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 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26/* #pragma ident "@(#)lockstat.c 1.12 08/01/16 SMI" */ 27 28 29#ifdef KERNEL 30#ifndef _KERNEL 31#define _KERNEL /* Solaris vs. Darwin */ 32#endif 33#endif 34 35#include <sys/param.h> 36#include <sys/systm.h> 37#include <sys/errno.h> 38#include <sys/stat.h> 39#include <sys/ioctl.h> 40#include <sys/conf.h> 41#include <sys/fcntl.h> 42#include <miscfs/devfs/devfs.h> 43 44#include <sys/dtrace.h> 45#include <sys/dtrace_impl.h> 46 47#include <sys/dtrace_glue.h> 48 49#include <sys/lockstat.h> 50 51#include <kern/processor.h> 52 53#define membar_producer dtrace_membar_producer 54 55/* 56 * Hot patch values, x86 57 */ 58#if defined(__x86_64__) 59#define NOP 0x90 60#define RET 0xc3 61#define LOCKSTAT_AFRAMES 1 62#else 63#error "not ported to this architecture" 64#endif 65 66typedef struct lockstat_probe { 67 const char *lsp_func; 68 const char *lsp_name; 69 int lsp_probe; 70 dtrace_id_t lsp_id; 71} lockstat_probe_t; 72 73lockstat_probe_t lockstat_probes[] = 74{ 75#if defined(__x86_64__) 76 /* Only provide implemented probes for each architecture */ 77 { LS_LCK_MTX_LOCK, LSA_ACQUIRE, LS_LCK_MTX_LOCK_ACQUIRE, DTRACE_IDNONE }, 78 { LS_LCK_MTX_LOCK, LSA_SPIN, LS_LCK_MTX_LOCK_SPIN, DTRACE_IDNONE }, 79 { LS_LCK_MTX_LOCK, LSA_BLOCK, LS_LCK_MTX_LOCK_BLOCK, DTRACE_IDNONE }, 80 { LS_LCK_MTX_TRY_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_LOCK_ACQUIRE, DTRACE_IDNONE }, 81 { LS_LCK_MTX_TRY_SPIN_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE }, 82 { LS_LCK_MTX_UNLOCK, LSA_RELEASE, LS_LCK_MTX_UNLOCK_RELEASE, DTRACE_IDNONE }, 83 { LS_LCK_MTX_EXT_LOCK, LSA_ACQUIRE, LS_LCK_MTX_EXT_LOCK_ACQUIRE, DTRACE_IDNONE }, 84 { LS_LCK_MTX_EXT_LOCK, LSA_SPIN, LS_LCK_MTX_EXT_LOCK_SPIN, DTRACE_IDNONE }, 85 { LS_LCK_MTX_EXT_LOCK, LSA_BLOCK, LS_LCK_MTX_EXT_LOCK_BLOCK, DTRACE_IDNONE }, 86// { LS_LCK_MTX_EXT_TRY_LOCK, LSA_ACQUIRE, LS_LCK_MTX_TRY_EXT_LOCK_ACQUIRE, DTRACE_IDNONE }, 87 { LS_LCK_MTX_EXT_UNLOCK, LSA_RELEASE, LS_LCK_MTX_EXT_UNLOCK_RELEASE, DTRACE_IDNONE }, 88 { LS_LCK_MTX_LOCK_SPIN_LOCK, LSA_ACQUIRE, LS_LCK_MTX_LOCK_SPIN_ACQUIRE, DTRACE_IDNONE }, 89 { LS_LCK_RW_LOCK_SHARED, LSR_ACQUIRE, LS_LCK_RW_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE }, 90 { LS_LCK_RW_LOCK_SHARED, LSR_BLOCK, LS_LCK_RW_LOCK_SHARED_BLOCK, DTRACE_IDNONE }, 91 { LS_LCK_RW_LOCK_SHARED, LSR_SPIN, LS_LCK_RW_LOCK_SHARED_SPIN, DTRACE_IDNONE }, 92 { LS_LCK_RW_LOCK_EXCL, LSR_ACQUIRE, LS_LCK_RW_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE }, 93 { LS_LCK_RW_LOCK_EXCL, LSR_BLOCK, LS_LCK_RW_LOCK_EXCL_BLOCK, DTRACE_IDNONE }, 94 { LS_LCK_RW_LOCK_EXCL, LSR_SPIN, LS_LCK_RW_LOCK_EXCL_SPIN, DTRACE_IDNONE }, 95 { LS_LCK_RW_DONE, LSR_RELEASE, LS_LCK_RW_DONE_RELEASE, DTRACE_IDNONE }, 96 { LS_LCK_RW_TRY_LOCK_SHARED, LSR_ACQUIRE, LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, DTRACE_IDNONE }, 97 { LS_LCK_RW_TRY_LOCK_EXCL, LSR_ACQUIRE, LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, DTRACE_IDNONE }, 98 { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_UPGRADE, LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE, DTRACE_IDNONE }, 99 { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_SPIN, LS_LCK_RW_LOCK_SHARED_TO_EXCL_SPIN, DTRACE_IDNONE }, 100 { LS_LCK_RW_LOCK_SHARED_TO_EXCL, LSR_BLOCK, LS_LCK_RW_LOCK_SHARED_TO_EXCL_BLOCK, DTRACE_IDNONE }, 101 { LS_LCK_RW_LOCK_EXCL_TO_SHARED, LSR_DOWNGRADE, LS_LCK_RW_LOCK_EXCL_TO_SHARED_DOWNGRADE, DTRACE_IDNONE }, 102#endif 103#ifdef LATER 104 /* Interlock and spinlock measurements would be nice, but later */ 105 { LS_LCK_SPIN_LOCK, LSS_ACQUIRE, LS_LCK_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE }, 106 { LS_LCK_SPIN_LOCK, LSS_SPIN, LS_LCK_SPIN_LOCK_SPIN, DTRACE_IDNONE }, 107 { LS_LCK_SPIN_UNLOCK, LSS_RELEASE, LS_LCK_SPIN_UNLOCK_RELEASE, DTRACE_IDNONE }, 108 109 { LS_LCK_RW_LOCK_EXCL_TO_SHARED, LSA_ILK_SPIN, LS_LCK_RW_LOCK_EXCL_TO_SHARED_ILK_SPIN, DTRACE_IDNONE }, 110 { LS_LCK_MTX_LOCK, LSA_ILK_SPIN, LS_LCK_MTX_LOCK_ILK_SPIN, DTRACE_IDNONE }, 111 { LS_LCK_MTX_EXT_LOCK, LSA_ILK_SPIN, LS_LCK_MTX_EXT_LOCK_ILK_SPIN, DTRACE_IDNONE }, 112 { LS_LCK_RW_TRY_LOCK_EXCL, LSA_ILK_SPIN, LS_LCK_RW_TRY_LOCK_EXCL_ILK_SPIN, DTRACE_IDNONE }, 113 { LS_LCK_RW_TRY_LOCK_SHARED, LSA_SPIN, LS_LCK_RW_TRY_LOCK_SHARED_SPIN, DTRACE_IDNONE }, 114#endif 115 116 { NULL, NULL, 0, 0 } 117}; 118 119dtrace_id_t lockstat_probemap[LS_NPROBES]; 120 121#if CONFIG_DTRACE 122#if defined(__x86_64__) 123extern void lck_mtx_lock_lockstat_patch_point(void); 124extern void lck_mtx_try_lock_lockstat_patch_point(void); 125extern void lck_mtx_try_lock_spin_lockstat_patch_point(void); 126extern void lck_mtx_unlock_lockstat_patch_point(void); 127extern void lck_mtx_lock_ext_lockstat_patch_point(void); 128extern void lck_mtx_ext_unlock_lockstat_patch_point(void); 129extern void lck_rw_lock_shared_lockstat_patch_point(void); 130extern void lck_rw_lock_exclusive_lockstat_patch_point(void); 131extern void lck_rw_lock_shared_to_exclusive_lockstat_patch_point(void); 132extern void lck_rw_try_lock_shared_lockstat_patch_point(void); 133extern void lck_rw_try_lock_exclusive_lockstat_patch_point(void); 134extern void lck_mtx_lock_spin_lockstat_patch_point(void); 135#endif 136 137#endif /* CONFIG_DTRACE */ 138 139typedef struct lockstat_assembly_probe { 140 int lsap_probe; 141 vm_offset_t * lsap_patch_point; 142} lockstat_assembly_probe_t; 143 144 145 lockstat_assembly_probe_t assembly_probes[] = 146 { 147#if CONFIG_DTRACE 148#if defined(__x86_64__) 149 /* 150 * On x86 these points are better done via hot patches, which ensure 151 * there is zero overhead when not in use. On x86 these patch points 152 * are swapped between the return instruction and a no-op, with the 153 * Dtrace call following the return. 154 */ 155 { LS_LCK_MTX_LOCK_ACQUIRE, (vm_offset_t *) lck_mtx_lock_lockstat_patch_point }, 156 { LS_LCK_MTX_TRY_LOCK_ACQUIRE, (vm_offset_t *) lck_mtx_try_lock_lockstat_patch_point }, 157 { LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE, (vm_offset_t *) lck_mtx_try_lock_spin_lockstat_patch_point }, 158 { LS_LCK_MTX_UNLOCK_RELEASE, (vm_offset_t *) lck_mtx_unlock_lockstat_patch_point }, 159 { LS_LCK_MTX_EXT_LOCK_ACQUIRE, (vm_offset_t *) lck_mtx_lock_ext_lockstat_patch_point }, 160 { LS_LCK_MTX_EXT_UNLOCK_RELEASE, (vm_offset_t *) lck_mtx_ext_unlock_lockstat_patch_point }, 161 { LS_LCK_RW_LOCK_SHARED_ACQUIRE, (vm_offset_t *) lck_rw_lock_shared_lockstat_patch_point }, 162 { LS_LCK_RW_LOCK_EXCL_ACQUIRE, (vm_offset_t *) lck_rw_lock_exclusive_lockstat_patch_point }, 163 { LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE,(vm_offset_t *) lck_rw_lock_shared_to_exclusive_lockstat_patch_point }, 164 { LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, (vm_offset_t *) lck_rw_try_lock_shared_lockstat_patch_point }, 165 { LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, (vm_offset_t *) lck_rw_try_lock_exclusive_lockstat_patch_point }, 166 { LS_LCK_MTX_LOCK_SPIN_ACQUIRE, (vm_offset_t *) lck_mtx_lock_spin_lockstat_patch_point }, 167#endif 168#endif /* CONFIG_DTRACE */ 169 { LS_LCK_INVALID, NULL } 170}; 171 172 173/* 174 * Hot patch switches back and forth the probe points between NOP and RET. 175 * The active argument indicates whether the probe point will turn on or off. 176 * on == plant a NOP and thus fall through to the probe call 177 * off == plant a RET and thus avoid the probe call completely 178 * The lsap_probe identifies which probe we will patch. 179 */ 180#if defined(__APPLE__) 181static 182void lockstat_hot_patch(boolean_t active, int ls_probe) 183{ 184#pragma unused(active) 185 int i; 186 187 /* 188 * Loop through entire table, in case there are 189 * multiple patch points per probe. 190 */ 191 for (i = 0; assembly_probes[i].lsap_patch_point; i++) { 192 if (ls_probe == assembly_probes[i].lsap_probe) 193#if defined(__x86_64__) 194 { 195 uint8_t instr; 196 instr = (active ? NOP : RET ); 197 (void) ml_nofault_copy( (vm_offset_t)&instr, *(assembly_probes[i].lsap_patch_point), 198 sizeof(instr)); 199 } 200#endif 201 } /* for */ 202} 203#endif /* __APPLE__*/ 204 205 206void (*lockstat_probe)(dtrace_id_t, uint64_t, uint64_t, 207 uint64_t, uint64_t, uint64_t); 208 209#if defined(__APPLE__) 210/* This wrapper is used by arm assembler hot patched probes */ 211void 212lockstat_probe_wrapper(int probe, uintptr_t lp, int rwflag) 213{ 214 dtrace_id_t id; 215 id = lockstat_probemap[probe]; 216 if (id != 0) 217 { 218 (*lockstat_probe)(id, (uintptr_t)lp, (uint64_t)rwflag, 0,0,0); 219 } 220} 221#endif /* __APPLE__ */ 222 223 224static dev_info_t *lockstat_devi; /* saved in xxattach() for xxinfo() */ 225static dtrace_provider_id_t lockstat_id; 226 227/*ARGSUSED*/ 228static int 229lockstat_enable(void *arg, dtrace_id_t id, void *parg) 230{ 231#pragma unused(arg) /* __APPLE__ */ 232 233 lockstat_probe_t *probe = parg; 234 235 ASSERT(!lockstat_probemap[probe->lsp_probe]); 236 237 lockstat_probemap[probe->lsp_probe] = id; 238 membar_producer(); 239 240 lockstat_hot_patch(TRUE, probe->lsp_probe); 241 membar_producer(); 242 return(0); 243 244} 245 246/*ARGSUSED*/ 247static void 248lockstat_disable(void *arg, dtrace_id_t id, void *parg) 249{ 250#pragma unused(arg, id) /* __APPLE__ */ 251 252 lockstat_probe_t *probe = parg; 253 int i; 254 255 ASSERT(lockstat_probemap[probe->lsp_probe]); 256 257 lockstat_probemap[probe->lsp_probe] = 0; 258 lockstat_hot_patch(FALSE, probe->lsp_probe); 259 membar_producer(); 260 261 /* 262 * See if we have any probes left enabled. 263 */ 264 for (i = 0; i < LS_NPROBES; i++) { 265 if (lockstat_probemap[i]) { 266 /* 267 * This probe is still enabled. We don't need to deal 268 * with waiting for all threads to be out of the 269 * lockstat critical sections; just return. 270 */ 271 return; 272 } 273 } 274 275} 276 277/*ARGSUSED*/ 278static void 279lockstat_provide(void *arg, const dtrace_probedesc_t *desc) 280{ 281#pragma unused(arg, desc) /* __APPLE__ */ 282 283 int i = 0; 284 285 for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) { 286 lockstat_probe_t *probe = &lockstat_probes[i]; 287 288 if (dtrace_probe_lookup(lockstat_id, "mach_kernel", 289 probe->lsp_func, probe->lsp_name) != 0) 290 continue; 291 292 ASSERT(!probe->lsp_id); 293 probe->lsp_id = dtrace_probe_create(lockstat_id, 294 "mach_kernel", probe->lsp_func, probe->lsp_name, 295 LOCKSTAT_AFRAMES, probe); 296 } 297} 298 299 300/*ARGSUSED*/ 301static void 302lockstat_destroy(void *arg, dtrace_id_t id, void *parg) 303{ 304#pragma unused(arg, id) /* __APPLE__ */ 305 306 lockstat_probe_t *probe = parg; 307 308 ASSERT(!lockstat_probemap[probe->lsp_probe]); 309 probe->lsp_id = 0; 310} 311 312static dtrace_pattr_t lockstat_attr = { 313{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 314{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 315{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 316{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 317{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 318}; 319 320static dtrace_pops_t lockstat_pops = { 321 lockstat_provide, 322 NULL, 323 lockstat_enable, 324 lockstat_disable, 325 NULL, 326 NULL, 327 NULL, 328 NULL, 329 NULL, 330 lockstat_destroy 331}; 332 333static int 334lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) 335{ 336 switch (cmd) { 337 case DDI_ATTACH: 338 break; 339 case DDI_RESUME: 340 return (DDI_SUCCESS); 341 default: 342 return (DDI_FAILURE); 343 } 344 345 if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0, 346 DDI_PSEUDO, 0) == DDI_FAILURE || 347 dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL, 348 NULL, &lockstat_pops, NULL, &lockstat_id) != 0) { 349 ddi_remove_minor_node(devi, NULL); 350 return (DDI_FAILURE); 351 } 352 353 lockstat_probe = dtrace_probe; 354 membar_producer(); 355 356 ddi_report_dev(devi); 357 lockstat_devi = devi; 358 return (DDI_SUCCESS); 359} 360 361d_open_t _lockstat_open; 362 363int _lockstat_open(dev_t dev, int flags, int devtype, struct proc *p) 364{ 365#pragma unused(dev,flags,devtype,p) 366 return 0; 367} 368 369#define LOCKSTAT_MAJOR -24 /* let the kernel pick the device number */ 370 371/* 372 * A struct describing which functions will get invoked for certain 373 * actions. 374 */ 375static struct cdevsw lockstat_cdevsw = 376{ 377 _lockstat_open, /* open */ 378 eno_opcl, /* close */ 379 eno_rdwrt, /* read */ 380 eno_rdwrt, /* write */ 381 eno_ioctl, /* ioctl */ 382 (stop_fcn_t *)nulldev, /* stop */ 383 (reset_fcn_t *)nulldev, /* reset */ 384 NULL, /* tty's */ 385 eno_select, /* select */ 386 eno_mmap, /* mmap */ 387 eno_strat, /* strategy */ 388 eno_getc, /* getc */ 389 eno_putc, /* putc */ 390 0 /* type */ 391}; 392 393static int gLockstatInited = 0; 394 395void lockstat_init( void ); 396 397void lockstat_init( void ) 398{ 399 if (0 == gLockstatInited) 400 { 401 int majdevno = cdevsw_add(LOCKSTAT_MAJOR, &lockstat_cdevsw); 402 403 if (majdevno < 0) { 404 printf("lockstat_init: failed to allocate a major number!\n"); 405 gLockstatInited = 0; 406 return; 407 } 408 409 lockstat_attach( (dev_info_t *)(uintptr_t)majdevno, DDI_ATTACH ); 410 gLockstatInited = 1; 411 } else 412 panic("lockstat_init: called twice!\n"); 413} 414#undef LOCKSTAT_MAJOR 415