summaryrefslogtreecommitdiffstats
path: root/arch/tile/kernel/single_step.c
blob: 266aae123632dfbea8900bf320e9e436e3902151 (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
/*
 * 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.
 *
 * A code-rewriter that enables instruction single-stepping.
 * Derived from iLib's single-stepping code.
 */

#ifndef __tilegx__   /* No support for single-step yet. */

/* These functions are only used on the TILE platform */
#include <linux/slab.h>
#include <linux/thread_info.h>
#include <linux/uaccess.h>
#include <linux/mman.h>
#include <linux/types.h>
#include <asm/cacheflush.h>
#include <asm/opcode-tile.h>
#include <asm/opcode_constants.h>
#include <arch/abi.h>

#define signExtend17(val) sign_extend((val), 17)
#define TILE_X1_MASK (0xffffffffULL << 31)

int unaligned_printk;

static int __init setup_unaligned_printk(char *str)
{
	long val;
	if (strict_strtol(str, 0, &val) != 0)
		return 0;
	unaligned_printk = val;
	printk("Printk for each unaligned data accesses is %s\n",
	       unaligned_printk ? "enabled" : "disabled");
	return 1;
}
__setup("unaligned_printk=", setup_unaligned_printk);

unsigned int unaligned_fixup_count;

enum mem_op {
	MEMOP_NONE,
	MEMOP_LOAD,
	MEMOP_STORE,
	MEMOP_LOAD_POSTINCR,
	MEMOP_STORE_POSTINCR
};

static inline tile_bundle_bits set_BrOff_X1(tile_bundle_bits n, int32_t offset)
{
	tile_bundle_bits result;

	/* mask out the old offset */
	tile_bundle_bits mask = create_BrOff_X1(-1);
	result = n & (~mask);

	/* or in the new offset */
	result |= create_BrOff_X1(offset);

	return result;
}

static inline tile_bundle_bits move_X1(tile_bundle_bits n, int dest, int src)
{
	tile_bundle_bits result;
	tile_bundle_bits op;

	result = n & (~TILE_X1_MASK);

	op = create_Opcode_X1(SPECIAL_0_OPCODE_X1) |
		create_RRROpcodeExtension_X1(OR_SPECIAL_0_OPCODE_X1) |
		create_Dest_X1(dest) |
		create_SrcB_X1(TREG_ZERO) |
		create_SrcA_X1(src) ;

	result |= op;
	return result;
}

static inline tile_bundle_bits nop_X1(tile_bundle_bits n)
{
	return move_X1(n, TREG_ZERO, TREG_ZERO);
}

static inline tile_bundle_bits addi_X1(
	tile_bundle_bits n, int dest, int src, int imm)
{
	n &= ~TILE_X1_MASK;

	n |=  (create_SrcA_X1(src) |
	       create_Dest_X1(dest) |
	       create_Imm8_X1(imm) |
	       create_S_X1(0) |
	       create_Opcode_X1(IMM_0_OPCODE_X1) |
	       create_ImmOpcodeExtension_X1(ADDI_IMM_0_OPCODE_X1));

	return n;
}

static tile_bundle_bits rewrite_load_store_unaligned(
	struct single_step_state *state,
	tile_bundle_bits bundle,
	struct pt_regs *regs,
	enum mem_op mem_op,
	int size, int sign_ext)
{
	unsigned char *addr;
	int val_reg, addr_reg, err, val;

	/* Get address and value registers */
	if (bundle & TILE_BUNDLE_Y_ENCODING_MASK) {
		addr_reg = get_SrcA_Y2(bundle);
		val_reg = get_SrcBDest_Y2(bundle);
	} else if (mem_op == MEMOP_LOAD || mem_op == MEMOP_LOAD_POSTINCR) {
		addr_reg = get_SrcA_X1(bundle);
		val_reg  = get_Dest_X1(bundle);
	} else {
		addr_reg = get_SrcA_X1(bundle);
		val_reg  = get_SrcB_X1(bundle);
	}

	/*
	 * If registers are not GPRs, don't try to handle it.
	 *
	 * FIXME: we could handle non-GPR loads by getting the real value
	 * from memory, writing it to the single step buffer, using a
	 * temp_reg to hold a pointer to that memory, then executing that
	 * instruction and resetting temp_reg.  For non-GPR stores, it's a
	 * little trickier; we could use the single step buffer for that
	 * too, but we'd have to add some more state bits so that we could
	 * call back in here to copy that value to the real target.  For
	 * now, we just handle the simple case.
	 */
	if ((val_reg >= PTREGS_NR_GPRS &&
	     (val_reg != TREG_ZERO ||
	      mem_op == MEMOP_LOAD ||
	      mem_op == MEMOP_LOAD_POSTINCR)) ||
	    addr_reg >= PTREGS_NR_GPRS)
		return bundle;

	/* If it's aligned, don't handle it specially */
	addr = (void *)regs->regs[addr_reg];
	if (((unsigned long)addr % size) == 0)
		return bundle;

#ifndef __LITTLE_ENDIAN
# error We assume little-endian representation with copy_xx_user size 2 here
#endif
	/* Handle unaligned load/store */
	if (mem_op == MEMOP_LOAD || mem_op == MEMOP_LOAD_POSTINCR) {
		unsigned short val_16;
		switch (size) {
		case 2:
			err = copy_from_user(&val_16, addr, sizeof(val_16));
			val = sign_ext ? ((short)val_16) : val_16;
			break;
		case 4:
			err = copy_from_user(&val, addr, sizeof(val));
			break;
		default:
			BUG();
		}
		if (err == 0) {
			state->update_reg = val_reg;
			state->update_value = val;
			state->update = 1;
		}
	} else {
		val = (val_reg == TREG_ZERO) ? 0 : regs->regs[val_reg];
		err = copy_to_user(addr, &val, size);
	}

	if (err) {
		siginfo_t info = {
			.si_signo = SIGSEGV,
			.si_code = SEGV_MAPERR,
			.si_addr = (void __user *)addr
		};
		force_sig_info(info.si_signo, &info, current);
		return (tile_bundle_bits) 0;
	}

	if (unaligned_fixup == 0) {
		siginfo_t info = {
			.si_signo = SIGBUS,
			.si_code = BUS_ADRALN,
			.si_addr = (void __user *)addr
		};
		force_sig_info(info.si_signo, &info, current);
		return (tile_bundle_bits) 0;
	}

	if (unaligned_printk || unaligned_fixup_count == 0) {
		printk("Process %d/%s: PC %#lx: Fixup of"
		       " unaligned %s at %#lx.\n",
		       current->pid, current->comm, regs->pc,
		       (mem_op == MEMOP_LOAD || mem_op == MEMOP_LOAD_POSTINCR) ?
			 "load" : "store",
		       (unsigned long)addr);
		if (!unaligned_printk) {
			printk("\n"
"Unaligned fixups in the kernel will slow your application considerably.\n"
"You can find them by writing \"1\" to /proc/sys/tile/unaligned_fixup/printk,\n"
"which requests the kernel show all unaligned fixups, or writing a \"0\"\n"
"to /proc/sys/tile/unaligned_fixup/enabled, in which case each unaligned\n"
"access will become a SIGBUS you can debug. No further warnings will be\n"
"shown so as to avoid additional slowdown, but you can track the number\n"
"of fixups performed via /proc/sys/tile/unaligned_fixup/count.\n"
"Use the tile-addr2line command (see \"info addr2line\") to decode PCs.\n"
				"\n");
		}
	}
	++unaligned_fixup_count;

	if (bundle & TILE_BUNDLE_Y_ENCODING_MASK) {
		/* Convert the Y2 instruction to a prefetch. */
		bundle &= ~(create_SrcBDest_Y2(-1) |
			    create_Opcode_Y2(-1));
		bundle |= (create_SrcBDest_Y2(TREG_ZERO) |
			   create_Opcode_Y2(LW_OPCODE_Y2));
	/* Replace the load postincr with an addi */
	} else if (mem_op == MEMOP_LOAD_POSTINCR) {
		bundle = addi_X1(bundle, addr_reg, addr_reg,
				 get_Imm8_X1(bundle));
	/* Replace the store postincr with an addi */
	} else if (mem_op == MEMOP_STORE_POSTINCR) {
		bundle = addi_X1(bundle, addr_reg, addr_reg,
				 get_Dest_Imm8_X1(bundle));
	} else {
		/* Convert the X1 instruction to a nop. */
		bundle &= ~(create_Opcode_X1(-1) |
			    create_UnShOpcodeExtension_X1(-1) |
			    create_UnOpcodeExtension_X1(-1));
		bundle |= (create_Opcode_X1(SHUN_0_OPCODE_X1) |
			   create_UnShOpcodeExtension_X1(
				   UN_0_SHUN_0_OPCODE_X1) |
			   create_UnOpcodeExtension_X1(
				   NOP_UN_0_SHUN_0_OPCODE_X1));
	}

	return bundle;
}

/**
 * single_step_once() - entry point when single stepping has been triggered.
 * @regs: The machine register state
 *
 *  When we arrive at this routine via a trampoline, the single step
 *  engine copies the executing bundle to the single step buffer.
 *  If the instruction is a condition branch, then the target is
 *  reset to one past the next instruction. If the instruction
 *  sets the lr, then that is noted. If the instruction is a jump
 *  or call, then the new target pc is preserved and the current
 *  bundle instruction set to null.
 *
 *  The necessary post-single-step rewriting information is stored in
 *  single_step_state->  We use data segment values because the
 *  stack will be rewound when we run the rewritten single-stepped
 *  instruction.
 */
void single_step_once(struct pt_regs *regs)
{
	extern tile_bundle_bits __single_step_ill_insn;
	extern tile_bundle_bits __single_step_j_insn;
	extern tile_bundle_bits __single_step_addli_insn;
	extern tile_bundle_bits __single_step_auli_insn;
	struct thread_info *info = (void *)current_thread_info();
	struct single_step_state *state = info->step_state;
	int is_single_step = test_ti_thread_flag(info, TIF_SINGLESTEP);
	tile_bundle_bits *buffer, *pc;
	tile_bundle_bits bundle;
	int temp_reg;
	int target_reg = TREG_LR;
	int err;
	enum mem_op mem_op = MEMOP_NONE;
	int size = 0, sign_ext = 0;  /* happy compiler */

	asm(
"    .pushsection .rodata.single_step\n"
"    .align 8\n"
"    .globl    __single_step_ill_insn\n"
"__single_step_ill_insn:\n"
"    ill\n"
"    .globl    __single_step_addli_insn\n"
"__single_step_addli_insn:\n"
"    { nop; addli r0, zero, 0 }\n"
"    .globl    __single_step_auli_insn\n"
"__single_step_auli_insn:\n"
"    { nop; auli r0, r0, 0 }\n"
"    .globl    __single_step_j_insn\n"
"__single_step_j_insn:\n"
"    j .\n"
"    .popsection\n"
	);

	if (state == NULL) {
		/* allocate a page of writable, executable memory */
		state = kmalloc(sizeof(struct single_step_state), GFP_KERNEL);
		if (state == NULL) {
			printk("Out of kernel memory trying to single-step\n");
			return;
		}

		/* allocate a cache line of writable, executable memory */
		down_write(&current->mm->mmap_sem);
		buffer = (void *) do_mmap(0, 0, 64,
					  PROT_EXEC | PROT_READ | PROT_WRITE,
					  MAP_PRIVATE | MAP_ANONYMOUS,
					  0);
		up_write(&current->mm->mmap_sem);

		if ((int)buffer < 0 && (int)buffer > -PAGE_SIZE) {
			kfree(state);
			printk("Out of kernel pages trying to single-step\n");
			return;
		}

		state->buffer = buffer;
		state->is_enabled = 0;

		info->step_state = state;

		/* Validate our stored instruction patterns */
		BUG_ON(get_Opcode_X1(__single_step_addli_insn) !=
		       ADDLI_OPCODE_X1);
		BUG_ON(get_Opcode_X1(__single_step_auli_insn) !=
		       AULI_OPCODE_X1);
		BUG_ON(get_SrcA_X1(__single_step_addli_insn) != TREG_ZERO);
		BUG_ON(get_Dest_X1(__single_step_addli_insn) != 0);
		BUG_ON(get_JOffLong_X1(__single_step_j_insn) != 0);
	}

	/*
	 * If we are returning from a syscall, we still haven't hit the
	 * "ill" for the swint1 instruction.  So back the PC up to be
	 * pointing at the swint1, but we'll actually return directly
	 * back to the "ill" so we come back in via SIGILL as if we
	 * had "executed" the swint1 without ever being in kernel space.
	 */
	if (regs->faultnum == INT_SWINT_1)
		regs->pc -= 8;

	pc = (tile_bundle_bits *)(regs->pc);
	bundle = pc[0];

	/* We'll follow the instruction with 2 ill op bundles */
	state->orig_pc = (unsigned long) pc;
	state->next_pc = (unsigned long)(pc + 1);
	state->branch_next_pc = 0;
	state->update = 0;

	if (!(bundle & TILE_BUNDLE_Y_ENCODING_MASK)) {
		/* two wide, check for control flow */
		int opcode = get_Opcode_X1(bundle);

		switch (opcode) {
		/* branches */
		case BRANCH_OPCODE_X1:
		{
			int32_t offset = signExtend17(get_BrOff_X1(bundle));

			/*
			 * For branches, we use a rewriting trick to let the
			 * hardware evaluate whether the branch is taken or
			 * untaken.  We record the target offset and then
			 * rewrite the branch instruction to target 1 insn
			 * ahead if the branch is taken.  We then follow the
			 * rewritten branch with two bundles, each containing
			 * an "ill" instruction. The supervisor examines the
			 * pc after the single step code is executed, and if
			 * the pc is the first ill instruction, then the
			 * branch (if any) was not taken.  If the pc is the
			 * second ill instruction, then the branch was
			 * taken. The new pc is computed for these cases, and
			 * inserted into the registers for the thread.  If
			 * the pc is the start of the single step code, then
			 * an exception or interrupt was taken before the
			 * code started processing, and the same "original"
			 * pc is restored.  This change, different from the
			 * original implementation, has the advantage of
			 * executing a single user instruction.
			 */
			state->branch_next_pc = (unsigned long)(pc + offset);

			/* rewrite branch offset to go forward one bundle */
			bundle = set_BrOff_X1(bundle, 2);
		}
		break;

		/* jumps */
		case JALB_OPCODE_X1:
		case JALF_OPCODE_X1:
			state->update = 1;
			state->next_pc =
				(unsigned long) (pc + get_JOffLong_X1(bundle));
			break;

		case JB_OPCODE_X1:
		case JF_OPCODE_X1:
			state->next_pc =
				(unsigned long) (pc + get_JOffLong_X1(bundle));
			bundle = nop_X1(bundle);
			break;

		case SPECIAL_0_OPCODE_X1:
			switch (get_RRROpcodeExtension_X1(bundle)) {
			/* jump-register */
			case JALRP_SPECIAL_0_OPCODE_X1:
			case JALR_SPECIAL_0_OPCODE_X1:
				state->update = 1;
				state->next_pc =
					regs->regs[get_SrcA_X1(bundle)];
				break;

			case JRP_SPECIAL_0_OPCODE_X1:
			case JR_SPECIAL_0_OPCODE_X1:
				state->next_pc =
					regs->regs[get_SrcA_X1(bundle)];
				bundle = nop_X1(bundle);
				break;

			case LNK_SPECIAL_0_OPCODE_X1:
				state->update = 1;
				target_reg = get_Dest_X1(bundle);
				break;

			/* stores */
			case SH_SPECIAL_0_OPCODE_X1:
				mem_op = MEMOP_STORE;
				size = 2;
				break;

			case SW_SPECIAL_0_OPCODE_X1:
				mem_op = MEMOP_STORE;
				size = 4;
				break;
			}
			break;

		/* loads and iret */
		case SHUN_0_OPCODE_X1:
			if (get_UnShOpcodeExtension_X1(bundle) ==
			    UN_0_SHUN_0_OPCODE_X1) {
				switch (get_UnOpcodeExtension_X1(bundle)) {
				case LH_UN_0_SHUN_0_OPCODE_X1:
					mem_op = MEMOP_LOAD;
					size = 2;
					sign_ext = 1;
					break;

				case LH_U_UN_0_SHUN_0_OPCODE_X1:
					mem_op = MEMOP_LOAD;
					size = 2;
					sign_ext = 0;
					break;

				case LW_UN_0_SHUN_0_OPCODE_X1:
					mem_op = MEMOP_LOAD;
					size = 4;
					break;

				case IRET_UN_0_SHUN_0_OPCODE_X1:
				{
					unsigned long ex0_0 = __insn_mfspr(
						SPR_EX_CONTEXT_0_0);
					unsigned long ex0_1 = __insn_mfspr(
						SPR_EX_CONTEXT_0_1);
					/*
					 * Special-case it if we're iret'ing
					 * to PL0 again.  Otherwise just let
					 * it run and it will generate SIGILL.
					 */
					if (EX1_PL(ex0_1) == USER_PL) {
						state->next_pc = ex0_0;
						regs->ex1 = ex0_1;
						bundle = nop_X1(bundle);
					}
				}
				}
			}
			break;

#if CHIP_HAS_WH64()
		/* postincrement operations */
		case IMM_0_OPCODE_X1:
			switch (get_ImmOpcodeExtension_X1(bundle)) {
			case LWADD_IMM_0_OPCODE_X1:
				mem_op = MEMOP_LOAD_POSTINCR;
				size = 4;
				break;

			case LHADD_IMM_0_OPCODE_X1:
				mem_op = MEMOP_LOAD_POSTINCR;
				size = 2;
				sign_ext = 1;
				break;

			case LHADD_U_IMM_0_OPCODE_X1:
				mem_op = MEMOP_LOAD_POSTINCR;
				size = 2;
				sign_ext = 0;
				break;

			case SWADD_IMM_0_OPCODE_X1:
				mem_op = MEMOP_STORE_POSTINCR;
				size = 4;
				break;

			case SHADD_IMM_0_OPCODE_X1:
				mem_op = MEMOP_STORE_POSTINCR;
				size = 2;
				break;

			default:
				break;
			}
			break;
#endif /* CHIP_HAS_WH64() */
		}

		if (state->update) {
			/*
			 * Get an available register.  We start with a
			 * bitmask with 1's for available registers.
			 * We truncate to the low 32 registers since
			 * we are guaranteed to have set bits in the
			 * low 32 bits, then use ctz to pick the first.
			 */
			u32 mask = (u32) ~((1ULL << get_Dest_X0(bundle)) |
					   (1ULL << get_SrcA_X0(bundle)) |
					   (1ULL << get_SrcB_X0(bundle)) |
					   (1ULL << target_reg));
			temp_reg = __builtin_ctz(mask);
			state->update_reg = temp_reg;
			state->update_value = regs->regs[temp_reg];
			regs->regs[temp_reg] = (unsigned long) (pc+1);
			regs->flags |= PT_FLAGS_RESTORE_REGS;
			bundle = move_X1(bundle, target_reg, temp_reg);
		}
	} else {
		int opcode = get_Opcode_Y2(bundle);

		switch (opcode) {
		/* loads */
		case LH_OPCODE_Y2:
			mem_op = MEMOP_LOAD;
			size = 2;
			sign_ext = 1;
			break;

		case LH_U_OPCODE_Y2:
			mem_op = MEMOP_LOAD;
			size = 2;
			sign_ext = 0;
			break;

		case LW_OPCODE_Y2:
			mem_op = MEMOP_LOAD;
			size = 4;
			break;

		/* stores */
		case SH_OPCODE_Y2:
			mem_op = MEMOP_STORE;
			size = 2;
			break;

		case SW_OPCODE_Y2:
			mem_op = MEMOP_STORE;
			size = 4;
			break;
		}
	}

	/*
	 * Check if we need to rewrite an unaligned load/store.
	 * Returning zero is a special value meaning we need to SIGSEGV.
	 */
	if (mem_op != MEMOP_NONE && unaligned_fixup >= 0) {
		bundle = rewrite_load_store_unaligned(state, bundle, regs,
						      mem_op, size, sign_ext);
		if (bundle == 0)
			return;
	}

	/* write the bundle to our execution area */
	buffer = state->buffer;
	err = __put_user(bundle, buffer++);

	/*
	 * If we're really single-stepping, we take an INT_ILL after.
	 * If we're just handling an unaligned access, we can just
	 * jump directly back to where we were in user code.
	 */
	if (is_single_step) {
		err |= __put_user(__single_step_ill_insn, buffer++);
		err |= __put_user(__single_step_ill_insn, buffer++);
	} else {
		long delta;

		if (state->update) {
			/* We have some state to update; do it inline */
			int ha16;
			bundle = __single_step_addli_insn;
			bundle |= create_Dest_X1(state->update_reg);
			bundle |= create_Imm16_X1(state->update_value);
			err |= __put_user(bundle, buffer++);
			bundle = __single_step_auli_insn;
			bundle |= create_Dest_X1(state->update_reg);
			bundle |= create_SrcA_X1(state->update_reg);
			ha16 = (state->update_value + 0x8000) >> 16;
			bundle |= create_Imm16_X1(ha16);
			err |= __put_user(bundle, buffer++);
			state->update = 0;
		}

		/* End with a jump back to the next instruction */
		delta = ((regs->pc + TILE_BUNDLE_SIZE_IN_BYTES) -
			(unsigned long)buffer) >>
			TILE_LOG2_BUNDLE_ALIGNMENT_IN_BYTES;
		bundle = __single_step_j_insn;
		bundle |= create_JOffLong_X1(delta);
		err |= __put_user(bundle, buffer++);
	}

	if (err) {
		printk("Fault when writing to single-step buffer\n");
		return;
	}

	/*
	 * Flush the buffer.
	 * We do a local flush only, since this is a thread-specific buffer.
	 */
	__flush_icache_range((unsigned long) state->buffer,
			     (unsigned long) buffer);

	/* Indicate enabled */
	state->is_enabled = is_single_step;
	regs->pc = (unsigned long) state->buffer;

	/* Fault immediately if we are coming back from a syscall. */
	if (regs->faultnum == INT_SWINT_1)
		regs->pc += 8;
}

#endif /* !__tilegx__ */