xref: /illumos-gate/usr/src/uts/common/io/i40e/i40e_sw.h (revision 8c69cc8fbe729fa7b091e901c4b50508ccc6bb33)
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  * 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  * Table 1-5 of the PRM notes that LSO supports up to 256 KB.
156  */
157 #define	I40E_LSO_MAXLEN	(256 * 1024)
158 
159 #define	I40E_CYCLIC_PERIOD NANOSEC	/* 1 second */
160 #define	I40E_DRAIN_RX_WAIT	(500 * MILLISEC)	/* In us */
161 
162 /*
163  * All the other queue types for are defined by the common code. However, this
164  * is the constant to indicate that it's terminated.
165  */
166 #define	I40E_QUEUE_TYPE_EOL	0x7FF
167 
168 /*
169  * See the comments in i40e_transceiver.c as to the purpose of this value and
170  * how it's used to ensure that the IP header is eventually aligned when it's
171  * received by the OS.
172  */
173 #define	I40E_BUF_IPHDR_ALIGNMENT	2
174 
175 /*
176  * The XL710 controller has a limit of eight buffers being allowed to be used
177  * for the transmission of a single frame. This is defined in 8.4.1 - Transmit
178  * Packet in System Memory.
179  */
180 #define	I40E_TX_MAX_COOKIE	8
181 
182 /*
183  * Sizing to determine the amount of available descriptors at which we'll
184  * consider ourselves blocked. Also, when we have these available, we'll then
185  * consider ourselves available to transmit to MAC again. Strictly speaking, the
186  * MAX is based on the ring size. The default sizing is based on ixgbe.
187  */
188 #define	I40E_MIN_TX_BLOCK_THRESH	I40E_TX_MAX_COOKIE
189 #define	I40E_DEF_TX_BLOCK_THRESH	I40E_MIN_TX_BLOCK_THRESH
190 
191 /*
192  * Sizing for DMA thresholds. These are used to indicate whether or not we
193  * should perform a bcopy or a DMA binding of a given message block. The range
194  * allows for setting things such that we'll always do a bcopy (a high value) or
195  * always perform a DMA binding (a low value).
196  */
197 #define	I40E_MIN_RX_DMA_THRESH		0
198 #define	I40E_DEF_RX_DMA_THRESH		256
199 #define	I40E_MAX_RX_DMA_THRESH		INT32_MAX
200 
201 #define	I40E_MIN_TX_DMA_THRESH		0
202 #define	I40E_DEF_TX_DMA_THRESH		256
203 #define	I40E_MAX_TX_DMA_THRESH		INT32_MAX
204 
205 /*
206  * Resource sizing counts. There are various aspects of hardware where we may
207  * have some variable number of elements that we need to handle. Such as the
208  * hardware capabilities and switch capacities. We cannot know a priori how many
209  * elements to do, so instead we take a starting guess and then will grow it up
210  * to an upper bound on a number of elements, to limit memory consumption in
211  * case of a hardware bug.
212  */
213 #define	I40E_HW_CAP_DEFAULT	40
214 #define	I40E_SWITCH_CAP_DEFAULT	25
215 
216 /*
217  * Host Memory Context related constants.
218  */
219 #define	I40E_HMC_RX_CTX_UNIT		128
220 #define	I40E_HMC_RX_DBUFF_MIN		1024
221 #define	I40E_HMC_RX_DBUFF_MAX		(16 * 1024 - 128)
222 #define	I40E_HMC_RX_DTYPE_NOSPLIT	0
223 #define	I40E_HMC_RX_DSIZE_32BYTE	1
224 #define	I40E_HMC_RX_CRCSTRIP_ENABLE	1
225 #define	I40E_HMC_RX_FC_DISABLE		0
226 #define	I40E_HMC_RX_L2TAGORDER		1
227 #define	I40E_HMC_RX_HDRSPLIT_DISABLE	0
228 #define	I40E_HMC_RX_INVLAN_DONTSTRIP	0
229 #define	I40E_HMC_RX_TPH_DISABLE		0
230 #define	I40E_HMC_RX_LOWRXQ_NOINTR	0
231 #define	I40E_HMC_RX_PREFENA		1
232 
233 #define	I40E_HMC_TX_CTX_UNIT		128
234 #define	I40E_HMC_TX_NEW_CONTEXT		1
235 #define	I40E_HMC_TX_FC_DISABLE		0
236 #define	I40E_HMC_TX_TS_DISABLE		0
237 #define	I40E_HMC_TX_FD_DISABLE		0
238 #define	I40E_HMC_TX_ALT_VLAN_DISABLE	0
239 #define	I40E_HMC_TX_WB_ENABLE		1
240 #define	I40E_HMC_TX_TPH_DISABLE		0
241 
242 /*
243  * Whenever we establish and create a VSI, we need to assign some number of
244  * queues that it's allowed to access from the PF. Because we only have a single
245  * VSI per PF at this time, we assign it all the queues.
246  *
247  * Many of the devices support what's called Data-center Bridging. Which is a
248  * feature that we don't have much use of at this time. However, we still need
249  * to fill in this information. We follow the guidance of the note in Table 7-80
250  * which talks about bytes 62-77. It says that if we don't want to assign
251  * anything to traffic classes, we should set the field to zero. Effectively
252  * this means that everything in the system is assigned to traffic class zero.
253  */
254 #define	I40E_ASSIGN_ALL_QUEUES		0
255 #define	I40E_TRAFFIC_CLASS_NO_QUEUES	0
256 
257 /*
258  * This defines the error mask that we care about from rx descriptors. Currently
259  * we're only concerned with the general errors and oversize errors.
260  */
261 #define	I40E_RX_ERR_BITS	((1 << I40E_RX_DESC_ERROR_RXE_SHIFT) | \
262 	(1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT))
263 
264 /*
265  * Property sizing macros for firmware versions, etc. They need to be large
266  * enough to hold 32-bit quantities transformed to strings as %d.%d or %x.
267  */
268 #define	I40E_DDI_PROP_LEN	64
269 
270 /*
271  * We currently consolidate some overrides that we use in the code here. These
272  * will be gone in the fullness of time, but as we're bringing up the device,
273  * this is what we use.
274  */
275 #define	I40E_GROUP_MAX		1
276 #define	I40E_TRQPAIR_MAX	1
277 
278 #define	I40E_GROUP_NOMSIX	1
279 #define	I40E_TRQPAIR_NOMSIX	1
280 
281 /*
282  * It seems reasonable to cast this to void because the only reason that we
283  * should be getting a DDI_FAILURE is due to the fact that we specify addresses
284  * out of range. Because we specify no offset or address, it shouldn't happen.
285  */
286 #ifdef	DEBUG
287 #define	I40E_DMA_SYNC(handle, flag)	ASSERT0(ddi_dma_sync( \
288 					    (handle)->dmab_dma_handle, 0, 0, \
289 					    (flag)))
290 #else	/* !DEBUG */
291 #define	I40E_DMA_SYNC(handle, flag)	((void) ddi_dma_sync( \
292 					    (handle)->dmab_dma_handle, 0, 0, \
293 					    (flag)))
294 #endif	/* DEBUG */
295 
296 /*
297  * Constants related to ring startup and teardown. These refer to the amount of
298  * time that we're willing to wait for a ring to spin up and spin down.
299  */
300 #define	I40E_RING_WAIT_NTRIES	10
301 #define	I40E_RING_WAIT_PAUSE	10	/* ms */
302 
303 /*
304  * Printed Board Assembly (PBA) length. These are derived from Table 6-2.
305  */
306 #define	I40E_PBANUM_LENGTH	12
307 #define	I40E_PBANUM_STRLEN	13
308 
309 /*
310  * Define the maximum number of queues for a traffic class. These values come
311  * from the 'Number and offset of queue pairs per TCs' section of the 'Add VSI
312  * Command Buffer' table. For the 710 controller family this is table 7-62
313  * (r2.5) and for the 722 this is table 38-216 (r2.0).
314  */
315 #define	I40E_710_MAX_TC_QUEUES	64
316 #define	I40E_722_MAX_TC_QUEUES	128
317 
318 /*
319  * Define the size of the HLUT table size. The HLUT table can either be 128 or
320  * 512 bytes. We always set the table size to be 512 bytes in i40e_chip_start().
321  * Note, this should not be confused with the common code's macro
322  * I40E_HASH_LUT_SIZE_512 which is the bit pattern needed to tell the card to
323  * use a 512 byte HLUT.
324  */
325 #define	I40E_HLUT_TABLE_SIZE	512
326 
327 /*
328  * Bit flags for attach_progress
329  */
330 typedef enum i40e_attach_state {
331 	I40E_ATTACH_PCI_CONFIG	= 0x0001,	/* PCI config setup */
332 	I40E_ATTACH_REGS_MAP	= 0x0002,	/* Registers mapped */
333 	I40E_ATTACH_PROPS	= 0x0004,	/* Properties initialized */
334 	I40E_ATTACH_ALLOC_INTR	= 0x0008,	/* Interrupts allocated */
335 	I40E_ATTACH_ALLOC_RINGSLOCKS	= 0x0010, /* Rings & locks allocated */
336 	I40E_ATTACH_ADD_INTR	= 0x0020,	/* Intr handlers added */
337 	I40E_ATTACH_COMMON_CODE	= 0x0040, 	/* Intel code initialized */
338 	I40E_ATTACH_INIT	= 0x0080,	/* Device initialized */
339 	I40E_ATTACH_STATS	= 0x0200,	/* Kstats created */
340 	I40E_ATTACH_MAC		= 0x0800,	/* MAC registered */
341 	I40E_ATTACH_ENABLE_INTR	= 0x1000,	/* DDI interrupts enabled */
342 	I40E_ATTACH_FM_INIT	= 0x2000,	/* FMA initialized */
343 	I40E_ATTACH_LINK_TIMER	= 0x4000,	/* link check timer */
344 } i40e_attach_state_t;
345 
346 
347 /*
348  * State flags that what's going on in in the device. Some of these state flags
349  * indicate some aspirational work that needs to happen in the driver.
350  *
351  * I40E_UNKNOWN:	The device has yet to be started.
352  * I40E_INITIALIZED:	The device has been fully attached.
353  * I40E_STARTED:	The device has come out of the GLDV3 start routine.
354  * I40E_SUSPENDED:	The device is suspended and I/O among other things
355  * 			should not occur. This happens because of an actual
356  * 			DDI_SUSPEND or interrupt adjustments.
357  * I40E_STALL:		The tx stall detection logic has found a stall.
358  * I40E_OVERTEMP:	The device has encountered a temperature alarm.
359  * I40E_INTR_ADJUST:	Our interrupts are being manipulated and therefore we
360  * 			shouldn't be manipulating their state.
361  * I40E_ERROR:		We've detected an FM error and degraded the device.
362  */
363 typedef enum i40e_state {
364 	I40E_UNKNOWN		= 0x00,
365 	I40E_INITIALIZED	= 0x01,
366 	I40E_STARTED		= 0x02,
367 	I40E_SUSPENDED		= 0x04,
368 	I40E_STALL		= 0x08,
369 	I40E_OVERTEMP		= 0x20,
370 	I40E_INTR_ADJUST	= 0x40,
371 	I40E_ERROR		= 0x80
372 } i40e_state_t;
373 
374 
375 /*
376  * Definitions for common Intel things that we use and some slightly more usable
377  * names.
378  */
379 typedef struct i40e_hw i40e_hw_t;
380 typedef struct i40e_aqc_switch_resource_alloc_element_resp i40e_switch_rsrc_t;
381 
382 /*
383  * Handles and addresses of DMA buffers.
384  */
385 typedef struct i40e_dma_buffer {
386 	caddr_t		dmab_address;		/* Virtual address */
387 	uint64_t	dmab_dma_address;	/* DMA (Hardware) address */
388 	ddi_acc_handle_t dmab_acc_handle;	/* Data access handle */
389 	ddi_dma_handle_t dmab_dma_handle;	/* DMA handle */
390 	size_t		dmab_size;		/* Buffer size */
391 	size_t		dmab_len;		/* Data length in the buffer */
392 } i40e_dma_buffer_t;
393 
394 /*
395  * RX Control Block
396  */
397 typedef struct i40e_rx_control_block {
398 	mblk_t			*rcb_mp;
399 	uint32_t		rcb_ref;
400 	i40e_dma_buffer_t	rcb_dma;
401 	frtn_t			rcb_free_rtn;
402 	struct i40e_rx_data	*rcb_rxd;
403 } i40e_rx_control_block_t;
404 
405 typedef enum {
406 	I40E_TX_NONE,
407 	I40E_TX_COPY,
408 	I40E_TX_DMA
409 } i40e_tx_type_t;
410 
411 typedef struct i40e_tx_desc i40e_tx_desc_t;
412 typedef union i40e_32byte_rx_desc i40e_rx_desc_t;
413 
414 typedef struct i40e_tx_control_block {
415 	struct i40e_tx_control_block	*tcb_next;
416 	mblk_t				*tcb_mp;
417 	i40e_tx_type_t			tcb_type;
418 	ddi_dma_handle_t		tcb_dma_handle;
419 	i40e_dma_buffer_t		tcb_dma;
420 } i40e_tx_control_block_t;
421 
422 /*
423  * Receive ring data (used below).
424  */
425 typedef struct i40e_rx_data {
426 	struct i40e	*rxd_i40e;
427 
428 	/*
429 	 * RX descriptor ring definitions
430 	 */
431 	i40e_dma_buffer_t rxd_desc_area;	/* DMA buffer of rx desc ring */
432 	i40e_rx_desc_t *rxd_desc_ring;		/* Rx desc ring */
433 	uint32_t rxd_desc_next;			/* Index of next rx desc */
434 
435 	/*
436 	 * RX control block list definitions
437 	 */
438 	kmutex_t		rxd_free_lock;	/* Lock to protect free data */
439 	i40e_rx_control_block_t	*rxd_rcb_area;	/* Array of control blocks */
440 	i40e_rx_control_block_t	**rxd_work_list; /* Work list of rcbs */
441 	i40e_rx_control_block_t	**rxd_free_list; /* Free list of rcbs */
442 	uint32_t		rxd_rcb_free;	/* Number of free rcbs */
443 
444 	/*
445 	 * RX software ring settings
446 	 */
447 	uint32_t	rxd_ring_size;		/* Rx descriptor ring size */
448 	uint32_t	rxd_free_list_size;	/* Rx free list size */
449 
450 	/*
451 	 * RX outstanding data. This is used to keep track of outstanding loaned
452 	 * descriptors after we've shut down receiving information. Note these
453 	 * are protected by the i40e_t`i40e_rx_pending_lock.
454 	 */
455 	uint32_t	rxd_rcb_pending;
456 	boolean_t	rxd_shutdown;
457 } i40e_rx_data_t;
458 
459 /*
460  * Structures for unicast and multicast addresses. Note that we keep the VSI id
461  * around for unicast addresses, since they may belong to different VSIs.
462  * However, since all multicast addresses belong to the default VSI, we don't
463  * duplicate that information.
464  */
465 typedef struct i40e_uaddr {
466 	uint8_t iua_mac[ETHERADDRL];
467 	int	iua_vsi;
468 } i40e_uaddr_t;
469 
470 typedef struct i40e_maddr {
471 	uint8_t ima_mac[ETHERADDRL];
472 } i40e_maddr_t;
473 
474 /*
475  * Collection of RX statistics on a given queue.
476  */
477 typedef struct i40e_rxq_stat {
478 	/*
479 	 * The i40e hardware does not maintain statistics on a per-ring basis,
480 	 * only on a per-PF and per-VSI level. As such, to satisfy the GLDv3, we
481 	 * need to maintain our own stats for packets and bytes.
482 	 */
483 	kstat_named_t	irxs_bytes;	/* Bytes in on queue */
484 	kstat_named_t	irxs_packets;	/* Packets in on queue */
485 
486 	/*
487 	 * The following set of stats cover non-checksum data path issues.
488 	 */
489 	kstat_named_t	irxs_rx_desc_error;	/* Error bit set on desc */
490 	kstat_named_t	irxs_rx_copy_nomem;	/* allocb failure for copy */
491 	kstat_named_t	irxs_rx_intr_limit;	/* Hit i40e_rx_limit_per_intr */
492 	kstat_named_t	irxs_rx_bind_norcb;	/* No replacement rcb free */
493 	kstat_named_t	irxs_rx_bind_nomp;	/* No mblk_t in bind rcb */
494 
495 	/*
496 	 * The following set of statistics covers rx checksum related activity.
497 	 * These are all primarily set in i40e_rx_hcksum. If rx checksum
498 	 * activity is disabled, then these should all be zero.
499 	 */
500 	kstat_named_t	irxs_hck_v4hdrok;	/* Valid IPv4 Header */
501 	kstat_named_t	irxs_hck_l4hdrok;	/* Valid L4 Header */
502 	kstat_named_t	irxs_hck_unknown;	/* !pinfo.known */
503 	kstat_named_t	irxs_hck_nol3l4p;	/* Missing L3L4P bit in desc */
504 	kstat_named_t	irxs_hck_iperr;		/* IPE error bit set */
505 	kstat_named_t	irxs_hck_eiperr;	/* EIPE error bit set */
506 	kstat_named_t	irxs_hck_l4err;		/* L4E error bit set */
507 	kstat_named_t	irxs_hck_v6skip;	/* IPv6 case hw fails on */
508 	kstat_named_t	irxs_hck_set;		/* Total times we set cksum */
509 	kstat_named_t	irxs_hck_miss;		/* Times with zero cksum bits */
510 } i40e_rxq_stat_t;
511 
512 /*
513  * Collection of TX Statistics on a given queue
514  */
515 typedef struct i40e_txq_stat {
516 	kstat_named_t	itxs_bytes;		/* Bytes out on queue */
517 	kstat_named_t	itxs_packets;		/* Packets out on queue */
518 	kstat_named_t	itxs_descriptors;	/* Descriptors issued */
519 	kstat_named_t	itxs_recycled;		/* Descriptors reclaimed */
520 	/*
521 	 * Various failure conditions.
522 	 */
523 	kstat_named_t	itxs_hck_meoifail;	/* ether offload failures */
524 	kstat_named_t	itxs_hck_nol2info;	/* Missing l2 info */
525 	kstat_named_t	itxs_hck_nol3info;	/* Missing l3 info */
526 	kstat_named_t	itxs_hck_nol4info;	/* Missing l4 info */
527 	kstat_named_t	itxs_hck_badl3;		/* Not IPv4/IPv6 */
528 	kstat_named_t	itxs_hck_badl4;		/* Bad L4 Paylaod */
529 
530 	kstat_named_t	itxs_err_notcb;		/* No tcb's available */
531 	kstat_named_t	itxs_err_nodescs;	/* No tcb's available */
532 	kstat_named_t	itxs_err_context;	/* Total context failures */
533 
534 	kstat_named_t	itxs_num_unblocked;	/* Number of MAC unblocks */
535 } i40e_txq_stat_t;
536 
537 /*
538  * An instance of an XL710 transmit/receive queue pair. This currently
539  * represents a combination of both a transmit and receive ring, though they
540  * should really be split apart into separate logical structures. Unfortunately,
541  * during initial work we mistakenly joined them together.
542  */
543 typedef struct i40e_trqpair {
544 	struct i40e *itrq_i40e;
545 
546 	/* Receive-side structures. */
547 	kmutex_t itrq_rx_lock;
548 	mac_ring_handle_t itrq_macrxring; /* Receive ring handle. */
549 	i40e_rx_data_t *itrq_rxdata;	/* Receive ring rx data. */
550 	uint64_t itrq_rxgen;		/* Generation number for mac/GLDv3. */
551 	uint32_t itrq_index;		/* Queue index in the PF */
552 	uint32_t itrq_rx_intrvec;	/* Receive interrupt vector. */
553 	boolean_t itrq_intr_poll;	/* True when polling */
554 
555 	/* Receive-side stats. */
556 	i40e_rxq_stat_t	itrq_rxstat;
557 	kstat_t	*itrq_rxkstat;
558 
559 	/* Transmit-side structures. */
560 	kmutex_t itrq_tx_lock;
561 	mac_ring_handle_t itrq_mactxring; /* Transmit ring handle. */
562 	uint32_t itrq_tx_intrvec;	/* Transmit interrupt vector. */
563 	boolean_t itrq_tx_blocked;	/* Does MAC think we're blocked? */
564 
565 	/*
566 	 * TX data sizing
567 	 */
568 	uint32_t		itrq_tx_ring_size;
569 	uint32_t		itrq_tx_free_list_size;
570 
571 	/*
572 	 * TX descriptor ring data
573 	 */
574 	i40e_dma_buffer_t	itrq_desc_area;	/* DMA buffer of tx desc ring */
575 	i40e_tx_desc_t		*itrq_desc_ring; /* TX Desc ring */
576 	volatile uint32_t 	*itrq_desc_wbhead; /* TX write-back index */
577 	uint32_t		itrq_desc_head;	/* Last index hw freed */
578 	uint32_t		itrq_desc_tail;	/* Index of next free desc */
579 	uint32_t		itrq_desc_free;	/* Number of free descriptors */
580 
581 	/*
582 	 * TX control block (tcb) data
583 	 */
584 	kmutex_t		itrq_tcb_lock;
585 	i40e_tx_control_block_t	*itrq_tcb_area;	/* Array of control blocks */
586 	i40e_tx_control_block_t	**itrq_tcb_work_list;	/* In use tcb */
587 	i40e_tx_control_block_t	**itrq_tcb_free_list;	/* Available tcb */
588 	uint32_t		itrq_tcb_free;	/* Count of free tcb */
589 
590 	/* Transmit-side stats. */
591 	i40e_txq_stat_t		itrq_txstat;
592 	kstat_t			*itrq_txkstat;
593 
594 } i40e_trqpair_t;
595 
596 /*
597  * VSI statistics.
598  *
599  * This mirrors the i40e_eth_stats structure but transforms it into a kstat.
600  * Note that the stock statistic structure also includes entries for tx
601  * discards. However, this is not actually implemented for the VSI (see Table
602  * 7-221), hence why we don't include the member which would always have a value
603  * of zero. This choice was made to minimize confusion to someone looking at
604  * these, as a value of zero does not necessarily equate to the fact that it's
605  * not implemented.
606  */
607 typedef struct i40e_vsi_stats {
608 	uint64_t ivs_rx_bytes;			/* gorc */
609 	uint64_t ivs_rx_unicast;		/* uprc */
610 	uint64_t ivs_rx_multicast;		/* mprc */
611 	uint64_t ivs_rx_broadcast;		/* bprc */
612 	uint64_t ivs_rx_discards;		/* rdpc */
613 	uint64_t ivs_rx_unknown_protocol;	/* rupp */
614 	uint64_t ivs_tx_bytes;			/* gotc */
615 	uint64_t ivs_tx_unicast;		/* uptc */
616 	uint64_t ivs_tx_multicast;		/* mptc */
617 	uint64_t ivs_tx_broadcast;		/* bptc */
618 	uint64_t ivs_tx_errors;			/* tepc */
619 } i40e_vsi_stats_t;
620 
621 typedef struct i40e_vsi_kstats {
622 	kstat_named_t	ivk_rx_bytes;
623 	kstat_named_t	ivk_rx_unicast;
624 	kstat_named_t	ivk_rx_multicast;
625 	kstat_named_t	ivk_rx_broadcast;
626 	kstat_named_t	ivk_rx_discards;
627 	kstat_named_t	ivk_rx_unknown_protocol;
628 	kstat_named_t	ivk_tx_bytes;
629 	kstat_named_t	ivk_tx_unicast;
630 	kstat_named_t	ivk_tx_multicast;
631 	kstat_named_t	ivk_tx_broadcast;
632 	kstat_named_t	ivk_tx_errors;
633 } i40e_vsi_kstats_t;
634 
635 /*
636  * For pf statistics, we opt not to use the standard statistics as defined by
637  * the Intel common code. This also currently combines statistics that are
638  * global across the entire device.
639  */
640 typedef struct i40e_pf_stats {
641 	uint64_t ips_rx_bytes;			/* gorc */
642 	uint64_t ips_rx_unicast;		/* uprc */
643 	uint64_t ips_rx_multicast;		/* mprc */
644 	uint64_t ips_rx_broadcast;		/* bprc */
645 	uint64_t ips_tx_bytes;			/* gotc */
646 	uint64_t ips_tx_unicast;		/* uptc */
647 	uint64_t ips_tx_multicast;		/* mptc */
648 	uint64_t ips_tx_broadcast;		/* bptc */
649 
650 	uint64_t ips_rx_size_64;		/* prc64 */
651 	uint64_t ips_rx_size_127;		/* prc127 */
652 	uint64_t ips_rx_size_255;		/* prc255 */
653 	uint64_t ips_rx_size_511;		/* prc511 */
654 	uint64_t ips_rx_size_1023;		/* prc1023 */
655 	uint64_t ips_rx_size_1522;		/* prc1522 */
656 	uint64_t ips_rx_size_9522;		/* prc9522 */
657 
658 	uint64_t ips_tx_size_64;		/* ptc64 */
659 	uint64_t ips_tx_size_127;		/* ptc127 */
660 	uint64_t ips_tx_size_255;		/* ptc255 */
661 	uint64_t ips_tx_size_511;		/* ptc511 */
662 	uint64_t ips_tx_size_1023;		/* ptc1023 */
663 	uint64_t ips_tx_size_1522;		/* ptc1522 */
664 	uint64_t ips_tx_size_9522;		/* ptc9522 */
665 
666 	uint64_t ips_link_xon_rx;		/* lxonrxc */
667 	uint64_t ips_link_xoff_rx;		/* lxoffrxc */
668 	uint64_t ips_link_xon_tx;		/* lxontxc */
669 	uint64_t ips_link_xoff_tx;		/* lxofftxc */
670 	uint64_t ips_priority_xon_rx[8];	/* pxonrxc[8] */
671 	uint64_t ips_priority_xoff_rx[8];	/* pxoffrxc[8] */
672 	uint64_t ips_priority_xon_tx[8];	/* pxontxc[8] */
673 	uint64_t ips_priority_xoff_tx[8];	/* pxofftxc[8] */
674 	uint64_t ips_priority_xon_2_xoff[8];	/* rxon2offcnt[8] */
675 
676 	uint64_t ips_crc_errors;		/* crcerrs */
677 	uint64_t ips_illegal_bytes;		/* illerrc */
678 	uint64_t ips_mac_local_faults;		/* mlfc */
679 	uint64_t ips_mac_remote_faults;		/* mrfc */
680 	uint64_t ips_rx_length_errors;		/* rlec */
681 	uint64_t ips_rx_undersize;		/* ruc */
682 	uint64_t ips_rx_fragments;		/* rfc */
683 	uint64_t ips_rx_oversize;		/* roc */
684 	uint64_t ips_rx_jabber;			/* rjc */
685 	uint64_t ips_rx_discards;		/* rdpc */
686 	uint64_t ips_rx_vm_discards;		/* ldpc */
687 	uint64_t ips_rx_short_discards;		/* mspdc */
688 	uint64_t ips_tx_dropped_link_down;	/* tdold */
689 	uint64_t ips_rx_unknown_protocol;	/* rupp */
690 	uint64_t ips_rx_err1;			/* rxerr1 */
691 	uint64_t ips_rx_err2;			/* rxerr2 */
692 } i40e_pf_stats_t;
693 
694 typedef struct i40e_pf_kstats {
695 	kstat_named_t ipk_rx_bytes;		/* gorc */
696 	kstat_named_t ipk_rx_unicast;		/* uprc */
697 	kstat_named_t ipk_rx_multicast;		/* mprc */
698 	kstat_named_t ipk_rx_broadcast;		/* bprc */
699 	kstat_named_t ipk_tx_bytes;		/* gotc */
700 	kstat_named_t ipk_tx_unicast;		/* uptc */
701 	kstat_named_t ipk_tx_multicast;		/* mptc */
702 	kstat_named_t ipk_tx_broadcast;		/* bptc */
703 
704 	kstat_named_t ipk_rx_size_64;		/* prc64 */
705 	kstat_named_t ipk_rx_size_127;		/* prc127 */
706 	kstat_named_t ipk_rx_size_255;		/* prc255 */
707 	kstat_named_t ipk_rx_size_511;		/* prc511 */
708 	kstat_named_t ipk_rx_size_1023;		/* prc1023 */
709 	kstat_named_t ipk_rx_size_1522;		/* prc1522 */
710 	kstat_named_t ipk_rx_size_9522;		/* prc9522 */
711 
712 	kstat_named_t ipk_tx_size_64;		/* ptc64 */
713 	kstat_named_t ipk_tx_size_127;		/* ptc127 */
714 	kstat_named_t ipk_tx_size_255;		/* ptc255 */
715 	kstat_named_t ipk_tx_size_511;		/* ptc511 */
716 	kstat_named_t ipk_tx_size_1023;		/* ptc1023 */
717 	kstat_named_t ipk_tx_size_1522;		/* ptc1522 */
718 	kstat_named_t ipk_tx_size_9522;		/* ptc9522 */
719 
720 	kstat_named_t ipk_link_xon_rx;		/* lxonrxc */
721 	kstat_named_t ipk_link_xoff_rx;		/* lxoffrxc */
722 	kstat_named_t ipk_link_xon_tx;		/* lxontxc */
723 	kstat_named_t ipk_link_xoff_tx;		/* lxofftxc */
724 	kstat_named_t ipk_priority_xon_rx[8];	/* pxonrxc[8] */
725 	kstat_named_t ipk_priority_xoff_rx[8];	/* pxoffrxc[8] */
726 	kstat_named_t ipk_priority_xon_tx[8];	/* pxontxc[8] */
727 	kstat_named_t ipk_priority_xoff_tx[8];	/* pxofftxc[8] */
728 	kstat_named_t ipk_priority_xon_2_xoff[8];	/* rxon2offcnt[8] */
729 
730 	kstat_named_t ipk_crc_errors;		/* crcerrs */
731 	kstat_named_t ipk_illegal_bytes;	/* illerrc */
732 	kstat_named_t ipk_mac_local_faults;	/* mlfc */
733 	kstat_named_t ipk_mac_remote_faults;	/* mrfc */
734 	kstat_named_t ipk_rx_length_errors;	/* rlec */
735 	kstat_named_t ipk_rx_undersize;		/* ruc */
736 	kstat_named_t ipk_rx_fragments;		/* rfc */
737 	kstat_named_t ipk_rx_oversize;		/* roc */
738 	kstat_named_t ipk_rx_jabber;		/* rjc */
739 	kstat_named_t ipk_rx_discards;		/* rdpc */
740 	kstat_named_t ipk_rx_vm_discards;	/* ldpc */
741 	kstat_named_t ipk_rx_short_discards;	/* mspdc */
742 	kstat_named_t ipk_tx_dropped_link_down;	/* tdold */
743 	kstat_named_t ipk_rx_unknown_protocol;	/* rupp */
744 	kstat_named_t ipk_rx_err1;		/* rxerr1 */
745 	kstat_named_t ipk_rx_err2;		/* rxerr2 */
746 } i40e_pf_kstats_t;
747 
748 /*
749  * Resources that are pooled and specific to a given i40e_t.
750  */
751 typedef struct i40e_func_rsrc {
752 	uint_t	ifr_nrx_queue;
753 	uint_t	ifr_nrx_queue_used;
754 	uint_t	ifr_ntx_queue;
755 	uint_t	ifr_trx_queue_used;
756 	uint_t	ifr_nvsis;
757 	uint_t	ifr_nvsis_used;
758 	uint_t	ifr_nmacfilt;
759 	uint_t	ifr_nmacfilt_used;
760 	uint_t	ifr_nmcastfilt;
761 	uint_t	ifr_nmcastfilt_used;
762 } i40e_func_rsrc_t;
763 
764 /*
765  * Main i40e per-instance state.
766  */
767 typedef struct i40e {
768 	list_node_t	i40e_glink;		/* Global list link */
769 	list_node_t	i40e_dlink;		/* Device list link */
770 	kmutex_t	i40e_general_lock;	/* General device lock */
771 
772 	/*
773 	 * General Data and management
774 	 */
775 	dev_info_t	*i40e_dip;
776 	int		i40e_instance;
777 	int		i40e_fm_capabilities;
778 	uint_t		i40e_state;
779 	i40e_attach_state_t i40e_attach_progress;
780 	mac_handle_t	i40e_mac_hdl;
781 	ddi_periodic_t	i40e_periodic_id;
782 
783 	/*
784 	 * Pointers to common code data structures and memory for the common
785 	 * code.
786 	 */
787 	struct i40e_hw				i40e_hw_space;
788 	struct i40e_osdep			i40e_osdep_space;
789 	struct i40e_aq_get_phy_abilities_resp	i40e_phy;
790 	void 					*i40e_aqbuf;
791 
792 	/*
793 	 * Device state, switch information, and resources.
794 	 */
795 	int			i40e_vsi_id;
796 	uint16_t		i40e_vsi_num;
797 	struct i40e_device	*i40e_device;
798 	i40e_func_rsrc_t	i40e_resources;
799 	uint16_t		i40e_switch_rsrc_alloc;
800 	uint16_t		i40e_switch_rsrc_actual;
801 	i40e_switch_rsrc_t	*i40e_switch_rsrcs;
802 	i40e_uaddr_t		*i40e_uaddrs;
803 	i40e_maddr_t		*i40e_maddrs;
804 	int			i40e_mcast_promisc_count;
805 	boolean_t		i40e_promisc_on;
806 	link_state_t		i40e_link_state;
807 	uint32_t		i40e_link_speed;	/* In Mbps */
808 	link_duplex_t		i40e_link_duplex;
809 	uint_t			i40e_sdu;
810 	uint_t			i40e_frame_max;
811 
812 	/*
813 	 * Transmit and receive information, tunables, and MAC info.
814 	 */
815 	i40e_trqpair_t	*i40e_trqpairs;
816 	boolean_t 	i40e_mr_enable;
817 	int		i40e_num_trqpairs;
818 	uint_t		i40e_other_itr;
819 
820 	int		i40e_num_rx_groups;
821 	int		i40e_num_rx_descs;
822 	mac_group_handle_t i40e_rx_group_handle;
823 	uint32_t	i40e_rx_ring_size;
824 	uint32_t	i40e_rx_buf_size;
825 	boolean_t	i40e_rx_hcksum_enable;
826 	uint32_t	i40e_rx_dma_min;
827 	uint32_t	i40e_rx_limit_per_intr;
828 	uint_t		i40e_rx_itr;
829 
830 	int		i40e_num_tx_descs;
831 	uint32_t	i40e_tx_ring_size;
832 	uint32_t	i40e_tx_buf_size;
833 	uint32_t	i40e_tx_block_thresh;
834 	boolean_t	i40e_tx_hcksum_enable;
835 	uint32_t	i40e_tx_dma_min;
836 	uint_t		i40e_tx_itr;
837 
838 	/*
839 	 * Interrupt state
840 	 */
841 	uint_t		i40e_intr_pri;
842 	uint_t		i40e_intr_force;
843 	uint_t		i40e_intr_type;
844 	int		i40e_intr_cap;
845 	uint32_t	i40e_intr_count;
846 	uint32_t	i40e_intr_count_max;
847 	uint32_t	i40e_intr_count_min;
848 	size_t		i40e_intr_size;
849 	ddi_intr_handle_t *i40e_intr_handles;
850 	ddi_cb_handle_t	i40e_callback_handle;
851 
852 	/*
853 	 * DMA attributes. See i40e_transceiver.c for why we have copies of them
854 	 * in the i40e_t.
855 	 */
856 	ddi_dma_attr_t		i40e_static_dma_attr;
857 	ddi_dma_attr_t		i40e_txbind_dma_attr;
858 	ddi_device_acc_attr_t	i40e_desc_acc_attr;
859 	ddi_device_acc_attr_t	i40e_buf_acc_attr;
860 
861 	/*
862 	 * The following two fields are used to protect and keep track of
863 	 * outstanding, loaned buffers to MAC. If we have these, we can't
864 	 * detach as we have active DMA memory outstanding.
865 	 */
866 	kmutex_t	i40e_rx_pending_lock;
867 	kcondvar_t	i40e_rx_pending_cv;
868 	uint32_t	i40e_rx_pending;
869 
870 	/*
871 	 * PF statistics and VSI statistics.
872 	 */
873 	kmutex_t		i40e_stat_lock;
874 	kstat_t			*i40e_pf_kstat;
875 	kstat_t			*i40e_vsi_kstat;
876 	i40e_pf_stats_t		i40e_pf_stat;
877 	i40e_vsi_stats_t	i40e_vsi_stat;
878 	uint16_t		i40e_vsi_stat_id;
879 
880 	/*
881 	 * Misc. stats and counters that should maybe one day be kstats.
882 	 */
883 	uint64_t	i40e_s_link_status_errs;
884 	uint32_t	i40e_s_link_status_lasterr;
885 } i40e_t;
886 
887 /*
888  * The i40e_device represents a PCI device which encapsulates multiple physical
889  * functions which are represented as an i40e_t. This is used to track the use
890  * of pooled resources throughout all of the various devices.
891  */
892 typedef struct i40e_device {
893 	list_node_t	id_link;
894 	dev_info_t	*id_parent;
895 	uint_t		id_pci_bus;
896 	uint_t		id_pci_device;
897 	uint_t		id_nfuncs;	/* Total number of functions */
898 	uint_t		id_nreg;	/* Total number present */
899 	list_t		id_i40e_list;	/* List of i40e_t's registered */
900 	i40e_switch_rsrc_t	*id_rsrcs; /* Switch resources for this PF */
901 	uint_t		id_rsrcs_alloc;	/* Total allocated resources */
902 	uint_t		id_rsrcs_act;	/* Actual number of resources */
903 } i40e_device_t;
904 
905 /* Values for the interrupt forcing on the NIC. */
906 #define	I40E_INTR_NONE			0
907 #define	I40E_INTR_MSIX			1
908 #define	I40E_INTR_MSI			2
909 #define	I40E_INTR_LEGACY		3
910 
911 /* Hint that we don't want to do any polling... */
912 #define	I40E_POLL_NULL			-1
913 
914 /*
915  * Logging functions.
916  */
917 /*PRINTFLIKE2*/
918 extern void i40e_error(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
919 /*PRINTFLIKE2*/
920 extern void i40e_notice(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
921 /*PRINTFLIKE2*/
922 extern void i40e_log(i40e_t *, const char *, ...) __KPRINTFLIKE(2);
923 
924 /*
925  * General link handling functions.
926  */
927 extern void i40e_link_check(i40e_t *);
928 extern void i40e_update_mtu(i40e_t *);
929 
930 /*
931  * FMA functions.
932  */
933 extern int i40e_check_acc_handle(ddi_acc_handle_t);
934 extern int i40e_check_dma_handle(ddi_dma_handle_t);
935 extern void i40e_fm_ereport(i40e_t *, char *);
936 
937 /*
938  * Interrupt handlers and interrupt handler setup.
939  */
940 extern void i40e_intr_chip_init(i40e_t *);
941 extern void i40e_intr_chip_fini(i40e_t *);
942 extern uint_t i40e_intr_msix(void *, void *);
943 extern uint_t i40e_intr_msi(void *, void *);
944 extern uint_t i40e_intr_legacy(void *, void *);
945 extern void i40e_intr_io_enable_all(i40e_t *);
946 extern void i40e_intr_io_disable_all(i40e_t *);
947 extern void i40e_intr_io_clear_cause(i40e_t *);
948 extern void i40e_intr_rx_queue_disable(i40e_trqpair_t *);
949 extern void i40e_intr_rx_queue_enable(i40e_trqpair_t *);
950 extern void i40e_intr_set_itr(i40e_t *, i40e_itr_index_t, uint_t);
951 
952 /*
953  * Receive-side functions
954  */
955 extern mblk_t *i40e_ring_rx(i40e_trqpair_t *, int);
956 extern mblk_t *i40e_ring_rx_poll(void *, int);
957 extern void i40e_rx_recycle(caddr_t);
958 
959 /*
960  * Transmit-side functions
961  */
962 mblk_t *i40e_ring_tx(void *, mblk_t *);
963 extern void i40e_tx_recycle_ring(i40e_trqpair_t *);
964 extern void i40e_tx_cleanup_ring(i40e_trqpair_t *);
965 
966 /*
967  * Statistics functions.
968  */
969 extern boolean_t i40e_stats_init(i40e_t *);
970 extern void i40e_stats_fini(i40e_t *);
971 extern boolean_t i40e_stat_vsi_init(i40e_t *);
972 extern void i40e_stat_vsi_fini(i40e_t *);
973 extern boolean_t i40e_stats_trqpair_init(i40e_trqpair_t *);
974 extern void i40e_stats_trqpair_fini(i40e_trqpair_t *);
975 extern int i40e_m_stat(void *, uint_t, uint64_t *);
976 extern int i40e_rx_ring_stat(mac_ring_driver_t, uint_t, uint64_t *);
977 extern int i40e_tx_ring_stat(mac_ring_driver_t, uint_t, uint64_t *);
978 
979 /*
980  * MAC/GLDv3 functions, and functions called by MAC/GLDv3 support code.
981  */
982 extern boolean_t i40e_register_mac(i40e_t *);
983 extern boolean_t i40e_start(i40e_t *, boolean_t);
984 extern void i40e_stop(i40e_t *, boolean_t);
985 
986 /*
987  * DMA & buffer functions and attributes
988  */
989 extern void i40e_init_dma_attrs(i40e_t *, boolean_t);
990 extern boolean_t i40e_alloc_ring_mem(i40e_t *);
991 extern void i40e_free_ring_mem(i40e_t *, boolean_t);
992 
993 #ifdef __cplusplus
994 }
995 #endif
996 
997 #endif /* _I40E_SW_H */
998