xref: /linux/drivers/infiniband/hw/hfi1/hfi.h (revision 38b8767462120c62a5046b529c80b06861f9ac85)
1 #ifndef _HFI1_KERNEL_H
2 #define _HFI1_KERNEL_H
3 /*
4  * Copyright(c) 2015, 2016 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 <rdma/rdma_vt.h>
68 
69 #include "chip_registers.h"
70 #include "common.h"
71 #include "verbs.h"
72 #include "pio.h"
73 #include "chip.h"
74 #include "mad.h"
75 #include "qsfp.h"
76 #include "platform.h"
77 #include "affinity.h"
78 
79 /* bumped 1 from s/w major version of TrueScale */
80 #define HFI1_CHIP_VERS_MAJ 3U
81 
82 /* don't care about this except printing */
83 #define HFI1_CHIP_VERS_MIN 0U
84 
85 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
86 #define HFI1_OUI 0x001175
87 #define HFI1_OUI_LSB 40
88 
89 #define DROP_PACKET_OFF		0
90 #define DROP_PACKET_ON		1
91 
92 extern unsigned long hfi1_cap_mask;
93 #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap)
94 #define HFI1_CAP_UGET_MASK(mask, cap) \
95 	(((mask) >> HFI1_CAP_USER_SHIFT) & HFI1_CAP_##cap)
96 #define HFI1_CAP_KGET(cap) (HFI1_CAP_KGET_MASK(hfi1_cap_mask, cap))
97 #define HFI1_CAP_UGET(cap) (HFI1_CAP_UGET_MASK(hfi1_cap_mask, cap))
98 #define HFI1_CAP_IS_KSET(cap) (!!HFI1_CAP_KGET(cap))
99 #define HFI1_CAP_IS_USET(cap) (!!HFI1_CAP_UGET(cap))
100 #define HFI1_MISC_GET() ((hfi1_cap_mask >> HFI1_CAP_MISC_SHIFT) & \
101 			HFI1_CAP_MISC_MASK)
102 /* Offline Disabled Reason is 4-bits */
103 #define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
104 
105 /*
106  * Control context is always 0 and handles the error packets.
107  * It also handles the VL15 and multicast packets.
108  */
109 #define HFI1_CTRL_CTXT    0
110 
111 /*
112  * Driver context will store software counters for each of the events
113  * associated with these status registers
114  */
115 #define NUM_CCE_ERR_STATUS_COUNTERS 41
116 #define NUM_RCV_ERR_STATUS_COUNTERS 64
117 #define NUM_MISC_ERR_STATUS_COUNTERS 13
118 #define NUM_SEND_PIO_ERR_STATUS_COUNTERS 36
119 #define NUM_SEND_DMA_ERR_STATUS_COUNTERS 4
120 #define NUM_SEND_EGRESS_ERR_STATUS_COUNTERS 64
121 #define NUM_SEND_ERR_STATUS_COUNTERS 3
122 #define NUM_SEND_CTXT_ERR_STATUS_COUNTERS 5
123 #define NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS 24
124 
125 /*
126  * per driver stats, either not device nor port-specific, or
127  * summed over all of the devices and ports.
128  * They are described by name via ipathfs filesystem, so layout
129  * and number of elements can change without breaking compatibility.
130  * If members are added or deleted hfi1_statnames[] in debugfs.c must
131  * change to match.
132  */
133 struct hfi1_ib_stats {
134 	__u64 sps_ints; /* number of interrupts handled */
135 	__u64 sps_errints; /* number of error interrupts */
136 	__u64 sps_txerrs; /* tx-related packet errors */
137 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
138 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
139 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
140 	__u64 sps_ctxts; /* number of contexts currently open */
141 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
142 	__u64 sps_buffull;
143 	__u64 sps_hdrfull;
144 };
145 
146 extern struct hfi1_ib_stats hfi1_stats;
147 extern const struct pci_error_handlers hfi1_pci_err_handler;
148 
149 /*
150  * First-cut criterion for "device is active" is
151  * two thousand dwords combined Tx, Rx traffic per
152  * 5-second interval. SMA packets are 64 dwords,
153  * and occur "a few per second", presumably each way.
154  */
155 #define HFI1_TRAFFIC_ACTIVE_THRESHOLD (2000)
156 
157 /*
158  * Below contains all data related to a single context (formerly called port).
159  */
160 
161 #ifdef CONFIG_DEBUG_FS
162 struct hfi1_opcode_stats_perctx;
163 #endif
164 
165 struct ctxt_eager_bufs {
166 	ssize_t size;            /* total size of eager buffers */
167 	u32 count;               /* size of buffers array */
168 	u32 numbufs;             /* number of buffers allocated */
169 	u32 alloced;             /* number of rcvarray entries used */
170 	u32 rcvtid_size;         /* size of each eager rcv tid */
171 	u32 threshold;           /* head update threshold */
172 	struct eager_buffer {
173 		void *addr;
174 		dma_addr_t phys;
175 		ssize_t len;
176 	} *buffers;
177 	struct {
178 		void *addr;
179 		dma_addr_t phys;
180 	} *rcvtids;
181 };
182 
183 struct exp_tid_set {
184 	struct list_head list;
185 	u32 count;
186 };
187 
188 struct hfi1_ctxtdata {
189 	/* shadow the ctxt's RcvCtrl register */
190 	u64 rcvctrl;
191 	/* rcvhdrq base, needs mmap before useful */
192 	void *rcvhdrq;
193 	/* kernel virtual address where hdrqtail is updated */
194 	volatile __le64 *rcvhdrtail_kvaddr;
195 	/*
196 	 * Shared page for kernel to signal user processes that send buffers
197 	 * need disarming.  The process should call HFI1_CMD_DISARM_BUFS
198 	 * or HFI1_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
199 	 */
200 	unsigned long *user_event_mask;
201 	/* when waiting for rcv or pioavail */
202 	wait_queue_head_t wait;
203 	/* rcvhdrq size (for freeing) */
204 	size_t rcvhdrq_size;
205 	/* number of rcvhdrq entries */
206 	u16 rcvhdrq_cnt;
207 	/* size of each of the rcvhdrq entries */
208 	u16 rcvhdrqentsize;
209 	/* mmap of hdrq, must fit in 44 bits */
210 	dma_addr_t rcvhdrq_phys;
211 	dma_addr_t rcvhdrqtailaddr_phys;
212 	struct ctxt_eager_bufs egrbufs;
213 	/* this receive context's assigned PIO ACK send context */
214 	struct send_context *sc;
215 
216 	/* dynamic receive available interrupt timeout */
217 	u32 rcvavail_timeout;
218 	/*
219 	 * number of opens (including slave sub-contexts) on this instance
220 	 * (ignoring forks, dup, etc. for now)
221 	 */
222 	int cnt;
223 	/*
224 	 * how much space to leave at start of eager TID entries for
225 	 * protocol use, on each TID
226 	 */
227 	/* instead of calculating it */
228 	unsigned ctxt;
229 	/* non-zero if ctxt is being shared. */
230 	u16 subctxt_cnt;
231 	/* non-zero if ctxt is being shared. */
232 	u16 subctxt_id;
233 	u8 uuid[16];
234 	/* job key */
235 	u16 jkey;
236 	/* number of RcvArray groups for this context. */
237 	u32 rcv_array_groups;
238 	/* index of first eager TID entry. */
239 	u32 eager_base;
240 	/* number of expected TID entries */
241 	u32 expected_count;
242 	/* index of first expected TID entry. */
243 	u32 expected_base;
244 
245 	struct exp_tid_set tid_group_list;
246 	struct exp_tid_set tid_used_list;
247 	struct exp_tid_set tid_full_list;
248 
249 	/* lock protecting all Expected TID data */
250 	struct mutex exp_lock;
251 	/* number of pio bufs for this ctxt (all procs, if shared) */
252 	u32 piocnt;
253 	/* first pio buffer for this ctxt */
254 	u32 pio_base;
255 	/* chip offset of PIO buffers for this ctxt */
256 	u32 piobufs;
257 	/* per-context configuration flags */
258 	unsigned long flags;
259 	/* per-context event flags for fileops/intr communication */
260 	unsigned long event_flags;
261 	/* WAIT_RCV that timed out, no interrupt */
262 	u32 rcvwait_to;
263 	/* WAIT_PIO that timed out, no interrupt */
264 	u32 piowait_to;
265 	/* WAIT_RCV already happened, no wait */
266 	u32 rcvnowait;
267 	/* WAIT_PIO already happened, no wait */
268 	u32 pionowait;
269 	/* total number of polled urgent packets */
270 	u32 urgent;
271 	/* saved total number of polled urgent packets for poll edge trigger */
272 	u32 urgent_poll;
273 	/* same size as task_struct .comm[], command that opened context */
274 	char comm[TASK_COMM_LEN];
275 	/* so file ops can get at unit */
276 	struct hfi1_devdata *dd;
277 	/* so functions that need physical port can get it easily */
278 	struct hfi1_pportdata *ppd;
279 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
280 	void *subctxt_uregbase;
281 	/* An array of pages for the eager receive buffers * N */
282 	void *subctxt_rcvegrbuf;
283 	/* An array of pages for the eager header queue entries * N */
284 	void *subctxt_rcvhdr_base;
285 	/* The version of the library which opened this ctxt */
286 	u32 userversion;
287 	/* Bitmask of active slaves */
288 	u32 active_slaves;
289 	/* Type of packets or conditions we want to poll for */
290 	u16 poll_type;
291 	/* receive packet sequence counter */
292 	u8 seq_cnt;
293 	u8 redirect_seq_cnt;
294 	/* ctxt rcvhdrq head offset */
295 	u32 head;
296 	u32 pkt_count;
297 	/* QPs waiting for context processing */
298 	struct list_head qp_wait_list;
299 	/* interrupt handling */
300 	u64 imask;	/* clear interrupt mask */
301 	int ireg;	/* clear interrupt register */
302 	unsigned numa_id; /* numa node of this context */
303 	/* verbs stats per CTX */
304 	struct hfi1_opcode_stats_perctx *opstats;
305 	/*
306 	 * This is the kernel thread that will keep making
307 	 * progress on the user sdma requests behind the scenes.
308 	 * There is one per context (shared contexts use the master's).
309 	 */
310 	struct task_struct *progress;
311 	struct list_head sdma_queues;
312 	/* protect sdma queues */
313 	spinlock_t sdma_qlock;
314 
315 	/* Is ASPM interrupt supported for this context */
316 	bool aspm_intr_supported;
317 	/* ASPM state (enabled/disabled) for this context */
318 	bool aspm_enabled;
319 	/* Timer for re-enabling ASPM if interrupt activity quietens down */
320 	struct timer_list aspm_timer;
321 	/* Lock to serialize between intr, timer intr and user threads */
322 	spinlock_t aspm_lock;
323 	/* Is ASPM processing enabled for this context (in intr context) */
324 	bool aspm_intr_enable;
325 	/* Last interrupt timestamp */
326 	ktime_t aspm_ts_last_intr;
327 	/* Last timestamp at which we scheduled a timer for this context */
328 	ktime_t aspm_ts_timer_sched;
329 
330 	/*
331 	 * The interrupt handler for a particular receive context can vary
332 	 * throughout it's lifetime. This is not a lock protected data member so
333 	 * it must be updated atomically and the prev and new value must always
334 	 * be valid. Worst case is we process an extra interrupt and up to 64
335 	 * packets with the wrong interrupt handler.
336 	 */
337 	int (*do_interrupt)(struct hfi1_ctxtdata *rcd, int threaded);
338 };
339 
340 /*
341  * Represents a single packet at a high level. Put commonly computed things in
342  * here so we do not have to keep doing them over and over. The rule of thumb is
343  * if something is used one time to derive some value, store that something in
344  * here. If it is used multiple times, then store the result of that derivation
345  * in here.
346  */
347 struct hfi1_packet {
348 	void *ebuf;
349 	void *hdr;
350 	struct hfi1_ctxtdata *rcd;
351 	__le32 *rhf_addr;
352 	struct rvt_qp *qp;
353 	struct hfi1_other_headers *ohdr;
354 	u64 rhf;
355 	u32 maxcnt;
356 	u32 rhqoff;
357 	u32 hdrqtail;
358 	int numpkt;
359 	u16 tlen;
360 	u16 hlen;
361 	s16 etail;
362 	u16 rsize;
363 	u8 updegr;
364 	u8 rcv_flags;
365 	u8 etype;
366 };
367 
368 /*
369  * Private data for snoop/capture support.
370  */
371 struct hfi1_snoop_data {
372 	int mode_flag;
373 	struct cdev cdev;
374 	struct device *class_dev;
375 	/* protect snoop data */
376 	spinlock_t snoop_lock;
377 	struct list_head queue;
378 	wait_queue_head_t waitq;
379 	void *filter_value;
380 	int (*filter_callback)(void *hdr, void *data, void *value);
381 	u64 dcc_cfg; /* saved value of DCC Cfg register */
382 };
383 
384 /* snoop mode_flag values */
385 #define HFI1_PORT_SNOOP_MODE     1U
386 #define HFI1_PORT_CAPTURE_MODE   2U
387 
388 struct rvt_sge_state;
389 
390 /*
391  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
392  * Mostly for MADs that set or query link parameters, also ipath
393  * config interfaces
394  */
395 #define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
396 #define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
397 #define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
398 #define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
399 #define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
400 #define HFI1_IB_CFG_SPD 5 /* current Link spd */
401 #define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
402 #define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
403 #define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
404 #define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
405 #define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
406 #define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
407 #define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
408 #define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
409 #define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
410 #define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
411 #define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
412 #define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
413 #define HFI1_IB_CFG_VL_HIGH_LIMIT 19
414 #define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
415 #define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
416 
417 /*
418  * HFI or Host Link States
419  *
420  * These describe the states the driver thinks the logical and physical
421  * states are in.  Used as an argument to set_link_state().  Implemented
422  * as bits for easy multi-state checking.  The actual state can only be
423  * one.
424  */
425 #define __HLS_UP_INIT_BP	0
426 #define __HLS_UP_ARMED_BP	1
427 #define __HLS_UP_ACTIVE_BP	2
428 #define __HLS_DN_DOWNDEF_BP	3	/* link down default */
429 #define __HLS_DN_POLL_BP	4
430 #define __HLS_DN_DISABLE_BP	5
431 #define __HLS_DN_OFFLINE_BP	6
432 #define __HLS_VERIFY_CAP_BP	7
433 #define __HLS_GOING_UP_BP	8
434 #define __HLS_GOING_OFFLINE_BP  9
435 #define __HLS_LINK_COOLDOWN_BP 10
436 
437 #define HLS_UP_INIT	  BIT(__HLS_UP_INIT_BP)
438 #define HLS_UP_ARMED	  BIT(__HLS_UP_ARMED_BP)
439 #define HLS_UP_ACTIVE	  BIT(__HLS_UP_ACTIVE_BP)
440 #define HLS_DN_DOWNDEF	  BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
441 #define HLS_DN_POLL	  BIT(__HLS_DN_POLL_BP)
442 #define HLS_DN_DISABLE	  BIT(__HLS_DN_DISABLE_BP)
443 #define HLS_DN_OFFLINE	  BIT(__HLS_DN_OFFLINE_BP)
444 #define HLS_VERIFY_CAP	  BIT(__HLS_VERIFY_CAP_BP)
445 #define HLS_GOING_UP	  BIT(__HLS_GOING_UP_BP)
446 #define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
447 #define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
448 
449 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
450 #define HLS_DOWN ~(HLS_UP)
451 
452 /* use this MTU size if none other is given */
453 #define HFI1_DEFAULT_ACTIVE_MTU 10240
454 /* use this MTU size as the default maximum */
455 #define HFI1_DEFAULT_MAX_MTU 10240
456 /* default partition key */
457 #define DEFAULT_PKEY 0xffff
458 
459 /*
460  * Possible fabric manager config parameters for fm_{get,set}_table()
461  */
462 #define FM_TBL_VL_HIGH_ARB		1 /* Get/set VL high prio weights */
463 #define FM_TBL_VL_LOW_ARB		2 /* Get/set VL low prio weights */
464 #define FM_TBL_BUFFER_CONTROL		3 /* Get/set Buffer Control */
465 #define FM_TBL_SC2VLNT			4 /* Get/set SC->VLnt */
466 #define FM_TBL_VL_PREEMPT_ELEMS		5 /* Get (no set) VL preempt elems */
467 #define FM_TBL_VL_PREEMPT_MATRIX	6 /* Get (no set) VL preempt matrix */
468 
469 /*
470  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
471  * these are bits so they can be combined, e.g.
472  * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
473  */
474 #define HFI1_RCVCTRL_TAILUPD_ENB 0x01
475 #define HFI1_RCVCTRL_TAILUPD_DIS 0x02
476 #define HFI1_RCVCTRL_CTXT_ENB 0x04
477 #define HFI1_RCVCTRL_CTXT_DIS 0x08
478 #define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
479 #define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
480 #define HFI1_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
481 #define HFI1_RCVCTRL_PKEY_DIS 0x80
482 #define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
483 #define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
484 #define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
485 #define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
486 #define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
487 #define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
488 #define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
489 #define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
490 
491 /* partition enforcement flags */
492 #define HFI1_PART_ENFORCE_IN	0x1
493 #define HFI1_PART_ENFORCE_OUT	0x2
494 
495 /* how often we check for synthetic counter wrap around */
496 #define SYNTH_CNT_TIME 2
497 
498 /* Counter flags */
499 #define CNTR_NORMAL		0x0 /* Normal counters, just read register */
500 #define CNTR_SYNTH		0x1 /* Synthetic counters, saturate at all 1s */
501 #define CNTR_DISABLED		0x2 /* Disable this counter */
502 #define CNTR_32BIT		0x4 /* Simulate 64 bits for this counter */
503 #define CNTR_VL			0x8 /* Per VL counter */
504 #define CNTR_SDMA              0x10
505 #define CNTR_INVALID_VL		-1  /* Specifies invalid VL */
506 #define CNTR_MODE_W		0x0
507 #define CNTR_MODE_R		0x1
508 
509 /* VLs Supported/Operational */
510 #define HFI1_MIN_VLS_SUPPORTED 1
511 #define HFI1_MAX_VLS_SUPPORTED 8
512 
513 static inline void incr_cntr64(u64 *cntr)
514 {
515 	if (*cntr < (u64)-1LL)
516 		(*cntr)++;
517 }
518 
519 static inline void incr_cntr32(u32 *cntr)
520 {
521 	if (*cntr < (u32)-1LL)
522 		(*cntr)++;
523 }
524 
525 #define MAX_NAME_SIZE 64
526 struct hfi1_msix_entry {
527 	enum irq_type type;
528 	struct msix_entry msix;
529 	void *arg;
530 	char name[MAX_NAME_SIZE];
531 	cpumask_t mask;
532 };
533 
534 /* per-SL CCA information */
535 struct cca_timer {
536 	struct hrtimer hrtimer;
537 	struct hfi1_pportdata *ppd; /* read-only */
538 	int sl; /* read-only */
539 	u16 ccti; /* read/write - current value of CCTI */
540 };
541 
542 struct link_down_reason {
543 	/*
544 	 * SMA-facing value.  Should be set from .latest when
545 	 * HLS_UP_* -> HLS_DN_* transition actually occurs.
546 	 */
547 	u8 sma;
548 	u8 latest;
549 };
550 
551 enum {
552 	LO_PRIO_TABLE,
553 	HI_PRIO_TABLE,
554 	MAX_PRIO_TABLE
555 };
556 
557 struct vl_arb_cache {
558 	/* protect vl arb cache */
559 	spinlock_t lock;
560 	struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
561 };
562 
563 /*
564  * The structure below encapsulates data relevant to a physical IB Port.
565  * Current chips support only one such port, but the separation
566  * clarifies things a bit. Note that to conform to IB conventions,
567  * port-numbers are one-based. The first or only port is port1.
568  */
569 struct hfi1_pportdata {
570 	struct hfi1_ibport ibport_data;
571 
572 	struct hfi1_devdata *dd;
573 	struct kobject pport_cc_kobj;
574 	struct kobject sc2vl_kobj;
575 	struct kobject sl2sc_kobj;
576 	struct kobject vl2mtu_kobj;
577 
578 	/* PHY support */
579 	u32 port_type;
580 	struct qsfp_data qsfp_info;
581 
582 	/* GUID for this interface, in host order */
583 	u64 guid;
584 	/* GUID for peer interface, in host order */
585 	u64 neighbor_guid;
586 
587 	/* up or down physical link state */
588 	u32 linkup;
589 
590 	/*
591 	 * this address is mapped read-only into user processes so they can
592 	 * get status cheaply, whenever they want.  One qword of status per port
593 	 */
594 	u64 *statusp;
595 
596 	/* SendDMA related entries */
597 
598 	struct workqueue_struct *hfi1_wq;
599 
600 	/* move out of interrupt context */
601 	struct work_struct link_vc_work;
602 	struct work_struct link_up_work;
603 	struct work_struct link_down_work;
604 	struct work_struct sma_message_work;
605 	struct work_struct freeze_work;
606 	struct work_struct link_downgrade_work;
607 	struct work_struct link_bounce_work;
608 	struct delayed_work start_link_work;
609 	/* host link state variables */
610 	struct mutex hls_lock;
611 	u32 host_link_state;
612 
613 	spinlock_t            sdma_alllock ____cacheline_aligned_in_smp;
614 
615 	u32 lstate;	/* logical link state */
616 
617 	/* these are the "32 bit" regs */
618 
619 	u32 ibmtu; /* The MTU programmed for this unit */
620 	/*
621 	 * Current max size IB packet (in bytes) including IB headers, that
622 	 * we can send. Changes when ibmtu changes.
623 	 */
624 	u32 ibmaxlen;
625 	u32 current_egress_rate; /* units [10^6 bits/sec] */
626 	/* LID programmed for this instance */
627 	u16 lid;
628 	/* list of pkeys programmed; 0 if not set */
629 	u16 pkeys[MAX_PKEY_VALUES];
630 	u16 link_width_supported;
631 	u16 link_width_downgrade_supported;
632 	u16 link_speed_supported;
633 	u16 link_width_enabled;
634 	u16 link_width_downgrade_enabled;
635 	u16 link_speed_enabled;
636 	u16 link_width_active;
637 	u16 link_width_downgrade_tx_active;
638 	u16 link_width_downgrade_rx_active;
639 	u16 link_speed_active;
640 	u8 vls_supported;
641 	u8 vls_operational;
642 	u8 actual_vls_operational;
643 	/* LID mask control */
644 	u8 lmc;
645 	/* Rx Polarity inversion (compensate for ~tx on partner) */
646 	u8 rx_pol_inv;
647 
648 	u8 hw_pidx;     /* physical port index */
649 	u8 port;        /* IB port number and index into dd->pports - 1 */
650 	/* type of neighbor node */
651 	u8 neighbor_type;
652 	u8 neighbor_normal;
653 	u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
654 	u8 neighbor_port_number;
655 	u8 is_sm_config_started;
656 	u8 offline_disabled_reason;
657 	u8 is_active_optimize_enabled;
658 	u8 driver_link_ready;	/* driver ready for active link */
659 	u8 link_enabled;	/* link enabled? */
660 	u8 linkinit_reason;
661 	u8 local_tx_rate;	/* rate given to 8051 firmware */
662 	u8 last_pstate;		/* info only */
663 	u8 qsfp_retry_count;
664 
665 	/* placeholders for IB MAD packet settings */
666 	u8 overrun_threshold;
667 	u8 phy_error_threshold;
668 
669 	/* Used to override LED behavior for things like maintenance beaconing*/
670 	/*
671 	 * Alternates per phase of blink
672 	 * [0] holds LED off duration, [1] holds LED on duration
673 	 */
674 	unsigned long led_override_vals[2];
675 	u8 led_override_phase; /* LSB picks from vals[] */
676 	atomic_t led_override_timer_active;
677 	/* Used to flash LEDs in override mode */
678 	struct timer_list led_override_timer;
679 
680 	u32 sm_trap_qp;
681 	u32 sa_qp;
682 
683 	/*
684 	 * cca_timer_lock protects access to the per-SL cca_timer
685 	 * structures (specifically the ccti member).
686 	 */
687 	spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
688 	struct cca_timer cca_timer[OPA_MAX_SLS];
689 
690 	/* List of congestion control table entries */
691 	struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
692 
693 	/* congestion entries, each entry corresponding to a SL */
694 	struct opa_congestion_setting_entry_shadow
695 		congestion_entries[OPA_MAX_SLS];
696 
697 	/*
698 	 * cc_state_lock protects (write) access to the per-port
699 	 * struct cc_state.
700 	 */
701 	spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
702 
703 	struct cc_state __rcu *cc_state;
704 
705 	/* Total number of congestion control table entries */
706 	u16 total_cct_entry;
707 
708 	/* Bit map identifying service level */
709 	u32 cc_sl_control_map;
710 
711 	/* CA's max number of 64 entry units in the congestion control table */
712 	u8 cc_max_table_entries;
713 
714 	/*
715 	 * begin congestion log related entries
716 	 * cc_log_lock protects all congestion log related data
717 	 */
718 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
719 	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
720 	u16 threshold_event_counter;
721 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
722 	int cc_log_idx; /* index for logging events */
723 	int cc_mad_idx; /* index for reporting events */
724 	/* end congestion log related entries */
725 
726 	struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
727 
728 	/* port relative counter buffer */
729 	u64 *cntrs;
730 	/* port relative synthetic counter buffer */
731 	u64 *scntrs;
732 	/* port_xmit_discards are synthesized from different egress errors */
733 	u64 port_xmit_discards;
734 	u64 port_xmit_discards_vl[C_VL_COUNT];
735 	u64 port_xmit_constraint_errors;
736 	u64 port_rcv_constraint_errors;
737 	/* count of 'link_err' interrupts from DC */
738 	u64 link_downed;
739 	/* number of times link retrained successfully */
740 	u64 link_up;
741 	/* number of times a link unknown frame was reported */
742 	u64 unknown_frame_count;
743 	/* port_ltp_crc_mode is returned in 'portinfo' MADs */
744 	u16 port_ltp_crc_mode;
745 	/* port_crc_mode_enabled is the crc we support */
746 	u8 port_crc_mode_enabled;
747 	/* mgmt_allowed is also returned in 'portinfo' MADs */
748 	u8 mgmt_allowed;
749 	u8 part_enforce; /* partition enforcement flags */
750 	struct link_down_reason local_link_down_reason;
751 	struct link_down_reason neigh_link_down_reason;
752 	/* Value to be sent to link peer on LinkDown .*/
753 	u8 remote_link_down_reason;
754 	/* Error events that will cause a port bounce. */
755 	u32 port_error_action;
756 	struct work_struct linkstate_active_work;
757 	/* Does this port need to prescan for FECNs */
758 	bool cc_prescan;
759 };
760 
761 typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
762 
763 typedef void (*opcode_handler)(struct hfi1_packet *packet);
764 
765 /* return values for the RHF receive functions */
766 #define RHF_RCV_CONTINUE  0	/* keep going */
767 #define RHF_RCV_DONE	  1	/* stop, this packet processed */
768 #define RHF_RCV_REPROCESS 2	/* stop. retain this packet */
769 
770 struct rcv_array_data {
771 	u8 group_size;
772 	u16 ngroups;
773 	u16 nctxt_extra;
774 };
775 
776 struct per_vl_data {
777 	u16 mtu;
778 	struct send_context *sc;
779 };
780 
781 /* 16 to directly index */
782 #define PER_VL_SEND_CONTEXTS 16
783 
784 struct err_info_rcvport {
785 	u8 status_and_code;
786 	u64 packet_flit1;
787 	u64 packet_flit2;
788 };
789 
790 struct err_info_constraint {
791 	u8 status;
792 	u16 pkey;
793 	u32 slid;
794 };
795 
796 struct hfi1_temp {
797 	unsigned int curr;       /* current temperature */
798 	unsigned int lo_lim;     /* low temperature limit */
799 	unsigned int hi_lim;     /* high temperature limit */
800 	unsigned int crit_lim;   /* critical temperature limit */
801 	u8 triggers;      /* temperature triggers */
802 };
803 
804 struct hfi1_i2c_bus {
805 	struct hfi1_devdata *controlling_dd; /* current controlling device */
806 	struct i2c_adapter adapter;	/* bus details */
807 	struct i2c_algo_bit_data algo;	/* bus algorithm details */
808 	int num;			/* bus number, 0 or 1 */
809 };
810 
811 /* common data between shared ASIC HFIs */
812 struct hfi1_asic_data {
813 	struct hfi1_devdata *dds[2];	/* back pointers */
814 	struct mutex asic_resource_mutex;
815 	struct hfi1_i2c_bus *i2c_bus0;
816 	struct hfi1_i2c_bus *i2c_bus1;
817 };
818 
819 /* device data struct now contains only "general per-device" info.
820  * fields related to a physical IB port are in a hfi1_pportdata struct.
821  */
822 struct sdma_engine;
823 struct sdma_vl_map;
824 
825 #define BOARD_VERS_MAX 96 /* how long the version string can be */
826 #define SERIAL_MAX 16 /* length of the serial number */
827 
828 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
829 struct hfi1_devdata {
830 	struct hfi1_ibdev verbs_dev;     /* must be first */
831 	struct list_head list;
832 	/* pointers to related structs for this device */
833 	/* pci access data structure */
834 	struct pci_dev *pcidev;
835 	struct cdev user_cdev;
836 	struct cdev diag_cdev;
837 	struct cdev ui_cdev;
838 	struct device *user_device;
839 	struct device *diag_device;
840 	struct device *ui_device;
841 
842 	/* mem-mapped pointer to base of chip regs */
843 	u8 __iomem *kregbase;
844 	/* end of mem-mapped chip space excluding sendbuf and user regs */
845 	u8 __iomem *kregend;
846 	/* physical address of chip for io_remap, etc. */
847 	resource_size_t physaddr;
848 	/* receive context data */
849 	struct hfi1_ctxtdata **rcd;
850 	/* send context data */
851 	struct send_context_info *send_contexts;
852 	/* map hardware send contexts to software index */
853 	u8 *hw_to_sw;
854 	/* spinlock for allocating and releasing send context resources */
855 	spinlock_t sc_lock;
856 	/* Per VL data. Enough for all VLs but not all elements are set/used. */
857 	struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
858 	/* lock for pio_map */
859 	spinlock_t pio_map_lock;
860 	/* array of kernel send contexts */
861 	struct send_context **kernel_send_context;
862 	/* array of vl maps */
863 	struct pio_vl_map __rcu *pio_map;
864 	/* seqlock for sc2vl */
865 	seqlock_t sc2vl_lock;
866 	u64 sc2vl[4];
867 	/* Send Context initialization lock. */
868 	spinlock_t sc_init_lock;
869 
870 	/* fields common to all SDMA engines */
871 
872 	/* default flags to last descriptor */
873 	u64 default_desc1;
874 	volatile __le64                    *sdma_heads_dma; /* DMA'ed by chip */
875 	dma_addr_t                          sdma_heads_phys;
876 	void                               *sdma_pad_dma; /* DMA'ed by chip */
877 	dma_addr_t                          sdma_pad_phys;
878 	/* for deallocation */
879 	size_t                              sdma_heads_size;
880 	/* number from the chip */
881 	u32                                 chip_sdma_engines;
882 	/* num used */
883 	u32                                 num_sdma;
884 	/* lock for sdma_map */
885 	spinlock_t                          sde_map_lock;
886 	/* array of engines sized by num_sdma */
887 	struct sdma_engine                 *per_sdma;
888 	/* array of vl maps */
889 	struct sdma_vl_map __rcu           *sdma_map;
890 	/* SPC freeze waitqueue and variable */
891 	wait_queue_head_t		  sdma_unfreeze_wq;
892 	atomic_t			  sdma_unfreeze_count;
893 
894 	/* common data between shared ASIC HFIs in this OS */
895 	struct hfi1_asic_data *asic_data;
896 
897 	/* hfi1_pportdata, points to array of (physical) port-specific
898 	 * data structs, indexed by pidx (0..n-1)
899 	 */
900 	struct hfi1_pportdata *pport;
901 
902 	/* mem-mapped pointer to base of PIO buffers */
903 	void __iomem *piobase;
904 	/*
905 	 * write-combining mem-mapped pointer to base of RcvArray
906 	 * memory.
907 	 */
908 	void __iomem *rcvarray_wc;
909 	/*
910 	 * credit return base - a per-NUMA range of DMA address that
911 	 * the chip will use to update the per-context free counter
912 	 */
913 	struct credit_return_base *cr_base;
914 
915 	/* send context numbers and sizes for each type */
916 	struct sc_config_sizes sc_sizes[SC_MAX];
917 
918 	u32 lcb_access_count;		/* count of LCB users */
919 
920 	char *boardname; /* human readable board info */
921 
922 	/* device (not port) flags, basically device capabilities */
923 	u32 flags;
924 
925 	/* reset value */
926 	u64 z_int_counter;
927 	u64 z_rcv_limit;
928 	u64 z_send_schedule;
929 	/* percpu int_counter */
930 	u64 __percpu *int_counter;
931 	u64 __percpu *rcv_limit;
932 	u64 __percpu *send_schedule;
933 	/* number of receive contexts in use by the driver */
934 	u32 num_rcv_contexts;
935 	/* number of pio send contexts in use by the driver */
936 	u32 num_send_contexts;
937 	/*
938 	 * number of ctxts available for PSM open
939 	 */
940 	u32 freectxts;
941 	/* total number of available user/PSM contexts */
942 	u32 num_user_contexts;
943 	/* base receive interrupt timeout, in CSR units */
944 	u32 rcv_intr_timeout_csr;
945 
946 	u64 __iomem *egrtidbase;
947 	spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
948 	spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
949 	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
950 	spinlock_t uctxt_lock; /* rcd and user context changes */
951 	/* exclusive access to 8051 */
952 	spinlock_t dc8051_lock;
953 	/* exclusive access to 8051 memory */
954 	spinlock_t dc8051_memlock;
955 	int dc8051_timed_out;	/* remember if the 8051 timed out */
956 	/*
957 	 * A page that will hold event notification bitmaps for all
958 	 * contexts. This page will be mapped into all processes.
959 	 */
960 	unsigned long *events;
961 	/*
962 	 * per unit status, see also portdata statusp
963 	 * mapped read-only into user processes so they can get unit and
964 	 * IB link status cheaply
965 	 */
966 	struct hfi1_status *status;
967 	u32 freezelen; /* max length of freezemsg */
968 
969 	/* revision register shadow */
970 	u64 revision;
971 	/* Base GUID for device (network order) */
972 	u64 base_guid;
973 
974 	/* these are the "32 bit" regs */
975 
976 	/* value we put in kr_rcvhdrsize */
977 	u32 rcvhdrsize;
978 	/* number of receive contexts the chip supports */
979 	u32 chip_rcv_contexts;
980 	/* number of receive array entries */
981 	u32 chip_rcv_array_count;
982 	/* number of PIO send contexts the chip supports */
983 	u32 chip_send_contexts;
984 	/* number of bytes in the PIO memory buffer */
985 	u32 chip_pio_mem_size;
986 	/* number of bytes in the SDMA memory buffer */
987 	u32 chip_sdma_mem_size;
988 
989 	/* size of each rcvegrbuffer */
990 	u32 rcvegrbufsize;
991 	/* log2 of above */
992 	u16 rcvegrbufsize_shift;
993 	/* both sides of the PCIe link are gen3 capable */
994 	u8 link_gen3_capable;
995 	/* localbus width (1, 2,4,8,16,32) from config space  */
996 	u32 lbus_width;
997 	/* localbus speed in MHz */
998 	u32 lbus_speed;
999 	int unit; /* unit # of this chip */
1000 	int node; /* home node of this chip */
1001 
1002 	/* save these PCI fields to restore after a reset */
1003 	u32 pcibar0;
1004 	u32 pcibar1;
1005 	u32 pci_rom;
1006 	u16 pci_command;
1007 	u16 pcie_devctl;
1008 	u16 pcie_lnkctl;
1009 	u16 pcie_devctl2;
1010 	u32 pci_msix0;
1011 	u32 pci_lnkctl3;
1012 	u32 pci_tph2;
1013 
1014 	/*
1015 	 * ASCII serial number, from flash, large enough for original
1016 	 * all digit strings, and longer serial number format
1017 	 */
1018 	u8 serial[SERIAL_MAX];
1019 	/* human readable board version */
1020 	u8 boardversion[BOARD_VERS_MAX];
1021 	u8 lbus_info[32]; /* human readable localbus info */
1022 	/* chip major rev, from CceRevision */
1023 	u8 majrev;
1024 	/* chip minor rev, from CceRevision */
1025 	u8 minrev;
1026 	/* hardware ID */
1027 	u8 hfi1_id;
1028 	/* implementation code */
1029 	u8 icode;
1030 	/* default link down value (poll/sleep) */
1031 	u8 link_default;
1032 	/* vAU of this device */
1033 	u8 vau;
1034 	/* vCU of this device */
1035 	u8 vcu;
1036 	/* link credits of this device */
1037 	u16 link_credits;
1038 	/* initial vl15 credits to use */
1039 	u16 vl15_init;
1040 
1041 	/* Misc small ints */
1042 	/* Number of physical ports available */
1043 	u8 num_pports;
1044 	/* Lowest context number which can be used by user processes */
1045 	u8 first_user_ctxt;
1046 	u8 n_krcv_queues;
1047 	u8 qos_shift;
1048 	u8 qpn_mask;
1049 
1050 	u16 rhf_offset; /* offset of RHF within receive header entry */
1051 	u16 irev;	/* implementation revision */
1052 	u16 dc8051_ver; /* 8051 firmware version */
1053 
1054 	struct platform_config platform_config;
1055 	struct platform_config_cache pcfg_cache;
1056 
1057 	struct diag_client *diag_client;
1058 	spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1059 
1060 	u8 psxmitwait_supported;
1061 	/* cycle length of PS* counters in HW (in picoseconds) */
1062 	u16 psxmitwait_check_rate;
1063 	/* high volume overflow errors deferred to tasklet */
1064 	struct tasklet_struct error_tasklet;
1065 
1066 	/* MSI-X information */
1067 	struct hfi1_msix_entry *msix_entries;
1068 	u32 num_msix_entries;
1069 
1070 	/* INTx information */
1071 	u32 requested_intx_irq;		/* did we request one? */
1072 	char intx_name[MAX_NAME_SIZE];	/* INTx name */
1073 
1074 	/* general interrupt: mask of handled interrupts */
1075 	u64 gi_mask[CCE_NUM_INT_CSRS];
1076 
1077 	struct rcv_array_data rcv_entries;
1078 
1079 	/*
1080 	 * 64 bit synthetic counters
1081 	 */
1082 	struct timer_list synth_stats_timer;
1083 
1084 	/*
1085 	 * device counters
1086 	 */
1087 	char *cntrnames;
1088 	size_t cntrnameslen;
1089 	size_t ndevcntrs;
1090 	u64 *cntrs;
1091 	u64 *scntrs;
1092 
1093 	/*
1094 	 * remembered values for synthetic counters
1095 	 */
1096 	u64 last_tx;
1097 	u64 last_rx;
1098 
1099 	/*
1100 	 * per-port counters
1101 	 */
1102 	size_t nportcntrs;
1103 	char *portcntrnames;
1104 	size_t portcntrnameslen;
1105 
1106 	struct hfi1_snoop_data hfi1_snoop;
1107 
1108 	struct err_info_rcvport err_info_rcvport;
1109 	struct err_info_constraint err_info_rcv_constraint;
1110 	struct err_info_constraint err_info_xmit_constraint;
1111 	u8 err_info_uncorrectable;
1112 	u8 err_info_fmconfig;
1113 
1114 	atomic_t drop_packet;
1115 	u8 do_drop;
1116 
1117 	/*
1118 	 * Software counters for the status bits defined by the
1119 	 * associated error status registers
1120 	 */
1121 	u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1122 	u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1123 	u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1124 	u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1125 	u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1126 	u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1127 	u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1128 
1129 	/* Software counter that spans all contexts */
1130 	u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1131 	/* Software counter that spans all DMA engines */
1132 	u64 sw_send_dma_eng_err_status_cnt[
1133 		NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1134 	/* Software counter that aggregates all cce_err_status errors */
1135 	u64 sw_cce_err_status_aggregate;
1136 	/* Software counter that aggregates all bypass packet rcv errors */
1137 	u64 sw_rcv_bypass_packet_errors;
1138 	/* receive interrupt functions */
1139 	rhf_rcv_function_ptr *rhf_rcv_function_map;
1140 	rhf_rcv_function_ptr normal_rhf_rcv_functions[8];
1141 
1142 	/*
1143 	 * Handlers for outgoing data so that snoop/capture does not
1144 	 * have to have its hooks in the send path
1145 	 */
1146 	send_routine process_pio_send;
1147 	send_routine process_dma_send;
1148 	void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1149 				u64 pbc, const void *from, size_t count);
1150 
1151 	/* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1152 	u8 oui1;
1153 	u8 oui2;
1154 	u8 oui3;
1155 	/* Timer and counter used to detect RcvBufOvflCnt changes */
1156 	struct timer_list rcverr_timer;
1157 	u32 rcv_ovfl_cnt;
1158 
1159 	wait_queue_head_t event_queue;
1160 
1161 	/* Save the enabled LCB error bits */
1162 	u64 lcb_err_en;
1163 	u8 dc_shutdown;
1164 
1165 	/* receive context tail dummy address */
1166 	__le64 *rcvhdrtail_dummy_kvaddr;
1167 	dma_addr_t rcvhdrtail_dummy_physaddr;
1168 
1169 	bool eprom_available;	/* true if EPROM is available for this device */
1170 	bool aspm_supported;	/* Does HW support ASPM */
1171 	bool aspm_enabled;	/* ASPM state: enabled/disabled */
1172 	/* Serialize ASPM enable/disable between multiple verbs contexts */
1173 	spinlock_t aspm_lock;
1174 	/* Number of verbs contexts which have disabled ASPM */
1175 	atomic_t aspm_disabled_cnt;
1176 
1177 	struct hfi1_affinity *affinity;
1178 	struct kobject kobj;
1179 };
1180 
1181 /* 8051 firmware version helper */
1182 #define dc8051_ver(a, b) ((a) << 8 | (b))
1183 #define dc8051_ver_maj(a) ((a & 0xff00) >> 8)
1184 #define dc8051_ver_min(a)  (a & 0x00ff)
1185 
1186 /* f_put_tid types */
1187 #define PT_EXPECTED 0
1188 #define PT_EAGER    1
1189 #define PT_INVALID  2
1190 
1191 struct tid_rb_node;
1192 struct mmu_rb_node;
1193 struct mmu_rb_handler;
1194 
1195 /* Private data for file operations */
1196 struct hfi1_filedata {
1197 	struct hfi1_ctxtdata *uctxt;
1198 	unsigned subctxt;
1199 	struct hfi1_user_sdma_comp_q *cq;
1200 	struct hfi1_user_sdma_pkt_q *pq;
1201 	/* for cpu affinity; -1 if none */
1202 	int rec_cpu_num;
1203 	u32 tid_n_pinned;
1204 	struct mmu_rb_handler *handler;
1205 	struct tid_rb_node **entry_to_rb;
1206 	spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1207 	u32 tid_limit;
1208 	u32 tid_used;
1209 	u32 *invalid_tids;
1210 	u32 invalid_tid_idx;
1211 	/* protect invalid_tids array and invalid_tid_idx */
1212 	spinlock_t invalid_lock;
1213 	struct mm_struct *mm;
1214 };
1215 
1216 extern struct list_head hfi1_dev_list;
1217 extern spinlock_t hfi1_devs_lock;
1218 struct hfi1_devdata *hfi1_lookup(int unit);
1219 extern u32 hfi1_cpulist_count;
1220 extern unsigned long *hfi1_cpulist;
1221 
1222 extern unsigned int snoop_drop_send;
1223 extern unsigned int snoop_force_capture;
1224 int hfi1_init(struct hfi1_devdata *, int);
1225 int hfi1_count_units(int *npresentp, int *nupp);
1226 int hfi1_count_active_units(void);
1227 
1228 int hfi1_diag_add(struct hfi1_devdata *);
1229 void hfi1_diag_remove(struct hfi1_devdata *);
1230 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1231 
1232 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1233 
1234 int hfi1_create_rcvhdrq(struct hfi1_devdata *, struct hfi1_ctxtdata *);
1235 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *);
1236 int hfi1_create_ctxts(struct hfi1_devdata *dd);
1237 struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *, u32, int);
1238 void hfi1_init_pportdata(struct pci_dev *, struct hfi1_pportdata *,
1239 			 struct hfi1_devdata *, u8, u8);
1240 void hfi1_free_ctxtdata(struct hfi1_devdata *, struct hfi1_ctxtdata *);
1241 
1242 int handle_receive_interrupt(struct hfi1_ctxtdata *, int);
1243 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *, int);
1244 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *, int);
1245 void set_all_slowpath(struct hfi1_devdata *dd);
1246 
1247 extern const struct pci_device_id hfi1_pci_tbl[];
1248 
1249 /* receive packet handler dispositions */
1250 #define RCV_PKT_OK      0x0 /* keep going */
1251 #define RCV_PKT_LIMIT   0x1 /* stop, hit limit, start thread */
1252 #define RCV_PKT_DONE    0x2 /* stop, no more packets detected */
1253 
1254 /* calculate the current RHF address */
1255 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1256 {
1257 	return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->dd->rhf_offset;
1258 }
1259 
1260 int hfi1_reset_device(int);
1261 
1262 /* return the driver's idea of the logical OPA port state */
1263 static inline u32 driver_lstate(struct hfi1_pportdata *ppd)
1264 {
1265 	return ppd->lstate; /* use the cached value */
1266 }
1267 
1268 void receive_interrupt_work(struct work_struct *work);
1269 
1270 /* extract service channel from header and rhf */
1271 static inline int hdr2sc(struct hfi1_message_header *hdr, u64 rhf)
1272 {
1273 	return ((be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf) |
1274 	       ((!!(rhf_dc_info(rhf))) << 4);
1275 }
1276 
1277 #define HFI1_JKEY_WIDTH       16
1278 #define HFI1_JKEY_MASK        (BIT(16) - 1)
1279 #define HFI1_ADMIN_JKEY_RANGE 32
1280 
1281 /*
1282  * J_KEYs are split and allocated in the following groups:
1283  *   0 - 31    - users with administrator privileges
1284  *  32 - 63    - kernel protocols using KDETH packets
1285  *  64 - 65535 - all other users using KDETH packets
1286  */
1287 static inline u16 generate_jkey(kuid_t uid)
1288 {
1289 	u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1290 
1291 	if (capable(CAP_SYS_ADMIN))
1292 		jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1293 	else if (jkey < 64)
1294 		jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1295 
1296 	return jkey;
1297 }
1298 
1299 /*
1300  * active_egress_rate
1301  *
1302  * returns the active egress rate in units of [10^6 bits/sec]
1303  */
1304 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1305 {
1306 	u16 link_speed = ppd->link_speed_active;
1307 	u16 link_width = ppd->link_width_active;
1308 	u32 egress_rate;
1309 
1310 	if (link_speed == OPA_LINK_SPEED_25G)
1311 		egress_rate = 25000;
1312 	else /* assume OPA_LINK_SPEED_12_5G */
1313 		egress_rate = 12500;
1314 
1315 	switch (link_width) {
1316 	case OPA_LINK_WIDTH_4X:
1317 		egress_rate *= 4;
1318 		break;
1319 	case OPA_LINK_WIDTH_3X:
1320 		egress_rate *= 3;
1321 		break;
1322 	case OPA_LINK_WIDTH_2X:
1323 		egress_rate *= 2;
1324 		break;
1325 	default:
1326 		/* assume IB_WIDTH_1X */
1327 		break;
1328 	}
1329 
1330 	return egress_rate;
1331 }
1332 
1333 /*
1334  * egress_cycles
1335  *
1336  * Returns the number of 'fabric clock cycles' to egress a packet
1337  * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1338  * rate is (approximately) 805 MHz, the units of the returned value
1339  * are (1/805 MHz).
1340  */
1341 static inline u32 egress_cycles(u32 len, u32 rate)
1342 {
1343 	u32 cycles;
1344 
1345 	/*
1346 	 * cycles is:
1347 	 *
1348 	 *          (length) [bits] / (rate) [bits/sec]
1349 	 *  ---------------------------------------------------
1350 	 *  fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1351 	 */
1352 
1353 	cycles = len * 8; /* bits */
1354 	cycles *= 805;
1355 	cycles /= rate;
1356 
1357 	return cycles;
1358 }
1359 
1360 void set_link_ipg(struct hfi1_pportdata *ppd);
1361 void process_becn(struct hfi1_pportdata *ppd, u8 sl,  u16 rlid, u32 lqpn,
1362 		  u32 rqpn, u8 svc_type);
1363 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1364 		u32 pkey, u32 slid, u32 dlid, u8 sc5,
1365 		const struct ib_grh *old_grh);
1366 #define PKEY_CHECK_INVALID -1
1367 int egress_pkey_check(struct hfi1_pportdata *ppd, __be16 *lrh, __be32 *bth,
1368 		      u8 sc5, int8_t s_pkey_index);
1369 
1370 #define PACKET_EGRESS_TIMEOUT 350
1371 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1372 {
1373 	/* Pause at least 1us, to ensure chip returns all credits */
1374 	u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1375 
1376 	udelay(usec ? usec : 1);
1377 }
1378 
1379 /**
1380  * sc_to_vlt() reverse lookup sc to vl
1381  * @dd - devdata
1382  * @sc5 - 5 bit sc
1383  */
1384 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1385 {
1386 	unsigned seq;
1387 	u8 rval;
1388 
1389 	if (sc5 >= OPA_MAX_SCS)
1390 		return (u8)(0xff);
1391 
1392 	do {
1393 		seq = read_seqbegin(&dd->sc2vl_lock);
1394 		rval = *(((u8 *)dd->sc2vl) + sc5);
1395 	} while (read_seqretry(&dd->sc2vl_lock, seq));
1396 
1397 	return rval;
1398 }
1399 
1400 #define PKEY_MEMBER_MASK 0x8000
1401 #define PKEY_LOW_15_MASK 0x7fff
1402 
1403 /*
1404  * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1405  * being an entry from the ingress partition key table), return 0
1406  * otherwise. Use the matching criteria for ingress partition keys
1407  * specified in the OPAv1 spec., section 9.10.14.
1408  */
1409 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1410 {
1411 	u16 mkey = pkey & PKEY_LOW_15_MASK;
1412 	u16 ment = ent & PKEY_LOW_15_MASK;
1413 
1414 	if (mkey == ment) {
1415 		/*
1416 		 * If pkey[15] is clear (limited partition member),
1417 		 * is bit 15 in the corresponding table element
1418 		 * clear (limited member)?
1419 		 */
1420 		if (!(pkey & PKEY_MEMBER_MASK))
1421 			return !!(ent & PKEY_MEMBER_MASK);
1422 		return 1;
1423 	}
1424 	return 0;
1425 }
1426 
1427 /*
1428  * ingress_pkey_table_search - search the entire pkey table for
1429  * an entry which matches 'pkey'. return 0 if a match is found,
1430  * and 1 otherwise.
1431  */
1432 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1433 {
1434 	int i;
1435 
1436 	for (i = 0; i < MAX_PKEY_VALUES; i++) {
1437 		if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1438 			return 0;
1439 	}
1440 	return 1;
1441 }
1442 
1443 /*
1444  * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1445  * i.e., increment port_rcv_constraint_errors for the port, and record
1446  * the 'error info' for this failure.
1447  */
1448 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1449 				    u16 slid)
1450 {
1451 	struct hfi1_devdata *dd = ppd->dd;
1452 
1453 	incr_cntr64(&ppd->port_rcv_constraint_errors);
1454 	if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1455 		dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1456 		dd->err_info_rcv_constraint.slid = slid;
1457 		dd->err_info_rcv_constraint.pkey = pkey;
1458 	}
1459 }
1460 
1461 /*
1462  * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1463  * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1464  * is a hint as to the best place in the partition key table to begin
1465  * searching. This function should not be called on the data path because
1466  * of performance reasons. On datapath pkey check is expected to be done
1467  * by HW and rcv_pkey_check function should be called instead.
1468  */
1469 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1470 				     u8 sc5, u8 idx, u16 slid)
1471 {
1472 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1473 		return 0;
1474 
1475 	/* If SC15, pkey[0:14] must be 0x7fff */
1476 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1477 		goto bad;
1478 
1479 	/* Is the pkey = 0x0, or 0x8000? */
1480 	if ((pkey & PKEY_LOW_15_MASK) == 0)
1481 		goto bad;
1482 
1483 	/* The most likely matching pkey has index 'idx' */
1484 	if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1485 		return 0;
1486 
1487 	/* no match - try the whole table */
1488 	if (!ingress_pkey_table_search(ppd, pkey))
1489 		return 0;
1490 
1491 bad:
1492 	ingress_pkey_table_fail(ppd, pkey, slid);
1493 	return 1;
1494 }
1495 
1496 /*
1497  * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1498  * otherwise. It only ensures pkey is vlid for QP0. This function
1499  * should be called on the data path instead of ingress_pkey_check
1500  * as on data path, pkey check is done by HW (except for QP0).
1501  */
1502 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1503 				 u8 sc5, u16 slid)
1504 {
1505 	if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1506 		return 0;
1507 
1508 	/* If SC15, pkey[0:14] must be 0x7fff */
1509 	if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1510 		goto bad;
1511 
1512 	return 0;
1513 bad:
1514 	ingress_pkey_table_fail(ppd, pkey, slid);
1515 	return 1;
1516 }
1517 
1518 /* MTU handling */
1519 
1520 /* MTU enumeration, 256-4k match IB */
1521 #define OPA_MTU_0     0
1522 #define OPA_MTU_256   1
1523 #define OPA_MTU_512   2
1524 #define OPA_MTU_1024  3
1525 #define OPA_MTU_2048  4
1526 #define OPA_MTU_4096  5
1527 
1528 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1529 int mtu_to_enum(u32 mtu, int default_if_bad);
1530 u16 enum_to_mtu(int);
1531 static inline int valid_ib_mtu(unsigned int mtu)
1532 {
1533 	return mtu == 256 || mtu == 512 ||
1534 		mtu == 1024 || mtu == 2048 ||
1535 		mtu == 4096;
1536 }
1537 
1538 static inline int valid_opa_max_mtu(unsigned int mtu)
1539 {
1540 	return mtu >= 2048 &&
1541 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1542 }
1543 
1544 int set_mtu(struct hfi1_pportdata *);
1545 
1546 int hfi1_set_lid(struct hfi1_pportdata *, u32, u8);
1547 void hfi1_disable_after_error(struct hfi1_devdata *);
1548 int hfi1_set_uevent_bits(struct hfi1_pportdata *, const int);
1549 int hfi1_rcvbuf_validate(u32, u8, u16 *);
1550 
1551 int fm_get_table(struct hfi1_pportdata *, int, void *);
1552 int fm_set_table(struct hfi1_pportdata *, int, void *);
1553 
1554 void set_up_vl15(struct hfi1_devdata *dd, u8 vau, u16 vl15buf);
1555 void reset_link_credits(struct hfi1_devdata *dd);
1556 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1557 
1558 int snoop_recv_handler(struct hfi1_packet *packet);
1559 int snoop_send_dma_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
1560 			   u64 pbc);
1561 int snoop_send_pio_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
1562 			   u64 pbc);
1563 void snoop_inline_pio_send(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1564 			   u64 pbc, const void *from, size_t count);
1565 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1566 
1567 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1568 {
1569 	return ppd->dd;
1570 }
1571 
1572 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1573 {
1574 	return container_of(dev, struct hfi1_devdata, verbs_dev);
1575 }
1576 
1577 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1578 {
1579 	return dd_from_dev(to_idev(ibdev));
1580 }
1581 
1582 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1583 {
1584 	return container_of(ibp, struct hfi1_pportdata, ibport_data);
1585 }
1586 
1587 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1588 {
1589 	return container_of(rdi, struct hfi1_ibdev, rdi);
1590 }
1591 
1592 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u8 port)
1593 {
1594 	struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1595 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1596 
1597 	WARN_ON(pidx >= dd->num_pports);
1598 	return &dd->pport[pidx].ibport_data;
1599 }
1600 
1601 void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
1602 			       bool do_cnp);
1603 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt,
1604 			       bool do_cnp)
1605 {
1606 	struct hfi1_other_headers *ohdr = pkt->ohdr;
1607 	u32 bth1;
1608 
1609 	bth1 = be32_to_cpu(ohdr->bth[1]);
1610 	if (unlikely(bth1 & (HFI1_BECN_SMASK | HFI1_FECN_SMASK))) {
1611 		hfi1_process_ecn_slowpath(qp, pkt, do_cnp);
1612 		return bth1 & HFI1_FECN_SMASK;
1613 	}
1614 	return false;
1615 }
1616 
1617 /*
1618  * Return the indexed PKEY from the port PKEY table.
1619  */
1620 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
1621 {
1622 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1623 	u16 ret;
1624 
1625 	if (index >= ARRAY_SIZE(ppd->pkeys))
1626 		ret = 0;
1627 	else
1628 		ret = ppd->pkeys[index];
1629 
1630 	return ret;
1631 }
1632 
1633 /*
1634  * Called by readers of cc_state only, must call under rcu_read_lock().
1635  */
1636 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
1637 {
1638 	return rcu_dereference(ppd->cc_state);
1639 }
1640 
1641 /*
1642  * Called by writers of cc_state only,  must call under cc_state_lock.
1643  */
1644 static inline
1645 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
1646 {
1647 	return rcu_dereference_protected(ppd->cc_state,
1648 					 lockdep_is_held(&ppd->cc_state_lock));
1649 }
1650 
1651 /*
1652  * values for dd->flags (_device_ related flags)
1653  */
1654 #define HFI1_INITTED           0x1    /* chip and driver up and initted */
1655 #define HFI1_PRESENT           0x2    /* chip accesses can be done */
1656 #define HFI1_FROZEN            0x4    /* chip in SPC freeze */
1657 #define HFI1_HAS_SDMA_TIMEOUT  0x8
1658 #define HFI1_HAS_SEND_DMA      0x10   /* Supports Send DMA */
1659 #define HFI1_FORCED_FREEZE     0x80   /* driver forced freeze mode */
1660 
1661 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
1662 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
1663 
1664 /* ctxt_flag bit offsets */
1665 		/* context has been setup */
1666 #define HFI1_CTXT_SETUP_DONE 1
1667 		/* waiting for a packet to arrive */
1668 #define HFI1_CTXT_WAITING_RCV   2
1669 		/* master has not finished initializing */
1670 #define HFI1_CTXT_MASTER_UNINIT 4
1671 		/* waiting for an urgent packet to arrive */
1672 #define HFI1_CTXT_WAITING_URG 5
1673 
1674 /* free up any allocated data at closes */
1675 struct hfi1_devdata *hfi1_init_dd(struct pci_dev *,
1676 				  const struct pci_device_id *);
1677 void hfi1_free_devdata(struct hfi1_devdata *);
1678 struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra);
1679 
1680 /* LED beaconing functions */
1681 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
1682 			     unsigned int timeoff);
1683 void shutdown_led_override(struct hfi1_pportdata *ppd);
1684 
1685 #define HFI1_CREDIT_RETURN_RATE (100)
1686 
1687 /*
1688  * The number of words for the KDETH protocol field.  If this is
1689  * larger then the actual field used, then part of the payload
1690  * will be in the header.
1691  *
1692  * Optimally, we want this sized so that a typical case will
1693  * use full cache lines.  The typical local KDETH header would
1694  * be:
1695  *
1696  *	Bytes	Field
1697  *	  8	LRH
1698  *	 12	BHT
1699  *	 ??	KDETH
1700  *	  8	RHF
1701  *	---
1702  *	 28 + KDETH
1703  *
1704  * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
1705  */
1706 #define DEFAULT_RCVHDRSIZE 9
1707 
1708 /*
1709  * Maximal header byte count:
1710  *
1711  *	Bytes	Field
1712  *	  8	LRH
1713  *	 40	GRH (optional)
1714  *	 12	BTH
1715  *	 ??	KDETH
1716  *	  8	RHF
1717  *	---
1718  *	 68 + KDETH
1719  *
1720  * We also want to maintain a cache line alignment to assist DMA'ing
1721  * of the header bytes.  Round up to a good size.
1722  */
1723 #define DEFAULT_RCVHDR_ENTSIZE 32
1724 
1725 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
1726 			u32 nlocked, u32 npages);
1727 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
1728 			    size_t npages, bool writable, struct page **pages);
1729 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
1730 			     size_t npages, bool dirty);
1731 
1732 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1733 {
1734 	*((u64 *)rcd->rcvhdrtail_kvaddr) = 0ULL;
1735 }
1736 
1737 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1738 {
1739 	/*
1740 	 * volatile because it's a DMA target from the chip, routine is
1741 	 * inlined, and don't want register caching or reordering.
1742 	 */
1743 	return (u32)le64_to_cpu(*rcd->rcvhdrtail_kvaddr);
1744 }
1745 
1746 /*
1747  * sysfs interface.
1748  */
1749 
1750 extern const char ib_hfi1_version[];
1751 
1752 int hfi1_device_create(struct hfi1_devdata *);
1753 void hfi1_device_remove(struct hfi1_devdata *);
1754 
1755 int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
1756 			   struct kobject *kobj);
1757 int hfi1_verbs_register_sysfs(struct hfi1_devdata *);
1758 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *);
1759 /* Hook for sysfs read of QSFP */
1760 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
1761 
1762 int hfi1_pcie_init(struct pci_dev *, const struct pci_device_id *);
1763 void hfi1_pcie_cleanup(struct pci_dev *);
1764 int hfi1_pcie_ddinit(struct hfi1_devdata *, struct pci_dev *,
1765 		     const struct pci_device_id *);
1766 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
1767 void hfi1_pcie_flr(struct hfi1_devdata *);
1768 int pcie_speeds(struct hfi1_devdata *);
1769 void request_msix(struct hfi1_devdata *, u32 *, struct hfi1_msix_entry *);
1770 void hfi1_enable_intx(struct pci_dev *);
1771 void restore_pci_variables(struct hfi1_devdata *dd);
1772 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
1773 int parse_platform_config(struct hfi1_devdata *dd);
1774 int get_platform_config_field(struct hfi1_devdata *dd,
1775 			      enum platform_config_table_type_encoding
1776 			      table_type, int table_index, int field_index,
1777 			      u32 *data, u32 len);
1778 
1779 const char *get_unit_name(int unit);
1780 const char *get_card_name(struct rvt_dev_info *rdi);
1781 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
1782 
1783 /*
1784  * Flush write combining store buffers (if present) and perform a write
1785  * barrier.
1786  */
1787 static inline void flush_wc(void)
1788 {
1789 	asm volatile("sfence" : : : "memory");
1790 }
1791 
1792 void handle_eflags(struct hfi1_packet *packet);
1793 int process_receive_ib(struct hfi1_packet *packet);
1794 int process_receive_bypass(struct hfi1_packet *packet);
1795 int process_receive_error(struct hfi1_packet *packet);
1796 int kdeth_process_expected(struct hfi1_packet *packet);
1797 int kdeth_process_eager(struct hfi1_packet *packet);
1798 int process_receive_invalid(struct hfi1_packet *packet);
1799 
1800 extern rhf_rcv_function_ptr snoop_rhf_rcv_functions[8];
1801 
1802 void update_sge(struct rvt_sge_state *ss, u32 length);
1803 
1804 /* global module parameter variables */
1805 extern unsigned int hfi1_max_mtu;
1806 extern unsigned int hfi1_cu;
1807 extern unsigned int user_credit_return_threshold;
1808 extern int num_user_contexts;
1809 extern unsigned long n_krcvqs;
1810 extern uint krcvqs[];
1811 extern int krcvqsset;
1812 extern uint kdeth_qp;
1813 extern uint loopback;
1814 extern uint quick_linkup;
1815 extern uint rcv_intr_timeout;
1816 extern uint rcv_intr_count;
1817 extern uint rcv_intr_dynamic;
1818 extern ushort link_crc_mask;
1819 
1820 extern struct mutex hfi1_mutex;
1821 
1822 /* Number of seconds before our card status check...  */
1823 #define STATUS_TIMEOUT 60
1824 
1825 #define DRIVER_NAME		"hfi1"
1826 #define HFI1_USER_MINOR_BASE     0
1827 #define HFI1_TRACE_MINOR         127
1828 #define HFI1_DIAGPKT_MINOR       128
1829 #define HFI1_DIAG_MINOR_BASE     129
1830 #define HFI1_SNOOP_CAPTURE_BASE  200
1831 #define HFI1_NMINORS             255
1832 
1833 #define PCI_VENDOR_ID_INTEL 0x8086
1834 #define PCI_DEVICE_ID_INTEL0 0x24f0
1835 #define PCI_DEVICE_ID_INTEL1 0x24f1
1836 
1837 #define HFI1_PKT_USER_SC_INTEGRITY					    \
1838 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK	    \
1839 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK		\
1840 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK		    \
1841 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
1842 
1843 #define HFI1_PKT_KERNEL_SC_INTEGRITY					    \
1844 	(SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
1845 
1846 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
1847 						  u16 ctxt_type)
1848 {
1849 	u64 base_sc_integrity =
1850 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1851 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
1852 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1853 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1854 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1855 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
1856 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1857 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1858 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1859 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
1860 	| SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1861 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1862 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
1863 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
1864 	| SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
1865 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
1866 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1867 
1868 	if (ctxt_type == SC_USER)
1869 		base_sc_integrity |= HFI1_PKT_USER_SC_INTEGRITY;
1870 	else
1871 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
1872 
1873 	if (is_ax(dd))
1874 		/* turn off send-side job key checks - A0 */
1875 		return base_sc_integrity &
1876 		       ~SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1877 	return base_sc_integrity;
1878 }
1879 
1880 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
1881 {
1882 	u64 base_sdma_integrity =
1883 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1884 	| SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
1885 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1886 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1887 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1888 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1889 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1890 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1891 	| SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
1892 	| SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1893 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1894 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
1895 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
1896 	| SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
1897 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
1898 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1899 
1900 	if (is_ax(dd))
1901 		/* turn off send-side job key checks - A0 */
1902 		return base_sdma_integrity &
1903 		       ~SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1904 	return base_sdma_integrity;
1905 }
1906 
1907 /*
1908  * hfi1_early_err is used (only!) to print early errors before devdata is
1909  * allocated, or when dd->pcidev may not be valid, and at the tail end of
1910  * cleanup when devdata may have been freed, etc.  hfi1_dev_porterr is
1911  * the same as dd_dev_err, but is used when the message really needs
1912  * the IB port# to be definitive as to what's happening..
1913  */
1914 #define hfi1_early_err(dev, fmt, ...) \
1915 	dev_err(dev, fmt, ##__VA_ARGS__)
1916 
1917 #define hfi1_early_info(dev, fmt, ...) \
1918 	dev_info(dev, fmt, ##__VA_ARGS__)
1919 
1920 #define dd_dev_emerg(dd, fmt, ...) \
1921 	dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
1922 		  get_unit_name((dd)->unit), ##__VA_ARGS__)
1923 #define dd_dev_err(dd, fmt, ...) \
1924 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1925 			get_unit_name((dd)->unit), ##__VA_ARGS__)
1926 #define dd_dev_warn(dd, fmt, ...) \
1927 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
1928 			get_unit_name((dd)->unit), ##__VA_ARGS__)
1929 
1930 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
1931 	dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
1932 			get_unit_name((dd)->unit), ##__VA_ARGS__)
1933 
1934 #define dd_dev_info(dd, fmt, ...) \
1935 	dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
1936 			get_unit_name((dd)->unit), ##__VA_ARGS__)
1937 
1938 #define dd_dev_dbg(dd, fmt, ...) \
1939 	dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
1940 		get_unit_name((dd)->unit), ##__VA_ARGS__)
1941 
1942 #define hfi1_dev_porterr(dd, port, fmt, ...) \
1943 	dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
1944 			get_unit_name((dd)->unit), (port), ##__VA_ARGS__)
1945 
1946 /*
1947  * this is used for formatting hw error messages...
1948  */
1949 struct hfi1_hwerror_msgs {
1950 	u64 mask;
1951 	const char *msg;
1952 	size_t sz;
1953 };
1954 
1955 /* in intr.c... */
1956 void hfi1_format_hwerrors(u64 hwerrs,
1957 			  const struct hfi1_hwerror_msgs *hwerrmsgs,
1958 			  size_t nhwerrmsgs, char *msg, size_t lmsg);
1959 
1960 #define USER_OPCODE_CHECK_VAL 0xC0
1961 #define USER_OPCODE_CHECK_MASK 0xC0
1962 #define OPCODE_CHECK_VAL_DISABLED 0x0
1963 #define OPCODE_CHECK_MASK_DISABLED 0x0
1964 
1965 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
1966 {
1967 	struct hfi1_pportdata *ppd;
1968 	int i;
1969 
1970 	dd->z_int_counter = get_all_cpu_total(dd->int_counter);
1971 	dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
1972 	dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
1973 
1974 	ppd = (struct hfi1_pportdata *)(dd + 1);
1975 	for (i = 0; i < dd->num_pports; i++, ppd++) {
1976 		ppd->ibport_data.rvp.z_rc_acks =
1977 			get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
1978 		ppd->ibport_data.rvp.z_rc_qacks =
1979 			get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
1980 	}
1981 }
1982 
1983 /* Control LED state */
1984 static inline void setextled(struct hfi1_devdata *dd, u32 on)
1985 {
1986 	if (on)
1987 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
1988 	else
1989 		write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
1990 }
1991 
1992 /* return the i2c resource given the target */
1993 static inline u32 i2c_target(u32 target)
1994 {
1995 	return target ? CR_I2C2 : CR_I2C1;
1996 }
1997 
1998 /* return the i2c chain chip resource that this HFI uses for QSFP */
1999 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2000 {
2001 	return i2c_target(dd->hfi1_id);
2002 }
2003 
2004 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2005 
2006 #define DD_DEV_ENTRY(dd)       __string(dev, dev_name(&(dd)->pcidev->dev))
2007 #define DD_DEV_ASSIGN(dd)      __assign_str(dev, dev_name(&(dd)->pcidev->dev))
2008 
2009 #define packettype_name(etype) { RHF_RCV_TYPE_##etype, #etype }
2010 #define show_packettype(etype)                  \
2011 __print_symbolic(etype,                         \
2012 	packettype_name(EXPECTED),              \
2013 	packettype_name(EAGER),                 \
2014 	packettype_name(IB),                    \
2015 	packettype_name(ERROR),                 \
2016 	packettype_name(BYPASS))
2017 
2018 #define ib_opcode_name(opcode) { IB_OPCODE_##opcode, #opcode  }
2019 #define show_ib_opcode(opcode)                             \
2020 __print_symbolic(opcode,                                   \
2021 	ib_opcode_name(RC_SEND_FIRST),                     \
2022 	ib_opcode_name(RC_SEND_MIDDLE),                    \
2023 	ib_opcode_name(RC_SEND_LAST),                      \
2024 	ib_opcode_name(RC_SEND_LAST_WITH_IMMEDIATE),       \
2025 	ib_opcode_name(RC_SEND_ONLY),                      \
2026 	ib_opcode_name(RC_SEND_ONLY_WITH_IMMEDIATE),       \
2027 	ib_opcode_name(RC_RDMA_WRITE_FIRST),               \
2028 	ib_opcode_name(RC_RDMA_WRITE_MIDDLE),              \
2029 	ib_opcode_name(RC_RDMA_WRITE_LAST),                \
2030 	ib_opcode_name(RC_RDMA_WRITE_LAST_WITH_IMMEDIATE), \
2031 	ib_opcode_name(RC_RDMA_WRITE_ONLY),                \
2032 	ib_opcode_name(RC_RDMA_WRITE_ONLY_WITH_IMMEDIATE), \
2033 	ib_opcode_name(RC_RDMA_READ_REQUEST),              \
2034 	ib_opcode_name(RC_RDMA_READ_RESPONSE_FIRST),       \
2035 	ib_opcode_name(RC_RDMA_READ_RESPONSE_MIDDLE),      \
2036 	ib_opcode_name(RC_RDMA_READ_RESPONSE_LAST),        \
2037 	ib_opcode_name(RC_RDMA_READ_RESPONSE_ONLY),        \
2038 	ib_opcode_name(RC_ACKNOWLEDGE),                    \
2039 	ib_opcode_name(RC_ATOMIC_ACKNOWLEDGE),             \
2040 	ib_opcode_name(RC_COMPARE_SWAP),                   \
2041 	ib_opcode_name(RC_FETCH_ADD),                      \
2042 	ib_opcode_name(UC_SEND_FIRST),                     \
2043 	ib_opcode_name(UC_SEND_MIDDLE),                    \
2044 	ib_opcode_name(UC_SEND_LAST),                      \
2045 	ib_opcode_name(UC_SEND_LAST_WITH_IMMEDIATE),       \
2046 	ib_opcode_name(UC_SEND_ONLY),                      \
2047 	ib_opcode_name(UC_SEND_ONLY_WITH_IMMEDIATE),       \
2048 	ib_opcode_name(UC_RDMA_WRITE_FIRST),               \
2049 	ib_opcode_name(UC_RDMA_WRITE_MIDDLE),              \
2050 	ib_opcode_name(UC_RDMA_WRITE_LAST),                \
2051 	ib_opcode_name(UC_RDMA_WRITE_LAST_WITH_IMMEDIATE), \
2052 	ib_opcode_name(UC_RDMA_WRITE_ONLY),                \
2053 	ib_opcode_name(UC_RDMA_WRITE_ONLY_WITH_IMMEDIATE), \
2054 	ib_opcode_name(UD_SEND_ONLY),                      \
2055 	ib_opcode_name(UD_SEND_ONLY_WITH_IMMEDIATE),       \
2056 	ib_opcode_name(CNP))
2057 #endif                          /* _HFI1_KERNEL_H */
2058