xref: /linux/drivers/infiniband/hw/hfi1/hfi.h (revision d4b996f9ef1fe83d9ce9ad5c1ca0bd8231638ce5)
1 #ifndef _HFI1_KERNEL_H
2 #define _HFI1_KERNEL_H
3 /*
4  * Copyright(c) 2020 Cornelis Networks, Inc.
5  * Copyright(c) 2015-2020 Intel Corporation.
6  *
7  * This file is provided under a dual BSD/GPLv2 license.  When using or
8  * redistributing this file, you may do so under either license.
9  *
10  * GPL LICENSE SUMMARY
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of version 2 of the GNU General Public License as
14  * published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful, but
17  * WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * BSD LICENSE
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  *
27  *  - Redistributions of source code must retain the above copyright
28  *    notice, this list of conditions and the following disclaimer.
29  *  - Redistributions in binary form must reproduce the above copyright
30  *    notice, this list of conditions and the following disclaimer in
31  *    the documentation and/or other materials provided with the
32  *    distribution.
33  *  - Neither the name of Intel Corporation nor the names of its
34  *    contributors may be used to endorse or promote products derived
35  *    from this software without specific prior written permission.
36  *
37  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
38  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
39  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
40  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
41  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
42  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
43  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
44  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
45  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
46  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
47  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
48  *
49  */
50 
51 #include <linux/interrupt.h>
52 #include <linux/pci.h>
53 #include <linux/dma-mapping.h>
54 #include <linux/mutex.h>
55 #include <linux/list.h>
56 #include <linux/scatterlist.h>
57 #include <linux/slab.h>
58 #include <linux/io.h>
59 #include <linux/fs.h>
60 #include <linux/completion.h>
61 #include <linux/kref.h>
62 #include <linux/sched.h>
63 #include <linux/cdev.h>
64 #include <linux/delay.h>
65 #include <linux/kthread.h>
66 #include <linux/i2c.h>
67 #include <linux/i2c-algo-bit.h>
68 #include <linux/xarray.h>
69 #include <rdma/ib_hdrs.h>
70 #include <rdma/opa_addr.h>
71 #include <linux/rhashtable.h>
72 #include <rdma/rdma_vt.h>
73 
74 #include "chip_registers.h"
75 #include "common.h"
76 #include "opfn.h"
77 #include "verbs.h"
78 #include "pio.h"
79 #include "chip.h"
80 #include "mad.h"
81 #include "qsfp.h"
82 #include "platform.h"
83 #include "affinity.h"
84 #include "msix.h"
85 
86 /* bumped 1 from s/w major version of TrueScale */
87 #define HFI1_CHIP_VERS_MAJ 3U
88 
89 /* don't care about this except printing */
90 #define HFI1_CHIP_VERS_MIN 0U
91 
92 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
93 #define HFI1_OUI 0x001175
94 #define HFI1_OUI_LSB 40
95 
96 #define DROP_PACKET_OFF		0
97 #define DROP_PACKET_ON		1
98 
99 #define NEIGHBOR_TYPE_HFI		0
100 #define NEIGHBOR_TYPE_SWITCH	1
101 
102 #define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5
103 
104 extern unsigned long hfi1_cap_mask;
105 #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap)
106 #define HFI1_CAP_UGET_MASK(mask, cap) \
107 	(((mask) >> HFI1_CAP_USER_SHIFT) & HFI1_CAP_##cap)
108 #define HFI1_CAP_KGET(cap) (HFI1_CAP_KGET_MASK(hfi1_cap_mask, cap))
109 #define HFI1_CAP_UGET(cap) (HFI1_CAP_UGET_MASK(hfi1_cap_mask, cap))
110 #define HFI1_CAP_IS_KSET(cap) (!!HFI1_CAP_KGET(cap))
111 #define HFI1_CAP_IS_USET(cap) (!!HFI1_CAP_UGET(cap))
112 #define HFI1_MISC_GET() ((hfi1_cap_mask >> HFI1_CAP_MISC_SHIFT) & \
113 			HFI1_CAP_MISC_MASK)
114 /* Offline Disabled Reason is 4-bits */
115 #define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
116 
117 /*
118  * Control context is always 0 and handles the error packets.
119  * It also handles the VL15 and multicast packets.
120  */
121 #define HFI1_CTRL_CTXT    0
122 
123 /*
124  * Driver context will store software counters for each of the events
125  * associated with these status registers
126  */
127 #define NUM_CCE_ERR_STATUS_COUNTERS 41
128 #define NUM_RCV_ERR_STATUS_COUNTERS 64
129 #define NUM_MISC_ERR_STATUS_COUNTERS 13
130 #define NUM_SEND_PIO_ERR_STATUS_COUNTERS 36
131 #define NUM_SEND_DMA_ERR_STATUS_COUNTERS 4
132 #define NUM_SEND_EGRESS_ERR_STATUS_COUNTERS 64
133 #define NUM_SEND_ERR_STATUS_COUNTERS 3
134 #define NUM_SEND_CTXT_ERR_STATUS_COUNTERS 5
135 #define NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS 24
136 
137 /*
138  * per driver stats, either not device nor port-specific, or
139  * summed over all of the devices and ports.
140  * They are described by name via ipathfs filesystem, so layout
141  * and number of elements can change without breaking compatibility.
142  * If members are added or deleted hfi1_statnames[] in debugfs.c must
143  * change to match.
144  */
145 struct hfi1_ib_stats {
146 	__u64 sps_ints; /* number of interrupts handled */
147 	__u64 sps_errints; /* number of error interrupts */
148 	__u64 sps_txerrs; /* tx-related packet errors */
149 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
150 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
151 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
152 	__u64 sps_ctxts; /* number of contexts currently open */
153 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
154 	__u64 sps_buffull;
155 	__u64 sps_hdrfull;
156 };
157 
158 extern struct hfi1_ib_stats hfi1_stats;
159 extern const struct pci_error_handlers hfi1_pci_err_handler;
160 
161 extern int num_driver_cntrs;
162 
163 /*
164  * First-cut criterion for "device is active" is
165  * two thousand dwords combined Tx, Rx traffic per
166  * 5-second interval. SMA packets are 64 dwords,
167  * and occur "a few per second", presumably each way.
168  */
169 #define HFI1_TRAFFIC_ACTIVE_THRESHOLD (2000)
170 
171 /*
172  * Below contains all data related to a single context (formerly called port).
173  */
174 
175 struct hfi1_opcode_stats_perctx;
176 
177 struct ctxt_eager_bufs {
178 	struct eager_buffer {
179 		void *addr;
180 		dma_addr_t dma;
181 		ssize_t len;
182 	} *buffers;
183 	struct {
184 		void *addr;
185 		dma_addr_t dma;
186 	} *rcvtids;
187 	u32 size;                /* total size of eager buffers */
188 	u32 rcvtid_size;         /* size of each eager rcv tid */
189 	u16 count;               /* size of buffers array */
190 	u16 numbufs;             /* number of buffers allocated */
191 	u16 alloced;             /* number of rcvarray entries used */
192 	u16 threshold;           /* head update threshold */
193 };
194 
195 struct exp_tid_set {
196 	struct list_head list;
197 	u32 count;
198 };
199 
200 struct hfi1_ctxtdata;
201 typedef int (*intr_handler)(struct hfi1_ctxtdata *rcd, int data);
202 typedef void (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
203 
204 struct tid_queue {
205 	struct list_head queue_head;
206 			/* queue head for QP TID resource waiters */
207 	u32 enqueue;	/* count of tid enqueues */
208 	u32 dequeue;	/* count of tid dequeues */
209 };
210 
211 struct hfi1_ctxtdata {
212 	/* rcvhdrq base, needs mmap before useful */
213 	void *rcvhdrq;
214 	/* kernel virtual address where hdrqtail is updated */
215 	volatile __le64 *rcvhdrtail_kvaddr;
216 	/* so functions that need physical port can get it easily */
217 	struct hfi1_pportdata *ppd;
218 	/* so file ops can get at unit */
219 	struct hfi1_devdata *dd;
220 	/* this receive context's assigned PIO ACK send context */
221 	struct send_context *sc;
222 	/* per context recv functions */
223 	const rhf_rcv_function_ptr *rhf_rcv_function_map;
224 	/*
225 	 * The interrupt handler for a particular receive context can vary
226 	 * throughout it's lifetime. This is not a lock protected data member so
227 	 * it must be updated atomically and the prev and new value must always
228 	 * be valid. Worst case is we process an extra interrupt and up to 64
229 	 * packets with the wrong interrupt handler.
230 	 */
231 	intr_handler do_interrupt;
232 	/** fast handler after autoactive */
233 	intr_handler fast_handler;
234 	/** slow handler */
235 	intr_handler slow_handler;
236 	/* napi pointer assiociated with netdev */
237 	struct napi_struct *napi;
238 	/* verbs rx_stats per rcd */
239 	struct hfi1_opcode_stats_perctx *opstats;
240 	/* clear interrupt mask */
241 	u64 imask;
242 	/* ctxt rcvhdrq head offset */
243 	u32 head;
244 	/* number of rcvhdrq entries */
245 	u16 rcvhdrq_cnt;
246 	u8 ireg;	/* clear interrupt register */
247 	/* receive packet sequence counter */
248 	u8 seq_cnt;
249 	/* size of each of the rcvhdrq entries */
250 	u8 rcvhdrqentsize;
251 	/* offset of RHF within receive header entry */
252 	u8 rhf_offset;
253 	/* dynamic receive available interrupt timeout */
254 	u8 rcvavail_timeout;
255 	/* Indicates that this is vnic context */
256 	bool is_vnic;
257 	/* vnic queue index this context is mapped to */
258 	u8 vnic_q_idx;
259 	/* Is ASPM interrupt supported for this context */
260 	bool aspm_intr_supported;
261 	/* ASPM state (enabled/disabled) for this context */
262 	bool aspm_enabled;
263 	/* Is ASPM processing enabled for this context (in intr context) */
264 	bool aspm_intr_enable;
265 	struct ctxt_eager_bufs egrbufs;
266 	/* QPs waiting for context processing */
267 	struct list_head qp_wait_list;
268 	/* tid allocation lists */
269 	struct exp_tid_set tid_group_list;
270 	struct exp_tid_set tid_used_list;
271 	struct exp_tid_set tid_full_list;
272 
273 	/* Timer for re-enabling ASPM if interrupt activity quiets down */
274 	struct timer_list aspm_timer;
275 	/* per-context configuration flags */
276 	unsigned long flags;
277 	/* array of tid_groups */
278 	struct tid_group  *groups;
279 	/* mmap of hdrq, must fit in 44 bits */
280 	dma_addr_t rcvhdrq_dma;
281 	dma_addr_t rcvhdrqtailaddr_dma;
282 	/* Last interrupt timestamp */
283 	ktime_t aspm_ts_last_intr;
284 	/* Last timestamp at which we scheduled a timer for this context */
285 	ktime_t aspm_ts_timer_sched;
286 	/* Lock to serialize between intr, timer intr and user threads */
287 	spinlock_t aspm_lock;
288 	/* Reference count the base context usage */
289 	struct kref kref;
290 	/* numa node of this context */
291 	int numa_id;
292 	/* associated msix interrupt. */
293 	s16 msix_intr;
294 	/* job key */
295 	u16 jkey;
296 	/* number of RcvArray groups for this context. */
297 	u16 rcv_array_groups;
298 	/* index of first eager TID entry. */
299 	u16 eager_base;
300 	/* number of expected TID entries */
301 	u16 expected_count;
302 	/* index of first expected TID entry. */
303 	u16 expected_base;
304 	/* Device context index */
305 	u8 ctxt;
306 
307 	/* PSM Specific fields */
308 	/* lock protecting all Expected TID data */
309 	struct mutex exp_mutex;
310 	/* lock protecting all Expected TID data of kernel contexts */
311 	spinlock_t exp_lock;
312 	/* Queue for QP's waiting for HW TID flows */
313 	struct tid_queue flow_queue;
314 	/* Queue for QP's waiting for HW receive array entries */
315 	struct tid_queue rarr_queue;
316 	/* when waiting for rcv or pioavail */
317 	wait_queue_head_t wait;
318 	/* uuid from PSM */
319 	u8 uuid[16];
320 	/* same size as task_struct .comm[], command that opened context */
321 	char comm[TASK_COMM_LEN];
322 	/* Bitmask of in use context(s) */
323 	DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
324 	/* per-context event flags for fileops/intr communication */
325 	unsigned long event_flags;
326 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
327 	void *subctxt_uregbase;
328 	/* An array of pages for the eager receive buffers * N */
329 	void *subctxt_rcvegrbuf;
330 	/* An array of pages for the eager header queue entries * N */
331 	void *subctxt_rcvhdr_base;
332 	/* total number of polled urgent packets */
333 	u32 urgent;
334 	/* saved total number of polled urgent packets for poll edge trigger */
335 	u32 urgent_poll;
336 	/* Type of packets or conditions we want to poll for */
337 	u16 poll_type;
338 	/* non-zero if ctxt is being shared. */
339 	u16 subctxt_id;
340 	/* The version of the library which opened this ctxt */
341 	u32 userversion;
342 	/*
343 	 * non-zero if ctxt can be shared, and defines the maximum number of
344 	 * sub-contexts for this device context.
345 	 */
346 	u8 subctxt_cnt;
347 
348 	/* Bit mask to track free TID RDMA HW flows */
349 	unsigned long flow_mask;
350 	struct tid_flow_state flows[RXE_NUM_TID_FLOWS];
351 };
352 
353 /**
354  * rcvhdrq_size - return total size in bytes for header queue
355  * @rcd: the receive context
356  *
357  * rcvhdrqentsize is in DWs, so we have to convert to bytes
358  *
359  */
360 static inline u32 rcvhdrq_size(struct hfi1_ctxtdata *rcd)
361 {
362 	return PAGE_ALIGN(rcd->rcvhdrq_cnt *
363 			  rcd->rcvhdrqentsize * sizeof(u32));
364 }
365 
366 /*
367  * Represents a single packet at a high level. Put commonly computed things in
368  * here so we do not have to keep doing them over and over. The rule of thumb is
369  * if something is used one time to derive some value, store that something in
370  * here. If it is used multiple times, then store the result of that derivation
371  * in here.
372  */
373 struct hfi1_packet {
374 	void *ebuf;
375 	void *hdr;
376 	void *payload;
377 	struct hfi1_ctxtdata *rcd;
378 	__le32 *rhf_addr;
379 	struct rvt_qp *qp;
380 	struct ib_other_headers *ohdr;
381 	struct ib_grh *grh;
382 	struct opa_16b_mgmt *mgmt;
383 	u64 rhf;
384 	u32 maxcnt;
385 	u32 rhqoff;
386 	u32 dlid;
387 	u32 slid;
388 	int numpkt;
389 	u16 tlen;
390 	s16 etail;
391 	u16 pkey;
392 	u8 hlen;
393 	u8 rsize;
394 	u8 updegr;
395 	u8 etype;
396 	u8 extra_byte;
397 	u8 pad;
398 	u8 sc;
399 	u8 sl;
400 	u8 opcode;
401 	bool migrated;
402 };
403 
404 /* Packet types */
405 #define HFI1_PKT_TYPE_9B  0
406 #define HFI1_PKT_TYPE_16B 1
407 
408 /*
409  * OPA 16B Header
410  */
411 #define OPA_16B_L4_MASK		0xFFull
412 #define OPA_16B_SC_MASK		0x1F00000ull
413 #define OPA_16B_SC_SHIFT	20
414 #define OPA_16B_LID_MASK	0xFFFFFull
415 #define OPA_16B_DLID_MASK	0xF000ull
416 #define OPA_16B_DLID_SHIFT	20
417 #define OPA_16B_DLID_HIGH_SHIFT	12
418 #define OPA_16B_SLID_MASK	0xF00ull
419 #define OPA_16B_SLID_SHIFT	20
420 #define OPA_16B_SLID_HIGH_SHIFT	8
421 #define OPA_16B_BECN_MASK       0x80000000ull
422 #define OPA_16B_BECN_SHIFT      31
423 #define OPA_16B_FECN_MASK       0x10000000ull
424 #define OPA_16B_FECN_SHIFT      28
425 #define OPA_16B_L2_MASK		0x60000000ull
426 #define OPA_16B_L2_SHIFT	29
427 #define OPA_16B_PKEY_MASK	0xFFFF0000ull
428 #define OPA_16B_PKEY_SHIFT	16
429 #define OPA_16B_LEN_MASK	0x7FF00000ull
430 #define OPA_16B_LEN_SHIFT	20
431 #define OPA_16B_RC_MASK		0xE000000ull
432 #define OPA_16B_RC_SHIFT	25
433 #define OPA_16B_AGE_MASK	0xFF0000ull
434 #define OPA_16B_AGE_SHIFT	16
435 #define OPA_16B_ENTROPY_MASK	0xFFFFull
436 
437 /*
438  * OPA 16B L2/L4 Encodings
439  */
440 #define OPA_16B_L4_9B		0x00
441 #define OPA_16B_L2_TYPE		0x02
442 #define OPA_16B_L4_FM		0x08
443 #define OPA_16B_L4_IB_LOCAL	0x09
444 #define OPA_16B_L4_IB_GLOBAL	0x0A
445 #define OPA_16B_L4_ETHR		OPA_VNIC_L4_ETHR
446 
447 /*
448  * OPA 16B Management
449  */
450 #define OPA_16B_L4_FM_PAD	3  /* fixed 3B pad */
451 #define OPA_16B_L4_FM_HLEN	24 /* 16B(16) + L4_FM(8) */
452 
453 static inline u8 hfi1_16B_get_l4(struct hfi1_16b_header *hdr)
454 {
455 	return (u8)(hdr->lrh[2] & OPA_16B_L4_MASK);
456 }
457 
458 static inline u8 hfi1_16B_get_sc(struct hfi1_16b_header *hdr)
459 {
460 	return (u8)((hdr->lrh[1] & OPA_16B_SC_MASK) >> OPA_16B_SC_SHIFT);
461 }
462 
463 static inline u32 hfi1_16B_get_dlid(struct hfi1_16b_header *hdr)
464 {
465 	return (u32)((hdr->lrh[1] & OPA_16B_LID_MASK) |
466 		     (((hdr->lrh[2] & OPA_16B_DLID_MASK) >>
467 		     OPA_16B_DLID_HIGH_SHIFT) << OPA_16B_DLID_SHIFT));
468 }
469 
470 static inline u32 hfi1_16B_get_slid(struct hfi1_16b_header *hdr)
471 {
472 	return (u32)((hdr->lrh[0] & OPA_16B_LID_MASK) |
473 		     (((hdr->lrh[2] & OPA_16B_SLID_MASK) >>
474 		     OPA_16B_SLID_HIGH_SHIFT) << OPA_16B_SLID_SHIFT));
475 }
476 
477 static inline u8 hfi1_16B_get_becn(struct hfi1_16b_header *hdr)
478 {
479 	return (u8)((hdr->lrh[0] & OPA_16B_BECN_MASK) >> OPA_16B_BECN_SHIFT);
480 }
481 
482 static inline u8 hfi1_16B_get_fecn(struct hfi1_16b_header *hdr)
483 {
484 	return (u8)((hdr->lrh[1] & OPA_16B_FECN_MASK) >> OPA_16B_FECN_SHIFT);
485 }
486 
487 static inline u8 hfi1_16B_get_l2(struct hfi1_16b_header *hdr)
488 {
489 	return (u8)((hdr->lrh[1] & OPA_16B_L2_MASK) >> OPA_16B_L2_SHIFT);
490 }
491 
492 static inline u16 hfi1_16B_get_pkey(struct hfi1_16b_header *hdr)
493 {
494 	return (u16)((hdr->lrh[2] & OPA_16B_PKEY_MASK) >> OPA_16B_PKEY_SHIFT);
495 }
496 
497 static inline u8 hfi1_16B_get_rc(struct hfi1_16b_header *hdr)
498 {
499 	return (u8)((hdr->lrh[1] & OPA_16B_RC_MASK) >> OPA_16B_RC_SHIFT);
500 }
501 
502 static inline u8 hfi1_16B_get_age(struct hfi1_16b_header *hdr)
503 {
504 	return (u8)((hdr->lrh[3] & OPA_16B_AGE_MASK) >> OPA_16B_AGE_SHIFT);
505 }
506 
507 static inline u16 hfi1_16B_get_len(struct hfi1_16b_header *hdr)
508 {
509 	return (u16)((hdr->lrh[0] & OPA_16B_LEN_MASK) >> OPA_16B_LEN_SHIFT);
510 }
511 
512 static inline u16 hfi1_16B_get_entropy(struct hfi1_16b_header *hdr)
513 {
514 	return (u16)(hdr->lrh[3] & OPA_16B_ENTROPY_MASK);
515 }
516 
517 #define OPA_16B_MAKE_QW(low_dw, high_dw) (((u64)(high_dw) << 32) | (low_dw))
518 
519 /*
520  * BTH
521  */
522 #define OPA_16B_BTH_PAD_MASK	7
523 static inline u8 hfi1_16B_bth_get_pad(struct ib_other_headers *ohdr)
524 {
525 	return (u8)((be32_to_cpu(ohdr->bth[0]) >> IB_BTH_PAD_SHIFT) &
526 		   OPA_16B_BTH_PAD_MASK);
527 }
528 
529 /*
530  * 16B Management
531  */
532 #define OPA_16B_MGMT_QPN_MASK	0xFFFFFF
533 static inline u32 hfi1_16B_get_dest_qpn(struct opa_16b_mgmt *mgmt)
534 {
535 	return be32_to_cpu(mgmt->dest_qpn) & OPA_16B_MGMT_QPN_MASK;
536 }
537 
538 static inline u32 hfi1_16B_get_src_qpn(struct opa_16b_mgmt *mgmt)
539 {
540 	return be32_to_cpu(mgmt->src_qpn) & OPA_16B_MGMT_QPN_MASK;
541 }
542 
543 static inline void hfi1_16B_set_qpn(struct opa_16b_mgmt *mgmt,
544 				    u32 dest_qp, u32 src_qp)
545 {
546 	mgmt->dest_qpn = cpu_to_be32(dest_qp & OPA_16B_MGMT_QPN_MASK);
547 	mgmt->src_qpn = cpu_to_be32(src_qp & OPA_16B_MGMT_QPN_MASK);
548 }
549 
550 /**
551  * hfi1_get_rc_ohdr - get extended header
552  * @opah - the opaheader
553  */
554 static inline struct ib_other_headers *
555 hfi1_get_rc_ohdr(struct hfi1_opa_header *opah)
556 {
557 	struct ib_other_headers *ohdr;
558 	struct ib_header *hdr = NULL;
559 	struct hfi1_16b_header *hdr_16b = NULL;
560 
561 	/* Find out where the BTH is */
562 	if (opah->hdr_type == HFI1_PKT_TYPE_9B) {
563 		hdr = &opah->ibh;
564 		if (ib_get_lnh(hdr) == HFI1_LRH_BTH)
565 			ohdr = &hdr->u.oth;
566 		else
567 			ohdr = &hdr->u.l.oth;
568 	} else {
569 		u8 l4;
570 
571 		hdr_16b = &opah->opah;
572 		l4  = hfi1_16B_get_l4(hdr_16b);
573 		if (l4 == OPA_16B_L4_IB_LOCAL)
574 			ohdr = &hdr_16b->u.oth;
575 		else
576 			ohdr = &hdr_16b->u.l.oth;
577 	}
578 	return ohdr;
579 }
580 
581 struct rvt_sge_state;
582 
583 /*
584  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
585  * Mostly for MADs that set or query link parameters, also ipath
586  * config interfaces
587  */
588 #define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
589 #define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
590 #define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
591 #define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
592 #define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
593 #define HFI1_IB_CFG_SPD 5 /* current Link spd */
594 #define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
595 #define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
596 #define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
597 #define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
598 #define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
599 #define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
600 #define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
601 #define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
602 #define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
603 #define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
604 #define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
605 #define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
606 #define HFI1_IB_CFG_VL_HIGH_LIMIT 19
607 #define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
608 #define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
609 
610 /*
611  * HFI or Host Link States
612  *
613  * These describe the states the driver thinks the logical and physical
614  * states are in.  Used as an argument to set_link_state().  Implemented
615  * as bits for easy multi-state checking.  The actual state can only be
616  * one.
617  */
618 #define __HLS_UP_INIT_BP	0
619 #define __HLS_UP_ARMED_BP	1
620 #define __HLS_UP_ACTIVE_BP	2
621 #define __HLS_DN_DOWNDEF_BP	3	/* link down default */
622 #define __HLS_DN_POLL_BP	4
623 #define __HLS_DN_DISABLE_BP	5
624 #define __HLS_DN_OFFLINE_BP	6
625 #define __HLS_VERIFY_CAP_BP	7
626 #define __HLS_GOING_UP_BP	8
627 #define __HLS_GOING_OFFLINE_BP  9
628 #define __HLS_LINK_COOLDOWN_BP 10
629 
630 #define HLS_UP_INIT	  BIT(__HLS_UP_INIT_BP)
631 #define HLS_UP_ARMED	  BIT(__HLS_UP_ARMED_BP)
632 #define HLS_UP_ACTIVE	  BIT(__HLS_UP_ACTIVE_BP)
633 #define HLS_DN_DOWNDEF	  BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
634 #define HLS_DN_POLL	  BIT(__HLS_DN_POLL_BP)
635 #define HLS_DN_DISABLE	  BIT(__HLS_DN_DISABLE_BP)
636 #define HLS_DN_OFFLINE	  BIT(__HLS_DN_OFFLINE_BP)
637 #define HLS_VERIFY_CAP	  BIT(__HLS_VERIFY_CAP_BP)
638 #define HLS_GOING_UP	  BIT(__HLS_GOING_UP_BP)
639 #define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
640 #define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
641 
642 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
643 #define HLS_DOWN ~(HLS_UP)
644 
645 #define HLS_DEFAULT HLS_DN_POLL
646 
647 /* use this MTU size if none other is given */
648 #define HFI1_DEFAULT_ACTIVE_MTU 10240
649 /* use this MTU size as the default maximum */
650 #define HFI1_DEFAULT_MAX_MTU 10240
651 /* default partition key */
652 #define DEFAULT_PKEY 0xffff
653 
654 /*
655  * Possible fabric manager config parameters for fm_{get,set}_table()
656  */
657 #define FM_TBL_VL_HIGH_ARB		1 /* Get/set VL high prio weights */
658 #define FM_TBL_VL_LOW_ARB		2 /* Get/set VL low prio weights */
659 #define FM_TBL_BUFFER_CONTROL		3 /* Get/set Buffer Control */
660 #define FM_TBL_SC2VLNT			4 /* Get/set SC->VLnt */
661 #define FM_TBL_VL_PREEMPT_ELEMS		5 /* Get (no set) VL preempt elems */
662 #define FM_TBL_VL_PREEMPT_MATRIX	6 /* Get (no set) VL preempt matrix */
663 
664 /*
665  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
666  * these are bits so they can be combined, e.g.
667  * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
668  */
669 #define HFI1_RCVCTRL_TAILUPD_ENB 0x01
670 #define HFI1_RCVCTRL_TAILUPD_DIS 0x02
671 #define HFI1_RCVCTRL_CTXT_ENB 0x04
672 #define HFI1_RCVCTRL_CTXT_DIS 0x08
673 #define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
674 #define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
675 #define HFI1_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
676 #define HFI1_RCVCTRL_PKEY_DIS 0x80
677 #define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
678 #define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
679 #define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
680 #define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
681 #define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
682 #define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
683 #define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
684 #define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
685 #define HFI1_RCVCTRL_URGENT_ENB 0x40000
686 #define HFI1_RCVCTRL_URGENT_DIS 0x80000
687 
688 /* partition enforcement flags */
689 #define HFI1_PART_ENFORCE_IN	0x1
690 #define HFI1_PART_ENFORCE_OUT	0x2
691 
692 /* how often we check for synthetic counter wrap around */
693 #define SYNTH_CNT_TIME 3
694 
695 /* Counter flags */
696 #define CNTR_NORMAL		0x0 /* Normal counters, just read register */
697 #define CNTR_SYNTH		0x1 /* Synthetic counters, saturate at all 1s */
698 #define CNTR_DISABLED		0x2 /* Disable this counter */
699 #define CNTR_32BIT		0x4 /* Simulate 64 bits for this counter */
700 #define CNTR_VL			0x8 /* Per VL counter */
701 #define CNTR_SDMA              0x10
702 #define CNTR_INVALID_VL		-1  /* Specifies invalid VL */
703 #define CNTR_MODE_W		0x0
704 #define CNTR_MODE_R		0x1
705 
706 /* VLs Supported/Operational */
707 #define HFI1_MIN_VLS_SUPPORTED 1
708 #define HFI1_MAX_VLS_SUPPORTED 8
709 
710 #define HFI1_GUIDS_PER_PORT  5
711 #define HFI1_PORT_GUID_INDEX 0
712 
713 static inline void incr_cntr64(u64 *cntr)
714 {
715 	if (*cntr < (u64)-1LL)
716 		(*cntr)++;
717 }
718 
719 #define MAX_NAME_SIZE 64
720 struct hfi1_msix_entry {
721 	enum irq_type type;
722 	int irq;
723 	void *arg;
724 	cpumask_t mask;
725 	struct irq_affinity_notify notify;
726 };
727 
728 struct hfi1_msix_info {
729 	/* lock to synchronize in_use_msix access */
730 	spinlock_t msix_lock;
731 	DECLARE_BITMAP(in_use_msix, CCE_NUM_MSIX_VECTORS);
732 	struct hfi1_msix_entry *msix_entries;
733 	u16 max_requested;
734 };
735 
736 /* per-SL CCA information */
737 struct cca_timer {
738 	struct hrtimer hrtimer;
739 	struct hfi1_pportdata *ppd; /* read-only */
740 	int sl; /* read-only */
741 	u16 ccti; /* read/write - current value of CCTI */
742 };
743 
744 struct link_down_reason {
745 	/*
746 	 * SMA-facing value.  Should be set from .latest when
747 	 * HLS_UP_* -> HLS_DN_* transition actually occurs.
748 	 */
749 	u8 sma;
750 	u8 latest;
751 };
752 
753 enum {
754 	LO_PRIO_TABLE,
755 	HI_PRIO_TABLE,
756 	MAX_PRIO_TABLE
757 };
758 
759 struct vl_arb_cache {
760 	/* protect vl arb cache */
761 	spinlock_t lock;
762 	struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
763 };
764 
765 /*
766  * The structure below encapsulates data relevant to a physical IB Port.
767  * Current chips support only one such port, but the separation
768  * clarifies things a bit. Note that to conform to IB conventions,
769  * port-numbers are one-based. The first or only port is port1.
770  */
771 struct hfi1_pportdata {
772 	struct hfi1_ibport ibport_data;
773 
774 	struct hfi1_devdata *dd;
775 
776 	/* PHY support */
777 	struct qsfp_data qsfp_info;
778 	/* Values for SI tuning of SerDes */
779 	u32 port_type;
780 	u32 tx_preset_eq;
781 	u32 tx_preset_noeq;
782 	u32 rx_preset;
783 	u8  local_atten;
784 	u8  remote_atten;
785 	u8  default_atten;
786 	u8  max_power_class;
787 
788 	/* did we read platform config from scratch registers? */
789 	bool config_from_scratch;
790 
791 	/* GUIDs for this interface, in host order, guids[0] is a port guid */
792 	u64 guids[HFI1_GUIDS_PER_PORT];
793 
794 	/* GUID for peer interface, in host order */
795 	u64 neighbor_guid;
796 
797 	/* up or down physical link state */
798 	u32 linkup;
799 
800 	/*
801 	 * this address is mapped read-only into user processes so they can
802 	 * get status cheaply, whenever they want.  One qword of status per port
803 	 */
804 	u64 *statusp;
805 
806 	/* SendDMA related entries */
807 
808 	struct workqueue_struct *hfi1_wq;
809 	struct workqueue_struct *link_wq;
810 
811 	/* move out of interrupt context */
812 	struct work_struct link_vc_work;
813 	struct work_struct link_up_work;
814 	struct work_struct link_down_work;
815 	struct work_struct sma_message_work;
816 	struct work_struct freeze_work;
817 	struct work_struct link_downgrade_work;
818 	struct work_struct link_bounce_work;
819 	struct delayed_work start_link_work;
820 	/* host link state variables */
821 	struct mutex hls_lock;
822 	u32 host_link_state;
823 
824 	/* these are the "32 bit" regs */
825 
826 	u32 ibmtu; /* The MTU programmed for this unit */
827 	/*
828 	 * Current max size IB packet (in bytes) including IB headers, that
829 	 * we can send. Changes when ibmtu changes.
830 	 */
831 	u32 ibmaxlen;
832 	u32 current_egress_rate; /* units [10^6 bits/sec] */
833 	/* LID programmed for this instance */
834 	u32 lid;
835 	/* list of pkeys programmed; 0 if not set */
836 	u16 pkeys[MAX_PKEY_VALUES];
837 	u16 link_width_supported;
838 	u16 link_width_downgrade_supported;
839 	u16 link_speed_supported;
840 	u16 link_width_enabled;
841 	u16 link_width_downgrade_enabled;
842 	u16 link_speed_enabled;
843 	u16 link_width_active;
844 	u16 link_width_downgrade_tx_active;
845 	u16 link_width_downgrade_rx_active;
846 	u16 link_speed_active;
847 	u8 vls_supported;
848 	u8 vls_operational;
849 	u8 actual_vls_operational;
850 	/* LID mask control */
851 	u8 lmc;
852 	/* Rx Polarity inversion (compensate for ~tx on partner) */
853 	u8 rx_pol_inv;
854 
855 	u8 hw_pidx;     /* physical port index */
856 	u32 port;        /* IB port number and index into dd->pports - 1 */
857 	/* type of neighbor node */
858 	u8 neighbor_type;
859 	u8 neighbor_normal;
860 	u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
861 	u8 neighbor_port_number;
862 	u8 is_sm_config_started;
863 	u8 offline_disabled_reason;
864 	u8 is_active_optimize_enabled;
865 	u8 driver_link_ready;	/* driver ready for active link */
866 	u8 link_enabled;	/* link enabled? */
867 	u8 linkinit_reason;
868 	u8 local_tx_rate;	/* rate given to 8051 firmware */
869 	u8 qsfp_retry_count;
870 
871 	/* placeholders for IB MAD packet settings */
872 	u8 overrun_threshold;
873 	u8 phy_error_threshold;
874 	unsigned int is_link_down_queued;
875 
876 	/* Used to override LED behavior for things like maintenance beaconing*/
877 	/*
878 	 * Alternates per phase of blink
879 	 * [0] holds LED off duration, [1] holds LED on duration
880 	 */
881 	unsigned long led_override_vals[2];
882 	u8 led_override_phase; /* LSB picks from vals[] */
883 	atomic_t led_override_timer_active;
884 	/* Used to flash LEDs in override mode */
885 	struct timer_list led_override_timer;
886 
887 	u32 sm_trap_qp;
888 	u32 sa_qp;
889 
890 	/*
891 	 * cca_timer_lock protects access to the per-SL cca_timer
892 	 * structures (specifically the ccti member).
893 	 */
894 	spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
895 	struct cca_timer cca_timer[OPA_MAX_SLS];
896 
897 	/* List of congestion control table entries */
898 	struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
899 
900 	/* congestion entries, each entry corresponding to a SL */
901 	struct opa_congestion_setting_entry_shadow
902 		congestion_entries[OPA_MAX_SLS];
903 
904 	/*
905 	 * cc_state_lock protects (write) access to the per-port
906 	 * struct cc_state.
907 	 */
908 	spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
909 
910 	struct cc_state __rcu *cc_state;
911 
912 	/* Total number of congestion control table entries */
913 	u16 total_cct_entry;
914 
915 	/* Bit map identifying service level */
916 	u32 cc_sl_control_map;
917 
918 	/* CA's max number of 64 entry units in the congestion control table */
919 	u8 cc_max_table_entries;
920 
921 	/*
922 	 * begin congestion log related entries
923 	 * cc_log_lock protects all congestion log related data
924 	 */
925 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
926 	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
927 	u16 threshold_event_counter;
928 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
929 	int cc_log_idx; /* index for logging events */
930 	int cc_mad_idx; /* index for reporting events */
931 	/* end congestion log related entries */
932 
933 	struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
934 
935 	/* port relative counter buffer */
936 	u64 *cntrs;
937 	/* port relative synthetic counter buffer */
938 	u64 *scntrs;
939 	/* port_xmit_discards are synthesized from different egress errors */
940 	u64 port_xmit_discards;
941 	u64 port_xmit_discards_vl[C_VL_COUNT];
942 	u64 port_xmit_constraint_errors;
943 	u64 port_rcv_constraint_errors;
944 	/* count of 'link_err' interrupts from DC */
945 	u64 link_downed;
946 	/* number of times link retrained successfully */
947 	u64 link_up;
948 	/* number of times a link unknown frame was reported */
949 	u64 unknown_frame_count;
950 	/* port_ltp_crc_mode is returned in 'portinfo' MADs */
951 	u16 port_ltp_crc_mode;
952 	/* port_crc_mode_enabled is the crc we support */
953 	u8 port_crc_mode_enabled;
954 	/* mgmt_allowed is also returned in 'portinfo' MADs */
955 	u8 mgmt_allowed;
956 	u8 part_enforce; /* partition enforcement flags */
957 	struct link_down_reason local_link_down_reason;
958 	struct link_down_reason neigh_link_down_reason;
959 	/* Value to be sent to link peer on LinkDown .*/
960 	u8 remote_link_down_reason;
961 	/* Error events that will cause a port bounce. */
962 	u32 port_error_action;
963 	struct work_struct linkstate_active_work;
964 	/* Does this port need to prescan for FECNs */
965 	bool cc_prescan;
966 	/*
967 	 * Sample sendWaitCnt & sendWaitVlCnt during link transition
968 	 * and counter request.
969 	 */
970 	u64 port_vl_xmit_wait_last[C_VL_COUNT + 1];
971 	u16 prev_link_width;
972 	u64 vl_xmit_flit_cnt[C_VL_COUNT + 1];
973 };
974 
975 typedef void (*opcode_handler)(struct hfi1_packet *packet);
976 typedef void (*hfi1_make_req)(struct rvt_qp *qp,
977 			      struct hfi1_pkt_state *ps,
978 			      struct rvt_swqe *wqe);
979 extern const rhf_rcv_function_ptr normal_rhf_rcv_functions[];
980 extern const rhf_rcv_function_ptr netdev_rhf_rcv_functions[];
981 
982 /* return values for the RHF receive functions */
983 #define RHF_RCV_CONTINUE  0	/* keep going */
984 #define RHF_RCV_DONE	  1	/* stop, this packet processed */
985 #define RHF_RCV_REPROCESS 2	/* stop. retain this packet */
986 
987 struct rcv_array_data {
988 	u16 ngroups;
989 	u16 nctxt_extra;
990 	u8 group_size;
991 };
992 
993 struct per_vl_data {
994 	u16 mtu;
995 	struct send_context *sc;
996 };
997 
998 /* 16 to directly index */
999 #define PER_VL_SEND_CONTEXTS 16
1000 
1001 struct err_info_rcvport {
1002 	u8 status_and_code;
1003 	u64 packet_flit1;
1004 	u64 packet_flit2;
1005 };
1006 
1007 struct err_info_constraint {
1008 	u8 status;
1009 	u16 pkey;
1010 	u32 slid;
1011 };
1012 
1013 struct hfi1_temp {
1014 	unsigned int curr;       /* current temperature */
1015 	unsigned int lo_lim;     /* low temperature limit */
1016 	unsigned int hi_lim;     /* high temperature limit */
1017 	unsigned int crit_lim;   /* critical temperature limit */
1018 	u8 triggers;      /* temperature triggers */
1019 };
1020 
1021 struct hfi1_i2c_bus {
1022 	struct hfi1_devdata *controlling_dd; /* current controlling device */
1023 	struct i2c_adapter adapter;	/* bus details */
1024 	struct i2c_algo_bit_data algo;	/* bus algorithm details */
1025 	int num;			/* bus number, 0 or 1 */
1026 };
1027 
1028 /* common data between shared ASIC HFIs */
1029 struct hfi1_asic_data {
1030 	struct hfi1_devdata *dds[2];	/* back pointers */
1031 	struct mutex asic_resource_mutex;
1032 	struct hfi1_i2c_bus *i2c_bus0;
1033 	struct hfi1_i2c_bus *i2c_bus1;
1034 };
1035 
1036 /* sizes for both the QP and RSM map tables */
1037 #define NUM_MAP_ENTRIES	 256
1038 #define NUM_MAP_REGS      32
1039 
1040 /* Virtual NIC information */
1041 struct hfi1_vnic_data {
1042 	struct kmem_cache *txreq_cache;
1043 	u8 num_vports;
1044 };
1045 
1046 struct hfi1_vnic_vport_info;
1047 
1048 /* device data struct now contains only "general per-device" info.
1049  * fields related to a physical IB port are in a hfi1_pportdata struct.
1050  */
1051 struct sdma_engine;
1052 struct sdma_vl_map;
1053 
1054 #define BOARD_VERS_MAX 96 /* how long the version string can be */
1055 #define SERIAL_MAX 16 /* length of the serial number */
1056 
1057 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
1058 struct hfi1_netdev_rx;
1059 struct hfi1_devdata {
1060 	struct hfi1_ibdev verbs_dev;     /* must be first */
1061 	/* pointers to related structs for this device */
1062 	/* pci access data structure */
1063 	struct pci_dev *pcidev;
1064 	struct cdev user_cdev;
1065 	struct cdev diag_cdev;
1066 	struct cdev ui_cdev;
1067 	struct device *user_device;
1068 	struct device *diag_device;
1069 	struct device *ui_device;
1070 
1071 	/* first mapping up to RcvArray */
1072 	u8 __iomem *kregbase1;
1073 	resource_size_t physaddr;
1074 
1075 	/* second uncached mapping from RcvArray to pio send buffers */
1076 	u8 __iomem *kregbase2;
1077 	/* for detecting offset above kregbase2 address */
1078 	u32 base2_start;
1079 
1080 	/* Per VL data. Enough for all VLs but not all elements are set/used. */
1081 	struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
1082 	/* send context data */
1083 	struct send_context_info *send_contexts;
1084 	/* map hardware send contexts to software index */
1085 	u8 *hw_to_sw;
1086 	/* spinlock for allocating and releasing send context resources */
1087 	spinlock_t sc_lock;
1088 	/* lock for pio_map */
1089 	spinlock_t pio_map_lock;
1090 	/* Send Context initialization lock. */
1091 	spinlock_t sc_init_lock;
1092 	/* lock for sdma_map */
1093 	spinlock_t                          sde_map_lock;
1094 	/* array of kernel send contexts */
1095 	struct send_context **kernel_send_context;
1096 	/* array of vl maps */
1097 	struct pio_vl_map __rcu *pio_map;
1098 	/* default flags to last descriptor */
1099 	u64 default_desc1;
1100 
1101 	/* fields common to all SDMA engines */
1102 
1103 	volatile __le64                    *sdma_heads_dma; /* DMA'ed by chip */
1104 	dma_addr_t                          sdma_heads_phys;
1105 	void                               *sdma_pad_dma; /* DMA'ed by chip */
1106 	dma_addr_t                          sdma_pad_phys;
1107 	/* for deallocation */
1108 	size_t                              sdma_heads_size;
1109 	/* num used */
1110 	u32                                 num_sdma;
1111 	/* array of engines sized by num_sdma */
1112 	struct sdma_engine                 *per_sdma;
1113 	/* array of vl maps */
1114 	struct sdma_vl_map __rcu           *sdma_map;
1115 	/* SPC freeze waitqueue and variable */
1116 	wait_queue_head_t		  sdma_unfreeze_wq;
1117 	atomic_t			  sdma_unfreeze_count;
1118 
1119 	u32 lcb_access_count;		/* count of LCB users */
1120 
1121 	/* common data between shared ASIC HFIs in this OS */
1122 	struct hfi1_asic_data *asic_data;
1123 
1124 	/* mem-mapped pointer to base of PIO buffers */
1125 	void __iomem *piobase;
1126 	/*
1127 	 * write-combining mem-mapped pointer to base of RcvArray
1128 	 * memory.
1129 	 */
1130 	void __iomem *rcvarray_wc;
1131 	/*
1132 	 * credit return base - a per-NUMA range of DMA address that
1133 	 * the chip will use to update the per-context free counter
1134 	 */
1135 	struct credit_return_base *cr_base;
1136 
1137 	/* send context numbers and sizes for each type */
1138 	struct sc_config_sizes sc_sizes[SC_MAX];
1139 
1140 	char *boardname; /* human readable board info */
1141 
1142 	u64 ctx0_seq_drop;
1143 
1144 	/* reset value */
1145 	u64 z_int_counter;
1146 	u64 z_rcv_limit;
1147 	u64 z_send_schedule;
1148 
1149 	u64 __percpu *send_schedule;
1150 	/* number of reserved contexts for netdev usage */
1151 	u16 num_netdev_contexts;
1152 	/* number of receive contexts in use by the driver */
1153 	u32 num_rcv_contexts;
1154 	/* number of pio send contexts in use by the driver */
1155 	u32 num_send_contexts;
1156 	/*
1157 	 * number of ctxts available for PSM open
1158 	 */
1159 	u32 freectxts;
1160 	/* total number of available user/PSM contexts */
1161 	u32 num_user_contexts;
1162 	/* base receive interrupt timeout, in CSR units */
1163 	u32 rcv_intr_timeout_csr;
1164 
1165 	spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
1166 	spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
1167 	spinlock_t uctxt_lock; /* protect rcd changes */
1168 	struct mutex dc8051_lock; /* exclusive access to 8051 */
1169 	struct workqueue_struct *update_cntr_wq;
1170 	struct work_struct update_cntr_work;
1171 	/* exclusive access to 8051 memory */
1172 	spinlock_t dc8051_memlock;
1173 	int dc8051_timed_out;	/* remember if the 8051 timed out */
1174 	/*
1175 	 * A page that will hold event notification bitmaps for all
1176 	 * contexts. This page will be mapped into all processes.
1177 	 */
1178 	unsigned long *events;
1179 	/*
1180 	 * per unit status, see also portdata statusp
1181 	 * mapped read-only into user processes so they can get unit and
1182 	 * IB link status cheaply
1183 	 */
1184 	struct hfi1_status *status;
1185 
1186 	/* revision register shadow */
1187 	u64 revision;
1188 	/* Base GUID for device (network order) */
1189 	u64 base_guid;
1190 
1191 	/* both sides of the PCIe link are gen3 capable */
1192 	u8 link_gen3_capable;
1193 	u8 dc_shutdown;
1194 	/* localbus width (1, 2,4,8,16,32) from config space  */
1195 	u32 lbus_width;
1196 	/* localbus speed in MHz */
1197 	u32 lbus_speed;
1198 	int unit; /* unit # of this chip */
1199 	int node; /* home node of this chip */
1200 
1201 	/* save these PCI fields to restore after a reset */
1202 	u32 pcibar0;
1203 	u32 pcibar1;
1204 	u32 pci_rom;
1205 	u16 pci_command;
1206 	u16 pcie_devctl;
1207 	u16 pcie_lnkctl;
1208 	u16 pcie_devctl2;
1209 	u32 pci_msix0;
1210 	u32 pci_tph2;
1211 
1212 	/*
1213 	 * ASCII serial number, from flash, large enough for original
1214 	 * all digit strings, and longer serial number format
1215 	 */
1216 	u8 serial[SERIAL_MAX];
1217 	/* human readable board version */
1218 	u8 boardversion[BOARD_VERS_MAX];
1219 	u8 lbus_info[32]; /* human readable localbus info */
1220 	/* chip major rev, from CceRevision */
1221 	u8 majrev;
1222 	/* chip minor rev, from CceRevision */
1223 	u8 minrev;
1224 	/* hardware ID */
1225 	u8 hfi1_id;
1226 	/* implementation code */
1227 	u8 icode;
1228 	/* vAU of this device */
1229 	u8 vau;
1230 	/* vCU of this device */
1231 	u8 vcu;
1232 	/* link credits of this device */
1233 	u16 link_credits;
1234 	/* initial vl15 credits to use */
1235 	u16 vl15_init;
1236 
1237 	/*
1238 	 * Cached value for vl15buf, read during verify cap interrupt. VL15
1239 	 * credits are to be kept at 0 and set when handling the link-up
1240 	 * interrupt. This removes the possibility of receiving VL15 MAD
1241 	 * packets before this HFI is ready.
1242 	 */
1243 	u16 vl15buf_cached;
1244 
1245 	/* Misc small ints */
1246 	u8 n_krcv_queues;
1247 	u8 qos_shift;
1248 
1249 	u16 irev;	/* implementation revision */
1250 	u32 dc8051_ver; /* 8051 firmware version */
1251 
1252 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1253 	struct platform_config platform_config;
1254 	struct platform_config_cache pcfg_cache;
1255 
1256 	struct diag_client *diag_client;
1257 
1258 	/* general interrupt: mask of handled interrupts */
1259 	u64 gi_mask[CCE_NUM_INT_CSRS];
1260 
1261 	struct rcv_array_data rcv_entries;
1262 
1263 	/* cycle length of PS* counters in HW (in picoseconds) */
1264 	u16 psxmitwait_check_rate;
1265 
1266 	/*
1267 	 * 64 bit synthetic counters
1268 	 */
1269 	struct timer_list synth_stats_timer;
1270 
1271 	/* MSI-X information */
1272 	struct hfi1_msix_info msix_info;
1273 
1274 	/*
1275 	 * device counters
1276 	 */
1277 	char *cntrnames;
1278 	size_t cntrnameslen;
1279 	size_t ndevcntrs;
1280 	u64 *cntrs;
1281 	u64 *scntrs;
1282 
1283 	/*
1284 	 * remembered values for synthetic counters
1285 	 */
1286 	u64 last_tx;
1287 	u64 last_rx;
1288 
1289 	/*
1290 	 * per-port counters
1291 	 */
1292 	size_t nportcntrs;
1293 	char *portcntrnames;
1294 	size_t portcntrnameslen;
1295 
1296 	struct err_info_rcvport err_info_rcvport;
1297 	struct err_info_constraint err_info_rcv_constraint;
1298 	struct err_info_constraint err_info_xmit_constraint;
1299 
1300 	atomic_t drop_packet;
1301 	bool do_drop;
1302 	u8 err_info_uncorrectable;
1303 	u8 err_info_fmconfig;
1304 
1305 	/*
1306 	 * Software counters for the status bits defined by the
1307 	 * associated error status registers
1308 	 */
1309 	u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1310 	u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1311 	u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1312 	u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1313 	u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1314 	u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1315 	u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1316 
1317 	/* Software counter that spans all contexts */
1318 	u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1319 	/* Software counter that spans all DMA engines */
1320 	u64 sw_send_dma_eng_err_status_cnt[
1321 		NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1322 	/* Software counter that aggregates all cce_err_status errors */
1323 	u64 sw_cce_err_status_aggregate;
1324 	/* Software counter that aggregates all bypass packet rcv errors */
1325 	u64 sw_rcv_bypass_packet_errors;
1326 
1327 	/* Save the enabled LCB error bits */
1328 	u64 lcb_err_en;
1329 	struct cpu_mask_set *comp_vect;
1330 	int *comp_vect_mappings;
1331 	u32 comp_vect_possible_cpus;
1332 
1333 	/*
1334 	 * Capability to have different send engines simply by changing a
1335 	 * pointer value.
1336 	 */
1337 	send_routine process_pio_send ____cacheline_aligned_in_smp;
1338 	send_routine process_dma_send;
1339 	void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1340 				u64 pbc, const void *from, size_t count);
1341 	int (*process_vnic_dma_send)(struct hfi1_devdata *dd, u8 q_idx,
1342 				     struct hfi1_vnic_vport_info *vinfo,
1343 				     struct sk_buff *skb, u64 pbc, u8 plen);
1344 	/* hfi1_pportdata, points to array of (physical) port-specific
1345 	 * data structs, indexed by pidx (0..n-1)
1346 	 */
1347 	struct hfi1_pportdata *pport;
1348 	/* receive context data */
1349 	struct hfi1_ctxtdata **rcd;
1350 	u64 __percpu *int_counter;
1351 	/* verbs tx opcode stats */
1352 	struct hfi1_opcode_stats_perctx __percpu *tx_opstats;
1353 	/* device (not port) flags, basically device capabilities */
1354 	u16 flags;
1355 	/* Number of physical ports available */
1356 	u8 num_pports;
1357 	/* Lowest context number which can be used by user processes or VNIC */
1358 	u8 first_dyn_alloc_ctxt;
1359 	/* adding a new field here would make it part of this cacheline */
1360 
1361 	/* seqlock for sc2vl */
1362 	seqlock_t sc2vl_lock ____cacheline_aligned_in_smp;
1363 	u64 sc2vl[4];
1364 	u64 __percpu *rcv_limit;
1365 	/* adding a new field here would make it part of this cacheline */
1366 
1367 	/* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1368 	u8 oui1;
1369 	u8 oui2;
1370 	u8 oui3;
1371 
1372 	/* Timer and counter used to detect RcvBufOvflCnt changes */
1373 	struct timer_list rcverr_timer;
1374 
1375 	wait_queue_head_t event_queue;
1376 
1377 	/* receive context tail dummy address */
1378 	__le64 *rcvhdrtail_dummy_kvaddr;
1379 	dma_addr_t rcvhdrtail_dummy_dma;
1380 
1381 	u32 rcv_ovfl_cnt;
1382 	/* Serialize ASPM enable/disable between multiple verbs contexts */
1383 	spinlock_t aspm_lock;
1384 	/* Number of verbs contexts which have disabled ASPM */
1385 	atomic_t aspm_disabled_cnt;
1386 	/* Keeps track of user space clients */
1387 	atomic_t user_refcount;
1388 	/* Used to wait for outstanding user space clients before dev removal */
1389 	struct completion user_comp;
1390 
1391 	bool eprom_available;	/* true if EPROM is available for this device */
1392 	bool aspm_supported;	/* Does HW support ASPM */
1393 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
1394 	struct rhashtable *sdma_rht;
1395 
1396 	/* vnic data */
1397 	struct hfi1_vnic_data vnic;
1398 	/* Lock to protect IRQ SRC register access */
1399 	spinlock_t irq_src_lock;
1400 	int vnic_num_vports;
1401 	struct hfi1_netdev_rx *netdev_rx;
1402 	struct hfi1_affinity_node *affinity_entry;
1403 
1404 	/* Keeps track of IPoIB RSM rule users */
1405 	atomic_t ipoib_rsm_usr_num;
1406 };
1407 
1408 /* 8051 firmware version helper */
1409 #define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
1410 #define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
1411 #define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
1412 #define dc8051_ver_patch(a) ((a) & 0x0000ff)
1413 
1414 /* f_put_tid types */
1415 #define PT_EXPECTED       0
1416 #define PT_EAGER          1
1417 #define PT_INVALID_FLUSH  2
1418 #define PT_INVALID        3
1419 
1420 struct tid_rb_node;
1421 struct mmu_rb_node;
1422 struct mmu_rb_handler;
1423 
1424 /* Private data for file operations */
1425 struct hfi1_filedata {
1426 	struct srcu_struct pq_srcu;
1427 	struct hfi1_devdata *dd;
1428 	struct hfi1_ctxtdata *uctxt;
1429 	struct hfi1_user_sdma_comp_q *cq;
1430 	/* update side lock for SRCU */
1431 	spinlock_t pq_rcu_lock;
1432 	struct hfi1_user_sdma_pkt_q __rcu *pq;
1433 	u16 subctxt;
1434 	/* for cpu affinity; -1 if none */
1435 	int rec_cpu_num;
1436 	u32 tid_n_pinned;
1437 	bool use_mn;
1438 	struct tid_rb_node **entry_to_rb;
1439 	spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1440 	u32 tid_limit;
1441 	u32 tid_used;
1442 	u32 *invalid_tids;
1443 	u32 invalid_tid_idx;
1444 	/* protect invalid_tids array and invalid_tid_idx */
1445 	spinlock_t invalid_lock;
1446 };
1447 
1448 extern struct xarray hfi1_dev_table;
1449 struct hfi1_devdata *hfi1_lookup(int unit);
1450 
1451 static inline unsigned long uctxt_offset(struct hfi1_ctxtdata *uctxt)
1452 {
1453 	return (uctxt->ctxt - uctxt->dd->first_dyn_alloc_ctxt) *
1454 		HFI1_MAX_SHARED_CTXTS;
1455 }
1456 
1457 int hfi1_init(struct hfi1_devdata *dd, int reinit);
1458 int hfi1_count_active_units(void);
1459 
1460 int hfi1_diag_add(struct hfi1_devdata *dd);
1461 void hfi1_diag_remove(struct hfi1_devdata *dd);
1462 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1463 
1464 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1465 
1466 int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1467 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
1468 int hfi1_create_kctxts(struct hfi1_devdata *dd);
1469 int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa,
1470 			 struct hfi1_ctxtdata **rcd);
1471 void hfi1_free_ctxt(struct hfi1_ctxtdata *rcd);
1472 void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
1473 			 struct hfi1_devdata *dd, u8 hw_pidx, u32 port);
1474 void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1475 int hfi1_rcd_put(struct hfi1_ctxtdata *rcd);
1476 int hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
1477 struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd,
1478 						 u16 ctxt);
1479 struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt);
1480 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
1481 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1482 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1483 int handle_receive_interrupt_napi_fp(struct hfi1_ctxtdata *rcd, int budget);
1484 int handle_receive_interrupt_napi_sp(struct hfi1_ctxtdata *rcd, int budget);
1485 void set_all_slowpath(struct hfi1_devdata *dd);
1486 
1487 extern const struct pci_device_id hfi1_pci_tbl[];
1488 void hfi1_make_ud_req_9B(struct rvt_qp *qp,
1489 			 struct hfi1_pkt_state *ps,
1490 			 struct rvt_swqe *wqe);
1491 
1492 void hfi1_make_ud_req_16B(struct rvt_qp *qp,
1493 			  struct hfi1_pkt_state *ps,
1494 			  struct rvt_swqe *wqe);
1495 
1496 /* receive packet handler dispositions */
1497 #define RCV_PKT_OK      0x0 /* keep going */
1498 #define RCV_PKT_LIMIT   0x1 /* stop, hit limit, start thread */
1499 #define RCV_PKT_DONE    0x2 /* stop, no more packets detected */
1500 
1501 /**
1502  * hfi1_rcd_head - add accessor for rcd head
1503  * @rcd: the context
1504  */
1505 static inline u32 hfi1_rcd_head(struct hfi1_ctxtdata *rcd)
1506 {
1507 	return rcd->head;
1508 }
1509 
1510 /**
1511  * hfi1_set_rcd_head - add accessor for rcd head
1512  * @rcd: the context
1513  * @head: the new head
1514  */
1515 static inline void hfi1_set_rcd_head(struct hfi1_ctxtdata *rcd, u32 head)
1516 {
1517 	rcd->head = head;
1518 }
1519 
1520 /* calculate the current RHF address */
1521 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1522 {
1523 	return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->rhf_offset;
1524 }
1525 
1526 /* return DMA_RTAIL configuration */
1527 static inline bool get_dma_rtail_setting(struct hfi1_ctxtdata *rcd)
1528 {
1529 	return !!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL);
1530 }
1531 
1532 /**
1533  * hfi1_seq_incr_wrap - wrapping increment for sequence
1534  * @seq: the current sequence number
1535  *
1536  * Returns: the incremented seq
1537  */
1538 static inline u8 hfi1_seq_incr_wrap(u8 seq)
1539 {
1540 	if (++seq > RHF_MAX_SEQ)
1541 		seq = 1;
1542 	return seq;
1543 }
1544 
1545 /**
1546  * hfi1_seq_cnt - return seq_cnt member
1547  * @rcd: the receive context
1548  *
1549  * Return seq_cnt member
1550  */
1551 static inline u8 hfi1_seq_cnt(struct hfi1_ctxtdata *rcd)
1552 {
1553 	return rcd->seq_cnt;
1554 }
1555 
1556 /**
1557  * hfi1_set_seq_cnt - return seq_cnt member
1558  * @rcd: the receive context
1559  *
1560  * Return seq_cnt member
1561  */
1562 static inline void hfi1_set_seq_cnt(struct hfi1_ctxtdata *rcd, u8 cnt)
1563 {
1564 	rcd->seq_cnt = cnt;
1565 }
1566 
1567 /**
1568  * last_rcv_seq - is last
1569  * @rcd: the receive context
1570  * @seq: sequence
1571  *
1572  * return true if last packet
1573  */
1574 static inline bool last_rcv_seq(struct hfi1_ctxtdata *rcd, u32 seq)
1575 {
1576 	return seq != rcd->seq_cnt;
1577 }
1578 
1579 /**
1580  * rcd_seq_incr - increment context sequence number
1581  * @rcd: the receive context
1582  * @seq: the current sequence number
1583  *
1584  * Returns: true if the this was the last packet
1585  */
1586 static inline bool hfi1_seq_incr(struct hfi1_ctxtdata *rcd, u32 seq)
1587 {
1588 	rcd->seq_cnt = hfi1_seq_incr_wrap(rcd->seq_cnt);
1589 	return last_rcv_seq(rcd, seq);
1590 }
1591 
1592 /**
1593  * get_hdrqentsize - return hdrq entry size
1594  * @rcd: the receive context
1595  */
1596 static inline u8 get_hdrqentsize(struct hfi1_ctxtdata *rcd)
1597 {
1598 	return rcd->rcvhdrqentsize;
1599 }
1600 
1601 /**
1602  * get_hdrq_cnt - return hdrq count
1603  * @rcd: the receive context
1604  */
1605 static inline u16 get_hdrq_cnt(struct hfi1_ctxtdata *rcd)
1606 {
1607 	return rcd->rcvhdrq_cnt;
1608 }
1609 
1610 /**
1611  * hfi1_is_slowpath - check if this context is slow path
1612  * @rcd: the receive context
1613  */
1614 static inline bool hfi1_is_slowpath(struct hfi1_ctxtdata *rcd)
1615 {
1616 	return rcd->do_interrupt == rcd->slow_handler;
1617 }
1618 
1619 /**
1620  * hfi1_is_fastpath - check if this context is fast path
1621  * @rcd: the receive context
1622  */
1623 static inline bool hfi1_is_fastpath(struct hfi1_ctxtdata *rcd)
1624 {
1625 	if (rcd->ctxt == HFI1_CTRL_CTXT)
1626 		return false;
1627 
1628 	return rcd->do_interrupt == rcd->fast_handler;
1629 }
1630 
1631 /**
1632  * hfi1_set_fast - change to the fast handler
1633  * @rcd: the receive context
1634  */
1635 static inline void hfi1_set_fast(struct hfi1_ctxtdata *rcd)
1636 {
1637 	if (unlikely(!rcd))
1638 		return;
1639 	if (unlikely(!hfi1_is_fastpath(rcd)))
1640 		rcd->do_interrupt = rcd->fast_handler;
1641 }
1642 
1643 int hfi1_reset_device(int);
1644 
1645 void receive_interrupt_work(struct work_struct *work);
1646 
1647 /* extract service channel from header and rhf */
1648 static inline int hfi1_9B_get_sc5(struct ib_header *hdr, u64 rhf)
1649 {
1650 	return ib_get_sc(hdr) | ((!!(rhf_dc_info(rhf))) << 4);
1651 }
1652 
1653 #define HFI1_JKEY_WIDTH       16
1654 #define HFI1_JKEY_MASK        (BIT(16) - 1)
1655 #define HFI1_ADMIN_JKEY_RANGE 32
1656 
1657 /*
1658  * J_KEYs are split and allocated in the following groups:
1659  *   0 - 31    - users with administrator privileges
1660  *  32 - 63    - kernel protocols using KDETH packets
1661  *  64 - 65535 - all other users using KDETH packets
1662  */
1663 static inline u16 generate_jkey(kuid_t uid)
1664 {
1665 	u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1666 
1667 	if (capable(CAP_SYS_ADMIN))
1668 		jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1669 	else if (jkey < 64)
1670 		jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1671 
1672 	return jkey;
1673 }
1674 
1675 /*
1676  * active_egress_rate
1677  *
1678  * returns the active egress rate in units of [10^6 bits/sec]
1679  */
1680 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1681 {
1682 	u16 link_speed = ppd->link_speed_active;
1683 	u16 link_width = ppd->link_width_active;
1684 	u32 egress_rate;
1685 
1686 	if (link_speed == OPA_LINK_SPEED_25G)
1687 		egress_rate = 25000;
1688 	else /* assume OPA_LINK_SPEED_12_5G */
1689 		egress_rate = 12500;
1690 
1691 	switch (link_width) {
1692 	case OPA_LINK_WIDTH_4X:
1693 		egress_rate *= 4;
1694 		break;
1695 	case OPA_LINK_WIDTH_3X:
1696 		egress_rate *= 3;
1697 		break;
1698 	case OPA_LINK_WIDTH_2X:
1699 		egress_rate *= 2;
1700 		break;
1701 	default:
1702 		/* assume IB_WIDTH_1X */
1703 		break;
1704 	}
1705 
1706 	return egress_rate;
1707 }
1708 
1709 /*
1710  * egress_cycles
1711  *
1712  * Returns the number of 'fabric clock cycles' to egress a packet
1713  * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1714  * rate is (approximately) 805 MHz, the units of the returned value
1715  * are (1/805 MHz).
1716  */
1717 static inline u32 egress_cycles(u32 len, u32 rate)
1718 {
1719 	u32 cycles;
1720 
1721 	/*
1722 	 * cycles is:
1723 	 *
1724 	 *          (length) [bits] / (rate) [bits/sec]
1725 	 *  ---------------------------------------------------
1726 	 *  fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1727 	 */
1728 
1729 	cycles = len * 8; /* bits */
1730 	cycles *= 805;
1731 	cycles /= rate;
1732 
1733 	return cycles;
1734 }
1735 
1736 void set_link_ipg(struct hfi1_pportdata *ppd);
1737 void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
1738 		  u32 rqpn, u8 svc_type);
1739 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1740 		u16 pkey, u32 slid, u32 dlid, u8 sc5,
1741 		const struct ib_grh *old_grh);
1742 void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1743 		    u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1744 		    u8 sc5, const struct ib_grh *old_grh);
1745 typedef void (*hfi1_handle_cnp)(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1746 				u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1747 				u8 sc5, const struct ib_grh *old_grh);
1748 
1749 #define PKEY_CHECK_INVALID -1
1750 int egress_pkey_check(struct hfi1_pportdata *ppd, u32 slid, u16 pkey,
1751 		      u8 sc5, int8_t s_pkey_index);
1752 
1753 #define PACKET_EGRESS_TIMEOUT 350
1754 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1755 {
1756 	/* Pause at least 1us, to ensure chip returns all credits */
1757 	u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1758 
1759 	udelay(usec ? usec : 1);
1760 }
1761 
1762 /**
1763  * sc_to_vlt() - reverse lookup sc to vl
1764  * @dd - devdata
1765  * @sc5 - 5 bit sc
1766  */
1767 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1768 {
1769 	unsigned seq;
1770 	u8 rval;
1771 
1772 	if (sc5 >= OPA_MAX_SCS)
1773 		return (u8)(0xff);
1774 
1775 	do {
1776 		seq = read_seqbegin(&dd->sc2vl_lock);
1777 		rval = *(((u8 *)dd->sc2vl) + sc5);
1778 	} while (read_seqretry(&dd->sc2vl_lock, seq));
1779 
1780 	return rval;
1781 }
1782 
1783 #define PKEY_MEMBER_MASK 0x8000
1784 #define PKEY_LOW_15_MASK 0x7fff
1785 
1786 /*
1787  * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1788  * being an entry from the ingress partition key table), return 0
1789  * otherwise. Use the matching criteria for ingress partition keys
1790  * specified in the OPAv1 spec., section 9.10.14.
1791  */
1792 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1793 {
1794 	u16 mkey = pkey & PKEY_LOW_15_MASK;
1795 	u16 ment = ent & PKEY_LOW_15_MASK;
1796 
1797 	if (mkey == ment) {
1798 		/*
1799 		 * If pkey[15] is clear (limited partition member),
1800 		 * is bit 15 in the corresponding table element
1801 		 * clear (limited member)?
1802 		 */
1803 		if (!(pkey & PKEY_MEMBER_MASK))
1804 			return !!(ent & PKEY_MEMBER_MASK);
1805 		return 1;
1806 	}
1807 	return 0;
1808 }
1809 
1810 /*
1811  * ingress_pkey_table_search - search the entire pkey table for
1812  * an entry which matches 'pkey'. return 0 if a match is found,
1813  * and 1 otherwise.
1814  */
1815 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1816 {
1817 	int i;
1818 
1819 	for (i = 0; i < MAX_PKEY_VALUES; i++) {
1820 		if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1821 			return 0;
1822 	}
1823 	return 1;
1824 }
1825 
1826 /*
1827  * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1828  * i.e., increment port_rcv_constraint_errors for the port, and record
1829  * the 'error info' for this failure.
1830  */
1831 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1832 				    u32 slid)
1833 {
1834 	struct hfi1_devdata *dd = ppd->dd;
1835 
1836 	incr_cntr64(&ppd->port_rcv_constraint_errors);
1837 	if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1838 		dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1839 		dd->err_info_rcv_constraint.slid = slid;
1840 		dd->err_info_rcv_constraint.pkey = pkey;
1841 	}
1842 }
1843 
1844 /*
1845  * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1846  * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1847  * is a hint as to the best place in the partition key table to begin
1848  * searching. This function should not be called on the data path because
1849  * of performance reasons. On datapath pkey check is expected to be done
1850  * by HW and rcv_pkey_check function should be called instead.
1851  */
1852 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1853 				     u8 sc5, u8 idx, u32 slid, bool force)
1854 {
1855 	if (!(force) && !(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1856 		return 0;
1857 
1858 	/* If SC15, pkey[0:14] must be 0x7fff */
1859 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1860 		goto bad;
1861 
1862 	/* Is the pkey = 0x0, or 0x8000? */
1863 	if ((pkey & PKEY_LOW_15_MASK) == 0)
1864 		goto bad;
1865 
1866 	/* The most likely matching pkey has index 'idx' */
1867 	if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1868 		return 0;
1869 
1870 	/* no match - try the whole table */
1871 	if (!ingress_pkey_table_search(ppd, pkey))
1872 		return 0;
1873 
1874 bad:
1875 	ingress_pkey_table_fail(ppd, pkey, slid);
1876 	return 1;
1877 }
1878 
1879 /*
1880  * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1881  * otherwise. It only ensures pkey is vlid for QP0. This function
1882  * should be called on the data path instead of ingress_pkey_check
1883  * as on data path, pkey check is done by HW (except for QP0).
1884  */
1885 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1886 				 u8 sc5, u16 slid)
1887 {
1888 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1889 		return 0;
1890 
1891 	/* If SC15, pkey[0:14] must be 0x7fff */
1892 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1893 		goto bad;
1894 
1895 	return 0;
1896 bad:
1897 	ingress_pkey_table_fail(ppd, pkey, slid);
1898 	return 1;
1899 }
1900 
1901 /* MTU handling */
1902 
1903 /* MTU enumeration, 256-4k match IB */
1904 #define OPA_MTU_0     0
1905 #define OPA_MTU_256   1
1906 #define OPA_MTU_512   2
1907 #define OPA_MTU_1024  3
1908 #define OPA_MTU_2048  4
1909 #define OPA_MTU_4096  5
1910 
1911 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1912 int mtu_to_enum(u32 mtu, int default_if_bad);
1913 u16 enum_to_mtu(int mtu);
1914 static inline int valid_ib_mtu(unsigned int mtu)
1915 {
1916 	return mtu == 256 || mtu == 512 ||
1917 		mtu == 1024 || mtu == 2048 ||
1918 		mtu == 4096;
1919 }
1920 
1921 static inline int valid_opa_max_mtu(unsigned int mtu)
1922 {
1923 	return mtu >= 2048 &&
1924 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1925 }
1926 
1927 int set_mtu(struct hfi1_pportdata *ppd);
1928 
1929 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
1930 void hfi1_disable_after_error(struct hfi1_devdata *dd);
1931 int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
1932 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
1933 
1934 int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
1935 int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
1936 
1937 void set_up_vau(struct hfi1_devdata *dd, u8 vau);
1938 void set_up_vl15(struct hfi1_devdata *dd, u16 vl15buf);
1939 void reset_link_credits(struct hfi1_devdata *dd);
1940 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1941 
1942 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1943 
1944 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1945 {
1946 	return ppd->dd;
1947 }
1948 
1949 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1950 {
1951 	return container_of(dev, struct hfi1_devdata, verbs_dev);
1952 }
1953 
1954 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1955 {
1956 	return dd_from_dev(to_idev(ibdev));
1957 }
1958 
1959 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1960 {
1961 	return container_of(ibp, struct hfi1_pportdata, ibport_data);
1962 }
1963 
1964 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1965 {
1966 	return container_of(rdi, struct hfi1_ibdev, rdi);
1967 }
1968 
1969 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u32 port)
1970 {
1971 	struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1972 	u32 pidx = port - 1; /* IB number port from 1, hdw from 0 */
1973 
1974 	WARN_ON(pidx >= dd->num_pports);
1975 	return &dd->pport[pidx].ibport_data;
1976 }
1977 
1978 static inline struct hfi1_ibport *rcd_to_iport(struct hfi1_ctxtdata *rcd)
1979 {
1980 	return &rcd->ppd->ibport_data;
1981 }
1982 
1983 /**
1984  * hfi1_may_ecn - Check whether FECN or BECN processing should be done
1985  * @pkt: the packet to be evaluated
1986  *
1987  * Check whether the FECN or BECN bits in the packet's header are
1988  * enabled, depending on packet type.
1989  *
1990  * This function only checks for FECN and BECN bits. Additional checks
1991  * are done in the slowpath (hfi1_process_ecn_slowpath()) in order to
1992  * ensure correct handling.
1993  */
1994 static inline bool hfi1_may_ecn(struct hfi1_packet *pkt)
1995 {
1996 	bool fecn, becn;
1997 
1998 	if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
1999 		fecn = hfi1_16B_get_fecn(pkt->hdr);
2000 		becn = hfi1_16B_get_becn(pkt->hdr);
2001 	} else {
2002 		fecn = ib_bth_get_fecn(pkt->ohdr);
2003 		becn = ib_bth_get_becn(pkt->ohdr);
2004 	}
2005 	return fecn || becn;
2006 }
2007 
2008 bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
2009 			       bool prescan);
2010 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt)
2011 {
2012 	bool do_work;
2013 
2014 	do_work = hfi1_may_ecn(pkt);
2015 	if (unlikely(do_work))
2016 		return hfi1_process_ecn_slowpath(qp, pkt, false);
2017 	return false;
2018 }
2019 
2020 /*
2021  * Return the indexed PKEY from the port PKEY table.
2022  */
2023 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
2024 {
2025 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2026 	u16 ret;
2027 
2028 	if (index >= ARRAY_SIZE(ppd->pkeys))
2029 		ret = 0;
2030 	else
2031 		ret = ppd->pkeys[index];
2032 
2033 	return ret;
2034 }
2035 
2036 /*
2037  * Return the indexed GUID from the port GUIDs table.
2038  */
2039 static inline __be64 get_sguid(struct hfi1_ibport *ibp, unsigned int index)
2040 {
2041 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2042 
2043 	WARN_ON(index >= HFI1_GUIDS_PER_PORT);
2044 	return cpu_to_be64(ppd->guids[index]);
2045 }
2046 
2047 /*
2048  * Called by readers of cc_state only, must call under rcu_read_lock().
2049  */
2050 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
2051 {
2052 	return rcu_dereference(ppd->cc_state);
2053 }
2054 
2055 /*
2056  * Called by writers of cc_state only,  must call under cc_state_lock.
2057  */
2058 static inline
2059 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
2060 {
2061 	return rcu_dereference_protected(ppd->cc_state,
2062 					 lockdep_is_held(&ppd->cc_state_lock));
2063 }
2064 
2065 /*
2066  * values for dd->flags (_device_ related flags)
2067  */
2068 #define HFI1_INITTED           0x1    /* chip and driver up and initted */
2069 #define HFI1_PRESENT           0x2    /* chip accesses can be done */
2070 #define HFI1_FROZEN            0x4    /* chip in SPC freeze */
2071 #define HFI1_HAS_SDMA_TIMEOUT  0x8
2072 #define HFI1_HAS_SEND_DMA      0x10   /* Supports Send DMA */
2073 #define HFI1_FORCED_FREEZE     0x80   /* driver forced freeze mode */
2074 #define HFI1_SHUTDOWN          0x100  /* device is shutting down */
2075 
2076 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
2077 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
2078 
2079 /* ctxt_flag bit offsets */
2080 		/* base context has not finished initializing */
2081 #define HFI1_CTXT_BASE_UNINIT 1
2082 		/* base context initaliation failed */
2083 #define HFI1_CTXT_BASE_FAILED 2
2084 		/* waiting for a packet to arrive */
2085 #define HFI1_CTXT_WAITING_RCV 3
2086 		/* waiting for an urgent packet to arrive */
2087 #define HFI1_CTXT_WAITING_URG 4
2088 
2089 /* free up any allocated data at closes */
2090 int hfi1_init_dd(struct hfi1_devdata *dd);
2091 void hfi1_free_devdata(struct hfi1_devdata *dd);
2092 
2093 /* LED beaconing functions */
2094 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
2095 			     unsigned int timeoff);
2096 void shutdown_led_override(struct hfi1_pportdata *ppd);
2097 
2098 #define HFI1_CREDIT_RETURN_RATE (100)
2099 
2100 /*
2101  * The number of words for the KDETH protocol field.  If this is
2102  * larger then the actual field used, then part of the payload
2103  * will be in the header.
2104  *
2105  * Optimally, we want this sized so that a typical case will
2106  * use full cache lines.  The typical local KDETH header would
2107  * be:
2108  *
2109  *	Bytes	Field
2110  *	  8	LRH
2111  *	 12	BHT
2112  *	 ??	KDETH
2113  *	  8	RHF
2114  *	---
2115  *	 28 + KDETH
2116  *
2117  * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
2118  */
2119 #define DEFAULT_RCVHDRSIZE 9
2120 
2121 /*
2122  * Maximal header byte count:
2123  *
2124  *	Bytes	Field
2125  *	  8	LRH
2126  *	 40	GRH (optional)
2127  *	 12	BTH
2128  *	 ??	KDETH
2129  *	  8	RHF
2130  *	---
2131  *	 68 + KDETH
2132  *
2133  * We also want to maintain a cache line alignment to assist DMA'ing
2134  * of the header bytes.  Round up to a good size.
2135  */
2136 #define DEFAULT_RCVHDR_ENTSIZE 32
2137 
2138 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
2139 			u32 nlocked, u32 npages);
2140 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
2141 			    size_t npages, bool writable, struct page **pages);
2142 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
2143 			     size_t npages, bool dirty);
2144 
2145 /**
2146  * hfi1_rcvhdrtail_kvaddr - return tail kvaddr
2147  * @rcd - the receive context
2148  */
2149 static inline __le64 *hfi1_rcvhdrtail_kvaddr(const struct hfi1_ctxtdata *rcd)
2150 {
2151 	return (__le64 *)rcd->rcvhdrtail_kvaddr;
2152 }
2153 
2154 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2155 {
2156 	u64 *kv = (u64 *)hfi1_rcvhdrtail_kvaddr(rcd);
2157 
2158 	if (kv)
2159 		*kv = 0ULL;
2160 }
2161 
2162 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2163 {
2164 	/*
2165 	 * volatile because it's a DMA target from the chip, routine is
2166 	 * inlined, and don't want register caching or reordering.
2167 	 */
2168 	return (u32)le64_to_cpu(*hfi1_rcvhdrtail_kvaddr(rcd));
2169 }
2170 
2171 static inline bool hfi1_packet_present(struct hfi1_ctxtdata *rcd)
2172 {
2173 	if (likely(!rcd->rcvhdrtail_kvaddr)) {
2174 		u32 seq = rhf_rcv_seq(rhf_to_cpu(get_rhf_addr(rcd)));
2175 
2176 		return !last_rcv_seq(rcd, seq);
2177 	}
2178 	return hfi1_rcd_head(rcd) != get_rcvhdrtail(rcd);
2179 }
2180 
2181 /*
2182  * sysfs interface.
2183  */
2184 
2185 extern const char ib_hfi1_version[];
2186 extern const struct attribute_group ib_hfi1_attr_group;
2187 extern const struct attribute_group *hfi1_attr_port_groups[];
2188 
2189 int hfi1_device_create(struct hfi1_devdata *dd);
2190 void hfi1_device_remove(struct hfi1_devdata *dd);
2191 
2192 int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
2193 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
2194 /* Hook for sysfs read of QSFP */
2195 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
2196 
2197 int hfi1_pcie_init(struct hfi1_devdata *dd);
2198 void hfi1_pcie_cleanup(struct pci_dev *pdev);
2199 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
2200 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
2201 int pcie_speeds(struct hfi1_devdata *dd);
2202 int restore_pci_variables(struct hfi1_devdata *dd);
2203 int save_pci_variables(struct hfi1_devdata *dd);
2204 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
2205 void tune_pcie_caps(struct hfi1_devdata *dd);
2206 int parse_platform_config(struct hfi1_devdata *dd);
2207 int get_platform_config_field(struct hfi1_devdata *dd,
2208 			      enum platform_config_table_type_encoding
2209 			      table_type, int table_index, int field_index,
2210 			      u32 *data, u32 len);
2211 
2212 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
2213 
2214 /*
2215  * Flush write combining store buffers (if present) and perform a write
2216  * barrier.
2217  */
2218 static inline void flush_wc(void)
2219 {
2220 	asm volatile("sfence" : : : "memory");
2221 }
2222 
2223 void handle_eflags(struct hfi1_packet *packet);
2224 void seqfile_dump_rcd(struct seq_file *s, struct hfi1_ctxtdata *rcd);
2225 
2226 /* global module parameter variables */
2227 extern unsigned int hfi1_max_mtu;
2228 extern unsigned int hfi1_cu;
2229 extern unsigned int user_credit_return_threshold;
2230 extern int num_user_contexts;
2231 extern unsigned long n_krcvqs;
2232 extern uint krcvqs[];
2233 extern int krcvqsset;
2234 extern uint loopback;
2235 extern uint quick_linkup;
2236 extern uint rcv_intr_timeout;
2237 extern uint rcv_intr_count;
2238 extern uint rcv_intr_dynamic;
2239 extern ushort link_crc_mask;
2240 
2241 extern struct mutex hfi1_mutex;
2242 
2243 /* Number of seconds before our card status check...  */
2244 #define STATUS_TIMEOUT 60
2245 
2246 #define DRIVER_NAME		"hfi1"
2247 #define HFI1_USER_MINOR_BASE     0
2248 #define HFI1_TRACE_MINOR         127
2249 #define HFI1_NMINORS             255
2250 
2251 #define PCI_VENDOR_ID_INTEL 0x8086
2252 #define PCI_DEVICE_ID_INTEL0 0x24f0
2253 #define PCI_DEVICE_ID_INTEL1 0x24f1
2254 
2255 #define HFI1_PKT_USER_SC_INTEGRITY					    \
2256 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
2257 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
2258 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
2259 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
2260 
2261 #define HFI1_PKT_KERNEL_SC_INTEGRITY					    \
2262 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
2263 
2264 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
2265 						  u16 ctxt_type)
2266 {
2267 	u64 base_sc_integrity;
2268 
2269 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2270 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2271 		return 0;
2272 
2273 	base_sc_integrity =
2274 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2275 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
2276 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2277 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2278 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2279 #ifndef CONFIG_FAULT_INJECTION
2280 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
2281 #endif
2282 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2283 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2284 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2285 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
2286 	| SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2287 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2288 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
2289 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
2290 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
2291 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2292 
2293 	if (ctxt_type == SC_USER)
2294 		base_sc_integrity |=
2295 #ifndef CONFIG_FAULT_INJECTION
2296 			SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK |
2297 #endif
2298 			HFI1_PKT_USER_SC_INTEGRITY;
2299 	else if (ctxt_type != SC_KERNEL)
2300 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
2301 
2302 	/* turn on send-side job key checks if !A0 */
2303 	if (!is_ax(dd))
2304 		base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2305 
2306 	return base_sc_integrity;
2307 }
2308 
2309 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
2310 {
2311 	u64 base_sdma_integrity;
2312 
2313 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2314 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2315 		return 0;
2316 
2317 	base_sdma_integrity =
2318 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2319 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2320 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2321 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2322 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2323 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2324 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2325 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
2326 	| SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2327 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2328 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
2329 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
2330 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
2331 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2332 
2333 	if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
2334 		base_sdma_integrity |=
2335 		SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
2336 
2337 	/* turn on send-side job key checks if !A0 */
2338 	if (!is_ax(dd))
2339 		base_sdma_integrity |=
2340 			SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2341 
2342 	return base_sdma_integrity;
2343 }
2344 
2345 #define dd_dev_emerg(dd, fmt, ...) \
2346 	dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
2347 		  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2348 
2349 #define dd_dev_err(dd, fmt, ...) \
2350 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
2351 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2352 
2353 #define dd_dev_err_ratelimited(dd, fmt, ...) \
2354 	dev_err_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2355 			    rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2356 			    ##__VA_ARGS__)
2357 
2358 #define dd_dev_warn(dd, fmt, ...) \
2359 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
2360 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2361 
2362 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
2363 	dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2364 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2365 			     ##__VA_ARGS__)
2366 
2367 #define dd_dev_info(dd, fmt, ...) \
2368 	dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
2369 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2370 
2371 #define dd_dev_info_ratelimited(dd, fmt, ...) \
2372 	dev_info_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2373 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2374 			     ##__VA_ARGS__)
2375 
2376 #define dd_dev_dbg(dd, fmt, ...) \
2377 	dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
2378 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2379 
2380 #define hfi1_dev_porterr(dd, port, fmt, ...) \
2381 	dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
2382 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), (port), ##__VA_ARGS__)
2383 
2384 /*
2385  * this is used for formatting hw error messages...
2386  */
2387 struct hfi1_hwerror_msgs {
2388 	u64 mask;
2389 	const char *msg;
2390 	size_t sz;
2391 };
2392 
2393 /* in intr.c... */
2394 void hfi1_format_hwerrors(u64 hwerrs,
2395 			  const struct hfi1_hwerror_msgs *hwerrmsgs,
2396 			  size_t nhwerrmsgs, char *msg, size_t lmsg);
2397 
2398 #define USER_OPCODE_CHECK_VAL 0xC0
2399 #define USER_OPCODE_CHECK_MASK 0xC0
2400 #define OPCODE_CHECK_VAL_DISABLED 0x0
2401 #define OPCODE_CHECK_MASK_DISABLED 0x0
2402 
2403 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
2404 {
2405 	struct hfi1_pportdata *ppd;
2406 	int i;
2407 
2408 	dd->z_int_counter = get_all_cpu_total(dd->int_counter);
2409 	dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
2410 	dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
2411 
2412 	ppd = (struct hfi1_pportdata *)(dd + 1);
2413 	for (i = 0; i < dd->num_pports; i++, ppd++) {
2414 		ppd->ibport_data.rvp.z_rc_acks =
2415 			get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
2416 		ppd->ibport_data.rvp.z_rc_qacks =
2417 			get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
2418 	}
2419 }
2420 
2421 /* Control LED state */
2422 static inline void setextled(struct hfi1_devdata *dd, u32 on)
2423 {
2424 	if (on)
2425 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
2426 	else
2427 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
2428 }
2429 
2430 /* return the i2c resource given the target */
2431 static inline u32 i2c_target(u32 target)
2432 {
2433 	return target ? CR_I2C2 : CR_I2C1;
2434 }
2435 
2436 /* return the i2c chain chip resource that this HFI uses for QSFP */
2437 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2438 {
2439 	return i2c_target(dd->hfi1_id);
2440 }
2441 
2442 /* Is this device integrated or discrete? */
2443 static inline bool is_integrated(struct hfi1_devdata *dd)
2444 {
2445 	return dd->pcidev->device == PCI_DEVICE_ID_INTEL1;
2446 }
2447 
2448 /**
2449  * hfi1_need_drop - detect need for drop
2450  * @dd: - the device
2451  *
2452  * In some cases, the first packet needs to be dropped.
2453  *
2454  * Return true is the current packet needs to be dropped and false otherwise.
2455  */
2456 static inline bool hfi1_need_drop(struct hfi1_devdata *dd)
2457 {
2458 	if (unlikely(dd->do_drop &&
2459 		     atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
2460 		     DROP_PACKET_ON)) {
2461 		dd->do_drop = false;
2462 		return true;
2463 	}
2464 	return false;
2465 }
2466 
2467 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2468 
2469 #define DD_DEV_ENTRY(dd)       __string(dev, dev_name(&(dd)->pcidev->dev))
2470 #define DD_DEV_ASSIGN(dd)      __assign_str(dev, dev_name(&(dd)->pcidev->dev))
2471 
2472 static inline void hfi1_update_ah_attr(struct ib_device *ibdev,
2473 				       struct rdma_ah_attr *attr)
2474 {
2475 	struct hfi1_pportdata *ppd;
2476 	struct hfi1_ibport *ibp;
2477 	u32 dlid = rdma_ah_get_dlid(attr);
2478 
2479 	/*
2480 	 * Kernel clients may not have setup GRH information
2481 	 * Set that here.
2482 	 */
2483 	ibp = to_iport(ibdev, rdma_ah_get_port_num(attr));
2484 	ppd = ppd_from_ibp(ibp);
2485 	if ((((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) ||
2486 	      (ppd->lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))) &&
2487 	    (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)) &&
2488 	    (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2489 	    (!(rdma_ah_get_ah_flags(attr) & IB_AH_GRH))) ||
2490 	    (rdma_ah_get_make_grd(attr))) {
2491 		rdma_ah_set_ah_flags(attr, IB_AH_GRH);
2492 		rdma_ah_set_interface_id(attr, OPA_MAKE_ID(dlid));
2493 		rdma_ah_set_subnet_prefix(attr, ibp->rvp.gid_prefix);
2494 	}
2495 }
2496 
2497 /*
2498  * hfi1_check_mcast- Check if the given lid is
2499  * in the OPA multicast range.
2500  *
2501  * The LID might either reside in ah.dlid or might be
2502  * in the GRH of the address handle as DGID if extended
2503  * addresses are in use.
2504  */
2505 static inline bool hfi1_check_mcast(u32 lid)
2506 {
2507 	return ((lid >= opa_get_mcast_base(OPA_MCAST_NR)) &&
2508 		(lid != be32_to_cpu(OPA_LID_PERMISSIVE)));
2509 }
2510 
2511 #define opa_get_lid(lid, format)	\
2512 	__opa_get_lid(lid, OPA_PORT_PACKET_FORMAT_##format)
2513 
2514 /* Convert a lid to a specific lid space */
2515 static inline u32 __opa_get_lid(u32 lid, u8 format)
2516 {
2517 	bool is_mcast = hfi1_check_mcast(lid);
2518 
2519 	switch (format) {
2520 	case OPA_PORT_PACKET_FORMAT_8B:
2521 	case OPA_PORT_PACKET_FORMAT_10B:
2522 		if (is_mcast)
2523 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2524 				0xF0000);
2525 		return lid & 0xFFFFF;
2526 	case OPA_PORT_PACKET_FORMAT_16B:
2527 		if (is_mcast)
2528 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2529 				0xF00000);
2530 		return lid & 0xFFFFFF;
2531 	case OPA_PORT_PACKET_FORMAT_9B:
2532 		if (is_mcast)
2533 			return (lid -
2534 				opa_get_mcast_base(OPA_MCAST_NR) +
2535 				be16_to_cpu(IB_MULTICAST_LID_BASE));
2536 		else
2537 			return lid & 0xFFFF;
2538 	default:
2539 		return lid;
2540 	}
2541 }
2542 
2543 /* Return true if the given lid is the OPA 16B multicast range */
2544 static inline bool hfi1_is_16B_mcast(u32 lid)
2545 {
2546 	return ((lid >=
2547 		opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 16B)) &&
2548 		(lid != opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B)));
2549 }
2550 
2551 static inline void hfi1_make_opa_lid(struct rdma_ah_attr *attr)
2552 {
2553 	const struct ib_global_route *grh = rdma_ah_read_grh(attr);
2554 	u32 dlid = rdma_ah_get_dlid(attr);
2555 
2556 	/* Modify ah_attr.dlid to be in the 32 bit LID space.
2557 	 * This is how the address will be laid out:
2558 	 * Assuming MCAST_NR to be 4,
2559 	 * 32 bit permissive LID = 0xFFFFFFFF
2560 	 * Multicast LID range = 0xFFFFFFFE to 0xF0000000
2561 	 * Unicast LID range = 0xEFFFFFFF to 1
2562 	 * Invalid LID = 0
2563 	 */
2564 	if (ib_is_opa_gid(&grh->dgid))
2565 		dlid = opa_get_lid_from_gid(&grh->dgid);
2566 	else if ((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
2567 		 (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2568 		 (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)))
2569 		dlid = dlid - be16_to_cpu(IB_MULTICAST_LID_BASE) +
2570 			opa_get_mcast_base(OPA_MCAST_NR);
2571 	else if (dlid == be16_to_cpu(IB_LID_PERMISSIVE))
2572 		dlid = be32_to_cpu(OPA_LID_PERMISSIVE);
2573 
2574 	rdma_ah_set_dlid(attr, dlid);
2575 }
2576 
2577 static inline u8 hfi1_get_packet_type(u32 lid)
2578 {
2579 	/* 9B if lid > 0xF0000000 */
2580 	if (lid >= opa_get_mcast_base(OPA_MCAST_NR))
2581 		return HFI1_PKT_TYPE_9B;
2582 
2583 	/* 16B if lid > 0xC000 */
2584 	if (lid >= opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 9B))
2585 		return HFI1_PKT_TYPE_16B;
2586 
2587 	return HFI1_PKT_TYPE_9B;
2588 }
2589 
2590 static inline bool hfi1_get_hdr_type(u32 lid, struct rdma_ah_attr *attr)
2591 {
2592 	/*
2593 	 * If there was an incoming 16B packet with permissive
2594 	 * LIDs, OPA GIDs would have been programmed when those
2595 	 * packets were received. A 16B packet will have to
2596 	 * be sent in response to that packet. Return a 16B
2597 	 * header type if that's the case.
2598 	 */
2599 	if (rdma_ah_get_dlid(attr) == be32_to_cpu(OPA_LID_PERMISSIVE))
2600 		return (ib_is_opa_gid(&rdma_ah_read_grh(attr)->dgid)) ?
2601 			HFI1_PKT_TYPE_16B : HFI1_PKT_TYPE_9B;
2602 
2603 	/*
2604 	 * Return a 16B header type if either the the destination
2605 	 * or source lid is extended.
2606 	 */
2607 	if (hfi1_get_packet_type(rdma_ah_get_dlid(attr)) == HFI1_PKT_TYPE_16B)
2608 		return HFI1_PKT_TYPE_16B;
2609 
2610 	return hfi1_get_packet_type(lid);
2611 }
2612 
2613 static inline void hfi1_make_ext_grh(struct hfi1_packet *packet,
2614 				     struct ib_grh *grh, u32 slid,
2615 				     u32 dlid)
2616 {
2617 	struct hfi1_ibport *ibp = &packet->rcd->ppd->ibport_data;
2618 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2619 
2620 	if (!ibp)
2621 		return;
2622 
2623 	grh->hop_limit = 1;
2624 	grh->sgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2625 	if (slid == opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B))
2626 		grh->sgid.global.interface_id =
2627 			OPA_MAKE_ID(be32_to_cpu(OPA_LID_PERMISSIVE));
2628 	else
2629 		grh->sgid.global.interface_id = OPA_MAKE_ID(slid);
2630 
2631 	/*
2632 	 * Upper layers (like mad) may compare the dgid in the
2633 	 * wc that is obtained here with the sgid_index in
2634 	 * the wr. Since sgid_index in wr is always 0 for
2635 	 * extended lids, set the dgid here to the default
2636 	 * IB gid.
2637 	 */
2638 	grh->dgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2639 	grh->dgid.global.interface_id =
2640 		cpu_to_be64(ppd->guids[HFI1_PORT_GUID_INDEX]);
2641 }
2642 
2643 static inline int hfi1_get_16b_padding(u32 hdr_size, u32 payload)
2644 {
2645 	return -(hdr_size + payload + (SIZE_OF_CRC << 2) +
2646 		     SIZE_OF_LT) & 0x7;
2647 }
2648 
2649 static inline void hfi1_make_ib_hdr(struct ib_header *hdr,
2650 				    u16 lrh0, u16 len,
2651 				    u16 dlid, u16 slid)
2652 {
2653 	hdr->lrh[0] = cpu_to_be16(lrh0);
2654 	hdr->lrh[1] = cpu_to_be16(dlid);
2655 	hdr->lrh[2] = cpu_to_be16(len);
2656 	hdr->lrh[3] = cpu_to_be16(slid);
2657 }
2658 
2659 static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr,
2660 				     u32 slid, u32 dlid,
2661 				     u16 len, u16 pkey,
2662 				     bool becn, bool fecn, u8 l4,
2663 				     u8 sc)
2664 {
2665 	u32 lrh0 = 0;
2666 	u32 lrh1 = 0x40000000;
2667 	u32 lrh2 = 0;
2668 	u32 lrh3 = 0;
2669 
2670 	lrh0 = (lrh0 & ~OPA_16B_BECN_MASK) | (becn << OPA_16B_BECN_SHIFT);
2671 	lrh0 = (lrh0 & ~OPA_16B_LEN_MASK) | (len << OPA_16B_LEN_SHIFT);
2672 	lrh0 = (lrh0 & ~OPA_16B_LID_MASK)  | (slid & OPA_16B_LID_MASK);
2673 	lrh1 = (lrh1 & ~OPA_16B_FECN_MASK) | (fecn << OPA_16B_FECN_SHIFT);
2674 	lrh1 = (lrh1 & ~OPA_16B_SC_MASK) | (sc << OPA_16B_SC_SHIFT);
2675 	lrh1 = (lrh1 & ~OPA_16B_LID_MASK) | (dlid & OPA_16B_LID_MASK);
2676 	lrh2 = (lrh2 & ~OPA_16B_SLID_MASK) |
2677 		((slid >> OPA_16B_SLID_SHIFT) << OPA_16B_SLID_HIGH_SHIFT);
2678 	lrh2 = (lrh2 & ~OPA_16B_DLID_MASK) |
2679 		((dlid >> OPA_16B_DLID_SHIFT) << OPA_16B_DLID_HIGH_SHIFT);
2680 	lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | ((u32)pkey << OPA_16B_PKEY_SHIFT);
2681 	lrh2 = (lrh2 & ~OPA_16B_L4_MASK) | l4;
2682 
2683 	hdr->lrh[0] = lrh0;
2684 	hdr->lrh[1] = lrh1;
2685 	hdr->lrh[2] = lrh2;
2686 	hdr->lrh[3] = lrh3;
2687 }
2688 #endif                          /* _HFI1_KERNEL_H */
2689