_transmit = ixgbe_mq_start;
1639 #if 0
1640 		ixv_start_locked = ixgbe_mq_start_locked;
1641 #endif
1642 	}
1643 	ifp->if_start = ixgbe_legacy_start;
1644 	IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 2);
1645 	IFQ_SET_READY(&ifp->if_snd);
1646 
1647 	rv = if_initialize(ifp);
1648 	if (rv != 0) {
1649 		aprint_error_dev(dev, "if_initialize failed(%d)\n", rv);
1650 		return rv;
1651 	}
1652 	adapter->ipq = if_percpuq_create(&adapter->osdep.ec.ec_if);
1653 	ether_ifattach(ifp, adapter->hw.mac.addr);
1654 	aprint_normal_dev(dev, "Ethernet address %s\n",
1655 	    ether_sprintf(adapter->hw.mac.addr));
1656 	/*
1657 	 * We use per TX queue softint, so if_deferred_start_init() isn't
1658 	 * used.
1659 	 */
1660 	ether_set_ifflags_cb(ec, ixv_ifflags_cb);
1661 
1662 	adapter->max_frame_size = ifp->if_mtu + IXGBE_MTU_HDR;
1663 
1664 	/*
1665 	 * Tell the upper layer(s) we support long frames.
1666 	 */
1667 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1668 
1669 	/* Set capability flags */
1670 	ifp->if_capabilities |= IFCAP_HWCSUM
1671 			     |	IFCAP_TSOv4
1672 			     |	IFCAP_TSOv6;
1673 	ifp->if_capenable = 0;
1674 
1675 	ec->ec_capabilities |= ETHERCAP_VLAN_HWFILTER
1676 			    |  ETHERCAP_VLAN_HWTAGGING
1677 			    |  ETHERCAP_VLAN_HWCSUM
1678 			    |  ETHERCAP_JUMBO_MTU
1679 			    |  ETHERCAP_VLAN_MTU;
1680 
1681 	/* Enable the above capabilities by default */
1682 	ec->ec_capenable = ec->ec_capabilities;
1683 
1684 	/* Don't enable LRO by default */
1685 #if 0
1686 	/* NetBSD doesn't support LRO yet */
1687 	ifp->if_capabilities |= IFCAP_LRO;
1688 #endif
1689 
1690 	/*
1691 	 * Specify the media types supported by this adapter and register
1692 	 * callbacks to update media and link information
1693 	 */
1694 	ec->ec_ifmedia = &adapter->media;
1695 	ifmedia_init_with_lock(&adapter->media, IFM_IMASK, ixv_media_change,
1696 	    ixv_media_status, &adapter->core_mtx);
1697 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1698 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
1699 
1700 	if_register(ifp);
1701 
1702 	return 0;
1703 } /* ixv_setup_interface */
1704 
1705 
1706 /************************************************************************
1707  * ixv_initialize_transmit_units - Enable transmit unit.
1708  ************************************************************************/
1709 static void
1710 ixv_initialize_transmit_units(struct adapter *adapter)
1711 {
1712 	struct tx_ring	*txr = adapter->tx_rings;
1713 	struct ixgbe_hw	*hw = &adapter->hw;
1714 	int i;
1715 
1716 	for (i = 0; i < adapter->num_queues; i++, txr++) {
1717 		u64 tdba = txr->txdma.dma_paddr;
1718 		u32 txctrl, txdctl;
1719 		int j = txr->me;
1720 
1721 		/* Set WTHRESH to 8, burst writeback */
1722 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1723 		txdctl |= (8 << 16);
1724 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1725 
1726 		/* Set the HW Tx Head and Tail indices */
1727 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDH(j), 0);
1728 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDT(j), 0);
1729 
1730 		/* Set Tx Tail register */
1731 		txr->tail = IXGBE_VFTDT(j);
1732 
1733 		txr->txr_no_space = false;
1734 
1735 		/* Set Ring parameters */
1736 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1737 		    (tdba & 0x00000000ffffffffULL));
1738 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1739 		IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j),
1740 		    adapter->num_tx_desc * sizeof(struct ixgbe_legacy_tx_desc));
1741 		txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1742 		txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
1743 		IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1744 
1745 		/* Now enable */
1746 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1747 		txdctl |= IXGBE_TXDCTL_ENABLE;
1748 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1749 	}
1750 
1751 	return;
1752 } /* ixv_initialize_transmit_units */
1753 
1754 
1755 /************************************************************************
1756  * ixv_initialize_rss_mapping
1757  ************************************************************************/
1758 static void
1759 ixv_initialize_rss_mapping(struct adapter *adapter)
1760 {
1761 	struct ixgbe_hw *hw = &adapter->hw;
1762 	u32		reta = 0, mrqc, rss_key[10];
1763 	int		queue_id;
1764 	int		i, j;
1765 	u32		rss_hash_config;
1766 
1767 	/* force use default RSS key. */
1768 #ifdef __NetBSD__
1769 	rss_getkey((uint8_t *) &rss_key);
1770 #else
1771 	if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1772 		/* Fetch the configured RSS key */
1773 		rss_getkey((uint8_t *)&rss_key);
1774 	} else {
1775 		/* set up random bits */
1776 		cprng_fast(&rss_key, sizeof(rss_key));
1777 	}
1778 #endif
1779 
1780 	/* Now fill out hash function seeds */
1781 	for (i = 0; i < 10; i++)
1782 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1783 
1784 	/* Set up the redirection table */
1785 	for (i = 0, j = 0; i < 64; i++, j++) {
1786 		if (j == adapter->num_queues)
1787 			j = 0;
1788 
1789 		if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1790 			/*
1791 			 * Fetch the RSS bucket id for the given indirection
1792 			 * entry. Cap it at the number of configured buckets
1793 			 * (which is num_queues.)
1794 			 */
1795 			queue_id = rss_get_indirection_to_bucket(i);
1796 			queue_id = queue_id % adapter->num_queues;
1797 		} else
1798 			queue_id = j;
1799 
1800 		/*
1801 		 * The low 8 bits are for hash value (n+0);
1802 		 * The next 8 bits are for hash value (n+1), etc.
1803 		 */
1804 		reta >>= 8;
1805 		reta |= ((uint32_t)queue_id) << 24;
1806 		if ((i & 3) == 3) {
1807 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), reta);
1808 			reta = 0;
1809 		}
1810 	}
1811 
1812 	/* Perform hash on these packet types */
1813 	if (adapter->feat_en & IXGBE_FEATURE_RSS)
1814 		rss_hash_config = rss_gethashconfig();
1815 	else {
1816 		/*
1817 		 * Disable UDP - IP fragments aren't currently being handled
1818 		 * and so we end up with a mix of 2-tuple and 4-tuple
1819 		 * traffic.
1820 		 */
1821 		rss_hash_config = RSS_HASHTYPE_RSS_IPV4
1822 				| RSS_HASHTYPE_RSS_TCP_IPV4
1823 				| RSS_HASHTYPE_RSS_IPV6
1824 				| RSS_HASHTYPE_RSS_TCP_IPV6;
1825 	}
1826 
1827 	mrqc = IXGBE_MRQC_RSSEN;
1828 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV4)
1829 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
1830 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV4)
1831 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
1832 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6)
1833 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
1834 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6)
1835 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
1836 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6_EX)
1837 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_IPV6_EX defined, but not supported\n",
1838 		    __func__);
1839 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6_EX)
1840 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_TCP_IPV6_EX defined, but not supported\n",
1841 		    __func__);
1842 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV4)
1843 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
1844 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6)
1845 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1846 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6_EX)
1847 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_UDP_IPV6_EX defined, but not supported\n",
1848 		    __func__);
1849 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, mrqc);
1850 } /* ixv_initialize_rss_mapping */
1851 
1852 
1853 /************************************************************************
1854  * ixv_initialize_receive_units - Setup receive registers and features.
1855  ************************************************************************/
1856 static void
1857 ixv_initialize_receive_units(struct adapter *adapter)
1858 {
1859 	struct	rx_ring	*rxr = adapter->rx_rings;
1860 	struct ixgbe_hw	*hw = &adapter->hw;
1861 	struct ifnet	*ifp = adapter->ifp;
1862 	u32		bufsz, psrtype;
1863 
1864 	if (ifp->if_mtu > ETHERMTU)
1865 		bufsz = 4096 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1866 	else
1867 		bufsz = 2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1868 
1869 	psrtype = IXGBE_PSRTYPE_TCPHDR
1870 		| IXGBE_PSRTYPE_UDPHDR
1871 		| IXGBE_PSRTYPE_IPV4HDR
1872 		| IXGBE_PSRTYPE_IPV6HDR
1873 		| IXGBE_PSRTYPE_L2HDR;
1874 
1875 	if (adapter->num_queues > 1)
1876 		psrtype |= 1 << 29;
1877 
1878 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1879 
1880 	/* Tell PF our max_frame size */
1881 	if (ixgbevf_rlpml_set_vf(hw, adapter->max_frame_size) != 0) {
1882 		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");
1883 	}
1884 
1885 	for (int i = 0; i < adapter->num_queues; i++, rxr++) {
1886 		u64 rdba = rxr->rxdma.dma_paddr;
1887 		u32 reg, rxdctl;
1888 		int j = rxr->me;
1889 
1890 		/* Disable the queue */
1891 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1892 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1893 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1894 		for (int k = 0; k < 10; k++) {
1895 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1896 			    IXGBE_RXDCTL_ENABLE)
1897 				msec_delay(1);
1898 			else
1899 				break;
1900 		}
1901 		IXGBE_WRITE_BARRIER(hw);
1902 		/* Setup the Base and Length of the Rx Descriptor Ring */
1903 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1904 		    (rdba & 0x00000000ffffffffULL));
1905 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1906 		IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j),
1907 		    adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc));
1908 
1909 		/* Reset the ring indices */
1910 		IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
1911 		IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), 0);
1912 
1913 		/* Set up the SRRCTL register */
1914 		reg = IXGBE_READ_REG(hw, IXGBE_VFSRRCTL(j));
1915 		reg &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1916 		reg &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1917 		reg |= bufsz;
1918 		reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1919 		IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(j), reg);
1920 
1921 		/* Capture Rx Tail index */
1922 		rxr->tail = IXGBE_VFRDT(rxr->me);
1923 
1924 		/* Do the queue enabling last */
1925 		rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1926 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1927 		for (int k = 0; k < 10; k++) {
1928 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1929 			    IXGBE_RXDCTL_ENABLE)
1930 				break;
1931 			msec_delay(1);
1932 		}
1933 		IXGBE_WRITE_BARRIER(hw);
1934 
1935 		/* Set the Tail Pointer */
1936 #ifdef DEV_NETMAP
1937 		/*
1938 		 * In netmap mode, we must preserve the buffers made
1939 		 * available to userspace before the if_init()
1940 		 * (this is true by default on the TX side, because
1941 		 * init makes all buffers available to userspace).
1942 		 *
1943 		 * netmap_reset() and the device specific routines
1944 		 * (e.g. ixgbe_setup_receive_rings()) map these
1945 		 * buffers at the end of the NIC ring, so here we
1946 		 * must set the RDT (tail) register to make sure
1947 		 * they are not overwritten.
1948 		 *
1949 		 * In this driver the NIC ring starts at RDH = 0,
1950 		 * RDT points to the last slot available for reception (?),
1951 		 * so RDT = num_rx_desc - 1 means the whole ring is available.
1952 		 */
1953 		if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) &&
1954 		    (ifp->if_capenable & IFCAP_NETMAP)) {
1955 			struct netmap_adapter *na = NA(adapter->ifp);
1956 			struct netmap_kring *kring = na->rx_rings[i];
1957 			int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring);
1958 
1959 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), t);
1960 		} else
1961 #endif /* DEV_NETMAP */
1962 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
1963 			    adapter->num_rx_desc - 1);
1964 	}
1965 
1966 	if (adapter->hw.mac.type >= ixgbe_mac_X550_vf)
1967 		ixv_initialize_rss_mapping(adapter);
1968 } /* ixv_initialize_receive_units */
1969 
1970 /************************************************************************
1971  * ixv_sysctl_tdh_handler - Transmit Descriptor Head handler function
1972  *
1973  *   Retrieves the TDH value from the hardware
1974  ************************************************************************/
1975 static int
1976 ixv_sysctl_tdh_handler(SYSCTLFN_ARGS)
1977 {
1978 	struct sysctlnode node = *rnode;
1979 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1980 	uint32_t val;
1981 
1982 	if (!txr)
1983 		return (0);
1984 
1985 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDH(txr->me));
1986 	node.sysctl_data = &val;
1987 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1988 } /* ixv_sysctl_tdh_handler */
1989 
1990 /************************************************************************
1991  * ixgbe_sysctl_tdt_handler - Transmit Descriptor Tail handler function
1992  *
1993  *   Retrieves the TDT value from the hardware
1994  ************************************************************************/
1995 static int
1996 ixv_sysctl_tdt_handler(SYSCTLFN_ARGS)
1997 {
1998 	struct sysctlnode node = *rnode;
1999 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
2000 	uint32_t val;
2001 
2002 	if (!txr)
2003 		return (0);
2004 
2005 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDT(txr->me));
2006 	node.sysctl_data = &val;
2007 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2008 } /* ixv_sysctl_tdt_handler */
2009 
2010 /************************************************************************
2011  * ixv_sysctl_next_to_check_handler - Receive Descriptor next to check
2012  * handler function
2013  *
2014  *   Retrieves the next_to_check value
2015  ************************************************************************/
2016 static int
2017 ixv_sysctl_next_to_check_handler(SYSCTLFN_ARGS)
2018 {
2019 	struct sysctlnode node = *rnode;
2020 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2021 	uint32_t val;
2022 
2023 	if (!rxr)
2024 		return (0);
2025 
2026 	val = rxr->next_to_check;
2027 	node.sysctl_data = &val;
2028 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2029 } /* ixv_sysctl_next_to_check_handler */
2030 
2031 /************************************************************************
2032  * ixv_sysctl_rdh_handler - Receive Descriptor Head handler function
2033  *
2034  *   Retrieves the RDH value from the hardware
2035  ************************************************************************/
2036 static int
2037 ixv_sysctl_rdh_handler(SYSCTLFN_ARGS)
2038 {
2039 	struct sysctlnode node = *rnode;
2040 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2041 	uint32_t val;
2042 
2043 	if (!rxr)
2044 		return (0);
2045 
2046 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDH(rxr->me));
2047 	node.sysctl_data = &val;
2048 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2049 } /* ixv_sysctl_rdh_handler */
2050 
2051 /************************************************************************
2052  * ixv_sysctl_rdt_handler - Receive Descriptor Tail handler function
2053  *
2054  *   Retrieves the RDT value from the hardware
2055  ************************************************************************/
2056 static int
2057 ixv_sysctl_rdt_handler(SYSCTLFN_ARGS)
2058 {
2059 	struct sysctlnode node = *rnode;
2060 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2061 	uint32_t val;
2062 
2063 	if (!rxr)
2064 		return (0);
2065 
2066 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDT(rxr->me));
2067 	node.sysctl_data = &val;
2068 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2069 } /* ixv_sysctl_rdt_handler */
2070 
2071 static void
2072 ixv_setup_vlan_tagging(struct adapter *adapter)
2073 {
2074 	struct ethercom *ec = &adapter->osdep.ec;
2075 	struct ixgbe_hw *hw = &adapter->hw;
2076 	struct rx_ring	*rxr;
2077 	u32		ctrl;
2078 	int		i;
2079 	bool		hwtagging;
2080 
2081 	/* Enable HW tagging only if any vlan is attached */
2082 	hwtagging = (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING)
2083 	    && VLAN_ATTACHED(ec);
2084 
2085 	/* Enable the queues */
2086 	for (i = 0; i < adapter->num_queues; i++) {
2087 		rxr = &adapter->rx_rings[i];
2088 		ctrl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(rxr->me));
2089 		if (hwtagging)
2090 			ctrl |= IXGBE_RXDCTL_VME;
2091 		else
2092 			ctrl &= ~IXGBE_RXDCTL_VME;
2093 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(rxr->me), ctrl);
2094 		/*
2095 		 * Let Rx path know that it needs to store VLAN tag
2096 		 * as part of extra mbuf info.
2097 		 */
2098 		rxr->vtag_strip = hwtagging ? TRUE : FALSE;
2099 	}
2100 } /* ixv_setup_vlan_tagging */
2101 
2102 /************************************************************************
2103  * ixv_setup_vlan_support
2104  ************************************************************************/
2105 static int
2106 ixv_setup_vlan_support(struct adapter *adapter)
2107 {
2108 	struct ethercom *ec = &adapter->osdep.ec;
2109 	struct ixgbe_hw *hw = &adapter->hw;
2110 	u32		vid, vfta, retry;
2111 	struct vlanid_list *vlanidp;
2112 	int rv, error = 0;
2113 
2114 	/*
2115 	 *  This function is called from both if_init and ifflags_cb()
2116 	 * on NetBSD.
2117 	 */
2118 
2119 	/*
2120 	 * Part 1:
2121 	 * Setup VLAN HW tagging
2122 	 */
2123 	ixv_setup_vlan_tagging(adapter);
2124 
2125 	if (!VLAN_ATTACHED(ec))
2126 		return 0;
2127 
2128 	/*
2129 	 * Part 2:
2130 	 * Setup VLAN HW filter
2131 	 */
2132 	/* Cleanup shadow_vfta */
2133 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++)
2134 		adapter->shadow_vfta[i] = 0;
2135 	/* Generate shadow_vfta from ec_vids */
2136 	ETHER_LOCK(ec);
2137 	SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
2138 		uint32_t idx;
2139 
2140 		idx = vlanidp->vid / 32;
2141 		KASSERT(idx < IXGBE_VFTA_SIZE);
2142 		adapter->shadow_vfta[idx] |= (u32)1 << (vlanidp->vid % 32);
2143 	}
2144 	ETHER_UNLOCK(ec);
2145 
2146 	/*
2147 	 * A soft reset zero's out the VFTA, so
2148 	 * we need to repopulate it now.
2149 	 */
2150 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++) {
2151 		if (adapter->shadow_vfta[i] == 0)
2152 			continue;
2153 		vfta = adapter->shadow_vfta[i];
2154 		/*
2155 		 * Reconstruct the vlan id's
2156 		 * based on the bits set in each
2157 		 * of the array ints.
2158 		 */
2159 		for (int j = 0; j < 32; j++) {
2160 			retry = 0;
2161 			if ((vfta & ((u32)1 << j)) == 0)
2162 				continue;
2163 			vid = (i * 32) + j;
2164 
2165 			/* Call the shared code mailbox routine */
2166 			while ((rv = hw->mac.ops.set_vfta(hw, vid, 0, TRUE,
2167 			    FALSE)) != 0) {
2168 				if (++retry > 5) {
2169 					device_printf(adapter->dev,
2170 					    "%s: max retry exceeded\n",
2171 						__func__);
2172 					break;
2173 				}
2174 			}
2175 			if (rv != 0) {
2176 				device_printf(adapter->dev,
2177 				    "failed to set vlan %d\n", vid);
2178 				error = EACCES;
2179 			}
2180 		}
2181 	}
2182 	return error;
2183 } /* ixv_setup_vlan_support */
2184 
2185 static int
2186 ixv_vlan_cb(struct ethercom *ec, uint16_t vid, bool set)
2187 {
2188 	struct ifnet *ifp = &ec->ec_if;
2189 	struct adapter *adapter = ifp->if_softc;
2190 	int rv;
2191 
2192 	if (set)
2193 		rv = ixv_register_vlan(adapter, vid);
2194 	else
2195 		rv = ixv_unregister_vlan(adapter, vid);
2196 
2197 	if (rv != 0)
2198 		return rv;
2199 
2200 	/*
2201 	 * Control VLAN HW tagging when ec_nvlan is changed from 1 to 0
2202 	 * or 0 to 1.
2203 	 */
2204 	if ((set && (ec->ec_nvlans == 1)) || (!set && (ec->ec_nvlans == 0)))
2205 		ixv_setup_vlan_tagging(adapter);
2206 
2207 	return rv;
2208 }
2209 
2210 /************************************************************************
2211  * ixv_register_vlan
2212  *
2213  *   Run via a vlan config EVENT, it enables us to use the
2214  *   HW Filter table since we can get the vlan id. This just
2215  *   creates the entry in the soft version of the VFTA, init
2216  *   will repopulate the real table.
2217  ************************************************************************/
2218 static int
2219 ixv_register_vlan(struct adapter *adapter, u16 vtag)
2220 {
2221 	struct ixgbe_hw *hw = &adapter->hw;
2222 	u16		index, bit;
2223 	int error;
2224 
2225 	if ((vtag == 0) || (vtag > 4095)) /* Invalid */
2226 		return EINVAL;
2227 	IXGBE_CORE_LOCK(adapter);
2228 	index = (vtag >> 5) & 0x7F;
2229 	bit = vtag & 0x1F;
2230 	adapter->shadow_vfta[index] |= ((u32)1 << bit);
2231 	error = hw->mac.ops.set_vfta(hw, vtag, 0, true, false);
2232 	IXGBE_CORE_UNLOCK(adapter);
2233 
2234 	if (error != 0) {
2235 		device_printf(adapter->dev, "failed to register vlan %hu\n",
2236 		    vtag);
2237 		error = EACCES;
2238 	}
2239 	return error;
2240 } /* ixv_register_vlan */
2241 
2242 /************************************************************************
2243  * ixv_unregister_vlan
2244  *
2245  *   Run via a vlan unconfig EVENT, remove our entry
2246  *   in the soft vfta.
2247  ************************************************************************/
2248 static int
2249 ixv_unregister_vlan(struct adapter *adapter, u16 vtag)
2250 {
2251 	struct ixgbe_hw *hw = &adapter->hw;
2252 	u16		index, bit;
2253 	int		error;
2254 
2255 	if ((vtag == 0) || (vtag > 4095))  /* Invalid */
2256 		return EINVAL;
2257 
2258 	IXGBE_CORE_LOCK(adapter);
2259 	index = (vtag >> 5) & 0x7F;
2260 	bit = vtag & 0x1F;
2261 	adapter->shadow_vfta[index] &= ~((u32)1 << bit);
2262 	error = hw->mac.ops.set_vfta(hw, vtag, 0, false, false);
2263 	IXGBE_CORE_UNLOCK(adapter);
2264 
2265 	if (error != 0) {
2266 		device_printf(adapter->dev, "failed to unregister vlan %hu\n",
2267 		    vtag);
2268 		error = EIO;
2269 	}
2270 	return error;
2271 } /* ixv_unregister_vlan */
2272 
2273 /************************************************************************
2274  * ixv_enable_intr
2275  ************************************************************************/
2276 static void
2277 ixv_enable_intr(struct adapter *adapter)
2278 {
2279 	struct ixgbe_hw *hw = &adapter->hw;
2280 	struct ix_queue *que = adapter->queues;
2281 	u32		mask;
2282 	int i;
2283 
2284 	/* For VTEIAC */
2285 	mask = (1 << adapter->vector);
2286 	for (i = 0; i < adapter->num_queues; i++, que++)
2287 		mask |= (1 << que->msix);
2288 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
2289 
2290 	/* For VTEIMS */
2291 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
2292 	que = adapter->queues;
2293 	for (i = 0; i < adapter->num_queues; i++, que++)
2294 		ixv_enable_queue(adapter, que->msix);
2295 
2296 	IXGBE_WRITE_FLUSH(hw);
2297 } /* ixv_enable_intr */
2298 
2299 /************************************************************************
2300  * ixv_disable_intr
2301  ************************************************************************/
2302 static void
2303 ixv_disable_intr(struct adapter *adapter)
2304 {
2305 	struct ix_queue	*que = adapter->queues;
2306 
2307 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIAC, 0);
2308 
2309 	/* disable interrupts other than queues */
2310 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIMC, adapter->vector);
2311 
2312 	for (int i = 0; i < adapter->num_queues; i++, que++)
2313 		ixv_disable_queue(adapter, que->msix);
2314 
2315 	IXGBE_WRITE_FLUSH(&adapter->hw);
2316 } /* ixv_disable_intr */
2317 
2318 /************************************************************************
2319  * ixv_set_ivar
2320  *
2321  *   Setup the correct IVAR register for a particular MSI-X interrupt
2322  *    - entry is the register array entry
2323  *    - vector is the MSI-X vector for this queue
2324  *    - type is RX/TX/MISC
2325  ************************************************************************/
2326 static void
2327 ixv_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type)
2328 {
2329 	struct ixgbe_hw *hw = &adapter->hw;
2330 	u32		ivar, index;
2331 
2332 	vector |= IXGBE_IVAR_ALLOC_VAL;
2333 
2334 	if (type == -1) { /* MISC IVAR */
2335 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
2336 		ivar &= ~0xFF;
2337 		ivar |= vector;
2338 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
2339 	} else {	  /* RX/TX IVARS */
2340 		index = (16 * (entry & 1)) + (8 * type);
2341 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(entry >> 1));
2342 		ivar &= ~(0xffUL << index);
2343 		ivar |= ((u32)vector << index);
2344 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(entry >> 1), ivar);
2345 	}
2346 } /* ixv_set_ivar */
2347 
2348 /************************************************************************
2349  * ixv_configure_ivars
2350  ************************************************************************/
2351 static void
2352 ixv_configure_ivars(struct adapter *adapter)
2353 {
2354 	struct ix_queue *que = adapter->queues;
2355 
2356 	/* XXX We should sync EITR value calculation with ixgbe.c? */
2357 
2358 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2359 		/* First the RX queue entry */
2360 		ixv_set_ivar(adapter, i, que->msix, 0);
2361 		/* ... and the TX */
2362 		ixv_set_ivar(adapter, i, que->msix, 1);
2363 		/* Set an initial value in EITR */
2364 		ixv_eitr_write(adapter, que->msix, IXGBE_EITR_DEFAULT);
2365 	}
2366 
2367 	/* For the mailbox interrupt */
2368 	ixv_set_ivar(adapter, 1, adapter->vector, -1);
2369 } /* ixv_configure_ivars */
2370 
2371 
2372 /************************************************************************
2373  * ixv_save_stats
2374  *
2375  *   The VF stats registers never have a truly virgin
2376  *   starting point, so this routine tries to make an
2377  *   artificial one, marking ground zero on attach as
2378  *   it were.
2379  ************************************************************************/
2380 static void
2381 ixv_save_stats(struct adapter *adapter)
2382 {
2383 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2384 
2385 	if (stats->vfgprc.ev_count || stats->vfgptc.ev_count) {
2386 		stats->saved_reset_vfgprc +=
2387 		    stats->vfgprc.ev_count - stats->base_vfgprc;
2388 		stats->saved_reset_vfgptc +=
2389 		    stats->vfgptc.ev_count - stats->base_vfgptc;
2390 		stats->saved_reset_vfgorc +=
2391 		    stats->vfgorc.ev_count - stats->base_vfgorc;
2392 		stats->saved_reset_vfgotc +=
2393 		    stats->vfgotc.ev_count - stats->base_vfgotc;
2394 		stats->saved_reset_vfmprc +=
2395 		    stats->vfmprc.ev_count - stats->base_vfmprc;
2396 	}
2397 } /* ixv_save_stats */
2398 
2399 /************************************************************************
2400  * ixv_init_stats
2401  ************************************************************************/
2402 static void
2403 ixv_init_stats(struct adapter *adapter)
2404 {
2405 	struct ixgbe_hw *hw = &adapter->hw;
2406 
2407 	adapter->stats.vf.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2408 	adapter->stats.vf.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2409 	adapter->stats.vf.last_vfgorc |=
2410 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2411 
2412 	adapter->stats.vf.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2413 	adapter->stats.vf.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2414 	adapter->stats.vf.last_vfgotc |=
2415 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2416 
2417 	adapter->stats.vf.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2418 
2419 	adapter->stats.vf.base_vfgprc = adapter->stats.vf.last_vfgprc;
2420 	adapter->stats.vf.base_vfgorc = adapter->stats.vf.last_vfgorc;
2421 	adapter->stats.vf.base_vfgptc = adapter->stats.vf.last_vfgptc;
2422 	adapter->stats.vf.base_vfgotc = adapter->stats.vf.last_vfgotc;
2423 	adapter->stats.vf.base_vfmprc = adapter->stats.vf.last_vfmprc;
2424 } /* ixv_init_stats */
2425 
2426 #define UPDATE_STAT_32(reg, last, count)		\
2427 {							\
2428 	u32 current = IXGBE_READ_REG(hw, (reg));	\
2429 	if (current < (last))				\
2430 		count.ev_count += 0x100000000LL;	\
2431 	(last) = current;				\
2432 	count.ev_count &= 0xFFFFFFFF00000000LL;		\
2433 	count.ev_count |= current;			\
2434 }
2435 
2436 #define UPDATE_STAT_36(lsb, msb, last, count)		\
2437 {							\
2438 	u64 cur_lsb = IXGBE_READ_REG(hw, (lsb));	\
2439 	u64 cur_msb = IXGBE_READ_REG(hw, (msb));	\
2440 	u64 current = ((cur_msb << 32) | cur_lsb);	\
2441 	if (current < (last))				\
2442 		count.ev_count += 0x1000000000LL;	\
2443 	(last) = current;				\
2444 	count.ev_count &= 0xFFFFFFF000000000LL;		\
2445 	count.ev_count |= current;			\
2446 }
2447 
2448 /************************************************************************
2449  * ixv_update_stats - Update the board statistics counters.
2450  ************************************************************************/
2451 void
2452 ixv_update_stats(struct adapter *adapter)
2453 {
2454 	struct ixgbe_hw *hw = &adapter->hw;
2455 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2456 
2457 	UPDATE_STAT_32(IXGBE_VFGPRC, stats->last_vfgprc, stats->vfgprc);
2458 	UPDATE_STAT_32(IXGBE_VFGPTC, stats->last_vfgptc, stats->vfgptc);
2459 	UPDATE_STAT_36(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB, stats->last_vfgorc,
2460 	    stats->vfgorc);
2461 	UPDATE_STAT_36(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB, stats->last_vfgotc,
2462 	    stats->vfgotc);
2463 	UPDATE_STAT_32(IXGBE_VFMPRC, stats->last_vfmprc, stats->vfmprc);
2464 
2465 	/* VF doesn't count errors by hardware */
2466 
2467 } /* ixv_update_stats */
2468 
2469 /************************************************************************
2470  * ixv_sysctl_interrupt_rate_handler
2471  ************************************************************************/
2472 static int
2473 ixv_sysctl_interrupt_rate_handler(SYSCTLFN_ARGS)
2474 {
2475 	struct sysctlnode node = *rnode;
2476 	struct ix_queue *que = (struct ix_queue *)node.sysctl_data;
2477 	struct adapter	*adapter = que->adapter;
2478 	uint32_t reg, usec, rate;
2479 	int error;
2480 
2481 	if (que == NULL)
2482 		return 0;
2483 	reg = IXGBE_READ_REG(&que->adapter->hw, IXGBE_VTEITR(que->msix));
2484 	usec = ((reg & 0x0FF8) >> 3);
2485 	if (usec > 0)
2486 		rate = 500000 / usec;
2487 	else
2488 		rate = 0;
2489 	node.sysctl_data = &rate;
2490 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2491 	if (error || newp == NULL)
2492 		return error;
2493 	reg &= ~0xfff; /* default, no limitation */
2494 	if (rate > 0 && rate < 500000) {
2495 		if (rate < 1000)
2496 			rate = 1000;
2497 		reg |= ((4000000 / rate) & 0xff8);
2498 		/*
2499 		 * When RSC is used, ITR interval must be larger than
2500 		 * RSC_DELAY. Currently, we use 2us for RSC_DELAY.
2501 		 * The minimum value is always greater than 2us on 100M
2502 		 * (and 10M?(not documented)), but it's not on 1G and higher.
2503 		 */
2504 		if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
2505 		    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
2506 			if ((adapter->num_queues > 1)
2507 			    && (reg < IXGBE_MIN_RSC_EITR_10G1G))
2508 				return EINVAL;
2509 		}
2510 		ixv_max_interrupt_rate = rate;
2511 	} else
2512 		ixv_max_interrupt_rate = 0;
2513 	ixv_eitr_write(adapter, que->msix, reg);
2514 
2515 	return (0);
2516 } /* ixv_sysctl_interrupt_rate_handler */
2517 
2518 const struct sysctlnode *
2519 ixv_sysctl_instance(struct adapter *adapter)
2520 {
2521 	const char *dvname;
2522 	struct sysctllog **log;
2523 	int rc;
2524 	const struct sysctlnode *rnode;
2525 
2526 	log = &adapter->sysctllog;
2527 	dvname = device_xname(adapter->dev);
2528 
2529 	if ((rc = sysctl_createv(log, 0, NULL, &rnode,
2530 	    0, CTLTYPE_NODE, dvname,
2531 	    SYSCTL_DESCR("ixv information and settings"),
2532 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0)
2533 		goto err;
2534 
2535 	return rnode;
2536 err:
2537 	device_printf(adapter->dev,
2538 	    "%s: sysctl_createv failed, rc = %d\n", __func__, rc);
2539 	return NULL;
2540 }
2541 
2542 static void
2543 ixv_add_device_sysctls(struct adapter *adapter)
2544 {
2545 	struct sysctllog **log;
2546 	const struct sysctlnode *rnode, *cnode;
2547 	device_t dev;
2548 
2549 	dev = adapter->dev;
2550 	log = &adapter->sysctllog;
2551 
2552 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2553 		aprint_error_dev(dev, "could not create sysctl root\n");
2554 		return;
2555 	}
2556 
2557 	if (sysctl_createv(log, 0, &rnode, &cnode,
2558 	    CTLFLAG_READWRITE, CTLTYPE_INT, "debug",
2559 	    SYSCTL_DESCR("Debug Info"),
2560 	    ixv_sysctl_debug, 0, (void *)adapter, 0, CTL_CREATE, CTL_EOL) != 0)
2561 		aprint_error_dev(dev, "could not create sysctl\n");
2562 
2563 	if (sysctl_createv(log, 0, &rnode, &cnode,
2564 	    CTLFLAG_READONLY, CTLTYPE_INT, "num_jcl_per_queue",
2565 	    SYSCTL_DESCR("Number of jumbo buffers per queue"),
2566 	    NULL, 0, &adapter->num_jcl, 0, CTL_CREATE,
2567 	    CTL_EOL) != 0)
2568 		aprint_error_dev(dev, "could not create sysctl\n");
2569 
2570 	if (sysctl_createv(log, 0, &rnode, &cnode,
2571 	    CTLFLAG_READWRITE, CTLTYPE_BOOL, "enable_aim",
2572 	    SYSCTL_DESCR("Interrupt Moderation"),
2573 	    NULL, 0, &adapter->enable_aim, 0, CTL_CREATE, CTL_EOL) != 0)
2574 		aprint_error_dev(dev, "could not create sysctl\n");
2575 
2576 	if (sysctl_createv(log, 0, &rnode, &cnode,
2577 	    CTLFLAG_READWRITE, CTLTYPE_BOOL, "txrx_workqueue",
2578 	    SYSCTL_DESCR("Use workqueue for packet processing"),
2579 	    NULL, 0, &adapter->txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL)
2580 	    != 0)
2581 		aprint_error_dev(dev, "could not create sysctl\n");
2582 }
2583 
2584 /************************************************************************
2585  * ixv_add_stats_sysctls - Add statistic sysctls for the VF.
2586  ************************************************************************/
2587 static void
2588 ixv_add_stats_sysctls(struct adapter *adapter)
2589 {
2590 	device_t		dev = adapter->dev;
2591 	struct tx_ring		*txr = adapter->tx_rings;
2592 	struct rx_ring		*rxr = adapter->rx_rings;
2593 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2594 	struct ixgbe_hw *hw = &adapter->hw;
2595 	const struct sysctlnode *rnode, *cnode;
2596 	struct sysctllog **log = &adapter->sysctllog;
2597 	const char *xname = device_xname(dev);
2598 
2599 	/* Driver Statistics */
2600 	evcnt_attach_dynamic(&adapter->efbig_tx_dma_setup, EVCNT_TYPE_MISC,
2601 	    NULL, xname, "Driver tx dma soft fail EFBIG");
2602 	evcnt_attach_dynamic(&adapter->mbuf_defrag_failed, EVCNT_TYPE_MISC,
2603 	    NULL, xname, "m_defrag() failed");
2604 	evcnt_attach_dynamic(&adapter->efbig2_tx_dma_setup, EVCNT_TYPE_MISC,
2605 	    NULL, xname, "Driver tx dma hard fail EFBIG");
2606 	evcnt_attach_dynamic(&adapter->einval_tx_dma_setup, EVCNT_TYPE_MISC,
2607 	    NULL, xname, "Driver tx dma hard fail EINVAL");
2608 	evcnt_attach_dynamic(&adapter->other_tx_dma_setup, EVCNT_TYPE_MISC,
2609 	    NULL, xname, "Driver tx dma hard fail other");
2610 	evcnt_attach_dynamic(&adapter->eagain_tx_dma_setup, EVCNT_TYPE_MISC,
2611 	    NULL, xname, "Driver tx dma soft fail EAGAIN");
2612 	evcnt_attach_dynamic(&adapter->enomem_tx_dma_setup, EVCNT_TYPE_MISC,
2613 	    NULL, xname, "Driver tx dma soft fail ENOMEM");
2614 	evcnt_attach_dynamic(&adapter->watchdog_events, EVCNT_TYPE_MISC,
2615 	    NULL, xname, "Watchdog timeouts");
2616 	evcnt_attach_dynamic(&adapter->tso_err, EVCNT_TYPE_MISC,
2617 	    NULL, xname, "TSO errors");
2618 	evcnt_attach_dynamic(&adapter->admin_irqev, EVCNT_TYPE_INTR,
2619 	    NULL, xname, "Admin MSI-X IRQ Handled");
2620 	evcnt_attach_dynamic(&adapter->link_workev, EVCNT_TYPE_INTR,
2621 	    NULL, xname, "Admin event");
2622 
2623 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2624 		snprintf(adapter->queues[i].evnamebuf,
2625 		    sizeof(adapter->queues[i].evnamebuf), "%s q%d",
2626 		    xname, i);
2627 		snprintf(adapter->queues[i].namebuf,
2628 		    sizeof(adapter->queues[i].namebuf), "q%d", i);
2629 
2630 		if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2631 			aprint_error_dev(dev, "could not create sysctl root\n");
2632 			break;
2633 		}
2634 
2635 		if (sysctl_createv(log, 0, &rnode, &rnode,
2636 		    0, CTLTYPE_NODE,
2637 		    adapter->queues[i].namebuf, SYSCTL_DESCR("Queue Name"),
2638 		    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL) != 0)
2639 			break;
2640 
2641 		if (sysctl_createv(log, 0, &rnode, &cnode,
2642 		    CTLFLAG_READWRITE, CTLTYPE_INT,
2643 		    "interrupt_rate", SYSCTL_DESCR("Interrupt Rate"),
2644 		    ixv_sysctl_interrupt_rate_handler, 0,
2645 		    (void *)&adapter->queues[i], 0, CTL_CREATE, CTL_EOL) != 0)
2646 			break;
2647 
2648 		if (sysctl_createv(log, 0, &rnode, &cnode,
2649 		    CTLFLAG_READONLY, CTLTYPE_INT,
2650 		    "txd_head", SYSCTL_DESCR("Transmit Descriptor Head"),
2651 		    ixv_sysctl_tdh_handler, 0, (void *)txr,
2652 		    0, CTL_CREATE, CTL_EOL) != 0)
2653 			break;
2654 
2655 		if (sysctl_createv(log, 0, &rnode, &cnode,
2656 		    CTLFLAG_READONLY, CTLTYPE_INT,
2657 		    "txd_tail", SYSCTL_DESCR("Transmit Descriptor Tail"),
2658 		    ixv_sysctl_tdt_handler, 0, (void *)txr,
2659 		    0, CTL_CREATE, CTL_EOL) != 0)
2660 			break;
2661 
2662 		evcnt_attach_dynamic(&adapter->queues[i].irqs, EVCNT_TYPE_INTR,
2663 		    NULL, adapter->queues[i].evnamebuf, "IRQs on queue");
2664 		evcnt_attach_dynamic(&adapter->queues[i].handleq,
2665 		    EVCNT_TYPE_MISC, NULL, adapter->queues[i].evnamebuf,
2666 		    "Handled queue in softint");
2667 		evcnt_attach_dynamic(&adapter->queues[i].req, EVCNT_TYPE_MISC,
2668 		    NULL, adapter->queues[i].evnamebuf, "Requeued in softint");
2669 		evcnt_attach_dynamic(&txr->tso_tx, EVCNT_TYPE_MISC,
2670 		    NULL, adapter->queues[i].evnamebuf, "TSO");
2671 		evcnt_attach_dynamic(&txr->no_desc_avail, EVCNT_TYPE_MISC,
2672 		    NULL, adapter->queues[i].evnamebuf,
2673 		    "TX Queue No Descriptor Available");
2674 		evcnt_attach_dynamic(&txr->total_packets, EVCNT_TYPE_MISC,
2675 		    NULL, adapter->queues[i].evnamebuf,
2676 		    "Queue Packets Transmitted");
2677 #ifndef IXGBE_LEGACY_TX
2678 		evcnt_attach_dynamic(&txr->pcq_drops, EVCNT_TYPE_MISC,
2679 		    NULL, adapter->queues[i].evnamebuf,
2680 		    "Packets dropped in pcq");
2681 #endif
2682 
2683 #ifdef LRO
2684 		struct lro_ctrl *lro = &rxr->lro;
2685 #endif /* LRO */
2686 
2687 		if (sysctl_createv(log, 0, &rnode, &cnode,
2688 		    CTLFLAG_READONLY, CTLTYPE_INT, "rxd_nxck",
2689 		    SYSCTL_DESCR("Receive Descriptor next to check"),
2690 		    ixv_sysctl_next_to_check_handler, 0, (void *)rxr, 0,
2691 		    CTL_CREATE, CTL_EOL) != 0)
2692 			break;
2693 
2694 		if (sysctl_createv(log, 0, &rnode, &cnode,
2695 		    CTLFLAG_READONLY, CTLTYPE_INT, "rxd_head",
2696 		    SYSCTL_DESCR("Receive Descriptor Head"),
2697 		    ixv_sysctl_rdh_handler, 0, (void *)rxr, 0,
2698 		    CTL_CREATE, CTL_EOL) != 0)
2699 			break;
2700 
2701 		if (sysctl_createv(log, 0, &rnode, &cnode,
2702 		    CTLFLAG_READONLY, CTLTYPE_INT, "rxd_tail",
2703 		    SYSCTL_DESCR("Receive Descriptor Tail"),
2704 		    ixv_sysctl_rdt_handler, 0, (void *)rxr, 0,
2705 		    CTL_CREATE, CTL_EOL) != 0)
2706 			break;
2707 
2708 		evcnt_attach_dynamic(&rxr->rx_packets, EVCNT_TYPE_MISC,
2709 		    NULL, adapter->queues[i].evnamebuf,
2710 		    "Queue Packets Received");
2711 		evcnt_attach_dynamic(&rxr->rx_bytes, EVCNT_TYPE_MISC,
2712 		    NULL, adapter->queues[i].evnamebuf,
2713 		    "Queue Bytes Received");
2714 		evcnt_attach_dynamic(&rxr->rx_copies, EVCNT_TYPE_MISC,
2715 		    NULL, adapter->queues[i].evnamebuf, "Copied RX Frames");
2716 		evcnt_attach_dynamic(&rxr->no_jmbuf, EVCNT_TYPE_MISC,
2717 		    NULL, adapter->queues[i].evnamebuf, "Rx no jumbo mbuf");
2718 		evcnt_attach_dynamic(&rxr->rx_discarded, EVCNT_TYPE_MISC,
2719 		    NULL, adapter->queues[i].evnamebuf, "Rx discarded");
2720 #ifdef LRO
2721 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_queued",
2722 				CTLFLAG_RD, &lro->lro_queued, 0,
2723 				"LRO Queued");
2724 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_flushed",
2725 				CTLFLAG_RD, &lro->lro_flushed, 0,
2726 				"LRO Flushed");
2727 #endif /* LRO */
2728 	}
2729 
2730 	/* MAC stats get their own sub node */
2731 
2732 	snprintf(stats->namebuf,
2733 	    sizeof(stats->namebuf), "%s MAC Statistics", xname);
2734 
2735 	evcnt_attach_dynamic(&stats->ipcs, EVCNT_TYPE_MISC, NULL,
2736 	    stats->namebuf, "rx csum offload - IP");
2737 	evcnt_attach_dynamic(&stats->l4cs, EVCNT_TYPE_MISC, NULL,
2738 	    stats->namebuf, "rx csum offload - L4");
2739 	evcnt_attach_dynamic(&stats->ipcs_bad, EVCNT_TYPE_MISC, NULL,
2740 	    stats->namebuf, "rx csum offload - IP bad");
2741 	evcnt_attach_dynamic(&stats->l4cs_bad, EVCNT_TYPE_MISC, NULL,
2742 	    stats->namebuf, "rx csum offload - L4 bad");
2743 
2744 	/* Packet Reception Stats */
2745 	evcnt_attach_dynamic(&stats->vfgprc, EVCNT_TYPE_MISC, NULL,
2746 	    xname, "Good Packets Received");
2747 	evcnt_attach_dynamic(&stats->vfgorc, EVCNT_TYPE_MISC, NULL,
2748 	    xname, "Good Octets Received");
2749 	evcnt_attach_dynamic(&stats->vfmprc, EVCNT_TYPE_MISC, NULL,
2750 	    xname, "Multicast Packets Received");
2751 	evcnt_attach_dynamic(&stats->vfgptc, EVCNT_TYPE_MISC, NULL,
2752 	    xname, "Good Packets Transmitted");
2753 	evcnt_attach_dynamic(&stats->vfgotc, EVCNT_TYPE_MISC, NULL,
2754 	    xname, "Good Octets Transmitted");
2755 
2756 	/* Mailbox Stats */
2757 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_tx, EVCNT_TYPE_MISC, NULL,
2758 	    xname, "message TXs");
2759 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_rx, EVCNT_TYPE_MISC, NULL,
2760 	    xname, "message RXs");
2761 	evcnt_attach_dynamic(&hw->mbx.stats.acks, EVCNT_TYPE_MISC, NULL,
2762 	    xname, "ACKs");
2763 	evcnt_attach_dynamic(&hw->mbx.stats.reqs, EVCNT_TYPE_MISC, NULL,
2764 	    xname, "REQs");
2765 	evcnt_attach_dynamic(&hw->mbx.stats.rsts, EVCNT_TYPE_MISC, NULL,
2766 	    xname, "RSTs");
2767 
2768 } /* ixv_add_stats_sysctls */
2769 
2770 static void
2771 ixv_clear_evcnt(struct adapter *adapter)
2772 {
2773 	struct tx_ring		*txr = adapter->tx_rings;
2774 	struct rx_ring		*rxr = adapter->rx_rings;
2775 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2776 	struct ixgbe_hw *hw = &adapter->hw;
2777 	int i;
2778 
2779 	/* Driver Statistics */
2780 	adapter->efbig_tx_dma_setup.ev_count = 0;
2781 	adapter->mbuf_defrag_failed.ev_count = 0;
2782 	adapter->efbig2_tx_dma_setup.ev_count = 0;
2783 	adapter->einval_tx_dma_setup.ev_count = 0;
2784 	adapter->other_tx_dma_setup.ev_count = 0;
2785 	adapter->eagain_tx_dma_setup.ev_count = 0;
2786 	adapter->enomem_tx_dma_setup.ev_count = 0;
2787 	adapter->watchdog_events.ev_count = 0;
2788 	adapter->tso_err.ev_count = 0;
2789 	adapter->admin_irqev.ev_count = 0;
2790 	adapter->link_workev.ev_count = 0;
2791 
2792 	for (i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2793 		adapter->queues[i].irqs.ev_count = 0;
2794 		adapter->queues[i].handleq.ev_count = 0;
2795 		adapter->queues[i].req.ev_count = 0;
2796 		txr->tso_tx.ev_count = 0;
2797 		txr->no_desc_avail.ev_count = 0;
2798 		txr->total_packets.ev_count = 0;
2799 #ifndef IXGBE_LEGACY_TX
2800 		txr->pcq_drops.ev_count = 0;
2801 #endif
2802 		txr->q_efbig_tx_dma_setup = 0;
2803 		txr->q_mbuf_defrag_failed = 0;
2804 		txr->q_efbig2_tx_dma_setup = 0;
2805 		txr->q_einval_tx_dma_setup = 0;
2806 		txr->q_other_tx_dma_setup = 0;
2807 		txr->q_eagain_tx_dma_setup = 0;
2808 		txr->q_enomem_tx_dma_setup = 0;
2809 		txr->q_tso_err = 0;
2810 
2811 		rxr->rx_packets.ev_count = 0;
2812 		rxr->rx_bytes.ev_count = 0;
2813 		rxr->rx_copies.ev_count = 0;
2814 		rxr->no_jmbuf.ev_count = 0;
2815 		rxr->rx_discarded.ev_count = 0;
2816 	}
2817 
2818 	/* MAC stats get their own sub node */
2819 
2820 	stats->ipcs.ev_count = 0;
2821 	stats->l4cs.ev_count = 0;
2822 	stats->ipcs_bad.ev_count = 0;
2823 	stats->l4cs_bad.ev_count = 0;
2824 
2825 	/* Packet Reception Stats */
2826 	stats->vfgprc.ev_count = 0;
2827 	stats->vfgorc.ev_count = 0;
2828 	stats->vfmprc.ev_count = 0;
2829 	stats->vfgptc.ev_count = 0;
2830 	stats->vfgotc.ev_count = 0;
2831 
2832 	/* Mailbox Stats */
2833 	hw->mbx.stats.msgs_tx.ev_count = 0;
2834 	hw->mbx.stats.msgs_rx.ev_count = 0;
2835 	hw->mbx.stats.acks.ev_count = 0;
2836 	hw->mbx.stats.reqs.ev_count = 0;
2837 	hw->mbx.stats.rsts.ev_count = 0;
2838 
2839 } /* ixv_clear_evcnt */
2840 
2841 /************************************************************************
2842  * ixv_set_sysctl_value
2843  ************************************************************************/
2844 static void
2845 ixv_set_sysctl_value(struct adapter *adapter, const char *name,
2846 	const char *description, int *limit, int value)
2847 {
2848 	device_t dev =	adapter->dev;
2849 	struct sysctllog **log;
2850 	const struct sysctlnode *rnode, *cnode;
2851 
2852 	log = &adapter->sysctllog;
2853 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2854 		aprint_error_dev(dev, "could not create sysctl root\n");
2855 		return;
2856 	}
2857 	if (sysctl_createv(log, 0, &rnode, &cnode,
2858 	    CTLFLAG_READWRITE, CTLTYPE_INT,
2859 	    name, SYSCTL_DESCR(description),
2860 	    NULL, 0, limit, 0, CTL_CREATE, CTL_EOL) != 0)
2861 		aprint_error_dev(dev, "could not create sysctl\n");
2862 	*limit = value;
2863 } /* ixv_set_sysctl_value */
2864 
2865 /************************************************************************
2866  * ixv_print_debug_info
2867  *
2868  *   Called only when em_display_debug_stats is enabled.
2869  *   Provides a way to take a look at important statistics
2870  *   maintained by the driver and hardware.
2871  ************************************************************************/
2872 static void
2873 ixv_print_debug_info(struct adapter *adapter)
2874 {
2875 	device_t	dev = adapter->dev;
2876 	struct ix_queue *que = adapter->queues;
2877 	struct rx_ring	*rxr;
2878 	struct tx_ring	*txr;
2879 #ifdef LRO
2880 	struct lro_ctrl *lro;
2881 #endif /* LRO */
2882 
2883 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2884 		txr = que->txr;
2885 		rxr = que->rxr;
2886 #ifdef LRO
2887 		lro = &rxr->lro;
2888 #endif /* LRO */
2889 		device_printf(dev, "QUE(%d) IRQs Handled: %lu\n",
2890 		    que->msix, (long)que->irqs.ev_count);
2891 		device_printf(dev, "RX(%d) Packets Received: %lld\n",
2892 		    rxr->me, (long long)rxr->rx_packets.ev_count);
2893 		device_printf(dev, "RX(%d) Bytes Received: %lu\n",
2894 		    rxr->me, (long)rxr->rx_bytes.ev_count);
2895 #ifdef LRO
2896 		device_printf(dev, "RX(%d) LRO Queued= %ju\n",
2897 		    rxr->me, (uintmax_t)lro->lro_queued);
2898 		device_printf(dev, "RX(%d) LRO Flushed= %ju\n",
2899 		    rxr->me, (uintmax_t)lro->lro_flushed);
2900 #endif /* LRO */
2901 		device_printf(dev, "TX(%d) Packets Sent: %lu\n",
2902 		    txr->me, (long)txr->total_packets.ev_count);
2903 		device_printf(dev, "TX(%d) NO Desc Avail: %lu\n",
2904 		    txr->me, (long)txr->no_desc_avail.ev_count);
2905 	}
2906 
2907 	device_printf(dev, "Admin IRQ Handled: %lu\n",
2908 	    (long)adapter->admin_irqev.ev_count);
2909 	device_printf(dev, "Admin work Handled: %lu\n",
2910 	    (long)adapter->link_workev.ev_count);
2911 } /* ixv_print_debug_info */
2912 
2913 /************************************************************************
2914  * ixv_sysctl_debug
2915  ************************************************************************/
2916 static int
2917 ixv_sysctl_debug(SYSCTLFN_ARGS)
2918 {
2919 	struct sysctlnode node = *rnode;
2920 	struct adapter *adapter = (struct adapter *)node.sysctl_data;
2921 	int	       error, result;
2922 
2923 	node.sysctl_data = &result;
2924 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2925 
2926 	if (error || newp == NULL)
2927 		return error;
2928 
2929 	if (result == 1)
2930 		ixv_print_debug_info(adapter);
2931 
2932 	return 0;
2933 } /* ixv_sysctl_debug */
2934 
2935 /************************************************************************
2936  * ixv_init_device_features
2937  ************************************************************************/
2938 static void
2939 ixv_init_device_features(struct adapter *adapter)
2940 {
2941 	adapter->feat_cap = IXGBE_FEATURE_NETMAP
2942 			  | IXGBE_FEATURE_VF
2943 			  | IXGBE_FEATURE_RSS
2944 			  | IXGBE_FEATURE_LEGACY_TX;
2945 
2946 	/* A tad short on feature flags for VFs, atm. */
2947 	switch (adapter->hw.mac.type) {
2948 	case ixgbe_mac_82599_vf:
2949 		break;
2950 	case ixgbe_mac_X540_vf:
2951 		break;
2952 	case ixgbe_mac_X550_vf:
2953 	case ixgbe_mac_X550EM_x_vf:
2954 	case ixgbe_mac_X550EM_a_vf:
2955 		adapter->feat_cap |= IXGBE_FEATURE_NEEDS_CTXD;
2956 		break;
2957 	default:
2958 		break;
2959 	}
2960 
2961 	/* Enabled by default... */
2962 	/* Is a virtual function (VF) */
2963 	if (adapter->feat_cap & IXGBE_FEATURE_VF)
2964 		adapter->feat_en |= IXGBE_FEATURE_VF;
2965 	/* Netmap */
2966 	if (adapter->feat_cap & IXGBE_FEATURE_NETMAP)
2967 		adapter->feat_en |= IXGBE_FEATURE_NETMAP;
2968 	/* Receive-Side Scaling (RSS) */
2969 	if (adapter->feat_cap & IXGBE_FEATURE_RSS)
2970 		adapter->feat_en |= IXGBE_FEATURE_RSS;
2971 	/* Needs advanced context descriptor regardless of offloads req'd */
2972 	if (adapter->feat_cap & IXGBE_FEATURE_NEEDS_CTXD)
2973 		adapter->feat_en |= IXGBE_FEATURE_NEEDS_CTXD;
2974 
2975 	/* Enabled via sysctl... */
2976 	/* Legacy (single queue) transmit */
2977 	if ((adapter->feat_cap & IXGBE_FEATURE_LEGACY_TX) &&
2978 	    ixv_enable_legacy_tx)
2979 		adapter->feat_en |= IXGBE_FEATURE_LEGACY_TX;
2980 } /* ixv_init_device_features */
2981 
2982 /************************************************************************
2983  * ixv_shutdown - Shutdown entry point
2984  ************************************************************************/
2985 #if 0 /* XXX NetBSD ought to register something like this through pmf(9) */
2986 static int
2987 ixv_shutdown(device_t dev)
2988 {
2989 	struct adapter *adapter = device_private(dev);
2990 	IXGBE_CORE_LOCK(adapter);
2991 	ixv_stop_locked(adapter);
2992 	IXGBE_CORE_UNLOCK(adapter);
2993 
2994 	return (0);
2995 } /* ixv_shutdown */
2996 #endif
2997 
2998 static int
2999 ixv_ifflags_cb(struct ethercom *ec)
3000 {
3001 	struct ifnet *ifp = &ec->ec_if;
3002 	struct adapter *adapter = ifp->if_softc;
3003 	u_short saved_flags;
3004 	u_short change;
3005 	int rv = 0;
3006 
3007 	IXGBE_CORE_LOCK(adapter);
3008 
3009 	saved_flags = adapter->if_flags;
3010 	change = ifp->if_flags ^ adapter->if_flags;
3011 	if (change != 0)
3012 		adapter->if_flags = ifp->if_flags;
3013 
3014 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
3015 		rv = ENETRESET;
3016 		goto out;
3017 	} else if ((change & IFF_PROMISC) != 0) {
3018 		rv = ixv_set_rxfilter(adapter);
3019 		if (rv != 0) {
3020 			/* Restore previous */
3021 			adapter->if_flags = saved_flags;
3022 			goto out;
3023 		}
3024 	}
3025 
3026 	/* Check for ec_capenable. */
3027 	change = ec->ec_capenable ^ adapter->ec_capenable;
3028 	adapter->ec_capenable = ec->ec_capenable;
3029 	if ((change & ~(ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING
3030 	    | ETHERCAP_VLAN_HWFILTER)) != 0) {
3031 		rv = ENETRESET;
3032 		goto out;
3033 	}
3034 
3035 	/*
3036 	 * Special handling is not required for ETHERCAP_VLAN_MTU.
3037 	 * PF's MAXFRS(MHADD) does not include the 4bytes of the VLAN header.
3038 	 */
3039 
3040 	/* Set up VLAN support and filter */
3041 	if ((change & (ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_HWFILTER)) != 0)
3042 		rv = ixv_setup_vlan_support(adapter);
3043 
3044 out:
3045 	IXGBE_CORE_UNLOCK(adapter);
3046 
3047 	return rv;
3048 }
3049 
3050 
3051 /************************************************************************
3052  * ixv_ioctl - Ioctl entry point
3053  *
3054  *   Called when the user wants to configure the interface.
3055  *
3056  *   return 0 on success, positive on failure
3057  ************************************************************************/
3058 static int
3059 ixv_ioctl(struct ifnet *ifp, u_long command, void *data)
3060 {
3061 	struct adapter	*adapter = ifp->if_softc;
3062 	struct ixgbe_hw *hw = &adapter->hw;
3063 	struct ifcapreq *ifcr = data;
3064 	int		error;
3065 	int l4csum_en;
3066 	const int l4csum = IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx |
3067 	     IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_UDPv6_Rx;
3068 
3069 	switch (command) {
3070 	case SIOCSIFFLAGS:
3071 		IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)");
3072 		break;
3073 	case SIOCADDMULTI: {
3074 		struct ether_multi *enm;
3075 		struct ether_multistep step;
3076 		struct ethercom *ec = &adapter->osdep.ec;
3077 		bool overflow = false;
3078 		int mcnt = 0;
3079 
3080 		/*
3081 		 * Check the number of multicast address. If it exceeds,
3082 		 * return ENOSPC.
3083 		 * Update this code when we support API 1.3.
3084 		 */
3085 		ETHER_LOCK(ec);
3086 		ETHER_FIRST_MULTI(step, ec, enm);
3087 		while (enm != NULL) {
3088 			mcnt++;
3089 
3090 			/*
3091 			 * This code is before adding, so one room is required
3092 			 * at least.
3093 			 */
3094 			if (mcnt > (IXGBE_MAX_VF_MC - 1)) {
3095 				overflow = true;
3096 				break;
3097 			}
3098 			ETHER_NEXT_MULTI(step, enm);
3099 		}
3100 		ETHER_UNLOCK(ec);
3101 		error = 0;
3102 		if (overflow && ((ec->ec_flags & ETHER_F_ALLMULTI) == 0)) {
3103 			error = hw->mac.ops.update_xcast_mode(hw,
3104 			    IXGBEVF_XCAST_MODE_ALLMULTI);
3105 			if (error == IXGBE_ERR_NOT_TRUSTED) {
3106 				device_printf(adapter->dev,
3107 				    "this interface is not trusted\n");
3108 				error = EPERM;
3109 			} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
3110 				device_printf(adapter->dev,
3111 				    "the PF doesn't support allmulti mode\n");
3112 				error = EOPNOTSUPP;
3113 			} else if (error) {
3114 				device_printf(adapter->dev,
3115 				    "number of Ethernet multicast addresses "
3116 				    "exceeds the limit (%d). error = %d\n",
3117 				    IXGBE_MAX_VF_MC, error);
3118 				error = ENOSPC;
3119 			} else
3120 				ec->ec_flags |= ETHER_F_ALLMULTI;
3121 		}
3122 		if (error)
3123 			return error;
3124 	}
3125 		/*FALLTHROUGH*/
3126 	case SIOCDELMULTI:
3127 		IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI");
3128 		break;
3129 	case SIOCSIFMEDIA:
3130 	case SIOCGIFMEDIA:
3131 		IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)");
3132 		break;
3133 	case SIOCSIFCAP:
3134 		IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)");
3135 		break;
3136 	case SIOCSIFMTU:
3137 		IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)");
3138 		break;
3139 	case SIOCZIFDATA:
3140 		IOCTL_DEBUGOUT("ioctl: SIOCZIFDATA (Zero counter)");
3141 		ixv_update_stats(adapter);
3142 		ixv_clear_evcnt(adapter);
3143 		break;
3144 	default:
3145 		IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)", (int)command);
3146 		break;
3147 	}
3148 
3149 	switch (command) {
3150 	case SIOCSIFCAP:
3151 		/* Layer-4 Rx checksum offload has to be turned on and
3152 		 * off as a unit.
3153 		 */
3154 		l4csum_en = ifcr->ifcr_capenable & l4csum;
3155 		if (l4csum_en != l4csum && l4csum_en != 0)
3156 			return EINVAL;
3157 		/*FALLTHROUGH*/
3158 	case SIOCADDMULTI:
3159 	case SIOCDELMULTI:
3160 	case SIOCSIFFLAGS:
3161 	case SIOCSIFMTU:
3162 	default:
3163 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
3164 			return error;
3165 		if ((ifp->if_flags & IFF_RUNNING) == 0)
3166 			;
3167 		else if (command == SIOCSIFCAP || command == SIOCSIFMTU) {
3168 			IXGBE_CORE_LOCK(adapter);
3169 			ixv_init_locked(adapter);
3170 			IXGBE_CORE_UNLOCK(adapter);
3171 		} else if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
3172 			/*
3173 			 * Multicast list has changed; set the hardware filter
3174 			 * accordingly.
3175 			 */
3176 			IXGBE_CORE_LOCK(adapter);
3177 			ixv_disable_intr(adapter);
3178 			ixv_set_rxfilter(adapter);
3179 			ixv_enable_intr(adapter);
3180 			IXGBE_CORE_UNLOCK(adapter);
3181 		}
3182 		return 0;
3183 	}
3184 } /* ixv_ioctl */
3185 
3186 /************************************************************************
3187  * ixv_init
3188  ************************************************************************/
3189 static int
3190 ixv_init(struct ifnet *ifp)
3191 {
3192 	struct adapter *adapter = ifp->if_softc;
3193 
3194 	IXGBE_CORE_LOCK(adapter);
3195 	ixv_init_locked(adapter);
3196 	IXGBE_CORE_UNLOCK(adapter);
3197 
3198 	return 0;
3199 } /* ixv_init */
3200 
3201 /************************************************************************
3202  * ixv_handle_que
3203  ************************************************************************/
3204 static void
3205 ixv_handle_que(void *context)
3206 {
3207 	struct ix_queue *que = context;
3208 	struct adapter	*adapter = que->adapter;
3209 	struct tx_ring	*txr = que->txr;
3210 	struct ifnet	*ifp = adapter->ifp;
3211 	bool		more;
3212 
3213 	que->handleq.ev_count++;
3214 
3215 	if (ifp->if_flags & IFF_RUNNING) {
3216 		more = ixgbe_rxeof(que);
3217 		IXGBE_TX_LOCK(txr);
3218 		more |= ixgbe_txeof(txr);
3219 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
3220 			if (!ixgbe_mq_ring_empty(ifp, txr->txr_interq))
3221 				ixgbe_mq_start_locked(ifp, txr);
3222 		/* Only for queue 0 */
3223 		/* NetBSD still needs this for CBQ */
3224 		if ((&adapter->queues[0] == que)
3225 		    && (!ixgbe_legacy_ring_empty(ifp, NULL)))
3226 			ixgbe_legacy_start_locked(ifp, txr);
3227 		IXGBE_TX_UNLOCK(txr);
3228 		if (more) {
3229 			que->req.ev_count++;
3230 			if (adapter->txrx_use_workqueue) {
3231 				/*
3232 				 * "enqueued flag" is not required here
3233 				 * the same as ixg(4). See ixgbe_msix_que().
3234 				 */
3235 				workqueue_enqueue(adapter->que_wq,
3236 				    &que->wq_cookie, curcpu());
3237 			} else
3238 				  softint_schedule(que->que_si);
3239 			return;
3240 		}
3241 	}
3242 
3243 	/* Re-enable this interrupt */
3244 	ixv_enable_queue(adapter, que->msix);
3245 
3246 	return;
3247 } /* ixv_handle_que */
3248 
3249 /************************************************************************
3250  * ixv_handle_que_work
3251  ************************************************************************/
3252 static void
3253 ixv_handle_que_work(struct work *wk, void *context)
3254 {
3255 	struct ix_queue *que = container_of(wk, struct ix_queue, wq_cookie);
3256 
3257 	/*
3258 	 * "enqueued flag" is not required here the same as ixg(4).
3259 	 * See ixgbe_msix_que().
3260 	 */
3261 	ixv_handle_que(que);
3262 }
3263 
3264 /************************************************************************
3265  * ixv_allocate_msix - Setup MSI-X Interrupt resources and handlers
3266  ************************************************************************/
3267 static int
3268 ixv_allocate_msix(struct adapter *adapter, const struct pci_attach_args *pa)
3269 {
3270 	device_t	dev = adapter->dev;
3271 	struct ix_queue *que = adapter->queues;
3272 	struct		tx_ring *txr = adapter->tx_rings;
3273 	int		error, msix_ctrl, rid, vector = 0;
3274 	pci_chipset_tag_t pc;
3275 	pcitag_t	tag;
3276 	char		intrbuf[PCI_INTRSTR_LEN];
3277 	char		wqname[MAXCOMLEN];
3278 	char		intr_xname[32];
3279 	const char	*intrstr = NULL;
3280 	kcpuset_t	*affinity;
3281 	int		cpu_id = 0;
3282 
3283 	pc = adapter->osdep.pc;
3284 	tag = adapter->osdep.tag;
3285 
3286 	adapter->osdep.nintrs = adapter->num_queues + 1;
3287 	if (pci_msix_alloc_exact(pa, &adapter->osdep.intrs,
3288 	    adapter->osdep.nintrs) != 0) {
3289 		aprint_error_dev(dev,
3290 		    "failed to allocate MSI-X interrupt\n");
3291 		return (ENXIO);
3292 	}
3293 
3294 	kcpuset_create(&affinity, false);
3295 	for (int i = 0; i < adapter->num_queues; i++, vector++, que++, txr++) {
3296 		snprintf(intr_xname, sizeof(intr_xname), "%s TXRX%d",
3297 		    device_xname(dev), i);
3298 		intrstr = pci_intr_string(pc, adapter->osdep.intrs[i], intrbuf,
3299 		    sizeof(intrbuf));
3300 #ifdef IXGBE_MPSAFE
3301 		pci_intr_setattr(pc, &adapter->osdep.intrs[i], PCI_INTR_MPSAFE,
3302 		    true);
3303 #endif
3304 		/* Set the handler function */
3305 		que->res = adapter->osdep.ihs[i] = pci_intr_establish_xname(pc,
3306 		    adapter->osdep.intrs[i], IPL_NET, ixv_msix_que, que,
3307 		    intr_xname);
3308 		if (que->res == NULL) {
3309 			pci_intr_release(pc, adapter->osdep.intrs,
3310 			    adapter->osdep.nintrs);
3311 			aprint_error_dev(dev,
3312 			    "Failed to register QUE handler\n");
3313 			kcpuset_destroy(affinity);
3314 			return (ENXIO);
3315 		}
3316 		que->msix = vector;
3317 		adapter->active_queues |= (u64)(1 << que->msix);
3318 
3319 		cpu_id = i;
3320 		/* Round-robin affinity */
3321 		kcpuset_zero(affinity);
3322 		kcpuset_set(affinity, cpu_id % ncpu);
3323 		error = interrupt_distribute(adapter->osdep.ihs[i], affinity,
3324 		    NULL);
3325 		aprint_normal_dev(dev, "for TX/RX, interrupting at %s",
3326 		    intrstr);
3327 		if (error == 0)
3328 			aprint_normal(", bound queue %d to cpu %d\n",
3329 			    i, cpu_id % ncpu);
3330 		else
3331 			aprint_normal("\n");
3332 
3333 #ifndef IXGBE_LEGACY_TX
3334 		txr->txr_si
3335 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINT_FLAGS,
3336 			ixgbe_deferred_mq_start, txr);
3337 #endif
3338 		que->que_si
3339 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINT_FLAGS,
3340 			ixv_handle_que, que);
3341 		if (que->que_si == NULL) {
3342 			aprint_error_dev(dev,
3343 			    "could not establish software interrupt\n");
3344 		}
3345 	}
3346 	snprintf(wqname, sizeof(wqname), "%sdeferTx", device_xname(dev));
3347 	error = workqueue_create(&adapter->txr_wq, wqname,
3348 	    ixgbe_deferred_mq_start_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3349 	    IXGBE_WORKQUEUE_FLAGS);
3350 	if (error) {
3351 		aprint_error_dev(dev,
3352 		    "couldn't create workqueue for deferred Tx\n");
3353 	}
3354 	adapter->txr_wq_enqueued = percpu_alloc(sizeof(u_int));
3355 
3356 	snprintf(wqname, sizeof(wqname), "%sTxRx", device_xname(dev));
3357 	error = workqueue_create(&adapter->que_wq, wqname,
3358 	    ixv_handle_que_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3359 	    IXGBE_WORKQUEUE_FLAGS);
3360 	if (error) {
3361 		aprint_error_dev(dev, "couldn't create workqueue for Tx/Rx\n");
3362 	}
3363 
3364 	/* and Mailbox */
3365 	cpu_id++;
3366 	snprintf(intr_xname, sizeof(intr_xname), "%s link", device_xname(dev));
3367 	adapter->vector = vector;
3368 	intrstr = pci_intr_string(pc, adapter->osdep.intrs[vector], intrbuf,
3369 	    sizeof(intrbuf));
3370 #ifdef IXGBE_MPSAFE
3371 	pci_intr_setattr(pc, &adapter->osdep.intrs[vector], PCI_INTR_MPSAFE,
3372 	    true);
3373 #endif
3374 	/* Set the mbx handler function */
3375 	adapter->osdep.ihs[vector] = pci_intr_establish_xname(pc,
3376 	    adapter->osdep.intrs[vector], IPL_NET, ixv_msix_mbx, adapter,
3377 	    intr_xname);
3378 	if (adapter->osdep.ihs[vector] == NULL) {
3379 		aprint_error_dev(dev, "Failed to register LINK handler\n");
3380 		kcpuset_destroy(affinity);
3381 		return (ENXIO);
3382 	}
3383 	/* Round-robin affinity */
3384 	kcpuset_zero(affinity);
3385 	kcpuset_set(affinity, cpu_id % ncpu);
3386 	error = interrupt_distribute(adapter->osdep.ihs[vector], affinity,
3387 	    NULL);
3388 
3389 	aprint_normal_dev(dev,
3390 	    "for link, interrupting at %s", intrstr);
3391 	if (error == 0)
3392 		aprint_normal(", affinity to cpu %d\n", cpu_id % ncpu);
3393 	else
3394 		aprint_normal("\n");
3395 
3396 	/* Tasklets for Mailbox */
3397 	snprintf(wqname, sizeof(wqname), "%s-admin", device_xname(dev));
3398 	error = workqueue_create(&adapter->admin_wq, wqname,
3399 	    ixv_handle_admin, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3400 	    IXGBE_TASKLET_WQ_FLAGS);
3401 	if (error) {
3402 		aprint_error_dev(dev,
3403 		    "could not create admin workqueue (%d)\n", error);
3404 		goto err_out;
3405 	}
3406 
3407 	/*
3408 	 * Due to a broken design QEMU will fail to properly
3409 	 * enable the guest for MSI-X unless the vectors in
3410 	 * the table are all set up, so we must rewrite the
3411 	 * ENABLE in the MSI-X control register again at this
3412 	 * point to cause it to successfully initialize us.
3413 	 */
3414 	if (adapter->hw.mac.type == ixgbe_mac_82599_vf) {
3415 		pci_get_capability(pc, tag, PCI_CAP_MSIX, &rid, NULL);
3416 		rid += PCI_MSIX_CTL;
3417 		msix_ctrl = pci_conf_read(pc, tag, rid);
3418 		msix_ctrl |= PCI_MSIX_CTL_ENABLE;
3419 		pci_conf_write(pc, tag, rid, msix_ctrl);
3420 	}
3421 
3422 	kcpuset_destroy(affinity);
3423 	return (0);
3424 err_out:
3425 	kcpuset_destroy(affinity);
3426 	ixv_free_deferred_handlers(adapter);
3427 	ixv_free_pci_resources(adapter);
3428 	return (error);
3429 } /* ixv_allocate_msix */
3430 
3431 /************************************************************************
3432  * ixv_configure_interrupts - Setup MSI-X resources
3433  *
3434  *   Note: The VF device MUST use MSI-X, there is no fallback.
3435  ************************************************************************/
3436 static int
3437 ixv_configure_interrupts(struct adapter *adapter)
3438 {
3439 	device_t dev = adapter->dev;
3440 	int want, queues, msgs;
3441 
3442 	/* Must have at least 2 MSI-X vectors */
3443 	msgs = pci_msix_count(adapter->osdep.pc, adapter->osdep.tag);
3444 	if (msgs < 2) {
3445 		aprint_error_dev(dev, "MSIX config error\n");
3446 		return (ENXIO);
3447 	}
3448 	msgs = MIN(msgs, IXG_MAX_NINTR);
3449 
3450 	/* Figure out a reasonable auto config value */
3451 	queues = (ncpu > (msgs - 1)) ? (msgs - 1) : ncpu;
3452 
3453 	if (ixv_num_queues != 0)
3454 		queues = ixv_num_queues;
3455 	else if ((ixv_num_queues == 0) && (queues > IXGBE_VF_MAX_TX_QUEUES))
3456 		queues = IXGBE_VF_MAX_TX_QUEUES;
3457 
3458 	/*
3459 	 * Want vectors for the queues,
3460 	 * plus an additional for mailbox.
3461 	 */
3462 	want = queues + 1;
3463 	if (msgs >= want)
3464 		msgs = want;
3465 	else {
3466 		aprint_error_dev(dev,
3467 		    "MSI-X Configuration Problem, "
3468 		    "%d vectors but %d queues wanted!\n",
3469 		    msgs, want);
3470 		return -1;
3471 	}
3472 
3473 	adapter->msix_mem = (void *)1; /* XXX */
3474 	aprint_normal_dev(dev,
3475 	    "Using MSI-X interrupts with %d vectors\n", msgs);
3476 	adapter->num_queues = queues;
3477 
3478 	return (0);
3479 } /* ixv_configure_interrupts */
3480 
3481 
3482 /************************************************************************
3483  * ixv_handle_admin - Tasklet handler for MSI-X MBX interrupts
3484  *
3485  *   Done outside of interrupt context since the driver might sleep
3486  ************************************************************************/
3487 static void
3488 ixv_handle_admin(struct work *wk, void *context)
3489 {
3490 	struct adapter *adapter = context;
3491 	struct ixgbe_hw	*hw = &adapter->hw;
3492 
3493 	IXGBE_CORE_LOCK(adapter);
3494 
3495 	++adapter->link_workev.ev_count;
3496 	adapter->hw.mac.ops.check_link(&adapter->hw, &adapter->link_speed,
3497 	    &adapter->link_up, FALSE);
3498 	ixv_update_link_status(adapter);
3499 
3500 	adapter->task_requests = 0;
3501 	atomic_store_relaxed(&adapter->admin_pending, 0);
3502 
3503 	/* Re-enable interrupts */
3504 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
3505 
3506 	IXGBE_CORE_UNLOCK(adapter);
3507 } /* ixv_handle_admin */
3508 
3509 /************************************************************************
3510  * ixv_check_link - Used in the local timer to poll for link changes
3511  ************************************************************************/
3512 static s32
3513 ixv_check_link(struct adapter *adapter)
3514 {
3515 	s32 error;
3516 
3517 	KASSERT(mutex_owned(&adapter->core_mtx));
3518 
3519 	adapter->hw.mac.get_link_status = TRUE;
3520 
3521 	error = adapter->hw.mac.ops.check_link(&adapter->hw,
3522 	    &adapter->link_speed, &adapter->link_up, FALSE);
3523 	ixv_update_link_status(adapter);
3524 
3525 	return error;
3526 } /* ixv_check_link */
3527