xref: /netbsd-src/sys/dev/pci/ixgbe/ixv.c (revision d90047b5d07facf36e6c01dcc0bded8997ce9cc2)
1 /*$NetBSD: ixv.c,v 1.151 2020/06/25 07:53:02 msaitoh Exp $*/
2 
3 /******************************************************************************
4 
5   Copyright (c) 2001-2017, Intel Corporation
6   All rights reserved.
7 
8   Redistribution and use in source and binary forms, with or without
9   modification, are permitted provided that the following conditions are met:
10 
11    1. Redistributions of source code must retain the above copyright notice,
12       this list of conditions and the following disclaimer.
13 
14    2. Redistributions in binary form must reproduce the above copyright
15       notice, this list of conditions and the following disclaimer in the
16       documentation and/or other materials provided with the distribution.
17 
18    3. Neither the name of the Intel Corporation nor the names of its
19       contributors may be used to endorse or promote products derived from
20       this software without specific prior written permission.
21 
22   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
23   AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24   IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25   ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26   LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27   CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28   SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29   INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30   CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31   ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32   POSSIBILITY OF SUCH DAMAGE.
33 
34 ******************************************************************************/
35 /*$FreeBSD: head/sys/dev/ixgbe/if_ixv.c 331224 2018-03-19 20:55:05Z erj $*/
36 
37 #ifdef _KERNEL_OPT
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_net_mpsafe.h"
41 #endif
42 
43 #include "ixgbe.h"
44 #include "vlan.h"
45 
46 /************************************************************************
47  * Driver version
48  ************************************************************************/
49 static const char ixv_driver_version[] = "2.0.1-k";
50 /* XXX NetBSD: + 1.5.17 */
51 
52 /************************************************************************
53  * PCI Device ID Table
54  *
55  *   Used by probe to select devices to load on
56  *   Last field stores an index into ixv_strings
57  *   Last entry must be all 0s
58  *
59  *   { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
60  ************************************************************************/
61 static const ixgbe_vendor_info_t ixv_vendor_info_array[] =
62 {
63 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_VF, 0, 0, 0},
64 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540_VF, 0, 0, 0},
65 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550_VF, 0, 0, 0},
66 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_VF, 0, 0, 0},
67 	{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_VF, 0, 0, 0},
68 	/* required last entry */
69 	{0, 0, 0, 0, 0}
70 };
71 
72 /************************************************************************
73  * Table of branding strings
74  ************************************************************************/
75 static const char *ixv_strings[] = {
76 	"Intel(R) PRO/10GbE Virtual Function Network Driver"
77 };
78 
79 /*********************************************************************
80  *  Function prototypes
81  *********************************************************************/
82 static int	ixv_probe(device_t, cfdata_t, void *);
83 static void	ixv_attach(device_t, device_t, void *);
84 static int	ixv_detach(device_t, int);
85 #if 0
86 static int	ixv_shutdown(device_t);
87 #endif
88 static int	ixv_ifflags_cb(struct ethercom *);
89 static int	ixv_ioctl(struct ifnet *, u_long, void *);
90 static int	ixv_init(struct ifnet *);
91 static void	ixv_init_locked(struct adapter *);
92 static void	ixv_ifstop(struct ifnet *, int);
93 static void	ixv_stop(void *);
94 static void	ixv_init_device_features(struct adapter *);
95 static void	ixv_media_status(struct ifnet *, struct ifmediareq *);
96 static int	ixv_media_change(struct ifnet *);
97 static int	ixv_allocate_pci_resources(struct adapter *,
98 		    const struct pci_attach_args *);
99 static void	ixv_free_workqueue(struct adapter *);
100 static int	ixv_allocate_msix(struct adapter *,
101 		    const struct pci_attach_args *);
102 static int	ixv_configure_interrupts(struct adapter *);
103 static void	ixv_free_pci_resources(struct adapter *);
104 static void	ixv_local_timer(void *);
105 static void	ixv_handle_timer(struct work *, void *);
106 static int	ixv_setup_interface(device_t, struct adapter *);
107 static void	ixv_schedule_admin_tasklet(struct adapter *);
108 static int	ixv_negotiate_api(struct adapter *);
109 
110 static void	ixv_initialize_transmit_units(struct adapter *);
111 static void	ixv_initialize_receive_units(struct adapter *);
112 static void	ixv_initialize_rss_mapping(struct adapter *);
113 static s32	ixv_check_link(struct adapter *);
114 
115 static void	ixv_enable_intr(struct adapter *);
116 static void	ixv_disable_intr(struct adapter *);
117 static int	ixv_set_rxfilter(struct adapter *);
118 static void	ixv_update_link_status(struct adapter *);
119 static int	ixv_sysctl_debug(SYSCTLFN_PROTO);
120 static void	ixv_set_ivar(struct adapter *, u8, u8, s8);
121 static void	ixv_configure_ivars(struct adapter *);
122 static u8 *	ixv_mc_array_itr(struct ixgbe_hw *, u8 **, u32 *);
123 static void	ixv_eitr_write(struct adapter *, uint32_t, uint32_t);
124 
125 static void	ixv_setup_vlan_tagging(struct adapter *);
126 static int	ixv_setup_vlan_support(struct adapter *);
127 static int	ixv_vlan_cb(struct ethercom *, uint16_t, bool);
128 static int	ixv_register_vlan(struct adapter *, u16);
129 static int	ixv_unregister_vlan(struct adapter *, u16);
130 
131 static void	ixv_add_device_sysctls(struct adapter *);
132 static void	ixv_save_stats(struct adapter *);
133 static void	ixv_init_stats(struct adapter *);
134 static void	ixv_update_stats(struct adapter *);
135 static void	ixv_add_stats_sysctls(struct adapter *);
136 static void	ixv_clear_evcnt(struct adapter *);
137 
138 /* Sysctl handlers */
139 static void	ixv_set_sysctl_value(struct adapter *, const char *,
140 		    const char *, int *, int);
141 static int	ixv_sysctl_interrupt_rate_handler(SYSCTLFN_PROTO);
142 static int	ixv_sysctl_next_to_check_handler(SYSCTLFN_PROTO);
143 static int	ixv_sysctl_rdh_handler(SYSCTLFN_PROTO);
144 static int	ixv_sysctl_rdt_handler(SYSCTLFN_PROTO);
145 static int	ixv_sysctl_tdt_handler(SYSCTLFN_PROTO);
146 static int	ixv_sysctl_tdh_handler(SYSCTLFN_PROTO);
147 
148 /* The MSI-X Interrupt handlers */
149 static int	ixv_msix_que(void *);
150 static int	ixv_msix_mbx(void *);
151 
152 /* Event handlers running on workqueue */
153 static void	ixv_handle_que(void *);
154 
155 /* Deferred workqueue handlers */
156 static void	ixv_handle_admin(struct work *, void *);
157 static void	ixv_handle_que_work(struct work *, void *);
158 
159 const struct sysctlnode *ixv_sysctl_instance(struct adapter *);
160 static const ixgbe_vendor_info_t *ixv_lookup(const struct pci_attach_args *);
161 
162 /************************************************************************
163  * NetBSD Device Interface Entry Points
164  ************************************************************************/
165 CFATTACH_DECL3_NEW(ixv, sizeof(struct adapter),
166     ixv_probe, ixv_attach, ixv_detach, NULL, NULL, NULL,
167     DVF_DETACH_SHUTDOWN);
168 
169 #if 0
170 static driver_t ixv_driver = {
171 	"ixv", ixv_methods, sizeof(struct adapter),
172 };
173 
174 devclass_t ixv_devclass;
175 DRIVER_MODULE(ixv, pci, ixv_driver, ixv_devclass, 0, 0);
176 MODULE_DEPEND(ixv, pci, 1, 1, 1);
177 MODULE_DEPEND(ixv, ether, 1, 1, 1);
178 #endif
179 
180 /*
181  * TUNEABLE PARAMETERS:
182  */
183 
184 /* Number of Queues - do not exceed MSI-X vectors - 1 */
185 static int ixv_num_queues = 0;
186 #define	TUNABLE_INT(__x, __y)
187 TUNABLE_INT("hw.ixv.num_queues", &ixv_num_queues);
188 
189 /*
190  * AIM: Adaptive Interrupt Moderation
191  * which means that the interrupt rate
192  * is varied over time based on the
193  * traffic for that interrupt vector
194  */
195 static bool ixv_enable_aim = false;
196 TUNABLE_INT("hw.ixv.enable_aim", &ixv_enable_aim);
197 
198 static int ixv_max_interrupt_rate = (4000000 / IXGBE_LOW_LATENCY);
199 TUNABLE_INT("hw.ixv.max_interrupt_rate", &ixv_max_interrupt_rate);
200 
201 /* How many packets rxeof tries to clean at a time */
202 static int ixv_rx_process_limit = 256;
203 TUNABLE_INT("hw.ixv.rx_process_limit", &ixv_rx_process_limit);
204 
205 /* How many packets txeof tries to clean at a time */
206 static int ixv_tx_process_limit = 256;
207 TUNABLE_INT("hw.ixv.tx_process_limit", &ixv_tx_process_limit);
208 
209 /* Which packet processing uses workqueue or softint */
210 static bool ixv_txrx_workqueue = false;
211 
212 /*
213  * Number of TX descriptors per ring,
214  * setting higher than RX as this seems
215  * the better performing choice.
216  */
217 static int ixv_txd = PERFORM_TXD;
218 TUNABLE_INT("hw.ixv.txd", &ixv_txd);
219 
220 /* Number of RX descriptors per ring */
221 static int ixv_rxd = PERFORM_RXD;
222 TUNABLE_INT("hw.ixv.rxd", &ixv_rxd);
223 
224 /* Legacy Transmit (single queue) */
225 static int ixv_enable_legacy_tx = 0;
226 TUNABLE_INT("hw.ixv.enable_legacy_tx", &ixv_enable_legacy_tx);
227 
228 #ifdef NET_MPSAFE
229 #define IXGBE_MPSAFE		1
230 #define IXGBE_CALLOUT_FLAGS	CALLOUT_MPSAFE
231 #define IXGBE_SOFTINT_FLAGS	SOFTINT_MPSAFE
232 #define IXGBE_WORKQUEUE_FLAGS	WQ_PERCPU | WQ_MPSAFE
233 #define IXGBE_TASKLET_WQ_FLAGS	WQ_MPSAFE
234 #else
235 #define IXGBE_CALLOUT_FLAGS	0
236 #define IXGBE_SOFTINT_FLAGS	0
237 #define IXGBE_WORKQUEUE_FLAGS	WQ_PERCPU
238 #define IXGBE_TASKLET_WQ_FLAGS	0
239 #endif
240 #define IXGBE_WORKQUEUE_PRI PRI_SOFTNET
241 
242 #if 0
243 static int (*ixv_start_locked)(struct ifnet *, struct tx_ring *);
244 static int (*ixv_ring_empty)(struct ifnet *, struct buf_ring *);
245 #endif
246 
247 /************************************************************************
248  * ixv_probe - Device identification routine
249  *
250  *   Determines if the driver should be loaded on
251  *   adapter based on its PCI vendor/device ID.
252  *
253  *   return BUS_PROBE_DEFAULT on success, positive on failure
254  ************************************************************************/
255 static int
256 ixv_probe(device_t dev, cfdata_t cf, void *aux)
257 {
258 #ifdef __HAVE_PCI_MSI_MSIX
259 	const struct pci_attach_args *pa = aux;
260 
261 	return (ixv_lookup(pa) != NULL) ? 1 : 0;
262 #else
263 	return 0;
264 #endif
265 } /* ixv_probe */
266 
267 static const ixgbe_vendor_info_t *
268 ixv_lookup(const struct pci_attach_args *pa)
269 {
270 	const ixgbe_vendor_info_t *ent;
271 	pcireg_t subid;
272 
273 	INIT_DEBUGOUT("ixv_lookup: begin");
274 
275 	if (PCI_VENDOR(pa->pa_id) != IXGBE_INTEL_VENDOR_ID)
276 		return NULL;
277 
278 	subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
279 
280 	for (ent = ixv_vendor_info_array; ent->vendor_id != 0; ent++) {
281 		if ((PCI_VENDOR(pa->pa_id) == ent->vendor_id) &&
282 		    (PCI_PRODUCT(pa->pa_id) == ent->device_id) &&
283 		    ((PCI_SUBSYS_VENDOR(subid) == ent->subvendor_id) ||
284 		     (ent->subvendor_id == 0)) &&
285 		    ((PCI_SUBSYS_ID(subid) == ent->subdevice_id) ||
286 		     (ent->subdevice_id == 0))) {
287 			return ent;
288 		}
289 	}
290 
291 	return NULL;
292 }
293 
294 /************************************************************************
295  * ixv_attach - Device initialization routine
296  *
297  *   Called when the driver is being loaded.
298  *   Identifies the type of hardware, allocates all resources
299  *   and initializes the hardware.
300  *
301  *   return 0 on success, positive on failure
302  ************************************************************************/
303 static void
304 ixv_attach(device_t parent, device_t dev, void *aux)
305 {
306 	struct adapter *adapter;
307 	struct ixgbe_hw *hw;
308 	int		error = 0;
309 	pcireg_t	id, subid;
310 	const ixgbe_vendor_info_t *ent;
311 	const struct pci_attach_args *pa = aux;
312 	const char *apivstr;
313 	const char *str;
314 	char wqname[MAXCOMLEN];
315 	char buf[256];
316 
317 	INIT_DEBUGOUT("ixv_attach: begin");
318 
319 	/*
320 	 * Make sure BUSMASTER is set, on a VM under
321 	 * KVM it may not be and will break things.
322 	 */
323 	ixgbe_pci_enable_busmaster(pa->pa_pc, pa->pa_tag);
324 
325 	/* Allocate, clear, and link in our adapter structure */
326 	adapter = device_private(dev);
327 	adapter->hw.back = adapter;
328 	adapter->dev = dev;
329 	hw = &adapter->hw;
330 
331 	adapter->init_locked = ixv_init_locked;
332 	adapter->stop_locked = ixv_stop;
333 
334 	adapter->osdep.pc = pa->pa_pc;
335 	adapter->osdep.tag = pa->pa_tag;
336 	if (pci_dma64_available(pa))
337 		adapter->osdep.dmat = pa->pa_dmat64;
338 	else
339 		adapter->osdep.dmat = pa->pa_dmat;
340 	adapter->osdep.attached = false;
341 
342 	ent = ixv_lookup(pa);
343 
344 	KASSERT(ent != NULL);
345 
346 	aprint_normal(": %s, Version - %s\n",
347 	    ixv_strings[ent->index], ixv_driver_version);
348 
349 	/* Core Lock Init */
350 	IXGBE_CORE_LOCK_INIT(adapter, device_xname(dev));
351 
352 	/* Do base PCI setup - map BAR0 */
353 	if (ixv_allocate_pci_resources(adapter, pa)) {
354 		aprint_error_dev(dev, "ixv_allocate_pci_resources() failed!\n");
355 		error = ENXIO;
356 		goto err_out;
357 	}
358 
359 	/* SYSCTL APIs */
360 	ixv_add_device_sysctls(adapter);
361 
362 	/* Set up the timer callout and workqueue */
363 	callout_init(&adapter->timer, IXGBE_CALLOUT_FLAGS);
364 	snprintf(wqname, sizeof(wqname), "%s-timer", device_xname(dev));
365 	error = workqueue_create(&adapter->timer_wq, wqname,
366 	    ixv_handle_timer, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
367 	    IXGBE_TASKLET_WQ_FLAGS);
368 	if (error) {
369 		aprint_error_dev(dev,
370 		    "could not create timer workqueue (%d)\n", error);
371 		goto err_out;
372 	}
373 
374 	/* Save off the information about this board */
375 	id = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_ID_REG);
376 	subid = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
377 	hw->vendor_id = PCI_VENDOR(id);
378 	hw->device_id = PCI_PRODUCT(id);
379 	hw->revision_id =
380 	    PCI_REVISION(pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_CLASS_REG));
381 	hw->subsystem_vendor_id = PCI_SUBSYS_VENDOR(subid);
382 	hw->subsystem_device_id = PCI_SUBSYS_ID(subid);
383 
384 	/* A subset of set_mac_type */
385 	switch (hw->device_id) {
386 	case IXGBE_DEV_ID_82599_VF:
387 		hw->mac.type = ixgbe_mac_82599_vf;
388 		str = "82599 VF";
389 		break;
390 	case IXGBE_DEV_ID_X540_VF:
391 		hw->mac.type = ixgbe_mac_X540_vf;
392 		str = "X540 VF";
393 		break;
394 	case IXGBE_DEV_ID_X550_VF:
395 		hw->mac.type = ixgbe_mac_X550_vf;
396 		str = "X550 VF";
397 		break;
398 	case IXGBE_DEV_ID_X550EM_X_VF:
399 		hw->mac.type = ixgbe_mac_X550EM_x_vf;
400 		str = "X550EM X VF";
401 		break;
402 	case IXGBE_DEV_ID_X550EM_A_VF:
403 		hw->mac.type = ixgbe_mac_X550EM_a_vf;
404 		str = "X550EM A VF";
405 		break;
406 	default:
407 		/* Shouldn't get here since probe succeeded */
408 		aprint_error_dev(dev, "Unknown device ID!\n");
409 		error = ENXIO;
410 		goto err_out;
411 		break;
412 	}
413 	aprint_normal_dev(dev, "device %s\n", str);
414 
415 	ixv_init_device_features(adapter);
416 
417 	/* Initialize the shared code */
418 	error = ixgbe_init_ops_vf(hw);
419 	if (error) {
420 		aprint_error_dev(dev, "ixgbe_init_ops_vf() failed!\n");
421 		error = EIO;
422 		goto err_out;
423 	}
424 
425 	/* Setup the mailbox */
426 	ixgbe_init_mbx_params_vf(hw);
427 
428 	/* Set the right number of segments */
429 	adapter->num_segs = IXGBE_82599_SCATTER;
430 
431 	/* Reset mbox api to 1.0 */
432 	error = hw->mac.ops.reset_hw(hw);
433 	if (error == IXGBE_ERR_RESET_FAILED)
434 		aprint_error_dev(dev, "...reset_hw() failure: Reset Failed!\n");
435 	else if (error)
436 		aprint_error_dev(dev, "...reset_hw() failed with error %d\n",
437 		    error);
438 	if (error) {
439 		error = EIO;
440 		goto err_out;
441 	}
442 
443 	error = hw->mac.ops.init_hw(hw);
444 	if (error) {
445 		aprint_error_dev(dev, "...init_hw() failed!\n");
446 		error = EIO;
447 		goto err_out;
448 	}
449 
450 	/* Negotiate mailbox API version */
451 	error = ixv_negotiate_api(adapter);
452 	if (error)
453 		aprint_normal_dev(dev,
454 		    "MBX API negotiation failed during attach!\n");
455 	switch (hw->api_version) {
456 	case ixgbe_mbox_api_10:
457 		apivstr = "1.0";
458 		break;
459 	case ixgbe_mbox_api_20:
460 		apivstr = "2.0";
461 		break;
462 	case ixgbe_mbox_api_11:
463 		apivstr = "1.1";
464 		break;
465 	case ixgbe_mbox_api_12:
466 		apivstr = "1.2";
467 		break;
468 	case ixgbe_mbox_api_13:
469 		apivstr = "1.3";
470 		break;
471 	default:
472 		apivstr = "unknown";
473 		break;
474 	}
475 	aprint_normal_dev(dev, "Mailbox API %s\n", apivstr);
476 
477 	/* If no mac address was assigned, make a random one */
478 	if (!ixv_check_ether_addr(hw->mac.addr)) {
479 		u8 addr[ETHER_ADDR_LEN];
480 		uint64_t rndval = cprng_strong64();
481 
482 		memcpy(addr, &rndval, sizeof(addr));
483 		addr[0] &= 0xFE;
484 		addr[0] |= 0x02;
485 		bcopy(addr, hw->mac.addr, sizeof(addr));
486 	}
487 
488 	/* Register for VLAN events */
489 	ether_set_vlan_cb(&adapter->osdep.ec, ixv_vlan_cb);
490 
491 	/* Sysctls for limiting the amount of work done in the taskqueues */
492 	ixv_set_sysctl_value(adapter, "rx_processing_limit",
493 	    "max number of rx packets to process",
494 	    &adapter->rx_process_limit, ixv_rx_process_limit);
495 
496 	ixv_set_sysctl_value(adapter, "tx_processing_limit",
497 	    "max number of tx packets to process",
498 	    &adapter->tx_process_limit, ixv_tx_process_limit);
499 
500 	/* Do descriptor calc and sanity checks */
501 	if (((ixv_txd * sizeof(union ixgbe_adv_tx_desc)) % DBA_ALIGN) != 0 ||
502 	    ixv_txd < MIN_TXD || ixv_txd > MAX_TXD) {
503 		aprint_error_dev(dev, "TXD config issue, using default!\n");
504 		adapter->num_tx_desc = DEFAULT_TXD;
505 	} else
506 		adapter->num_tx_desc = ixv_txd;
507 
508 	if (((ixv_rxd * sizeof(union ixgbe_adv_rx_desc)) % DBA_ALIGN) != 0 ||
509 	    ixv_rxd < MIN_RXD || ixv_rxd > MAX_RXD) {
510 		aprint_error_dev(dev, "RXD config issue, using default!\n");
511 		adapter->num_rx_desc = DEFAULT_RXD;
512 	} else
513 		adapter->num_rx_desc = ixv_rxd;
514 
515 	/* Setup MSI-X */
516 	error = ixv_configure_interrupts(adapter);
517 	if (error)
518 		goto err_out;
519 
520 	/* Allocate our TX/RX Queues */
521 	if (ixgbe_allocate_queues(adapter)) {
522 		aprint_error_dev(dev, "ixgbe_allocate_queues() failed!\n");
523 		error = ENOMEM;
524 		goto err_out;
525 	}
526 
527 	/* hw.ix defaults init */
528 	adapter->enable_aim = ixv_enable_aim;
529 
530 	adapter->txrx_use_workqueue = ixv_txrx_workqueue;
531 
532 	error = ixv_allocate_msix(adapter, pa);
533 	if (error) {
534 		aprint_error_dev(dev, "ixv_allocate_msix() failed!\n");
535 		goto err_late;
536 	}
537 
538 	/* Setup OS specific network interface */
539 	error = ixv_setup_interface(dev, adapter);
540 	if (error != 0) {
541 		aprint_error_dev(dev, "ixv_setup_interface() failed!\n");
542 		goto err_late;
543 	}
544 
545 	/* Do the stats setup */
546 	ixv_save_stats(adapter);
547 	ixv_init_stats(adapter);
548 	ixv_add_stats_sysctls(adapter);
549 
550 	if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
551 		ixgbe_netmap_attach(adapter);
552 
553 	snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_cap);
554 	aprint_verbose_dev(dev, "feature cap %s\n", buf);
555 	snprintb(buf, sizeof(buf), IXGBE_FEATURE_FLAGS, adapter->feat_en);
556 	aprint_verbose_dev(dev, "feature ena %s\n", buf);
557 
558 	INIT_DEBUGOUT("ixv_attach: end");
559 	adapter->osdep.attached = true;
560 
561 	return;
562 
563 err_late:
564 	ixgbe_free_queues(adapter);
565 err_out:
566 	ixv_free_pci_resources(adapter);
567 	IXGBE_CORE_LOCK_DESTROY(adapter);
568 
569 	return;
570 } /* ixv_attach */
571 
572 /************************************************************************
573  * ixv_detach - Device removal routine
574  *
575  *   Called when the driver is being removed.
576  *   Stops the adapter and deallocates all the resources
577  *   that were allocated for driver operation.
578  *
579  *   return 0 on success, positive on failure
580  ************************************************************************/
581 static int
582 ixv_detach(device_t dev, int flags)
583 {
584 	struct adapter	*adapter = device_private(dev);
585 	struct ixgbe_hw *hw = &adapter->hw;
586 	struct tx_ring *txr = adapter->tx_rings;
587 	struct rx_ring *rxr = adapter->rx_rings;
588 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
589 
590 	INIT_DEBUGOUT("ixv_detach: begin");
591 	if (adapter->osdep.attached == false)
592 		return 0;
593 
594 	/* Stop the interface. Callouts are stopped in it. */
595 	ixv_ifstop(adapter->ifp, 1);
596 
597 #if NVLAN > 0
598 	/* Make sure VLANs are not using driver */
599 	if (!VLAN_ATTACHED(&adapter->osdep.ec))
600 		;	/* nothing to do: no VLANs */
601 	else if ((flags & (DETACH_SHUTDOWN | DETACH_FORCE)) != 0)
602 		vlan_ifdetach(adapter->ifp);
603 	else {
604 		aprint_error_dev(dev, "VLANs in use, detach first\n");
605 		return EBUSY;
606 	}
607 #endif
608 
609 	ether_ifdetach(adapter->ifp);
610 	callout_halt(&adapter->timer, NULL);
611 	ixv_free_workqueue(adapter);
612 
613 	if (adapter->feat_en & IXGBE_FEATURE_NETMAP)
614 		netmap_detach(adapter->ifp);
615 
616 	ixv_free_pci_resources(adapter);
617 #if 0 /* XXX the NetBSD port is probably missing something here */
618 	bus_generic_detach(dev);
619 #endif
620 	if_detach(adapter->ifp);
621 	ifmedia_fini(&adapter->media);
622 	if_percpuq_destroy(adapter->ipq);
623 
624 	sysctl_teardown(&adapter->sysctllog);
625 	evcnt_detach(&adapter->efbig_tx_dma_setup);
626 	evcnt_detach(&adapter->mbuf_defrag_failed);
627 	evcnt_detach(&adapter->efbig2_tx_dma_setup);
628 	evcnt_detach(&adapter->einval_tx_dma_setup);
629 	evcnt_detach(&adapter->other_tx_dma_setup);
630 	evcnt_detach(&adapter->eagain_tx_dma_setup);
631 	evcnt_detach(&adapter->enomem_tx_dma_setup);
632 	evcnt_detach(&adapter->watchdog_events);
633 	evcnt_detach(&adapter->tso_err);
634 	evcnt_detach(&adapter->admin_irqev);
635 	evcnt_detach(&adapter->link_workev);
636 
637 	txr = adapter->tx_rings;
638 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
639 		evcnt_detach(&adapter->queues[i].irqs);
640 		evcnt_detach(&adapter->queues[i].handleq);
641 		evcnt_detach(&adapter->queues[i].req);
642 		evcnt_detach(&txr->no_desc_avail);
643 		evcnt_detach(&txr->total_packets);
644 		evcnt_detach(&txr->tso_tx);
645 #ifndef IXGBE_LEGACY_TX
646 		evcnt_detach(&txr->pcq_drops);
647 #endif
648 
649 		evcnt_detach(&rxr->rx_packets);
650 		evcnt_detach(&rxr->rx_bytes);
651 		evcnt_detach(&rxr->rx_copies);
652 		evcnt_detach(&rxr->no_jmbuf);
653 		evcnt_detach(&rxr->rx_discarded);
654 	}
655 	evcnt_detach(&stats->ipcs);
656 	evcnt_detach(&stats->l4cs);
657 	evcnt_detach(&stats->ipcs_bad);
658 	evcnt_detach(&stats->l4cs_bad);
659 
660 	/* Packet Reception Stats */
661 	evcnt_detach(&stats->vfgorc);
662 	evcnt_detach(&stats->vfgprc);
663 	evcnt_detach(&stats->vfmprc);
664 
665 	/* Packet Transmission Stats */
666 	evcnt_detach(&stats->vfgotc);
667 	evcnt_detach(&stats->vfgptc);
668 
669 	/* Mailbox Stats */
670 	evcnt_detach(&hw->mbx.stats.msgs_tx);
671 	evcnt_detach(&hw->mbx.stats.msgs_rx);
672 	evcnt_detach(&hw->mbx.stats.acks);
673 	evcnt_detach(&hw->mbx.stats.reqs);
674 	evcnt_detach(&hw->mbx.stats.rsts);
675 
676 	ixgbe_free_queues(adapter);
677 
678 	IXGBE_CORE_LOCK_DESTROY(adapter);
679 
680 	return (0);
681 } /* ixv_detach */
682 
683 /************************************************************************
684  * ixv_init_locked - Init entry point
685  *
686  *   Used in two ways: It is used by the stack as an init entry
687  *   point in network interface structure. It is also used
688  *   by the driver as a hw/sw initialization routine to get
689  *   to a consistent state.
690  *
691  *   return 0 on success, positive on failure
692  ************************************************************************/
693 static void
694 ixv_init_locked(struct adapter *adapter)
695 {
696 	struct ifnet	*ifp = adapter->ifp;
697 	device_t	dev = adapter->dev;
698 	struct ixgbe_hw *hw = &adapter->hw;
699 	struct ix_queue	*que;
700 	int		error = 0;
701 	uint32_t mask;
702 	int i;
703 
704 	INIT_DEBUGOUT("ixv_init_locked: begin");
705 	KASSERT(mutex_owned(&adapter->core_mtx));
706 	hw->adapter_stopped = FALSE;
707 	hw->mac.ops.stop_adapter(hw);
708 	callout_stop(&adapter->timer);
709 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
710 		que->disabled_count = 0;
711 
712 	adapter->max_frame_size =
713 	    ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
714 
715 	/* reprogram the RAR[0] in case user changed it. */
716 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
717 
718 	/* Get the latest mac address, User can use a LAA */
719 	memcpy(hw->mac.addr, CLLADDR(ifp->if_sadl),
720 	     IXGBE_ETH_LENGTH_OF_ADDRESS);
721 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, 1);
722 
723 	/* Prepare transmit descriptors and buffers */
724 	if (ixgbe_setup_transmit_structures(adapter)) {
725 		aprint_error_dev(dev, "Could not setup transmit structures\n");
726 		ixv_stop(adapter);
727 		return;
728 	}
729 
730 	/* Reset VF and renegotiate mailbox API version */
731 	hw->mac.ops.reset_hw(hw);
732 	hw->mac.ops.start_hw(hw);
733 	error = ixv_negotiate_api(adapter);
734 	if (error)
735 		device_printf(dev,
736 		    "Mailbox API negotiation failed in init_locked!\n");
737 
738 	ixv_initialize_transmit_units(adapter);
739 
740 	/* Setup Multicast table */
741 	ixv_set_rxfilter(adapter);
742 
743 	/*
744 	 * Determine the correct mbuf pool
745 	 * for doing jumbo/headersplit
746 	 */
747 	if (adapter->max_frame_size <= MCLBYTES)
748 		adapter->rx_mbuf_sz = MCLBYTES;
749 	else
750 		adapter->rx_mbuf_sz = MJUMPAGESIZE;
751 
752 	/* Prepare receive descriptors and buffers */
753 	if (ixgbe_setup_receive_structures(adapter)) {
754 		device_printf(dev, "Could not setup receive structures\n");
755 		ixv_stop(adapter);
756 		return;
757 	}
758 
759 	/* Configure RX settings */
760 	ixv_initialize_receive_units(adapter);
761 
762 	/* Initialize variable holding task enqueue requests interrupts */
763 	adapter->task_requests = 0;
764 
765 	/* Set up VLAN offload and filter */
766 	ixv_setup_vlan_support(adapter);
767 
768 	/* Set up MSI-X routing */
769 	ixv_configure_ivars(adapter);
770 
771 	/* Set up auto-mask */
772 	mask = (1 << adapter->vector);
773 	for (i = 0, que = adapter->queues; i < adapter->num_queues; i++, que++)
774 		mask |= (1 << que->msix);
775 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, mask);
776 
777 	/* Set moderation on the Link interrupt */
778 	ixv_eitr_write(adapter, adapter->vector, IXGBE_LINK_ITR);
779 
780 	/* Stats init */
781 	ixv_init_stats(adapter);
782 
783 	/* Config/Enable Link */
784 	hw->mac.get_link_status = TRUE;
785 	hw->mac.ops.check_link(hw, &adapter->link_speed, &adapter->link_up,
786 	    FALSE);
787 
788 	/* Start watchdog */
789 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
790 	atomic_and_uint(&adapter->timer_pending, ~1);
791 
792 	/* OK to schedule workqueues. */
793 	adapter->schedule_wqs_ok = true;
794 
795 	/* And now turn on interrupts */
796 	ixv_enable_intr(adapter);
797 
798 	/* Update saved flags. See ixgbe_ifflags_cb() */
799 	adapter->if_flags = ifp->if_flags;
800 	adapter->ec_capenable = adapter->osdep.ec.ec_capenable;
801 
802 	/* Now inform the stack we're ready */
803 	ifp->if_flags |= IFF_RUNNING;
804 	ifp->if_flags &= ~IFF_OACTIVE;
805 
806 	return;
807 } /* ixv_init_locked */
808 
809 /************************************************************************
810  * ixv_enable_queue
811  ************************************************************************/
812 static inline void
813 ixv_enable_queue(struct adapter *adapter, u32 vector)
814 {
815 	struct ixgbe_hw *hw = &adapter->hw;
816 	struct ix_queue *que = &adapter->queues[vector];
817 	u32		queue = 1UL << vector;
818 	u32		mask;
819 
820 	mutex_enter(&que->dc_mtx);
821 	if (que->disabled_count > 0 && --que->disabled_count > 0)
822 		goto out;
823 
824 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
825 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
826 out:
827 	mutex_exit(&que->dc_mtx);
828 } /* ixv_enable_queue */
829 
830 /************************************************************************
831  * ixv_disable_queue
832  ************************************************************************/
833 static inline void
834 ixv_disable_queue(struct adapter *adapter, u32 vector)
835 {
836 	struct ixgbe_hw *hw = &adapter->hw;
837 	struct ix_queue *que = &adapter->queues[vector];
838 	u32		queue = 1UL << vector;
839 	u32		mask;
840 
841 	mutex_enter(&que->dc_mtx);
842 	if (que->disabled_count++ > 0)
843 		goto  out;
844 
845 	mask = (IXGBE_EIMS_RTX_QUEUE & queue);
846 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, mask);
847 out:
848 	mutex_exit(&que->dc_mtx);
849 } /* ixv_disable_queue */
850 
851 #if 0
852 static inline void
853 ixv_rearm_queues(struct adapter *adapter, u64 queues)
854 {
855 	u32 mask = (IXGBE_EIMS_RTX_QUEUE & queues);
856 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEICS, mask);
857 } /* ixv_rearm_queues */
858 #endif
859 
860 
861 /************************************************************************
862  * ixv_msix_que - MSI-X Queue Interrupt Service routine
863  ************************************************************************/
864 static int
865 ixv_msix_que(void *arg)
866 {
867 	struct ix_queue	*que = arg;
868 	struct adapter	*adapter = que->adapter;
869 	struct tx_ring	*txr = que->txr;
870 	struct rx_ring	*rxr = que->rxr;
871 	bool		more;
872 	u32		newitr = 0;
873 
874 	ixv_disable_queue(adapter, que->msix);
875 	++que->irqs.ev_count;
876 
877 #ifdef __NetBSD__
878 	/* Don't run ixgbe_rxeof in interrupt context */
879 	more = true;
880 #else
881 	more = ixgbe_rxeof(que);
882 #endif
883 
884 	IXGBE_TX_LOCK(txr);
885 	ixgbe_txeof(txr);
886 	IXGBE_TX_UNLOCK(txr);
887 
888 	/* Do AIM now? */
889 
890 	if (adapter->enable_aim == false)
891 		goto no_calc;
892 	/*
893 	 * Do Adaptive Interrupt Moderation:
894 	 *  - Write out last calculated setting
895 	 *  - Calculate based on average size over
896 	 *    the last interval.
897 	 */
898 	if (que->eitr_setting)
899 		ixv_eitr_write(adapter, que->msix, que->eitr_setting);
900 
901 	que->eitr_setting = 0;
902 
903 	/* Idle, do nothing */
904 	if ((txr->bytes == 0) && (rxr->bytes == 0))
905 		goto no_calc;
906 
907 	if ((txr->bytes) && (txr->packets))
908 		newitr = txr->bytes/txr->packets;
909 	if ((rxr->bytes) && (rxr->packets))
910 		newitr = uimax(newitr, (rxr->bytes / rxr->packets));
911 	newitr += 24; /* account for hardware frame, crc */
912 
913 	/* set an upper boundary */
914 	newitr = uimin(newitr, 3000);
915 
916 	/* Be nice to the mid range */
917 	if ((newitr > 300) && (newitr < 1200))
918 		newitr = (newitr / 3);
919 	else
920 		newitr = (newitr / 2);
921 
922 	/*
923 	 * When RSC is used, ITR interval must be larger than RSC_DELAY.
924 	 * Currently, we use 2us for RSC_DELAY. The minimum value is always
925 	 * greater than 2us on 100M (and 10M?(not documented)), but it's not
926 	 * on 1G and higher.
927 	 */
928 	if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
929 	    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
930 		if (newitr < IXGBE_MIN_RSC_EITR_10G1G)
931 			newitr = IXGBE_MIN_RSC_EITR_10G1G;
932 	}
933 
934 	/* save for next interrupt */
935 	que->eitr_setting = newitr;
936 
937 	/* Reset state */
938 	txr->bytes = 0;
939 	txr->packets = 0;
940 	rxr->bytes = 0;
941 	rxr->packets = 0;
942 
943 no_calc:
944 	if (more)
945 		softint_schedule(que->que_si);
946 	else /* Re-enable this interrupt */
947 		ixv_enable_queue(adapter, que->msix);
948 
949 	return 1;
950 } /* ixv_msix_que */
951 
952 /************************************************************************
953  * ixv_msix_mbx
954  ************************************************************************/
955 static int
956 ixv_msix_mbx(void *arg)
957 {
958 	struct adapter	*adapter = arg;
959 	struct ixgbe_hw *hw = &adapter->hw;
960 
961 	++adapter->admin_irqev.ev_count;
962 	/* NetBSD: We use auto-clear, so it's not required to write VTEICR */
963 
964 	/* Link status change */
965 	hw->mac.get_link_status = TRUE;
966 	atomic_or_32(&adapter->task_requests, IXGBE_REQUEST_TASK_MBX);
967 	ixv_schedule_admin_tasklet(adapter);
968 
969 	return 1;
970 } /* ixv_msix_mbx */
971 
972 static void
973 ixv_eitr_write(struct adapter *adapter, uint32_t index, uint32_t itr)
974 {
975 
976 	/*
977 	 * Newer devices than 82598 have VF function, so this function is
978 	 * simple.
979 	 */
980 	itr |= IXGBE_EITR_CNT_WDIS;
981 
982 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEITR(index), itr);
983 }
984 
985 
986 /************************************************************************
987  * ixv_media_status - Media Ioctl callback
988  *
989  *   Called whenever the user queries the status of
990  *   the interface using ifconfig.
991  ************************************************************************/
992 static void
993 ixv_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
994 {
995 	struct adapter *adapter = ifp->if_softc;
996 
997 	INIT_DEBUGOUT("ixv_media_status: begin");
998 	ixv_update_link_status(adapter);
999 
1000 	ifmr->ifm_status = IFM_AVALID;
1001 	ifmr->ifm_active = IFM_ETHER;
1002 
1003 	if (adapter->link_active != LINK_STATE_UP) {
1004 		ifmr->ifm_active |= IFM_NONE;
1005 		return;
1006 	}
1007 
1008 	ifmr->ifm_status |= IFM_ACTIVE;
1009 
1010 	switch (adapter->link_speed) {
1011 		case IXGBE_LINK_SPEED_10GB_FULL:
1012 			ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
1013 			break;
1014 		case IXGBE_LINK_SPEED_5GB_FULL:
1015 			ifmr->ifm_active |= IFM_5000_T | IFM_FDX;
1016 			break;
1017 		case IXGBE_LINK_SPEED_2_5GB_FULL:
1018 			ifmr->ifm_active |= IFM_2500_T | IFM_FDX;
1019 			break;
1020 		case IXGBE_LINK_SPEED_1GB_FULL:
1021 			ifmr->ifm_active |= IFM_1000_T | IFM_FDX;
1022 			break;
1023 		case IXGBE_LINK_SPEED_100_FULL:
1024 			ifmr->ifm_active |= IFM_100_TX | IFM_FDX;
1025 			break;
1026 		case IXGBE_LINK_SPEED_10_FULL:
1027 			ifmr->ifm_active |= IFM_10_T | IFM_FDX;
1028 			break;
1029 	}
1030 
1031 	ifp->if_baudrate = ifmedia_baudrate(ifmr->ifm_active);
1032 } /* ixv_media_status */
1033 
1034 /************************************************************************
1035  * ixv_media_change - Media Ioctl callback
1036  *
1037  *   Called when the user changes speed/duplex using
1038  *   media/mediopt option with ifconfig.
1039  ************************************************************************/
1040 static int
1041 ixv_media_change(struct ifnet *ifp)
1042 {
1043 	struct adapter *adapter = ifp->if_softc;
1044 	struct ifmedia *ifm = &adapter->media;
1045 
1046 	INIT_DEBUGOUT("ixv_media_change: begin");
1047 
1048 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1049 		return (EINVAL);
1050 
1051 	switch (IFM_SUBTYPE(ifm->ifm_media)) {
1052 	case IFM_AUTO:
1053 		break;
1054 	default:
1055 		device_printf(adapter->dev, "Only auto media type\n");
1056 		return (EINVAL);
1057 	}
1058 
1059 	return (0);
1060 } /* ixv_media_change */
1061 
1062 static void
1063 ixv_schedule_admin_tasklet(struct adapter *adapter)
1064 {
1065 	if (adapter->schedule_wqs_ok) {
1066 		if (!adapter->admin_pending) {
1067 			atomic_or_uint(&adapter->admin_pending, 1);
1068 			workqueue_enqueue(adapter->admin_wq,
1069 			    &adapter->admin_wc, NULL);
1070 		}
1071 	}
1072 }
1073 
1074 /************************************************************************
1075  * ixv_negotiate_api
1076  *
1077  *   Negotiate the Mailbox API with the PF;
1078  *   start with the most featured API first.
1079  ************************************************************************/
1080 static int
1081 ixv_negotiate_api(struct adapter *adapter)
1082 {
1083 	struct ixgbe_hw *hw = &adapter->hw;
1084 	int		mbx_api[] = { ixgbe_mbox_api_13,
1085 				      ixgbe_mbox_api_12,
1086 				      ixgbe_mbox_api_11,
1087 				      ixgbe_mbox_api_10,
1088 				      ixgbe_mbox_api_unknown };
1089 	int		i = 0;
1090 
1091 	while (mbx_api[i] != ixgbe_mbox_api_unknown) {
1092 		if (ixgbevf_negotiate_api_version(hw, mbx_api[i]) == 0)
1093 			return (0);
1094 		i++;
1095 	}
1096 
1097 	return (EINVAL);
1098 } /* ixv_negotiate_api */
1099 
1100 
1101 /************************************************************************
1102  * ixv_set_rxfilter - Multicast Update
1103  *
1104  *   Called whenever multicast address list is updated.
1105  ************************************************************************/
1106 static int
1107 ixv_set_rxfilter(struct adapter *adapter)
1108 {
1109 	u8	mta[IXGBE_MAX_VF_MC * IXGBE_ETH_LENGTH_OF_ADDRESS];
1110 	struct ifnet		*ifp = adapter->ifp;
1111 	struct ixgbe_hw		*hw = &adapter->hw;
1112 	u8			*update_ptr;
1113 	int			mcnt = 0;
1114 	struct ethercom		*ec = &adapter->osdep.ec;
1115 	struct ether_multi	*enm;
1116 	struct ether_multistep	step;
1117 	bool			overflow = false;
1118 	int			error, rc = 0;
1119 
1120 	KASSERT(mutex_owned(&adapter->core_mtx));
1121 	IOCTL_DEBUGOUT("ixv_set_rxfilter: begin");
1122 
1123 	/* 1: For PROMISC */
1124 	if (ifp->if_flags & IFF_PROMISC) {
1125 		error = hw->mac.ops.update_xcast_mode(hw,
1126 		    IXGBEVF_XCAST_MODE_PROMISC);
1127 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1128 			device_printf(adapter->dev,
1129 			    "this interface is not trusted\n");
1130 			error = EPERM;
1131 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1132 			device_printf(adapter->dev,
1133 			    "the PF doesn't support promisc mode\n");
1134 			error = EOPNOTSUPP;
1135 		} else if (error == IXGBE_ERR_NOT_IN_PROMISC) {
1136 			device_printf(adapter->dev,
1137 			    "the PF may not in promisc mode\n");
1138 			error = EINVAL;
1139 		} else if (error) {
1140 			device_printf(adapter->dev,
1141 			    "failed to set promisc mode. error = %d\n",
1142 			    error);
1143 			error = EIO;
1144 		} else
1145 			return 0;
1146 		rc = error;
1147 	}
1148 
1149 	/* 2: For ALLMULTI or normal */
1150 	ETHER_LOCK(ec);
1151 	ETHER_FIRST_MULTI(step, ec, enm);
1152 	while (enm != NULL) {
1153 		if ((mcnt >= IXGBE_MAX_VF_MC) ||
1154 		    (memcmp(enm->enm_addrlo, enm->enm_addrhi,
1155 			ETHER_ADDR_LEN) != 0)) {
1156 			overflow = true;
1157 			break;
1158 		}
1159 		bcopy(enm->enm_addrlo,
1160 		    &mta[mcnt * IXGBE_ETH_LENGTH_OF_ADDRESS],
1161 		    IXGBE_ETH_LENGTH_OF_ADDRESS);
1162 		mcnt++;
1163 		ETHER_NEXT_MULTI(step, enm);
1164 	}
1165 	ETHER_UNLOCK(ec);
1166 
1167 	/* 3: For ALLMULTI */
1168 	if (overflow) {
1169 		error = hw->mac.ops.update_xcast_mode(hw,
1170 		    IXGBEVF_XCAST_MODE_ALLMULTI);
1171 		if (error == IXGBE_ERR_NOT_TRUSTED) {
1172 			device_printf(adapter->dev,
1173 			    "this interface is not trusted\n");
1174 			error = EPERM;
1175 		} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
1176 			device_printf(adapter->dev,
1177 			    "the PF doesn't support allmulti mode\n");
1178 			error = EOPNOTSUPP;
1179 		} else if (error) {
1180 			device_printf(adapter->dev,
1181 			    "number of Ethernet multicast addresses "
1182 			    "exceeds the limit (%d). error = %d\n",
1183 			    IXGBE_MAX_VF_MC, error);
1184 			error = ENOSPC;
1185 		} else {
1186 			ETHER_LOCK(ec);
1187 			ec->ec_flags |= ETHER_F_ALLMULTI;
1188 			ETHER_UNLOCK(ec);
1189 			return rc; /* Promisc might have failed */
1190 		}
1191 
1192 		if (rc == 0)
1193 			rc = error;
1194 
1195 		/* Continue to update the multicast table as many as we can */
1196 	}
1197 
1198 	/* 4: For normal operation */
1199 	error = hw->mac.ops.update_xcast_mode(hw, IXGBEVF_XCAST_MODE_MULTI);
1200 	if ((error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) || (error == 0)) {
1201 		/* Normal operation */
1202 		ETHER_LOCK(ec);
1203 		ec->ec_flags &= ~ETHER_F_ALLMULTI;
1204 		ETHER_UNLOCK(ec);
1205 		error = 0;
1206 	} else if (error) {
1207 		device_printf(adapter->dev,
1208 		    "failed to set Ethernet multicast address "
1209 		    "operation to normal. error = %d\n", error);
1210 	}
1211 
1212 	update_ptr = mta;
1213 
1214 	error = adapter->hw.mac.ops.update_mc_addr_list(&adapter->hw,
1215 	    update_ptr, mcnt, ixv_mc_array_itr, TRUE);
1216 	if (rc == 0)
1217 		rc = error;
1218 
1219 	return rc;
1220 } /* ixv_set_rxfilter */
1221 
1222 /************************************************************************
1223  * ixv_mc_array_itr
1224  *
1225  *   An iterator function needed by the multicast shared code.
1226  *   It feeds the shared code routine the addresses in the
1227  *   array of ixv_set_rxfilter() one by one.
1228  ************************************************************************/
1229 static u8 *
1230 ixv_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq)
1231 {
1232 	u8 *addr = *update_ptr;
1233 	u8 *newptr;
1234 
1235 	*vmdq = 0;
1236 
1237 	newptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
1238 	*update_ptr = newptr;
1239 
1240 	return addr;
1241 } /* ixv_mc_array_itr */
1242 
1243 /************************************************************************
1244  * ixv_local_timer - Timer routine
1245  *
1246  *   Checks for link status, updates statistics,
1247  *   and runs the watchdog check.
1248  ************************************************************************/
1249 static void
1250 ixv_local_timer(void *arg)
1251 {
1252 	struct adapter *adapter = arg;
1253 
1254 	if (adapter->schedule_wqs_ok) {
1255 		if (!adapter->timer_pending) {
1256 			atomic_or_uint(&adapter->timer_pending, 1);
1257 			workqueue_enqueue(adapter->timer_wq,
1258 			    &adapter->timer_wc, NULL);
1259 		}
1260 	}
1261 }
1262 
1263 static void
1264 ixv_handle_timer(struct work *wk, void *context)
1265 {
1266 	struct adapter	*adapter = context;
1267 	device_t	dev = adapter->dev;
1268 	struct ix_queue	*que = adapter->queues;
1269 	u64		queues = 0;
1270 	u64		v0, v1, v2, v3, v4, v5, v6, v7;
1271 	int		hung = 0;
1272 	int		i;
1273 
1274 	IXGBE_CORE_LOCK(adapter);
1275 
1276 	if (ixv_check_link(adapter)) {
1277 		ixv_init_locked(adapter);
1278 		IXGBE_CORE_UNLOCK(adapter);
1279 		return;
1280 	}
1281 
1282 	/* Stats Update */
1283 	ixv_update_stats(adapter);
1284 
1285 	/* Update some event counters */
1286 	v0 = v1 = v2 = v3 = v4 = v5 = v6 = v7 = 0;
1287 	que = adapter->queues;
1288 	for (i = 0; i < adapter->num_queues; i++, que++) {
1289 		struct tx_ring	*txr = que->txr;
1290 
1291 		v0 += txr->q_efbig_tx_dma_setup;
1292 		v1 += txr->q_mbuf_defrag_failed;
1293 		v2 += txr->q_efbig2_tx_dma_setup;
1294 		v3 += txr->q_einval_tx_dma_setup;
1295 		v4 += txr->q_other_tx_dma_setup;
1296 		v5 += txr->q_eagain_tx_dma_setup;
1297 		v6 += txr->q_enomem_tx_dma_setup;
1298 		v7 += txr->q_tso_err;
1299 	}
1300 	adapter->efbig_tx_dma_setup.ev_count = v0;
1301 	adapter->mbuf_defrag_failed.ev_count = v1;
1302 	adapter->efbig2_tx_dma_setup.ev_count = v2;
1303 	adapter->einval_tx_dma_setup.ev_count = v3;
1304 	adapter->other_tx_dma_setup.ev_count = v4;
1305 	adapter->eagain_tx_dma_setup.ev_count = v5;
1306 	adapter->enomem_tx_dma_setup.ev_count = v6;
1307 	adapter->tso_err.ev_count = v7;
1308 
1309 	/*
1310 	 * Check the TX queues status
1311 	 *	- mark hung queues so we don't schedule on them
1312 	 *	- watchdog only if all queues show hung
1313 	 */
1314 	que = adapter->queues;
1315 	for (i = 0; i < adapter->num_queues; i++, que++) {
1316 		/* Keep track of queues with work for soft irq */
1317 		if (que->txr->busy)
1318 			queues |= ((u64)1 << que->me);
1319 		/*
1320 		 * Each time txeof runs without cleaning, but there
1321 		 * are uncleaned descriptors it increments busy. If
1322 		 * we get to the MAX we declare it hung.
1323 		 */
1324 		if (que->busy == IXGBE_QUEUE_HUNG) {
1325 			++hung;
1326 			/* Mark the queue as inactive */
1327 			adapter->active_queues &= ~((u64)1 << que->me);
1328 			continue;
1329 		} else {
1330 			/* Check if we've come back from hung */
1331 			if ((adapter->active_queues & ((u64)1 << que->me)) == 0)
1332 				adapter->active_queues |= ((u64)1 << que->me);
1333 		}
1334 		if (que->busy >= IXGBE_MAX_TX_BUSY) {
1335 			device_printf(dev,
1336 			    "Warning queue %d appears to be hung!\n", i);
1337 			que->txr->busy = IXGBE_QUEUE_HUNG;
1338 			++hung;
1339 		}
1340 	}
1341 
1342 	/* Only truly watchdog if all queues show hung */
1343 	if (hung == adapter->num_queues)
1344 		goto watchdog;
1345 #if 0
1346 	else if (queues != 0) { /* Force an IRQ on queues with work */
1347 		ixv_rearm_queues(adapter, queues);
1348 	}
1349 #endif
1350 
1351 	atomic_and_uint(&adapter->timer_pending, ~1);
1352 	IXGBE_CORE_UNLOCK(adapter);
1353 	callout_reset(&adapter->timer, hz, ixv_local_timer, adapter);
1354 
1355 	return;
1356 
1357 watchdog:
1358 	device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
1359 	adapter->ifp->if_flags &= ~IFF_RUNNING;
1360 	adapter->watchdog_events.ev_count++;
1361 	ixv_init_locked(adapter);
1362 	IXGBE_CORE_UNLOCK(adapter);
1363 } /* ixv_handle_timer */
1364 
1365 /************************************************************************
1366  * ixv_update_link_status - Update OS on link state
1367  *
1368  * Note: Only updates the OS on the cached link state.
1369  *	 The real check of the hardware only happens with
1370  *	 a link interrupt.
1371  ************************************************************************/
1372 static void
1373 ixv_update_link_status(struct adapter *adapter)
1374 {
1375 	struct ifnet *ifp = adapter->ifp;
1376 	device_t     dev = adapter->dev;
1377 
1378 	KASSERT(mutex_owned(&adapter->core_mtx));
1379 
1380 	if (adapter->link_up) {
1381 		if (adapter->link_active != LINK_STATE_UP) {
1382 			if (bootverbose) {
1383 				const char *bpsmsg;
1384 
1385 				switch (adapter->link_speed) {
1386 				case IXGBE_LINK_SPEED_10GB_FULL:
1387 					bpsmsg = "10 Gbps";
1388 					break;
1389 				case IXGBE_LINK_SPEED_5GB_FULL:
1390 					bpsmsg = "5 Gbps";
1391 					break;
1392 				case IXGBE_LINK_SPEED_2_5GB_FULL:
1393 					bpsmsg = "2.5 Gbps";
1394 					break;
1395 				case IXGBE_LINK_SPEED_1GB_FULL:
1396 					bpsmsg = "1 Gbps";
1397 					break;
1398 				case IXGBE_LINK_SPEED_100_FULL:
1399 					bpsmsg = "100 Mbps";
1400 					break;
1401 				case IXGBE_LINK_SPEED_10_FULL:
1402 					bpsmsg = "10 Mbps";
1403 					break;
1404 				default:
1405 					bpsmsg = "unknown speed";
1406 					break;
1407 				}
1408 				device_printf(dev, "Link is up %s %s \n",
1409 				    bpsmsg, "Full Duplex");
1410 			}
1411 			adapter->link_active = LINK_STATE_UP;
1412 			if_link_state_change(ifp, LINK_STATE_UP);
1413 		}
1414 	} else {
1415 		/*
1416 		 * Do it when link active changes to DOWN. i.e.
1417 		 * a) LINK_STATE_UNKNOWN -> LINK_STATE_DOWN
1418 		 * b) LINK_STATE_UP	 -> LINK_STATE_DOWN
1419 		 */
1420 		if (adapter->link_active != LINK_STATE_DOWN) {
1421 			if (bootverbose)
1422 				device_printf(dev, "Link is Down\n");
1423 			if_link_state_change(ifp, LINK_STATE_DOWN);
1424 			adapter->link_active = LINK_STATE_DOWN;
1425 		}
1426 	}
1427 } /* ixv_update_link_status */
1428 
1429 
1430 /************************************************************************
1431  * ixv_stop - Stop the hardware
1432  *
1433  *   Disables all traffic on the adapter by issuing a
1434  *   global reset on the MAC and deallocates TX/RX buffers.
1435  ************************************************************************/
1436 static void
1437 ixv_ifstop(struct ifnet *ifp, int disable)
1438 {
1439 	struct adapter *adapter = ifp->if_softc;
1440 
1441 	IXGBE_CORE_LOCK(adapter);
1442 	ixv_stop(adapter);
1443 	IXGBE_CORE_UNLOCK(adapter);
1444 
1445 	workqueue_wait(adapter->admin_wq, &adapter->admin_wc);
1446 	atomic_and_uint(&adapter->admin_pending, ~1);
1447 	workqueue_wait(adapter->timer_wq, &adapter->timer_wc);
1448 	atomic_and_uint(&adapter->timer_pending, ~1);
1449 }
1450 
1451 static void
1452 ixv_stop(void *arg)
1453 {
1454 	struct ifnet	*ifp;
1455 	struct adapter	*adapter = arg;
1456 	struct ixgbe_hw *hw = &adapter->hw;
1457 
1458 	ifp = adapter->ifp;
1459 
1460 	KASSERT(mutex_owned(&adapter->core_mtx));
1461 
1462 	INIT_DEBUGOUT("ixv_stop: begin\n");
1463 	ixv_disable_intr(adapter);
1464 
1465 	/* Tell the stack that the interface is no longer active */
1466 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1467 
1468 	hw->mac.ops.reset_hw(hw);
1469 	adapter->hw.adapter_stopped = FALSE;
1470 	hw->mac.ops.stop_adapter(hw);
1471 	callout_stop(&adapter->timer);
1472 
1473 	/* Don't schedule workqueues. */
1474 	adapter->schedule_wqs_ok = false;
1475 
1476 	/* reprogram the RAR[0] in case user changed it. */
1477 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
1478 
1479 	return;
1480 } /* ixv_stop */
1481 
1482 
1483 /************************************************************************
1484  * ixv_allocate_pci_resources
1485  ************************************************************************/
1486 static int
1487 ixv_allocate_pci_resources(struct adapter *adapter,
1488     const struct pci_attach_args *pa)
1489 {
1490 	pcireg_t	memtype, csr;
1491 	device_t	dev = adapter->dev;
1492 	bus_addr_t addr;
1493 	int flags;
1494 
1495 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, PCI_BAR(0));
1496 	switch (memtype) {
1497 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
1498 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
1499 		adapter->osdep.mem_bus_space_tag = pa->pa_memt;
1500 		if (pci_mapreg_info(pa->pa_pc, pa->pa_tag, PCI_BAR(0),
1501 		      memtype, &addr, &adapter->osdep.mem_size, &flags) != 0)
1502 			goto map_err;
1503 		if ((flags & BUS_SPACE_MAP_PREFETCHABLE) != 0) {
1504 			aprint_normal_dev(dev, "clearing prefetchable bit\n");
1505 			flags &= ~BUS_SPACE_MAP_PREFETCHABLE;
1506 		}
1507 		if (bus_space_map(adapter->osdep.mem_bus_space_tag, addr,
1508 		     adapter->osdep.mem_size, flags,
1509 		     &adapter->osdep.mem_bus_space_handle) != 0) {
1510 map_err:
1511 			adapter->osdep.mem_size = 0;
1512 			aprint_error_dev(dev, "unable to map BAR0\n");
1513 			return ENXIO;
1514 		}
1515 		/*
1516 		 * Enable address decoding for memory range in case it's not
1517 		 * set.
1518 		 */
1519 		csr = pci_conf_read(pa->pa_pc, pa->pa_tag,
1520 		    PCI_COMMAND_STATUS_REG);
1521 		csr |= PCI_COMMAND_MEM_ENABLE;
1522 		pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
1523 		    csr);
1524 		break;
1525 	default:
1526 		aprint_error_dev(dev, "unexpected type on BAR0\n");
1527 		return ENXIO;
1528 	}
1529 
1530 	/* Pick up the tuneable queues */
1531 	adapter->num_queues = ixv_num_queues;
1532 
1533 	return (0);
1534 } /* ixv_allocate_pci_resources */
1535 
1536 static void
1537 ixv_free_workqueue(struct adapter *adapter)
1538 {
1539 	struct ix_queue *que = adapter->queues;
1540 	struct tx_ring *txr = adapter->tx_rings;
1541 	int i;
1542 
1543 	for (i = 0; i < adapter->num_queues; i++, que++, txr++) {
1544 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX)) {
1545 			if (txr->txr_si != NULL)
1546 				softint_disestablish(txr->txr_si);
1547 		}
1548 		if (que->que_si != NULL)
1549 			softint_disestablish(que->que_si);
1550 	}
1551 	if (adapter->txr_wq != NULL)
1552 		workqueue_destroy(adapter->txr_wq);
1553 	if (adapter->txr_wq_enqueued != NULL)
1554 		percpu_free(adapter->txr_wq_enqueued, sizeof(u_int));
1555 	if (adapter->que_wq != NULL)
1556 		workqueue_destroy(adapter->que_wq);
1557 
1558 	/* Drain the Mailbox(link) queue */
1559 	if (adapter->admin_wq != NULL) {
1560 		workqueue_destroy(adapter->admin_wq);
1561 		adapter->admin_wq = NULL;
1562 	}
1563 	if (adapter->timer_wq != NULL) {
1564 		workqueue_destroy(adapter->timer_wq);
1565 		adapter->timer_wq = NULL;
1566 	}
1567 } /* ixv_free_workqueue */
1568 
1569 /************************************************************************
1570  * ixv_free_pci_resources
1571  ************************************************************************/
1572 static void
1573 ixv_free_pci_resources(struct adapter * adapter)
1574 {
1575 	struct		ix_queue *que = adapter->queues;
1576 	int		rid;
1577 
1578 	/*
1579 	 *  Release all msix queue resources:
1580 	 */
1581 	for (int i = 0; i < adapter->num_queues; i++, que++) {
1582 		if (que->res != NULL)
1583 			pci_intr_disestablish(adapter->osdep.pc,
1584 			    adapter->osdep.ihs[i]);
1585 	}
1586 
1587 
1588 	/* Clean the Mailbox interrupt last */
1589 	rid = adapter->vector;
1590 
1591 	if (adapter->osdep.ihs[rid] != NULL) {
1592 		pci_intr_disestablish(adapter->osdep.pc,
1593 		    adapter->osdep.ihs[rid]);
1594 		adapter->osdep.ihs[rid] = NULL;
1595 	}
1596 
1597 	pci_intr_release(adapter->osdep.pc, adapter->osdep.intrs,
1598 	    adapter->osdep.nintrs);
1599 
1600 	if (adapter->osdep.mem_size != 0) {
1601 		bus_space_unmap(adapter->osdep.mem_bus_space_tag,
1602 		    adapter->osdep.mem_bus_space_handle,
1603 		    adapter->osdep.mem_size);
1604 	}
1605 
1606 	return;
1607 } /* ixv_free_pci_resources */
1608 
1609 /************************************************************************
1610  * ixv_setup_interface
1611  *
1612  *   Setup networking device structure and register an interface.
1613  ************************************************************************/
1614 static int
1615 ixv_setup_interface(device_t dev, struct adapter *adapter)
1616 {
1617 	struct ethercom *ec = &adapter->osdep.ec;
1618 	struct ifnet   *ifp;
1619 	int rv;
1620 
1621 	INIT_DEBUGOUT("ixv_setup_interface: begin");
1622 
1623 	ifp = adapter->ifp = &ec->ec_if;
1624 	strlcpy(ifp->if_xname, device_xname(dev), IFNAMSIZ);
1625 	ifp->if_baudrate = IF_Gbps(10);
1626 	ifp->if_init = ixv_init;
1627 	ifp->if_stop = ixv_ifstop;
1628 	ifp->if_softc = adapter;
1629 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1630 #ifdef IXGBE_MPSAFE
1631 	ifp->if_extflags = IFEF_MPSAFE;
1632 #endif
1633 	ifp->if_ioctl = ixv_ioctl;
1634 	if (adapter->feat_en & IXGBE_FEATURE_LEGACY_TX) {
1635 #if 0
1636 		ixv_start_locked = ixgbe_legacy_start_locked;
1637 #endif
1638 	} else {
1639 		ifp->if_transmit = ixgbe_mq_start;
1640 #if 0
1641 		ixv_start_locked = ixgbe_mq_start_locked;
1642 #endif
1643 	}
1644 	ifp->if_start = ixgbe_legacy_start;
1645 	IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 2);
1646 	IFQ_SET_READY(&ifp->if_snd);
1647 
1648 	rv = if_initialize(ifp);
1649 	if (rv != 0) {
1650 		aprint_error_dev(dev, "if_initialize failed(%d)\n", rv);
1651 		return rv;
1652 	}
1653 	adapter->ipq = if_percpuq_create(&adapter->osdep.ec.ec_if);
1654 	ether_ifattach(ifp, adapter->hw.mac.addr);
1655 	aprint_normal_dev(dev, "Ethernet address %s\n",
1656 	    ether_sprintf(adapter->hw.mac.addr));
1657 	/*
1658 	 * We use per TX queue softint, so if_deferred_start_init() isn't
1659 	 * used.
1660 	 */
1661 	ether_set_ifflags_cb(ec, ixv_ifflags_cb);
1662 
1663 	adapter->max_frame_size = ifp->if_mtu + IXGBE_MTU_HDR;
1664 
1665 	/*
1666 	 * Tell the upper layer(s) we support long frames.
1667 	 */
1668 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
1669 
1670 	/* Set capability flags */
1671 	ifp->if_capabilities |= IFCAP_HWCSUM
1672 			     |	IFCAP_TSOv4
1673 			     |	IFCAP_TSOv6;
1674 	ifp->if_capenable = 0;
1675 
1676 	ec->ec_capabilities |= ETHERCAP_VLAN_HWFILTER
1677 			    |  ETHERCAP_VLAN_HWTAGGING
1678 			    |  ETHERCAP_VLAN_HWCSUM
1679 			    |  ETHERCAP_JUMBO_MTU
1680 			    |  ETHERCAP_VLAN_MTU;
1681 
1682 	/* Enable the above capabilities by default */
1683 	ec->ec_capenable = ec->ec_capabilities;
1684 
1685 	/* Don't enable LRO by default */
1686 #if 0
1687 	/* NetBSD doesn't support LRO yet */
1688 	ifp->if_capabilities |= IFCAP_LRO;
1689 #endif
1690 
1691 	/*
1692 	 * Specify the media types supported by this adapter and register
1693 	 * callbacks to update media and link information
1694 	 */
1695 	ec->ec_ifmedia = &adapter->media;
1696 	ifmedia_init_with_lock(&adapter->media, IFM_IMASK, ixv_media_change,
1697 	    ixv_media_status, &adapter->core_mtx);
1698 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1699 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
1700 
1701 	if_register(ifp);
1702 
1703 	return 0;
1704 } /* ixv_setup_interface */
1705 
1706 
1707 /************************************************************************
1708  * ixv_initialize_transmit_units - Enable transmit unit.
1709  ************************************************************************/
1710 static void
1711 ixv_initialize_transmit_units(struct adapter *adapter)
1712 {
1713 	struct tx_ring	*txr = adapter->tx_rings;
1714 	struct ixgbe_hw	*hw = &adapter->hw;
1715 	int i;
1716 
1717 	for (i = 0; i < adapter->num_queues; i++, txr++) {
1718 		u64 tdba = txr->txdma.dma_paddr;
1719 		u32 txctrl, txdctl;
1720 		int j = txr->me;
1721 
1722 		/* Set WTHRESH to 8, burst writeback */
1723 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1724 		txdctl |= (8 << 16);
1725 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1726 
1727 		/* Set the HW Tx Head and Tail indices */
1728 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDH(j), 0);
1729 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_VFTDT(j), 0);
1730 
1731 		/* Set Tx Tail register */
1732 		txr->tail = IXGBE_VFTDT(j);
1733 
1734 		txr->txr_no_space = false;
1735 
1736 		/* Set Ring parameters */
1737 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1738 		    (tdba & 0x00000000ffffffffULL));
1739 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1740 		IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j),
1741 		    adapter->num_tx_desc * sizeof(struct ixgbe_legacy_tx_desc));
1742 		txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1743 		txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
1744 		IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1745 
1746 		/* Now enable */
1747 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1748 		txdctl |= IXGBE_TXDCTL_ENABLE;
1749 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1750 	}
1751 
1752 	return;
1753 } /* ixv_initialize_transmit_units */
1754 
1755 
1756 /************************************************************************
1757  * ixv_initialize_rss_mapping
1758  ************************************************************************/
1759 static void
1760 ixv_initialize_rss_mapping(struct adapter *adapter)
1761 {
1762 	struct ixgbe_hw *hw = &adapter->hw;
1763 	u32		reta = 0, mrqc, rss_key[10];
1764 	int		queue_id;
1765 	int		i, j;
1766 	u32		rss_hash_config;
1767 
1768 	/* force use default RSS key. */
1769 #ifdef __NetBSD__
1770 	rss_getkey((uint8_t *) &rss_key);
1771 #else
1772 	if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1773 		/* Fetch the configured RSS key */
1774 		rss_getkey((uint8_t *)&rss_key);
1775 	} else {
1776 		/* set up random bits */
1777 		cprng_fast(&rss_key, sizeof(rss_key));
1778 	}
1779 #endif
1780 
1781 	/* Now fill out hash function seeds */
1782 	for (i = 0; i < 10; i++)
1783 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1784 
1785 	/* Set up the redirection table */
1786 	for (i = 0, j = 0; i < 64; i++, j++) {
1787 		if (j == adapter->num_queues)
1788 			j = 0;
1789 
1790 		if (adapter->feat_en & IXGBE_FEATURE_RSS) {
1791 			/*
1792 			 * Fetch the RSS bucket id for the given indirection
1793 			 * entry. Cap it at the number of configured buckets
1794 			 * (which is num_queues.)
1795 			 */
1796 			queue_id = rss_get_indirection_to_bucket(i);
1797 			queue_id = queue_id % adapter->num_queues;
1798 		} else
1799 			queue_id = j;
1800 
1801 		/*
1802 		 * The low 8 bits are for hash value (n+0);
1803 		 * The next 8 bits are for hash value (n+1), etc.
1804 		 */
1805 		reta >>= 8;
1806 		reta |= ((uint32_t)queue_id) << 24;
1807 		if ((i & 3) == 3) {
1808 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), reta);
1809 			reta = 0;
1810 		}
1811 	}
1812 
1813 	/* Perform hash on these packet types */
1814 	if (adapter->feat_en & IXGBE_FEATURE_RSS)
1815 		rss_hash_config = rss_gethashconfig();
1816 	else {
1817 		/*
1818 		 * Disable UDP - IP fragments aren't currently being handled
1819 		 * and so we end up with a mix of 2-tuple and 4-tuple
1820 		 * traffic.
1821 		 */
1822 		rss_hash_config = RSS_HASHTYPE_RSS_IPV4
1823 				| RSS_HASHTYPE_RSS_TCP_IPV4
1824 				| RSS_HASHTYPE_RSS_IPV6
1825 				| RSS_HASHTYPE_RSS_TCP_IPV6;
1826 	}
1827 
1828 	mrqc = IXGBE_MRQC_RSSEN;
1829 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV4)
1830 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
1831 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV4)
1832 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
1833 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6)
1834 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
1835 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6)
1836 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
1837 	if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6_EX)
1838 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_IPV6_EX defined, but not supported\n",
1839 		    __func__);
1840 	if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6_EX)
1841 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_TCP_IPV6_EX defined, but not supported\n",
1842 		    __func__);
1843 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV4)
1844 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
1845 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6)
1846 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1847 	if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6_EX)
1848 		device_printf(adapter->dev, "%s: RSS_HASHTYPE_RSS_UDP_IPV6_EX defined, but not supported\n",
1849 		    __func__);
1850 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, mrqc);
1851 } /* ixv_initialize_rss_mapping */
1852 
1853 
1854 /************************************************************************
1855  * ixv_initialize_receive_units - Setup receive registers and features.
1856  ************************************************************************/
1857 static void
1858 ixv_initialize_receive_units(struct adapter *adapter)
1859 {
1860 	struct	rx_ring	*rxr = adapter->rx_rings;
1861 	struct ixgbe_hw	*hw = &adapter->hw;
1862 	struct ifnet	*ifp = adapter->ifp;
1863 	u32		bufsz, psrtype;
1864 
1865 	if (ifp->if_mtu > ETHERMTU)
1866 		bufsz = 4096 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1867 	else
1868 		bufsz = 2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1869 
1870 	psrtype = IXGBE_PSRTYPE_TCPHDR
1871 		| IXGBE_PSRTYPE_UDPHDR
1872 		| IXGBE_PSRTYPE_IPV4HDR
1873 		| IXGBE_PSRTYPE_IPV6HDR
1874 		| IXGBE_PSRTYPE_L2HDR;
1875 
1876 	if (adapter->num_queues > 1)
1877 		psrtype |= 1 << 29;
1878 
1879 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1880 
1881 	/* Tell PF our max_frame size */
1882 	if (ixgbevf_rlpml_set_vf(hw, adapter->max_frame_size) != 0) {
1883 		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");
1884 	}
1885 
1886 	for (int i = 0; i < adapter->num_queues; i++, rxr++) {
1887 		u64 rdba = rxr->rxdma.dma_paddr;
1888 		u32 reg, rxdctl;
1889 		int j = rxr->me;
1890 
1891 		/* Disable the queue */
1892 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1893 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1894 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1895 		for (int k = 0; k < 10; k++) {
1896 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1897 			    IXGBE_RXDCTL_ENABLE)
1898 				msec_delay(1);
1899 			else
1900 				break;
1901 		}
1902 		IXGBE_WRITE_BARRIER(hw);
1903 		/* Setup the Base and Length of the Rx Descriptor Ring */
1904 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1905 		    (rdba & 0x00000000ffffffffULL));
1906 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1907 		IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j),
1908 		    adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc));
1909 
1910 		/* Reset the ring indices */
1911 		IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
1912 		IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), 0);
1913 
1914 		/* Set up the SRRCTL register */
1915 		reg = IXGBE_READ_REG(hw, IXGBE_VFSRRCTL(j));
1916 		reg &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1917 		reg &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1918 		reg |= bufsz;
1919 		reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1920 		IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(j), reg);
1921 
1922 		/* Capture Rx Tail index */
1923 		rxr->tail = IXGBE_VFRDT(rxr->me);
1924 
1925 		/* Do the queue enabling last */
1926 		rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1927 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1928 		for (int k = 0; k < 10; k++) {
1929 			if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) &
1930 			    IXGBE_RXDCTL_ENABLE)
1931 				break;
1932 			msec_delay(1);
1933 		}
1934 		IXGBE_WRITE_BARRIER(hw);
1935 
1936 		/* Set the Tail Pointer */
1937 #ifdef DEV_NETMAP
1938 		/*
1939 		 * In netmap mode, we must preserve the buffers made
1940 		 * available to userspace before the if_init()
1941 		 * (this is true by default on the TX side, because
1942 		 * init makes all buffers available to userspace).
1943 		 *
1944 		 * netmap_reset() and the device specific routines
1945 		 * (e.g. ixgbe_setup_receive_rings()) map these
1946 		 * buffers at the end of the NIC ring, so here we
1947 		 * must set the RDT (tail) register to make sure
1948 		 * they are not overwritten.
1949 		 *
1950 		 * In this driver the NIC ring starts at RDH = 0,
1951 		 * RDT points to the last slot available for reception (?),
1952 		 * so RDT = num_rx_desc - 1 means the whole ring is available.
1953 		 */
1954 		if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) &&
1955 		    (ifp->if_capenable & IFCAP_NETMAP)) {
1956 			struct netmap_adapter *na = NA(adapter->ifp);
1957 			struct netmap_kring *kring = na->rx_rings[i];
1958 			int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring);
1959 
1960 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me), t);
1961 		} else
1962 #endif /* DEV_NETMAP */
1963 			IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
1964 			    adapter->num_rx_desc - 1);
1965 	}
1966 
1967 	if (adapter->hw.mac.type >= ixgbe_mac_X550_vf)
1968 		ixv_initialize_rss_mapping(adapter);
1969 } /* ixv_initialize_receive_units */
1970 
1971 /************************************************************************
1972  * ixv_sysctl_tdh_handler - Transmit Descriptor Head handler function
1973  *
1974  *   Retrieves the TDH value from the hardware
1975  ************************************************************************/
1976 static int
1977 ixv_sysctl_tdh_handler(SYSCTLFN_ARGS)
1978 {
1979 	struct sysctlnode node = *rnode;
1980 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
1981 	uint32_t val;
1982 
1983 	if (!txr)
1984 		return (0);
1985 
1986 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDH(txr->me));
1987 	node.sysctl_data = &val;
1988 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1989 } /* ixv_sysctl_tdh_handler */
1990 
1991 /************************************************************************
1992  * ixgbe_sysctl_tdt_handler - Transmit Descriptor Tail handler function
1993  *
1994  *   Retrieves the TDT value from the hardware
1995  ************************************************************************/
1996 static int
1997 ixv_sysctl_tdt_handler(SYSCTLFN_ARGS)
1998 {
1999 	struct sysctlnode node = *rnode;
2000 	struct tx_ring *txr = (struct tx_ring *)node.sysctl_data;
2001 	uint32_t val;
2002 
2003 	if (!txr)
2004 		return (0);
2005 
2006 	val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_VFTDT(txr->me));
2007 	node.sysctl_data = &val;
2008 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2009 } /* ixv_sysctl_tdt_handler */
2010 
2011 /************************************************************************
2012  * ixv_sysctl_next_to_check_handler - Receive Descriptor next to check
2013  * handler function
2014  *
2015  *   Retrieves the next_to_check value
2016  ************************************************************************/
2017 static int
2018 ixv_sysctl_next_to_check_handler(SYSCTLFN_ARGS)
2019 {
2020 	struct sysctlnode node = *rnode;
2021 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2022 	uint32_t val;
2023 
2024 	if (!rxr)
2025 		return (0);
2026 
2027 	val = rxr->next_to_check;
2028 	node.sysctl_data = &val;
2029 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2030 } /* ixv_sysctl_next_to_check_handler */
2031 
2032 /************************************************************************
2033  * ixv_sysctl_rdh_handler - Receive Descriptor Head handler function
2034  *
2035  *   Retrieves the RDH value from the hardware
2036  ************************************************************************/
2037 static int
2038 ixv_sysctl_rdh_handler(SYSCTLFN_ARGS)
2039 {
2040 	struct sysctlnode node = *rnode;
2041 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2042 	uint32_t val;
2043 
2044 	if (!rxr)
2045 		return (0);
2046 
2047 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDH(rxr->me));
2048 	node.sysctl_data = &val;
2049 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2050 } /* ixv_sysctl_rdh_handler */
2051 
2052 /************************************************************************
2053  * ixv_sysctl_rdt_handler - Receive Descriptor Tail handler function
2054  *
2055  *   Retrieves the RDT value from the hardware
2056  ************************************************************************/
2057 static int
2058 ixv_sysctl_rdt_handler(SYSCTLFN_ARGS)
2059 {
2060 	struct sysctlnode node = *rnode;
2061 	struct rx_ring *rxr = (struct rx_ring *)node.sysctl_data;
2062 	uint32_t val;
2063 
2064 	if (!rxr)
2065 		return (0);
2066 
2067 	val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_VFRDT(rxr->me));
2068 	node.sysctl_data = &val;
2069 	return sysctl_lookup(SYSCTLFN_CALL(&node));
2070 } /* ixv_sysctl_rdt_handler */
2071 
2072 static void
2073 ixv_setup_vlan_tagging(struct adapter *adapter)
2074 {
2075 	struct ethercom *ec = &adapter->osdep.ec;
2076 	struct ixgbe_hw *hw = &adapter->hw;
2077 	struct rx_ring	*rxr;
2078 	u32		ctrl;
2079 	int		i;
2080 	bool		hwtagging;
2081 
2082 	/* Enable HW tagging only if any vlan is attached */
2083 	hwtagging = (ec->ec_capenable & ETHERCAP_VLAN_HWTAGGING)
2084 	    && VLAN_ATTACHED(ec);
2085 
2086 	/* Enable the queues */
2087 	for (i = 0; i < adapter->num_queues; i++) {
2088 		rxr = &adapter->rx_rings[i];
2089 		ctrl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(rxr->me));
2090 		if (hwtagging)
2091 			ctrl |= IXGBE_RXDCTL_VME;
2092 		else
2093 			ctrl &= ~IXGBE_RXDCTL_VME;
2094 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(rxr->me), ctrl);
2095 		/*
2096 		 * Let Rx path know that it needs to store VLAN tag
2097 		 * as part of extra mbuf info.
2098 		 */
2099 		rxr->vtag_strip = hwtagging ? TRUE : FALSE;
2100 	}
2101 } /* ixv_setup_vlan_tagging */
2102 
2103 /************************************************************************
2104  * ixv_setup_vlan_support
2105  ************************************************************************/
2106 static int
2107 ixv_setup_vlan_support(struct adapter *adapter)
2108 {
2109 	struct ethercom *ec = &adapter->osdep.ec;
2110 	struct ixgbe_hw *hw = &adapter->hw;
2111 	u32		vid, vfta, retry;
2112 	struct vlanid_list *vlanidp;
2113 	int rv, error = 0;
2114 
2115 	/*
2116 	 *  This function is called from both if_init and ifflags_cb()
2117 	 * on NetBSD.
2118 	 */
2119 
2120 	/*
2121 	 * Part 1:
2122 	 * Setup VLAN HW tagging
2123 	 */
2124 	ixv_setup_vlan_tagging(adapter);
2125 
2126 	if (!VLAN_ATTACHED(ec))
2127 		return 0;
2128 
2129 	/*
2130 	 * Part 2:
2131 	 * Setup VLAN HW filter
2132 	 */
2133 	/* Cleanup shadow_vfta */
2134 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++)
2135 		adapter->shadow_vfta[i] = 0;
2136 	/* Generate shadow_vfta from ec_vids */
2137 	ETHER_LOCK(ec);
2138 	SIMPLEQ_FOREACH(vlanidp, &ec->ec_vids, vid_list) {
2139 		uint32_t idx;
2140 
2141 		idx = vlanidp->vid / 32;
2142 		KASSERT(idx < IXGBE_VFTA_SIZE);
2143 		adapter->shadow_vfta[idx] |= (u32)1 << (vlanidp->vid % 32);
2144 	}
2145 	ETHER_UNLOCK(ec);
2146 
2147 	/*
2148 	 * A soft reset zero's out the VFTA, so
2149 	 * we need to repopulate it now.
2150 	 */
2151 	for (int i = 0; i < IXGBE_VFTA_SIZE; i++) {
2152 		if (adapter->shadow_vfta[i] == 0)
2153 			continue;
2154 		vfta = adapter->shadow_vfta[i];
2155 		/*
2156 		 * Reconstruct the vlan id's
2157 		 * based on the bits set in each
2158 		 * of the array ints.
2159 		 */
2160 		for (int j = 0; j < 32; j++) {
2161 			retry = 0;
2162 			if ((vfta & ((u32)1 << j)) == 0)
2163 				continue;
2164 			vid = (i * 32) + j;
2165 
2166 			/* Call the shared code mailbox routine */
2167 			while ((rv = hw->mac.ops.set_vfta(hw, vid, 0, TRUE,
2168 			    FALSE)) != 0) {
2169 				if (++retry > 5) {
2170 					device_printf(adapter->dev,
2171 					    "%s: max retry exceeded\n",
2172 						__func__);
2173 					break;
2174 				}
2175 			}
2176 			if (rv != 0) {
2177 				device_printf(adapter->dev,
2178 				    "failed to set vlan %d\n", vid);
2179 				error = EACCES;
2180 			}
2181 		}
2182 	}
2183 	return error;
2184 } /* ixv_setup_vlan_support */
2185 
2186 static int
2187 ixv_vlan_cb(struct ethercom *ec, uint16_t vid, bool set)
2188 {
2189 	struct ifnet *ifp = &ec->ec_if;
2190 	struct adapter *adapter = ifp->if_softc;
2191 	int rv;
2192 
2193 	if (set)
2194 		rv = ixv_register_vlan(adapter, vid);
2195 	else
2196 		rv = ixv_unregister_vlan(adapter, vid);
2197 
2198 	if (rv != 0)
2199 		return rv;
2200 
2201 	/*
2202 	 * Control VLAN HW tagging when ec_nvlan is changed from 1 to 0
2203 	 * or 0 to 1.
2204 	 */
2205 	if ((set && (ec->ec_nvlans == 1)) || (!set && (ec->ec_nvlans == 0)))
2206 		ixv_setup_vlan_tagging(adapter);
2207 
2208 	return rv;
2209 }
2210 
2211 /************************************************************************
2212  * ixv_register_vlan
2213  *
2214  *   Run via a vlan config EVENT, it enables us to use the
2215  *   HW Filter table since we can get the vlan id. This just
2216  *   creates the entry in the soft version of the VFTA, init
2217  *   will repopulate the real table.
2218  ************************************************************************/
2219 static int
2220 ixv_register_vlan(struct adapter *adapter, u16 vtag)
2221 {
2222 	struct ixgbe_hw *hw = &adapter->hw;
2223 	u16		index, bit;
2224 	int error;
2225 
2226 	if ((vtag == 0) || (vtag > 4095)) /* Invalid */
2227 		return EINVAL;
2228 	IXGBE_CORE_LOCK(adapter);
2229 	index = (vtag >> 5) & 0x7F;
2230 	bit = vtag & 0x1F;
2231 	adapter->shadow_vfta[index] |= ((u32)1 << bit);
2232 	error = hw->mac.ops.set_vfta(hw, vtag, 0, true, false);
2233 	IXGBE_CORE_UNLOCK(adapter);
2234 
2235 	if (error != 0) {
2236 		device_printf(adapter->dev, "failed to register vlan %hu\n",
2237 		    vtag);
2238 		error = EACCES;
2239 	}
2240 	return error;
2241 } /* ixv_register_vlan */
2242 
2243 /************************************************************************
2244  * ixv_unregister_vlan
2245  *
2246  *   Run via a vlan unconfig EVENT, remove our entry
2247  *   in the soft vfta.
2248  ************************************************************************/
2249 static int
2250 ixv_unregister_vlan(struct adapter *adapter, u16 vtag)
2251 {
2252 	struct ixgbe_hw *hw = &adapter->hw;
2253 	u16		index, bit;
2254 	int		error;
2255 
2256 	if ((vtag == 0) || (vtag > 4095))  /* Invalid */
2257 		return EINVAL;
2258 
2259 	IXGBE_CORE_LOCK(adapter);
2260 	index = (vtag >> 5) & 0x7F;
2261 	bit = vtag & 0x1F;
2262 	adapter->shadow_vfta[index] &= ~((u32)1 << bit);
2263 	error = hw->mac.ops.set_vfta(hw, vtag, 0, false, false);
2264 	IXGBE_CORE_UNLOCK(adapter);
2265 
2266 	if (error != 0) {
2267 		device_printf(adapter->dev, "failed to unregister vlan %hu\n",
2268 		    vtag);
2269 		error = EIO;
2270 	}
2271 	return error;
2272 } /* ixv_unregister_vlan */
2273 
2274 /************************************************************************
2275  * ixv_enable_intr
2276  ************************************************************************/
2277 static void
2278 ixv_enable_intr(struct adapter *adapter)
2279 {
2280 	struct ixgbe_hw *hw = &adapter->hw;
2281 	struct ix_queue *que = adapter->queues;
2282 	u32		mask;
2283 	int i;
2284 
2285 	/* For VTEIAC */
2286 	mask = (1 << adapter->vector);
2287 	for (i = 0; i < adapter->num_queues; i++, que++)
2288 		mask |= (1 << que->msix);
2289 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
2290 
2291 	/* For VTEIMS */
2292 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
2293 	que = adapter->queues;
2294 	for (i = 0; i < adapter->num_queues; i++, que++)
2295 		ixv_enable_queue(adapter, que->msix);
2296 
2297 	IXGBE_WRITE_FLUSH(hw);
2298 } /* ixv_enable_intr */
2299 
2300 /************************************************************************
2301  * ixv_disable_intr
2302  ************************************************************************/
2303 static void
2304 ixv_disable_intr(struct adapter *adapter)
2305 {
2306 	struct ix_queue	*que = adapter->queues;
2307 
2308 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIAC, 0);
2309 
2310 	/* disable interrupts other than queues */
2311 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_VTEIMC, adapter->vector);
2312 
2313 	for (int i = 0; i < adapter->num_queues; i++, que++)
2314 		ixv_disable_queue(adapter, que->msix);
2315 
2316 	IXGBE_WRITE_FLUSH(&adapter->hw);
2317 } /* ixv_disable_intr */
2318 
2319 /************************************************************************
2320  * ixv_set_ivar
2321  *
2322  *   Setup the correct IVAR register for a particular MSI-X interrupt
2323  *    - entry is the register array entry
2324  *    - vector is the MSI-X vector for this queue
2325  *    - type is RX/TX/MISC
2326  ************************************************************************/
2327 static void
2328 ixv_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type)
2329 {
2330 	struct ixgbe_hw *hw = &adapter->hw;
2331 	u32		ivar, index;
2332 
2333 	vector |= IXGBE_IVAR_ALLOC_VAL;
2334 
2335 	if (type == -1) { /* MISC IVAR */
2336 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
2337 		ivar &= ~0xFF;
2338 		ivar |= vector;
2339 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
2340 	} else {	  /* RX/TX IVARS */
2341 		index = (16 * (entry & 1)) + (8 * type);
2342 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(entry >> 1));
2343 		ivar &= ~(0xffUL << index);
2344 		ivar |= ((u32)vector << index);
2345 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(entry >> 1), ivar);
2346 	}
2347 } /* ixv_set_ivar */
2348 
2349 /************************************************************************
2350  * ixv_configure_ivars
2351  ************************************************************************/
2352 static void
2353 ixv_configure_ivars(struct adapter *adapter)
2354 {
2355 	struct ix_queue *que = adapter->queues;
2356 
2357 	/* XXX We should sync EITR value calculation with ixgbe.c? */
2358 
2359 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2360 		/* First the RX queue entry */
2361 		ixv_set_ivar(adapter, i, que->msix, 0);
2362 		/* ... and the TX */
2363 		ixv_set_ivar(adapter, i, que->msix, 1);
2364 		/* Set an initial value in EITR */
2365 		ixv_eitr_write(adapter, que->msix, IXGBE_EITR_DEFAULT);
2366 	}
2367 
2368 	/* For the mailbox interrupt */
2369 	ixv_set_ivar(adapter, 1, adapter->vector, -1);
2370 } /* ixv_configure_ivars */
2371 
2372 
2373 /************************************************************************
2374  * ixv_save_stats
2375  *
2376  *   The VF stats registers never have a truly virgin
2377  *   starting point, so this routine tries to make an
2378  *   artificial one, marking ground zero on attach as
2379  *   it were.
2380  ************************************************************************/
2381 static void
2382 ixv_save_stats(struct adapter *adapter)
2383 {
2384 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2385 
2386 	if (stats->vfgprc.ev_count || stats->vfgptc.ev_count) {
2387 		stats->saved_reset_vfgprc +=
2388 		    stats->vfgprc.ev_count - stats->base_vfgprc;
2389 		stats->saved_reset_vfgptc +=
2390 		    stats->vfgptc.ev_count - stats->base_vfgptc;
2391 		stats->saved_reset_vfgorc +=
2392 		    stats->vfgorc.ev_count - stats->base_vfgorc;
2393 		stats->saved_reset_vfgotc +=
2394 		    stats->vfgotc.ev_count - stats->base_vfgotc;
2395 		stats->saved_reset_vfmprc +=
2396 		    stats->vfmprc.ev_count - stats->base_vfmprc;
2397 	}
2398 } /* ixv_save_stats */
2399 
2400 /************************************************************************
2401  * ixv_init_stats
2402  ************************************************************************/
2403 static void
2404 ixv_init_stats(struct adapter *adapter)
2405 {
2406 	struct ixgbe_hw *hw = &adapter->hw;
2407 
2408 	adapter->stats.vf.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2409 	adapter->stats.vf.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2410 	adapter->stats.vf.last_vfgorc |=
2411 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2412 
2413 	adapter->stats.vf.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2414 	adapter->stats.vf.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2415 	adapter->stats.vf.last_vfgotc |=
2416 	    (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2417 
2418 	adapter->stats.vf.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2419 
2420 	adapter->stats.vf.base_vfgprc = adapter->stats.vf.last_vfgprc;
2421 	adapter->stats.vf.base_vfgorc = adapter->stats.vf.last_vfgorc;
2422 	adapter->stats.vf.base_vfgptc = adapter->stats.vf.last_vfgptc;
2423 	adapter->stats.vf.base_vfgotc = adapter->stats.vf.last_vfgotc;
2424 	adapter->stats.vf.base_vfmprc = adapter->stats.vf.last_vfmprc;
2425 } /* ixv_init_stats */
2426 
2427 #define UPDATE_STAT_32(reg, last, count)		\
2428 {							\
2429 	u32 current = IXGBE_READ_REG(hw, (reg));	\
2430 	if (current < (last))				\
2431 		count.ev_count += 0x100000000LL;	\
2432 	(last) = current;				\
2433 	count.ev_count &= 0xFFFFFFFF00000000LL;		\
2434 	count.ev_count |= current;			\
2435 }
2436 
2437 #define UPDATE_STAT_36(lsb, msb, last, count)		\
2438 {							\
2439 	u64 cur_lsb = IXGBE_READ_REG(hw, (lsb));	\
2440 	u64 cur_msb = IXGBE_READ_REG(hw, (msb));	\
2441 	u64 current = ((cur_msb << 32) | cur_lsb);	\
2442 	if (current < (last))				\
2443 		count.ev_count += 0x1000000000LL;	\
2444 	(last) = current;				\
2445 	count.ev_count &= 0xFFFFFFF000000000LL;		\
2446 	count.ev_count |= current;			\
2447 }
2448 
2449 /************************************************************************
2450  * ixv_update_stats - Update the board statistics counters.
2451  ************************************************************************/
2452 void
2453 ixv_update_stats(struct adapter *adapter)
2454 {
2455 	struct ixgbe_hw *hw = &adapter->hw;
2456 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2457 
2458 	UPDATE_STAT_32(IXGBE_VFGPRC, stats->last_vfgprc, stats->vfgprc);
2459 	UPDATE_STAT_32(IXGBE_VFGPTC, stats->last_vfgptc, stats->vfgptc);
2460 	UPDATE_STAT_36(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB, stats->last_vfgorc,
2461 	    stats->vfgorc);
2462 	UPDATE_STAT_36(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB, stats->last_vfgotc,
2463 	    stats->vfgotc);
2464 	UPDATE_STAT_32(IXGBE_VFMPRC, stats->last_vfmprc, stats->vfmprc);
2465 
2466 	/* VF doesn't count errors by hardware */
2467 
2468 } /* ixv_update_stats */
2469 
2470 /************************************************************************
2471  * ixv_sysctl_interrupt_rate_handler
2472  ************************************************************************/
2473 static int
2474 ixv_sysctl_interrupt_rate_handler(SYSCTLFN_ARGS)
2475 {
2476 	struct sysctlnode node = *rnode;
2477 	struct ix_queue *que = (struct ix_queue *)node.sysctl_data;
2478 	struct adapter	*adapter = que->adapter;
2479 	uint32_t reg, usec, rate;
2480 	int error;
2481 
2482 	if (que == NULL)
2483 		return 0;
2484 	reg = IXGBE_READ_REG(&que->adapter->hw, IXGBE_VTEITR(que->msix));
2485 	usec = ((reg & 0x0FF8) >> 3);
2486 	if (usec > 0)
2487 		rate = 500000 / usec;
2488 	else
2489 		rate = 0;
2490 	node.sysctl_data = &rate;
2491 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2492 	if (error || newp == NULL)
2493 		return error;
2494 	reg &= ~0xfff; /* default, no limitation */
2495 	if (rate > 0 && rate < 500000) {
2496 		if (rate < 1000)
2497 			rate = 1000;
2498 		reg |= ((4000000 / rate) & 0xff8);
2499 		/*
2500 		 * When RSC is used, ITR interval must be larger than
2501 		 * RSC_DELAY. Currently, we use 2us for RSC_DELAY.
2502 		 * The minimum value is always greater than 2us on 100M
2503 		 * (and 10M?(not documented)), but it's not on 1G and higher.
2504 		 */
2505 		if ((adapter->link_speed != IXGBE_LINK_SPEED_100_FULL)
2506 		    && (adapter->link_speed != IXGBE_LINK_SPEED_10_FULL)) {
2507 			if ((adapter->num_queues > 1)
2508 			    && (reg < IXGBE_MIN_RSC_EITR_10G1G))
2509 				return EINVAL;
2510 		}
2511 		ixv_max_interrupt_rate = rate;
2512 	} else
2513 		ixv_max_interrupt_rate = 0;
2514 	ixv_eitr_write(adapter, que->msix, reg);
2515 
2516 	return (0);
2517 } /* ixv_sysctl_interrupt_rate_handler */
2518 
2519 const struct sysctlnode *
2520 ixv_sysctl_instance(struct adapter *adapter)
2521 {
2522 	const char *dvname;
2523 	struct sysctllog **log;
2524 	int rc;
2525 	const struct sysctlnode *rnode;
2526 
2527 	log = &adapter->sysctllog;
2528 	dvname = device_xname(adapter->dev);
2529 
2530 	if ((rc = sysctl_createv(log, 0, NULL, &rnode,
2531 	    0, CTLTYPE_NODE, dvname,
2532 	    SYSCTL_DESCR("ixv information and settings"),
2533 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0)
2534 		goto err;
2535 
2536 	return rnode;
2537 err:
2538 	device_printf(adapter->dev,
2539 	    "%s: sysctl_createv failed, rc = %d\n", __func__, rc);
2540 	return NULL;
2541 }
2542 
2543 static void
2544 ixv_add_device_sysctls(struct adapter *adapter)
2545 {
2546 	struct sysctllog **log;
2547 	const struct sysctlnode *rnode, *cnode;
2548 	device_t dev;
2549 
2550 	dev = adapter->dev;
2551 	log = &adapter->sysctllog;
2552 
2553 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2554 		aprint_error_dev(dev, "could not create sysctl root\n");
2555 		return;
2556 	}
2557 
2558 	if (sysctl_createv(log, 0, &rnode, &cnode,
2559 	    CTLFLAG_READWRITE, CTLTYPE_INT,
2560 	    "debug", SYSCTL_DESCR("Debug Info"),
2561 	    ixv_sysctl_debug, 0, (void *)adapter, 0, CTL_CREATE, CTL_EOL) != 0)
2562 		aprint_error_dev(dev, "could not create sysctl\n");
2563 
2564 	if (sysctl_createv(log, 0, &rnode, &cnode,
2565 	    CTLFLAG_READWRITE, CTLTYPE_BOOL,
2566 	    "enable_aim", SYSCTL_DESCR("Interrupt Moderation"),
2567 	    NULL, 0, &adapter->enable_aim, 0, CTL_CREATE, 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,
2572 	    "txrx_workqueue", SYSCTL_DESCR("Use workqueue for packet processing"),
2573 		NULL, 0, &adapter->txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL) != 0)
2574 		aprint_error_dev(dev, "could not create sysctl\n");
2575 }
2576 
2577 /************************************************************************
2578  * ixv_add_stats_sysctls - Add statistic sysctls for the VF.
2579  ************************************************************************/
2580 static void
2581 ixv_add_stats_sysctls(struct adapter *adapter)
2582 {
2583 	device_t		dev = adapter->dev;
2584 	struct tx_ring		*txr = adapter->tx_rings;
2585 	struct rx_ring		*rxr = adapter->rx_rings;
2586 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2587 	struct ixgbe_hw *hw = &adapter->hw;
2588 	const struct sysctlnode *rnode, *cnode;
2589 	struct sysctllog **log = &adapter->sysctllog;
2590 	const char *xname = device_xname(dev);
2591 
2592 	/* Driver Statistics */
2593 	evcnt_attach_dynamic(&adapter->efbig_tx_dma_setup, EVCNT_TYPE_MISC,
2594 	    NULL, xname, "Driver tx dma soft fail EFBIG");
2595 	evcnt_attach_dynamic(&adapter->mbuf_defrag_failed, EVCNT_TYPE_MISC,
2596 	    NULL, xname, "m_defrag() failed");
2597 	evcnt_attach_dynamic(&adapter->efbig2_tx_dma_setup, EVCNT_TYPE_MISC,
2598 	    NULL, xname, "Driver tx dma hard fail EFBIG");
2599 	evcnt_attach_dynamic(&adapter->einval_tx_dma_setup, EVCNT_TYPE_MISC,
2600 	    NULL, xname, "Driver tx dma hard fail EINVAL");
2601 	evcnt_attach_dynamic(&adapter->other_tx_dma_setup, EVCNT_TYPE_MISC,
2602 	    NULL, xname, "Driver tx dma hard fail other");
2603 	evcnt_attach_dynamic(&adapter->eagain_tx_dma_setup, EVCNT_TYPE_MISC,
2604 	    NULL, xname, "Driver tx dma soft fail EAGAIN");
2605 	evcnt_attach_dynamic(&adapter->enomem_tx_dma_setup, EVCNT_TYPE_MISC,
2606 	    NULL, xname, "Driver tx dma soft fail ENOMEM");
2607 	evcnt_attach_dynamic(&adapter->watchdog_events, EVCNT_TYPE_MISC,
2608 	    NULL, xname, "Watchdog timeouts");
2609 	evcnt_attach_dynamic(&adapter->tso_err, EVCNT_TYPE_MISC,
2610 	    NULL, xname, "TSO errors");
2611 	evcnt_attach_dynamic(&adapter->admin_irqev, EVCNT_TYPE_INTR,
2612 	    NULL, xname, "Admin MSI-X IRQ Handled");
2613 	evcnt_attach_dynamic(&adapter->link_workev, EVCNT_TYPE_INTR,
2614 	    NULL, xname, "Admin event");
2615 
2616 	for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2617 		snprintf(adapter->queues[i].evnamebuf,
2618 		    sizeof(adapter->queues[i].evnamebuf), "%s q%d",
2619 		    xname, i);
2620 		snprintf(adapter->queues[i].namebuf,
2621 		    sizeof(adapter->queues[i].namebuf), "q%d", i);
2622 
2623 		if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2624 			aprint_error_dev(dev, "could not create sysctl root\n");
2625 			break;
2626 		}
2627 
2628 		if (sysctl_createv(log, 0, &rnode, &rnode,
2629 		    0, CTLTYPE_NODE,
2630 		    adapter->queues[i].namebuf, SYSCTL_DESCR("Queue Name"),
2631 		    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL) != 0)
2632 			break;
2633 
2634 		if (sysctl_createv(log, 0, &rnode, &cnode,
2635 		    CTLFLAG_READWRITE, CTLTYPE_INT,
2636 		    "interrupt_rate", SYSCTL_DESCR("Interrupt Rate"),
2637 		    ixv_sysctl_interrupt_rate_handler, 0,
2638 		    (void *)&adapter->queues[i], 0, CTL_CREATE, CTL_EOL) != 0)
2639 			break;
2640 
2641 		if (sysctl_createv(log, 0, &rnode, &cnode,
2642 		    CTLFLAG_READONLY, CTLTYPE_INT,
2643 		    "txd_head", SYSCTL_DESCR("Transmit Descriptor Head"),
2644 		    ixv_sysctl_tdh_handler, 0, (void *)txr,
2645 		    0, CTL_CREATE, CTL_EOL) != 0)
2646 			break;
2647 
2648 		if (sysctl_createv(log, 0, &rnode, &cnode,
2649 		    CTLFLAG_READONLY, CTLTYPE_INT,
2650 		    "txd_tail", SYSCTL_DESCR("Transmit Descriptor Tail"),
2651 		    ixv_sysctl_tdt_handler, 0, (void *)txr,
2652 		    0, CTL_CREATE, CTL_EOL) != 0)
2653 			break;
2654 
2655 		evcnt_attach_dynamic(&adapter->queues[i].irqs, EVCNT_TYPE_INTR,
2656 		    NULL, adapter->queues[i].evnamebuf, "IRQs on queue");
2657 		evcnt_attach_dynamic(&adapter->queues[i].handleq,
2658 		    EVCNT_TYPE_MISC, NULL, adapter->queues[i].evnamebuf,
2659 		    "Handled queue in softint");
2660 		evcnt_attach_dynamic(&adapter->queues[i].req, EVCNT_TYPE_MISC,
2661 		    NULL, adapter->queues[i].evnamebuf, "Requeued in softint");
2662 		evcnt_attach_dynamic(&txr->tso_tx, EVCNT_TYPE_MISC,
2663 		    NULL, adapter->queues[i].evnamebuf, "TSO");
2664 		evcnt_attach_dynamic(&txr->no_desc_avail, EVCNT_TYPE_MISC,
2665 		    NULL, adapter->queues[i].evnamebuf,
2666 		    "Queue No Descriptor Available");
2667 		evcnt_attach_dynamic(&txr->total_packets, EVCNT_TYPE_MISC,
2668 		    NULL, adapter->queues[i].evnamebuf,
2669 		    "Queue Packets Transmitted");
2670 #ifndef IXGBE_LEGACY_TX
2671 		evcnt_attach_dynamic(&txr->pcq_drops, EVCNT_TYPE_MISC,
2672 		    NULL, adapter->queues[i].evnamebuf,
2673 		    "Packets dropped in pcq");
2674 #endif
2675 
2676 #ifdef LRO
2677 		struct lro_ctrl *lro = &rxr->lro;
2678 #endif /* LRO */
2679 
2680 		if (sysctl_createv(log, 0, &rnode, &cnode,
2681 		    CTLFLAG_READONLY,
2682 		    CTLTYPE_INT,
2683 		    "rxd_nxck", SYSCTL_DESCR("Receive Descriptor next to check"),
2684 			ixv_sysctl_next_to_check_handler, 0, (void *)rxr, 0,
2685 		    CTL_CREATE, CTL_EOL) != 0)
2686 			break;
2687 
2688 		if (sysctl_createv(log, 0, &rnode, &cnode,
2689 		    CTLFLAG_READONLY,
2690 		    CTLTYPE_INT,
2691 		    "rxd_head", SYSCTL_DESCR("Receive Descriptor Head"),
2692 		    ixv_sysctl_rdh_handler, 0, (void *)rxr, 0,
2693 		    CTL_CREATE, CTL_EOL) != 0)
2694 			break;
2695 
2696 		if (sysctl_createv(log, 0, &rnode, &cnode,
2697 		    CTLFLAG_READONLY,
2698 		    CTLTYPE_INT,
2699 		    "rxd_tail", SYSCTL_DESCR("Receive Descriptor Tail"),
2700 		    ixv_sysctl_rdt_handler, 0, (void *)rxr, 0,
2701 		    CTL_CREATE, CTL_EOL) != 0)
2702 			break;
2703 
2704 		evcnt_attach_dynamic(&rxr->rx_packets, EVCNT_TYPE_MISC,
2705 		    NULL, adapter->queues[i].evnamebuf, "Queue Packets Received");
2706 		evcnt_attach_dynamic(&rxr->rx_bytes, EVCNT_TYPE_MISC,
2707 		    NULL, adapter->queues[i].evnamebuf, "Queue Bytes Received");
2708 		evcnt_attach_dynamic(&rxr->rx_copies, EVCNT_TYPE_MISC,
2709 		    NULL, adapter->queues[i].evnamebuf, "Copied RX Frames");
2710 		evcnt_attach_dynamic(&rxr->no_jmbuf, EVCNT_TYPE_MISC,
2711 		    NULL, adapter->queues[i].evnamebuf, "Rx no jumbo mbuf");
2712 		evcnt_attach_dynamic(&rxr->rx_discarded, EVCNT_TYPE_MISC,
2713 		    NULL, adapter->queues[i].evnamebuf, "Rx discarded");
2714 #ifdef LRO
2715 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_queued",
2716 				CTLFLAG_RD, &lro->lro_queued, 0,
2717 				"LRO Queued");
2718 		SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, "lro_flushed",
2719 				CTLFLAG_RD, &lro->lro_flushed, 0,
2720 				"LRO Flushed");
2721 #endif /* LRO */
2722 	}
2723 
2724 	/* MAC stats get their own sub node */
2725 
2726 	snprintf(stats->namebuf,
2727 	    sizeof(stats->namebuf), "%s MAC Statistics", xname);
2728 
2729 	evcnt_attach_dynamic(&stats->ipcs, EVCNT_TYPE_MISC, NULL,
2730 	    stats->namebuf, "rx csum offload - IP");
2731 	evcnt_attach_dynamic(&stats->l4cs, EVCNT_TYPE_MISC, NULL,
2732 	    stats->namebuf, "rx csum offload - L4");
2733 	evcnt_attach_dynamic(&stats->ipcs_bad, EVCNT_TYPE_MISC, NULL,
2734 	    stats->namebuf, "rx csum offload - IP bad");
2735 	evcnt_attach_dynamic(&stats->l4cs_bad, EVCNT_TYPE_MISC, NULL,
2736 	    stats->namebuf, "rx csum offload - L4 bad");
2737 
2738 	/* Packet Reception Stats */
2739 	evcnt_attach_dynamic(&stats->vfgprc, EVCNT_TYPE_MISC, NULL,
2740 	    xname, "Good Packets Received");
2741 	evcnt_attach_dynamic(&stats->vfgorc, EVCNT_TYPE_MISC, NULL,
2742 	    xname, "Good Octets Received");
2743 	evcnt_attach_dynamic(&stats->vfmprc, EVCNT_TYPE_MISC, NULL,
2744 	    xname, "Multicast Packets Received");
2745 	evcnt_attach_dynamic(&stats->vfgptc, EVCNT_TYPE_MISC, NULL,
2746 	    xname, "Good Packets Transmitted");
2747 	evcnt_attach_dynamic(&stats->vfgotc, EVCNT_TYPE_MISC, NULL,
2748 	    xname, "Good Octets Transmitted");
2749 
2750 	/* Mailbox Stats */
2751 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_tx, EVCNT_TYPE_MISC, NULL,
2752 	    xname, "message TXs");
2753 	evcnt_attach_dynamic(&hw->mbx.stats.msgs_rx, EVCNT_TYPE_MISC, NULL,
2754 	    xname, "message RXs");
2755 	evcnt_attach_dynamic(&hw->mbx.stats.acks, EVCNT_TYPE_MISC, NULL,
2756 	    xname, "ACKs");
2757 	evcnt_attach_dynamic(&hw->mbx.stats.reqs, EVCNT_TYPE_MISC, NULL,
2758 	    xname, "REQs");
2759 	evcnt_attach_dynamic(&hw->mbx.stats.rsts, EVCNT_TYPE_MISC, NULL,
2760 	    xname, "RSTs");
2761 
2762 } /* ixv_add_stats_sysctls */
2763 
2764 static void
2765 ixv_clear_evcnt(struct adapter *adapter)
2766 {
2767 	struct tx_ring		*txr = adapter->tx_rings;
2768 	struct rx_ring		*rxr = adapter->rx_rings;
2769 	struct ixgbevf_hw_stats *stats = &adapter->stats.vf;
2770 	struct ixgbe_hw *hw = &adapter->hw;
2771 	int i;
2772 
2773 	/* Driver Statistics */
2774 	adapter->efbig_tx_dma_setup.ev_count = 0;
2775 	adapter->mbuf_defrag_failed.ev_count = 0;
2776 	adapter->efbig2_tx_dma_setup.ev_count = 0;
2777 	adapter->einval_tx_dma_setup.ev_count = 0;
2778 	adapter->other_tx_dma_setup.ev_count = 0;
2779 	adapter->eagain_tx_dma_setup.ev_count = 0;
2780 	adapter->enomem_tx_dma_setup.ev_count = 0;
2781 	adapter->watchdog_events.ev_count = 0;
2782 	adapter->tso_err.ev_count = 0;
2783 	adapter->admin_irqev.ev_count = 0;
2784 	adapter->link_workev.ev_count = 0;
2785 
2786 	for (i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
2787 		adapter->queues[i].irqs.ev_count = 0;
2788 		adapter->queues[i].handleq.ev_count = 0;
2789 		adapter->queues[i].req.ev_count = 0;
2790 		txr->tso_tx.ev_count = 0;
2791 		txr->no_desc_avail.ev_count = 0;
2792 		txr->total_packets.ev_count = 0;
2793 #ifndef IXGBE_LEGACY_TX
2794 		txr->pcq_drops.ev_count = 0;
2795 #endif
2796 		txr->q_efbig_tx_dma_setup = 0;
2797 		txr->q_mbuf_defrag_failed = 0;
2798 		txr->q_efbig2_tx_dma_setup = 0;
2799 		txr->q_einval_tx_dma_setup = 0;
2800 		txr->q_other_tx_dma_setup = 0;
2801 		txr->q_eagain_tx_dma_setup = 0;
2802 		txr->q_enomem_tx_dma_setup = 0;
2803 		txr->q_tso_err = 0;
2804 
2805 		rxr->rx_packets.ev_count = 0;
2806 		rxr->rx_bytes.ev_count = 0;
2807 		rxr->rx_copies.ev_count = 0;
2808 		rxr->no_jmbuf.ev_count = 0;
2809 		rxr->rx_discarded.ev_count = 0;
2810 	}
2811 
2812 	/* MAC stats get their own sub node */
2813 
2814 	stats->ipcs.ev_count = 0;
2815 	stats->l4cs.ev_count = 0;
2816 	stats->ipcs_bad.ev_count = 0;
2817 	stats->l4cs_bad.ev_count = 0;
2818 
2819 	/* Packet Reception Stats */
2820 	stats->vfgprc.ev_count = 0;
2821 	stats->vfgorc.ev_count = 0;
2822 	stats->vfmprc.ev_count = 0;
2823 	stats->vfgptc.ev_count = 0;
2824 	stats->vfgotc.ev_count = 0;
2825 
2826 	/* Mailbox Stats */
2827 	hw->mbx.stats.msgs_tx.ev_count = 0;
2828 	hw->mbx.stats.msgs_rx.ev_count = 0;
2829 	hw->mbx.stats.acks.ev_count = 0;
2830 	hw->mbx.stats.reqs.ev_count = 0;
2831 	hw->mbx.stats.rsts.ev_count = 0;
2832 
2833 } /* ixv_clear_evcnt */
2834 
2835 /************************************************************************
2836  * ixv_set_sysctl_value
2837  ************************************************************************/
2838 static void
2839 ixv_set_sysctl_value(struct adapter *adapter, const char *name,
2840 	const char *description, int *limit, int value)
2841 {
2842 	device_t dev =	adapter->dev;
2843 	struct sysctllog **log;
2844 	const struct sysctlnode *rnode, *cnode;
2845 
2846 	log = &adapter->sysctllog;
2847 	if ((rnode = ixv_sysctl_instance(adapter)) == NULL) {
2848 		aprint_error_dev(dev, "could not create sysctl root\n");
2849 		return;
2850 	}
2851 	if (sysctl_createv(log, 0, &rnode, &cnode,
2852 	    CTLFLAG_READWRITE, CTLTYPE_INT,
2853 	    name, SYSCTL_DESCR(description),
2854 	    NULL, 0, limit, 0, CTL_CREATE, CTL_EOL) != 0)
2855 		aprint_error_dev(dev, "could not create sysctl\n");
2856 	*limit = value;
2857 } /* ixv_set_sysctl_value */
2858 
2859 /************************************************************************
2860  * ixv_print_debug_info
2861  *
2862  *   Called only when em_display_debug_stats is enabled.
2863  *   Provides a way to take a look at important statistics
2864  *   maintained by the driver and hardware.
2865  ************************************************************************/
2866 static void
2867 ixv_print_debug_info(struct adapter *adapter)
2868 {
2869 	device_t	dev = adapter->dev;
2870 	struct ix_queue *que = adapter->queues;
2871 	struct rx_ring	*rxr;
2872 	struct tx_ring	*txr;
2873 #ifdef LRO
2874 	struct lro_ctrl *lro;
2875 #endif /* LRO */
2876 
2877 	for (int i = 0; i < adapter->num_queues; i++, que++) {
2878 		txr = que->txr;
2879 		rxr = que->rxr;
2880 #ifdef LRO
2881 		lro = &rxr->lro;
2882 #endif /* LRO */
2883 		device_printf(dev, "QUE(%d) IRQs Handled: %lu\n",
2884 		    que->msix, (long)que->irqs.ev_count);
2885 		device_printf(dev, "RX(%d) Packets Received: %lld\n",
2886 		    rxr->me, (long long)rxr->rx_packets.ev_count);
2887 		device_printf(dev, "RX(%d) Bytes Received: %lu\n",
2888 		    rxr->me, (long)rxr->rx_bytes.ev_count);
2889 #ifdef LRO
2890 		device_printf(dev, "RX(%d) LRO Queued= %ju\n",
2891 		    rxr->me, (uintmax_t)lro->lro_queued);
2892 		device_printf(dev, "RX(%d) LRO Flushed= %ju\n",
2893 		    rxr->me, (uintmax_t)lro->lro_flushed);
2894 #endif /* LRO */
2895 		device_printf(dev, "TX(%d) Packets Sent: %lu\n",
2896 		    txr->me, (long)txr->total_packets.ev_count);
2897 		device_printf(dev, "TX(%d) NO Desc Avail: %lu\n",
2898 		    txr->me, (long)txr->no_desc_avail.ev_count);
2899 	}
2900 
2901 	device_printf(dev, "Admin IRQ Handled: %lu\n",
2902 	    (long)adapter->admin_irqev.ev_count);
2903 	device_printf(dev, "Admin work Handled: %lu\n",
2904 	    (long)adapter->link_workev.ev_count);
2905 } /* ixv_print_debug_info */
2906 
2907 /************************************************************************
2908  * ixv_sysctl_debug
2909  ************************************************************************/
2910 static int
2911 ixv_sysctl_debug(SYSCTLFN_ARGS)
2912 {
2913 	struct sysctlnode node = *rnode;
2914 	struct adapter *adapter = (struct adapter *)node.sysctl_data;
2915 	int	       error, result;
2916 
2917 	node.sysctl_data = &result;
2918 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2919 
2920 	if (error || newp == NULL)
2921 		return error;
2922 
2923 	if (result == 1)
2924 		ixv_print_debug_info(adapter);
2925 
2926 	return 0;
2927 } /* ixv_sysctl_debug */
2928 
2929 /************************************************************************
2930  * ixv_init_device_features
2931  ************************************************************************/
2932 static void
2933 ixv_init_device_features(struct adapter *adapter)
2934 {
2935 	adapter->feat_cap = IXGBE_FEATURE_NETMAP
2936 			  | IXGBE_FEATURE_VF
2937 			  | IXGBE_FEATURE_RSS
2938 			  | IXGBE_FEATURE_LEGACY_TX;
2939 
2940 	/* A tad short on feature flags for VFs, atm. */
2941 	switch (adapter->hw.mac.type) {
2942 	case ixgbe_mac_82599_vf:
2943 		break;
2944 	case ixgbe_mac_X540_vf:
2945 		break;
2946 	case ixgbe_mac_X550_vf:
2947 	case ixgbe_mac_X550EM_x_vf:
2948 	case ixgbe_mac_X550EM_a_vf:
2949 		adapter->feat_cap |= IXGBE_FEATURE_NEEDS_CTXD;
2950 		break;
2951 	default:
2952 		break;
2953 	}
2954 
2955 	/* Enabled by default... */
2956 	/* Is a virtual function (VF) */
2957 	if (adapter->feat_cap & IXGBE_FEATURE_VF)
2958 		adapter->feat_en |= IXGBE_FEATURE_VF;
2959 	/* Netmap */
2960 	if (adapter->feat_cap & IXGBE_FEATURE_NETMAP)
2961 		adapter->feat_en |= IXGBE_FEATURE_NETMAP;
2962 	/* Receive-Side Scaling (RSS) */
2963 	if (adapter->feat_cap & IXGBE_FEATURE_RSS)
2964 		adapter->feat_en |= IXGBE_FEATURE_RSS;
2965 	/* Needs advanced context descriptor regardless of offloads req'd */
2966 	if (adapter->feat_cap & IXGBE_FEATURE_NEEDS_CTXD)
2967 		adapter->feat_en |= IXGBE_FEATURE_NEEDS_CTXD;
2968 
2969 	/* Enabled via sysctl... */
2970 	/* Legacy (single queue) transmit */
2971 	if ((adapter->feat_cap & IXGBE_FEATURE_LEGACY_TX) &&
2972 	    ixv_enable_legacy_tx)
2973 		adapter->feat_en |= IXGBE_FEATURE_LEGACY_TX;
2974 } /* ixv_init_device_features */
2975 
2976 /************************************************************************
2977  * ixv_shutdown - Shutdown entry point
2978  ************************************************************************/
2979 #if 0 /* XXX NetBSD ought to register something like this through pmf(9) */
2980 static int
2981 ixv_shutdown(device_t dev)
2982 {
2983 	struct adapter *adapter = device_private(dev);
2984 	IXGBE_CORE_LOCK(adapter);
2985 	ixv_stop(adapter);
2986 	IXGBE_CORE_UNLOCK(adapter);
2987 
2988 	return (0);
2989 } /* ixv_shutdown */
2990 #endif
2991 
2992 static int
2993 ixv_ifflags_cb(struct ethercom *ec)
2994 {
2995 	struct ifnet *ifp = &ec->ec_if;
2996 	struct adapter *adapter = ifp->if_softc;
2997 	u_short saved_flags;
2998 	u_short change;
2999 	int rv = 0;
3000 
3001 	IXGBE_CORE_LOCK(adapter);
3002 
3003 	saved_flags = adapter->if_flags;
3004 	change = ifp->if_flags ^ adapter->if_flags;
3005 	if (change != 0)
3006 		adapter->if_flags = ifp->if_flags;
3007 
3008 	if ((change & ~(IFF_CANTCHANGE | IFF_DEBUG)) != 0) {
3009 		rv = ENETRESET;
3010 		goto out;
3011 	} else if ((change & IFF_PROMISC) != 0) {
3012 		rv = ixv_set_rxfilter(adapter);
3013 		if (rv != 0) {
3014 			/* Restore previous */
3015 			adapter->if_flags = saved_flags;
3016 			goto out;
3017 		}
3018 	}
3019 
3020 	/* Check for ec_capenable. */
3021 	change = ec->ec_capenable ^ adapter->ec_capenable;
3022 	adapter->ec_capenable = ec->ec_capenable;
3023 	if ((change & ~(ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING
3024 	    | ETHERCAP_VLAN_HWFILTER)) != 0) {
3025 		rv = ENETRESET;
3026 		goto out;
3027 	}
3028 
3029 	/*
3030 	 * Special handling is not required for ETHERCAP_VLAN_MTU.
3031 	 * PF's MAXFRS(MHADD) does not include the 4bytes of the VLAN header.
3032 	 */
3033 
3034 	/* Set up VLAN support and filter */
3035 	if ((change & (ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_HWFILTER)) != 0)
3036 		rv = ixv_setup_vlan_support(adapter);
3037 
3038 out:
3039 	IXGBE_CORE_UNLOCK(adapter);
3040 
3041 	return rv;
3042 }
3043 
3044 
3045 /************************************************************************
3046  * ixv_ioctl - Ioctl entry point
3047  *
3048  *   Called when the user wants to configure the interface.
3049  *
3050  *   return 0 on success, positive on failure
3051  ************************************************************************/
3052 static int
3053 ixv_ioctl(struct ifnet *ifp, u_long command, void *data)
3054 {
3055 	struct adapter	*adapter = ifp->if_softc;
3056 	struct ixgbe_hw *hw = &adapter->hw;
3057 	struct ifcapreq *ifcr = data;
3058 	int		error;
3059 	int l4csum_en;
3060 	const int l4csum = IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx |
3061 	     IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_UDPv6_Rx;
3062 
3063 	switch (command) {
3064 	case SIOCSIFFLAGS:
3065 		IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)");
3066 		break;
3067 	case SIOCADDMULTI: {
3068 		struct ether_multi *enm;
3069 		struct ether_multistep step;
3070 		struct ethercom *ec = &adapter->osdep.ec;
3071 		bool overflow = false;
3072 		int mcnt = 0;
3073 
3074 		/*
3075 		 * Check the number of multicast address. If it exceeds,
3076 		 * return ENOSPC.
3077 		 * Update this code when we support API 1.3.
3078 		 */
3079 		ETHER_LOCK(ec);
3080 		ETHER_FIRST_MULTI(step, ec, enm);
3081 		while (enm != NULL) {
3082 			mcnt++;
3083 
3084 			/*
3085 			 * This code is before adding, so one room is required
3086 			 * at least.
3087 			 */
3088 			if (mcnt > (IXGBE_MAX_VF_MC - 1)) {
3089 				overflow = true;
3090 				break;
3091 			}
3092 			ETHER_NEXT_MULTI(step, enm);
3093 		}
3094 		ETHER_UNLOCK(ec);
3095 		error = 0;
3096 		if (overflow && ((ec->ec_flags & ETHER_F_ALLMULTI) == 0)) {
3097 			error = hw->mac.ops.update_xcast_mode(hw,
3098 			    IXGBEVF_XCAST_MODE_ALLMULTI);
3099 			if (error == IXGBE_ERR_NOT_TRUSTED) {
3100 				device_printf(adapter->dev,
3101 				    "this interface is not trusted\n");
3102 				error = EPERM;
3103 			} else if (error == IXGBE_ERR_FEATURE_NOT_SUPPORTED) {
3104 				device_printf(adapter->dev,
3105 				    "the PF doesn't support allmulti mode\n");
3106 				error = EOPNOTSUPP;
3107 			} else if (error) {
3108 				device_printf(adapter->dev,
3109 				    "number of Ethernet multicast addresses "
3110 				    "exceeds the limit (%d). error = %d\n",
3111 				    IXGBE_MAX_VF_MC, error);
3112 				error = ENOSPC;
3113 			} else
3114 				ec->ec_flags |= ETHER_F_ALLMULTI;
3115 		}
3116 		if (error)
3117 			return error;
3118 	}
3119 		/*FALLTHROUGH*/
3120 	case SIOCDELMULTI:
3121 		IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI");
3122 		break;
3123 	case SIOCSIFMEDIA:
3124 	case SIOCGIFMEDIA:
3125 		IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)");
3126 		break;
3127 	case SIOCSIFCAP:
3128 		IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)");
3129 		break;
3130 	case SIOCSIFMTU:
3131 		IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)");
3132 		break;
3133 	case SIOCZIFDATA:
3134 		IOCTL_DEBUGOUT("ioctl: SIOCZIFDATA (Zero counter)");
3135 		ixv_update_stats(adapter);
3136 		ixv_clear_evcnt(adapter);
3137 		break;
3138 	default:
3139 		IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)", (int)command);
3140 		break;
3141 	}
3142 
3143 	switch (command) {
3144 	case SIOCSIFCAP:
3145 		/* Layer-4 Rx checksum offload has to be turned on and
3146 		 * off as a unit.
3147 		 */
3148 		l4csum_en = ifcr->ifcr_capenable & l4csum;
3149 		if (l4csum_en != l4csum && l4csum_en != 0)
3150 			return EINVAL;
3151 		/*FALLTHROUGH*/
3152 	case SIOCADDMULTI:
3153 	case SIOCDELMULTI:
3154 	case SIOCSIFFLAGS:
3155 	case SIOCSIFMTU:
3156 	default:
3157 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
3158 			return error;
3159 		if ((ifp->if_flags & IFF_RUNNING) == 0)
3160 			;
3161 		else if (command == SIOCSIFCAP || command == SIOCSIFMTU) {
3162 			IXGBE_CORE_LOCK(adapter);
3163 			ixv_init_locked(adapter);
3164 			IXGBE_CORE_UNLOCK(adapter);
3165 		} else if (command == SIOCADDMULTI || command == SIOCDELMULTI) {
3166 			/*
3167 			 * Multicast list has changed; set the hardware filter
3168 			 * accordingly.
3169 			 */
3170 			IXGBE_CORE_LOCK(adapter);
3171 			ixv_disable_intr(adapter);
3172 			ixv_set_rxfilter(adapter);
3173 			ixv_enable_intr(adapter);
3174 			IXGBE_CORE_UNLOCK(adapter);
3175 		}
3176 		return 0;
3177 	}
3178 } /* ixv_ioctl */
3179 
3180 /************************************************************************
3181  * ixv_init
3182  ************************************************************************/
3183 static int
3184 ixv_init(struct ifnet *ifp)
3185 {
3186 	struct adapter *adapter = ifp->if_softc;
3187 
3188 	IXGBE_CORE_LOCK(adapter);
3189 	ixv_init_locked(adapter);
3190 	IXGBE_CORE_UNLOCK(adapter);
3191 
3192 	return 0;
3193 } /* ixv_init */
3194 
3195 /************************************************************************
3196  * ixv_handle_que
3197  ************************************************************************/
3198 static void
3199 ixv_handle_que(void *context)
3200 {
3201 	struct ix_queue *que = context;
3202 	struct adapter	*adapter = que->adapter;
3203 	struct tx_ring	*txr = que->txr;
3204 	struct ifnet	*ifp = adapter->ifp;
3205 	bool		more;
3206 
3207 	que->handleq.ev_count++;
3208 
3209 	if (ifp->if_flags & IFF_RUNNING) {
3210 		more = ixgbe_rxeof(que);
3211 		IXGBE_TX_LOCK(txr);
3212 		more |= ixgbe_txeof(txr);
3213 		if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX))
3214 			if (!ixgbe_mq_ring_empty(ifp, txr->txr_interq))
3215 				ixgbe_mq_start_locked(ifp, txr);
3216 		/* Only for queue 0 */
3217 		/* NetBSD still needs this for CBQ */
3218 		if ((&adapter->queues[0] == que)
3219 		    && (!ixgbe_legacy_ring_empty(ifp, NULL)))
3220 			ixgbe_legacy_start_locked(ifp, txr);
3221 		IXGBE_TX_UNLOCK(txr);
3222 		if (more) {
3223 			que->req.ev_count++;
3224 			if (adapter->txrx_use_workqueue) {
3225 				/*
3226 				 * "enqueued flag" is not required here
3227 				 * the same as ixg(4). See ixgbe_msix_que().
3228 				 */
3229 				workqueue_enqueue(adapter->que_wq,
3230 				    &que->wq_cookie, curcpu());
3231 			} else
3232 				  softint_schedule(que->que_si);
3233 			return;
3234 		}
3235 	}
3236 
3237 	/* Re-enable this interrupt */
3238 	ixv_enable_queue(adapter, que->msix);
3239 
3240 	return;
3241 } /* ixv_handle_que */
3242 
3243 /************************************************************************
3244  * ixv_handle_que_work
3245  ************************************************************************/
3246 static void
3247 ixv_handle_que_work(struct work *wk, void *context)
3248 {
3249 	struct ix_queue *que = container_of(wk, struct ix_queue, wq_cookie);
3250 
3251 	/*
3252 	 * "enqueued flag" is not required here the same as ixg(4).
3253 	 * See ixgbe_msix_que().
3254 	 */
3255 	ixv_handle_que(que);
3256 }
3257 
3258 /************************************************************************
3259  * ixv_allocate_msix - Setup MSI-X Interrupt resources and handlers
3260  ************************************************************************/
3261 static int
3262 ixv_allocate_msix(struct adapter *adapter, const struct pci_attach_args *pa)
3263 {
3264 	device_t	dev = adapter->dev;
3265 	struct ix_queue *que = adapter->queues;
3266 	struct		tx_ring *txr = adapter->tx_rings;
3267 	int		error, msix_ctrl, rid, vector = 0;
3268 	pci_chipset_tag_t pc;
3269 	pcitag_t	tag;
3270 	char		intrbuf[PCI_INTRSTR_LEN];
3271 	char		wqname[MAXCOMLEN];
3272 	char		intr_xname[32];
3273 	const char	*intrstr = NULL;
3274 	kcpuset_t	*affinity;
3275 	int		cpu_id = 0;
3276 
3277 	pc = adapter->osdep.pc;
3278 	tag = adapter->osdep.tag;
3279 
3280 	adapter->osdep.nintrs = adapter->num_queues + 1;
3281 	if (pci_msix_alloc_exact(pa, &adapter->osdep.intrs,
3282 	    adapter->osdep.nintrs) != 0) {
3283 		aprint_error_dev(dev,
3284 		    "failed to allocate MSI-X interrupt\n");
3285 		return (ENXIO);
3286 	}
3287 
3288 	kcpuset_create(&affinity, false);
3289 	for (int i = 0; i < adapter->num_queues; i++, vector++, que++, txr++) {
3290 		snprintf(intr_xname, sizeof(intr_xname), "%s TXRX%d",
3291 		    device_xname(dev), i);
3292 		intrstr = pci_intr_string(pc, adapter->osdep.intrs[i], intrbuf,
3293 		    sizeof(intrbuf));
3294 #ifdef IXGBE_MPSAFE
3295 		pci_intr_setattr(pc, &adapter->osdep.intrs[i], PCI_INTR_MPSAFE,
3296 		    true);
3297 #endif
3298 		/* Set the handler function */
3299 		que->res = adapter->osdep.ihs[i] = pci_intr_establish_xname(pc,
3300 		    adapter->osdep.intrs[i], IPL_NET, ixv_msix_que, que,
3301 		    intr_xname);
3302 		if (que->res == NULL) {
3303 			pci_intr_release(pc, adapter->osdep.intrs,
3304 			    adapter->osdep.nintrs);
3305 			aprint_error_dev(dev,
3306 			    "Failed to register QUE handler\n");
3307 			kcpuset_destroy(affinity);
3308 			return (ENXIO);
3309 		}
3310 		que->msix = vector;
3311 		adapter->active_queues |= (u64)(1 << que->msix);
3312 
3313 		cpu_id = i;
3314 		/* Round-robin affinity */
3315 		kcpuset_zero(affinity);
3316 		kcpuset_set(affinity, cpu_id % ncpu);
3317 		error = interrupt_distribute(adapter->osdep.ihs[i], affinity,
3318 		    NULL);
3319 		aprint_normal_dev(dev, "for TX/RX, interrupting at %s",
3320 		    intrstr);
3321 		if (error == 0)
3322 			aprint_normal(", bound queue %d to cpu %d\n",
3323 			    i, cpu_id % ncpu);
3324 		else
3325 			aprint_normal("\n");
3326 
3327 #ifndef IXGBE_LEGACY_TX
3328 		txr->txr_si
3329 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINT_FLAGS,
3330 			ixgbe_deferred_mq_start, txr);
3331 #endif
3332 		que->que_si
3333 		    = softint_establish(SOFTINT_NET | IXGBE_SOFTINT_FLAGS,
3334 			ixv_handle_que, que);
3335 		if (que->que_si == NULL) {
3336 			aprint_error_dev(dev,
3337 			    "could not establish software interrupt\n");
3338 		}
3339 	}
3340 	snprintf(wqname, sizeof(wqname), "%sdeferTx", device_xname(dev));
3341 	error = workqueue_create(&adapter->txr_wq, wqname,
3342 	    ixgbe_deferred_mq_start_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3343 	    IXGBE_WORKQUEUE_FLAGS);
3344 	if (error) {
3345 		aprint_error_dev(dev, "couldn't create workqueue for deferred Tx\n");
3346 	}
3347 	adapter->txr_wq_enqueued = percpu_alloc(sizeof(u_int));
3348 
3349 	snprintf(wqname, sizeof(wqname), "%sTxRx", device_xname(dev));
3350 	error = workqueue_create(&adapter->que_wq, wqname,
3351 	    ixv_handle_que_work, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3352 	    IXGBE_WORKQUEUE_FLAGS);
3353 	if (error) {
3354 		aprint_error_dev(dev,
3355 		    "couldn't create workqueue\n");
3356 	}
3357 
3358 	/* and Mailbox */
3359 	cpu_id++;
3360 	snprintf(intr_xname, sizeof(intr_xname), "%s link", device_xname(dev));
3361 	adapter->vector = vector;
3362 	intrstr = pci_intr_string(pc, adapter->osdep.intrs[vector], intrbuf,
3363 	    sizeof(intrbuf));
3364 #ifdef IXGBE_MPSAFE
3365 	pci_intr_setattr(pc, &adapter->osdep.intrs[vector], PCI_INTR_MPSAFE,
3366 	    true);
3367 #endif
3368 	/* Set the mbx handler function */
3369 	adapter->osdep.ihs[vector] = pci_intr_establish_xname(pc,
3370 	    adapter->osdep.intrs[vector], IPL_NET, ixv_msix_mbx, adapter,
3371 	    intr_xname);
3372 	if (adapter->osdep.ihs[vector] == NULL) {
3373 		aprint_error_dev(dev, "Failed to register LINK handler\n");
3374 		kcpuset_destroy(affinity);
3375 		return (ENXIO);
3376 	}
3377 	/* Round-robin affinity */
3378 	kcpuset_zero(affinity);
3379 	kcpuset_set(affinity, cpu_id % ncpu);
3380 	error = interrupt_distribute(adapter->osdep.ihs[vector], affinity,
3381 	    NULL);
3382 
3383 	aprint_normal_dev(dev,
3384 	    "for link, interrupting at %s", intrstr);
3385 	if (error == 0)
3386 		aprint_normal(", affinity to cpu %d\n", cpu_id % ncpu);
3387 	else
3388 		aprint_normal("\n");
3389 
3390 	/* Tasklets for Mailbox */
3391 	snprintf(wqname, sizeof(wqname), "%s-admin", device_xname(dev));
3392 	error = workqueue_create(&adapter->admin_wq, wqname,
3393 	    ixv_handle_admin, adapter, IXGBE_WORKQUEUE_PRI, IPL_NET,
3394 	    IXGBE_TASKLET_WQ_FLAGS);
3395 	if (error) {
3396 		aprint_error_dev(dev,
3397 		    "could not create admin workqueue (%d)\n", error);
3398 		goto err_out;
3399 	}
3400 
3401 	/*
3402 	 * Due to a broken design QEMU will fail to properly
3403 	 * enable the guest for MSI-X unless the vectors in
3404 	 * the table are all set up, so we must rewrite the
3405 	 * ENABLE in the MSI-X control register again at this
3406 	 * point to cause it to successfully initialize us.
3407 	 */
3408 	if (adapter->hw.mac.type == ixgbe_mac_82599_vf) {
3409 		pci_get_capability(pc, tag, PCI_CAP_MSIX, &rid, NULL);
3410 		rid += PCI_MSIX_CTL;
3411 		msix_ctrl = pci_conf_read(pc, tag, rid);
3412 		msix_ctrl |= PCI_MSIX_CTL_ENABLE;
3413 		pci_conf_write(pc, tag, rid, msix_ctrl);
3414 	}
3415 
3416 	kcpuset_destroy(affinity);
3417 	return (0);
3418 err_out:
3419 	kcpuset_destroy(affinity);
3420 	ixv_free_workqueue(adapter);
3421 	ixv_free_pci_resources(adapter);
3422 	return (error);
3423 } /* ixv_allocate_msix */
3424 
3425 /************************************************************************
3426  * ixv_configure_interrupts - Setup MSI-X resources
3427  *
3428  *   Note: The VF device MUST use MSI-X, there is no fallback.
3429  ************************************************************************/
3430 static int
3431 ixv_configure_interrupts(struct adapter *adapter)
3432 {
3433 	device_t dev = adapter->dev;
3434 	int want, queues, msgs;
3435 
3436 	/* Must have at least 2 MSI-X vectors */
3437 	msgs = pci_msix_count(adapter->osdep.pc, adapter->osdep.tag);
3438 	if (msgs < 2) {
3439 		aprint_error_dev(dev, "MSIX config error\n");
3440 		return (ENXIO);
3441 	}
3442 	msgs = MIN(msgs, IXG_MAX_NINTR);
3443 
3444 	/* Figure out a reasonable auto config value */
3445 	queues = (ncpu > (msgs - 1)) ? (msgs - 1) : ncpu;
3446 
3447 	if (ixv_num_queues != 0)
3448 		queues = ixv_num_queues;
3449 	else if ((ixv_num_queues == 0) && (queues > IXGBE_VF_MAX_TX_QUEUES))
3450 		queues = IXGBE_VF_MAX_TX_QUEUES;
3451 
3452 	/*
3453 	 * Want vectors for the queues,
3454 	 * plus an additional for mailbox.
3455 	 */
3456 	want = queues + 1;
3457 	if (msgs >= want)
3458 		msgs = want;
3459 	else {
3460 		aprint_error_dev(dev,
3461 		    "MSI-X Configuration Problem, "
3462 		    "%d vectors but %d queues wanted!\n",
3463 		    msgs, want);
3464 		return -1;
3465 	}
3466 
3467 	adapter->msix_mem = (void *)1; /* XXX */
3468 	aprint_normal_dev(dev,
3469 	    "Using MSI-X interrupts with %d vectors\n", msgs);
3470 	adapter->num_queues = queues;
3471 
3472 	return (0);
3473 } /* ixv_configure_interrupts */
3474 
3475 
3476 /************************************************************************
3477  * ixv_handle_admin - Tasklet handler for MSI-X MBX interrupts
3478  *
3479  *   Done outside of interrupt context since the driver might sleep
3480  ************************************************************************/
3481 static void
3482 ixv_handle_admin(struct work *wk, void *context)
3483 {
3484 	struct adapter *adapter = context;
3485 	struct ixgbe_hw	*hw = &adapter->hw;
3486 
3487 	IXGBE_CORE_LOCK(adapter);
3488 
3489 	++adapter->link_workev.ev_count;
3490 	adapter->hw.mac.ops.check_link(&adapter->hw, &adapter->link_speed,
3491 	    &adapter->link_up, FALSE);
3492 	ixv_update_link_status(adapter);
3493 
3494 	adapter->task_requests = 0;
3495 	atomic_and_uint(&adapter->admin_pending, ~1);
3496 
3497 	/* Re-enable interrupts */
3498 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, (1 << adapter->vector));
3499 
3500 	IXGBE_CORE_UNLOCK(adapter);
3501 } /* ixv_handle_admin */
3502 
3503 /************************************************************************
3504  * ixv_check_link - Used in the local timer to poll for link changes
3505  ************************************************************************/
3506 static s32
3507 ixv_check_link(struct adapter *adapter)
3508 {
3509 	s32 error;
3510 
3511 	KASSERT(mutex_owned(&adapter->core_mtx));
3512 
3513 	adapter->hw.mac.get_link_status = TRUE;
3514 
3515 	error = adapter->hw.mac.ops.check_link(&adapter->hw,
3516 	    &adapter->link_speed, &adapter->link_up, FALSE);
3517 	ixv_update_link_status(adapter);
3518 
3519 	return error;
3520 } /* ixv_check_link */
3521