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 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/types.h> 36 #include <sys/kernel.h> 37 #include <sys/malloc.h> 38 #include <sys/kmem.h> 39 #include <sys/smp.h> 40 #include <sys/dtrace_impl.h> 41 #include <sys/dtrace_bsd.h> 42 #include <machine/clock.h> 43 #include <machine/frame.h> 44 #include <machine/trap.h> 45 #include <vm/pmap.h> 46 47 #define DELAYBRANCH(x) ((int)(x) < 0) 48 49 extern uintptr_t dtrace_in_probe_addr; 50 extern int dtrace_in_probe; 51 extern dtrace_id_t dtrace_probeid_error; 52 extern int (*dtrace_invop_jump_addr)(struct trapframe *); 53 54 extern void dtrace_getnanotime(struct timespec *tsp); 55 56 int dtrace_invop(uintptr_t, uintptr_t *, uintptr_t); 57 void dtrace_invop_init(void); 58 void dtrace_invop_uninit(void); 59 60 typedef struct dtrace_invop_hdlr { 61 int (*dtih_func)(uintptr_t, uintptr_t *, uintptr_t); 62 struct dtrace_invop_hdlr *dtih_next; 63 } dtrace_invop_hdlr_t; 64 65 dtrace_invop_hdlr_t *dtrace_invop_hdlr; 66 67 int 68 dtrace_invop(uintptr_t addr, uintptr_t *stack, uintptr_t arg0) 69 { 70 dtrace_invop_hdlr_t *hdlr; 71 int rval; 72 73 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) 74 if ((rval = hdlr->dtih_func(addr, stack, arg0)) != 0) 75 return (rval); 76 77 return (0); 78 } 79 80 void 81 dtrace_invop_add(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) 82 { 83 dtrace_invop_hdlr_t *hdlr; 84 85 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); 86 hdlr->dtih_func = func; 87 hdlr->dtih_next = dtrace_invop_hdlr; 88 dtrace_invop_hdlr = hdlr; 89 } 90 91 void 92 dtrace_invop_remove(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) 93 { 94 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; 95 96 for (;;) { 97 if (hdlr == NULL) 98 panic("attempt to remove non-existent invop handler"); 99 100 if (hdlr->dtih_func == func) 101 break; 102 103 prev = hdlr; 104 hdlr = hdlr->dtih_next; 105 } 106 107 if (prev == NULL) { 108 ASSERT(dtrace_invop_hdlr == hdlr); 109 dtrace_invop_hdlr = hdlr->dtih_next; 110 } else { 111 ASSERT(dtrace_invop_hdlr != hdlr); 112 prev->dtih_next = hdlr->dtih_next; 113 } 114 115 kmem_free(hdlr, 0); 116 } 117 118 119 /*ARGSUSED*/ 120 void 121 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) 122 { 123 /* 124 * No toxic regions? 125 */ 126 } 127 128 void 129 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) 130 { 131 cpuset_t cpus; 132 133 if (cpu == DTRACE_CPUALL) 134 cpus = all_cpus; 135 else 136 CPU_SETOF(cpu, &cpus); 137 138 smp_rendezvous_cpus(cpus, smp_no_rendevous_barrier, func, 139 smp_no_rendevous_barrier, arg); 140 } 141 142 static void 143 dtrace_sync_func(void) 144 { 145 } 146 147 void 148 dtrace_sync(void) 149 { 150 dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); 151 } 152 153 static int64_t tgt_cpu_tsc; 154 static int64_t hst_cpu_tsc; 155 static int64_t timebase_skew[MAXCPU]; 156 static uint64_t nsec_scale; 157 158 /* See below for the explanation of this macro. */ 159 /* This is taken from the amd64 dtrace_subr, to provide a synchronized timer 160 * between multiple processors in dtrace. Since PowerPC Timebases can be much 161 * lower than x86, the scale shift is 26 instead of 28, allowing for a 15.63MHz 162 * timebase. 163 */ 164 #define SCALE_SHIFT 26 165 166 static void 167 dtrace_gethrtime_init_cpu(void *arg) 168 { 169 uintptr_t cpu = (uintptr_t) arg; 170 171 if (cpu == curcpu) 172 tgt_cpu_tsc = mftb(); 173 else 174 hst_cpu_tsc = mftb(); 175 } 176 177 static void 178 dtrace_gethrtime_init(void *arg) 179 { 180 struct pcpu *pc; 181 uint64_t tb_f; 182 cpuset_t map; 183 int i; 184 185 tb_f = cpu_tickrate(); 186 187 /* 188 * The following line checks that nsec_scale calculated below 189 * doesn't overflow 32-bit unsigned integer, so that it can multiply 190 * another 32-bit integer without overflowing 64-bit. 191 * Thus minimum supported Timebase frequency is 15.63MHz. 192 */ 193 KASSERT(tb_f > (NANOSEC >> (32 - SCALE_SHIFT)), ("Timebase frequency is too low")); 194 195 /* 196 * We scale up NANOSEC/tb_f ratio to preserve as much precision 197 * as possible. 198 * 2^26 factor was chosen quite arbitrarily from practical 199 * considerations: 200 * - it supports TSC frequencies as low as 15.63MHz (see above); 201 */ 202 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tb_f; 203 204 /* The current CPU is the reference one. */ 205 sched_pin(); 206 timebase_skew[curcpu] = 0; 207 CPU_FOREACH(i) { 208 if (i == curcpu) 209 continue; 210 211 pc = pcpu_find(i); 212 CPU_SETOF(PCPU_GET(cpuid), &map); 213 CPU_SET(pc->pc_cpuid, &map); 214 215 smp_rendezvous_cpus(map, NULL, 216 dtrace_gethrtime_init_cpu, 217 smp_no_rendevous_barrier, (void *)(uintptr_t) i); 218 219 timebase_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; 220 } 221 sched_unpin(); 222 } 223 224 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, NULL); 225 226 /* 227 * DTrace needs a high resolution time function which can 228 * be called from a probe context and guaranteed not to have 229 * instrumented with probes itself. 230 * 231 * Returns nanoseconds since boot. 232 */ 233 uint64_t 234 dtrace_gethrtime() 235 { 236 uint64_t timebase; 237 uint32_t lo; 238 uint32_t hi; 239 240 /* 241 * We split timebase value into lower and higher 32-bit halves and separately 242 * scale them with nsec_scale, then we scale them down by 2^28 243 * (see nsec_scale calculations) taking into account 32-bit shift of 244 * the higher half and finally add. 245 */ 246 timebase = mftb() - timebase_skew[curcpu]; 247 lo = timebase; 248 hi = timebase >> 32; 249 return (((lo * nsec_scale) >> SCALE_SHIFT) + 250 ((hi * nsec_scale) << (32 - SCALE_SHIFT))); 251 } 252 253 uint64_t 254 dtrace_gethrestime(void) 255 { 256 struct timespec curtime; 257 258 dtrace_getnanotime(&curtime); 259 260 return (curtime.tv_sec * 1000000000UL + curtime.tv_nsec); 261 } 262 263 /* Function to handle DTrace traps during probes. See powerpc/powerpc/trap.c */ 264 int 265 dtrace_trap(struct trapframe *frame, u_int type) 266 { 267 /* 268 * A trap can occur while DTrace executes a probe. Before 269 * executing the probe, DTrace blocks re-scheduling and sets 270 * a flag in it's per-cpu flags to indicate that it doesn't 271 * want to fault. On returning from the probe, the no-fault 272 * flag is cleared and finally re-scheduling is enabled. 273 * 274 * Check if DTrace has enabled 'no-fault' mode: 275 * 276 */ 277 if ((cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT) != 0) { 278 /* 279 * There are only a couple of trap types that are expected. 280 * All the rest will be handled in the usual way. 281 */ 282 switch (type) { 283 /* Page fault. */ 284 case EXC_DSI: 285 case EXC_DSE: 286 /* Flag a bad address. */ 287 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 288 cpu_core[curcpu].cpuc_dtrace_illval = frame->cpu.aim.dar; 289 290 /* 291 * Offset the instruction pointer to the instruction 292 * following the one causing the fault. 293 */ 294 frame->srr0 += sizeof(int); 295 return (1); 296 case EXC_ISI: 297 case EXC_ISE: 298 /* Flag a bad address. */ 299 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 300 cpu_core[curcpu].cpuc_dtrace_illval = frame->srr0; 301 302 /* 303 * Offset the instruction pointer to the instruction 304 * following the one causing the fault. 305 */ 306 frame->srr0 += sizeof(int); 307 return (1); 308 default: 309 /* Handle all other traps in the usual way. */ 310 break; 311 } 312 } 313 314 /* Handle the trap in the usual way. */ 315 return (0); 316 } 317 318 void 319 dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, 320 int fault, int fltoffs, uintptr_t illval) 321 { 322 323 dtrace_probe(dtrace_probeid_error, (uint64_t)(uintptr_t)state, 324 (uintptr_t)epid, 325 (uintptr_t)which, (uintptr_t)fault, (uintptr_t)fltoffs); 326 } 327 328 static int 329 dtrace_invop_start(struct trapframe *frame) 330 { 331 switch (dtrace_invop(frame->srr0, (uintptr_t *)frame, frame->fixreg[3])) { 332 case DTRACE_INVOP_JUMP: 333 break; 334 case DTRACE_INVOP_BCTR: 335 frame->srr0 = frame->ctr; 336 break; 337 case DTRACE_INVOP_BLR: 338 frame->srr0 = frame->lr; 339 break; 340 case DTRACE_INVOP_MFLR_R0: 341 frame->fixreg[0] = frame->lr; 342 frame->srr0 = frame->srr0 + 4; 343 break; 344 default: 345 return (-1); 346 break; 347 } 348 349 return (0); 350 } 351 352 void dtrace_invop_init(void) 353 { 354 dtrace_invop_jump_addr = dtrace_invop_start; 355 } 356 357 void dtrace_invop_uninit(void) 358 { 359 dtrace_invop_jump_addr = 0; 360 } 361