xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_sw.h (revision 396505af)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source.  A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * Copyright 2015 OmniTI Computer Consulting, Inc. All rights reserved.
14  * Copyright 2016 Joyent, Inc.
15  * Copyright 2017 Tegile Systems, Inc.  All rights reserved.
16  */
17 
18 /*
19  * Please see i40e_main.c for an introduction to the device driver, its layout,
20  * and more.
21  */
22 
23 #ifndef	_I40E_SW_H
24 #define	_I40E_SW_H
25 
26 #ifdef __cplusplus
27 extern "C" {
28 #endif
29 
30 #include <sys/types.h>
31 #include <sys/conf.h>
32 #include <sys/debug.h>
33 #include <sys/stropts.h>
34 #include <sys/stream.h>
35 #include <sys/strsun.h>
36 #include <sys/strlog.h>
37 #include <sys/kmem.h>
38 #include <sys/stat.h>
39 #include <sys/kstat.h>
40 #include <sys/modctl.h>
41 #include <sys/errno.h>
42 #include <sys/dlpi.h>
43 #include <sys/mac_provider.h>
44 #include <sys/mac_ether.h>
45 #include <sys/vlan.h>
46 #include <sys/ddi.h>
47 #include <sys/sunddi.h>
48 #include <sys/pci.h>
49 #include <sys/pcie.h>
50 #include <sys/sdt.h>
51 #include <sys/ethernet.h>
52 #include <sys/pattr.h>
53 #include <sys/strsubr.h>
54 #include <sys/netlb.h>
55 #include <sys/random.h>
56 #include <inet/common.h>
57 #include <inet/tcp.h>
58 #include <inet/ip.h>
59 #include <inet/mi.h>
60 #include <inet/nd.h>
61 #include <netinet/udp.h>
62 #include <netinet/sctp.h>
63 #include <sys/bitmap.h>
64 #include <sys/cpuvar.h>
65 #include <sys/ddifm.h>
66 #include <sys/fm/protocol.h>
67 #include <sys/fm/util.h>
68 #include <sys/disp.h>
69 #include <sys/fm/io/ddi.h>
70 #include <sys/list.h>
71 #include <sys/debug.h>
72 #include <sys/sdt.h>
73 #include "i40e_type.h"
74 #include "i40e_osdep.h"
75 #include "i40e_prototype.h"
76 #include "i40e_xregs.h"
77 
78 #define	I40E_MODULE_NAME "i40e"
79 
80 #define	I40E_ADAPTER_REGSET	1
81 
82 /*
83  * Configuration constants. Note that the hardware defines a minimum bound of 32
84  * descriptors and requires that the programming of the descriptor lengths be
85  * aligned in units of 32 descriptors.
86  */
87 #define	I40E_MIN_TX_RING_SIZE	64
88 #define	I40E_MAX_TX_RING_SIZE	4096
89 #define	I40E_DEF_TX_RING_SIZE	1024
90 
91 #define	I40E_MIN_RX_RING_SIZE	64
92 #define	I40E_MAX_RX_RING_SIZE	4096
93 #define	I40E_DEF_RX_RING_SIZE	1024
94 
95 #define	I40E_DESC_ALIGN		32
96 
97 /*
98  * Sizes used for asynchronous processing of the adminq. We allocate a fixed
99  * size buffer for each instance of the device during attach time, rather than
100  * allocating and freeing one during interrupt processing.
101  *
102  * We also define the descriptor size of the admin queue here.
103  */
104 #define	I40E_ADMINQ_BUFSZ	4096
105 #define	I40E_MAX_ADMINQ_SIZE	1024
106 #define	I40E_DEF_ADMINQ_SIZE	256
107 
108 /*
109  * Note, while the min and maximum values are based upon the sizing of the ring
110  * itself, the default is taken from ixgbe without much thought. It's basically
111  * been cargo culted. See i40e_transceiver.c for a bit more information.
112  */
113 #define	I40E_MIN_RX_LIMIT_PER_INTR	16
114 #define	I40E_MAX_RX_LIMIT_PER_INTR	4096
115 #define	I40E_DEF_RX_LIMIT_PER_INTR	256
116 
117 /*
118  * Valid MTU ranges. Note that the XL710's maximum payload is actually 9728.
119  * However, we need to adjust for the ETHERFCSL (4 bytes) and the Ethernet VLAN
120  * header size (18 bytes) to get the actual maximum frame we can use. If
121  * different adapters end up with different sizes, we should make this value a
122  * bit more dynamic.
123  */
124 #define	I40E_MAX_MTU	9706
125 #define	I40E_MIN_MTU	ETHERMIN
126 #define	I40E_DEF_MTU	ETHERMTU
127 
128 /*
129  * Interrupt throttling related values. Interrupt throttling values are defined
130  * in two microsecond increments. Note that a value of zero basically says do no
131  * ITR activity. A helpful way to think about these is that setting the ITR to a
132  * value will allow a certain number of interrupts per second.
133  *
134  * Our default values for RX allow 20k interrupts per second while our default
135  * values for TX allow for 5k interrupts per second. For other class interrupts,
136  * we limit ourselves to a rate of 2k/s.
137  */
138 #define	I40E_MIN_ITR		0x0000
139 #define	I40E_MAX_ITR		0x0FF0
140 #define	I40E_DEF_RX_ITR		0x0019
141 #define	I40E_DEF_TX_ITR		0x0064
142 #define	I40E_DEF_OTHER_ITR	0x00FA
143 
144 /*
145  * Indexes into the three ITR registers that we have.
146  */
147 typedef enum i40e_itr_index {
148 	I40E_ITR_INDEX_RX	= 0x0,
149 	I40E_ITR_INDEX_TX	= 0x1,
150 	I40E_ITR_INDEX_OTHER	= 0x2,
151 	I40E_ITR_INDEX_NONE 	= 0x3
152 } i40e_itr_index_t;
153 
154 
155 /*
156  * Table 1-5 of the PRM notes that LSO supports up to 256 KB.
157  */
158 #define	I40E_LSO_MAXLEN	(256 * 1024)
159 
160 #define	I40E_CYCLIC_PERIOD NANOSEC	/* 1 second */
161 #define	I40E_DRAIN_RX_WAIT	(500 * MILLISEC)	/* In us */
162 
163 /*
164  * All the other queue types for are defined by the common code. However, this
165  * is the constant to indicate that it's terminated.
166  */
167 #define	I40E_QUEUE_TYPE_EOL	0x7FF
168 
169 /*
170  * See the comments in i40e_transceiver.c as to the purpose of this value and
171  * how it's used to ensure that the IP header is eventually aligned when it's
172  * received by the OS.
173  */
174 #define	I40E_BUF_IPHDR_ALIGNMENT	2
175 
176 /*
177  * The XL710 controller has a limit of eight buffers being allowed to be used
178  * for the transmission of a single frame. This is defined in 8.4.1 - Transmit
179  * Packet in System Memory.
180  */
181 #define	I40E_TX_MAX_COOKIE	8
182 
183 /*
184  * Sizing to determine the amount of available descriptors at which we'll
185  * consider ourselves blocked. Also, when we have these available, we'll then
186  * consider ourselves available to transmit to MAC again. Strictly speaking, the
187  * MAX is based on the ring size. The default sizing is based on ixgbe.
188  */
189 #define	I40E_MIN_TX_BLOCK_THRESH	I40E_TX_MAX_COOKIE
190 #define	I40E_DEF_TX_BLOCK_THRESH	I40E_MIN_TX_BLOCK_THRESH
191 
192 /*
193  * Sizing for DMA thresholds. These are used to indicate whether or not we
194  * should perform a bcopy or a DMA binding of a given message block. The range
195  * allows for setting things such that we'll always do a bcopy (a high value) or
196  * always perform a DMA binding (a low value).
197  */
198 #define	I40E_MIN_RX_DMA_THRESH		0
199 #define	I40E_DEF_RX_DMA_THRESH		256
200 #define	I40E_MAX_RX_DMA_THRESH		INT32_MAX
201 
202 #define	I40E_MIN_TX_DMA_THRESH		0
203 #define	I40E_DEF_TX_DMA_THRESH		256
204 #define	I40E_MAX_TX_DMA_THRESH		INT32_MAX
205 
206 /*
207  * Resource sizing counts. There are various aspects of hardware where we may
208  * have some variable number of elements that we need to handle. Such as the
209  * hardware capabilities and switch capacities. We cannot know a priori how many
210  * elements to do, so instead we take a starting guess and then will grow it up
211  * to an upper bound on a number of elements, to limit memory consumption in
212  * case of a hardware bug.
213  */
214 #define	I40E_HW_CAP_DEFAULT	40
215 #define	I40E_SWITCH_CAP_DEFAULT	25
216 
217 /*
218  * Host Memory Context related constants.
219  */
220 #define	I40E_HMC_RX_CTX_UNIT		128
221 #define	I40E_HMC_RX_DBUFF_MIN		1024
222 #define	I40E_HMC_RX_DBUFF_MAX		(16 * 1024 - 128)
223 #define	I40E_HMC_RX_DTYPE_NOSPLIT	0
224 #define	I40E_HMC_RX_DSIZE_32BYTE	1
225 #define	I40E_HMC_RX_CRCSTRIP_ENABLE	1
226 #define	I40E_HMC_RX_FC_DISABLE		0
227 #define	I40E_HMC_RX_L2TAGORDER		1
228 #define	I40E_HMC_RX_HDRSPLIT_DISABLE	0
229 #define	I40E_HMC_RX_INVLAN_DONTSTRIP	0
230 #define	I40E_HMC_RX_TPH_DISABLE		0
231 #define	I40E_HMC_RX_LOWRXQ_NOINTR	0
232 #define	I40E_HMC_RX_PREFENA		1
233 
234 #define	I40E_HMC_TX_CTX_UNIT		128
235 #define	I40E_HMC_TX_NEW_CONTEXT		1
236 #define	I40E_HMC_TX_FC_DISABLE		0
237 #define	I40E_HMC_TX_TS_DISABLE		0
238 #define	I40E_HMC_TX_FD_DISABLE		0
239 #define	I40E_HMC_TX_ALT_VLAN_DISABLE	0
240 #define	I40E_HMC_TX_WB_ENABLE		1
241 #define	I40E_HMC_TX_TPH_DISABLE		0
242 
243 /*
244  * Whenever we establish and create a VSI, we need to assign some number of
245  * queues that it's allowed to access from the PF. Because we only have a single
246  * VSI per PF at this time, we assign it all the queues.
247  *
248  * Many of the devices support what's called Data-center Bridging. Which is a
249  * feature that we don't have much use of at this time. However, we still need
250  * to fill in this information. We follow the guidance of the note in Table 7-80
251  * which talks about bytes 62-77. It says that if we don't want to assign
252  * anything to traffic classes, we should set the field to zero. Effectively
253  * this means that everything in the system is assigned to traffic class zero.
254  */
255 #define	I40E_ASSIGN_ALL_QUEUES		0
256 #define	I40E_TRAFFIC_CLASS_NO_QUEUES	0
257 
258 /*
259  * This defines the error mask that we care about from rx descriptors. Currently
260  * we're only concerned with the general errors and oversize errors.
261  */
262 #define	I40E_RX_ERR_BITS	((1 << I40E_RX_DESC_ERROR_RXE_SHIFT) | \
263 	(1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT))
264 
265 /*
266  * Property sizing macros for firmware versions, etc. They need to be large
267  * enough to hold 32-bit quantities transformed to strings as %d.%d or %x.
268  */
269 #define	I40E_DDI_PROP_LEN	64
270 
271 /*
272  * We currently consolidate some overrides that we use in the code here. These
273  * will be gone in the fullness of time, but as we're bringing up the device,
274  * this is what we use.
275  */
276 #define	I40E_GROUP_MAX		1
277 #define	I40E_TRQPAIR_MAX	1
278 
279 #define	I40E_GROUP_NOMSIX	1
280 #define	I40E_TRQPAIR_NOMSIX	1
281 
282 /*
283  * It seems reasonable to cast this to void because the only reason that we
284  * should be getting a DDI_FAILURE is due to the fact that we specify addresses
285  * out of range. Because we specify no offset or address, it shouldn't happen.
286  */
287 #ifdef	DEBUG
288 #define	I40E_DMA_SYNC(handle, flag)	ASSERT0(ddi_dma_sync( \
289 					    (handle)->dmab_dma_handle, 0, 0, \
290 					    (flag)))
291 #else	/* !DEBUG */
292 #define	I40E_DMA_SYNC(handle, flag)	((void) ddi_dma_sync( \
293 					    (handle)->dmab_dma_handle, 0, 0, \
294 					    (flag)))
295 #endif	/* DEBUG */
296 
297 /*
298  * Constants related to ring startup and teardown. These refer to the amount of
299  * time that we're willing to wait for a ring to spin up and spin down.
300  */
301 #define	I40E_RING_WAIT_NTRIES	10
302 #define	I40E_RING_WAIT_PAUSE	10	/* ms */
303 
304 /*
305  * Bit flags for attach_progress
306  */
307 typedef enum i40e_attach_state {
308 	I40E_ATTACH_PCI_CONFIG	= 0x0001,	/* PCI config setup */
309 	I40E_ATTACH_REGS_MAP	= 0x0002,	/* Registers mapped */
310 	I40E_ATTACH_PROPS	= 0x0004,	/* Properties initialized */
311 	I40E_ATTACH_ALLOC_INTR	= 0x0008,	/* Interrupts allocated */
312 	I40E_ATTACH_ALLOC_RINGSLOCKS	= 0x0010, /* Rings & locks allocated */
313 	I40E_ATTACH_ADD_INTR	= 0x0020,	/* Intr handlers added */
314 	I40E_ATTACH_COMMON_CODE	= 0x0040, 	/* Intel code initialized */
315 	I40E_ATTACH_INIT	= 0x0080,	/* Device initialized */
316 	I40E_ATTACH_STATS	= 0x0200,	/* Kstats created */
317 	I40E_ATTACH_MAC		= 0x0800,	/* MAC registered */
318 	I40E_ATTACH_ENABLE_INTR	= 0x1000,	/* DDI interrupts enabled */
319 	I40E_ATTACH_FM_INIT	= 0x2000,	/* FMA initialized */
320 	I40E_ATTACH_LINK_TIMER	= 0x4000,	/* link check timer */
321 } i40e_attach_state_t;
322 
323 
324 /*
325  * State flags that what's going on in in the device. Some of these state flags
326  * indicate some aspirational work that needs to happen in the driver.
327  *
328  * I40E_UNKNOWN:	The device has yet to be started.
329  * I40E_INITIALIZED:	The device has been fully attached.
330  * I40E_STARTED:	The device has come out of the GLDV3 start routine.
331  * I40E_SUSPENDED:	The device is suspended and I/O among other things
332  * 			should not occur. This happens because of an actual
333  * 			DDI_SUSPEND or interrupt adjustments.
334  * I40E_STALL:		The tx stall detection logic has found a stall.
335  * I40E_OVERTEMP:	The device has encountered a temperature alarm.
336  * I40E_INTR_ADJUST:	Our interrupts are being manipulated and therefore we
337  * 			shouldn't be manipulating their state.
338  * I40E_ERROR:		We've detected an FM error and degraded the device.
339  */
340 typedef enum i40e_state {
341 	I40E_UNKNOWN		= 0x00,
342 	I40E_INITIALIZED	= 0x01,
343 	I40E_STARTED		= 0x02,
344 	I40E_SUSPENDED		= 0x04,
345 	I40E_STALL		= 0x08,
346 	I40E_OVERTEMP		= 0x20,
347 	I40E_INTR_ADJUST	= 0x40,
348 	I40E_ERROR		= 0x80
349 } i40e_state_t;
350 
351 
352 /*
353  * Definitions for common Intel things that we use and some slightly more usable
354  * names.
355  */
356 typedef struct i40e_hw i40e_hw_t;
357 typedef struct i40e_aqc_switch_resource_alloc_element_resp i40e_switch_rsrc_t;
358 
359 /*
360  * Handles and addresses of DMA buffers.
361  */
362 typedef struct i40e_dma_buffer {
363 	caddr_t		dmab_address;		/* Virtual address */
364 	uint64_t	dmab_dma_address;	/* DMA (Hardware) address */
365 	ddi_acc_handle_t dmab_acc_handle;	/* Data access handle */
366 	ddi_dma_handle_t dmab_dma_handle;	/* DMA handle */
367 	size_t		dmab_size;		/* Buffer size */
368 	size_t		dmab_len;		/* Data length in the buffer */
369 } i40e_dma_buffer_t;
370 
371 /*
372  * RX Control Block
373  */
374 typedef struct i40e_rx_control_block {
375 	mblk_t			*rcb_mp;
376 	uint32_t		rcb_ref;
377 	i40e_dma_buffer_t	rcb_dma;
378 	frtn_t			rcb_free_rtn;
379 	struct i40e_rx_data	*rcb_rxd;
380 } i40e_rx_control_block_t;
381 
382 typedef enum {
383 	I40E_TX_NONE,
384 	I40E_TX_COPY,
385 	I40E_TX_DMA
386 } i40e_tx_type_t;
387 
388 typedef struct i40e_tx_desc i40e_tx_desc_t;
389 typedef union i40e_32byte_rx_desc i40e_rx_desc_t;
390 
391 typedef struct i40e_tx_control_block {
392 	struct i40e_tx_control_block	*tcb_next;
393 	mblk_t				*tcb_mp;
394 	i40e_tx_type_t			tcb_type;
395 	ddi_dma_handle_t		tcb_dma_handle;
396 	i40e_dma_buffer_t		tcb_dma;
397 } i40e_tx_control_block_t;
398 
399 /*
400  * Receive ring data (used below).
401  */
402 typedef struct i40e_rx_data {
403 	struct i40e	*rxd_i40e;
404 
405 	/*
406 	 * RX descriptor ring definitions
407 	 */
408 	i40e_dma_buffer_t rxd_desc_area;	/* DMA buffer of rx desc ring */
409 	i40e_rx_desc_t *rxd_desc_ring;		/* Rx desc ring */
410 	uint32_t rxd_desc_next;			/* Index of next rx desc */
411 
412 	/*
413 	 * RX control block list definitions
414 	 */
415 	kmutex_t		rxd_free_lock;	/* Lock to protect free data */
416 	i40e_rx_control_block_t	*rxd_rcb_area;	/* Array of control blocks */
417 	i40e_rx_control_block_t	**rxd_work_list; /* Work list of rcbs */
418 	i40e_rx_control_block_t	**rxd_free_list; /* Free list of rcbs */
419 	uint32_t		rxd_rcb_free;	/* Number of free rcbs */
420 
421 	/*
422 	 * RX software ring settings
423 	 */
424 	uint32_t	rxd_ring_size;		/* Rx descriptor ring size */
425 	uint32_t	rxd_free_list_size;	/* Rx free list size */
426 
427 	/*
428 	 * RX outstanding data. This is used to keep track of outstanding loaned
429 	 * descriptors after we've shut down receiving information. Note these
430 	 * are protected by the i40e_t`i40e_rx_pending_lock.
431 	 */
432 	uint32_t	rxd_rcb_pending;
433 	boolean_t	rxd_shutdown;
434 } i40e_rx_data_t;
435 
436 /*
437  * Structures for unicast and multicast addresses. Note that we keep the VSI id
438  * around for unicast addresses, since they may belong to different VSIs.
439  * However, since all multicast addresses belong to the default VSI, we don't
440  * duplicate that information.
441  */
442 typedef struct i40e_uaddr {
443 	uint8_t iua_mac[ETHERADDRL];
444 	int	iua_vsi;
445 } i40e_uaddr_t;
446 
447 typedef struct i40e_maddr {
448 	uint8_t ima_mac[ETHERADDRL];
449 } i40e_maddr_t;
450 
451 /*
452  * Collection of RX statistics on a given queue.
453  */
454 typedef struct i40e_rxq_stat {
455 	/*
456 	 * The i40e hardware does not maintain statistics on a per-ring basis,
457 	 * only on a per-PF and per-VSI level. As such, to satisfy the GLDv3, we
458 	 * need to maintain our own stats for packets and bytes.
459 	 */
460 	kstat_named_t	irxs_bytes;	/* Bytes in on queue */
461 	kstat_named_t	irxs_packets;	/* Packets in on queue */
462 
463 	/*
464 	 * The following set of stats cover non-checksum data path issues.
465 	 */
466 	kstat_named_t	irxs_rx_desc_error;	/* Error bit set on desc */
467 	kstat_named_t	irxs_rx_copy_nomem;	/* allocb failure for copy */
468 	kstat_named_t	irxs_rx_intr_limit;	/* Hit i40e_rx_limit_per_intr */
469 	kstat_named_t	irxs_rx_bind_norcb;	/* No replacement rcb free */
470 	kstat_named_t	irxs_rx_bind_nomp;	/* No mblk_t in bind rcb */
471 
472 	/*
473 	 * The following set of statistics covers rx checksum related activity.
474 	 * These are all primarily set in i40e_rx_hcksum. If rx checksum
475 	 * activity is disabled, then these should all be zero.
476 	 */
477 	kstat_named_t	irxs_hck_v4hdrok;	/* Valid IPv4 Header */
478 	kstat_named_t	irxs_hck_l4hdrok;	/* Valid L4 Header */
479 	kstat_named_t	irxs_hck_unknown;	/* !pinfo.known */
480 	kstat_named_t	irxs_hck_nol3l4p;	/* Missing L3L4P bit in desc */
481 	kstat_named_t	irxs_hck_iperr;		/* IPE error bit set */
482 	kstat_named_t	irxs_hck_eiperr;	/* EIPE error bit set */
483 	kstat_named_t	irxs_hck_l4err;		/* L4E error bit set */
484 	kstat_named_t	irxs_hck_v6skip;	/* IPv6 case hw fails on */
485 	kstat_named_t	irxs_hck_set;		/* Total times we set cksum */
486 	kstat_named_t	irxs_hck_miss;		/* Times with zero cksum bits */
487 } i40e_rxq_stat_t;
488 
489 /*
490  * Collection of TX Statistics on a given queue
491  */
492 typedef struct i40e_txq_stat {
493 	kstat_named_t	itxs_bytes;		/* Bytes out on queue */
494 	kstat_named_t	itxs_packets;		/* Packets out on queue */
495 	kstat_named_t	itxs_descriptors;	/* Descriptors issued */
496 	kstat_named_t	itxs_recycled;		/* Descriptors reclaimed */
497 	/*
498 	 * Various failure conditions.
499 	 */
500 	kstat_named_t	itxs_hck_meoifail;	/* ether offload failures */
501 	kstat_named_t	itxs_hck_nol2info;	/* Missing l2 info */
502 	kstat_named_t	itxs_hck_nol3info;	/* Missing l3 info */
503 	kstat_named_t	itxs_hck_nol4info;	/* Missing l4 info */
504 	kstat_named_t	itxs_hck_badl3;		/* Not IPv4/IPv6 */
505 	kstat_named_t	itxs_hck_badl4;		/* Bad L4 Paylaod */
506 
507 	kstat_named_t	itxs_err_notcb;		/* No tcb's available */
508 	kstat_named_t	itxs_err_nodescs;	/* No tcb's available */
509 	kstat_named_t	itxs_err_context;	/* Total context failures */
510 
511 	kstat_named_t	itxs_num_unblocked;	/* Number of MAC unblocks */
512 } i40e_txq_stat_t;
513 
514 /*
515  * An instance of an XL710 transmit/receive queue pair. This currently
516  * represents a combination of both a transmit and receive ring, though they
517  * should really be split apart into separate logical structures. Unfortunately,
518  * during initial work we mistakenly joined them together.
519  */
520 typedef struct i40e_trqpair {
521 	struct i40e *itrq_i40e;
522 
523 	/* Receive-side structures. */
524 	kmutex_t itrq_rx_lock;
525 	mac_ring_handle_t itrq_macrxring; /* Receive ring handle. */
526 	i40e_rx_data_t *itrq_rxdata;	/* Receive ring rx data. */
527 	uint64_t itrq_rxgen;		/* Generation number for mac/GLDv3. */
528 	uint32_t itrq_index;		/* Queue index in the PF */
529 	uint32_t itrq_rx_intrvec;	/* Receive interrupt vector. */
530 	boolean_t itrq_intr_poll;	/* True when polling */
531 
532 	/* Receive-side stats. */
533 	i40e_rxq_stat_t	itrq_rxstat;
534 	kstat_t	*itrq_rxkstat;
535 
536 	/* Transmit-side structures. */
537 	kmutex_t itrq_tx_lock;
538 	mac_ring_handle_t itrq_mactxring; /* Transmit ring handle. */
539 	uint32_t itrq_tx_intrvec;	/* Transmit interrupt vector. */
540 	boolean_t itrq_tx_blocked;	/* Does MAC think we're blocked? */
541 
542 	/*
543 	 * TX data sizing
544 	 */
545 	uint32_t		itrq_tx_ring_size;
546 	uint32_t		itrq_tx_free_list_size;
547 
548 	/*
549 	 * TX descriptor ring data
550 	 */
551 	i40e_dma_buffer_t	itrq_desc_area;	/* DMA buffer of tx desc ring */
552 	i40e_tx_desc_t		*itrq_desc_ring; /* TX Desc ring */
553 	volatile uint32_t 	*itrq_desc_wbhead; /* TX write-back index */
554 	uint32_t		itrq_desc_head;	/* Last index hw freed */
555 	uint32_t		itrq_desc_tail;	/* Index of next free desc */
556 	uint32_t		itrq_desc_free;	/* Number of free descriptors */
557 
558 	/*
559 	 * TX control block (tcb) data
560 	 */
561 	kmutex_t		itrq_tcb_lock;
562 	i40e_tx_control_block_t	*itrq_tcb_area;	/* Array of control blocks */
563 	i40e_tx_control_block_t	**itrq_tcb_work_list;	/* In use tcb */
564 	i40e_tx_control_block_t	**itrq_tcb_free_list;	/* Available tcb */
565 	uint32_t		itrq_tcb_free;	/* Count of free tcb */
566 
567 	/* Transmit-side stats. */
568 	i40e_txq_stat_t		itrq_txstat;
569 	kstat_t			*itrq_txkstat;
570 
571 } i40e_trqpair_t;
572 
573 /*
574  * VSI statistics.
575  *
576  * This mirrors the i40e_eth_stats structure but transforms it into a kstat.
577  * Note that the stock statistic structure also includes entries for tx
578  * discards. However, this is not actually implemented for the VSI (see Table
579  * 7-221), hence why we don't include the member which would always have a value
580  * of zero. This choice was made to minimize confusion to someone looking at
581  * these, as a value of zero does not necessarily equate to the fact that it's
582  * not implemented.
583  */
584 typedef struct i40e_vsi_stats {
585 	uint64_t ivs_rx_bytes;			/* gorc */
586 	uint64_t ivs_rx_unicast;		/* uprc */
587 	uint64_t ivs_rx_multicast;		/* mprc */
588 	uint64_t ivs_rx_broadcast;		/* bprc */
589 	uint64_t ivs_rx_discards;		/* rdpc */
590 	uint64_t ivs_rx_unknown_protocol;	/* rupp */
591 	uint64_t ivs_tx_bytes;			/* gotc */
592 	uint64_t ivs_tx_unicast;		/* uptc */
593 	uint64_t ivs_tx_multicast;		/* mptc */
594 	uint64_t ivs_tx_broadcast;		/* bptc */
595 	uint64_t ivs_tx_errors;			/* tepc */
596 } i40e_vsi_stats_t;
597 
598 typedef struct i40e_vsi_kstats {
599 	kstat_named_t	ivk_rx_bytes;
600 	kstat_named_t	ivk_rx_unicast;
601 	kstat_named_t	ivk_rx_multicast;
602 	kstat_named_t	ivk_rx_broadcast;
603 	kstat_named_t	ivk_rx_discards;
604 	kstat_named_t	ivk_rx_unknown_protocol;
605 	kstat_named_t	ivk_tx_bytes;
606 	kstat_named_t	ivk_tx_unicast;
607 	kstat_named_t	ivk_tx_multicast;
608 	kstat_named_t	ivk_tx_broadcast;
609 	kstat_named_t	ivk_tx_errors;
610 } i40e_vsi_kstats_t;
611 
612 /*
613  * For pf statistics, we opt not to use the standard statistics as defined by
614  * the Intel common code. This also currently combines statistics that are
615  * global across the entire device.
616  */
617 typedef struct i40e_pf_stats {
618 	uint64_t ips_rx_bytes;			/* gorc */
619 	uint64_t ips_rx_unicast;		/* uprc */
620 	uint64_t ips_rx_multicast;		/* mprc */
621 	uint64_t ips_rx_broadcast;		/* bprc */
622 	uint64_t ips_tx_bytes;			/* gotc */
623 	uint64_t ips_tx_unicast;		/* uptc */
624 	uint64_t ips_tx_multicast;		/* mptc */
625 	uint64_t ips_tx_broadcast;		/* bptc */
626 
627 	uint64_t ips_rx_size_64;		/* prc64 */
628 	uint64_t ips_rx_size_127;		/* prc127 */
629 	uint64_t ips_rx_size_255;		/* prc255 */
630 	uint64_t ips_rx_size_511;		/* prc511 */
631 	uint64_t ips_rx_size_1023;		/* prc1023 */
632 	uint64_t ips_rx_size_1522;		/* prc1522 */
633 	uint64_t ips_rx_size_9522;		/* prc9522 */
634 
635 	uint64_t ips_tx_size_64;		/* ptc64 */
636 	uint64_t ips_tx_size_127;		/* ptc127 */
637 	uint64_t ips_tx_size_255;		/* ptc255 */
638 	uint64_t ips_tx_size_511;		/* ptc511 */
639 	uint64_t ips_tx_size_1023;		/* ptc1023 */
640 	uint64_t ips_tx_size_1522;		/* ptc1522 */
641 	uint64_t ips_tx_size_9522;		/* ptc9522 */
642 
643 	uint64_t ips_link_xon_rx;		/* lxonrxc */
644 	uint64_t ips_link_xoff_rx;		/* lxoffrxc */
645 	uint64_t ips_link_xon_tx;		/* lxontxc */
646 	uint64_t ips_link_xoff_tx;		/* lxofftxc */
647 	uint64_t ips_priority_xon_rx[8];	/* pxonrxc[8] */
648 	uint64_t ips_priority_xoff_rx[8];	/* pxoffrxc[8] */
649 	uint64_t ips_priority_xon_tx[8];	/* pxontxc[8] */
650 	uint64_t ips_priority_xoff_tx[8];	/* pxofftxc[8] */
651 	uint64_t ips_priority_xon_2_xoff[8];	/* rxon2offcnt[8] */
652 
653 	uint64_t ips_crc_errors;		/* crcerrs */
654 	uint64_t ips_illegal_bytes;		/* illerrc */
655 	uint64_t ips_mac_local_faults;		/* mlfc */
656 	uint64_t ips_mac_remote_faults;		/* mrfc */
657 	uint64_t ips_rx_length_errors;		/* rlec */
658 	uint64_t ips_rx_undersize;		/* ruc */
659 	uint64_t ips_rx_fragments;		/* rfc */
660 	uint64_t ips_rx_oversize;		/* roc */
661 	uint64_t ips_rx_jabber;			/* rjc */
662 	uint64_t ips_rx_discards;		/* rdpc */
663 	uint64_t ips_rx_vm_discards;		/* ldpc */
664 	uint64_t ips_rx_short_discards;		/* mspdc */
665 	uint64_t ips_tx_dropped_link_down;	/* tdold */
666 	uint64_t ips_rx_unknown_protocol;	/* rupp */
667 	uint64_t ips_rx_err1;			/* rxerr1 */
668 	uint64_t ips_rx_err2;			/* rxerr2 */
669 } i40e_pf_stats_t;
670 
671 typedef struct i40e_pf_kstats {
672 	kstat_named_t ipk_rx_bytes;		/* gorc */
673 	kstat_named_t ipk_rx_unicast;		/* uprc */
674 	kstat_named_t ipk_rx_multicast;		/* mprc */
675 	kstat_named_t ipk_rx_broadcast;		/* bprc */
676 	kstat_named_t ipk_tx_bytes;		/* gotc */
677 	kstat_named_t ipk_tx_unicast;		/* uptc */
678 	kstat_named_t ipk_tx_multicast;		/* mptc */
679 	kstat_named_t ipk_tx_broadcast;		/* bptc */
680 
681 	kstat_named_t ipk_rx_size_64;		/* prc64 */
682 	kstat_named_t ipk_rx_size_127;		/* prc127 */
683 	kstat_named_t ipk_rx_size_255;		/* prc255 */
684 	kstat_named_t ipk_rx_size_511;		/* prc511 */
685 	kstat_named_t ipk_rx_size_1023;		/* prc1023 */
686 	kstat_named_t ipk_rx_size_1522;		/* prc1522 */
687 	kstat_named_t ipk_rx_size_9522;		/* prc9522 */
688 
689 	kstat_named_t ipk_tx_size_64;		/* ptc64 */
690 	kstat_named_t ipk_tx_size_127;		/* ptc127 */
691 	kstat_named_t ipk_tx_size_255;		/* ptc255 */
692 	kstat_named_t ipk_tx_size_511;		/* ptc511 */
693 	kstat_named_t ipk_tx_size_1023;		/* ptc1023 */
694 	kstat_named_t ipk_tx_size_1522;		/* ptc1522 */
695 	kstat_named_t ipk_tx_size_9522;		/* ptc9522 */
696 
697 	kstat_named_t ipk_link_xon_rx;		/* lxonrxc */
698 	kstat_named_t ipk_link_xoff_rx;		/* lxoffrxc */
699 	kstat_named_t ipk_link_xon_tx;		/* lxontxc */
700 	kstat_named_t ipk_link_xoff_tx;		/* lxofftxc */
701 	kstat_named_t ipk_priority_xon_rx[8];	/* pxonrxc[8] */
702 	kstat_named_t ipk_priority_xoff_rx[8];	/* pxoffrxc[8] */
703 	kstat_named_t ipk_priority_xon_tx[8];	/* pxontxc[8] */
704 	kstat_named_t ipk_priority_xoff_tx[8];	/* pxofftxc[8] */
705 	kstat_named_t ipk_priority_xon_2_xoff[8];	/* rxon2offcnt[8] */
706 
707 	kstat_named_t ipk_crc_errors;		/* crcerrs */
708 	kstat_named_t ipk_illegal_bytes;	/* illerrc */
709 	kstat_named_t ipk_mac_local_faults;	/* mlfc */
710 	kstat_named_t ipk_mac_remote_faults;	/* mrfc */
711 	kstat_named_t ipk_rx_length_errors;	/* rlec */
712 	kstat_named_t ipk_rx_undersize;		/* ruc */
713 	kstat_named_t ipk_rx_fragments;		/* rfc */
714 	kstat_named_t ipk_rx_oversize;		/* roc */
715 	kstat_named_t ipk_rx_jabber;		/* rjc */
716 	kstat_named_t ipk_rx_discards;		/* rdpc */
717 	kstat_named_t ipk_rx_vm_discards;	/* ldpc */
718 	kstat_named_t ipk_rx_short_discards;	/* mspdc */
719 	kstat_named_t ipk_tx_dropped_link_down;	/* tdold */
720 	kstat_named_t ipk_rx_unknown_protocol;	/* rupp */
721 	kstat_named_t ipk_rx_err1;		/* rxerr1 */
722 	kstat_named_t ipk_rx_err2;		/* rxerr2 */
723 } i40e_pf_kstats_t;
724 
725 /*
726  * Resources that are pooled and specific to a given i40e_t.
727  */
728 typedef struct i40e_func_rsrc {
729 	uint_t	ifr_nrx_queue;
730 	uint_t	ifr_nrx_queue_used;
731 	uint_t	ifr_ntx_queue;
732 	uint_t	ifr_trx_queue_used;
733 	uint_t	ifr_nvsis;
734 	uint_t	ifr_nvsis_used;
735 	uint_t	ifr_nmacfilt;
736 	uint_t	ifr_nmacfilt_used;
737 	uint_t	ifr_nmcastfilt;
738 	uint_t	ifr_nmcastfilt_used;
739 } i40e_func_rsrc_t;
740 
741 /*
742  * Main i40e per-instance state.
743  */
744 typedef struct i40e {
745 	list_node_t	i40e_glink;		/* Global list link */
746 	list_node_t	i40e_dlink;		/* Device list link */
747 	kmutex_t	i40e_general_lock;	/* General device lock */
748 
749 	/*
750 	 * General Data and management
751 	 */
752 	dev_info_t	*i40e_dip;
753 	int		i40e_instance;
754 	int		i40e_fm_capabilities;
755 	uint_t		i40e_state;
756 	i40e_attach_state_t i40e_attach_progress;
757 	mac_handle_t	i40e_mac_hdl;
758 	ddi_periodic_t	i40e_periodic_id;
759 
760 	/*
761 	 * Pointers to common code data structures and memory for the common
762 	 * code.
763 	 */
764 	struct i40e_hw				i40e_hw_space;
765 	struct i40e_osdep			i40e_osdep_space;
766 	struct i40e_aq_get_phy_abilities_resp	i40e_phy;
767 	void 					*i40e_aqbuf;
768 
769 	/*
770 	 * Device state, switch information, and resources.
771 	 */
772 	int			i40e_vsi_id;
773 	uint16_t		i40e_vsi_num;
774 	struct i40e_device	*i40e_device;
775 	i40e_func_rsrc_t	i40e_resources;
776 	uint16_t		i40e_switch_rsrc_alloc;
777 	uint16_t		i40e_switch_rsrc_actual;
778 	i40e_switch_rsrc_t	*i40e_switch_rsrcs;
779 	i40e_uaddr_t		*i40e_uaddrs;
780 	i40e_maddr_t		*i40e_maddrs;
781 	int			i40e_mcast_promisc_count;
782 	boolean_t		i40e_promisc_on;
783 	link_state_t		i40e_link_state;
784 	uint32_t		i40e_link_speed;	/* In Mbps */
785 	link_duplex_t		i40e_link_duplex;
786 	uint_t			i40e_sdu;
787 	uint_t			i40e_frame_max;
788 
789 	/*
790 	 * Transmit and receive information, tunables, and MAC info.
791 	 */
792 	i40e_trqpair_t	*i40e_trqpairs;
793 	boolean_t 	i40e_mr_enable;
794 	int		i40e_num_trqpairs;
795 	uint_t		i40e_other_itr;
796 
797 	int		i40e_num_rx_groups;
798 	int		i40e_num_rx_descs;
799 	mac_group_handle_t i40e_rx_group_handle;
800 	uint32_t	i40e_rx_ring_size;
801 	uint32_t	i40e_rx_buf_size;
802 	boolean_t	i40e_rx_hcksum_enable;
803 	uint32_t	i40e_rx_dma_min;
804 	uint32_t	i40e_rx_limit_per_intr;
805 	uint_t		i40e_rx_itr;
806 
807 	int		i40e_num_tx_descs;
808 	uint32_t	i40e_tx_ring_size;
809 	uint32_t	i40e_tx_buf_size;
810 	uint32_t	i40e_tx_block_thresh;
811 	boolean_t	i40e_tx_hcksum_enable;
812 	uint32_t	i40e_tx_dma_min;
813 	uint_t		i40e_tx_itr;
814 
815 	/*
816 	 * Interrupt state
817 	 */
818 	uint_t		i40e_intr_pri;
819 	uint_t		i40e_intr_force;
820 	uint_t		i40e_intr_type;
821 	int		i40e_intr_cap;
822 	uint32_t	i40e_intr_count;
823 	uint32_t	i40e_intr_count_max;
824 	uint32_t	i40e_intr_count_min;
825 	size_t		i40e_intr_size;
826 	ddi_intr_handle_t *i40e_intr_handles;
827 	ddi_cb_handle_t	i40e_callback_handle;
828 
829 	/*
830 	 * DMA attributes. See i40e_transceiver.c for why we have copies of them
831 	 * in the i40e_t.
832 	 */
833 	ddi_dma_attr_t		i40e_static_dma_attr;
834 	ddi_dma_attr_t		i40e_txbind_dma_attr;
835 	ddi_device_acc_attr_t	i40e_desc_acc_attr;
836 	ddi_device_acc_attr_t	i40e_buf_acc_attr;
837 
838 	/*
839 	 * The following two fields are used to protect and keep track of
840 	 * outstanding, loaned buffers to MAC. If we have these, we can't
841 	 * detach as we have active DMA memory outstanding.
842 	 */
843 	kmutex_t	i40e_rx_pending_lock;
844 	kcondvar_t	i40e_rx_pending_cv;
845 	uint32_t	i40e_rx_pending;
846 
847 	/*
848 	 * PF statistics and VSI statistics.
849 	 */
850 	kmutex_t		i40e_stat_lock;
851 	kstat_t			*i40e_pf_kstat;
852 	kstat_t			*i40e_vsi_kstat;
853 	i40e_pf_stats_t		i40e_pf_stat;
854 	i40e_vsi_stats_t	i40e_vsi_stat;
855 	uint16_t		i40e_vsi_stat_id;
856 
857 	/*
858 	 * Misc. stats and counters that should maybe one day be kstats.
859 	 */
860 	uint64_t	i40e_s_link_status_errs;
861 	uint32_t	i40e_s_link_status_lasterr;
862 } i40e_t;
863 
864 /*
865  * The i40e_device represents a PCI device which encapsulates multiple physical
866  * functions which are represented as an i40e_t. This is used to track the use
867  * of pooled resources throughout all of the various devices.
868  */
869 typedef struct i40e_device {
870 	list_node_t	id_link;
871 	dev_info_t	*id_parent;
872 	uint_t		id_pci_bus;
873 	uint_t		id_pci_device;
874 	uint_t		id_nfuncs;	/* Total number of functions */
875 	uint_t		id_nreg;	/* Total number present */
876 	list_t		id_i40e_list;	/* List of i40e_t's registered */
877 	i40e_switch_rsrc_t	*id_rsrcs; /* Switch resources for this PF */
878 	uint_t		id_rsrcs_alloc;	/* Total allocated resources */
879 	uint_t		id_rsrcs_act;	/* Actual number of resources */
880 } i40e_device_t;
881 
882 /* Values for the interrupt forcing on the NIC. */
883 #define	I40E_INTR_NONE			0
884 #define	I40E_INTR_MSIX			1
885 #define	I40E_INTR_MSI			2
886 #define	I40E_INTR_LEGACY		3
887 
888 /* Hint that we don't want to do any polling... */
889 #define	I40E_POLL_NULL			-1
890 
891 /*
892  * Logging functions.
893  */
894 /*PRINTFLIKE2*/
895 extern void i40e_error(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
896 /*PRINTFLIKE2*/
897 extern void i40e_notice(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
898 /*PRINTFLIKE2*/
899 extern void i40e_log(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
900 
901 /*
902  * General link handling functions.
903  */
904 extern void i40e_link_check(i40e_t *);
905 extern void i40e_update_mtu(i40e_t *);
906 
907 /*
908  * FMA functions.
909  */
910 extern int i40e_check_acc_handle(ddi_acc_handle_t);
911 extern int i40e_check_dma_handle(ddi_dma_handle_t);
912 extern void i40e_fm_ereport(i40e_t *, char *);
913 
914 /*
915  * Interrupt handlers and interrupt handler setup.
916  */
917 extern void i40e_intr_chip_init(i40e_t *);
918 extern void i40e_intr_chip_fini(i40e_t *);
919 extern uint_t i40e_intr_msix(void *, void *);
920 extern uint_t i40e_intr_msi(void *, void *);
921 extern uint_t i40e_intr_legacy(void *, void *);
922 extern void i40e_intr_io_enable_all(i40e_t *);
923 extern void i40e_intr_io_disable_all(i40e_t *);
924 extern void i40e_intr_io_clear_cause(i40e_t *);
925 extern void i40e_intr_rx_queue_disable(i40e_t *, uint_t);
926 extern void i40e_intr_rx_queue_enable(i40e_t *, uint_t);
927 extern void i40e_intr_set_itr(i40e_t *, i40e_itr_index_t, uint_t);
928 
929 /*
930  * Receive-side functions
931  */
932 extern mblk_t *i40e_ring_rx(i40e_trqpair_t *, int);
933 extern mblk_t *i40e_ring_rx_poll(void *, int);
934 extern void i40e_rx_recycle(caddr_t);
935 
936 /*
937  * Transmit-side functions
938  */
939 mblk_t *i40e_ring_tx(void *, mblk_t *);
940 extern void i40e_tx_recycle_ring(i40e_trqpair_t *);
941 extern void i40e_tx_cleanup_ring(i40e_trqpair_t *);
942 
943 /*
944  * Statistics functions.
945  */
946 extern boolean_t i40e_stats_init(i40e_t *);
947 extern void i40e_stats_fini(i40e_t *);
948 extern boolean_t i40e_stat_vsi_init(i40e_t *);
949 extern void i40e_stat_vsi_fini(i40e_t *);
950 extern boolean_t i40e_stats_trqpair_init(i40e_trqpair_t *);
951 extern void i40e_stats_trqpair_fini(i40e_trqpair_t *);
952 extern int i40e_m_stat(void *, uint_t, uint64_t *);
953 extern int i40e_rx_ring_stat(mac_ring_driver_t, uint_t, uint64_t *);
954 extern int i40e_tx_ring_stat(mac_ring_driver_t, uint_t, uint64_t *);
955 
956 /*
957  * MAC/GLDv3 functions, and functions called by MAC/GLDv3 support code.
958  */
959 extern boolean_t i40e_register_mac(i40e_t *);
960 extern boolean_t i40e_start(i40e_t *, boolean_t);
961 extern void i40e_stop(i40e_t *, boolean_t);
962 
963 /*
964  * DMA & buffer functions and attributes
965  */
966 extern void i40e_init_dma_attrs(i40e_t *, boolean_t);
967 extern boolean_t i40e_alloc_ring_mem(i40e_t *);
968 extern void i40e_free_ring_mem(i40e_t *, boolean_t);
969 
970 #ifdef __cplusplus
971 }
972 #endif
973 
974 #endif /* _I40E_SW_H */
975