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/sunddi.h>
29 #include <sys/dtrace.h>
30 #include <sys/kobj.h>
31 #include <sys/stat.h>
32 #include <sys/conf.h>
33 #include <vm/seg_kmem.h>
34 #include <sys/stack.h>
35 #include <sys/frame.h>
36 #include <sys/dtrace_impl.h>
37 #include <sys/cmn_err.h>
38 #include <sys/sysmacros.h>
39 #include <sys/privregs.h>
40 #include <sys/sdt_impl.h>
41
42 #define SDT_PATCHVAL 0xf0
43 #define SDT_ADDR2NDX(addr) ((((uintptr_t)(addr)) >> 4) & sdt_probetab_mask)
44 #define SDT_PROBETAB_SIZE 0x1000 /* 4k entries -- 16K total */
45
46 static dev_info_t *sdt_devi;
47 static int sdt_verbose = 0;
48 static sdt_probe_t **sdt_probetab;
49 static int sdt_probetab_size;
50 static int sdt_probetab_mask;
51
52 /*ARGSUSED*/
53 static int
54 sdt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax)
55 {
56 uintptr_t stack0, stack1, stack2, stack3, stack4;
57 int i = 0;
58 sdt_probe_t *sdt = sdt_probetab[SDT_ADDR2NDX(addr)];
59
60 #ifdef __amd64
61 /*
62 * On amd64, stack[0] contains the dereferenced stack pointer,
63 * stack[1] contains savfp, stack[2] contains savpc. We want
64 * to step over these entries.
65 */
66 i += 3;
67 #endif
68
69 for (; sdt != NULL; sdt = sdt->sdp_hashnext) {
70 if ((uintptr_t)sdt->sdp_patchpoint == addr) {
71 /*
72 * When accessing the arguments on the stack, we must
73 * protect against accessing beyond the stack. We can
74 * safely set NOFAULT here -- we know that interrupts
75 * are already disabled.
76 */
77 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
78 stack0 = stack[i++];
79 stack1 = stack[i++];
80 stack2 = stack[i++];
81 stack3 = stack[i++];
82 stack4 = stack[i++];
83 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT |
84 CPU_DTRACE_BADADDR);
85
86 dtrace_probe(sdt->sdp_id, stack0, stack1,
87 stack2, stack3, stack4);
88
89 return (DTRACE_INVOP_NOP);
90 }
91 }
92
93 return (0);
94 }
95
96 /*ARGSUSED*/
97 static void
98 sdt_provide_module(void *arg, struct modctl *ctl)
99 {
100 struct module *mp = ctl->mod_mp;
101 char *modname = ctl->mod_modname;
102 sdt_probedesc_t *sdpd;
103 sdt_probe_t *sdp, *old;
104 sdt_provider_t *prov;
105 int len;
106
107 /*
108 * One for all, and all for one: if we haven't yet registered all of
109 * our providers, we'll refuse to provide anything.
110 */
111 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) {
112 if (prov->sdtp_id == DTRACE_PROVNONE)
113 return;
114 }
115
116 if (mp->sdt_nprobes != 0 || (sdpd = mp->sdt_probes) == NULL)
117 return;
118
119 for (sdpd = mp->sdt_probes; sdpd != NULL; sdpd = sdpd->sdpd_next) {
120 char *name = sdpd->sdpd_name, *func, *nname;
121 int i, j;
122 sdt_provider_t *prov;
123 ulong_t offs;
124 dtrace_id_t id;
125
126 for (prov = sdt_providers; prov->sdtp_prefix != NULL; prov++) {
127 char *prefix = prov->sdtp_prefix;
128
129 if (strncmp(name, prefix, strlen(prefix)) == 0) {
130 name += strlen(prefix);
131 break;
132 }
133 }
134
135 nname = kmem_alloc(len = strlen(name) + 1, KM_SLEEP);
136
137 for (i = 0, j = 0; name[j] != '\0'; i++) {
138 if (name[j] == '_' && name[j + 1] == '_') {
139 nname[i] = '-';
140 j += 2;
141 } else {
142 nname[i] = name[j++];
143 }
144 }
145
146 nname[i] = '\0';
147
148 sdp = kmem_zalloc(sizeof (sdt_probe_t), KM_SLEEP);
149 sdp->sdp_loadcnt = ctl->mod_loadcnt;
150 sdp->sdp_ctl = ctl;
151 sdp->sdp_name = nname;
152 sdp->sdp_namelen = len;
153 sdp->sdp_provider = prov;
154
155 func = kobj_searchsym(mp, sdpd->sdpd_offset, &offs);
156
157 if (func == NULL)
158 func = "<unknown>";
159
160 /*
161 * We have our provider. Now create the probe.
162 */
163 if ((id = dtrace_probe_lookup(prov->sdtp_id, modname,
164 func, nname)) != DTRACE_IDNONE) {
165 old = dtrace_probe_arg(prov->sdtp_id, id);
166 ASSERT(old != NULL);
167
168 sdp->sdp_next = old->sdp_next;
169 sdp->sdp_id = id;
170 old->sdp_next = sdp;
171 } else {
172 sdp->sdp_id = dtrace_probe_create(prov->sdtp_id,
173 modname, func, nname, 3, sdp);
174
175 mp->sdt_nprobes++;
176 }
177
178 sdp->sdp_hashnext =
179 sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)];
180 sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)] = sdp;
181
182 sdp->sdp_patchval = SDT_PATCHVAL;
183 sdp->sdp_patchpoint = (uint8_t *)sdpd->sdpd_offset;
184 sdp->sdp_savedval = *sdp->sdp_patchpoint;
185 }
186 }
187
188 /*ARGSUSED*/
189 static void
190 sdt_destroy(void *arg, dtrace_id_t id, void *parg)
191 {
192 sdt_probe_t *sdp = parg, *old, *last, *hash;
193 struct modctl *ctl = sdp->sdp_ctl;
194 int ndx;
195
196 if (ctl != NULL && ctl->mod_loadcnt == sdp->sdp_loadcnt) {
197 if ((ctl->mod_loadcnt == sdp->sdp_loadcnt &&
198 ctl->mod_loaded)) {
199 ((struct module *)(ctl->mod_mp))->sdt_nprobes--;
200 }
201 }
202
203 while (sdp != NULL) {
204 old = sdp;
205
206 /*
207 * Now we need to remove this probe from the sdt_probetab.
208 */
209 ndx = SDT_ADDR2NDX(sdp->sdp_patchpoint);
210 last = NULL;
211 hash = sdt_probetab[ndx];
212
213 while (hash != sdp) {
214 ASSERT(hash != NULL);
215 last = hash;
216 hash = hash->sdp_hashnext;
217 }
218
219 if (last != NULL) {
220 last->sdp_hashnext = sdp->sdp_hashnext;
221 } else {
222 sdt_probetab[ndx] = sdp->sdp_hashnext;
223 }
224
225 kmem_free(sdp->sdp_name, sdp->sdp_namelen);
226 sdp = sdp->sdp_next;
227 kmem_free(old, sizeof (sdt_probe_t));
228 }
229 }
230
231 /*ARGSUSED*/
232 static int
233 sdt_enable(void *arg, dtrace_id_t id, void *parg)
234 {
235 sdt_probe_t *sdp = parg;
236 struct modctl *ctl = sdp->sdp_ctl;
237
238 ctl->mod_nenabled++;
239
240 /*
241 * If this module has disappeared since we discovered its probes,
242 * refuse to enable it.
243 */
244 if (!ctl->mod_loaded) {
245 if (sdt_verbose) {
246 cmn_err(CE_NOTE, "sdt is failing for probe %s "
247 "(module %s unloaded)",
248 sdp->sdp_name, ctl->mod_modname);
249 }
250 goto err;
251 }
252
253 /*
254 * Now check that our modctl has the expected load count. If it
255 * doesn't, this module must have been unloaded and reloaded -- and
256 * we're not going to touch it.
257 */
258 if (ctl->mod_loadcnt != sdp->sdp_loadcnt) {
259 if (sdt_verbose) {
260 cmn_err(CE_NOTE, "sdt is failing for probe %s "
261 "(module %s reloaded)",
262 sdp->sdp_name, ctl->mod_modname);
263 }
264 goto err;
265 }
266
267 while (sdp != NULL) {
268 *sdp->sdp_patchpoint = sdp->sdp_patchval;
269 sdp = sdp->sdp_next;
270 }
271 err:
272 return (0);
273 }
274
275 /*ARGSUSED*/
276 static void
277 sdt_disable(void *arg, dtrace_id_t id, void *parg)
278 {
279 sdt_probe_t *sdp = parg;
280 struct modctl *ctl = sdp->sdp_ctl;
281
282 ctl->mod_nenabled--;
283
284 if (!ctl->mod_loaded || ctl->mod_loadcnt != sdp->sdp_loadcnt)
285 goto err;
286
287 while (sdp != NULL) {
288 *sdp->sdp_patchpoint = sdp->sdp_savedval;
289 sdp = sdp->sdp_next;
290 }
291
292 err:
293 ;
294 }
295
296 /*ARGSUSED*/
297 uint64_t
298 sdt_getarg(void *arg, dtrace_id_t id, void *parg, int argno, int aframes)
299 {
300 uintptr_t val;
301 struct frame *fp = (struct frame *)dtrace_getfp();
302 uintptr_t *stack;
303 int i;
304 #if defined(__amd64)
305 /*
306 * A total of 6 arguments are passed via registers; any argument with
307 * index of 5 or lower is therefore in a register.
308 */
309 int inreg = 5;
310 #endif
311
312 for (i = 1; i <= aframes; i++) {
313 fp = (struct frame *)(fp->fr_savfp);
314
315 if (fp->fr_savpc == (pc_t)dtrace_invop_callsite) {
316 #if !defined(__amd64)
317 /*
318 * If we pass through the invalid op handler, we will
319 * use the pointer that it passed to the stack as the
320 * second argument to dtrace_invop() as the pointer to
321 * the stack.
322 */
323 stack = ((uintptr_t **)&fp[1])[1];
324 #else
325 /*
326 * In the case of amd64, we will use the pointer to the
327 * regs structure that was pushed when we took the
328 * trap. To get this structure, we must increment
329 * beyond the frame structure. If the argument that
330 * we're seeking is passed on the stack, we'll pull
331 * the true stack pointer out of the saved registers
332 * and decrement our argument by the number of
333 * arguments passed in registers; if the argument
334 * we're seeking is passed in regsiters, we can just
335 * load it directly.
336 */
337 struct regs *rp = (struct regs *)((uintptr_t)&fp[1] +
338 sizeof (uintptr_t));
339
340 if (argno <= inreg) {
341 stack = (uintptr_t *)&rp->r_rdi;
342 } else {
343 stack = (uintptr_t *)(rp->r_rsp);
344 argno -= (inreg + 1);
345 }
346 #endif
347 goto load;
348 }
349 }
350
351 /*
352 * We know that we did not come through a trap to get into
353 * dtrace_probe() -- the provider simply called dtrace_probe()
354 * directly. As this is the case, we need to shift the argument
355 * that we're looking for: the probe ID is the first argument to
356 * dtrace_probe(), so the argument n will actually be found where
357 * one would expect to find argument (n + 1).
358 */
359 argno++;
360
361 #if defined(__amd64)
362 if (argno <= inreg) {
363 /*
364 * This shouldn't happen. If the argument is passed in a
365 * register then it should have been, well, passed in a
366 * register...
367 */
368 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
369 return (0);
370 }
371
372 argno -= (inreg + 1);
373 #endif
374 stack = (uintptr_t *)&fp[1];
375
376 load:
377 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
378 val = stack[argno];
379 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
380
381 return (val);
382 }
383
384 static dtrace_pops_t sdt_pops = {
385 NULL,
386 sdt_provide_module,
387 sdt_enable,
388 sdt_disable,
389 NULL,
390 NULL,
391 sdt_getargdesc,
392 sdt_getarg,
393 NULL,
394 sdt_destroy
395 };
396
397 /*ARGSUSED*/
398 static int
399 sdt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
400 {
401 sdt_provider_t *prov;
402
403 if (ddi_create_minor_node(devi, "sdt", S_IFCHR,
404 0, DDI_PSEUDO, NULL) == DDI_FAILURE) {
405 cmn_err(CE_NOTE, "/dev/sdt couldn't create minor node");
406 ddi_remove_minor_node(devi, NULL);
407 return (DDI_FAILURE);
408 }
409
410 ddi_report_dev(devi);
411 sdt_devi = devi;
412
413 if (sdt_probetab_size == 0)
414 sdt_probetab_size = SDT_PROBETAB_SIZE;
415
416 sdt_probetab_mask = sdt_probetab_size - 1;
417 sdt_probetab =
418 kmem_zalloc(sdt_probetab_size * sizeof (sdt_probe_t *), KM_SLEEP);
419 dtrace_invop_add(sdt_invop);
420
421 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) {
422 if (dtrace_register(prov->sdtp_name, prov->sdtp_attr,
423 DTRACE_PRIV_KERNEL, NULL,
424 &sdt_pops, prov, &prov->sdtp_id) != 0) {
425 cmn_err(CE_WARN, "failed to register sdt provider %s",
426 prov->sdtp_name);
427 }
428 }
429
430 return (DDI_SUCCESS);
431 }
432
433 /*ARGSUSED*/
434 static int
435 sdt_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
436 {
437 sdt_provider_t *prov;
438
439 switch (cmd) {
440 case DDI_DETACH:
441 break;
442
443 case DDI_SUSPEND:
444 return (DDI_SUCCESS);
445
446 default:
447 return (DDI_FAILURE);
448 }
449
450 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) {
451 if (prov->sdtp_id != DTRACE_PROVNONE) {
452 if (dtrace_unregister(prov->sdtp_id) != 0)
453 return (DDI_FAILURE);
454
455 prov->sdtp_id = DTRACE_PROVNONE;
456 }
457 }
458
459 dtrace_invop_remove(sdt_invop);
460 kmem_free(sdt_probetab, sdt_probetab_size * sizeof (sdt_probe_t *));
461
462 return (DDI_SUCCESS);
463 }
464
465 /*ARGSUSED*/
466 static int
467 sdt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
468 {
469 int error;
470
471 switch (infocmd) {
472 case DDI_INFO_DEVT2DEVINFO:
473 *result = (void *)sdt_devi;
474 error = DDI_SUCCESS;
475 break;
476 case DDI_INFO_DEVT2INSTANCE:
477 *result = (void *)0;
478 error = DDI_SUCCESS;
479 break;
480 default:
481 error = DDI_FAILURE;
482 }
483 return (error);
484 }
485
486 /*ARGSUSED*/
487 static int
488 sdt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
489 {
490 return (0);
491 }
492
493 static struct cb_ops sdt_cb_ops = {
494 sdt_open, /* open */
495 nodev, /* close */
496 nulldev, /* strategy */
497 nulldev, /* print */
498 nodev, /* dump */
499 nodev, /* read */
500 nodev, /* write */
501 nodev, /* ioctl */
502 nodev, /* devmap */
503 nodev, /* mmap */
504 nodev, /* segmap */
505 nochpoll, /* poll */
506 ddi_prop_op, /* cb_prop_op */
507 0, /* streamtab */
508 D_NEW | D_MP /* Driver compatibility flag */
509 };
510
511 static struct dev_ops sdt_ops = {
512 DEVO_REV, /* devo_rev, */
513 0, /* refcnt */
514 sdt_info, /* get_dev_info */
515 nulldev, /* identify */
516 nulldev, /* probe */
517 sdt_attach, /* attach */
518 sdt_detach, /* detach */
519 nodev, /* reset */
520 &sdt_cb_ops, /* driver operations */
521 NULL, /* bus operations */
522 nodev, /* dev power */
523 ddi_quiesce_not_needed, /* quiesce */
524 };
525
526 /*
527 * Module linkage information for the kernel.
528 */
529 static struct modldrv modldrv = {
530 &mod_driverops, /* module type (this is a pseudo driver) */
531 "Statically Defined Tracing", /* name of module */
532 &sdt_ops, /* driver ops */
533 };
534
535 static struct modlinkage modlinkage = {
536 MODREV_1,
537 (void *)&modldrv,
538 NULL
539 };
540
541 int
542 _init(void)
543 {
544 return (mod_install(&modlinkage));
545 }
546
547 int
548 _info(struct modinfo *modinfop)
549 {
550 return (mod_info(&modlinkage, modinfop));
551 }
552
553 int
554 _fini(void)
555 {
556 return (mod_remove(&modlinkage));
557 }