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 */ 23 /* 24 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 25 * Use is subject to license terms. 26 */ 27 28 #include <sys/param.h> 29 #include <sys/systm.h> 30 #include <sys/kernel.h> 31 #include <sys/malloc.h> 32 #include <sys/kmem.h> 33 #include <sys/proc.h> 34 #include <sys/smp.h> 35 #include <sys/dtrace_impl.h> 36 #include <sys/dtrace_bsd.h> 37 #include <cddl/dev/dtrace/dtrace_cddl.h> 38 #include <machine/clock.h> 39 #include <machine/frame.h> 40 #include <machine/trap.h> 41 #include <vm/pmap.h> 42 43 #define DELAYBRANCH(x) ((int)(x) < 0) 44 45 extern dtrace_id_t dtrace_probeid_error; 46 extern int (*dtrace_invop_jump_addr)(struct trapframe *); 47 48 extern void dtrace_getnanotime(struct timespec *tsp); 49 50 int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t); 51 void dtrace_invop_init(void); 52 void dtrace_invop_uninit(void); 53 54 typedef struct dtrace_invop_hdlr { 55 int (*dtih_func)(uintptr_t, struct trapframe *, uintptr_t); 56 struct dtrace_invop_hdlr *dtih_next; 57 } dtrace_invop_hdlr_t; 58 59 dtrace_invop_hdlr_t *dtrace_invop_hdlr; 60 61 int 62 dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t arg0) 63 { 64 struct thread *td; 65 dtrace_invop_hdlr_t *hdlr; 66 int rval; 67 68 rval = 0; 69 td = curthread; 70 td->t_dtrace_trapframe = frame; 71 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) 72 if ((rval = hdlr->dtih_func(addr, frame, arg0)) != 0) 73 break; 74 td->t_dtrace_trapframe = NULL; 75 return (rval); 76 } 77 78 void 79 dtrace_invop_add(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 80 { 81 dtrace_invop_hdlr_t *hdlr; 82 83 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); 84 hdlr->dtih_func = func; 85 hdlr->dtih_next = dtrace_invop_hdlr; 86 dtrace_invop_hdlr = hdlr; 87 } 88 89 void 90 dtrace_invop_remove(int (*func)(uintptr_t, struct trapframe *, uintptr_t)) 91 { 92 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; 93 94 for (;;) { 95 if (hdlr == NULL) 96 panic("attempt to remove non-existent invop handler"); 97 98 if (hdlr->dtih_func == func) 99 break; 100 101 prev = hdlr; 102 hdlr = hdlr->dtih_next; 103 } 104 105 if (prev == NULL) { 106 ASSERT(dtrace_invop_hdlr == hdlr); 107 dtrace_invop_hdlr = hdlr->dtih_next; 108 } else { 109 ASSERT(dtrace_invop_hdlr != hdlr); 110 prev->dtih_next = hdlr->dtih_next; 111 } 112 113 kmem_free(hdlr, 0); 114 } 115 116 117 /*ARGSUSED*/ 118 void 119 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) 120 { 121 /* 122 * No toxic regions? 123 */ 124 } 125 126 void 127 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) 128 { 129 cpuset_t cpus; 130 131 if (cpu == DTRACE_CPUALL) 132 cpus = all_cpus; 133 else 134 CPU_SETOF(cpu, &cpus); 135 136 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, func, 137 smp_no_rendezvous_barrier, arg); 138 } 139 140 static void 141 dtrace_sync_func(void) 142 { 143 } 144 145 void 146 dtrace_sync(void) 147 { 148 dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); 149 } 150 151 static int64_t tgt_cpu_tsc; 152 static int64_t hst_cpu_tsc; 153 static int64_t timebase_skew[MAXCPU]; 154 static uint64_t nsec_scale; 155 156 /* See below for the explanation of this macro. */ 157 /* This is taken from the amd64 dtrace_subr, to provide a synchronized timer 158 * between multiple processors in dtrace. Since PowerPC Timebases can be much 159 * lower than x86, the scale shift is 26 instead of 28, allowing for a 15.63MHz 160 * timebase. 161 */ 162 #define SCALE_SHIFT 26 163 164 static void 165 dtrace_gethrtime_init_cpu(void *arg) 166 { 167 uintptr_t cpu = (uintptr_t) arg; 168 169 if (cpu == curcpu) 170 tgt_cpu_tsc = mftb(); 171 else 172 hst_cpu_tsc = mftb(); 173 } 174 175 static void 176 dtrace_gethrtime_init(void *arg) 177 { 178 struct pcpu *pc; 179 uint64_t tb_f; 180 cpuset_t map; 181 int i; 182 183 tb_f = cpu_tickrate(); 184 185 /* 186 * The following line checks that nsec_scale calculated below 187 * doesn't overflow 32-bit unsigned integer, so that it can multiply 188 * another 32-bit integer without overflowing 64-bit. 189 * Thus minimum supported Timebase frequency is 15.63MHz. 190 */ 191 KASSERT(tb_f > (NANOSEC >> (32 - SCALE_SHIFT)), ("Timebase frequency is too low")); 192 193 /* 194 * We scale up NANOSEC/tb_f ratio to preserve as much precision 195 * as possible. 196 * 2^26 factor was chosen quite arbitrarily from practical 197 * considerations: 198 * - it supports TSC frequencies as low as 15.63MHz (see above); 199 */ 200 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tb_f; 201 202 /* The current CPU is the reference one. */ 203 sched_pin(); 204 timebase_skew[curcpu] = 0; 205 CPU_FOREACH(i) { 206 if (i == curcpu) 207 continue; 208 209 pc = pcpu_find(i); 210 CPU_SETOF(PCPU_GET(cpuid), &map); 211 CPU_SET(pc->pc_cpuid, &map); 212 213 smp_rendezvous_cpus(map, NULL, 214 dtrace_gethrtime_init_cpu, 215 smp_no_rendezvous_barrier, (void *)(uintptr_t) i); 216 217 timebase_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; 218 } 219 sched_unpin(); 220 } 221 #ifdef EARLY_AP_STARTUP 222 SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY, 223 dtrace_gethrtime_init, NULL); 224 #else 225 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, 226 NULL); 227 #endif 228 229 /* 230 * DTrace needs a high resolution time function which can 231 * be called from a probe context and guaranteed not to have 232 * instrumented with probes itself. 233 * 234 * Returns nanoseconds since boot. 235 */ 236 uint64_t 237 dtrace_gethrtime(void) 238 { 239 uint64_t timebase; 240 uint32_t lo; 241 uint32_t hi; 242 243 /* 244 * We split timebase value into lower and higher 32-bit halves and separately 245 * scale them with nsec_scale, then we scale them down by 2^28 246 * (see nsec_scale calculations) taking into account 32-bit shift of 247 * the higher half and finally add. 248 */ 249 timebase = mftb() - timebase_skew[curcpu]; 250 lo = timebase; 251 hi = timebase >> 32; 252 return (((lo * nsec_scale) >> SCALE_SHIFT) + 253 ((hi * nsec_scale) << (32 - SCALE_SHIFT))); 254 } 255 256 uint64_t 257 dtrace_gethrestime(void) 258 { 259 struct timespec curtime; 260 261 dtrace_getnanotime(&curtime); 262 263 return (curtime.tv_sec * 1000000000UL + curtime.tv_nsec); 264 } 265 266 /* Function to handle DTrace traps during probes. See powerpc/powerpc/trap.c */ 267 int 268 dtrace_trap(struct trapframe *frame, u_int type) 269 { 270 uint16_t nofault; 271 272 /* 273 * A trap can occur while DTrace executes a probe. Before 274 * executing the probe, DTrace blocks re-scheduling and sets 275 * a flag in its per-cpu flags to indicate that it doesn't 276 * want to fault. On returning from the probe, the no-fault 277 * flag is cleared and finally re-scheduling is enabled. 278 * 279 * Check if DTrace has enabled 'no-fault' mode: 280 */ 281 sched_pin(); 282 nofault = cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT; 283 sched_unpin(); 284 if (nofault) { 285 KASSERT((frame->srr1 & PSL_EE) == 0, ("interrupts enabled")); 286 /* 287 * There are only a couple of trap types that are expected. 288 * All the rest will be handled in the usual way. 289 */ 290 switch (type) { 291 /* Page fault. */ 292 case EXC_DSI: 293 case EXC_DSE: 294 /* Flag a bad address. */ 295 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 296 cpu_core[curcpu].cpuc_dtrace_illval = frame->dar; 297 298 /* 299 * Offset the instruction pointer to the instruction 300 * following the one causing the fault. 301 */ 302 frame->srr0 += sizeof(int); 303 return (1); 304 case EXC_ISI: 305 case EXC_ISE: 306 /* Flag a bad address. */ 307 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; 308 cpu_core[curcpu].cpuc_dtrace_illval = frame->srr0; 309 310 /* 311 * Offset the instruction pointer to the instruction 312 * following the one causing the fault. 313 */ 314 frame->srr0 += sizeof(int); 315 return (1); 316 default: 317 /* Handle all other traps in the usual way. */ 318 break; 319 } 320 } 321 322 /* Handle the trap in the usual way. */ 323 return (0); 324 } 325 326 void 327 dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, 328 int fault, int fltoffs, uintptr_t illval) 329 { 330 331 dtrace_probe(dtrace_probeid_error, (uint64_t)(uintptr_t)state, 332 (uintptr_t)epid, 333 (uintptr_t)which, (uintptr_t)fault, (uintptr_t)fltoffs); 334 } 335 336 static int 337 dtrace_invop_start(struct trapframe *frame) 338 { 339 340 switch (dtrace_invop(frame->srr0, frame, frame->fixreg[3])) { 341 case DTRACE_INVOP_JUMP: 342 break; 343 case DTRACE_INVOP_BCTR: 344 frame->srr0 = frame->ctr; 345 break; 346 case DTRACE_INVOP_BLR: 347 frame->srr0 = frame->lr; 348 break; 349 case DTRACE_INVOP_MFLR_R0: 350 frame->fixreg[0] = frame->lr; 351 frame->srr0 = frame->srr0 + 4; 352 break; 353 default: 354 return (-1); 355 } 356 return (0); 357 } 358 359 void dtrace_invop_init(void) 360 { 361 dtrace_invop_jump_addr = dtrace_invop_start; 362 } 363 364 void dtrace_invop_uninit(void) 365 { 366 dtrace_invop_jump_addr = 0; 367 } 368