staging/rdma/hfi1: HFI reports wrong offline disabled reason when cable removed
[deliverable/linux.git] / drivers / staging / rdma / hfi1 / hfi.h
CommitLineData
77241056
MM
1#ifndef _HFI1_KERNEL_H
2#define _HFI1_KERNEL_H
3/*
4 *
5 * This file is provided under a dual BSD/GPLv2 license. When using or
6 * redistributing this file, you may do so under either license.
7 *
8 * GPL LICENSE SUMMARY
9 *
10 * Copyright(c) 2015 Intel Corporation.
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of version 2 of the GNU General Public License as
14 * published by the Free Software Foundation.
15 *
16 * This program is distributed in the hope that it will be useful, but
17 * WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * BSD LICENSE
22 *
23 * Copyright(c) 2015 Intel Corporation.
24 *
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
27 * are met:
28 *
29 * - Redistributions of source code must retain the above copyright
30 * notice, this list of conditions and the following disclaimer.
31 * - Redistributions in binary form must reproduce the above copyright
32 * notice, this list of conditions and the following disclaimer in
33 * the documentation and/or other materials provided with the
34 * distribution.
35 * - Neither the name of Intel Corporation nor the names of its
36 * contributors may be used to endorse or promote products derived
37 * from this software without specific prior written permission.
38 *
39 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
40 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
41 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
42 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
43 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
44 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
45 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
46 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
47 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
48 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
49 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
50 *
51 */
52
53#include <linux/interrupt.h>
54#include <linux/pci.h>
55#include <linux/dma-mapping.h>
56#include <linux/mutex.h>
57#include <linux/list.h>
58#include <linux/scatterlist.h>
59#include <linux/slab.h>
60#include <linux/io.h>
61#include <linux/fs.h>
62#include <linux/completion.h>
63#include <linux/kref.h>
64#include <linux/sched.h>
65#include <linux/cdev.h>
66#include <linux/delay.h>
67#include <linux/kthread.h>
ec3f2c12 68#include <rdma/rdma_vt.h>
77241056
MM
69
70#include "chip_registers.h"
71#include "common.h"
72#include "verbs.h"
73#include "pio.h"
74#include "chip.h"
75#include "mad.h"
76#include "qsfp.h"
77#include "platform_config.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
92extern 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)
a9c05e35
BM
102/* Offline Disabled Reason is 4-bits */
103#define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
77241056 104
82c2611d
NV
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
2c5b521a
JR
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
77241056
MM
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 */
133struct 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
146extern struct hfi1_ib_stats hfi1_stats;
147extern 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
162struct hfi1_opcode_stats_perctx;
163#endif
164
77241056
MM
165struct 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
a86cd357
MH
183struct exp_tid_set {
184 struct list_head list;
185 u32 count;
186};
187
77241056
MM
188struct 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;
a86cd357
MH
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
77241056 249 /* lock protecting all Expected TID data */
463e6ebc 250 struct mutex exp_lock;
77241056
MM
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 */
82c2611d 258 u32 flags;
77241056
MM
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 /* pid of process using this ctxt */
274 pid_t pid;
275 pid_t subpid[HFI1_MAX_SHARED_CTXTS];
276 /* same size as task_struct .comm[], command that opened context */
c3af8a28 277 char comm[TASK_COMM_LEN];
77241056
MM
278 /* so file ops can get at unit */
279 struct hfi1_devdata *dd;
280 /* so functions that need physical port can get it easily */
281 struct hfi1_pportdata *ppd;
282 /* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
283 void *subctxt_uregbase;
284 /* An array of pages for the eager receive buffers * N */
285 void *subctxt_rcvegrbuf;
286 /* An array of pages for the eager header queue entries * N */
287 void *subctxt_rcvhdr_base;
288 /* The version of the library which opened this ctxt */
289 u32 userversion;
290 /* Bitmask of active slaves */
291 u32 active_slaves;
292 /* Type of packets or conditions we want to poll for */
293 u16 poll_type;
294 /* receive packet sequence counter */
295 u8 seq_cnt;
296 u8 redirect_seq_cnt;
297 /* ctxt rcvhdrq head offset */
298 u32 head;
299 u32 pkt_count;
300 /* QPs waiting for context processing */
301 struct list_head qp_wait_list;
302 /* interrupt handling */
303 u64 imask; /* clear interrupt mask */
304 int ireg; /* clear interrupt register */
305 unsigned numa_id; /* numa node of this context */
306 /* verbs stats per CTX */
307 struct hfi1_opcode_stats_perctx *opstats;
308 /*
309 * This is the kernel thread that will keep making
310 * progress on the user sdma requests behind the scenes.
311 * There is one per context (shared contexts use the master's).
312 */
313 struct task_struct *progress;
314 struct list_head sdma_queues;
315 spinlock_t sdma_qlock;
316
77241056
MM
317 /*
318 * The interrupt handler for a particular receive context can vary
319 * throughout it's lifetime. This is not a lock protected data member so
320 * it must be updated atomically and the prev and new value must always
321 * be valid. Worst case is we process an extra interrupt and up to 64
322 * packets with the wrong interrupt handler.
323 */
f4f30031 324 int (*do_interrupt)(struct hfi1_ctxtdata *rcd, int threaded);
77241056
MM
325};
326
327/*
328 * Represents a single packet at a high level. Put commonly computed things in
329 * here so we do not have to keep doing them over and over. The rule of thumb is
330 * if something is used one time to derive some value, store that something in
331 * here. If it is used multiple times, then store the result of that derivation
332 * in here.
333 */
334struct hfi1_packet {
335 void *ebuf;
336 void *hdr;
337 struct hfi1_ctxtdata *rcd;
338 __le32 *rhf_addr;
895420dd 339 struct rvt_qp *qp;
77241056
MM
340 struct hfi1_other_headers *ohdr;
341 u64 rhf;
342 u32 maxcnt;
343 u32 rhqoff;
344 u32 hdrqtail;
345 int numpkt;
346 u16 tlen;
347 u16 hlen;
348 s16 etail;
349 u16 rsize;
350 u8 updegr;
351 u8 rcv_flags;
352 u8 etype;
353};
354
355static inline bool has_sc4_bit(struct hfi1_packet *p)
356{
357 return !!rhf_dc_info(p->rhf);
358}
359
360/*
361 * Private data for snoop/capture support.
362 */
363struct hfi1_snoop_data {
364 int mode_flag;
365 struct cdev cdev;
366 struct device *class_dev;
367 spinlock_t snoop_lock;
368 struct list_head queue;
369 wait_queue_head_t waitq;
370 void *filter_value;
371 int (*filter_callback)(void *hdr, void *data, void *value);
372 u64 dcc_cfg; /* saved value of DCC Cfg register */
373};
374
375/* snoop mode_flag values */
376#define HFI1_PORT_SNOOP_MODE 1U
377#define HFI1_PORT_CAPTURE_MODE 2U
378
895420dd 379struct rvt_sge_state;
77241056
MM
380
381/*
382 * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
383 * Mostly for MADs that set or query link parameters, also ipath
384 * config interfaces
385 */
386#define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
387#define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
388#define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
389#define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
390#define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
391#define HFI1_IB_CFG_SPD 5 /* current Link spd */
392#define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
393#define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
394#define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
395#define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
396#define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
397#define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
398#define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
399#define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
400#define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
401#define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
402#define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
403#define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
404#define HFI1_IB_CFG_VL_HIGH_LIMIT 19
405#define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
406#define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
407
408/*
409 * HFI or Host Link States
410 *
411 * These describe the states the driver thinks the logical and physical
412 * states are in. Used as an argument to set_link_state(). Implemented
413 * as bits for easy multi-state checking. The actual state can only be
414 * one.
415 */
416#define __HLS_UP_INIT_BP 0
417#define __HLS_UP_ARMED_BP 1
418#define __HLS_UP_ACTIVE_BP 2
419#define __HLS_DN_DOWNDEF_BP 3 /* link down default */
420#define __HLS_DN_POLL_BP 4
421#define __HLS_DN_DISABLE_BP 5
422#define __HLS_DN_OFFLINE_BP 6
423#define __HLS_VERIFY_CAP_BP 7
424#define __HLS_GOING_UP_BP 8
425#define __HLS_GOING_OFFLINE_BP 9
426#define __HLS_LINK_COOLDOWN_BP 10
427
349ac71f 428#define HLS_UP_INIT BIT(__HLS_UP_INIT_BP)
429#define HLS_UP_ARMED BIT(__HLS_UP_ARMED_BP)
430#define HLS_UP_ACTIVE BIT(__HLS_UP_ACTIVE_BP)
431#define HLS_DN_DOWNDEF BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
432#define HLS_DN_POLL BIT(__HLS_DN_POLL_BP)
433#define HLS_DN_DISABLE BIT(__HLS_DN_DISABLE_BP)
434#define HLS_DN_OFFLINE BIT(__HLS_DN_OFFLINE_BP)
435#define HLS_VERIFY_CAP BIT(__HLS_VERIFY_CAP_BP)
436#define HLS_GOING_UP BIT(__HLS_GOING_UP_BP)
437#define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
438#define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
77241056
MM
439
440#define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
441
442/* use this MTU size if none other is given */
443#define HFI1_DEFAULT_ACTIVE_MTU 8192
444/* use this MTU size as the default maximum */
445#define HFI1_DEFAULT_MAX_MTU 8192
446/* default partition key */
447#define DEFAULT_PKEY 0xffff
448
449/*
450 * Possible fabric manager config parameters for fm_{get,set}_table()
451 */
452#define FM_TBL_VL_HIGH_ARB 1 /* Get/set VL high prio weights */
453#define FM_TBL_VL_LOW_ARB 2 /* Get/set VL low prio weights */
454#define FM_TBL_BUFFER_CONTROL 3 /* Get/set Buffer Control */
455#define FM_TBL_SC2VLNT 4 /* Get/set SC->VLnt */
456#define FM_TBL_VL_PREEMPT_ELEMS 5 /* Get (no set) VL preempt elems */
457#define FM_TBL_VL_PREEMPT_MATRIX 6 /* Get (no set) VL preempt matrix */
458
459/*
460 * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
461 * these are bits so they can be combined, e.g.
462 * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
463 */
464#define HFI1_RCVCTRL_TAILUPD_ENB 0x01
465#define HFI1_RCVCTRL_TAILUPD_DIS 0x02
466#define HFI1_RCVCTRL_CTXT_ENB 0x04
467#define HFI1_RCVCTRL_CTXT_DIS 0x08
468#define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
469#define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
470#define HFI1_RCVCTRL_PKEY_ENB 0x40 /* Note, default is enabled */
471#define HFI1_RCVCTRL_PKEY_DIS 0x80
472#define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
473#define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
474#define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
475#define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
476#define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
477#define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
478#define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
479#define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
480
481/* partition enforcement flags */
482#define HFI1_PART_ENFORCE_IN 0x1
483#define HFI1_PART_ENFORCE_OUT 0x2
484
485/* how often we check for synthetic counter wrap around */
486#define SYNTH_CNT_TIME 2
487
488/* Counter flags */
489#define CNTR_NORMAL 0x0 /* Normal counters, just read register */
490#define CNTR_SYNTH 0x1 /* Synthetic counters, saturate at all 1s */
491#define CNTR_DISABLED 0x2 /* Disable this counter */
492#define CNTR_32BIT 0x4 /* Simulate 64 bits for this counter */
493#define CNTR_VL 0x8 /* Per VL counter */
a699c6c2 494#define CNTR_SDMA 0x10
77241056
MM
495#define CNTR_INVALID_VL -1 /* Specifies invalid VL */
496#define CNTR_MODE_W 0x0
497#define CNTR_MODE_R 0x1
498
499/* VLs Supported/Operational */
500#define HFI1_MIN_VLS_SUPPORTED 1
501#define HFI1_MAX_VLS_SUPPORTED 8
502
503static inline void incr_cntr64(u64 *cntr)
504{
505 if (*cntr < (u64)-1LL)
506 (*cntr)++;
507}
508
509static inline void incr_cntr32(u32 *cntr)
510{
511 if (*cntr < (u32)-1LL)
512 (*cntr)++;
513}
514
515#define MAX_NAME_SIZE 64
516struct hfi1_msix_entry {
517 struct msix_entry msix;
518 void *arg;
519 char name[MAX_NAME_SIZE];
520 cpumask_var_t mask;
521};
522
523/* per-SL CCA information */
524struct cca_timer {
525 struct hrtimer hrtimer;
526 struct hfi1_pportdata *ppd; /* read-only */
527 int sl; /* read-only */
528 u16 ccti; /* read/write - current value of CCTI */
529};
530
531struct link_down_reason {
532 /*
533 * SMA-facing value. Should be set from .latest when
534 * HLS_UP_* -> HLS_DN_* transition actually occurs.
535 */
536 u8 sma;
537 u8 latest;
538};
539
540enum {
541 LO_PRIO_TABLE,
542 HI_PRIO_TABLE,
543 MAX_PRIO_TABLE
544};
545
546struct 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 */
557struct 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 /* QSFP support */
567 struct qsfp_data qsfp_info;
568
569 /* GUID for this interface, in host order */
570 u64 guid;
571 /* GUID for peer interface, in host order */
572 u64 neighbor_guid;
573
574 /* up or down physical link state */
575 u32 linkup;
576
577 /*
578 * this address is mapped read-only into user processes so they can
579 * get status cheaply, whenever they want. One qword of status per port
580 */
581 u64 *statusp;
582
583 /* SendDMA related entries */
584
585 struct workqueue_struct *hfi1_wq;
586
587 /* move out of interrupt context */
588 struct work_struct link_vc_work;
589 struct work_struct link_up_work;
590 struct work_struct link_down_work;
591 struct work_struct sma_message_work;
592 struct work_struct freeze_work;
593 struct work_struct link_downgrade_work;
594 struct work_struct link_bounce_work;
595 /* host link state variables */
596 struct mutex hls_lock;
597 u32 host_link_state;
598
599 spinlock_t sdma_alllock ____cacheline_aligned_in_smp;
600
601 u32 lstate; /* logical link state */
602
603 /* these are the "32 bit" regs */
604
605 u32 ibmtu; /* The MTU programmed for this unit */
606 /*
607 * Current max size IB packet (in bytes) including IB headers, that
608 * we can send. Changes when ibmtu changes.
609 */
610 u32 ibmaxlen;
611 u32 current_egress_rate; /* units [10^6 bits/sec] */
612 /* LID programmed for this instance */
613 u16 lid;
614 /* list of pkeys programmed; 0 if not set */
615 u16 pkeys[MAX_PKEY_VALUES];
616 u16 link_width_supported;
617 u16 link_width_downgrade_supported;
618 u16 link_speed_supported;
619 u16 link_width_enabled;
620 u16 link_width_downgrade_enabled;
621 u16 link_speed_enabled;
622 u16 link_width_active;
623 u16 link_width_downgrade_tx_active;
624 u16 link_width_downgrade_rx_active;
625 u16 link_speed_active;
626 u8 vls_supported;
627 u8 vls_operational;
628 /* LID mask control */
629 u8 lmc;
630 /* Rx Polarity inversion (compensate for ~tx on partner) */
631 u8 rx_pol_inv;
632
633 u8 hw_pidx; /* physical port index */
634 u8 port; /* IB port number and index into dd->pports - 1 */
635 /* type of neighbor node */
636 u8 neighbor_type;
637 u8 neighbor_normal;
638 u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
639 u8 neighbor_port_number;
640 u8 is_sm_config_started;
641 u8 offline_disabled_reason;
642 u8 is_active_optimize_enabled;
643 u8 driver_link_ready; /* driver ready for active link */
644 u8 link_enabled; /* link enabled? */
645 u8 linkinit_reason;
646 u8 local_tx_rate; /* rate given to 8051 firmware */
647
648 /* placeholders for IB MAD packet settings */
649 u8 overrun_threshold;
650 u8 phy_error_threshold;
651
652 /* used to override LED behavior */
653 u8 led_override; /* Substituted for normal value, if non-zero */
654 u16 led_override_timeoff; /* delta to next timer event */
655 u8 led_override_vals[2]; /* Alternates per blink-frame */
656 u8 led_override_phase; /* Just counts, LSB picks from vals[] */
657 atomic_t led_override_timer_active;
658 /* Used to flash LEDs in override mode */
659 struct timer_list led_override_timer;
660 u32 sm_trap_qp;
661 u32 sa_qp;
662
663 /*
664 * cca_timer_lock protects access to the per-SL cca_timer
665 * structures (specifically the ccti member).
666 */
667 spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
668 struct cca_timer cca_timer[OPA_MAX_SLS];
669
670 /* List of congestion control table entries */
671 struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
672
673 /* congestion entries, each entry corresponding to a SL */
674 struct opa_congestion_setting_entry_shadow
675 congestion_entries[OPA_MAX_SLS];
676
677 /*
678 * cc_state_lock protects (write) access to the per-port
679 * struct cc_state.
680 */
681 spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
682
683 struct cc_state __rcu *cc_state;
684
685 /* Total number of congestion control table entries */
686 u16 total_cct_entry;
687
688 /* Bit map identifying service level */
689 u32 cc_sl_control_map;
690
691 /* CA's max number of 64 entry units in the congestion control table */
692 u8 cc_max_table_entries;
693
694 /* begin congestion log related entries
695 * cc_log_lock protects all congestion log related data */
696 spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
697 u8 threshold_cong_event_map[OPA_MAX_SLS/8];
698 u16 threshold_event_counter;
699 struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
700 int cc_log_idx; /* index for logging events */
701 int cc_mad_idx; /* index for reporting events */
702 /* end congestion log related entries */
703
704 struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
705
706 /* port relative counter buffer */
707 u64 *cntrs;
708 /* port relative synthetic counter buffer */
709 u64 *scntrs;
710 /* we synthesize port_xmit_discards from several egress errors */
711 u64 port_xmit_discards;
712 u64 port_xmit_constraint_errors;
713 u64 port_rcv_constraint_errors;
714 /* count of 'link_err' interrupts from DC */
715 u64 link_downed;
716 /* number of times link retrained successfully */
717 u64 link_up;
6d014530
DL
718 /* number of times a link unknown frame was reported */
719 u64 unknown_frame_count;
77241056
MM
720 /* port_ltp_crc_mode is returned in 'portinfo' MADs */
721 u16 port_ltp_crc_mode;
722 /* port_crc_mode_enabled is the crc we support */
723 u8 port_crc_mode_enabled;
724 /* mgmt_allowed is also returned in 'portinfo' MADs */
725 u8 mgmt_allowed;
726 u8 part_enforce; /* partition enforcement flags */
727 struct link_down_reason local_link_down_reason;
728 struct link_down_reason neigh_link_down_reason;
729 /* Value to be sent to link peer on LinkDown .*/
730 u8 remote_link_down_reason;
731 /* Error events that will cause a port bounce. */
732 u32 port_error_action;
fb9036dd 733 struct work_struct linkstate_active_work;
77241056
MM
734};
735
736typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
737
738typedef void (*opcode_handler)(struct hfi1_packet *packet);
739
740/* return values for the RHF receive functions */
741#define RHF_RCV_CONTINUE 0 /* keep going */
742#define RHF_RCV_DONE 1 /* stop, this packet processed */
743#define RHF_RCV_REPROCESS 2 /* stop. retain this packet */
744
745struct rcv_array_data {
746 u8 group_size;
747 u16 ngroups;
748 u16 nctxt_extra;
749};
750
751struct per_vl_data {
752 u16 mtu;
753 struct send_context *sc;
754};
755
756/* 16 to directly index */
757#define PER_VL_SEND_CONTEXTS 16
758
759struct err_info_rcvport {
760 u8 status_and_code;
761 u64 packet_flit1;
762 u64 packet_flit2;
763};
764
765struct err_info_constraint {
766 u8 status;
767 u16 pkey;
768 u32 slid;
769};
770
771struct hfi1_temp {
772 unsigned int curr; /* current temperature */
773 unsigned int lo_lim; /* low temperature limit */
774 unsigned int hi_lim; /* high temperature limit */
775 unsigned int crit_lim; /* critical temperature limit */
776 u8 triggers; /* temperature triggers */
777};
778
779/* device data struct now contains only "general per-device" info.
780 * fields related to a physical IB port are in a hfi1_pportdata struct.
781 */
782struct sdma_engine;
783struct sdma_vl_map;
784
785#define BOARD_VERS_MAX 96 /* how long the version string can be */
786#define SERIAL_MAX 16 /* length of the serial number */
787
788struct hfi1_devdata {
789 struct hfi1_ibdev verbs_dev; /* must be first */
790 struct list_head list;
791 /* pointers to related structs for this device */
792 /* pci access data structure */
793 struct pci_dev *pcidev;
794 struct cdev user_cdev;
795 struct cdev diag_cdev;
796 struct cdev ui_cdev;
797 struct device *user_device;
798 struct device *diag_device;
799 struct device *ui_device;
800
801 /* mem-mapped pointer to base of chip regs */
802 u8 __iomem *kregbase;
803 /* end of mem-mapped chip space excluding sendbuf and user regs */
804 u8 __iomem *kregend;
805 /* physical address of chip for io_remap, etc. */
806 resource_size_t physaddr;
807 /* receive context data */
808 struct hfi1_ctxtdata **rcd;
809 /* send context data */
810 struct send_context_info *send_contexts;
811 /* map hardware send contexts to software index */
812 u8 *hw_to_sw;
813 /* spinlock for allocating and releasing send context resources */
814 spinlock_t sc_lock;
815 /* Per VL data. Enough for all VLs but not all elements are set/used. */
816 struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
817 /* seqlock for sc2vl */
818 seqlock_t sc2vl_lock;
819 u64 sc2vl[4];
820 /* Send Context initialization lock. */
821 spinlock_t sc_init_lock;
822
823 /* fields common to all SDMA engines */
824
825 /* default flags to last descriptor */
826 u64 default_desc1;
827 volatile __le64 *sdma_heads_dma; /* DMA'ed by chip */
828 dma_addr_t sdma_heads_phys;
829 void *sdma_pad_dma; /* DMA'ed by chip */
830 dma_addr_t sdma_pad_phys;
831 /* for deallocation */
832 size_t sdma_heads_size;
833 /* number from the chip */
834 u32 chip_sdma_engines;
835 /* num used */
836 u32 num_sdma;
837 /* lock for sdma_map */
838 spinlock_t sde_map_lock;
839 /* array of engines sized by num_sdma */
840 struct sdma_engine *per_sdma;
841 /* array of vl maps */
842 struct sdma_vl_map __rcu *sdma_map;
843 /* SPC freeze waitqueue and variable */
844 wait_queue_head_t sdma_unfreeze_wq;
845 atomic_t sdma_unfreeze_count;
846
847
848 /* hfi1_pportdata, points to array of (physical) port-specific
849 * data structs, indexed by pidx (0..n-1)
850 */
851 struct hfi1_pportdata *pport;
852
853 /* mem-mapped pointer to base of PIO buffers */
854 void __iomem *piobase;
855 /*
856 * write-combining mem-mapped pointer to base of RcvArray
857 * memory.
858 */
859 void __iomem *rcvarray_wc;
860 /*
861 * credit return base - a per-NUMA range of DMA address that
862 * the chip will use to update the per-context free counter
863 */
864 struct credit_return_base *cr_base;
865
866 /* send context numbers and sizes for each type */
867 struct sc_config_sizes sc_sizes[SC_MAX];
868
869 u32 lcb_access_count; /* count of LCB users */
870
871 char *boardname; /* human readable board info */
872
873 /* device (not port) flags, basically device capabilities */
874 u32 flags;
875
876 /* reset value */
877 u64 z_int_counter;
878 u64 z_rcv_limit;
879 /* percpu int_counter */
880 u64 __percpu *int_counter;
881 u64 __percpu *rcv_limit;
882
883 /* number of receive contexts in use by the driver */
884 u32 num_rcv_contexts;
885 /* number of pio send contexts in use by the driver */
886 u32 num_send_contexts;
887 /*
888 * number of ctxts available for PSM open
889 */
890 u32 freectxts;
891 /* base receive interrupt timeout, in CSR units */
892 u32 rcv_intr_timeout_csr;
893
894 u64 __iomem *egrtidbase;
895 spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
896 spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
897 /* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
898 spinlock_t uctxt_lock; /* rcd and user context changes */
899 /* exclusive access to 8051 */
900 spinlock_t dc8051_lock;
901 /* exclusive access to 8051 memory */
902 spinlock_t dc8051_memlock;
903 int dc8051_timed_out; /* remember if the 8051 timed out */
904 /*
905 * A page that will hold event notification bitmaps for all
906 * contexts. This page will be mapped into all processes.
907 */
908 unsigned long *events;
909 /*
910 * per unit status, see also portdata statusp
911 * mapped read-only into user processes so they can get unit and
912 * IB link status cheaply
913 */
914 struct hfi1_status *status;
915 u32 freezelen; /* max length of freezemsg */
916
917 /* revision register shadow */
918 u64 revision;
919 /* Base GUID for device (network order) */
920 u64 base_guid;
921
922 /* these are the "32 bit" regs */
923
924 /* value we put in kr_rcvhdrsize */
925 u32 rcvhdrsize;
926 /* number of receive contexts the chip supports */
927 u32 chip_rcv_contexts;
928 /* number of receive array entries */
929 u32 chip_rcv_array_count;
930 /* number of PIO send contexts the chip supports */
931 u32 chip_send_contexts;
932 /* number of bytes in the PIO memory buffer */
933 u32 chip_pio_mem_size;
934 /* number of bytes in the SDMA memory buffer */
935 u32 chip_sdma_mem_size;
936
937 /* size of each rcvegrbuffer */
938 u32 rcvegrbufsize;
939 /* log2 of above */
940 u16 rcvegrbufsize_shift;
941 /* both sides of the PCIe link are gen3 capable */
942 u8 link_gen3_capable;
943 /* localbus width (1, 2,4,8,16,32) from config space */
944 u32 lbus_width;
945 /* localbus speed in MHz */
946 u32 lbus_speed;
947 int unit; /* unit # of this chip */
948 int node; /* home node of this chip */
949
950 /* save these PCI fields to restore after a reset */
951 u32 pcibar0;
952 u32 pcibar1;
953 u32 pci_rom;
954 u16 pci_command;
955 u16 pcie_devctl;
956 u16 pcie_lnkctl;
957 u16 pcie_devctl2;
958 u32 pci_msix0;
959 u32 pci_lnkctl3;
960 u32 pci_tph2;
961
962 /*
963 * ASCII serial number, from flash, large enough for original
964 * all digit strings, and longer serial number format
965 */
966 u8 serial[SERIAL_MAX];
967 /* human readable board version */
968 u8 boardversion[BOARD_VERS_MAX];
969 u8 lbus_info[32]; /* human readable localbus info */
970 /* chip major rev, from CceRevision */
971 u8 majrev;
972 /* chip minor rev, from CceRevision */
973 u8 minrev;
974 /* hardware ID */
975 u8 hfi1_id;
976 /* implementation code */
977 u8 icode;
978 /* default link down value (poll/sleep) */
979 u8 link_default;
980 /* vAU of this device */
981 u8 vau;
982 /* vCU of this device */
983 u8 vcu;
984 /* link credits of this device */
985 u16 link_credits;
986 /* initial vl15 credits to use */
987 u16 vl15_init;
988
989 /* Misc small ints */
990 /* Number of physical ports available */
991 u8 num_pports;
992 /* Lowest context number which can be used by user processes */
993 u8 first_user_ctxt;
994 u8 n_krcv_queues;
995 u8 qos_shift;
996 u8 qpn_mask;
997
998 u16 rhf_offset; /* offset of RHF within receive header entry */
999 u16 irev; /* implementation revision */
1000 u16 dc8051_ver; /* 8051 firmware version */
1001
1002 struct platform_config_cache pcfg_cache;
1003 /* control high-level access to qsfp */
1004 struct mutex qsfp_i2c_mutex;
1005
1006 struct diag_client *diag_client;
1007 spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1008
1009 u8 psxmitwait_supported;
1010 /* cycle length of PS* counters in HW (in picoseconds) */
1011 u16 psxmitwait_check_rate;
1012 /* high volume overflow errors deferred to tasklet */
1013 struct tasklet_struct error_tasklet;
77241056
MM
1014
1015 /* MSI-X information */
1016 struct hfi1_msix_entry *msix_entries;
1017 u32 num_msix_entries;
1018
1019 /* INTx information */
1020 u32 requested_intx_irq; /* did we request one? */
1021 char intx_name[MAX_NAME_SIZE]; /* INTx name */
1022
1023 /* general interrupt: mask of handled interrupts */
1024 u64 gi_mask[CCE_NUM_INT_CSRS];
1025
1026 struct rcv_array_data rcv_entries;
1027
1028 /*
1029 * 64 bit synthetic counters
1030 */
1031 struct timer_list synth_stats_timer;
1032
1033 /*
1034 * device counters
1035 */
1036 char *cntrnames;
1037 size_t cntrnameslen;
1038 size_t ndevcntrs;
1039 u64 *cntrs;
1040 u64 *scntrs;
1041
1042 /*
1043 * remembered values for synthetic counters
1044 */
1045 u64 last_tx;
1046 u64 last_rx;
1047
1048 /*
1049 * per-port counters
1050 */
1051 size_t nportcntrs;
1052 char *portcntrnames;
1053 size_t portcntrnameslen;
1054
1055 struct hfi1_snoop_data hfi1_snoop;
1056
1057 struct err_info_rcvport err_info_rcvport;
1058 struct err_info_constraint err_info_rcv_constraint;
1059 struct err_info_constraint err_info_xmit_constraint;
1060 u8 err_info_uncorrectable;
1061 u8 err_info_fmconfig;
1062
1063 atomic_t drop_packet;
1064 u8 do_drop;
1065
2c5b521a
JR
1066 /*
1067 * Software counters for the status bits defined by the
1068 * associated error status registers
1069 */
1070 u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1071 u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1072 u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1073 u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1074 u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1075 u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1076 u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1077
1078 /* Software counter that spans all contexts */
1079 u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1080 /* Software counter that spans all DMA engines */
1081 u64 sw_send_dma_eng_err_status_cnt[
1082 NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1083 /* Software counter that aggregates all cce_err_status errors */
1084 u64 sw_cce_err_status_aggregate;
1085
77241056
MM
1086 /* receive interrupt functions */
1087 rhf_rcv_function_ptr *rhf_rcv_function_map;
1088 rhf_rcv_function_ptr normal_rhf_rcv_functions[8];
1089
1090 /*
1091 * Handlers for outgoing data so that snoop/capture does not
1092 * have to have its hooks in the send path
1093 */
895420dd 1094 int (*process_pio_send)(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
d46e5144 1095 u64 pbc);
895420dd 1096 int (*process_dma_send)(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
d46e5144 1097 u64 pbc);
77241056
MM
1098 void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1099 u64 pbc, const void *from, size_t count);
1100
1101 /* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1102 u8 oui1;
1103 u8 oui2;
1104 u8 oui3;
1105 /* Timer and counter used to detect RcvBufOvflCnt changes */
1106 struct timer_list rcverr_timer;
1107 u32 rcv_ovfl_cnt;
1108
1109 int assigned_node_id;
1110 wait_queue_head_t event_queue;
1111
1112 /* Save the enabled LCB error bits */
1113 u64 lcb_err_en;
1114 u8 dc_shutdown;
46b010d3
MB
1115
1116 /* receive context tail dummy address */
1117 __le64 *rcvhdrtail_dummy_kvaddr;
1118 dma_addr_t rcvhdrtail_dummy_physaddr;
77241056
MM
1119};
1120
1121/* 8051 firmware version helper */
1122#define dc8051_ver(a, b) ((a) << 8 | (b))
1123
1124/* f_put_tid types */
1125#define PT_EXPECTED 0
1126#define PT_EAGER 1
1127#define PT_INVALID 2
1128
f727a0c3
MH
1129struct mmu_rb_node;
1130
77241056
MM
1131/* Private data for file operations */
1132struct hfi1_filedata {
1133 struct hfi1_ctxtdata *uctxt;
1134 unsigned subctxt;
1135 struct hfi1_user_sdma_comp_q *cq;
1136 struct hfi1_user_sdma_pkt_q *pq;
1137 /* for cpu affinity; -1 if none */
1138 int rec_cpu_num;
a86cd357
MH
1139 struct mmu_notifier mn;
1140 struct rb_root tid_rb_root;
1141 spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1142 u32 tid_limit;
1143 u32 tid_used;
1144 spinlock_t rb_lock; /* protect tid_rb_root RB tree */
1145 u32 *invalid_tids;
1146 u32 invalid_tid_idx;
1147 spinlock_t invalid_lock; /* protect the invalid_tids array */
1148 int (*mmu_rb_insert)(struct rb_root *, struct mmu_rb_node *);
77241056
MM
1149};
1150
1151extern struct list_head hfi1_dev_list;
1152extern spinlock_t hfi1_devs_lock;
1153struct hfi1_devdata *hfi1_lookup(int unit);
1154extern u32 hfi1_cpulist_count;
1155extern unsigned long *hfi1_cpulist;
1156
1157extern unsigned int snoop_drop_send;
1158extern unsigned int snoop_force_capture;
1159int hfi1_init(struct hfi1_devdata *, int);
1160int hfi1_count_units(int *npresentp, int *nupp);
1161int hfi1_count_active_units(void);
1162
1163int hfi1_diag_add(struct hfi1_devdata *);
1164void hfi1_diag_remove(struct hfi1_devdata *);
1165void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1166
1167void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1168
1169int hfi1_create_rcvhdrq(struct hfi1_devdata *, struct hfi1_ctxtdata *);
1170int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *);
1171int hfi1_create_ctxts(struct hfi1_devdata *dd);
1172struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *, u32);
1173void hfi1_init_pportdata(struct pci_dev *, struct hfi1_pportdata *,
1174 struct hfi1_devdata *, u8, u8);
1175void hfi1_free_ctxtdata(struct hfi1_devdata *, struct hfi1_ctxtdata *);
1176
f4f30031
DL
1177int handle_receive_interrupt(struct hfi1_ctxtdata *, int);
1178int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *, int);
1179int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *, int);
fb9036dd 1180void set_all_slowpath(struct hfi1_devdata *dd);
f4f30031
DL
1181
1182/* receive packet handler dispositions */
1183#define RCV_PKT_OK 0x0 /* keep going */
1184#define RCV_PKT_LIMIT 0x1 /* stop, hit limit, start thread */
1185#define RCV_PKT_DONE 0x2 /* stop, no more packets detected */
1186
1187/* calculate the current RHF address */
1188static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1189{
1190 return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->dd->rhf_offset;
1191}
1192
77241056
MM
1193int hfi1_reset_device(int);
1194
1195/* return the driver's idea of the logical OPA port state */
1196static inline u32 driver_lstate(struct hfi1_pportdata *ppd)
1197{
1198 return ppd->lstate; /* use the cached value */
1199}
1200
fb9036dd
JS
1201void receive_interrupt_work(struct work_struct *work);
1202
1203/* extract service channel from header and rhf */
1204static inline int hdr2sc(struct hfi1_message_header *hdr, u64 rhf)
1205{
1206 return ((be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf) |
1207 ((!!(rhf & RHF_DC_INFO_MASK)) << 4);
1208}
1209
77241056
MM
1210static inline u16 generate_jkey(kuid_t uid)
1211{
1212 return from_kuid(current_user_ns(), uid) & 0xffff;
1213}
1214
1215/*
1216 * active_egress_rate
1217 *
1218 * returns the active egress rate in units of [10^6 bits/sec]
1219 */
1220static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1221{
1222 u16 link_speed = ppd->link_speed_active;
1223 u16 link_width = ppd->link_width_active;
1224 u32 egress_rate;
1225
1226 if (link_speed == OPA_LINK_SPEED_25G)
1227 egress_rate = 25000;
1228 else /* assume OPA_LINK_SPEED_12_5G */
1229 egress_rate = 12500;
1230
1231 switch (link_width) {
1232 case OPA_LINK_WIDTH_4X:
1233 egress_rate *= 4;
1234 break;
1235 case OPA_LINK_WIDTH_3X:
1236 egress_rate *= 3;
1237 break;
1238 case OPA_LINK_WIDTH_2X:
1239 egress_rate *= 2;
1240 break;
1241 default:
1242 /* assume IB_WIDTH_1X */
1243 break;
1244 }
1245
1246 return egress_rate;
1247}
1248
1249/*
1250 * egress_cycles
1251 *
1252 * Returns the number of 'fabric clock cycles' to egress a packet
1253 * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1254 * rate is (approximately) 805 MHz, the units of the returned value
1255 * are (1/805 MHz).
1256 */
1257static inline u32 egress_cycles(u32 len, u32 rate)
1258{
1259 u32 cycles;
1260
1261 /*
1262 * cycles is:
1263 *
1264 * (length) [bits] / (rate) [bits/sec]
1265 * ---------------------------------------------------
1266 * fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1267 */
1268
1269 cycles = len * 8; /* bits */
1270 cycles *= 805;
1271 cycles /= rate;
1272
1273 return cycles;
1274}
1275
1276void set_link_ipg(struct hfi1_pportdata *ppd);
1277void process_becn(struct hfi1_pportdata *ppd, u8 sl, u16 rlid, u32 lqpn,
1278 u32 rqpn, u8 svc_type);
895420dd 1279void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
77241056
MM
1280 u32 pkey, u32 slid, u32 dlid, u8 sc5,
1281 const struct ib_grh *old_grh);
1282
1283#define PACKET_EGRESS_TIMEOUT 350
1284static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1285{
1286 /* Pause at least 1us, to ensure chip returns all credits */
1287 u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1288
1289 udelay(usec ? usec : 1);
1290}
1291
1292/**
1293 * sc_to_vlt() reverse lookup sc to vl
1294 * @dd - devdata
1295 * @sc5 - 5 bit sc
1296 */
1297static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1298{
1299 unsigned seq;
1300 u8 rval;
1301
1302 if (sc5 >= OPA_MAX_SCS)
1303 return (u8)(0xff);
1304
1305 do {
1306 seq = read_seqbegin(&dd->sc2vl_lock);
1307 rval = *(((u8 *)dd->sc2vl) + sc5);
1308 } while (read_seqretry(&dd->sc2vl_lock, seq));
1309
1310 return rval;
1311}
1312
1313#define PKEY_MEMBER_MASK 0x8000
1314#define PKEY_LOW_15_MASK 0x7fff
1315
1316/*
1317 * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1318 * being an entry from the ingress partition key table), return 0
1319 * otherwise. Use the matching criteria for ingress partition keys
1320 * specified in the OPAv1 spec., section 9.10.14.
1321 */
1322static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1323{
1324 u16 mkey = pkey & PKEY_LOW_15_MASK;
1325 u16 ment = ent & PKEY_LOW_15_MASK;
1326
1327 if (mkey == ment) {
1328 /*
1329 * If pkey[15] is clear (limited partition member),
1330 * is bit 15 in the corresponding table element
1331 * clear (limited member)?
1332 */
1333 if (!(pkey & PKEY_MEMBER_MASK))
1334 return !!(ent & PKEY_MEMBER_MASK);
1335 return 1;
1336 }
1337 return 0;
1338}
1339
1340/*
1341 * ingress_pkey_table_search - search the entire pkey table for
1342 * an entry which matches 'pkey'. return 0 if a match is found,
1343 * and 1 otherwise.
1344 */
1345static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1346{
1347 int i;
1348
1349 for (i = 0; i < MAX_PKEY_VALUES; i++) {
1350 if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1351 return 0;
1352 }
1353 return 1;
1354}
1355
1356/*
1357 * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1358 * i.e., increment port_rcv_constraint_errors for the port, and record
1359 * the 'error info' for this failure.
1360 */
1361static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1362 u16 slid)
1363{
1364 struct hfi1_devdata *dd = ppd->dd;
1365
1366 incr_cntr64(&ppd->port_rcv_constraint_errors);
1367 if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1368 dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1369 dd->err_info_rcv_constraint.slid = slid;
1370 dd->err_info_rcv_constraint.pkey = pkey;
1371 }
1372}
1373
1374/*
1375 * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1376 * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1377 * is a hint as to the best place in the partition key table to begin
1378 * searching. This function should not be called on the data path because
1379 * of performance reasons. On datapath pkey check is expected to be done
1380 * by HW and rcv_pkey_check function should be called instead.
1381 */
1382static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1383 u8 sc5, u8 idx, u16 slid)
1384{
1385 if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1386 return 0;
1387
1388 /* If SC15, pkey[0:14] must be 0x7fff */
1389 if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1390 goto bad;
1391
1392 /* Is the pkey = 0x0, or 0x8000? */
1393 if ((pkey & PKEY_LOW_15_MASK) == 0)
1394 goto bad;
1395
1396 /* The most likely matching pkey has index 'idx' */
1397 if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1398 return 0;
1399
1400 /* no match - try the whole table */
1401 if (!ingress_pkey_table_search(ppd, pkey))
1402 return 0;
1403
1404bad:
1405 ingress_pkey_table_fail(ppd, pkey, slid);
1406 return 1;
1407}
1408
1409/*
1410 * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1411 * otherwise. It only ensures pkey is vlid for QP0. This function
1412 * should be called on the data path instead of ingress_pkey_check
1413 * as on data path, pkey check is done by HW (except for QP0).
1414 */
1415static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1416 u8 sc5, u16 slid)
1417{
1418 if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1419 return 0;
1420
1421 /* If SC15, pkey[0:14] must be 0x7fff */
1422 if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1423 goto bad;
1424
1425 return 0;
1426bad:
1427 ingress_pkey_table_fail(ppd, pkey, slid);
1428 return 1;
1429}
1430
1431/* MTU handling */
1432
1433/* MTU enumeration, 256-4k match IB */
1434#define OPA_MTU_0 0
1435#define OPA_MTU_256 1
1436#define OPA_MTU_512 2
1437#define OPA_MTU_1024 3
1438#define OPA_MTU_2048 4
1439#define OPA_MTU_4096 5
1440
1441u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1442int mtu_to_enum(u32 mtu, int default_if_bad);
1443u16 enum_to_mtu(int);
1444static inline int valid_ib_mtu(unsigned int mtu)
1445{
1446 return mtu == 256 || mtu == 512 ||
1447 mtu == 1024 || mtu == 2048 ||
1448 mtu == 4096;
1449}
1450static inline int valid_opa_max_mtu(unsigned int mtu)
1451{
1452 return mtu >= 2048 &&
1453 (valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1454}
1455
1456int set_mtu(struct hfi1_pportdata *);
1457
1458int hfi1_set_lid(struct hfi1_pportdata *, u32, u8);
1459void hfi1_disable_after_error(struct hfi1_devdata *);
1460int hfi1_set_uevent_bits(struct hfi1_pportdata *, const int);
1461int hfi1_rcvbuf_validate(u32, u8, u16 *);
1462
1463int fm_get_table(struct hfi1_pportdata *, int, void *);
1464int fm_set_table(struct hfi1_pportdata *, int, void *);
1465
1466void set_up_vl15(struct hfi1_devdata *dd, u8 vau, u16 vl15buf);
1467void reset_link_credits(struct hfi1_devdata *dd);
1468void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1469
1470int snoop_recv_handler(struct hfi1_packet *packet);
895420dd 1471int snoop_send_dma_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
d46e5144 1472 u64 pbc);
895420dd 1473int snoop_send_pio_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
d46e5144 1474 u64 pbc);
77241056
MM
1475void snoop_inline_pio_send(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1476 u64 pbc, const void *from, size_t count);
1477
77241056
MM
1478static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1479{
1480 return ppd->dd;
1481}
1482
1483static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1484{
1485 return container_of(dev, struct hfi1_devdata, verbs_dev);
1486}
1487
1488static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1489{
1490 return dd_from_dev(to_idev(ibdev));
1491}
1492
1493static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1494{
1495 return container_of(ibp, struct hfi1_pportdata, ibport_data);
1496}
1497
1498static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u8 port)
1499{
1500 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1501 unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1502
1503 WARN_ON(pidx >= dd->num_pports);
1504 return &dd->pport[pidx].ibport_data;
1505}
1506
1507/*
1508 * Return the indexed PKEY from the port PKEY table.
1509 */
1510static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
1511{
1512 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1513 u16 ret;
1514
1515 if (index >= ARRAY_SIZE(ppd->pkeys))
1516 ret = 0;
1517 else
1518 ret = ppd->pkeys[index];
1519
1520 return ret;
1521}
1522
1523/*
1524 * Readers of cc_state must call get_cc_state() under rcu_read_lock().
1525 * Writers of cc_state must call get_cc_state() under cc_state_lock.
1526 */
1527static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
1528{
1529 return rcu_dereference(ppd->cc_state);
1530}
1531
1532/*
1533 * values for dd->flags (_device_ related flags)
1534 */
1535#define HFI1_INITTED 0x1 /* chip and driver up and initted */
1536#define HFI1_PRESENT 0x2 /* chip accesses can be done */
1537#define HFI1_FROZEN 0x4 /* chip in SPC freeze */
1538#define HFI1_HAS_SDMA_TIMEOUT 0x8
1539#define HFI1_HAS_SEND_DMA 0x10 /* Supports Send DMA */
1540#define HFI1_FORCED_FREEZE 0x80 /* driver forced freeze mode */
1541#define HFI1_DO_INIT_ASIC 0x100 /* This device will init the ASIC */
1542
1543/* IB dword length mask in PBC (lower 11 bits); same for all chips */
1544#define HFI1_PBC_LENGTH_MASK ((1 << 11) - 1)
1545
1546
1547/* ctxt_flag bit offsets */
1548 /* context has been setup */
1549#define HFI1_CTXT_SETUP_DONE 1
1550 /* waiting for a packet to arrive */
1551#define HFI1_CTXT_WAITING_RCV 2
1552 /* master has not finished initializing */
1553#define HFI1_CTXT_MASTER_UNINIT 4
1554 /* waiting for an urgent packet to arrive */
1555#define HFI1_CTXT_WAITING_URG 5
1556
1557/* free up any allocated data at closes */
1558struct hfi1_devdata *hfi1_init_dd(struct pci_dev *,
1559 const struct pci_device_id *);
1560void hfi1_free_devdata(struct hfi1_devdata *);
1561void cc_state_reclaim(struct rcu_head *rcu);
1562struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra);
1563
1564/*
1565 * Set LED override, only the two LSBs have "public" meaning, but
1566 * any non-zero value substitutes them for the Link and LinkTrain
1567 * LED states.
1568 */
1569#define HFI1_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
1570#define HFI1_LED_LOG 2 /* Logical (link) YELLOW LED */
1571void hfi1_set_led_override(struct hfi1_pportdata *ppd, unsigned int val);
1572
1573#define HFI1_CREDIT_RETURN_RATE (100)
1574
1575/*
1576 * The number of words for the KDETH protocol field. If this is
1577 * larger then the actual field used, then part of the payload
1578 * will be in the header.
1579 *
1580 * Optimally, we want this sized so that a typical case will
1581 * use full cache lines. The typical local KDETH header would
1582 * be:
1583 *
1584 * Bytes Field
1585 * 8 LRH
1586 * 12 BHT
1587 * ?? KDETH
1588 * 8 RHF
1589 * ---
1590 * 28 + KDETH
1591 *
1592 * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
1593 */
1594#define DEFAULT_RCVHDRSIZE 9
1595
1596/*
1597 * Maximal header byte count:
1598 *
1599 * Bytes Field
1600 * 8 LRH
1601 * 40 GRH (optional)
1602 * 12 BTH
1603 * ?? KDETH
1604 * 8 RHF
1605 * ---
1606 * 68 + KDETH
1607 *
1608 * We also want to maintain a cache line alignment to assist DMA'ing
1609 * of the header bytes. Round up to a good size.
1610 */
1611#define DEFAULT_RCVHDR_ENTSIZE 32
1612
def82284
MH
1613int hfi1_acquire_user_pages(unsigned long, size_t, bool, struct page **);
1614void hfi1_release_user_pages(struct page **, size_t, bool);
77241056
MM
1615
1616static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1617{
1618 *((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
1619}
1620
1621static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
1622{
1623 /*
1624 * volatile because it's a DMA target from the chip, routine is
1625 * inlined, and don't want register caching or reordering.
1626 */
1627 return (u32) le64_to_cpu(*rcd->rcvhdrtail_kvaddr);
1628}
1629
1630/*
1631 * sysfs interface.
1632 */
1633
1634extern const char ib_hfi1_version[];
1635
1636int hfi1_device_create(struct hfi1_devdata *);
1637void hfi1_device_remove(struct hfi1_devdata *);
1638
1639int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
1640 struct kobject *kobj);
1641int hfi1_verbs_register_sysfs(struct hfi1_devdata *);
1642void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *);
1643/* Hook for sysfs read of QSFP */
1644int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
1645
1646int hfi1_pcie_init(struct pci_dev *, const struct pci_device_id *);
1647void hfi1_pcie_cleanup(struct pci_dev *);
1648int hfi1_pcie_ddinit(struct hfi1_devdata *, struct pci_dev *,
1649 const struct pci_device_id *);
1650void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
1651void hfi1_pcie_flr(struct hfi1_devdata *);
1652int pcie_speeds(struct hfi1_devdata *);
1653void request_msix(struct hfi1_devdata *, u32 *, struct hfi1_msix_entry *);
1654void hfi1_enable_intx(struct pci_dev *);
77241056
MM
1655void restore_pci_variables(struct hfi1_devdata *dd);
1656int do_pcie_gen3_transition(struct hfi1_devdata *dd);
1657int parse_platform_config(struct hfi1_devdata *dd);
1658int get_platform_config_field(struct hfi1_devdata *dd,
1659 enum platform_config_table_type_encoding table_type,
1660 int table_index, int field_index, u32 *data, u32 len);
1661
77241056 1662const char *get_unit_name(int unit);
49dbb6cf
DD
1663const char *get_card_name(struct rvt_dev_info *rdi);
1664struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
77241056
MM
1665
1666/*
1667 * Flush write combining store buffers (if present) and perform a write
1668 * barrier.
1669 */
1670static inline void flush_wc(void)
1671{
1672 asm volatile("sfence" : : : "memory");
1673}
1674
1675void handle_eflags(struct hfi1_packet *packet);
1676int process_receive_ib(struct hfi1_packet *packet);
1677int process_receive_bypass(struct hfi1_packet *packet);
1678int process_receive_error(struct hfi1_packet *packet);
1679int kdeth_process_expected(struct hfi1_packet *packet);
1680int kdeth_process_eager(struct hfi1_packet *packet);
1681int process_receive_invalid(struct hfi1_packet *packet);
1682
1683extern rhf_rcv_function_ptr snoop_rhf_rcv_functions[8];
1684
895420dd 1685void update_sge(struct rvt_sge_state *ss, u32 length);
77241056
MM
1686
1687/* global module parameter variables */
1688extern unsigned int hfi1_max_mtu;
1689extern unsigned int hfi1_cu;
1690extern unsigned int user_credit_return_threshold;
2ce6bf22 1691extern int num_user_contexts;
77241056 1692extern unsigned n_krcvqs;
5b55ea3b 1693extern uint krcvqs[];
77241056
MM
1694extern int krcvqsset;
1695extern uint kdeth_qp;
1696extern uint loopback;
1697extern uint quick_linkup;
1698extern uint rcv_intr_timeout;
1699extern uint rcv_intr_count;
1700extern uint rcv_intr_dynamic;
1701extern ushort link_crc_mask;
1702
1703extern struct mutex hfi1_mutex;
1704
1705/* Number of seconds before our card status check... */
1706#define STATUS_TIMEOUT 60
1707
1708#define DRIVER_NAME "hfi1"
1709#define HFI1_USER_MINOR_BASE 0
1710#define HFI1_TRACE_MINOR 127
1711#define HFI1_DIAGPKT_MINOR 128
1712#define HFI1_DIAG_MINOR_BASE 129
1713#define HFI1_SNOOP_CAPTURE_BASE 200
1714#define HFI1_NMINORS 255
1715
1716#define PCI_VENDOR_ID_INTEL 0x8086
1717#define PCI_DEVICE_ID_INTEL0 0x24f0
1718#define PCI_DEVICE_ID_INTEL1 0x24f1
1719
1720#define HFI1_PKT_USER_SC_INTEGRITY \
1721 (SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK \
1722 | SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK \
1723 | SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
1724
1725#define HFI1_PKT_KERNEL_SC_INTEGRITY \
1726 (SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
1727
1728static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
1729 u16 ctxt_type)
1730{
1731 u64 base_sc_integrity =
1732 SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1733 | SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
1734 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1735 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1736 | SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1737 | SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
1738 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1739 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1740 | SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1741 | SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
1742 | SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1743 | SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1744 | SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
1745 | SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
1746 | SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
1747 | SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
1748 | SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1749
1750 if (ctxt_type == SC_USER)
1751 base_sc_integrity |= HFI1_PKT_USER_SC_INTEGRITY;
1752 else
1753 base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
1754
995deafa 1755 if (is_ax(dd))
624be1db 1756 /* turn off send-side job key checks - A0 */
77241056
MM
1757 return base_sc_integrity &
1758 ~SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1759 return base_sc_integrity;
1760}
1761
1762static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
1763{
1764 u64 base_sdma_integrity =
1765 SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
1766 | SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
1767 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
1768 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
1769 | SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
1770 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
1771 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
1772 | SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
1773 | SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
1774 | SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
1775 | SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
1776 | SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
1777 | SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
1778 | SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
1779 | SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
1780 | SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
1781
995deafa 1782 if (is_ax(dd))
624be1db 1783 /* turn off send-side job key checks - A0 */
77241056
MM
1784 return base_sdma_integrity &
1785 ~SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
1786 return base_sdma_integrity;
1787}
1788
1789/*
1790 * hfi1_early_err is used (only!) to print early errors before devdata is
1791 * allocated, or when dd->pcidev may not be valid, and at the tail end of
1792 * cleanup when devdata may have been freed, etc. hfi1_dev_porterr is
1793 * the same as dd_dev_err, but is used when the message really needs
1794 * the IB port# to be definitive as to what's happening..
1795 */
1796#define hfi1_early_err(dev, fmt, ...) \
1797 dev_err(dev, fmt, ##__VA_ARGS__)
1798
1799#define hfi1_early_info(dev, fmt, ...) \
1800 dev_info(dev, fmt, ##__VA_ARGS__)
1801
1802#define dd_dev_emerg(dd, fmt, ...) \
1803 dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
1804 get_unit_name((dd)->unit), ##__VA_ARGS__)
1805#define dd_dev_err(dd, fmt, ...) \
1806 dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1807 get_unit_name((dd)->unit), ##__VA_ARGS__)
1808#define dd_dev_warn(dd, fmt, ...) \
1809 dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
1810 get_unit_name((dd)->unit), ##__VA_ARGS__)
1811
1812#define dd_dev_warn_ratelimited(dd, fmt, ...) \
1813 dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
1814 get_unit_name((dd)->unit), ##__VA_ARGS__)
1815
1816#define dd_dev_info(dd, fmt, ...) \
1817 dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
1818 get_unit_name((dd)->unit), ##__VA_ARGS__)
1819
a1edc18a
IW
1820#define dd_dev_dbg(dd, fmt, ...) \
1821 dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
1822 get_unit_name((dd)->unit), ##__VA_ARGS__)
1823
77241056
MM
1824#define hfi1_dev_porterr(dd, port, fmt, ...) \
1825 dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
1826 get_unit_name((dd)->unit), (dd)->unit, (port), \
1827 ##__VA_ARGS__)
1828
1829/*
1830 * this is used for formatting hw error messages...
1831 */
1832struct hfi1_hwerror_msgs {
1833 u64 mask;
1834 const char *msg;
1835 size_t sz;
1836};
1837
1838/* in intr.c... */
1839void hfi1_format_hwerrors(u64 hwerrs,
1840 const struct hfi1_hwerror_msgs *hwerrmsgs,
1841 size_t nhwerrmsgs, char *msg, size_t lmsg);
1842
1843#define USER_OPCODE_CHECK_VAL 0xC0
1844#define USER_OPCODE_CHECK_MASK 0xC0
1845#define OPCODE_CHECK_VAL_DISABLED 0x0
1846#define OPCODE_CHECK_MASK_DISABLED 0x0
1847
1848static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
1849{
1850 struct hfi1_pportdata *ppd;
1851 int i;
1852
1853 dd->z_int_counter = get_all_cpu_total(dd->int_counter);
1854 dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
1855
1856 ppd = (struct hfi1_pportdata *)(dd + 1);
1857 for (i = 0; i < dd->num_pports; i++, ppd++) {
4eb06882
DD
1858 ppd->ibport_data.rvp.z_rc_acks =
1859 get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
1860 ppd->ibport_data.rvp.z_rc_qacks =
1861 get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
77241056
MM
1862 }
1863}
1864
1865/* Control LED state */
1866static inline void setextled(struct hfi1_devdata *dd, u32 on)
1867{
1868 if (on)
1869 write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
1870 else
1871 write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
1872}
1873
1874int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
1875
1876#endif /* _HFI1_KERNEL_H */
This page took 0.174074 seconds and 5 git commands to generate.