dtrace_isa.c revision 269557
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, Version 1.0 only
6 * (the "License").  You may not use this file except in compliance
7 * with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or http://www.opensolaris.org/os/licensing.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 *
22 * $FreeBSD: stable/10/sys/cddl/dev/dtrace/amd64/dtrace_isa.c 269557 2014-08-05 01:53:15Z markj $
23 */
24/*
25 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
26 * Use is subject to license terms.
27 */
28#include <sys/cdefs.h>
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/kernel.h>
33#include <sys/stack.h>
34#include <sys/pcpu.h>
35
36#include <machine/frame.h>
37#include <machine/md_var.h>
38#include <machine/reg.h>
39#include <machine/stack.h>
40
41#include <vm/vm.h>
42#include <vm/vm_param.h>
43#include <vm/pmap.h>
44
45#include "regset.h"
46
47uint8_t dtrace_fuword8_nocheck(void *);
48uint16_t dtrace_fuword16_nocheck(void *);
49uint32_t dtrace_fuword32_nocheck(void *);
50uint64_t dtrace_fuword64_nocheck(void *);
51
52void
53dtrace_getpcstack(pc_t *pcstack, int pcstack_limit, int aframes,
54    uint32_t *intrpc)
55{
56	int depth = 0;
57	register_t rbp;
58	struct amd64_frame *frame;
59	vm_offset_t callpc;
60	pc_t caller = (pc_t) solaris_cpu[curcpu].cpu_dtrace_caller;
61
62	if (intrpc != 0)
63		pcstack[depth++] = (pc_t) intrpc;
64
65	aframes++;
66
67	__asm __volatile("movq %%rbp,%0" : "=r" (rbp));
68
69	frame = (struct amd64_frame *)rbp;
70	while (depth < pcstack_limit) {
71		if (!INKERNEL((long) frame))
72			break;
73
74		callpc = frame->f_retaddr;
75
76		if (!INKERNEL(callpc))
77			break;
78
79		if (aframes > 0) {
80			aframes--;
81			if ((aframes == 0) && (caller != 0)) {
82				pcstack[depth++] = caller;
83			}
84		}
85		else {
86			pcstack[depth++] = callpc;
87		}
88
89		if (frame->f_frame <= frame ||
90		    (vm_offset_t)frame->f_frame >=
91		    (vm_offset_t)rbp + KSTACK_PAGES * PAGE_SIZE)
92			break;
93		frame = frame->f_frame;
94	}
95
96	for (; depth < pcstack_limit; depth++) {
97		pcstack[depth] = 0;
98	}
99}
100
101static int
102dtrace_getustack_common(uint64_t *pcstack, int pcstack_limit, uintptr_t pc,
103    uintptr_t sp)
104{
105	volatile uint16_t *flags =
106	    (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
107	int ret = 0;
108
109	ASSERT(pcstack == NULL || pcstack_limit > 0);
110
111	while (pc != 0) {
112		ret++;
113		if (pcstack != NULL) {
114			*pcstack++ = (uint64_t)pc;
115			pcstack_limit--;
116			if (pcstack_limit <= 0)
117				break;
118		}
119
120		if (sp == 0)
121			break;
122
123		pc = dtrace_fuword64((void *)(sp +
124			offsetof(struct amd64_frame, f_retaddr)));
125		sp = dtrace_fuword64((void *)sp);
126
127		/*
128		 * This is totally bogus:  if we faulted, we're going to clear
129		 * the fault and break.  This is to deal with the apparently
130		 * broken Java stacks on x86.
131		 */
132		if (*flags & CPU_DTRACE_FAULT) {
133			*flags &= ~CPU_DTRACE_FAULT;
134			break;
135		}
136	}
137
138	return (ret);
139}
140
141void
142dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit)
143{
144	proc_t *p = curproc;
145	struct trapframe *tf;
146	uintptr_t pc, sp, fp;
147	volatile uint16_t *flags =
148	    (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
149	int n;
150
151	if (*flags & CPU_DTRACE_FAULT)
152		return;
153
154	if (pcstack_limit <= 0)
155		return;
156
157	/*
158	 * If there's no user context we still need to zero the stack.
159	 */
160	if (p == NULL || (tf = curthread->td_frame) == NULL)
161		goto zero;
162
163	*pcstack++ = (uint64_t)p->p_pid;
164	pcstack_limit--;
165
166	if (pcstack_limit <= 0)
167		return;
168
169	pc = tf->tf_rip;
170	fp = tf->tf_rbp;
171	sp = tf->tf_rsp;
172
173	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) {
174		/*
175		 * In an entry probe.  The frame pointer has not yet been
176		 * pushed (that happens in the function prologue).  The
177		 * best approach is to add the current pc as a missing top
178		 * of stack and back the pc up to the caller, which is stored
179		 * at the current stack pointer address since the call
180		 * instruction puts it there right before the branch.
181		 */
182
183		*pcstack++ = (uint64_t)pc;
184		pcstack_limit--;
185		if (pcstack_limit <= 0)
186			return;
187
188		pc = dtrace_fuword64((void *) sp);
189	}
190
191	n = dtrace_getustack_common(pcstack, pcstack_limit, pc, fp);
192	ASSERT(n >= 0);
193	ASSERT(n <= pcstack_limit);
194
195	pcstack += n;
196	pcstack_limit -= n;
197
198zero:
199	while (pcstack_limit-- > 0)
200		*pcstack++ = 0;
201}
202
203int
204dtrace_getustackdepth(void)
205{
206	proc_t *p = curproc;
207	struct trapframe *tf;
208	uintptr_t pc, fp, sp;
209	int n = 0;
210
211	if (p == NULL || (tf = curthread->td_frame) == NULL)
212		return (0);
213
214	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
215		return (-1);
216
217	pc = tf->tf_rip;
218	fp = tf->tf_rbp;
219	sp = tf->tf_rsp;
220
221	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) {
222		/*
223		 * In an entry probe.  The frame pointer has not yet been
224		 * pushed (that happens in the function prologue).  The
225		 * best approach is to add the current pc as a missing top
226		 * of stack and back the pc up to the caller, which is stored
227		 * at the current stack pointer address since the call
228		 * instruction puts it there right before the branch.
229		 */
230
231		pc = dtrace_fuword64((void *) sp);
232		n++;
233	}
234
235	n += dtrace_getustack_common(NULL, 0, pc, fp);
236
237	return (n);
238}
239
240void
241dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, int pcstack_limit)
242{
243	proc_t *p = curproc;
244	struct trapframe *tf;
245	uintptr_t pc, sp, fp;
246	volatile uint16_t *flags =
247	    (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
248#ifdef notyet	/* XXX signal stack */
249	uintptr_t oldcontext;
250	size_t s1, s2;
251#endif
252
253	if (*flags & CPU_DTRACE_FAULT)
254		return;
255
256	if (pcstack_limit <= 0)
257		return;
258
259	/*
260	 * If there's no user context we still need to zero the stack.
261	 */
262	if (p == NULL || (tf = curthread->td_frame) == NULL)
263		goto zero;
264
265	*pcstack++ = (uint64_t)p->p_pid;
266	pcstack_limit--;
267
268	if (pcstack_limit <= 0)
269		return;
270
271	pc = tf->tf_rip;
272	sp = tf->tf_rsp;
273	fp = tf->tf_rbp;
274
275#ifdef notyet /* XXX signal stack */
276	oldcontext = lwp->lwp_oldcontext;
277	s1 = sizeof (struct xframe) + 2 * sizeof (long);
278	s2 = s1 + sizeof (siginfo_t);
279#endif
280
281	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) {
282		*pcstack++ = (uint64_t)pc;
283		*fpstack++ = 0;
284		pcstack_limit--;
285		if (pcstack_limit <= 0)
286			return;
287
288		pc = dtrace_fuword64((void *)sp);
289	}
290
291	while (pc != 0) {
292		*pcstack++ = (uint64_t)pc;
293		*fpstack++ = fp;
294		pcstack_limit--;
295		if (pcstack_limit <= 0)
296			break;
297
298		if (fp == 0)
299			break;
300
301#ifdef notyet /* XXX signal stack */
302		if (oldcontext == sp + s1 || oldcontext == sp + s2) {
303			ucontext_t *ucp = (ucontext_t *)oldcontext;
304			greg_t *gregs = ucp->uc_mcontext.gregs;
305
306			sp = dtrace_fulword(&gregs[REG_FP]);
307			pc = dtrace_fulword(&gregs[REG_PC]);
308
309			oldcontext = dtrace_fulword(&ucp->uc_link);
310		} else
311#endif /* XXX */
312		{
313			pc = dtrace_fuword64((void *)(fp +
314				offsetof(struct amd64_frame, f_retaddr)));
315			fp = dtrace_fuword64((void *)fp);
316		}
317
318		/*
319		 * This is totally bogus:  if we faulted, we're going to clear
320		 * the fault and break.  This is to deal with the apparently
321		 * broken Java stacks on x86.
322		 */
323		if (*flags & CPU_DTRACE_FAULT) {
324			*flags &= ~CPU_DTRACE_FAULT;
325			break;
326		}
327	}
328
329zero:
330	while (pcstack_limit-- > 0)
331		*pcstack++ = 0;
332}
333
334/*ARGSUSED*/
335uint64_t
336dtrace_getarg(int arg, int aframes)
337{
338	uintptr_t val;
339	struct amd64_frame *fp = (struct amd64_frame *)dtrace_getfp();
340	uintptr_t *stack;
341	int i;
342
343	/*
344	 * A total of 6 arguments are passed via registers; any argument with
345	 * index of 5 or lower is therefore in a register.
346	 */
347	int inreg = 5;
348
349	for (i = 1; i <= aframes; i++) {
350		fp = fp->f_frame;
351
352		if (P2ROUNDUP(fp->f_retaddr, 16) ==
353		    (long)dtrace_invop_callsite) {
354			/*
355			 * In the case of amd64, we will use the pointer to the
356			 * regs structure that was pushed when we took the
357			 * trap.  To get this structure, we must increment
358			 * beyond the frame structure, and then again beyond
359			 * the calling RIP stored in dtrace_invop().  If the
360			 * argument that we're seeking is passed on the stack,
361			 * we'll pull the true stack pointer out of the saved
362			 * registers and decrement our argument by the number
363			 * of arguments passed in registers; if the argument
364			 * we're seeking is passed in regsiters, we can just
365			 * load it directly.
366			 */
367			struct trapframe *tf =
368			    (struct trapframe *)((uintptr_t)&fp[1]);
369
370			if (arg <= inreg) {
371				switch (arg) {
372				case 0:
373					stack = (uintptr_t *)&tf->tf_rdi;
374					break;
375				case 1:
376					stack = (uintptr_t *)&tf->tf_rsi;
377					break;
378				case 2:
379					stack = (uintptr_t *)&tf->tf_rdx;
380					break;
381				case 3:
382					stack = (uintptr_t *)&tf->tf_rcx;
383					break;
384				case 4:
385					stack = (uintptr_t *)&tf->tf_r8;
386					break;
387				case 5:
388					stack = (uintptr_t *)&tf->tf_r9;
389					break;
390				}
391				arg = 0;
392			} else {
393				stack = (uintptr_t *)(tf->tf_rsp);
394				arg -= inreg;
395			}
396			goto load;
397		}
398
399	}
400
401	/*
402	 * We know that we did not come through a trap to get into
403	 * dtrace_probe() -- the provider simply called dtrace_probe()
404	 * directly.  As this is the case, we need to shift the argument
405	 * that we're looking for:  the probe ID is the first argument to
406	 * dtrace_probe(), so the argument n will actually be found where
407	 * one would expect to find argument (n + 1).
408	 */
409	arg++;
410
411	if (arg <= inreg) {
412		/*
413		 * This shouldn't happen.  If the argument is passed in a
414		 * register then it should have been, well, passed in a
415		 * register...
416		 */
417		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
418		return (0);
419	}
420
421	arg -= (inreg + 1);
422	stack = (uintptr_t *)fp + 2;
423
424load:
425	DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
426	val = stack[arg];
427	DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
428
429	return (val);
430	return (0);
431}
432
433int
434dtrace_getstackdepth(int aframes)
435{
436	int depth = 0;
437	struct amd64_frame *frame;
438	vm_offset_t rbp;
439
440	aframes++;
441	rbp = dtrace_getfp();
442	frame = (struct amd64_frame *)rbp;
443	depth++;
444	for(;;) {
445		if (!INKERNEL((long) frame))
446			break;
447		if (!INKERNEL((long) frame->f_frame))
448			break;
449		depth++;
450		if (frame->f_frame <= frame ||
451		    (vm_offset_t)frame->f_frame >=
452		    (vm_offset_t)rbp + KSTACK_PAGES * PAGE_SIZE)
453			break;
454		frame = frame->f_frame;
455	}
456	if (depth < aframes)
457		return 0;
458	else
459		return depth - aframes;
460}
461
462ulong_t
463dtrace_getreg(struct trapframe *rp, uint_t reg)
464{
465	/* This table is dependent on reg.d. */
466	int regmap[] = {
467		REG_GS,		/* 0  GS */
468		REG_FS,		/* 1  FS */
469		REG_ES,		/* 2  ES */
470		REG_DS,		/* 3  DS */
471		REG_RDI,	/* 4  EDI */
472		REG_RSI,	/* 5  ESI */
473		REG_RBP,	/* 6  EBP, REG_FP */
474		REG_RSP,	/* 7  ESP */
475		REG_RBX,	/* 8  EBX, REG_R1 */
476		REG_RDX,	/* 9  EDX */
477		REG_RCX,	/* 10 ECX */
478		REG_RAX,	/* 11 EAX, REG_R0 */
479		REG_TRAPNO,	/* 12 TRAPNO */
480		REG_ERR,	/* 13 ERR */
481		REG_RIP,	/* 14 EIP, REG_PC */
482		REG_CS,		/* 15 CS */
483		REG_RFL,	/* 16 EFL, REG_PS */
484		REG_RSP,	/* 17 UESP, REG_SP */
485		REG_SS		/* 18 SS */
486	};
487
488	if (reg <= SS) {
489		if (reg >= sizeof (regmap) / sizeof (int)) {
490			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
491			return (0);
492		}
493
494		reg = regmap[reg];
495	} else {
496		/* This is dependent on reg.d. */
497		reg -= SS + 1;
498	}
499
500	switch (reg) {
501	case REG_RDI:
502		return (rp->tf_rdi);
503	case REG_RSI:
504		return (rp->tf_rsi);
505	case REG_RDX:
506		return (rp->tf_rdx);
507	case REG_RCX:
508		return (rp->tf_rcx);
509	case REG_R8:
510		return (rp->tf_r8);
511	case REG_R9:
512		return (rp->tf_r9);
513	case REG_RAX:
514		return (rp->tf_rax);
515	case REG_RBX:
516		return (rp->tf_rbx);
517	case REG_RBP:
518		return (rp->tf_rbp);
519	case REG_R10:
520		return (rp->tf_r10);
521	case REG_R11:
522		return (rp->tf_r11);
523	case REG_R12:
524		return (rp->tf_r12);
525	case REG_R13:
526		return (rp->tf_r13);
527	case REG_R14:
528		return (rp->tf_r14);
529	case REG_R15:
530		return (rp->tf_r15);
531	case REG_DS:
532		return (rp->tf_ds);
533	case REG_ES:
534		return (rp->tf_es);
535	case REG_FS:
536		return (rp->tf_fs);
537	case REG_GS:
538		return (rp->tf_gs);
539	case REG_TRAPNO:
540		return (rp->tf_trapno);
541	case REG_ERR:
542		return (rp->tf_err);
543	case REG_RIP:
544		return (rp->tf_rip);
545	case REG_CS:
546		return (rp->tf_cs);
547	case REG_SS:
548		return (rp->tf_ss);
549	case REG_RFL:
550		return (rp->tf_rflags);
551	case REG_RSP:
552		return (rp->tf_rsp);
553	default:
554		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
555		return (0);
556	}
557}
558
559static int
560dtrace_copycheck(uintptr_t uaddr, uintptr_t kaddr, size_t size)
561{
562	ASSERT(INKERNEL(kaddr) && kaddr + size >= kaddr);
563
564	if (uaddr + size > VM_MAXUSER_ADDRESS || uaddr + size < uaddr) {
565		DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
566		cpu_core[curcpu].cpuc_dtrace_illval = uaddr;
567		return (0);
568	}
569
570	return (1);
571}
572
573void
574dtrace_copyin(uintptr_t uaddr, uintptr_t kaddr, size_t size,
575    volatile uint16_t *flags)
576{
577	if (dtrace_copycheck(uaddr, kaddr, size))
578		dtrace_copy(uaddr, kaddr, size);
579}
580
581void
582dtrace_copyout(uintptr_t kaddr, uintptr_t uaddr, size_t size,
583    volatile uint16_t *flags)
584{
585	if (dtrace_copycheck(uaddr, kaddr, size))
586		dtrace_copy(kaddr, uaddr, size);
587}
588
589void
590dtrace_copyinstr(uintptr_t uaddr, uintptr_t kaddr, size_t size,
591    volatile uint16_t *flags)
592{
593	if (dtrace_copycheck(uaddr, kaddr, size))
594		dtrace_copystr(uaddr, kaddr, size, flags);
595}
596
597void
598dtrace_copyoutstr(uintptr_t kaddr, uintptr_t uaddr, size_t size,
599    volatile uint16_t *flags)
600{
601	if (dtrace_copycheck(uaddr, kaddr, size))
602		dtrace_copystr(kaddr, uaddr, size, flags);
603}
604
605uint8_t
606dtrace_fuword8(void *uaddr)
607{
608	if ((uintptr_t)uaddr > VM_MAXUSER_ADDRESS) {
609		DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
610		cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr;
611		return (0);
612	}
613	return (dtrace_fuword8_nocheck(uaddr));
614}
615
616uint16_t
617dtrace_fuword16(void *uaddr)
618{
619	if ((uintptr_t)uaddr > VM_MAXUSER_ADDRESS) {
620		DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
621		cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr;
622		return (0);
623	}
624	return (dtrace_fuword16_nocheck(uaddr));
625}
626
627uint32_t
628dtrace_fuword32(void *uaddr)
629{
630	if ((uintptr_t)uaddr > VM_MAXUSER_ADDRESS) {
631		DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
632		cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr;
633		return (0);
634	}
635	return (dtrace_fuword32_nocheck(uaddr));
636}
637
638uint64_t
639dtrace_fuword64(void *uaddr)
640{
641	if ((uintptr_t)uaddr > VM_MAXUSER_ADDRESS) {
642		DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
643		cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr;
644		return (0);
645	}
646	return (dtrace_fuword64_nocheck(uaddr));
647}
648