xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_main.c (revision b1e2e3fb17324e9ddf43db264a0c64da7756d9e6)
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 (c) 2017, 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  * have 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 promiscuous 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 /*
453  * Various parts of the driver need to know if the controller is from the X722
454  * family, which has a few additional capabilities and different programming
455  * means. We don't consider virtual functions as part of this as they are quite
456  * different and will require substantially more work.
457  */
458 static boolean_t
459 i40e_is_x722(i40e_t *i40e)
460 {
461 	return (i40e->i40e_hw_space.mac.type == I40E_MAC_X722);
462 }
463 
464 static void
465 i40e_device_rele(i40e_t *i40e)
466 {
467 	i40e_device_t *idp = i40e->i40e_device;
468 
469 	if (idp == NULL)
470 		return;
471 
472 	mutex_enter(&i40e_glock);
473 	VERIFY(idp->id_nreg > 0);
474 	list_remove(&idp->id_i40e_list, i40e);
475 	idp->id_nreg--;
476 	if (idp->id_nreg == 0) {
477 		list_remove(&i40e_dlist, idp);
478 		list_destroy(&idp->id_i40e_list);
479 		kmem_free(idp->id_rsrcs, sizeof (i40e_switch_rsrc_t) *
480 		    idp->id_rsrcs_alloc);
481 		kmem_free(idp, sizeof (i40e_device_t));
482 	}
483 	i40e->i40e_device = NULL;
484 	mutex_exit(&i40e_glock);
485 }
486 
487 static i40e_device_t *
488 i40e_device_find(i40e_t *i40e, dev_info_t *parent, uint_t bus, uint_t device)
489 {
490 	i40e_device_t *idp;
491 	mutex_enter(&i40e_glock);
492 	for (idp = list_head(&i40e_dlist); idp != NULL;
493 	    idp = list_next(&i40e_dlist, idp)) {
494 		if (idp->id_parent == parent && idp->id_pci_bus == bus &&
495 		    idp->id_pci_device == device) {
496 			break;
497 		}
498 	}
499 
500 	if (idp != NULL) {
501 		VERIFY(idp->id_nreg < idp->id_nfuncs);
502 		idp->id_nreg++;
503 	} else {
504 		i40e_hw_t *hw = &i40e->i40e_hw_space;
505 		ASSERT(hw->num_ports > 0);
506 		ASSERT(hw->num_partitions > 0);
507 
508 		/*
509 		 * The Intel common code doesn't exactly keep the number of PCI
510 		 * functions. But it calculates it during discovery of
511 		 * partitions and ports. So what we do is undo the calculation
512 		 * that it does originally, as functions are evenly spread
513 		 * across ports in the rare case of partitions.
514 		 */
515 		idp = kmem_alloc(sizeof (i40e_device_t), KM_SLEEP);
516 		idp->id_parent = parent;
517 		idp->id_pci_bus = bus;
518 		idp->id_pci_device = device;
519 		idp->id_nfuncs = hw->num_ports * hw->num_partitions;
520 		idp->id_nreg = 1;
521 		idp->id_rsrcs_alloc = i40e->i40e_switch_rsrc_alloc;
522 		idp->id_rsrcs_act = i40e->i40e_switch_rsrc_actual;
523 		idp->id_rsrcs = kmem_alloc(sizeof (i40e_switch_rsrc_t) *
524 		    idp->id_rsrcs_alloc, KM_SLEEP);
525 		bcopy(i40e->i40e_switch_rsrcs, idp->id_rsrcs,
526 		    sizeof (i40e_switch_rsrc_t) * idp->id_rsrcs_alloc);
527 		list_create(&idp->id_i40e_list, sizeof (i40e_t),
528 		    offsetof(i40e_t, i40e_dlink));
529 
530 		list_insert_tail(&i40e_dlist, idp);
531 	}
532 
533 	list_insert_tail(&idp->id_i40e_list, i40e);
534 	mutex_exit(&i40e_glock);
535 
536 	return (idp);
537 }
538 
539 static void
540 i40e_link_state_set(i40e_t *i40e, link_state_t state)
541 {
542 	if (i40e->i40e_link_state == state)
543 		return;
544 
545 	i40e->i40e_link_state = state;
546 	mac_link_update(i40e->i40e_mac_hdl, i40e->i40e_link_state);
547 }
548 
549 /*
550  * This is a basic link check routine. Mostly we're using this just to see
551  * if we can get any accurate information about the state of the link being
552  * up or down, as well as updating the link state, speed, etc. information.
553  */
554 void
555 i40e_link_check(i40e_t *i40e)
556 {
557 	i40e_hw_t *hw = &i40e->i40e_hw_space;
558 	boolean_t ls;
559 	int ret;
560 
561 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
562 
563 	hw->phy.get_link_info = B_TRUE;
564 	if ((ret = i40e_get_link_status(hw, &ls)) != I40E_SUCCESS) {
565 		i40e->i40e_s_link_status_errs++;
566 		i40e->i40e_s_link_status_lasterr = ret;
567 		return;
568 	}
569 
570 	/*
571 	 * Firmware abstracts all of the mac and phy information for us, so we
572 	 * can use i40e_get_link_status to determine the current state.
573 	 */
574 	if (ls == B_TRUE) {
575 		enum i40e_aq_link_speed speed;
576 
577 		speed = i40e_get_link_speed(hw);
578 
579 		/*
580 		 * Translate from an i40e value to a value in Mbits/s.
581 		 */
582 		switch (speed) {
583 		case I40E_LINK_SPEED_100MB:
584 			i40e->i40e_link_speed = 100;
585 			break;
586 		case I40E_LINK_SPEED_1GB:
587 			i40e->i40e_link_speed = 1000;
588 			break;
589 		case I40E_LINK_SPEED_10GB:
590 			i40e->i40e_link_speed = 10000;
591 			break;
592 		case I40E_LINK_SPEED_20GB:
593 			i40e->i40e_link_speed = 20000;
594 			break;
595 		case I40E_LINK_SPEED_40GB:
596 			i40e->i40e_link_speed = 40000;
597 			break;
598 		case I40E_LINK_SPEED_25GB:
599 			i40e->i40e_link_speed = 25000;
600 			break;
601 		default:
602 			i40e->i40e_link_speed = 0;
603 			break;
604 		}
605 
606 		/*
607 		 * At this time, hardware does not support half-duplex
608 		 * operation, hence why we don't ask the hardware about our
609 		 * current speed.
610 		 */
611 		i40e->i40e_link_duplex = LINK_DUPLEX_FULL;
612 		i40e_link_state_set(i40e, LINK_STATE_UP);
613 	} else {
614 		i40e->i40e_link_speed = 0;
615 		i40e->i40e_link_duplex = 0;
616 		i40e_link_state_set(i40e, LINK_STATE_DOWN);
617 	}
618 }
619 
620 static void
621 i40e_rem_intrs(i40e_t *i40e)
622 {
623 	int i, rc;
624 
625 	for (i = 0; i < i40e->i40e_intr_count; i++) {
626 		rc = ddi_intr_free(i40e->i40e_intr_handles[i]);
627 		if (rc != DDI_SUCCESS) {
628 			i40e_log(i40e, "failed to free interrupt %d: %d",
629 			    i, rc);
630 		}
631 	}
632 
633 	kmem_free(i40e->i40e_intr_handles, i40e->i40e_intr_size);
634 	i40e->i40e_intr_handles = NULL;
635 }
636 
637 static void
638 i40e_rem_intr_handlers(i40e_t *i40e)
639 {
640 	int i, rc;
641 
642 	for (i = 0; i < i40e->i40e_intr_count; i++) {
643 		rc = ddi_intr_remove_handler(i40e->i40e_intr_handles[i]);
644 		if (rc != DDI_SUCCESS) {
645 			i40e_log(i40e, "failed to remove interrupt %d: %d",
646 			    i, rc);
647 		}
648 	}
649 }
650 
651 /*
652  * illumos Fault Management Architecture (FMA) support.
653  */
654 
655 int
656 i40e_check_acc_handle(ddi_acc_handle_t handle)
657 {
658 	ddi_fm_error_t de;
659 
660 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
661 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
662 	return (de.fme_status);
663 }
664 
665 int
666 i40e_check_dma_handle(ddi_dma_handle_t handle)
667 {
668 	ddi_fm_error_t de;
669 
670 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
671 	return (de.fme_status);
672 }
673 
674 /*
675  * Fault service error handling callback function.
676  */
677 /* ARGSUSED */
678 static int
679 i40e_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
680 {
681 	pci_ereport_post(dip, err, NULL);
682 	return (err->fme_status);
683 }
684 
685 static void
686 i40e_fm_init(i40e_t *i40e)
687 {
688 	ddi_iblock_cookie_t iblk;
689 
690 	i40e->i40e_fm_capabilities = ddi_prop_get_int(DDI_DEV_T_ANY,
691 	    i40e->i40e_dip, DDI_PROP_DONTPASS, "fm_capable",
692 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
693 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
694 
695 	if (i40e->i40e_fm_capabilities < 0) {
696 		i40e->i40e_fm_capabilities = 0;
697 	} else if (i40e->i40e_fm_capabilities > 0xf) {
698 		i40e->i40e_fm_capabilities = DDI_FM_EREPORT_CAPABLE |
699 		    DDI_FM_ACCCHK_CAPABLE | DDI_FM_DMACHK_CAPABLE |
700 		    DDI_FM_ERRCB_CAPABLE;
701 	}
702 
703 	/*
704 	 * Only register with IO Fault Services if we have some capability
705 	 */
706 	if (i40e->i40e_fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
707 		i40e_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
708 	} else {
709 		i40e_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
710 	}
711 
712 	if (i40e->i40e_fm_capabilities) {
713 		ddi_fm_init(i40e->i40e_dip, &i40e->i40e_fm_capabilities, &iblk);
714 
715 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
716 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
717 			pci_ereport_setup(i40e->i40e_dip);
718 		}
719 
720 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities)) {
721 			ddi_fm_handler_register(i40e->i40e_dip,
722 			    i40e_fm_error_cb, (void*)i40e);
723 		}
724 	}
725 
726 	if (i40e->i40e_fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
727 		i40e_init_dma_attrs(i40e, B_TRUE);
728 	} else {
729 		i40e_init_dma_attrs(i40e, B_FALSE);
730 	}
731 }
732 
733 static void
734 i40e_fm_fini(i40e_t *i40e)
735 {
736 	if (i40e->i40e_fm_capabilities) {
737 
738 		if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities) ||
739 		    DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
740 			pci_ereport_teardown(i40e->i40e_dip);
741 
742 		if (DDI_FM_ERRCB_CAP(i40e->i40e_fm_capabilities))
743 			ddi_fm_handler_unregister(i40e->i40e_dip);
744 
745 		ddi_fm_fini(i40e->i40e_dip);
746 	}
747 }
748 
749 void
750 i40e_fm_ereport(i40e_t *i40e, char *detail)
751 {
752 	uint64_t ena;
753 	char buf[FM_MAX_CLASS];
754 
755 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
756 	ena = fm_ena_generate(0, FM_ENA_FMT1);
757 	if (DDI_FM_EREPORT_CAP(i40e->i40e_fm_capabilities)) {
758 		ddi_fm_ereport_post(i40e->i40e_dip, buf, ena, DDI_NOSLEEP,
759 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
760 	}
761 }
762 
763 /*
764  * Here we're trying to get the ID of the default VSI. In general, when we come
765  * through and look at this shortly after attach, we expect there to only be a
766  * single element present, which is the default VSI. Importantly, each PF seems
767  * to not see any other devices, in part because of the simple switch mode that
768  * we're using. If for some reason, we see more artifact, we'll need to revisit
769  * what we're doing here.
770  */
771 static int
772 i40e_get_vsi_id(i40e_t *i40e)
773 {
774 	i40e_hw_t *hw = &i40e->i40e_hw_space;
775 	struct i40e_aqc_get_switch_config_resp *sw_config;
776 	uint8_t aq_buf[I40E_AQ_LARGE_BUF];
777 	uint16_t next = 0;
778 	int rc;
779 
780 	/* LINTED: E_BAD_PTR_CAST_ALIGN */
781 	sw_config = (struct i40e_aqc_get_switch_config_resp *)aq_buf;
782 	rc = i40e_aq_get_switch_config(hw, sw_config, sizeof (aq_buf), &next,
783 	    NULL);
784 	if (rc != I40E_SUCCESS) {
785 		i40e_error(i40e, "i40e_aq_get_switch_config() failed %d: %d",
786 		    rc, hw->aq.asq_last_status);
787 		return (-1);
788 	}
789 
790 	if (LE_16(sw_config->header.num_reported) != 1) {
791 		i40e_error(i40e, "encountered multiple (%d) switching units "
792 		    "during attach, not proceeding",
793 		    LE_16(sw_config->header.num_reported));
794 		return (-1);
795 	}
796 
797 	return (sw_config->element[0].seid);
798 }
799 
800 /*
801  * We need to fill the i40e_hw_t structure with the capabilities of this PF. We
802  * must also provide the memory for it; however, we don't need to keep it around
803  * to the call to the common code. It takes it and parses it into an internal
804  * structure.
805  */
806 static boolean_t
807 i40e_get_hw_capabilities(i40e_t *i40e, i40e_hw_t *hw)
808 {
809 	struct i40e_aqc_list_capabilities_element_resp *buf;
810 	int rc;
811 	size_t len;
812 	uint16_t needed;
813 	int nelems = I40E_HW_CAP_DEFAULT;
814 
815 	len = nelems * sizeof (*buf);
816 
817 	for (;;) {
818 		ASSERT(len > 0);
819 		buf = kmem_alloc(len, KM_SLEEP);
820 		rc = i40e_aq_discover_capabilities(hw, buf, len,
821 		    &needed, i40e_aqc_opc_list_func_capabilities, NULL);
822 		kmem_free(buf, len);
823 
824 		if (hw->aq.asq_last_status == I40E_AQ_RC_ENOMEM &&
825 		    nelems == I40E_HW_CAP_DEFAULT) {
826 			if (nelems == needed) {
827 				i40e_error(i40e, "Capability discovery failed "
828 				    "due to byzantine common code");
829 				return (B_FALSE);
830 			}
831 			len = needed;
832 			continue;
833 		} else if (rc != I40E_SUCCESS ||
834 		    hw->aq.asq_last_status != I40E_AQ_RC_OK) {
835 			i40e_error(i40e, "Capability discovery failed: %d", rc);
836 			return (B_FALSE);
837 		}
838 
839 		break;
840 	}
841 
842 	return (B_TRUE);
843 }
844 
845 /*
846  * Obtain the switch's capabilities as seen by this PF and keep it around for
847  * our later use.
848  */
849 static boolean_t
850 i40e_get_switch_resources(i40e_t *i40e)
851 {
852 	i40e_hw_t *hw = &i40e->i40e_hw_space;
853 	uint8_t cnt = 2;
854 	uint8_t act;
855 	size_t size;
856 	i40e_switch_rsrc_t *buf;
857 
858 	for (;;) {
859 		enum i40e_status_code ret;
860 		size = cnt * sizeof (i40e_switch_rsrc_t);
861 		ASSERT(size > 0);
862 		if (size > UINT16_MAX)
863 			return (B_FALSE);
864 		buf = kmem_alloc(size, KM_SLEEP);
865 
866 		ret = i40e_aq_get_switch_resource_alloc(hw, &act, buf,
867 		    cnt, NULL);
868 		if (ret == I40E_ERR_ADMIN_QUEUE_ERROR &&
869 		    hw->aq.asq_last_status == I40E_AQ_RC_EINVAL) {
870 			kmem_free(buf, size);
871 			cnt += I40E_SWITCH_CAP_DEFAULT;
872 			continue;
873 		} else if (ret != I40E_SUCCESS) {
874 			kmem_free(buf, size);
875 			i40e_error(i40e,
876 			    "failed to retrieve switch statistics: %d", ret);
877 			return (B_FALSE);
878 		}
879 
880 		break;
881 	}
882 
883 	i40e->i40e_switch_rsrc_alloc = cnt;
884 	i40e->i40e_switch_rsrc_actual = act;
885 	i40e->i40e_switch_rsrcs = buf;
886 
887 	return (B_TRUE);
888 }
889 
890 static void
891 i40e_cleanup_resources(i40e_t *i40e)
892 {
893 	if (i40e->i40e_uaddrs != NULL) {
894 		kmem_free(i40e->i40e_uaddrs, sizeof (i40e_uaddr_t) *
895 		    i40e->i40e_resources.ifr_nmacfilt);
896 		i40e->i40e_uaddrs = NULL;
897 	}
898 
899 	if (i40e->i40e_maddrs != NULL) {
900 		kmem_free(i40e->i40e_maddrs, sizeof (i40e_maddr_t) *
901 		    i40e->i40e_resources.ifr_nmcastfilt);
902 		i40e->i40e_maddrs = NULL;
903 	}
904 
905 	if (i40e->i40e_switch_rsrcs != NULL) {
906 		size_t sz = sizeof (i40e_switch_rsrc_t) *
907 		    i40e->i40e_switch_rsrc_alloc;
908 		ASSERT(sz > 0);
909 		kmem_free(i40e->i40e_switch_rsrcs, sz);
910 		i40e->i40e_switch_rsrcs = NULL;
911 	}
912 
913 	if (i40e->i40e_device != NULL)
914 		i40e_device_rele(i40e);
915 }
916 
917 static boolean_t
918 i40e_get_available_resources(i40e_t *i40e)
919 {
920 	dev_info_t *parent;
921 	uint16_t bus, device, func;
922 	uint_t nregs;
923 	int *regs, i;
924 	i40e_device_t *idp;
925 	i40e_hw_t *hw = &i40e->i40e_hw_space;
926 
927 	parent = ddi_get_parent(i40e->i40e_dip);
928 
929 	if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, i40e->i40e_dip, 0, "reg",
930 	    &regs, &nregs) != DDI_PROP_SUCCESS) {
931 		return (B_FALSE);
932 	}
933 
934 	if (nregs < 1) {
935 		ddi_prop_free(regs);
936 		return (B_FALSE);
937 	}
938 
939 	bus = PCI_REG_BUS_G(regs[0]);
940 	device = PCI_REG_DEV_G(regs[0]);
941 	func = PCI_REG_FUNC_G(regs[0]);
942 	ddi_prop_free(regs);
943 
944 	i40e->i40e_hw_space.bus.func = func;
945 	i40e->i40e_hw_space.bus.device = device;
946 
947 	if (i40e_get_switch_resources(i40e) == B_FALSE) {
948 		return (B_FALSE);
949 	}
950 
951 	/*
952 	 * To calculate the total amount of a resource we have available, we
953 	 * need to add how many our i40e_t thinks it has guaranteed, if any, and
954 	 * then we need to go through and divide the number of available on the
955 	 * device, which was snapshotted before anyone should have allocated
956 	 * anything, and use that to derive how many are available from the
957 	 * pool. Longer term, we may want to turn this into something that's
958 	 * more of a pool-like resource that everything can share (though that
959 	 * may require some more assistance from MAC).
960 	 *
961 	 * Though for transmit and receive queue pairs, we just have to ask
962 	 * firmware instead.
963 	 */
964 	idp = i40e_device_find(i40e, parent, bus, device);
965 	i40e->i40e_device = idp;
966 	i40e->i40e_resources.ifr_nvsis = 0;
967 	i40e->i40e_resources.ifr_nvsis_used = 0;
968 	i40e->i40e_resources.ifr_nmacfilt = 0;
969 	i40e->i40e_resources.ifr_nmacfilt_used = 0;
970 	i40e->i40e_resources.ifr_nmcastfilt = 0;
971 	i40e->i40e_resources.ifr_nmcastfilt_used = 0;
972 
973 	for (i = 0; i < i40e->i40e_switch_rsrc_actual; i++) {
974 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
975 
976 		switch (srp->resource_type) {
977 		case I40E_AQ_RESOURCE_TYPE_VSI:
978 			i40e->i40e_resources.ifr_nvsis +=
979 			    LE_16(srp->guaranteed);
980 			i40e->i40e_resources.ifr_nvsis_used = LE_16(srp->used);
981 			break;
982 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
983 			i40e->i40e_resources.ifr_nmacfilt +=
984 			    LE_16(srp->guaranteed);
985 			i40e->i40e_resources.ifr_nmacfilt_used =
986 			    LE_16(srp->used);
987 			break;
988 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
989 			i40e->i40e_resources.ifr_nmcastfilt +=
990 			    LE_16(srp->guaranteed);
991 			i40e->i40e_resources.ifr_nmcastfilt_used =
992 			    LE_16(srp->used);
993 			break;
994 		default:
995 			break;
996 		}
997 	}
998 
999 	for (i = 0; i < idp->id_rsrcs_act; i++) {
1000 		i40e_switch_rsrc_t *srp = &i40e->i40e_switch_rsrcs[i];
1001 		switch (srp->resource_type) {
1002 		case I40E_AQ_RESOURCE_TYPE_VSI:
1003 			i40e->i40e_resources.ifr_nvsis +=
1004 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1005 			break;
1006 		case I40E_AQ_RESOURCE_TYPE_MACADDR:
1007 			i40e->i40e_resources.ifr_nmacfilt +=
1008 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1009 			break;
1010 		case I40E_AQ_RESOURCE_TYPE_MULTICAST_HASH:
1011 			i40e->i40e_resources.ifr_nmcastfilt +=
1012 			    LE_16(srp->total_unalloced) / idp->id_nfuncs;
1013 		default:
1014 			break;
1015 		}
1016 	}
1017 
1018 	i40e->i40e_resources.ifr_nrx_queue = hw->func_caps.num_rx_qp;
1019 	i40e->i40e_resources.ifr_ntx_queue = hw->func_caps.num_tx_qp;
1020 
1021 	i40e->i40e_uaddrs = kmem_zalloc(sizeof (i40e_uaddr_t) *
1022 	    i40e->i40e_resources.ifr_nmacfilt, KM_SLEEP);
1023 	i40e->i40e_maddrs = kmem_zalloc(sizeof (i40e_maddr_t) *
1024 	    i40e->i40e_resources.ifr_nmcastfilt, KM_SLEEP);
1025 
1026 	/*
1027 	 * Initialize these as multicast addresses to indicate it's invalid for
1028 	 * sanity purposes. Think of it like 0xdeadbeef.
1029 	 */
1030 	for (i = 0; i < i40e->i40e_resources.ifr_nmacfilt; i++)
1031 		i40e->i40e_uaddrs[i].iua_mac[0] = 0x01;
1032 
1033 	return (B_TRUE);
1034 }
1035 
1036 static boolean_t
1037 i40e_enable_interrupts(i40e_t *i40e)
1038 {
1039 	int i, rc;
1040 
1041 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1042 		rc = ddi_intr_block_enable(i40e->i40e_intr_handles,
1043 		    i40e->i40e_intr_count);
1044 		if (rc != DDI_SUCCESS) {
1045 			i40e_error(i40e, "Interrupt block-enable failed: %d",
1046 			    rc);
1047 			return (B_FALSE);
1048 		}
1049 	} else {
1050 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1051 			rc = ddi_intr_enable(i40e->i40e_intr_handles[i]);
1052 			if (rc != DDI_SUCCESS) {
1053 				i40e_error(i40e,
1054 				    "Failed to enable interrupt %d: %d", i, rc);
1055 				while (--i >= 0) {
1056 					(void) ddi_intr_disable(
1057 					    i40e->i40e_intr_handles[i]);
1058 				}
1059 				return (B_FALSE);
1060 			}
1061 		}
1062 	}
1063 
1064 	return (B_TRUE);
1065 }
1066 
1067 static boolean_t
1068 i40e_disable_interrupts(i40e_t *i40e)
1069 {
1070 	int i, rc;
1071 
1072 	if (i40e->i40e_intr_cap & DDI_INTR_FLAG_BLOCK) {
1073 		rc = ddi_intr_block_disable(i40e->i40e_intr_handles,
1074 		    i40e->i40e_intr_count);
1075 		if (rc != DDI_SUCCESS) {
1076 			i40e_error(i40e,
1077 			    "Interrupt block-disabled failed: %d", rc);
1078 			return (B_FALSE);
1079 		}
1080 	} else {
1081 		for (i = 0; i < i40e->i40e_intr_count; i++) {
1082 			rc = ddi_intr_disable(i40e->i40e_intr_handles[i]);
1083 			if (rc != DDI_SUCCESS) {
1084 				i40e_error(i40e,
1085 				    "Failed to disable interrupt %d: %d",
1086 				    i, rc);
1087 				return (B_FALSE);
1088 			}
1089 		}
1090 	}
1091 
1092 	return (B_TRUE);
1093 }
1094 
1095 /*
1096  * Free receive & transmit rings.
1097  */
1098 static void
1099 i40e_free_trqpairs(i40e_t *i40e)
1100 {
1101 	int i;
1102 	i40e_trqpair_t *itrq;
1103 
1104 	if (i40e->i40e_trqpairs != NULL) {
1105 		for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1106 			itrq = &i40e->i40e_trqpairs[i];
1107 			mutex_destroy(&itrq->itrq_rx_lock);
1108 			mutex_destroy(&itrq->itrq_tx_lock);
1109 			mutex_destroy(&itrq->itrq_tcb_lock);
1110 
1111 			/*
1112 			 * Should have already been cleaned up by start/stop,
1113 			 * etc.
1114 			 */
1115 			ASSERT(itrq->itrq_txkstat == NULL);
1116 			ASSERT(itrq->itrq_rxkstat == NULL);
1117 		}
1118 
1119 		kmem_free(i40e->i40e_trqpairs,
1120 		    sizeof (i40e_trqpair_t) * i40e->i40e_num_trqpairs);
1121 		i40e->i40e_trqpairs = NULL;
1122 	}
1123 
1124 	cv_destroy(&i40e->i40e_rx_pending_cv);
1125 	mutex_destroy(&i40e->i40e_rx_pending_lock);
1126 	mutex_destroy(&i40e->i40e_general_lock);
1127 }
1128 
1129 /*
1130  * Allocate transmit and receive rings, as well as other data structures that we
1131  * need.
1132  */
1133 static boolean_t
1134 i40e_alloc_trqpairs(i40e_t *i40e)
1135 {
1136 	int i;
1137 	void *mutexpri = DDI_INTR_PRI(i40e->i40e_intr_pri);
1138 
1139 	/*
1140 	 * Now that we have the priority for the interrupts, initialize
1141 	 * all relevant locks.
1142 	 */
1143 	mutex_init(&i40e->i40e_general_lock, NULL, MUTEX_DRIVER, mutexpri);
1144 	mutex_init(&i40e->i40e_rx_pending_lock, NULL, MUTEX_DRIVER, mutexpri);
1145 	cv_init(&i40e->i40e_rx_pending_cv, NULL, CV_DRIVER, NULL);
1146 
1147 	i40e->i40e_trqpairs = kmem_zalloc(sizeof (i40e_trqpair_t) *
1148 	    i40e->i40e_num_trqpairs, KM_SLEEP);
1149 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1150 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
1151 
1152 		itrq->itrq_i40e = i40e;
1153 		mutex_init(&itrq->itrq_rx_lock, NULL, MUTEX_DRIVER, mutexpri);
1154 		mutex_init(&itrq->itrq_tx_lock, NULL, MUTEX_DRIVER, mutexpri);
1155 		mutex_init(&itrq->itrq_tcb_lock, NULL, MUTEX_DRIVER, mutexpri);
1156 		itrq->itrq_index = i;
1157 	}
1158 
1159 	return (B_TRUE);
1160 }
1161 
1162 
1163 
1164 /*
1165  * Unless a .conf file already overrode i40e_t structure values, they will
1166  * be 0, and need to be set in conjunction with the now-available HW report.
1167  *
1168  * However, at the moment, we cap all of these resources as we only support a
1169  * single receive ring and a single group.
1170  */
1171 /* ARGSUSED */
1172 static void
1173 i40e_hw_to_instance(i40e_t *i40e, i40e_hw_t *hw)
1174 {
1175 	if (i40e->i40e_num_trqpairs == 0) {
1176 		i40e->i40e_num_trqpairs = I40E_TRQPAIR_MAX;
1177 	}
1178 
1179 	if (i40e->i40e_num_rx_groups == 0) {
1180 		i40e->i40e_num_rx_groups = I40E_GROUP_MAX;
1181 	}
1182 }
1183 
1184 /*
1185  * Free any resources required by, or setup by, the Intel common code.
1186  */
1187 static void
1188 i40e_common_code_fini(i40e_t *i40e)
1189 {
1190 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1191 	int rc;
1192 
1193 	rc = i40e_shutdown_lan_hmc(hw);
1194 	if (rc != I40E_SUCCESS)
1195 		i40e_error(i40e, "failed to shutdown LAN hmc: %d", rc);
1196 
1197 	rc = i40e_shutdown_adminq(hw);
1198 	if (rc != I40E_SUCCESS)
1199 		i40e_error(i40e, "failed to shutdown admin queue: %d", rc);
1200 }
1201 
1202 /*
1203  * Initialize and call Intel common-code routines, includes some setup
1204  * the common code expects from the driver.  Also prints on failure, so
1205  * the caller doesn't have to.
1206  */
1207 static boolean_t
1208 i40e_common_code_init(i40e_t *i40e, i40e_hw_t *hw)
1209 {
1210 	int rc;
1211 
1212 	i40e_clear_hw(hw);
1213 	rc = i40e_pf_reset(hw);
1214 	if (rc != 0) {
1215 		i40e_error(i40e, "failed to reset hardware: %d", rc);
1216 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_NO_RESPONSE);
1217 		return (B_FALSE);
1218 	}
1219 
1220 	rc = i40e_init_shared_code(hw);
1221 	if (rc != 0) {
1222 		i40e_error(i40e, "failed to initialize i40e core: %d", rc);
1223 		return (B_FALSE);
1224 	}
1225 
1226 	hw->aq.num_arq_entries = I40E_DEF_ADMINQ_SIZE;
1227 	hw->aq.num_asq_entries =  I40E_DEF_ADMINQ_SIZE;
1228 	hw->aq.arq_buf_size = I40E_ADMINQ_BUFSZ;
1229 	hw->aq.asq_buf_size = I40E_ADMINQ_BUFSZ;
1230 
1231 	rc = i40e_init_adminq(hw);
1232 	if (rc != 0) {
1233 		i40e_error(i40e, "failed to initialize firmware admin queue: "
1234 		    "%d, potential firmware version mismatch", rc);
1235 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
1236 		return (B_FALSE);
1237 	}
1238 
1239 	if (hw->aq.api_maj_ver == I40E_FW_API_VERSION_MAJOR &&
1240 	    hw->aq.api_min_ver > I40E_FW_API_VERSION_MINOR) {
1241 		i40e_log(i40e, "The driver for the device detected a newer "
1242 		    "version of the NVM image (%d.%d) than expected (%d.%d).\n"
1243 		    "Please install the most recent version of the network "
1244 		    "driver.\n", hw->aq.api_maj_ver, hw->aq.api_min_ver,
1245 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_API_VERSION_MINOR);
1246 	} else if (hw->aq.api_maj_ver < I40E_FW_API_VERSION_MAJOR ||
1247 	    hw->aq.api_min_ver < (I40E_FW_API_VERSION_MINOR - 1)) {
1248 		i40e_log(i40e, "The driver for the device detected an older"
1249 		    " version of the NVM image (%d.%d) than expected (%d.%d)."
1250 		    "\nPlease update the NVM image.\n",
1251 		    hw->aq.api_maj_ver, hw->aq.api_min_ver,
1252 		    I40E_FW_API_VERSION_MAJOR, I40E_FW_API_VERSION_MINOR - 1);
1253 	}
1254 
1255 	i40e_clear_pxe_mode(hw);
1256 
1257 	/*
1258 	 * We need to call this so that the common code can discover
1259 	 * capabilities of the hardware, which it uses throughout the rest.
1260 	 */
1261 	if (!i40e_get_hw_capabilities(i40e, hw)) {
1262 		i40e_error(i40e, "failed to obtain hardware capabilities");
1263 		return (B_FALSE);
1264 	}
1265 
1266 	if (i40e_get_available_resources(i40e) == B_FALSE) {
1267 		i40e_error(i40e, "failed to obtain hardware resources");
1268 		return (B_FALSE);
1269 	}
1270 
1271 	i40e_hw_to_instance(i40e, hw);
1272 
1273 	rc = i40e_init_lan_hmc(hw, hw->func_caps.num_tx_qp,
1274 	    hw->func_caps.num_rx_qp, 0, 0);
1275 	if (rc != 0) {
1276 		i40e_error(i40e, "failed to initialize hardware memory cache: "
1277 		    "%d", rc);
1278 		return (B_FALSE);
1279 	}
1280 
1281 	rc = i40e_configure_lan_hmc(hw, I40E_HMC_MODEL_DIRECT_ONLY);
1282 	if (rc != 0) {
1283 		i40e_error(i40e, "failed to configure hardware memory cache: "
1284 		    "%d", rc);
1285 		return (B_FALSE);
1286 	}
1287 
1288 	(void) i40e_aq_stop_lldp(hw, TRUE, NULL);
1289 
1290 	rc = i40e_get_mac_addr(hw, hw->mac.addr);
1291 	if (rc != I40E_SUCCESS) {
1292 		i40e_error(i40e, "failed to retrieve hardware mac address: %d",
1293 		    rc);
1294 		return (B_FALSE);
1295 	}
1296 
1297 	rc = i40e_validate_mac_addr(hw->mac.addr);
1298 	if (rc != 0) {
1299 		i40e_error(i40e, "failed to validate internal mac address: "
1300 		    "%d", rc);
1301 		return (B_FALSE);
1302 	}
1303 	bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
1304 	if ((rc = i40e_get_port_mac_addr(hw, hw->mac.port_addr)) !=
1305 	    I40E_SUCCESS) {
1306 		i40e_error(i40e, "failed to retrieve port mac address: %d",
1307 		    rc);
1308 		return (B_FALSE);
1309 	}
1310 
1311 	/*
1312 	 * We need to obtain the Virtual Station ID (VSI) before we can
1313 	 * perform other operations on the device.
1314 	 */
1315 	i40e->i40e_vsi_id = i40e_get_vsi_id(i40e);
1316 	if (i40e->i40e_vsi_id == -1) {
1317 		i40e_error(i40e, "failed to obtain VSI ID");
1318 		return (B_FALSE);
1319 	}
1320 
1321 	return (B_TRUE);
1322 }
1323 
1324 static void
1325 i40e_unconfigure(dev_info_t *devinfo, i40e_t *i40e)
1326 {
1327 	int rc;
1328 
1329 	if (i40e->i40e_attach_progress & I40E_ATTACH_ENABLE_INTR)
1330 		(void) i40e_disable_interrupts(i40e);
1331 
1332 	if ((i40e->i40e_attach_progress & I40E_ATTACH_LINK_TIMER) &&
1333 	    i40e->i40e_periodic_id != 0) {
1334 		ddi_periodic_delete(i40e->i40e_periodic_id);
1335 		i40e->i40e_periodic_id = 0;
1336 	}
1337 
1338 	if (i40e->i40e_attach_progress & I40E_ATTACH_MAC) {
1339 		rc = mac_unregister(i40e->i40e_mac_hdl);
1340 		if (rc != 0) {
1341 			i40e_error(i40e, "failed to unregister from mac: %d",
1342 			    rc);
1343 		}
1344 	}
1345 
1346 	if (i40e->i40e_attach_progress & I40E_ATTACH_STATS) {
1347 		i40e_stats_fini(i40e);
1348 	}
1349 
1350 	if (i40e->i40e_attach_progress & I40E_ATTACH_ADD_INTR)
1351 		i40e_rem_intr_handlers(i40e);
1352 
1353 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_RINGSLOCKS)
1354 		i40e_free_trqpairs(i40e);
1355 
1356 	if (i40e->i40e_attach_progress & I40E_ATTACH_ALLOC_INTR)
1357 		i40e_rem_intrs(i40e);
1358 
1359 	if (i40e->i40e_attach_progress & I40E_ATTACH_COMMON_CODE)
1360 		i40e_common_code_fini(i40e);
1361 
1362 	i40e_cleanup_resources(i40e);
1363 
1364 	if (i40e->i40e_attach_progress & I40E_ATTACH_PROPS)
1365 		(void) ddi_prop_remove_all(devinfo);
1366 
1367 	if (i40e->i40e_attach_progress & I40E_ATTACH_REGS_MAP &&
1368 	    i40e->i40e_osdep_space.ios_reg_handle != NULL) {
1369 		ddi_regs_map_free(&i40e->i40e_osdep_space.ios_reg_handle);
1370 		i40e->i40e_osdep_space.ios_reg_handle = NULL;
1371 	}
1372 
1373 	if ((i40e->i40e_attach_progress & I40E_ATTACH_PCI_CONFIG) &&
1374 	    i40e->i40e_osdep_space.ios_cfg_handle != NULL) {
1375 		pci_config_teardown(&i40e->i40e_osdep_space.ios_cfg_handle);
1376 		i40e->i40e_osdep_space.ios_cfg_handle = NULL;
1377 	}
1378 
1379 	if (i40e->i40e_attach_progress & I40E_ATTACH_FM_INIT)
1380 		i40e_fm_fini(i40e);
1381 
1382 	kmem_free(i40e->i40e_aqbuf, I40E_ADMINQ_BUFSZ);
1383 	kmem_free(i40e, sizeof (i40e_t));
1384 
1385 	ddi_set_driver_private(devinfo, NULL);
1386 }
1387 
1388 static boolean_t
1389 i40e_final_init(i40e_t *i40e)
1390 {
1391 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1392 	struct i40e_osdep *osdep = OS_DEP(hw);
1393 	uint8_t pbanum[I40E_PBANUM_STRLEN];
1394 	enum i40e_status_code irc;
1395 	char buf[I40E_DDI_PROP_LEN];
1396 
1397 	pbanum[0] = '\0';
1398 	irc = i40e_read_pba_string(hw, pbanum, sizeof (pbanum));
1399 	if (irc != I40E_SUCCESS) {
1400 		i40e_log(i40e, "failed to read PBA string: %d", irc);
1401 	} else {
1402 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1403 		    "printed-board-assembly", (char *)pbanum);
1404 	}
1405 
1406 #ifdef	DEBUG
1407 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.fw_maj_ver,
1408 	    hw->aq.fw_min_ver) < sizeof (buf));
1409 	ASSERT(snprintf(NULL, 0, "%x", hw->aq.fw_build) < sizeof (buf));
1410 	ASSERT(snprintf(NULL, 0, "%d.%d", hw->aq.api_maj_ver,
1411 	    hw->aq.api_min_ver) < sizeof (buf));
1412 #endif
1413 
1414 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.fw_maj_ver,
1415 	    hw->aq.fw_min_ver);
1416 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1417 	    "firmware-version", buf);
1418 	(void) snprintf(buf, sizeof (buf), "%x", hw->aq.fw_build);
1419 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1420 	    "firmware-build", buf);
1421 	(void) snprintf(buf, sizeof (buf), "%d.%d", hw->aq.api_maj_ver,
1422 	    hw->aq.api_min_ver);
1423 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, i40e->i40e_dip,
1424 	    "api-version", buf);
1425 
1426 	if (!i40e_set_hw_bus_info(hw))
1427 		return (B_FALSE);
1428 
1429 	if (i40e_check_acc_handle(osdep->ios_reg_handle) != DDI_FM_OK) {
1430 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
1431 		return (B_FALSE);
1432 	}
1433 
1434 	return (B_TRUE);
1435 }
1436 
1437 static void
1438 i40e_identify_hardware(i40e_t *i40e)
1439 {
1440 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1441 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1442 
1443 	hw->vendor_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_VENID);
1444 	hw->device_id = pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_DEVID);
1445 	hw->revision_id = pci_config_get8(osdep->ios_cfg_handle,
1446 	    PCI_CONF_REVID);
1447 	hw->subsystem_device_id =
1448 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBSYSID);
1449 	hw->subsystem_vendor_id =
1450 	    pci_config_get16(osdep->ios_cfg_handle, PCI_CONF_SUBVENID);
1451 
1452 	/*
1453 	 * Note that we set the hardware's bus information later on, in
1454 	 * i40e_get_available_resources(). The common code doesn't seem to
1455 	 * require that it be set in any ways, it seems to be mostly for
1456 	 * book-keeping.
1457 	 */
1458 }
1459 
1460 static boolean_t
1461 i40e_regs_map(i40e_t *i40e)
1462 {
1463 	dev_info_t *devinfo = i40e->i40e_dip;
1464 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1465 	struct i40e_osdep *osdep = &i40e->i40e_osdep_space;
1466 	off_t memsize;
1467 	int ret;
1468 
1469 	if (ddi_dev_regsize(devinfo, I40E_ADAPTER_REGSET, &memsize) !=
1470 	    DDI_SUCCESS) {
1471 		i40e_error(i40e, "Used invalid register set to map PCIe regs");
1472 		return (B_FALSE);
1473 	}
1474 
1475 	if ((ret = ddi_regs_map_setup(devinfo, I40E_ADAPTER_REGSET,
1476 	    (caddr_t *)&hw->hw_addr, 0, memsize, &i40e_regs_acc_attr,
1477 	    &osdep->ios_reg_handle)) != DDI_SUCCESS) {
1478 		i40e_error(i40e, "failed to map device registers: %d", ret);
1479 		return (B_FALSE);
1480 	}
1481 
1482 	osdep->ios_reg_size = memsize;
1483 	return (B_TRUE);
1484 }
1485 
1486 /*
1487  * Update parameters required when a new MTU has been configured.  Calculate the
1488  * maximum frame size, as well as, size our DMA buffers which we size in
1489  * increments of 1K.
1490  */
1491 void
1492 i40e_update_mtu(i40e_t *i40e)
1493 {
1494 	uint32_t rx, tx;
1495 
1496 	i40e->i40e_frame_max = i40e->i40e_sdu +
1497 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
1498 
1499 	rx = i40e->i40e_frame_max + I40E_BUF_IPHDR_ALIGNMENT;
1500 	i40e->i40e_rx_buf_size = ((rx >> 10) +
1501 	    ((rx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1502 
1503 	tx = i40e->i40e_frame_max;
1504 	i40e->i40e_tx_buf_size = ((tx >> 10) +
1505 	    ((tx & (((uint32_t)1 << 10) -1)) > 0 ? 1 : 0)) << 10;
1506 }
1507 
1508 static int
1509 i40e_get_prop(i40e_t *i40e, char *prop, int min, int max, int def)
1510 {
1511 	int val;
1512 
1513 	val = ddi_prop_get_int(DDI_DEV_T_ANY, i40e->i40e_dip, DDI_PROP_DONTPASS,
1514 	    prop, def);
1515 	if (val > max)
1516 		val = max;
1517 	if (val < min)
1518 		val = min;
1519 	return (val);
1520 }
1521 
1522 static void
1523 i40e_init_properties(i40e_t *i40e)
1524 {
1525 	i40e->i40e_sdu = i40e_get_prop(i40e, "default_mtu",
1526 	    I40E_MIN_MTU, I40E_MAX_MTU, I40E_DEF_MTU);
1527 
1528 	i40e->i40e_intr_force = i40e_get_prop(i40e, "intr_force",
1529 	    I40E_INTR_NONE, I40E_INTR_LEGACY, I40E_INTR_NONE);
1530 
1531 	i40e->i40e_mr_enable = i40e_get_prop(i40e, "mr_enable",
1532 	    B_FALSE, B_TRUE, B_TRUE);
1533 
1534 	i40e->i40e_tx_ring_size = i40e_get_prop(i40e, "tx_ring_size",
1535 	    I40E_MIN_TX_RING_SIZE, I40E_MAX_TX_RING_SIZE,
1536 	    I40E_DEF_TX_RING_SIZE);
1537 	if ((i40e->i40e_tx_ring_size % I40E_DESC_ALIGN) != 0) {
1538 		i40e->i40e_tx_ring_size = P2ROUNDUP(i40e->i40e_tx_ring_size,
1539 		    I40E_DESC_ALIGN);
1540 	}
1541 
1542 	i40e->i40e_tx_block_thresh = i40e_get_prop(i40e, "tx_resched_threshold",
1543 	    I40E_MIN_TX_BLOCK_THRESH,
1544 	    i40e->i40e_tx_ring_size - I40E_TX_MAX_COOKIE,
1545 	    I40E_DEF_TX_BLOCK_THRESH);
1546 
1547 	i40e->i40e_rx_ring_size = i40e_get_prop(i40e, "rx_ring_size",
1548 	    I40E_MIN_RX_RING_SIZE, I40E_MAX_RX_RING_SIZE,
1549 	    I40E_DEF_RX_RING_SIZE);
1550 	if ((i40e->i40e_rx_ring_size % I40E_DESC_ALIGN) != 0) {
1551 		i40e->i40e_rx_ring_size = P2ROUNDUP(i40e->i40e_rx_ring_size,
1552 		    I40E_DESC_ALIGN);
1553 	}
1554 
1555 	i40e->i40e_rx_limit_per_intr = i40e_get_prop(i40e, "rx_limit_per_intr",
1556 	    I40E_MIN_RX_LIMIT_PER_INTR,	I40E_MAX_RX_LIMIT_PER_INTR,
1557 	    I40E_DEF_RX_LIMIT_PER_INTR);
1558 
1559 	i40e->i40e_tx_hcksum_enable = i40e_get_prop(i40e, "tx_hcksum_enable",
1560 	    B_FALSE, B_TRUE, B_TRUE);
1561 
1562 	i40e->i40e_rx_hcksum_enable = i40e_get_prop(i40e, "rx_hcksum_enable",
1563 	    B_FALSE, B_TRUE, B_TRUE);
1564 
1565 	i40e->i40e_rx_dma_min = i40e_get_prop(i40e, "rx_dma_threshold",
1566 	    I40E_MIN_RX_DMA_THRESH, I40E_MAX_RX_DMA_THRESH,
1567 	    I40E_DEF_RX_DMA_THRESH);
1568 
1569 	i40e->i40e_tx_dma_min = i40e_get_prop(i40e, "tx_dma_threshold",
1570 	    I40E_MIN_TX_DMA_THRESH, I40E_MAX_TX_DMA_THRESH,
1571 	    I40E_DEF_TX_DMA_THRESH);
1572 
1573 	i40e->i40e_tx_itr = i40e_get_prop(i40e, "tx_intr_throttle",
1574 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_TX_ITR);
1575 
1576 	i40e->i40e_rx_itr = i40e_get_prop(i40e, "rx_intr_throttle",
1577 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_RX_ITR);
1578 
1579 	i40e->i40e_other_itr = i40e_get_prop(i40e, "other_intr_throttle",
1580 	    I40E_MIN_ITR, I40E_MAX_ITR, I40E_DEF_OTHER_ITR);
1581 
1582 	if (!i40e->i40e_mr_enable) {
1583 		i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1584 		i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1585 	}
1586 
1587 	i40e_update_mtu(i40e);
1588 }
1589 
1590 /*
1591  * There are a few constraints on interrupts that we're currently imposing, some
1592  * of which are restrictions from hardware. For a fuller treatment, see
1593  * i40e_intr.c.
1594  *
1595  * Currently, to use MSI-X we require two interrupts be available though in
1596  * theory we should participate in IRM and happily use more interrupts.
1597  *
1598  * Hardware only supports a single MSI being programmed and therefore if we
1599  * don't have MSI-X interrupts available at this time, then we ratchet down the
1600  * number of rings and groups available. Obviously, we only bother with a single
1601  * fixed interrupt.
1602  */
1603 static boolean_t
1604 i40e_alloc_intr_handles(i40e_t *i40e, dev_info_t *devinfo, int intr_type)
1605 {
1606 	i40e_hw_t *hw = &i40e->i40e_hw_space;
1607 	ddi_acc_handle_t rh = i40e->i40e_osdep_space.ios_reg_handle;
1608 	int request, count, actual, rc, min;
1609 	uint32_t reg;
1610 
1611 	switch (intr_type) {
1612 	case DDI_INTR_TYPE_FIXED:
1613 	case DDI_INTR_TYPE_MSI:
1614 		request = 1;
1615 		min = 1;
1616 		break;
1617 	case DDI_INTR_TYPE_MSIX:
1618 		min = 2;
1619 		if (!i40e->i40e_mr_enable) {
1620 			request = 2;
1621 			break;
1622 		}
1623 		reg = I40E_READ_REG(hw, I40E_GLPCI_CNF2);
1624 		/*
1625 		 * Should this read fail, we will drop back to using
1626 		 * MSI or fixed interrupts.
1627 		 */
1628 		if (i40e_check_acc_handle(rh) != DDI_FM_OK) {
1629 			ddi_fm_service_impact(i40e->i40e_dip,
1630 			    DDI_SERVICE_DEGRADED);
1631 			return (B_FALSE);
1632 		}
1633 		request = (reg & I40E_GLPCI_CNF2_MSI_X_PF_N_MASK) >>
1634 		    I40E_GLPCI_CNF2_MSI_X_PF_N_SHIFT;
1635 		request++;	/* the register value is n - 1 */
1636 		break;
1637 	default:
1638 		panic("bad interrupt type passed to i40e_alloc_intr_handles: "
1639 		    "%d", intr_type);
1640 		return (B_FALSE);
1641 	}
1642 
1643 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
1644 	if (rc != DDI_SUCCESS || count < min) {
1645 		i40e_log(i40e, "Get interrupt number failed, "
1646 		    "returned %d, count %d", rc, count);
1647 		return (B_FALSE);
1648 	}
1649 
1650 	rc = ddi_intr_get_navail(devinfo, intr_type, &count);
1651 	if (rc != DDI_SUCCESS || count < min) {
1652 		i40e_log(i40e, "Get AVAILABLE interrupt number failed, "
1653 		    "returned %d, count %d", rc, count);
1654 		return (B_FALSE);
1655 	}
1656 
1657 	actual = 0;
1658 	i40e->i40e_intr_count = 0;
1659 	i40e->i40e_intr_count_max = 0;
1660 	i40e->i40e_intr_count_min = 0;
1661 
1662 	i40e->i40e_intr_size = request * sizeof (ddi_intr_handle_t);
1663 	ASSERT(i40e->i40e_intr_size != 0);
1664 	i40e->i40e_intr_handles = kmem_alloc(i40e->i40e_intr_size, KM_SLEEP);
1665 
1666 	rc = ddi_intr_alloc(devinfo, i40e->i40e_intr_handles, intr_type, 0,
1667 	    min(request, count), &actual, DDI_INTR_ALLOC_NORMAL);
1668 	if (rc != DDI_SUCCESS) {
1669 		i40e_log(i40e, "Interrupt allocation failed with %d.", rc);
1670 		goto alloc_handle_fail;
1671 	}
1672 
1673 	i40e->i40e_intr_count = actual;
1674 	i40e->i40e_intr_count_max = request;
1675 	i40e->i40e_intr_count_min = min;
1676 
1677 	if (actual < min) {
1678 		i40e_log(i40e, "actual (%d) is less than minimum (%d).",
1679 		    actual, min);
1680 		goto alloc_handle_fail;
1681 	}
1682 
1683 	/*
1684 	 * Record the priority and capabilities for our first vector.  Once
1685 	 * we have it, that's our priority until detach time.  Even if we
1686 	 * eventually participate in IRM, our priority shouldn't change.
1687 	 */
1688 	rc = ddi_intr_get_pri(i40e->i40e_intr_handles[0], &i40e->i40e_intr_pri);
1689 	if (rc != DDI_SUCCESS) {
1690 		i40e_log(i40e,
1691 		    "Getting interrupt priority failed with %d.", rc);
1692 		goto alloc_handle_fail;
1693 	}
1694 
1695 	rc = ddi_intr_get_cap(i40e->i40e_intr_handles[0], &i40e->i40e_intr_cap);
1696 	if (rc != DDI_SUCCESS) {
1697 		i40e_log(i40e,
1698 		    "Getting interrupt capabilities failed with %d.", rc);
1699 		goto alloc_handle_fail;
1700 	}
1701 
1702 	i40e->i40e_intr_type = intr_type;
1703 	return (B_TRUE);
1704 
1705 alloc_handle_fail:
1706 
1707 	i40e_rem_intrs(i40e);
1708 	return (B_FALSE);
1709 }
1710 
1711 static boolean_t
1712 i40e_alloc_intrs(i40e_t *i40e, dev_info_t *devinfo)
1713 {
1714 	int intr_types, rc;
1715 	uint_t max_trqpairs;
1716 
1717 	if (i40e_is_x722(i40e)) {
1718 		max_trqpairs = I40E_722_MAX_TC_QUEUES;
1719 	} else {
1720 		max_trqpairs = I40E_710_MAX_TC_QUEUES;
1721 	}
1722 
1723 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
1724 	if (rc != DDI_SUCCESS) {
1725 		i40e_error(i40e, "failed to get supported interrupt types: %d",
1726 		    rc);
1727 		return (B_FALSE);
1728 	}
1729 
1730 	i40e->i40e_intr_type = 0;
1731 
1732 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
1733 	    i40e->i40e_intr_force <= I40E_INTR_MSIX) {
1734 		if (i40e_alloc_intr_handles(i40e, devinfo,
1735 		    DDI_INTR_TYPE_MSIX)) {
1736 			i40e->i40e_num_trqpairs =
1737 			    MIN(i40e->i40e_intr_count - 1, max_trqpairs);
1738 			return (B_TRUE);
1739 		}
1740 	}
1741 
1742 	/*
1743 	 * We only use multiple transmit/receive pairs when MSI-X interrupts are
1744 	 * available due to the fact that the device basically only supports a
1745 	 * single MSI interrupt.
1746 	 */
1747 	i40e->i40e_num_trqpairs = I40E_TRQPAIR_NOMSIX;
1748 	i40e->i40e_num_rx_groups = I40E_GROUP_NOMSIX;
1749 
1750 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
1751 	    (i40e->i40e_intr_force <= I40E_INTR_MSI)) {
1752 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_MSI))
1753 			return (B_TRUE);
1754 	}
1755 
1756 	if (intr_types & DDI_INTR_TYPE_FIXED) {
1757 		if (i40e_alloc_intr_handles(i40e, devinfo, DDI_INTR_TYPE_FIXED))
1758 			return (B_TRUE);
1759 	}
1760 
1761 	return (B_FALSE);
1762 }
1763 
1764 /*
1765  * Map different interrupts to MSI-X vectors.
1766  */
1767 static boolean_t
1768 i40e_map_intrs_to_vectors(i40e_t *i40e)
1769 {
1770 	int i;
1771 
1772 	if (i40e->i40e_intr_type != DDI_INTR_TYPE_MSIX) {
1773 		return (B_TRUE);
1774 	}
1775 
1776 	/*
1777 	 * Each queue pair is mapped to a single interrupt, so transmit
1778 	 * and receive interrupts for a given queue share the same vector.
1779 	 * The number of queue pairs is one less than the number of interrupt
1780 	 * vectors and is assigned the vector one higher than its index.
1781 	 * Vector zero is reserved for the admin queue.
1782 	 */
1783 	ASSERT(i40e->i40e_intr_count == i40e->i40e_num_trqpairs + 1);
1784 
1785 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
1786 		i40e->i40e_trqpairs[i].itrq_rx_intrvec = i + 1;
1787 		i40e->i40e_trqpairs[i].itrq_tx_intrvec = i + 1;
1788 	}
1789 
1790 	return (B_TRUE);
1791 }
1792 
1793 static boolean_t
1794 i40e_add_intr_handlers(i40e_t *i40e)
1795 {
1796 	int rc, vector;
1797 
1798 	switch (i40e->i40e_intr_type) {
1799 	case DDI_INTR_TYPE_MSIX:
1800 		for (vector = 0; vector < i40e->i40e_intr_count; vector++) {
1801 			rc = ddi_intr_add_handler(
1802 			    i40e->i40e_intr_handles[vector],
1803 			    (ddi_intr_handler_t *)i40e_intr_msix, i40e,
1804 			    (void *)(uintptr_t)vector);
1805 			if (rc != DDI_SUCCESS) {
1806 				i40e_log(i40e, "Add interrupt handler (MSI-X) "
1807 				    "failed: return %d, vector %d", rc, vector);
1808 				for (vector--; vector >= 0; vector--) {
1809 					(void) ddi_intr_remove_handler(
1810 					    i40e->i40e_intr_handles[vector]);
1811 				}
1812 				return (B_FALSE);
1813 			}
1814 		}
1815 		break;
1816 	case DDI_INTR_TYPE_MSI:
1817 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1818 		    (ddi_intr_handler_t *)i40e_intr_msi, i40e, NULL);
1819 		if (rc != DDI_SUCCESS) {
1820 			i40e_log(i40e, "Add interrupt handler (MSI) failed: "
1821 			    "return %d", rc);
1822 			return (B_FALSE);
1823 		}
1824 		break;
1825 	case DDI_INTR_TYPE_FIXED:
1826 		rc = ddi_intr_add_handler(i40e->i40e_intr_handles[0],
1827 		    (ddi_intr_handler_t *)i40e_intr_legacy, i40e, NULL);
1828 		if (rc != DDI_SUCCESS) {
1829 			i40e_log(i40e, "Add interrupt handler (legacy) failed:"
1830 			    " return %d", rc);
1831 			return (B_FALSE);
1832 		}
1833 		break;
1834 	default:
1835 		/* Cast to pacify lint */
1836 		panic("i40e_intr_type %p contains an unknown type: %d",
1837 		    (void *)i40e, i40e->i40e_intr_type);
1838 	}
1839 
1840 	return (B_TRUE);
1841 }
1842 
1843 /*
1844  * Perform periodic checks. Longer term, we should be thinking about additional
1845  * things here:
1846  *
1847  * o Stall Detection
1848  * o Temperature sensor detection
1849  * o Device resetting
1850  * o Statistics updating to avoid wraparound
1851  */
1852 static void
1853 i40e_timer(void *arg)
1854 {
1855 	i40e_t *i40e = arg;
1856 
1857 	mutex_enter(&i40e->i40e_general_lock);
1858 	i40e_link_check(i40e);
1859 	mutex_exit(&i40e->i40e_general_lock);
1860 }
1861 
1862 /*
1863  * Get the hardware state, and scribble away anything that needs scribbling.
1864  */
1865 static void
1866 i40e_get_hw_state(i40e_t *i40e, i40e_hw_t *hw)
1867 {
1868 	int rc;
1869 
1870 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
1871 
1872 	(void) i40e_aq_get_link_info(hw, TRUE, NULL, NULL);
1873 	i40e_link_check(i40e);
1874 
1875 	/*
1876 	 * Try and determine our PHY. Note that we may have to retry to and
1877 	 * delay to detect fiber correctly.
1878 	 */
1879 	rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE, &i40e->i40e_phy,
1880 	    NULL);
1881 	if (rc == I40E_ERR_UNKNOWN_PHY) {
1882 		i40e_msec_delay(200);
1883 		rc = i40e_aq_get_phy_capabilities(hw, B_FALSE, B_TRUE,
1884 		    &i40e->i40e_phy, NULL);
1885 	}
1886 
1887 	if (rc != I40E_SUCCESS) {
1888 		if (rc == I40E_ERR_UNKNOWN_PHY) {
1889 			i40e_error(i40e, "encountered unknown PHY type, "
1890 			    "not attaching.");
1891 		} else {
1892 			i40e_error(i40e, "error getting physical capabilities: "
1893 			    "%d, %d", rc, hw->aq.asq_last_status);
1894 		}
1895 	}
1896 
1897 	rc = i40e_update_link_info(hw);
1898 	if (rc != I40E_SUCCESS) {
1899 		i40e_error(i40e, "failed to update link information: %d", rc);
1900 	}
1901 
1902 	/*
1903 	 * In general, we don't want to mask off (as in stop from being a cause)
1904 	 * any of the interrupts that the phy might be able to generate.
1905 	 */
1906 	rc = i40e_aq_set_phy_int_mask(hw, 0, NULL);
1907 	if (rc != I40E_SUCCESS) {
1908 		i40e_error(i40e, "failed to update phy link mask: %d", rc);
1909 	}
1910 }
1911 
1912 /*
1913  * Go through and re-initialize any existing filters that we may have set up for
1914  * this device. Note that we would only expect them to exist if hardware had
1915  * already been initialized and we had just reset it. While we're not
1916  * implementing this yet, we're keeping this around for when we add reset
1917  * capabilities, so this isn't forgotten.
1918  */
1919 /* ARGSUSED */
1920 static void
1921 i40e_init_macaddrs(i40e_t *i40e, i40e_hw_t *hw)
1922 {
1923 }
1924 
1925 /*
1926  * Configure the hardware for the Virtual Station Interface (VSI).  Currently
1927  * we only support one, but in the future we could instantiate more than one
1928  * per attach-point.
1929  */
1930 static boolean_t
1931 i40e_config_vsi(i40e_t *i40e, i40e_hw_t *hw)
1932 {
1933 	struct i40e_vsi_context	context;
1934 	int err, tc_queues;
1935 
1936 	bzero(&context, sizeof (struct i40e_vsi_context));
1937 	context.seid = i40e->i40e_vsi_id;
1938 	context.pf_num = hw->pf_id;
1939 	err = i40e_aq_get_vsi_params(hw, &context, NULL);
1940 	if (err != I40E_SUCCESS) {
1941 		i40e_error(i40e, "get VSI params failed with %d", err);
1942 		return (B_FALSE);
1943 	}
1944 
1945 	i40e->i40e_vsi_num = context.vsi_number;
1946 
1947 	/*
1948 	 * Set the queue and traffic class bits.  Keep it simple for now.
1949 	 */
1950 	context.info.valid_sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
1951 	context.info.mapping_flags = I40E_AQ_VSI_QUE_MAP_CONTIG;
1952 	context.info.queue_mapping[0] = I40E_ASSIGN_ALL_QUEUES;
1953 
1954 	/*
1955 	 * tc_queues determines the size of the traffic class, where the size is
1956 	 * 2^^tc_queues to a maximum of 64 for the X710 and 128 for the X722.
1957 	 *
1958 	 * Some examples:
1959 	 * 	i40e_num_trqpairs == 1 =>  tc_queues = 0, 2^^0 = 1.
1960 	 * 	i40e_num_trqpairs == 7 =>  tc_queues = 3, 2^^3 = 8.
1961 	 * 	i40e_num_trqpairs == 8 =>  tc_queues = 3, 2^^3 = 8.
1962 	 * 	i40e_num_trqpairs == 9 =>  tc_queues = 4, 2^^4 = 16.
1963 	 * 	i40e_num_trqpairs == 17 => tc_queues = 5, 2^^5 = 32.
1964 	 * 	i40e_num_trqpairs == 64 => tc_queues = 6, 2^^6 = 64.
1965 	 */
1966 	tc_queues = ddi_fls(i40e->i40e_num_trqpairs - 1);
1967 
1968 	context.info.tc_mapping[0] = ((0 << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) &
1969 	    I40E_AQ_VSI_TC_QUE_OFFSET_MASK) |
1970 	    ((tc_queues << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT) &
1971 	    I40E_AQ_VSI_TC_QUE_NUMBER_MASK);
1972 
1973 	context.info.valid_sections |= I40E_AQ_VSI_PROP_VLAN_VALID;
1974 	context.info.port_vlan_flags = I40E_AQ_VSI_PVLAN_MODE_ALL |
1975 	    I40E_AQ_VSI_PVLAN_EMOD_NOTHING;
1976 
1977 	context.flags = LE16_TO_CPU(I40E_AQ_VSI_TYPE_PF);
1978 
1979 	i40e->i40e_vsi_stat_id = LE16_TO_CPU(context.info.stat_counter_idx);
1980 	if (i40e_stat_vsi_init(i40e) == B_FALSE)
1981 		return (B_FALSE);
1982 
1983 	err = i40e_aq_update_vsi_params(hw, &context, NULL);
1984 	if (err != I40E_SUCCESS) {
1985 		i40e_error(i40e, "Update VSI params failed with %d", err);
1986 		return (B_FALSE);
1987 	}
1988 
1989 
1990 	return (B_TRUE);
1991 }
1992 
1993 /*
1994  * Configure the RSS key. For the X710 controller family, this is set on a
1995  * per-PF basis via registers. For the X722, this is done on a per-VSI basis
1996  * through the admin queue.
1997  */
1998 static boolean_t
1999 i40e_config_rss_key(i40e_t *i40e, i40e_hw_t *hw)
2000 {
2001 	uint32_t seed[I40E_PFQF_HKEY_MAX_INDEX + 1];
2002 
2003 	(void) random_get_pseudo_bytes((uint8_t *)seed, sizeof (seed));
2004 
2005 	if (i40e_is_x722(i40e)) {
2006 		struct i40e_aqc_get_set_rss_key_data key;
2007 		const char *u8seed = (char *)seed;
2008 		enum i40e_status_code status;
2009 
2010 		CTASSERT(sizeof (key) >= (sizeof (key.standard_rss_key) +
2011 		    sizeof (key.extended_hash_key)));
2012 
2013 		bcopy(u8seed, key.standard_rss_key,
2014 		    sizeof (key.standard_rss_key));
2015 		bcopy(&u8seed[sizeof (key.standard_rss_key)],
2016 		    key.extended_hash_key, sizeof (key.extended_hash_key));
2017 
2018 		status = i40e_aq_set_rss_key(hw, i40e->i40e_vsi_num, &key);
2019 		if (status != I40E_SUCCESS) {
2020 			i40e_error(i40e, "failed to set rss key: %d", status);
2021 			return (B_FALSE);
2022 		}
2023 	} else {
2024 		uint_t i;
2025 		for (i = 0; i <= I40E_PFQF_HKEY_MAX_INDEX; i++)
2026 			i40e_write_rx_ctl(hw, I40E_PFQF_HKEY(i), seed[i]);
2027 	}
2028 
2029 	return (B_TRUE);
2030 }
2031 
2032 /*
2033  * Populate the LUT. The size of each entry in the LUT depends on the controller
2034  * family, with the X722 using a known 7-bit width. On the X710 controller, this
2035  * is programmed through its control registers where as on the X722 this is
2036  * configured through the admin queue. Also of note, the X722 allows the LUT to
2037  * be set on a per-PF or VSI basis. At this time, as we only have a single VSI,
2038  * we use the PF setting as it is the primary VSI.
2039  *
2040  * We populate the LUT in a round robin fashion with the rx queue indices from 0
2041  * to i40e_num_trqpairs - 1.
2042  */
2043 static boolean_t
2044 i40e_config_rss_hlut(i40e_t *i40e, i40e_hw_t *hw)
2045 {
2046 	uint32_t *hlut;
2047 	uint8_t lut_mask;
2048 	uint_t i;
2049 	boolean_t ret = B_FALSE;
2050 
2051 	/*
2052 	 * We always configure the PF with a table size of 512 bytes in
2053 	 * i40e_chip_start().
2054 	 */
2055 	hlut = kmem_alloc(I40E_HLUT_TABLE_SIZE, KM_NOSLEEP);
2056 	if (hlut == NULL) {
2057 		i40e_error(i40e, "i40e_config_rss() buffer allocation failed");
2058 		return (B_FALSE);
2059 	}
2060 
2061 	/*
2062 	 * The width of the X722 is apparently defined to be 7 bits, regardless
2063 	 * of the capability.
2064 	 */
2065 	if (i40e_is_x722(i40e)) {
2066 		lut_mask = (1 << 7) - 1;
2067 	} else {
2068 		lut_mask = (1 << hw->func_caps.rss_table_entry_width) - 1;
2069 	}
2070 
2071 	for (i = 0; i < I40E_HLUT_TABLE_SIZE; i++)
2072 		((uint8_t *)hlut)[i] = (i % i40e->i40e_num_trqpairs) & lut_mask;
2073 
2074 	if (i40e_is_x722(i40e)) {
2075 		enum i40e_status_code status;
2076 		status = i40e_aq_set_rss_lut(hw, i40e->i40e_vsi_num, B_TRUE,
2077 		    (uint8_t *)hlut, I40E_HLUT_TABLE_SIZE);
2078 		if (status != I40E_SUCCESS) {
2079 			i40e_error(i40e, "failed to set RSS LUT: %d", status);
2080 			goto out;
2081 		}
2082 	} else {
2083 		for (i = 0; i < I40E_HLUT_TABLE_SIZE >> 2; i++) {
2084 			I40E_WRITE_REG(hw, I40E_PFQF_HLUT(i), hlut[i]);
2085 		}
2086 	}
2087 	ret = B_TRUE;
2088 out:
2089 	kmem_free(hlut, I40E_HLUT_TABLE_SIZE);
2090 	return (ret);
2091 }
2092 
2093 /*
2094  * Set up RSS.
2095  * 	1. Seed the hash key.
2096  *	2. Enable PCTYPEs for the hash filter.
2097  *	3. Populate the LUT.
2098  */
2099 static boolean_t
2100 i40e_config_rss(i40e_t *i40e, i40e_hw_t *hw)
2101 {
2102 	uint64_t hena;
2103 
2104 	/*
2105 	 * 1. Seed the hash key
2106 	 */
2107 	if (!i40e_config_rss_key(i40e, hw))
2108 		return (B_FALSE);
2109 
2110 	/*
2111 	 * 2. Configure PCTYPES
2112 	 */
2113 	hena = (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) |
2114 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP) |
2115 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) |
2116 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_UDP) |
2117 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV4) |
2118 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) |
2119 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP) |
2120 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) |
2121 	    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_UDP) |
2122 	    (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV6) |
2123 	    (1ULL << I40E_FILTER_PCTYPE_L2_PAYLOAD);
2124 
2125 	/*
2126 	 * Add additional types supported by the X722 controller.
2127 	 */
2128 	if (i40e_is_x722(i40e)) {
2129 		hena |= (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) |
2130 		    (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) |
2131 		    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) |
2132 		    (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) |
2133 		    (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP) |
2134 		    (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK);
2135 	}
2136 
2137 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(0), (uint32_t)hena);
2138 	i40e_write_rx_ctl(hw, I40E_PFQF_HENA(1), (uint32_t)(hena >> 32));
2139 
2140 	/*
2141 	 * 3. Populate LUT
2142 	 */
2143 	return (i40e_config_rss_hlut(i40e, hw));
2144 }
2145 
2146 /*
2147  * Wrapper to kick the chipset on.
2148  */
2149 static boolean_t
2150 i40e_chip_start(i40e_t *i40e)
2151 {
2152 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2153 	struct i40e_filter_control_settings filter;
2154 	int rc;
2155 
2156 	if (((hw->aq.fw_maj_ver == 4) && (hw->aq.fw_min_ver < 33)) ||
2157 	    (hw->aq.fw_maj_ver < 4)) {
2158 		i40e_msec_delay(75);
2159 		if (i40e_aq_set_link_restart_an(hw, TRUE, NULL) !=
2160 		    I40E_SUCCESS) {
2161 			i40e_error(i40e, "failed to restart link: admin queue "
2162 			    "error: %d", hw->aq.asq_last_status);
2163 			return (B_FALSE);
2164 		}
2165 	}
2166 
2167 	/* Determine hardware state */
2168 	i40e_get_hw_state(i40e, hw);
2169 
2170 	/* Initialize mac addresses. */
2171 	i40e_init_macaddrs(i40e, hw);
2172 
2173 	/*
2174 	 * Set up the filter control. If the hash lut size is changed from
2175 	 * I40E_HASH_LUT_SIZE_512 then I40E_HLUT_TABLE_SIZE and
2176 	 * i40e_config_rss_hlut() will need to be updated.
2177 	 */
2178 	bzero(&filter, sizeof (filter));
2179 	filter.enable_ethtype = TRUE;
2180 	filter.enable_macvlan = TRUE;
2181 	filter.hash_lut_size = I40E_HASH_LUT_SIZE_512;
2182 
2183 	rc = i40e_set_filter_control(hw, &filter);
2184 	if (rc != I40E_SUCCESS) {
2185 		i40e_error(i40e, "i40e_set_filter_control() returned %d", rc);
2186 		return (B_FALSE);
2187 	}
2188 
2189 	i40e_intr_chip_init(i40e);
2190 
2191 	if (!i40e_config_vsi(i40e, hw))
2192 		return (B_FALSE);
2193 
2194 	if (!i40e_config_rss(i40e, hw))
2195 		return (B_FALSE);
2196 
2197 	i40e_flush(hw);
2198 
2199 	return (B_TRUE);
2200 }
2201 
2202 /*
2203  * Take care of tearing down the rx ring. See 8.3.3.1.2 for more information.
2204  */
2205 static void
2206 i40e_shutdown_rx_rings(i40e_t *i40e)
2207 {
2208 	int i;
2209 	uint32_t reg;
2210 
2211 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2212 
2213 	/*
2214 	 * Step 1. The interrupt linked list (see i40e_intr.c for more
2215 	 * information) should have already been cleared before calling this
2216 	 * function.
2217 	 */
2218 #ifdef	DEBUG
2219 	if (i40e->i40e_intr_type == DDI_INTR_TYPE_MSIX) {
2220 		for (i = 1; i < i40e->i40e_intr_count; i++) {
2221 			reg = I40E_READ_REG(hw, I40E_PFINT_LNKLSTN(i - 1));
2222 			VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2223 		}
2224 	} else {
2225 		reg = I40E_READ_REG(hw, I40E_PFINT_LNKLST0);
2226 		VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2227 	}
2228 
2229 #endif	/* DEBUG */
2230 
2231 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2232 		/*
2233 		 * Step 1. Request the queue by clearing QENA_REQ. It may not be
2234 		 * set due to unwinding from failures and a partially enabled
2235 		 * ring set.
2236 		 */
2237 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2238 		if (!(reg & I40E_QRX_ENA_QENA_REQ_MASK))
2239 			continue;
2240 		VERIFY((reg & I40E_QRX_ENA_QENA_REQ_MASK) ==
2241 		    I40E_QRX_ENA_QENA_REQ_MASK);
2242 		reg &= ~I40E_QRX_ENA_QENA_REQ_MASK;
2243 		I40E_WRITE_REG(hw, I40E_QRX_ENA(i), reg);
2244 	}
2245 
2246 	/*
2247 	 * Step 2. Wait for the disable to take, by having QENA_STAT in the FPM
2248 	 * be cleared. Note that we could still receive data in the queue during
2249 	 * this time. We don't actually wait for this now and instead defer this
2250 	 * to i40e_shutdown_rings_wait(), after we've interleaved disabling the
2251 	 * TX queues as well.
2252 	 */
2253 }
2254 
2255 static void
2256 i40e_shutdown_tx_rings(i40e_t *i40e)
2257 {
2258 	int i;
2259 	uint32_t reg;
2260 
2261 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2262 
2263 	/*
2264 	 * Step 1. The interrupt linked list should already have been cleared.
2265 	 */
2266 #ifdef DEBUG
2267 	if (i40e->i40e_intr_type == DDI_INTR_TYPE_MSIX) {
2268 		for (i = 1; i < i40e->i40e_intr_count; i++) {
2269 			reg = I40E_READ_REG(hw, I40E_PFINT_LNKLSTN(i - 1));
2270 			VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2271 		}
2272 	} else {
2273 		reg = I40E_READ_REG(hw, I40E_PFINT_LNKLST0);
2274 		VERIFY3U(reg, ==, I40E_QUEUE_TYPE_EOL);
2275 
2276 	}
2277 #endif	/* DEBUG */
2278 
2279 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2280 		/*
2281 		 * Step 2. Set the SET_QDIS flag for every queue.
2282 		 */
2283 		i40e_pre_tx_queue_cfg(hw, i, B_FALSE);
2284 	}
2285 
2286 	/*
2287 	 * Step 3. Wait at least 400 usec (can be done once for all queues).
2288 	 */
2289 	drv_usecwait(500);
2290 
2291 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2292 		/*
2293 		 * Step 4. Clear the QENA_REQ flag which tells hardware to
2294 		 * quiesce. If QENA_REQ is not already set then that means that
2295 		 * we likely already tried to disable this queue.
2296 		 */
2297 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2298 		if (!(reg & I40E_QTX_ENA_QENA_REQ_MASK))
2299 			continue;
2300 		reg &= ~I40E_QTX_ENA_QENA_REQ_MASK;
2301 		I40E_WRITE_REG(hw, I40E_QTX_ENA(i), reg);
2302 	}
2303 
2304 	/*
2305 	 * Step 5. Wait for all drains to finish. This will be done by the
2306 	 * hardware removing the QENA_STAT flag from the queue. Rather than
2307 	 * waiting here, we interleave it with all the others in
2308 	 * i40e_shutdown_rings_wait().
2309 	 */
2310 }
2311 
2312 /*
2313  * Wait for all the rings to be shut down. e.g. Steps 2 and 5 from the above
2314  * functions.
2315  */
2316 static boolean_t
2317 i40e_shutdown_rings_wait(i40e_t *i40e)
2318 {
2319 	int i, try;
2320 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2321 
2322 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2323 		uint32_t reg;
2324 
2325 		for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2326 			reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2327 			if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0)
2328 				break;
2329 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2330 		}
2331 
2332 		if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) != 0) {
2333 			i40e_error(i40e, "timed out disabling rx queue %d",
2334 			    i);
2335 			return (B_FALSE);
2336 		}
2337 
2338 		for (try = 0; try < I40E_RING_WAIT_NTRIES; try++) {
2339 			reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2340 			if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0)
2341 				break;
2342 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2343 		}
2344 
2345 		if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) != 0) {
2346 			i40e_error(i40e, "timed out disabling tx queue %d",
2347 			    i);
2348 			return (B_FALSE);
2349 		}
2350 	}
2351 
2352 	return (B_TRUE);
2353 }
2354 
2355 static boolean_t
2356 i40e_shutdown_rings(i40e_t *i40e)
2357 {
2358 	i40e_shutdown_rx_rings(i40e);
2359 	i40e_shutdown_tx_rings(i40e);
2360 	return (i40e_shutdown_rings_wait(i40e));
2361 }
2362 
2363 static void
2364 i40e_setup_rx_descs(i40e_trqpair_t *itrq)
2365 {
2366 	int i;
2367 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2368 
2369 	for (i = 0; i < rxd->rxd_ring_size; i++) {
2370 		i40e_rx_control_block_t *rcb;
2371 		i40e_rx_desc_t *rdesc;
2372 
2373 		rcb = rxd->rxd_work_list[i];
2374 		rdesc = &rxd->rxd_desc_ring[i];
2375 
2376 		rdesc->read.pkt_addr =
2377 		    CPU_TO_LE64((uintptr_t)rcb->rcb_dma.dmab_dma_address);
2378 		rdesc->read.hdr_addr = 0;
2379 	}
2380 }
2381 
2382 static boolean_t
2383 i40e_setup_rx_hmc(i40e_trqpair_t *itrq)
2384 {
2385 	i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2386 	i40e_t *i40e = itrq->itrq_i40e;
2387 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2388 
2389 	struct i40e_hmc_obj_rxq rctx;
2390 	int err;
2391 
2392 	bzero(&rctx, sizeof (struct i40e_hmc_obj_rxq));
2393 	rctx.base = rxd->rxd_desc_area.dmab_dma_address /
2394 	    I40E_HMC_RX_CTX_UNIT;
2395 	rctx.qlen = rxd->rxd_ring_size;
2396 	VERIFY(i40e->i40e_rx_buf_size >= I40E_HMC_RX_DBUFF_MIN);
2397 	VERIFY(i40e->i40e_rx_buf_size <= I40E_HMC_RX_DBUFF_MAX);
2398 	rctx.dbuff = i40e->i40e_rx_buf_size >> I40E_RXQ_CTX_DBUFF_SHIFT;
2399 	rctx.hbuff = 0 >> I40E_RXQ_CTX_HBUFF_SHIFT;
2400 	rctx.dtype = I40E_HMC_RX_DTYPE_NOSPLIT;
2401 	rctx.dsize = I40E_HMC_RX_DSIZE_32BYTE;
2402 	rctx.crcstrip = I40E_HMC_RX_CRCSTRIP_ENABLE;
2403 	rctx.fc_ena = I40E_HMC_RX_FC_DISABLE;
2404 	rctx.l2tsel = I40E_HMC_RX_L2TAGORDER;
2405 	rctx.hsplit_0 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2406 	rctx.hsplit_1 = I40E_HMC_RX_HDRSPLIT_DISABLE;
2407 	rctx.showiv = I40E_HMC_RX_INVLAN_DONTSTRIP;
2408 	rctx.rxmax = i40e->i40e_frame_max;
2409 	rctx.tphrdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2410 	rctx.tphwdesc_ena = I40E_HMC_RX_TPH_DISABLE;
2411 	rctx.tphdata_ena = I40E_HMC_RX_TPH_DISABLE;
2412 	rctx.tphhead_ena = I40E_HMC_RX_TPH_DISABLE;
2413 	rctx.lrxqthresh = I40E_HMC_RX_LOWRXQ_NOINTR;
2414 
2415 	/*
2416 	 * This must be set to 0x1, see Table 8-12 in section 8.3.3.2.2.
2417 	 */
2418 	rctx.prefena = I40E_HMC_RX_PREFENA;
2419 
2420 	err = i40e_clear_lan_rx_queue_context(hw, itrq->itrq_index);
2421 	if (err != I40E_SUCCESS) {
2422 		i40e_error(i40e, "failed to clear rx queue %d context: %d",
2423 		    itrq->itrq_index, err);
2424 		return (B_FALSE);
2425 	}
2426 
2427 	err = i40e_set_lan_rx_queue_context(hw, itrq->itrq_index, &rctx);
2428 	if (err != I40E_SUCCESS) {
2429 		i40e_error(i40e, "failed to set rx queue %d context: %d",
2430 		    itrq->itrq_index, err);
2431 		return (B_FALSE);
2432 	}
2433 
2434 	return (B_TRUE);
2435 }
2436 
2437 /*
2438  * Take care of setting up the descriptor rings and actually programming the
2439  * device. See 8.3.3.1.1 for the full list of steps we need to do to enable the
2440  * rx rings.
2441  */
2442 static boolean_t
2443 i40e_setup_rx_rings(i40e_t *i40e)
2444 {
2445 	int i;
2446 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2447 
2448 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2449 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
2450 		i40e_rx_data_t *rxd = itrq->itrq_rxdata;
2451 		uint32_t reg;
2452 
2453 		/*
2454 		 * Step 1. Program all receive ring descriptors.
2455 		 */
2456 		i40e_setup_rx_descs(itrq);
2457 
2458 		/*
2459 		 * Step 2. Program the queue's FPM/HMC context.
2460 		 */
2461 		if (i40e_setup_rx_hmc(itrq) == B_FALSE)
2462 			return (B_FALSE);
2463 
2464 		/*
2465 		 * Step 3. Clear the queue's tail pointer and set it to the end
2466 		 * of the space.
2467 		 */
2468 		I40E_WRITE_REG(hw, I40E_QRX_TAIL(i), 0);
2469 		I40E_WRITE_REG(hw, I40E_QRX_TAIL(i), rxd->rxd_ring_size - 1);
2470 
2471 		/*
2472 		 * Step 4. Enable the queue via the QENA_REQ.
2473 		 */
2474 		reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2475 		VERIFY0(reg & (I40E_QRX_ENA_QENA_REQ_MASK |
2476 		    I40E_QRX_ENA_QENA_STAT_MASK));
2477 		reg |= I40E_QRX_ENA_QENA_REQ_MASK;
2478 		I40E_WRITE_REG(hw, I40E_QRX_ENA(i), reg);
2479 	}
2480 
2481 	/*
2482 	 * Note, we wait for every queue to be enabled before we start checking.
2483 	 * This will hopefully cause most queues to be enabled at this point.
2484 	 */
2485 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2486 		uint32_t j, reg;
2487 
2488 		/*
2489 		 * Step 5. Verify that QENA_STAT has been set. It's promised
2490 		 * that this should occur within about 10 us, but like other
2491 		 * systems, we give the card a bit more time.
2492 		 */
2493 		for (j = 0; j < I40E_RING_WAIT_NTRIES; j++) {
2494 			reg = I40E_READ_REG(hw, I40E_QRX_ENA(i));
2495 
2496 			if (reg & I40E_QRX_ENA_QENA_STAT_MASK)
2497 				break;
2498 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2499 		}
2500 
2501 		if ((reg & I40E_QRX_ENA_QENA_STAT_MASK) == 0) {
2502 			i40e_error(i40e, "failed to enable rx queue %d, timed "
2503 			    "out.", i);
2504 			return (B_FALSE);
2505 		}
2506 	}
2507 
2508 	return (B_TRUE);
2509 }
2510 
2511 static boolean_t
2512 i40e_setup_tx_hmc(i40e_trqpair_t *itrq)
2513 {
2514 	i40e_t *i40e = itrq->itrq_i40e;
2515 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2516 
2517 	struct i40e_hmc_obj_txq tctx;
2518 	struct i40e_vsi_context	context;
2519 	int err;
2520 
2521 	bzero(&tctx, sizeof (struct i40e_hmc_obj_txq));
2522 	tctx.new_context = I40E_HMC_TX_NEW_CONTEXT;
2523 	tctx.base = itrq->itrq_desc_area.dmab_dma_address /
2524 	    I40E_HMC_TX_CTX_UNIT;
2525 	tctx.fc_ena = I40E_HMC_TX_FC_DISABLE;
2526 	tctx.timesync_ena = I40E_HMC_TX_TS_DISABLE;
2527 	tctx.fd_ena = I40E_HMC_TX_FD_DISABLE;
2528 	tctx.alt_vlan_ena = I40E_HMC_TX_ALT_VLAN_DISABLE;
2529 	tctx.head_wb_ena = I40E_HMC_TX_WB_ENABLE;
2530 	tctx.qlen = itrq->itrq_tx_ring_size;
2531 	tctx.tphrdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2532 	tctx.tphrpacket_ena = I40E_HMC_TX_TPH_DISABLE;
2533 	tctx.tphwdesc_ena = I40E_HMC_TX_TPH_DISABLE;
2534 	tctx.head_wb_addr = itrq->itrq_desc_area.dmab_dma_address +
2535 	    sizeof (i40e_tx_desc_t) * itrq->itrq_tx_ring_size;
2536 
2537 	/*
2538 	 * This field isn't actually documented, like crc, but it suggests that
2539 	 * it should be zeroed. We leave both of these here because of that for
2540 	 * now. We should check with Intel on why these are here even.
2541 	 */
2542 	tctx.crc = 0;
2543 	tctx.rdylist_act = 0;
2544 
2545 	/*
2546 	 * We're supposed to assign the rdylist field with the value of the
2547 	 * traffic class index for the first device. We query the VSI parameters
2548 	 * again to get what the handle is. Note that every queue is always
2549 	 * assigned to traffic class zero, because we don't actually use them.
2550 	 */
2551 	bzero(&context, sizeof (struct i40e_vsi_context));
2552 	context.seid = i40e->i40e_vsi_id;
2553 	context.pf_num = hw->pf_id;
2554 	err = i40e_aq_get_vsi_params(hw, &context, NULL);
2555 	if (err != I40E_SUCCESS) {
2556 		i40e_error(i40e, "get VSI params failed with %d", err);
2557 		return (B_FALSE);
2558 	}
2559 	tctx.rdylist = LE_16(context.info.qs_handle[0]);
2560 
2561 	err = i40e_clear_lan_tx_queue_context(hw, itrq->itrq_index);
2562 	if (err != I40E_SUCCESS) {
2563 		i40e_error(i40e, "failed to clear tx queue %d context: %d",
2564 		    itrq->itrq_index, err);
2565 		return (B_FALSE);
2566 	}
2567 
2568 	err = i40e_set_lan_tx_queue_context(hw, itrq->itrq_index, &tctx);
2569 	if (err != I40E_SUCCESS) {
2570 		i40e_error(i40e, "failed to set tx queue %d context: %d",
2571 		    itrq->itrq_index, err);
2572 		return (B_FALSE);
2573 	}
2574 
2575 	return (B_TRUE);
2576 }
2577 
2578 /*
2579  * Take care of setting up the descriptor rings and actually programming the
2580  * device. See 8.4.3.1.1 for what we need to do here.
2581  */
2582 static boolean_t
2583 i40e_setup_tx_rings(i40e_t *i40e)
2584 {
2585 	int i;
2586 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2587 
2588 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2589 		i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[i];
2590 		uint32_t reg;
2591 
2592 		/*
2593 		 * Step 1. Clear the queue disable flag and verify that the
2594 		 * index is set correctly.
2595 		 */
2596 		i40e_pre_tx_queue_cfg(hw, i, B_TRUE);
2597 
2598 		/*
2599 		 * Step 2. Prepare the queue's FPM/HMC context.
2600 		 */
2601 		if (i40e_setup_tx_hmc(itrq) == B_FALSE)
2602 			return (B_FALSE);
2603 
2604 		/*
2605 		 * Step 3. Verify that it's clear that this PF owns this queue.
2606 		 */
2607 		reg = I40E_QTX_CTL_PF_QUEUE;
2608 		reg |= (hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2609 		    I40E_QTX_CTL_PF_INDX_MASK;
2610 		I40E_WRITE_REG(hw, I40E_QTX_CTL(itrq->itrq_index), reg);
2611 		i40e_flush(hw);
2612 
2613 		/*
2614 		 * Step 4. Set the QENA_REQ flag.
2615 		 */
2616 		reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2617 		VERIFY0(reg & (I40E_QTX_ENA_QENA_REQ_MASK |
2618 		    I40E_QTX_ENA_QENA_STAT_MASK));
2619 		reg |= I40E_QTX_ENA_QENA_REQ_MASK;
2620 		I40E_WRITE_REG(hw, I40E_QTX_ENA(i), reg);
2621 	}
2622 
2623 	/*
2624 	 * Note, we wait for every queue to be enabled before we start checking.
2625 	 * This will hopefully cause most queues to be enabled at this point.
2626 	 */
2627 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2628 		uint32_t j, reg;
2629 
2630 		/*
2631 		 * Step 5. Verify that QENA_STAT has been set. It's promised
2632 		 * that this should occur within about 10 us, but like BSD,
2633 		 * we'll try for up to 100 ms for this queue.
2634 		 */
2635 		for (j = 0; j < I40E_RING_WAIT_NTRIES; j++) {
2636 			reg = I40E_READ_REG(hw, I40E_QTX_ENA(i));
2637 
2638 			if (reg & I40E_QTX_ENA_QENA_STAT_MASK)
2639 				break;
2640 			i40e_msec_delay(I40E_RING_WAIT_PAUSE);
2641 		}
2642 
2643 		if ((reg & I40E_QTX_ENA_QENA_STAT_MASK) == 0) {
2644 			i40e_error(i40e, "failed to enable tx queue %d, timed "
2645 			    "out", i);
2646 			return (B_FALSE);
2647 		}
2648 	}
2649 
2650 	return (B_TRUE);
2651 }
2652 
2653 void
2654 i40e_stop(i40e_t *i40e, boolean_t free_allocations)
2655 {
2656 	int i;
2657 
2658 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
2659 
2660 	/*
2661 	 * Shutdown and drain the tx and rx pipeline. We do this using the
2662 	 * following steps.
2663 	 *
2664 	 * 1) Shutdown interrupts to all the queues (trying to keep the admin
2665 	 *    queue alive).
2666 	 *
2667 	 * 2) Remove all of the interrupt tx and rx causes by setting the
2668 	 *    interrupt linked lists to zero.
2669 	 *
2670 	 * 2) Shutdown the tx and rx rings. Because i40e_shutdown_rings() should
2671 	 *    wait for all the queues to be disabled, once we reach that point
2672 	 *    it should be safe to free associated data.
2673 	 *
2674 	 * 4) Wait 50ms after all that is done. This ensures that the rings are
2675 	 *    ready for programming again and we don't have to think about this
2676 	 *    in other parts of the driver.
2677 	 *
2678 	 * 5) Disable remaining chip interrupts, (admin queue, etc.)
2679 	 *
2680 	 * 6) Verify that FM is happy with all the register accesses we
2681 	 *    performed.
2682 	 */
2683 	i40e_intr_io_disable_all(i40e);
2684 	i40e_intr_io_clear_cause(i40e);
2685 
2686 	if (i40e_shutdown_rings(i40e) == B_FALSE) {
2687 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2688 	}
2689 
2690 	delay(50 * drv_usectohz(1000));
2691 
2692 	i40e_intr_chip_fini(i40e);
2693 
2694 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2695 		mutex_enter(&i40e->i40e_trqpairs[i].itrq_rx_lock);
2696 		mutex_enter(&i40e->i40e_trqpairs[i].itrq_tx_lock);
2697 	}
2698 
2699 	/*
2700 	 * We should consider refactoring this to be part of the ring start /
2701 	 * stop routines at some point.
2702 	 */
2703 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2704 		i40e_stats_trqpair_fini(&i40e->i40e_trqpairs[i]);
2705 	}
2706 
2707 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
2708 	    DDI_FM_OK) {
2709 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2710 	}
2711 
2712 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2713 		i40e_tx_cleanup_ring(&i40e->i40e_trqpairs[i]);
2714 	}
2715 
2716 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2717 		mutex_exit(&i40e->i40e_trqpairs[i].itrq_rx_lock);
2718 		mutex_exit(&i40e->i40e_trqpairs[i].itrq_tx_lock);
2719 	}
2720 
2721 	i40e_stat_vsi_fini(i40e);
2722 
2723 	i40e->i40e_link_speed = 0;
2724 	i40e->i40e_link_duplex = 0;
2725 	i40e_link_state_set(i40e, LINK_STATE_UNKNOWN);
2726 
2727 	if (free_allocations) {
2728 		i40e_free_ring_mem(i40e, B_FALSE);
2729 	}
2730 }
2731 
2732 boolean_t
2733 i40e_start(i40e_t *i40e, boolean_t alloc)
2734 {
2735 	i40e_hw_t *hw = &i40e->i40e_hw_space;
2736 	boolean_t rc = B_TRUE;
2737 	int i, err;
2738 
2739 	ASSERT(MUTEX_HELD(&i40e->i40e_general_lock));
2740 
2741 	if (alloc) {
2742 		if (i40e_alloc_ring_mem(i40e) == B_FALSE) {
2743 			i40e_error(i40e,
2744 			    "Failed to allocate ring memory");
2745 			return (B_FALSE);
2746 		}
2747 	}
2748 
2749 	/*
2750 	 * This should get refactored to be part of ring start and stop at
2751 	 * some point, along with most of the logic here.
2752 	 */
2753 	for (i = 0; i < i40e->i40e_num_trqpairs; i++) {
2754 		if (i40e_stats_trqpair_init(&i40e->i40e_trqpairs[i]) ==
2755 		    B_FALSE) {
2756 			int j;
2757 
2758 			for (j = 0; j < i; j++) {
2759 				i40e_trqpair_t *itrq = &i40e->i40e_trqpairs[j];
2760 				i40e_stats_trqpair_fini(itrq);
2761 			}
2762 			return (B_FALSE);
2763 		}
2764 	}
2765 
2766 	if (!i40e_chip_start(i40e)) {
2767 		i40e_fm_ereport(i40e, DDI_FM_DEVICE_INVAL_STATE);
2768 		rc = B_FALSE;
2769 		goto done;
2770 	}
2771 
2772 	if (i40e_setup_rx_rings(i40e) == B_FALSE) {
2773 		rc = B_FALSE;
2774 		goto done;
2775 	}
2776 
2777 	if (i40e_setup_tx_rings(i40e) == B_FALSE) {
2778 		rc = B_FALSE;
2779 		goto done;
2780 	}
2781 
2782 	/*
2783 	 * Enable broadcast traffic; however, do not enable multicast traffic.
2784 	 * That's handle exclusively through MAC's mc_multicst routines.
2785 	 */
2786 	err = i40e_aq_set_vsi_broadcast(hw, i40e->i40e_vsi_id, B_TRUE, NULL);
2787 	if (err != I40E_SUCCESS) {
2788 		i40e_error(i40e, "failed to set default VSI: %d", err);
2789 		rc = B_FALSE;
2790 		goto done;
2791 	}
2792 
2793 	err = i40e_aq_set_mac_config(hw, i40e->i40e_frame_max, B_TRUE, 0, NULL);
2794 	if (err != I40E_SUCCESS) {
2795 		i40e_error(i40e, "failed to set MAC config: %d", err);
2796 		rc = B_FALSE;
2797 		goto done;
2798 	}
2799 
2800 	/*
2801 	 * Finally, make sure that we're happy from an FM perspective.
2802 	 */
2803 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_reg_handle) !=
2804 	    DDI_FM_OK) {
2805 		rc = B_FALSE;
2806 		goto done;
2807 	}
2808 
2809 	/* Clear state bits prior to final interrupt enabling. */
2810 	atomic_and_32(&i40e->i40e_state,
2811 	    ~(I40E_ERROR | I40E_STALL | I40E_OVERTEMP));
2812 
2813 	i40e_intr_io_enable_all(i40e);
2814 
2815 done:
2816 	if (rc == B_FALSE) {
2817 		i40e_stop(i40e, B_FALSE);
2818 		if (alloc == B_TRUE) {
2819 			i40e_free_ring_mem(i40e, B_TRUE);
2820 		}
2821 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2822 	}
2823 
2824 	return (rc);
2825 }
2826 
2827 /*
2828  * We may have loaned up descriptors to the stack. As such, if we still have
2829  * them outstanding, then we will not continue with detach.
2830  */
2831 static boolean_t
2832 i40e_drain_rx(i40e_t *i40e)
2833 {
2834 	mutex_enter(&i40e->i40e_rx_pending_lock);
2835 	while (i40e->i40e_rx_pending > 0) {
2836 		if (cv_reltimedwait(&i40e->i40e_rx_pending_cv,
2837 		    &i40e->i40e_rx_pending_lock,
2838 		    drv_usectohz(I40E_DRAIN_RX_WAIT), TR_CLOCK_TICK) == -1) {
2839 			mutex_exit(&i40e->i40e_rx_pending_lock);
2840 			return (B_FALSE);
2841 		}
2842 	}
2843 	mutex_exit(&i40e->i40e_rx_pending_lock);
2844 
2845 	return (B_TRUE);
2846 }
2847 
2848 static int
2849 i40e_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
2850 {
2851 	i40e_t *i40e;
2852 	struct i40e_osdep *osdep;
2853 	i40e_hw_t *hw;
2854 	int instance;
2855 
2856 	if (cmd != DDI_ATTACH)
2857 		return (DDI_FAILURE);
2858 
2859 	instance = ddi_get_instance(devinfo);
2860 	i40e = kmem_zalloc(sizeof (i40e_t), KM_SLEEP);
2861 
2862 	i40e->i40e_aqbuf = kmem_zalloc(I40E_ADMINQ_BUFSZ, KM_SLEEP);
2863 	i40e->i40e_instance = instance;
2864 	i40e->i40e_dip = devinfo;
2865 
2866 	hw = &i40e->i40e_hw_space;
2867 	osdep = &i40e->i40e_osdep_space;
2868 	hw->back = osdep;
2869 	osdep->ios_i40e = i40e;
2870 
2871 	ddi_set_driver_private(devinfo, i40e);
2872 
2873 	i40e_fm_init(i40e);
2874 	i40e->i40e_attach_progress |= I40E_ATTACH_FM_INIT;
2875 
2876 	if (pci_config_setup(devinfo, &osdep->ios_cfg_handle) != DDI_SUCCESS) {
2877 		i40e_error(i40e, "Failed to map PCI configurations.");
2878 		goto attach_fail;
2879 	}
2880 	i40e->i40e_attach_progress |= I40E_ATTACH_PCI_CONFIG;
2881 
2882 	i40e_identify_hardware(i40e);
2883 
2884 	if (!i40e_regs_map(i40e)) {
2885 		i40e_error(i40e, "Failed to map device registers.");
2886 		goto attach_fail;
2887 	}
2888 	i40e->i40e_attach_progress |= I40E_ATTACH_REGS_MAP;
2889 
2890 	i40e_init_properties(i40e);
2891 	i40e->i40e_attach_progress |= I40E_ATTACH_PROPS;
2892 
2893 	if (!i40e_common_code_init(i40e, hw))
2894 		goto attach_fail;
2895 	i40e->i40e_attach_progress |= I40E_ATTACH_COMMON_CODE;
2896 
2897 	/*
2898 	 * When we participate in IRM, we should make sure that we register
2899 	 * ourselves with it before callbacks.
2900 	 */
2901 	if (!i40e_alloc_intrs(i40e, devinfo)) {
2902 		i40e_error(i40e, "Failed to allocate interrupts.");
2903 		goto attach_fail;
2904 	}
2905 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_INTR;
2906 
2907 	if (!i40e_alloc_trqpairs(i40e)) {
2908 		i40e_error(i40e,
2909 		    "Failed to allocate receive & transmit rings.");
2910 		goto attach_fail;
2911 	}
2912 	i40e->i40e_attach_progress |= I40E_ATTACH_ALLOC_RINGSLOCKS;
2913 
2914 	if (!i40e_map_intrs_to_vectors(i40e)) {
2915 		i40e_error(i40e, "Failed to map interrupts to vectors.");
2916 		goto attach_fail;
2917 	}
2918 
2919 	if (!i40e_add_intr_handlers(i40e)) {
2920 		i40e_error(i40e, "Failed to add the interrupt handlers.");
2921 		goto attach_fail;
2922 	}
2923 	i40e->i40e_attach_progress |= I40E_ATTACH_ADD_INTR;
2924 
2925 	if (!i40e_final_init(i40e)) {
2926 		i40e_error(i40e, "Final initialization failed.");
2927 		goto attach_fail;
2928 	}
2929 	i40e->i40e_attach_progress |= I40E_ATTACH_INIT;
2930 
2931 	if (i40e_check_acc_handle(i40e->i40e_osdep_space.ios_cfg_handle) !=
2932 	    DDI_FM_OK) {
2933 		ddi_fm_service_impact(i40e->i40e_dip, DDI_SERVICE_LOST);
2934 		goto attach_fail;
2935 	}
2936 
2937 	if (!i40e_stats_init(i40e)) {
2938 		i40e_error(i40e, "Stats initialization failed.");
2939 		goto attach_fail;
2940 	}
2941 	i40e->i40e_attach_progress |= I40E_ATTACH_STATS;
2942 
2943 	if (!i40e_register_mac(i40e)) {
2944 		i40e_error(i40e, "Failed to register to MAC/GLDv3");
2945 		goto attach_fail;
2946 	}
2947 	i40e->i40e_attach_progress |= I40E_ATTACH_MAC;
2948 
2949 	i40e->i40e_periodic_id = ddi_periodic_add(i40e_timer, i40e,
2950 	    I40E_CYCLIC_PERIOD, DDI_IPL_0);
2951 	if (i40e->i40e_periodic_id == 0) {
2952 		i40e_error(i40e, "Failed to add the link-check timer");
2953 		goto attach_fail;
2954 	}
2955 	i40e->i40e_attach_progress |= I40E_ATTACH_LINK_TIMER;
2956 
2957 	if (!i40e_enable_interrupts(i40e)) {
2958 		i40e_error(i40e, "Failed to enable DDI interrupts");
2959 		goto attach_fail;
2960 	}
2961 	i40e->i40e_attach_progress |= I40E_ATTACH_ENABLE_INTR;
2962 
2963 	atomic_or_32(&i40e->i40e_state, I40E_INITIALIZED);
2964 
2965 	mutex_enter(&i40e_glock);
2966 	list_insert_tail(&i40e_glist, i40e);
2967 	mutex_exit(&i40e_glock);
2968 
2969 	return (DDI_SUCCESS);
2970 
2971 attach_fail:
2972 	i40e_unconfigure(devinfo, i40e);
2973 	return (DDI_FAILURE);
2974 }
2975 
2976 static int
2977 i40e_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
2978 {
2979 	i40e_t *i40e;
2980 
2981 	if (cmd != DDI_DETACH)
2982 		return (DDI_FAILURE);
2983 
2984 	i40e = (i40e_t *)ddi_get_driver_private(devinfo);
2985 	if (i40e == NULL) {
2986 		i40e_log(NULL, "i40e_detach() called with no i40e pointer!");
2987 		return (DDI_FAILURE);
2988 	}
2989 
2990 	if (i40e_drain_rx(i40e) == B_FALSE) {
2991 		i40e_log(i40e, "timed out draining DMA resources, %d buffers "
2992 		    "remain", i40e->i40e_rx_pending);
2993 		return (DDI_FAILURE);
2994 	}
2995 
2996 	mutex_enter(&i40e_glock);
2997 	list_remove(&i40e_glist, i40e);
2998 	mutex_exit(&i40e_glock);
2999 
3000 	i40e_unconfigure(devinfo, i40e);
3001 
3002 	return (DDI_SUCCESS);
3003 }
3004 
3005 static struct cb_ops i40e_cb_ops = {
3006 	nulldev,		/* cb_open */
3007 	nulldev,		/* cb_close */
3008 	nodev,			/* cb_strategy */
3009 	nodev,			/* cb_print */
3010 	nodev,			/* cb_dump */
3011 	nodev,			/* cb_read */
3012 	nodev,			/* cb_write */
3013 	nodev,			/* cb_ioctl */
3014 	nodev,			/* cb_devmap */
3015 	nodev,			/* cb_mmap */
3016 	nodev,			/* cb_segmap */
3017 	nochpoll,		/* cb_chpoll */
3018 	ddi_prop_op,		/* cb_prop_op */
3019 	NULL,			/* cb_stream */
3020 	D_MP | D_HOTPLUG,	/* cb_flag */
3021 	CB_REV,			/* cb_rev */
3022 	nodev,			/* cb_aread */
3023 	nodev			/* cb_awrite */
3024 };
3025 
3026 static struct dev_ops i40e_dev_ops = {
3027 	DEVO_REV,		/* devo_rev */
3028 	0,			/* devo_refcnt */
3029 	NULL,			/* devo_getinfo */
3030 	nulldev,		/* devo_identify */
3031 	nulldev,		/* devo_probe */
3032 	i40e_attach,		/* devo_attach */
3033 	i40e_detach,		/* devo_detach */
3034 	nodev,			/* devo_reset */
3035 	&i40e_cb_ops,		/* devo_cb_ops */
3036 	NULL,			/* devo_bus_ops */
3037 	ddi_power,		/* devo_power */
3038 	ddi_quiesce_not_supported /* devo_quiesce */
3039 };
3040 
3041 static struct modldrv i40e_modldrv = {
3042 	&mod_driverops,
3043 	i40e_ident,
3044 	&i40e_dev_ops
3045 };
3046 
3047 static struct modlinkage i40e_modlinkage = {
3048 	MODREV_1,
3049 	&i40e_modldrv,
3050 	NULL
3051 };
3052 
3053 /*
3054  * Module Initialization Functions.
3055  */
3056 int
3057 _init(void)
3058 {
3059 	int status;
3060 
3061 	list_create(&i40e_glist, sizeof (i40e_t), offsetof(i40e_t, i40e_glink));
3062 	list_create(&i40e_dlist, sizeof (i40e_device_t),
3063 	    offsetof(i40e_device_t, id_link));
3064 	mutex_init(&i40e_glock, NULL, MUTEX_DRIVER, NULL);
3065 	mac_init_ops(&i40e_dev_ops, I40E_MODULE_NAME);
3066 
3067 	status = mod_install(&i40e_modlinkage);
3068 	if (status != DDI_SUCCESS) {
3069 		mac_fini_ops(&i40e_dev_ops);
3070 		mutex_destroy(&i40e_glock);
3071 		list_destroy(&i40e_dlist);
3072 		list_destroy(&i40e_glist);
3073 	}
3074 
3075 	return (status);
3076 }
3077 
3078 int
3079 _info(struct modinfo *modinfop)
3080 {
3081 	return (mod_info(&i40e_modlinkage, modinfop));
3082 }
3083 
3084 int
3085 _fini(void)
3086 {
3087 	int status;
3088 
3089 	status = mod_remove(&i40e_modlinkage);
3090 	if (status == DDI_SUCCESS) {
3091 		mac_fini_ops(&i40e_dev_ops);
3092 		mutex_destroy(&i40e_glock);
3093 		list_destroy(&i40e_dlist);
3094 		list_destroy(&i40e_glist);
3095 	}
3096 
3097 	return (status);
3098 }
3099