xref: /openbmc/linux/drivers/infiniband/hw/hfi1/hfi.h (revision a977d045)
1 #ifndef _HFI1_KERNEL_H
2 #define _HFI1_KERNEL_H
3 /*
4  * Copyright(c) 2015-2017 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/idr.h>
58 #include <linux/io.h>
59 #include <linux/fs.h>
60 #include <linux/completion.h>
61 #include <linux/kref.h>
62 #include <linux/sched.h>
63 #include <linux/cdev.h>
64 #include <linux/delay.h>
65 #include <linux/kthread.h>
66 #include <linux/i2c.h>
67 #include <linux/i2c-algo-bit.h>
68 #include <rdma/ib_hdrs.h>
69 #include <linux/rhashtable.h>
70 #include <linux/netdevice.h>
71 #include <rdma/rdma_vt.h>
72 
73 #include "chip_registers.h"
74 #include "common.h"
75 #include "verbs.h"
76 #include "pio.h"
77 #include "chip.h"
78 #include "mad.h"
79 #include "qsfp.h"
80 #include "platform.h"
81 #include "affinity.h"
82 
83 /* bumped 1 from s/w major version of TrueScale */
84 #define HFI1_CHIP_VERS_MAJ 3U
85 
86 /* don't care about this except printing */
87 #define HFI1_CHIP_VERS_MIN 0U
88 
89 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
90 #define HFI1_OUI 0x001175
91 #define HFI1_OUI_LSB 40
92 
93 #define DROP_PACKET_OFF		0
94 #define DROP_PACKET_ON		1
95 
96 extern unsigned long hfi1_cap_mask;
97 #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap)
98 #define HFI1_CAP_UGET_MASK(mask, cap) \
99 	(((mask) >> HFI1_CAP_USER_SHIFT) & HFI1_CAP_##cap)
100 #define HFI1_CAP_KGET(cap) (HFI1_CAP_KGET_MASK(hfi1_cap_mask, cap))
101 #define HFI1_CAP_UGET(cap) (HFI1_CAP_UGET_MASK(hfi1_cap_mask, cap))
102 #define HFI1_CAP_IS_KSET(cap) (!!HFI1_CAP_KGET(cap))
103 #define HFI1_CAP_IS_USET(cap) (!!HFI1_CAP_UGET(cap))
104 #define HFI1_MISC_GET() ((hfi1_cap_mask >> HFI1_CAP_MISC_SHIFT) & \
105 			HFI1_CAP_MISC_MASK)
106 /* Offline Disabled Reason is 4-bits */
107 #define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
108 
109 /*
110  * Control context is always 0 and handles the error packets.
111  * It also handles the VL15 and multicast packets.
112  */
113 #define HFI1_CTRL_CTXT    0
114 
115 /*
116  * Driver context will store software counters for each of the events
117  * associated with these status registers
118  */
119 #define NUM_CCE_ERR_STATUS_COUNTERS 41
120 #define NUM_RCV_ERR_STATUS_COUNTERS 64
121 #define NUM_MISC_ERR_STATUS_COUNTERS 13
122 #define NUM_SEND_PIO_ERR_STATUS_COUNTERS 36
123 #define NUM_SEND_DMA_ERR_STATUS_COUNTERS 4
124 #define NUM_SEND_EGRESS_ERR_STATUS_COUNTERS 64
125 #define NUM_SEND_ERR_STATUS_COUNTERS 3
126 #define NUM_SEND_CTXT_ERR_STATUS_COUNTERS 5
127 #define NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS 24
128 
129 /*
130  * per driver stats, either not device nor port-specific, or
131  * summed over all of the devices and ports.
132  * They are described by name via ipathfs filesystem, so layout
133  * and number of elements can change without breaking compatibility.
134  * If members are added or deleted hfi1_statnames[] in debugfs.c must
135  * change to match.
136  */
137 struct hfi1_ib_stats {
138 	__u64 sps_ints; /* number of interrupts handled */
139 	__u64 sps_errints; /* number of error interrupts */
140 	__u64 sps_txerrs; /* tx-related packet errors */
141 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
142 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
143 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
144 	__u64 sps_ctxts; /* number of contexts currently open */
145 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
146 	__u64 sps_buffull;
147 	__u64 sps_hdrfull;
148 };
149 
150 extern struct hfi1_ib_stats hfi1_stats;
151 extern const struct pci_error_handlers hfi1_pci_err_handler;
152 
153 /*
154  * First-cut criterion for "device is active" is
155  * two thousand dwords combined Tx, Rx traffic per
156  * 5-second interval. SMA packets are 64 dwords,
157  * and occur "a few per second", presumably each way.
158  */
159 #define HFI1_TRAFFIC_ACTIVE_THRESHOLD (2000)
160 
161 /*
162  * Below contains all data related to a single context (formerly called port).
163  */
164 
165 #ifdef CONFIG_DEBUG_FS
166 struct hfi1_opcode_stats_perctx;
167 #endif
168 
169 struct ctxt_eager_bufs {
170 	ssize_t size;            /* total size of eager buffers */
171 	u32 count;               /* size of buffers array */
172 	u32 numbufs;             /* number of buffers allocated */
173 	u32 alloced;             /* number of rcvarray entries used */
174 	u32 rcvtid_size;         /* size of each eager rcv tid */
175 	u32 threshold;           /* head update threshold */
176 	struct eager_buffer {
177 		void *addr;
178 		dma_addr_t dma;
179 		ssize_t len;
180 	} *buffers;
181 	struct {
182 		void *addr;
183 		dma_addr_t dma;
184 	} *rcvtids;
185 };
186 
187 struct exp_tid_set {
188 	struct list_head list;
189 	u32 count;
190 };
191 
192 struct hfi1_ctxtdata {
193 	/* shadow the ctxt's RcvCtrl register */
194 	u64 rcvctrl;
195 	/* rcvhdrq base, needs mmap before useful */
196 	void *rcvhdrq;
197 	/* kernel virtual address where hdrqtail is updated */
198 	volatile __le64 *rcvhdrtail_kvaddr;
199 	/* when waiting for rcv or pioavail */
200 	wait_queue_head_t wait;
201 	/* rcvhdrq size (for freeing) */
202 	size_t rcvhdrq_size;
203 	/* number of rcvhdrq entries */
204 	u16 rcvhdrq_cnt;
205 	/* size of each of the rcvhdrq entries */
206 	u16 rcvhdrqentsize;
207 	/* mmap of hdrq, must fit in 44 bits */
208 	dma_addr_t rcvhdrq_dma;
209 	dma_addr_t rcvhdrqtailaddr_dma;
210 	struct ctxt_eager_bufs egrbufs;
211 	/* this receive context's assigned PIO ACK send context */
212 	struct send_context *sc;
213 
214 	/* dynamic receive available interrupt timeout */
215 	u32 rcvavail_timeout;
216 	/*
217 	 * number of opens (including slave sub-contexts) on this instance
218 	 * (ignoring forks, dup, etc. for now)
219 	 */
220 	int cnt;
221 	/* Device context index */
222 	unsigned ctxt;
223 	/*
224 	 * non-zero if ctxt can be shared, and defines the maximum number of
225 	 * sub-contexts for this device context.
226 	 */
227 	u16 subctxt_cnt;
228 	/* non-zero if ctxt is being shared. */
229 	u16 subctxt_id;
230 	u8 uuid[16];
231 	/* job key */
232 	u16 jkey;
233 	/* number of RcvArray groups for this context. */
234 	u32 rcv_array_groups;
235 	/* index of first eager TID entry. */
236 	u32 eager_base;
237 	/* number of expected TID entries */
238 	u32 expected_count;
239 	/* index of first expected TID entry. */
240 	u32 expected_base;
241 
242 	struct exp_tid_set tid_group_list;
243 	struct exp_tid_set tid_used_list;
244 	struct exp_tid_set tid_full_list;
245 
246 	/* lock protecting all Expected TID data */
247 	struct mutex exp_lock;
248 	/* number of pio bufs for this ctxt (all procs, if shared) */
249 	u32 piocnt;
250 	/* first pio buffer for this ctxt */
251 	u32 pio_base;
252 	/* chip offset of PIO buffers for this ctxt */
253 	u32 piobufs;
254 	/* per-context configuration flags */
255 	unsigned long flags;
256 	/* per-context event flags for fileops/intr communication */
257 	unsigned long event_flags;
258 	/* WAIT_RCV that timed out, no interrupt */
259 	u32 rcvwait_to;
260 	/* WAIT_PIO that timed out, no interrupt */
261 	u32 piowait_to;
262 	/* WAIT_RCV already happened, no wait */
263 	u32 rcvnowait;
264 	/* WAIT_PIO already happened, no wait */
265 	u32 pionowait;
266 	/* total number of polled urgent packets */
267 	u32 urgent;
268 	/* saved total number of polled urgent packets for poll edge trigger */
269 	u32 urgent_poll;
270 	/* same size as task_struct .comm[], command that opened context */
271 	char comm[TASK_COMM_LEN];
272 	/* so file ops can get at unit */
273 	struct hfi1_devdata *dd;
274 	/* so functions that need physical port can get it easily */
275 	struct hfi1_pportdata *ppd;
276 	/* associated msix interrupt */
277 	u32 msix_intr;
278 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
279 	void *subctxt_uregbase;
280 	/* An array of pages for the eager receive buffers * N */
281 	void *subctxt_rcvegrbuf;
282 	/* An array of pages for the eager header queue entries * N */
283 	void *subctxt_rcvhdr_base;
284 	/* Bitmask of in use context(s) */
285 	DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
286 	/* The version of the library which opened this ctxt */
287 	u32 userversion;
288 	/* Type of packets or conditions we want to poll for */
289 	u16 poll_type;
290 	/* receive packet sequence counter */
291 	u8 seq_cnt;
292 	u8 redirect_seq_cnt;
293 	/* ctxt rcvhdrq head offset */
294 	u32 head;
295 	u32 pkt_count;
296 	/* QPs waiting for context processing */
297 	struct list_head qp_wait_list;
298 	/* interrupt handling */
299 	u64 imask;	/* clear interrupt mask */
300 	int ireg;	/* clear interrupt register */
301 	unsigned numa_id; /* numa node of this context */
302 	/* verbs stats per CTX */
303 	struct hfi1_opcode_stats_perctx *opstats;
304 	/*
305 	 * This is the kernel thread that will keep making
306 	 * progress on the user sdma requests behind the scenes.
307 	 * There is one per context (shared contexts use the master's).
308 	 */
309 	struct task_struct *progress;
310 	struct list_head sdma_queues;
311 	/* protect sdma queues */
312 	spinlock_t sdma_qlock;
313 
314 	/* Is ASPM interrupt supported for this context */
315 	bool aspm_intr_supported;
316 	/* ASPM state (enabled/disabled) for this context */
317 	bool aspm_enabled;
318 	/* Timer for re-enabling ASPM if interrupt activity quietens down */
319 	struct timer_list aspm_timer;
320 	/* Lock to serialize between intr, timer intr and user threads */
321 	spinlock_t aspm_lock;
322 	/* Is ASPM processing enabled for this context (in intr context) */
323 	bool aspm_intr_enable;
324 	/* Last interrupt timestamp */
325 	ktime_t aspm_ts_last_intr;
326 	/* Last timestamp at which we scheduled a timer for this context */
327 	ktime_t aspm_ts_timer_sched;
328 
329 	/*
330 	 * The interrupt handler for a particular receive context can vary
331 	 * throughout it's lifetime. This is not a lock protected data member so
332 	 * it must be updated atomically and the prev and new value must always
333 	 * be valid. Worst case is we process an extra interrupt and up to 64
334 	 * packets with the wrong interrupt handler.
335 	 */
336 	int (*do_interrupt)(struct hfi1_ctxtdata *rcd, int threaded);
337 
338 	/* Indicates that this is vnic context */
339 	bool is_vnic;
340 
341 	/* vnic queue index this context is mapped to */
342 	u8 vnic_q_idx;
343 };
344 
345 /*
346  * Represents a single packet at a high level. Put commonly computed things in
347  * here so we do not have to keep doing them over and over. The rule of thumb is
348  * if something is used one time to derive some value, store that something in
349  * here. If it is used multiple times, then store the result of that derivation
350  * in here.
351  */
352 struct hfi1_packet {
353 	void *ebuf;
354 	void *hdr;
355 	struct hfi1_ctxtdata *rcd;
356 	__le32 *rhf_addr;
357 	struct rvt_qp *qp;
358 	struct ib_other_headers *ohdr;
359 	u64 rhf;
360 	u32 maxcnt;
361 	u32 rhqoff;
362 	u16 tlen;
363 	s16 etail;
364 	u8 hlen;
365 	u8 numpkt;
366 	u8 rsize;
367 	u8 updegr;
368 	u8 rcv_flags;
369 	u8 etype;
370 };
371 
372 struct rvt_sge_state;
373 
374 /*
375  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
376  * Mostly for MADs that set or query link parameters, also ipath
377  * config interfaces
378  */
379 #define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
380 #define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
381 #define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
382 #define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
383 #define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
384 #define HFI1_IB_CFG_SPD 5 /* current Link spd */
385 #define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
386 #define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
387 #define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
388 #define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
389 #define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
390 #define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
391 #define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
392 #define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
393 #define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
394 #define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
395 #define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
396 #define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
397 #define HFI1_IB_CFG_VL_HIGH_LIMIT 19
398 #define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
399 #define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
400 
401 /*
402  * HFI or Host Link States
403  *
404  * These describe the states the driver thinks the logical and physical
405  * states are in.  Used as an argument to set_link_state().  Implemented
406  * as bits for easy multi-state checking.  The actual state can only be
407  * one.
408  */
409 #define __HLS_UP_INIT_BP	0
410 #define __HLS_UP_ARMED_BP	1
411 #define __HLS_UP_ACTIVE_BP	2
412 #define __HLS_DN_DOWNDEF_BP	3	/* link down default */
413 #define __HLS_DN_POLL_BP	4
414 #define __HLS_DN_DISABLE_BP	5
415 #define __HLS_DN_OFFLINE_BP	6
416 #define __HLS_VERIFY_CAP_BP	7
417 #define __HLS_GOING_UP_BP	8
418 #define __HLS_GOING_OFFLINE_BP  9
419 #define __HLS_LINK_COOLDOWN_BP 10
420 
421 #define HLS_UP_INIT	  BIT(__HLS_UP_INIT_BP)
422 #define HLS_UP_ARMED	  BIT(__HLS_UP_ARMED_BP)
423 #define HLS_UP_ACTIVE	  BIT(__HLS_UP_ACTIVE_BP)
424 #define HLS_DN_DOWNDEF	  BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
425 #define HLS_DN_POLL	  BIT(__HLS_DN_POLL_BP)
426 #define HLS_DN_DISABLE	  BIT(__HLS_DN_DISABLE_BP)
427 #define HLS_DN_OFFLINE	  BIT(__HLS_DN_OFFLINE_BP)
428 #define HLS_VERIFY_CAP	  BIT(__HLS_VERIFY_CAP_BP)
429 #define HLS_GOING_UP	  BIT(__HLS_GOING_UP_BP)
430 #define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
431 #define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
432 
433 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
434 #define HLS_DOWN ~(HLS_UP)
435 
436 /* use this MTU size if none other is given */
437 #define HFI1_DEFAULT_ACTIVE_MTU 10240
438 /* use this MTU size as the default maximum */
439 #define HFI1_DEFAULT_MAX_MTU 10240
440 /* default partition key */
441 #define DEFAULT_PKEY 0xffff
442 
443 /*
444  * Possible fabric manager config parameters for fm_{get,set}_table()
445  */
446 #define FM_TBL_VL_HIGH_ARB		1 /* Get/set VL high prio weights */
447 #define FM_TBL_VL_LOW_ARB		2 /* Get/set VL low prio weights */
448 #define FM_TBL_BUFFER_CONTROL		3 /* Get/set Buffer Control */
449 #define FM_TBL_SC2VLNT			4 /* Get/set SC->VLnt */
450 #define FM_TBL_VL_PREEMPT_ELEMS		5 /* Get (no set) VL preempt elems */
451 #define FM_TBL_VL_PREEMPT_MATRIX	6 /* Get (no set) VL preempt matrix */
452 
453 /*
454  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
455  * these are bits so they can be combined, e.g.
456  * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
457  */
458 #define HFI1_RCVCTRL_TAILUPD_ENB 0x01
459 #define HFI1_RCVCTRL_TAILUPD_DIS 0x02
460 #define HFI1_RCVCTRL_CTXT_ENB 0x04
461 #define HFI1_RCVCTRL_CTXT_DIS 0x08
462 #define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
463 #define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
464 #define HFI1_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
465 #define HFI1_RCVCTRL_PKEY_DIS 0x80
466 #define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
467 #define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
468 #define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
469 #define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
470 #define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
471 #define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
472 #define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
473 #define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
474 
475 /* partition enforcement flags */
476 #define HFI1_PART_ENFORCE_IN	0x1
477 #define HFI1_PART_ENFORCE_OUT	0x2
478 
479 /* how often we check for synthetic counter wrap around */
480 #define SYNTH_CNT_TIME 3
481 
482 /* Counter flags */
483 #define CNTR_NORMAL		0x0 /* Normal counters, just read register */
484 #define CNTR_SYNTH		0x1 /* Synthetic counters, saturate at all 1s */
485 #define CNTR_DISABLED		0x2 /* Disable this counter */
486 #define CNTR_32BIT		0x4 /* Simulate 64 bits for this counter */
487 #define CNTR_VL			0x8 /* Per VL counter */
488 #define CNTR_SDMA              0x10
489 #define CNTR_INVALID_VL		-1  /* Specifies invalid VL */
490 #define CNTR_MODE_W		0x0
491 #define CNTR_MODE_R		0x1
492 
493 /* VLs Supported/Operational */
494 #define HFI1_MIN_VLS_SUPPORTED 1
495 #define HFI1_MAX_VLS_SUPPORTED 8
496 
497 #define HFI1_GUIDS_PER_PORT  5
498 #define HFI1_PORT_GUID_INDEX 0
499 
500 static inline void incr_cntr64(u64 *cntr)
501 {
502 	if (*cntr < (u64)-1LL)
503 		(*cntr)++;
504 }
505 
506 static inline void incr_cntr32(u32 *cntr)
507 {
508 	if (*cntr < (u32)-1LL)
509 		(*cntr)++;
510 }
511 
512 #define MAX_NAME_SIZE 64
513 struct hfi1_msix_entry {
514 	enum irq_type type;
515 	struct msix_entry msix;
516 	void *arg;
517 	char name[MAX_NAME_SIZE];
518 	cpumask_t mask;
519 	struct irq_affinity_notify notify;
520 };
521 
522 /* per-SL CCA information */
523 struct cca_timer {
524 	struct hrtimer hrtimer;
525 	struct hfi1_pportdata *ppd; /* read-only */
526 	int sl; /* read-only */
527 	u16 ccti; /* read/write - current value of CCTI */
528 };
529 
530 struct link_down_reason {
531 	/*
532 	 * SMA-facing value.  Should be set from .latest when
533 	 * HLS_UP_* -> HLS_DN_* transition actually occurs.
534 	 */
535 	u8 sma;
536 	u8 latest;
537 };
538 
539 enum {
540 	LO_PRIO_TABLE,
541 	HI_PRIO_TABLE,
542 	MAX_PRIO_TABLE
543 };
544 
545 struct vl_arb_cache {
546 	/* protect vl arb cache */
547 	spinlock_t lock;
548 	struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
549 };
550 
551 /*
552  * The structure below encapsulates data relevant to a physical IB Port.
553  * Current chips support only one such port, but the separation
554  * clarifies things a bit. Note that to conform to IB conventions,
555  * port-numbers are one-based. The first or only port is port1.
556  */
557 struct hfi1_pportdata {
558 	struct hfi1_ibport ibport_data;
559 
560 	struct hfi1_devdata *dd;
561 	struct kobject pport_cc_kobj;
562 	struct kobject sc2vl_kobj;
563 	struct kobject sl2sc_kobj;
564 	struct kobject vl2mtu_kobj;
565 
566 	/* PHY support */
567 	struct qsfp_data qsfp_info;
568 	/* Values for SI tuning of SerDes */
569 	u32 port_type;
570 	u32 tx_preset_eq;
571 	u32 tx_preset_noeq;
572 	u32 rx_preset;
573 	u8  local_atten;
574 	u8  remote_atten;
575 	u8  default_atten;
576 	u8  max_power_class;
577 
578 	/* GUIDs for this interface, in host order, guids[0] is a port guid */
579 	u64 guids[HFI1_GUIDS_PER_PORT];
580 
581 	/* GUID for peer interface, in host order */
582 	u64 neighbor_guid;
583 
584 	/* up or down physical link state */
585 	u32 linkup;
586 
587 	/*
588 	 * this address is mapped read-only into user processes so they can
589 	 * get status cheaply, whenever they want.  One qword of status per port
590 	 */
591 	u64 *statusp;
592 
593 	/* SendDMA related entries */
594 
595 	struct workqueue_struct *hfi1_wq;
596 
597 	/* move out of interrupt context */
598 	struct work_struct link_vc_work;
599 	struct work_struct link_up_work;
600 	struct work_struct link_down_work;
601 	struct work_struct sma_message_work;
602 	struct work_struct freeze_work;
603 	struct work_struct link_downgrade_work;
604 	struct work_struct link_bounce_work;
605 	struct delayed_work start_link_work;
606 	/* host link state variables */
607 	struct mutex hls_lock;
608 	u32 host_link_state;
609 
610 	u32 lstate;	/* logical link state */
611 
612 	/* these are the "32 bit" regs */
613 
614 	u32 ibmtu; /* The MTU programmed for this unit */
615 	/*
616 	 * Current max size IB packet (in bytes) including IB headers, that
617 	 * we can send. Changes when ibmtu changes.
618 	 */
619 	u32 ibmaxlen;
620 	u32 current_egress_rate; /* units [10^6 bits/sec] */
621 	/* LID programmed for this instance */
622 	u16 lid;
623 	/* list of pkeys programmed; 0 if not set */
624 	u16 pkeys[MAX_PKEY_VALUES];
625 	u16 link_width_supported;
626 	u16 link_width_downgrade_supported;
627 	u16 link_speed_supported;
628 	u16 link_width_enabled;
629 	u16 link_width_downgrade_enabled;
630 	u16 link_speed_enabled;
631 	u16 link_width_active;
632 	u16 link_width_downgrade_tx_active;
633 	u16 link_width_downgrade_rx_active;
634 	u16 link_speed_active;
635 	u8 vls_supported;
636 	u8 vls_operational;
637 	u8 actual_vls_operational;
638 	/* LID mask control */
639 	u8 lmc;
640 	/* Rx Polarity inversion (compensate for ~tx on partner) */
641 	u8 rx_pol_inv;
642 
643 	u8 hw_pidx;     /* physical port index */
644 	u8 port;        /* IB port number and index into dd->pports - 1 */
645 	/* type of neighbor node */
646 	u8 neighbor_type;
647 	u8 neighbor_normal;
648 	u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
649 	u8 neighbor_port_number;
650 	u8 is_sm_config_started;
651 	u8 offline_disabled_reason;
652 	u8 is_active_optimize_enabled;
653 	u8 driver_link_ready;	/* driver ready for active link */
654 	u8 link_enabled;	/* link enabled? */
655 	u8 linkinit_reason;
656 	u8 local_tx_rate;	/* rate given to 8051 firmware */
657 	u8 last_pstate;		/* info only */
658 	u8 qsfp_retry_count;
659 
660 	/* placeholders for IB MAD packet settings */
661 	u8 overrun_threshold;
662 	u8 phy_error_threshold;
663 
664 	/* Used to override LED behavior for things like maintenance beaconing*/
665 	/*
666 	 * Alternates per phase of blink
667 	 * [0] holds LED off duration, [1] holds LED on duration
668 	 */
669 	unsigned long led_override_vals[2];
670 	u8 led_override_phase; /* LSB picks from vals[] */
671 	atomic_t led_override_timer_active;
672 	/* Used to flash LEDs in override mode */
673 	struct timer_list led_override_timer;
674 
675 	u32 sm_trap_qp;
676 	u32 sa_qp;
677 
678 	/*
679 	 * cca_timer_lock protects access to the per-SL cca_timer
680 	 * structures (specifically the ccti member).
681 	 */
682 	spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
683 	struct cca_timer cca_timer[OPA_MAX_SLS];
684 
685 	/* List of congestion control table entries */
686 	struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
687 
688 	/* congestion entries, each entry corresponding to a SL */
689 	struct opa_congestion_setting_entry_shadow
690 		congestion_entries[OPA_MAX_SLS];
691 
692 	/*
693 	 * cc_state_lock protects (write) access to the per-port
694 	 * struct cc_state.
695 	 */
696 	spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
697 
698 	struct cc_state __rcu *cc_state;
699 
700 	/* Total number of congestion control table entries */
701 	u16 total_cct_entry;
702 
703 	/* Bit map identifying service level */
704 	u32 cc_sl_control_map;
705 
706 	/* CA's max number of 64 entry units in the congestion control table */
707 	u8 cc_max_table_entries;
708 
709 	/*
710 	 * begin congestion log related entries
711 	 * cc_log_lock protects all congestion log related data
712 	 */
713 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
714 	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
715 	u16 threshold_event_counter;
716 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
717 	int cc_log_idx; /* index for logging events */
718 	int cc_mad_idx; /* index for reporting events */
719 	/* end congestion log related entries */
720 
721 	struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
722 
723 	/* port relative counter buffer */
724 	u64 *cntrs;
725 	/* port relative synthetic counter buffer */
726 	u64 *scntrs;
727 	/* port_xmit_discards are synthesized from different egress errors */
728 	u64 port_xmit_discards;
729 	u64 port_xmit_discards_vl[C_VL_COUNT];
730 	u64 port_xmit_constraint_errors;
731 	u64 port_rcv_constraint_errors;
732 	/* count of 'link_err' interrupts from DC */
733 	u64 link_downed;
734 	/* number of times link retrained successfully */
735 	u64 link_up;
736 	/* number of times a link unknown frame was reported */
737 	u64 unknown_frame_count;
738 	/* port_ltp_crc_mode is returned in 'portinfo' MADs */
739 	u16 port_ltp_crc_mode;
740 	/* port_crc_mode_enabled is the crc we support */
741 	u8 port_crc_mode_enabled;
742 	/* mgmt_allowed is also returned in 'portinfo' MADs */
743 	u8 mgmt_allowed;
744 	u8 part_enforce; /* partition enforcement flags */
745 	struct link_down_reason local_link_down_reason;
746 	struct link_down_reason neigh_link_down_reason;
747 	/* Value to be sent to link peer on LinkDown .*/
748 	u8 remote_link_down_reason;
749 	/* Error events that will cause a port bounce. */
750 	u32 port_error_action;
751 	struct work_struct linkstate_active_work;
752 	/* Does this port need to prescan for FECNs */
753 	bool cc_prescan;
754 };
755 
756 typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
757 
758 typedef void (*opcode_handler)(struct hfi1_packet *packet);
759 
760 /* return values for the RHF receive functions */
761 #define RHF_RCV_CONTINUE  0	/* keep going */
762 #define RHF_RCV_DONE	  1	/* stop, this packet processed */
763 #define RHF_RCV_REPROCESS 2	/* stop. retain this packet */
764 
765 struct rcv_array_data {
766 	u8 group_size;
767 	u16 ngroups;
768 	u16 nctxt_extra;
769 };
770 
771 struct per_vl_data {
772 	u16 mtu;
773 	struct send_context *sc;
774 };
775 
776 /* 16 to directly index */
777 #define PER_VL_SEND_CONTEXTS 16
778 
779 struct err_info_rcvport {
780 	u8 status_and_code;
781 	u64 packet_flit1;
782 	u64 packet_flit2;
783 };
784 
785 struct err_info_constraint {
786 	u8 status;
787 	u16 pkey;
788 	u32 slid;
789 };
790 
791 struct hfi1_temp {
792 	unsigned int curr;       /* current temperature */
793 	unsigned int lo_lim;     /* low temperature limit */
794 	unsigned int hi_lim;     /* high temperature limit */
795 	unsigned int crit_lim;   /* critical temperature limit */
796 	u8 triggers;      /* temperature triggers */
797 };
798 
799 struct hfi1_i2c_bus {
800 	struct hfi1_devdata *controlling_dd; /* current controlling device */
801 	struct i2c_adapter adapter;	/* bus details */
802 	struct i2c_algo_bit_data algo;	/* bus algorithm details */
803 	int num;			/* bus number, 0 or 1 */
804 };
805 
806 /* common data between shared ASIC HFIs */
807 struct hfi1_asic_data {
808 	struct hfi1_devdata *dds[2];	/* back pointers */
809 	struct mutex asic_resource_mutex;
810 	struct hfi1_i2c_bus *i2c_bus0;
811 	struct hfi1_i2c_bus *i2c_bus1;
812 };
813 
814 /* sizes for both the QP and RSM map tables */
815 #define NUM_MAP_ENTRIES	 256
816 #define NUM_MAP_REGS      32
817 
818 /*
819  * Number of VNIC contexts used. Ensure it is less than or equal to
820  * max queues supported by VNIC (HFI1_VNIC_MAX_QUEUE).
821  */
822 #define HFI1_NUM_VNIC_CTXT   8
823 
824 /* Number of VNIC RSM entries */
825 #define NUM_VNIC_MAP_ENTRIES 8
826 
827 /* Virtual NIC information */
828 struct hfi1_vnic_data {
829 	struct hfi1_ctxtdata *ctxt[HFI1_NUM_VNIC_CTXT];
830 	struct kmem_cache *txreq_cache;
831 	u8 num_vports;
832 	struct idr vesw_idr;
833 	u8 rmt_start;
834 	u8 num_ctxt;
835 	u32 msix_idx;
836 };
837 
838 struct hfi1_vnic_vport_info;
839 
840 /* device data struct now contains only "general per-device" info.
841  * fields related to a physical IB port are in a hfi1_pportdata struct.
842  */
843 struct sdma_engine;
844 struct sdma_vl_map;
845 
846 #define BOARD_VERS_MAX 96 /* how long the version string can be */
847 #define SERIAL_MAX 16 /* length of the serial number */
848 
849 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
850 struct hfi1_devdata {
851 	struct hfi1_ibdev verbs_dev;     /* must be first */
852 	struct list_head list;
853 	/* pointers to related structs for this device */
854 	/* pci access data structure */
855 	struct pci_dev *pcidev;
856 	struct cdev user_cdev;
857 	struct cdev diag_cdev;
858 	struct cdev ui_cdev;
859 	struct device *user_device;
860 	struct device *diag_device;
861 	struct device *ui_device;
862 
863 	/* mem-mapped pointer to base of chip regs */
864 	u8 __iomem *kregbase;
865 	/* end of mem-mapped chip space excluding sendbuf and user regs */
866 	u8 __iomem *kregend;
867 	/* physical address of chip for io_remap, etc. */
868 	resource_size_t physaddr;
869 	/* Per VL data. Enough for all VLs but not all elements are set/used. */
870 	struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
871 	/* send context data */
872 	struct send_context_info *send_contexts;
873 	/* map hardware send contexts to software index */
874 	u8 *hw_to_sw;
875 	/* spinlock for allocating and releasing send context resources */
876 	spinlock_t sc_lock;
877 	/* lock for pio_map */
878 	spinlock_t pio_map_lock;
879 	/* Send Context initialization lock. */
880 	spinlock_t sc_init_lock;
881 	/* lock for sdma_map */
882 	spinlock_t                          sde_map_lock;
883 	/* array of kernel send contexts */
884 	struct send_context **kernel_send_context;
885 	/* array of vl maps */
886 	struct pio_vl_map __rcu *pio_map;
887 	/* default flags to last descriptor */
888 	u64 default_desc1;
889 
890 	/* fields common to all SDMA engines */
891 
892 	volatile __le64                    *sdma_heads_dma; /* DMA'ed by chip */
893 	dma_addr_t                          sdma_heads_phys;
894 	void                               *sdma_pad_dma; /* DMA'ed by chip */
895 	dma_addr_t                          sdma_pad_phys;
896 	/* for deallocation */
897 	size_t                              sdma_heads_size;
898 	/* number from the chip */
899 	u32                                 chip_sdma_engines;
900 	/* num used */
901 	u32                                 num_sdma;
902 	/* array of engines sized by num_sdma */
903 	struct sdma_engine                 *per_sdma;
904 	/* array of vl maps */
905 	struct sdma_vl_map __rcu           *sdma_map;
906 	/* SPC freeze waitqueue and variable */
907 	wait_queue_head_t		  sdma_unfreeze_wq;
908 	atomic_t			  sdma_unfreeze_count;
909 
910 	u32 lcb_access_count;		/* count of LCB users */
911 
912 	/* common data between shared ASIC HFIs in this OS */
913 	struct hfi1_asic_data *asic_data;
914 
915 	/* mem-mapped pointer to base of PIO buffers */
916 	void __iomem *piobase;
917 	/*
918 	 * write-combining mem-mapped pointer to base of RcvArray
919 	 * memory.
920 	 */
921 	void __iomem *rcvarray_wc;
922 	/*
923 	 * credit return base - a per-NUMA range of DMA address that
924 	 * the chip will use to update the per-context free counter
925 	 */
926 	struct credit_return_base *cr_base;
927 
928 	/* send context numbers and sizes for each type */
929 	struct sc_config_sizes sc_sizes[SC_MAX];
930 
931 	char *boardname; /* human readable board info */
932 
933 	/* reset value */
934 	u64 z_int_counter;
935 	u64 z_rcv_limit;
936 	u64 z_send_schedule;
937 
938 	u64 __percpu *send_schedule;
939 	/* number of receive contexts in use by the driver */
940 	u32 num_rcv_contexts;
941 	/* number of pio send contexts in use by the driver */
942 	u32 num_send_contexts;
943 	/*
944 	 * number of ctxts available for PSM open
945 	 */
946 	u32 freectxts;
947 	/* total number of available user/PSM contexts */
948 	u32 num_user_contexts;
949 	/* base receive interrupt timeout, in CSR units */
950 	u32 rcv_intr_timeout_csr;
951 
952 	u32 freezelen; /* max length of freezemsg */
953 	u64 __iomem *egrtidbase;
954 	spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
955 	spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
956 	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
957 	spinlock_t uctxt_lock; /* rcd and user context changes */
958 	struct mutex dc8051_lock; /* exclusive access to 8051 */
959 	struct workqueue_struct *update_cntr_wq;
960 	struct work_struct update_cntr_work;
961 	/* exclusive access to 8051 memory */
962 	spinlock_t dc8051_memlock;
963 	int dc8051_timed_out;	/* remember if the 8051 timed out */
964 	/*
965 	 * A page that will hold event notification bitmaps for all
966 	 * contexts. This page will be mapped into all processes.
967 	 */
968 	unsigned long *events;
969 	/*
970 	 * per unit status, see also portdata statusp
971 	 * mapped read-only into user processes so they can get unit and
972 	 * IB link status cheaply
973 	 */
974 	struct hfi1_status *status;
975 
976 	/* revision register shadow */
977 	u64 revision;
978 	/* Base GUID for device (network order) */
979 	u64 base_guid;
980 
981 	/* these are the "32 bit" regs */
982 
983 	/* value we put in kr_rcvhdrsize */
984 	u32 rcvhdrsize;
985 	/* number of receive contexts the chip supports */
986 	u32 chip_rcv_contexts;
987 	/* number of receive array entries */
988 	u32 chip_rcv_array_count;
989 	/* number of PIO send contexts the chip supports */
990 	u32 chip_send_contexts;
991 	/* number of bytes in the PIO memory buffer */
992 	u32 chip_pio_mem_size;
993 	/* number of bytes in the SDMA memory buffer */
994 	u32 chip_sdma_mem_size;
995 
996 	/* size of each rcvegrbuffer */
997 	u32 rcvegrbufsize;
998 	/* log2 of above */
999 	u16 rcvegrbufsize_shift;
1000 	/* both sides of the PCIe link are gen3 capable */
1001 	u8 link_gen3_capable;
1002 	/* default link down value (poll/sleep) */
1003 	u8 link_default;
1004 	/* localbus width (1, 2,4,8,16,32) from config space  */
1005 	u32 lbus_width;
1006 	/* localbus speed in MHz */
1007 	u32 lbus_speed;
1008 	int unit; /* unit # of this chip */
1009 	int node; /* home node of this chip */
1010 
1011 	/* save these PCI fields to restore after a reset */
1012 	u32 pcibar0;
1013 	u32 pcibar1;
1014 	u32 pci_rom;
1015 	u16 pci_command;
1016 	u16 pcie_devctl;
1017 	u16 pcie_lnkctl;
1018 	u16 pcie_devctl2;
1019 	u32 pci_msix0;
1020 	u32 pci_lnkctl3;
1021 	u32 pci_tph2;
1022 
1023 	/*
1024 	 * ASCII serial number, from flash, large enough for original
1025 	 * all digit strings, and longer serial number format
1026 	 */
1027 	u8 serial[SERIAL_MAX];
1028 	/* human readable board version */
1029 	u8 boardversion[BOARD_VERS_MAX];
1030 	u8 lbus_info[32]; /* human readable localbus info */
1031 	/* chip major rev, from CceRevision */
1032 	u8 majrev;
1033 	/* chip minor rev, from CceRevision */
1034 	u8 minrev;
1035 	/* hardware ID */
1036 	u8 hfi1_id;
1037 	/* implementation code */
1038 	u8 icode;
1039 	/* vAU of this device */
1040 	u8 vau;
1041 	/* vCU of this device */
1042 	u8 vcu;
1043 	/* link credits of this device */
1044 	u16 link_credits;
1045 	/* initial vl15 credits to use */
1046 	u16 vl15_init;
1047 
1048 	/*
1049 	 * Cached value for vl15buf, read during verify cap interrupt. VL15
1050 	 * credits are to be kept at 0 and set when handling the link-up
1051 	 * interrupt. This removes the possibility of receiving VL15 MAD
1052 	 * packets before this HFI is ready.
1053 	 */
1054 	u16 vl15buf_cached;
1055 
1056 	/* Misc small ints */
1057 	u8 n_krcv_queues;
1058 	u8 qos_shift;
1059 
1060 	u16 irev;	/* implementation revision */
1061 	u32 dc8051_ver; /* 8051 firmware version */
1062 
1063 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1064 	struct platform_config platform_config;
1065 	struct platform_config_cache pcfg_cache;
1066 
1067 	struct diag_client *diag_client;
1068 
1069 	/* MSI-X information */
1070 	struct hfi1_msix_entry *msix_entries;
1071 	u32 num_msix_entries;
1072 	u32 first_dyn_msix_idx;
1073 
1074 	/* INTx information */
1075 	u32 requested_intx_irq;		/* did we request one? */
1076 	char intx_name[MAX_NAME_SIZE];	/* INTx name */
1077 
1078 	/* general interrupt: mask of handled interrupts */
1079 	u64 gi_mask[CCE_NUM_INT_CSRS];
1080 
1081 	struct rcv_array_data rcv_entries;
1082 
1083 	/* cycle length of PS* counters in HW (in picoseconds) */
1084 	u16 psxmitwait_check_rate;
1085 
1086 	/*
1087 	 * 64 bit synthetic counters
1088 	 */
1089 	struct timer_list synth_stats_timer;
1090 
1091 	/*
1092 	 * device counters
1093 	 */
1094 	char *cntrnames;
1095 	size_t cntrnameslen;
1096 	size_t ndevcntrs;
1097 	u64 *cntrs;
1098 	u64 *scntrs;
1099 
1100 	/*
1101 	 * remembered values for synthetic counters
1102 	 */
1103 	u64 last_tx;
1104 	u64 last_rx;
1105 
1106 	/*
1107 	 * per-port counters
1108 	 */
1109 	size_t nportcntrs;
1110 	char *portcntrnames;
1111 	size_t portcntrnameslen;
1112 
1113 	struct err_info_rcvport err_info_rcvport;
1114 	struct err_info_constraint err_info_rcv_constraint;
1115 	struct err_info_constraint err_info_xmit_constraint;
1116 
1117 	atomic_t drop_packet;
1118 	u8 do_drop;
1119 	u8 err_info_uncorrectable;
1120 	u8 err_info_fmconfig;
1121 
1122 	/*
1123 	 * Software counters for the status bits defined by the
1124 	 * associated error status registers
1125 	 */
1126 	u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1127 	u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1128 	u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1129 	u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1130 	u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1131 	u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1132 	u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1133 
1134 	/* Software counter that spans all contexts */
1135 	u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1136 	/* Software counter that spans all DMA engines */
1137 	u64 sw_send_dma_eng_err_status_cnt[
1138 		NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1139 	/* Software counter that aggregates all cce_err_status errors */
1140 	u64 sw_cce_err_status_aggregate;
1141 	/* Software counter that aggregates all bypass packet rcv errors */
1142 	u64 sw_rcv_bypass_packet_errors;
1143 	/* receive interrupt function */
1144 	rhf_rcv_function_ptr normal_rhf_rcv_functions[8];
1145 
1146 	/* Save the enabled LCB error bits */
1147 	u64 lcb_err_en;
1148 
1149 	/*
1150 	 * Capability to have different send engines simply by changing a
1151 	 * pointer value.
1152 	 */
1153 	send_routine process_pio_send ____cacheline_aligned_in_smp;
1154 	send_routine process_dma_send;
1155 	void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1156 				u64 pbc, const void *from, size_t count);
1157 	int (*process_vnic_dma_send)(struct hfi1_devdata *dd, u8 q_idx,
1158 				     struct hfi1_vnic_vport_info *vinfo,
1159 				     struct sk_buff *skb, u64 pbc, u8 plen);
1160 	/* hfi1_pportdata, points to array of (physical) port-specific
1161 	 * data structs, indexed by pidx (0..n-1)
1162 	 */
1163 	struct hfi1_pportdata *pport;
1164 	/* receive context data */
1165 	struct hfi1_ctxtdata **rcd;
1166 	u64 __percpu *int_counter;
1167 	/* device (not port) flags, basically device capabilities */
1168 	u16 flags;
1169 	/* Number of physical ports available */
1170 	u8 num_pports;
1171 	/* Lowest context number which can be used by user processes or VNIC */
1172 	u8 first_dyn_alloc_ctxt;
1173 	/* adding a new field here would make it part of this cacheline */
1174 
1175 	/* seqlock for sc2vl */
1176 	seqlock_t sc2vl_lock ____cacheline_aligned_in_smp;
1177 	u64 sc2vl[4];
1178 	/* receive interrupt functions */
1179 	rhf_rcv_function_ptr *rhf_rcv_function_map;
1180 	u64 __percpu *rcv_limit;
1181 	u16 rhf_offset; /* offset of RHF within receive header entry */
1182 	/* adding a new field here would make it part of this cacheline */
1183 
1184 	/* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1185 	u8 oui1;
1186 	u8 oui2;
1187 	u8 oui3;
1188 	u8 dc_shutdown;
1189 
1190 	/* Timer and counter used to detect RcvBufOvflCnt changes */
1191 	struct timer_list rcverr_timer;
1192 
1193 	wait_queue_head_t event_queue;
1194 
1195 	/* receive context tail dummy address */
1196 	__le64 *rcvhdrtail_dummy_kvaddr;
1197 	dma_addr_t rcvhdrtail_dummy_dma;
1198 
1199 	u32 rcv_ovfl_cnt;
1200 	/* Serialize ASPM enable/disable between multiple verbs contexts */
1201 	spinlock_t aspm_lock;
1202 	/* Number of verbs contexts which have disabled ASPM */
1203 	atomic_t aspm_disabled_cnt;
1204 	/* Keeps track of user space clients */
1205 	atomic_t user_refcount;
1206 	/* Used to wait for outstanding user space clients before dev removal */
1207 	struct completion user_comp;
1208 
1209 	bool eprom_available;	/* true if EPROM is available for this device */
1210 	bool aspm_supported;	/* Does HW support ASPM */
1211 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
1212 	struct rhashtable *sdma_rht;
1213 
1214 	struct kobject kobj;
1215 
1216 	/* vnic data */
1217 	struct hfi1_vnic_data vnic;
1218 };
1219 
1220 static inline bool hfi1_vnic_is_rsm_full(struct hfi1_devdata *dd, int spare)
1221 {
1222 	return (dd->vnic.rmt_start + spare) > NUM_MAP_ENTRIES;
1223 }
1224 
1225 /* 8051 firmware version helper */
1226 #define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
1227 #define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
1228 #define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
1229 #define dc8051_ver_patch(a) ((a) & 0x0000ff)
1230 
1231 /* f_put_tid types */
1232 #define PT_EXPECTED 0
1233 #define PT_EAGER    1
1234 #define PT_INVALID  2
1235 
1236 struct tid_rb_node;
1237 struct mmu_rb_node;
1238 struct mmu_rb_handler;
1239 
1240 /* Private data for file operations */
1241 struct hfi1_filedata {
1242 	struct hfi1_devdata *dd;
1243 	struct hfi1_ctxtdata *uctxt;
1244 	struct hfi1_user_sdma_comp_q *cq;
1245 	struct hfi1_user_sdma_pkt_q *pq;
1246 	u16 subctxt;
1247 	/* for cpu affinity; -1 if none */
1248 	int rec_cpu_num;
1249 	u32 tid_n_pinned;
1250 	struct mmu_rb_handler *handler;
1251 	struct tid_rb_node **entry_to_rb;
1252 	spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1253 	u32 tid_limit;
1254 	u32 tid_used;
1255 	u32 *invalid_tids;
1256 	u32 invalid_tid_idx;
1257 	/* protect invalid_tids array and invalid_tid_idx */
1258 	spinlock_t invalid_lock;
1259 	struct mm_struct *mm;
1260 };
1261 
1262 extern struct list_head hfi1_dev_list;
1263 extern spinlock_t hfi1_devs_lock;
1264 struct hfi1_devdata *hfi1_lookup(int unit);
1265 extern u32 hfi1_cpulist_count;
1266 extern unsigned long *hfi1_cpulist;
1267 
1268 int hfi1_init(struct hfi1_devdata *dd, int reinit);
1269 int hfi1_count_active_units(void);
1270 
1271 int hfi1_diag_add(struct hfi1_devdata *dd);
1272 void hfi1_diag_remove(struct hfi1_devdata *dd);
1273 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1274 
1275 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1276 
1277 int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1278 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
1279 int hfi1_create_ctxts(struct hfi1_devdata *dd);
1280 struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt,
1281 					   int numa);
1282 void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
1283 			 struct hfi1_devdata *dd, u8 hw_pidx, u8 port);
1284 void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1285 
1286 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
1287 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1288 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1289 void set_all_slowpath(struct hfi1_devdata *dd);
1290 void hfi1_vnic_synchronize_irq(struct hfi1_devdata *dd);
1291 void hfi1_set_vnic_msix_info(struct hfi1_ctxtdata *rcd);
1292 void hfi1_reset_vnic_msix_info(struct hfi1_ctxtdata *rcd);
1293 
1294 extern const struct pci_device_id hfi1_pci_tbl[];
1295 
1296 /* receive packet handler dispositions */
1297 #define RCV_PKT_OK      0x0 /* keep going */
1298 #define RCV_PKT_LIMIT   0x1 /* stop, hit limit, start thread */
1299 #define RCV_PKT_DONE    0x2 /* stop, no more packets detected */
1300 
1301 /* calculate the current RHF address */
1302 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1303 {
1304 	return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->dd->rhf_offset;
1305 }
1306 
1307 int hfi1_reset_device(int);
1308 
1309 /* return the driver's idea of the logical OPA port state */
1310 static inline u32 driver_lstate(struct hfi1_pportdata *ppd)
1311 {
1312 	/*
1313 	 * The driver does some processing from the time the logical
1314 	 * link state is at INIT to the time the SM can be notified
1315 	 * as such. Return IB_PORT_DOWN until the software state
1316 	 * is ready.
1317 	 */
1318 	if (ppd->lstate == IB_PORT_INIT && !(ppd->host_link_state & HLS_UP))
1319 		return IB_PORT_DOWN;
1320 	else
1321 		return ppd->lstate;
1322 }
1323 
1324 void receive_interrupt_work(struct work_struct *work);
1325 
1326 /* extract service channel from header and rhf */
1327 static inline int hfi1_9B_get_sc5(struct ib_header *hdr, u64 rhf)
1328 {
1329 	return ib_get_sc(hdr) | ((!!(rhf_dc_info(rhf))) << 4);
1330 }
1331 
1332 #define HFI1_JKEY_WIDTH       16
1333 #define HFI1_JKEY_MASK        (BIT(16) - 1)
1334 #define HFI1_ADMIN_JKEY_RANGE 32
1335 
1336 /*
1337  * J_KEYs are split and allocated in the following groups:
1338  *   0 - 31    - users with administrator privileges
1339  *  32 - 63    - kernel protocols using KDETH packets
1340  *  64 - 65535 - all other users using KDETH packets
1341  */
1342 static inline u16 generate_jkey(kuid_t uid)
1343 {
1344 	u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1345 
1346 	if (capable(CAP_SYS_ADMIN))
1347 		jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1348 	else if (jkey < 64)
1349 		jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1350 
1351 	return jkey;
1352 }
1353 
1354 /*
1355  * active_egress_rate
1356  *
1357  * returns the active egress rate in units of [10^6 bits/sec]
1358  */
1359 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1360 {
1361 	u16 link_speed = ppd->link_speed_active;
1362 	u16 link_width = ppd->link_width_active;
1363 	u32 egress_rate;
1364 
1365 	if (link_speed == OPA_LINK_SPEED_25G)
1366 		egress_rate = 25000;
1367 	else /* assume OPA_LINK_SPEED_12_5G */
1368 		egress_rate = 12500;
1369 
1370 	switch (link_width) {
1371 	case OPA_LINK_WIDTH_4X:
1372 		egress_rate *= 4;
1373 		break;
1374 	case OPA_LINK_WIDTH_3X:
1375 		egress_rate *= 3;
1376 		break;
1377 	case OPA_LINK_WIDTH_2X:
1378 		egress_rate *= 2;
1379 		break;
1380 	default:
1381 		/* assume IB_WIDTH_1X */
1382 		break;
1383 	}
1384 
1385 	return egress_rate;
1386 }
1387 
1388 /*
1389  * egress_cycles
1390  *
1391  * Returns the number of 'fabric clock cycles' to egress a packet
1392  * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1393  * rate is (approximately) 805 MHz, the units of the returned value
1394  * are (1/805 MHz).
1395  */
1396 static inline u32 egress_cycles(u32 len, u32 rate)
1397 {
1398 	u32 cycles;
1399 
1400 	/*
1401 	 * cycles is:
1402 	 *
1403 	 *          (length) [bits] / (rate) [bits/sec]
1404 	 *  ---------------------------------------------------
1405 	 *  fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1406 	 */
1407 
1408 	cycles = len * 8; /* bits */
1409 	cycles *= 805;
1410 	cycles /= rate;
1411 
1412 	return cycles;
1413 }
1414 
1415 void set_link_ipg(struct hfi1_pportdata *ppd);
1416 void process_becn(struct hfi1_pportdata *ppd, u8 sl,  u16 rlid, u32 lqpn,
1417 		  u32 rqpn, u8 svc_type);
1418 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1419 		u32 pkey, u32 slid, u32 dlid, u8 sc5,
1420 		const struct ib_grh *old_grh);
1421 #define PKEY_CHECK_INVALID -1
1422 int egress_pkey_check(struct hfi1_pportdata *ppd, __be16 *lrh, __be32 *bth,
1423 		      u8 sc5, int8_t s_pkey_index);
1424 
1425 #define PACKET_EGRESS_TIMEOUT 350
1426 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1427 {
1428 	/* Pause at least 1us, to ensure chip returns all credits */
1429 	u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1430 
1431 	udelay(usec ? usec : 1);
1432 }
1433 
1434 /**
1435  * sc_to_vlt() reverse lookup sc to vl
1436  * @dd - devdata
1437  * @sc5 - 5 bit sc
1438  */
1439 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1440 {
1441 	unsigned seq;
1442 	u8 rval;
1443 
1444 	if (sc5 >= OPA_MAX_SCS)
1445 		return (u8)(0xff);
1446 
1447 	do {
1448 		seq = read_seqbegin(&dd->sc2vl_lock);
1449 		rval = *(((u8 *)dd->sc2vl) + sc5);
1450 	} while (read_seqretry(&dd->sc2vl_lock, seq));
1451 
1452 	return rval;
1453 }
1454 
1455 #define PKEY_MEMBER_MASK 0x8000
1456 #define PKEY_LOW_15_MASK 0x7fff
1457 
1458 /*
1459  * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1460  * being an entry from the ingress partition key table), return 0
1461  * otherwise. Use the matching criteria for ingress partition keys
1462  * specified in the OPAv1 spec., section 9.10.14.
1463  */
1464 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1465 {
1466 	u16 mkey = pkey & PKEY_LOW_15_MASK;
1467 	u16 ment = ent & PKEY_LOW_15_MASK;
1468 
1469 	if (mkey == ment) {
1470 		/*
1471 		 * If pkey[15] is clear (limited partition member),
1472 		 * is bit 15 in the corresponding table element
1473 		 * clear (limited member)?
1474 		 */
1475 		if (!(pkey & PKEY_MEMBER_MASK))
1476 			return !!(ent & PKEY_MEMBER_MASK);
1477 		return 1;
1478 	}
1479 	return 0;
1480 }
1481 
1482 /*
1483  * ingress_pkey_table_search - search the entire pkey table for
1484  * an entry which matches 'pkey'. return 0 if a match is found,
1485  * and 1 otherwise.
1486  */
1487 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1488 {
1489 	int i;
1490 
1491 	for (i = 0; i < MAX_PKEY_VALUES; i++) {
1492 		if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1493 			return 0;
1494 	}
1495 	return 1;
1496 }
1497 
1498 /*
1499  * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1500  * i.e., increment port_rcv_constraint_errors for the port, and record
1501  * the 'error info' for this failure.
1502  */
1503 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1504 				    u16 slid)
1505 {
1506 	struct hfi1_devdata *dd = ppd->dd;
1507 
1508 	incr_cntr64(&ppd->port_rcv_constraint_errors);
1509 	if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1510 		dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1511 		dd->err_info_rcv_constraint.slid = slid;
1512 		dd->err_info_rcv_constraint.pkey = pkey;
1513 	}
1514 }
1515 
1516 /*
1517  * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1518  * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1519  * is a hint as to the best place in the partition key table to begin
1520  * searching. This function should not be called on the data path because
1521  * of performance reasons. On datapath pkey check is expected to be done
1522  * by HW and rcv_pkey_check function should be called instead.
1523  */
1524 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1525 				     u8 sc5, u8 idx, u16 slid)
1526 {
1527 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1528 		return 0;
1529 
1530 	/* If SC15, pkey[0:14] must be 0x7fff */
1531 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1532 		goto bad;
1533 
1534 	/* Is the pkey = 0x0, or 0x8000? */
1535 	if ((pkey & PKEY_LOW_15_MASK) == 0)
1536 		goto bad;
1537 
1538 	/* The most likely matching pkey has index 'idx' */
1539 	if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1540 		return 0;
1541 
1542 	/* no match - try the whole table */
1543 	if (!ingress_pkey_table_search(ppd, pkey))
1544 		return 0;
1545 
1546 bad:
1547 	ingress_pkey_table_fail(ppd, pkey, slid);
1548 	return 1;
1549 }
1550 
1551 /*
1552  * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1553  * otherwise. It only ensures pkey is vlid for QP0. This function
1554  * should be called on the data path instead of ingress_pkey_check
1555  * as on data path, pkey check is done by HW (except for QP0).
1556  */
1557 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1558 				 u8 sc5, u16 slid)
1559 {
1560 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1561 		return 0;
1562 
1563 	/* If SC15, pkey[0:14] must be 0x7fff */
1564 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1565 		goto bad;
1566 
1567 	return 0;
1568 bad:
1569 	ingress_pkey_table_fail(ppd, pkey, slid);
1570 	return 1;
1571 }
1572 
1573 /* MTU handling */
1574 
1575 /* MTU enumeration, 256-4k match IB */
1576 #define OPA_MTU_0     0
1577 #define OPA_MTU_256   1
1578 #define OPA_MTU_512   2
1579 #define OPA_MTU_1024  3
1580 #define OPA_MTU_2048  4
1581 #define OPA_MTU_4096  5
1582 
1583 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1584 int mtu_to_enum(u32 mtu, int default_if_bad);
1585 u16 enum_to_mtu(int mtu);
1586 static inline int valid_ib_mtu(unsigned int mtu)
1587 {
1588 	return mtu == 256 || mtu == 512 ||
1589 		mtu == 1024 || mtu == 2048 ||
1590 		mtu == 4096;
1591 }
1592 
1593 static inline int valid_opa_max_mtu(unsigned int mtu)
1594 {
1595 	return mtu >= 2048 &&
1596 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1597 }
1598 
1599 int set_mtu(struct hfi1_pportdata *ppd);
1600 
1601 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
1602 void hfi1_disable_after_error(struct hfi1_devdata *dd);
1603 int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
1604 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
1605 
1606 int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
1607 int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
1608 
1609 void set_up_vau(struct hfi1_devdata *dd, u8 vau);
1610 void set_up_vl15(struct hfi1_devdata *dd, u16 vl15buf);
1611 void reset_link_credits(struct hfi1_devdata *dd);
1612 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1613 
1614 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1615 
1616 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1617 {
1618 	return ppd->dd;
1619 }
1620 
1621 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1622 {
1623 	return container_of(dev, struct hfi1_devdata, verbs_dev);
1624 }
1625 
1626 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1627 {
1628 	return dd_from_dev(to_idev(ibdev));
1629 }
1630 
1631 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1632 {
1633 	return container_of(ibp, struct hfi1_pportdata, ibport_data);
1634 }
1635 
1636 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1637 {
1638 	return container_of(rdi, struct hfi1_ibdev, rdi);
1639 }
1640 
1641 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u8 port)
1642 {
1643 	struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1644 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1645 
1646 	WARN_ON(pidx >= dd->num_pports);
1647 	return &dd->pport[pidx].ibport_data;
1648 }
1649 
1650 static inline struct hfi1_ibport *rcd_to_iport(struct hfi1_ctxtdata *rcd)
1651 {
1652 	return &rcd->ppd->ibport_data;
1653 }
1654 
1655 void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
1656 			       bool do_cnp);
1657 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt,
1658 			       bool do_cnp)
1659 {
1660 	struct ib_other_headers *ohdr = pkt->ohdr;
1661 	u32 bth1;
1662 
1663 	bth1 = be32_to_cpu(ohdr->bth[1]);
1664 	if (unlikely(bth1 & (IB_BECN_SMASK | IB_FECN_SMASK))) {
1665 		hfi1_process_ecn_slowpath(qp, pkt, do_cnp);
1666 		return !!(bth1 & IB_FECN_SMASK);
1667 	}
1668 	return false;
1669 }
1670 
1671 /*
1672  * Return the indexed PKEY from the port PKEY table.
1673  */
1674 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
1675 {
1676 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1677 	u16 ret;
1678 
1679 	if (index >= ARRAY_SIZE(ppd->pkeys))
1680 		ret = 0;
1681 	else
1682 		ret = ppd->pkeys[index];
1683 
1684 	return ret;
1685 }
1686 
1687 /*
1688  * Return the indexed GUID from the port GUIDs table.
1689  */
1690 static inline __be64 get_sguid(struct hfi1_ibport *ibp, unsigned int index)
1691 {
1692 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1693 
1694 	WARN_ON(index >= HFI1_GUIDS_PER_PORT);
1695 	return cpu_to_be64(ppd->guids[index]);
1696 }
1697 
1698 /*
1699  * Called by readers of cc_state only, must call under rcu_read_lock().
1700  */
1701 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
1702 {
1703 	return rcu_dereference(ppd->cc_state);
1704 }
1705 
1706 /*
1707  * Called by writers of cc_state only,  must call under cc_state_lock.
1708  */
1709 static inline
1710 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
1711 {
1712 	return rcu_dereference_protected(ppd->cc_state,
1713 					 lockdep_is_held(&ppd->cc_state_lock));
1714 }
1715 
1716 /*
1717  * values for dd->flags (_device_ related flags)
1718  */
1719 #define HFI1_INITTED           0x1    /* chip and driver up and initted */
1720 #define HFI1_PRESENT           0x2    /* chip accesses can be done */
1721 #define HFI1_FROZEN            0x4    /* chip in SPC freeze */
1722 #define HFI1_HAS_SDMA_TIMEOUT  0x8
1723 #define HFI1_HAS_SEND_DMA      0x10   /* Supports Send DMA */
1724 #define HFI1_FORCED_FREEZE     0x80   /* driver forced freeze mode */
1725 
1726 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
1727 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
1728 
1729 /* ctxt_flag bit offsets */
1730 		/* base context has not finished initializing */
1731 #define HFI1_CTXT_BASE_UNINIT 1
1732 		/* base context initaliation failed */
1733 #define HFI1_CTXT_BASE_FAILED 2
1734 		/* waiting for a packet to arrive */
1735 #define HFI1_CTXT_WAITING_RCV 3
1736 		/* waiting for an urgent packet to arrive */
1737 #define HFI1_CTXT_WAITING_URG 4
1738 
1739 /* free up any allocated data at closes */
1740 struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
1741 				  const struct pci_device_id *ent);
1742 void hfi1_free_devdata(struct hfi1_devdata *dd);
1743 struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra);
1744 
1745 /* LED beaconing functions */
1746 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
1747 			     unsigned int timeoff);
1748 void shutdown_led_override(struct hfi1_pportdata *ppd);
1749 
1750 #define HFI1_CREDIT_RETURN_RATE (100)
1751 
1752 /*
1753  * The number of words for the KDETH protocol field.  If this is
1754  * larger then the actual field used, then part of the payload
1755  * will be in the header.
1756  *
1757  * Optimally, we want this sized so that a typical case will
1758  * use full cache lines.  The typical local KDETH header would
1759  * be:
1760  *
1761  *	Bytes	Field
1762  *	  8	LRH
1763  *	 12	BHT
1764  *	 ??	KDETH
1765  *	  8	RHF
1766  *	---
1767  *	 28 + KDETH
1768  *
1769  * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
1770  */
1771 #define DEFAULT_RCVHDRSIZE 9
1772 
1773 /*
1774  * Maximal header byte count:
1775  *
1776  *	Bytes	Field
1777  *	  8	LRH
1778  *	 40	GRH (optional)
1779  *	 12	BTH
1780  *	 ??	KDETH
1781  *	  8	RHF
1782  *	---
1783  *	 68 + KDETH
1784  *
1785  * We also want to maintain a cache line alignment to assist DMA'ing
1786  * of the header bytes.  Round up to a good size.
1787  */
1788 #define DEFAULT_RCVHDR_ENTSIZE 32
1789 
1790 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
1791 			u32 nlocked, u32 npages);
1792 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
1793 			    size_t npages, bool writable, struct page **pages);
1794 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
1795 			     size_t npages, bool dirty);
1796 
1797 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1798 {
1799 	*((u64 *)rcd->rcvhdrtail_kvaddr) = 0ULL;
1800 }
1801 
1802 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1803 {
1804 	/*
1805 	 * volatile because it's a DMA target from the chip, routine is
1806 	 * inlined, and don't want register caching or reordering.
1807 	 */
1808 	return (u32)le64_to_cpu(*rcd->rcvhdrtail_kvaddr);
1809 }
1810 
1811 /*
1812  * sysfs interface.
1813  */
1814 
1815 extern const char ib_hfi1_version[];
1816 
1817 int hfi1_device_create(struct hfi1_devdata *dd);
1818 void hfi1_device_remove(struct hfi1_devdata *dd);
1819 
1820 int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
1821 			   struct kobject *kobj);
1822 int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
1823 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
1824 /* Hook for sysfs read of QSFP */
1825 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
1826 
1827 int hfi1_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent);
1828 void hfi1_pcie_cleanup(struct pci_dev *pdev);
1829 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
1830 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
1831 int pcie_speeds(struct hfi1_devdata *dd);
1832 void request_msix(struct hfi1_devdata *dd, u32 *nent,
1833 		  struct hfi1_msix_entry *entry);
1834 void hfi1_enable_intx(struct pci_dev *pdev);
1835 void restore_pci_variables(struct hfi1_devdata *dd);
1836 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
1837 int parse_platform_config(struct hfi1_devdata *dd);
1838 int get_platform_config_field(struct hfi1_devdata *dd,
1839 			      enum platform_config_table_type_encoding
1840 			      table_type, int table_index, int field_index,
1841 			      u32 *data, u32 len);
1842 
1843 const char *get_unit_name(int unit);
1844 const char *get_card_name(struct rvt_dev_info *rdi);
1845 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
1846 
1847 /*
1848  * Flush write combining store buffers (if present) and perform a write
1849  * barrier.
1850  */
1851 static inline void flush_wc(void)
1852 {
1853 	asm volatile("sfence" : : : "memory");
1854 }
1855 
1856 void handle_eflags(struct hfi1_packet *packet);
1857 int process_receive_ib(struct hfi1_packet *packet);
1858 int process_receive_bypass(struct hfi1_packet *packet);
1859 int process_receive_error(struct hfi1_packet *packet);
1860 int kdeth_process_expected(struct hfi1_packet *packet);
1861 int kdeth_process_eager(struct hfi1_packet *packet);
1862 int process_receive_invalid(struct hfi1_packet *packet);
1863 
1864 /* global module parameter variables */
1865 extern unsigned int hfi1_max_mtu;
1866 extern unsigned int hfi1_cu;
1867 extern unsigned int user_credit_return_threshold;
1868 extern int num_user_contexts;
1869 extern unsigned long n_krcvqs;
1870 extern uint krcvqs[];
1871 extern int krcvqsset;
1872 extern uint kdeth_qp;
1873 extern uint loopback;
1874 extern uint quick_linkup;
1875 extern uint rcv_intr_timeout;
1876 extern uint rcv_intr_count;
1877 extern uint rcv_intr_dynamic;
1878 extern ushort link_crc_mask;
1879 
1880 extern struct mutex hfi1_mutex;
1881 
1882 /* Number of seconds before our card status check...  */
1883 #define STATUS_TIMEOUT 60
1884 
1885 #define DRIVER_NAME		"hfi1"
1886 #define HFI1_USER_MINOR_BASE     0
1887 #define HFI1_TRACE_MINOR         127
1888 #define HFI1_NMINORS             255
1889 
1890 #define PCI_VENDOR_ID_INTEL 0x8086
1891 #define PCI_DEVICE_ID_INTEL0 0x24f0
1892 #define PCI_DEVICE_ID_INTEL1 0x24f1
1893 
1894 #define HFI1_PKT_USER_SC_INTEGRITY					    \
1895 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
1896 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
1897 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
1898 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
1899 
1900 #define HFI1_PKT_KERNEL_SC_INTEGRITY					    \
1901 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
1902 
1903 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
1904 						  u16 ctxt_type)
1905 {
1906 	u64 base_sc_integrity;
1907 
1908 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
1909 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
1910 		return 0;
1911 
1912 	base_sc_integrity =
1913 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1914 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
1915 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1916 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1917 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1918 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
1919 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1920 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1921 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1922 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
1923 	| SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1924 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1925 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
1926 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
1927 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
1928 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1929 
1930 	if (ctxt_type == SC_USER)
1931 		base_sc_integrity |= HFI1_PKT_USER_SC_INTEGRITY;
1932 	else
1933 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
1934 
1935 	/* turn on send-side job key checks if !A0 */
1936 	if (!is_ax(dd))
1937 		base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1938 
1939 	return base_sc_integrity;
1940 }
1941 
1942 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
1943 {
1944 	u64 base_sdma_integrity;
1945 
1946 	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
1947 	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
1948 		return 0;
1949 
1950 	base_sdma_integrity =
1951 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1952 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1953 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1954 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1955 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1956 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1957 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1958 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
1959 	| SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1960 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1961 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
1962 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
1963 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
1964 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1965 
1966 	if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
1967 		base_sdma_integrity |=
1968 		SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
1969 
1970 	/* turn on send-side job key checks if !A0 */
1971 	if (!is_ax(dd))
1972 		base_sdma_integrity |=
1973 			SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1974 
1975 	return base_sdma_integrity;
1976 }
1977 
1978 /*
1979  * hfi1_early_err is used (only!) to print early errors before devdata is
1980  * allocated, or when dd->pcidev may not be valid, and at the tail end of
1981  * cleanup when devdata may have been freed, etc.  hfi1_dev_porterr is
1982  * the same as dd_dev_err, but is used when the message really needs
1983  * the IB port# to be definitive as to what's happening..
1984  */
1985 #define hfi1_early_err(dev, fmt, ...) \
1986 	dev_err(dev, fmt, ##__VA_ARGS__)
1987 
1988 #define hfi1_early_info(dev, fmt, ...) \
1989 	dev_info(dev, fmt, ##__VA_ARGS__)
1990 
1991 #define dd_dev_emerg(dd, fmt, ...) \
1992 	dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
1993 		  get_unit_name((dd)->unit), ##__VA_ARGS__)
1994 #define dd_dev_err(dd, fmt, ...) \
1995 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1996 			get_unit_name((dd)->unit), ##__VA_ARGS__)
1997 #define dd_dev_warn(dd, fmt, ...) \
1998 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
1999 			get_unit_name((dd)->unit), ##__VA_ARGS__)
2000 
2001 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
2002 	dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2003 			get_unit_name((dd)->unit), ##__VA_ARGS__)
2004 
2005 #define dd_dev_info(dd, fmt, ...) \
2006 	dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
2007 			get_unit_name((dd)->unit), ##__VA_ARGS__)
2008 
2009 #define dd_dev_info_ratelimited(dd, fmt, ...) \
2010 	dev_info_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2011 			get_unit_name((dd)->unit), ##__VA_ARGS__)
2012 
2013 #define dd_dev_dbg(dd, fmt, ...) \
2014 	dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
2015 		get_unit_name((dd)->unit), ##__VA_ARGS__)
2016 
2017 #define hfi1_dev_porterr(dd, port, fmt, ...) \
2018 	dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
2019 			get_unit_name((dd)->unit), (port), ##__VA_ARGS__)
2020 
2021 /*
2022  * this is used for formatting hw error messages...
2023  */
2024 struct hfi1_hwerror_msgs {
2025 	u64 mask;
2026 	const char *msg;
2027 	size_t sz;
2028 };
2029 
2030 /* in intr.c... */
2031 void hfi1_format_hwerrors(u64 hwerrs,
2032 			  const struct hfi1_hwerror_msgs *hwerrmsgs,
2033 			  size_t nhwerrmsgs, char *msg, size_t lmsg);
2034 
2035 #define USER_OPCODE_CHECK_VAL 0xC0
2036 #define USER_OPCODE_CHECK_MASK 0xC0
2037 #define OPCODE_CHECK_VAL_DISABLED 0x0
2038 #define OPCODE_CHECK_MASK_DISABLED 0x0
2039 
2040 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
2041 {
2042 	struct hfi1_pportdata *ppd;
2043 	int i;
2044 
2045 	dd->z_int_counter = get_all_cpu_total(dd->int_counter);
2046 	dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
2047 	dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
2048 
2049 	ppd = (struct hfi1_pportdata *)(dd + 1);
2050 	for (i = 0; i < dd->num_pports; i++, ppd++) {
2051 		ppd->ibport_data.rvp.z_rc_acks =
2052 			get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
2053 		ppd->ibport_data.rvp.z_rc_qacks =
2054 			get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
2055 	}
2056 }
2057 
2058 /* Control LED state */
2059 static inline void setextled(struct hfi1_devdata *dd, u32 on)
2060 {
2061 	if (on)
2062 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
2063 	else
2064 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
2065 }
2066 
2067 /* return the i2c resource given the target */
2068 static inline u32 i2c_target(u32 target)
2069 {
2070 	return target ? CR_I2C2 : CR_I2C1;
2071 }
2072 
2073 /* return the i2c chain chip resource that this HFI uses for QSFP */
2074 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2075 {
2076 	return i2c_target(dd->hfi1_id);
2077 }
2078 
2079 /* Is this device integrated or discrete? */
2080 static inline bool is_integrated(struct hfi1_devdata *dd)
2081 {
2082 	return dd->pcidev->device == PCI_DEVICE_ID_INTEL1;
2083 }
2084 
2085 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2086 
2087 #define DD_DEV_ENTRY(dd)       __string(dev, dev_name(&(dd)->pcidev->dev))
2088 #define DD_DEV_ASSIGN(dd)      __assign_str(dev, dev_name(&(dd)->pcidev->dev))
2089 
2090 #define packettype_name(etype) { RHF_RCV_TYPE_##etype, #etype }
2091 #define show_packettype(etype)                  \
2092 __print_symbolic(etype,                         \
2093 	packettype_name(EXPECTED),              \
2094 	packettype_name(EAGER),                 \
2095 	packettype_name(IB),                    \
2096 	packettype_name(ERROR),                 \
2097 	packettype_name(BYPASS))
2098 
2099 #define ib_opcode_name(opcode) { IB_OPCODE_##opcode, #opcode  }
2100 #define show_ib_opcode(opcode)                             \
2101 __print_symbolic(opcode,                                   \
2102 	ib_opcode_name(RC_SEND_FIRST),                     \
2103 	ib_opcode_name(RC_SEND_MIDDLE),                    \
2104 	ib_opcode_name(RC_SEND_LAST),                      \
2105 	ib_opcode_name(RC_SEND_LAST_WITH_IMMEDIATE),       \
2106 	ib_opcode_name(RC_SEND_ONLY),                      \
2107 	ib_opcode_name(RC_SEND_ONLY_WITH_IMMEDIATE),       \
2108 	ib_opcode_name(RC_RDMA_WRITE_FIRST),               \
2109 	ib_opcode_name(RC_RDMA_WRITE_MIDDLE),              \
2110 	ib_opcode_name(RC_RDMA_WRITE_LAST),                \
2111 	ib_opcode_name(RC_RDMA_WRITE_LAST_WITH_IMMEDIATE), \
2112 	ib_opcode_name(RC_RDMA_WRITE_ONLY),                \
2113 	ib_opcode_name(RC_RDMA_WRITE_ONLY_WITH_IMMEDIATE), \
2114 	ib_opcode_name(RC_RDMA_READ_REQUEST),              \
2115 	ib_opcode_name(RC_RDMA_READ_RESPONSE_FIRST),       \
2116 	ib_opcode_name(RC_RDMA_READ_RESPONSE_MIDDLE),      \
2117 	ib_opcode_name(RC_RDMA_READ_RESPONSE_LAST),        \
2118 	ib_opcode_name(RC_RDMA_READ_RESPONSE_ONLY),        \
2119 	ib_opcode_name(RC_ACKNOWLEDGE),                    \
2120 	ib_opcode_name(RC_ATOMIC_ACKNOWLEDGE),             \
2121 	ib_opcode_name(RC_COMPARE_SWAP),                   \
2122 	ib_opcode_name(RC_FETCH_ADD),                      \
2123 	ib_opcode_name(UC_SEND_FIRST),                     \
2124 	ib_opcode_name(UC_SEND_MIDDLE),                    \
2125 	ib_opcode_name(UC_SEND_LAST),                      \
2126 	ib_opcode_name(UC_SEND_LAST_WITH_IMMEDIATE),       \
2127 	ib_opcode_name(UC_SEND_ONLY),                      \
2128 	ib_opcode_name(UC_SEND_ONLY_WITH_IMMEDIATE),       \
2129 	ib_opcode_name(UC_RDMA_WRITE_FIRST),               \
2130 	ib_opcode_name(UC_RDMA_WRITE_MIDDLE),              \
2131 	ib_opcode_name(UC_RDMA_WRITE_LAST),                \
2132 	ib_opcode_name(UC_RDMA_WRITE_LAST_WITH_IMMEDIATE), \
2133 	ib_opcode_name(UC_RDMA_WRITE_ONLY),                \
2134 	ib_opcode_name(UC_RDMA_WRITE_ONLY_WITH_IMMEDIATE), \
2135 	ib_opcode_name(UD_SEND_ONLY),                      \
2136 	ib_opcode_name(UD_SEND_ONLY_WITH_IMMEDIATE),       \
2137 	ib_opcode_name(CNP))
2138 #endif                          /* _HFI1_KERNEL_H */
2139