xref: /illumos-gate/usr/src/uts/i86pc/os/dtrace_subr.c (revision 5279807d7e1818eac6f90ac640b7a89cdb37522d)
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 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/dtrace.h>
28 #include <sys/fasttrap.h>
29 #include <sys/x_call.h>
30 #include <sys/cmn_err.h>
31 #include <sys/trap.h>
32 #include <sys/psw.h>
33 #include <sys/privregs.h>
34 #include <sys/machsystm.h>
35 #include <vm/seg_kmem.h>
36 
37 typedef struct dtrace_invop_hdlr {
38 	int (*dtih_func)(uintptr_t, uintptr_t *, uintptr_t);
39 	struct dtrace_invop_hdlr *dtih_next;
40 } dtrace_invop_hdlr_t;
41 
42 dtrace_invop_hdlr_t *dtrace_invop_hdlr;
43 
44 int
45 dtrace_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax)
46 {
47 	dtrace_invop_hdlr_t *hdlr;
48 	int rval;
49 
50 	for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) {
51 		if ((rval = hdlr->dtih_func(addr, stack, eax)) != 0)
52 			return (rval);
53 	}
54 
55 	return (0);
56 }
57 
58 void
59 dtrace_invop_add(int (*func)(uintptr_t, uintptr_t *, uintptr_t))
60 {
61 	dtrace_invop_hdlr_t *hdlr;
62 
63 	hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP);
64 	hdlr->dtih_func = func;
65 	hdlr->dtih_next = dtrace_invop_hdlr;
66 	dtrace_invop_hdlr = hdlr;
67 }
68 
69 void
70 dtrace_invop_remove(int (*func)(uintptr_t, uintptr_t *, uintptr_t))
71 {
72 	dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL;
73 
74 	for (;;) {
75 		if (hdlr == NULL)
76 			panic("attempt to remove non-existent invop handler");
77 
78 		if (hdlr->dtih_func == func)
79 			break;
80 
81 		prev = hdlr;
82 		hdlr = hdlr->dtih_next;
83 	}
84 
85 	if (prev == NULL) {
86 		ASSERT(dtrace_invop_hdlr == hdlr);
87 		dtrace_invop_hdlr = hdlr->dtih_next;
88 	} else {
89 		ASSERT(dtrace_invop_hdlr != hdlr);
90 		prev->dtih_next = hdlr->dtih_next;
91 	}
92 
93 	kmem_free(hdlr, sizeof (dtrace_invop_hdlr_t));
94 }
95 
96 int
97 dtrace_getipl(void)
98 {
99 	return (CPU->cpu_pri);
100 }
101 
102 /*ARGSUSED*/
103 void
104 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit))
105 {
106 #ifdef __amd64
107 	extern uintptr_t toxic_addr;
108 	extern size_t toxic_size;
109 
110 	(*func)(0, _userlimit);
111 
112 	if (hole_end > hole_start)
113 		(*func)(hole_start, hole_end);
114 	(*func)(toxic_addr, toxic_addr + toxic_size);
115 #else
116 	extern void *device_arena_contains(void *, size_t, size_t *);
117 	caddr_t	vaddr;
118 	size_t	len;
119 
120 	for (vaddr = (caddr_t)kernelbase; vaddr < (caddr_t)KERNEL_TEXT;
121 	    vaddr += len) {
122 		len = (caddr_t)KERNEL_TEXT - vaddr;
123 		vaddr = device_arena_contains(vaddr, len, &len);
124 		if (vaddr == NULL)
125 			break;
126 		(*func)((uintptr_t)vaddr, (uintptr_t)vaddr + len);
127 	}
128 #endif
129 	(*func)(0, _userlimit);
130 }
131 
132 static int
133 dtrace_xcall_func(dtrace_xcall_t func, void *arg)
134 {
135 	(*func)(arg);
136 
137 	return (0);
138 }
139 
140 /*ARGSUSED*/
141 void
142 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg)
143 {
144 	cpuset_t set;
145 
146 	CPUSET_ZERO(set);
147 
148 	if (cpu == DTRACE_CPUALL) {
149 		CPUSET_ALL(set);
150 	} else {
151 		CPUSET_ADD(set, cpu);
152 	}
153 
154 	kpreempt_disable();
155 	xc_sync((xc_arg_t)func, (xc_arg_t)arg, 0, CPUSET2BV(set),
156 	    (xc_func_t)dtrace_xcall_func);
157 	kpreempt_enable();
158 }
159 
160 void
161 dtrace_sync_func(void)
162 {}
163 
164 void
165 dtrace_sync(void)
166 {
167 	dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL);
168 }
169 
170 int (*dtrace_pid_probe_ptr)(struct regs *);
171 int (*dtrace_return_probe_ptr)(struct regs *);
172 
173 void
174 dtrace_user_probe(struct regs *rp, caddr_t addr, processorid_t cpuid)
175 {
176 	krwlock_t *rwp;
177 	proc_t *p = curproc;
178 	extern void trap(struct regs *, caddr_t, processorid_t);
179 
180 	if (USERMODE(rp->r_cs) || (rp->r_ps & PS_VM)) {
181 		if (curthread->t_cred != p->p_cred) {
182 			cred_t *oldcred = curthread->t_cred;
183 			/*
184 			 * DTrace accesses t_cred in probe context.  t_cred
185 			 * must always be either NULL, or point to a valid,
186 			 * allocated cred structure.
187 			 */
188 			curthread->t_cred = crgetcred();
189 			crfree(oldcred);
190 		}
191 	}
192 
193 	if (rp->r_trapno == T_DTRACE_RET) {
194 		uint8_t step = curthread->t_dtrace_step;
195 		uint8_t ret = curthread->t_dtrace_ret;
196 		uintptr_t npc = curthread->t_dtrace_npc;
197 
198 		if (curthread->t_dtrace_ast) {
199 			aston(curthread);
200 			curthread->t_sig_check = 1;
201 		}
202 
203 		/*
204 		 * Clear all user tracing flags.
205 		 */
206 		curthread->t_dtrace_ft = 0;
207 
208 		/*
209 		 * If we weren't expecting to take a return probe trap, kill
210 		 * the process as though it had just executed an unassigned
211 		 * trap instruction.
212 		 */
213 		if (step == 0) {
214 			tsignal(curthread, SIGILL);
215 			return;
216 		}
217 
218 		/*
219 		 * If we hit this trap unrelated to a return probe, we're
220 		 * just here to reset the AST flag since we deferred a signal
221 		 * until after we logically single-stepped the instruction we
222 		 * copied out.
223 		 */
224 		if (ret == 0) {
225 			rp->r_pc = npc;
226 			return;
227 		}
228 
229 		/*
230 		 * We need to wait until after we've called the
231 		 * dtrace_return_probe_ptr function pointer to set %pc.
232 		 */
233 		rwp = &CPU->cpu_ft_lock;
234 		rw_enter(rwp, RW_READER);
235 		if (dtrace_return_probe_ptr != NULL)
236 			(void) (*dtrace_return_probe_ptr)(rp);
237 		rw_exit(rwp);
238 		rp->r_pc = npc;
239 
240 	} else if (rp->r_trapno == T_BPTFLT) {
241 		uint8_t instr, instr2;
242 		caddr_t linearpc;
243 		rwp = &CPU->cpu_ft_lock;
244 
245 		/*
246 		 * The DTrace fasttrap provider uses the breakpoint trap
247 		 * (int 3). We let DTrace take the first crack at handling
248 		 * this trap; if it's not a probe that DTrace knowns about,
249 		 * we call into the trap() routine to handle it like a
250 		 * breakpoint placed by a conventional debugger.
251 		 */
252 		rw_enter(rwp, RW_READER);
253 		if (dtrace_pid_probe_ptr != NULL &&
254 		    (*dtrace_pid_probe_ptr)(rp) == 0) {
255 			rw_exit(rwp);
256 			return;
257 		}
258 		rw_exit(rwp);
259 
260 		if (dtrace_linear_pc(rp, p, &linearpc) != 0) {
261 			trap(rp, addr, cpuid);
262 			return;
263 		}
264 
265 		/*
266 		 * If the instruction that caused the breakpoint trap doesn't
267 		 * look like an int 3 anymore, it may be that this tracepoint
268 		 * was removed just after the user thread executed it. In
269 		 * that case, return to user land to retry the instuction.
270 		 * Note that we assume the length of the instruction to retry
271 		 * is 1 byte because that's the length of FASTTRAP_INSTR.
272 		 * We check for r_pc > 0 and > 2 so that we don't have to
273 		 * deal with segment wraparound.
274 		 */
275 		if (rp->r_pc > 0 && fuword8(linearpc - 1, &instr) == 0 &&
276 		    instr != FASTTRAP_INSTR &&
277 		    (instr != 3 || (rp->r_pc >= 2 &&
278 		    (fuword8(linearpc - 2, &instr2) != 0 || instr2 != 0xCD)))) {
279 			rp->r_pc--;
280 			return;
281 		}
282 
283 		trap(rp, addr, cpuid);
284 
285 	} else {
286 		trap(rp, addr, cpuid);
287 	}
288 }
289 
290 void
291 dtrace_safe_synchronous_signal(void)
292 {
293 	kthread_t *t = curthread;
294 	struct regs *rp = lwptoregs(ttolwp(t));
295 	size_t isz = t->t_dtrace_npc - t->t_dtrace_pc;
296 
297 	ASSERT(t->t_dtrace_on);
298 
299 	/*
300 	 * If we're not in the range of scratch addresses, we're not actually
301 	 * tracing user instructions so turn off the flags. If the instruction
302 	 * we copied out caused a synchonous trap, reset the pc back to its
303 	 * original value and turn off the flags.
304 	 */
305 	if (rp->r_pc < t->t_dtrace_scrpc ||
306 	    rp->r_pc > t->t_dtrace_astpc + isz) {
307 		t->t_dtrace_ft = 0;
308 	} else if (rp->r_pc == t->t_dtrace_scrpc ||
309 	    rp->r_pc == t->t_dtrace_astpc) {
310 		rp->r_pc = t->t_dtrace_pc;
311 		t->t_dtrace_ft = 0;
312 	}
313 }
314 
315 int
316 dtrace_safe_defer_signal(void)
317 {
318 	kthread_t *t = curthread;
319 	struct regs *rp = lwptoregs(ttolwp(t));
320 	size_t isz = t->t_dtrace_npc - t->t_dtrace_pc;
321 
322 	ASSERT(t->t_dtrace_on);
323 
324 	/*
325 	 * If we're not in the range of scratch addresses, we're not actually
326 	 * tracing user instructions so turn off the flags.
327 	 */
328 	if (rp->r_pc < t->t_dtrace_scrpc ||
329 	    rp->r_pc > t->t_dtrace_astpc + isz) {
330 		t->t_dtrace_ft = 0;
331 		return (0);
332 	}
333 
334 	/*
335 	 * If we've executed the original instruction, but haven't performed
336 	 * the jmp back to t->t_dtrace_npc or the clean up of any registers
337 	 * used to emulate %rip-relative instructions in 64-bit mode, do that
338 	 * here and take the signal right away. We detect this condition by
339 	 * seeing if the program counter is the range [scrpc + isz, astpc).
340 	 */
341 	if (t->t_dtrace_astpc - rp->r_pc <
342 	    t->t_dtrace_astpc - t->t_dtrace_scrpc - isz) {
343 #ifdef __amd64
344 		/*
345 		 * If there is a scratch register and we're on the
346 		 * instruction immediately after the modified instruction,
347 		 * restore the value of that scratch register.
348 		 */
349 		if (t->t_dtrace_reg != 0 &&
350 		    rp->r_pc == t->t_dtrace_scrpc + isz) {
351 			switch (t->t_dtrace_reg) {
352 			case REG_RAX:
353 				rp->r_rax = t->t_dtrace_regv;
354 				break;
355 			case REG_RCX:
356 				rp->r_rcx = t->t_dtrace_regv;
357 				break;
358 			case REG_R8:
359 				rp->r_r8 = t->t_dtrace_regv;
360 				break;
361 			case REG_R9:
362 				rp->r_r9 = t->t_dtrace_regv;
363 				break;
364 			}
365 		}
366 #endif
367 		rp->r_pc = t->t_dtrace_npc;
368 		t->t_dtrace_ft = 0;
369 		return (0);
370 	}
371 
372 	/*
373 	 * Otherwise, make sure we'll return to the kernel after executing
374 	 * the copied out instruction and defer the signal.
375 	 */
376 	if (!t->t_dtrace_step) {
377 		ASSERT(rp->r_pc < t->t_dtrace_astpc);
378 		rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc;
379 		t->t_dtrace_step = 1;
380 	}
381 
382 	t->t_dtrace_ast = 1;
383 
384 	return (1);
385 }
386 
387 /*
388  * Additional artificial frames for the machine type. For i86pc, we're already
389  * accounted for, so return 0. On the hypervisor, we have an additional frame
390  * (xen_callback_handler).
391  */
392 int
393 dtrace_mach_aframes(void)
394 {
395 #ifdef __xpv
396 	return (1);
397 #else
398 	return (0);
399 #endif
400 }
401