xref: /illumos-gate/usr/src/uts/intel/dtrace/fbt.c (revision dbdc225a81ccef01e9d416169099b09ddbc06ea1)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 
27 #include <sys/modctl.h>
28 #include <sys/dtrace.h>
29 #include <sys/kobj.h>
30 #include <sys/stat.h>
31 #include <sys/ddi.h>
32 #include <sys/sunddi.h>
33 #include <sys/conf.h>
34 
35 #define	FBT_PUSHL_EBP		0x55
36 #define	FBT_MOVL_ESP_EBP0_V0	0x8b
37 #define	FBT_MOVL_ESP_EBP1_V0	0xec
38 #define	FBT_MOVL_ESP_EBP0_V1	0x89
39 #define	FBT_MOVL_ESP_EBP1_V1	0xe5
40 #define	FBT_REX_RSP_RBP		0x48
41 
42 #define	FBT_POPL_EBP		0x5d
43 #define	FBT_RET			0xc3
44 #define	FBT_RET_IMM16		0xc2
45 #define	FBT_LEAVE		0xc9
46 
47 #ifdef __amd64
48 #define	FBT_PATCHVAL		0xcc
49 #else
50 #define	FBT_PATCHVAL		0xf0
51 #endif
52 
53 #define	FBT_ENTRY	"entry"
54 #define	FBT_RETURN	"return"
55 #define	FBT_ADDR2NDX(addr)	((((uintptr_t)(addr)) >> 4) & fbt_probetab_mask)
56 #define	FBT_PROBETAB_SIZE	0x8000		/* 32k entries -- 128K total */
57 
58 typedef struct fbt_probe {
59 	struct fbt_probe *fbtp_hashnext;
60 	uint8_t		*fbtp_patchpoint;
61 	int8_t		fbtp_rval;
62 	uint8_t		fbtp_patchval;
63 	uint8_t		fbtp_savedval;
64 	uintptr_t	fbtp_roffset;
65 	dtrace_id_t	fbtp_id;
66 	char		*fbtp_name;
67 	struct modctl	*fbtp_ctl;
68 	int		fbtp_loadcnt;
69 	int		fbtp_symndx;
70 	int		fbtp_primary;
71 	struct fbt_probe *fbtp_next;
72 } fbt_probe_t;
73 
74 static dev_info_t		*fbt_devi;
75 static dtrace_provider_id_t	fbt_id;
76 static fbt_probe_t		**fbt_probetab;
77 static int			fbt_probetab_size;
78 static int			fbt_probetab_mask;
79 static int			fbt_verbose = 0;
80 
81 static int
82 fbt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t rval)
83 {
84 	uintptr_t stack0, stack1, stack2, stack3, stack4;
85 	fbt_probe_t *fbt = fbt_probetab[FBT_ADDR2NDX(addr)];
86 
87 	for (; fbt != NULL; fbt = fbt->fbtp_hashnext) {
88 		if ((uintptr_t)fbt->fbtp_patchpoint == addr) {
89 			if (fbt->fbtp_roffset == 0) {
90 				int i = 0;
91 				/*
92 				 * When accessing the arguments on the stack,
93 				 * we must protect against accessing beyond
94 				 * the stack.  We can safely set NOFAULT here
95 				 * -- we know that interrupts are already
96 				 * disabled.
97 				 */
98 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
99 				CPU->cpu_dtrace_caller = stack[i++];
100 #ifdef __amd64
101 				/*
102 				 * On amd64, stack[0] contains the dereferenced
103 				 * stack pointer, stack[1] contains savfp,
104 				 * stack[2] contains savpc.  We want to step
105 				 * over these entries.
106 				 */
107 				i += 2;
108 #endif
109 				stack0 = stack[i++];
110 				stack1 = stack[i++];
111 				stack2 = stack[i++];
112 				stack3 = stack[i++];
113 				stack4 = stack[i++];
114 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT |
115 				    CPU_DTRACE_BADADDR);
116 
117 				dtrace_probe(fbt->fbtp_id, stack0, stack1,
118 				    stack2, stack3, stack4);
119 
120 				CPU->cpu_dtrace_caller = 0;
121 			} else {
122 #ifdef __amd64
123 				/*
124 				 * On amd64, we instrument the ret, not the
125 				 * leave.  We therefore need to set the caller
126 				 * to assure that the top frame of a stack()
127 				 * action is correct.
128 				 */
129 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
130 				CPU->cpu_dtrace_caller = stack[0];
131 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT |
132 				    CPU_DTRACE_BADADDR);
133 #endif
134 
135 				dtrace_probe(fbt->fbtp_id, fbt->fbtp_roffset,
136 				    rval, 0, 0, 0);
137 				CPU->cpu_dtrace_caller = 0;
138 			}
139 
140 			return (fbt->fbtp_rval);
141 		}
142 	}
143 
144 	return (0);
145 }
146 
147 /*ARGSUSED*/
148 static void
149 fbt_provide_module(void *arg, struct modctl *ctl)
150 {
151 	struct module *mp = ctl->mod_mp;
152 	char *str = mp->strings;
153 	int nsyms = mp->nsyms;
154 	Shdr *symhdr = mp->symhdr;
155 	char *modname = ctl->mod_modname;
156 	char *name;
157 	fbt_probe_t *fbt, *retfbt;
158 	size_t symsize;
159 	int i, size;
160 
161 	/*
162 	 * Employees of dtrace and their families are ineligible.  Void
163 	 * where prohibited.
164 	 */
165 	if (strcmp(modname, "dtrace") == 0)
166 		return;
167 
168 	if (ctl->mod_requisites != NULL) {
169 		struct modctl_list *list;
170 
171 		list = (struct modctl_list *)ctl->mod_requisites;
172 
173 		for (; list != NULL; list = list->modl_next) {
174 			if (strcmp(list->modl_modp->mod_modname, "dtrace") == 0)
175 				return;
176 		}
177 	}
178 
179 	/*
180 	 * KMDB is ineligible for instrumentation -- it may execute in
181 	 * any context, including probe context.
182 	 */
183 	if (strcmp(modname, "kmdbmod") == 0)
184 		return;
185 
186 	if (str == NULL || symhdr == NULL || symhdr->sh_addr == 0) {
187 		/*
188 		 * If this module doesn't (yet) have its string or symbol
189 		 * table allocated, clear out.
190 		 */
191 		return;
192 	}
193 
194 	symsize = symhdr->sh_entsize;
195 
196 	if (mp->fbt_nentries) {
197 		/*
198 		 * This module has some FBT entries allocated; we're afraid
199 		 * to screw with it.
200 		 */
201 		return;
202 	}
203 
204 	for (i = 1; i < nsyms; i++) {
205 		uint8_t *instr, *limit;
206 		Sym *sym = (Sym *)(symhdr->sh_addr + i * symsize);
207 		int j;
208 
209 		if (ELF_ST_TYPE(sym->st_info) != STT_FUNC)
210 			continue;
211 
212 		/*
213 		 * Weak symbols are not candidates.  This could be made to
214 		 * work (where weak functions and their underlying function
215 		 * appear as two disjoint probes), but it's not simple.
216 		 */
217 		if (ELF_ST_BIND(sym->st_info) == STB_WEAK)
218 			continue;
219 
220 		name = str + sym->st_name;
221 
222 		if (strstr(name, "dtrace_") == name &&
223 		    strstr(name, "dtrace_safe_") != name) {
224 			/*
225 			 * Anything beginning with "dtrace_" may be called
226 			 * from probe context unless it explitly indicates
227 			 * that it won't be called from probe context by
228 			 * using the prefix "dtrace_safe_".
229 			 */
230 			continue;
231 		}
232 
233 		if (strstr(name, "kdi_") == name ||
234 		    strstr(name, "_kdi_") != NULL) {
235 			/*
236 			 * Any function name beginning with "kdi_" or
237 			 * containing the string "_kdi_" is a part of the
238 			 * kernel debugger interface and may be called in
239 			 * arbitrary context -- including probe context.
240 			 */
241 			continue;
242 		}
243 
244 		/*
245 		 * Due to 4524008, _init and _fini may have a bloated st_size.
246 		 * While this bug was fixed quite some time ago, old drivers
247 		 * may be lurking.  We need to develop a better solution to
248 		 * this problem, such that correct _init and _fini functions
249 		 * (the vast majority) may be correctly traced.  One solution
250 		 * may be to scan through the entire symbol table to see if
251 		 * any symbol overlaps with _init.  If none does, set a bit in
252 		 * the module structure that this module has correct _init and
253 		 * _fini sizes.  This will cause some pain the first time a
254 		 * module is scanned, but at least it would be O(N) instead of
255 		 * O(N log N)...
256 		 */
257 		if (strcmp(name, "_init") == 0)
258 			continue;
259 
260 		if (strcmp(name, "_fini") == 0)
261 			continue;
262 
263 		/*
264 		 * In order to be eligible, the function must begin with the
265 		 * following sequence:
266 		 *
267 		 *	pushl	%esp
268 		 *	movl	%esp, %ebp
269 		 *
270 		 * Note that there are two variants of encodings that generate
271 		 * the movl; we must check for both.  For 64-bit, we would
272 		 * normally insist that a function begin with the following
273 		 * sequence:
274 		 *
275 		 *	pushq	%rbp
276 		 *	movq	%rsp, %rbp
277 		 *
278 		 * However, the compiler for 64-bit often splits these two
279 		 * instructions -- and the first instruction in the function
280 		 * is often not the pushq.  As a result, on 64-bit we look
281 		 * for any "pushq %rbp" in the function and we instrument
282 		 * this with a breakpoint instruction.
283 		 */
284 		instr = (uint8_t *)sym->st_value;
285 		limit = (uint8_t *)(sym->st_value + sym->st_size);
286 
287 #ifdef __amd64
288 		while (instr < limit) {
289 			if (*instr == FBT_PUSHL_EBP)
290 				break;
291 
292 			if ((size = dtrace_instr_size(instr)) <= 0)
293 				break;
294 
295 			instr += size;
296 		}
297 
298 		if (instr >= limit || *instr != FBT_PUSHL_EBP) {
299 			/*
300 			 * We either don't save the frame pointer in this
301 			 * function, or we ran into some disassembly
302 			 * screw-up.  Either way, we bail.
303 			 */
304 			continue;
305 		}
306 #else
307 		if (instr[0] != FBT_PUSHL_EBP)
308 			continue;
309 
310 		if (!(instr[1] == FBT_MOVL_ESP_EBP0_V0 &&
311 		    instr[2] == FBT_MOVL_ESP_EBP1_V0) &&
312 		    !(instr[1] == FBT_MOVL_ESP_EBP0_V1 &&
313 		    instr[2] == FBT_MOVL_ESP_EBP1_V1))
314 			continue;
315 #endif
316 
317 		fbt = kmem_zalloc(sizeof (fbt_probe_t), KM_SLEEP);
318 		fbt->fbtp_name = name;
319 		fbt->fbtp_id = dtrace_probe_create(fbt_id, modname,
320 		    name, FBT_ENTRY, 3, fbt);
321 		fbt->fbtp_patchpoint = instr;
322 		fbt->fbtp_ctl = ctl;
323 		fbt->fbtp_loadcnt = ctl->mod_loadcnt;
324 		fbt->fbtp_rval = DTRACE_INVOP_PUSHL_EBP;
325 		fbt->fbtp_savedval = *instr;
326 		fbt->fbtp_patchval = FBT_PATCHVAL;
327 
328 		fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)];
329 		fbt->fbtp_symndx = i;
330 		fbt_probetab[FBT_ADDR2NDX(instr)] = fbt;
331 
332 		mp->fbt_nentries++;
333 
334 		retfbt = NULL;
335 again:
336 		if (instr >= limit)
337 			continue;
338 
339 		/*
340 		 * If this disassembly fails, then we've likely walked off into
341 		 * a jump table or some other unsuitable area.  Bail out of the
342 		 * disassembly now.
343 		 */
344 		if ((size = dtrace_instr_size(instr)) <= 0)
345 			continue;
346 
347 #ifdef __amd64
348 		/*
349 		 * We only instrument "ret" on amd64 -- we don't yet instrument
350 		 * ret imm16, largely because the compiler doesn't seem to
351 		 * (yet) emit them in the kernel...
352 		 */
353 		if (*instr != FBT_RET) {
354 			instr += size;
355 			goto again;
356 		}
357 #else
358 		if (!(size == 1 &&
359 		    (*instr == FBT_POPL_EBP || *instr == FBT_LEAVE) &&
360 		    (*(instr + 1) == FBT_RET ||
361 		    *(instr + 1) == FBT_RET_IMM16))) {
362 			instr += size;
363 			goto again;
364 		}
365 #endif
366 
367 		/*
368 		 * We (desperately) want to avoid erroneously instrumenting a
369 		 * jump table, especially given that our markers are pretty
370 		 * short:  two bytes on x86, and just one byte on amd64.  To
371 		 * determine if we're looking at a true instruction sequence
372 		 * or an inline jump table that happens to contain the same
373 		 * byte sequences, we resort to some heuristic sleeze:  we
374 		 * treat this instruction as being contained within a pointer,
375 		 * and see if that pointer points to within the body of the
376 		 * function.  If it does, we refuse to instrument it.
377 		 */
378 		for (j = 0; j < sizeof (uintptr_t); j++) {
379 			uintptr_t check = (uintptr_t)instr - j;
380 			uint8_t *ptr;
381 
382 			if (check < sym->st_value)
383 				break;
384 
385 			if (check + sizeof (uintptr_t) > (uintptr_t)limit)
386 				continue;
387 
388 			ptr = *(uint8_t **)check;
389 
390 			if (ptr >= (uint8_t *)sym->st_value && ptr < limit) {
391 				instr += size;
392 				goto again;
393 			}
394 		}
395 
396 		/*
397 		 * We have a winner!
398 		 */
399 		fbt = kmem_zalloc(sizeof (fbt_probe_t), KM_SLEEP);
400 		fbt->fbtp_name = name;
401 
402 		if (retfbt == NULL) {
403 			fbt->fbtp_id = dtrace_probe_create(fbt_id, modname,
404 			    name, FBT_RETURN, 3, fbt);
405 		} else {
406 			retfbt->fbtp_next = fbt;
407 			fbt->fbtp_id = retfbt->fbtp_id;
408 		}
409 
410 		retfbt = fbt;
411 		fbt->fbtp_patchpoint = instr;
412 		fbt->fbtp_ctl = ctl;
413 		fbt->fbtp_loadcnt = ctl->mod_loadcnt;
414 
415 #ifndef __amd64
416 		if (*instr == FBT_POPL_EBP) {
417 			fbt->fbtp_rval = DTRACE_INVOP_POPL_EBP;
418 		} else {
419 			ASSERT(*instr == FBT_LEAVE);
420 			fbt->fbtp_rval = DTRACE_INVOP_LEAVE;
421 		}
422 		fbt->fbtp_roffset =
423 		    (uintptr_t)(instr - (uint8_t *)sym->st_value) + 1;
424 
425 #else
426 		ASSERT(*instr == FBT_RET);
427 		fbt->fbtp_rval = DTRACE_INVOP_RET;
428 		fbt->fbtp_roffset =
429 		    (uintptr_t)(instr - (uint8_t *)sym->st_value);
430 #endif
431 
432 		fbt->fbtp_savedval = *instr;
433 		fbt->fbtp_patchval = FBT_PATCHVAL;
434 		fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)];
435 		fbt->fbtp_symndx = i;
436 		fbt_probetab[FBT_ADDR2NDX(instr)] = fbt;
437 
438 		mp->fbt_nentries++;
439 
440 		instr += size;
441 		goto again;
442 	}
443 }
444 
445 /*ARGSUSED*/
446 static void
447 fbt_destroy(void *arg, dtrace_id_t id, void *parg)
448 {
449 	fbt_probe_t *fbt = parg, *next, *hash, *last;
450 	struct modctl *ctl = fbt->fbtp_ctl;
451 	int ndx;
452 
453 	do {
454 		if (ctl != NULL && ctl->mod_loadcnt == fbt->fbtp_loadcnt) {
455 			if ((ctl->mod_loadcnt == fbt->fbtp_loadcnt &&
456 			    ctl->mod_loaded)) {
457 				((struct module *)
458 				    (ctl->mod_mp))->fbt_nentries--;
459 			}
460 		}
461 
462 		/*
463 		 * Now we need to remove this probe from the fbt_probetab.
464 		 */
465 		ndx = FBT_ADDR2NDX(fbt->fbtp_patchpoint);
466 		last = NULL;
467 		hash = fbt_probetab[ndx];
468 
469 		while (hash != fbt) {
470 			ASSERT(hash != NULL);
471 			last = hash;
472 			hash = hash->fbtp_hashnext;
473 		}
474 
475 		if (last != NULL) {
476 			last->fbtp_hashnext = fbt->fbtp_hashnext;
477 		} else {
478 			fbt_probetab[ndx] = fbt->fbtp_hashnext;
479 		}
480 
481 		next = fbt->fbtp_next;
482 		kmem_free(fbt, sizeof (fbt_probe_t));
483 
484 		fbt = next;
485 	} while (fbt != NULL);
486 }
487 
488 /*ARGSUSED*/
489 static int
490 fbt_enable(void *arg, dtrace_id_t id, void *parg)
491 {
492 	fbt_probe_t *fbt = parg;
493 	struct modctl *ctl = fbt->fbtp_ctl;
494 
495 	ctl->mod_nenabled++;
496 
497 	if (!ctl->mod_loaded) {
498 		if (fbt_verbose) {
499 			cmn_err(CE_NOTE, "fbt is failing for probe %s "
500 			    "(module %s unloaded)",
501 			    fbt->fbtp_name, ctl->mod_modname);
502 		}
503 
504 		return (0);
505 	}
506 
507 	/*
508 	 * Now check that our modctl has the expected load count.  If it
509 	 * doesn't, this module must have been unloaded and reloaded -- and
510 	 * we're not going to touch it.
511 	 */
512 	if (ctl->mod_loadcnt != fbt->fbtp_loadcnt) {
513 		if (fbt_verbose) {
514 			cmn_err(CE_NOTE, "fbt is failing for probe %s "
515 			    "(module %s reloaded)",
516 			    fbt->fbtp_name, ctl->mod_modname);
517 		}
518 
519 		return (0);
520 	}
521 
522 	for (; fbt != NULL; fbt = fbt->fbtp_next)
523 		*fbt->fbtp_patchpoint = fbt->fbtp_patchval;
524 
525 	return (0);
526 }
527 
528 /*ARGSUSED*/
529 static void
530 fbt_disable(void *arg, dtrace_id_t id, void *parg)
531 {
532 	fbt_probe_t *fbt = parg;
533 	struct modctl *ctl = fbt->fbtp_ctl;
534 
535 	ASSERT(ctl->mod_nenabled > 0);
536 	ctl->mod_nenabled--;
537 
538 	if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt))
539 		return;
540 
541 	for (; fbt != NULL; fbt = fbt->fbtp_next)
542 		*fbt->fbtp_patchpoint = fbt->fbtp_savedval;
543 }
544 
545 /*ARGSUSED*/
546 static void
547 fbt_suspend(void *arg, dtrace_id_t id, void *parg)
548 {
549 	fbt_probe_t *fbt = parg;
550 	struct modctl *ctl = fbt->fbtp_ctl;
551 
552 	ASSERT(ctl->mod_nenabled > 0);
553 
554 	if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt))
555 		return;
556 
557 	for (; fbt != NULL; fbt = fbt->fbtp_next)
558 		*fbt->fbtp_patchpoint = fbt->fbtp_savedval;
559 }
560 
561 /*ARGSUSED*/
562 static void
563 fbt_resume(void *arg, dtrace_id_t id, void *parg)
564 {
565 	fbt_probe_t *fbt = parg;
566 	struct modctl *ctl = fbt->fbtp_ctl;
567 
568 	ASSERT(ctl->mod_nenabled > 0);
569 
570 	if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt))
571 		return;
572 
573 	for (; fbt != NULL; fbt = fbt->fbtp_next)
574 		*fbt->fbtp_patchpoint = fbt->fbtp_patchval;
575 }
576 
577 /*ARGSUSED*/
578 static void
579 fbt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc)
580 {
581 	fbt_probe_t *fbt = parg;
582 	struct modctl *ctl = fbt->fbtp_ctl;
583 	struct module *mp = ctl->mod_mp;
584 	ctf_file_t *fp = NULL, *pfp;
585 	ctf_funcinfo_t f;
586 	int error;
587 	ctf_id_t argv[32], type;
588 	int argc = sizeof (argv) / sizeof (ctf_id_t);
589 	const char *parent;
590 
591 	if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt))
592 		goto err;
593 
594 	if (fbt->fbtp_roffset != 0 && desc->dtargd_ndx == 0) {
595 		(void) strcpy(desc->dtargd_native, "int");
596 		return;
597 	}
598 
599 	if ((fp = ctf_modopen(mp, &error)) == NULL) {
600 		/*
601 		 * We have no CTF information for this module -- and therefore
602 		 * no args[] information.
603 		 */
604 		goto err;
605 	}
606 
607 	/*
608 	 * If we have a parent container, we must manually import it.
609 	 */
610 	if ((parent = ctf_parent_name(fp)) != NULL) {
611 		struct modctl *mp = &modules;
612 		struct modctl *mod = NULL;
613 
614 		/*
615 		 * We must iterate over all modules to find the module that
616 		 * is our parent.
617 		 */
618 		do {
619 			if (strcmp(mp->mod_modname, parent) == 0) {
620 				mod = mp;
621 				break;
622 			}
623 		} while ((mp = mp->mod_next) != &modules);
624 
625 		if (mod == NULL)
626 			goto err;
627 
628 		if ((pfp = ctf_modopen(mod->mod_mp, &error)) == NULL) {
629 			goto err;
630 		}
631 
632 		if (ctf_import(fp, pfp) != 0) {
633 			ctf_close(pfp);
634 			goto err;
635 		}
636 
637 		ctf_close(pfp);
638 	}
639 
640 	if (ctf_func_info(fp, fbt->fbtp_symndx, &f) == CTF_ERR)
641 		goto err;
642 
643 	if (fbt->fbtp_roffset != 0) {
644 		if (desc->dtargd_ndx > 1)
645 			goto err;
646 
647 		ASSERT(desc->dtargd_ndx == 1);
648 		type = f.ctc_return;
649 	} else {
650 		if (desc->dtargd_ndx + 1 > f.ctc_argc)
651 			goto err;
652 
653 		if (ctf_func_args(fp, fbt->fbtp_symndx, argc, argv) == CTF_ERR)
654 			goto err;
655 
656 		type = argv[desc->dtargd_ndx];
657 	}
658 
659 	if (ctf_type_name(fp, type, desc->dtargd_native,
660 	    DTRACE_ARGTYPELEN) != NULL) {
661 		ctf_close(fp);
662 		return;
663 	}
664 err:
665 	if (fp != NULL)
666 		ctf_close(fp);
667 
668 	desc->dtargd_ndx = DTRACE_ARGNONE;
669 }
670 
671 static dtrace_pattr_t fbt_attr = {
672 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA },
673 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
674 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
675 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA },
676 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
677 };
678 
679 static dtrace_pops_t fbt_pops = {
680 	NULL,
681 	fbt_provide_module,
682 	fbt_enable,
683 	fbt_disable,
684 	fbt_suspend,
685 	fbt_resume,
686 	fbt_getargdesc,
687 	NULL,
688 	NULL,
689 	fbt_destroy
690 };
691 
692 static void
693 fbt_cleanup(dev_info_t *devi)
694 {
695 	dtrace_invop_remove(fbt_invop);
696 	ddi_remove_minor_node(devi, NULL);
697 	kmem_free(fbt_probetab, fbt_probetab_size * sizeof (fbt_probe_t *));
698 	fbt_probetab = NULL;
699 	fbt_probetab_mask = 0;
700 }
701 
702 static int
703 fbt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
704 {
705 	switch (cmd) {
706 	case DDI_ATTACH:
707 		break;
708 	case DDI_RESUME:
709 		return (DDI_SUCCESS);
710 	default:
711 		return (DDI_FAILURE);
712 	}
713 
714 	if (fbt_probetab_size == 0)
715 		fbt_probetab_size = FBT_PROBETAB_SIZE;
716 
717 	fbt_probetab_mask = fbt_probetab_size - 1;
718 	fbt_probetab =
719 	    kmem_zalloc(fbt_probetab_size * sizeof (fbt_probe_t *), KM_SLEEP);
720 
721 	dtrace_invop_add(fbt_invop);
722 
723 	if (ddi_create_minor_node(devi, "fbt", S_IFCHR, 0,
724 	    DDI_PSEUDO, 0) == DDI_FAILURE ||
725 	    dtrace_register("fbt", &fbt_attr, DTRACE_PRIV_KERNEL, NULL,
726 	    &fbt_pops, NULL, &fbt_id) != 0) {
727 		fbt_cleanup(devi);
728 		return (DDI_FAILURE);
729 	}
730 
731 	ddi_report_dev(devi);
732 	fbt_devi = devi;
733 
734 	return (DDI_SUCCESS);
735 }
736 
737 static int
738 fbt_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
739 {
740 	switch (cmd) {
741 	case DDI_DETACH:
742 		break;
743 	case DDI_SUSPEND:
744 		return (DDI_SUCCESS);
745 	default:
746 		return (DDI_FAILURE);
747 	}
748 
749 	if (dtrace_unregister(fbt_id) != 0)
750 		return (DDI_FAILURE);
751 
752 	fbt_cleanup(devi);
753 
754 	return (DDI_SUCCESS);
755 }
756 
757 /*ARGSUSED*/
758 static int
759 fbt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
760 {
761 	int error;
762 
763 	switch (infocmd) {
764 	case DDI_INFO_DEVT2DEVINFO:
765 		*result = (void *)fbt_devi;
766 		error = DDI_SUCCESS;
767 		break;
768 	case DDI_INFO_DEVT2INSTANCE:
769 		*result = (void *)0;
770 		error = DDI_SUCCESS;
771 		break;
772 	default:
773 		error = DDI_FAILURE;
774 	}
775 	return (error);
776 }
777 
778 /*ARGSUSED*/
779 static int
780 fbt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
781 {
782 	return (0);
783 }
784 
785 static struct cb_ops fbt_cb_ops = {
786 	fbt_open,		/* open */
787 	nodev,			/* close */
788 	nulldev,		/* strategy */
789 	nulldev,		/* print */
790 	nodev,			/* dump */
791 	nodev,			/* read */
792 	nodev,			/* write */
793 	nodev,			/* ioctl */
794 	nodev,			/* devmap */
795 	nodev,			/* mmap */
796 	nodev,			/* segmap */
797 	nochpoll,		/* poll */
798 	ddi_prop_op,		/* cb_prop_op */
799 	0,			/* streamtab  */
800 	D_NEW | D_MP		/* Driver compatibility flag */
801 };
802 
803 static struct dev_ops fbt_ops = {
804 	DEVO_REV,		/* devo_rev */
805 	0,			/* refcnt */
806 	fbt_info,		/* get_dev_info */
807 	nulldev,		/* identify */
808 	nulldev,		/* probe */
809 	fbt_attach,		/* attach */
810 	fbt_detach,		/* detach */
811 	nodev,			/* reset */
812 	&fbt_cb_ops,		/* driver operations */
813 	NULL,			/* bus operations */
814 	nodev,			/* dev power */
815 	ddi_quiesce_not_needed,		/* quiesce */
816 };
817 
818 /*
819  * Module linkage information for the kernel.
820  */
821 static struct modldrv modldrv = {
822 	&mod_driverops,		/* module type (this is a pseudo driver) */
823 	"Function Boundary Tracing",	/* name of module */
824 	&fbt_ops,		/* driver ops */
825 };
826 
827 static struct modlinkage modlinkage = {
828 	MODREV_1,
829 	(void *)&modldrv,
830 	NULL
831 };
832 
833 int
834 _init(void)
835 {
836 	return (mod_install(&modlinkage));
837 }
838 
839 int
840 _info(struct modinfo *modinfop)
841 {
842 	return (mod_info(&modlinkage, modinfop));
843 }
844 
845 int
846 _fini(void)
847 {
848 	return (mod_remove(&modlinkage));
849 }
850