aboutsummaryrefslogtreecommitdiff
path: root/arch/tile/kernel/hardwall.c
blob: 2fd1694ac1d03ee6b40524cae990716b761fafe6 (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
/*
 * Copyright 2010 Tilera Corporation. 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, version 2.
 *
 *   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, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 */

#include <linux/fs.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/rwsem.h>
#include <linux/kprobes.h>
#include <linux/sched.h>
#include <linux/hardirq.h>
#include <linux/uaccess.h>
#include <linux/smp.h>
#include <linux/cdev.h>
#include <linux/compat.h>
#include <asm/hardwall.h>
#include <asm/traps.h>
#include <asm/siginfo.h>
#include <asm/irq_regs.h>

#include <arch/interrupts.h>
#include <arch/spr_def.h>


/*
 * Implement a per-cpu "hardwall" resource class such as UDN or IPI.
 * We use "hardwall" nomenclature throughout for historical reasons.
 * The lock here controls access to the list data structure as well as
 * to the items on the list.
 */
struct hardwall_type {
	int index;
	int is_xdn;
	int is_idn;
	int disabled;
	const char *name;
	struct list_head list;
	spinlock_t lock;
	struct proc_dir_entry *proc_dir;
};

enum hardwall_index {
	HARDWALL_UDN = 0,
#ifndef __tilepro__
	HARDWALL_IDN = 1,
	HARDWALL_IPI = 2,
#endif
	_HARDWALL_TYPES
};

static struct hardwall_type hardwall_types[] = {
	{  /* user-space access to UDN */
		0,
		1,
		0,
		0,
		"udn",
		LIST_HEAD_INIT(hardwall_types[HARDWALL_UDN].list),
		__SPIN_LOCK_UNLOCKED(hardwall_types[HARDWALL_UDN].lock),
		NULL
	},
#ifndef __tilepro__
	{  /* user-space access to IDN */
		1,
		1,
		1,
		1,  /* disabled pending hypervisor support */
		"idn",
		LIST_HEAD_INIT(hardwall_types[HARDWALL_IDN].list),
		__SPIN_LOCK_UNLOCKED(hardwall_types[HARDWALL_IDN].lock),
		NULL
	},
	{  /* access to user-space IPI */
		2,
		0,
		0,
		0,
		"ipi",
		LIST_HEAD_INIT(hardwall_types[HARDWALL_IPI].list),
		__SPIN_LOCK_UNLOCKED(hardwall_types[HARDWALL_IPI].lock),
		NULL
	},
#endif
};

/*
 * This data structure tracks the cpu data, etc., associated
 * one-to-one with a "struct file *" from opening a hardwall device file.
 * Note that the file's private data points back to this structure.
 */
struct hardwall_info {
	struct list_head list;             /* for hardwall_types.list */
	struct list_head task_head;        /* head of tasks in this hardwall */
	struct hardwall_type *type;        /* type of this resource */
	struct cpumask cpumask;            /* cpus reserved */
	int id;                            /* integer id for this hardwall */
	int teardown_in_progress;          /* are we tearing this one down? */

	/* Remaining fields only valid for user-network resources. */
	int ulhc_x;                        /* upper left hand corner x coord */
	int ulhc_y;                        /* upper left hand corner y coord */
	int width;                         /* rectangle width */
	int height;                        /* rectangle height */
#if CHIP_HAS_REV1_XDN()
	atomic_t xdn_pending_count;        /* cores in phase 1 of drain */
#endif
};


/* /proc/tile/hardwall */
static struct proc_dir_entry *hardwall_proc_dir;

/* Functions to manage files in /proc/tile/hardwall. */
static void hardwall_add_proc(struct hardwall_info *);
static void hardwall_remove_proc(struct hardwall_info *);

/* Allow disabling UDN access. */
static int __init noudn(char *str)
{
	pr_info("User-space UDN access is disabled\n");
	hardwall_types[HARDWALL_UDN].disabled = 1;
	return 0;
}
early_param("noudn", noudn);

#ifndef __tilepro__
/* Allow disabling IDN access. */
static int __init noidn(char *str)
{
	pr_info("User-space IDN access is disabled\n");
	hardwall_types[HARDWALL_IDN].disabled = 1;
	return 0;
}
early_param("noidn", noidn);

/* Allow disabling IPI access. */
static int __init noipi(char *str)
{
	pr_info("User-space IPI access is disabled\n");
	hardwall_types[HARDWALL_IPI].disabled = 1;
	return 0;
}
early_param("noipi", noipi);
#endif


/*
 * Low-level primitives for UDN/IDN
 */

#ifdef __tilepro__
#define mtspr_XDN(hwt, name, val) \
	do { (void)(hwt); __insn_mtspr(SPR_UDN_##name, (val)); } while (0)
#define mtspr_MPL_XDN(hwt, name, val) \
	do { (void)(hwt); __insn_mtspr(SPR_MPL_UDN_##name, (val)); } while (0)
#define mfspr_XDN(hwt, name) \
	((void)(hwt), __insn_mfspr(SPR_UDN_##name))
#else
#define mtspr_XDN(hwt, name, val)					\
	do {								\
		if ((hwt)->is_idn)					\
			__insn_mtspr(SPR_IDN_##name, (val));		\
		else							\
			__insn_mtspr(SPR_UDN_##name, (val));		\
	} while (0)
#define mtspr_MPL_XDN(hwt, name, val)					\
	do {								\
		if ((hwt)->is_idn)					\
			__insn_mtspr(SPR_MPL_IDN_##name, (val));	\
		else							\
			__insn_mtspr(SPR_MPL_UDN_##name, (val));	\
	} while (0)
#define mfspr_XDN(hwt, name) \
  ((hwt)->is_idn ? __insn_mfspr(SPR_IDN_##name) : __insn_mfspr(SPR_UDN_##name))
#endif

/* Set a CPU bit if the CPU is online. */
#define cpu_online_set(cpu, dst) do { \
	if (cpu_online(cpu))          \
		cpumask_set_cpu(cpu, dst);    \
} while (0)


/* Does the given rectangle contain the given x,y coordinate? */
static int contains(struct hardwall_info *r, int x, int y)
{
	return (x >= r->ulhc_x && x < r->ulhc_x + r->width) &&
		(y >= r->ulhc_y && y < r->ulhc_y + r->height);
}

/* Compute the rectangle parameters and validate the cpumask. */
static int check_rectangle(struct hardwall_info *r, struct cpumask *mask)
{
	int x, y, cpu, ulhc, lrhc;

	/* The first cpu is the ULHC, the last the LRHC. */
	ulhc = find_first_bit(cpumask_bits(mask), nr_cpumask_bits);
	lrhc = find_last_bit(cpumask_bits(mask), nr_cpumask_bits);

	/* Compute the rectangle attributes from the cpus. */
	r->ulhc_x = cpu_x(ulhc);
	r->ulhc_y = cpu_y(ulhc);
	r->width = cpu_x(lrhc) - r->ulhc_x + 1;
	r->height = cpu_y(lrhc) - r->ulhc_y + 1;

	/* Width and height must be positive */
	if (r->width <= 0 || r->height <= 0)
		return -EINVAL;

	/* Confirm that the cpumask is exactly the rectangle. */
	for (y = 0, cpu = 0; y < smp_height; ++y)
		for (x = 0; x < smp_width; ++x, ++cpu)
			if (cpumask_test_cpu(cpu, mask) != contains(r, x, y))
				return -EINVAL;

	/*
	 * Note that offline cpus can't be drained when this user network
	 * rectangle eventually closes.  We used to detect this
	 * situation and print a warning, but it annoyed users and
	 * they ignored it anyway, so now we just return without a
	 * warning.
	 */
	return 0;
}

/*
 * Hardware management of hardwall setup, teardown, trapping,
 * and enabling/disabling PL0 access to the networks.
 */

/* Bit field values to mask together for writes to SPR_XDN_DIRECTION_PROTECT */
enum direction_protect {
	N_PROTECT = (1 << 0),
	E_PROTECT = (1 << 1),
	S_PROTECT = (1 << 2),
	W_PROTECT = (1 << 3),
	C_PROTECT = (1 << 4),
};

static inline int xdn_which_interrupt(struct hardwall_type *hwt)
{
#ifndef __tilepro__
	if (hwt->is_idn)
		return INT_IDN_FIREWALL;
#endif
	return INT_UDN_FIREWALL;
}

static void enable_firewall_interrupts(struct hardwall_type *hwt)
{
	arch_local_irq_unmask_now(xdn_which_interrupt(hwt));
}

static void disable_firewall_interrupts(struct hardwall_type *hwt)
{
	arch_local_irq_mask_now(xdn_which_interrupt(hwt));
}

/* Set up hardwall on this cpu based on the passed hardwall_info. */
static void hardwall_setup_func(void *info)
{
	struct hardwall_info *r = info;
	struct hardwall_type *hwt = r->type;

	int cpu = smp_processor_id();  /* on_each_cpu disables preemption */
	int x = cpu_x(cpu);
	int y = cpu_y(cpu);
	int bits = 0;
	if (x == r->ulhc_x)
		bits |= W_PROTECT;
	if (x == r->ulhc_x + r->width - 1)
		bits |= E_PROTECT;
	if (y == r->ulhc_y)
		bits |= N_PROTECT;
	if (y == r->ulhc_y + r->height - 1)
		bits |= S_PROTECT;
	BUG_ON(bits == 0);
	mtspr_XDN(hwt, DIRECTION_PROTECT, bits);
	enable_firewall_interrupts(hwt);
}

/* Set up all cpus on edge of rectangle to enable/disable hardwall SPRs. */
static void hardwall_protect_rectangle(struct hardwall_info *r)
{
	int x, y, cpu, delta;
	struct cpumask rect_cpus;

	cpumask_clear(&rect_cpus);

	/* First include the top and bottom edges */
	cpu = r->ulhc_y * smp_width + r->ulhc_x;
	delta = (r->height - 1) * smp_width;
	for (x = 0; x < r->width; ++x, ++cpu) {
		cpu_online_set(cpu, &rect_cpus);
		cpu_online_set(cpu + delta, &rect_cpus);
	}

	/* Then the left and right edges */
	cpu -= r->width;
	delta = r->width - 1;
	for (y = 0; y < r->height; ++y, cpu += smp_width) {
		cpu_online_set(cpu, &rect_cpus);
		cpu_online_set(cpu + delta, &rect_cpus);
	}

	/* Then tell all the cpus to set up their protection SPR */
	on_each_cpu_mask(&rect_cpus, hardwall_setup_func, r, 1);
}

/* Entered from INT_xDN_FIREWALL interrupt vector with irqs disabled. */
void __kprobes do_hardwall_trap(struct pt_regs* regs, int fault_num)
{
	struct hardwall_info *rect;
	struct hardwall_type *hwt;
	struct task_struct *p;
	struct siginfo info;
	int cpu = smp_processor_id();
	int found_processes;
	struct pt_regs *old_regs = set_irq_regs(regs);

	irq_enter();

	/* Figure out which network trapped. */
	switch (fault_num) {
#ifndef __tilepro__
	case INT_IDN_FIREWALL:
		hwt = &hardwall_types[HARDWALL_IDN];
		break;
#endif
	case INT_UDN_FIREWALL:
		hwt = &hardwall_types[HARDWALL_UDN];
		break;
	default:
		BUG();
	}
	BUG_ON(hwt->disabled);

	/* This tile trapped a network access; find the rectangle. */
	spin_lock(&hwt->lock);
	list_for_each_entry(rect, &hwt->list, list) {
		if (cpumask_test_cpu(cpu, &rect->cpumask))
			break;
	}

	/*
	 * It shouldn't be possible not to find this cpu on the
	 * rectangle list, since only cpus in rectangles get hardwalled.
	 * The hardwall is only removed after the user network is drained.
	 */
	BUG_ON(&rect->list == &hwt->list);

	/*
	 * If we already started teardown on this hardwall, don't worry;
	 * the abort signal has been sent and we are just waiting for things
	 * to quiesce.
	 */
	if (rect->teardown_in_progress) {
		pr_notice("cpu %d: detected %s hardwall violation %#lx while teardown already in progress\n",
			  cpu, hwt->name,
			  (long)mfspr_XDN(hwt, DIRECTION_PROTECT));
		goto done;
	}

	/*
	 * Kill off any process that is activated in this rectangle.
	 * We bypass security to deliver the signal, since it must be
	 * one of the activated processes that generated the user network
	 * message that caused this trap, and all the activated
	 * processes shared a single open file so are pretty tightly
	 * bound together from a security point of view to begin with.
	 */
	rect->teardown_in_progress = 1;
	wmb(); /* Ensure visibility of rectangle before notifying processes. */
	pr_notice("cpu %d: detected %s hardwall violation %#lx...\n",
		  cpu, hwt->name, (long)mfspr_XDN(hwt, DIRECTION_PROTECT));
	info.si_signo = SIGILL;
	info.si_errno = 0;
	info.si_code = ILL_HARDWALL;
	found_processes = 0;
	list_for_each_entry(p, &rect->task_head,
			    thread.hardwall[hwt->index].list) {
		BUG_ON(p->thread.hardwall[hwt->index].info != rect);
		if (!(p->flags & PF_EXITING)) {
			found_processes = 1;
			pr_notice("hardwall: killing %d\n", p->pid);
			do_send_sig_info(info.si_signo, &info, p, false);
		}
	}
	if (!found_processes)
		pr_notice("hardwall: no associated processes!\n");

 done:
	spin_unlock(&hwt->lock);

	/*
	 * We have to disable firewall interrupts now, or else when we
	 * return from this handler, we will simply re-interrupt back to
	 * it.  However, we can't clear the protection bits, since we
	 * haven't yet drained the network, and that would allow packets
	 * to cross out of the hardwall region.
	 */
	disable_firewall_interrupts(hwt);

	irq_exit();
	set_irq_regs(old_regs);
}

/* Allow access from user space to the user network. */
void grant_hardwall_mpls(struct hardwall_type *hwt)
{
#ifndef __tilepro__
	if (!hwt->is_xdn) {
		__insn_mtspr(SPR_MPL_IPI_0_SET_0, 1);
		return;
	}
#endif
	mtspr_MPL_XDN(hwt, ACCESS_SET_0, 1);
	mtspr_MPL_XDN(hwt, AVAIL_SET_0, 1);
	mtspr_MPL_XDN(hwt, COMPLETE_SET_0, 1);
	mtspr_MPL_XDN(hwt, TIMER_SET_0, 1);
#if !CHIP_HAS_REV1_XDN()
	mtspr_MPL_XDN(hwt, REFILL_SET_0, 1);
	mtspr_MPL_XDN(hwt, CA_SET_0, 1);
#endif
}

/* Deny access from user space to the user network. */
void restrict_hardwall_mpls(struct hardwall_type *hwt)
{
#ifndef __tilepro__
	if (!hwt->is_xdn) {
		__insn_mtspr(SPR_MPL_IPI_0_SET_1, 1);
		return;
	}
#endif
	mtspr_MPL_XDN(hwt, ACCESS_SET_1, 1);
	mtspr_MPL_XDN(hwt, AVAIL_SET_1, 1);
	mtspr_MPL_XDN(hwt, COMPLETE_SET_1, 1);
	mtspr_MPL_XDN(hwt, TIMER_SET_1, 1);
#if !CHIP_HAS_REV1_XDN()
	mtspr_MPL_XDN(hwt, REFILL_SET_1, 1);
	mtspr_MPL_XDN(hwt, CA_SET_1, 1);
#endif
}

/* Restrict or deny as necessary for the task we're switching to. */
void hardwall_switch_tasks(struct task_struct *prev,
			   struct task_struct *next)
{
	int i;
	for (i = 0; i < HARDWALL_TYPES; ++i) {
		if (prev->thread.hardwall[i].info != NULL) {
			if (next->thread.hardwall[i].info == NULL)
				restrict_hardwall_mpls(&hardwall_types[i]);
		} else if (next->thread.hardwall[i].info != NULL) {
			grant_hardwall_mpls(&hardwall_types[i]);
		}
	}
}

/* Does this task have the right to IPI the given cpu? */
int hardwall_ipi_valid(int cpu)
{
#ifdef __tilegx__
	struct hardwall_info *info =
		current->thread.hardwall[HARDWALL_IPI].info;
	return info && cpumask_test_cpu(cpu, &info->cpumask);
#else
	return 0;
#endif
}

/*
 * Code to create, activate, deactivate, and destroy hardwall resources.
 */

/* Create a hardwall for the given resource */
static struct hardwall_info *hardwall_create(struct hardwall_type *hwt,
					     size_t size,
					     const unsigned char __user *bits)
{
	struct hardwall_info *iter, *info;
	struct cpumask mask;
	unsigned long flags;
	int rc;

	/* Reject crazy sizes out of hand, a la sys_mbind(). */
	if (size > PAGE_SIZE)
		return ERR_PTR(-EINVAL);

	/* Copy whatever fits into a cpumask. */
	if (copy_from_user(&mask, bits, min(sizeof(struct cpumask), size)))
		return ERR_PTR(-EFAULT);

	/*
	 * If the size was short, clear the rest of the mask;
	 * otherwise validate that the rest of the user mask was zero
	 * (we don't try hard to be efficient when validating huge masks).
	 */
	if (size < sizeof(struct cpumask)) {
		memset((char *)&mask + size, 0, sizeof(struct cpumask) - size);
	} else if (size > sizeof(struct cpumask)) {
		size_t i;
		for (i = sizeof(struct cpumask); i < size; ++i) {
			char c;
			if (get_user(c, &bits[i]))
				return ERR_PTR(-EFAULT);
			if (c)
				return ERR_PTR(-EINVAL);
		}
	}

	/* Allocate a new hardwall_info optimistically. */
	info = kmalloc(sizeof(struct hardwall_info),
			GFP_KERNEL | __GFP_ZERO);
	if (info == NULL)
		return ERR_PTR(-ENOMEM);
	INIT_LIST_HEAD(&info->task_head);
	info->type = hwt;

	/* Compute the rectangle size and validate that it's plausible. */
	cpumask_copy(&info->cpumask, &mask);
	info->id = find_first_bit(cpumask_bits(&mask), nr_cpumask_bits);
	if (hwt->is_xdn) {
		rc = check_rectangle(info, &mask);
		if (rc != 0) {
			kfree(info);
			return ERR_PTR(rc);
		}
	}

	/*
	 * Eliminate cpus that are not part of this Linux client.
	 * Note that this allows for configurations that we might not want to
	 * support, such as one client on every even cpu, another client on
	 * every odd cpu.
	 */
	cpumask_and(&info->cpumask, &info->cpumask, cpu_online_mask);

	/* Confirm it doesn't overlap and add it to the list. */
	spin_lock_irqsave(&hwt->lock, flags);
	list_for_each_entry(iter, &hwt->list, list) {
		if (cpumask_intersects(&iter->cpumask, &info->cpumask)) {
			spin_unlock_irqrestore(&hwt->lock, flags);
			kfree(info);
			return ERR_PTR(-EBUSY);
		}
	}
	list_add_tail(&info->list, &hwt->list);
	spin_unlock_irqrestore(&hwt->lock, flags);

	/* Set up appropriate hardwalling on all affected cpus. */
	if (hwt->is_xdn)
		hardwall_protect_rectangle(info);

	/* Create a /proc/tile/hardwall entry. */
	hardwall_add_proc(info);

	return info;
}

/* Activate a given hardwall on this cpu for this process. */
static int hardwall_activate(struct hardwall_info *info)
{
	int cpu;
	unsigned long flags;
	struct task_struct *p = current;
	struct thread_struct *ts = &p->thread;
	struct hardwall_type *hwt;

	/* Require a hardwall. */
	if (info == NULL)
		return -ENODATA;

	/* Not allowed to activate a hardwall that is being torn down. */
	if (info->teardown_in_progress)
		return -EINVAL;

	/*
	 * Get our affinity; if we're not bound to this tile uniquely,
	 * we can't access the network registers.
	 */
	if (cpumask_weight(&p->cpus_allowed) != 1)
		return -EPERM;

	/* Make sure we are bound to a cpu assigned to this resource. */
	cpu = smp_processor_id();
	BUG_ON(cpumask_first(&p->cpus_allowed) != cpu);
	if (!cpumask_test_cpu(cpu, &info->cpumask))
		return -EINVAL;

	/* If we are already bound to this hardwall, it's a no-op. */
	hwt = info->type;
	if (ts->hardwall[hwt->index].info) {
		BUG_ON(ts->hardwall[hwt->index].info != info);
		return 0;
	}

	/* Success!  This process gets to use the resource on this cpu. */
	ts->hardwall[hwt->index].info = info;
	spin_lock_irqsave(&hwt->lock, flags);
	list_add(&ts->hardwall[hwt->index].list, &info->task_head);
	spin_unlock_irqrestore(&hwt->lock, flags);
	grant_hardwall_mpls(hwt);
	printk(KERN_DEBUG "Pid %d (%s) activated for %s hardwall: cpu %d\n",
	       p->pid, p->comm, hwt->name, cpu);
	return 0;
}

/*
 * Deactivate a task's hardwall.  Must hold lock for hardwall_type.
 * This method may be called from exit_thread(), so we don't want to
 * rely on too many fields of struct task_struct still being valid.
 * We assume the cpus_allowed, pid, and comm fields are still valid.
 */
static void _hardwall_deactivate(struct hardwall_type *hwt,
				 struct task_struct *task)
{
	struct thread_struct *ts = &task->thread;

	if (cpumask_weight(&task->cpus_allowed) != 1) {
		pr_err("pid %d (%s) releasing %s hardwall with an affinity mask containing %d cpus!\n",
		       task->pid, task->comm, hwt->name,
		       cpumask_weight(&task->cpus_allowed));
		BUG();
	}

	BUG_ON(ts->hardwall[hwt->index].info == NULL);
	ts->hardwall[hwt->index].info = NULL;
	list_del(&ts->hardwall[hwt->index].list);
	if (task == current)
		restrict_hardwall_mpls(hwt);
}

/* Deactivate a task's hardwall. */
static int hardwall_deactivate(struct hardwall_type *hwt,
			       struct task_struct *task)
{
	unsigned long flags;
	int activated;

	spin_lock_irqsave(&hwt->lock, flags);
	activated = (task->thread.hardwall[hwt->index].info != NULL);
	if (activated)
		_hardwall_deactivate(hwt, task);
	spin_unlock_irqrestore(&hwt->lock, flags);

	if (!activated)
		return -EINVAL;

	printk(KERN_DEBUG "Pid %d (%s) deactivated for %s hardwall: cpu %d\n",
	       task->pid, task->comm, hwt->name, raw_smp_processor_id());
	return 0;
}

void hardwall_deactivate_all(struct task_struct *task)
{
	int i;
	for (i = 0; i < HARDWALL_TYPES; ++i)
		if (task->thread.hardwall[i].info)
			hardwall_deactivate(&hardwall_types[i], task);
}

/* Stop the switch before draining the network. */
static void stop_xdn_switch(void *arg)
{
#if !CHIP_HAS_REV1_XDN()
	/* Freeze the switch and the demux. */
	__insn_mtspr(SPR_UDN_SP_FREEZE,
		     SPR_UDN_SP_FREEZE__SP_FRZ_MASK |
		     SPR_UDN_SP_FREEZE__DEMUX_FRZ_MASK |
		     SPR_UDN_SP_FREEZE__NON_DEST_EXT_MASK);
#else
	/*
	 * Drop all packets bound for the core or off the edge.
	 * We rely on the normal hardwall protection setup code
	 * to have set the low four bits to trigger firewall interrupts,
	 * and shift those bits up to trigger "drop on send" semantics,
	 * plus adding "drop on send to core" for all switches.
	 * In practice it seems the switches latch the DIRECTION_PROTECT
	 * SPR so they won't start dropping if they're already
	 * delivering the last message to the core, but it doesn't
	 * hurt to enable it here.
	 */
	struct hardwall_type *hwt = arg;
	unsigned long protect = mfspr_XDN(hwt, DIRECTION_PROTECT);
	mtspr_XDN(hwt, DIRECTION_PROTECT, (protect | C_PROTECT) << 5);
#endif
}

static void empty_xdn_demuxes(struct hardwall_type *hwt)
{
#ifndef __tilepro__
	if (hwt->is_idn) {
		while (__insn_mfspr(SPR_IDN_DATA_AVAIL) & (1 << 0))
			(void) __tile_idn0_receive();
		while (__insn_mfspr(SPR_IDN_DATA_AVAIL) & (1 << 1))
			(void) __tile_idn1_receive();
		return;
	}
#endif
	while (__insn_mfspr(SPR_UDN_DATA_AVAIL) & (1 << 0))
		(void) __tile_udn0_receive();
	while (__insn_mfspr(SPR_UDN_DATA_AVAIL) & (1 << 1))
		(void) __tile_udn1_receive();
	while (__insn_mfspr(SPR_UDN_DATA_AVAIL) & (1 << 2))
		(void) __tile_udn2_receive();
	while (__insn_mfspr(SPR_UDN_DATA_AVAIL) & (1 << 3))
		(void) __tile_udn3_receive();
}

/* Drain all the state from a stopped switch. */
static void drain_xdn_switch(void *arg)
{
	struct hardwall_info *info = arg;
	struct hardwall_type *hwt = info->type;

#if CHIP_HAS_REV1_XDN()
	/*
	 * The switches have been configured to drop any messages
	 * destined for cores (or off the edge of the rectangle).
	 * But the current message may continue to be delivered,
	 * so we wait until all the cores have finished any pending
	 * messages before we stop draining.
	 */
	int pending = mfspr_XDN(hwt, PENDING);
	while (pending--) {
		empty_xdn_demuxes(hwt);
		if (hwt->is_idn)
			__tile_idn_send(0);
		else
			__tile_udn_send(0);
	}
	atomic_dec(&info->xdn_pending_count);
	while (atomic_read(&info->xdn_pending_count))
		empty_xdn_demuxes(hwt);
#else
	int i;
	int from_tile_words, ca_count;

	/* Empty out the 5 switch point fifos. */
	for (i = 0; i < 5; i++) {
		int words, j;
		__insn_mtspr(SPR_UDN_SP_FIFO_SEL, i);
		words = __insn_mfspr(SPR_UDN_SP_STATE) & 0xF;
		for (j = 0; j < words; j++)
			(void) __insn_mfspr(SPR_UDN_SP_FIFO_DATA);
		BUG_ON((__insn_mfspr(SPR_UDN_SP_STATE) & 0xF) != 0);
	}

	/* Dump out the 3 word fifo at top. */
	from_tile_words = (__insn_mfspr(SPR_UDN_DEMUX_STATUS) >> 10) & 0x3;
	for (i = 0; i < from_tile_words; i++)
		(void) __insn_mfspr(SPR_UDN_DEMUX_WRITE_FIFO);

	/* Empty out demuxes. */
	empty_xdn_demuxes(hwt);

	/* Empty out catch all. */
	ca_count = __insn_mfspr(SPR_UDN_DEMUX_CA_COUNT);
	for (i = 0; i < ca_count; i++)
		(void) __insn_mfspr(SPR_UDN_CA_DATA);
	BUG_ON(__insn_mfspr(SPR_UDN_DEMUX_CA_COUNT) != 0);

	/* Clear demux logic. */
	__insn_mtspr(SPR_UDN_DEMUX_CTL, 1);

	/*
	 * Write switch state; experimentation indicates that 0xc3000
	 * is an idle switch point.
	 */
	for (i = 0; i < 5; i++) {
		__insn_mtspr(SPR_UDN_SP_FIFO_SEL, i);
		__insn_mtspr(SPR_UDN_SP_STATE, 0xc3000);
	}
#endif
}

/* Reset random XDN state registers at boot up and during hardwall teardown. */
static void reset_xdn_network_state(struct hardwall_type *hwt)
{
	if (hwt->disabled)
		return;

	/* Clear out other random registers so we have a clean slate. */
	mtspr_XDN(hwt, DIRECTION_PROTECT, 0);
	mtspr_XDN(hwt, AVAIL_EN, 0);
	mtspr_XDN(hwt, DEADLOCK_TIMEOUT, 0);

#if !CHIP_HAS_REV1_XDN()
	/* Reset UDN coordinates to their standard value */
	{
		unsigned int cpu = smp_processor_id();
		unsigned int x = cpu_x(cpu);
		unsigned int y = cpu_y(cpu);
		__insn_mtspr(SPR_UDN_TILE_COORD, (x << 18) | (y << 7));
	}

	/* Set demux tags to predefined values and enable them. */
	__insn_mtspr(SPR_UDN_TAG_VALID, 0xf);
	__insn_mtspr(SPR_UDN_TAG_0, (1 << 0));
	__insn_mtspr(SPR_UDN_TAG_1, (1 << 1));
	__insn_mtspr(SPR_UDN_TAG_2, (1 << 2));
	__insn_mtspr(SPR_UDN_TAG_3, (1 << 3));

	/* Set other rev0 random registers to a clean state. */
	__insn_mtspr(SPR_UDN_REFILL_EN, 0);
	__insn_mtspr(SPR_UDN_DEMUX_QUEUE_SEL, 0);
	__insn_mtspr(SPR_UDN_SP_FIFO_SEL, 0);

	/* Start the switch and demux. */
	__insn_mtspr(SPR_UDN_SP_FREEZE, 0);
#endif
}

void reset_network_state(void)
{
	reset_xdn_network_state(&hardwall_types[HARDWALL_UDN]);
#ifndef __tilepro__
	reset_xdn_network_state(&hardwall_types[HARDWALL_IDN]);
#endif
}

/* Restart an XDN switch after draining. */
static void restart_xdn_switch(void *arg)
{
	struct hardwall_type *hwt = arg;

#if CHIP_HAS_REV1_XDN()
	/* One last drain step to avoid races with injection and draining. */
	empty_xdn_demuxes(hwt);
#endif

	reset_xdn_network_state(hwt);

	/* Disable firewall interrupts. */
	disable_firewall_interrupts(hwt);
}

/* Last reference to a hardwall is gone, so clear the network. */
static void hardwall_destroy(struct hardwall_info *info)
{
	struct task_struct *task;
	struct hardwall_type *hwt;
	unsigned long flags;

	/* Make sure this file actually represents a hardwall. */
	if (info == NULL)
		return;

	/*
	 * Deactivate any remaining tasks.  It's possible to race with
	 * some other thread that is exiting and hasn't yet called
	 * deactivate (when freeing its thread_info), so we carefully
	 * deactivate any remaining tasks before freeing the
	 * hardwall_info object itself.
	 */
	hwt = info->type;
	info->teardown_in_progress = 1;
	spin_lock_irqsave(&hwt->lock, flags);
	list_for_each_entry(task, &info->task_head,
			    thread.hardwall[hwt->index].list)
		_hardwall_deactivate(hwt, task);
	spin_unlock_irqrestore(&hwt->lock, flags);

	if (hwt->is_xdn) {
		/* Configure the switches for draining the user network. */
		printk(KERN_DEBUG
		       "Clearing %s hardwall rectangle %dx%d %d,%d\n",
		       hwt->name, info->width, info->height,
		       info->ulhc_x, info->ulhc_y);
		on_each_cpu_mask(&info->cpumask, stop_xdn_switch, hwt, 1);

		/* Drain the network. */
#if CHIP_HAS_REV1_XDN()
		atomic_set(&info->xdn_pending_count,
			   cpumask_weight(&info->cpumask));
		on_each_cpu_mask(&info->cpumask, drain_xdn_switch, info, 0);
#else
		on_each_cpu_mask(&info->cpumask, drain_xdn_switch, info, 1);
#endif

		/* Restart switch and disable firewall. */
		on_each_cpu_mask(&info->cpumask, restart_xdn_switch, hwt, 1);
	}

	/* Remove the /proc/tile/hardwall entry. */
	hardwall_remove_proc(info);

	/* Now free the hardwall from the list. */
	spin_lock_irqsave(&hwt->lock, flags);
	BUG_ON(!list_empty(&info->task_head));
	list_del(&info->list);
	spin_unlock_irqrestore(&hwt->lock, flags);
	kfree(info);
}


static int hardwall_proc_show(struct seq_file *sf, void *v)
{
	struct hardwall_info *info = sf->private;

	seq_printf(sf, "%*pbl\n", cpumask_pr_args(&info->cpumask));
	return 0;
}

static int hardwall_proc_open(struct inode *inode,
			      struct file *file)
{
	return single_open(file, hardwall_proc_show, PDE_DATA(inode));
}

static const struct file_operations hardwall_proc_fops = {
	.open		= hardwall_proc_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
};

static void hardwall_add_proc(struct hardwall_info *info)
{
	char buf[64];
	snprintf(buf, sizeof(buf), "%d", info->id);
	proc_create_data(buf, 0444, info->type->proc_dir,
			 &hardwall_proc_fops, info);
}

static void hardwall_remove_proc(struct hardwall_info *info)
{
	char buf[64];
	snprintf(buf, sizeof(buf), "%d", info->id);
	remove_proc_entry(buf, info->type->proc_dir);
}

int proc_pid_hardwall(struct seq_file *m, struct pid_namespace *ns,
		      struct pid *pid, struct task_struct *task)
{
	int i;
	int n = 0;
	for (i = 0; i < HARDWALL_TYPES; ++i) {
		struct hardwall_info *info = task->thread.hardwall[i].info;
		if (info)
			seq_printf(m, "%s: %d\n", info->type->name, info->id);
	}
	return n;
}

void proc_tile_hardwall_init(struct proc_dir_entry *root)
{
	int i;
	for (i = 0; i < HARDWALL_TYPES; ++i) {
		struct hardwall_type *hwt = &hardwall_types[i];
		if (hwt->disabled)
			continue;
		if (hardwall_proc_dir == NULL)
			hardwall_proc_dir = proc_mkdir("hardwall", root);
		hwt->proc_dir = proc_mkdir(hwt->name, hardwall_proc_dir);
	}
}


/*
 * Character device support via ioctl/close.
 */

static long hardwall_ioctl(struct file *file, unsigned int a, unsigned long b)
{
	struct hardwall_info *info = file->private_data;
	int minor = iminor(file->f_mapping->host);
	struct hardwall_type* hwt;

	if (_IOC_TYPE(a) != HARDWALL_IOCTL_BASE)
		return -EINVAL;

	BUILD_BUG_ON(HARDWALL_TYPES != _HARDWALL_TYPES);
	BUILD_BUG_ON(HARDWALL_TYPES !=
		     sizeof(hardwall_types)/sizeof(hardwall_types[0]));

	if (minor < 0 || minor >= HARDWALL_TYPES)
		return -EINVAL;
	hwt = &hardwall_types[minor];
	WARN_ON(info && hwt != info->type);

	switch (_IOC_NR(a)) {
	case _HARDWALL_CREATE:
		if (hwt->disabled)
			return -ENOSYS;
		if (info != NULL)
			return -EALREADY;
		info = hardwall_create(hwt, _IOC_SIZE(a),
				       (const unsigned char __user *)b);
		if (IS_ERR(info))
			return PTR_ERR(info);
		file->private_data = info;
		return 0;

	case _HARDWALL_ACTIVATE:
		return hardwall_activate(info);

	case _HARDWALL_DEACTIVATE:
		if (current->thread.hardwall[hwt->index].info != info)
			return -EINVAL;
		return hardwall_deactivate(hwt, current);

	case _HARDWALL_GET_ID:
		return info ? info->id : -EINVAL;

	default:
		return -EINVAL;
	}
}

#ifdef CONFIG_COMPAT
static long hardwall_compat_ioctl(struct file *file,
				  unsigned int a, unsigned long b)
{
	/* Sign-extend the argument so it can be used as a pointer. */
	return hardwall_ioctl(file, a, (unsigned long)compat_ptr(b));
}
#endif

/* The user process closed the file; revoke access to user networks. */
static int hardwall_flush(struct file *file, fl_owner_t owner)
{
	struct hardwall_info *info = file->private_data;
	struct task_struct *task, *tmp;
	unsigned long flags;

	if (info) {
		/*
		 * NOTE: if multiple threads are activated on this hardwall
		 * file, the other threads will continue having access to the
		 * user network until they are context-switched out and back
		 * in again.
		 *
		 * NOTE: A NULL files pointer means the task is being torn
		 * down, so in that case we also deactivate it.
		 */
		struct hardwall_type *hwt = info->type;
		spin_lock_irqsave(&hwt->lock, flags);
		list_for_each_entry_safe(task, tmp, &info->task_head,
					 thread.hardwall[hwt->index].list) {
			if (task->files == owner || task->files == NULL)
				_hardwall_deactivate(hwt, task);
		}
		spin_unlock_irqrestore(&hwt->lock, flags);
	}

	return 0;
}

/* This hardwall is gone, so destroy it. */
static int hardwall_release(struct inode *inode, struct file *file)
{
	hardwall_destroy(file->private_data);
	return 0;
}

static const struct file_operations dev_hardwall_fops = {
	.open           = nonseekable_open,
	.unlocked_ioctl = hardwall_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl   = hardwall_compat_ioctl,
#endif
	.flush          = hardwall_flush,
	.release        = hardwall_release,
};

static struct cdev hardwall_dev;

static int __init dev_hardwall_init(void)
{
	int rc;
	dev_t dev;

	rc = alloc_chrdev_region(&dev, 0, HARDWALL_TYPES, "hardwall");
	if (rc < 0)
		return rc;
	cdev_init(&hardwall_dev, &dev_hardwall_fops);
	rc = cdev_add(&hardwall_dev, dev, HARDWALL_TYPES);
	if (rc < 0)
		return rc;

	return 0;
}
late_initcall(dev_hardwall_init);