xref: /linux/drivers/infiniband/hw/hfi1/hfi.h (revision 3f41368fbfe1b3d5922d317fe1a0a0cab6846802)
1 /* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */
2 /*
3  * Copyright(c) 2020-2023 Cornelis Networks, Inc.
4  * Copyright(c) 2015-2020 Intel Corporation.
5  */
6 
7 #ifndef _HFI1_KERNEL_H
8 #define _HFI1_KERNEL_H
9 
10 #include <linux/refcount.h>
11 #include <linux/interrupt.h>
12 #include <linux/pci.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/mutex.h>
15 #include <linux/list.h>
16 #include <linux/scatterlist.h>
17 #include <linux/slab.h>
18 #include <linux/io.h>
19 #include <linux/fs.h>
20 #include <linux/completion.h>
21 #include <linux/kref.h>
22 #include <linux/sched.h>
23 #include <linux/cdev.h>
24 #include <linux/delay.h>
25 #include <linux/kthread.h>
26 #include <linux/i2c.h>
27 #include <linux/i2c-algo-bit.h>
28 #include <linux/xarray.h>
29 #include <rdma/ib_hdrs.h>
30 #include <rdma/opa_addr.h>
31 #include <linux/rhashtable.h>
32 #include <rdma/rdma_vt.h>
33 
34 #include "chip_registers.h"
35 #include "common.h"
36 #include "opfn.h"
37 #include "verbs.h"
38 #include "pio.h"
39 #include "chip.h"
40 #include "mad.h"
41 #include "qsfp.h"
42 #include "platform.h"
43 #include "affinity.h"
44 #include "msix.h"
45 
46 /* bumped 1 from s/w major version of TrueScale */
47 #define HFI1_CHIP_VERS_MAJ 3U
48 
49 /* don't care about this except printing */
50 #define HFI1_CHIP_VERS_MIN 0U
51 
52 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
53 #define HFI1_OUI 0x001175
54 #define HFI1_OUI_LSB 40
55 
56 #define DROP_PACKET_OFF		0
57 #define DROP_PACKET_ON		1
58 
59 #define NEIGHBOR_TYPE_HFI		0
60 #define NEIGHBOR_TYPE_SWITCH	1
61 
62 #define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5
63 
64 extern unsigned long hfi1_cap_mask;
65 #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap)
66 #define HFI1_CAP_UGET_MASK(mask, cap) \
67 	(((mask) >> HFI1_CAP_USER_SHIFT) & HFI1_CAP_##cap)
68 #define HFI1_CAP_KGET(cap) (HFI1_CAP_KGET_MASK(hfi1_cap_mask, cap))
69 #define HFI1_CAP_UGET(cap) (HFI1_CAP_UGET_MASK(hfi1_cap_mask, cap))
70 #define HFI1_CAP_IS_KSET(cap) (!!HFI1_CAP_KGET(cap))
71 #define HFI1_CAP_IS_USET(cap) (!!HFI1_CAP_UGET(cap))
72 #define HFI1_MISC_GET() ((hfi1_cap_mask >> HFI1_CAP_MISC_SHIFT) & \
73 			HFI1_CAP_MISC_MASK)
74 /* Offline Disabled Reason is 4-bits */
75 #define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
76 
77 /*
78  * Control context is always 0 and handles the error packets.
79  * It also handles the VL15 and multicast packets.
80  */
81 #define HFI1_CTRL_CTXT    0
82 
83 /*
84  * Driver context will store software counters for each of the events
85  * associated with these status registers
86  */
87 #define NUM_CCE_ERR_STATUS_COUNTERS 41
88 #define NUM_RCV_ERR_STATUS_COUNTERS 64
89 #define NUM_MISC_ERR_STATUS_COUNTERS 13
90 #define NUM_SEND_PIO_ERR_STATUS_COUNTERS 36
91 #define NUM_SEND_DMA_ERR_STATUS_COUNTERS 4
92 #define NUM_SEND_EGRESS_ERR_STATUS_COUNTERS 64
93 #define NUM_SEND_ERR_STATUS_COUNTERS 3
94 #define NUM_SEND_CTXT_ERR_STATUS_COUNTERS 5
95 #define NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS 24
96 
97 /*
98  * per driver stats, either not device nor port-specific, or
99  * summed over all of the devices and ports.
100  * They are described by name via ipathfs filesystem, so layout
101  * and number of elements can change without breaking compatibility.
102  * If members are added or deleted hfi1_statnames[] in debugfs.c must
103  * change to match.
104  */
105 struct hfi1_ib_stats {
106 	__u64 sps_ints; /* number of interrupts handled */
107 	__u64 sps_errints; /* number of error interrupts */
108 	__u64 sps_txerrs; /* tx-related packet errors */
109 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
110 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
111 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
112 	__u64 sps_ctxts; /* number of contexts currently open */
113 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
114 	__u64 sps_buffull;
115 	__u64 sps_hdrfull;
116 };
117 
118 extern struct hfi1_ib_stats hfi1_stats;
119 extern const struct pci_error_handlers hfi1_pci_err_handler;
120 
121 extern int num_driver_cntrs;
122 
123 /*
124  * First-cut criterion for "device is active" is
125  * two thousand dwords combined Tx, Rx traffic per
126  * 5-second interval. SMA packets are 64 dwords,
127  * and occur "a few per second", presumably each way.
128  */
129 #define HFI1_TRAFFIC_ACTIVE_THRESHOLD (2000)
130 
131 /*
132  * Below contains all data related to a single context (formerly called port).
133  */
134 
135 struct hfi1_opcode_stats_perctx;
136 
137 struct ctxt_eager_bufs {
138 	struct eager_buffer {
139 		void *addr;
140 		dma_addr_t dma;
141 		ssize_t len;
142 	} *buffers;
143 	struct {
144 		void *addr;
145 		dma_addr_t dma;
146 	} *rcvtids;
147 	u32 size;                /* total size of eager buffers */
148 	u32 rcvtid_size;         /* size of each eager rcv tid */
149 	u16 count;               /* size of buffers array */
150 	u16 numbufs;             /* number of buffers allocated */
151 	u16 alloced;             /* number of rcvarray entries used */
152 	u16 threshold;           /* head update threshold */
153 };
154 
155 struct exp_tid_set {
156 	struct list_head list;
157 	u32 count;
158 };
159 
160 struct hfi1_ctxtdata;
161 typedef int (*intr_handler)(struct hfi1_ctxtdata *rcd, int data);
162 typedef void (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
163 
164 struct tid_queue {
165 	struct list_head queue_head;
166 			/* queue head for QP TID resource waiters */
167 	u32 enqueue;	/* count of tid enqueues */
168 	u32 dequeue;	/* count of tid dequeues */
169 };
170 
171 struct hfi1_ctxtdata {
172 	/* rcvhdrq base, needs mmap before useful */
173 	void *rcvhdrq;
174 	/* kernel virtual address where hdrqtail is updated */
175 	volatile __le64 *rcvhdrtail_kvaddr;
176 	/* so functions that need physical port can get it easily */
177 	struct hfi1_pportdata *ppd;
178 	/* so file ops can get at unit */
179 	struct hfi1_devdata *dd;
180 	/* this receive context's assigned PIO ACK send context */
181 	struct send_context *sc;
182 	/* per context recv functions */
183 	const rhf_rcv_function_ptr *rhf_rcv_function_map;
184 	/*
185 	 * The interrupt handler for a particular receive context can vary
186 	 * throughout it's lifetime. This is not a lock protected data member so
187 	 * it must be updated atomically and the prev and new value must always
188 	 * be valid. Worst case is we process an extra interrupt and up to 64
189 	 * packets with the wrong interrupt handler.
190 	 */
191 	intr_handler do_interrupt;
192 	/** fast handler after autoactive */
193 	intr_handler fast_handler;
194 	/** slow handler */
195 	intr_handler slow_handler;
196 	/* napi pointer assiociated with netdev */
197 	struct napi_struct *napi;
198 	/* verbs rx_stats per rcd */
199 	struct hfi1_opcode_stats_perctx *opstats;
200 	/* clear interrupt mask */
201 	u64 imask;
202 	/* ctxt rcvhdrq head offset */
203 	u32 head;
204 	/* number of rcvhdrq entries */
205 	u16 rcvhdrq_cnt;
206 	u8 ireg;	/* clear interrupt register */
207 	/* receive packet sequence counter */
208 	u8 seq_cnt;
209 	/* size of each of the rcvhdrq entries */
210 	u8 rcvhdrqentsize;
211 	/* offset of RHF within receive header entry */
212 	u8 rhf_offset;
213 	/* dynamic receive available interrupt timeout */
214 	u8 rcvavail_timeout;
215 	/* Indicates that this is vnic context */
216 	bool is_vnic;
217 	/* vnic queue index this context is mapped to */
218 	u8 vnic_q_idx;
219 	/* Is ASPM interrupt supported for this context */
220 	bool aspm_intr_supported;
221 	/* ASPM state (enabled/disabled) for this context */
222 	bool aspm_enabled;
223 	/* Is ASPM processing enabled for this context (in intr context) */
224 	bool aspm_intr_enable;
225 	struct ctxt_eager_bufs egrbufs;
226 	/* QPs waiting for context processing */
227 	struct list_head qp_wait_list;
228 	/* tid allocation lists */
229 	struct exp_tid_set tid_group_list;
230 	struct exp_tid_set tid_used_list;
231 	struct exp_tid_set tid_full_list;
232 
233 	/* Timer for re-enabling ASPM if interrupt activity quiets down */
234 	struct timer_list aspm_timer;
235 	/* per-context configuration flags */
236 	unsigned long flags;
237 	/* array of tid_groups */
238 	struct tid_group  *groups;
239 	/* mmap of hdrq, must fit in 44 bits */
240 	dma_addr_t rcvhdrq_dma;
241 	dma_addr_t rcvhdrqtailaddr_dma;
242 	/* Last interrupt timestamp */
243 	ktime_t aspm_ts_last_intr;
244 	/* Last timestamp at which we scheduled a timer for this context */
245 	ktime_t aspm_ts_timer_sched;
246 	/* Lock to serialize between intr, timer intr and user threads */
247 	spinlock_t aspm_lock;
248 	/* Reference count the base context usage */
249 	struct kref kref;
250 	/* numa node of this context */
251 	int numa_id;
252 	/* associated msix interrupt. */
253 	s16 msix_intr;
254 	/* job key */
255 	u16 jkey;
256 	/* number of RcvArray groups for this context. */
257 	u16 rcv_array_groups;
258 	/* index of first eager TID entry. */
259 	u16 eager_base;
260 	/* number of expected TID entries */
261 	u16 expected_count;
262 	/* index of first expected TID entry. */
263 	u16 expected_base;
264 	/* Device context index */
265 	u8 ctxt;
266 
267 	/* PSM Specific fields */
268 	/* lock protecting all Expected TID data */
269 	struct mutex exp_mutex;
270 	/* lock protecting all Expected TID data of kernel contexts */
271 	spinlock_t exp_lock;
272 	/* Queue for QP's waiting for HW TID flows */
273 	struct tid_queue flow_queue;
274 	/* Queue for QP's waiting for HW receive array entries */
275 	struct tid_queue rarr_queue;
276 	/* when waiting for rcv or pioavail */
277 	wait_queue_head_t wait;
278 	/* uuid from PSM */
279 	u8 uuid[16];
280 	/* same size as task_struct .comm[], command that opened context */
281 	char comm[TASK_COMM_LEN];
282 	/* Bitmask of in use context(s) */
283 	DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
284 	/* per-context event flags for fileops/intr communication */
285 	unsigned long event_flags;
286 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
287 	void *subctxt_uregbase;
288 	/* An array of pages for the eager receive buffers * N */
289 	void *subctxt_rcvegrbuf;
290 	/* An array of pages for the eager header queue entries * N */
291 	void *subctxt_rcvhdr_base;
292 	/* total number of polled urgent packets */
293 	u32 urgent;
294 	/* saved total number of polled urgent packets for poll edge trigger */
295 	u32 urgent_poll;
296 	/* Type of packets or conditions we want to poll for */
297 	u16 poll_type;
298 	/* non-zero if ctxt is being shared. */
299 	u16 subctxt_id;
300 	/* The version of the library which opened this ctxt */
301 	u32 userversion;
302 	/*
303 	 * non-zero if ctxt can be shared, and defines the maximum number of
304 	 * sub-contexts for this device context.
305 	 */
306 	u8 subctxt_cnt;
307 
308 	/* Bit mask to track free TID RDMA HW flows */
309 	unsigned long flow_mask;
310 	struct tid_flow_state flows[RXE_NUM_TID_FLOWS];
311 };
312 
313 /**
314  * rcvhdrq_size - return total size in bytes for header queue
315  * @rcd: the receive context
316  *
317  * rcvhdrqentsize is in DWs, so we have to convert to bytes
318  *
319  */
320 static inline u32 rcvhdrq_size(struct hfi1_ctxtdata *rcd)
321 {
322 	return PAGE_ALIGN(rcd->rcvhdrq_cnt *
323 			  rcd->rcvhdrqentsize * sizeof(u32));
324 }
325 
326 /*
327  * Represents a single packet at a high level. Put commonly computed things in
328  * here so we do not have to keep doing them over and over. The rule of thumb is
329  * if something is used one time to derive some value, store that something in
330  * here. If it is used multiple times, then store the result of that derivation
331  * in here.
332  */
333 struct hfi1_packet {
334 	void *ebuf;
335 	void *hdr;
336 	void *payload;
337 	struct hfi1_ctxtdata *rcd;
338 	__le32 *rhf_addr;
339 	struct rvt_qp *qp;
340 	struct ib_other_headers *ohdr;
341 	struct ib_grh *grh;
342 	struct opa_16b_mgmt *mgmt;
343 	u64 rhf;
344 	u32 maxcnt;
345 	u32 rhqoff;
346 	u32 dlid;
347 	u32 slid;
348 	int numpkt;
349 	u16 tlen;
350 	s16 etail;
351 	u16 pkey;
352 	u8 hlen;
353 	u8 rsize;
354 	u8 updegr;
355 	u8 etype;
356 	u8 extra_byte;
357 	u8 pad;
358 	u8 sc;
359 	u8 sl;
360 	u8 opcode;
361 	bool migrated;
362 };
363 
364 /* Packet types */
365 #define HFI1_PKT_TYPE_9B  0
366 #define HFI1_PKT_TYPE_16B 1
367 
368 /*
369  * OPA 16B Header
370  */
371 #define OPA_16B_L4_MASK		0xFFull
372 #define OPA_16B_SC_MASK		0x1F00000ull
373 #define OPA_16B_SC_SHIFT	20
374 #define OPA_16B_LID_MASK	0xFFFFFull
375 #define OPA_16B_DLID_MASK	0xF000ull
376 #define OPA_16B_DLID_SHIFT	20
377 #define OPA_16B_DLID_HIGH_SHIFT	12
378 #define OPA_16B_SLID_MASK	0xF00ull
379 #define OPA_16B_SLID_SHIFT	20
380 #define OPA_16B_SLID_HIGH_SHIFT	8
381 #define OPA_16B_BECN_MASK       0x80000000ull
382 #define OPA_16B_BECN_SHIFT      31
383 #define OPA_16B_FECN_MASK       0x10000000ull
384 #define OPA_16B_FECN_SHIFT      28
385 #define OPA_16B_L2_MASK		0x60000000ull
386 #define OPA_16B_L2_SHIFT	29
387 #define OPA_16B_PKEY_MASK	0xFFFF0000ull
388 #define OPA_16B_PKEY_SHIFT	16
389 #define OPA_16B_LEN_MASK	0x7FF00000ull
390 #define OPA_16B_LEN_SHIFT	20
391 #define OPA_16B_RC_MASK		0xE000000ull
392 #define OPA_16B_RC_SHIFT	25
393 #define OPA_16B_AGE_MASK	0xFF0000ull
394 #define OPA_16B_AGE_SHIFT	16
395 #define OPA_16B_ENTROPY_MASK	0xFFFFull
396 
397 /*
398  * OPA 16B L2/L4 Encodings
399  */
400 #define OPA_16B_L4_9B		0x00
401 #define OPA_16B_L2_TYPE		0x02
402 #define OPA_16B_L4_FM		0x08
403 #define OPA_16B_L4_IB_LOCAL	0x09
404 #define OPA_16B_L4_IB_GLOBAL	0x0A
405 #define OPA_16B_L4_ETHR		OPA_VNIC_L4_ETHR
406 
407 /*
408  * OPA 16B Management
409  */
410 #define OPA_16B_L4_FM_PAD	3  /* fixed 3B pad */
411 #define OPA_16B_L4_FM_HLEN	24 /* 16B(16) + L4_FM(8) */
412 
413 static inline u8 hfi1_16B_get_l4(struct hfi1_16b_header *hdr)
414 {
415 	return (u8)(hdr->lrh[2] & OPA_16B_L4_MASK);
416 }
417 
418 static inline u8 hfi1_16B_get_sc(struct hfi1_16b_header *hdr)
419 {
420 	return (u8)((hdr->lrh[1] & OPA_16B_SC_MASK) >> OPA_16B_SC_SHIFT);
421 }
422 
423 static inline u32 hfi1_16B_get_dlid(struct hfi1_16b_header *hdr)
424 {
425 	return (u32)((hdr->lrh[1] & OPA_16B_LID_MASK) |
426 		     (((hdr->lrh[2] & OPA_16B_DLID_MASK) >>
427 		     OPA_16B_DLID_HIGH_SHIFT) << OPA_16B_DLID_SHIFT));
428 }
429 
430 static inline u32 hfi1_16B_get_slid(struct hfi1_16b_header *hdr)
431 {
432 	return (u32)((hdr->lrh[0] & OPA_16B_LID_MASK) |
433 		     (((hdr->lrh[2] & OPA_16B_SLID_MASK) >>
434 		     OPA_16B_SLID_HIGH_SHIFT) << OPA_16B_SLID_SHIFT));
435 }
436 
437 static inline u8 hfi1_16B_get_becn(struct hfi1_16b_header *hdr)
438 {
439 	return (u8)((hdr->lrh[0] & OPA_16B_BECN_MASK) >> OPA_16B_BECN_SHIFT);
440 }
441 
442 static inline u8 hfi1_16B_get_fecn(struct hfi1_16b_header *hdr)
443 {
444 	return (u8)((hdr->lrh[1] & OPA_16B_FECN_MASK) >> OPA_16B_FECN_SHIFT);
445 }
446 
447 static inline u8 hfi1_16B_get_l2(struct hfi1_16b_header *hdr)
448 {
449 	return (u8)((hdr->lrh[1] & OPA_16B_L2_MASK) >> OPA_16B_L2_SHIFT);
450 }
451 
452 static inline u16 hfi1_16B_get_pkey(struct hfi1_16b_header *hdr)
453 {
454 	return (u16)((hdr->lrh[2] & OPA_16B_PKEY_MASK) >> OPA_16B_PKEY_SHIFT);
455 }
456 
457 static inline u8 hfi1_16B_get_rc(struct hfi1_16b_header *hdr)
458 {
459 	return (u8)((hdr->lrh[1] & OPA_16B_RC_MASK) >> OPA_16B_RC_SHIFT);
460 }
461 
462 static inline u8 hfi1_16B_get_age(struct hfi1_16b_header *hdr)
463 {
464 	return (u8)((hdr->lrh[3] & OPA_16B_AGE_MASK) >> OPA_16B_AGE_SHIFT);
465 }
466 
467 static inline u16 hfi1_16B_get_len(struct hfi1_16b_header *hdr)
468 {
469 	return (u16)((hdr->lrh[0] & OPA_16B_LEN_MASK) >> OPA_16B_LEN_SHIFT);
470 }
471 
472 static inline u16 hfi1_16B_get_entropy(struct hfi1_16b_header *hdr)
473 {
474 	return (u16)(hdr->lrh[3] & OPA_16B_ENTROPY_MASK);
475 }
476 
477 #define OPA_16B_MAKE_QW(low_dw, high_dw) (((u64)(high_dw) << 32) | (low_dw))
478 
479 /*
480  * BTH
481  */
482 #define OPA_16B_BTH_PAD_MASK	7
483 static inline u8 hfi1_16B_bth_get_pad(struct ib_other_headers *ohdr)
484 {
485 	return (u8)((be32_to_cpu(ohdr->bth[0]) >> IB_BTH_PAD_SHIFT) &
486 		   OPA_16B_BTH_PAD_MASK);
487 }
488 
489 /*
490  * 16B Management
491  */
492 #define OPA_16B_MGMT_QPN_MASK	0xFFFFFF
493 static inline u32 hfi1_16B_get_dest_qpn(struct opa_16b_mgmt *mgmt)
494 {
495 	return be32_to_cpu(mgmt->dest_qpn) & OPA_16B_MGMT_QPN_MASK;
496 }
497 
498 static inline u32 hfi1_16B_get_src_qpn(struct opa_16b_mgmt *mgmt)
499 {
500 	return be32_to_cpu(mgmt->src_qpn) & OPA_16B_MGMT_QPN_MASK;
501 }
502 
503 static inline void hfi1_16B_set_qpn(struct opa_16b_mgmt *mgmt,
504 				    u32 dest_qp, u32 src_qp)
505 {
506 	mgmt->dest_qpn = cpu_to_be32(dest_qp & OPA_16B_MGMT_QPN_MASK);
507 	mgmt->src_qpn = cpu_to_be32(src_qp & OPA_16B_MGMT_QPN_MASK);
508 }
509 
510 /**
511  * hfi1_get_rc_ohdr - get extended header
512  * @opah - the opaheader
513  */
514 static inline struct ib_other_headers *
515 hfi1_get_rc_ohdr(struct hfi1_opa_header *opah)
516 {
517 	struct ib_other_headers *ohdr;
518 	struct ib_header *hdr = NULL;
519 	struct hfi1_16b_header *hdr_16b = NULL;
520 
521 	/* Find out where the BTH is */
522 	if (opah->hdr_type == HFI1_PKT_TYPE_9B) {
523 		hdr = &opah->ibh;
524 		if (ib_get_lnh(hdr) == HFI1_LRH_BTH)
525 			ohdr = &hdr->u.oth;
526 		else
527 			ohdr = &hdr->u.l.oth;
528 	} else {
529 		u8 l4;
530 
531 		hdr_16b = &opah->opah;
532 		l4  = hfi1_16B_get_l4(hdr_16b);
533 		if (l4 == OPA_16B_L4_IB_LOCAL)
534 			ohdr = &hdr_16b->u.oth;
535 		else
536 			ohdr = &hdr_16b->u.l.oth;
537 	}
538 	return ohdr;
539 }
540 
541 struct rvt_sge_state;
542 
543 /*
544  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
545  * Mostly for MADs that set or query link parameters, also ipath
546  * config interfaces
547  */
548 #define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
549 #define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
550 #define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
551 #define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
552 #define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
553 #define HFI1_IB_CFG_SPD 5 /* current Link spd */
554 #define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
555 #define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
556 #define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
557 #define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
558 #define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
559 #define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
560 #define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
561 #define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
562 #define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
563 #define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
564 #define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
565 #define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
566 #define HFI1_IB_CFG_VL_HIGH_LIMIT 19
567 #define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
568 #define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
569 
570 /*
571  * HFI or Host Link States
572  *
573  * These describe the states the driver thinks the logical and physical
574  * states are in.  Used as an argument to set_link_state().  Implemented
575  * as bits for easy multi-state checking.  The actual state can only be
576  * one.
577  */
578 #define __HLS_UP_INIT_BP	0
579 #define __HLS_UP_ARMED_BP	1
580 #define __HLS_UP_ACTIVE_BP	2
581 #define __HLS_DN_DOWNDEF_BP	3	/* link down default */
582 #define __HLS_DN_POLL_BP	4
583 #define __HLS_DN_DISABLE_BP	5
584 #define __HLS_DN_OFFLINE_BP	6
585 #define __HLS_VERIFY_CAP_BP	7
586 #define __HLS_GOING_UP_BP	8
587 #define __HLS_GOING_OFFLINE_BP  9
588 #define __HLS_LINK_COOLDOWN_BP 10
589 
590 #define HLS_UP_INIT	  BIT(__HLS_UP_INIT_BP)
591 #define HLS_UP_ARMED	  BIT(__HLS_UP_ARMED_BP)
592 #define HLS_UP_ACTIVE	  BIT(__HLS_UP_ACTIVE_BP)
593 #define HLS_DN_DOWNDEF	  BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
594 #define HLS_DN_POLL	  BIT(__HLS_DN_POLL_BP)
595 #define HLS_DN_DISABLE	  BIT(__HLS_DN_DISABLE_BP)
596 #define HLS_DN_OFFLINE	  BIT(__HLS_DN_OFFLINE_BP)
597 #define HLS_VERIFY_CAP	  BIT(__HLS_VERIFY_CAP_BP)
598 #define HLS_GOING_UP	  BIT(__HLS_GOING_UP_BP)
599 #define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
600 #define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
601 
602 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
603 #define HLS_DOWN ~(HLS_UP)
604 
605 #define HLS_DEFAULT HLS_DN_POLL
606 
607 /* use this MTU size if none other is given */
608 #define HFI1_DEFAULT_ACTIVE_MTU 10240
609 /* use this MTU size as the default maximum */
610 #define HFI1_DEFAULT_MAX_MTU 10240
611 /* default partition key */
612 #define DEFAULT_PKEY 0xffff
613 
614 /*
615  * Possible fabric manager config parameters for fm_{get,set}_table()
616  */
617 #define FM_TBL_VL_HIGH_ARB		1 /* Get/set VL high prio weights */
618 #define FM_TBL_VL_LOW_ARB		2 /* Get/set VL low prio weights */
619 #define FM_TBL_BUFFER_CONTROL		3 /* Get/set Buffer Control */
620 #define FM_TBL_SC2VLNT			4 /* Get/set SC->VLnt */
621 #define FM_TBL_VL_PREEMPT_ELEMS		5 /* Get (no set) VL preempt elems */
622 #define FM_TBL_VL_PREEMPT_MATRIX	6 /* Get (no set) VL preempt matrix */
623 
624 /*
625  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
626  * these are bits so they can be combined, e.g.
627  * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
628  */
629 #define HFI1_RCVCTRL_TAILUPD_ENB 0x01
630 #define HFI1_RCVCTRL_TAILUPD_DIS 0x02
631 #define HFI1_RCVCTRL_CTXT_ENB 0x04
632 #define HFI1_RCVCTRL_CTXT_DIS 0x08
633 #define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
634 #define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
635 #define HFI1_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
636 #define HFI1_RCVCTRL_PKEY_DIS 0x80
637 #define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
638 #define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
639 #define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
640 #define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
641 #define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
642 #define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
643 #define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
644 #define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
645 #define HFI1_RCVCTRL_URGENT_ENB 0x40000
646 #define HFI1_RCVCTRL_URGENT_DIS 0x80000
647 
648 /* partition enforcement flags */
649 #define HFI1_PART_ENFORCE_IN	0x1
650 #define HFI1_PART_ENFORCE_OUT	0x2
651 
652 /* how often we check for synthetic counter wrap around */
653 #define SYNTH_CNT_TIME 3
654 
655 /* Counter flags */
656 #define CNTR_NORMAL		0x0 /* Normal counters, just read register */
657 #define CNTR_SYNTH		0x1 /* Synthetic counters, saturate at all 1s */
658 #define CNTR_DISABLED		0x2 /* Disable this counter */
659 #define CNTR_32BIT		0x4 /* Simulate 64 bits for this counter */
660 #define CNTR_VL			0x8 /* Per VL counter */
661 #define CNTR_SDMA              0x10
662 #define CNTR_INVALID_VL		-1  /* Specifies invalid VL */
663 #define CNTR_MODE_W		0x0
664 #define CNTR_MODE_R		0x1
665 
666 /* VLs Supported/Operational */
667 #define HFI1_MIN_VLS_SUPPORTED 1
668 #define HFI1_MAX_VLS_SUPPORTED 8
669 
670 #define HFI1_GUIDS_PER_PORT  5
671 #define HFI1_PORT_GUID_INDEX 0
672 
673 static inline void incr_cntr64(u64 *cntr)
674 {
675 	if (*cntr < (u64)-1LL)
676 		(*cntr)++;
677 }
678 
679 #define MAX_NAME_SIZE 64
680 struct hfi1_msix_entry {
681 	enum irq_type type;
682 	int irq;
683 	void *arg;
684 	cpumask_t mask;
685 	struct irq_affinity_notify notify;
686 };
687 
688 struct hfi1_msix_info {
689 	/* lock to synchronize in_use_msix access */
690 	spinlock_t msix_lock;
691 	DECLARE_BITMAP(in_use_msix, CCE_NUM_MSIX_VECTORS);
692 	struct hfi1_msix_entry *msix_entries;
693 	u16 max_requested;
694 };
695 
696 /* per-SL CCA information */
697 struct cca_timer {
698 	struct hrtimer hrtimer;
699 	struct hfi1_pportdata *ppd; /* read-only */
700 	int sl; /* read-only */
701 	u16 ccti; /* read/write - current value of CCTI */
702 };
703 
704 struct link_down_reason {
705 	/*
706 	 * SMA-facing value.  Should be set from .latest when
707 	 * HLS_UP_* -> HLS_DN_* transition actually occurs.
708 	 */
709 	u8 sma;
710 	u8 latest;
711 };
712 
713 enum {
714 	LO_PRIO_TABLE,
715 	HI_PRIO_TABLE,
716 	MAX_PRIO_TABLE
717 };
718 
719 struct vl_arb_cache {
720 	/* protect vl arb cache */
721 	spinlock_t lock;
722 	struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
723 };
724 
725 /*
726  * The structure below encapsulates data relevant to a physical IB Port.
727  * Current chips support only one such port, but the separation
728  * clarifies things a bit. Note that to conform to IB conventions,
729  * port-numbers are one-based. The first or only port is port1.
730  */
731 struct hfi1_pportdata {
732 	struct hfi1_ibport ibport_data;
733 
734 	struct hfi1_devdata *dd;
735 
736 	/* PHY support */
737 	struct qsfp_data qsfp_info;
738 	/* Values for SI tuning of SerDes */
739 	u32 port_type;
740 	u32 tx_preset_eq;
741 	u32 tx_preset_noeq;
742 	u32 rx_preset;
743 	u8  local_atten;
744 	u8  remote_atten;
745 	u8  default_atten;
746 	u8  max_power_class;
747 
748 	/* did we read platform config from scratch registers? */
749 	bool config_from_scratch;
750 
751 	/* GUIDs for this interface, in host order, guids[0] is a port guid */
752 	u64 guids[HFI1_GUIDS_PER_PORT];
753 
754 	/* GUID for peer interface, in host order */
755 	u64 neighbor_guid;
756 
757 	/* up or down physical link state */
758 	u32 linkup;
759 
760 	/*
761 	 * this address is mapped read-only into user processes so they can
762 	 * get status cheaply, whenever they want.  One qword of status per port
763 	 */
764 	u64 *statusp;
765 
766 	/* SendDMA related entries */
767 
768 	struct workqueue_struct *hfi1_wq;
769 	struct workqueue_struct *link_wq;
770 
771 	/* move out of interrupt context */
772 	struct work_struct link_vc_work;
773 	struct work_struct link_up_work;
774 	struct work_struct link_down_work;
775 	struct work_struct sma_message_work;
776 	struct work_struct freeze_work;
777 	struct work_struct link_downgrade_work;
778 	struct work_struct link_bounce_work;
779 	struct delayed_work start_link_work;
780 	/* host link state variables */
781 	struct mutex hls_lock;
782 	u32 host_link_state;
783 
784 	/* these are the "32 bit" regs */
785 
786 	u32 ibmtu; /* The MTU programmed for this unit */
787 	/*
788 	 * Current max size IB packet (in bytes) including IB headers, that
789 	 * we can send. Changes when ibmtu changes.
790 	 */
791 	u32 ibmaxlen;
792 	u32 current_egress_rate; /* units [10^6 bits/sec] */
793 	/* LID programmed for this instance */
794 	u32 lid;
795 	/* list of pkeys programmed; 0 if not set */
796 	u16 pkeys[MAX_PKEY_VALUES];
797 	u16 link_width_supported;
798 	u16 link_width_downgrade_supported;
799 	u16 link_speed_supported;
800 	u16 link_width_enabled;
801 	u16 link_width_downgrade_enabled;
802 	u16 link_speed_enabled;
803 	u16 link_width_active;
804 	u16 link_width_downgrade_tx_active;
805 	u16 link_width_downgrade_rx_active;
806 	u16 link_speed_active;
807 	u8 vls_supported;
808 	u8 vls_operational;
809 	u8 actual_vls_operational;
810 	/* LID mask control */
811 	u8 lmc;
812 	/* Rx Polarity inversion (compensate for ~tx on partner) */
813 	u8 rx_pol_inv;
814 
815 	u8 hw_pidx;     /* physical port index */
816 	u32 port;        /* IB port number and index into dd->pports - 1 */
817 	/* type of neighbor node */
818 	u8 neighbor_type;
819 	u8 neighbor_normal;
820 	u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
821 	u8 neighbor_port_number;
822 	u8 is_sm_config_started;
823 	u8 offline_disabled_reason;
824 	u8 is_active_optimize_enabled;
825 	u8 driver_link_ready;	/* driver ready for active link */
826 	u8 link_enabled;	/* link enabled? */
827 	u8 linkinit_reason;
828 	u8 local_tx_rate;	/* rate given to 8051 firmware */
829 	u8 qsfp_retry_count;
830 
831 	/* placeholders for IB MAD packet settings */
832 	u8 overrun_threshold;
833 	u8 phy_error_threshold;
834 	unsigned int is_link_down_queued;
835 
836 	/* Used to override LED behavior for things like maintenance beaconing*/
837 	/*
838 	 * Alternates per phase of blink
839 	 * [0] holds LED off duration, [1] holds LED on duration
840 	 */
841 	unsigned long led_override_vals[2];
842 	u8 led_override_phase; /* LSB picks from vals[] */
843 	atomic_t led_override_timer_active;
844 	/* Used to flash LEDs in override mode */
845 	struct timer_list led_override_timer;
846 
847 	u32 sm_trap_qp;
848 	u32 sa_qp;
849 
850 	/*
851 	 * cca_timer_lock protects access to the per-SL cca_timer
852 	 * structures (specifically the ccti member).
853 	 */
854 	spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
855 	struct cca_timer cca_timer[OPA_MAX_SLS];
856 
857 	/* List of congestion control table entries */
858 	struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
859 
860 	/* congestion entries, each entry corresponding to a SL */
861 	struct opa_congestion_setting_entry_shadow
862 		congestion_entries[OPA_MAX_SLS];
863 
864 	/*
865 	 * cc_state_lock protects (write) access to the per-port
866 	 * struct cc_state.
867 	 */
868 	spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
869 
870 	struct cc_state __rcu *cc_state;
871 
872 	/* Total number of congestion control table entries */
873 	u16 total_cct_entry;
874 
875 	/* Bit map identifying service level */
876 	u32 cc_sl_control_map;
877 
878 	/* CA's max number of 64 entry units in the congestion control table */
879 	u8 cc_max_table_entries;
880 
881 	/*
882 	 * begin congestion log related entries
883 	 * cc_log_lock protects all congestion log related data
884 	 */
885 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
886 	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
887 	u16 threshold_event_counter;
888 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
889 	int cc_log_idx; /* index for logging events */
890 	int cc_mad_idx; /* index for reporting events */
891 	/* end congestion log related entries */
892 
893 	struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
894 
895 	/* port relative counter buffer */
896 	u64 *cntrs;
897 	/* port relative synthetic counter buffer */
898 	u64 *scntrs;
899 	/* port_xmit_discards are synthesized from different egress errors */
900 	u64 port_xmit_discards;
901 	u64 port_xmit_discards_vl[C_VL_COUNT];
902 	u64 port_xmit_constraint_errors;
903 	u64 port_rcv_constraint_errors;
904 	/* count of 'link_err' interrupts from DC */
905 	u64 link_downed;
906 	/* number of times link retrained successfully */
907 	u64 link_up;
908 	/* number of times a link unknown frame was reported */
909 	u64 unknown_frame_count;
910 	/* port_ltp_crc_mode is returned in 'portinfo' MADs */
911 	u16 port_ltp_crc_mode;
912 	/* port_crc_mode_enabled is the crc we support */
913 	u8 port_crc_mode_enabled;
914 	/* mgmt_allowed is also returned in 'portinfo' MADs */
915 	u8 mgmt_allowed;
916 	u8 part_enforce; /* partition enforcement flags */
917 	struct link_down_reason local_link_down_reason;
918 	struct link_down_reason neigh_link_down_reason;
919 	/* Value to be sent to link peer on LinkDown .*/
920 	u8 remote_link_down_reason;
921 	/* Error events that will cause a port bounce. */
922 	u32 port_error_action;
923 	struct work_struct linkstate_active_work;
924 	/* Does this port need to prescan for FECNs */
925 	bool cc_prescan;
926 	/*
927 	 * Sample sendWaitCnt & sendWaitVlCnt during link transition
928 	 * and counter request.
929 	 */
930 	u64 port_vl_xmit_wait_last[C_VL_COUNT + 1];
931 	u16 prev_link_width;
932 	u64 vl_xmit_flit_cnt[C_VL_COUNT + 1];
933 };
934 
935 typedef void (*opcode_handler)(struct hfi1_packet *packet);
936 typedef void (*hfi1_make_req)(struct rvt_qp *qp,
937 			      struct hfi1_pkt_state *ps,
938 			      struct rvt_swqe *wqe);
939 extern const rhf_rcv_function_ptr normal_rhf_rcv_functions[];
940 extern const rhf_rcv_function_ptr netdev_rhf_rcv_functions[];
941 
942 /* return values for the RHF receive functions */
943 #define RHF_RCV_CONTINUE  0	/* keep going */
944 #define RHF_RCV_DONE	  1	/* stop, this packet processed */
945 #define RHF_RCV_REPROCESS 2	/* stop. retain this packet */
946 
947 struct rcv_array_data {
948 	u16 ngroups;
949 	u16 nctxt_extra;
950 	u8 group_size;
951 };
952 
953 struct per_vl_data {
954 	u16 mtu;
955 	struct send_context *sc;
956 };
957 
958 /* 16 to directly index */
959 #define PER_VL_SEND_CONTEXTS 16
960 
961 struct err_info_rcvport {
962 	u8 status_and_code;
963 	u64 packet_flit1;
964 	u64 packet_flit2;
965 };
966 
967 struct err_info_constraint {
968 	u8 status;
969 	u16 pkey;
970 	u32 slid;
971 };
972 
973 struct hfi1_temp {
974 	unsigned int curr;       /* current temperature */
975 	unsigned int lo_lim;     /* low temperature limit */
976 	unsigned int hi_lim;     /* high temperature limit */
977 	unsigned int crit_lim;   /* critical temperature limit */
978 	u8 triggers;      /* temperature triggers */
979 };
980 
981 struct hfi1_i2c_bus {
982 	struct hfi1_devdata *controlling_dd; /* current controlling device */
983 	struct i2c_adapter adapter;	/* bus details */
984 	struct i2c_algo_bit_data algo;	/* bus algorithm details */
985 	int num;			/* bus number, 0 or 1 */
986 };
987 
988 /* common data between shared ASIC HFIs */
989 struct hfi1_asic_data {
990 	struct hfi1_devdata *dds[2];	/* back pointers */
991 	struct mutex asic_resource_mutex;
992 	struct hfi1_i2c_bus *i2c_bus0;
993 	struct hfi1_i2c_bus *i2c_bus1;
994 };
995 
996 /* sizes for both the QP and RSM map tables */
997 #define NUM_MAP_ENTRIES	 256
998 #define NUM_MAP_REGS      32
999 
1000 /* Virtual NIC information */
1001 struct hfi1_vnic_data {
1002 	struct kmem_cache *txreq_cache;
1003 	u8 num_vports;
1004 };
1005 
1006 struct hfi1_vnic_vport_info;
1007 
1008 /* device data struct now contains only "general per-device" info.
1009  * fields related to a physical IB port are in a hfi1_pportdata struct.
1010  */
1011 struct sdma_engine;
1012 struct sdma_vl_map;
1013 
1014 #define BOARD_VERS_MAX 96 /* how long the version string can be */
1015 #define SERIAL_MAX 16 /* length of the serial number */
1016 
1017 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
1018 struct hfi1_netdev_rx;
1019 struct hfi1_devdata {
1020 	struct hfi1_ibdev verbs_dev;     /* must be first */
1021 	/* pointers to related structs for this device */
1022 	/* pci access data structure */
1023 	struct pci_dev *pcidev;
1024 	struct cdev user_cdev;
1025 	struct cdev diag_cdev;
1026 	struct cdev ui_cdev;
1027 	struct device *user_device;
1028 	struct device *diag_device;
1029 	struct device *ui_device;
1030 
1031 	/* first mapping up to RcvArray */
1032 	u8 __iomem *kregbase1;
1033 	resource_size_t physaddr;
1034 
1035 	/* second uncached mapping from RcvArray to pio send buffers */
1036 	u8 __iomem *kregbase2;
1037 	/* for detecting offset above kregbase2 address */
1038 	u32 base2_start;
1039 
1040 	/* Per VL data. Enough for all VLs but not all elements are set/used. */
1041 	struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
1042 	/* send context data */
1043 	struct send_context_info *send_contexts;
1044 	/* map hardware send contexts to software index */
1045 	u8 *hw_to_sw;
1046 	/* spinlock for allocating and releasing send context resources */
1047 	spinlock_t sc_lock;
1048 	/* lock for pio_map */
1049 	spinlock_t pio_map_lock;
1050 	/* Send Context initialization lock. */
1051 	spinlock_t sc_init_lock;
1052 	/* lock for sdma_map */
1053 	spinlock_t                          sde_map_lock;
1054 	/* array of kernel send contexts */
1055 	struct send_context **kernel_send_context;
1056 	/* array of vl maps */
1057 	struct pio_vl_map __rcu *pio_map;
1058 	/* default flags to last descriptor */
1059 	u64 default_desc1;
1060 
1061 	/* fields common to all SDMA engines */
1062 
1063 	volatile __le64                    *sdma_heads_dma; /* DMA'ed by chip */
1064 	dma_addr_t                          sdma_heads_phys;
1065 	void                               *sdma_pad_dma; /* DMA'ed by chip */
1066 	dma_addr_t                          sdma_pad_phys;
1067 	/* for deallocation */
1068 	size_t                              sdma_heads_size;
1069 	/* num used */
1070 	u32                                 num_sdma;
1071 	/* array of engines sized by num_sdma */
1072 	struct sdma_engine                 *per_sdma;
1073 	/* array of vl maps */
1074 	struct sdma_vl_map __rcu           *sdma_map;
1075 	/* SPC freeze waitqueue and variable */
1076 	wait_queue_head_t		  sdma_unfreeze_wq;
1077 	atomic_t			  sdma_unfreeze_count;
1078 
1079 	u32 lcb_access_count;		/* count of LCB users */
1080 
1081 	/* common data between shared ASIC HFIs in this OS */
1082 	struct hfi1_asic_data *asic_data;
1083 
1084 	/* mem-mapped pointer to base of PIO buffers */
1085 	void __iomem *piobase;
1086 	/*
1087 	 * write-combining mem-mapped pointer to base of RcvArray
1088 	 * memory.
1089 	 */
1090 	void __iomem *rcvarray_wc;
1091 	/*
1092 	 * credit return base - a per-NUMA range of DMA address that
1093 	 * the chip will use to update the per-context free counter
1094 	 */
1095 	struct credit_return_base *cr_base;
1096 
1097 	/* send context numbers and sizes for each type */
1098 	struct sc_config_sizes sc_sizes[SC_MAX];
1099 
1100 	char *boardname; /* human readable board info */
1101 
1102 	u64 ctx0_seq_drop;
1103 
1104 	/* reset value */
1105 	u64 z_int_counter;
1106 	u64 z_rcv_limit;
1107 	u64 z_send_schedule;
1108 
1109 	u64 __percpu *send_schedule;
1110 	/* number of reserved contexts for netdev usage */
1111 	u16 num_netdev_contexts;
1112 	/* number of receive contexts in use by the driver */
1113 	u32 num_rcv_contexts;
1114 	/* number of pio send contexts in use by the driver */
1115 	u32 num_send_contexts;
1116 	/*
1117 	 * number of ctxts available for PSM open
1118 	 */
1119 	u32 freectxts;
1120 	/* total number of available user/PSM contexts */
1121 	u32 num_user_contexts;
1122 	/* base receive interrupt timeout, in CSR units */
1123 	u32 rcv_intr_timeout_csr;
1124 
1125 	spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
1126 	spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
1127 	spinlock_t uctxt_lock; /* protect rcd changes */
1128 	struct mutex dc8051_lock; /* exclusive access to 8051 */
1129 	struct workqueue_struct *update_cntr_wq;
1130 	struct work_struct update_cntr_work;
1131 	/* exclusive access to 8051 memory */
1132 	spinlock_t dc8051_memlock;
1133 	int dc8051_timed_out;	/* remember if the 8051 timed out */
1134 	/*
1135 	 * A page that will hold event notification bitmaps for all
1136 	 * contexts. This page will be mapped into all processes.
1137 	 */
1138 	unsigned long *events;
1139 	/*
1140 	 * per unit status, see also portdata statusp
1141 	 * mapped read-only into user processes so they can get unit and
1142 	 * IB link status cheaply
1143 	 */
1144 	struct hfi1_status *status;
1145 
1146 	/* revision register shadow */
1147 	u64 revision;
1148 	/* Base GUID for device (network order) */
1149 	u64 base_guid;
1150 
1151 	/* both sides of the PCIe link are gen3 capable */
1152 	u8 link_gen3_capable;
1153 	u8 dc_shutdown;
1154 	/* localbus width (1, 2,4,8,16,32) from config space  */
1155 	u32 lbus_width;
1156 	/* localbus speed in MHz */
1157 	u32 lbus_speed;
1158 	int unit; /* unit # of this chip */
1159 	int node; /* home node of this chip */
1160 
1161 	/* save these PCI fields to restore after a reset */
1162 	u32 pcibar0;
1163 	u32 pcibar1;
1164 	u32 pci_rom;
1165 	u16 pci_command;
1166 	u16 pcie_devctl;
1167 	u16 pcie_lnkctl;
1168 	u16 pcie_devctl2;
1169 	u32 pci_msix0;
1170 	u32 pci_tph2;
1171 
1172 	/*
1173 	 * ASCII serial number, from flash, large enough for original
1174 	 * all digit strings, and longer serial number format
1175 	 */
1176 	u8 serial[SERIAL_MAX];
1177 	/* human readable board version */
1178 	u8 boardversion[BOARD_VERS_MAX];
1179 	u8 lbus_info[32]; /* human readable localbus info */
1180 	/* chip major rev, from CceRevision */
1181 	u8 majrev;
1182 	/* chip minor rev, from CceRevision */
1183 	u8 minrev;
1184 	/* hardware ID */
1185 	u8 hfi1_id;
1186 	/* implementation code */
1187 	u8 icode;
1188 	/* vAU of this device */
1189 	u8 vau;
1190 	/* vCU of this device */
1191 	u8 vcu;
1192 	/* link credits of this device */
1193 	u16 link_credits;
1194 	/* initial vl15 credits to use */
1195 	u16 vl15_init;
1196 
1197 	/*
1198 	 * Cached value for vl15buf, read during verify cap interrupt. VL15
1199 	 * credits are to be kept at 0 and set when handling the link-up
1200 	 * interrupt. This removes the possibility of receiving VL15 MAD
1201 	 * packets before this HFI is ready.
1202 	 */
1203 	u16 vl15buf_cached;
1204 
1205 	/* Misc small ints */
1206 	u8 n_krcv_queues;
1207 	u8 qos_shift;
1208 
1209 	u16 irev;	/* implementation revision */
1210 	u32 dc8051_ver; /* 8051 firmware version */
1211 
1212 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1213 	struct platform_config platform_config;
1214 	struct platform_config_cache pcfg_cache;
1215 
1216 	struct diag_client *diag_client;
1217 
1218 	/* general interrupt: mask of handled interrupts */
1219 	u64 gi_mask[CCE_NUM_INT_CSRS];
1220 
1221 	struct rcv_array_data rcv_entries;
1222 
1223 	/* cycle length of PS* counters in HW (in picoseconds) */
1224 	u16 psxmitwait_check_rate;
1225 
1226 	/*
1227 	 * 64 bit synthetic counters
1228 	 */
1229 	struct timer_list synth_stats_timer;
1230 
1231 	/* MSI-X information */
1232 	struct hfi1_msix_info msix_info;
1233 
1234 	/*
1235 	 * device counters
1236 	 */
1237 	char *cntrnames;
1238 	size_t cntrnameslen;
1239 	size_t ndevcntrs;
1240 	u64 *cntrs;
1241 	u64 *scntrs;
1242 
1243 	/*
1244 	 * remembered values for synthetic counters
1245 	 */
1246 	u64 last_tx;
1247 	u64 last_rx;
1248 
1249 	/*
1250 	 * per-port counters
1251 	 */
1252 	size_t nportcntrs;
1253 	char *portcntrnames;
1254 	size_t portcntrnameslen;
1255 
1256 	struct err_info_rcvport err_info_rcvport;
1257 	struct err_info_constraint err_info_rcv_constraint;
1258 	struct err_info_constraint err_info_xmit_constraint;
1259 
1260 	atomic_t drop_packet;
1261 	bool do_drop;
1262 	u8 err_info_uncorrectable;
1263 	u8 err_info_fmconfig;
1264 
1265 	/*
1266 	 * Software counters for the status bits defined by the
1267 	 * associated error status registers
1268 	 */
1269 	u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1270 	u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1271 	u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1272 	u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1273 	u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1274 	u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1275 	u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1276 
1277 	/* Software counter that spans all contexts */
1278 	u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1279 	/* Software counter that spans all DMA engines */
1280 	u64 sw_send_dma_eng_err_status_cnt[
1281 		NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1282 	/* Software counter that aggregates all cce_err_status errors */
1283 	u64 sw_cce_err_status_aggregate;
1284 	/* Software counter that aggregates all bypass packet rcv errors */
1285 	u64 sw_rcv_bypass_packet_errors;
1286 
1287 	/* Save the enabled LCB error bits */
1288 	u64 lcb_err_en;
1289 	struct cpu_mask_set *comp_vect;
1290 	int *comp_vect_mappings;
1291 	u32 comp_vect_possible_cpus;
1292 
1293 	/*
1294 	 * Capability to have different send engines simply by changing a
1295 	 * pointer value.
1296 	 */
1297 	send_routine process_pio_send ____cacheline_aligned_in_smp;
1298 	send_routine process_dma_send;
1299 	void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1300 				u64 pbc, const void *from, size_t count);
1301 	int (*process_vnic_dma_send)(struct hfi1_devdata *dd, u8 q_idx,
1302 				     struct hfi1_vnic_vport_info *vinfo,
1303 				     struct sk_buff *skb, u64 pbc, u8 plen);
1304 	/* hfi1_pportdata, points to array of (physical) port-specific
1305 	 * data structs, indexed by pidx (0..n-1)
1306 	 */
1307 	struct hfi1_pportdata *pport;
1308 	/* receive context data */
1309 	struct hfi1_ctxtdata **rcd;
1310 	u64 __percpu *int_counter;
1311 	/* verbs tx opcode stats */
1312 	struct hfi1_opcode_stats_perctx __percpu *tx_opstats;
1313 	/* device (not port) flags, basically device capabilities */
1314 	u16 flags;
1315 	/* Number of physical ports available */
1316 	u8 num_pports;
1317 	/* Lowest context number which can be used by user processes or VNIC */
1318 	u8 first_dyn_alloc_ctxt;
1319 	/* adding a new field here would make it part of this cacheline */
1320 
1321 	/* seqlock for sc2vl */
1322 	seqlock_t sc2vl_lock ____cacheline_aligned_in_smp;
1323 	u64 sc2vl[4];
1324 	u64 __percpu *rcv_limit;
1325 	/* adding a new field here would make it part of this cacheline */
1326 
1327 	/* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1328 	u8 oui1;
1329 	u8 oui2;
1330 	u8 oui3;
1331 
1332 	/* Timer and counter used to detect RcvBufOvflCnt changes */
1333 	struct timer_list rcverr_timer;
1334 
1335 	wait_queue_head_t event_queue;
1336 
1337 	/* receive context tail dummy address */
1338 	__le64 *rcvhdrtail_dummy_kvaddr;
1339 	dma_addr_t rcvhdrtail_dummy_dma;
1340 
1341 	u32 rcv_ovfl_cnt;
1342 	/* Serialize ASPM enable/disable between multiple verbs contexts */
1343 	spinlock_t aspm_lock;
1344 	/* Number of verbs contexts which have disabled ASPM */
1345 	atomic_t aspm_disabled_cnt;
1346 	/* Keeps track of user space clients */
1347 	refcount_t user_refcount;
1348 	/* Used to wait for outstanding user space clients before dev removal */
1349 	struct completion user_comp;
1350 
1351 	bool eprom_available;	/* true if EPROM is available for this device */
1352 	bool aspm_supported;	/* Does HW support ASPM */
1353 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
1354 	struct rhashtable *sdma_rht;
1355 
1356 	/* vnic data */
1357 	struct hfi1_vnic_data vnic;
1358 	/* Lock to protect IRQ SRC register access */
1359 	spinlock_t irq_src_lock;
1360 	int vnic_num_vports;
1361 	struct hfi1_netdev_rx *netdev_rx;
1362 	struct hfi1_affinity_node *affinity_entry;
1363 
1364 	/* Keeps track of IPoIB RSM rule users */
1365 	atomic_t ipoib_rsm_usr_num;
1366 };
1367 
1368 /* 8051 firmware version helper */
1369 #define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
1370 #define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
1371 #define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
1372 #define dc8051_ver_patch(a) ((a) & 0x0000ff)
1373 
1374 /* f_put_tid types */
1375 #define PT_EXPECTED       0
1376 #define PT_EAGER          1
1377 #define PT_INVALID_FLUSH  2
1378 #define PT_INVALID        3
1379 
1380 struct tid_rb_node;
1381 
1382 /* Private data for file operations */
1383 struct hfi1_filedata {
1384 	struct srcu_struct pq_srcu;
1385 	struct hfi1_devdata *dd;
1386 	struct hfi1_ctxtdata *uctxt;
1387 	struct hfi1_user_sdma_comp_q *cq;
1388 	/* update side lock for SRCU */
1389 	spinlock_t pq_rcu_lock;
1390 	struct hfi1_user_sdma_pkt_q __rcu *pq;
1391 	u16 subctxt;
1392 	/* for cpu affinity; -1 if none */
1393 	int rec_cpu_num;
1394 	u32 tid_n_pinned;
1395 	bool use_mn;
1396 	struct tid_rb_node **entry_to_rb;
1397 	spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1398 	u32 tid_limit;
1399 	u32 tid_used;
1400 	u32 *invalid_tids;
1401 	u32 invalid_tid_idx;
1402 	/* protect invalid_tids array and invalid_tid_idx */
1403 	spinlock_t invalid_lock;
1404 };
1405 
1406 extern struct xarray hfi1_dev_table;
1407 struct hfi1_devdata *hfi1_lookup(int unit);
1408 
1409 static inline unsigned long uctxt_offset(struct hfi1_ctxtdata *uctxt)
1410 {
1411 	return (uctxt->ctxt - uctxt->dd->first_dyn_alloc_ctxt) *
1412 		HFI1_MAX_SHARED_CTXTS;
1413 }
1414 
1415 int hfi1_init(struct hfi1_devdata *dd, int reinit);
1416 int hfi1_count_active_units(void);
1417 
1418 int hfi1_diag_add(struct hfi1_devdata *dd);
1419 void hfi1_diag_remove(struct hfi1_devdata *dd);
1420 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1421 
1422 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1423 
1424 int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1425 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
1426 int hfi1_create_kctxts(struct hfi1_devdata *dd);
1427 int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa,
1428 			 struct hfi1_ctxtdata **rcd);
1429 void hfi1_free_ctxt(struct hfi1_ctxtdata *rcd);
1430 void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
1431 			 struct hfi1_devdata *dd, u8 hw_pidx, u32 port);
1432 void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1433 int hfi1_rcd_put(struct hfi1_ctxtdata *rcd);
1434 int hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
1435 struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd,
1436 						 u16 ctxt);
1437 struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt);
1438 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
1439 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1440 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1441 int handle_receive_interrupt_napi_fp(struct hfi1_ctxtdata *rcd, int budget);
1442 int handle_receive_interrupt_napi_sp(struct hfi1_ctxtdata *rcd, int budget);
1443 void set_all_slowpath(struct hfi1_devdata *dd);
1444 
1445 extern const struct pci_device_id hfi1_pci_tbl[];
1446 void hfi1_make_ud_req_9B(struct rvt_qp *qp,
1447 			 struct hfi1_pkt_state *ps,
1448 			 struct rvt_swqe *wqe);
1449 
1450 void hfi1_make_ud_req_16B(struct rvt_qp *qp,
1451 			  struct hfi1_pkt_state *ps,
1452 			  struct rvt_swqe *wqe);
1453 
1454 /* receive packet handler dispositions */
1455 #define RCV_PKT_OK      0x0 /* keep going */
1456 #define RCV_PKT_LIMIT   0x1 /* stop, hit limit, start thread */
1457 #define RCV_PKT_DONE    0x2 /* stop, no more packets detected */
1458 
1459 /**
1460  * hfi1_rcd_head - add accessor for rcd head
1461  * @rcd: the context
1462  */
1463 static inline u32 hfi1_rcd_head(struct hfi1_ctxtdata *rcd)
1464 {
1465 	return rcd->head;
1466 }
1467 
1468 /**
1469  * hfi1_set_rcd_head - add accessor for rcd head
1470  * @rcd: the context
1471  * @head: the new head
1472  */
1473 static inline void hfi1_set_rcd_head(struct hfi1_ctxtdata *rcd, u32 head)
1474 {
1475 	rcd->head = head;
1476 }
1477 
1478 /* calculate the current RHF address */
1479 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1480 {
1481 	return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->rhf_offset;
1482 }
1483 
1484 /* return DMA_RTAIL configuration */
1485 static inline bool get_dma_rtail_setting(struct hfi1_ctxtdata *rcd)
1486 {
1487 	return !!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL);
1488 }
1489 
1490 /**
1491  * hfi1_seq_incr_wrap - wrapping increment for sequence
1492  * @seq: the current sequence number
1493  *
1494  * Returns: the incremented seq
1495  */
1496 static inline u8 hfi1_seq_incr_wrap(u8 seq)
1497 {
1498 	if (++seq > RHF_MAX_SEQ)
1499 		seq = 1;
1500 	return seq;
1501 }
1502 
1503 /**
1504  * hfi1_seq_cnt - return seq_cnt member
1505  * @rcd: the receive context
1506  *
1507  * Return seq_cnt member
1508  */
1509 static inline u8 hfi1_seq_cnt(struct hfi1_ctxtdata *rcd)
1510 {
1511 	return rcd->seq_cnt;
1512 }
1513 
1514 /**
1515  * hfi1_set_seq_cnt - return seq_cnt member
1516  * @rcd: the receive context
1517  *
1518  * Return seq_cnt member
1519  */
1520 static inline void hfi1_set_seq_cnt(struct hfi1_ctxtdata *rcd, u8 cnt)
1521 {
1522 	rcd->seq_cnt = cnt;
1523 }
1524 
1525 /**
1526  * last_rcv_seq - is last
1527  * @rcd: the receive context
1528  * @seq: sequence
1529  *
1530  * return true if last packet
1531  */
1532 static inline bool last_rcv_seq(struct hfi1_ctxtdata *rcd, u32 seq)
1533 {
1534 	return seq != rcd->seq_cnt;
1535 }
1536 
1537 /**
1538  * rcd_seq_incr - increment context sequence number
1539  * @rcd: the receive context
1540  * @seq: the current sequence number
1541  *
1542  * Returns: true if the this was the last packet
1543  */
1544 static inline bool hfi1_seq_incr(struct hfi1_ctxtdata *rcd, u32 seq)
1545 {
1546 	rcd->seq_cnt = hfi1_seq_incr_wrap(rcd->seq_cnt);
1547 	return last_rcv_seq(rcd, seq);
1548 }
1549 
1550 /**
1551  * get_hdrqentsize - return hdrq entry size
1552  * @rcd: the receive context
1553  */
1554 static inline u8 get_hdrqentsize(struct hfi1_ctxtdata *rcd)
1555 {
1556 	return rcd->rcvhdrqentsize;
1557 }
1558 
1559 /**
1560  * get_hdrq_cnt - return hdrq count
1561  * @rcd: the receive context
1562  */
1563 static inline u16 get_hdrq_cnt(struct hfi1_ctxtdata *rcd)
1564 {
1565 	return rcd->rcvhdrq_cnt;
1566 }
1567 
1568 /**
1569  * hfi1_is_slowpath - check if this context is slow path
1570  * @rcd: the receive context
1571  */
1572 static inline bool hfi1_is_slowpath(struct hfi1_ctxtdata *rcd)
1573 {
1574 	return rcd->do_interrupt == rcd->slow_handler;
1575 }
1576 
1577 /**
1578  * hfi1_is_fastpath - check if this context is fast path
1579  * @rcd: the receive context
1580  */
1581 static inline bool hfi1_is_fastpath(struct hfi1_ctxtdata *rcd)
1582 {
1583 	if (rcd->ctxt == HFI1_CTRL_CTXT)
1584 		return false;
1585 
1586 	return rcd->do_interrupt == rcd->fast_handler;
1587 }
1588 
1589 /**
1590  * hfi1_set_fast - change to the fast handler
1591  * @rcd: the receive context
1592  */
1593 static inline void hfi1_set_fast(struct hfi1_ctxtdata *rcd)
1594 {
1595 	if (unlikely(!rcd))
1596 		return;
1597 	if (unlikely(!hfi1_is_fastpath(rcd)))
1598 		rcd->do_interrupt = rcd->fast_handler;
1599 }
1600 
1601 int hfi1_reset_device(int);
1602 
1603 void receive_interrupt_work(struct work_struct *work);
1604 
1605 /* extract service channel from header and rhf */
1606 static inline int hfi1_9B_get_sc5(struct ib_header *hdr, u64 rhf)
1607 {
1608 	return ib_get_sc(hdr) | ((!!(rhf_dc_info(rhf))) << 4);
1609 }
1610 
1611 #define HFI1_JKEY_WIDTH       16
1612 #define HFI1_JKEY_MASK        (BIT(16) - 1)
1613 #define HFI1_ADMIN_JKEY_RANGE 32
1614 
1615 /*
1616  * J_KEYs are split and allocated in the following groups:
1617  *   0 - 31    - users with administrator privileges
1618  *  32 - 63    - kernel protocols using KDETH packets
1619  *  64 - 65535 - all other users using KDETH packets
1620  */
1621 static inline u16 generate_jkey(kuid_t uid)
1622 {
1623 	u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1624 
1625 	if (capable(CAP_SYS_ADMIN))
1626 		jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1627 	else if (jkey < 64)
1628 		jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1629 
1630 	return jkey;
1631 }
1632 
1633 /*
1634  * active_egress_rate
1635  *
1636  * returns the active egress rate in units of [10^6 bits/sec]
1637  */
1638 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1639 {
1640 	u16 link_speed = ppd->link_speed_active;
1641 	u16 link_width = ppd->link_width_active;
1642 	u32 egress_rate;
1643 
1644 	if (link_speed == OPA_LINK_SPEED_25G)
1645 		egress_rate = 25000;
1646 	else /* assume OPA_LINK_SPEED_12_5G */
1647 		egress_rate = 12500;
1648 
1649 	switch (link_width) {
1650 	case OPA_LINK_WIDTH_4X:
1651 		egress_rate *= 4;
1652 		break;
1653 	case OPA_LINK_WIDTH_3X:
1654 		egress_rate *= 3;
1655 		break;
1656 	case OPA_LINK_WIDTH_2X:
1657 		egress_rate *= 2;
1658 		break;
1659 	default:
1660 		/* assume IB_WIDTH_1X */
1661 		break;
1662 	}
1663 
1664 	return egress_rate;
1665 }
1666 
1667 /*
1668  * egress_cycles
1669  *
1670  * Returns the number of 'fabric clock cycles' to egress a packet
1671  * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1672  * rate is (approximately) 805 MHz, the units of the returned value
1673  * are (1/805 MHz).
1674  */
1675 static inline u32 egress_cycles(u32 len, u32 rate)
1676 {
1677 	u32 cycles;
1678 
1679 	/*
1680 	 * cycles is:
1681 	 *
1682 	 *          (length) [bits] / (rate) [bits/sec]
1683 	 *  ---------------------------------------------------
1684 	 *  fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1685 	 */
1686 
1687 	cycles = len * 8; /* bits */
1688 	cycles *= 805;
1689 	cycles /= rate;
1690 
1691 	return cycles;
1692 }
1693 
1694 void set_link_ipg(struct hfi1_pportdata *ppd);
1695 void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
1696 		  u32 rqpn, u8 svc_type);
1697 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1698 		u16 pkey, u32 slid, u32 dlid, u8 sc5,
1699 		const struct ib_grh *old_grh);
1700 void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1701 		    u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1702 		    u8 sc5, const struct ib_grh *old_grh);
1703 typedef void (*hfi1_handle_cnp)(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1704 				u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1705 				u8 sc5, const struct ib_grh *old_grh);
1706 
1707 #define PKEY_CHECK_INVALID -1
1708 int egress_pkey_check(struct hfi1_pportdata *ppd, u32 slid, u16 pkey,
1709 		      u8 sc5, int8_t s_pkey_index);
1710 
1711 #define PACKET_EGRESS_TIMEOUT 350
1712 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1713 {
1714 	/* Pause at least 1us, to ensure chip returns all credits */
1715 	u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1716 
1717 	udelay(usec ? usec : 1);
1718 }
1719 
1720 /**
1721  * sc_to_vlt() - reverse lookup sc to vl
1722  * @dd - devdata
1723  * @sc5 - 5 bit sc
1724  */
1725 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1726 {
1727 	unsigned seq;
1728 	u8 rval;
1729 
1730 	if (sc5 >= OPA_MAX_SCS)
1731 		return (u8)(0xff);
1732 
1733 	do {
1734 		seq = read_seqbegin(&dd->sc2vl_lock);
1735 		rval = *(((u8 *)dd->sc2vl) + sc5);
1736 	} while (read_seqretry(&dd->sc2vl_lock, seq));
1737 
1738 	return rval;
1739 }
1740 
1741 #define PKEY_MEMBER_MASK 0x8000
1742 #define PKEY_LOW_15_MASK 0x7fff
1743 
1744 /*
1745  * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1746  * being an entry from the ingress partition key table), return 0
1747  * otherwise. Use the matching criteria for ingress partition keys
1748  * specified in the OPAv1 spec., section 9.10.14.
1749  */
1750 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1751 {
1752 	u16 mkey = pkey & PKEY_LOW_15_MASK;
1753 	u16 ment = ent & PKEY_LOW_15_MASK;
1754 
1755 	if (mkey == ment) {
1756 		/*
1757 		 * If pkey[15] is clear (limited partition member),
1758 		 * is bit 15 in the corresponding table element
1759 		 * clear (limited member)?
1760 		 */
1761 		if (!(pkey & PKEY_MEMBER_MASK))
1762 			return !!(ent & PKEY_MEMBER_MASK);
1763 		return 1;
1764 	}
1765 	return 0;
1766 }
1767 
1768 /*
1769  * ingress_pkey_table_search - search the entire pkey table for
1770  * an entry which matches 'pkey'. return 0 if a match is found,
1771  * and 1 otherwise.
1772  */
1773 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1774 {
1775 	int i;
1776 
1777 	for (i = 0; i < MAX_PKEY_VALUES; i++) {
1778 		if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1779 			return 0;
1780 	}
1781 	return 1;
1782 }
1783 
1784 /*
1785  * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1786  * i.e., increment port_rcv_constraint_errors for the port, and record
1787  * the 'error info' for this failure.
1788  */
1789 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1790 				    u32 slid)
1791 {
1792 	struct hfi1_devdata *dd = ppd->dd;
1793 
1794 	incr_cntr64(&ppd->port_rcv_constraint_errors);
1795 	if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1796 		dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1797 		dd->err_info_rcv_constraint.slid = slid;
1798 		dd->err_info_rcv_constraint.pkey = pkey;
1799 	}
1800 }
1801 
1802 /*
1803  * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1804  * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1805  * is a hint as to the best place in the partition key table to begin
1806  * searching. This function should not be called on the data path because
1807  * of performance reasons. On datapath pkey check is expected to be done
1808  * by HW and rcv_pkey_check function should be called instead.
1809  */
1810 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1811 				     u8 sc5, u8 idx, u32 slid, bool force)
1812 {
1813 	if (!(force) && !(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1814 		return 0;
1815 
1816 	/* If SC15, pkey[0:14] must be 0x7fff */
1817 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1818 		goto bad;
1819 
1820 	/* Is the pkey = 0x0, or 0x8000? */
1821 	if ((pkey & PKEY_LOW_15_MASK) == 0)
1822 		goto bad;
1823 
1824 	/* The most likely matching pkey has index 'idx' */
1825 	if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1826 		return 0;
1827 
1828 	/* no match - try the whole table */
1829 	if (!ingress_pkey_table_search(ppd, pkey))
1830 		return 0;
1831 
1832 bad:
1833 	ingress_pkey_table_fail(ppd, pkey, slid);
1834 	return 1;
1835 }
1836 
1837 /*
1838  * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1839  * otherwise. It only ensures pkey is vlid for QP0. This function
1840  * should be called on the data path instead of ingress_pkey_check
1841  * as on data path, pkey check is done by HW (except for QP0).
1842  */
1843 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1844 				 u8 sc5, u16 slid)
1845 {
1846 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1847 		return 0;
1848 
1849 	/* If SC15, pkey[0:14] must be 0x7fff */
1850 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1851 		goto bad;
1852 
1853 	return 0;
1854 bad:
1855 	ingress_pkey_table_fail(ppd, pkey, slid);
1856 	return 1;
1857 }
1858 
1859 /* MTU handling */
1860 
1861 /* MTU enumeration, 256-4k match IB */
1862 #define OPA_MTU_0     0
1863 #define OPA_MTU_256   1
1864 #define OPA_MTU_512   2
1865 #define OPA_MTU_1024  3
1866 #define OPA_MTU_2048  4
1867 #define OPA_MTU_4096  5
1868 
1869 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1870 int mtu_to_enum(u32 mtu, int default_if_bad);
1871 u16 enum_to_mtu(int mtu);
1872 static inline int valid_ib_mtu(unsigned int mtu)
1873 {
1874 	return mtu == 256 || mtu == 512 ||
1875 		mtu == 1024 || mtu == 2048 ||
1876 		mtu == 4096;
1877 }
1878 
1879 static inline int valid_opa_max_mtu(unsigned int mtu)
1880 {
1881 	return mtu >= 2048 &&
1882 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1883 }
1884 
1885 int set_mtu(struct hfi1_pportdata *ppd);
1886 
1887 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
1888 void hfi1_disable_after_error(struct hfi1_devdata *dd);
1889 int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
1890 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
1891 
1892 int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
1893 int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
1894 
1895 void set_up_vau(struct hfi1_devdata *dd, u8 vau);
1896 void set_up_vl15(struct hfi1_devdata *dd, u16 vl15buf);
1897 void reset_link_credits(struct hfi1_devdata *dd);
1898 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1899 
1900 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1901 
1902 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1903 {
1904 	return ppd->dd;
1905 }
1906 
1907 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1908 {
1909 	return container_of(dev, struct hfi1_devdata, verbs_dev);
1910 }
1911 
1912 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1913 {
1914 	return dd_from_dev(to_idev(ibdev));
1915 }
1916 
1917 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1918 {
1919 	return container_of(ibp, struct hfi1_pportdata, ibport_data);
1920 }
1921 
1922 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1923 {
1924 	return container_of(rdi, struct hfi1_ibdev, rdi);
1925 }
1926 
1927 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u32 port)
1928 {
1929 	struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1930 	u32 pidx = port - 1; /* IB number port from 1, hdw from 0 */
1931 
1932 	WARN_ON(pidx >= dd->num_pports);
1933 	return &dd->pport[pidx].ibport_data;
1934 }
1935 
1936 static inline struct hfi1_ibport *rcd_to_iport(struct hfi1_ctxtdata *rcd)
1937 {
1938 	return &rcd->ppd->ibport_data;
1939 }
1940 
1941 /**
1942  * hfi1_may_ecn - Check whether FECN or BECN processing should be done
1943  * @pkt: the packet to be evaluated
1944  *
1945  * Check whether the FECN or BECN bits in the packet's header are
1946  * enabled, depending on packet type.
1947  *
1948  * This function only checks for FECN and BECN bits. Additional checks
1949  * are done in the slowpath (hfi1_process_ecn_slowpath()) in order to
1950  * ensure correct handling.
1951  */
1952 static inline bool hfi1_may_ecn(struct hfi1_packet *pkt)
1953 {
1954 	bool fecn, becn;
1955 
1956 	if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
1957 		fecn = hfi1_16B_get_fecn(pkt->hdr);
1958 		becn = hfi1_16B_get_becn(pkt->hdr);
1959 	} else {
1960 		fecn = ib_bth_get_fecn(pkt->ohdr);
1961 		becn = ib_bth_get_becn(pkt->ohdr);
1962 	}
1963 	return fecn || becn;
1964 }
1965 
1966 bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
1967 			       bool prescan);
1968 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt)
1969 {
1970 	bool do_work;
1971 
1972 	do_work = hfi1_may_ecn(pkt);
1973 	if (unlikely(do_work))
1974 		return hfi1_process_ecn_slowpath(qp, pkt, false);
1975 	return false;
1976 }
1977 
1978 /*
1979  * Return the indexed PKEY from the port PKEY table.
1980  */
1981 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
1982 {
1983 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1984 	u16 ret;
1985 
1986 	if (index >= ARRAY_SIZE(ppd->pkeys))
1987 		ret = 0;
1988 	else
1989 		ret = ppd->pkeys[index];
1990 
1991 	return ret;
1992 }
1993 
1994 /*
1995  * Return the indexed GUID from the port GUIDs table.
1996  */
1997 static inline __be64 get_sguid(struct hfi1_ibport *ibp, unsigned int index)
1998 {
1999 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2000 
2001 	WARN_ON(index >= HFI1_GUIDS_PER_PORT);
2002 	return cpu_to_be64(ppd->guids[index]);
2003 }
2004 
2005 /*
2006  * Called by readers of cc_state only, must call under rcu_read_lock().
2007  */
2008 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
2009 {
2010 	return rcu_dereference(ppd->cc_state);
2011 }
2012 
2013 /*
2014  * Called by writers of cc_state only,  must call under cc_state_lock.
2015  */
2016 static inline
2017 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
2018 {
2019 	return rcu_dereference_protected(ppd->cc_state,
2020 					 lockdep_is_held(&ppd->cc_state_lock));
2021 }
2022 
2023 /*
2024  * values for dd->flags (_device_ related flags)
2025  */
2026 #define HFI1_INITTED           0x1    /* chip and driver up and initted */
2027 #define HFI1_PRESENT           0x2    /* chip accesses can be done */
2028 #define HFI1_FROZEN            0x4    /* chip in SPC freeze */
2029 #define HFI1_HAS_SDMA_TIMEOUT  0x8
2030 #define HFI1_HAS_SEND_DMA      0x10   /* Supports Send DMA */
2031 #define HFI1_FORCED_FREEZE     0x80   /* driver forced freeze mode */
2032 #define HFI1_SHUTDOWN          0x100  /* device is shutting down */
2033 
2034 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
2035 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
2036 
2037 /* ctxt_flag bit offsets */
2038 		/* base context has not finished initializing */
2039 #define HFI1_CTXT_BASE_UNINIT 1
2040 		/* base context initaliation failed */
2041 #define HFI1_CTXT_BASE_FAILED 2
2042 		/* waiting for a packet to arrive */
2043 #define HFI1_CTXT_WAITING_RCV 3
2044 		/* waiting for an urgent packet to arrive */
2045 #define HFI1_CTXT_WAITING_URG 4
2046 
2047 /* free up any allocated data at closes */
2048 int hfi1_init_dd(struct hfi1_devdata *dd);
2049 void hfi1_free_devdata(struct hfi1_devdata *dd);
2050 
2051 /* LED beaconing functions */
2052 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
2053 			     unsigned int timeoff);
2054 void shutdown_led_override(struct hfi1_pportdata *ppd);
2055 
2056 #define HFI1_CREDIT_RETURN_RATE (100)
2057 
2058 /*
2059  * The number of words for the KDETH protocol field.  If this is
2060  * larger then the actual field used, then part of the payload
2061  * will be in the header.
2062  *
2063  * Optimally, we want this sized so that a typical case will
2064  * use full cache lines.  The typical local KDETH header would
2065  * be:
2066  *
2067  *	Bytes	Field
2068  *	  8	LRH
2069  *	 12	BHT
2070  *	 ??	KDETH
2071  *	  8	RHF
2072  *	---
2073  *	 28 + KDETH
2074  *
2075  * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
2076  */
2077 #define DEFAULT_RCVHDRSIZE 9
2078 
2079 /*
2080  * Maximal header byte count:
2081  *
2082  *	Bytes	Field
2083  *	  8	LRH
2084  *	 40	GRH (optional)
2085  *	 12	BTH
2086  *	 ??	KDETH
2087  *	  8	RHF
2088  *	---
2089  *	 68 + KDETH
2090  *
2091  * We also want to maintain a cache line alignment to assist DMA'ing
2092  * of the header bytes.  Round up to a good size.
2093  */
2094 #define DEFAULT_RCVHDR_ENTSIZE 32
2095 
2096 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
2097 			u32 nlocked, u32 npages);
2098 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
2099 			    size_t npages, bool writable, struct page **pages);
2100 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
2101 			     size_t npages, bool dirty);
2102 
2103 /**
2104  * hfi1_rcvhdrtail_kvaddr - return tail kvaddr
2105  * @rcd - the receive context
2106  */
2107 static inline __le64 *hfi1_rcvhdrtail_kvaddr(const struct hfi1_ctxtdata *rcd)
2108 {
2109 	return (__le64 *)rcd->rcvhdrtail_kvaddr;
2110 }
2111 
2112 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2113 {
2114 	u64 *kv = (u64 *)hfi1_rcvhdrtail_kvaddr(rcd);
2115 
2116 	if (kv)
2117 		*kv = 0ULL;
2118 }
2119 
2120 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2121 {
2122 	/*
2123 	 * volatile because it's a DMA target from the chip, routine is
2124 	 * inlined, and don't want register caching or reordering.
2125 	 */
2126 	return (u32)le64_to_cpu(*hfi1_rcvhdrtail_kvaddr(rcd));
2127 }
2128 
2129 static inline bool hfi1_packet_present(struct hfi1_ctxtdata *rcd)
2130 {
2131 	if (likely(!rcd->rcvhdrtail_kvaddr)) {
2132 		u32 seq = rhf_rcv_seq(rhf_to_cpu(get_rhf_addr(rcd)));
2133 
2134 		return !last_rcv_seq(rcd, seq);
2135 	}
2136 	return hfi1_rcd_head(rcd) != get_rcvhdrtail(rcd);
2137 }
2138 
2139 /*
2140  * sysfs interface.
2141  */
2142 
2143 extern const char ib_hfi1_version[];
2144 extern const struct attribute_group ib_hfi1_attr_group;
2145 extern const struct attribute_group *hfi1_attr_port_groups[];
2146 
2147 int hfi1_device_create(struct hfi1_devdata *dd);
2148 void hfi1_device_remove(struct hfi1_devdata *dd);
2149 
2150 int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
2151 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
2152 /* Hook for sysfs read of QSFP */
2153 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
2154 
2155 int hfi1_pcie_init(struct hfi1_devdata *dd);
2156 void hfi1_pcie_cleanup(struct pci_dev *pdev);
2157 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
2158 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
2159 int pcie_speeds(struct hfi1_devdata *dd);
2160 int restore_pci_variables(struct hfi1_devdata *dd);
2161 int save_pci_variables(struct hfi1_devdata *dd);
2162 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
2163 void tune_pcie_caps(struct hfi1_devdata *dd);
2164 int parse_platform_config(struct hfi1_devdata *dd);
2165 int get_platform_config_field(struct hfi1_devdata *dd,
2166 			      enum platform_config_table_type_encoding
2167 			      table_type, int table_index, int field_index,
2168 			      u32 *data, u32 len);
2169 
2170 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
2171 
2172 /*
2173  * Flush write combining store buffers (if present) and perform a write
2174  * barrier.
2175  */
2176 static inline void flush_wc(void)
2177 {
2178 	asm volatile("sfence" : : : "memory");
2179 }
2180 
2181 void handle_eflags(struct hfi1_packet *packet);
2182 void seqfile_dump_rcd(struct seq_file *s, struct hfi1_ctxtdata *rcd);
2183 
2184 /* global module parameter variables */
2185 extern unsigned int hfi1_max_mtu;
2186 extern unsigned int hfi1_cu;
2187 extern unsigned int user_credit_return_threshold;
2188 extern int num_user_contexts;
2189 extern unsigned long n_krcvqs;
2190 extern uint krcvqs[];
2191 extern int krcvqsset;
2192 extern uint loopback;
2193 extern uint quick_linkup;
2194 extern uint rcv_intr_timeout;
2195 extern uint rcv_intr_count;
2196 extern uint rcv_intr_dynamic;
2197 extern ushort link_crc_mask;
2198 
2199 extern struct mutex hfi1_mutex;
2200 
2201 /* Number of seconds before our card status check...  */
2202 #define STATUS_TIMEOUT 60
2203 
2204 #define DRIVER_NAME		"hfi1"
2205 #define HFI1_USER_MINOR_BASE     0
2206 #define HFI1_TRACE_MINOR         127
2207 #define HFI1_NMINORS             255
2208 
2209 #define PCI_VENDOR_ID_INTEL 0x8086
2210 #define PCI_DEVICE_ID_INTEL0 0x24f0
2211 #define PCI_DEVICE_ID_INTEL1 0x24f1
2212 
2213 #define HFI1_PKT_USER_SC_INTEGRITY					    \
2214 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
2215 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
2216 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
2217 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
2218 
2219 #define HFI1_PKT_KERNEL_SC_INTEGRITY					    \
2220 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
2221 
2222 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
2223 						  u16 ctxt_type)
2224 {
2225 	u64 base_sc_integrity;
2226 
2227 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2228 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2229 		return 0;
2230 
2231 	base_sc_integrity =
2232 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2233 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
2234 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2235 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2236 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2237 #ifndef CONFIG_FAULT_INJECTION
2238 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
2239 #endif
2240 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2241 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2242 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2243 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
2244 	| SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2245 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2246 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
2247 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
2248 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
2249 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2250 
2251 	if (ctxt_type == SC_USER)
2252 		base_sc_integrity |=
2253 #ifndef CONFIG_FAULT_INJECTION
2254 			SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK |
2255 #endif
2256 			HFI1_PKT_USER_SC_INTEGRITY;
2257 	else if (ctxt_type != SC_KERNEL)
2258 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
2259 
2260 	/* turn on send-side job key checks if !A0 */
2261 	if (!is_ax(dd))
2262 		base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2263 
2264 	return base_sc_integrity;
2265 }
2266 
2267 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
2268 {
2269 	u64 base_sdma_integrity;
2270 
2271 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2272 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2273 		return 0;
2274 
2275 	base_sdma_integrity =
2276 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2277 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2278 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2279 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2280 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2281 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2282 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2283 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
2284 	| SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2285 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2286 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
2287 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
2288 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
2289 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2290 
2291 	if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
2292 		base_sdma_integrity |=
2293 		SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
2294 
2295 	/* turn on send-side job key checks if !A0 */
2296 	if (!is_ax(dd))
2297 		base_sdma_integrity |=
2298 			SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2299 
2300 	return base_sdma_integrity;
2301 }
2302 
2303 #define dd_dev_emerg(dd, fmt, ...) \
2304 	dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
2305 		  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2306 
2307 #define dd_dev_err(dd, fmt, ...) \
2308 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
2309 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2310 
2311 #define dd_dev_err_ratelimited(dd, fmt, ...) \
2312 	dev_err_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2313 			    rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2314 			    ##__VA_ARGS__)
2315 
2316 #define dd_dev_warn(dd, fmt, ...) \
2317 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
2318 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2319 
2320 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
2321 	dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2322 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2323 			     ##__VA_ARGS__)
2324 
2325 #define dd_dev_info(dd, fmt, ...) \
2326 	dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
2327 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2328 
2329 #define dd_dev_info_ratelimited(dd, fmt, ...) \
2330 	dev_info_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2331 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2332 			     ##__VA_ARGS__)
2333 
2334 #define dd_dev_dbg(dd, fmt, ...) \
2335 	dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
2336 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2337 
2338 #define hfi1_dev_porterr(dd, port, fmt, ...) \
2339 	dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
2340 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), (port), ##__VA_ARGS__)
2341 
2342 /*
2343  * this is used for formatting hw error messages...
2344  */
2345 struct hfi1_hwerror_msgs {
2346 	u64 mask;
2347 	const char *msg;
2348 	size_t sz;
2349 };
2350 
2351 /* in intr.c... */
2352 void hfi1_format_hwerrors(u64 hwerrs,
2353 			  const struct hfi1_hwerror_msgs *hwerrmsgs,
2354 			  size_t nhwerrmsgs, char *msg, size_t lmsg);
2355 
2356 #define USER_OPCODE_CHECK_VAL 0xC0
2357 #define USER_OPCODE_CHECK_MASK 0xC0
2358 #define OPCODE_CHECK_VAL_DISABLED 0x0
2359 #define OPCODE_CHECK_MASK_DISABLED 0x0
2360 
2361 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
2362 {
2363 	struct hfi1_pportdata *ppd;
2364 	int i;
2365 
2366 	dd->z_int_counter = get_all_cpu_total(dd->int_counter);
2367 	dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
2368 	dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
2369 
2370 	ppd = (struct hfi1_pportdata *)(dd + 1);
2371 	for (i = 0; i < dd->num_pports; i++, ppd++) {
2372 		ppd->ibport_data.rvp.z_rc_acks =
2373 			get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
2374 		ppd->ibport_data.rvp.z_rc_qacks =
2375 			get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
2376 	}
2377 }
2378 
2379 /* Control LED state */
2380 static inline void setextled(struct hfi1_devdata *dd, u32 on)
2381 {
2382 	if (on)
2383 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
2384 	else
2385 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
2386 }
2387 
2388 /* return the i2c resource given the target */
2389 static inline u32 i2c_target(u32 target)
2390 {
2391 	return target ? CR_I2C2 : CR_I2C1;
2392 }
2393 
2394 /* return the i2c chain chip resource that this HFI uses for QSFP */
2395 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2396 {
2397 	return i2c_target(dd->hfi1_id);
2398 }
2399 
2400 /* Is this device integrated or discrete? */
2401 static inline bool is_integrated(struct hfi1_devdata *dd)
2402 {
2403 	return dd->pcidev->device == PCI_DEVICE_ID_INTEL1;
2404 }
2405 
2406 /**
2407  * hfi1_need_drop - detect need for drop
2408  * @dd: - the device
2409  *
2410  * In some cases, the first packet needs to be dropped.
2411  *
2412  * Return true is the current packet needs to be dropped and false otherwise.
2413  */
2414 static inline bool hfi1_need_drop(struct hfi1_devdata *dd)
2415 {
2416 	if (unlikely(dd->do_drop &&
2417 		     atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
2418 		     DROP_PACKET_ON)) {
2419 		dd->do_drop = false;
2420 		return true;
2421 	}
2422 	return false;
2423 }
2424 
2425 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2426 
2427 #define DD_DEV_ENTRY(dd)       __string(dev, dev_name(&(dd)->pcidev->dev))
2428 #define DD_DEV_ASSIGN(dd)      __assign_str(dev)
2429 
2430 static inline void hfi1_update_ah_attr(struct ib_device *ibdev,
2431 				       struct rdma_ah_attr *attr)
2432 {
2433 	struct hfi1_pportdata *ppd;
2434 	struct hfi1_ibport *ibp;
2435 	u32 dlid = rdma_ah_get_dlid(attr);
2436 
2437 	/*
2438 	 * Kernel clients may not have setup GRH information
2439 	 * Set that here.
2440 	 */
2441 	ibp = to_iport(ibdev, rdma_ah_get_port_num(attr));
2442 	ppd = ppd_from_ibp(ibp);
2443 	if ((((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) ||
2444 	      (ppd->lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))) &&
2445 	    (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)) &&
2446 	    (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2447 	    (!(rdma_ah_get_ah_flags(attr) & IB_AH_GRH))) ||
2448 	    (rdma_ah_get_make_grd(attr))) {
2449 		rdma_ah_set_ah_flags(attr, IB_AH_GRH);
2450 		rdma_ah_set_interface_id(attr, OPA_MAKE_ID(dlid));
2451 		rdma_ah_set_subnet_prefix(attr, ibp->rvp.gid_prefix);
2452 	}
2453 }
2454 
2455 /*
2456  * hfi1_check_mcast- Check if the given lid is
2457  * in the OPA multicast range.
2458  *
2459  * The LID might either reside in ah.dlid or might be
2460  * in the GRH of the address handle as DGID if extended
2461  * addresses are in use.
2462  */
2463 static inline bool hfi1_check_mcast(u32 lid)
2464 {
2465 	return ((lid >= opa_get_mcast_base(OPA_MCAST_NR)) &&
2466 		(lid != be32_to_cpu(OPA_LID_PERMISSIVE)));
2467 }
2468 
2469 #define opa_get_lid(lid, format)	\
2470 	__opa_get_lid(lid, OPA_PORT_PACKET_FORMAT_##format)
2471 
2472 /* Convert a lid to a specific lid space */
2473 static inline u32 __opa_get_lid(u32 lid, u8 format)
2474 {
2475 	bool is_mcast = hfi1_check_mcast(lid);
2476 
2477 	switch (format) {
2478 	case OPA_PORT_PACKET_FORMAT_8B:
2479 	case OPA_PORT_PACKET_FORMAT_10B:
2480 		if (is_mcast)
2481 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2482 				0xF0000);
2483 		return lid & 0xFFFFF;
2484 	case OPA_PORT_PACKET_FORMAT_16B:
2485 		if (is_mcast)
2486 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2487 				0xF00000);
2488 		return lid & 0xFFFFFF;
2489 	case OPA_PORT_PACKET_FORMAT_9B:
2490 		if (is_mcast)
2491 			return (lid -
2492 				opa_get_mcast_base(OPA_MCAST_NR) +
2493 				be16_to_cpu(IB_MULTICAST_LID_BASE));
2494 		else
2495 			return lid & 0xFFFF;
2496 	default:
2497 		return lid;
2498 	}
2499 }
2500 
2501 /* Return true if the given lid is the OPA 16B multicast range */
2502 static inline bool hfi1_is_16B_mcast(u32 lid)
2503 {
2504 	return ((lid >=
2505 		opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 16B)) &&
2506 		(lid != opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B)));
2507 }
2508 
2509 static inline void hfi1_make_opa_lid(struct rdma_ah_attr *attr)
2510 {
2511 	const struct ib_global_route *grh = rdma_ah_read_grh(attr);
2512 	u32 dlid = rdma_ah_get_dlid(attr);
2513 
2514 	/* Modify ah_attr.dlid to be in the 32 bit LID space.
2515 	 * This is how the address will be laid out:
2516 	 * Assuming MCAST_NR to be 4,
2517 	 * 32 bit permissive LID = 0xFFFFFFFF
2518 	 * Multicast LID range = 0xFFFFFFFE to 0xF0000000
2519 	 * Unicast LID range = 0xEFFFFFFF to 1
2520 	 * Invalid LID = 0
2521 	 */
2522 	if (ib_is_opa_gid(&grh->dgid))
2523 		dlid = opa_get_lid_from_gid(&grh->dgid);
2524 	else if ((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
2525 		 (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2526 		 (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)))
2527 		dlid = dlid - be16_to_cpu(IB_MULTICAST_LID_BASE) +
2528 			opa_get_mcast_base(OPA_MCAST_NR);
2529 	else if (dlid == be16_to_cpu(IB_LID_PERMISSIVE))
2530 		dlid = be32_to_cpu(OPA_LID_PERMISSIVE);
2531 
2532 	rdma_ah_set_dlid(attr, dlid);
2533 }
2534 
2535 static inline u8 hfi1_get_packet_type(u32 lid)
2536 {
2537 	/* 9B if lid > 0xF0000000 */
2538 	if (lid >= opa_get_mcast_base(OPA_MCAST_NR))
2539 		return HFI1_PKT_TYPE_9B;
2540 
2541 	/* 16B if lid > 0xC000 */
2542 	if (lid >= opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 9B))
2543 		return HFI1_PKT_TYPE_16B;
2544 
2545 	return HFI1_PKT_TYPE_9B;
2546 }
2547 
2548 static inline bool hfi1_get_hdr_type(u32 lid, struct rdma_ah_attr *attr)
2549 {
2550 	/*
2551 	 * If there was an incoming 16B packet with permissive
2552 	 * LIDs, OPA GIDs would have been programmed when those
2553 	 * packets were received. A 16B packet will have to
2554 	 * be sent in response to that packet. Return a 16B
2555 	 * header type if that's the case.
2556 	 */
2557 	if (rdma_ah_get_dlid(attr) == be32_to_cpu(OPA_LID_PERMISSIVE))
2558 		return (ib_is_opa_gid(&rdma_ah_read_grh(attr)->dgid)) ?
2559 			HFI1_PKT_TYPE_16B : HFI1_PKT_TYPE_9B;
2560 
2561 	/*
2562 	 * Return a 16B header type if either the destination
2563 	 * or source lid is extended.
2564 	 */
2565 	if (hfi1_get_packet_type(rdma_ah_get_dlid(attr)) == HFI1_PKT_TYPE_16B)
2566 		return HFI1_PKT_TYPE_16B;
2567 
2568 	return hfi1_get_packet_type(lid);
2569 }
2570 
2571 static inline void hfi1_make_ext_grh(struct hfi1_packet *packet,
2572 				     struct ib_grh *grh, u32 slid,
2573 				     u32 dlid)
2574 {
2575 	struct hfi1_ibport *ibp = &packet->rcd->ppd->ibport_data;
2576 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2577 
2578 	if (!ibp)
2579 		return;
2580 
2581 	grh->hop_limit = 1;
2582 	grh->sgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2583 	if (slid == opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B))
2584 		grh->sgid.global.interface_id =
2585 			OPA_MAKE_ID(be32_to_cpu(OPA_LID_PERMISSIVE));
2586 	else
2587 		grh->sgid.global.interface_id = OPA_MAKE_ID(slid);
2588 
2589 	/*
2590 	 * Upper layers (like mad) may compare the dgid in the
2591 	 * wc that is obtained here with the sgid_index in
2592 	 * the wr. Since sgid_index in wr is always 0 for
2593 	 * extended lids, set the dgid here to the default
2594 	 * IB gid.
2595 	 */
2596 	grh->dgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2597 	grh->dgid.global.interface_id =
2598 		cpu_to_be64(ppd->guids[HFI1_PORT_GUID_INDEX]);
2599 }
2600 
2601 static inline int hfi1_get_16b_padding(u32 hdr_size, u32 payload)
2602 {
2603 	return -(hdr_size + payload + (SIZE_OF_CRC << 2) +
2604 		     SIZE_OF_LT) & 0x7;
2605 }
2606 
2607 static inline void hfi1_make_ib_hdr(struct ib_header *hdr,
2608 				    u16 lrh0, u16 len,
2609 				    u16 dlid, u16 slid)
2610 {
2611 	hdr->lrh[0] = cpu_to_be16(lrh0);
2612 	hdr->lrh[1] = cpu_to_be16(dlid);
2613 	hdr->lrh[2] = cpu_to_be16(len);
2614 	hdr->lrh[3] = cpu_to_be16(slid);
2615 }
2616 
2617 static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr,
2618 				     u32 slid, u32 dlid,
2619 				     u16 len, u16 pkey,
2620 				     bool becn, bool fecn, u8 l4,
2621 				     u8 sc)
2622 {
2623 	u32 lrh0 = 0;
2624 	u32 lrh1 = 0x40000000;
2625 	u32 lrh2 = 0;
2626 	u32 lrh3 = 0;
2627 
2628 	lrh0 = (lrh0 & ~OPA_16B_BECN_MASK) | (becn << OPA_16B_BECN_SHIFT);
2629 	lrh0 = (lrh0 & ~OPA_16B_LEN_MASK) | (len << OPA_16B_LEN_SHIFT);
2630 	lrh0 = (lrh0 & ~OPA_16B_LID_MASK)  | (slid & OPA_16B_LID_MASK);
2631 	lrh1 = (lrh1 & ~OPA_16B_FECN_MASK) | (fecn << OPA_16B_FECN_SHIFT);
2632 	lrh1 = (lrh1 & ~OPA_16B_SC_MASK) | (sc << OPA_16B_SC_SHIFT);
2633 	lrh1 = (lrh1 & ~OPA_16B_LID_MASK) | (dlid & OPA_16B_LID_MASK);
2634 	lrh2 = (lrh2 & ~OPA_16B_SLID_MASK) |
2635 		((slid >> OPA_16B_SLID_SHIFT) << OPA_16B_SLID_HIGH_SHIFT);
2636 	lrh2 = (lrh2 & ~OPA_16B_DLID_MASK) |
2637 		((dlid >> OPA_16B_DLID_SHIFT) << OPA_16B_DLID_HIGH_SHIFT);
2638 	lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | ((u32)pkey << OPA_16B_PKEY_SHIFT);
2639 	lrh2 = (lrh2 & ~OPA_16B_L4_MASK) | l4;
2640 
2641 	hdr->lrh[0] = lrh0;
2642 	hdr->lrh[1] = lrh1;
2643 	hdr->lrh[2] = lrh2;
2644 	hdr->lrh[3] = lrh3;
2645 }
2646 #endif                          /* _HFI1_KERNEL_H */
2647