xref: /openbmc/linux/drivers/infiniband/hw/hfi1/hfi.h (revision e8ec0493)
1 #ifndef _HFI1_KERNEL_H
2 #define _HFI1_KERNEL_H
3 /*
4  * Copyright(c) 2015-2020 Intel Corporation.
5  *
6  * This file is provided under a dual BSD/GPLv2 license.  When using or
7  * redistributing this file, you may do so under either license.
8  *
9  * GPL LICENSE SUMMARY
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of version 2 of the GNU General Public License as
13  * published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * BSD LICENSE
21  *
22  * Redistribution and use in source and binary forms, with or without
23  * modification, are permitted provided that the following conditions
24  * are met:
25  *
26  *  - Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  *  - Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in
30  *    the documentation and/or other materials provided with the
31  *    distribution.
32  *  - Neither the name of Intel Corporation nor the names of its
33  *    contributors may be used to endorse or promote products derived
34  *    from this software without specific prior written permission.
35  *
36  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
37  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
38  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
39  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
40  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
41  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
42  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
43  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
44  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
45  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
46  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
47  *
48  */
49 
50 #include <linux/interrupt.h>
51 #include <linux/pci.h>
52 #include <linux/dma-mapping.h>
53 #include <linux/mutex.h>
54 #include <linux/list.h>
55 #include <linux/scatterlist.h>
56 #include <linux/slab.h>
57 #include <linux/io.h>
58 #include <linux/fs.h>
59 #include <linux/completion.h>
60 #include <linux/kref.h>
61 #include <linux/sched.h>
62 #include <linux/cdev.h>
63 #include <linux/delay.h>
64 #include <linux/kthread.h>
65 #include <linux/i2c.h>
66 #include <linux/i2c-algo-bit.h>
67 #include <linux/xarray.h>
68 #include <rdma/ib_hdrs.h>
69 #include <rdma/opa_addr.h>
70 #include <linux/rhashtable.h>
71 #include <linux/netdevice.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 static inline void incr_cntr32(u32 *cntr)
720 {
721 	if (*cntr < (u32)-1LL)
722 		(*cntr)++;
723 }
724 
725 #define MAX_NAME_SIZE 64
726 struct hfi1_msix_entry {
727 	enum irq_type type;
728 	int irq;
729 	void *arg;
730 	cpumask_t mask;
731 	struct irq_affinity_notify notify;
732 };
733 
734 struct hfi1_msix_info {
735 	/* lock to synchronize in_use_msix access */
736 	spinlock_t msix_lock;
737 	DECLARE_BITMAP(in_use_msix, CCE_NUM_MSIX_VECTORS);
738 	struct hfi1_msix_entry *msix_entries;
739 	u16 max_requested;
740 };
741 
742 /* per-SL CCA information */
743 struct cca_timer {
744 	struct hrtimer hrtimer;
745 	struct hfi1_pportdata *ppd; /* read-only */
746 	int sl; /* read-only */
747 	u16 ccti; /* read/write - current value of CCTI */
748 };
749 
750 struct link_down_reason {
751 	/*
752 	 * SMA-facing value.  Should be set from .latest when
753 	 * HLS_UP_* -> HLS_DN_* transition actually occurs.
754 	 */
755 	u8 sma;
756 	u8 latest;
757 };
758 
759 enum {
760 	LO_PRIO_TABLE,
761 	HI_PRIO_TABLE,
762 	MAX_PRIO_TABLE
763 };
764 
765 struct vl_arb_cache {
766 	/* protect vl arb cache */
767 	spinlock_t lock;
768 	struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
769 };
770 
771 /*
772  * The structure below encapsulates data relevant to a physical IB Port.
773  * Current chips support only one such port, but the separation
774  * clarifies things a bit. Note that to conform to IB conventions,
775  * port-numbers are one-based. The first or only port is port1.
776  */
777 struct hfi1_pportdata {
778 	struct hfi1_ibport ibport_data;
779 
780 	struct hfi1_devdata *dd;
781 	struct kobject pport_cc_kobj;
782 	struct kobject sc2vl_kobj;
783 	struct kobject sl2sc_kobj;
784 	struct kobject vl2mtu_kobj;
785 
786 	/* PHY support */
787 	struct qsfp_data qsfp_info;
788 	/* Values for SI tuning of SerDes */
789 	u32 port_type;
790 	u32 tx_preset_eq;
791 	u32 tx_preset_noeq;
792 	u32 rx_preset;
793 	u8  local_atten;
794 	u8  remote_atten;
795 	u8  default_atten;
796 	u8  max_power_class;
797 
798 	/* did we read platform config from scratch registers? */
799 	bool config_from_scratch;
800 
801 	/* GUIDs for this interface, in host order, guids[0] is a port guid */
802 	u64 guids[HFI1_GUIDS_PER_PORT];
803 
804 	/* GUID for peer interface, in host order */
805 	u64 neighbor_guid;
806 
807 	/* up or down physical link state */
808 	u32 linkup;
809 
810 	/*
811 	 * this address is mapped read-only into user processes so they can
812 	 * get status cheaply, whenever they want.  One qword of status per port
813 	 */
814 	u64 *statusp;
815 
816 	/* SendDMA related entries */
817 
818 	struct workqueue_struct *hfi1_wq;
819 	struct workqueue_struct *link_wq;
820 
821 	/* move out of interrupt context */
822 	struct work_struct link_vc_work;
823 	struct work_struct link_up_work;
824 	struct work_struct link_down_work;
825 	struct work_struct sma_message_work;
826 	struct work_struct freeze_work;
827 	struct work_struct link_downgrade_work;
828 	struct work_struct link_bounce_work;
829 	struct delayed_work start_link_work;
830 	/* host link state variables */
831 	struct mutex hls_lock;
832 	u32 host_link_state;
833 
834 	/* these are the "32 bit" regs */
835 
836 	u32 ibmtu; /* The MTU programmed for this unit */
837 	/*
838 	 * Current max size IB packet (in bytes) including IB headers, that
839 	 * we can send. Changes when ibmtu changes.
840 	 */
841 	u32 ibmaxlen;
842 	u32 current_egress_rate; /* units [10^6 bits/sec] */
843 	/* LID programmed for this instance */
844 	u32 lid;
845 	/* list of pkeys programmed; 0 if not set */
846 	u16 pkeys[MAX_PKEY_VALUES];
847 	u16 link_width_supported;
848 	u16 link_width_downgrade_supported;
849 	u16 link_speed_supported;
850 	u16 link_width_enabled;
851 	u16 link_width_downgrade_enabled;
852 	u16 link_speed_enabled;
853 	u16 link_width_active;
854 	u16 link_width_downgrade_tx_active;
855 	u16 link_width_downgrade_rx_active;
856 	u16 link_speed_active;
857 	u8 vls_supported;
858 	u8 vls_operational;
859 	u8 actual_vls_operational;
860 	/* LID mask control */
861 	u8 lmc;
862 	/* Rx Polarity inversion (compensate for ~tx on partner) */
863 	u8 rx_pol_inv;
864 
865 	u8 hw_pidx;     /* physical port index */
866 	u8 port;        /* IB port number and index into dd->pports - 1 */
867 	/* type of neighbor node */
868 	u8 neighbor_type;
869 	u8 neighbor_normal;
870 	u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
871 	u8 neighbor_port_number;
872 	u8 is_sm_config_started;
873 	u8 offline_disabled_reason;
874 	u8 is_active_optimize_enabled;
875 	u8 driver_link_ready;	/* driver ready for active link */
876 	u8 link_enabled;	/* link enabled? */
877 	u8 linkinit_reason;
878 	u8 local_tx_rate;	/* rate given to 8051 firmware */
879 	u8 qsfp_retry_count;
880 
881 	/* placeholders for IB MAD packet settings */
882 	u8 overrun_threshold;
883 	u8 phy_error_threshold;
884 	unsigned int is_link_down_queued;
885 
886 	/* Used to override LED behavior for things like maintenance beaconing*/
887 	/*
888 	 * Alternates per phase of blink
889 	 * [0] holds LED off duration, [1] holds LED on duration
890 	 */
891 	unsigned long led_override_vals[2];
892 	u8 led_override_phase; /* LSB picks from vals[] */
893 	atomic_t led_override_timer_active;
894 	/* Used to flash LEDs in override mode */
895 	struct timer_list led_override_timer;
896 
897 	u32 sm_trap_qp;
898 	u32 sa_qp;
899 
900 	/*
901 	 * cca_timer_lock protects access to the per-SL cca_timer
902 	 * structures (specifically the ccti member).
903 	 */
904 	spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
905 	struct cca_timer cca_timer[OPA_MAX_SLS];
906 
907 	/* List of congestion control table entries */
908 	struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
909 
910 	/* congestion entries, each entry corresponding to a SL */
911 	struct opa_congestion_setting_entry_shadow
912 		congestion_entries[OPA_MAX_SLS];
913 
914 	/*
915 	 * cc_state_lock protects (write) access to the per-port
916 	 * struct cc_state.
917 	 */
918 	spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
919 
920 	struct cc_state __rcu *cc_state;
921 
922 	/* Total number of congestion control table entries */
923 	u16 total_cct_entry;
924 
925 	/* Bit map identifying service level */
926 	u32 cc_sl_control_map;
927 
928 	/* CA's max number of 64 entry units in the congestion control table */
929 	u8 cc_max_table_entries;
930 
931 	/*
932 	 * begin congestion log related entries
933 	 * cc_log_lock protects all congestion log related data
934 	 */
935 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
936 	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
937 	u16 threshold_event_counter;
938 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
939 	int cc_log_idx; /* index for logging events */
940 	int cc_mad_idx; /* index for reporting events */
941 	/* end congestion log related entries */
942 
943 	struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
944 
945 	/* port relative counter buffer */
946 	u64 *cntrs;
947 	/* port relative synthetic counter buffer */
948 	u64 *scntrs;
949 	/* port_xmit_discards are synthesized from different egress errors */
950 	u64 port_xmit_discards;
951 	u64 port_xmit_discards_vl[C_VL_COUNT];
952 	u64 port_xmit_constraint_errors;
953 	u64 port_rcv_constraint_errors;
954 	/* count of 'link_err' interrupts from DC */
955 	u64 link_downed;
956 	/* number of times link retrained successfully */
957 	u64 link_up;
958 	/* number of times a link unknown frame was reported */
959 	u64 unknown_frame_count;
960 	/* port_ltp_crc_mode is returned in 'portinfo' MADs */
961 	u16 port_ltp_crc_mode;
962 	/* port_crc_mode_enabled is the crc we support */
963 	u8 port_crc_mode_enabled;
964 	/* mgmt_allowed is also returned in 'portinfo' MADs */
965 	u8 mgmt_allowed;
966 	u8 part_enforce; /* partition enforcement flags */
967 	struct link_down_reason local_link_down_reason;
968 	struct link_down_reason neigh_link_down_reason;
969 	/* Value to be sent to link peer on LinkDown .*/
970 	u8 remote_link_down_reason;
971 	/* Error events that will cause a port bounce. */
972 	u32 port_error_action;
973 	struct work_struct linkstate_active_work;
974 	/* Does this port need to prescan for FECNs */
975 	bool cc_prescan;
976 	/*
977 	 * Sample sendWaitCnt & sendWaitVlCnt during link transition
978 	 * and counter request.
979 	 */
980 	u64 port_vl_xmit_wait_last[C_VL_COUNT + 1];
981 	u16 prev_link_width;
982 	u64 vl_xmit_flit_cnt[C_VL_COUNT + 1];
983 };
984 
985 typedef void (*opcode_handler)(struct hfi1_packet *packet);
986 typedef void (*hfi1_make_req)(struct rvt_qp *qp,
987 			      struct hfi1_pkt_state *ps,
988 			      struct rvt_swqe *wqe);
989 extern const rhf_rcv_function_ptr normal_rhf_rcv_functions[];
990 extern const rhf_rcv_function_ptr netdev_rhf_rcv_functions[];
991 
992 /* return values for the RHF receive functions */
993 #define RHF_RCV_CONTINUE  0	/* keep going */
994 #define RHF_RCV_DONE	  1	/* stop, this packet processed */
995 #define RHF_RCV_REPROCESS 2	/* stop. retain this packet */
996 
997 struct rcv_array_data {
998 	u16 ngroups;
999 	u16 nctxt_extra;
1000 	u8 group_size;
1001 };
1002 
1003 struct per_vl_data {
1004 	u16 mtu;
1005 	struct send_context *sc;
1006 };
1007 
1008 /* 16 to directly index */
1009 #define PER_VL_SEND_CONTEXTS 16
1010 
1011 struct err_info_rcvport {
1012 	u8 status_and_code;
1013 	u64 packet_flit1;
1014 	u64 packet_flit2;
1015 };
1016 
1017 struct err_info_constraint {
1018 	u8 status;
1019 	u16 pkey;
1020 	u32 slid;
1021 };
1022 
1023 struct hfi1_temp {
1024 	unsigned int curr;       /* current temperature */
1025 	unsigned int lo_lim;     /* low temperature limit */
1026 	unsigned int hi_lim;     /* high temperature limit */
1027 	unsigned int crit_lim;   /* critical temperature limit */
1028 	u8 triggers;      /* temperature triggers */
1029 };
1030 
1031 struct hfi1_i2c_bus {
1032 	struct hfi1_devdata *controlling_dd; /* current controlling device */
1033 	struct i2c_adapter adapter;	/* bus details */
1034 	struct i2c_algo_bit_data algo;	/* bus algorithm details */
1035 	int num;			/* bus number, 0 or 1 */
1036 };
1037 
1038 /* common data between shared ASIC HFIs */
1039 struct hfi1_asic_data {
1040 	struct hfi1_devdata *dds[2];	/* back pointers */
1041 	struct mutex asic_resource_mutex;
1042 	struct hfi1_i2c_bus *i2c_bus0;
1043 	struct hfi1_i2c_bus *i2c_bus1;
1044 };
1045 
1046 /* sizes for both the QP and RSM map tables */
1047 #define NUM_MAP_ENTRIES	 256
1048 #define NUM_MAP_REGS      32
1049 
1050 /* Virtual NIC information */
1051 struct hfi1_vnic_data {
1052 	struct kmem_cache *txreq_cache;
1053 	u8 num_vports;
1054 };
1055 
1056 struct hfi1_vnic_vport_info;
1057 
1058 /* device data struct now contains only "general per-device" info.
1059  * fields related to a physical IB port are in a hfi1_pportdata struct.
1060  */
1061 struct sdma_engine;
1062 struct sdma_vl_map;
1063 
1064 #define BOARD_VERS_MAX 96 /* how long the version string can be */
1065 #define SERIAL_MAX 16 /* length of the serial number */
1066 
1067 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
1068 struct hfi1_devdata {
1069 	struct hfi1_ibdev verbs_dev;     /* must be first */
1070 	/* pointers to related structs for this device */
1071 	/* pci access data structure */
1072 	struct pci_dev *pcidev;
1073 	struct cdev user_cdev;
1074 	struct cdev diag_cdev;
1075 	struct cdev ui_cdev;
1076 	struct device *user_device;
1077 	struct device *diag_device;
1078 	struct device *ui_device;
1079 
1080 	/* first mapping up to RcvArray */
1081 	u8 __iomem *kregbase1;
1082 	resource_size_t physaddr;
1083 
1084 	/* second uncached mapping from RcvArray to pio send buffers */
1085 	u8 __iomem *kregbase2;
1086 	/* for detecting offset above kregbase2 address */
1087 	u32 base2_start;
1088 
1089 	/* Per VL data. Enough for all VLs but not all elements are set/used. */
1090 	struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
1091 	/* send context data */
1092 	struct send_context_info *send_contexts;
1093 	/* map hardware send contexts to software index */
1094 	u8 *hw_to_sw;
1095 	/* spinlock for allocating and releasing send context resources */
1096 	spinlock_t sc_lock;
1097 	/* lock for pio_map */
1098 	spinlock_t pio_map_lock;
1099 	/* Send Context initialization lock. */
1100 	spinlock_t sc_init_lock;
1101 	/* lock for sdma_map */
1102 	spinlock_t                          sde_map_lock;
1103 	/* array of kernel send contexts */
1104 	struct send_context **kernel_send_context;
1105 	/* array of vl maps */
1106 	struct pio_vl_map __rcu *pio_map;
1107 	/* default flags to last descriptor */
1108 	u64 default_desc1;
1109 
1110 	/* fields common to all SDMA engines */
1111 
1112 	volatile __le64                    *sdma_heads_dma; /* DMA'ed by chip */
1113 	dma_addr_t                          sdma_heads_phys;
1114 	void                               *sdma_pad_dma; /* DMA'ed by chip */
1115 	dma_addr_t                          sdma_pad_phys;
1116 	/* for deallocation */
1117 	size_t                              sdma_heads_size;
1118 	/* num used */
1119 	u32                                 num_sdma;
1120 	/* array of engines sized by num_sdma */
1121 	struct sdma_engine                 *per_sdma;
1122 	/* array of vl maps */
1123 	struct sdma_vl_map __rcu           *sdma_map;
1124 	/* SPC freeze waitqueue and variable */
1125 	wait_queue_head_t		  sdma_unfreeze_wq;
1126 	atomic_t			  sdma_unfreeze_count;
1127 
1128 	u32 lcb_access_count;		/* count of LCB users */
1129 
1130 	/* common data between shared ASIC HFIs in this OS */
1131 	struct hfi1_asic_data *asic_data;
1132 
1133 	/* mem-mapped pointer to base of PIO buffers */
1134 	void __iomem *piobase;
1135 	/*
1136 	 * write-combining mem-mapped pointer to base of RcvArray
1137 	 * memory.
1138 	 */
1139 	void __iomem *rcvarray_wc;
1140 	/*
1141 	 * credit return base - a per-NUMA range of DMA address that
1142 	 * the chip will use to update the per-context free counter
1143 	 */
1144 	struct credit_return_base *cr_base;
1145 
1146 	/* send context numbers and sizes for each type */
1147 	struct sc_config_sizes sc_sizes[SC_MAX];
1148 
1149 	char *boardname; /* human readable board info */
1150 
1151 	u64 ctx0_seq_drop;
1152 
1153 	/* reset value */
1154 	u64 z_int_counter;
1155 	u64 z_rcv_limit;
1156 	u64 z_send_schedule;
1157 
1158 	u64 __percpu *send_schedule;
1159 	/* number of reserved contexts for netdev usage */
1160 	u16 num_netdev_contexts;
1161 	/* number of receive contexts in use by the driver */
1162 	u32 num_rcv_contexts;
1163 	/* number of pio send contexts in use by the driver */
1164 	u32 num_send_contexts;
1165 	/*
1166 	 * number of ctxts available for PSM open
1167 	 */
1168 	u32 freectxts;
1169 	/* total number of available user/PSM contexts */
1170 	u32 num_user_contexts;
1171 	/* base receive interrupt timeout, in CSR units */
1172 	u32 rcv_intr_timeout_csr;
1173 
1174 	spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
1175 	spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
1176 	spinlock_t uctxt_lock; /* protect rcd changes */
1177 	struct mutex dc8051_lock; /* exclusive access to 8051 */
1178 	struct workqueue_struct *update_cntr_wq;
1179 	struct work_struct update_cntr_work;
1180 	/* exclusive access to 8051 memory */
1181 	spinlock_t dc8051_memlock;
1182 	int dc8051_timed_out;	/* remember if the 8051 timed out */
1183 	/*
1184 	 * A page that will hold event notification bitmaps for all
1185 	 * contexts. This page will be mapped into all processes.
1186 	 */
1187 	unsigned long *events;
1188 	/*
1189 	 * per unit status, see also portdata statusp
1190 	 * mapped read-only into user processes so they can get unit and
1191 	 * IB link status cheaply
1192 	 */
1193 	struct hfi1_status *status;
1194 
1195 	/* revision register shadow */
1196 	u64 revision;
1197 	/* Base GUID for device (network order) */
1198 	u64 base_guid;
1199 
1200 	/* both sides of the PCIe link are gen3 capable */
1201 	u8 link_gen3_capable;
1202 	u8 dc_shutdown;
1203 	/* localbus width (1, 2,4,8,16,32) from config space  */
1204 	u32 lbus_width;
1205 	/* localbus speed in MHz */
1206 	u32 lbus_speed;
1207 	int unit; /* unit # of this chip */
1208 	int node; /* home node of this chip */
1209 
1210 	/* save these PCI fields to restore after a reset */
1211 	u32 pcibar0;
1212 	u32 pcibar1;
1213 	u32 pci_rom;
1214 	u16 pci_command;
1215 	u16 pcie_devctl;
1216 	u16 pcie_lnkctl;
1217 	u16 pcie_devctl2;
1218 	u32 pci_msix0;
1219 	u32 pci_tph2;
1220 
1221 	/*
1222 	 * ASCII serial number, from flash, large enough for original
1223 	 * all digit strings, and longer serial number format
1224 	 */
1225 	u8 serial[SERIAL_MAX];
1226 	/* human readable board version */
1227 	u8 boardversion[BOARD_VERS_MAX];
1228 	u8 lbus_info[32]; /* human readable localbus info */
1229 	/* chip major rev, from CceRevision */
1230 	u8 majrev;
1231 	/* chip minor rev, from CceRevision */
1232 	u8 minrev;
1233 	/* hardware ID */
1234 	u8 hfi1_id;
1235 	/* implementation code */
1236 	u8 icode;
1237 	/* vAU of this device */
1238 	u8 vau;
1239 	/* vCU of this device */
1240 	u8 vcu;
1241 	/* link credits of this device */
1242 	u16 link_credits;
1243 	/* initial vl15 credits to use */
1244 	u16 vl15_init;
1245 
1246 	/*
1247 	 * Cached value for vl15buf, read during verify cap interrupt. VL15
1248 	 * credits are to be kept at 0 and set when handling the link-up
1249 	 * interrupt. This removes the possibility of receiving VL15 MAD
1250 	 * packets before this HFI is ready.
1251 	 */
1252 	u16 vl15buf_cached;
1253 
1254 	/* Misc small ints */
1255 	u8 n_krcv_queues;
1256 	u8 qos_shift;
1257 
1258 	u16 irev;	/* implementation revision */
1259 	u32 dc8051_ver; /* 8051 firmware version */
1260 
1261 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1262 	struct platform_config platform_config;
1263 	struct platform_config_cache pcfg_cache;
1264 
1265 	struct diag_client *diag_client;
1266 
1267 	/* general interrupt: mask of handled interrupts */
1268 	u64 gi_mask[CCE_NUM_INT_CSRS];
1269 
1270 	struct rcv_array_data rcv_entries;
1271 
1272 	/* cycle length of PS* counters in HW (in picoseconds) */
1273 	u16 psxmitwait_check_rate;
1274 
1275 	/*
1276 	 * 64 bit synthetic counters
1277 	 */
1278 	struct timer_list synth_stats_timer;
1279 
1280 	/* MSI-X information */
1281 	struct hfi1_msix_info msix_info;
1282 
1283 	/*
1284 	 * device counters
1285 	 */
1286 	char *cntrnames;
1287 	size_t cntrnameslen;
1288 	size_t ndevcntrs;
1289 	u64 *cntrs;
1290 	u64 *scntrs;
1291 
1292 	/*
1293 	 * remembered values for synthetic counters
1294 	 */
1295 	u64 last_tx;
1296 	u64 last_rx;
1297 
1298 	/*
1299 	 * per-port counters
1300 	 */
1301 	size_t nportcntrs;
1302 	char *portcntrnames;
1303 	size_t portcntrnameslen;
1304 
1305 	struct err_info_rcvport err_info_rcvport;
1306 	struct err_info_constraint err_info_rcv_constraint;
1307 	struct err_info_constraint err_info_xmit_constraint;
1308 
1309 	atomic_t drop_packet;
1310 	bool do_drop;
1311 	u8 err_info_uncorrectable;
1312 	u8 err_info_fmconfig;
1313 
1314 	/*
1315 	 * Software counters for the status bits defined by the
1316 	 * associated error status registers
1317 	 */
1318 	u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1319 	u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1320 	u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1321 	u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1322 	u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1323 	u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1324 	u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1325 
1326 	/* Software counter that spans all contexts */
1327 	u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1328 	/* Software counter that spans all DMA engines */
1329 	u64 sw_send_dma_eng_err_status_cnt[
1330 		NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1331 	/* Software counter that aggregates all cce_err_status errors */
1332 	u64 sw_cce_err_status_aggregate;
1333 	/* Software counter that aggregates all bypass packet rcv errors */
1334 	u64 sw_rcv_bypass_packet_errors;
1335 
1336 	/* Save the enabled LCB error bits */
1337 	u64 lcb_err_en;
1338 	struct cpu_mask_set *comp_vect;
1339 	int *comp_vect_mappings;
1340 	u32 comp_vect_possible_cpus;
1341 
1342 	/*
1343 	 * Capability to have different send engines simply by changing a
1344 	 * pointer value.
1345 	 */
1346 	send_routine process_pio_send ____cacheline_aligned_in_smp;
1347 	send_routine process_dma_send;
1348 	void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1349 				u64 pbc, const void *from, size_t count);
1350 	int (*process_vnic_dma_send)(struct hfi1_devdata *dd, u8 q_idx,
1351 				     struct hfi1_vnic_vport_info *vinfo,
1352 				     struct sk_buff *skb, u64 pbc, u8 plen);
1353 	/* hfi1_pportdata, points to array of (physical) port-specific
1354 	 * data structs, indexed by pidx (0..n-1)
1355 	 */
1356 	struct hfi1_pportdata *pport;
1357 	/* receive context data */
1358 	struct hfi1_ctxtdata **rcd;
1359 	u64 __percpu *int_counter;
1360 	/* verbs tx opcode stats */
1361 	struct hfi1_opcode_stats_perctx __percpu *tx_opstats;
1362 	/* device (not port) flags, basically device capabilities */
1363 	u16 flags;
1364 	/* Number of physical ports available */
1365 	u8 num_pports;
1366 	/* Lowest context number which can be used by user processes or VNIC */
1367 	u8 first_dyn_alloc_ctxt;
1368 	/* adding a new field here would make it part of this cacheline */
1369 
1370 	/* seqlock for sc2vl */
1371 	seqlock_t sc2vl_lock ____cacheline_aligned_in_smp;
1372 	u64 sc2vl[4];
1373 	u64 __percpu *rcv_limit;
1374 	/* adding a new field here would make it part of this cacheline */
1375 
1376 	/* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1377 	u8 oui1;
1378 	u8 oui2;
1379 	u8 oui3;
1380 
1381 	/* Timer and counter used to detect RcvBufOvflCnt changes */
1382 	struct timer_list rcverr_timer;
1383 
1384 	wait_queue_head_t event_queue;
1385 
1386 	/* receive context tail dummy address */
1387 	__le64 *rcvhdrtail_dummy_kvaddr;
1388 	dma_addr_t rcvhdrtail_dummy_dma;
1389 
1390 	u32 rcv_ovfl_cnt;
1391 	/* Serialize ASPM enable/disable between multiple verbs contexts */
1392 	spinlock_t aspm_lock;
1393 	/* Number of verbs contexts which have disabled ASPM */
1394 	atomic_t aspm_disabled_cnt;
1395 	/* Keeps track of user space clients */
1396 	atomic_t user_refcount;
1397 	/* Used to wait for outstanding user space clients before dev removal */
1398 	struct completion user_comp;
1399 
1400 	bool eprom_available;	/* true if EPROM is available for this device */
1401 	bool aspm_supported;	/* Does HW support ASPM */
1402 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
1403 	struct rhashtable *sdma_rht;
1404 
1405 	/* vnic data */
1406 	struct hfi1_vnic_data vnic;
1407 	/* Lock to protect IRQ SRC register access */
1408 	spinlock_t irq_src_lock;
1409 	int vnic_num_vports;
1410 	struct net_device *dummy_netdev;
1411 
1412 	/* Keeps track of IPoIB RSM rule users */
1413 	atomic_t ipoib_rsm_usr_num;
1414 };
1415 
1416 /* 8051 firmware version helper */
1417 #define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
1418 #define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
1419 #define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
1420 #define dc8051_ver_patch(a) ((a) & 0x0000ff)
1421 
1422 /* f_put_tid types */
1423 #define PT_EXPECTED       0
1424 #define PT_EAGER          1
1425 #define PT_INVALID_FLUSH  2
1426 #define PT_INVALID        3
1427 
1428 struct tid_rb_node;
1429 struct mmu_rb_node;
1430 struct mmu_rb_handler;
1431 
1432 /* Private data for file operations */
1433 struct hfi1_filedata {
1434 	struct srcu_struct pq_srcu;
1435 	struct hfi1_devdata *dd;
1436 	struct hfi1_ctxtdata *uctxt;
1437 	struct hfi1_user_sdma_comp_q *cq;
1438 	/* update side lock for SRCU */
1439 	spinlock_t pq_rcu_lock;
1440 	struct hfi1_user_sdma_pkt_q __rcu *pq;
1441 	u16 subctxt;
1442 	/* for cpu affinity; -1 if none */
1443 	int rec_cpu_num;
1444 	u32 tid_n_pinned;
1445 	bool use_mn;
1446 	struct tid_rb_node **entry_to_rb;
1447 	spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1448 	u32 tid_limit;
1449 	u32 tid_used;
1450 	u32 *invalid_tids;
1451 	u32 invalid_tid_idx;
1452 	/* protect invalid_tids array and invalid_tid_idx */
1453 	spinlock_t invalid_lock;
1454 	struct mm_struct *mm;
1455 };
1456 
1457 extern struct xarray hfi1_dev_table;
1458 struct hfi1_devdata *hfi1_lookup(int unit);
1459 
1460 static inline unsigned long uctxt_offset(struct hfi1_ctxtdata *uctxt)
1461 {
1462 	return (uctxt->ctxt - uctxt->dd->first_dyn_alloc_ctxt) *
1463 		HFI1_MAX_SHARED_CTXTS;
1464 }
1465 
1466 int hfi1_init(struct hfi1_devdata *dd, int reinit);
1467 int hfi1_count_active_units(void);
1468 
1469 int hfi1_diag_add(struct hfi1_devdata *dd);
1470 void hfi1_diag_remove(struct hfi1_devdata *dd);
1471 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1472 
1473 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1474 
1475 int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1476 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
1477 int hfi1_create_kctxts(struct hfi1_devdata *dd);
1478 int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa,
1479 			 struct hfi1_ctxtdata **rcd);
1480 void hfi1_free_ctxt(struct hfi1_ctxtdata *rcd);
1481 void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
1482 			 struct hfi1_devdata *dd, u8 hw_pidx, u8 port);
1483 void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1484 int hfi1_rcd_put(struct hfi1_ctxtdata *rcd);
1485 int hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
1486 struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd,
1487 						 u16 ctxt);
1488 struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt);
1489 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
1490 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1491 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1492 int handle_receive_interrupt_napi_fp(struct hfi1_ctxtdata *rcd, int budget);
1493 int handle_receive_interrupt_napi_sp(struct hfi1_ctxtdata *rcd, int budget);
1494 void set_all_slowpath(struct hfi1_devdata *dd);
1495 
1496 extern const struct pci_device_id hfi1_pci_tbl[];
1497 void hfi1_make_ud_req_9B(struct rvt_qp *qp,
1498 			 struct hfi1_pkt_state *ps,
1499 			 struct rvt_swqe *wqe);
1500 
1501 void hfi1_make_ud_req_16B(struct rvt_qp *qp,
1502 			  struct hfi1_pkt_state *ps,
1503 			  struct rvt_swqe *wqe);
1504 
1505 /* receive packet handler dispositions */
1506 #define RCV_PKT_OK      0x0 /* keep going */
1507 #define RCV_PKT_LIMIT   0x1 /* stop, hit limit, start thread */
1508 #define RCV_PKT_DONE    0x2 /* stop, no more packets detected */
1509 
1510 /**
1511  * hfi1_rcd_head - add accessor for rcd head
1512  * @rcd: the context
1513  */
1514 static inline u32 hfi1_rcd_head(struct hfi1_ctxtdata *rcd)
1515 {
1516 	return rcd->head;
1517 }
1518 
1519 /**
1520  * hfi1_set_rcd_head - add accessor for rcd head
1521  * @rcd: the context
1522  * @head: the new head
1523  */
1524 static inline void hfi1_set_rcd_head(struct hfi1_ctxtdata *rcd, u32 head)
1525 {
1526 	rcd->head = head;
1527 }
1528 
1529 /* calculate the current RHF address */
1530 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1531 {
1532 	return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->rhf_offset;
1533 }
1534 
1535 /* return DMA_RTAIL configuration */
1536 static inline bool get_dma_rtail_setting(struct hfi1_ctxtdata *rcd)
1537 {
1538 	return !!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL);
1539 }
1540 
1541 /**
1542  * hfi1_seq_incr_wrap - wrapping increment for sequence
1543  * @seq: the current sequence number
1544  *
1545  * Returns: the incremented seq
1546  */
1547 static inline u8 hfi1_seq_incr_wrap(u8 seq)
1548 {
1549 	if (++seq > RHF_MAX_SEQ)
1550 		seq = 1;
1551 	return seq;
1552 }
1553 
1554 /**
1555  * hfi1_seq_cnt - return seq_cnt member
1556  * @rcd: the receive context
1557  *
1558  * Return seq_cnt member
1559  */
1560 static inline u8 hfi1_seq_cnt(struct hfi1_ctxtdata *rcd)
1561 {
1562 	return rcd->seq_cnt;
1563 }
1564 
1565 /**
1566  * hfi1_set_seq_cnt - return seq_cnt member
1567  * @rcd: the receive context
1568  *
1569  * Return seq_cnt member
1570  */
1571 static inline void hfi1_set_seq_cnt(struct hfi1_ctxtdata *rcd, u8 cnt)
1572 {
1573 	rcd->seq_cnt = cnt;
1574 }
1575 
1576 /**
1577  * last_rcv_seq - is last
1578  * @rcd: the receive context
1579  * @seq: sequence
1580  *
1581  * return true if last packet
1582  */
1583 static inline bool last_rcv_seq(struct hfi1_ctxtdata *rcd, u32 seq)
1584 {
1585 	return seq != rcd->seq_cnt;
1586 }
1587 
1588 /**
1589  * rcd_seq_incr - increment context sequence number
1590  * @rcd: the receive context
1591  * @seq: the current sequence number
1592  *
1593  * Returns: true if the this was the last packet
1594  */
1595 static inline bool hfi1_seq_incr(struct hfi1_ctxtdata *rcd, u32 seq)
1596 {
1597 	rcd->seq_cnt = hfi1_seq_incr_wrap(rcd->seq_cnt);
1598 	return last_rcv_seq(rcd, seq);
1599 }
1600 
1601 /**
1602  * get_hdrqentsize - return hdrq entry size
1603  * @rcd: the receive context
1604  */
1605 static inline u8 get_hdrqentsize(struct hfi1_ctxtdata *rcd)
1606 {
1607 	return rcd->rcvhdrqentsize;
1608 }
1609 
1610 /**
1611  * get_hdrq_cnt - return hdrq count
1612  * @rcd: the receive context
1613  */
1614 static inline u16 get_hdrq_cnt(struct hfi1_ctxtdata *rcd)
1615 {
1616 	return rcd->rcvhdrq_cnt;
1617 }
1618 
1619 /**
1620  * hfi1_is_slowpath - check if this context is slow path
1621  * @rcd: the receive context
1622  */
1623 static inline bool hfi1_is_slowpath(struct hfi1_ctxtdata *rcd)
1624 {
1625 	return rcd->do_interrupt == rcd->slow_handler;
1626 }
1627 
1628 /**
1629  * hfi1_is_fastpath - check if this context is fast path
1630  * @rcd: the receive context
1631  */
1632 static inline bool hfi1_is_fastpath(struct hfi1_ctxtdata *rcd)
1633 {
1634 	if (rcd->ctxt == HFI1_CTRL_CTXT)
1635 		return false;
1636 
1637 	return rcd->do_interrupt == rcd->fast_handler;
1638 }
1639 
1640 /**
1641  * hfi1_set_fast - change to the fast handler
1642  * @rcd: the receive context
1643  */
1644 static inline void hfi1_set_fast(struct hfi1_ctxtdata *rcd)
1645 {
1646 	if (unlikely(!rcd))
1647 		return;
1648 	if (unlikely(!hfi1_is_fastpath(rcd)))
1649 		rcd->do_interrupt = rcd->fast_handler;
1650 }
1651 
1652 int hfi1_reset_device(int);
1653 
1654 void receive_interrupt_work(struct work_struct *work);
1655 
1656 /* extract service channel from header and rhf */
1657 static inline int hfi1_9B_get_sc5(struct ib_header *hdr, u64 rhf)
1658 {
1659 	return ib_get_sc(hdr) | ((!!(rhf_dc_info(rhf))) << 4);
1660 }
1661 
1662 #define HFI1_JKEY_WIDTH       16
1663 #define HFI1_JKEY_MASK        (BIT(16) - 1)
1664 #define HFI1_ADMIN_JKEY_RANGE 32
1665 
1666 /*
1667  * J_KEYs are split and allocated in the following groups:
1668  *   0 - 31    - users with administrator privileges
1669  *  32 - 63    - kernel protocols using KDETH packets
1670  *  64 - 65535 - all other users using KDETH packets
1671  */
1672 static inline u16 generate_jkey(kuid_t uid)
1673 {
1674 	u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1675 
1676 	if (capable(CAP_SYS_ADMIN))
1677 		jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1678 	else if (jkey < 64)
1679 		jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1680 
1681 	return jkey;
1682 }
1683 
1684 /*
1685  * active_egress_rate
1686  *
1687  * returns the active egress rate in units of [10^6 bits/sec]
1688  */
1689 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1690 {
1691 	u16 link_speed = ppd->link_speed_active;
1692 	u16 link_width = ppd->link_width_active;
1693 	u32 egress_rate;
1694 
1695 	if (link_speed == OPA_LINK_SPEED_25G)
1696 		egress_rate = 25000;
1697 	else /* assume OPA_LINK_SPEED_12_5G */
1698 		egress_rate = 12500;
1699 
1700 	switch (link_width) {
1701 	case OPA_LINK_WIDTH_4X:
1702 		egress_rate *= 4;
1703 		break;
1704 	case OPA_LINK_WIDTH_3X:
1705 		egress_rate *= 3;
1706 		break;
1707 	case OPA_LINK_WIDTH_2X:
1708 		egress_rate *= 2;
1709 		break;
1710 	default:
1711 		/* assume IB_WIDTH_1X */
1712 		break;
1713 	}
1714 
1715 	return egress_rate;
1716 }
1717 
1718 /*
1719  * egress_cycles
1720  *
1721  * Returns the number of 'fabric clock cycles' to egress a packet
1722  * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1723  * rate is (approximately) 805 MHz, the units of the returned value
1724  * are (1/805 MHz).
1725  */
1726 static inline u32 egress_cycles(u32 len, u32 rate)
1727 {
1728 	u32 cycles;
1729 
1730 	/*
1731 	 * cycles is:
1732 	 *
1733 	 *          (length) [bits] / (rate) [bits/sec]
1734 	 *  ---------------------------------------------------
1735 	 *  fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1736 	 */
1737 
1738 	cycles = len * 8; /* bits */
1739 	cycles *= 805;
1740 	cycles /= rate;
1741 
1742 	return cycles;
1743 }
1744 
1745 void set_link_ipg(struct hfi1_pportdata *ppd);
1746 void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
1747 		  u32 rqpn, u8 svc_type);
1748 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1749 		u16 pkey, u32 slid, u32 dlid, u8 sc5,
1750 		const struct ib_grh *old_grh);
1751 void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1752 		    u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1753 		    u8 sc5, const struct ib_grh *old_grh);
1754 typedef void (*hfi1_handle_cnp)(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1755 				u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1756 				u8 sc5, const struct ib_grh *old_grh);
1757 
1758 #define PKEY_CHECK_INVALID -1
1759 int egress_pkey_check(struct hfi1_pportdata *ppd, u32 slid, u16 pkey,
1760 		      u8 sc5, int8_t s_pkey_index);
1761 
1762 #define PACKET_EGRESS_TIMEOUT 350
1763 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1764 {
1765 	/* Pause at least 1us, to ensure chip returns all credits */
1766 	u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1767 
1768 	udelay(usec ? usec : 1);
1769 }
1770 
1771 /**
1772  * sc_to_vlt() reverse lookup sc to vl
1773  * @dd - devdata
1774  * @sc5 - 5 bit sc
1775  */
1776 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1777 {
1778 	unsigned seq;
1779 	u8 rval;
1780 
1781 	if (sc5 >= OPA_MAX_SCS)
1782 		return (u8)(0xff);
1783 
1784 	do {
1785 		seq = read_seqbegin(&dd->sc2vl_lock);
1786 		rval = *(((u8 *)dd->sc2vl) + sc5);
1787 	} while (read_seqretry(&dd->sc2vl_lock, seq));
1788 
1789 	return rval;
1790 }
1791 
1792 #define PKEY_MEMBER_MASK 0x8000
1793 #define PKEY_LOW_15_MASK 0x7fff
1794 
1795 /*
1796  * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1797  * being an entry from the ingress partition key table), return 0
1798  * otherwise. Use the matching criteria for ingress partition keys
1799  * specified in the OPAv1 spec., section 9.10.14.
1800  */
1801 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1802 {
1803 	u16 mkey = pkey & PKEY_LOW_15_MASK;
1804 	u16 ment = ent & PKEY_LOW_15_MASK;
1805 
1806 	if (mkey == ment) {
1807 		/*
1808 		 * If pkey[15] is clear (limited partition member),
1809 		 * is bit 15 in the corresponding table element
1810 		 * clear (limited member)?
1811 		 */
1812 		if (!(pkey & PKEY_MEMBER_MASK))
1813 			return !!(ent & PKEY_MEMBER_MASK);
1814 		return 1;
1815 	}
1816 	return 0;
1817 }
1818 
1819 /*
1820  * ingress_pkey_table_search - search the entire pkey table for
1821  * an entry which matches 'pkey'. return 0 if a match is found,
1822  * and 1 otherwise.
1823  */
1824 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1825 {
1826 	int i;
1827 
1828 	for (i = 0; i < MAX_PKEY_VALUES; i++) {
1829 		if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1830 			return 0;
1831 	}
1832 	return 1;
1833 }
1834 
1835 /*
1836  * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1837  * i.e., increment port_rcv_constraint_errors for the port, and record
1838  * the 'error info' for this failure.
1839  */
1840 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1841 				    u32 slid)
1842 {
1843 	struct hfi1_devdata *dd = ppd->dd;
1844 
1845 	incr_cntr64(&ppd->port_rcv_constraint_errors);
1846 	if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1847 		dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1848 		dd->err_info_rcv_constraint.slid = slid;
1849 		dd->err_info_rcv_constraint.pkey = pkey;
1850 	}
1851 }
1852 
1853 /*
1854  * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1855  * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1856  * is a hint as to the best place in the partition key table to begin
1857  * searching. This function should not be called on the data path because
1858  * of performance reasons. On datapath pkey check is expected to be done
1859  * by HW and rcv_pkey_check function should be called instead.
1860  */
1861 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1862 				     u8 sc5, u8 idx, u32 slid, bool force)
1863 {
1864 	if (!(force) && !(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1865 		return 0;
1866 
1867 	/* If SC15, pkey[0:14] must be 0x7fff */
1868 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1869 		goto bad;
1870 
1871 	/* Is the pkey = 0x0, or 0x8000? */
1872 	if ((pkey & PKEY_LOW_15_MASK) == 0)
1873 		goto bad;
1874 
1875 	/* The most likely matching pkey has index 'idx' */
1876 	if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1877 		return 0;
1878 
1879 	/* no match - try the whole table */
1880 	if (!ingress_pkey_table_search(ppd, pkey))
1881 		return 0;
1882 
1883 bad:
1884 	ingress_pkey_table_fail(ppd, pkey, slid);
1885 	return 1;
1886 }
1887 
1888 /*
1889  * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1890  * otherwise. It only ensures pkey is vlid for QP0. This function
1891  * should be called on the data path instead of ingress_pkey_check
1892  * as on data path, pkey check is done by HW (except for QP0).
1893  */
1894 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1895 				 u8 sc5, u16 slid)
1896 {
1897 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1898 		return 0;
1899 
1900 	/* If SC15, pkey[0:14] must be 0x7fff */
1901 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1902 		goto bad;
1903 
1904 	return 0;
1905 bad:
1906 	ingress_pkey_table_fail(ppd, pkey, slid);
1907 	return 1;
1908 }
1909 
1910 /* MTU handling */
1911 
1912 /* MTU enumeration, 256-4k match IB */
1913 #define OPA_MTU_0     0
1914 #define OPA_MTU_256   1
1915 #define OPA_MTU_512   2
1916 #define OPA_MTU_1024  3
1917 #define OPA_MTU_2048  4
1918 #define OPA_MTU_4096  5
1919 
1920 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1921 int mtu_to_enum(u32 mtu, int default_if_bad);
1922 u16 enum_to_mtu(int mtu);
1923 static inline int valid_ib_mtu(unsigned int mtu)
1924 {
1925 	return mtu == 256 || mtu == 512 ||
1926 		mtu == 1024 || mtu == 2048 ||
1927 		mtu == 4096;
1928 }
1929 
1930 static inline int valid_opa_max_mtu(unsigned int mtu)
1931 {
1932 	return mtu >= 2048 &&
1933 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1934 }
1935 
1936 int set_mtu(struct hfi1_pportdata *ppd);
1937 
1938 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
1939 void hfi1_disable_after_error(struct hfi1_devdata *dd);
1940 int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
1941 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
1942 
1943 int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
1944 int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
1945 
1946 void set_up_vau(struct hfi1_devdata *dd, u8 vau);
1947 void set_up_vl15(struct hfi1_devdata *dd, u16 vl15buf);
1948 void reset_link_credits(struct hfi1_devdata *dd);
1949 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1950 
1951 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1952 
1953 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1954 {
1955 	return ppd->dd;
1956 }
1957 
1958 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1959 {
1960 	return container_of(dev, struct hfi1_devdata, verbs_dev);
1961 }
1962 
1963 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1964 {
1965 	return dd_from_dev(to_idev(ibdev));
1966 }
1967 
1968 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1969 {
1970 	return container_of(ibp, struct hfi1_pportdata, ibport_data);
1971 }
1972 
1973 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1974 {
1975 	return container_of(rdi, struct hfi1_ibdev, rdi);
1976 }
1977 
1978 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u8 port)
1979 {
1980 	struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1981 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1982 
1983 	WARN_ON(pidx >= dd->num_pports);
1984 	return &dd->pport[pidx].ibport_data;
1985 }
1986 
1987 static inline struct hfi1_ibport *rcd_to_iport(struct hfi1_ctxtdata *rcd)
1988 {
1989 	return &rcd->ppd->ibport_data;
1990 }
1991 
1992 /**
1993  * hfi1_may_ecn - Check whether FECN or BECN processing should be done
1994  * @pkt: the packet to be evaluated
1995  *
1996  * Check whether the FECN or BECN bits in the packet's header are
1997  * enabled, depending on packet type.
1998  *
1999  * This function only checks for FECN and BECN bits. Additional checks
2000  * are done in the slowpath (hfi1_process_ecn_slowpath()) in order to
2001  * ensure correct handling.
2002  */
2003 static inline bool hfi1_may_ecn(struct hfi1_packet *pkt)
2004 {
2005 	bool fecn, becn;
2006 
2007 	if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
2008 		fecn = hfi1_16B_get_fecn(pkt->hdr);
2009 		becn = hfi1_16B_get_becn(pkt->hdr);
2010 	} else {
2011 		fecn = ib_bth_get_fecn(pkt->ohdr);
2012 		becn = ib_bth_get_becn(pkt->ohdr);
2013 	}
2014 	return fecn || becn;
2015 }
2016 
2017 bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
2018 			       bool prescan);
2019 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt)
2020 {
2021 	bool do_work;
2022 
2023 	do_work = hfi1_may_ecn(pkt);
2024 	if (unlikely(do_work))
2025 		return hfi1_process_ecn_slowpath(qp, pkt, false);
2026 	return false;
2027 }
2028 
2029 /*
2030  * Return the indexed PKEY from the port PKEY table.
2031  */
2032 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
2033 {
2034 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2035 	u16 ret;
2036 
2037 	if (index >= ARRAY_SIZE(ppd->pkeys))
2038 		ret = 0;
2039 	else
2040 		ret = ppd->pkeys[index];
2041 
2042 	return ret;
2043 }
2044 
2045 /*
2046  * Return the indexed GUID from the port GUIDs table.
2047  */
2048 static inline __be64 get_sguid(struct hfi1_ibport *ibp, unsigned int index)
2049 {
2050 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2051 
2052 	WARN_ON(index >= HFI1_GUIDS_PER_PORT);
2053 	return cpu_to_be64(ppd->guids[index]);
2054 }
2055 
2056 /*
2057  * Called by readers of cc_state only, must call under rcu_read_lock().
2058  */
2059 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
2060 {
2061 	return rcu_dereference(ppd->cc_state);
2062 }
2063 
2064 /*
2065  * Called by writers of cc_state only,  must call under cc_state_lock.
2066  */
2067 static inline
2068 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
2069 {
2070 	return rcu_dereference_protected(ppd->cc_state,
2071 					 lockdep_is_held(&ppd->cc_state_lock));
2072 }
2073 
2074 /*
2075  * values for dd->flags (_device_ related flags)
2076  */
2077 #define HFI1_INITTED           0x1    /* chip and driver up and initted */
2078 #define HFI1_PRESENT           0x2    /* chip accesses can be done */
2079 #define HFI1_FROZEN            0x4    /* chip in SPC freeze */
2080 #define HFI1_HAS_SDMA_TIMEOUT  0x8
2081 #define HFI1_HAS_SEND_DMA      0x10   /* Supports Send DMA */
2082 #define HFI1_FORCED_FREEZE     0x80   /* driver forced freeze mode */
2083 #define HFI1_SHUTDOWN          0x100  /* device is shutting down */
2084 
2085 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
2086 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
2087 
2088 /* ctxt_flag bit offsets */
2089 		/* base context has not finished initializing */
2090 #define HFI1_CTXT_BASE_UNINIT 1
2091 		/* base context initaliation failed */
2092 #define HFI1_CTXT_BASE_FAILED 2
2093 		/* waiting for a packet to arrive */
2094 #define HFI1_CTXT_WAITING_RCV 3
2095 		/* waiting for an urgent packet to arrive */
2096 #define HFI1_CTXT_WAITING_URG 4
2097 
2098 /* free up any allocated data at closes */
2099 int hfi1_init_dd(struct hfi1_devdata *dd);
2100 void hfi1_free_devdata(struct hfi1_devdata *dd);
2101 
2102 /* LED beaconing functions */
2103 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
2104 			     unsigned int timeoff);
2105 void shutdown_led_override(struct hfi1_pportdata *ppd);
2106 
2107 #define HFI1_CREDIT_RETURN_RATE (100)
2108 
2109 /*
2110  * The number of words for the KDETH protocol field.  If this is
2111  * larger then the actual field used, then part of the payload
2112  * will be in the header.
2113  *
2114  * Optimally, we want this sized so that a typical case will
2115  * use full cache lines.  The typical local KDETH header would
2116  * be:
2117  *
2118  *	Bytes	Field
2119  *	  8	LRH
2120  *	 12	BHT
2121  *	 ??	KDETH
2122  *	  8	RHF
2123  *	---
2124  *	 28 + KDETH
2125  *
2126  * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
2127  */
2128 #define DEFAULT_RCVHDRSIZE 9
2129 
2130 /*
2131  * Maximal header byte count:
2132  *
2133  *	Bytes	Field
2134  *	  8	LRH
2135  *	 40	GRH (optional)
2136  *	 12	BTH
2137  *	 ??	KDETH
2138  *	  8	RHF
2139  *	---
2140  *	 68 + KDETH
2141  *
2142  * We also want to maintain a cache line alignment to assist DMA'ing
2143  * of the header bytes.  Round up to a good size.
2144  */
2145 #define DEFAULT_RCVHDR_ENTSIZE 32
2146 
2147 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
2148 			u32 nlocked, u32 npages);
2149 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
2150 			    size_t npages, bool writable, struct page **pages);
2151 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
2152 			     size_t npages, bool dirty);
2153 
2154 /**
2155  * hfi1_rcvhdrtail_kvaddr - return tail kvaddr
2156  * @rcd - the receive context
2157  */
2158 static inline __le64 *hfi1_rcvhdrtail_kvaddr(const struct hfi1_ctxtdata *rcd)
2159 {
2160 	return (__le64 *)rcd->rcvhdrtail_kvaddr;
2161 }
2162 
2163 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2164 {
2165 	u64 *kv = (u64 *)hfi1_rcvhdrtail_kvaddr(rcd);
2166 
2167 	if (kv)
2168 		*kv = 0ULL;
2169 }
2170 
2171 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2172 {
2173 	/*
2174 	 * volatile because it's a DMA target from the chip, routine is
2175 	 * inlined, and don't want register caching or reordering.
2176 	 */
2177 	return (u32)le64_to_cpu(*hfi1_rcvhdrtail_kvaddr(rcd));
2178 }
2179 
2180 static inline bool hfi1_packet_present(struct hfi1_ctxtdata *rcd)
2181 {
2182 	if (likely(!rcd->rcvhdrtail_kvaddr)) {
2183 		u32 seq = rhf_rcv_seq(rhf_to_cpu(get_rhf_addr(rcd)));
2184 
2185 		return !last_rcv_seq(rcd, seq);
2186 	}
2187 	return hfi1_rcd_head(rcd) != get_rcvhdrtail(rcd);
2188 }
2189 
2190 /*
2191  * sysfs interface.
2192  */
2193 
2194 extern const char ib_hfi1_version[];
2195 extern const struct attribute_group ib_hfi1_attr_group;
2196 
2197 int hfi1_device_create(struct hfi1_devdata *dd);
2198 void hfi1_device_remove(struct hfi1_devdata *dd);
2199 
2200 int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
2201 			   struct kobject *kobj);
2202 int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
2203 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
2204 /* Hook for sysfs read of QSFP */
2205 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
2206 
2207 int hfi1_pcie_init(struct hfi1_devdata *dd);
2208 void hfi1_pcie_cleanup(struct pci_dev *pdev);
2209 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
2210 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
2211 int pcie_speeds(struct hfi1_devdata *dd);
2212 int restore_pci_variables(struct hfi1_devdata *dd);
2213 int save_pci_variables(struct hfi1_devdata *dd);
2214 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
2215 void tune_pcie_caps(struct hfi1_devdata *dd);
2216 int parse_platform_config(struct hfi1_devdata *dd);
2217 int get_platform_config_field(struct hfi1_devdata *dd,
2218 			      enum platform_config_table_type_encoding
2219 			      table_type, int table_index, int field_index,
2220 			      u32 *data, u32 len);
2221 
2222 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
2223 
2224 /*
2225  * Flush write combining store buffers (if present) and perform a write
2226  * barrier.
2227  */
2228 static inline void flush_wc(void)
2229 {
2230 	asm volatile("sfence" : : : "memory");
2231 }
2232 
2233 void handle_eflags(struct hfi1_packet *packet);
2234 void seqfile_dump_rcd(struct seq_file *s, struct hfi1_ctxtdata *rcd);
2235 
2236 /* global module parameter variables */
2237 extern unsigned int hfi1_max_mtu;
2238 extern unsigned int hfi1_cu;
2239 extern unsigned int user_credit_return_threshold;
2240 extern int num_user_contexts;
2241 extern unsigned long n_krcvqs;
2242 extern uint krcvqs[];
2243 extern int krcvqsset;
2244 extern uint loopback;
2245 extern uint quick_linkup;
2246 extern uint rcv_intr_timeout;
2247 extern uint rcv_intr_count;
2248 extern uint rcv_intr_dynamic;
2249 extern ushort link_crc_mask;
2250 
2251 extern struct mutex hfi1_mutex;
2252 
2253 /* Number of seconds before our card status check...  */
2254 #define STATUS_TIMEOUT 60
2255 
2256 #define DRIVER_NAME		"hfi1"
2257 #define HFI1_USER_MINOR_BASE     0
2258 #define HFI1_TRACE_MINOR         127
2259 #define HFI1_NMINORS             255
2260 
2261 #define PCI_VENDOR_ID_INTEL 0x8086
2262 #define PCI_DEVICE_ID_INTEL0 0x24f0
2263 #define PCI_DEVICE_ID_INTEL1 0x24f1
2264 
2265 #define HFI1_PKT_USER_SC_INTEGRITY					    \
2266 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
2267 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
2268 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
2269 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
2270 
2271 #define HFI1_PKT_KERNEL_SC_INTEGRITY					    \
2272 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
2273 
2274 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
2275 						  u16 ctxt_type)
2276 {
2277 	u64 base_sc_integrity;
2278 
2279 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2280 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2281 		return 0;
2282 
2283 	base_sc_integrity =
2284 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2285 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
2286 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2287 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2288 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2289 #ifndef CONFIG_FAULT_INJECTION
2290 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
2291 #endif
2292 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2293 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2294 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2295 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
2296 	| SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2297 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2298 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
2299 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
2300 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
2301 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2302 
2303 	if (ctxt_type == SC_USER)
2304 		base_sc_integrity |=
2305 #ifndef CONFIG_FAULT_INJECTION
2306 			SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK |
2307 #endif
2308 			HFI1_PKT_USER_SC_INTEGRITY;
2309 	else if (ctxt_type != SC_KERNEL)
2310 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
2311 
2312 	/* turn on send-side job key checks if !A0 */
2313 	if (!is_ax(dd))
2314 		base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2315 
2316 	return base_sc_integrity;
2317 }
2318 
2319 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
2320 {
2321 	u64 base_sdma_integrity;
2322 
2323 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2324 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2325 		return 0;
2326 
2327 	base_sdma_integrity =
2328 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2329 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2330 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2331 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2332 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2333 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2334 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2335 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
2336 	| SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2337 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2338 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
2339 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
2340 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
2341 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2342 
2343 	if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
2344 		base_sdma_integrity |=
2345 		SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
2346 
2347 	/* turn on send-side job key checks if !A0 */
2348 	if (!is_ax(dd))
2349 		base_sdma_integrity |=
2350 			SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2351 
2352 	return base_sdma_integrity;
2353 }
2354 
2355 #define dd_dev_emerg(dd, fmt, ...) \
2356 	dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
2357 		  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2358 
2359 #define dd_dev_err(dd, fmt, ...) \
2360 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
2361 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2362 
2363 #define dd_dev_err_ratelimited(dd, fmt, ...) \
2364 	dev_err_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2365 			    rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2366 			    ##__VA_ARGS__)
2367 
2368 #define dd_dev_warn(dd, fmt, ...) \
2369 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
2370 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2371 
2372 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
2373 	dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2374 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2375 			     ##__VA_ARGS__)
2376 
2377 #define dd_dev_info(dd, fmt, ...) \
2378 	dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
2379 		 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2380 
2381 #define dd_dev_info_ratelimited(dd, fmt, ...) \
2382 	dev_info_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2383 			     rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2384 			     ##__VA_ARGS__)
2385 
2386 #define dd_dev_dbg(dd, fmt, ...) \
2387 	dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
2388 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2389 
2390 #define hfi1_dev_porterr(dd, port, fmt, ...) \
2391 	dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
2392 		rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), (port), ##__VA_ARGS__)
2393 
2394 /*
2395  * this is used for formatting hw error messages...
2396  */
2397 struct hfi1_hwerror_msgs {
2398 	u64 mask;
2399 	const char *msg;
2400 	size_t sz;
2401 };
2402 
2403 /* in intr.c... */
2404 void hfi1_format_hwerrors(u64 hwerrs,
2405 			  const struct hfi1_hwerror_msgs *hwerrmsgs,
2406 			  size_t nhwerrmsgs, char *msg, size_t lmsg);
2407 
2408 #define USER_OPCODE_CHECK_VAL 0xC0
2409 #define USER_OPCODE_CHECK_MASK 0xC0
2410 #define OPCODE_CHECK_VAL_DISABLED 0x0
2411 #define OPCODE_CHECK_MASK_DISABLED 0x0
2412 
2413 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
2414 {
2415 	struct hfi1_pportdata *ppd;
2416 	int i;
2417 
2418 	dd->z_int_counter = get_all_cpu_total(dd->int_counter);
2419 	dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
2420 	dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
2421 
2422 	ppd = (struct hfi1_pportdata *)(dd + 1);
2423 	for (i = 0; i < dd->num_pports; i++, ppd++) {
2424 		ppd->ibport_data.rvp.z_rc_acks =
2425 			get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
2426 		ppd->ibport_data.rvp.z_rc_qacks =
2427 			get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
2428 	}
2429 }
2430 
2431 /* Control LED state */
2432 static inline void setextled(struct hfi1_devdata *dd, u32 on)
2433 {
2434 	if (on)
2435 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
2436 	else
2437 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
2438 }
2439 
2440 /* return the i2c resource given the target */
2441 static inline u32 i2c_target(u32 target)
2442 {
2443 	return target ? CR_I2C2 : CR_I2C1;
2444 }
2445 
2446 /* return the i2c chain chip resource that this HFI uses for QSFP */
2447 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2448 {
2449 	return i2c_target(dd->hfi1_id);
2450 }
2451 
2452 /* Is this device integrated or discrete? */
2453 static inline bool is_integrated(struct hfi1_devdata *dd)
2454 {
2455 	return dd->pcidev->device == PCI_DEVICE_ID_INTEL1;
2456 }
2457 
2458 /**
2459  * hfi1_need_drop - detect need for drop
2460  * @dd: - the device
2461  *
2462  * In some cases, the first packet needs to be dropped.
2463  *
2464  * Return true is the current packet needs to be dropped and false otherwise.
2465  */
2466 static inline bool hfi1_need_drop(struct hfi1_devdata *dd)
2467 {
2468 	if (unlikely(dd->do_drop &&
2469 		     atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
2470 		     DROP_PACKET_ON)) {
2471 		dd->do_drop = false;
2472 		return true;
2473 	}
2474 	return false;
2475 }
2476 
2477 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2478 
2479 #define DD_DEV_ENTRY(dd)       __string(dev, dev_name(&(dd)->pcidev->dev))
2480 #define DD_DEV_ASSIGN(dd)      __assign_str(dev, dev_name(&(dd)->pcidev->dev))
2481 
2482 static inline void hfi1_update_ah_attr(struct ib_device *ibdev,
2483 				       struct rdma_ah_attr *attr)
2484 {
2485 	struct hfi1_pportdata *ppd;
2486 	struct hfi1_ibport *ibp;
2487 	u32 dlid = rdma_ah_get_dlid(attr);
2488 
2489 	/*
2490 	 * Kernel clients may not have setup GRH information
2491 	 * Set that here.
2492 	 */
2493 	ibp = to_iport(ibdev, rdma_ah_get_port_num(attr));
2494 	ppd = ppd_from_ibp(ibp);
2495 	if ((((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) ||
2496 	      (ppd->lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))) &&
2497 	    (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)) &&
2498 	    (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2499 	    (!(rdma_ah_get_ah_flags(attr) & IB_AH_GRH))) ||
2500 	    (rdma_ah_get_make_grd(attr))) {
2501 		rdma_ah_set_ah_flags(attr, IB_AH_GRH);
2502 		rdma_ah_set_interface_id(attr, OPA_MAKE_ID(dlid));
2503 		rdma_ah_set_subnet_prefix(attr, ibp->rvp.gid_prefix);
2504 	}
2505 }
2506 
2507 /*
2508  * hfi1_check_mcast- Check if the given lid is
2509  * in the OPA multicast range.
2510  *
2511  * The LID might either reside in ah.dlid or might be
2512  * in the GRH of the address handle as DGID if extended
2513  * addresses are in use.
2514  */
2515 static inline bool hfi1_check_mcast(u32 lid)
2516 {
2517 	return ((lid >= opa_get_mcast_base(OPA_MCAST_NR)) &&
2518 		(lid != be32_to_cpu(OPA_LID_PERMISSIVE)));
2519 }
2520 
2521 #define opa_get_lid(lid, format)	\
2522 	__opa_get_lid(lid, OPA_PORT_PACKET_FORMAT_##format)
2523 
2524 /* Convert a lid to a specific lid space */
2525 static inline u32 __opa_get_lid(u32 lid, u8 format)
2526 {
2527 	bool is_mcast = hfi1_check_mcast(lid);
2528 
2529 	switch (format) {
2530 	case OPA_PORT_PACKET_FORMAT_8B:
2531 	case OPA_PORT_PACKET_FORMAT_10B:
2532 		if (is_mcast)
2533 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2534 				0xF0000);
2535 		return lid & 0xFFFFF;
2536 	case OPA_PORT_PACKET_FORMAT_16B:
2537 		if (is_mcast)
2538 			return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2539 				0xF00000);
2540 		return lid & 0xFFFFFF;
2541 	case OPA_PORT_PACKET_FORMAT_9B:
2542 		if (is_mcast)
2543 			return (lid -
2544 				opa_get_mcast_base(OPA_MCAST_NR) +
2545 				be16_to_cpu(IB_MULTICAST_LID_BASE));
2546 		else
2547 			return lid & 0xFFFF;
2548 	default:
2549 		return lid;
2550 	}
2551 }
2552 
2553 /* Return true if the given lid is the OPA 16B multicast range */
2554 static inline bool hfi1_is_16B_mcast(u32 lid)
2555 {
2556 	return ((lid >=
2557 		opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 16B)) &&
2558 		(lid != opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B)));
2559 }
2560 
2561 static inline void hfi1_make_opa_lid(struct rdma_ah_attr *attr)
2562 {
2563 	const struct ib_global_route *grh = rdma_ah_read_grh(attr);
2564 	u32 dlid = rdma_ah_get_dlid(attr);
2565 
2566 	/* Modify ah_attr.dlid to be in the 32 bit LID space.
2567 	 * This is how the address will be laid out:
2568 	 * Assuming MCAST_NR to be 4,
2569 	 * 32 bit permissive LID = 0xFFFFFFFF
2570 	 * Multicast LID range = 0xFFFFFFFE to 0xF0000000
2571 	 * Unicast LID range = 0xEFFFFFFF to 1
2572 	 * Invalid LID = 0
2573 	 */
2574 	if (ib_is_opa_gid(&grh->dgid))
2575 		dlid = opa_get_lid_from_gid(&grh->dgid);
2576 	else if ((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
2577 		 (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2578 		 (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)))
2579 		dlid = dlid - be16_to_cpu(IB_MULTICAST_LID_BASE) +
2580 			opa_get_mcast_base(OPA_MCAST_NR);
2581 	else if (dlid == be16_to_cpu(IB_LID_PERMISSIVE))
2582 		dlid = be32_to_cpu(OPA_LID_PERMISSIVE);
2583 
2584 	rdma_ah_set_dlid(attr, dlid);
2585 }
2586 
2587 static inline u8 hfi1_get_packet_type(u32 lid)
2588 {
2589 	/* 9B if lid > 0xF0000000 */
2590 	if (lid >= opa_get_mcast_base(OPA_MCAST_NR))
2591 		return HFI1_PKT_TYPE_9B;
2592 
2593 	/* 16B if lid > 0xC000 */
2594 	if (lid >= opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 9B))
2595 		return HFI1_PKT_TYPE_16B;
2596 
2597 	return HFI1_PKT_TYPE_9B;
2598 }
2599 
2600 static inline bool hfi1_get_hdr_type(u32 lid, struct rdma_ah_attr *attr)
2601 {
2602 	/*
2603 	 * If there was an incoming 16B packet with permissive
2604 	 * LIDs, OPA GIDs would have been programmed when those
2605 	 * packets were received. A 16B packet will have to
2606 	 * be sent in response to that packet. Return a 16B
2607 	 * header type if that's the case.
2608 	 */
2609 	if (rdma_ah_get_dlid(attr) == be32_to_cpu(OPA_LID_PERMISSIVE))
2610 		return (ib_is_opa_gid(&rdma_ah_read_grh(attr)->dgid)) ?
2611 			HFI1_PKT_TYPE_16B : HFI1_PKT_TYPE_9B;
2612 
2613 	/*
2614 	 * Return a 16B header type if either the the destination
2615 	 * or source lid is extended.
2616 	 */
2617 	if (hfi1_get_packet_type(rdma_ah_get_dlid(attr)) == HFI1_PKT_TYPE_16B)
2618 		return HFI1_PKT_TYPE_16B;
2619 
2620 	return hfi1_get_packet_type(lid);
2621 }
2622 
2623 static inline void hfi1_make_ext_grh(struct hfi1_packet *packet,
2624 				     struct ib_grh *grh, u32 slid,
2625 				     u32 dlid)
2626 {
2627 	struct hfi1_ibport *ibp = &packet->rcd->ppd->ibport_data;
2628 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2629 
2630 	if (!ibp)
2631 		return;
2632 
2633 	grh->hop_limit = 1;
2634 	grh->sgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2635 	if (slid == opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B))
2636 		grh->sgid.global.interface_id =
2637 			OPA_MAKE_ID(be32_to_cpu(OPA_LID_PERMISSIVE));
2638 	else
2639 		grh->sgid.global.interface_id = OPA_MAKE_ID(slid);
2640 
2641 	/*
2642 	 * Upper layers (like mad) may compare the dgid in the
2643 	 * wc that is obtained here with the sgid_index in
2644 	 * the wr. Since sgid_index in wr is always 0 for
2645 	 * extended lids, set the dgid here to the default
2646 	 * IB gid.
2647 	 */
2648 	grh->dgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2649 	grh->dgid.global.interface_id =
2650 		cpu_to_be64(ppd->guids[HFI1_PORT_GUID_INDEX]);
2651 }
2652 
2653 static inline int hfi1_get_16b_padding(u32 hdr_size, u32 payload)
2654 {
2655 	return -(hdr_size + payload + (SIZE_OF_CRC << 2) +
2656 		     SIZE_OF_LT) & 0x7;
2657 }
2658 
2659 static inline void hfi1_make_ib_hdr(struct ib_header *hdr,
2660 				    u16 lrh0, u16 len,
2661 				    u16 dlid, u16 slid)
2662 {
2663 	hdr->lrh[0] = cpu_to_be16(lrh0);
2664 	hdr->lrh[1] = cpu_to_be16(dlid);
2665 	hdr->lrh[2] = cpu_to_be16(len);
2666 	hdr->lrh[3] = cpu_to_be16(slid);
2667 }
2668 
2669 static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr,
2670 				     u32 slid, u32 dlid,
2671 				     u16 len, u16 pkey,
2672 				     bool becn, bool fecn, u8 l4,
2673 				     u8 sc)
2674 {
2675 	u32 lrh0 = 0;
2676 	u32 lrh1 = 0x40000000;
2677 	u32 lrh2 = 0;
2678 	u32 lrh3 = 0;
2679 
2680 	lrh0 = (lrh0 & ~OPA_16B_BECN_MASK) | (becn << OPA_16B_BECN_SHIFT);
2681 	lrh0 = (lrh0 & ~OPA_16B_LEN_MASK) | (len << OPA_16B_LEN_SHIFT);
2682 	lrh0 = (lrh0 & ~OPA_16B_LID_MASK)  | (slid & OPA_16B_LID_MASK);
2683 	lrh1 = (lrh1 & ~OPA_16B_FECN_MASK) | (fecn << OPA_16B_FECN_SHIFT);
2684 	lrh1 = (lrh1 & ~OPA_16B_SC_MASK) | (sc << OPA_16B_SC_SHIFT);
2685 	lrh1 = (lrh1 & ~OPA_16B_LID_MASK) | (dlid & OPA_16B_LID_MASK);
2686 	lrh2 = (lrh2 & ~OPA_16B_SLID_MASK) |
2687 		((slid >> OPA_16B_SLID_SHIFT) << OPA_16B_SLID_HIGH_SHIFT);
2688 	lrh2 = (lrh2 & ~OPA_16B_DLID_MASK) |
2689 		((dlid >> OPA_16B_DLID_SHIFT) << OPA_16B_DLID_HIGH_SHIFT);
2690 	lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | ((u32)pkey << OPA_16B_PKEY_SHIFT);
2691 	lrh2 = (lrh2 & ~OPA_16B_L4_MASK) | l4;
2692 
2693 	hdr->lrh[0] = lrh0;
2694 	hdr->lrh[1] = lrh1;
2695 	hdr->lrh[2] = lrh2;
2696 	hdr->lrh[3] = lrh3;
2697 }
2698 #endif                          /* _HFI1_KERNEL_H */
2699