1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
|
/*
* SN Platform GRU Driver
*
* GRU DRIVER TABLES, MACROS, externs, etc
*
* Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#ifndef __GRUTABLES_H__
#define __GRUTABLES_H__
/*
* GRU Chiplet:
* The GRU is a user addressible memory accelerator. It provides
* several forms of load, store, memset, bcopy instructions. In addition, it
* contains special instructions for AMOs, sending messages to message
* queues, etc.
*
* The GRU is an integral part of the node controller. It connects
* directly to the cpu socket. In its current implementation, there are 2
* GRU chiplets in the node controller on each blade (~node).
*
* The entire GRU memory space is fully coherent and cacheable by the cpus.
*
* Each GRU chiplet has a physical memory map that looks like the following:
*
* +-----------------+
* |/////////////////|
* |/////////////////|
* |/////////////////|
* |/////////////////|
* |/////////////////|
* |/////////////////|
* |/////////////////|
* |/////////////////|
* +-----------------+
* | system control |
* +-----------------+ _______ +-------------+
* |/////////////////| / | |
* |/////////////////| / | |
* |/////////////////| / | instructions|
* |/////////////////| / | |
* |/////////////////| / | |
* |/////////////////| / |-------------|
* |/////////////////| / | |
* +-----------------+ | |
* | context 15 | | data |
* +-----------------+ | |
* | ...... | \ | |
* +-----------------+ \____________ +-------------+
* | context 1 |
* +-----------------+
* | context 0 |
* +-----------------+
*
* Each of the "contexts" is a chunk of memory that can be mmaped into user
* space. The context consists of 2 parts:
*
* - an instruction space that can be directly accessed by the user
* to issue GRU instructions and to check instruction status.
*
* - a data area that acts as normal RAM.
*
* User instructions contain virtual addresses of data to be accessed by the
* GRU. The GRU contains a TLB that is used to convert these user virtual
* addresses to physical addresses.
*
* The "system control" area of the GRU chiplet is used by the kernel driver
* to manage user contexts and to perform functions such as TLB dropin and
* purging.
*
* One context may be reserved for the kernel and used for cross-partition
* communication. The GRU will also be used to asynchronously zero out
* large blocks of memory (not currently implemented).
*
*
* Tables:
*
* VDATA-VMA Data - Holds a few parameters. Head of linked list of
* GTS tables for threads using the GSEG
* GTS - Gru Thread State - contains info for managing a GSEG context. A
* GTS is allocated for each thread accessing a
* GSEG.
* GTD - GRU Thread Data - contains shadow copy of GRU data when GSEG is
* not loaded into a GRU
* GMS - GRU Memory Struct - Used to manage TLB shootdowns. Tracks GRUs
* where a GSEG has been loaded. Similar to
* an mm_struct but for GRU.
*
* GS - GRU State - Used to manage the state of a GRU chiplet
* BS - Blade State - Used to manage state of all GRU chiplets
* on a blade
*
*
* Normal task tables for task using GRU.
* - 2 threads in process
* - 2 GSEGs open in process
* - GSEG1 is being used by both threads
* - GSEG2 is used only by thread 2
*
* task -->|
* task ---+---> mm ->------ (notifier) -------+-> gms
* | |
* |--> vma -> vdata ---> gts--->| GSEG1 (thread1)
* | | |
* | +-> gts--->| GSEG1 (thread2)
* | |
* |--> vma -> vdata ---> gts--->| GSEG2 (thread2)
* .
* .
*
* GSEGs are marked DONTCOPY on fork
*
* At open
* file.private_data -> NULL
*
* At mmap,
* vma -> vdata
*
* After gseg reference
* vma -> vdata ->gts
*
* After fork
* parent
* vma -> vdata -> gts
* child
* (vma is not copied)
*
*/
#include <linux/rmap.h>
#include <linux/interrupt.h>
#include <linux/mutex.h>
#include <linux/wait.h>
#include <linux/mmu_notifier.h>
#include "gru.h"
#include "gruhandles.h"
extern struct gru_stats_s gru_stats;
extern struct gru_blade_state *gru_base[];
extern unsigned long gru_start_paddr, gru_end_paddr;
extern unsigned int gru_max_gids;
#define GRU_MAX_BLADES MAX_NUMNODES
#define GRU_MAX_GRUS (GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE)
#define GRU_DRIVER_ID_STR "SGI GRU Device Driver"
#define GRU_DRIVER_VERSION_STR "0.80"
/*
* GRU statistics.
*/
struct gru_stats_s {
atomic_long_t vdata_alloc;
atomic_long_t vdata_free;
atomic_long_t gts_alloc;
atomic_long_t gts_free;
atomic_long_t vdata_double_alloc;
atomic_long_t gts_double_allocate;
atomic_long_t assign_context;
atomic_long_t assign_context_failed;
atomic_long_t free_context;
atomic_long_t load_context;
atomic_long_t unload_context;
atomic_long_t steal_context;
atomic_long_t steal_context_failed;
atomic_long_t nopfn;
atomic_long_t break_cow;
atomic_long_t asid_new;
atomic_long_t asid_next;
atomic_long_t asid_wrap;
atomic_long_t asid_reuse;
atomic_long_t intr;
atomic_long_t intr_mm_lock_failed;
atomic_long_t call_os;
atomic_long_t call_os_offnode_reference;
atomic_long_t call_os_check_for_bug;
atomic_long_t call_os_wait_queue;
atomic_long_t user_flush_tlb;
atomic_long_t user_unload_context;
atomic_long_t user_exception;
atomic_long_t set_task_slice;
atomic_long_t migrate_check;
atomic_long_t migrated_retarget;
atomic_long_t migrated_unload;
atomic_long_t migrated_unload_delay;
atomic_long_t migrated_nopfn_retarget;
atomic_long_t migrated_nopfn_unload;
atomic_long_t tlb_dropin;
atomic_long_t tlb_dropin_fail_no_asid;
atomic_long_t tlb_dropin_fail_upm;
atomic_long_t tlb_dropin_fail_invalid;
atomic_long_t tlb_dropin_fail_range_active;
atomic_long_t tlb_dropin_fail_idle;
atomic_long_t tlb_dropin_fail_fmm;
atomic_long_t tlb_dropin_fail_no_exception;
atomic_long_t tlb_dropin_fail_no_exception_war;
atomic_long_t mmu_invalidate_range;
atomic_long_t mmu_invalidate_page;
atomic_long_t mmu_clear_flush_young;
atomic_long_t flush_tlb;
atomic_long_t flush_tlb_gru;
atomic_long_t flush_tlb_gru_tgh;
atomic_long_t flush_tlb_gru_zero_asid;
atomic_long_t copy_gpa;
atomic_long_t mesq_receive;
atomic_long_t mesq_receive_none;
atomic_long_t mesq_send;
atomic_long_t mesq_send_failed;
atomic_long_t mesq_noop;
atomic_long_t mesq_send_unexpected_error;
atomic_long_t mesq_send_lb_overflow;
atomic_long_t mesq_send_qlimit_reached;
atomic_long_t mesq_send_amo_nacked;
atomic_long_t mesq_send_put_nacked;
atomic_long_t mesq_qf_not_full;
atomic_long_t mesq_qf_locked;
atomic_long_t mesq_qf_noop_not_full;
atomic_long_t mesq_qf_switch_head_failed;
atomic_long_t mesq_qf_unexpected_error;
atomic_long_t mesq_noop_unexpected_error;
atomic_long_t mesq_noop_lb_overflow;
atomic_long_t mesq_noop_qlimit_reached;
atomic_long_t mesq_noop_amo_nacked;
atomic_long_t mesq_noop_put_nacked;
};
enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
cchop_deallocate, tghop_invalidate, mcsop_last};
struct mcs_op_statistic {
atomic_long_t count;
atomic_long_t total;
unsigned long max;
};
extern struct mcs_op_statistic mcs_op_statistics[mcsop_last];
#define OPT_DPRINT 1
#define OPT_STATS 2
#define GRU_QUICKLOOK 4
#define IRQ_GRU 110 /* Starting IRQ number for interrupts */
/* Delay in jiffies between attempts to assign a GRU context */
#define GRU_ASSIGN_DELAY ((HZ * 20) / 1000)
/*
* If a process has it's context stolen, min delay in jiffies before trying to
* steal a context from another process.
*/
#define GRU_STEAL_DELAY ((HZ * 200) / 1000)
#define STAT(id) do { \
if (gru_options & OPT_STATS) \
atomic_long_inc(&gru_stats.id); \
} while (0)
#ifdef CONFIG_SGI_GRU_DEBUG
#define gru_dbg(dev, fmt, x...) \
do { \
if (gru_options & OPT_DPRINT) \
dev_dbg(dev, "%s: " fmt, __func__, x); \
} while (0)
#else
#define gru_dbg(x...)
#endif
/*-----------------------------------------------------------------------------
* ASID management
*/
#define MAX_ASID 0xfffff0
#define MIN_ASID 8
#define ASID_INC 8 /* number of regions */
/* Generate a GRU asid value from a GRU base asid & a virtual address. */
#if defined CONFIG_IA64
#define VADDR_HI_BIT 64
#elif defined CONFIG_X86_64
#define VADDR_HI_BIT 48
#else
#error "Unsupported architecture"
#endif
#define GRUREGION(addr) ((addr) >> (VADDR_HI_BIT - 3) & 3)
#define GRUASID(asid, addr) ((asid) + GRUREGION(addr))
/*------------------------------------------------------------------------------
* File & VMS Tables
*/
struct gru_state;
/*
* This structure is pointed to from the mmstruct via the notifier pointer.
* There is one of these per address space.
*/
struct gru_mm_tracker { /* pack to reduce size */
unsigned int mt_asid_gen:24; /* ASID wrap count */
unsigned int mt_asid:24; /* current base ASID for gru */
unsigned short mt_ctxbitmap:16;/* bitmap of contexts using
asid */
} __attribute__ ((packed));
struct gru_mm_struct {
struct mmu_notifier ms_notifier;
atomic_t ms_refcnt;
spinlock_t ms_asid_lock; /* protects ASID assignment */
atomic_t ms_range_active;/* num range_invals active */
char ms_released;
wait_queue_head_t ms_wait_queue;
DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS);
struct gru_mm_tracker ms_asids[GRU_MAX_GRUS];
};
/*
* One of these structures is allocated when a GSEG is mmaped. The
* structure is pointed to by the vma->vm_private_data field in the vma struct.
*/
struct gru_vma_data {
spinlock_t vd_lock; /* Serialize access to vma */
struct list_head vd_head; /* head of linked list of gts */
long vd_user_options;/* misc user option flags */
int vd_cbr_au_count;
int vd_dsr_au_count;
};
/*
* One of these is allocated for each thread accessing a mmaped GRU. A linked
* list of these structure is hung off the struct gru_vma_data in the mm_struct.
*/
struct gru_thread_state {
struct list_head ts_next; /* list - head at vma-private */
struct mutex ts_ctxlock; /* load/unload CTX lock */
struct mm_struct *ts_mm; /* mm currently mapped to
context */
struct vm_area_struct *ts_vma; /* vma of GRU context */
struct gru_state *ts_gru; /* GRU where the context is
loaded */
struct gru_mm_struct *ts_gms; /* asid & ioproc struct */
unsigned long ts_cbr_map; /* map of allocated CBRs */
unsigned long ts_dsr_map; /* map of allocated DATA
resources */
unsigned long ts_steal_jiffies;/* jiffies when context last
stolen */
long ts_user_options;/* misc user option flags */
pid_t ts_tgid_owner; /* task that is using the
context - for migration */
unsigned short ts_sizeavail; /* Pagesizes in use */
int ts_tsid; /* thread that owns the
structure */
int ts_tlb_int_select;/
|