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/cpuset.h> 38 #include <sys/kernel.h> 39 #include <sys/malloc.h> 40 #include <sys/kmem.h> 41 #include <sys/smp.h> 42 #include <sys/dtrace_impl.h> 43 #include <sys/dtrace_bsd.h> 44 #include <machine/clock.h> 45 #include <machine/cpufunc.h> 46 #include <machine/frame.h> 47 #include <machine/psl.h> 48 #include <machine/trap.h> 49 #include <vm/pmap.h> 50 51 extern uintptr_t kernelbase; 52 53 extern void dtrace_getnanotime(struct timespec *tsp); 54 55 int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t); 56 57 typedef struct dtrace_invop_hdlr { 58 int (*dtih_func)(uintptr_t, struct trapframe *, uintptr_t); 59 struct dtrace_invop_hdlr *dtih_next; 60 } dtrace_invop_hdlr_t; 61 62 dtrace_invop_hdlr_t *dtrace_invop_hdlr; 63 64 int 65 dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t eax) 66 { 67 dtrace_invop_hdlr_t *hdlr; 68 int rval; 69 70 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) 71 if ((rval = hdlr->dtih_func(addr, frame, eax)) != 0) 72 return (rval); 73 74 return (0); 75 } 76 77 void 78 dtrace_invop_add(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 79 { 80 dtrace_invop_hdlr_t *hdlr; 81 82 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); 83 hdlr->dtih_func = func; 84 hdlr->dtih_next = dtrace_invop_hdlr; 85 dtrace_invop_hdlr = hdlr; 86 } 87 88 void 89 dtrace_invop_remove(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 90 { 91 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; 92 93 for (;;) { 94 if (hdlr == NULL) 95 panic("attempt to remove non-existent invop handler"); 96 97 if (hdlr->dtih_func == func) 98 break; 99 100 prev = hdlr; 101 hdlr = hdlr->dtih_next; 102 } 103 104 if (prev == NULL) { 105 ASSERT(dtrace_invop_hdlr == hdlr); 106 dtrace_invop_hdlr = hdlr->dtih_next; 107 } else { 108 ASSERT(dtrace_invop_hdlr != hdlr); 109 prev->dtih_next = hdlr->dtih_next; 110 } 111 112 kmem_free(hdlr, 0); 113 } 114 115 void 116 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) 117 { 118 (*func)(0, kernelbase); 119 } 120 121 void 122 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) 123 { 124 cpuset_t cpus; 125 126 if (cpu == DTRACE_CPUALL) 127 cpus = all_cpus; 128 else 129 CPU_SETOF(cpu, &cpus); 130 131 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, func, 132 smp_no_rendezvous_barrier, arg); 133 } 134 135 static void 136 dtrace_sync_func(void) 137 { 138 } 139 140 void 141 dtrace_sync(void) 142 { 143 dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); 144 } 145 146 #ifdef notyet 147 void 148 dtrace_safe_synchronous_signal(void) 149 { 150 kthread_t *t = curthread; 151 struct regs *rp = lwptoregs(ttolwp(t)); 152 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; 153 154 ASSERT(t->t_dtrace_on); 155 156 /* 157 * If we're not in the range of scratch addresses, we're not actually 158 * tracing user instructions so turn off the flags. If the instruction 159 * we copied out caused a synchonous trap, reset the pc back to its 160 * original value and turn off the flags. 161 */ 162 if (rp->r_pc < t->t_dtrace_scrpc || 163 rp->r_pc > t->t_dtrace_astpc + isz) { 164 t->t_dtrace_ft = 0; 165 } else if (rp->r_pc == t->t_dtrace_scrpc || 166 rp->r_pc == t->t_dtrace_astpc) { 167 rp->r_pc = t->t_dtrace_pc; 168 t->t_dtrace_ft = 0; 169 } 170 } 171 172 int 173 dtrace_safe_defer_signal(void) 174 { 175 kthread_t *t = curthread; 176 struct regs *rp = lwptoregs(ttolwp(t)); 177 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; 178 179 ASSERT(t->t_dtrace_on); 180 181 /* 182 * If we're not in the range of scratch addresses, we're not actually 183 * tracing user instructions so turn off the flags. 184 */ 185 if (rp->r_pc < t->t_dtrace_scrpc || 186 rp->r_pc > t->t_dtrace_astpc + isz) { 187 t->t_dtrace_ft = 0; 188 return (0); 189 } 190 191 /* 192 * If we have executed the original instruction, but we have performed 193 * neither the jmp back to t->t_dtrace_npc nor the clean up of any 194 * registers used to emulate %rip-relative instructions in 64-bit mode, 195 * we'll save ourselves some effort by doing that here and taking the 196 * signal right away. We detect this condition by seeing if the program 197 * counter is the range [scrpc + isz, astpc). 198 */ 199 if (rp->r_pc >= t->t_dtrace_scrpc + isz && 200 rp->r_pc < t->t_dtrace_astpc) { 201 #ifdef __amd64 202 /* 203 * If there is a scratch register and we're on the 204 * instruction immediately after the modified instruction, 205 * restore the value of that scratch register. 206 */ 207 if (t->t_dtrace_reg != 0 && 208 rp->r_pc == t->t_dtrace_scrpc + isz) { 209 switch (t->t_dtrace_reg) { 210 case REG_RAX: 211 rp->r_rax = t->t_dtrace_regv; 212 break; 213 case REG_RCX: 214 rp->r_rcx = t->t_dtrace_regv; 215 break; 216 case REG_R8: 217 rp->r_r8 = t->t_dtrace_regv; 218 break; 219 case REG_R9: 220 rp->r_r9 = t->t_dtrace_regv; 221 break; 222 } 223 } 224 #endif 225 rp->r_pc = t->t_dtrace_npc; 226 t->t_dtrace_ft = 0; 227 return (0); 228 } 229 230 /* 231 * Otherwise, make sure we'll return to the kernel after executing 232 * the copied out instruction and defer the signal. 233 */ 234 if (!t->t_dtrace_step) { 235 ASSERT(rp->r_pc < t->t_dtrace_astpc); 236 rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc; 237 t->t_dtrace_step = 1; 238 } 239 240 t->t_dtrace_ast = 1; 241 242 return (1); 243 } 244 #endif 245 246 static int64_t tgt_cpu_tsc; 247 static int64_t hst_cpu_tsc; 248 static int64_t tsc_skew[MAXCPU]; 249 static uint64_t nsec_scale; 250 251 /* See below for the explanation of this macro. */ 252 #define SCALE_SHIFT 28 253 254 static void 255 dtrace_gethrtime_init_cpu(void *arg) 256 { 257 uintptr_t cpu = (uintptr_t) arg; 258 259 if (cpu == curcpu) 260 tgt_cpu_tsc = rdtsc(); 261 else 262 hst_cpu_tsc = rdtsc(); 263 } 264 265 #ifdef EARLY_AP_STARTUP 266 static void 267 dtrace_gethrtime_init(void *arg) 268 { 269 struct pcpu *pc; 270 uint64_t tsc_f; 271 cpuset_t map; 272 int i; 273 #else 274 /* 275 * Get the frequency and scale factor as early as possible so that they can be 276 * used for boot-time tracing. 277 */ 278 static void 279 dtrace_gethrtime_init_early(void *arg) 280 { 281 uint64_t tsc_f; 282 #endif 283 284 /* 285 * Get TSC frequency known at this moment. 286 * This should be constant if TSC is invariant. 287 * Otherwise tick->time conversion will be inaccurate, but 288 * will preserve monotonic property of TSC. 289 */ 290 tsc_f = atomic_load_acq_64(&tsc_freq); 291 292 /* 293 * The following line checks that nsec_scale calculated below 294 * doesn't overflow 32-bit unsigned integer, so that it can multiply 295 * another 32-bit integer without overflowing 64-bit. 296 * Thus minimum supported TSC frequency is 62.5MHz. 297 */ 298 KASSERT(tsc_f > (NANOSEC >> (32 - SCALE_SHIFT)), 299 ("TSC frequency is too low")); 300 301 /* 302 * We scale up NANOSEC/tsc_f ratio to preserve as much precision 303 * as possible. 304 * 2^28 factor was chosen quite arbitrarily from practical 305 * considerations: 306 * - it supports TSC frequencies as low as 62.5MHz (see above); 307 * - it provides quite good precision (e < 0.01%) up to THz 308 * (terahertz) values; 309 */ 310 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tsc_f; 311 #ifndef EARLY_AP_STARTUP 312 } 313 SYSINIT(dtrace_gethrtime_init_early, SI_SUB_CPU, SI_ORDER_ANY, 314 dtrace_gethrtime_init_early, NULL); 315 316 static void 317 dtrace_gethrtime_init(void *arg) 318 { 319 cpuset_t map; 320 struct pcpu *pc; 321 int i; 322 #endif 323 324 /* The current CPU is the reference one. */ 325 sched_pin(); 326 tsc_skew[curcpu] = 0; 327 CPU_FOREACH(i) { 328 if (i == curcpu) 329 continue; 330 331 pc = pcpu_find(i); 332 CPU_SETOF(PCPU_GET(cpuid), &map); 333 CPU_SET(pc->pc_cpuid, &map); 334 335 smp_rendezvous_cpus(map, NULL, 336 dtrace_gethrtime_init_cpu, 337 smp_no_rendezvous_barrier, (void *)(uintptr_t) i); 338 339 tsc_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; 340 } 341 sched_unpin(); 342 } 343 #ifdef EARLY_AP_STARTUP 344 SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY, 345 dtrace_gethrtime_init, NULL); 346 #else 347 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, 348 NULL); 349 #endif 350 351 /* 352 * DTrace needs a high resolution time function which can 353 * be called from a probe context and guaranteed not to have 354 * instrumented with probes itself. 355 * 356 * Returns nanoseconds since boot. 357 */ 358 uint64_t 359 dtrace_gethrtime(void) 360 { 361 uint64_t tsc; 362 uint32_t lo, hi; 363 register_t eflags; 364 365 /* 366 * We split TSC value into lower and higher 32-bit halves and separately 367 * scale them with nsec_scale, then we scale them down by 2^28 368 * (see nsec_scale calculations) taking into account 32-bit shift of 369 * the higher half and finally add. 370 */ 371 eflags = intr_disable(); 372 tsc = rdtsc() - tsc_skew[curcpu]; 373 intr_restore(eflags); 374 375 lo = tsc; 376 hi = tsc >> 32; 377 return (((lo * nsec_scale) >> SCALE_SHIFT) + 378 ((hi * nsec_scale) << (32 - SCALE_SHIFT))); 379 } 380 381 uint64_t 382 dtrace_gethrestime(void) 383 { 384 struct timespec current_time; 385 386 dtrace_getnanotime(¤t_time); 387 388 return (current_time.tv_sec * 1000000000ULL + current_time.tv_nsec); 389 } 390 391 /* Function to handle DTrace traps during probes. See i386/i386/trap.c */ 392 int 393 dtrace_trap(struct trapframe *frame, u_int type) 394 { 395 uint16_t nofault; 396 397 /* 398 * A trap can occur while DTrace executes a probe. Before 399 * executing the probe, DTrace blocks re-scheduling and sets 400 * a flag in its per-cpu flags to indicate that it doesn't 401 * want to fault. On returning from the probe, the no-fault 402 * flag is cleared and finally re-scheduling is enabled. 403 * 404 * Check if DTrace has enabled 'no-fault' mode: 405 */ 406 sched_pin(); 407 nofault = cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT; 408 sched_unpin(); 409 if (nofault) { 410 KASSERT((read_eflags() & PSL_I) == 0, ("interrupts enabled")); 411 412 /* 413 * There are only a couple of trap types that are expected. 414 * All the rest will be handled in the usual way. 415 */ 416 switch (type) { 417 /* General protection fault. */ 418 case T_PROTFLT: 419 /* Flag an illegal operation. */ 420 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 421 422 /* 423 * Offset the instruction pointer to the instruction 424 * following the one causing the fault. 425 */ 426 frame->tf_eip += dtrace_instr_size((u_char *) frame->tf_eip); 427 return (1); 428 /* Page fault. */ 429 case T_PAGEFLT: 430 /* Flag a bad address. */ 431 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 432 cpu_core[curcpu].cpuc_dtrace_illval = rcr2(); 433 434 /* 435 * Offset the instruction pointer to the instruction 436 * following the one causing the fault. 437 */ 438 frame->tf_eip += dtrace_instr_size((u_char *) frame->tf_eip); 439 return (1); 440 default: 441 /* Handle all other traps in the usual way. */ 442 break; 443 } 444 } 445 446 /* Handle the trap in the usual way. */ 447 return (0); 448 } 449