xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_main.c (revision 51396a8ee7fb52fe0ab33bfe7b4f495ad431904a)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source.  A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * Copyright 2015 OmniTI Computer Consulting, Inc. All rights reserved.
14  * Copyright 2019 Joyent, Inc.
15  * Copyright 2017 Tegile Systems, Inc.  All rights reserved.
16  * Copyright 2020 RackTop Systems, Inc.
17  * Copyright 2020 Ryan Zezeski
18  */
19 
20 /*
21  * i40e - Intel 10/40 Gb Ethernet driver
22  *
23  * The i40e driver is the main software device driver for the Intel 40 Gb family
24  * of devices. Note that these devices come in many flavors with both 40 GbE
25  * ports and 10 GbE ports. This device is the successor to the 82599 family of
26  * devices (ixgbe).
27  *
28  * Unlike previous generations of Intel 1 GbE and 10 GbE devices, the 40 GbE
29  * devices defined in the XL710 controller (previously known as Fortville) are a
30  * rather different beast and have a small switch embedded inside of them. In
31  * addition, the way that most of the programming is done has been overhauled.
32  * As opposed to just using PCIe memory mapped registers, it also has an
33  * administrative queue which is used to communicate with firmware running on
34  * the chip.
35  *
36  * Each physical function in the hardware shows up as a device that this driver
37  * will bind to. The hardware splits many resources evenly across all of the
38  * physical functions present on the device, while other resources are instead
39  * shared across the entire card and its up to the device driver to
40  * intelligently partition them.
41  *
42  * ------------
43  * Organization
44  * ------------
45  *
46  * This driver is made up of several files which have their own theory
47  * statements spread across them. We'll touch on the high level purpose of each
48  * file here, and then we'll get into more discussion on how the device is
49  * generally modelled with respect to the interfaces in illumos.
50  *
51  * i40e_gld.c: This file contains all of the bindings to MAC and the networking
52  *             stack.
53  *
54  * i40e_intr.c: This file contains all of the interrupt service routines and
55  *              contains logic to enable and disable interrupts on the hardware.
56  *              It also contains the logic to map hardware resources such as the
57  *              rings to and from interrupts and controls their ability to fire.
58  *
59  *              There is a big theory statement on interrupts present there.
60  *
61  * i40e_main.c: The file that you're currently in. It interfaces with the
62  *              traditional OS DDI interfaces and is in charge of configuring
63  *              the device.
64  *
65  * i40e_osdep.[ch]: These files contain interfaces and definitions needed to
66  *                  work with Intel's common code for the device.
67  *
68  * i40e_stats.c: This file contains the general work and logic around our
69  *               kstats. A theory statement on their organization and use of the
70  *               hardware exists there.
71  *
72  * i40e_sw.h: This header file contains all of the primary structure definitions
73  *            and constants that are used across the entire driver.
74  *
75  * i40e_transceiver.c: This file contains all of the logic for sending and
76  *                     receiving data. It contains all of the ring and DMA
77  *                     allocation logic, as well as, the actual interfaces to
78  *                     send and receive data.
79  *
80  *                     A big theory statement on ring management, descriptors,
81  *                     and how it ties into the OS is present there.
82  *
83  * --------------
84  * General Design
85  * --------------
86  *
87  * Before we go too far into the general way we've laid out data structures and
88  * the like, it's worth taking some time to explain how the hardware is
89  * organized. This organization informs a lot of how we do things at this time
90  * in the driver.
91  *
92  * Each physical device consists of a number of one or more ports, which are
93  * considered physical functions in the PCI sense and thus each get enumerated
94  * by the system, resulting in an instance being created and attached to. While
95  * there are many resources that are unique to each physical function eg.
96  * instance of the device, there are many that are shared across all of them.
97  * Several resources have an amount reserved for each Virtual Station Interface
98  * (VSI) and then a static pool of resources, available for all functions on the
99  * card.
100  *
101  * The most important resource in hardware are its transmit and receive queue
102  * pairs (i40e_trqpair_t). These should be thought of as rings in GLDv3
103  * parlance. There are a set number of these on each device; however, they are
104  * statically partitioned among all of the different physical functions.
105  *
106  * 'Fortville' (the code name for this device family) is basically a switch. To
107  * map MAC addresses and other things to queues, we end up having to create
108  * Virtual Station Interfaces (VSIs) and establish forwarding rules that direct
109  * traffic to a queue. A VSI owns a collection of queues and has a series of
110  * forwarding rules that point to it. One way to think of this is to treat it
111  * like MAC does a VNIC. When MAC refers to a group, a collection of rings and
112  * classification resources, that is a VSI in i40e.
113  *
114  * The sets of VSIs is shared across the entire device, though there may be some
115  * amount that are reserved to each PF. Because the GLDv3 does not let us change
116  * the number of groups dynamically, we instead statically divide this amount
117  * evenly between all the functions that exist. In addition, we have the same
118  * problem with the mac address forwarding rules. There are a static number that
119  * exist shared across all the functions.
120  *
121  * To handle both of these resources, what we end up doing is going through and
122  * determining which functions belong to the same device. Nominally one might do
123  * this by having a nexus driver; however, a prime requirement for a nexus
124  * driver is identifying the various children and activating them. While it is
125  * possible to get this information from NVRAM, we would end up duplicating a
126  * lot of the PCI enumeration logic. Really, at the end of the day, the device
127  * doesn't give us the traditional identification properties we want from a
128  * nexus driver.
129  *
130  * Instead, we rely on some properties that are guaranteed to be unique. While
131  * it might be tempting to leverage the PBA or serial number of the device from
132  * NVRAM, there is nothing that says that two devices can't be mis-programmed to
133  * have the same values in NVRAM. Instead, we uniquely identify a group of
134  * functions based on their parent in the /devices tree, their PCI bus and PCI
135  * function identifiers. Using either on their own may not be sufficient.
136  *
137  * For each unique PCI device that we encounter, we'll create a i40e_device_t.
138  * From there, because we don't have a good way to tell the GLDv3 about sharing
139  * resources between everything, we'll end up just dividing the resources
140  * evenly between all of the functions. Longer term, if we don't have to declare
141  * to the GLDv3 that these resources are shared, then we'll maintain a pool and
142  * have each PF allocate from the pool in the device, thus if only two of four
143  * ports are being used, for example, then all of the resources can still be
144  * used.
145  *
146  * -------------------------------------------
147  * Transmit and Receive Queue Pair Allocations
148  * -------------------------------------------
149  *
150  * NVRAM ends up assigning each PF its own share of the transmit and receive LAN
151  * queue pairs, we have no way of modifying it, only observing it. From there,
152  * it's up to us to map these queues to VSIs and VFs. Since we don't support any
153  * VFs at this time, we only focus on assignments to VSIs.
154  *
155  * At the moment, we used a static mapping of transmit/receive queue pairs to a
156  * given VSI (eg. rings to a group). Though in the fullness of time, we want to
157  * make this something which is fully dynamic and take advantage of documented,
158  * but not yet available functionality for adding filters based on VXLAN and
159  * other encapsulation technologies.
160  *
161  * -------------------------------------
162  * Broadcast, Multicast, and Promiscuous
163  * -------------------------------------
164  *
165  * As part of the GLDv3, we need to make sure that we can handle receiving
166  * broadcast and multicast traffic. As well as enabling promiscuous mode when
167  * requested. GLDv3 requires that all broadcast and multicast traffic be
168  * retrieved by the default group, eg. the first one. This is the same thing as
169  * the default VSI.
170  *
171  * To receieve broadcast traffic, we enable it through the admin queue, rather
172  * than use one of our filters for it. For multicast traffic, we reserve a
173  * certain number of the hash filters and assign them to a given PF. When we
174  * exceed those, we then switch to using promiscuous mode for multicast traffic.
175  *
176  * More specifically, once we exceed the number of filters (indicated because
177  * the i40e_t`i40e_resources.ifr_nmcastfilt ==
178  * i40e_t`i40e_resources.ifr_nmcastfilt_used), we then instead need to toggle
179  * promiscuous mode. If promiscuous mode is toggled then we keep track of the
180  * number of MACs added to it by incrementing i40e_t`i40e_mcast_promisc_count.
181  * That will stay enabled until that count reaches zero indicating that we have
182  * only added multicast addresses that we have a corresponding entry for.
183  *
184  * Because MAC itself wants to toggle promiscuous mode, which includes both
185  * unicast and multicast traffic, we go through and keep track of that
186  * ourselves. That is maintained through the use of the i40e_t`i40e_promisc_on
187  * member.
188  *
189  * --------------
190  * VSI Management
191  * --------------
192  *
193  * The PFs share 384 VSIs. The firmware creates one VSI per PF by default.
194  * During chip start we retrieve the SEID of this VSI and assign it as the
195  * default VSI for our VEB (one VEB per PF). We then add additional VSIs to
196  * the VEB up to the determined number of rx groups: i40e_t`i40e_num_rx_groups.
197  * We currently cap this number to I40E_GROUP_MAX to a) make sure all PFs can
198  * allocate the same number of VSIs, and b) to keep the interrupt multiplexing
199  * under control. In the future, when we improve the interrupt allocation, we
200  * may want to revisit this cap to make better use of the available VSIs. The
201  * VSI allocation and configuration can be found in i40e_chip_start().
202  *
203  * ----------------
204  * Structure Layout
205  * ----------------
206  *
207  * The following images relates the core data structures together. The primary
208  * structure in the system is the i40e_t. It itself contains multiple rings,
209  * i40e_trqpair_t's which contain the various transmit and receive data. The
210  * receive data is stored outside of the i40e_trqpair_t and instead in the
211  * i40e_rx_data_t. The i40e_t has a corresponding i40e_device_t which keeps
212  * track of per-physical device state. Finally, for every active descriptor,
213  * there is a corresponding control block, which is where the
214  * i40e_rx_control_block_t and the i40e_tx_control_block_t come from.
215  *
216  *   +-----------------------+       +-----------------------+
217  *   | Global i40e_t list    |       | Global Device list    |
218  *   |                       |    +--|                       |
219  *   | i40e_glist            |    |  | i40e_dlist            |
220  *   +-----------------------+    |  +-----------------------+
221  *       |                        v
222  *       |      +------------------------+      +-----------------------+
223  *       |      | Device-wide Structure  |----->| Device-wide Structure |--> ...
224  *       |      | i40e_device_t          |      | i40e_device_t         |
225  *       |      |                        |      +-----------------------+
226  *       |      | dev_info_t *     ------+--> Parent in devices tree.
227  *       |      | uint_t           ------+--> PCI bus number
228  *       |      | uint_t           ------+--> PCI device number
229  *       |      | uint_t           ------+--> Number of functions
230  *       |      | i40e_switch_rsrcs_t ---+--> Captured total switch resources
231  *       |      | list_t           ------+-------------+
232  *       |      +------------------------+             |
233  *       |                           ^                 |
234  *       |                           +--------+        |
235  *       |                                    |        v
236  *       |  +---------------------------+     |   +-------------------+
237  *       +->| GLDv3 Device, per PF      |-----|-->| GLDv3 Device (PF) |--> ...
238  *          | i40e_t                    |     |   | i40e_t            |
239  *          | **Primary Structure**     |     |   +-------------------+
240  *          |                           |     |
241  *          | i40e_device_t *         --+-----+
242  *          | i40e_state_t            --+---> Device State
243  *          | i40e_hw_t               --+---> Intel common code structure
244  *          | mac_handle_t            --+---> GLDv3 handle to MAC
245  *          | ddi_periodic_t          --+---> Link activity timer
246  *          | i40e_vsi_t *            --+---> Array of VSIs
247  *          | i40e_func_rsrc_t        --+---> Available hardware resources
248  *          | i40e_switch_rsrc_t *    --+---> Switch resource snapshot
249  *          | i40e_sdu                --+---> Current MTU
250  *          | i40e_frame_max          --+---> Current HW frame size
251  *          | i40e_uaddr_t *          --+---> Array of assigned unicast MACs
252  *          | i40e_maddr_t *          --+---> Array of assigned multicast MACs
253  *          | i40e_mcast_promisccount --+---> Active multicast state
254  *          | i40e_promisc_on         --+---> Current promiscuous mode state
255  *          | uint_t                  --+---> Number of transmit/receive pairs
256  *          | i40e_rx_group_t *       --+---> Array of Rx groups
257  *          | kstat_t *               --+---> PF kstats
258  *          | i40e_pf_stats_t         --+---> PF kstat backing data
259  *          | i40e_trqpair_t *        --+---------+
260  *          +---------------------------+         |
261  *                                                |
262  *                                                v
263  *  +-------------------------------+       +-----------------------------+
264  *  | Transmit/Receive Queue Pair   |-------| Transmit/Receive Queue Pair |->...
265  *  | i40e_trqpair_t                |       | i40e_trqpair_t              |
266  *  + Ring Data Structure           |       +-----------------------------+
267  *  |                               |
268  *  | mac_ring_handle_t             +--> MAC RX ring handle
269  *  | mac_ring_handle_t             +--> MAC TX ring handle
270  *  | i40e_rxq_stat_t             --+--> RX Queue stats
271  *  | i40e_txq_stat_t             --+--> TX Queue stats
272  *  | uint32_t (tx ring size)       +--> TX Ring Size
273  *  | uint32_t (tx free list size)  +--> TX Free List Size
274  *  | i40e_dma_buffer_t     --------+--> TX Descriptor ring DMA
275  *  | i40e_tx_desc_t *      --------+--> TX descriptor ring
276  *  | volatile unt32_t *            +--> TX Write back head
277  *  | uint32_t               -------+--> TX ring head
278  *  | uint32_t               -------+--> TX ring tail
279  *  | uint32_t               -------+--> Num TX desc free
280  *  | i40e_tx_control_block_t *   --+--> TX control block array  ---+
281  *  | i40e_tx_control_block_t **  --+--> TCB work list          ----+
282  *  | i40e_tx_control_block_t **  --+--> TCB free list           ---+
283  *  | uint32_t               -------+--> Free TCB count             |
284  *  | i40e_rx_data_t *       -------+--+                            v
285  *  +-------------------------------+  |          +---------------------------+
286  *                                     |          | Per-TX Frame Metadata     |
287  *                                     |          | i40e_tx_control_block_t   |
288  *                +--------------------+          |                           |
289  *                |           mblk to transmit <--+---      mblk_t *          |
290  *                |           type of transmit <--+---      i40e_tx_type_t    |
291  *                |              TX DMA handle <--+---      ddi_dma_handle_t  |
292  *                v              TX DMA buffer <--+---      i40e_dma_buffer_t |
293  *    +------------------------------+            +---------------------------+
294  *    | Core Receive Data            |
295  *    | i40e_rx_data_t               |
296  *    |                              |
297  *    | i40e_dma_buffer_t          --+--> RX descriptor DMA Data
298  *    | i40e_rx_desc_t             --+--> RX descriptor ring
299  *    | uint32_t                   --+--> Next free desc.
300  *    | i40e_rx_control_block_t *  --+--> RX Control Block Array  ---+
301  *    | i40e_rx_control_block_t ** --+--> RCB work list           ---+
302  *    | i40e_rx_control_block_t ** --+--> RCB free list           ---+
303  *    +------------------------------+                               |
304  *                ^                                                  |
305  *                |     +---------------------------+                |
306  *                |     | Per-RX Frame Metadata     |<---------------+
307  *                |     | i40e_rx_control_block_t   |
308  *                |     |                           |
309  *                |     | mblk_t *              ----+--> Received mblk_t data
310  *                |     | uint32_t              ----+--> Reference count
311  *                |     | i40e_dma_buffer_t     ----+--> Receive data DMA info
312  *                |     | frtn_t                ----+--> mblk free function info
313  *                +-----+-- i40e_rx_data_t *        |
314  *                      +---------------------------+
315  *
316  * -------------
317  * Lock Ordering
318  * -------------
319  *
320  * In order to ensure that we don't deadlock, the following represents the
321  * lock order being used. When grabbing locks, follow the following order. Lower
322  * numbers are more important. Thus, the i40e_glock which is number 0, must be
323  * taken before any other locks in the driver. On the other hand, the
324  * i40e_t`i40e_stat_lock, has the highest number because it's the least
325  * important lock. Note, that just because one lock is higher than another does
326  * not mean that all intermediary locks are required.
327  *
328  * 0) i40e_glock
329  * 1) i40e_t`i40e_general_lock
330  *
331  * 2) i40e_trqpair_t`itrq_rx_lock
332  * 3) i40e_trqpair_t`itrq_tx_lock
333  * 4) i40e_trqpair_t`itrq_intr_lock
334  * 5) i40e_t`i40e_rx_pending_lock
335  * 6) i40e_trqpair_t`itrq_tcb_lock
336  *
337  * 7) i40e_t`i40e_stat_lock
338  *
339  * Rules and expectations:
340  *
341  * 1) A thread holding locks belong to one PF should not hold locks belonging to
342  * a second. If for some reason this becomes necessary, locks should be grabbed
343  * based on the list order in the i40e_device_t, which implies that the
344  * i40e_glock is held.
345  *
346  * 2) When grabbing locks between multiple transmit and receive queues, the
347  * locks for the lowest number transmit/receive queue should be grabbed first.
348  *
349  * 3) When grabbing both the transmit and receive lock for a given queue, always
350  * grab i40e_trqpair_t`itrq_rx_lock before the i40e_trqpair_t`itrq_tx_lock.
351  *
352  * 4) The following pairs of locks are not expected to be held at the same time:
353  *
354  * o i40e_t`i40e_rx_pending_lock and i40e_trqpair_t`itrq_tcb_lock
355  * o i40e_trqpair_t`itrq_intr_lock is not expected to be held with any
356  *   other lock except i40e_t`i40e_general_lock in mc_start(9E) and
357  *   mc_stop(9e).
358  *
359  * -----------
360  * Future Work
361  * -----------
362  *
363  * At the moment the i40e_t driver is rather bare bones, allowing us to start
364  * getting data flowing and folks using it while we develop additional features.
365  * While bugs have been filed to cover this future work, the following gives an
366  * overview of expected work:
367  *
368  *  o DMA binding and breaking up the locking in ring recycling.
369  *  o Enhanced detection of device errors
370  *  o Participation in IRM
371  *  o FMA device reset
372  *  o Stall detection, temperature error detection, etc.
373  *  o More dynamic resource pools
374  */
375 
376 #include "i40e_sw.h"
377 
378 static char i40e_ident[] = "Intel 10/40Gb Ethernet v1.0.3";
379 
380 /*
381  * The i40e_glock primarily protects the lists below and the i40e_device_t
382  * structures.
383  */
384 static kmutex_t i40e_glock;
385 static list_t i40e_glist;
386 static list_t i40e_dlist;
387 
388 /*
389  * Access attributes for register mapping.
390  */
391 static ddi_device_acc_attr_t i40e_regs_acc_attr = {
392 	DDI_DEVICE_ATTR_V1,
393 	DDI_STRUCTURE_LE_ACC,
394 	DDI_STRICTORDER_ACC,
395 	DDI_FLAGERR_ACC
396 };
397 
398 /*
399  * Logging function for this driver.
400  */
401 static void
402 i40e_dev_err(i40e_t *i40e, int level, boolean_t console, const char *fmt,
403     va_list ap)
404 {
405 	char buf[1024];
406 
407 	(void) vsnprintf(buf, sizeof (buf), fmt, ap);
408 
409 	if (i40e == NULL) {
410 		cmn_err(level, (console) ? "%s: %s" : "!%s: %s",
411 		    I40E_MODULE_NAME, buf);
412 	} else {
413 		dev_err(i40e->i40e_dip, level, (console) ? "%s" : "!%s",
414 		    buf);
415 	}
416 }
417 
418 /*
419  * Because there's the stupid trailing-comma problem with the C preprocessor
420  * and variable arguments, I need to instantiate these.	 Pardon the redundant
421  * code.
422  */
423 /*PRINTFLIKE2*/
424 void
425 i40e_error(i40e_t *i40e, const char *fmt, ...)
426 {
427 	va_list ap;
428 
429 	va_start(ap, fmt);
430 	i40e_dev_err(i40e, CE_WARN, B_FALSE, fmt, ap);
431 	va_end(ap);
432 }
433 
434 /*PRINTFLIKE2*/
435 void
436 i40e_log(i40e_t *i40e, const char *fmt, ...)
437 {
438 	va_list ap;
439 
440 	va_start(ap, fmt);
441 	i40e_dev_err(i40e, CE_NOTE, B_FALSE, fmt, ap);
442 	va_end(ap);
443 }
444 
445 /*PRINTFLIKE2*/
446 void
447 i40e_notice(i40e_t *i40e, const char *fmt, ...)
448 {
449 	va_list ap;
450 
451 	va_start(ap, fmt);
452 	i40e_dev_err(i40e, CE_NOTE, B_TRUE, fmt, ap);
453 	va_end(ap);
454 }
455 
456 /*
457  * Various parts of the driver need to know if the controller is from the X722
458  * family, which has a few additional capabilities and different programming
459  * means. We don't consider virtual functions as part of this as they are quite
460  * different and will require substantially more work.
461  */
462 static boolean_t
463 i40e_is_x722(i40e_t *i40e)
464 {
465 	return (i40e->i40e_hw_space.mac.type == I40E_MAC_X722);
466 }
467 
468 static void
469 i40e_device_rele(i40e_t *i40e)
470 {
471 	i40e_device_t *idp = i40e->i40e_device;
472 
473 	if (idp == NULL)
474 		return;
475 
476 	mutex_enter(&i40e_glock);
477 	VERIFY(idp->id_nreg > 0);
478 	list_remove(&idp->id_i40e_list, i40e);
479 	idp->id_nreg--;
480 	if (idp->id_nreg == 0) {
481 		list_remove(&i40e_dlist, idp);
482 		list_destroy(&idp->id_i40e_list);
483 		kmem_free(idp->id_rsrcs, sizeof (i40e_switch_rsrc_t) *
484 		    idp->id_rsrcs_alloc);
485 		kmem_free(idp, sizeof (i40e_device_t));
486 	}
487 	i40e->i40e_device = NULL;
488 	mutex_exit(&i40e_glock);
489 }
490 
491 static i40e_device_t *
492 i40e_device_find(i40e_t *i40e, dev_info_t *parent, uint_t bus, uint_t device)
493 {
494 	i40e_device_t *idp;
495 	mutex_enter(&i40e_glock);
496 	for (idp = list_head(&i40e_dlist); idp != NULL;
497 	    idp = list_next(&i40e_dlist, idp)) {
498 		if (idp->id_parent == parent && idp->id_pci_bus == bus &&
499 		    idp->id_pci_device == device) {
500 			break;
501 		}
502 	}
503 
504 	if (idp != NULL) {
505 		VERIFY(idp->id_nreg < idp->id_nfuncs);
506 		idp->id_nreg++;
507 	} else {
508 		i40e_hw_t *hw = &i40e->i40e_hw_space;
509 		ASSERT(hw->num_ports > 0);
510 		ASSERT(hw->num_partitions > 0);
511 
512 		/*
513 		 * The Intel common code doesn't exactly keep the number of PCI
514 		 * functions. But it calculates it during discovery of
515 		 * partitions and ports. So what we do is undo the calculation
516 		 * that it does originally, as functions are evenly spread
517 		 * across ports in the rare case of partitions.
518 		 */
519 		idp = kmem_alloc(sizeof (i40e_device_t), KM_SLEEP);
520 		idp->id_parent = parent;
521 		idp->id_pci_bus = bus;
522 		idp->id_pci_device = device;
523 		idp->id_nfuncs = hw->num_ports * hw->num_partitions;
524 		idp->id_nreg = 1;
525 		idp->id_rsrcs_alloc = i40e->i40e_switch_rsrc_alloc;
526 		idp->id_rsrcs_act = i40e->i40e_switch_rsrc_actual;
527 		idp->id_rsrcs = kmem_alloc(sizeof (i40e_switch_rsrc_t) *
528 		    idp->id_rsrcs_alloc, KM_SLEEP);
529 		bcopy(i40e->i40e_switch_rsrcs, idp->id_rsrcs,
530 		    sizeof (i40e_switch_rsrc_t) * idp->id_rsrcs_alloc);
531 		list_create(&idp->id_i40e_list, sizeof (i40e_t),
532 		    offsetof(i40e_t, i40e_dlink));
533 
534 		list_insert_tail(&i40e_dlist, idp);
535 	}
536 
537 	list_insert_tail(&idp->id_i40e_list, i40e);
538 	mutex_exit(&i40e_glock);
539 
540 	return (idp);
541 }
542 
543 static void
544 i40e_link_state_set(i40e_t *i40e, link_state_t state)
545 {
546 	if (i40e->i40e_link_state == state)
547 		return;
548 
549 	i40e->i40e_link_state = state;
550 	mac_link_update(i40e->i40e_mac_hdl, i40e->i40e_link_state);
551 }
552 
553 /*
554  * This is a basic link check routine. Mostly we're using this just to see
555  * if we can get any accurate information about the state of the link being
556  * up or down, as well as updating the link state, speed, etc. information.
557  */
558 void
559 i40e_link_check(i40e_t *i40e)
560 {
561 	i40e_hw_t *hw = &i40e->i40e_hw_space;
562 	boolean_t ls;
563 	int ret;
564 
565 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
566 
567 	hw->phy.get_link_info = B_TRUE;
568 	if ((ret = i40e_get_link_status(hw, &ls)) != I40E_SUCCESS) {
569 		i40e->i40e_s_link_status_errs++;
570 		i40e->i40e_s_link_status_lasterr = ret;
571 		return;
572 	}
573 
574 	/*
575 	 * Firmware abstracts all of the mac and phy information for us, so we
576 	 * can use i40e_get_link_status to determine the current state.
577 	 */
578 	if (ls == B_TRUE) {
579 		enum i40e_aq_link_speed speed;
580 
581 		speed = i40e_get_link_speed(hw);
582 
583 		/*
584 		 * Translate from an i40e value to a value in Mbits/s.
585 		 */
586 		switch (speed) {
587 		case I40E_LINK_SPEED_100MB:
588 			i40e->i40e_link_speed = 100;
589 			break;
590 		case I40E_LINK_SPEED_1GB:
591 			i40e->i40e_link_speed = 1000;
592 			break;
593 		case I40E_LINK_SPEED_10GB:
594 			i40e->i40e_link_speed = 10000;
595 			break;
596 		case I40E_LINK_SPEED_20GB:
597 			i40e->i40e_link_speed = 20000;
598 			break;
599 		case I40E_LINK_SPEED_40GB:
600 			i40e->i40e_link_speed = 40000;
601 			break;
602 		case I40E_LINK_SPEED_25GB:
603 			i40e->i40e_link_speed = 25000;
604 			break;
605 		default:
606 			i40e->i40e_link_speed = 0;
607 			break;
608 		}
609 
610 		/*
611 		 * At this time, hardware does not support half-duplex
612 		 * operation, hence why we don't ask the hardware about our
613 		 * current speed.
614 		 */
615 		i40e->i40e_link_duplex = LINK_DUPLEX_FULL;
616 		i40e_link_state_set(i40e, LINK_STATE_UP);
617 	} else {
618 		i40e->i40e_link_speed = 0;
619 		i40e->i40e_link_duplex = 0;
620 		i40e_link_state_set(i40e, LINK_STATE_DOWN);
621 	}
622 }
623 
624 static void
625 i40e_rem_intrs(i40e_t *i40e)
626 {
627 	int i, rc;
628 
629 	for (i = 0; i < i40e->i40e_intr_count; i++) {
630 		rc = ddi_intr_free(i40e->i40e_intr_handles[i]);
631 		if (rc != DDI_SUCCESS) {
632 			i40e_log(i40e, "failed to free interrupt %d: %d",
633 			    i, rc);
634 		}
635 	}
636 
637 	kmem_free(i40e->i40e_intr_handles, i40e->i40e_intr_size);
638 	i40e->i40e_intr_handles = NULL;
639 }
640 
641 static void
642 i40e_rem_intr_handlers(i40e_t *i40e)
643 {
644 	int i, rc;
645 
646 	for (i = 0; i < i40e->i40e_intr_count; i++) {
647 		rc = ddi_intr_remove_handler(i40e->i40e_intr_handles[i]);
648 		if (rc != DDI_SUCCESS) {
649 			i40e_log(i40e, "failed to remove interrupt %d: %d",
650 			    i, rc);
651 		}
652 	}
653 }
654 
655 /*
656  * illumos Fault Management Architecture (FMA) support.
657  */
658 
659 int
660 i40e_check_acc_handle(ddi_acc_handle_t handle)
661 {
662 	ddi_fm_error_t de;
663 
664 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
665 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
666 	return (de.fme_status);
667 }
668 
669 int
670 i40e_check_dma_handle(ddi_dma_handle_t handle)
671 {
672 	ddi_fm_error_t de;
673 
674 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
675 	return (de.fme_status);
676 }
677 
678 /*
679  * Fault service error handling callback function.
680  */
681 /* ARGSUSED */
682 static int
683 i40e_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
684 {
685 	pci_ereport_post(dip, err, NULL);
686 	return (err->fme_status);
687 }
688 
689 static void
690 i40e_fm_init(i40e_t *i40e)
691 {
692 	ddi_iblock_cookie_t iblk;
693 
694 	i40e->i40e_fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY,
695 	    i40e->i40e_dip, DDI_PROP_DONTPASS, "fm_capable",
696 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
697 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
698 
699 	if (i40e->i40e_fm_capabilities < 0) {
700 		i40e->i40e_fm_capabilities = 0;
701 	} else if (i40e->i40e_fm_capabilities > 0xf) {
702 		i40e->i40e_fm_capabilities = DDI_FM_EREPORT_CAPABLE |
703 		    DDI_FM_ACCCHK_CAPABLE | DDI_FM_DMACHK_CAPABLE |
704 		    DDI_FM_ERRCB_CAPABLE;
705 	}
706 
707 	/*
708 	 * Only register with IO Fault Services if we have some capability
709 	 */
710 	if (i40e->i40e_fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
711 		i40e_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
712 	} else {
713 		i40e_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
714 	}
715 
716 	if (i40e->i40e_fm_capabilities) {
717 		ddi_fm_init(i40e->i40e_dip, &i40e->i40e_fm_capabilities, &iblk);
718 
719 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
720 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
721 			pci_ereport_setup(i40e->i40e_dip);
722 		}
723 
724 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
725 			ddi_fm_handler_register(i40e->i40e_dip,
726 			    i40e_fm_error_cb, (void*)i40e);
727 		}
728 	}
729 
730 	if (i40e->i40e_fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
731 		i40e_init_dma_attrs(i40e, B_TRUE);
732 	} else {
733 		i40e_init_dma_attrs(i40e, B_FALSE);
734 	}
735 }
736 
737 static void
738 i40e_fm_fini(i40e_t *i40e)
739 {
740 	if (i40e->i40e_fm_capabilities) {
741 
742 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
743 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
744 			pci_ereport_teardown(i40e->i40e_dip);
745 
746 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
747 			ddi_fm_handler_unregister(i40e->i40e_dip);
748 
749 		ddi_fm_fini(i40e->i40e_dip);
750 	}
751 }
752 
753 void
754 i40e_fm_ereport(i40e_t *i40e, char *detail)
755 {
756 	uint64_t ena;
757 	char buf[FM_MAX_CLASS];
758 
759 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
760 	ena = fm_ena_generate(0, FM_ENA_FMT1);
761 	if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities)) {
762 		ddi_fm_ereport_post(i40e->i40e_dip, buf, ena, DDI_NOSLEEP,
763 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
764 	}
765 }
766 
767 /*
768  * Here we're trying to set the SEID of the default VSI. In general,
769  * when we come through and look at this shortly after attach, we
770  * expect there to only be a single element present, which is the
771  * default VSI. Importantly, each PF seems to not see any other
772  * devices, in part because of the simple switch mode that we're
773  * using. If for some reason, we see more artifacts, we'll need to
774  * revisit what we're doing here.
775  */
776 static boolean_t
777 i40e_set_def_vsi_seid(i40e_t *i40e)
778 {
779 	i40e_hw_t *hw = &i40e->i40e_hw_space;
780 	struct i40e_aqc_get_switch_config_resp *sw_config;
781 	uint8_t aq_buf[I40E_AQ_LARGE_BUF];
782 	uint16_t next = 0;
783 	int rc;
784 
785 	/* LINTED: E_BAD_PTR_CAST_ALIGN */
786 	sw_config = (struct i40e_aqc_get_switch_config_resp *)aq_buf;
787 	rc = i40e_aq_get_switch_config(hw, sw_config, sizeof (aq_buf), &next,
788 	    NULL);
789 	if (rc != I40E_SUCCESS) {
790 		i40e_error(i40e, "i40e_aq_get_switch_config() failed %d: %d",
791 		    rc, hw->aq.asq_last_status);
792 		return (B_FALSE);
793 	}
794 
795 	if (LE_16(sw_config->header.num_reported) != 1) {
796 		i40e_error(i40e, "encountered multiple (%d) switching units "
797 		    "during attach, not proceeding",
798 		    LE_16(sw_config->header.num_reported));
799 		return (B_FALSE);
800 	}
801 
802 	I40E_DEF_VSI_SEID(i40e) = sw_config->element[0].seid;
803 	return (B_TRUE);
804 }
805 
806 /*
807  * Get the SEID of the uplink MAC.
808  */
809 static int
810 i40e_get_mac_seid(i40e_t *i40e)
811 {
812 	i40e_hw_t *hw = &i40e->i40e_hw_space;
813 	struct i40e_aqc_get_switch_config_resp *sw_config;
814 	uint8_t aq_buf[I40E_AQ_LARGE_BUF];
815 	uint16_t next = 0;
816 	int rc;
817 
818 	/* LINTED: E_BAD_PTR_CAST_ALIGN */
819 	sw_config = (struct i40e_aqc_get_switch_config_resp *)aq_buf;
820 	rc = i40e_aq_get_switch_config(hw, sw_config, sizeof (aq_buf), &next,
821 	    NULL);
822 	if (rc != I40E_SUCCESS) {
823 		i40e_error(i40e, "i40e_aq_get_switch_config() failed %d: %d",
824 		    rc, hw->aq.asq_last_status);
825 		return (-1);
826 	}
827 
828 	return (LE_16(sw_config->element[0].uplink_seid));
829 }
830 
831 /*
832  * We need to fill the i40e_hw_t structure with the capabilities of this PF. We
833  * must also provide the memory for it; however, we don't need to keep it around
834  * to the call to the common code. It takes it and parses it into an internal
835  * structure.
836  */
837 static boolean_t
838 i40e_get_hw_capabilities(i40e_t *i40e, i40e_hw_t *hw)
839 {
840 	struct i40e_aqc_list_capabilities_element_resp *buf;
841 	int rc;
842 	size_t len;
843 	uint16_t needed;
844 	int nelems = I40E_HW_CAP_DEFAULT;
845 
846 	len = nelems * sizeof (*buf);
847 
848 	for (;;) {
849 		ASSERT(len > 0);
850 		buf = kmem_alloc(len, KM_SLEEP);
851 		rc = i40e_aq_discover_capabilities(hw, buf, len,
852 		    &needed, i40e_aqc_opc_list_func_capabilities, NULL);
853 		kmem_free(buf, len);
854 
855 		if (hw->aq.asq_last_status == I40E_AQ_RC_ENOMEM &&
856 		    nelems == I40E_HW_CAP_DEFAULT) {
857 			if (nelems == needed) {
858 				i40e_error(i40e, "Capability discovery failed "
859 				    "due to byzantine common code");
860 				return (B_FALSE);
861 			}
862 			len = needed;
863 			continue;
864 		} else if (rc != I40E_SUCCESS ||
865 		    hw->aq.asq_last_status != I40E_AQ_RC_OK) {
866 			i40e_error(i40e, "Capability discovery failed: %d", rc);
867 			return (B_FALSE);
868 		}
869 
870 		break;
871 	}
872 
873 	return (B_TRUE);
874 }
875 
876 /*
877  * Obtain the switch's capabilities as seen by this PF and keep it around for
878  * our later use.
879  */
880 static boolean_t
881 i40e_get_switch_resources(i40e_t *i40e)
882 {
883 	i40e_hw_t *hw = &i40e->i40e_hw_space;
884 	uint8_t cnt = 2;
885 	uint8_t act;
886 	size_t size;
887 	i40e_switch_rsrc_t *buf;
888 
889 	for (;;) {
890 		enum i40e_status_code ret;
891 		size = cnt * sizeof (i40e_switch_rsrc_t);
892 		ASSERT(size > 0);
893 		if (size > UINT16_MAX)
894 			return (B_FALSE);
895 		buf = kmem_alloc(size, KM_SLEEP);
896 
897 		ret = i40e_aq_get_switch_resource_alloc(hw, &act, buf,
898 		    cnt, NULL);
899 		if (ret == I40E_ERR_ADMIN_QUEUE_ERROR &&
900 		    hw->aq.asq_last_status == I40E_AQ_RC_EINVAL) {
901 			kmem_free(buf, size);
902 			cnt += I40E_SWITCH_CAP_DEFAULT;
903 			continue;
904 		} else if (ret != I40E_SUCCESS) {
905 			kmem_free(buf, size);
906 			i40e_error(i40e,
907 			    "failed to retrieve switch statistics: %d", ret);
908 			return (B_FALSE);
909 		}
910 
911 		break;
912 	}
913 
914 	i40e->i40e_switch_rsrc_alloc = cnt;
915 	i40e->i40e_switch_rsrc_actual = act;
916 	i40e->i40e_switch_rsrcs = buf;
917 
918 	return (B_TRUE);
919 }
920 
921 static void
922 i40e_cleanup_resources(i40e_t *i40e)
923 {
924 	if (i40e->i40e_uaddrs != NULL) {
925 		kmem_free(i40e->i40e_uaddrs, sizeof (i40e_uaddr_t) *
926 		    i40e->i40e_resources.ifr_nmacfilt);
927 		i40e->i40e_uaddrs = NULL;
928 	}
929 
930 	if (i40e->i40e_maddrs != NULL) {
931 		kmem_free(i40e->i40e_maddrs, sizeof (i40e_maddr_t) *
932 		    i40e->i40e_resources.ifr_nmcastfilt);
933 		i40e->i40e_maddrs = NULL;
934 	}
935 
936 	if (i40e->i40e_switch_rsrcs != NULL) {
937 		size_t sz = sizeof (i40e_switch_rsrc_t) *
938 		    i40e->i40e_switch_rsrc_alloc;
939 		ASSERT(sz > 0);
940 		kmem_free(i40e->i40e_switch_rsrcs, sz);
941 		i40e->i40e_switch_rsrcs = NULL;
942 	}
943 
944 	if (i40e->i40e_device != NULL)
945 		i40e_device_rele(i40e);
946 }
947 
948 static boolean_t
949 i40e_get_available_resources(i40e_t *i40e)
950 {
951 	dev_info_t *parent;
952 	uint16_t bus, device, func;
953 	uint_t nregs;
954 	int *regs, i;
955 	i40e_device_t *idp;
956 	i40e_hw_t *hw = &i40e->i40e_hw_space;
957 
958 	parent = ddi_get_parent(i40e->i40e_dip);
959 
960 	if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, i40e->i40e_dip, 0, "reg",
961 	    &regs, &nregs) != DDI_PROP_SUCCESS) {
962 		return (B_FALSE);
963 	}
964 
965 	if (nregs < 1) {
966 		ddi_prop_free(regs);
967 		return (B_FALSE);
968 	}
969 
970 	bus = PCI_REG_BUS_G(regs[0]);
971 	device = PCI_REG_DEV_G(regs[0]);
972 	func = PCI_REG_FUNC_G(regs[0]);
973 	ddi_prop_free(regs);
974 
975 	i40e->i40e_hw_space.bus.func = func;
976 	i40e->i40e_hw_space.bus.device = device;
977 
978 	if (i40e_get_switch_resources(i40e) == B_FALSE) {
979 		return (B_FALSE);
980 	}
981 
982 	/*
983 	 * To calculate the total amount of a resource we have available, we
984 	 * need to add how many our i40e_t thinks it has guaranteed, if any, and
985 	 * then we need to go through and divide the number of available on the
986 	 * device, which was snapshotted before anyone should have allocated
987 	 * anything, and use that to derive how many are available from the
988 	 * pool. Longer term, we may want to turn this into something that's
989 	 * more of a pool-like resource that everything can share (though that
990 	 * may require some more assistance from MAC).
991 	 *
992 	 * Though for transmit and receive queue pairs, we just have to ask
993 	 * firmware instead.
994 	 */
995 	idp = i40e_device_find(i40e, parent, bus, device);
996 	i40e->i40e_device = idp;
997 	i40e->i40e_resources.ifr_nvsis = 0;
998 	i40e->i40e_resources.ifr_nvsis_used = 0;
999 	i40e->i40e_resources.ifr_nmacfilt = 0;
1000 	i40e->i40e_resources.ifr_nmacfilt_used = 0;
1001 	i40e->i40e_resources.ifr_nmcastfilt = 0;
1002 	i40e->i40e_resources.ifr_nmcastfilt_used = 0;
1003 
1004 	for (i = 0; i < i40e->i40e_switch_rsrc_actual; i++) {
1005 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
1006 
1007 		switch (srp->resource_type) {
1008 		case I40E_AQ_RESOURCE_TYPE_VSI:
1009 			i40e->i40e_resources.ifr_nvsis +=
1010 			    LE_16(srp->guaranteed);
1011 			i40e->i40e_resources.ifr_nvsis_used = LE_16(srp->used);
1012 			break;
1013 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
1014 			i40e->i40e_resources.ifr_nmacfilt +=
1015 			    LE_16(srp->guaranteed);
1016 			i40e->i40e_resources.ifr_nmacfilt_used =
1017 			    LE_16(srp->used);
1018 			break;
1019 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
1020 			i40e->i40e_resources.ifr_nmcastfilt +=
1021 			    LE_16(srp->guaranteed);
1022 			i40e->i40e_resources.ifr_nmcastfilt_used =
1023 			    LE_16(srp->used);
1024 			break;
1025 		default:
1026 			break;
1027 		}
1028 	}
1029 
1030 	for (i = 0; i < idp->id_rsrcs_act; i++) {
1031 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
1032 		switch (srp->resource_type) {
1033 		case I40E_AQ_RESOURCE_TYPE_VSI:
1034 			i40e->i40e_resources.ifr_nvsis +=
1035 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1036 			break;
1037 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
1038 			i40e->i40e_resources.ifr_nmacfilt +=
1039 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1040 			break;
1041 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
1042 			i40e->i40e_resources.ifr_nmcastfilt +=
1043 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1044 		default:
1045 			break;
1046 		}
1047 	}
1048 
1049 	i40e->i40e_resources.ifr_nrx_queue = hw->func_caps.num_rx_qp;
1050 	i40e->i40e_resources.ifr_ntx_queue = hw->func_caps.num_tx_qp;
1051 
1052 	i40e->i40e_uaddrs = kmem_zalloc(sizeof (i40e_uaddr_t) *
1053 	    i40e->i40e_resources.ifr_nmacfilt, KM_SLEEP);
1054 	i40e->i40e_maddrs = kmem_zalloc(sizeof (i40e_maddr_t) *
1055 	    i40e->i40e_resources.ifr_nmcastfilt, KM_SLEEP);
1056 
1057 	/*
1058 	 * Initialize these as multicast addresses to indicate it's invalid for
1059 	 * sanity purposes. Think of it like 0xdeadbeef.
1060 	 */
1061 	for (i = 0; i < i40e->i40e_resources.ifr_nmacfilt; i++)
1062 		i40e->i40e_uaddrs[i].iua_mac[0] = 0x01;
1063 
1064 	return (B_TRUE);
1065 }
1066 
1067 static boolean_t
1068 i40e_enable_interrupts(i40e_t *i40e)
1069 {
1070 	int i, rc;
1071 
1072 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1073 		rc = ddi_intr_block_enable(i40e->i40e_intr_handles,
1074 		    i40e->i40e_intr_count);
1075 		if (rc != DDI_SUCCESS) {
1076 			i40e_error(i40e, "Interrupt block-enable failed: %d",
1077 			    rc);
1078 			return (B_FALSE);
1079 		}
1080 	} else {
1081 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1082 			rc = ddi_intr_enable(i40e->i40e_intr_handles[i]);
1083 			if (rc != DDI_SUCCESS) {
1084 				i40e_error(i40e,
1085 				    "Failed to enable interrupt %d: %d", i, rc);
1086 				while (--i >= 0) {
1087 					(void) ddi_intr_disable(
1088 					    i40e->i40e_intr_handles[i]);
1089 				}
1090 				return (B_FALSE);
1091 			}
1092 		}
1093 	}
1094 
1095 	return (B_TRUE);
1096 }
1097 
1098 static boolean_t
1099 i40e_disable_interrupts(i40e_t *i40e)
1100 {
1101 	int i, rc;
1102 
1103 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1104 		rc = ddi_intr_block_disable(i40e->i40e_intr_handles,
1105 		    i40e->i40e_intr_count);
1106 		if (rc != DDI_SUCCESS) {
1107 			i40e_error(i40e,
1108 			    "Interrupt block-disabled failed: %d", rc);
1109 			return (B_FALSE);
1110 		}
1111 	} else {
1112 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1113 			rc = ddi_intr_disable(i40e->i40e_intr_handles[i]);
1114 			if (rc != DDI_SUCCESS) {
1115 				i40e_error(i40e,
1116 				    "Failed to disable interrupt %d: %d",
1117 				    i, rc);
1118 				return (B_FALSE);
1119 			}
1120 		}
1121 	}
1122 
1123 	return (B_TRUE);
1124 }
1125 
1126 /*
1127  * Free receive & transmit rings.
1128  */
1129 static void
1130 i40e_free_trqpairs(i40e_t *i40e)
1131 {
1132 	i40e_trqpair_t *itrq;
1133 
1134 	if (i40e->i40e_rx_groups != NULL) {
1135 		kmem_free(i40e->i40e_rx_groups,
1136 		    sizeof (i40e_rx_group_t) * i40e->i40e_num_rx_groups);
1137 		i40e->i40e_rx_groups = NULL;
1138 	}
1139 
1140 	if (i40e->i40e_trqpairs != NULL) {
1141 		for (uint_t i = 0; i < i40e->i40e_num_trqpairs; i++) {
1142 			itrq = &i40e->i40e_trqpairs[i];
1143 			mutex_destroy(&itrq->itrq_intr_lock);
1144 			mutex_destroy(&itrq->itrq_rx_lock);
1145 			mutex_destroy(&itrq->itrq_tx_lock);
1146 			mutex_destroy(&itrq->itrq_tcb_lock);
1147 			cv_destroy(&itrq->itrq_intr_cv);
1148 			cv_destroy(&itrq->itrq_tx_cv);
1149 
1150 			i40e_stats_trqpair_fini(itrq);
1151 		}
1152 
1153 		kmem_free(i40e->i40e_trqpairs,
1154 		    sizeof (i40e_trqpair_t) * i40e->i40e_num_trqpairs);
1155 		i40e->i40e_trqpairs = NULL;
1156 	}
1157 
1158 	cv_destroy(&i40e->i40e_rx_pending_cv);
1159 	mutex_destroy(&i40e->i40e_rx_pending_lock);
1160 	mutex_destroy(&i40e->i40e_general_lock);
1161 }
1162 
1163 /*
1164  * Allocate transmit and receive rings, as well as other data structures that we
1165  * need.
1166  */
1167 static boolean_t
1168 i40e_alloc_trqpairs(i40e_t *i40e)
1169 {
1170 	void *mutexpri = DDI_INTR_PRI(i40e->i40e_intr_pri);
1171 
1172 	/*
1173 	 * Now that we have the priority for the interrupts, initialize
1174 	 * all relevant locks.
1175 	 */
1176 	mutex_init(&i40e->i40e_general_lock, NULL, MUTEX_DRIVER, mutexpri);
1177 	mutex_init(&i40e->i40e_rx_pending_lock, NULL, MUTEX_DRIVER, mutexpri);
1178 	cv_init(&i40e->i40e_rx_pending_cv, NULL, CV_DRIVER, NULL);
1179 
1180 	i40e->i40e_trqpairs = kmem_zalloc(sizeof (i40e_trqpair_t) *
1181 	    i40e->i40e_num_trqpairs, KM_SLEEP);
1182 	for (uint_t i = 0; i < i40e->i40e_num_trqpairs; i++) {
1183 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
1184 
1185 		itrq->itrq_i40e = i40e;
1186 		mutex_init(&itrq->itrq_intr_lock, NULL, MUTEX_DRIVER, mutexpri);
1187 		mutex_init(&itrq->itrq_rx_lock, NULL, MUTEX_DRIVER, mutexpri);
1188 		mutex_init(&itrq->itrq_tx_lock, NULL, MUTEX_DRIVER, mutexpri);
1189 		mutex_init(&itrq->itrq_tcb_lock, NULL, MUTEX_DRIVER, mutexpri);
1190 		cv_init(&itrq->itrq_intr_cv, NULL, CV_DRIVER, NULL);
1191 		cv_init(&itrq->itrq_tx_cv, NULL, CV_DRIVER, NULL);
1192 		itrq->itrq_index = i;
1193 		itrq->itrq_intr_quiesce = B_TRUE;
1194 		itrq->itrq_tx_quiesce = B_TRUE;
1195 	}
1196 
1197 	for (uint_t i = 0; i < i40e->i40e_num_trqpairs; i++) {
1198 		/*
1199 		 * Keeping this in a separate iteration makes the
1200 		 * clean up path safe.
1201 		 */
1202 		if (!i40e_stats_trqpair_init(&i40e->i40e_trqpairs[i])) {
1203 			i40e_free_trqpairs(i40e);
1204 			return (B_FALSE);
1205 		}
1206 	}
1207 
1208 	i40e->i40e_rx_groups = kmem_zalloc(sizeof (i40e_rx_group_t) *
1209 	    i40e->i40e_num_rx_groups, KM_SLEEP);
1210 
1211 	for (uint_t i = 0; i < i40e->i40e_num_rx_groups; i++) {
1212 		i40e_rx_group_t *rxg = &i40e->i40e_rx_groups[i];
1213 
1214 		rxg->irg_index = i;
1215 		rxg->irg_i40e = i40e;
1216 	}
1217 
1218 	return (B_TRUE);
1219 }
1220 
1221 
1222 
1223 /*
1224  * Unless a .conf file already overrode i40e_t structure values, they will
1225  * be 0, and need to be set in conjunction with the now-available HW report.
1226  */
1227 /* ARGSUSED */
1228 static void
1229 i40e_hw_to_instance(i40e_t *i40e, i40e_hw_t *hw)
1230 {
1231 	if (i40e->i40e_num_trqpairs_per_vsi == 0) {
1232 		if (i40e_is_x722(i40e)) {
1233 			i40e->i40e_num_trqpairs_per_vsi =
1234 			    I40E_722_MAX_TC_QUEUES;
1235 		} else {
1236 			i40e->i40e_num_trqpairs_per_vsi =
1237 			    I40E_710_MAX_TC_QUEUES;
1238 		}
1239 	}
1240 
1241 	if (i40e->i40e_num_rx_groups == 0) {
1242 		i40e->i40e_num_rx_groups = I40E_DEF_NUM_RX_GROUPS;
1243 	}
1244 }
1245 
1246 /*
1247  * Free any resources required by, or setup by, the Intel common code.
1248  */
1249 static void
1250 i40e_common_code_fini(i40e_t *i40e)
1251 {
1252 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1253 	int rc;
1254 
1255 	rc = i40e_shutdown_lan_hmc(hw);
1256 	if (rc != I40E_SUCCESS)
1257 		i40e_error(i40e, "failed to shutdown LAN hmc: %d", rc);
1258 
1259 	rc = i40e_shutdown_adminq(hw);
1260 	if (rc != I40E_SUCCESS)
1261 		i40e_error(i40e, "failed to shutdown admin queue: %d", rc);
1262 }
1263 
1264 /*
1265  * Initialize and call Intel common-code routines, includes some setup
1266  * the common code expects from the driver.  Also prints on failure, so
1267  * the caller doesn't have to.
1268  */
1269 static boolean_t
1270 i40e_common_code_init(i40e_t *i40e, i40e_hw_t *hw)
1271 {
1272 	int rc;
1273 
1274 	i40e_clear_hw(hw);
1275 	rc = i40e_pf_reset(hw);
1276 	if (rc != 0) {
1277 		i40e_error(i40e, "failed to reset hardware: %d", rc);
1278 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_NO_RESPONSE);
1279 		return (B_FALSE);
1280 	}
1281 
1282 	rc = i40e_init_shared_code(hw);
1283 	if (rc != 0) {
1284 		i40e_error(i40e, "failed to initialize i40e core: %d", rc);
1285 		return (B_FALSE);
1286 	}
1287 
1288 	hw->aq.num_arq_entries = I40E_DEF_ADMINQ_SIZE;
1289 	hw->aq.num_asq_entries =  I40E_DEF_ADMINQ_SIZE;
1290 	hw->aq.arq_buf_size = I40E_ADMINQ_BUFSZ;
1291 	hw->aq.asq_buf_size = I40E_ADMINQ_BUFSZ;
1292 
1293 	rc = i40e_init_adminq(hw);
1294 	if (rc != 0) {
1295 		i40e_error(i40e, "failed to initialize firmware admin queue: "
1296 		    "%d, potential firmware version mismatch", rc);
1297 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
1298 		return (B_FALSE);
1299 	}
1300 
1301 	if (hw->aq.api_maj_ver == I40E_FW_API_VERSION_MAJOR &&
1302 	    hw->aq.api_min_ver > I40E_FW_MINOR_VERSION(hw)) {
1303 		i40e_log(i40e, "The driver for the device detected a newer "
1304 		    "version of the NVM image (%d.%d) than expected (%d.%d).\n"
1305 		    "Please install the most recent version of the network "
1306 		    "driver.\n", hw->aq.api_maj_ver, hw->aq.api_min_ver,
1307 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_MINOR_VERSION(hw));
1308 	} else if (hw->aq.api_maj_ver < I40E_FW_API_VERSION_MAJOR ||
1309 	    hw->aq.api_min_ver < (I40E_FW_MINOR_VERSION(hw) - 1)) {
1310 		i40e_log(i40e, "The driver for the device detected an older"
1311 		    " version of the NVM image (%d.%d) than expected (%d.%d)."
1312 		    "\nPlease update the NVM image.\n",
1313 		    hw->aq.api_maj_ver, hw->aq.api_min_ver,
1314 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_MINOR_VERSION(hw) - 1);
1315 	}
1316 
1317 	i40e_clear_pxe_mode(hw);
1318 
1319 	/*
1320 	 * We need to call this so that the common code can discover
1321 	 * capabilities of the hardware, which it uses throughout the rest.
1322 	 */
1323 	if (!i40e_get_hw_capabilities(i40e, hw)) {
1324 		i40e_error(i40e, "failed to obtain hardware capabilities");
1325 		return (B_FALSE);
1326 	}
1327 
1328 	if (i40e_get_available_resources(i40e) == B_FALSE) {
1329 		i40e_error(i40e, "failed to obtain hardware resources");
1330 		return (B_FALSE);
1331 	}
1332 
1333 	i40e_hw_to_instance(i40e, hw);
1334 
1335 	rc = i40e_init_lan_hmc(hw, hw->func_caps.num_tx_qp,
1336 	    hw->func_caps.num_rx_qp, 0, 0);
1337 	if (rc != 0) {
1338 		i40e_error(i40e, "failed to initialize hardware memory cache: "
1339 		    "%d", rc);
1340 		return (B_FALSE);
1341 	}
1342 
1343 	rc = i40e_configure_lan_hmc(hw, I40E_HMC_MODEL_DIRECT_ONLY);
1344 	if (rc != 0) {
1345 		i40e_error(i40e, "failed to configure hardware memory cache: "
1346 		    "%d", rc);
1347 		return (B_FALSE);
1348 	}
1349 
1350 	(void) i40e_aq_stop_lldp(hw, TRUE, NULL);
1351 
1352 	rc = i40e_get_mac_addr(hw, hw->mac.addr);
1353 	if (rc != I40E_SUCCESS) {
1354 		i40e_error(i40e, "failed to retrieve hardware mac address: %d",
1355 		    rc);
1356 		return (B_FALSE);
1357 	}
1358 
1359 	rc = i40e_validate_mac_addr(hw->mac.addr);
1360 	if (rc != 0) {
1361 		i40e_error(i40e, "failed to validate internal mac address: "
1362 		    "%d", rc);
1363 		return (B_FALSE);
1364 	}
1365 	bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
1366 	if ((rc = i40e_get_port_mac_addr(hw, hw->mac.port_addr)) !=
1367 	    I40E_SUCCESS) {
1368 		i40e_error(i40e, "failed to retrieve port mac address: %d",
1369 		    rc);
1370 		return (B_FALSE);
1371 	}
1372 
1373 	/*
1374 	 * We need to obtain the Default Virtual Station SEID (VSI)
1375 	 * before we can perform other operations on the device.
1376 	 */
1377 	if (!i40e_set_def_vsi_seid(i40e)) {
1378 		i40e_error(i40e, "failed to obtain Default VSI SEID");
1379 		return (B_FALSE);
1380 	}
1381 
1382 	return (B_TRUE);
1383 }
1384 
1385 static void
1386 i40e_unconfigure(dev_info_t *devinfo, i40e_t *i40e)
1387 {
1388 	int rc;
1389 
1390 	if (i40e->i40e_attach_progress & I40E_ATTACH_ENABLE_INTR)
1391 		(void) i40e_disable_interrupts(i40e);
1392 
1393 	if ((i40e->i40e_attach_progress & I40E_ATTACH_LINK_TIMER) &&
1394 	    i40e->i40e_periodic_id != 0) {
1395 		ddi_periodic_delete(i40e->i40e_periodic_id);
1396 		i40e->i40e_periodic_id = 0;
1397 	}
1398 
1399 	if (i40e->i40e_attach_progress & I40E_ATTACH_UFM_INIT)
1400 		ddi_ufm_fini(i40e->i40e_ufmh);
1401 
1402 	if (i40e->i40e_attach_progress & I40E_ATTACH_MAC) {
1403 		rc = mac_unregister(i40e->i40e_mac_hdl);
1404 		if (rc != 0) {
1405 			i40e_error(i40e, "failed to unregister from mac: %d",
1406 			    rc);
1407 		}
1408 	}
1409 
1410 	if (i40e->i40e_attach_progress & I40E_ATTACH_STATS) {
1411 		i40e_stats_fini(i40e);
1412 	}
1413 
1414 	if (i40e->i40e_attach_progress & I40E_ATTACH_ADD_INTR)
1415 		i40e_rem_intr_handlers(i40e);
1416 
1417 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_RINGSLOCKS)
1418 		i40e_free_trqpairs(i40e);
1419 
1420 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_INTR)
1421 		i40e_rem_intrs(i40e);
1422 
1423 	if (i40e->i40e_attach_progress & I40E_ATTACH_COMMON_CODE)
1424 		i40e_common_code_fini(i40e);
1425 
1426 	i40e_cleanup_resources(i40e);
1427 
1428 	if (i40e->i40e_attach_progress & I40E_ATTACH_PROPS)
1429 		(void) ddi_prop_remove_all(devinfo);
1430 
1431 	if (i40e->i40e_attach_progress & I40E_ATTACH_REGS_MAP &&
1432 	    i40e->i40e_osdep_space.ios_reg_handle != NULL) {
1433 		ddi_regs_map_free(&i40e->i40e_osdep_space.ios_reg_handle);
1434 		i40e->i40e_osdep_space.ios_reg_handle = NULL;
1435 	}
1436 
1437 	if ((i40e->i40e_attach_progress & I40E_ATTACH_PCI_CONFIG) &&
1438 	    i40e->i40e_osdep_space.ios_cfg_handle != NULL) {
1439 		pci_config_teardown(&i40e->i40e_osdep_space.ios_cfg_handle);
1440 		i40e->i40e_osdep_space.ios_cfg_handle = NULL;
1441 	}
1442 
1443 	if (i40e->i40e_attach_progress & I40E_ATTACH_FM_INIT)
1444 		i40e_fm_fini(i40e);
1445 
1446 	kmem_free(i40e->i40e_aqbuf, I40E_ADMINQ_BUFSZ);
1447 	kmem_free(i40e, sizeof (i40e_t));
1448 
1449 	ddi_set_driver_private(devinfo, NULL);
1450 }
1451 
1452 static boolean_t
1453 i40e_final_init(i40e_t *i40e)
1454 {
1455 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1456 	struct i40e_osdep *osdep = OS_DEP(hw);
1457 	uint8_t pbanum[I40E_PBANUM_STRLEN];
1458 	enum i40e_status_code irc;
1459 	char buf[I40E_DDI_PROP_LEN];
1460 
1461 	pbanum[0] = '\0';
1462 	irc = i40e_read_pba_string(hw, pbanum, sizeof (pbanum));
1463 	if (irc != I40E_SUCCESS) {
1464 		i40e_log(i40e, "failed to read PBA string: %d", irc);
1465 	} else {
1466 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1467 		    "printed-board-assembly", (char *)pbanum);
1468 	}
1469 
1470 #ifdef	DEBUG
1471 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.fw_maj_ver,
1472 	    hw->aq.fw_min_ver) < sizeof (buf));
1473 	ASSERT(snprintf(NULL, 0, "%x", hw->aq.fw_build) < sizeof (buf));
1474 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.api_maj_ver,
1475 	    hw->aq.api_min_ver) < sizeof (buf));
1476 #endif
1477 
1478 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.fw_maj_ver,
1479 	    hw->aq.fw_min_ver);
1480 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1481 	    "firmware-version", buf);
1482 	(void) snprintf(buf, sizeof (buf), "%x", hw->aq.fw_build);
1483 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1484 	    "firmware-build", buf);
1485 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.api_maj_ver,
1486 	    hw->aq.api_min_ver);
1487 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1488 	    "api-version", buf);
1489 
1490 	if (!i40e_set_hw_bus_info(hw))
1491 		return (B_FALSE);
1492 
1493 	if (i40e_check_acc_handle(osdep->ios_reg_handle) != DDI_FM_OK) {
1494 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
1495 		return (B_FALSE);
1496 	}
1497 
1498 	return (B_TRUE);
1499 }
1500 
1501 static void
1502 i40e_identify_hardware(i40e_t *i40e)
1503 {
1504 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1505 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1506 
1507 	hw->vendor_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_VENID);
1508 	hw->device_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_DEVID);
1509 	hw->revision_id = pci_config_get8(osdep->ios_cfg_handle,
1510 	    PCI_CONF_REVID);
1511 	hw->subsystem_device_id =
1512 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBSYSID);
1513 	hw->subsystem_vendor_id =
1514 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBVENID);
1515 
1516 	/*
1517 	 * Note that we set the hardware's bus information later on, in
1518 	 * i40e_get_available_resources(). The common code doesn't seem to
1519 	 * require that it be set in any ways, it seems to be mostly for
1520 	 * book-keeping.
1521 	 */
1522 }
1523 
1524 static boolean_t
1525 i40e_regs_map(i40e_t *i40e)
1526 {
1527 	dev_info_t *devinfo = i40e->i40e_dip;
1528 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1529 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1530 	off_t memsize;
1531 	int ret;
1532 
1533 	if (ddi_dev_regsize(devinfo, I40E_ADAPTER_REGSET, &memsize) !=
1534 	    DDI_SUCCESS) {
1535 		i40e_error(i40e, "Used invalid register set to map PCIe regs");
1536 		return (B_FALSE);
1537 	}
1538 
1539 	if ((ret = ddi_regs_map_setup(devinfo, I40E_ADAPTER_REGSET,
1540 	    (caddr_t *)&hw->hw_addr, 0, memsize, &i40e_regs_acc_attr,
1541 	    &osdep->ios_reg_handle)) != DDI_SUCCESS) {
1542 		i40e_error(i40e, "failed to map device registers: %d", ret);
1543 		return (B_FALSE);
1544 	}
1545 
1546 	osdep->ios_reg_size = memsize;
1547 	return (B_TRUE);
1548 }
1549 
1550 /*
1551  * Update parameters required when a new MTU has been configured.  Calculate the
1552  * maximum frame size, as well as, size our DMA buffers which we size in
1553  * increments of 1K.
1554  */
1555 void
1556 i40e_update_mtu(i40e_t *i40e)
1557 {
1558 	uint32_t rx, tx;
1559 
1560 	i40e->i40e_frame_max = i40e->i40e_sdu +
1561 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
1562 
1563 	rx = i40e->i40e_frame_max + I40E_BUF_IPHDR_ALIGNMENT;
1564 	i40e->i40e_rx_buf_size = ((rx >> 10) +
1565 	    ((rx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1566 
1567 	tx = i40e->i40e_frame_max;
1568 	i40e->i40e_tx_buf_size = ((tx >> 10) +
1569 	    ((tx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1570 }
1571 
1572 static int
1573 i40e_get_prop(i40e_t *i40e, char *prop, int min, int max, int def)
1574 {
1575 	int val;
1576 
1577 	val = ddi_prop_get_int(DDI_DEV_T_ANY, i40e->i40e_dip, DDI_PROP_DONTPASS,
1578 	    prop, def);
1579 	if (val > max)
1580 		val = max;
1581 	if (val < min)
1582 		val = min;
1583 	return (val);
1584 }
1585 
1586 static void
1587 i40e_init_properties(i40e_t *i40e)
1588 {
1589 	i40e->i40e_sdu = i40e_get_prop(i40e, "default_mtu",
1590 	    I40E_MIN_MTU, I40E_MAX_MTU, I40E_DEF_MTU);
1591 
1592 	i40e->i40e_intr_force = i40e_get_prop(i40e, "intr_force",
1593 	    I40E_INTR_NONE, I40E_INTR_LEGACY, I40E_INTR_NONE);
1594 
1595 	i40e->i40e_mr_enable = i40e_get_prop(i40e, "mr_enable",
1596 	    B_FALSE, B_TRUE, B_TRUE);
1597 
1598 	i40e->i40e_tx_ring_size = i40e_get_prop(i40e, "tx_ring_size",
1599 	    I40E_MIN_TX_RING_SIZE, I40E_MAX_TX_RING_SIZE,
1600 	    I40E_DEF_TX_RING_SIZE);
1601 	if ((i40e->i40e_tx_ring_size % I40E_DESC_ALIGN) != 0) {
1602 		i40e->i40e_tx_ring_size = P2ROUNDUP(i40e->i40e_tx_ring_size,
1603 		    I40E_DESC_ALIGN);
1604 	}
1605 
1606 	i40e->i40e_tx_block_thresh = i40e_get_prop(i40e, "tx_resched_threshold",
1607 	    I40E_MIN_TX_BLOCK_THRESH,
1608 	    i40e->i40e_tx_ring_size - I40E_TX_MAX_COOKIE,
1609 	    I40E_DEF_TX_BLOCK_THRESH);
1610 
1611 	i40e->i40e_num_rx_groups = i40e_get_prop(i40e, "rx_num_groups",
1612 	    I40E_MIN_NUM_RX_GROUPS, I40E_MAX_NUM_RX_GROUPS,
1613 	    I40E_DEF_NUM_RX_GROUPS);
1614 
1615 	i40e->i40e_rx_ring_size = i40e_get_prop(i40e, "rx_ring_size",
1616 	    I40E_MIN_RX_RING_SIZE, I40E_MAX_RX_RING_SIZE,
1617 	    I40E_DEF_RX_RING_SIZE);
1618 	if ((i40e->i40e_rx_ring_size % I40E_DESC_ALIGN) != 0) {
1619 		i40e->i40e_rx_ring_size = P2ROUNDUP(i40e->i40e_rx_ring_size,
1620 		    I40E_DESC_ALIGN);
1621 	}
1622 
1623 	i40e->i40e_rx_limit_per_intr = i40e_get_prop(i40e, "rx_limit_per_intr",
1624 	    I40E_MIN_RX_LIMIT_PER_INTR,	I40E_MAX_RX_LIMIT_PER_INTR,
1625 	    I40E_DEF_RX_LIMIT_PER_INTR);
1626 
1627 	i40e->i40e_tx_hcksum_enable = i40e_get_prop(i40e, "tx_hcksum_enable",
1628 	    B_FALSE, B_TRUE, B_TRUE);
1629 
1630 	i40e->i40e_tx_lso_enable = i40e_get_prop(i40e, "tx_lso_enable",
1631 	    B_FALSE, B_TRUE, B_TRUE);
1632 
1633 	i40e->i40e_rx_hcksum_enable = i40e_get_prop(i40e, "rx_hcksum_enable",
1634 	    B_FALSE, B_TRUE, B_TRUE);
1635 
1636 	i40e->i40e_rx_dma_min = i40e_get_prop(i40e, "rx_dma_threshold",
1637 	    I40E_MIN_RX_DMA_THRESH, I40E_MAX_RX_DMA_THRESH,
1638 	    I40E_DEF_RX_DMA_THRESH);
1639 
1640 	i40e->i40e_tx_dma_min = i40e_get_prop(i40e, "tx_dma_threshold",
1641 	    I40E_MIN_TX_DMA_THRESH, I40E_MAX_TX_DMA_THRESH,
1642 	    I40E_DEF_TX_DMA_THRESH);
1643 
1644 	i40e->i40e_tx_itr = i40e_get_prop(i40e, "tx_intr_throttle",
1645 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_TX_ITR);
1646 
1647 	i40e->i40e_rx_itr = i40e_get_prop(i40e, "rx_intr_throttle",
1648 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_RX_ITR);
1649 
1650 	i40e->i40e_other_itr = i40e_get_prop(i40e, "other_intr_throttle",
1651 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_OTHER_ITR);
1652 
1653 	if (!i40e->i40e_mr_enable) {
1654 		i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1655 		i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1656 	}
1657 
1658 	i40e_update_mtu(i40e);
1659 }
1660 
1661 /*
1662  * There are a few constraints on interrupts that we're currently imposing, some
1663  * of which are restrictions from hardware. For a fuller treatment, see
1664  * i40e_intr.c.
1665  *
1666  * Currently, to use MSI-X we require two interrupts be available though in
1667  * theory we should participate in IRM and happily use more interrupts.
1668  *
1669  * Hardware only supports a single MSI being programmed and therefore if we
1670  * don't have MSI-X interrupts available at this time, then we ratchet down the
1671  * number of rings and groups available. Obviously, we only bother with a single
1672  * fixed interrupt.
1673  */
1674 static boolean_t
1675 i40e_alloc_intr_handles(i40e_t *i40e, dev_info_t *devinfo, int intr_type)
1676 {
1677 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1678 	ddi_acc_handle_t rh = i40e->i40e_osdep_space.ios_reg_handle;
1679 	int request, count, actual, rc, min;
1680 	uint32_t reg;
1681 
1682 	switch (intr_type) {
1683 	case DDI_INTR_TYPE_FIXED:
1684 	case DDI_INTR_TYPE_MSI:
1685 		request = 1;
1686 		min = 1;
1687 		break;
1688 	case DDI_INTR_TYPE_MSIX:
1689 		min = 2;
1690 		if (!i40e->i40e_mr_enable) {
1691 			request = 2;
1692 			break;
1693 		}
1694 		reg = I40E_READ_REG(hw, I40E_GLPCI_CNF2);
1695 		/*
1696 		 * Should this read fail, we will drop back to using
1697 		 * MSI or fixed interrupts.
1698 		 */
1699 		if (i40e_check_acc_handle(rh) != DDI_FM_OK) {
1700 			ddi_fm_service_impact(i40e->i40e_dip,
1701 			    DDI_SERVICE_DEGRADED);
1702 			return (B_FALSE);
1703 		}
1704 		request = (reg & I40E_GLPCI_CNF2_MSI_X_PF_N_MASK) >>
1705 		    I40E_GLPCI_CNF2_MSI_X_PF_N_SHIFT;
1706 		request++;	/* the register value is n - 1 */
1707 		break;
1708 	default:
1709 		panic("bad interrupt type passed to i40e_alloc_intr_handles: "
1710 		    "%d", intr_type);
1711 	}
1712 
1713 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
1714 	if (rc != DDI_SUCCESS || count < min) {
1715 		i40e_log(i40e, "Get interrupt number failed, "
1716 		    "returned %d, count %d", rc, count);
1717 		return (B_FALSE);
1718 	}
1719 
1720 	rc = ddi_intr_get_navail(devinfo, intr_type, &count);
1721 	if (rc != DDI_SUCCESS || count < min) {
1722 		i40e_log(i40e, "Get AVAILABLE interrupt number failed, "
1723 		    "returned %d, count %d", rc, count);
1724 		return (B_FALSE);
1725 	}
1726 
1727 	actual = 0;
1728 	i40e->i40e_intr_count = 0;
1729 	i40e->i40e_intr_count_max = 0;
1730 	i40e->i40e_intr_count_min = 0;
1731 
1732 	i40e->i40e_intr_size = request * sizeof (ddi_intr_handle_t);
1733 	ASSERT(i40e->i40e_intr_size != 0);
1734 	i40e->i40e_intr_handles = kmem_alloc(i40e->i40e_intr_size, KM_SLEEP);
1735 
1736 	rc = ddi_intr_alloc(devinfo, i40e->i40e_intr_handles, intr_type, 0,
1737 	    min(request, count), &actual, DDI_INTR_ALLOC_NORMAL);
1738 	if (rc != DDI_SUCCESS) {
1739 		i40e_log(i40e, "Interrupt allocation failed with %d.", rc);
1740 		goto alloc_handle_fail;
1741 	}
1742 
1743 	i40e->i40e_intr_count = actual;
1744 	i40e->i40e_intr_count_max = request;
1745 	i40e->i40e_intr_count_min = min;
1746 
1747 	if (actual < min) {
1748 		i40e_log(i40e, "actual (%d) is less than minimum (%d).",
1749 		    actual, min);
1750 		goto alloc_handle_fail;
1751 	}
1752 
1753 	/*
1754 	 * Record the priority and capabilities for our first vector.  Once
1755 	 * we have it, that's our priority until detach time.  Even if we
1756 	 * eventually participate in IRM, our priority shouldn't change.
1757 	 */
1758 	rc = ddi_intr_get_pri(i40e->i40e_intr_handles[0], &i40e->i40e_intr_pri);
1759 	if (rc != DDI_SUCCESS) {
1760 		i40e_log(i40e,
1761 		    "Getting interrupt priority failed with %d.", rc);
1762 		goto alloc_handle_fail;
1763 	}
1764 
1765 	rc = ddi_intr_get_cap(i40e->i40e_intr_handles[0], &i40e->i40e_intr_cap);
1766 	if (rc != DDI_SUCCESS) {
1767 		i40e_log(i40e,
1768 		    "Getting interrupt capabilities failed with %d.", rc);
1769 		goto alloc_handle_fail;
1770 	}
1771 
1772 	i40e->i40e_intr_type = intr_type;
1773 	return (B_TRUE);
1774 
1775 alloc_handle_fail:
1776 
1777 	i40e_rem_intrs(i40e);
1778 	return (B_FALSE);
1779 }
1780 
1781 static boolean_t
1782 i40e_alloc_intrs(i40e_t *i40e, dev_info_t *devinfo)
1783 {
1784 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1785 	int intr_types, rc;
1786 	uint_t max_trqpairs;
1787 
1788 	if (i40e_is_x722(i40e)) {
1789 		max_trqpairs = I40E_722_MAX_TC_QUEUES;
1790 	} else {
1791 		max_trqpairs = I40E_710_MAX_TC_QUEUES;
1792 	}
1793 
1794 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
1795 	if (rc != DDI_SUCCESS) {
1796 		i40e_error(i40e, "failed to get supported interrupt types: %d",
1797 		    rc);
1798 		return (B_FALSE);
1799 	}
1800 
1801 	i40e->i40e_intr_type = 0;
1802 
1803 	/*
1804 	 * We need to determine the number of queue pairs per traffic
1805 	 * class. We only have one traffic class (TC0), so we'll base
1806 	 * this off the number of interrupts provided. Furthermore,
1807 	 * since we only use one traffic class, the number of queues
1808 	 * per traffic class and per VSI are the same.
1809 	 */
1810 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
1811 	    (i40e->i40e_intr_force <= I40E_INTR_MSIX) &&
1812 	    (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_MSIX))) {
1813 		uint32_t n, qp_cap, num_trqpairs;
1814 
1815 		/*
1816 		 * While we want the number of queue pairs to match
1817 		 * the number of interrupts, we must keep stay in
1818 		 * bounds of the maximum number of queues per traffic
1819 		 * class. We subtract one from i40e_intr_count to
1820 		 * account for interrupt zero; which is currently
1821 		 * restricted to admin queue commands and other
1822 		 * interrupt causes.
1823 		 */
1824 		n = MIN(i40e->i40e_intr_count - 1, max_trqpairs);
1825 		ASSERT3U(n, >, 0);
1826 
1827 		/*
1828 		 * Round up to the nearest power of two to ensure that
1829 		 * the QBASE aligns with the TC size which must be
1830 		 * programmed as a power of two. See the queue mapping
1831 		 * description in section 7.4.9.5.5.1.
1832 		 *
1833 		 * If i40e_intr_count - 1 is not a power of two then
1834 		 * some queue pairs on the same VSI will have to share
1835 		 * an interrupt.
1836 		 *
1837 		 * We may want to revisit this logic in a future where
1838 		 * we have more interrupts and more VSIs. Otherwise,
1839 		 * each VSI will use as many interrupts as possible.
1840 		 * Using more QPs per VSI means better RSS for each
1841 		 * group, but at the same time may require more
1842 		 * sharing of interrupts across VSIs. This may be a
1843 		 * good candidate for a .conf tunable.
1844 		 */
1845 		n = 0x1 << ddi_fls(n);
1846 		i40e->i40e_num_trqpairs_per_vsi = n;
1847 
1848 		/*
1849 		 * Make sure the number of tx/rx qpairs does not exceed
1850 		 * the device's capabilities.
1851 		 */
1852 		ASSERT3U(i40e->i40e_num_rx_groups, >, 0);
1853 		qp_cap = MIN(hw->func_caps.num_rx_qp, hw->func_caps.num_tx_qp);
1854 		num_trqpairs = i40e->i40e_num_trqpairs_per_vsi *
1855 		    i40e->i40e_num_rx_groups;
1856 		if (num_trqpairs > qp_cap) {
1857 			i40e->i40e_num_rx_groups = MAX(1, qp_cap /
1858 			    i40e->i40e_num_trqpairs_per_vsi);
1859 			num_trqpairs = i40e->i40e_num_trqpairs_per_vsi *
1860 			    i40e->i40e_num_rx_groups;
1861 			i40e_log(i40e, "Rx groups restricted to %u",
1862 			    i40e->i40e_num_rx_groups);
1863 		}
1864 		ASSERT3U(num_trqpairs, >, 0);
1865 		i40e->i40e_num_trqpairs = num_trqpairs;
1866 		return (B_TRUE);
1867 	}
1868 
1869 	/*
1870 	 * We only use multiple transmit/receive pairs when MSI-X interrupts are
1871 	 * available due to the fact that the device basically only supports a
1872 	 * single MSI interrupt.
1873 	 */
1874 	i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1875 	i40e->i40e_num_trqpairs_per_vsi = i40e->i40e_num_trqpairs;
1876 	i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1877 
1878 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
1879 	    (i40e->i40e_intr_force <= I40E_INTR_MSI)) {
1880 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_MSI))
1881 			return (B_TRUE);
1882 	}
1883 
1884 	if (intr_types & DDI_INTR_TYPE_FIXED) {
1885 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_FIXED))
1886 			return (B_TRUE);
1887 	}
1888 
1889 	return (B_FALSE);
1890 }
1891 
1892 /*
1893  * Map different interrupts to MSI-X vectors.
1894  */
1895 static boolean_t
1896 i40e_map_intrs_to_vectors(i40e_t *i40e)
1897 {
1898 	if (i40e->i40e_intr_type != DDI_INTR_TYPE_MSIX) {
1899 		return (B_TRUE);
1900 	}
1901 
1902 	/*
1903 	 * Each queue pair is mapped to a single interrupt, so
1904 	 * transmit and receive interrupts for a given queue share the
1905 	 * same vector. Vector zero is reserved for the admin queue.
1906 	 */
1907 	for (uint_t i = 0; i < i40e->i40e_num_trqpairs; i++) {
1908 		uint_t vector = i % (i40e->i40e_intr_count - 1);
1909 
1910 		i40e->i40e_trqpairs[i].itrq_rx_intrvec = vector + 1;
1911 		i40e->i40e_trqpairs[i].itrq_tx_intrvec = vector + 1;
1912 	}
1913 
1914 	return (B_TRUE);
1915 }
1916 
1917 static boolean_t
1918 i40e_add_intr_handlers(i40e_t *i40e)
1919 {
1920 	int rc, vector;
1921 
1922 	switch (i40e->i40e_intr_type) {
1923 	case DDI_INTR_TYPE_MSIX:
1924 		for (vector = 0; vector < i40e->i40e_intr_count; vector++) {
1925 			rc = ddi_intr_add_handler(
1926 			    i40e->i40e_intr_handles[vector],
1927 			    (ddi_intr_handler_t *)i40e_intr_msix, i40e,
1928 			    (void *)(uintptr_t)vector);
1929 			if (rc != DDI_SUCCESS) {
1930 				i40e_log(i40e, "Add interrupt handler (MSI-X) "
1931 				    "failed: return %d, vector %d", rc, vector);
1932 				for (vector--; vector >= 0; vector--) {
1933 					(void) ddi_intr_remove_handler(
1934 					    i40e->i40e_intr_handles[vector]);
1935 				}
1936 				return (B_FALSE);
1937 			}
1938 		}
1939 		break;
1940 	case DDI_INTR_TYPE_MSI:
1941 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1942 		    (ddi_intr_handler_t *)i40e_intr_msi, i40e, NULL);
1943 		if (rc != DDI_SUCCESS) {
1944 			i40e_log(i40e, "Add interrupt handler (MSI) failed: "
1945 			    "return %d", rc);
1946 			return (B_FALSE);
1947 		}
1948 		break;
1949 	case DDI_INTR_TYPE_FIXED:
1950 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1951 		    (ddi_intr_handler_t *)i40e_intr_legacy, i40e, NULL);
1952 		if (rc != DDI_SUCCESS) {
1953 			i40e_log(i40e, "Add interrupt handler (legacy) failed:"
1954 			    " return %d", rc);
1955 			return (B_FALSE);
1956 		}
1957 		break;
1958 	default:
1959 		/* Cast to pacify lint */
1960 		panic("i40e_intr_type %p contains an unknown type: %d",
1961 		    (void *)i40e, i40e->i40e_intr_type);
1962 	}
1963 
1964 	return (B_TRUE);
1965 }
1966 
1967 /*
1968  * Perform periodic checks. Longer term, we should be thinking about additional
1969  * things here:
1970  *
1971  * o Stall Detection
1972  * o Temperature sensor detection
1973  * o Device resetting
1974  * o Statistics updating to avoid wraparound
1975  */
1976 static void
1977 i40e_timer(void *arg)
1978 {
1979 	i40e_t *i40e = arg;
1980 
1981 	mutex_enter(&i40e->i40e_general_lock);
1982 	i40e_link_check(i40e);
1983 	mutex_exit(&i40e->i40e_general_lock);
1984 }
1985 
1986 /*
1987  * Get the hardware state, and scribble away anything that needs scribbling.
1988  */
1989 static void
1990 i40e_get_hw_state(i40e_t *i40e, i40e_hw_t *hw)
1991 {
1992 	int rc;
1993 
1994 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
1995 
1996 	(void) i40e_aq_get_link_info(hw, TRUE, NULL, NULL);
1997 	i40e_link_check(i40e);
1998 
1999 	/*
2000 	 * Try and determine our PHY. Note that we may have to retry to and
2001 	 * delay to detect fiber correctly.
2002 	 */
2003 	rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE, &i40e->i40e_phy,
2004 	    NULL);
2005 	if (rc == I40E_ERR_UNKNOWN_PHY) {
2006 		i40e_msec_delay(200);
2007 		rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE,
2008 		    &i40e->i40e_phy, NULL);
2009 	}
2010 
2011 	if (rc != I40E_SUCCESS) {
2012 		if (rc == I40E_ERR_UNKNOWN_PHY) {
2013 			i40e_error(i40e, "encountered unknown PHY type, "
2014 			    "not attaching.");
2015 		} else {
2016 			i40e_error(i40e, "error getting physical capabilities: "
2017 			    "%d, %d", rc, hw->aq.asq_last_status);
2018 		}
2019 	}
2020 
2021 	rc = i40e_update_link_info(hw);
2022 	if (rc != I40E_SUCCESS) {
2023 		i40e_error(i40e, "failed to update link information: %d", rc);
2024 	}
2025 
2026 	/*
2027 	 * In general, we don't want to mask off (as in stop from being a cause)
2028 	 * any of the interrupts that the phy might be able to generate.
2029 	 */
2030 	rc = i40e_aq_set_phy_int_mask(hw, 0, NULL);
2031 	if (rc != I40E_SUCCESS) {
2032 		i40e_error(i40e, "failed to update phy link mask: %d", rc);
2033 	}
2034 }
2035 
2036 /*
2037  * Go through and re-initialize any existing filters that we may have set up for
2038  * this device. Note that we would only expect them to exist if hardware had
2039  * already been initialized and we had just reset it. While we're not
2040  * implementing this yet, we're keeping this around for when we add reset
2041  * capabilities, so this isn't forgotten.
2042  */
2043 /* ARGSUSED */
2044 static void
2045 i40e_init_macaddrs(i40e_t *i40e, i40e_hw_t *hw)
2046 {
2047 }
2048 
2049 /*
2050  * Set the properties which have common values across all the VSIs.
2051  * Consult the "Add VSI" command section (7.4.9.5.5.1) for a
2052  * complete description of these properties.
2053  */
2054 static void
2055 i40e_set_shared_vsi_props(i40e_t *i40e,
2056     struct i40e_aqc_vsi_properties_data *info, uint_t vsi_idx)
2057 {
2058 	uint_t tc_queues;
2059 	uint16_t vsi_qp_base;
2060 
2061 	/*
2062 	 * It's important that we use bitwise-OR here; callers to this
2063 	 * function might enable other sections before calling this
2064 	 * function.
2065 	 */
2066 	info->valid_sections |= LE_16(I40E_AQ_VSI_PROP_QUEUE_MAP_VALID |
2067 	    I40E_AQ_VSI_PROP_VLAN_VALID);
2068 
2069 	/*
2070 	 * Calculate the starting QP index for this VSI. This base is
2071 	 * relative to the PF queue space; so a value of 0 for PF#1
2072 	 * represents the absolute index PFLAN_QALLOC_FIRSTQ for PF#1.
2073 	 */
2074 	vsi_qp_base = vsi_idx * i40e->i40e_num_trqpairs_per_vsi;
2075 	info->mapping_flags = LE_16(I40E_AQ_VSI_QUE_MAP_CONTIG);
2076 	info->queue_mapping[0] =
2077 	    LE_16((vsi_qp_base << I40E_AQ_VSI_QUEUE_SHIFT) &
2078 	    I40E_AQ_VSI_QUEUE_MASK);
2079 
2080 	/*
2081 	 * tc_queues determines the size of the traffic class, where
2082 	 * the size is 2^^tc_queues to a maximum of 64 for the X710
2083 	 * and 128 for the X722.
2084 	 *
2085 	 * Some examples:
2086 	 *	i40e_num_trqpairs_per_vsi == 1 =>  tc_queues = 0, 2^^0 = 1.
2087 	 *	i40e_num_trqpairs_per_vsi == 7 =>  tc_queues = 3, 2^^3 = 8.
2088 	 *	i40e_num_trqpairs_per_vsi == 8 =>  tc_queues = 3, 2^^3 = 8.
2089 	 *	i40e_num_trqpairs_per_vsi == 9 =>  tc_queues = 4, 2^^4 = 16.
2090 	 *	i40e_num_trqpairs_per_vsi == 17 => tc_queues = 5, 2^^5 = 32.
2091 	 *	i40e_num_trqpairs_per_vsi == 64 => tc_queues = 6, 2^^6 = 64.
2092 	 */
2093 	tc_queues = ddi_fls(i40e->i40e_num_trqpairs_per_vsi - 1);
2094 
2095 	/*
2096 	 * The TC queue mapping is in relation to the VSI queue space.
2097 	 * Since we are only using one traffic class (TC0) we always
2098 	 * start at queue offset 0.
2099 	 */
2100 	info->tc_mapping[0] =
2101 	    LE_16(((0 << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) &
2102 	    I40E_AQ_VSI_TC_QUE_OFFSET_MASK) |
2103 	    ((tc_queues << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT) &
2104 	    I40E_AQ_VSI_TC_QUE_NUMBER_MASK));
2105 
2106 	/*
2107 	 * I40E_AQ_VSI_PVLAN_MODE_ALL ("VLAN driver insertion mode")
2108 	 *
2109 	 *	Allow tagged and untagged packets to be sent to this
2110 	 *	VSI from the host.
2111 	 *
2112 	 * I40E_AQ_VSI_PVLAN_EMOD_NOTHING ("VLAN and UP expose mode")
2113 	 *
2114 	 *	Leave the tag on the frame and place no VLAN
2115 	 *	information in the descriptor. We want this mode
2116 	 *	because our MAC layer will take care of the VLAN tag,
2117 	 *	if there is one.
2118 	 */
2119 	info->port_vlan_flags = I40E_AQ_VSI_PVLAN_MODE_ALL |
2120 	    I40E_AQ_VSI_PVLAN_EMOD_NOTHING;
2121 }
2122 
2123 /*
2124  * Delete the VSI at this index, if one exists. We assume there is no
2125  * action we can take if this command fails but to log the failure.
2126  */
2127 static void
2128 i40e_delete_vsi(i40e_t *i40e, uint_t idx)
2129 {
2130 	i40e_hw_t	*hw = &i40e->i40e_hw_space;
2131 	uint16_t	seid = i40e->i40e_vsis[idx].iv_seid;
2132 
2133 	if (seid != 0) {
2134 		int rc;
2135 
2136 		rc = i40e_aq_delete_element(hw, seid, NULL);
2137 
2138 		if (rc != I40E_SUCCESS) {
2139 			i40e_error(i40e, "Failed to delete VSI %d: %d",
2140 			    rc, hw->aq.asq_last_status);
2141 		}
2142 
2143 		i40e->i40e_vsis[idx].iv_seid = 0;
2144 	}
2145 }
2146 
2147 /*
2148  * Add a new VSI.
2149  */
2150 static boolean_t
2151 i40e_add_vsi(i40e_t *i40e, i40e_hw_t *hw, uint_t idx)
2152 {
2153 	struct i40e_vsi_context	ctx;
2154 	i40e_rx_group_t		*rxg;
2155 	int			rc;
2156 
2157 	/*
2158 	 * The default VSI is created by the controller. This function
2159 	 * creates new, non-defualt VSIs only.
2160 	 */
2161 	ASSERT3U(idx, !=, 0);
2162 
2163 	bzero(&ctx, sizeof (struct i40e_vsi_context));
2164 	ctx.uplink_seid = i40e->i40e_veb_seid;
2165 	ctx.pf_num = hw->pf_id;
2166 	ctx.flags = I40E_AQ_VSI_TYPE_PF;
2167 	ctx.connection_type = I40E_AQ_VSI_CONN_TYPE_NORMAL;
2168 	i40e_set_shared_vsi_props(i40e, &ctx.info, idx);
2169 
2170 	rc = i40e_aq_add_vsi(hw, &ctx, NULL);
2171 	if (rc != I40E_SUCCESS) {
2172 		i40e_error(i40e, "i40e_aq_add_vsi() failed %d: %d", rc,
2173 		    hw->aq.asq_last_status);
2174 		return (B_FALSE);
2175 	}
2176 
2177 	rxg = &i40e->i40e_rx_groups[idx];
2178 	rxg->irg_vsi_seid = ctx.seid;
2179 	i40e->i40e_vsis[idx].iv_number = ctx.vsi_number;
2180 	i40e->i40e_vsis[idx].iv_seid = ctx.seid;
2181 	i40e->i40e_vsis[idx].iv_stats_id = LE_16(ctx.info.stat_counter_idx);
2182 
2183 	if (i40e_stat_vsi_init(i40e, idx) == B_FALSE)
2184 		return (B_FALSE);
2185 
2186 	return (B_TRUE);
2187 }
2188 
2189 /*
2190  * Configure the hardware for the Default Virtual Station Interface (VSI).
2191  */
2192 static boolean_t
2193 i40e_config_def_vsi(i40e_t *i40e, i40e_hw_t *hw)
2194 {
2195 	struct i40e_vsi_context	ctx;
2196 	i40e_rx_group_t *def_rxg;
2197 	int err;
2198 	struct i40e_aqc_remove_macvlan_element_data filt;
2199 
2200 	bzero(&ctx, sizeof (struct i40e_vsi_context));
2201 	ctx.seid = I40E_DEF_VSI_SEID(i40e);
2202 	ctx.pf_num = hw->pf_id;
2203 	err = i40e_aq_get_vsi_params(hw, &ctx, NULL);
2204 	if (err != I40E_SUCCESS) {
2205 		i40e_error(i40e, "get VSI params failed with %d", err);
2206 		return (B_FALSE);
2207 	}
2208 
2209 	ctx.info.valid_sections = 0;
2210 	i40e->i40e_vsis[0].iv_number = ctx.vsi_number;
2211 	i40e->i40e_vsis[0].iv_stats_id = LE_16(ctx.info.stat_counter_idx);
2212 	if (i40e_stat_vsi_init(i40e, 0) == B_FALSE)
2213 		return (B_FALSE);
2214 
2215 	i40e_set_shared_vsi_props(i40e, &ctx.info, I40E_DEF_VSI_IDX);
2216 
2217 	err = i40e_aq_update_vsi_params(hw, &ctx, NULL);
2218 	if (err != I40E_SUCCESS) {
2219 		i40e_error(i40e, "Update VSI params failed with %d", err);
2220 		return (B_FALSE);
2221 	}
2222 
2223 	def_rxg = &i40e->i40e_rx_groups[0];
2224 	def_rxg->irg_vsi_seid = I40E_DEF_VSI_SEID(i40e);
2225 
2226 	/*
2227 	 * We have seen three different behaviors in regards to the
2228 	 * Default VSI and its implicit L2 MAC+VLAN filter.
2229 	 *
2230 	 * 1. It has an implicit filter for the factory MAC address
2231 	 *    and this filter counts against 'ifr_nmacfilt_used'.
2232 	 *
2233 	 * 2. It has an implicit filter for the factory MAC address
2234 	 *    and this filter DOES NOT count against 'ifr_nmacfilt_used'.
2235 	 *
2236 	 * 3. It DOES NOT have an implicit filter.
2237 	 *
2238 	 * All three of these cases are accounted for below. If we
2239 	 * fail to remove the L2 filter (ENOENT) then we assume there
2240 	 * wasn't one. Otherwise, if we successfully remove the
2241 	 * filter, we make sure to update the 'ifr_nmacfilt_used'
2242 	 * count accordingly.
2243 	 *
2244 	 * We remove this filter to prevent duplicate delivery of
2245 	 * packets destined for the primary MAC address as DLS will
2246 	 * create the same filter on a non-default VSI for the primary
2247 	 * MAC client.
2248 	 *
2249 	 * If you change the following code please test it across as
2250 	 * many X700 series controllers and firmware revisions as you
2251 	 * can.
2252 	 */
2253 	bzero(&filt, sizeof (filt));
2254 	bcopy(hw->mac.port_addr, filt.mac_addr, ETHERADDRL);
2255 	filt.flags = I40E_AQC_MACVLAN_DEL_PERFECT_MATCH;
2256 	filt.vlan_tag = 0;
2257 
2258 	ASSERT3U(i40e->i40e_resources.ifr_nmacfilt_used, <=, 1);
2259 	i40e_log(i40e, "Num L2 filters: %u",
2260 	    i40e->i40e_resources.ifr_nmacfilt_used);
2261 
2262 	err = i40e_aq_remove_macvlan(hw, I40E_DEF_VSI_SEID(i40e), &filt, 1,
2263 	    NULL);
2264 	if (err == I40E_SUCCESS) {
2265 		i40e_log(i40e,
2266 		    "Removed L2 filter from Default VSI with SEID %u",
2267 		    I40E_DEF_VSI_SEID(i40e));
2268 	} else if (hw->aq.asq_last_status == ENOENT) {
2269 		i40e_log(i40e,
2270 		    "No L2 filter for Default VSI with SEID %u",
2271 		    I40E_DEF_VSI_SEID(i40e));
2272 	} else {
2273 		i40e_error(i40e, "Failed to remove L2 filter from"
2274 		    " Default VSI with SEID %u: %d (%d)",
2275 		    I40E_DEF_VSI_SEID(i40e), err, hw->aq.asq_last_status);
2276 
2277 		return (B_FALSE);
2278 	}
2279 
2280 	/*
2281 	 *  As mentioned above, the controller created an implicit L2
2282 	 *  filter for the primary MAC. We want to remove both the
2283 	 *  filter and decrement the filter count. However, not all
2284 	 *  controllers count this implicit filter against the total
2285 	 *  MAC filter count. So here we are making sure it is either
2286 	 *  one or zero. If it is one, then we know it is for the
2287 	 *  implicit filter and we should decrement since we just
2288 	 *  removed the filter above. If it is zero then we know the
2289 	 *  controller that does not count the implicit filter, and it
2290 	 *  was enough to just remove it; we leave the count alone.
2291 	 *  But if it is neither, then we have never seen a controller
2292 	 *  like this before and we should fail to attach.
2293 	 *
2294 	 *  It is unfortunate that this code must exist but the
2295 	 *  behavior of this implicit L2 filter and its corresponding
2296 	 *  count were dicovered through empirical testing. The
2297 	 *  programming manuals hint at this filter but do not
2298 	 *  explicitly call out the exact behavior.
2299 	 */
2300 	if (i40e->i40e_resources.ifr_nmacfilt_used == 1) {
2301 		i40e->i40e_resources.ifr_nmacfilt_used--;
2302 	} else {
2303 		if (i40e->i40e_resources.ifr_nmacfilt_used != 0) {
2304 			i40e_error(i40e, "Unexpected L2 filter count: %u"
2305 			    " (expected 0)",
2306 			    i40e->i40e_resources.ifr_nmacfilt_used);
2307 			return (B_FALSE);
2308 		}
2309 	}
2310 
2311 	return (B_TRUE);
2312 }
2313 
2314 static boolean_t
2315 i40e_config_rss_key_x722(i40e_t *i40e, i40e_hw_t *hw)
2316 {
2317 	for (uint_t i = 0; i < i40e->i40e_num_rx_groups; i++) {
2318 		uint32_t seed[I40E_PFQF_HKEY_MAX_INDEX + 1];
2319 		struct i40e_aqc_get_set_rss_key_data key;
2320 		const char *u8seed;
2321 		enum i40e_status_code status;
2322 		uint16_t vsi_number = i40e->i40e_vsis[i].iv_number;
2323 
2324 		(void) random_get_pseudo_bytes((uint8_t *)seed, sizeof (seed));
2325 		u8seed = (char *)seed;
2326 
2327 		CTASSERT(sizeof (key) >= (sizeof (key.standard_rss_key) +
2328 		    sizeof (key.extended_hash_key)));
2329 
2330 		bcopy(u8seed, key.standard_rss_key,
2331 		    sizeof (key.standard_rss_key));
2332 		bcopy(&u8seed[sizeof (key.standard_rss_key)],
2333 		    key.extended_hash_key, sizeof (key.extended_hash_key));
2334 
2335 		ASSERT3U(vsi_number, !=, 0);
2336 		status = i40e_aq_set_rss_key(hw, vsi_number, &key);
2337 
2338 		if (status != I40E_SUCCESS) {
2339 			i40e_error(i40e, "failed to set RSS key for VSI %u: %d",
2340 			    vsi_number, status);
2341 			return (B_FALSE);
2342 		}
2343 	}
2344 
2345 	return (B_TRUE);
2346 }
2347 
2348 /*
2349  * Configure the RSS key. For the X710 controller family, this is set on a
2350  * per-PF basis via registers. For the X722, this is done on a per-VSI basis
2351  * through the admin queue.
2352  */
2353 static boolean_t
2354 i40e_config_rss_key(i40e_t *i40e, i40e_hw_t *hw)
2355 {
2356 	if (i40e_is_x722(i40e)) {
2357 		if (!i40e_config_rss_key_x722(i40e, hw))
2358 			return (B_FALSE);
2359 	} else {
2360 		uint32_t seed[I40E_PFQF_HKEY_MAX_INDEX + 1];
2361 
2362 		(void) random_get_pseudo_bytes((uint8_t *)seed, sizeof (seed));
2363 		for (uint_t i = 0; i <= I40E_PFQF_HKEY_MAX_INDEX; i++)
2364 			i40e_write_rx_ctl(hw, I40E_PFQF_HKEY(i), seed[i]);
2365 	}
2366 
2367 	return (B_TRUE);
2368 }
2369 
2370 /*
2371  * Populate the LUT. The size of each entry in the LUT depends on the controller
2372  * family, with the X722 using a known 7-bit width. On the X710 controller, this
2373  * is programmed through its control registers where as on the X722 this is
2374  * configured through the admin queue. Also of note, the X722 allows the LUT to
2375  * be set on a per-PF or VSI basis. At this time we use the PF setting. If we
2376  * decide to use the per-VSI LUT in the future, then we will need to modify the
2377  * i40e_add_vsi() function to set the RSS LUT bits in the queueing section.
2378  *
2379  * We populate the LUT in a round robin fashion with the rx queue indices from 0
2380  * to i40e_num_trqpairs_per_vsi - 1.
2381  */
2382 static boolean_t
2383 i40e_config_rss_hlut(i40e_t *i40e, i40e_hw_t *hw)
2384 {
2385 	uint32_t *hlut;
2386 	uint8_t lut_mask;
2387 	uint_t i;
2388 	boolean_t ret = B_FALSE;
2389 
2390 	/*
2391 	 * We always configure the PF with a table size of 512 bytes in
2392 	 * i40e_chip_start().
2393 	 */
2394 	hlut = kmem_alloc(I40E_HLUT_TABLE_SIZE, KM_NOSLEEP);
2395 	if (hlut == NULL) {
2396 		i40e_error(i40e, "i40e_config_rss() buffer allocation failed");
2397 		return (B_FALSE);
2398 	}
2399 
2400 	/*
2401 	 * The width of the X722 is apparently defined to be 7 bits, regardless
2402 	 * of the capability.
2403 	 */
2404 	if (i40e_is_x722(i40e)) {
2405 		lut_mask = (1 << 7) - 1;
2406 	} else {
2407 		lut_mask = (1 << hw->func_caps.rss_table_entry_width) - 1;
2408 	}
2409 
2410 	for (i = 0; i < I40E_HLUT_TABLE_SIZE; i++) {
2411 		((uint8_t *)hlut)[i] =
2412 		    (i % i40e->i40e_num_trqpairs_per_vsi) & lut_mask;
2413 	}
2414 
2415 	if (i40e_is_x722(i40e)) {
2416 		enum i40e_status_code status;
2417 
2418 		status = i40e_aq_set_rss_lut(hw, 0, B_TRUE, (uint8_t *)hlut,
2419 		    I40E_HLUT_TABLE_SIZE);
2420 
2421 		if (status != I40E_SUCCESS) {
2422 			i40e_error(i40e, "failed to set RSS LUT %d: %d",
2423 			    status, hw->aq.asq_last_status);
2424 			goto out;
2425 		}
2426 	} else {
2427 		for (i = 0; i < I40E_HLUT_TABLE_SIZE >> 2; i++) {
2428 			I40E_WRITE_REG(hw, I40E_PFQF_HLUT(i), hlut[i]);
2429 		}
2430 	}
2431 	ret = B_TRUE;
2432 out:
2433 	kmem_free(hlut, I40E_HLUT_TABLE_SIZE);
2434 	return (ret);
2435 }
2436 
2437 /*
2438  * Set up RSS.
2439  *	1. Seed the hash key.
2440  *	2. Enable PCTYPEs for the hash filter.
2441  *	3. Populate the LUT.
2442  */
2443 static boolean_t
2444 i40e_config_rss(i40e_t *i40e, i40e_hw_t *hw)
2445 {
2446 	uint64_t hena;
2447 
2448 	/*
2449 	 * 1. Seed the hash key
2450 	 */
2451 	if (!i40e_config_rss_key(i40e, hw))
2452 		return (B_FALSE);
2453 
2454 	/*
2455 	 * 2. Configure PCTYPES
2456 	 */
2457 	hena = (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) |
2458 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP) |
2459 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) |
2460 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_UDP) |
2461 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV4) |
2462 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) |
2463 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP) |
2464 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) |
2465 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_UDP) |
2466 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV6) |
2467 	    (1ULL << I40E_FILTER_PCTYPE_L2_PAYLOAD);
2468 
2469 	/*
2470 	 * Add additional types supported by the X722 controller.
2471 	 */
2472 	if (i40e_is_x722(i40e)) {
2473 		hena |= (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) |
2474 		    (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) |
2475 		    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) |
2476 		    (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) |
2477 		    (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP) |
2478 		    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK);
2479 	}
2480 
2481 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(0), (uint32_t)hena);
2482 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(1), (uint32_t)(hena >> 32));
2483 
2484 	/*
2485 	 * 3. Populate LUT
2486 	 */
2487 	return (i40e_config_rss_hlut(i40e, hw));
2488 }
2489 
2490 /*
2491  * Wrapper to kick the chipset on.
2492  */
2493 static boolean_t
2494 i40e_chip_start(i40e_t *i40e)
2495 {
2496 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2497 	struct i40e_filter_control_settings filter;
2498 	int rc;
2499 	uint8_t err;
2500 
2501 	if (((hw->aq.fw_maj_ver == 4) && (hw->aq.fw_min_ver < 33)) ||
2502 	    (hw->aq.fw_maj_ver < 4)) {
2503 		i40e_msec_delay(75);
2504 		if (i40e_aq_set_link_restart_an(hw, TRUE, NULL) !=
2505 		    I40E_SUCCESS) {
2506 			i40e_error(i40e, "failed to restart link: admin queue "
2507 			    "error: %d", hw->aq.asq_last_status);
2508 			return (B_FALSE);
2509 		}
2510 	}
2511 
2512 	/* Determine hardware state */
2513 	i40e_get_hw_state(i40e, hw);
2514 
2515 	/* For now, we always disable Ethernet Flow Control. */
2516 	hw->fc.requested_mode = I40E_FC_NONE;
2517 	rc = i40e_set_fc(hw, &err, B_TRUE);
2518 	if (rc != I40E_SUCCESS) {
2519 		i40e_error(i40e, "Setting flow control failed, returned %d"
2520 		    " with error: 0x%x", rc, err);
2521 		return (B_FALSE);
2522 	}
2523 
2524 	/* Initialize mac addresses. */
2525 	i40e_init_macaddrs(i40e, hw);
2526 
2527 	/*
2528 	 * Set up the filter control. If the hash lut size is changed from
2529 	 * I40E_HASH_LUT_SIZE_512 then I40E_HLUT_TABLE_SIZE and
2530 	 * i40e_config_rss_hlut() will need to be updated.
2531 	 */
2532 	bzero(&filter, sizeof (filter));
2533 	filter.enable_ethtype = TRUE;
2534 	filter.enable_macvlan = TRUE;
2535 	filter.hash_lut_size = I40E_HASH_LUT_SIZE_512;
2536 
2537 	rc = i40e_set_filter_control(hw, &filter);
2538 	if (rc != I40E_SUCCESS) {
2539 		i40e_error(i40e, "i40e_set_filter_control() returned %d", rc);
2540 		return (B_FALSE);
2541 	}
2542 
2543 	i40e_intr_chip_init(i40e);
2544 
2545 	rc = i40e_get_mac_seid(i40e);
2546 	if (rc == -1) {
2547 		i40e_error(i40e, "failed to obtain MAC Uplink SEID");
2548 		return (B_FALSE);
2549 	}
2550 	i40e->i40e_mac_seid = (uint16_t)rc;
2551 
2552 	/*
2553 	 * Create a VEB in order to support multiple VSIs. Each VSI
2554 	 * functions as a MAC group. This call sets the PF's MAC as
2555 	 * the uplink port and the PF's default VSI as the default
2556 	 * downlink port.
2557 	 */
2558 	rc = i40e_aq_add_veb(hw, i40e->i40e_mac_seid, I40E_DEF_VSI_SEID(i40e),
2559 	    0x1, B_TRUE, &i40e->i40e_veb_seid, B_FALSE, NULL);
2560 	if (rc != I40E_SUCCESS) {
2561 		i40e_error(i40e, "i40e_aq_add_veb() failed %d: %d", rc,
2562 		    hw->aq.asq_last_status);
2563 		return (B_FALSE);
2564 	}
2565 
2566 	if (!i40e_config_def_vsi(i40e, hw))
2567 		return (B_FALSE);
2568 
2569 	for (uint_t i = 1; i < i40e->i40e_num_rx_groups; i++) {
2570 		if (!i40e_add_vsi(i40e, hw, i))
2571 			return (B_FALSE);
2572 	}
2573 
2574 	if (!i40e_config_rss(i40e, hw))
2575 		return (B_FALSE);
2576 
2577 	i40e_flush(hw);
2578 
2579 	return (B_TRUE);
2580 }
2581 
2582 /*
2583  * Take care of tearing down the rx ring. See 8.3.3.1.2 for more information.
2584  */
2585 static void
2586 i40e_shutdown_rx_ring(i40e_trqpair_t *itrq)
2587 {
2588 	i40e_t *i40e = itrq->itrq_i40e;
2589 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2590 	uint32_t reg;
2591 
2592 	/*
2593 	 * Step 1. 8.3.3.1.2 suggests the interrupt is removed from the
2594 	 * hardware interrupt linked list (see i40e_intr.c) but for
2595 	 * simplicity we keep this list immutable until the device
2596 	 * (distinct from an individual ring) is stopped.
2597 	 */
2598 
2599 	/*
2600 	 * Step 2. Request the queue by clearing QENA_REQ. It may not be
2601 	 * set due to unwinding from failures and a partially enabled
2602 	 * ring set.
2603 	 */
2604 	reg = I40E_READ_REG(hw, I40E_QRX_ENA(itrq->itrq_index));
2605 	if (!(reg & I40E_QRX_ENA_QENA_REQ_MASK))
2606 		return;
2607 	VERIFY((reg & I40E_QRX_ENA_QENA_REQ_MASK) ==
2608 	    I40E_QRX_ENA_QENA_REQ_MASK);
2609 	reg &= ~I40E_QRX_ENA_QENA_REQ_MASK;
2610 	I40E_WRITE_REG(hw, I40E_QRX_ENA(itrq->itrq_index), reg);
2611 
2612 	/*
2613 	 * Step 3. Wait for the disable to take, by having QENA_STAT in the FPM
2614 	 * be cleared. Note that we could still receive data in the queue during
2615 	 * this time. We don't actually wait for this now and instead defer this
2616 	 * to i40e_shutdown_ring_wait(), after we've interleaved disabling the
2617 	 * TX queue as well.
2618 	 */
2619 }
2620 
2621 static void
2622 i40e_shutdown_tx_ring(i40e_trqpair_t *itrq)
2623 {
2624 	i40e_t *i40e = itrq->itrq_i40e;
2625 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2626 	uint32_t reg;
2627 
2628 	/*
2629 	 * Step 2. Set the SET_QDIS flag for the queue.
2630 	 */
2631 	i40e_pre_tx_queue_cfg(hw, itrq->itrq_index, B_FALSE);
2632 
2633 	/*
2634 	 * Step 3. Wait at least 400 usec.
2635 	 */
2636 	drv_usecwait(500);
2637 
2638 	/*
2639 	 * Step 4. Clear the QENA_REQ flag which tells hardware to
2640 	 * quiesce. If QENA_REQ is not already set then that means that
2641 	 * we likely already tried to disable this queue.
2642 	 */
2643 	reg = I40E_READ_REG(hw, I40E_QTX_ENA(itrq->itrq_index));
2644 	if ((reg & I40E_QTX_ENA_QENA_REQ_MASK) != 0) {
2645 		reg &= ~I40E_QTX_ENA_QENA_REQ_MASK;
2646 		I40E_WRITE_REG(hw, I40E_QTX_ENA(itrq->itrq_index), reg);
2647 	}
2648 
2649 	/*
2650 	 * Step 5. Wait for the drain to finish. This will be done by the
2651 	 * hardware removing the QENA_STAT flag from the queue. Rather than
2652 	 * waiting here, we interleave it with the receive shutdown in
2653 	 * i40e_shutdown_ring_wait().
2654 	 */
2655 }
2656 
2657 /*
2658  * Wait for a ring to be shut down. e.g. Steps 2 and 5 from the above
2659  * functions.
2660  */
2661 static boolean_t
2662 i40e_shutdown_ring_wait(i40e_trqpair_t *itrq)
2663 {
2664 	i40e_t *i40e = itrq->itrq_i40e;
2665 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2666 	uint32_t reg;
2667 	int try;
2668 
2669 	for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2670 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(itrq->itrq_index));
2671 		if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0)
2672 			break;
2673 		i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2674 	}
2675 
2676 	if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) != 0) {
2677 		i40e_error(i40e, "timed out disabling rx queue %d",
2678 		    itrq->itrq_index);
2679 		return (B_FALSE);
2680 	}
2681 
2682 	for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2683 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(itrq->itrq_index));
2684 		if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0)
2685 			break;
2686 		i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2687 	}
2688 
2689 	if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) != 0) {
2690 		i40e_error(i40e, "timed out disabling tx queue %d",
2691 		    itrq->itrq_index);
2692 		return (B_FALSE);
2693 	}
2694 
2695 	return (B_TRUE);
2696 }
2697 
2698 
2699 /*
2700  * Shutdown an individual ring and release any memory.
2701  */
2702 boolean_t
2703 i40e_shutdown_ring(i40e_trqpair_t *itrq)
2704 {
2705 	boolean_t rv = B_TRUE;
2706 
2707 	/*
2708 	 * Tell transmit path to quiesce, and wait until done.
2709 	 */
2710 	if (i40e_ring_tx_quiesce(itrq)) {
2711 		/* Already quiesced. */
2712 		return (B_TRUE);
2713 	}
2714 
2715 	i40e_shutdown_rx_ring(itrq);
2716 	i40e_shutdown_tx_ring(itrq);
2717 	if (!i40e_shutdown_ring_wait(itrq))
2718 		rv = B_FALSE;
2719 
2720 	/*
2721 	 * After the ring has stopped, we need to wait 50ms before
2722 	 * programming it again. Rather than wait here, we'll record
2723 	 * the time the ring was stopped. When the ring is started, we'll
2724 	 * check if enough time has expired and then wait if necessary.
2725 	 */
2726 	itrq->irtq_time_stopped = gethrtime();
2727 
2728 	/*
2729 	 * The rings have been stopped in the hardware, now wait for
2730 	 * a possibly active interrupt thread.
2731 	 */
2732 	i40e_intr_quiesce(itrq);
2733 
2734 	mutex_enter(&itrq->itrq_tx_lock);
2735 	i40e_tx_cleanup_ring(itrq);
2736 	mutex_exit(&itrq->itrq_tx_lock);
2737 
2738 	i40e_free_ring_mem(itrq, B_FALSE);
2739 
2740 	return (rv);
2741 }
2742 
2743 /*
2744  * Shutdown all the rings.
2745  * Called from i40e_stop(), and hopefully the mac layer has already
2746  * called ring stop for each ring, which would make this almost a no-op.
2747  */
2748 static boolean_t
2749 i40e_shutdown_rings(i40e_t *i40e)
2750 {
2751 	boolean_t rv = B_TRUE;
2752 	int i;
2753 
2754 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2755 		if (!i40e_shutdown_ring(&i40e->i40e_trqpairs[i]))
2756 			rv = B_FALSE;
2757 	}
2758 
2759 	return (rv);
2760 }
2761 
2762 static void
2763 i40e_setup_rx_descs(i40e_trqpair_t *itrq)
2764 {
2765 	int i;
2766 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2767 
2768 	for (i = 0; i < rxd->rxd_ring_size; i++) {
2769 		i40e_rx_control_block_t *rcb;
2770 		i40e_rx_desc_t *rdesc;
2771 
2772 		rcb = rxd->rxd_work_list[i];
2773 		rdesc = &rxd->rxd_desc_ring[i];
2774 
2775 		rdesc->read.pkt_addr =
2776 		    CPU_TO_LE64((uintptr_t)rcb->rcb_dma.dmab_dma_address);
2777 		rdesc->read.hdr_addr = 0;
2778 	}
2779 }
2780 
2781 static boolean_t
2782 i40e_setup_rx_hmc(i40e_trqpair_t *itrq)
2783 {
2784 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2785 	i40e_t *i40e = itrq->itrq_i40e;
2786 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2787 
2788 	struct i40e_hmc_obj_rxq rctx;
2789 	int err;
2790 
2791 	bzero(&rctx, sizeof (struct i40e_hmc_obj_rxq));
2792 	rctx.base = rxd->rxd_desc_area.dmab_dma_address /
2793 	    I40E_HMC_RX_CTX_UNIT;
2794 	rctx.qlen = rxd->rxd_ring_size;
2795 	VERIFY(i40e->i40e_rx_buf_size >= I40E_HMC_RX_DBUFF_MIN);
2796 	VERIFY(i40e->i40e_rx_buf_size <= I40E_HMC_RX_DBUFF_MAX);
2797 	rctx.dbuff = i40e->i40e_rx_buf_size >> I40E_RXQ_CTX_DBUFF_SHIFT;
2798 	rctx.hbuff = 0 >> I40E_RXQ_CTX_HBUFF_SHIFT;
2799 	rctx.dtype = I40E_HMC_RX_DTYPE_NOSPLIT;
2800 	rctx.dsize = I40E_HMC_RX_DSIZE_32BYTE;
2801 	rctx.crcstrip = I40E_HMC_RX_CRCSTRIP_ENABLE;
2802 	rctx.fc_ena = I40E_HMC_RX_FC_DISABLE;
2803 	rctx.l2tsel = I40E_HMC_RX_L2TAGORDER;
2804 	rctx.hsplit_0 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2805 	rctx.hsplit_1 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2806 	rctx.showiv = I40E_HMC_RX_INVLAN_DONTSTRIP;
2807 	rctx.rxmax = i40e->i40e_frame_max;
2808 	rctx.tphrdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2809 	rctx.tphwdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2810 	rctx.tphdata_ena = I40E_HMC_RX_TPH_DISABLE;
2811 	rctx.tphhead_ena = I40E_HMC_RX_TPH_DISABLE;
2812 	rctx.lrxqthresh = I40E_HMC_RX_LOWRXQ_NOINTR;
2813 
2814 	/*
2815 	 * This must be set to 0x1, see Table 8-12 in section 8.3.3.2.2.
2816 	 */
2817 	rctx.prefena = I40E_HMC_RX_PREFENA;
2818 
2819 	err = i40e_clear_lan_rx_queue_context(hw, itrq->itrq_index);
2820 	if (err != I40E_SUCCESS) {
2821 		i40e_error(i40e, "failed to clear rx queue %d context: %d",
2822 		    itrq->itrq_index, err);
2823 		return (B_FALSE);
2824 	}
2825 
2826 	err = i40e_set_lan_rx_queue_context(hw, itrq->itrq_index, &rctx);
2827 	if (err != I40E_SUCCESS) {
2828 		i40e_error(i40e, "failed to set rx queue %d context: %d",
2829 		    itrq->itrq_index, err);
2830 		return (B_FALSE);
2831 	}
2832 
2833 	return (B_TRUE);
2834 }
2835 
2836 /*
2837  * Take care of setting up the descriptor ring and actually programming the
2838  * device. See 8.3.3.1.1 for the full list of steps we need to do to enable the
2839  * rx rings.
2840  */
2841 static boolean_t
2842 i40e_setup_rx_ring(i40e_trqpair_t *itrq)
2843 {
2844 	i40e_t *i40e = itrq->itrq_i40e;
2845 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2846 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2847 	uint32_t reg;
2848 	int i;
2849 
2850 	/*
2851 	 * Step 1. Program all receive ring descriptors.
2852 	 */
2853 	i40e_setup_rx_descs(itrq);
2854 
2855 	/*
2856 	 * Step 2. Program the queue's FPM/HMC context.
2857 	 */
2858 	if (!i40e_setup_rx_hmc(itrq))
2859 		return (B_FALSE);
2860 
2861 	/*
2862 	 * Step 3. Clear the queue's tail pointer and set it to the end
2863 	 * of the space.
2864 	 */
2865 	I40E_WRITE_REG(hw, I40E_QRX_TAIL(itrq->itrq_index), 0);
2866 	I40E_WRITE_REG(hw, I40E_QRX_TAIL(itrq->itrq_index),
2867 	    rxd->rxd_ring_size - 1);
2868 
2869 	/*
2870 	 * Step 4. Enable the queue via the QENA_REQ.
2871 	 */
2872 	reg = I40E_READ_REG(hw, I40E_QRX_ENA(itrq->itrq_index));
2873 	VERIFY0(reg & (I40E_QRX_ENA_QENA_REQ_MASK |
2874 	    I40E_QRX_ENA_QENA_STAT_MASK));
2875 	reg |= I40E_QRX_ENA_QENA_REQ_MASK;
2876 	I40E_WRITE_REG(hw, I40E_QRX_ENA(itrq->itrq_index), reg);
2877 
2878 	/*
2879 	 * Step 5. Verify that QENA_STAT has been set. It's promised
2880 	 * that this should occur within about 10 us, but like other
2881 	 * systems, we give the card a bit more time.
2882 	 */
2883 	for (i = 0; i < I40E_RING_WAIT_NTRIES; i++) {
2884 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(itrq->itrq_index));
2885 
2886 		if (reg & I40E_QRX_ENA_QENA_STAT_MASK)
2887 			break;
2888 		i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2889 	}
2890 
2891 	if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0) {
2892 		i40e_error(i40e, "failed to enable rx queue %d, timed "
2893 		    "out.", itrq->itrq_index);
2894 		return (B_FALSE);
2895 	}
2896 
2897 	return (B_TRUE);
2898 }
2899 
2900 static boolean_t
2901 i40e_setup_tx_hmc(i40e_trqpair_t *itrq)
2902 {
2903 	i40e_t *i40e = itrq->itrq_i40e;
2904 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2905 
2906 	struct i40e_hmc_obj_txq tctx;
2907 	struct i40e_vsi_context	context;
2908 	int err;
2909 
2910 	bzero(&tctx, sizeof (struct i40e_hmc_obj_txq));
2911 	tctx.new_context = I40E_HMC_TX_NEW_CONTEXT;
2912 	tctx.base = itrq->itrq_desc_area.dmab_dma_address /
2913 	    I40E_HMC_TX_CTX_UNIT;
2914 	tctx.fc_ena = I40E_HMC_TX_FC_DISABLE;
2915 	tctx.timesync_ena = I40E_HMC_TX_TS_DISABLE;
2916 	tctx.fd_ena = I40E_HMC_TX_FD_DISABLE;
2917 	tctx.alt_vlan_ena = I40E_HMC_TX_ALT_VLAN_DISABLE;
2918 	tctx.head_wb_ena = I40E_HMC_TX_WB_ENABLE;
2919 	tctx.qlen = itrq->itrq_tx_ring_size;
2920 	tctx.tphrdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2921 	tctx.tphrpacket_ena = I40E_HMC_TX_TPH_DISABLE;
2922 	tctx.tphwdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2923 	tctx.head_wb_addr = itrq->itrq_desc_area.dmab_dma_address +
2924 	    sizeof (i40e_tx_desc_t) * itrq->itrq_tx_ring_size;
2925 
2926 	/*
2927 	 * This field isn't actually documented, like crc, but it suggests that
2928 	 * it should be zeroed. We leave both of these here because of that for
2929 	 * now. We should check with Intel on why these are here even.
2930 	 */
2931 	tctx.crc = 0;
2932 	tctx.rdylist_act = 0;
2933 
2934 	/*
2935 	 * We're supposed to assign the rdylist field with the value of the
2936 	 * traffic class index for the first device. We query the VSI parameters
2937 	 * again to get what the handle is. Note that every queue is always
2938 	 * assigned to traffic class zero, because we don't actually use them.
2939 	 */
2940 	bzero(&context, sizeof (struct i40e_vsi_context));
2941 	context.seid = I40E_DEF_VSI_SEID(i40e);
2942 	context.pf_num = hw->pf_id;
2943 	err = i40e_aq_get_vsi_params(hw, &context, NULL);
2944 	if (err != I40E_SUCCESS) {
2945 		i40e_error(i40e, "get VSI params failed with %d", err);
2946 		return (B_FALSE);
2947 	}
2948 	tctx.rdylist = LE_16(context.info.qs_handle[0]);
2949 
2950 	err = i40e_clear_lan_tx_queue_context(hw, itrq->itrq_index);
2951 	if (err != I40E_SUCCESS) {
2952 		i40e_error(i40e, "failed to clear tx queue %d context: %d",
2953 		    itrq->itrq_index, err);
2954 		return (B_FALSE);
2955 	}
2956 
2957 	err = i40e_set_lan_tx_queue_context(hw, itrq->itrq_index, &tctx);
2958 	if (err != I40E_SUCCESS) {
2959 		i40e_error(i40e, "failed to set tx queue %d context: %d",
2960 		    itrq->itrq_index, err);
2961 		return (B_FALSE);
2962 	}
2963 
2964 	return (B_TRUE);
2965 }
2966 
2967 /*
2968  * Take care of setting up the descriptor ring and actually programming the
2969  * device. See 8.4.3.1.1 for what we need to do here.
2970  */
2971 static boolean_t
2972 i40e_setup_tx_ring(i40e_trqpair_t *itrq)
2973 {
2974 	i40e_t *i40e = itrq->itrq_i40e;
2975 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2976 	uint32_t reg;
2977 	int i;
2978 
2979 	/*
2980 	 * Step 1. Clear the queue disable flag and verify that the
2981 	 * index is set correctly.
2982 	 */
2983 	i40e_pre_tx_queue_cfg(hw, itrq->itrq_index, B_TRUE);
2984 
2985 	/*
2986 	 * Step 2. Prepare the queue's FPM/HMC context.
2987 	 */
2988 	if (!i40e_setup_tx_hmc(itrq))
2989 		return (B_FALSE);
2990 
2991 	/*
2992 	 * Step 3. Verify that it's clear that this PF owns this queue.
2993 	 */
2994 	reg = I40E_QTX_CTL_PF_QUEUE;
2995 	reg |= (hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2996 	    I40E_QTX_CTL_PF_INDX_MASK;
2997 	I40E_WRITE_REG(hw, I40E_QTX_CTL(itrq->itrq_index), reg);
2998 	i40e_flush(hw);
2999 
3000 	/*
3001 	 * Step 4. Set the QENA_REQ flag.
3002 	 */
3003 	reg = I40E_READ_REG(hw, I40E_QTX_ENA(itrq->itrq_index));
3004 	VERIFY0(reg & (I40E_QTX_ENA_QENA_REQ_MASK |
3005 	    I40E_QTX_ENA_QENA_STAT_MASK));
3006 	reg |= I40E_QTX_ENA_QENA_REQ_MASK;
3007 	I40E_WRITE_REG(hw, I40E_QTX_ENA(itrq->itrq_index), reg);
3008 
3009 	/*
3010 	 * Step 5. Verify that QENA_STAT has been set. It's promised
3011 	 * that this should occur within about 10 us, but like BSD,
3012 	 * we'll try for up to 100 ms for this queue.
3013 	 */
3014 	for (i = 0; i < I40E_RING_WAIT_NTRIES; i++) {
3015 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(itrq->itrq_index));
3016 
3017 		if (reg & I40E_QTX_ENA_QENA_STAT_MASK)
3018 			break;
3019 		i40e_msec_delay(I40E_RING_WAIT_PAUSE);
3020 	}
3021 
3022 	if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0) {
3023 		i40e_error(i40e, "failed to enable tx queue %d, timed "
3024 		    "out", itrq->itrq_index);
3025 		return (B_FALSE);
3026 	}
3027 
3028 	return (B_TRUE);
3029 }
3030 
3031 int
3032 i40e_setup_ring(i40e_trqpair_t *itrq)
3033 {
3034 	i40e_t *i40e = itrq->itrq_i40e;
3035 	hrtime_t now, gap;
3036 
3037 	if (!i40e_alloc_ring_mem(itrq)) {
3038 		i40e_error(i40e, "Failed to allocate ring memory");
3039 		return (ENOMEM);
3040 	}
3041 
3042 	/*
3043 	 * 8.3.3.1.1 Receive Queue Enable Flow states software should
3044 	 * wait at least 50ms between ring disable and enable. See how
3045 	 * long we need to wait, and wait only if required.
3046 	 */
3047 	now = gethrtime();
3048 	gap = NSEC2MSEC(now - itrq->irtq_time_stopped);
3049 	if (gap < I40E_RING_ENABLE_GAP && gap != 0)
3050 		delay(drv_usectohz(gap * 1000));
3051 
3052 	mutex_enter(&itrq->itrq_intr_lock);
3053 	if (!i40e_setup_rx_ring(itrq))
3054 		goto failed;
3055 
3056 	if (!i40e_setup_tx_ring(itrq))
3057 		goto failed;
3058 
3059 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_reg_handle) !=
3060 	    DDI_FM_OK)
3061 		goto failed;
3062 
3063 	itrq->itrq_intr_quiesce = B_FALSE;
3064 	mutex_exit(&itrq->itrq_intr_lock);
3065 
3066 	mutex_enter(&itrq->itrq_tx_lock);
3067 	itrq->itrq_tx_quiesce = B_FALSE;
3068 	mutex_exit(&itrq->itrq_tx_lock);
3069 
3070 	return (0);
3071 
3072 failed:
3073 	mutex_exit(&itrq->itrq_intr_lock);
3074 	i40e_free_ring_mem(itrq, B_TRUE);
3075 	ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
3076 
3077 	return (EIO);
3078 }
3079 
3080 void
3081 i40e_stop(i40e_t *i40e)
3082 {
3083 	uint_t i;
3084 	i40e_hw_t *hw = &i40e->i40e_hw_space;
3085 
3086 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
3087 
3088 	/*
3089 	 * Shutdown and drain the tx and rx pipeline. We do this using the
3090 	 * following steps.
3091 	 *
3092 	 * 1) Shutdown interrupts to all the queues (trying to keep the admin
3093 	 *    queue alive).
3094 	 *
3095 	 * 2) Remove all of the interrupt tx and rx causes by setting the
3096 	 *    interrupt linked lists to zero.
3097 	 *
3098 	 * 2) Shutdown the tx and rx rings. Because i40e_shutdown_rings() should
3099 	 *    wait for all the queues to be disabled, once we reach that point
3100 	 *    it should be safe to free associated data.
3101 	 *
3102 	 * 4) Wait 50ms after all that is done. This ensures that the rings are
3103 	 *    ready for programming again and we don't have to think about this
3104 	 *    in other parts of the driver.
3105 	 *
3106 	 * 5) Disable remaining chip interrupts, (admin queue, etc.)
3107 	 *
3108 	 * 6) Verify that FM is happy with all the register accesses we
3109 	 *    performed.
3110 	 */
3111 	i40e_intr_io_disable_all(i40e);
3112 	i40e_intr_io_clear_cause(i40e);
3113 
3114 	if (!i40e_shutdown_rings(i40e))
3115 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
3116 
3117 	/*
3118 	 * We don't delete the default VSI because it replaces the VEB
3119 	 * after VEB deletion (see the "Delete Element" section).
3120 	 * Furthermore, since the default VSI is provided by the
3121 	 * firmware, we never attempt to delete it.
3122 	 */
3123 	for (i = 1; i < i40e->i40e_num_rx_groups; i++) {
3124 		i40e_delete_vsi(i40e, i);
3125 	}
3126 
3127 	if (i40e->i40e_veb_seid != 0) {
3128 		int rc = i40e_aq_delete_element(hw, i40e->i40e_veb_seid, NULL);
3129 
3130 		if (rc != I40E_SUCCESS) {
3131 			i40e_error(i40e, "Failed to delete VEB %d: %d", rc,
3132 			    hw->aq.asq_last_status);
3133 		}
3134 
3135 		i40e->i40e_veb_seid = 0;
3136 	}
3137 
3138 	i40e_intr_chip_fini(i40e);
3139 
3140 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
3141 	    DDI_FM_OK) {
3142 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
3143 	}
3144 
3145 	for (i = 0; i < i40e->i40e_num_rx_groups; i++) {
3146 		i40e_stat_vsi_fini(i40e, i);
3147 	}
3148 
3149 	i40e->i40e_link_speed = 0;
3150 	i40e->i40e_link_duplex = 0;
3151 	i40e_link_state_set(i40e, LINK_STATE_UNKNOWN);
3152 }
3153 
3154 boolean_t
3155 i40e_start(i40e_t *i40e)
3156 {
3157 	i40e_hw_t *hw = &i40e->i40e_hw_space;
3158 	boolean_t rc = B_TRUE;
3159 	int err;
3160 
3161 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
3162 
3163 	if (!i40e_chip_start(i40e)) {
3164 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
3165 		rc = B_FALSE;
3166 		goto done;
3167 	}
3168 
3169 	/*
3170 	 * Enable broadcast traffic; however, do not enable multicast traffic.
3171 	 * That's handle exclusively through MAC's mc_multicst routines.
3172 	 */
3173 	err = i40e_aq_set_vsi_broadcast(hw, I40E_DEF_VSI_SEID(i40e), B_TRUE,
3174 	    NULL);
3175 	if (err != I40E_SUCCESS) {
3176 		i40e_error(i40e, "failed to set default VSI: %d", err);
3177 		rc = B_FALSE;
3178 		goto done;
3179 	}
3180 
3181 	err = i40e_aq_set_mac_config(hw, i40e->i40e_frame_max, B_TRUE, 0, NULL);
3182 	if (err != I40E_SUCCESS) {
3183 		i40e_error(i40e, "failed to set MAC config: %d", err);
3184 		rc = B_FALSE;
3185 		goto done;
3186 	}
3187 
3188 	/*
3189 	 * Finally, make sure that we're happy from an FM perspective.
3190 	 */
3191 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_reg_handle) !=
3192 	    DDI_FM_OK) {
3193 		rc = B_FALSE;
3194 		goto done;
3195 	}
3196 
3197 	/* Clear state bits prior to final interrupt enabling. */
3198 	atomic_and_32(&i40e->i40e_state,
3199 	    ~(I40E_ERROR | I40E_STALL | I40E_OVERTEMP));
3200 
3201 	i40e_intr_io_enable_all(i40e);
3202 
3203 done:
3204 	if (rc == B_FALSE) {
3205 		i40e_stop(i40e);
3206 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
3207 	}
3208 
3209 	return (rc);
3210 }
3211 
3212 /*
3213  * We may have loaned up descriptors to the stack. As such, if we still have
3214  * them outstanding, then we will not continue with detach.
3215  */
3216 static boolean_t
3217 i40e_drain_rx(i40e_t *i40e)
3218 {
3219 	mutex_enter(&i40e->i40e_rx_pending_lock);
3220 	while (i40e->i40e_rx_pending > 0) {
3221 		if (cv_reltimedwait(&i40e->i40e_rx_pending_cv,
3222 		    &i40e->i40e_rx_pending_lock,
3223 		    drv_usectohz(I40E_DRAIN_RX_WAIT), TR_CLOCK_TICK) == -1) {
3224 			mutex_exit(&i40e->i40e_rx_pending_lock);
3225 			return (B_FALSE);
3226 		}
3227 	}
3228 	mutex_exit(&i40e->i40e_rx_pending_lock);
3229 
3230 	return (B_TRUE);
3231 }
3232 
3233 /*
3234  * DDI UFM Callbacks
3235  */
3236 static int
3237 i40e_ufm_fill_image(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
3238     ddi_ufm_image_t *img)
3239 {
3240 	if (imgno != 0)
3241 		return (EINVAL);
3242 
3243 	ddi_ufm_image_set_desc(img, "Firmware");
3244 	ddi_ufm_image_set_nslots(img, 1);
3245 
3246 	return (0);
3247 }
3248 
3249 static int
3250 i40e_ufm_fill_slot(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
3251     uint_t slotno, ddi_ufm_slot_t *slot)
3252 {
3253 	i40e_t *i40e = (i40e_t *)arg;
3254 	char *fw_ver = NULL, *fw_bld = NULL, *api_ver = NULL;
3255 	nvlist_t *misc = NULL;
3256 	uint_t flags = DDI_PROP_DONTPASS;
3257 	int err;
3258 
3259 	if (imgno != 0 || slotno != 0 ||
3260 	    ddi_prop_lookup_string(DDI_DEV_T_ANY, i40e->i40e_dip, flags,
3261 	    "firmware-version", &fw_ver) != DDI_PROP_SUCCESS ||
3262 	    ddi_prop_lookup_string(DDI_DEV_T_ANY, i40e->i40e_dip, flags,
3263 	    "firmware-build", &fw_bld) != DDI_PROP_SUCCESS ||
3264 	    ddi_prop_lookup_string(DDI_DEV_T_ANY, i40e->i40e_dip, flags,
3265 	    "api-version", &api_ver) != DDI_PROP_SUCCESS) {
3266 		err = EINVAL;
3267 		goto err;
3268 	}
3269 
3270 	ddi_ufm_slot_set_attrs(slot, DDI_UFM_ATTR_ACTIVE);
3271 	ddi_ufm_slot_set_version(slot, fw_ver);
3272 
3273 	(void) nvlist_alloc(&misc, NV_UNIQUE_NAME, KM_SLEEP);
3274 	if ((err = nvlist_add_string(misc, "firmware-build", fw_bld)) != 0 ||
3275 	    (err = nvlist_add_string(misc, "api-version", api_ver)) != 0) {
3276 		goto err;
3277 	}
3278 	ddi_ufm_slot_set_misc(slot, misc);
3279 
3280 	ddi_prop_free(fw_ver);
3281 	ddi_prop_free(fw_bld);
3282 	ddi_prop_free(api_ver);
3283 
3284 	return (0);
3285 err:
3286 	nvlist_free(misc);
3287 	if (fw_ver != NULL)
3288 		ddi_prop_free(fw_ver);
3289 	if (fw_bld != NULL)
3290 		ddi_prop_free(fw_bld);
3291 	if (api_ver != NULL)
3292 		ddi_prop_free(api_ver);
3293 
3294 	return (err);
3295 }
3296 
3297 static int
3298 i40e_ufm_getcaps(ddi_ufm_handle_t *ufmh, void *arg, ddi_ufm_cap_t *caps)
3299 {
3300 	*caps = DDI_UFM_CAP_REPORT;
3301 
3302 	return (0);
3303 }
3304 
3305 static ddi_ufm_ops_t i40e_ufm_ops = {
3306 	NULL,
3307 	i40e_ufm_fill_image,
3308 	i40e_ufm_fill_slot,
3309 	i40e_ufm_getcaps
3310 };
3311 
3312 static int
3313 i40e_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
3314 {
3315 	i40e_t *i40e;
3316 	struct i40e_osdep *osdep;
3317 	i40e_hw_t *hw;
3318 	int instance;
3319 
3320 	if (cmd != DDI_ATTACH)
3321 		return (DDI_FAILURE);
3322 
3323 	instance = ddi_get_instance(devinfo);
3324 	i40e = kmem_zalloc(sizeof (i40e_t), KM_SLEEP);
3325 
3326 	i40e->i40e_aqbuf = kmem_zalloc(I40E_ADMINQ_BUFSZ, KM_SLEEP);
3327 	i40e->i40e_instance = instance;
3328 	i40e->i40e_dip = devinfo;
3329 
3330 	hw = &i40e->i40e_hw_space;
3331 	osdep = &i40e->i40e_osdep_space;
3332 	hw->back = osdep;
3333 	osdep->ios_i40e = i40e;
3334 
3335 	ddi_set_driver_private(devinfo, i40e);
3336 
3337 	i40e_fm_init(i40e);
3338 	i40e->i40e_attach_progress |= I40E_ATTACH_FM_INIT;
3339 
3340 	if (pci_config_setup(devinfo, &osdep->ios_cfg_handle) != DDI_SUCCESS) {
3341 		i40e_error(i40e, "Failed to map PCI configurations.");
3342 		goto attach_fail;
3343 	}
3344 	i40e->i40e_attach_progress |= I40E_ATTACH_PCI_CONFIG;
3345 
3346 	i40e_identify_hardware(i40e);
3347 
3348 	if (!i40e_regs_map(i40e)) {
3349 		i40e_error(i40e, "Failed to map device registers.");
3350 		goto attach_fail;
3351 	}
3352 	i40e->i40e_attach_progress |= I40E_ATTACH_REGS_MAP;
3353 
3354 	i40e_init_properties(i40e);
3355 	i40e->i40e_attach_progress |= I40E_ATTACH_PROPS;
3356 
3357 	if (!i40e_common_code_init(i40e, hw))
3358 		goto attach_fail;
3359 	i40e->i40e_attach_progress |= I40E_ATTACH_COMMON_CODE;
3360 
3361 	/*
3362 	 * When we participate in IRM, we should make sure that we register
3363 	 * ourselves with it before callbacks.
3364 	 */
3365 	if (!i40e_alloc_intrs(i40e, devinfo)) {
3366 		i40e_error(i40e, "Failed to allocate interrupts.");
3367 		goto attach_fail;
3368 	}
3369 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_INTR;
3370 
3371 	if (!i40e_alloc_trqpairs(i40e)) {
3372 		i40e_error(i40e,
3373 		    "Failed to allocate receive & transmit rings.");
3374 		goto attach_fail;
3375 	}
3376 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_RINGSLOCKS;
3377 
3378 	if (!i40e_map_intrs_to_vectors(i40e)) {
3379 		i40e_error(i40e, "Failed to map interrupts to vectors.");
3380 		goto attach_fail;
3381 	}
3382 
3383 	if (!i40e_add_intr_handlers(i40e)) {
3384 		i40e_error(i40e, "Failed to add the interrupt handlers.");
3385 		goto attach_fail;
3386 	}
3387 	i40e->i40e_attach_progress |= I40E_ATTACH_ADD_INTR;
3388 
3389 	if (!i40e_final_init(i40e)) {
3390 		i40e_error(i40e, "Final initialization failed.");
3391 		goto attach_fail;
3392 	}
3393 	i40e->i40e_attach_progress |= I40E_ATTACH_INIT;
3394 
3395 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
3396 	    DDI_FM_OK) {
3397 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
3398 		goto attach_fail;
3399 	}
3400 
3401 	if (!i40e_stats_init(i40e)) {
3402 		i40e_error(i40e, "Stats initialization failed.");
3403 		goto attach_fail;
3404 	}
3405 	i40e->i40e_attach_progress |= I40E_ATTACH_STATS;
3406 
3407 	if (!i40e_register_mac(i40e)) {
3408 		i40e_error(i40e, "Failed to register to MAC/GLDv3");
3409 		goto attach_fail;
3410 	}
3411 	i40e->i40e_attach_progress |= I40E_ATTACH_MAC;
3412 
3413 	i40e->i40e_periodic_id = ddi_periodic_add(i40e_timer, i40e,
3414 	    I40E_CYCLIC_PERIOD, DDI_IPL_0);
3415 	if (i40e->i40e_periodic_id == 0) {
3416 		i40e_error(i40e, "Failed to add the link-check timer");
3417 		goto attach_fail;
3418 	}
3419 	i40e->i40e_attach_progress |= I40E_ATTACH_LINK_TIMER;
3420 
3421 	if (!i40e_enable_interrupts(i40e)) {
3422 		i40e_error(i40e, "Failed to enable DDI interrupts");
3423 		goto attach_fail;
3424 	}
3425 	i40e->i40e_attach_progress |= I40E_ATTACH_ENABLE_INTR;
3426 
3427 	if (i40e->i40e_hw_space.bus.func == 0) {
3428 		if (ddi_ufm_init(i40e->i40e_dip, DDI_UFM_CURRENT_VERSION,
3429 		    &i40e_ufm_ops, &i40e->i40e_ufmh, i40e) != 0) {
3430 			i40e_error(i40e, "failed to initialize UFM subsystem");
3431 			goto attach_fail;
3432 		}
3433 		ddi_ufm_update(i40e->i40e_ufmh);
3434 		i40e->i40e_attach_progress |= I40E_ATTACH_UFM_INIT;
3435 	}
3436 
3437 	atomic_or_32(&i40e->i40e_state, I40E_INITIALIZED);
3438 
3439 	mutex_enter(&i40e_glock);
3440 	list_insert_tail(&i40e_glist, i40e);
3441 	mutex_exit(&i40e_glock);
3442 
3443 	return (DDI_SUCCESS);
3444 
3445 attach_fail:
3446 	i40e_unconfigure(devinfo, i40e);
3447 	return (DDI_FAILURE);
3448 }
3449 
3450 static int
3451 i40e_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
3452 {
3453 	i40e_t *i40e;
3454 
3455 	if (cmd != DDI_DETACH)
3456 		return (DDI_FAILURE);
3457 
3458 	i40e = (i40e_t *)ddi_get_driver_private(devinfo);
3459 	if (i40e == NULL) {
3460 		i40e_log(NULL, "i40e_detach() called with no i40e pointer!");
3461 		return (DDI_FAILURE);
3462 	}
3463 
3464 	if (i40e_drain_rx(i40e) == B_FALSE) {
3465 		i40e_log(i40e, "timed out draining DMA resources, %d buffers "
3466 		    "remain", i40e->i40e_rx_pending);
3467 		return (DDI_FAILURE);
3468 	}
3469 
3470 	mutex_enter(&i40e_glock);
3471 	list_remove(&i40e_glist, i40e);
3472 	mutex_exit(&i40e_glock);
3473 
3474 	i40e_unconfigure(devinfo, i40e);
3475 
3476 	return (DDI_SUCCESS);
3477 }
3478 
3479 static struct cb_ops i40e_cb_ops = {
3480 	nulldev,		/* cb_open */
3481 	nulldev,		/* cb_close */
3482 	nodev,			/* cb_strategy */
3483 	nodev,			/* cb_print */
3484 	nodev,			/* cb_dump */
3485 	nodev,			/* cb_read */
3486 	nodev,			/* cb_write */
3487 	nodev,			/* cb_ioctl */
3488 	nodev,			/* cb_devmap */
3489 	nodev,			/* cb_mmap */
3490 	nodev,			/* cb_segmap */
3491 	nochpoll,		/* cb_chpoll */
3492 	ddi_prop_op,		/* cb_prop_op */
3493 	NULL,			/* cb_stream */
3494 	D_MP | D_HOTPLUG,	/* cb_flag */
3495 	CB_REV,			/* cb_rev */
3496 	nodev,			/* cb_aread */
3497 	nodev			/* cb_awrite */
3498 };
3499 
3500 static struct dev_ops i40e_dev_ops = {
3501 	DEVO_REV,		/* devo_rev */
3502 	0,			/* devo_refcnt */
3503 	NULL,			/* devo_getinfo */
3504 	nulldev,		/* devo_identify */
3505 	nulldev,		/* devo_probe */
3506 	i40e_attach,		/* devo_attach */
3507 	i40e_detach,		/* devo_detach */
3508 	nodev,			/* devo_reset */
3509 	&i40e_cb_ops,		/* devo_cb_ops */
3510 	NULL,			/* devo_bus_ops */
3511 	nulldev,		/* devo_power */
3512 	ddi_quiesce_not_supported /* devo_quiesce */
3513 };
3514 
3515 static struct modldrv i40e_modldrv = {
3516 	&mod_driverops,
3517 	i40e_ident,
3518 	&i40e_dev_ops
3519 };
3520 
3521 static struct modlinkage i40e_modlinkage = {
3522 	MODREV_1,
3523 	&i40e_modldrv,
3524 	NULL
3525 };
3526 
3527 /*
3528  * Module Initialization Functions.
3529  */
3530 int
3531 _init(void)
3532 {
3533 	int status;
3534 
3535 	list_create(&i40e_glist, sizeof (i40e_t), offsetof(i40e_t, i40e_glink));
3536 	list_create(&i40e_dlist, sizeof (i40e_device_t),
3537 	    offsetof(i40e_device_t, id_link));
3538 	mutex_init(&i40e_glock, NULL, MUTEX_DRIVER, NULL);
3539 	mac_init_ops(&i40e_dev_ops, I40E_MODULE_NAME);
3540 
3541 	status = mod_install(&i40e_modlinkage);
3542 	if (status != DDI_SUCCESS) {
3543 		mac_fini_ops(&i40e_dev_ops);
3544 		mutex_destroy(&i40e_glock);
3545 		list_destroy(&i40e_dlist);
3546 		list_destroy(&i40e_glist);
3547 	}
3548 
3549 	return (status);
3550 }
3551 
3552 int
3553 _info(struct modinfo *modinfop)
3554 {
3555 	return (mod_info(&i40e_modlinkage, modinfop));
3556 }
3557 
3558 int
3559 _fini(void)
3560 {
3561 	int status;
3562 
3563 	status = mod_remove(&i40e_modlinkage);
3564 	if (status == DDI_SUCCESS) {
3565 		mac_fini_ops(&i40e_dev_ops);
3566 		mutex_destroy(&i40e_glock);
3567 		list_destroy(&i40e_dlist);
3568 		list_destroy(&i40e_glist);
3569 	}
3570 
3571 	return (status);
3572 }
3573