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 * $FreeBSD$ 23 * 24 */ 25 /* 26 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 27 * Use is subject to license terms. 28 */ 29 30 /* 31 * Copyright (c) 2011, Joyent, Inc. All rights reserved. 32 */ 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/types.h> 37 #include <sys/kernel.h> 38 #include <sys/malloc.h> 39 #include <sys/kmem.h> 40 #include <sys/smp.h> 41 #include <sys/dtrace_impl.h> 42 #include <sys/dtrace_bsd.h> 43 #include <machine/clock.h> 44 #include <machine/cpufunc.h> 45 #include <machine/frame.h> 46 #include <machine/psl.h> 47 #include <vm/pmap.h> 48 49 extern void dtrace_getnanotime(struct timespec *tsp); 50 51 int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t); 52 53 typedef struct dtrace_invop_hdlr { 54 int (*dtih_func)(uintptr_t, struct trapframe *, uintptr_t); 55 struct dtrace_invop_hdlr *dtih_next; 56 } dtrace_invop_hdlr_t; 57 58 dtrace_invop_hdlr_t *dtrace_invop_hdlr; 59 60 int 61 dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t eax) 62 { 63 dtrace_invop_hdlr_t *hdlr; 64 int rval; 65 66 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) 67 if ((rval = hdlr->dtih_func(addr, frame, eax)) != 0) 68 return (rval); 69 70 return (0); 71 } 72 73 void 74 dtrace_invop_add(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 75 { 76 dtrace_invop_hdlr_t *hdlr; 77 78 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); 79 hdlr->dtih_func = func; 80 hdlr->dtih_next = dtrace_invop_hdlr; 81 dtrace_invop_hdlr = hdlr; 82 } 83 84 void 85 dtrace_invop_remove(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 86 { 87 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; 88 89 for (;;) { 90 if (hdlr == NULL) 91 panic("attempt to remove non-existent invop handler"); 92 93 if (hdlr->dtih_func == func) 94 break; 95 96 prev = hdlr; 97 hdlr = hdlr->dtih_next; 98 } 99 100 if (prev == NULL) { 101 ASSERT(dtrace_invop_hdlr == hdlr); 102 dtrace_invop_hdlr = hdlr->dtih_next; 103 } else { 104 ASSERT(dtrace_invop_hdlr != hdlr); 105 prev->dtih_next = hdlr->dtih_next; 106 } 107 108 kmem_free(hdlr, 0); 109 } 110 111 /*ARGSUSED*/ 112 void 113 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) 114 { 115 (*func)(0, (uintptr_t) addr_PTmap); 116 } 117 118 void 119 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) 120 { 121 cpuset_t cpus; 122 123 if (cpu == DTRACE_CPUALL) 124 cpus = all_cpus; 125 else 126 CPU_SETOF(cpu, &cpus); 127 128 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, func, 129 smp_no_rendezvous_barrier, arg); 130 } 131 132 static void 133 dtrace_sync_func(void) 134 { 135 } 136 137 void 138 dtrace_sync(void) 139 { 140 dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); 141 } 142 143 #ifdef notyet 144 void 145 dtrace_safe_synchronous_signal(void) 146 { 147 kthread_t *t = curthread; 148 struct regs *rp = lwptoregs(ttolwp(t)); 149 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; 150 151 ASSERT(t->t_dtrace_on); 152 153 /* 154 * If we're not in the range of scratch addresses, we're not actually 155 * tracing user instructions so turn off the flags. If the instruction 156 * we copied out caused a synchonous trap, reset the pc back to its 157 * original value and turn off the flags. 158 */ 159 if (rp->r_pc < t->t_dtrace_scrpc || 160 rp->r_pc > t->t_dtrace_astpc + isz) { 161 t->t_dtrace_ft = 0; 162 } else if (rp->r_pc == t->t_dtrace_scrpc || 163 rp->r_pc == t->t_dtrace_astpc) { 164 rp->r_pc = t->t_dtrace_pc; 165 t->t_dtrace_ft = 0; 166 } 167 } 168 169 int 170 dtrace_safe_defer_signal(void) 171 { 172 kthread_t *t = curthread; 173 struct regs *rp = lwptoregs(ttolwp(t)); 174 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; 175 176 ASSERT(t->t_dtrace_on); 177 178 /* 179 * If we're not in the range of scratch addresses, we're not actually 180 * tracing user instructions so turn off the flags. 181 */ 182 if (rp->r_pc < t->t_dtrace_scrpc || 183 rp->r_pc > t->t_dtrace_astpc + isz) { 184 t->t_dtrace_ft = 0; 185 return (0); 186 } 187 188 /* 189 * If we have executed the original instruction, but we have performed 190 * neither the jmp back to t->t_dtrace_npc nor the clean up of any 191 * registers used to emulate %rip-relative instructions in 64-bit mode, 192 * we'll save ourselves some effort by doing that here and taking the 193 * signal right away. We detect this condition by seeing if the program 194 * counter is the range [scrpc + isz, astpc). 195 */ 196 if (rp->r_pc >= t->t_dtrace_scrpc + isz && 197 rp->r_pc < t->t_dtrace_astpc) { 198 #ifdef __amd64 199 /* 200 * If there is a scratch register and we're on the 201 * instruction immediately after the modified instruction, 202 * restore the value of that scratch register. 203 */ 204 if (t->t_dtrace_reg != 0 && 205 rp->r_pc == t->t_dtrace_scrpc + isz) { 206 switch (t->t_dtrace_reg) { 207 case REG_RAX: 208 rp->r_rax = t->t_dtrace_regv; 209 break; 210 case REG_RCX: 211 rp->r_rcx = t->t_dtrace_regv; 212 break; 213 case REG_R8: 214 rp->r_r8 = t->t_dtrace_regv; 215 break; 216 case REG_R9: 217 rp->r_r9 = t->t_dtrace_regv; 218 break; 219 } 220 } 221 #endif 222 rp->r_pc = t->t_dtrace_npc; 223 t->t_dtrace_ft = 0; 224 return (0); 225 } 226 227 /* 228 * Otherwise, make sure we'll return to the kernel after executing 229 * the copied out instruction and defer the signal. 230 */ 231 if (!t->t_dtrace_step) { 232 ASSERT(rp->r_pc < t->t_dtrace_astpc); 233 rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc; 234 t->t_dtrace_step = 1; 235 } 236 237 t->t_dtrace_ast = 1; 238 239 return (1); 240 } 241 #endif 242 243 static int64_t tgt_cpu_tsc; 244 static int64_t hst_cpu_tsc; 245 static int64_t tsc_skew[MAXCPU]; 246 static uint64_t nsec_scale; 247 248 /* See below for the explanation of this macro. */ 249 #define SCALE_SHIFT 28 250 251 static void 252 dtrace_gethrtime_init_cpu(void *arg) 253 { 254 uintptr_t cpu = (uintptr_t) arg; 255 256 if (cpu == curcpu) 257 tgt_cpu_tsc = rdtsc(); 258 else 259 hst_cpu_tsc = rdtsc(); 260 } 261 262 #ifdef EARLY_AP_STARTUP 263 static void 264 dtrace_gethrtime_init(void *arg) 265 { 266 struct pcpu *pc; 267 uint64_t tsc_f; 268 cpuset_t map; 269 int i; 270 #else 271 /* 272 * Get the frequency and scale factor as early as possible so that they can be 273 * used for boot-time tracing. 274 */ 275 static void 276 dtrace_gethrtime_init_early(void *arg) 277 { 278 uint64_t tsc_f; 279 #endif 280 281 /* 282 * Get TSC frequency known at this moment. 283 * This should be constant if TSC is invariant. 284 * Otherwise tick->time conversion will be inaccurate, but 285 * will preserve monotonic property of TSC. 286 */ 287 tsc_f = atomic_load_acq_64(&tsc_freq); 288 289 /* 290 * The following line checks that nsec_scale calculated below 291 * doesn't overflow 32-bit unsigned integer, so that it can multiply 292 * another 32-bit integer without overflowing 64-bit. 293 * Thus minimum supported TSC frequency is 62.5MHz. 294 */ 295 KASSERT(tsc_f > (NANOSEC >> (32 - SCALE_SHIFT)), 296 ("TSC frequency is too low")); 297 298 /* 299 * We scale up NANOSEC/tsc_f ratio to preserve as much precision 300 * as possible. 301 * 2^28 factor was chosen quite arbitrarily from practical 302 * considerations: 303 * - it supports TSC frequencies as low as 62.5MHz (see above); 304 * - it provides quite good precision (e < 0.01%) up to THz 305 * (terahertz) values; 306 */ 307 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tsc_f; 308 #ifndef EARLY_AP_STARTUP 309 } 310 SYSINIT(dtrace_gethrtime_init_early, SI_SUB_CPU, SI_ORDER_ANY, 311 dtrace_gethrtime_init_early, NULL); 312 313 static void 314 dtrace_gethrtime_init(void *arg) 315 { 316 struct pcpu *pc; 317 cpuset_t map; 318 int i; 319 #endif 320 321 /* The current CPU is the reference one. */ 322 sched_pin(); 323 tsc_skew[curcpu] = 0; 324 CPU_FOREACH(i) { 325 if (i == curcpu) 326 continue; 327 328 pc = pcpu_find(i); 329 CPU_SETOF(PCPU_GET(cpuid), &map); 330 CPU_SET(pc->pc_cpuid, &map); 331 332 smp_rendezvous_cpus(map, NULL, 333 dtrace_gethrtime_init_cpu, 334 smp_no_rendezvous_barrier, (void *)(uintptr_t) i); 335 336 tsc_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; 337 } 338 sched_unpin(); 339 } 340 #ifdef EARLY_AP_STARTUP 341 SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY, 342 dtrace_gethrtime_init, NULL); 343 #else 344 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, 345 NULL); 346 #endif 347 348 /* 349 * DTrace needs a high resolution time function which can 350 * be called from a probe context and guaranteed not to have 351 * instrumented with probes itself. 352 * 353 * Returns nanoseconds since boot. 354 */ 355 uint64_t 356 dtrace_gethrtime() 357 { 358 uint64_t tsc; 359 uint32_t lo; 360 uint32_t hi; 361 362 /* 363 * We split TSC value into lower and higher 32-bit halves and separately 364 * scale them with nsec_scale, then we scale them down by 2^28 365 * (see nsec_scale calculations) taking into account 32-bit shift of 366 * the higher half and finally add. 367 */ 368 tsc = rdtsc() - tsc_skew[curcpu]; 369 lo = tsc; 370 hi = tsc >> 32; 371 return (((lo * nsec_scale) >> SCALE_SHIFT) + 372 ((hi * nsec_scale) << (32 - SCALE_SHIFT))); 373 } 374 375 uint64_t 376 dtrace_gethrestime(void) 377 { 378 struct timespec current_time; 379 380 dtrace_getnanotime(¤t_time); 381 382 return (current_time.tv_sec * 1000000000ULL + current_time.tv_nsec); 383 } 384 385 /* Function to handle DTrace traps during probes. See amd64/amd64/trap.c. */ 386 int 387 dtrace_trap(struct trapframe *frame, u_int type) 388 { 389 uint16_t nofault; 390 391 /* 392 * A trap can occur while DTrace executes a probe. Before 393 * executing the probe, DTrace blocks re-scheduling and sets 394 * a flag in its per-cpu flags to indicate that it doesn't 395 * want to fault. On returning from the probe, the no-fault 396 * flag is cleared and finally re-scheduling is enabled. 397 * 398 * Check if DTrace has enabled 'no-fault' mode: 399 */ 400 sched_pin(); 401 nofault = cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT; 402 sched_unpin(); 403 if (nofault) { 404 KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled")); 405 406 /* 407 * There are only a couple of trap types that are expected. 408 * All the rest will be handled in the usual way. 409 */ 410 switch (type) { 411 /* General protection fault. */ 412 case T_PROTFLT: 413 /* Flag an illegal operation. */ 414 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 415 416 /* 417 * Offset the instruction pointer to the instruction 418 * following the one causing the fault. 419 */ 420 frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip); 421 return (1); 422 /* Page fault. */ 423 case T_PAGEFLT: 424 /* Flag a bad address. */ 425 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 426 cpu_core[curcpu].cpuc_dtrace_illval = frame->tf_addr; 427 428 /* 429 * Offset the instruction pointer to the instruction 430 * following the one causing the fault. 431 */ 432 frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip); 433 return (1); 434 default: 435 /* Handle all other traps in the usual way. */ 436 break; 437 } 438 } 439 440 /* Handle the trap in the usual way. */ 441 return (0); 442 } 443