xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_main.c (revision c160bf3613805cfb4a89a0433ae896d3594f551f)
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 2016 Joyent, Inc.
15  * Copyright 2017 Tegile Systems, Inc.  All rights reserved.
16  */
17 
18 /*
19  * i40e - Intel 10/40 Gb Ethernet driver
20  *
21  * The i40e driver is the main software device driver for the Intel 40 Gb family
22  * of devices. Note that these devices come in many flavors with both 40 GbE
23  * ports and 10 GbE ports. This device is the successor to the 82599 family of
24  * devices (ixgbe).
25  *
26  * Unlike previous generations of Intel 1 GbE and 10 GbE devices, the 40 GbE
27  * devices defined in the XL710 controller (previously known as Fortville) are a
28  * rather different beast and have a small switch embedded inside of them. In
29  * addition, the way that most of the programming is done has been overhauled.
30  * As opposed to just using PCIe memory mapped registers, it also has an
31  * administrative queue which is used to communicate with firmware running on
32  * the chip.
33  *
34  * Each physical function in the hardware shows up as a device that this driver
35  * will bind to. The hardware splits many resources evenly across all of the
36  * physical functions present on the device, while other resources are instead
37  * shared across the entire card and its up to the device driver to
38  * intelligently partition them.
39  *
40  * ------------
41  * Organization
42  * ------------
43  *
44  * This driver is made up of several files which have their own theory
45  * statements spread across them. We'll touch on the high level purpose of each
46  * file here, and then we'll get into more discussion on how the device is
47  * generally modelled with respect to the interfaces in illumos.
48  *
49  * i40e_gld.c: This file contains all of the bindings to MAC and the networking
50  *             stack.
51  *
52  * i40e_intr.c: This file contains all of the interrupt service routines and
53  *              contains logic to enable and disable interrupts on the hardware.
54  *              It also contains the logic to map hardware resources such as the
55  *              rings to and from interrupts and controls their ability to fire.
56  *
57  *              There is a big theory statement on interrupts present there.
58  *
59  * i40e_main.c: The file that you're currently in. It interfaces with the
60  *              traditional OS DDI interfaces and is in charge of configuring
61  *              the device.
62  *
63  * i40e_osdep.[ch]: These files contain interfaces and definitions needed to
64  *                  work with Intel's common code for the device.
65  *
66  * i40e_stats.c: This file contains the general work and logic around our
67  *               kstats. A theory statement on their organization and use of the
68  *               hardware exists there.
69  *
70  * i40e_sw.h: This header file contains all of the primary structure definitions
71  *            and constants that are used across the entire driver.
72  *
73  * i40e_transceiver.c: This file contains all of the logic for sending and
74  *                     receiving data. It contains all of the ring and DMA
75  *                     allocation logic, as well as, the actual interfaces to
76  *                     send and receive data.
77  *
78  *                     A big theory statement on ring management, descriptors,
79  *                     and how it ties into the OS is present there.
80  *
81  * --------------
82  * General Design
83  * --------------
84  *
85  * Before we go too far into the general way we've laid out data structures and
86  * the like, it's worth taking some time to explain how the hardware is
87  * organized. This organization informs a lot of how we do things at this time
88  * in the driver.
89  *
90  * Each physical device consists of a number of one or more ports, which are
91  * considered physical functions in the PCI sense and thus each get enumerated
92  * by the system, resulting in an instance being created and attached to. While
93  * there are many resources that are unique to each physical function eg.
94  * instance of the device, there are many that are shared across all of them.
95  * Several resources have an amount reserved for each Virtual Station Interface
96  * (VSI) and then a static pool of resources, available for all functions on the
97  * card.
98  *
99  * The most important resource in hardware are its transmit and receive queue
100  * pairs (i40e_trqpair_t). These should be thought of as rings in GLDv3
101  * parlance. There are a set number of these on each device; however, they are
102  * statically partitioned among all of the different physical functions.
103  *
104  * 'Fortville' (the code name for this device family) is basically a switch. To
105  * map MAC addresses and other things to queues, we end up having to create
106  * Virtual Station Interfaces (VSIs) and establish forwarding rules that direct
107  * traffic to a queue. A VSI owns a collection of queues and has a series of
108  * forwarding rules that point to it. One way to think of this is to treat it
109  * like MAC does a VNIC. When MAC refers to a group, a collection of rings and
110  * classification resources, that is a VSI in i40e.
111  *
112  * The sets of VSIs is shared across the entire device, though there may be some
113  * amount that are reserved to each PF. Because the GLDv3 does not let us change
114  * the number of groups dynamically, we instead statically divide this amount
115  * evenly between all the functions that exist. In addition, we have the same
116  * problem with the mac address forwarding rules. There are a static number that
117  * exist shared across all the functions.
118  *
119  * To handle both of these resources, what we end up doing is going through and
120  * determining which functions belong to the same device. Nominally one might do
121  * this by having a nexus driver; however, a prime requirement for a nexus
122  * driver is identifying the various children and activating them. While it is
123  * possible to get this information from NVRAM, we would end up duplicating a
124  * lot of the PCI enumeration logic. Really, at the end of the day, the device
125  * doesn't give us the traditional identification properties we want from a
126  * nexus driver.
127  *
128  * Instead, we rely on some properties that are guaranteed to be unique. While
129  * it might be tempting to leverage the PBA or serial number of the device from
130  * NVRAM, there is nothing that says that two devices can't be mis-programmed to
131  * have the same values in NVRAM. Instead, we uniquely identify a group of
132  * functions based on their parent in the /devices tree, their PCI bus and PCI
133  * function identifiers. Using either on their own may not be sufficient.
134  *
135  * For each unique PCI device that we encounter, we'll create a i40e_device_t.
136  * From there, because we don't have a good way to tell the GLDv3 about sharing
137  * resources between everything, we'll end up just dividing the resources
138  * evenly between all of the functions. Longer term, if we don't have to declare
139  * to the GLDv3 that these resources are shared, then we'll maintain a pool and
140  * hae each PF allocate from the pool in the device, thus if only two of four
141  * ports are being used, for example, then all of the resources can still be
142  * used.
143  *
144  * -------------------------------------------
145  * Transmit and Receive Queue Pair Allocations
146  * -------------------------------------------
147  *
148  * NVRAM ends up assigning each PF its own share of the transmit and receive LAN
149  * queue pairs, we have no way of modifying it, only observing it. From there,
150  * it's up to us to map these queues to VSIs and VFs. Since we don't support any
151  * VFs at this time, we only focus on assignments to VSIs.
152  *
153  * At the moment, we used a static mapping of transmit/receive queue pairs to a
154  * given VSI (eg. rings to a group). Though in the fullness of time, we want to
155  * make this something which is fully dynamic and take advantage of documented,
156  * but not yet available functionality for adding filters based on VXLAN and
157  * other encapsulation technologies.
158  *
159  * -------------------------------------
160  * Broadcast, Multicast, and Promiscuous
161  * -------------------------------------
162  *
163  * As part of the GLDv3, we need to make sure that we can handle receiving
164  * broadcast and multicast traffic. As well as enabling promiscuous mode when
165  * requested. GLDv3 requires that all broadcast and multicast traffic be
166  * retrieved by the default group, eg. the first one. This is the same thing as
167  * the default VSI.
168  *
169  * To receieve broadcast traffic, we enable it through the admin queue, rather
170  * than use one of our filters for it. For multicast traffic, we reserve a
171  * certain number of the hash filters and assign them to a given PF. When we
172  * exceed those, we then switch to using promicuous mode for multicast traffic.
173  *
174  * More specifically, once we exceed the number of filters (indicated because
175  * the i40e_t`i40e_resources.ifr_nmcastfilt ==
176  * i40e_t`i40e_resources.ifr_nmcastfilt_used), we then instead need to toggle
177  * promiscuous mode. If promiscuous mode is toggled then we keep track of the
178  * number of MACs added to it by incrementing i40e_t`i40e_mcast_promisc_count.
179  * That will stay enabled until that count reaches zero indicating that we have
180  * only added multicast addresses that we have a corresponding entry for.
181  *
182  * Because MAC itself wants to toggle promiscuous mode, which includes both
183  * unicast and multicast traffic, we go through and keep track of that
184  * ourselves. That is maintained through the use of the i40e_t`i40e_promisc_on
185  * member.
186  *
187  * --------------
188  * VSI Management
189  * --------------
190  *
191  * At this time, we currently only support a single MAC group, and thus a single
192  * VSI. This VSI is considered the default VSI and should be the only one that
193  * exists after a reset. Currently it is stored as the member
194  * i40e_t`i40e_vsi_id. While this works for the moment and for an initial
195  * driver, it's not sufficient for the longer-term path of the driver. Instead,
196  * we'll want to actually have a unique i40e_vsi_t structure which is used
197  * everywhere. Note that this means that every place that uses the
198  * i40e_t`i40e_vsi_id will need to be refactored.
199  *
200  * ----------------
201  * Structure Layout
202  * ----------------
203  *
204  * The following images relates the core data structures together. The primary
205  * structure in the system is the i40e_t. It itself contains multiple rings,
206  * i40e_trqpair_t's which contain the various transmit and receive data. The
207  * receive data is stored outside of the i40e_trqpair_t and instead in the
208  * i40e_rx_data_t. The i40e_t has a corresponding i40e_device_t which keeps
209  * track of per-physical device state. Finally, for every active descriptor,
210  * there is a corresponding control block, which is where the
211  * i40e_rx_control_block_t and the i40e_tx_control_block_t come from.
212  *
213  *   +-----------------------+       +-----------------------+
214  *   | Global i40e_t list    |       | Global Device list    |
215  *   |                       |    +--|                       |
216  *   | i40e_glist            |    |  | i40e_dlist            |
217  *   +-----------------------+    |  +-----------------------+
218  *       |                        v
219  *       |      +------------------------+      +-----------------------+
220  *       |      | Device-wide Structure  |----->| Device-wide Structure |--> ...
221  *       |      | i40e_device_t          |      | i40e_device_t         |
222  *       |      |                        |      +-----------------------+
223  *       |      | dev_info_t *     ------+--> Parent in devices tree.
224  *       |      | uint_t           ------+--> PCI bus number
225  *       |      | uint_t           ------+--> PCI device number
226  *       |      | uint_t           ------+--> Number of functions
227  *       |      | i40e_switch_rsrcs_t ---+--> Captured total switch resources
228  *       |      | list_t           ------+-------------+
229  *       |      +------------------------+             |
230  *       |                           ^                 |
231  *       |                           +--------+        |
232  *       |                                    |        v
233  *       |  +---------------------------+     |   +-------------------+
234  *       +->| GLDv3 Device, per PF      |-----|-->| GLDv3 Device (PF) |--> ...
235  *          | i40e_t                    |     |   | i40e_t            |
236  *          | **Primary Structure**     |     |   +-------------------+
237  *          |                           |     |
238  *          | i40e_device_t *         --+-----+
239  *          | i40e_state_t            --+---> Device State
240  *          | i40e_hw_t               --+---> Intel common code structure
241  *          | mac_handle_t            --+---> GLDv3 handle to MAC
242  *          | ddi_periodic_t          --+---> Link activity timer
243  *          | int (vsi_id)            --+---> VSI ID, main identifier
244  *          | i40e_func_rsrc_t        --+---> Available hardware resources
245  *          | i40e_switch_rsrc_t *    --+---> Switch resource snapshot
246  *          | i40e_sdu                --+---> Current MTU
247  *          | i40e_frame_max          --+---> Current HW frame size
248  *          | i40e_uaddr_t *          --+---> Array of assigned unicast MACs
249  *          | i40e_maddr_t *          --+---> Array of assigned multicast MACs
250  *          | i40e_mcast_promisccount --+---> Active multicast state
251  *          | i40e_promisc_on         --+---> Current promiscuous mode state
252  *          | int                     --+---> Number of transmit/receive pairs
253  *          | kstat_t *               --+---> PF kstats
254  *          | kstat_t *               --+---> VSI kstats
255  *          | i40e_pf_stats_t         --+---> PF kstat backing data
256  *          | i40e_vsi_stats_t        --+---> VSI kstat backing data
257  *          | i40e_trqpair_t *        --+---------+
258  *          +---------------------------+         |
259  *                                                |
260  *                                                v
261  *  +-------------------------------+       +-----------------------------+
262  *  | Transmit/Receive Queue Pair   |-------| Transmit/Receive Queue Pair |->...
263  *  | i40e_trqpair_t                |       | i40e_trqpair_t              |
264  *  + Ring Data Structure           |       +-----------------------------+
265  *  |                               |
266  *  | mac_ring_handle_t             +--> MAC RX ring handle
267  *  | mac_ring_handle_t             +--> MAC TX ring handle
268  *  | i40e_rxq_stat_t             --+--> RX Queue stats
269  *  | i40e_txq_stat_t             --+--> TX Queue stats
270  *  | uint32_t (tx ring size)       +--> TX Ring Size
271  *  | uint32_t (tx free list size)  +--> TX Free List Size
272  *  | i40e_dma_buffer_t     --------+--> TX Descriptor ring DMA
273  *  | i40e_tx_desc_t *      --------+--> TX descriptor ring
274  *  | volatile unt32_t *            +--> TX Write back head
275  *  | uint32_t               -------+--> TX ring head
276  *  | uint32_t               -------+--> TX ring tail
277  *  | uint32_t               -------+--> Num TX desc free
278  *  | i40e_tx_control_block_t *   --+--> TX control block array  ---+
279  *  | i40e_tx_control_block_t **  --+--> TCB work list          ----+
280  *  | i40e_tx_control_block_t **  --+--> TCB free list           ---+
281  *  | uint32_t               -------+--> Free TCB count             |
282  *  | i40e_rx_data_t *       -------+--+                            v
283  *  +-------------------------------+  |          +---------------------------+
284  *                                     |          | Per-TX Frame Metadata     |
285  *                                     |          | i40e_tx_control_block_t   |
286  *                +--------------------+          |                           |
287  *                |           mblk to transmit <--+---      mblk_t *          |
288  *                |           type of transmit <--+---      i40e_tx_type_t    |
289  *                |              TX DMA handle <--+---      ddi_dma_handle_t  |
290  *                v              TX DMA buffer <--+---      i40e_dma_buffer_t |
291  *    +------------------------------+            +---------------------------+
292  *    | Core Receive Data            |
293  *    | i40e_rx_data_t               |
294  *    |                              |
295  *    | i40e_dma_buffer_t          --+--> RX descriptor DMA Data
296  *    | i40e_rx_desc_t             --+--> RX descriptor ring
297  *    | uint32_t                   --+--> Next free desc.
298  *    | i40e_rx_control_block_t *  --+--> RX Control Block Array  ---+
299  *    | i40e_rx_control_block_t ** --+--> RCB work list           ---+
300  *    | i40e_rx_control_block_t ** --+--> RCB free list           ---+
301  *    +------------------------------+                               |
302  *                ^                                                  |
303  *                |     +---------------------------+                |
304  *                |     | Per-RX Frame Metadata     |<---------------+
305  *                |     | i40e_rx_control_block_t   |
306  *                |     |                           |
307  *                |     | mblk_t *              ----+--> Received mblk_t data
308  *                |     | uint32_t              ----+--> Reference count
309  *                |     | i40e_dma_buffer_t     ----+--> Receive data DMA info
310  *                |     | frtn_t                ----+--> mblk free function info
311  *                +-----+-- i40e_rx_data_t *        |
312  *                      +---------------------------+
313  *
314  * -------------
315  * Lock Ordering
316  * -------------
317  *
318  * In order to ensure that we don't deadlock, the following represents the
319  * lock order being used. When grabbing locks, follow the following order. Lower
320  * numbers are more important. Thus, the i40e_glock which is number 0, must be
321  * taken before any other locks in the driver. On the other hand, the
322  * i40e_t`i40e_stat_lock, has the highest number because it's the least
323  * important lock. Note, that just because one lock is higher than another does
324  * not mean that all intermediary locks are required.
325  *
326  * 0) i40e_glock
327  * 1) i40e_t`i40e_general_lock
328  *
329  * 2) i40e_trqpair_t`itrq_rx_lock
330  * 3) i40e_trqpair_t`itrq_tx_lock
331  * 4) i40e_t`i40e_rx_pending_lock
332  * 5) i40e_trqpair_t`itrq_tcb_lock
333  *
334  * 6) i40e_t`i40e_stat_lock
335  *
336  * Rules and expectations:
337  *
338  * 1) A thread holding locks belong to one PF should not hold locks belonging to
339  * a second. If for some reason this becomes necessary, locks should be grabbed
340  * based on the list order in the i40e_device_t, which implies that the
341  * i40e_glock is held.
342  *
343  * 2) When grabbing locks between multiple transmit and receive queues, the
344  * locks for the lowest number transmit/receive queue should be grabbed first.
345  *
346  * 3) When grabbing both the transmit and receive lock for a given queue, always
347  * grab i40e_trqpair_t`itrq_rx_lock before the i40e_trqpair_t`itrq_tx_lock.
348  *
349  * 4) The following pairs of locks are not expected to be held at the same time:
350  *
351  * o i40e_t`i40e_rx_pending_lock and i40e_trqpair_t`itrq_tcb_lock
352  *
353  * -----------
354  * Future Work
355  * -----------
356  *
357  * At the moment the i40e_t driver is rather bare bones, allowing us to start
358  * getting data flowing and folks using it while we develop additional features.
359  * While bugs have been filed to cover this future work, the following gives an
360  * overview of expected work:
361  *
362  *  o TSO support
363  *  o Multiple group support
364  *  o DMA binding and breaking up the locking in ring recycling.
365  *  o Enhanced detection of device errors
366  *  o Participation in IRM
367  *  o FMA device reset
368  *  o Stall detection, temperature error detection, etc.
369  *  o More dynamic resource pools
370  */
371 
372 #include "i40e_sw.h"
373 
374 static char i40e_ident[] = "Intel 10/40Gb Ethernet v1.0.1";
375 
376 /*
377  * The i40e_glock primarily protects the lists below and the i40e_device_t
378  * structures.
379  */
380 static kmutex_t i40e_glock;
381 static list_t i40e_glist;
382 static list_t i40e_dlist;
383 
384 /*
385  * Access attributes for register mapping.
386  */
387 static ddi_device_acc_attr_t i40e_regs_acc_attr = {
388 	DDI_DEVICE_ATTR_V1,
389 	DDI_STRUCTURE_LE_ACC,
390 	DDI_STRICTORDER_ACC,
391 	DDI_FLAGERR_ACC
392 };
393 
394 /*
395  * Logging function for this driver.
396  */
397 static void
398 i40e_dev_err(i40e_t *i40e, int level, boolean_t console, const char *fmt,
399     va_list ap)
400 {
401 	char buf[1024];
402 
403 	(void) vsnprintf(buf, sizeof (buf), fmt, ap);
404 
405 	if (i40e == NULL) {
406 		cmn_err(level, (console) ? "%s: %s" : "!%s: %s",
407 		    I40E_MODULE_NAME, buf);
408 	} else {
409 		dev_err(i40e->i40e_dip, level, (console) ? "%s" : "!%s",
410 		    buf);
411 	}
412 }
413 
414 /*
415  * Because there's the stupid trailing-comma problem with the C preprocessor
416  * and variable arguments, I need to instantiate these.	 Pardon the redundant
417  * code.
418  */
419 /*PRINTFLIKE2*/
420 void
421 i40e_error(i40e_t *i40e, const char *fmt, ...)
422 {
423 	va_list ap;
424 
425 	va_start(ap, fmt);
426 	i40e_dev_err(i40e, CE_WARN, B_FALSE, fmt, ap);
427 	va_end(ap);
428 }
429 
430 /*PRINTFLIKE2*/
431 void
432 i40e_log(i40e_t *i40e, const char *fmt, ...)
433 {
434 	va_list ap;
435 
436 	va_start(ap, fmt);
437 	i40e_dev_err(i40e, CE_NOTE, B_FALSE, fmt, ap);
438 	va_end(ap);
439 }
440 
441 /*PRINTFLIKE2*/
442 void
443 i40e_notice(i40e_t *i40e, const char *fmt, ...)
444 {
445 	va_list ap;
446 
447 	va_start(ap, fmt);
448 	i40e_dev_err(i40e, CE_NOTE, B_TRUE, fmt, ap);
449 	va_end(ap);
450 }
451 
452 static void
453 i40e_device_rele(i40e_t *i40e)
454 {
455 	i40e_device_t *idp = i40e->i40e_device;
456 
457 	if (idp == NULL)
458 		return;
459 
460 	mutex_enter(&i40e_glock);
461 	VERIFY(idp->id_nreg > 0);
462 	list_remove(&idp->id_i40e_list, i40e);
463 	idp->id_nreg--;
464 	if (idp->id_nreg == 0) {
465 		list_remove(&i40e_dlist, idp);
466 		list_destroy(&idp->id_i40e_list);
467 		kmem_free(idp->id_rsrcs, sizeof (i40e_switch_rsrc_t) *
468 		    idp->id_rsrcs_alloc);
469 		kmem_free(idp, sizeof (i40e_device_t));
470 	}
471 	i40e->i40e_device = NULL;
472 	mutex_exit(&i40e_glock);
473 }
474 
475 static i40e_device_t *
476 i40e_device_find(i40e_t *i40e, dev_info_t *parent, uint_t bus, uint_t device)
477 {
478 	i40e_device_t *idp;
479 	mutex_enter(&i40e_glock);
480 	for (idp = list_head(&i40e_dlist); idp != NULL;
481 	    idp = list_next(&i40e_dlist, idp)) {
482 		if (idp->id_parent == parent && idp->id_pci_bus == bus &&
483 		    idp->id_pci_device == device) {
484 			break;
485 		}
486 	}
487 
488 	if (idp != NULL) {
489 		VERIFY(idp->id_nreg < idp->id_nfuncs);
490 		idp->id_nreg++;
491 	} else {
492 		i40e_hw_t *hw = &i40e->i40e_hw_space;
493 		ASSERT(hw->num_ports > 0);
494 		ASSERT(hw->num_partitions > 0);
495 
496 		/*
497 		 * The Intel common code doesn't exactly keep the number of PCI
498 		 * functions. But it calculates it during discovery of
499 		 * partitions and ports. So what we do is undo the calculation
500 		 * that it does originally, as functions are evenly spread
501 		 * across ports in the rare case of partitions.
502 		 */
503 		idp = kmem_alloc(sizeof (i40e_device_t), KM_SLEEP);
504 		idp->id_parent = parent;
505 		idp->id_pci_bus = bus;
506 		idp->id_pci_device = device;
507 		idp->id_nfuncs = hw->num_ports * hw->num_partitions;
508 		idp->id_nreg = 1;
509 		idp->id_rsrcs_alloc = i40e->i40e_switch_rsrc_alloc;
510 		idp->id_rsrcs_act = i40e->i40e_switch_rsrc_actual;
511 		idp->id_rsrcs = kmem_alloc(sizeof (i40e_switch_rsrc_t) *
512 		    idp->id_rsrcs_alloc, KM_SLEEP);
513 		bcopy(i40e->i40e_switch_rsrcs, idp->id_rsrcs,
514 		    sizeof (i40e_switch_rsrc_t) * idp->id_rsrcs_alloc);
515 		list_create(&idp->id_i40e_list, sizeof (i40e_t),
516 		    offsetof(i40e_t, i40e_dlink));
517 
518 		list_insert_tail(&i40e_dlist, idp);
519 	}
520 
521 	list_insert_tail(&idp->id_i40e_list, i40e);
522 	mutex_exit(&i40e_glock);
523 
524 	return (idp);
525 }
526 
527 static void
528 i40e_link_state_set(i40e_t *i40e, link_state_t state)
529 {
530 	if (i40e->i40e_link_state == state)
531 		return;
532 
533 	i40e->i40e_link_state = state;
534 	mac_link_update(i40e->i40e_mac_hdl, i40e->i40e_link_state);
535 }
536 
537 /*
538  * This is a basic link check routine. Mostly we're using this just to see
539  * if we can get any accurate information about the state of the link being
540  * up or down, as well as updating the link state, speed, etc. information.
541  */
542 void
543 i40e_link_check(i40e_t *i40e)
544 {
545 	i40e_hw_t *hw = &i40e->i40e_hw_space;
546 	boolean_t ls;
547 	int ret;
548 
549 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
550 
551 	hw->phy.get_link_info = B_TRUE;
552 	if ((ret = i40e_get_link_status(hw, &ls)) != I40E_SUCCESS) {
553 		i40e->i40e_s_link_status_errs++;
554 		i40e->i40e_s_link_status_lasterr = ret;
555 		return;
556 	}
557 
558 	/*
559 	 * Firmware abstracts all of the mac and phy information for us, so we
560 	 * can use i40e_get_link_status to determine the current state.
561 	 */
562 	if (ls == B_TRUE) {
563 		enum i40e_aq_link_speed speed;
564 
565 		speed = i40e_get_link_speed(hw);
566 
567 		/*
568 		 * Translate from an i40e value to a value in Mbits/s.
569 		 */
570 		switch (speed) {
571 		case I40E_LINK_SPEED_100MB:
572 			i40e->i40e_link_speed = 100;
573 			break;
574 		case I40E_LINK_SPEED_1GB:
575 			i40e->i40e_link_speed = 1000;
576 			break;
577 		case I40E_LINK_SPEED_10GB:
578 			i40e->i40e_link_speed = 10000;
579 			break;
580 		case I40E_LINK_SPEED_20GB:
581 			i40e->i40e_link_speed = 20000;
582 			break;
583 		case I40E_LINK_SPEED_40GB:
584 			i40e->i40e_link_speed = 40000;
585 			break;
586 		case I40E_LINK_SPEED_25GB:
587 			i40e->i40e_link_speed = 25000;
588 			break;
589 		default:
590 			i40e->i40e_link_speed = 0;
591 			break;
592 		}
593 
594 		/*
595 		 * At this time, hardware does not support half-duplex
596 		 * operation, hence why we don't ask the hardware about our
597 		 * current speed.
598 		 */
599 		i40e->i40e_link_duplex = LINK_DUPLEX_FULL;
600 		i40e_link_state_set(i40e, LINK_STATE_UP);
601 	} else {
602 		i40e->i40e_link_speed = 0;
603 		i40e->i40e_link_duplex = 0;
604 		i40e_link_state_set(i40e, LINK_STATE_DOWN);
605 	}
606 }
607 
608 static void
609 i40e_rem_intrs(i40e_t *i40e)
610 {
611 	int i, rc;
612 
613 	for (i = 0; i < i40e->i40e_intr_count; i++) {
614 		rc = ddi_intr_free(i40e->i40e_intr_handles[i]);
615 		if (rc != DDI_SUCCESS) {
616 			i40e_log(i40e, "failed to free interrupt %d: %d",
617 			    i, rc);
618 		}
619 	}
620 
621 	kmem_free(i40e->i40e_intr_handles, i40e->i40e_intr_size);
622 	i40e->i40e_intr_handles = NULL;
623 }
624 
625 static void
626 i40e_rem_intr_handlers(i40e_t *i40e)
627 {
628 	int i, rc;
629 
630 	for (i = 0; i < i40e->i40e_intr_count; i++) {
631 		rc = ddi_intr_remove_handler(i40e->i40e_intr_handles[i]);
632 		if (rc != DDI_SUCCESS) {
633 			i40e_log(i40e, "failed to remove interrupt %d: %d",
634 			    i, rc);
635 		}
636 	}
637 }
638 
639 /*
640  * illumos Fault Management Architecture (FMA) support.
641  */
642 
643 int
644 i40e_check_acc_handle(ddi_acc_handle_t handle)
645 {
646 	ddi_fm_error_t de;
647 
648 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
649 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
650 	return (de.fme_status);
651 }
652 
653 int
654 i40e_check_dma_handle(ddi_dma_handle_t handle)
655 {
656 	ddi_fm_error_t de;
657 
658 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
659 	return (de.fme_status);
660 }
661 
662 /*
663  * Fault service error handling callback function.
664  */
665 /* ARGSUSED */
666 static int
667 i40e_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
668 {
669 	pci_ereport_post(dip, err, NULL);
670 	return (err->fme_status);
671 }
672 
673 static void
674 i40e_fm_init(i40e_t *i40e)
675 {
676 	ddi_iblock_cookie_t iblk;
677 
678 	i40e->i40e_fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY,
679 	    i40e->i40e_dip, DDI_PROP_DONTPASS, "fm_capable",
680 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
681 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
682 
683 	if (i40e->i40e_fm_capabilities < 0) {
684 		i40e->i40e_fm_capabilities = 0;
685 	} else if (i40e->i40e_fm_capabilities > 0xf) {
686 		i40e->i40e_fm_capabilities = DDI_FM_EREPORT_CAPABLE |
687 		    DDI_FM_ACCCHK_CAPABLE | DDI_FM_DMACHK_CAPABLE |
688 		    DDI_FM_ERRCB_CAPABLE;
689 	}
690 
691 	/*
692 	 * Only register with IO Fault Services if we have some capability
693 	 */
694 	if (i40e->i40e_fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
695 		i40e_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
696 	} else {
697 		i40e_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
698 	}
699 
700 	if (i40e->i40e_fm_capabilities) {
701 		ddi_fm_init(i40e->i40e_dip, &i40e->i40e_fm_capabilities, &iblk);
702 
703 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
704 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
705 			pci_ereport_setup(i40e->i40e_dip);
706 		}
707 
708 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
709 			ddi_fm_handler_register(i40e->i40e_dip,
710 			    i40e_fm_error_cb, (void*)i40e);
711 		}
712 	}
713 
714 	if (i40e->i40e_fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
715 		i40e_init_dma_attrs(i40e, B_TRUE);
716 	} else {
717 		i40e_init_dma_attrs(i40e, B_FALSE);
718 	}
719 }
720 
721 static void
722 i40e_fm_fini(i40e_t *i40e)
723 {
724 	if (i40e->i40e_fm_capabilities) {
725 
726 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
727 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
728 			pci_ereport_teardown(i40e->i40e_dip);
729 
730 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
731 			ddi_fm_handler_unregister(i40e->i40e_dip);
732 
733 		ddi_fm_fini(i40e->i40e_dip);
734 	}
735 }
736 
737 void
738 i40e_fm_ereport(i40e_t *i40e, char *detail)
739 {
740 	uint64_t ena;
741 	char buf[FM_MAX_CLASS];
742 
743 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
744 	ena = fm_ena_generate(0, FM_ENA_FMT1);
745 	if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities)) {
746 		ddi_fm_ereport_post(i40e->i40e_dip, buf, ena, DDI_NOSLEEP,
747 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
748 	}
749 }
750 
751 /*
752  * Here we're trying to get the ID of the default VSI. In general, when we come
753  * through and look at this shortly after attach, we expect there to only be a
754  * single element present, which is the default VSI. Importantly, each PF seems
755  * to not see any other devices, in part because of the simple switch mode that
756  * we're using. If for some reason, we see more artifact, we'll need to revisit
757  * what we're doing here.
758  */
759 static int
760 i40e_get_vsi_id(i40e_t *i40e)
761 {
762 	i40e_hw_t *hw = &i40e->i40e_hw_space;
763 	struct i40e_aqc_get_switch_config_resp *sw_config;
764 	uint8_t aq_buf[I40E_AQ_LARGE_BUF];
765 	uint16_t next = 0;
766 	int rc;
767 
768 	/* LINTED: E_BAD_PTR_CAST_ALIGN */
769 	sw_config = (struct i40e_aqc_get_switch_config_resp *)aq_buf;
770 	rc = i40e_aq_get_switch_config(hw, sw_config, sizeof (aq_buf), &next,
771 	    NULL);
772 	if (rc != I40E_SUCCESS) {
773 		i40e_error(i40e, "i40e_aq_get_switch_config() failed %d: %d",
774 		    rc, hw->aq.asq_last_status);
775 		return (-1);
776 	}
777 
778 	if (LE_16(sw_config->header.num_reported) != 1) {
779 		i40e_error(i40e, "encountered multiple (%d) switching units "
780 		    "during attach, not proceeding",
781 		    LE_16(sw_config->header.num_reported));
782 		return (-1);
783 	}
784 
785 	return (sw_config->element[0].seid);
786 }
787 
788 /*
789  * We need to fill the i40e_hw_t structure with the capabilities of this PF. We
790  * must also provide the memory for it; however, we don't need to keep it around
791  * to the call to the common code. It takes it and parses it into an internal
792  * structure.
793  */
794 static boolean_t
795 i40e_get_hw_capabilities(i40e_t *i40e, i40e_hw_t *hw)
796 {
797 	struct i40e_aqc_list_capabilities_element_resp *buf;
798 	int rc;
799 	size_t len;
800 	uint16_t needed;
801 	int nelems = I40E_HW_CAP_DEFAULT;
802 
803 	len = nelems * sizeof (*buf);
804 
805 	for (;;) {
806 		ASSERT(len > 0);
807 		buf = kmem_alloc(len, KM_SLEEP);
808 		rc = i40e_aq_discover_capabilities(hw, buf, len,
809 		    &needed, i40e_aqc_opc_list_func_capabilities, NULL);
810 		kmem_free(buf, len);
811 
812 		if (hw->aq.asq_last_status == I40E_AQ_RC_ENOMEM &&
813 		    nelems == I40E_HW_CAP_DEFAULT) {
814 			if (nelems == needed) {
815 				i40e_error(i40e, "Capability discovery failed "
816 				    "due to byzantine common code");
817 				return (B_FALSE);
818 			}
819 			len = needed;
820 			continue;
821 		} else if (rc != I40E_SUCCESS ||
822 		    hw->aq.asq_last_status != I40E_AQ_RC_OK) {
823 			i40e_error(i40e, "Capability discovery failed: %d", rc);
824 			return (B_FALSE);
825 		}
826 
827 		break;
828 	}
829 
830 	return (B_TRUE);
831 }
832 
833 /*
834  * Obtain the switch's capabilities as seen by this PF and keep it around for
835  * our later use.
836  */
837 static boolean_t
838 i40e_get_switch_resources(i40e_t *i40e)
839 {
840 	i40e_hw_t *hw = &i40e->i40e_hw_space;
841 	uint8_t cnt = 2;
842 	uint8_t act;
843 	size_t size;
844 	i40e_switch_rsrc_t *buf;
845 
846 	for (;;) {
847 		enum i40e_status_code ret;
848 		size = cnt * sizeof (i40e_switch_rsrc_t);
849 		ASSERT(size > 0);
850 		if (size > UINT16_MAX)
851 			return (B_FALSE);
852 		buf = kmem_alloc(size, KM_SLEEP);
853 
854 		ret = i40e_aq_get_switch_resource_alloc(hw, &act, buf,
855 		    cnt, NULL);
856 		if (ret == I40E_ERR_ADMIN_QUEUE_ERROR &&
857 		    hw->aq.asq_last_status == I40E_AQ_RC_EINVAL) {
858 			kmem_free(buf, size);
859 			cnt += I40E_SWITCH_CAP_DEFAULT;
860 			continue;
861 		} else if (ret != I40E_SUCCESS) {
862 			kmem_free(buf, size);
863 			i40e_error(i40e,
864 			    "failed to retrieve switch statistics: %d", ret);
865 			return (B_FALSE);
866 		}
867 
868 		break;
869 	}
870 
871 	i40e->i40e_switch_rsrc_alloc = cnt;
872 	i40e->i40e_switch_rsrc_actual = act;
873 	i40e->i40e_switch_rsrcs = buf;
874 
875 	return (B_TRUE);
876 }
877 
878 static void
879 i40e_cleanup_resources(i40e_t *i40e)
880 {
881 	if (i40e->i40e_uaddrs != NULL) {
882 		kmem_free(i40e->i40e_uaddrs, sizeof (i40e_uaddr_t) *
883 		    i40e->i40e_resources.ifr_nmacfilt);
884 		i40e->i40e_uaddrs = NULL;
885 	}
886 
887 	if (i40e->i40e_maddrs != NULL) {
888 		kmem_free(i40e->i40e_maddrs, sizeof (i40e_maddr_t) *
889 		    i40e->i40e_resources.ifr_nmcastfilt);
890 		i40e->i40e_maddrs = NULL;
891 	}
892 
893 	if (i40e->i40e_switch_rsrcs != NULL) {
894 		size_t sz = sizeof (i40e_switch_rsrc_t) *
895 		    i40e->i40e_switch_rsrc_alloc;
896 		ASSERT(sz > 0);
897 		kmem_free(i40e->i40e_switch_rsrcs, sz);
898 		i40e->i40e_switch_rsrcs = NULL;
899 	}
900 
901 	if (i40e->i40e_device != NULL)
902 		i40e_device_rele(i40e);
903 }
904 
905 static boolean_t
906 i40e_get_available_resources(i40e_t *i40e)
907 {
908 	dev_info_t *parent;
909 	uint16_t bus, device, func;
910 	uint_t nregs;
911 	int *regs, i;
912 	i40e_device_t *idp;
913 	i40e_hw_t *hw = &i40e->i40e_hw_space;
914 
915 	parent = ddi_get_parent(i40e->i40e_dip);
916 
917 	if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, i40e->i40e_dip, 0, "reg",
918 	    &regs, &nregs) != DDI_PROP_SUCCESS) {
919 		return (B_FALSE);
920 	}
921 
922 	if (nregs < 1) {
923 		ddi_prop_free(regs);
924 		return (B_FALSE);
925 	}
926 
927 	bus = PCI_REG_BUS_G(regs[0]);
928 	device = PCI_REG_DEV_G(regs[0]);
929 	func = PCI_REG_FUNC_G(regs[0]);
930 	ddi_prop_free(regs);
931 
932 	i40e->i40e_hw_space.bus.func = func;
933 	i40e->i40e_hw_space.bus.device = device;
934 
935 	if (i40e_get_switch_resources(i40e) == B_FALSE) {
936 		return (B_FALSE);
937 	}
938 
939 	/*
940 	 * To calculate the total amount of a resource we have available, we
941 	 * need to add how many our i40e_t thinks it has guaranteed, if any, and
942 	 * then we need to go through and divide the number of available on the
943 	 * device, which was snapshotted before anyone should have allocated
944 	 * anything, and use that to derive how many are available from the
945 	 * pool. Longer term, we may want to turn this into something that's
946 	 * more of a pool-like resource that everything can share (though that
947 	 * may require some more assistance from MAC).
948 	 *
949 	 * Though for transmit and receive queue pairs, we just have to ask
950 	 * firmware instead.
951 	 */
952 	idp = i40e_device_find(i40e, parent, bus, device);
953 	i40e->i40e_device = idp;
954 	i40e->i40e_resources.ifr_nvsis = 0;
955 	i40e->i40e_resources.ifr_nvsis_used = 0;
956 	i40e->i40e_resources.ifr_nmacfilt = 0;
957 	i40e->i40e_resources.ifr_nmacfilt_used = 0;
958 	i40e->i40e_resources.ifr_nmcastfilt = 0;
959 	i40e->i40e_resources.ifr_nmcastfilt_used = 0;
960 
961 	for (i = 0; i < i40e->i40e_switch_rsrc_actual; i++) {
962 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
963 
964 		switch (srp->resource_type) {
965 		case I40E_AQ_RESOURCE_TYPE_VSI:
966 			i40e->i40e_resources.ifr_nvsis +=
967 			    LE_16(srp->guaranteed);
968 			i40e->i40e_resources.ifr_nvsis_used = LE_16(srp->used);
969 			break;
970 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
971 			i40e->i40e_resources.ifr_nmacfilt +=
972 			    LE_16(srp->guaranteed);
973 			i40e->i40e_resources.ifr_nmacfilt_used =
974 			    LE_16(srp->used);
975 			break;
976 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
977 			i40e->i40e_resources.ifr_nmcastfilt +=
978 			    LE_16(srp->guaranteed);
979 			i40e->i40e_resources.ifr_nmcastfilt_used =
980 			    LE_16(srp->used);
981 			break;
982 		default:
983 			break;
984 		}
985 	}
986 
987 	for (i = 0; i < idp->id_rsrcs_act; i++) {
988 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
989 		switch (srp->resource_type) {
990 		case I40E_AQ_RESOURCE_TYPE_VSI:
991 			i40e->i40e_resources.ifr_nvsis +=
992 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
993 			break;
994 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
995 			i40e->i40e_resources.ifr_nmacfilt +=
996 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
997 			break;
998 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
999 			i40e->i40e_resources.ifr_nmcastfilt +=
1000 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1001 		default:
1002 			break;
1003 		}
1004 	}
1005 
1006 	i40e->i40e_resources.ifr_nrx_queue = hw->func_caps.num_rx_qp;
1007 	i40e->i40e_resources.ifr_ntx_queue = hw->func_caps.num_tx_qp;
1008 
1009 	i40e->i40e_uaddrs = kmem_zalloc(sizeof (i40e_uaddr_t) *
1010 	    i40e->i40e_resources.ifr_nmacfilt, KM_SLEEP);
1011 	i40e->i40e_maddrs = kmem_zalloc(sizeof (i40e_maddr_t) *
1012 	    i40e->i40e_resources.ifr_nmcastfilt, KM_SLEEP);
1013 
1014 	/*
1015 	 * Initialize these as multicast addresses to indicate it's invalid for
1016 	 * sanity purposes. Think of it like 0xdeadbeef.
1017 	 */
1018 	for (i = 0; i < i40e->i40e_resources.ifr_nmacfilt; i++)
1019 		i40e->i40e_uaddrs[i].iua_mac[0] = 0x01;
1020 
1021 	return (B_TRUE);
1022 }
1023 
1024 static boolean_t
1025 i40e_enable_interrupts(i40e_t *i40e)
1026 {
1027 	int i, rc;
1028 
1029 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1030 		rc = ddi_intr_block_enable(i40e->i40e_intr_handles,
1031 		    i40e->i40e_intr_count);
1032 		if (rc != DDI_SUCCESS) {
1033 			i40e_error(i40e, "Interrupt block-enable failed: %d",
1034 			    rc);
1035 			return (B_FALSE);
1036 		}
1037 	} else {
1038 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1039 			rc = ddi_intr_enable(i40e->i40e_intr_handles[i]);
1040 			if (rc != DDI_SUCCESS) {
1041 				i40e_error(i40e,
1042 				    "Failed to enable interrupt %d: %d", i, rc);
1043 				while (--i >= 0) {
1044 					(void) ddi_intr_disable(
1045 					    i40e->i40e_intr_handles[i]);
1046 				}
1047 				return (B_FALSE);
1048 			}
1049 		}
1050 	}
1051 
1052 	return (B_TRUE);
1053 }
1054 
1055 static boolean_t
1056 i40e_disable_interrupts(i40e_t *i40e)
1057 {
1058 	int i, rc;
1059 
1060 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1061 		rc = ddi_intr_block_disable(i40e->i40e_intr_handles,
1062 		    i40e->i40e_intr_count);
1063 		if (rc != DDI_SUCCESS) {
1064 			i40e_error(i40e,
1065 			    "Interrupt block-disabled failed: %d", rc);
1066 			return (B_FALSE);
1067 		}
1068 	} else {
1069 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1070 			rc = ddi_intr_disable(i40e->i40e_intr_handles[i]);
1071 			if (rc != DDI_SUCCESS) {
1072 				i40e_error(i40e,
1073 				    "Failed to disable interrupt %d: %d",
1074 				    i, rc);
1075 				return (B_FALSE);
1076 			}
1077 		}
1078 	}
1079 
1080 	return (B_TRUE);
1081 }
1082 
1083 /*
1084  * Free receive & transmit rings.
1085  */
1086 static void
1087 i40e_free_trqpairs(i40e_t *i40e)
1088 {
1089 	int i;
1090 	i40e_trqpair_t *itrq;
1091 
1092 	if (i40e->i40e_trqpairs != NULL) {
1093 		for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1094 			itrq = &i40e->i40e_trqpairs[i];
1095 			mutex_destroy(&itrq->itrq_rx_lock);
1096 			mutex_destroy(&itrq->itrq_tx_lock);
1097 			mutex_destroy(&itrq->itrq_tcb_lock);
1098 
1099 			/*
1100 			 * Should have already been cleaned up by start/stop,
1101 			 * etc.
1102 			 */
1103 			ASSERT(itrq->itrq_txkstat == NULL);
1104 			ASSERT(itrq->itrq_rxkstat == NULL);
1105 		}
1106 
1107 		kmem_free(i40e->i40e_trqpairs,
1108 		    sizeof (i40e_trqpair_t) * i40e->i40e_num_trqpairs);
1109 		i40e->i40e_trqpairs = NULL;
1110 	}
1111 
1112 	cv_destroy(&i40e->i40e_rx_pending_cv);
1113 	mutex_destroy(&i40e->i40e_rx_pending_lock);
1114 	mutex_destroy(&i40e->i40e_general_lock);
1115 }
1116 
1117 /*
1118  * Allocate transmit and receive rings, as well as other data structures that we
1119  * need.
1120  */
1121 static boolean_t
1122 i40e_alloc_trqpairs(i40e_t *i40e)
1123 {
1124 	int i;
1125 	void *mutexpri = DDI_INTR_PRI(i40e->i40e_intr_pri);
1126 
1127 	/*
1128 	 * Now that we have the priority for the interrupts, initialize
1129 	 * all relevant locks.
1130 	 */
1131 	mutex_init(&i40e->i40e_general_lock, NULL, MUTEX_DRIVER, mutexpri);
1132 	mutex_init(&i40e->i40e_rx_pending_lock, NULL, MUTEX_DRIVER, mutexpri);
1133 	cv_init(&i40e->i40e_rx_pending_cv, NULL, CV_DRIVER, NULL);
1134 
1135 	i40e->i40e_trqpairs = kmem_zalloc(sizeof (i40e_trqpair_t) *
1136 	    i40e->i40e_num_trqpairs, KM_SLEEP);
1137 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1138 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
1139 
1140 		itrq->itrq_i40e = i40e;
1141 		mutex_init(&itrq->itrq_rx_lock, NULL, MUTEX_DRIVER, mutexpri);
1142 		mutex_init(&itrq->itrq_tx_lock, NULL, MUTEX_DRIVER, mutexpri);
1143 		mutex_init(&itrq->itrq_tcb_lock, NULL, MUTEX_DRIVER, mutexpri);
1144 		itrq->itrq_index = i;
1145 	}
1146 
1147 	return (B_TRUE);
1148 }
1149 
1150 
1151 
1152 /*
1153  * Unless a .conf file already overrode i40e_t structure values, they will
1154  * be 0, and need to be set in conjunction with the now-available HW report.
1155  *
1156  * However, at the moment, we cap all of these resources as we only support a
1157  * single receive ring and a single group.
1158  */
1159 /* ARGSUSED */
1160 static void
1161 i40e_hw_to_instance(i40e_t *i40e, i40e_hw_t *hw)
1162 {
1163 	if (i40e->i40e_num_trqpairs == 0) {
1164 		i40e->i40e_num_trqpairs = I40E_TRQPAIR_MAX;
1165 	}
1166 
1167 	if (i40e->i40e_num_rx_groups == 0) {
1168 		i40e->i40e_num_rx_groups = I40E_GROUP_MAX;
1169 	}
1170 }
1171 
1172 /*
1173  * Free any resources required by, or setup by, the Intel common code.
1174  */
1175 static void
1176 i40e_common_code_fini(i40e_t *i40e)
1177 {
1178 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1179 	int rc;
1180 
1181 	rc = i40e_shutdown_lan_hmc(hw);
1182 	if (rc != I40E_SUCCESS)
1183 		i40e_error(i40e, "failed to shutdown LAN hmc: %d", rc);
1184 
1185 	rc = i40e_shutdown_adminq(hw);
1186 	if (rc != I40E_SUCCESS)
1187 		i40e_error(i40e, "failed to shutdown admin queue: %d", rc);
1188 }
1189 
1190 /*
1191  * Initialize and call Intel common-code routines, includes some setup
1192  * the common code expects from the driver.  Also prints on failure, so
1193  * the caller doesn't have to.
1194  */
1195 static boolean_t
1196 i40e_common_code_init(i40e_t *i40e, i40e_hw_t *hw)
1197 {
1198 	int rc;
1199 
1200 	i40e_clear_hw(hw);
1201 	rc = i40e_pf_reset(hw);
1202 	if (rc != 0) {
1203 		i40e_error(i40e, "failed to reset hardware: %d", rc);
1204 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_NO_RESPONSE);
1205 		return (B_FALSE);
1206 	}
1207 
1208 	rc = i40e_init_shared_code(hw);
1209 	if (rc != 0) {
1210 		i40e_error(i40e, "failed to initialize i40e core: %d", rc);
1211 		return (B_FALSE);
1212 	}
1213 
1214 	hw->aq.num_arq_entries = I40E_DEF_ADMINQ_SIZE;
1215 	hw->aq.num_asq_entries =  I40E_DEF_ADMINQ_SIZE;
1216 	hw->aq.arq_buf_size = I40E_ADMINQ_BUFSZ;
1217 	hw->aq.asq_buf_size = I40E_ADMINQ_BUFSZ;
1218 
1219 	rc = i40e_init_adminq(hw);
1220 	if (rc != 0) {
1221 		i40e_error(i40e, "failed to initialize firmware admin queue: "
1222 		    "%d, potential firmware version mismatch", rc);
1223 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
1224 		return (B_FALSE);
1225 	}
1226 
1227 	if (hw->aq.api_maj_ver == I40E_FW_API_VERSION_MAJOR &&
1228 	    hw->aq.api_min_ver > I40E_FW_API_VERSION_MINOR) {
1229 		i40e_log(i40e, "The driver for the device detected a newer "
1230 		    "version of the NVM image (%d.%d) than expected (%d.%d).\n"
1231 		    "Please install the most recent version of the network "
1232 		    "driver.\n", hw->aq.api_maj_ver, hw->aq.api_min_ver,
1233 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_API_VERSION_MINOR);
1234 	} else if (hw->aq.api_maj_ver < I40E_FW_API_VERSION_MAJOR ||
1235 	    hw->aq.api_min_ver < (I40E_FW_API_VERSION_MINOR - 1)) {
1236 		i40e_log(i40e, "The driver for the device detected an older"
1237 		    " version of the NVM image (%d.%d) than expected (%d.%d)."
1238 		    "\nPlease update the NVM image.\n",
1239 		    hw->aq.api_maj_ver, hw->aq.api_min_ver,
1240 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_API_VERSION_MINOR - 1);
1241 	}
1242 
1243 	i40e_clear_pxe_mode(hw);
1244 
1245 	/*
1246 	 * We need to call this so that the common code can discover
1247 	 * capabilities of the hardware, which it uses throughout the rest.
1248 	 */
1249 	if (!i40e_get_hw_capabilities(i40e, hw)) {
1250 		i40e_error(i40e, "failed to obtain hardware capabilities");
1251 		return (B_FALSE);
1252 	}
1253 
1254 	if (i40e_get_available_resources(i40e) == B_FALSE) {
1255 		i40e_error(i40e, "failed to obtain hardware resources");
1256 		return (B_FALSE);
1257 	}
1258 
1259 	i40e_hw_to_instance(i40e, hw);
1260 
1261 	rc = i40e_init_lan_hmc(hw, hw->func_caps.num_tx_qp,
1262 	    hw->func_caps.num_rx_qp, 0, 0);
1263 	if (rc != 0) {
1264 		i40e_error(i40e, "failed to initialize hardware memory cache: "
1265 		    "%d", rc);
1266 		return (B_FALSE);
1267 	}
1268 
1269 	rc = i40e_configure_lan_hmc(hw, I40E_HMC_MODEL_DIRECT_ONLY);
1270 	if (rc != 0) {
1271 		i40e_error(i40e, "failed to configure hardware memory cache: "
1272 		    "%d", rc);
1273 		return (B_FALSE);
1274 	}
1275 
1276 	(void) i40e_aq_stop_lldp(hw, TRUE, NULL);
1277 
1278 	rc = i40e_get_mac_addr(hw, hw->mac.addr);
1279 	if (rc != I40E_SUCCESS) {
1280 		i40e_error(i40e, "failed to retrieve hardware mac address: %d",
1281 		    rc);
1282 		return (B_FALSE);
1283 	}
1284 
1285 	rc = i40e_validate_mac_addr(hw->mac.addr);
1286 	if (rc != 0) {
1287 		i40e_error(i40e, "failed to validate internal mac address: "
1288 		    "%d", rc);
1289 		return (B_FALSE);
1290 	}
1291 	bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
1292 	if ((rc = i40e_get_port_mac_addr(hw, hw->mac.port_addr)) !=
1293 	    I40E_SUCCESS) {
1294 		i40e_error(i40e, "failed to retrieve port mac address: %d",
1295 		    rc);
1296 		return (B_FALSE);
1297 	}
1298 
1299 	/*
1300 	 * We need to obtain the Virtual Station ID (VSI) before we can
1301 	 * perform other operations on the device.
1302 	 */
1303 	i40e->i40e_vsi_id = i40e_get_vsi_id(i40e);
1304 	if (i40e->i40e_vsi_id == -1) {
1305 		i40e_error(i40e, "failed to obtain VSI ID");
1306 		return (B_FALSE);
1307 	}
1308 
1309 	return (B_TRUE);
1310 }
1311 
1312 static void
1313 i40e_unconfigure(dev_info_t *devinfo, i40e_t *i40e)
1314 {
1315 	int rc;
1316 
1317 	if (i40e->i40e_attach_progress & I40E_ATTACH_ENABLE_INTR)
1318 		(void) i40e_disable_interrupts(i40e);
1319 
1320 	if ((i40e->i40e_attach_progress & I40E_ATTACH_LINK_TIMER) &&
1321 	    i40e->i40e_periodic_id != 0) {
1322 		ddi_periodic_delete(i40e->i40e_periodic_id);
1323 		i40e->i40e_periodic_id = 0;
1324 	}
1325 
1326 	if (i40e->i40e_attach_progress & I40E_ATTACH_MAC) {
1327 		rc = mac_unregister(i40e->i40e_mac_hdl);
1328 		if (rc != 0) {
1329 			i40e_error(i40e, "failed to unregister from mac: %d",
1330 			    rc);
1331 		}
1332 	}
1333 
1334 	if (i40e->i40e_attach_progress & I40E_ATTACH_STATS) {
1335 		i40e_stats_fini(i40e);
1336 	}
1337 
1338 	if (i40e->i40e_attach_progress & I40E_ATTACH_ADD_INTR)
1339 		i40e_rem_intr_handlers(i40e);
1340 
1341 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_RINGSLOCKS)
1342 		i40e_free_trqpairs(i40e);
1343 
1344 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_INTR)
1345 		i40e_rem_intrs(i40e);
1346 
1347 	if (i40e->i40e_attach_progress & I40E_ATTACH_COMMON_CODE)
1348 		i40e_common_code_fini(i40e);
1349 
1350 	i40e_cleanup_resources(i40e);
1351 
1352 	if (i40e->i40e_attach_progress & I40E_ATTACH_PROPS)
1353 		(void) ddi_prop_remove_all(devinfo);
1354 
1355 	if (i40e->i40e_attach_progress & I40E_ATTACH_REGS_MAP &&
1356 	    i40e->i40e_osdep_space.ios_reg_handle != NULL) {
1357 		ddi_regs_map_free(&i40e->i40e_osdep_space.ios_reg_handle);
1358 		i40e->i40e_osdep_space.ios_reg_handle = NULL;
1359 	}
1360 
1361 	if ((i40e->i40e_attach_progress & I40E_ATTACH_PCI_CONFIG) &&
1362 	    i40e->i40e_osdep_space.ios_cfg_handle != NULL) {
1363 		pci_config_teardown(&i40e->i40e_osdep_space.ios_cfg_handle);
1364 		i40e->i40e_osdep_space.ios_cfg_handle = NULL;
1365 	}
1366 
1367 	if (i40e->i40e_attach_progress & I40E_ATTACH_FM_INIT)
1368 		i40e_fm_fini(i40e);
1369 
1370 	kmem_free(i40e->i40e_aqbuf, I40E_ADMINQ_BUFSZ);
1371 	kmem_free(i40e, sizeof (i40e_t));
1372 
1373 	ddi_set_driver_private(devinfo, NULL);
1374 }
1375 
1376 static boolean_t
1377 i40e_final_init(i40e_t *i40e)
1378 {
1379 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1380 	struct i40e_osdep *osdep = OS_DEP(hw);
1381 	uint8_t pbanum[I40E_PBANUM_STRLEN];
1382 	enum i40e_status_code irc;
1383 	char buf[I40E_DDI_PROP_LEN];
1384 
1385 	pbanum[0] = '\0';
1386 	irc = i40e_read_pba_string(hw, pbanum, sizeof (pbanum));
1387 	if (irc != I40E_SUCCESS) {
1388 		i40e_log(i40e, "failed to read PBA string: %d", irc);
1389 	} else {
1390 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1391 		    "printed-board-assembly", (char *)pbanum);
1392 	}
1393 
1394 #ifdef	DEBUG
1395 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.fw_maj_ver,
1396 	    hw->aq.fw_min_ver) < sizeof (buf));
1397 	ASSERT(snprintf(NULL, 0, "%x", hw->aq.fw_build) < sizeof (buf));
1398 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.api_maj_ver,
1399 	    hw->aq.api_min_ver) < sizeof (buf));
1400 #endif
1401 
1402 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.fw_maj_ver,
1403 	    hw->aq.fw_min_ver);
1404 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1405 	    "firmware-version", buf);
1406 	(void) snprintf(buf, sizeof (buf), "%x", hw->aq.fw_build);
1407 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1408 	    "firmware-build", buf);
1409 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.api_maj_ver,
1410 	    hw->aq.api_min_ver);
1411 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1412 	    "api-version", buf);
1413 
1414 	if (!i40e_set_hw_bus_info(hw))
1415 		return (B_FALSE);
1416 
1417 	if (i40e_check_acc_handle(osdep->ios_reg_handle) != DDI_FM_OK) {
1418 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
1419 		return (B_FALSE);
1420 	}
1421 
1422 	return (B_TRUE);
1423 }
1424 
1425 static void
1426 i40e_identify_hardware(i40e_t *i40e)
1427 {
1428 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1429 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1430 
1431 	hw->vendor_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_VENID);
1432 	hw->device_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_DEVID);
1433 	hw->revision_id = pci_config_get8(osdep->ios_cfg_handle,
1434 	    PCI_CONF_REVID);
1435 	hw->subsystem_device_id =
1436 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBSYSID);
1437 	hw->subsystem_vendor_id =
1438 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBVENID);
1439 
1440 	/*
1441 	 * Note that we set the hardware's bus information later on, in
1442 	 * i40e_get_available_resources(). The common code doesn't seem to
1443 	 * require that it be set in any ways, it seems to be mostly for
1444 	 * book-keeping.
1445 	 */
1446 }
1447 
1448 static boolean_t
1449 i40e_regs_map(i40e_t *i40e)
1450 {
1451 	dev_info_t *devinfo = i40e->i40e_dip;
1452 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1453 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1454 	off_t memsize;
1455 	int ret;
1456 
1457 	if (ddi_dev_regsize(devinfo, I40E_ADAPTER_REGSET, &memsize) !=
1458 	    DDI_SUCCESS) {
1459 		i40e_error(i40e, "Used invalid register set to map PCIe regs");
1460 		return (B_FALSE);
1461 	}
1462 
1463 	if ((ret = ddi_regs_map_setup(devinfo, I40E_ADAPTER_REGSET,
1464 	    (caddr_t *)&hw->hw_addr, 0, memsize, &i40e_regs_acc_attr,
1465 	    &osdep->ios_reg_handle)) != DDI_SUCCESS) {
1466 		i40e_error(i40e, "failed to map device registers: %d", ret);
1467 		return (B_FALSE);
1468 	}
1469 
1470 	osdep->ios_reg_size = memsize;
1471 	return (B_TRUE);
1472 }
1473 
1474 /*
1475  * Update parameters required when a new MTU has been configured.  Calculate the
1476  * maximum frame size, as well as, size our DMA buffers which we size in
1477  * increments of 1K.
1478  */
1479 void
1480 i40e_update_mtu(i40e_t *i40e)
1481 {
1482 	uint32_t rx, tx;
1483 
1484 	i40e->i40e_frame_max = i40e->i40e_sdu +
1485 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
1486 
1487 	rx = i40e->i40e_frame_max + I40E_BUF_IPHDR_ALIGNMENT;
1488 	i40e->i40e_rx_buf_size = ((rx >> 10) +
1489 	    ((rx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1490 
1491 	tx = i40e->i40e_frame_max;
1492 	i40e->i40e_tx_buf_size = ((tx >> 10) +
1493 	    ((tx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1494 }
1495 
1496 static int
1497 i40e_get_prop(i40e_t *i40e, char *prop, int min, int max, int def)
1498 {
1499 	int val;
1500 
1501 	val = ddi_prop_get_int(DDI_DEV_T_ANY, i40e->i40e_dip, DDI_PROP_DONTPASS,
1502 	    prop, def);
1503 	if (val > max)
1504 		val = max;
1505 	if (val < min)
1506 		val = min;
1507 	return (val);
1508 }
1509 
1510 static void
1511 i40e_init_properties(i40e_t *i40e)
1512 {
1513 	i40e->i40e_sdu = i40e_get_prop(i40e, "default_mtu",
1514 	    I40E_MIN_MTU, I40E_MAX_MTU, I40E_DEF_MTU);
1515 
1516 	i40e->i40e_intr_force = i40e_get_prop(i40e, "intr_force",
1517 	    I40E_INTR_NONE, I40E_INTR_LEGACY, I40E_INTR_NONE);
1518 
1519 	i40e->i40e_mr_enable = i40e_get_prop(i40e, "mr_enable",
1520 	    B_FALSE, B_TRUE, B_TRUE);
1521 
1522 	i40e->i40e_tx_ring_size = i40e_get_prop(i40e, "tx_ring_size",
1523 	    I40E_MIN_TX_RING_SIZE, I40E_MAX_TX_RING_SIZE,
1524 	    I40E_DEF_TX_RING_SIZE);
1525 	if ((i40e->i40e_tx_ring_size % I40E_DESC_ALIGN) != 0) {
1526 		i40e->i40e_tx_ring_size = P2ROUNDUP(i40e->i40e_tx_ring_size,
1527 		    I40E_DESC_ALIGN);
1528 	}
1529 
1530 	i40e->i40e_tx_block_thresh = i40e_get_prop(i40e, "tx_resched_threshold",
1531 	    I40E_MIN_TX_BLOCK_THRESH,
1532 	    i40e->i40e_tx_ring_size - I40E_TX_MAX_COOKIE,
1533 	    I40E_DEF_TX_BLOCK_THRESH);
1534 
1535 	i40e->i40e_rx_ring_size = i40e_get_prop(i40e, "rx_ring_size",
1536 	    I40E_MIN_RX_RING_SIZE, I40E_MAX_RX_RING_SIZE,
1537 	    I40E_DEF_RX_RING_SIZE);
1538 	if ((i40e->i40e_rx_ring_size % I40E_DESC_ALIGN) != 0) {
1539 		i40e->i40e_rx_ring_size = P2ROUNDUP(i40e->i40e_rx_ring_size,
1540 		    I40E_DESC_ALIGN);
1541 	}
1542 
1543 	i40e->i40e_rx_limit_per_intr = i40e_get_prop(i40e, "rx_limit_per_intr",
1544 	    I40E_MIN_RX_LIMIT_PER_INTR,	I40E_MAX_RX_LIMIT_PER_INTR,
1545 	    I40E_DEF_RX_LIMIT_PER_INTR);
1546 
1547 	i40e->i40e_tx_hcksum_enable = i40e_get_prop(i40e, "tx_hcksum_enable",
1548 	    B_FALSE, B_TRUE, B_TRUE);
1549 
1550 	i40e->i40e_rx_hcksum_enable = i40e_get_prop(i40e, "rx_hcksum_enable",
1551 	    B_FALSE, B_TRUE, B_TRUE);
1552 
1553 	i40e->i40e_rx_dma_min = i40e_get_prop(i40e, "rx_dma_threshold",
1554 	    I40E_MIN_RX_DMA_THRESH, I40E_MAX_RX_DMA_THRESH,
1555 	    I40E_DEF_RX_DMA_THRESH);
1556 
1557 	i40e->i40e_tx_dma_min = i40e_get_prop(i40e, "tx_dma_threshold",
1558 	    I40E_MIN_TX_DMA_THRESH, I40E_MAX_TX_DMA_THRESH,
1559 	    I40E_DEF_TX_DMA_THRESH);
1560 
1561 	i40e->i40e_tx_itr = i40e_get_prop(i40e, "tx_intr_throttle",
1562 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_TX_ITR);
1563 
1564 	i40e->i40e_rx_itr = i40e_get_prop(i40e, "rx_intr_throttle",
1565 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_RX_ITR);
1566 
1567 	i40e->i40e_other_itr = i40e_get_prop(i40e, "other_intr_throttle",
1568 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_OTHER_ITR);
1569 
1570 	if (!i40e->i40e_mr_enable) {
1571 		i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1572 		i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1573 	}
1574 
1575 	i40e_update_mtu(i40e);
1576 }
1577 
1578 /*
1579  * There are a few constraints on interrupts that we're currently imposing, some
1580  * of which are restrictions from hardware. For a fuller treatment, see
1581  * i40e_intr.c.
1582  *
1583  * Currently, to use MSI-X we require two interrupts be available though in
1584  * theory we should participate in IRM and happily use more interrupts.
1585  *
1586  * Hardware only supports a single MSI being programmed and therefore if we
1587  * don't have MSI-X interrupts available at this time, then we ratchet down the
1588  * number of rings and groups available. Obviously, we only bother with a single
1589  * fixed interrupt.
1590  */
1591 static boolean_t
1592 i40e_alloc_intr_handles(i40e_t *i40e, dev_info_t *devinfo, int intr_type)
1593 {
1594 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1595 	ddi_acc_handle_t rh = i40e->i40e_osdep_space.ios_reg_handle;
1596 	int request, count, actual, rc, min;
1597 	uint32_t reg;
1598 
1599 	switch (intr_type) {
1600 	case DDI_INTR_TYPE_FIXED:
1601 	case DDI_INTR_TYPE_MSI:
1602 		request = 1;
1603 		min = 1;
1604 		break;
1605 	case DDI_INTR_TYPE_MSIX:
1606 		min = 2;
1607 		if (!i40e->i40e_mr_enable) {
1608 			request = 2;
1609 			break;
1610 		}
1611 		reg = I40E_READ_REG(hw, I40E_GLPCI_CNF2);
1612 		/*
1613 		 * Should this read fail, we will drop back to using
1614 		 * MSI or fixed interrupts.
1615 		 */
1616 		if (i40e_check_acc_handle(rh) != DDI_FM_OK) {
1617 			ddi_fm_service_impact(i40e->i40e_dip,
1618 			    DDI_SERVICE_DEGRADED);
1619 			return (B_FALSE);
1620 		}
1621 		request = (reg & I40E_GLPCI_CNF2_MSI_X_PF_N_MASK) >>
1622 		    I40E_GLPCI_CNF2_MSI_X_PF_N_SHIFT;
1623 		request++;	/* the register value is n - 1 */
1624 		break;
1625 	default:
1626 		panic("bad interrupt type passed to i40e_alloc_intr_handles: "
1627 		    "%d", intr_type);
1628 		return (B_FALSE);
1629 	}
1630 
1631 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
1632 	if (rc != DDI_SUCCESS || count < min) {
1633 		i40e_log(i40e, "Get interrupt number failed, "
1634 		    "returned %d, count %d", rc, count);
1635 		return (B_FALSE);
1636 	}
1637 
1638 	rc = ddi_intr_get_navail(devinfo, intr_type, &count);
1639 	if (rc != DDI_SUCCESS || count < min) {
1640 		i40e_log(i40e, "Get AVAILABLE interrupt number failed, "
1641 		    "returned %d, count %d", rc, count);
1642 		return (B_FALSE);
1643 	}
1644 
1645 	actual = 0;
1646 	i40e->i40e_intr_count = 0;
1647 	i40e->i40e_intr_count_max = 0;
1648 	i40e->i40e_intr_count_min = 0;
1649 
1650 	i40e->i40e_intr_size = request * sizeof (ddi_intr_handle_t);
1651 	ASSERT(i40e->i40e_intr_size != 0);
1652 	i40e->i40e_intr_handles = kmem_alloc(i40e->i40e_intr_size, KM_SLEEP);
1653 
1654 	rc = ddi_intr_alloc(devinfo, i40e->i40e_intr_handles, intr_type, 0,
1655 	    min(request, count), &actual, DDI_INTR_ALLOC_NORMAL);
1656 	if (rc != DDI_SUCCESS) {
1657 		i40e_log(i40e, "Interrupt allocation failed with %d.", rc);
1658 		goto alloc_handle_fail;
1659 	}
1660 
1661 	i40e->i40e_intr_count = actual;
1662 	i40e->i40e_intr_count_max = request;
1663 	i40e->i40e_intr_count_min = min;
1664 
1665 	if (actual < min) {
1666 		i40e_log(i40e, "actual (%d) is less than minimum (%d).",
1667 		    actual, min);
1668 		goto alloc_handle_fail;
1669 	}
1670 
1671 	/*
1672 	 * Record the priority and capabilities for our first vector.  Once
1673 	 * we have it, that's our priority until detach time.  Even if we
1674 	 * eventually participate in IRM, our priority shouldn't change.
1675 	 */
1676 	rc = ddi_intr_get_pri(i40e->i40e_intr_handles[0], &i40e->i40e_intr_pri);
1677 	if (rc != DDI_SUCCESS) {
1678 		i40e_log(i40e,
1679 		    "Getting interrupt priority failed with %d.", rc);
1680 		goto alloc_handle_fail;
1681 	}
1682 
1683 	rc = ddi_intr_get_cap(i40e->i40e_intr_handles[0], &i40e->i40e_intr_cap);
1684 	if (rc != DDI_SUCCESS) {
1685 		i40e_log(i40e,
1686 		    "Getting interrupt capabilities failed with %d.", rc);
1687 		goto alloc_handle_fail;
1688 	}
1689 
1690 	i40e->i40e_intr_type = intr_type;
1691 	return (B_TRUE);
1692 
1693 alloc_handle_fail:
1694 
1695 	i40e_rem_intrs(i40e);
1696 	return (B_FALSE);
1697 }
1698 
1699 static boolean_t
1700 i40e_alloc_intrs(i40e_t *i40e, dev_info_t *devinfo)
1701 {
1702 	int intr_types, rc;
1703 
1704 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
1705 	if (rc != DDI_SUCCESS) {
1706 		i40e_error(i40e, "failed to get supported interrupt types: %d",
1707 		    rc);
1708 		return (B_FALSE);
1709 	}
1710 
1711 	i40e->i40e_intr_type = 0;
1712 
1713 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
1714 	    i40e->i40e_intr_force <= I40E_INTR_MSIX) {
1715 		if (i40e_alloc_intr_handles(i40e, devinfo,
1716 		    DDI_INTR_TYPE_MSIX)) {
1717 			i40e->i40e_num_trqpairs =
1718 			    MIN(i40e->i40e_intr_count - 1,
1719 			    I40E_AQ_VSI_TC_QUE_SIZE_MAX);
1720 			return (B_TRUE);
1721 		}
1722 	}
1723 
1724 	/*
1725 	 * We only use multiple transmit/receive pairs when MSI-X interrupts are
1726 	 * available due to the fact that the device basically only supports a
1727 	 * single MSI interrupt.
1728 	 */
1729 	i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1730 	i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1731 
1732 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
1733 	    (i40e->i40e_intr_force <= I40E_INTR_MSI)) {
1734 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_MSI))
1735 			return (B_TRUE);
1736 	}
1737 
1738 	if (intr_types & DDI_INTR_TYPE_FIXED) {
1739 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_FIXED))
1740 			return (B_TRUE);
1741 	}
1742 
1743 	return (B_FALSE);
1744 }
1745 
1746 /*
1747  * Map different interrupts to MSI-X vectors.
1748  */
1749 static boolean_t
1750 i40e_map_intrs_to_vectors(i40e_t *i40e)
1751 {
1752 	int i;
1753 
1754 	if (i40e->i40e_intr_type != DDI_INTR_TYPE_MSIX) {
1755 		return (B_TRUE);
1756 	}
1757 
1758 	/*
1759 	 * Each queue pair is mapped to a single interrupt, so transmit
1760 	 * and receive interrupts for a given queue share the same vector.
1761 	 * The number of queue pairs is one less than the number of interrupt
1762 	 * vectors and is assigned the vector one higher than its index.
1763 	 * Vector zero is reserved for the admin queue.
1764 	 */
1765 	ASSERT(i40e->i40e_intr_count == i40e->i40e_num_trqpairs + 1);
1766 
1767 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1768 		i40e->i40e_trqpairs[i].itrq_rx_intrvec = i + 1;
1769 		i40e->i40e_trqpairs[i].itrq_tx_intrvec = i + 1;
1770 	}
1771 
1772 	return (B_TRUE);
1773 }
1774 
1775 static boolean_t
1776 i40e_add_intr_handlers(i40e_t *i40e)
1777 {
1778 	int rc, vector;
1779 
1780 	switch (i40e->i40e_intr_type) {
1781 	case DDI_INTR_TYPE_MSIX:
1782 		for (vector = 0; vector < i40e->i40e_intr_count; vector++) {
1783 			rc = ddi_intr_add_handler(
1784 			    i40e->i40e_intr_handles[vector],
1785 			    (ddi_intr_handler_t *)i40e_intr_msix, i40e,
1786 			    (void *)(uintptr_t)vector);
1787 			if (rc != DDI_SUCCESS) {
1788 				i40e_log(i40e, "Add interrupt handler (MSI-X) "
1789 				    "failed: return %d, vector %d", rc, vector);
1790 				for (vector--; vector >= 0; vector--) {
1791 					(void) ddi_intr_remove_handler(
1792 					    i40e->i40e_intr_handles[vector]);
1793 				}
1794 				return (B_FALSE);
1795 			}
1796 		}
1797 		break;
1798 	case DDI_INTR_TYPE_MSI:
1799 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1800 		    (ddi_intr_handler_t *)i40e_intr_msi, i40e, NULL);
1801 		if (rc != DDI_SUCCESS) {
1802 			i40e_log(i40e, "Add interrupt handler (MSI) failed: "
1803 			    "return %d", rc);
1804 			return (B_FALSE);
1805 		}
1806 		break;
1807 	case DDI_INTR_TYPE_FIXED:
1808 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1809 		    (ddi_intr_handler_t *)i40e_intr_legacy, i40e, NULL);
1810 		if (rc != DDI_SUCCESS) {
1811 			i40e_log(i40e, "Add interrupt handler (legacy) failed:"
1812 			    " return %d", rc);
1813 			return (B_FALSE);
1814 		}
1815 		break;
1816 	default:
1817 		/* Cast to pacify lint */
1818 		panic("i40e_intr_type %p contains an unknown type: %d",
1819 		    (void *)i40e, i40e->i40e_intr_type);
1820 	}
1821 
1822 	return (B_TRUE);
1823 }
1824 
1825 /*
1826  * Perform periodic checks. Longer term, we should be thinking about additional
1827  * things here:
1828  *
1829  * o Stall Detection
1830  * o Temperature sensor detection
1831  * o Device resetting
1832  * o Statistics updating to avoid wraparound
1833  */
1834 static void
1835 i40e_timer(void *arg)
1836 {
1837 	i40e_t *i40e = arg;
1838 
1839 	mutex_enter(&i40e->i40e_general_lock);
1840 	i40e_link_check(i40e);
1841 	mutex_exit(&i40e->i40e_general_lock);
1842 }
1843 
1844 /*
1845  * Get the hardware state, and scribble away anything that needs scribbling.
1846  */
1847 static void
1848 i40e_get_hw_state(i40e_t *i40e, i40e_hw_t *hw)
1849 {
1850 	int rc;
1851 
1852 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
1853 
1854 	(void) i40e_aq_get_link_info(hw, TRUE, NULL, NULL);
1855 	i40e_link_check(i40e);
1856 
1857 	/*
1858 	 * Try and determine our PHY. Note that we may have to retry to and
1859 	 * delay to detect fiber correctly.
1860 	 */
1861 	rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE, &i40e->i40e_phy,
1862 	    NULL);
1863 	if (rc == I40E_ERR_UNKNOWN_PHY) {
1864 		i40e_msec_delay(200);
1865 		rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE,
1866 		    &i40e->i40e_phy, NULL);
1867 	}
1868 
1869 	if (rc != I40E_SUCCESS) {
1870 		if (rc == I40E_ERR_UNKNOWN_PHY) {
1871 			i40e_error(i40e, "encountered unknown PHY type, "
1872 			    "not attaching.");
1873 		} else {
1874 			i40e_error(i40e, "error getting physical capabilities: "
1875 			    "%d, %d", rc, hw->aq.asq_last_status);
1876 		}
1877 	}
1878 
1879 	rc = i40e_update_link_info(hw);
1880 	if (rc != I40E_SUCCESS) {
1881 		i40e_error(i40e, "failed to update link information: %d", rc);
1882 	}
1883 
1884 	/*
1885 	 * In general, we don't want to mask off (as in stop from being a cause)
1886 	 * any of the interrupts that the phy might be able to generate.
1887 	 */
1888 	rc = i40e_aq_set_phy_int_mask(hw, 0, NULL);
1889 	if (rc != I40E_SUCCESS) {
1890 		i40e_error(i40e, "failed to update phy link mask: %d", rc);
1891 	}
1892 }
1893 
1894 /*
1895  * Go through and re-initialize any existing filters that we may have set up for
1896  * this device. Note that we would only expect them to exist if hardware had
1897  * already been initialized and we had just reset it. While we're not
1898  * implementing this yet, we're keeping this around for when we add reset
1899  * capabilities, so this isn't forgotten.
1900  */
1901 /* ARGSUSED */
1902 static void
1903 i40e_init_macaddrs(i40e_t *i40e, i40e_hw_t *hw)
1904 {
1905 }
1906 
1907 /*
1908  * Configure the hardware for the Virtual Station Interface (VSI).  Currently
1909  * we only support one, but in the future we could instantiate more than one
1910  * per attach-point.
1911  */
1912 static boolean_t
1913 i40e_config_vsi(i40e_t *i40e, i40e_hw_t *hw)
1914 {
1915 	struct i40e_vsi_context	context;
1916 	int err, tc_queues;
1917 
1918 	bzero(&context, sizeof (struct i40e_vsi_context));
1919 	context.seid = i40e->i40e_vsi_id;
1920 	context.pf_num = hw->pf_id;
1921 	err = i40e_aq_get_vsi_params(hw, &context, NULL);
1922 	if (err != I40E_SUCCESS) {
1923 		i40e_error(i40e, "get VSI params failed with %d", err);
1924 		return (B_FALSE);
1925 	}
1926 
1927 	i40e->i40e_vsi_num = context.vsi_number;
1928 
1929 	/*
1930 	 * Set the queue and traffic class bits.  Keep it simple for now.
1931 	 */
1932 	context.info.valid_sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
1933 	context.info.mapping_flags = I40E_AQ_VSI_QUE_MAP_CONTIG;
1934 	context.info.queue_mapping[0] = I40E_ASSIGN_ALL_QUEUES;
1935 
1936 	/*
1937 	 * tc_queues determines the size of the traffic class, where the
1938 	 * size is 2^^tc_queues to a maximum of 64.
1939 	 * Some examples:
1940 	 * 	i40e_num_trqpairs == 1 =>  tc_queues = 0, 2^^0 = 1.
1941 	 * 	i40e_num_trqpairs == 7 =>  tc_queues = 3, 2^^3 = 8.
1942 	 * 	i40e_num_trqpairs == 8 =>  tc_queues = 3, 2^^3 = 8.
1943 	 * 	i40e_num_trqpairs == 9 =>  tc_queues = 4, 2^^4 = 16.
1944 	 * 	i40e_num_trqpairs == 17 => tc_queues = 5, 2^^5 = 32.
1945 	 * 	i40e_num_trqpairs == 64 => tc_queues = 6, 2^^6 = 64.
1946 	 */
1947 	tc_queues = ddi_fls(i40e->i40e_num_trqpairs - 1);
1948 
1949 	context.info.tc_mapping[0] = ((0 << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) &
1950 	    I40E_AQ_VSI_TC_QUE_OFFSET_MASK) |
1951 	    ((tc_queues << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT) &
1952 	    I40E_AQ_VSI_TC_QUE_NUMBER_MASK);
1953 
1954 	context.info.valid_sections |= I40E_AQ_VSI_PROP_VLAN_VALID;
1955 	context.info.port_vlan_flags = I40E_AQ_VSI_PVLAN_MODE_ALL |
1956 	    I40E_AQ_VSI_PVLAN_EMOD_NOTHING;
1957 
1958 	context.flags = LE16_TO_CPU(I40E_AQ_VSI_TYPE_PF);
1959 
1960 	i40e->i40e_vsi_stat_id = LE16_TO_CPU(context.info.stat_counter_idx);
1961 	if (i40e_stat_vsi_init(i40e) == B_FALSE)
1962 		return (B_FALSE);
1963 
1964 	err = i40e_aq_update_vsi_params(hw, &context, NULL);
1965 	if (err != I40E_SUCCESS) {
1966 		i40e_error(i40e, "Update VSI params failed with %d", err);
1967 		return (B_FALSE);
1968 	}
1969 
1970 
1971 	return (B_TRUE);
1972 }
1973 
1974 /*
1975  * Set up RSS.
1976  * 	1. Seed the hash key.
1977  *	2. Enable PCTYPEs for the hash filter.
1978  *	3. Populate the LUT.
1979  *
1980  * Note: When/if X722 support is added the hash key is seeded via a call
1981  *	 to i40e_aq_set_rss_key(), and the LUT is populated using
1982  *	 i40e_aq_set_rss_lut().
1983  */
1984 static boolean_t
1985 i40e_config_rss(i40e_t *i40e, i40e_hw_t *hw)
1986 {
1987 	int i;
1988 	uint8_t lut_mask;
1989 	uint32_t *hlut;
1990 	uint64_t hena;
1991 	boolean_t rv = B_TRUE;
1992 	uint32_t seed[I40E_PFQF_HKEY_MAX_INDEX + 1];
1993 
1994 	/*
1995 	 * 1. Seed the hash key
1996 	 */
1997 	(void) random_get_pseudo_bytes((uint8_t *)seed, sizeof (seed));
1998 
1999 	for (i = 0; i <= I40E_PFQF_HKEY_MAX_INDEX; i++)
2000 		i40e_write_rx_ctl(hw, I40E_PFQF_HKEY(i), seed[i]);
2001 
2002 	/*
2003 	 * 2. Configure PCTYPES
2004 	 */
2005 	hena = (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) |
2006 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP) |
2007 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_UDP) |
2008 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV4) |
2009 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) |
2010 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP) |
2011 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_UDP) |
2012 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV6) |
2013 	    (1ULL << I40E_FILTER_PCTYPE_L2_PAYLOAD);
2014 
2015 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(0), (uint32_t)hena);
2016 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(1), (uint32_t)(hena >> 32));
2017 
2018 	/*
2019 	 * 3. Populate LUT
2020 	 *
2021 	 * Each entry in the LUT is 8 bits and is used to index
2022 	 * the rx queue. Populate the LUT in a round robin fashion
2023 	 * with rx queue indices from 0 to i40e_num_trqpairs - 1.
2024 	 */
2025 	hlut = kmem_alloc(hw->func_caps.rss_table_size, KM_NOSLEEP);
2026 	if (hlut == NULL) {
2027 		i40e_error(i40e, "i40e_config_rss() buffer allocation failed");
2028 		return (B_FALSE);
2029 	}
2030 
2031 	lut_mask = (1 << hw->func_caps.rss_table_entry_width) - 1;
2032 
2033 	for (i = 0; i < hw->func_caps.rss_table_size; i++)
2034 		((uint8_t *)hlut)[i] = (i % i40e->i40e_num_trqpairs) & lut_mask;
2035 
2036 	for (i = 0; i < hw->func_caps.rss_table_size >> 2; i++)
2037 		I40E_WRITE_REG(hw, I40E_PFQF_HLUT(i), hlut[i]);
2038 
2039 	kmem_free(hlut, hw->func_caps.rss_table_size);
2040 
2041 	return (rv);
2042 }
2043 
2044 /*
2045  * Wrapper to kick the chipset on.
2046  */
2047 static boolean_t
2048 i40e_chip_start(i40e_t *i40e)
2049 {
2050 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2051 	struct i40e_filter_control_settings filter;
2052 	int rc;
2053 
2054 	if (((hw->aq.fw_maj_ver == 4) && (hw->aq.fw_min_ver < 33)) ||
2055 	    (hw->aq.fw_maj_ver < 4)) {
2056 		i40e_msec_delay(75);
2057 		if (i40e_aq_set_link_restart_an(hw, TRUE, NULL) !=
2058 		    I40E_SUCCESS) {
2059 			i40e_error(i40e, "failed to restart link: admin queue "
2060 			    "error: %d", hw->aq.asq_last_status);
2061 			return (B_FALSE);
2062 		}
2063 	}
2064 
2065 	/* Determine hardware state */
2066 	i40e_get_hw_state(i40e, hw);
2067 
2068 	/* Initialize mac addresses. */
2069 	i40e_init_macaddrs(i40e, hw);
2070 
2071 	/*
2072 	 * Set up the filter control.
2073 	 */
2074 	bzero(&filter, sizeof (filter));
2075 	filter.enable_ethtype = TRUE;
2076 	filter.enable_macvlan = TRUE;
2077 	filter.hash_lut_size = I40E_HASH_LUT_SIZE_512;
2078 
2079 	rc = i40e_set_filter_control(hw, &filter);
2080 	if (rc != I40E_SUCCESS) {
2081 		i40e_error(i40e, "i40e_set_filter_control() returned %d", rc);
2082 		return (B_FALSE);
2083 	}
2084 
2085 	i40e_intr_chip_init(i40e);
2086 
2087 	if (!i40e_config_vsi(i40e, hw))
2088 		return (B_FALSE);
2089 
2090 	if (!i40e_config_rss(i40e, hw))
2091 		return (B_FALSE);
2092 
2093 	i40e_flush(hw);
2094 
2095 	return (B_TRUE);
2096 }
2097 
2098 /*
2099  * Take care of tearing down the rx ring. See 8.3.3.1.2 for more information.
2100  */
2101 static void
2102 i40e_shutdown_rx_rings(i40e_t *i40e)
2103 {
2104 	int i;
2105 	uint32_t reg;
2106 
2107 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2108 
2109 	/*
2110 	 * Step 1. The interrupt linked list (see i40e_intr.c for more
2111 	 * information) should have already been cleared before calling this
2112 	 * function.
2113 	 */
2114 #ifdef	DEBUG
2115 	if (i40e->i40e_intr_type == DDI_INTR_TYPE_MSIX) {
2116 		for (i = 1; i < i40e->i40e_intr_count; i++) {
2117 			reg = I40E_READ_REG(hw, I40E_PFINT_LNKLSTN(i - 1));
2118 			VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2119 		}
2120 	} else {
2121 		reg = I40E_READ_REG(hw, I40E_PFINT_LNKLST0);
2122 		VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2123 	}
2124 
2125 #endif	/* DEBUG */
2126 
2127 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2128 		/*
2129 		 * Step 1. Request the queue by clearing QENA_REQ. It may not be
2130 		 * set due to unwinding from failures and a partially enabled
2131 		 * ring set.
2132 		 */
2133 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2134 		if (!(reg & I40E_QRX_ENA_QENA_REQ_MASK))
2135 			continue;
2136 		VERIFY((reg & I40E_QRX_ENA_QENA_REQ_MASK) ==
2137 		    I40E_QRX_ENA_QENA_REQ_MASK);
2138 		reg &= ~I40E_QRX_ENA_QENA_REQ_MASK;
2139 		I40E_WRITE_REG(hw, I40E_QRX_ENA(i), reg);
2140 	}
2141 
2142 	/*
2143 	 * Step 2. Wait for the disable to take, by having QENA_STAT in the FPM
2144 	 * be cleared. Note that we could still receive data in the queue during
2145 	 * this time. We don't actually wait for this now and instead defer this
2146 	 * to i40e_shutdown_rings_wait(), after we've interleaved disabling the
2147 	 * TX queues as well.
2148 	 */
2149 }
2150 
2151 static void
2152 i40e_shutdown_tx_rings(i40e_t *i40e)
2153 {
2154 	int i;
2155 	uint32_t reg;
2156 
2157 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2158 
2159 	/*
2160 	 * Step 1. The interrupt linked list should already have been cleared.
2161 	 */
2162 #ifdef DEBUG
2163 	if (i40e->i40e_intr_type == DDI_INTR_TYPE_MSIX) {
2164 		for (i = 1; i < i40e->i40e_intr_count; i++) {
2165 			reg = I40E_READ_REG(hw, I40E_PFINT_LNKLSTN(i - 1));
2166 			VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2167 		}
2168 	} else {
2169 		reg = I40E_READ_REG(hw, I40E_PFINT_LNKLST0);
2170 		VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2171 
2172 	}
2173 #endif	/* DEBUG */
2174 
2175 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2176 		/*
2177 		 * Step 2. Set the SET_QDIS flag for every queue.
2178 		 */
2179 		i40e_pre_tx_queue_cfg(hw, i, B_FALSE);
2180 	}
2181 
2182 	/*
2183 	 * Step 3. Wait at least 400 usec (can be done once for all queues).
2184 	 */
2185 	drv_usecwait(500);
2186 
2187 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2188 		/*
2189 		 * Step 4. Clear the QENA_REQ flag which tells hardware to
2190 		 * quiesce. If QENA_REQ is not already set then that means that
2191 		 * we likely already tried to disable this queue.
2192 		 */
2193 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2194 		if (!(reg & I40E_QTX_ENA_QENA_REQ_MASK))
2195 			continue;
2196 		reg &= ~I40E_QTX_ENA_QENA_REQ_MASK;
2197 		I40E_WRITE_REG(hw, I40E_QTX_ENA(i), reg);
2198 	}
2199 
2200 	/*
2201 	 * Step 5. Wait for all drains to finish. This will be done by the
2202 	 * hardware removing the QENA_STAT flag from the queue. Rather than
2203 	 * waiting here, we interleave it with all the others in
2204 	 * i40e_shutdown_rings_wait().
2205 	 */
2206 }
2207 
2208 /*
2209  * Wait for all the rings to be shut down. e.g. Steps 2 and 5 from the above
2210  * functions.
2211  */
2212 static boolean_t
2213 i40e_shutdown_rings_wait(i40e_t *i40e)
2214 {
2215 	int i, try;
2216 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2217 
2218 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2219 		uint32_t reg;
2220 
2221 		for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2222 			reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2223 			if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0)
2224 				break;
2225 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2226 		}
2227 
2228 		if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) != 0) {
2229 			i40e_error(i40e, "timed out disabling rx queue %d",
2230 			    i);
2231 			return (B_FALSE);
2232 		}
2233 
2234 		for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2235 			reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2236 			if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0)
2237 				break;
2238 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2239 		}
2240 
2241 		if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) != 0) {
2242 			i40e_error(i40e, "timed out disabling tx queue %d",
2243 			    i);
2244 			return (B_FALSE);
2245 		}
2246 	}
2247 
2248 	return (B_TRUE);
2249 }
2250 
2251 static boolean_t
2252 i40e_shutdown_rings(i40e_t *i40e)
2253 {
2254 	i40e_shutdown_rx_rings(i40e);
2255 	i40e_shutdown_tx_rings(i40e);
2256 	return (i40e_shutdown_rings_wait(i40e));
2257 }
2258 
2259 static void
2260 i40e_setup_rx_descs(i40e_trqpair_t *itrq)
2261 {
2262 	int i;
2263 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2264 
2265 	for (i = 0; i < rxd->rxd_ring_size; i++) {
2266 		i40e_rx_control_block_t *rcb;
2267 		i40e_rx_desc_t *rdesc;
2268 
2269 		rcb = rxd->rxd_work_list[i];
2270 		rdesc = &rxd->rxd_desc_ring[i];
2271 
2272 		rdesc->read.pkt_addr =
2273 		    CPU_TO_LE64((uintptr_t)rcb->rcb_dma.dmab_dma_address);
2274 		rdesc->read.hdr_addr = 0;
2275 	}
2276 }
2277 
2278 static boolean_t
2279 i40e_setup_rx_hmc(i40e_trqpair_t *itrq)
2280 {
2281 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2282 	i40e_t *i40e = itrq->itrq_i40e;
2283 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2284 
2285 	struct i40e_hmc_obj_rxq rctx;
2286 	int err;
2287 
2288 	bzero(&rctx, sizeof (struct i40e_hmc_obj_rxq));
2289 	rctx.base = rxd->rxd_desc_area.dmab_dma_address /
2290 	    I40E_HMC_RX_CTX_UNIT;
2291 	rctx.qlen = rxd->rxd_ring_size;
2292 	VERIFY(i40e->i40e_rx_buf_size >= I40E_HMC_RX_DBUFF_MIN);
2293 	VERIFY(i40e->i40e_rx_buf_size <= I40E_HMC_RX_DBUFF_MAX);
2294 	rctx.dbuff = i40e->i40e_rx_buf_size >> I40E_RXQ_CTX_DBUFF_SHIFT;
2295 	rctx.hbuff = 0 >> I40E_RXQ_CTX_HBUFF_SHIFT;
2296 	rctx.dtype = I40E_HMC_RX_DTYPE_NOSPLIT;
2297 	rctx.dsize = I40E_HMC_RX_DSIZE_32BYTE;
2298 	rctx.crcstrip = I40E_HMC_RX_CRCSTRIP_ENABLE;
2299 	rctx.fc_ena = I40E_HMC_RX_FC_DISABLE;
2300 	rctx.l2tsel = I40E_HMC_RX_L2TAGORDER;
2301 	rctx.hsplit_0 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2302 	rctx.hsplit_1 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2303 	rctx.showiv = I40E_HMC_RX_INVLAN_DONTSTRIP;
2304 	rctx.rxmax = i40e->i40e_frame_max;
2305 	rctx.tphrdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2306 	rctx.tphwdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2307 	rctx.tphdata_ena = I40E_HMC_RX_TPH_DISABLE;
2308 	rctx.tphhead_ena = I40E_HMC_RX_TPH_DISABLE;
2309 	rctx.lrxqthresh = I40E_HMC_RX_LOWRXQ_NOINTR;
2310 
2311 	/*
2312 	 * This must be set to 0x1, see Table 8-12 in section 8.3.3.2.2.
2313 	 */
2314 	rctx.prefena = I40E_HMC_RX_PREFENA;
2315 
2316 	err = i40e_clear_lan_rx_queue_context(hw, itrq->itrq_index);
2317 	if (err != I40E_SUCCESS) {
2318 		i40e_error(i40e, "failed to clear rx queue %d context: %d",
2319 		    itrq->itrq_index, err);
2320 		return (B_FALSE);
2321 	}
2322 
2323 	err = i40e_set_lan_rx_queue_context(hw, itrq->itrq_index, &rctx);
2324 	if (err != I40E_SUCCESS) {
2325 		i40e_error(i40e, "failed to set rx queue %d context: %d",
2326 		    itrq->itrq_index, err);
2327 		return (B_FALSE);
2328 	}
2329 
2330 	return (B_TRUE);
2331 }
2332 
2333 /*
2334  * Take care of setting up the descriptor rings and actually programming the
2335  * device. See 8.3.3.1.1 for the full list of steps we need to do to enable the
2336  * rx rings.
2337  */
2338 static boolean_t
2339 i40e_setup_rx_rings(i40e_t *i40e)
2340 {
2341 	int i;
2342 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2343 
2344 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2345 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
2346 		i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2347 		uint32_t reg;
2348 
2349 		/*
2350 		 * Step 1. Program all receive ring descriptors.
2351 		 */
2352 		i40e_setup_rx_descs(itrq);
2353 
2354 		/*
2355 		 * Step 2. Program the queue's FPM/HMC context.
2356 		 */
2357 		if (i40e_setup_rx_hmc(itrq) == B_FALSE)
2358 			return (B_FALSE);
2359 
2360 		/*
2361 		 * Step 3. Clear the queue's tail pointer and set it to the end
2362 		 * of the space.
2363 		 */
2364 		I40E_WRITE_REG(hw, I40E_QRX_TAIL(i), 0);
2365 		I40E_WRITE_REG(hw, I40E_QRX_TAIL(i), rxd->rxd_ring_size - 1);
2366 
2367 		/*
2368 		 * Step 4. Enable the queue via the QENA_REQ.
2369 		 */
2370 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2371 		VERIFY0(reg & (I40E_QRX_ENA_QENA_REQ_MASK |
2372 		    I40E_QRX_ENA_QENA_STAT_MASK));
2373 		reg |= I40E_QRX_ENA_QENA_REQ_MASK;
2374 		I40E_WRITE_REG(hw, I40E_QRX_ENA(i), reg);
2375 	}
2376 
2377 	/*
2378 	 * Note, we wait for every queue to be enabled before we start checking.
2379 	 * This will hopefully cause most queues to be enabled at this point.
2380 	 */
2381 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2382 		uint32_t j, reg;
2383 
2384 		/*
2385 		 * Step 5. Verify that QENA_STAT has been set. It's promised
2386 		 * that this should occur within about 10 us, but like other
2387 		 * systems, we give the card a bit more time.
2388 		 */
2389 		for (j = 0; j < I40E_RING_WAIT_NTRIES; j++) {
2390 			reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2391 
2392 			if (reg & I40E_QRX_ENA_QENA_STAT_MASK)
2393 				break;
2394 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2395 		}
2396 
2397 		if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0) {
2398 			i40e_error(i40e, "failed to enable rx queue %d, timed "
2399 			    "out.", i);
2400 			return (B_FALSE);
2401 		}
2402 	}
2403 
2404 	return (B_TRUE);
2405 }
2406 
2407 static boolean_t
2408 i40e_setup_tx_hmc(i40e_trqpair_t *itrq)
2409 {
2410 	i40e_t *i40e = itrq->itrq_i40e;
2411 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2412 
2413 	struct i40e_hmc_obj_txq tctx;
2414 	struct i40e_vsi_context	context;
2415 	int err;
2416 
2417 	bzero(&tctx, sizeof (struct i40e_hmc_obj_txq));
2418 	tctx.new_context = I40E_HMC_TX_NEW_CONTEXT;
2419 	tctx.base = itrq->itrq_desc_area.dmab_dma_address /
2420 	    I40E_HMC_TX_CTX_UNIT;
2421 	tctx.fc_ena = I40E_HMC_TX_FC_DISABLE;
2422 	tctx.timesync_ena = I40E_HMC_TX_TS_DISABLE;
2423 	tctx.fd_ena = I40E_HMC_TX_FD_DISABLE;
2424 	tctx.alt_vlan_ena = I40E_HMC_TX_ALT_VLAN_DISABLE;
2425 	tctx.head_wb_ena = I40E_HMC_TX_WB_ENABLE;
2426 	tctx.qlen = itrq->itrq_tx_ring_size;
2427 	tctx.tphrdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2428 	tctx.tphrpacket_ena = I40E_HMC_TX_TPH_DISABLE;
2429 	tctx.tphwdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2430 	tctx.head_wb_addr = itrq->itrq_desc_area.dmab_dma_address +
2431 	    sizeof (i40e_tx_desc_t) * itrq->itrq_tx_ring_size;
2432 
2433 	/*
2434 	 * This field isn't actually documented, like crc, but it suggests that
2435 	 * it should be zeroed. We leave both of these here because of that for
2436 	 * now. We should check with Intel on why these are here even.
2437 	 */
2438 	tctx.crc = 0;
2439 	tctx.rdylist_act = 0;
2440 
2441 	/*
2442 	 * We're supposed to assign the rdylist field with the value of the
2443 	 * traffic class index for the first device. We query the VSI parameters
2444 	 * again to get what the handle is. Note that every queue is always
2445 	 * assigned to traffic class zero, because we don't actually use them.
2446 	 */
2447 	bzero(&context, sizeof (struct i40e_vsi_context));
2448 	context.seid = i40e->i40e_vsi_id;
2449 	context.pf_num = hw->pf_id;
2450 	err = i40e_aq_get_vsi_params(hw, &context, NULL);
2451 	if (err != I40E_SUCCESS) {
2452 		i40e_error(i40e, "get VSI params failed with %d", err);
2453 		return (B_FALSE);
2454 	}
2455 	tctx.rdylist = LE_16(context.info.qs_handle[0]);
2456 
2457 	err = i40e_clear_lan_tx_queue_context(hw, itrq->itrq_index);
2458 	if (err != I40E_SUCCESS) {
2459 		i40e_error(i40e, "failed to clear tx queue %d context: %d",
2460 		    itrq->itrq_index, err);
2461 		return (B_FALSE);
2462 	}
2463 
2464 	err = i40e_set_lan_tx_queue_context(hw, itrq->itrq_index, &tctx);
2465 	if (err != I40E_SUCCESS) {
2466 		i40e_error(i40e, "failed to set tx queue %d context: %d",
2467 		    itrq->itrq_index, err);
2468 		return (B_FALSE);
2469 	}
2470 
2471 	return (B_TRUE);
2472 }
2473 
2474 /*
2475  * Take care of setting up the descriptor rings and actually programming the
2476  * device. See 8.4.3.1.1 for what we need to do here.
2477  */
2478 static boolean_t
2479 i40e_setup_tx_rings(i40e_t *i40e)
2480 {
2481 	int i;
2482 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2483 
2484 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2485 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
2486 		uint32_t reg;
2487 
2488 		/*
2489 		 * Step 1. Clear the queue disable flag and verify that the
2490 		 * index is set correctly.
2491 		 */
2492 		i40e_pre_tx_queue_cfg(hw, i, B_TRUE);
2493 
2494 		/*
2495 		 * Step 2. Prepare the queue's FPM/HMC context.
2496 		 */
2497 		if (i40e_setup_tx_hmc(itrq) == B_FALSE)
2498 			return (B_FALSE);
2499 
2500 		/*
2501 		 * Step 3. Verify that it's clear that this PF owns this queue.
2502 		 */
2503 		reg = I40E_QTX_CTL_PF_QUEUE;
2504 		reg |= (hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2505 		    I40E_QTX_CTL_PF_INDX_MASK;
2506 		I40E_WRITE_REG(hw, I40E_QTX_CTL(itrq->itrq_index), reg);
2507 		i40e_flush(hw);
2508 
2509 		/*
2510 		 * Step 4. Set the QENA_REQ flag.
2511 		 */
2512 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2513 		VERIFY0(reg & (I40E_QTX_ENA_QENA_REQ_MASK |
2514 		    I40E_QTX_ENA_QENA_STAT_MASK));
2515 		reg |= I40E_QTX_ENA_QENA_REQ_MASK;
2516 		I40E_WRITE_REG(hw, I40E_QTX_ENA(i), reg);
2517 	}
2518 
2519 	/*
2520 	 * Note, we wait for every queue to be enabled before we start checking.
2521 	 * This will hopefully cause most queues to be enabled at this point.
2522 	 */
2523 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2524 		uint32_t j, reg;
2525 
2526 		/*
2527 		 * Step 5. Verify that QENA_STAT has been set. It's promised
2528 		 * that this should occur within about 10 us, but like BSD,
2529 		 * we'll try for up to 100 ms for this queue.
2530 		 */
2531 		for (j = 0; j < I40E_RING_WAIT_NTRIES; j++) {
2532 			reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2533 
2534 			if (reg & I40E_QTX_ENA_QENA_STAT_MASK)
2535 				break;
2536 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2537 		}
2538 
2539 		if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0) {
2540 			i40e_error(i40e, "failed to enable tx queue %d, timed "
2541 			    "out", i);
2542 			return (B_FALSE);
2543 		}
2544 	}
2545 
2546 	return (B_TRUE);
2547 }
2548 
2549 void
2550 i40e_stop(i40e_t *i40e, boolean_t free_allocations)
2551 {
2552 	int i;
2553 
2554 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
2555 
2556 	/*
2557 	 * Shutdown and drain the tx and rx pipeline. We do this using the
2558 	 * following steps.
2559 	 *
2560 	 * 1) Shutdown interrupts to all the queues (trying to keep the admin
2561 	 *    queue alive).
2562 	 *
2563 	 * 2) Remove all of the interrupt tx and rx causes by setting the
2564 	 *    interrupt linked lists to zero.
2565 	 *
2566 	 * 2) Shutdown the tx and rx rings. Because i40e_shutdown_rings() should
2567 	 *    wait for all the queues to be disabled, once we reach that point
2568 	 *    it should be safe to free associated data.
2569 	 *
2570 	 * 4) Wait 50ms after all that is done. This ensures that the rings are
2571 	 *    ready for programming again and we don't have to think about this
2572 	 *    in other parts of the driver.
2573 	 *
2574 	 * 5) Disable remaining chip interrupts, (admin queue, etc.)
2575 	 *
2576 	 * 6) Verify that FM is happy with all the register accesses we
2577 	 *    performed.
2578 	 */
2579 	i40e_intr_io_disable_all(i40e);
2580 	i40e_intr_io_clear_cause(i40e);
2581 
2582 	if (i40e_shutdown_rings(i40e) == B_FALSE) {
2583 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2584 	}
2585 
2586 	delay(50 * drv_usectohz(1000));
2587 
2588 	i40e_intr_chip_fini(i40e);
2589 
2590 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2591 		mutex_enter(&i40e->i40e_trqpairs[i].itrq_rx_lock);
2592 		mutex_enter(&i40e->i40e_trqpairs[i].itrq_tx_lock);
2593 	}
2594 
2595 	/*
2596 	 * We should consider refactoring this to be part of the ring start /
2597 	 * stop routines at some point.
2598 	 */
2599 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2600 		i40e_stats_trqpair_fini(&i40e->i40e_trqpairs[i]);
2601 	}
2602 
2603 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
2604 	    DDI_FM_OK) {
2605 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2606 	}
2607 
2608 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2609 		i40e_tx_cleanup_ring(&i40e->i40e_trqpairs[i]);
2610 	}
2611 
2612 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2613 		mutex_exit(&i40e->i40e_trqpairs[i].itrq_rx_lock);
2614 		mutex_exit(&i40e->i40e_trqpairs[i].itrq_tx_lock);
2615 	}
2616 
2617 	i40e_stat_vsi_fini(i40e);
2618 
2619 	i40e->i40e_link_speed = 0;
2620 	i40e->i40e_link_duplex = 0;
2621 	i40e_link_state_set(i40e, LINK_STATE_UNKNOWN);
2622 
2623 	if (free_allocations) {
2624 		i40e_free_ring_mem(i40e, B_FALSE);
2625 	}
2626 }
2627 
2628 boolean_t
2629 i40e_start(i40e_t *i40e, boolean_t alloc)
2630 {
2631 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2632 	boolean_t rc = B_TRUE;
2633 	int i, err;
2634 
2635 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
2636 
2637 	if (alloc) {
2638 		if (i40e_alloc_ring_mem(i40e) == B_FALSE) {
2639 			i40e_error(i40e,
2640 			    "Failed to allocate ring memory");
2641 			return (B_FALSE);
2642 		}
2643 	}
2644 
2645 	/*
2646 	 * This should get refactored to be part of ring start and stop at
2647 	 * some point, along with most of the logic here.
2648 	 */
2649 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2650 		if (i40e_stats_trqpair_init(&i40e->i40e_trqpairs[i]) ==
2651 		    B_FALSE) {
2652 			int j;
2653 
2654 			for (j = 0; j < i; j++) {
2655 				i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[j];
2656 				i40e_stats_trqpair_fini(itrq);
2657 			}
2658 			return (B_FALSE);
2659 		}
2660 	}
2661 
2662 	if (!i40e_chip_start(i40e)) {
2663 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
2664 		rc = B_FALSE;
2665 		goto done;
2666 	}
2667 
2668 	if (i40e_setup_rx_rings(i40e) == B_FALSE) {
2669 		rc = B_FALSE;
2670 		goto done;
2671 	}
2672 
2673 	if (i40e_setup_tx_rings(i40e) == B_FALSE) {
2674 		rc = B_FALSE;
2675 		goto done;
2676 	}
2677 
2678 	/*
2679 	 * Enable broadcast traffic; however, do not enable multicast traffic.
2680 	 * That's handle exclusively through MAC's mc_multicst routines.
2681 	 */
2682 	err = i40e_aq_set_vsi_broadcast(hw, i40e->i40e_vsi_id, B_TRUE, NULL);
2683 	if (err != I40E_SUCCESS) {
2684 		i40e_error(i40e, "failed to set default VSI: %d", err);
2685 		rc = B_FALSE;
2686 		goto done;
2687 	}
2688 
2689 	err = i40e_aq_set_mac_config(hw, i40e->i40e_frame_max, B_TRUE, 0, NULL);
2690 	if (err != I40E_SUCCESS) {
2691 		i40e_error(i40e, "failed to set MAC config: %d", err);
2692 		rc = B_FALSE;
2693 		goto done;
2694 	}
2695 
2696 	/*
2697 	 * Finally, make sure that we're happy from an FM perspective.
2698 	 */
2699 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_reg_handle) !=
2700 	    DDI_FM_OK) {
2701 		rc = B_FALSE;
2702 		goto done;
2703 	}
2704 
2705 	/* Clear state bits prior to final interrupt enabling. */
2706 	atomic_and_32(&i40e->i40e_state,
2707 	    ~(I40E_ERROR | I40E_STALL | I40E_OVERTEMP));
2708 
2709 	i40e_intr_io_enable_all(i40e);
2710 
2711 done:
2712 	if (rc == B_FALSE) {
2713 		i40e_stop(i40e, B_FALSE);
2714 		if (alloc == B_TRUE) {
2715 			i40e_free_ring_mem(i40e, B_TRUE);
2716 		}
2717 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2718 	}
2719 
2720 	return (rc);
2721 }
2722 
2723 /*
2724  * We may have loaned up descriptors to the stack. As such, if we still have
2725  * them outstanding, then we will not continue with detach.
2726  */
2727 static boolean_t
2728 i40e_drain_rx(i40e_t *i40e)
2729 {
2730 	mutex_enter(&i40e->i40e_rx_pending_lock);
2731 	while (i40e->i40e_rx_pending > 0) {
2732 		if (cv_reltimedwait(&i40e->i40e_rx_pending_cv,
2733 		    &i40e->i40e_rx_pending_lock,
2734 		    drv_usectohz(I40E_DRAIN_RX_WAIT), TR_CLOCK_TICK) == -1) {
2735 			mutex_exit(&i40e->i40e_rx_pending_lock);
2736 			return (B_FALSE);
2737 		}
2738 	}
2739 	mutex_exit(&i40e->i40e_rx_pending_lock);
2740 
2741 	return (B_TRUE);
2742 }
2743 
2744 static int
2745 i40e_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
2746 {
2747 	i40e_t *i40e;
2748 	struct i40e_osdep *osdep;
2749 	i40e_hw_t *hw;
2750 	int instance;
2751 
2752 	if (cmd != DDI_ATTACH)
2753 		return (DDI_FAILURE);
2754 
2755 	instance = ddi_get_instance(devinfo);
2756 	i40e = kmem_zalloc(sizeof (i40e_t), KM_SLEEP);
2757 
2758 	i40e->i40e_aqbuf = kmem_zalloc(I40E_ADMINQ_BUFSZ, KM_SLEEP);
2759 	i40e->i40e_instance = instance;
2760 	i40e->i40e_dip = devinfo;
2761 
2762 	hw = &i40e->i40e_hw_space;
2763 	osdep = &i40e->i40e_osdep_space;
2764 	hw->back = osdep;
2765 	osdep->ios_i40e = i40e;
2766 
2767 	ddi_set_driver_private(devinfo, i40e);
2768 
2769 	i40e_fm_init(i40e);
2770 	i40e->i40e_attach_progress |= I40E_ATTACH_FM_INIT;
2771 
2772 	if (pci_config_setup(devinfo, &osdep->ios_cfg_handle) != DDI_SUCCESS) {
2773 		i40e_error(i40e, "Failed to map PCI configurations.");
2774 		goto attach_fail;
2775 	}
2776 	i40e->i40e_attach_progress |= I40E_ATTACH_PCI_CONFIG;
2777 
2778 	i40e_identify_hardware(i40e);
2779 
2780 	if (!i40e_regs_map(i40e)) {
2781 		i40e_error(i40e, "Failed to map device registers.");
2782 		goto attach_fail;
2783 	}
2784 	i40e->i40e_attach_progress |= I40E_ATTACH_REGS_MAP;
2785 
2786 	i40e_init_properties(i40e);
2787 	i40e->i40e_attach_progress |= I40E_ATTACH_PROPS;
2788 
2789 	if (!i40e_common_code_init(i40e, hw))
2790 		goto attach_fail;
2791 	i40e->i40e_attach_progress |= I40E_ATTACH_COMMON_CODE;
2792 
2793 	/*
2794 	 * When we participate in IRM, we should make sure that we register
2795 	 * ourselves with it before callbacks.
2796 	 */
2797 	if (!i40e_alloc_intrs(i40e, devinfo)) {
2798 		i40e_error(i40e, "Failed to allocate interrupts.");
2799 		goto attach_fail;
2800 	}
2801 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_INTR;
2802 
2803 	if (!i40e_alloc_trqpairs(i40e)) {
2804 		i40e_error(i40e,
2805 		    "Failed to allocate receive & transmit rings.");
2806 		goto attach_fail;
2807 	}
2808 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_RINGSLOCKS;
2809 
2810 	if (!i40e_map_intrs_to_vectors(i40e)) {
2811 		i40e_error(i40e, "Failed to map interrupts to vectors.");
2812 		goto attach_fail;
2813 	}
2814 
2815 	if (!i40e_add_intr_handlers(i40e)) {
2816 		i40e_error(i40e, "Failed to add the interrupt handlers.");
2817 		goto attach_fail;
2818 	}
2819 	i40e->i40e_attach_progress |= I40E_ATTACH_ADD_INTR;
2820 
2821 	if (!i40e_final_init(i40e)) {
2822 		i40e_error(i40e, "Final initialization failed.");
2823 		goto attach_fail;
2824 	}
2825 	i40e->i40e_attach_progress |= I40E_ATTACH_INIT;
2826 
2827 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
2828 	    DDI_FM_OK) {
2829 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2830 		goto attach_fail;
2831 	}
2832 
2833 	if (!i40e_stats_init(i40e)) {
2834 		i40e_error(i40e, "Stats initialization failed.");
2835 		goto attach_fail;
2836 	}
2837 	i40e->i40e_attach_progress |= I40E_ATTACH_STATS;
2838 
2839 	if (!i40e_register_mac(i40e)) {
2840 		i40e_error(i40e, "Failed to register to MAC/GLDv3");
2841 		goto attach_fail;
2842 	}
2843 	i40e->i40e_attach_progress |= I40E_ATTACH_MAC;
2844 
2845 	i40e->i40e_periodic_id = ddi_periodic_add(i40e_timer, i40e,
2846 	    I40E_CYCLIC_PERIOD, DDI_IPL_0);
2847 	if (i40e->i40e_periodic_id == 0) {
2848 		i40e_error(i40e, "Failed to add the link-check timer");
2849 		goto attach_fail;
2850 	}
2851 	i40e->i40e_attach_progress |= I40E_ATTACH_LINK_TIMER;
2852 
2853 	if (!i40e_enable_interrupts(i40e)) {
2854 		i40e_error(i40e, "Failed to enable DDI interrupts");
2855 		goto attach_fail;
2856 	}
2857 	i40e->i40e_attach_progress |= I40E_ATTACH_ENABLE_INTR;
2858 
2859 	atomic_or_32(&i40e->i40e_state, I40E_INITIALIZED);
2860 
2861 	mutex_enter(&i40e_glock);
2862 	list_insert_tail(&i40e_glist, i40e);
2863 	mutex_exit(&i40e_glock);
2864 
2865 	return (DDI_SUCCESS);
2866 
2867 attach_fail:
2868 	i40e_unconfigure(devinfo, i40e);
2869 	return (DDI_FAILURE);
2870 }
2871 
2872 static int
2873 i40e_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
2874 {
2875 	i40e_t *i40e;
2876 
2877 	if (cmd != DDI_DETACH)
2878 		return (DDI_FAILURE);
2879 
2880 	i40e = (i40e_t *)ddi_get_driver_private(devinfo);
2881 	if (i40e == NULL) {
2882 		i40e_log(NULL, "i40e_detach() called with no i40e pointer!");
2883 		return (DDI_FAILURE);
2884 	}
2885 
2886 	if (i40e_drain_rx(i40e) == B_FALSE) {
2887 		i40e_log(i40e, "timed out draining DMA resources, %d buffers "
2888 		    "remain", i40e->i40e_rx_pending);
2889 		return (DDI_FAILURE);
2890 	}
2891 
2892 	mutex_enter(&i40e_glock);
2893 	list_remove(&i40e_glist, i40e);
2894 	mutex_exit(&i40e_glock);
2895 
2896 	i40e_unconfigure(devinfo, i40e);
2897 
2898 	return (DDI_SUCCESS);
2899 }
2900 
2901 static struct cb_ops i40e_cb_ops = {
2902 	nulldev,		/* cb_open */
2903 	nulldev,		/* cb_close */
2904 	nodev,			/* cb_strategy */
2905 	nodev,			/* cb_print */
2906 	nodev,			/* cb_dump */
2907 	nodev,			/* cb_read */
2908 	nodev,			/* cb_write */
2909 	nodev,			/* cb_ioctl */
2910 	nodev,			/* cb_devmap */
2911 	nodev,			/* cb_mmap */
2912 	nodev,			/* cb_segmap */
2913 	nochpoll,		/* cb_chpoll */
2914 	ddi_prop_op,		/* cb_prop_op */
2915 	NULL,			/* cb_stream */
2916 	D_MP | D_HOTPLUG,	/* cb_flag */
2917 	CB_REV,			/* cb_rev */
2918 	nodev,			/* cb_aread */
2919 	nodev			/* cb_awrite */
2920 };
2921 
2922 static struct dev_ops i40e_dev_ops = {
2923 	DEVO_REV,		/* devo_rev */
2924 	0,			/* devo_refcnt */
2925 	NULL,			/* devo_getinfo */
2926 	nulldev,		/* devo_identify */
2927 	nulldev,		/* devo_probe */
2928 	i40e_attach,		/* devo_attach */
2929 	i40e_detach,		/* devo_detach */
2930 	nodev,			/* devo_reset */
2931 	&i40e_cb_ops,		/* devo_cb_ops */
2932 	NULL,			/* devo_bus_ops */
2933 	ddi_power,		/* devo_power */
2934 	ddi_quiesce_not_supported /* devo_quiesce */
2935 };
2936 
2937 static struct modldrv i40e_modldrv = {
2938 	&mod_driverops,
2939 	i40e_ident,
2940 	&i40e_dev_ops
2941 };
2942 
2943 static struct modlinkage i40e_modlinkage = {
2944 	MODREV_1,
2945 	&i40e_modldrv,
2946 	NULL
2947 };
2948 
2949 /*
2950  * Module Initialization Functions.
2951  */
2952 int
2953 _init(void)
2954 {
2955 	int status;
2956 
2957 	list_create(&i40e_glist, sizeof (i40e_t), offsetof(i40e_t, i40e_glink));
2958 	list_create(&i40e_dlist, sizeof (i40e_device_t),
2959 	    offsetof(i40e_device_t, id_link));
2960 	mutex_init(&i40e_glock, NULL, MUTEX_DRIVER, NULL);
2961 	mac_init_ops(&i40e_dev_ops, I40E_MODULE_NAME);
2962 
2963 	status = mod_install(&i40e_modlinkage);
2964 	if (status != DDI_SUCCESS) {
2965 		mac_fini_ops(&i40e_dev_ops);
2966 		mutex_destroy(&i40e_glock);
2967 		list_destroy(&i40e_dlist);
2968 		list_destroy(&i40e_glist);
2969 	}
2970 
2971 	return (status);
2972 }
2973 
2974 int
2975 _info(struct modinfo *modinfop)
2976 {
2977 	return (mod_info(&i40e_modlinkage, modinfop));
2978 }
2979 
2980 int
2981 _fini(void)
2982 {
2983 	int status;
2984 
2985 	status = mod_remove(&i40e_modlinkage);
2986 	if (status == DDI_SUCCESS) {
2987 		mac_fini_ops(&i40e_dev_ops);
2988 		mutex_destroy(&i40e_glock);
2989 		list_destroy(&i40e_dlist);
2990 		list_destroy(&i40e_glist);
2991 	}
2992 
2993 	return (status);
2994 }
2995