summaryrefslogtreecommitdiff
path: root/drivers/misc/sgi-gru/grukservices.c
blob: 030769018461b5ddca59e5dc62ab6fd3ab35d97f (plain)
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
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
/*
 * SN Platform GRU Driver
 *
 *              KERNEL SERVICES THAT USE THE GRU
 *
 *  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
 */

#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/spinlock.h>
#include <linux/device.h>
#include <linux/miscdevice.h>
#include <linux/proc_fs.h>
#include <linux/interrupt.h>
#include <linux/uaccess.h>
#include <linux/delay.h>
#include <linux/export.h>
#include <asm/io_apic.h>
#include "gru.h"
#include "grulib.h"
#include "grutables.h"
#include "grukservices.h"
#include "gru_instructions.h"
#include <asm/uv/uv_hub.h>

/*
 * Kernel GRU Usage
 *
 * The following is an interim algorithm for management of kernel GRU
 * resources. This will likely be replaced when we better understand the
 * kernel/user requirements.
 *
 * Blade percpu resources reserved for kernel use. These resources are
 * reserved whenever the the kernel context for the blade is loaded. Note
 * that the kernel context is not guaranteed to be always available. It is
 * loaded on demand & can be stolen by a user if the user demand exceeds the
 * kernel demand. The kernel can always reload the kernel context but
 * a SLEEP may be required!!!.
 *
 * Async Overview:
 *
 * 	Each blade has one "kernel context" that owns GRU kernel resources
 * 	located on the blade. Kernel drivers use GRU resources in this context
 * 	for sending messages, zeroing memory, etc.
 *
 * 	The kernel context is dynamically loaded on demand. If it is not in
 * 	use by the kernel, the kernel context can be unloaded & given to a user.
 * 	The kernel context will be reloaded when needed. This may require that
 * 	a context be stolen from a user.
 * 		NOTE: frequent unloading/reloading of the kernel context is
 * 		expensive. We are depending on batch schedulers, cpusets, sane
 * 		drivers or some other mechanism to prevent the need for frequent
 *	 	stealing/reloading.
 *
 * 	The kernel context consists of two parts:
 * 		- 1 CB & a few DSRs that are reserved for each cpu on the blade.
 * 		  Each cpu has it's own private resources & does not share them
 * 		  with other cpus. These resources are used serially, ie,
 * 		  locked, used & unlocked  on each call to a function in
 * 		  grukservices.
 * 		  	(Now that we have dynamic loading of kernel contexts, I
 * 		  	 may rethink this & allow sharing between cpus....)
 *
 *		- Additional resources can be reserved long term & used directly
 *		  by UV drivers located in the kernel. Drivers using these GRU
 *		  resources can use asynchronous GRU instructions that send
 *		  interrupts on completion.
 *		  	- these resources must be explicitly locked/unlocked
 *		  	- locked resources prevent (obviously) the kernel
 *		  	  context from being unloaded.
 *			- drivers using these resource directly issue their own
 *			  GRU instruction and must wait/check completion.
 *
 * 		  When these resources are reserved, the caller can optionally
 * 		  associate a wait_queue with the resources and use asynchronous
 * 		  GRU instructions. When an async GRU instruction completes, the
 * 		  driver will do a wakeup on the event.
 *
 */


#define ASYNC_HAN_TO_BID(h)	((h) - 1)
#define ASYNC_BID_TO_HAN(b)	((b) + 1)
#define ASYNC_HAN_TO_BS(h)	gru_base[ASYNC_HAN_TO_BID(h)]

#define GRU_NUM_KERNEL_CBR	1
#define GRU_NUM_KERNEL_DSR_BYTES 256
#define GRU_NUM_KERNEL_DSR_CL	(GRU_NUM_KERNEL_DSR_BYTES /		\
					GRU_CACHE_LINE_BYTES)

/* GRU instruction attributes for all instructions */
#define IMA			IMA_CB_DELAY

/* GRU cacheline size is always 64 bytes - even on arches with 128 byte lines */
#define __gru_cacheline_aligned__                               \
	__attribute__((__aligned__(GRU_CACHE_LINE_BYTES)))

#define MAGIC	0x1234567887654321UL

/* Default retry count for GRU errors on kernel instructions */
#define EXCEPTION_RETRY_LIMIT	3

/* Status of message queue sections */
#define MQS_EMPTY		0
#define MQS_FULL		1
#define MQS_NOOP		2

/*----------------- RESOURCE MANAGEMENT -------------------------------------*/
/* optimized for x86_64 */
struct message_queue {
	union gru_mesqhead	head __gru_cacheline_aligned__;	/* CL 0 */
	int			qlines;				/* DW 1 */
	long 			hstatus[2];
	void 			*next __gru_cacheline_aligned__;/* CL 1 */
	void 			*limit;
	void 			*start;
	void 			*start2;
	char			data ____cacheline_aligned;	/* CL 2 */
};

/* First word in every message - used by mesq interface */
struct message_header {
	char	present;
	char	present2;
	char 	lines;
	char	fill;
};

#define HSTATUS(mq, h)	((mq) + offsetof(struct message_queue, hstatus[h]))

/*
 * Reload the blade's kernel context into a GRU chiplet. Called holding
 * the bs_kgts_sema for READ. Will steal user contexts if necessary.
 */
static void gru_load_kernel_context(struct gru_blade_state *bs, int blade_id)
{
	struct gru_state *gru;
	struct gru_thread_state *kgts;
	void *vaddr;
	int ctxnum, ncpus;

	up_read(&bs->bs_kgts_sema);
	down_write(&bs->bs_kgts_sema);

	if (!bs->bs_kgts) {
		do {
			bs->bs_kgts = gru_alloc_gts(NULL, 0, 0, 0, 0, 0);
			if (!IS_ERR(bs->bs_kgts))
				break;
			msleep(1);
		} while (true);
		bs->bs_kgts->ts_user_blade_id = blade_id;
	}
	kgts = bs->bs_kgts;

	if (!kgts->ts_gru) {
		STAT(load_kernel_context);
		ncpus = uv_blade_nr_possible_cpus(blade_id);
		kgts->ts_cbr_au_count = GRU_CB_COUNT_TO_AU(
			GRU_NUM_KERNEL_CBR * ncpus + bs->bs_async_cbrs);
		kgts->ts_dsr_au_count = GRU_DS_BYTES_TO_AU(
			GRU_NUM_KERNEL_DSR_BYTES * ncpus +
				bs->bs_async_dsr_bytes);
		while (!gru_assign_gru_context(kgts)) {
			msleep(1);
			gru_steal_context(kgts);
		}
		gru_load_context(kgts);
		gru = bs->bs_kgts->ts_gru;
		vaddr = gru->gs_gru_base_vaddr;
		ctxnum = kgts->ts_ctxnum;
		bs->kernel_cb = get_gseg_base_address_cb(vaddr, ctxnum, 0);
		bs->kernel_dsr = get_gseg_base_address_ds(vaddr, ctxnum, 0);
	}
	downgrade_write(&bs->bs_kgts_sema);
}

/*
 * Free all kernel contexts that are not currently in use.
 *   Returns 0 if all freed, else number of inuse context.
 */
static int gru_free_kernel_contexts(void)
{
	struct gru_blade_state *bs;
	struct gru_thread_state *kgts;
	int bid, ret = 0;

	for (bid = 0; bid < GRU_MAX_BLADES; bid++) {
		bs = gru_base[bid];
		if (!bs)
			continue;

		/* Ignore busy contexts. Don't want to block here.  */
		if (down_write_trylock(&bs->bs_kgts_sema)) {
			kgts = bs->bs_kgts;
			if (kgts && kgts->ts_gru)
				gru_unload_context(kgts, 0);
			bs->bs_kgts = NULL;
			up_write(&bs->bs_kgts_sema);
			kfree(kgts);
		} else {
			ret++;
		}
	}
	return ret;
}

/*
 * Lock & load the kernel context for the specified blade.
 */
static struct gru_blade_state *gru_lock_kernel_context(int blade_id)
{
	struct gru_blade_state *bs;
	int bid;

	STAT(lock_kernel_context);
again:
	bid = blade_id < 0 ? uv_numa_blade_id() : blade_id;
	bs = gru_base[bid];

	/* Handle the case where migration occurred while waiting for the sema */
	down_read(&bs->bs_kgts_sema);
	if (blade_id < 0 && bid != uv_numa_blade_id()) {
		up_read(&bs->bs_kgts_sema);
		goto again;
	}
	if (!bs->bs_kgts || !bs->bs_kgts->ts_gru)
		gru_load_kernel_context(bs, bid);
	return bs;

}

/*
 * Unlock the kernel context for the specified blade. Context is not
 * unloaded but may be stolen before next use.
 */
static void gru_unlock_kernel_context(int blade_id)
{
	struct gru_blade_state *bs;

	bs = gru_base[blade_id];
	up_read(&bs->bs_kgts_sema);
	STAT(unlock_kernel_context);
}

/*
 * Reserve & get pointers to the DSR/CBRs reserved for the current cpu.
 * 	- returns with preemption disabled
 */
static int gru_get_cpu_resources(int dsr_bytes, void **cb, void **dsr)
{
	struct gru_blade_state *bs;
	int lcpu;

	BUG_ON(dsr_bytes > GRU_NUM_KERNEL_DSR_BYTES);
	preempt_disable();
	bs = gru_lock_kernel_context(-1);
	lcpu = uv_blade_processor_id();
	*cb = bs->kernel_cb + lcpu * GRU_HANDLE_STRIDE;
	*dsr = bs->kernel_dsr + lcpu * GRU_NUM_KERNEL_DSR_BYTES;
	return 0;
}

/*
 * Free the current cpus reserved DSR/CBR resources.
 */
static void gru_free_cpu_resources(void *cb, void *dsr)
{
	gru_unlock_kernel_context(uv_numa_blade_id());
	preempt_enable();
}

/*
 * Reserve GRU resources to be used asynchronously.
 *   Note: currently supports only 1 reservation per blade.
 *
 * 	input:
 * 		blade_id  - blade on which resources should be reserved
 * 		cbrs	  - number of CBRs
 * 		dsr_bytes - number of DSR bytes needed
 *	output:
 *		handle to identify resource
 *		(0 = async resources already reserved)
 */
unsigned long gru_reserve_async_resources(int blade_id, int cbrs, int dsr_bytes,
			struct completion *cmp)
{
	struct gru_blade_state *bs;
	struct gru_thread_state *kgts;
	int ret = 0;

	bs = gru_base[blade_id];

	down_write(&bs->bs_kgts_sema);

	/* Verify no resources already reserved */
	if (bs->bs_async_dsr_bytes + bs->bs_async_cbrs)
		goto done;
	bs->bs_async_dsr_bytes = dsr_bytes;
	bs->bs_async_cbrs = cbrs;
	bs->bs_async_wq = cmp;
	kgts = bs->bs_kgts;

	/* Resources changed. Unload context if already loaded */
	if (kgts && kgts->ts_gru)
		gru_unload_context(kgts, 0);
	ret = ASYNC_BID_TO_HAN(blade_id);

done:
	up_write(&bs->bs_kgts_sema);
	return ret;
}

/*
 * Release async resources previously reserved.
 *
 *	input:
 *		han - handle to identify resources
 */
void gru_release_async_resources(unsigned long han)
{
	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);

	down_write(&bs->bs_kgts_sema);
	bs->bs_async_dsr_bytes = 0;
	bs->bs_async_cbrs = 0;
	bs->bs_async_wq = NULL;
	up_write(&bs->bs_kgts_sema);
}

/*
 * Wait for async GRU instructions to complete.
 *
 *	input:
 *		han - handle to identify resources
 */
void gru_wait_async_cbr(unsigned long han)
{
	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);

	wait_for_completion(bs->bs_async_wq);
	mb();
}

/*
 * Lock previous reserved async GRU resources
 *
 *	input:
 *		han - handle to identify resources
 *	output:
 *		cb  - pointer to first CBR
 *		dsr - pointer to first DSR
 */
void gru_lock_async_resource(unsigned long han,  void **cb, void **dsr)
{
	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);
	int blade_id = ASYNC_HAN_TO_BID(han);
	int ncpus;

	gru_lock_kernel_context(blade_id);
	ncpus = uv_blade_nr_possible_cpus(blade_id);
	if (cb)
		*cb = bs->kernel_cb + ncpus * GRU_HANDLE_STRIDE;
	if (dsr)
		*dsr = bs->kernel_dsr + ncpus * GRU_NUM_KERNEL_DSR_BYTES;
}

/*
 * Unlock previous reserved async GRU resources
 *
 *	input:
 *		han - handle to identify resources
 */
void gru_unlock_async_resource(unsigned long han)
{
	int blade_id = ASYNC_HAN_TO_BID(han);

	gru_unlock_kernel_context(blade_id);
}

/*----------------------------------------------------------------------*/
int gru_get_cb_exception_detail(void *cb,
		struct control_block_extended_exc_detail *excdet)
{
	struct gru_control_block_extended *cbe;
	struct gru_thread_state *kgts = NULL;
	unsigned long off;
	int cbrnum, bid;

	/*
	 * Locate kgts for cb. This algorithm is SLOW but
	 * this function is rarely called (ie., almost never).
	 * Performance does not matter.
	 */
	for_each_possible_blade(bid) {
		if (!gru_base[bid])
			break;
		kgts = gru_base[bid]->bs_kgts;
		if (!kgts || !kgts->ts_gru)
			continue;
		off = cb - kgts->ts_gru->gs_gru_base_vaddr;
		if (off < GRU_SIZE)
			break;
		kgts = NULL;
	}
	BUG_ON(!kgts);
	cbrnum = thread_cbr_number(kgts, get_cb_number(cb));
	cbe = get_cbe(GRUBASE(cb), cbrnum);
	gru_flush_cache(cbe);	/* CBE not coherent */
	sync_core();
	excdet->opc = cbe->opccpy;
	excdet->exopc = cbe->exopccpy;
	excdet->ecause = cbe->ecause;
	excdet->exceptdet0 = cbe->idef1upd;
	excdet->exceptdet1 = cbe->idef3upd;
	gru_flush_cache(cbe);
	return 0;
}

static char *gru_get_cb_exception_detail_str(int ret, void *cb,
					     char *buf, int size)
{
	struct gru_control_block_status *gen = (void *)cb;
	struct control_block_extended_exc_detail excdet;

	if (ret > 0 && gen->istatus == CBS_EXCEPTION) {
		gru_get_cb_exception_detail(cb, &excdet);
		snprintf(buf, size,
			"GRU:%d exception: cb %p, opc %d, exopc %d, ecause 0x%x,"
			"excdet0 0x%lx, excdet1 0x%x", smp_processor_id(),
			gen, excdet.opc, excdet.exopc, excdet.ecause,
			excdet.exceptdet0, excdet.exceptdet1);
	} else {
		snprintf(buf, size, "No exception");
	}
	return buf;
}

static int gru_wait_idle_or_exception(struct gru_control_block_status *gen)
{
	while (gen->istatus >= CBS_ACTIVE) {
		cpu_relax();
		barrier();
	}
	return gen->istatus;
}

static int gru_retry_exception(void *cb)
{
	struct gru_control_block_status *gen = (void *)cb;
	struct control_block_extended_exc_detail excdet;
	int retry = EXCEPTION_RETRY_LIMIT;

	while (1)  {
		if (gru_wait_idle_or_exception(gen) == CBS_IDLE)
			return CBS_IDLE;
		if (gru_get_cb_message_queue_substatus(cb))
			return CBS_EXCEPTION;
		gru_get_cb_exception_detail(cb, &excdet);
		if ((excdet.ecause & ~EXCEPTION_RETRY_BITS) ||
				(excdet.cbrexecstatus & CBR_EXS_ABORT_OCC))
			break;
		if (retry-- == 0)
			break;
		gen->icmd = 1;
		gru_flush_cache(gen);
	}
	return CBS_EXCEPTION;
}

int gru_check_status_proc(void *cb)
{
	struct gru_control_block_status *gen = (void *)cb;
	int ret;

	ret = gen->istatus;
	if (ret == CBS_EXCEPTION)
		ret = gru_retry_exception(cb);
	rmb();
	return ret;

}

int gru_wait_proc(void *cb)
{
	struct gru_control_block_status *gen = (void *)cb;
	int ret;

	ret = gru_wait_idle_or_exception(gen);
	if (ret == CBS_EXCEPTION)
		ret = gru_retry_exception(cb);
	rmb();
	return ret;
}

static void gru_abort(int ret, void *cb, char *str)
{
	char buf[GRU_EXC_STR_SIZE];

	panic("GRU FATAL ERROR: %s - %s\n", str,
	      gru_get_cb_exception_detail_str(ret, cb, buf, sizeof(buf)));
}

void gru_wait_abort_proc(void *cb)
{
	int ret;

	ret = gru_wait_proc(cb);
	if (ret)
		gru_abort(ret, cb, "gru_wait_abort");
}


/*------------------------------ MESSAGE QUEUES -----------------------------*/

/* Internal status . These are NOT returned to the user. */
#define MQIE_AGAIN		-1	/* try again */


/*
 * Save/restore the "present" flag that is in the second line of 2-line
 * messages
 */
static inline int get_present2(void *p)
{
	struct message_header *mhdr = p + GRU_CACHE_LINE_BYTES;
	return mhdr->present;
}

static inline void restore_present2(void *p, int val)
{
	struct message_header *mhdr = p + GRU_CACHE_LINE_BYTES;
	mhdr->present = val;
}

/*
 * Create a message queue.
 * 	qlines - message queue size in cache lines. Includes 2-line header.
 */
int gru_create_message_queue(struct gru_message_queue_desc *mqd,
		void *p, unsigned int bytes, int nasid, int vector, int apicid)
{
	struct message_queue *mq = p;
	unsigned int qlines;

	qlines = bytes / GRU_CACHE_LINE_BYTES - 2;
	memset(mq, 0, bytes);
	mq->start = &mq->data;
	mq->start2 = &mq->data + (qlines / 2 - 1) * GRU_CACHE_LINE_BYTES;
	mq->next = &mq->data;
	mq->limit = &mq->data + (qlines - 2) * GRU_CACHE_LINE_BYTES;
	mq->qlines = qlines;
	mq->hstatus[0] = 0;
	mq->hstatus[1] = 1;
	mq->head = gru_mesq_head(2, qlines / 2 + 1);
	mqd->mq = mq;
	mqd->mq_gpa = uv_gpa(mq);
	mqd->qlines = qlines;
	mqd->interrupt_pnode = nasid >> 1;
	mqd->interrupt_vector = vector;
	mqd->interrupt_apicid = apicid;
	return 0;
}
EXPORT_SYMBOL_GPL(gru_create_message_queue);

/*
 * Send a NOOP message to a message queue
 * 	Returns:
 * 		 0 - if queue is full after the send. This is the normal case
 * 		     but various races can change this.
 *		-1 - if mesq sent successfully but queue not full
 *		>0 - unexpected error. MQE_xxx returned
 */
static int send_noop_message(void *cb, struct gru_message_queue_desc *mqd,
				void *mesg)
{
	const struct message_header noop_header = {
					.present = MQS_NOOP, .lines = 1};
	unsigned long m;
	int substatus, ret;
	struct message_header save_mhdr, *mhdr = mesg;

	STAT(mesq_noop);
	save_mhdr = *mhdr;
	*mhdr = noop_header;
	gru_mesq(cb, mqd->mq_gpa, gru_get_tri(mhdr), 1, IMA);
	ret = gru_wait(cb);

	if (ret) {
		substatus = gru_get_cb_message_queue_substatus(cb);
		switch (substatus) {
		case CBSS_NO_ERROR:
			STAT(mesq_noop_unexpected_error);
			ret = MQE_UNEXPECTED_CB_ERR;
			break;
		case CBSS_LB_OVERFLOWED:
			STAT(mesq_noop_lb_overflow);
			ret = MQE_CONGESTION;
			break;
		case CBSS_QLIMIT_REACHED:
			STAT(mesq_noop_qlimit_reached);
			ret = 0;
			break;
		case CBSS_AMO_NACKED:
			STAT(mesq_noop_amo_nacked);
			ret = MQE_CONGESTION;
			break;
		case CBSS_PUT_NACKED:
			STAT(mesq_noop_put_nacked);
			m = mqd->mq_gpa + (gru_get_amo_value_head(cb) << 6);
			gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, 1, 1,
						IMA);
			if (gru_wait(cb) == CBS_IDLE)
				ret = MQIE_AGAIN;
			else
				ret = MQE_UNEXPECTED_CB_ERR;
			break;
		case CBSS_PAGE_OVERFLOW:
			STAT(mesq_noop_page_overflow);
			/* fallthru */
		default:
			BUG();
		}
	}
	*mhdr = save_mhdr;
	return ret;
}

/*
 * Handle a gru_mesq full.
 */
static int send_message_queue_full(void *cb, struct gru_message_queue_desc *mqd,
				void *mesg, int lines)
{
	union gru_mesqhead mqh;
	unsigned int limit, head;
	unsigned long avalue;
	int half, qlines;

	/* Determine if switching to first/second half of q */
	avalue = gru_get_amo_value(cb);
	head = gru_get_amo_value_head(cb);
	limit = gru_get_amo_value_limit(cb);

	qlines = mqd->qlines;
	half = (limit != qlines);

	if (half)
		mqh = gru_mesq_head(qlines / 2 + 1, qlines);
	else
		mqh = gru_mesq_head(2, qlines / 2 + 1);

	/* Try to get lock for switching head pointer */
	gru_gamir(cb, EOP_IR_CLR, HSTATUS(mqd->mq_gpa, half), XTYPE_DW, IMA);
	if (gru_wait(cb) != CBS_IDLE)
		goto cberr;
	if (!gru_get_amo_value(cb)) {
		STAT(mesq_qf_locked);
		return MQE_QUEUE_FULL;
	}

	/* Got the lock. Send optional NOP if queue not full, */
	if (head != limit) {
		if (send_noop_message(cb, mqd, mesg)) {
			gru_gamir(cb, EOP_IR_INC, HSTATUS(mqd->mq_gpa, half),
					XTYPE_DW, IMA);
			if (gru_wait(cb) != CBS_IDLE)
				goto cberr;
			STAT(mesq_qf_noop_not_full);
			return MQIE_AGAIN;
		}
		avalue++;
	}

	/* Then flip queuehead to other half of queue. */
	gru_gamer(cb, EOP_ERR_CSWAP, mqd->mq_gpa, XTYPE_DW, mqh.val, avalue,
							IMA);
	if (gru_wait(cb) != CBS_IDLE)
		goto cberr;

	/* If not successfully in swapping queue head, clear the hstatus lock */
	if (gru_get_amo_value(cb) != avalue) {
		STAT(mesq_qf_switch_head_failed);
		gru_gamir(cb, EOP_IR_INC, HSTATUS(mqd->mq_gpa, half), XTYPE_DW,
							IMA);
		if (gru_wait(cb) != CBS_IDLE)
			goto cberr;
	}
	return MQIE_AGAIN;
cberr:
	STAT(mesq_qf_unexpected_error);
	return MQE_UNEXPECTED_CB_ERR;
}

/*
 * Handle a PUT failure. Note: if message was a 2-line message, one of the
 * lines might have successfully have been written. Before sending the
 * message, "present" must be cleared in BOTH lines to prevent the receiver
 * from prematurely seeing the full message.
 */
static int send_message_put_nacked(void *cb, struct gru_message_queue_desc *mqd,
			void *mesg, int lines)
{
	unsigned long m;
	int ret, loops = 200;	/* experimentally determined */

	m = mqd->mq_gpa + (gru_get_amo_value_head(cb) << 6);
	if (lines == 2) {
		gru_vset(cb, m, 0, XTYPE_CL, lines, 1, IMA);
		if (gru_wait(cb) != CBS_IDLE)
			return MQE_UNEXPECTED_CB_ERR;
	}
	gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, lines, 1, IMA);
	if (gru_wait(cb) != CBS_IDLE)
		return MQE_UNEXPECTED_CB_ERR;

	if (!mqd->interrupt_vector)
		return MQE_OK;

	/*
	 * Send a noop message in order to deliver a cross-partition interrupt
	 * to the SSI that contains the target message queue. Normally, the
	 * interrupt is automatically delivered by hardware following mesq
	 * operations, but some error conditions require explicit delivery.
	 * The noop message will trigger delivery. Otherwise partition failures
	 * could cause unrecovered errors.
	 */
	do {
		ret = send_noop_message(cb, mqd, mesg);
	} while ((ret == MQIE_AGAIN || ret == MQE_CONGESTION) && (loops-- > 0));

	if (ret == MQIE_AGAIN || ret == MQE_CONGESTION) {
		/*
		 * Don't indicate to the app to resend the message, as it's
		 * already been successfully sent.  We simply send an OK
		 * (rather than fail the send with MQE_UNEXPECTED_CB_ERR),
		 * assuming that the other side is receiving enough
		 * interrupts to get this message processed anyway.
		 */
		ret = MQE_OK;
	}
	return ret;
}

/*
 * Handle a gru_mesq failure. Some of these failures are software recoverable
 * or retryable.
 */
static int send_message_failure(void *cb, struct gru_message_queue_desc *mqd,
				void *mesg, int lines)
{
	int substatus, ret = 0;

	substatus = gru_get_cb_message_queue_substatus(cb);
	switch (substatus) {
	case CBSS_NO_ERROR:
		STAT(mesq_send_unexpected_error);
		ret = MQE_UNEXPECTED_CB_ERR;
		break;
	case CBSS_LB_OVERFLOWED:
		STAT(mesq_send_lb_overflow);
		ret = MQE_CONGESTION;
		break;
	case CBSS_QLIMIT_REACHED:
		STAT(mesq_send_qlimit_reached);
		ret = send_message_queue_full(cb, mqd, mesg, lines);
		break;
	case CBSS_AMO_NACKED:
		STAT(mesq_send_amo_nacked);
		ret = MQE_CONGESTION;
		break;
	case CBSS_PUT_NACKED:
		STAT(mesq_send_put_nacked);
		ret = send_message_put_nacked(cb, mqd, mesg, lines);
		break;
	case CBSS_PAGE_OVERFLOW:
		STAT(mesq_page_overflow);
		/* fallthru */
	default:
		BUG();
	}
	return ret;
}

/*
 * Send a message to a message queue
 * 	mqd	message queue descriptor
 * 	mesg	message. ust be vaddr within a GSEG
 * 	bytes	message size (<= 2 CL)
 */
int gru_send_message_gpa(struct gru_message_queue_desc *mqd, void *mesg,
				unsigned int bytes)
{
	struct message_header *mhdr;
	void *cb;
	void *dsr;
	int istatus, clines, ret;

	STAT(mesq_send);
	BUG_ON(bytes < sizeof(int) || bytes > 2 * GRU_CACHE_LINE_BYTES);

	clines = DIV_ROUND_UP(bytes, GRU_CACHE_LINE_BYTES);
	if (gru_get_cpu_resources(bytes, &cb, &dsr))
		return MQE_BUG_NO_RESOURCES;
	memcpy(dsr, mesg, bytes);
	mhdr = dsr;
	mhdr->present = MQS_FULL;
	mhdr->lines = clines;
	if (clines == 2) {
		mhdr->present2 = get_present2(mhdr);
		restore_present2(mhdr, MQS_FULL);
	}

	do {
		ret = MQE_OK;
		gru_mesq(cb, mqd->mq_gpa, gru_get_tri(mhdr), clines, IMA);
		istatus = gru_wait(cb);
		if (istatus != CBS_IDLE)
			ret = send_message_failure(cb, mqd, dsr, clines);
	} while (ret == MQIE_AGAIN);
	gru_free_cpu_resources(cb, dsr);

	if (ret)
		STAT(mesq_send_failed);
	return ret;
}
EXPORT_SYMBOL_GPL(gru_send_message_gpa);

/*
 * Advance the receive pointer for the queue to the next message.
 */
void gru_free_message(struct gru_message_queue_desc *mqd, void *mesg)
{
	struct message_queue *mq = mqd->mq;
	struct message_header *mhdr = mq->next;
	void *next, *pnext;
	int half = -1;
	int lines = mhdr->lines;

	if (lines == 2)
		restore_present2(mhdr, MQS_EMPTY);
	mhdr->present = MQS_EMPTY;

	pnext = mq->next;
	next = pnext + GRU_CACHE_LINE_BYTES * lines;
	if (next == mq->limit) {
		next = mq->start;
		half = 1;
	} else if (pnext < mq->start2 && next >= mq->start2) {
		half = 0;
	}

	if (half >= 0)
		mq->hstatus[half] = 1;
	mq->next = next;
}
EXPORT_SYMBOL_GPL(gru_free_message);

/*
 * Get next message from message queue. Return NULL if no message
 * present. User must call next_message() to move to next message.
 * 	rmq	message queue
 */
void *gru_get_next_message(struct gru_message_queue_desc *mqd)
{
	struct message_queue *mq = mqd->mq;
	struct message_header *mhdr = mq->next;
	int present = mhdr->present;

	/* skip NOOP messages */
	while (present == MQS_NOOP) {
		gru_free_message(mqd, mhdr);
		mhdr = mq->next;
		present = mhdr->present;
	}

	/* Wait for both halves of 2 line messages */
	if (present == MQS_FULL && mhdr->lines == 2 &&
				get_present2(mhdr) == MQS_EMPTY)
		present = MQS_EMPTY;

	if (!present) {
		STAT(mesq_receive_none);
		return NULL;
	}

	if (mhdr->lines == 2)
		restore_present2(mhdr, mhdr->present2);

	STAT(mesq_receive);
	return mhdr;
}
EXPORT_SYMBOL_GPL(gru_get_next_message);

/* ---------------------- GRU DATA COPY FUNCTIONS ---------------------------*/

/*
 * Load a DW from a global GPA. The GPA can be a memory or MMR address.
 */
int gru_read_gpa(unsigned long *value, unsigned long gpa)
{
	void *cb;
	void *dsr;
	int ret, iaa;

	STAT(read_gpa);
	if (gru_get_cpu_resources(GRU_NUM_KERNEL_DSR_BYTES, &cb, &dsr))
		return MQE_BUG_NO_RESOURCES;
	iaa = gpa >> 62;
	gru_vload_phys(cb, gpa, gru_get_tri(dsr), iaa, IMA);
	ret = gru_wait(cb);
	if (ret == CBS_IDLE)
		*value = *(unsigned long *)dsr;
	gru_free_cpu_resources(cb, dsr);
	return ret;
}
EXPORT_SYMBOL_GPL(gru_read_gpa);


/*
 * Copy a block of data using the GRU resources
 */
int gru_copy_gpa(unsigned long dest_gpa, unsigned long src_gpa,
				unsigned int bytes)
{
	void *cb;
	void *dsr;
	int ret;

	STAT(copy_gpa);
	if (gru_get_cpu_resources(GRU_NUM_KERNEL_DSR_BYTES, &cb, &dsr))
		return MQE_BUG_NO_RESOURCES;
	gru_bcopy(cb, src_gpa, dest_gpa, gru_get_tri(dsr),
		  XTYPE_B, bytes, GRU_NUM_KERNEL_DSR_CL, IMA);
	ret = gru_wait(cb);
	gru_free_cpu_resources(cb, dsr);
	return ret;
}
EXPORT_SYMBOL_GPL(gru_copy_gpa);

/* ------------------- KERNEL QUICKTESTS RUN AT STARTUP ----------------*/
/* 	Temp - will delete after we gain confidence in the GRU		*/

static int quicktest0(unsigned long arg)
{
	unsigned long word0;
	unsigned long word1;
	void *cb;
	void *dsr;
	unsigned long *p;
	int ret = -EIO;

	if (gru_get_cpu_resources(GRU_CACHE_LINE_BYTES, &cb, &dsr))
		return MQE_BUG_NO_RESOURCES;
	p = dsr;
	word0 = MAGIC;
	word1 = 0;

	gru_vload(cb, uv_gpa(&word0), gru_get_tri(dsr), XTYPE_DW, 1, 1, IMA);
	if (gru_wait(cb) != CBS_IDLE) {
		printk(KERN_DEBUG "GRU:%d quicktest0: CBR failure 1\n", smp_processor_id());
		goto done;
	}

	if (*p != MAGIC) {
		printk(KERN_DEBUG "GRU:%d quicktest0 bad magic 0x%lx\n", smp_processor_id(), *p);
		goto done;
	}
	gru_vstore(cb, uv_gpa(&word1), gru_get_tri(dsr), XTYPE_DW, 1, 1, IMA);
	if (gru_wait(cb) != CBS_IDLE) {
		printk(KERN_DEBUG "GRU:%d quicktest0: CBR failure 2\n", smp_processor_id());
		goto done;
	}

	if (word0 != word1 || word1 != MAGIC) {
		printk(KERN_DEBUG
		       "GRU:%d quicktest0 err: found 0x%lx, expected 0x%lx\n",
		     smp_processor_id(), word1, MAGIC);
		goto done;
	}
	ret = 0;

done:
	gru_free_cpu_resources(cb, dsr);
	return ret;
}

#define ALIGNUP(p, q)	((void *)(((unsigned long)(p) + (q) - 1) & ~(q - 1)))

static int quicktest1(unsigned long arg)
{
	struct gru_message_queue_desc mqd;
	void *p, *mq;
	int i, ret = -EIO;
	char mes[GRU_CACHE_LINE_BYTES], *m;

	/* Need  1K cacheline aligned that does not cross page boundary */
	p = kmalloc(4096, 0);
	if (p == NULL)
		return -ENOMEM;
	mq = ALIGNUP(p, 1024);
	memset(mes, 0xee, sizeof(mes));

	gru_create_message_queue(&mqd, mq, 8 * GRU_CACHE_LINE_BYTES, 0, 0, 0);
	for (i = 0; i < 6; i++) {
		mes[8] = i;
		do {
			ret = gru_send_message_gpa(&mqd, mes, sizeof(mes));
		} while (ret == MQE_CONGESTION);
		if (ret)
			break;
	}
	if (ret != MQE_QUEUE_FULL || i != 4) {
		printk(KERN_DEBUG "GRU:%d quicktest1: unexpect status %d, i %d\n",
		       smp_processor_id(), ret, i);
		goto done;
	}

	for (i = 0; i < 6; i++) {
		m = gru_get_next_message(&mqd);
		if (!m || m[8] != i)
			break;
		gru_free_message(&mqd, m);
	}
	if (i != 4) {
		printk(KERN_DEBUG "GRU:%d quicktest2: bad message, i %d, m %p, m8 %d\n",
			smp_processor_id(), i, m, m ? m[8] : -1);
		goto done;
	}
	ret = 0;

done:
	kfree(p);
	return ret;
}

static int quicktest2(unsigned long arg)
{
	static DECLARE_COMPLETION(cmp);
	unsigned long han;
	int blade_id = 0;
	int numcb = 4;
	int ret = 0;
	unsigned long *buf;
	void *cb0, *cb;
	struct gru_control_block_status *gen;
	int i, k, istatus, bytes;

	bytes = numcb * 4 * 8;
	buf = kmalloc(bytes, GFP_KERNEL);
	if (!buf)
		return -ENOMEM;

	ret = -EBUSY;
	han = gru_reserve_async_resources(blade_id, numcb, 0, &cmp);
	if (!han)
		goto done;

	gru_lock_async_resource(han, &cb0, NULL);
	memset(buf, 0xee, bytes);
	for (i = 0; i < numcb; i++)
		gru_vset(cb0 + i * GRU_HANDLE_STRIDE, uv_gpa(&buf[i * 4]), 0,
				XTYPE_DW, 4, 1, IMA_INTERRUPT);

	ret = 0;
	k = numcb;
	do {
		gru_wait_async_cbr(han);
		for (i = 0; i < numcb; i++) {
			cb = cb0 + i * GRU_HANDLE_STRIDE;
			istatus = gru_check_status(cb);
			if (istatus != CBS_ACTIVE && istatus != CBS_CALL_OS)
				break;
		}
		if (i == numcb)
			continue;
		if (istatus != CBS_IDLE) {
			printk(KERN_DEBUG "GRU:%d quicktest2: cb %d, exception\n", smp_processor_id(), i);
			ret = -EFAULT;
		} else if (buf[4 * i] || buf[4 * i + 1] || buf[4 * i + 2] ||
				buf[4 * i + 3]) {
			printk(KERN_DEBUG "GRU:%d quicktest2:cb %d,  buf 0x%lx, 0x%lx, 0x%lx, 0x%lx\n",
			       smp_processor_id(), i, buf[4 * i], buf[4 * i + 1], buf[4 * i + 2], buf[4 * i + 3]);
			ret = -EIO;
		}
		k--;
		gen = cb;
		gen->istatus = CBS_CALL_OS; /* don't handle this CBR again */
	} while (k);
	BUG_ON(cmp.done);

	gru_unlock_async_resource(han);
	gru_release_async_resources(han);
done:
	kfree(buf);
	return ret;
}

#define BUFSIZE 200
static int quicktest3(unsigned long arg)
{
	char buf1[BUFSIZE], buf2[BUFSIZE];
	int ret = 0;

	memset(buf2, 0, sizeof(buf2));
	memset(buf1, get_cycles() & 255, sizeof(buf1));
	gru_copy_gpa(uv_gpa(buf2), uv_gpa(buf1), BUFSIZE);
	if (memcmp(buf1, buf2, BUFSIZE)) {
		printk(KERN_DEBUG "GRU:%d quicktest3 error\n", smp_processor_id());
		ret = -EIO;
	}
	return ret;
}

/*
 * Debugging only. User hook for various kernel tests
 * of driver & gru.
 */
int gru_ktest(unsigned long arg)
{
	int ret = -EINVAL;

	switch (arg & 0xff) {
	case 0:
		ret = quicktest0(arg);
		break;
	case 1:
		ret = quicktest1(arg);
		break;
	case 2:
		ret = quicktest2(arg);
		break;
	case 3:
		ret = quicktest3(arg);
		break;
	case 99:
		ret = gru_free_kernel_contexts();
		break;
	}
	return ret;

}

int gru_kservices_init(void)
{
	return 0;
}

void gru_kservices_exit(void)
{
	if (gru_free_kernel_contexts())
		BUG();
}