Mercurial > hg > truffle
annotate src/os/linux/vm/os_linux.cpp @ 79:82db0859acbe
6642862: Code cache allocation fails with large pages after 6588638
Reviewed-by: apetrusenko
| author | jcoomes |
|---|---|
| date | Fri, 28 Mar 2008 23:35:42 -0700 |
| parents | cd0742ba123c |
| children | fcbfc50865ab |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. | |
| 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
| 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 // do not include precompiled header file | |
| 26 # include "incls/_os_linux.cpp.incl" | |
| 27 | |
| 28 // put OS-includes here | |
| 29 # include <sys/types.h> | |
| 30 # include <sys/mman.h> | |
| 31 # include <pthread.h> | |
| 32 # include <signal.h> | |
| 33 # include <errno.h> | |
| 34 # include <dlfcn.h> | |
| 35 # include <stdio.h> | |
| 36 # include <unistd.h> | |
| 37 # include <sys/resource.h> | |
| 38 # include <pthread.h> | |
| 39 # include <sys/stat.h> | |
| 40 # include <sys/time.h> | |
| 41 # include <sys/times.h> | |
| 42 # include <sys/utsname.h> | |
| 43 # include <sys/socket.h> | |
| 44 # include <sys/wait.h> | |
| 45 # include <pwd.h> | |
| 46 # include <poll.h> | |
| 47 # include <semaphore.h> | |
| 48 # include <fcntl.h> | |
| 49 # include <string.h> | |
| 50 # include <syscall.h> | |
| 51 # include <sys/sysinfo.h> | |
| 52 # include <gnu/libc-version.h> | |
| 53 # include <sys/ipc.h> | |
| 54 # include <sys/shm.h> | |
| 55 # include <link.h> | |
| 56 | |
| 57 #define MAX_PATH (2 * K) | |
| 58 | |
| 59 // for timer info max values which include all bits | |
| 60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) | |
| 61 #define SEC_IN_NANOSECS 1000000000LL | |
| 62 | |
| 63 //////////////////////////////////////////////////////////////////////////////// | |
| 64 // global variables | |
| 65 julong os::Linux::_physical_memory = 0; | |
| 66 | |
| 67 address os::Linux::_initial_thread_stack_bottom = NULL; | |
| 68 uintptr_t os::Linux::_initial_thread_stack_size = 0; | |
| 69 | |
| 70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; | |
| 71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; | |
| 72 Mutex* os::Linux::_createThread_lock = NULL; | |
| 73 pthread_t os::Linux::_main_thread; | |
| 74 int os::Linux::_page_size = -1; | |
| 75 bool os::Linux::_is_floating_stack = false; | |
| 76 bool os::Linux::_is_NPTL = false; | |
| 77 bool os::Linux::_supports_fast_thread_cpu_time = false; | |
| 78 char * os::Linux::_glibc_version = NULL; | |
| 79 char * os::Linux::_libpthread_version = NULL; | |
| 80 | |
| 81 static jlong initial_time_count=0; | |
| 82 | |
| 83 static int clock_tics_per_sec = 100; | |
| 84 | |
| 85 // For diagnostics to print a message once. see run_periodic_checks | |
| 86 static sigset_t check_signal_done; | |
| 87 static bool check_signals = true;; | |
| 88 | |
| 89 static pid_t _initial_pid = 0; | |
| 90 | |
| 91 /* Signal number used to suspend/resume a thread */ | |
| 92 | |
| 93 /* do not use any signal number less than SIGSEGV, see 4355769 */ | |
| 94 static int SR_signum = SIGUSR2; | |
| 95 sigset_t SR_sigset; | |
| 96 | |
| 97 //////////////////////////////////////////////////////////////////////////////// | |
| 98 // utility functions | |
| 99 | |
| 100 static int SR_initialize(); | |
| 101 static int SR_finalize(); | |
| 102 | |
| 103 julong os::available_memory() { | |
| 104 return Linux::available_memory(); | |
| 105 } | |
| 106 | |
| 107 julong os::Linux::available_memory() { | |
| 108 // values in struct sysinfo are "unsigned long" | |
| 109 struct sysinfo si; | |
| 110 sysinfo(&si); | |
| 111 | |
| 112 return (julong)si.freeram * si.mem_unit; | |
| 113 } | |
| 114 | |
| 115 julong os::physical_memory() { | |
| 116 return Linux::physical_memory(); | |
| 117 } | |
| 118 | |
|
20
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
119 julong os::allocatable_physical_memory(julong size) { |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
120 #ifdef _LP64 |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
121 return size; |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
122 #else |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
123 julong result = MIN2(size, (julong)3800*M); |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
124 if (!is_allocatable(result)) { |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
125 // See comments under solaris for alignment considerations |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
126 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
127 result = MIN2(size, reasonable_size); |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
128 } |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
129 return result; |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
130 #endif // _LP64 |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
131 } |
|
e195fe4c40c7
6629887: 64-bit windows should not restrict default heap size to 1400m
phh
parents:
0
diff
changeset
|
132 |
| 0 | 133 //////////////////////////////////////////////////////////////////////////////// |
| 134 // environment support | |
| 135 | |
| 136 bool os::getenv(const char* name, char* buf, int len) { | |
| 137 const char* val = ::getenv(name); | |
| 138 if (val != NULL && strlen(val) < (size_t)len) { | |
| 139 strcpy(buf, val); | |
| 140 return true; | |
| 141 } | |
| 142 if (len > 0) buf[0] = 0; // return a null string | |
| 143 return false; | |
| 144 } | |
| 145 | |
| 146 | |
| 147 // Return true if user is running as root. | |
| 148 | |
| 149 bool os::have_special_privileges() { | |
| 150 static bool init = false; | |
| 151 static bool privileges = false; | |
| 152 if (!init) { | |
| 153 privileges = (getuid() != geteuid()) || (getgid() != getegid()); | |
| 154 init = true; | |
| 155 } | |
| 156 return privileges; | |
| 157 } | |
| 158 | |
| 159 | |
| 160 #ifndef SYS_gettid | |
| 161 // i386: 224, ia64: 1105, amd64: 186, sparc 143 | |
| 162 #ifdef __ia64__ | |
| 163 #define SYS_gettid 1105 | |
| 164 #elif __i386__ | |
| 165 #define SYS_gettid 224 | |
| 166 #elif __amd64__ | |
| 167 #define SYS_gettid 186 | |
| 168 #elif __sparc__ | |
| 169 #define SYS_gettid 143 | |
| 170 #else | |
| 171 #error define gettid for the arch | |
| 172 #endif | |
| 173 #endif | |
| 174 | |
| 175 // Cpu architecture string | |
| 176 #if defined(IA64) | |
| 177 static char cpu_arch[] = "ia64"; | |
| 178 #elif defined(IA32) | |
| 179 static char cpu_arch[] = "i386"; | |
| 180 #elif defined(AMD64) | |
| 181 static char cpu_arch[] = "amd64"; | |
| 182 #elif defined(SPARC) | |
| 183 # ifdef _LP64 | |
| 184 static char cpu_arch[] = "sparcv9"; | |
| 185 # else | |
| 186 static char cpu_arch[] = "sparc"; | |
| 187 # endif | |
| 188 #else | |
| 189 #error Add appropriate cpu_arch setting | |
| 190 #endif | |
| 191 | |
| 192 | |
| 193 // pid_t gettid() | |
| 194 // | |
| 195 // Returns the kernel thread id of the currently running thread. Kernel | |
| 196 // thread id is used to access /proc. | |
| 197 // | |
| 198 // (Note that getpid() on LinuxThreads returns kernel thread id too; but | |
| 199 // on NPTL, it returns the same pid for all threads, as required by POSIX.) | |
| 200 // | |
| 201 pid_t os::Linux::gettid() { | |
| 202 int rslt = syscall(SYS_gettid); | |
| 203 if (rslt == -1) { | |
| 204 // old kernel, no NPTL support | |
| 205 return getpid(); | |
| 206 } else { | |
| 207 return (pid_t)rslt; | |
| 208 } | |
| 209 } | |
| 210 | |
| 211 // Most versions of linux have a bug where the number of processors are | |
| 212 // determined by looking at the /proc file system. In a chroot environment, | |
| 213 // the system call returns 1. This causes the VM to act as if it is | |
| 214 // a single processor and elide locking (see is_MP() call). | |
| 215 static bool unsafe_chroot_detected = false; | |
| 216 static char *unstable_chroot_error = "/proc file system not found.\n" | |
| 217 "Java may be unstable running multithreaded in a chroot " | |
| 218 "environment on Linux when /proc filesystem is not mounted."; | |
| 219 | |
| 220 void os::Linux::initialize_system_info() { | |
| 221 _processor_count = sysconf(_SC_NPROCESSORS_CONF); | |
| 222 if (_processor_count == 1) { | |
| 223 pid_t pid = os::Linux::gettid(); | |
| 224 char fname[32]; | |
| 225 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); | |
| 226 FILE *fp = fopen(fname, "r"); | |
| 227 if (fp == NULL) { | |
| 228 unsafe_chroot_detected = true; | |
| 229 } else { | |
| 230 fclose(fp); | |
| 231 } | |
| 232 } | |
| 233 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); | |
| 234 assert(_processor_count > 0, "linux error"); | |
| 235 } | |
| 236 | |
| 237 void os::init_system_properties_values() { | |
| 238 // char arch[12]; | |
| 239 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); | |
| 240 | |
| 241 // The next steps are taken in the product version: | |
| 242 // | |
| 243 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. | |
| 244 // This library should be located at: | |
| 245 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. | |
| 246 // | |
| 247 // If "/jre/lib/" appears at the right place in the path, then we | |
| 248 // assume libjvm[_g].so is installed in a JDK and we use this path. | |
| 249 // | |
| 250 // Otherwise exit with message: "Could not create the Java virtual machine." | |
| 251 // | |
| 252 // The following extra steps are taken in the debugging version: | |
| 253 // | |
| 254 // If "/jre/lib/" does NOT appear at the right place in the path | |
| 255 // instead of exit check for $JAVA_HOME environment variable. | |
| 256 // | |
| 257 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, | |
| 258 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so | |
| 259 // it looks like libjvm[_g].so is installed there | |
| 260 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. | |
| 261 // | |
| 262 // Otherwise exit. | |
| 263 // | |
| 264 // Important note: if the location of libjvm.so changes this | |
| 265 // code needs to be changed accordingly. | |
| 266 | |
| 267 // The next few definitions allow the code to be verbatim: | |
| 268 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) | |
| 269 #define getenv(n) ::getenv(n) | |
| 270 | |
| 271 /* | |
| 272 * See ld(1): | |
| 273 * The linker uses the following search paths to locate required | |
| 274 * shared libraries: | |
| 275 * 1: ... | |
| 276 * ... | |
| 277 * 7: The default directories, normally /lib and /usr/lib. | |
| 278 */ | |
| 279 #define DEFAULT_LIBPATH "/lib:/usr/lib" | |
| 280 | |
| 281 #define EXTENSIONS_DIR "/lib/ext" | |
| 282 #define ENDORSED_DIR "/lib/endorsed" | |
| 283 #define REG_DIR "/usr/java/packages" | |
| 284 | |
| 285 { | |
| 286 /* sysclasspath, java_home, dll_dir */ | |
| 287 { | |
| 288 char *home_path; | |
| 289 char *dll_path; | |
| 290 char *pslash; | |
| 291 char buf[MAXPATHLEN]; | |
| 292 os::jvm_path(buf, sizeof(buf)); | |
| 293 | |
| 294 // Found the full path to libjvm.so. | |
| 295 // Now cut the path to <java_home>/jre if we can. | |
| 296 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ | |
| 297 pslash = strrchr(buf, '/'); | |
| 298 if (pslash != NULL) | |
| 299 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ | |
| 300 dll_path = malloc(strlen(buf) + 1); | |
| 301 if (dll_path == NULL) | |
| 302 return; | |
| 303 strcpy(dll_path, buf); | |
| 304 Arguments::set_dll_dir(dll_path); | |
| 305 | |
| 306 if (pslash != NULL) { | |
| 307 pslash = strrchr(buf, '/'); | |
| 308 if (pslash != NULL) { | |
| 309 *pslash = '\0'; /* get rid of /<arch> */ | |
| 310 pslash = strrchr(buf, '/'); | |
| 311 if (pslash != NULL) | |
| 312 *pslash = '\0'; /* get rid of /lib */ | |
| 313 } | |
| 314 } | |
| 315 | |
| 316 home_path = malloc(strlen(buf) + 1); | |
| 317 if (home_path == NULL) | |
| 318 return; | |
| 319 strcpy(home_path, buf); | |
| 320 Arguments::set_java_home(home_path); | |
| 321 | |
| 322 if (!set_boot_path('/', ':')) | |
| 323 return; | |
| 324 } | |
| 325 | |
| 326 /* | |
| 327 * Where to look for native libraries | |
| 328 * | |
| 329 * Note: Due to a legacy implementation, most of the library path | |
| 330 * is set in the launcher. This was to accomodate linking restrictions | |
| 331 * on legacy Linux implementations (which are no longer supported). | |
| 332 * Eventually, all the library path setting will be done here. | |
| 333 * | |
| 334 * However, to prevent the proliferation of improperly built native | |
| 335 * libraries, the new path component /usr/java/packages is added here. | |
| 336 * Eventually, all the library path setting will be done here. | |
| 337 */ | |
| 338 { | |
| 339 char *ld_library_path; | |
| 340 | |
| 341 /* | |
| 342 * Construct the invariant part of ld_library_path. Note that the | |
| 343 * space for the colon and the trailing null are provided by the | |
| 344 * nulls included by the sizeof operator (so actually we allocate | |
| 345 * a byte more than necessary). | |
| 346 */ | |
| 347 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + | |
| 348 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); | |
| 349 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); | |
| 350 | |
| 351 /* | |
| 352 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It | |
| 353 * should always exist (until the legacy problem cited above is | |
| 354 * addressed). | |
| 355 */ | |
| 356 char *v = getenv("LD_LIBRARY_PATH"); | |
| 357 if (v != NULL) { | |
| 358 char *t = ld_library_path; | |
| 359 /* That's +1 for the colon and +1 for the trailing '\0' */ | |
| 360 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); | |
| 361 sprintf(ld_library_path, "%s:%s", v, t); | |
| 362 } | |
| 363 Arguments::set_library_path(ld_library_path); | |
| 364 } | |
| 365 | |
| 366 /* | |
| 367 * Extensions directories. | |
| 368 * | |
| 369 * Note that the space for the colon and the trailing null are provided | |
| 370 * by the nulls included by the sizeof operator (so actually one byte more | |
| 371 * than necessary is allocated). | |
| 372 */ | |
| 373 { | |
| 374 char *buf = malloc(strlen(Arguments::get_java_home()) + | |
| 375 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); | |
| 376 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, | |
| 377 Arguments::get_java_home()); | |
| 378 Arguments::set_ext_dirs(buf); | |
| 379 } | |
| 380 | |
| 381 /* Endorsed standards default directory. */ | |
| 382 { | |
| 383 char * buf; | |
| 384 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); | |
| 385 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); | |
| 386 Arguments::set_endorsed_dirs(buf); | |
| 387 } | |
| 388 } | |
| 389 | |
| 390 #undef malloc | |
| 391 #undef getenv | |
| 392 #undef EXTENSIONS_DIR | |
| 393 #undef ENDORSED_DIR | |
| 394 | |
| 395 // Done | |
| 396 return; | |
| 397 } | |
| 398 | |
| 399 //////////////////////////////////////////////////////////////////////////////// | |
| 400 // breakpoint support | |
| 401 | |
| 402 void os::breakpoint() { | |
| 403 BREAKPOINT; | |
| 404 } | |
| 405 | |
| 406 extern "C" void breakpoint() { | |
| 407 // use debugger to set breakpoint here | |
| 408 } | |
| 409 | |
| 410 //////////////////////////////////////////////////////////////////////////////// | |
| 411 // signal support | |
| 412 | |
| 413 debug_only(static bool signal_sets_initialized = false); | |
| 414 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; | |
| 415 | |
| 416 bool os::Linux::is_sig_ignored(int sig) { | |
| 417 struct sigaction oact; | |
| 418 sigaction(sig, (struct sigaction*)NULL, &oact); | |
| 419 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) | |
| 420 : CAST_FROM_FN_PTR(void*, oact.sa_handler); | |
| 421 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) | |
| 422 return true; | |
| 423 else | |
| 424 return false; | |
| 425 } | |
| 426 | |
| 427 void os::Linux::signal_sets_init() { | |
| 428 // Should also have an assertion stating we are still single-threaded. | |
| 429 assert(!signal_sets_initialized, "Already initialized"); | |
| 430 // Fill in signals that are necessarily unblocked for all threads in | |
| 431 // the VM. Currently, we unblock the following signals: | |
| 432 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden | |
| 433 // by -Xrs (=ReduceSignalUsage)); | |
| 434 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all | |
| 435 // other threads. The "ReduceSignalUsage" boolean tells us not to alter | |
| 436 // the dispositions or masks wrt these signals. | |
| 437 // Programs embedding the VM that want to use the above signals for their | |
| 438 // own purposes must, at this time, use the "-Xrs" option to prevent | |
| 439 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. | |
| 440 // (See bug 4345157, and other related bugs). | |
| 441 // In reality, though, unblocking these signals is really a nop, since | |
| 442 // these signals are not blocked by default. | |
| 443 sigemptyset(&unblocked_sigs); | |
| 444 sigemptyset(&allowdebug_blocked_sigs); | |
| 445 sigaddset(&unblocked_sigs, SIGILL); | |
| 446 sigaddset(&unblocked_sigs, SIGSEGV); | |
| 447 sigaddset(&unblocked_sigs, SIGBUS); | |
| 448 sigaddset(&unblocked_sigs, SIGFPE); | |
| 449 sigaddset(&unblocked_sigs, SR_signum); | |
| 450 | |
| 451 if (!ReduceSignalUsage) { | |
| 452 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { | |
| 453 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); | |
| 454 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); | |
| 455 } | |
| 456 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { | |
| 457 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); | |
| 458 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); | |
| 459 } | |
| 460 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { | |
| 461 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); | |
| 462 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); | |
| 463 } | |
| 464 } | |
| 465 // Fill in signals that are blocked by all but the VM thread. | |
| 466 sigemptyset(&vm_sigs); | |
| 467 if (!ReduceSignalUsage) | |
| 468 sigaddset(&vm_sigs, BREAK_SIGNAL); | |
| 469 debug_only(signal_sets_initialized = true); | |
| 470 | |
| 471 } | |
| 472 | |
| 473 // These are signals that are unblocked while a thread is running Java. | |
| 474 // (For some reason, they get blocked by default.) | |
| 475 sigset_t* os::Linux::unblocked_signals() { | |
| 476 assert(signal_sets_initialized, "Not initialized"); | |
| 477 return &unblocked_sigs; | |
| 478 } | |
| 479 | |
| 480 // These are the signals that are blocked while a (non-VM) thread is | |
| 481 // running Java. Only the VM thread handles these signals. | |
| 482 sigset_t* os::Linux::vm_signals() { | |
| 483 assert(signal_sets_initialized, "Not initialized"); | |
| 484 return &vm_sigs; | |
| 485 } | |
| 486 | |
| 487 // These are signals that are blocked during cond_wait to allow debugger in | |
| 488 sigset_t* os::Linux::allowdebug_blocked_signals() { | |
| 489 assert(signal_sets_initialized, "Not initialized"); | |
| 490 return &allowdebug_blocked_sigs; | |
| 491 } | |
| 492 | |
| 493 void os::Linux::hotspot_sigmask(Thread* thread) { | |
| 494 | |
| 495 //Save caller's signal mask before setting VM signal mask | |
| 496 sigset_t caller_sigmask; | |
| 497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); | |
| 498 | |
| 499 OSThread* osthread = thread->osthread(); | |
| 500 osthread->set_caller_sigmask(caller_sigmask); | |
| 501 | |
| 502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); | |
| 503 | |
| 504 if (!ReduceSignalUsage) { | |
| 505 if (thread->is_VM_thread()) { | |
| 506 // Only the VM thread handles BREAK_SIGNAL ... | |
| 507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); | |
| 508 } else { | |
| 509 // ... all other threads block BREAK_SIGNAL | |
| 510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); | |
| 511 } | |
| 512 } | |
| 513 } | |
| 514 | |
| 515 ////////////////////////////////////////////////////////////////////////////// | |
| 516 // detecting pthread library | |
| 517 | |
| 518 void os::Linux::libpthread_init() { | |
| 519 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION | |
| 520 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a | |
| 521 // generic name for earlier versions. | |
| 522 // Define macros here so we can build HotSpot on old systems. | |
| 523 # ifndef _CS_GNU_LIBC_VERSION | |
| 524 # define _CS_GNU_LIBC_VERSION 2 | |
| 525 # endif | |
| 526 # ifndef _CS_GNU_LIBPTHREAD_VERSION | |
| 527 # define _CS_GNU_LIBPTHREAD_VERSION 3 | |
| 528 # endif | |
| 529 | |
| 530 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); | |
| 531 if (n > 0) { | |
| 532 char *str = (char *)malloc(n); | |
| 533 confstr(_CS_GNU_LIBC_VERSION, str, n); | |
| 534 os::Linux::set_glibc_version(str); | |
| 535 } else { | |
| 536 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() | |
| 537 static char _gnu_libc_version[32]; | |
| 538 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), | |
| 539 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); | |
| 540 os::Linux::set_glibc_version(_gnu_libc_version); | |
| 541 } | |
| 542 | |
| 543 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); | |
| 544 if (n > 0) { | |
| 545 char *str = (char *)malloc(n); | |
| 546 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); | |
| 547 | |
| 548 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells | |
| 549 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this | |
| 550 // is the case: | |
| 551 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && | |
| 552 strstr(str, "NPTL")) { | |
| 553 // LinuxThreads has a hard limit on max number of threads. So | |
| 554 // sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. | |
| 555 // On the other hand, NPTL does not have such a limit, sysconf() | |
| 556 // will return -1 and errno is not changed. Check if it is really | |
| 557 // NPTL: | |
| 558 if (sysconf(_SC_THREAD_THREADS_MAX) > 0) { | |
| 559 free(str); | |
| 560 str = "linuxthreads"; | |
| 561 } | |
| 562 } | |
| 563 os::Linux::set_libpthread_version(str); | |
| 564 } else { | |
| 565 // glibc before 2.3.2 only has LinuxThreads. | |
| 566 os::Linux::set_libpthread_version("linuxthreads"); | |
| 567 } | |
| 568 | |
| 569 if (strstr(libpthread_version(), "NPTL")) { | |
| 570 os::Linux::set_is_NPTL(); | |
| 571 } else { | |
| 572 os::Linux::set_is_LinuxThreads(); | |
| 573 } | |
| 574 | |
| 575 // LinuxThreads have two flavors: floating-stack mode, which allows variable | |
| 576 // stack size; and fixed-stack mode. NPTL is always floating-stack. | |
| 577 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { | |
| 578 os::Linux::set_is_floating_stack(); | |
| 579 } | |
| 580 } | |
| 581 | |
| 582 ///////////////////////////////////////////////////////////////////////////// | |
| 583 // thread stack | |
| 584 | |
| 585 // Force Linux kernel to expand current thread stack. If "bottom" is close | |
| 586 // to the stack guard, caller should block all signals. | |
| 587 // | |
| 588 // MAP_GROWSDOWN: | |
| 589 // A special mmap() flag that is used to implement thread stacks. It tells | |
| 590 // kernel that the memory region should extend downwards when needed. This | |
| 591 // allows early versions of LinuxThreads to only mmap the first few pages | |
| 592 // when creating a new thread. Linux kernel will automatically expand thread | |
| 593 // stack as needed (on page faults). | |
| 594 // | |
| 595 // However, because the memory region of a MAP_GROWSDOWN stack can grow on | |
| 596 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN | |
| 597 // region, it's hard to tell if the fault is due to a legitimate stack | |
| 598 // access or because of reading/writing non-exist memory (e.g. buffer | |
| 599 // overrun). As a rule, if the fault happens below current stack pointer, | |
| 600 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the | |
| 601 // application (see Linux kernel fault.c). | |
| 602 // | |
| 603 // This Linux feature can cause SIGSEGV when VM bangs thread stack for | |
| 604 // stack overflow detection. | |
| 605 // | |
| 606 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do | |
| 607 // not use this flag. However, the stack of initial thread is not created | |
| 608 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though | |
| 609 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack | |
| 610 // and then attach the thread to JVM. | |
| 611 // | |
| 612 // To get around the problem and allow stack banging on Linux, we need to | |
| 613 // manually expand thread stack after receiving the SIGSEGV. | |
| 614 // | |
| 615 // There are two ways to expand thread stack to address "bottom", we used | |
| 616 // both of them in JVM before 1.5: | |
| 617 // 1. adjust stack pointer first so that it is below "bottom", and then | |
| 618 // touch "bottom" | |
| 619 // 2. mmap() the page in question | |
| 620 // | |
| 621 // Now alternate signal stack is gone, it's harder to use 2. For instance, | |
| 622 // if current sp is already near the lower end of page 101, and we need to | |
| 623 // call mmap() to map page 100, it is possible that part of the mmap() frame | |
| 624 // will be placed in page 100. When page 100 is mapped, it is zero-filled. | |
| 625 // That will destroy the mmap() frame and cause VM to crash. | |
| 626 // | |
| 627 // The following code works by adjusting sp first, then accessing the "bottom" | |
| 628 // page to force a page fault. Linux kernel will then automatically expand the | |
| 629 // stack mapping. | |
| 630 // | |
| 631 // _expand_stack_to() assumes its frame size is less than page size, which | |
| 632 // should always be true if the function is not inlined. | |
| 633 | |
| 634 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute | |
| 635 #define NOINLINE | |
| 636 #else | |
| 637 #define NOINLINE __attribute__ ((noinline)) | |
| 638 #endif | |
| 639 | |
| 640 static void _expand_stack_to(address bottom) NOINLINE; | |
| 641 | |
| 642 static void _expand_stack_to(address bottom) { | |
| 643 address sp; | |
| 644 size_t size; | |
| 645 volatile char *p; | |
| 646 | |
| 647 // Adjust bottom to point to the largest address within the same page, it | |
| 648 // gives us a one-page buffer if alloca() allocates slightly more memory. | |
| 649 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); | |
| 650 bottom += os::Linux::page_size() - 1; | |
| 651 | |
| 652 // sp might be slightly above current stack pointer; if that's the case, we | |
| 653 // will alloca() a little more space than necessary, which is OK. Don't use | |
| 654 // os::current_stack_pointer(), as its result can be slightly below current | |
| 655 // stack pointer, causing us to not alloca enough to reach "bottom". | |
| 656 sp = (address)&sp; | |
| 657 | |
| 658 if (sp > bottom) { | |
| 659 size = sp - bottom; | |
| 660 p = (volatile char *)alloca(size); | |
| 661 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); | |
| 662 p[0] = '\0'; | |
| 663 } | |
| 664 } | |
| 665 | |
| 666 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { | |
| 667 assert(t!=NULL, "just checking"); | |
| 668 assert(t->osthread()->expanding_stack(), "expand should be set"); | |
| 669 assert(t->stack_base() != NULL, "stack_base was not initialized"); | |
| 670 | |
| 671 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { | |
| 672 sigset_t mask_all, old_sigset; | |
| 673 sigfillset(&mask_all); | |
| 674 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); | |
| 675 _expand_stack_to(addr); | |
| 676 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); | |
| 677 return true; | |
| 678 } | |
| 679 return false; | |
| 680 } | |
| 681 | |
| 682 ////////////////////////////////////////////////////////////////////////////// | |
| 683 // create new thread | |
| 684 | |
| 685 static address highest_vm_reserved_address(); | |
| 686 | |
| 687 // check if it's safe to start a new thread | |
| 688 static bool _thread_safety_check(Thread* thread) { | |
| 689 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { | |
| 690 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) | |
| 691 // Heap is mmap'ed at lower end of memory space. Thread stacks are | |
| 692 // allocated (MAP_FIXED) from high address space. Every thread stack | |
| 693 // occupies a fixed size slot (usually 2Mbytes, but user can change | |
| 694 // it to other values if they rebuild LinuxThreads). | |
| 695 // | |
| 696 // Problem with MAP_FIXED is that mmap() can still succeed even part of | |
| 697 // the memory region has already been mmap'ed. That means if we have too | |
| 698 // many threads and/or very large heap, eventually thread stack will | |
| 699 // collide with heap. | |
| 700 // | |
| 701 // Here we try to prevent heap/stack collision by comparing current | |
| 702 // stack bottom with the highest address that has been mmap'ed by JVM | |
| 703 // plus a safety margin for memory maps created by native code. | |
| 704 // | |
| 705 // This feature can be disabled by setting ThreadSafetyMargin to 0 | |
| 706 // | |
| 707 if (ThreadSafetyMargin > 0) { | |
| 708 address stack_bottom = os::current_stack_base() - os::current_stack_size(); | |
| 709 | |
| 710 // not safe if our stack extends below the safety margin | |
| 711 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); | |
| 712 } else { | |
| 713 return true; | |
| 714 } | |
| 715 } else { | |
| 716 // Floating stack LinuxThreads or NPTL: | |
| 717 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When | |
| 718 // there's not enough space left, pthread_create() will fail. If we come | |
| 719 // here, that means enough space has been reserved for stack. | |
| 720 return true; | |
| 721 } | |
| 722 } | |
| 723 | |
| 724 // Thread start routine for all newly created threads | |
| 725 static void *java_start(Thread *thread) { | |
| 726 // Try to randomize the cache line index of hot stack frames. | |
| 727 // This helps when threads of the same stack traces evict each other's | |
| 728 // cache lines. The threads can be either from the same JVM instance, or | |
| 729 // from different JVM instances. The benefit is especially true for | |
| 730 // processors with hyperthreading technology. | |
| 731 static int counter = 0; | |
| 732 int pid = os::current_process_id(); | |
| 733 alloca(((pid ^ counter++) & 7) * 128); | |
| 734 | |
| 735 ThreadLocalStorage::set_thread(thread); | |
| 736 | |
| 737 OSThread* osthread = thread->osthread(); | |
| 738 Monitor* sync = osthread->startThread_lock(); | |
| 739 | |
| 740 // non floating stack LinuxThreads needs extra check, see above | |
| 741 if (!_thread_safety_check(thread)) { | |
| 742 // notify parent thread | |
| 743 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); | |
| 744 osthread->set_state(ZOMBIE); | |
| 745 sync->notify_all(); | |
| 746 return NULL; | |
| 747 } | |
| 748 | |
| 749 // thread_id is kernel thread id (similar to Solaris LWP id) | |
| 750 osthread->set_thread_id(os::Linux::gettid()); | |
| 751 | |
| 752 if (UseNUMA) { | |
| 753 int lgrp_id = os::numa_get_group_id(); | |
| 754 if (lgrp_id != -1) { | |
| 755 thread->set_lgrp_id(lgrp_id); | |
| 756 } | |
| 757 } | |
| 758 // initialize signal mask for this thread | |
| 759 os::Linux::hotspot_sigmask(thread); | |
| 760 | |
| 761 // initialize floating point control register | |
| 762 os::Linux::init_thread_fpu_state(); | |
| 763 | |
| 764 // handshaking with parent thread | |
| 765 { | |
| 766 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); | |
| 767 | |
| 768 // notify parent thread | |
| 769 osthread->set_state(INITIALIZED); | |
| 770 sync->notify_all(); | |
| 771 | |
| 772 // wait until os::start_thread() | |
| 773 while (osthread->get_state() == INITIALIZED) { | |
| 774 sync->wait(Mutex::_no_safepoint_check_flag); | |
| 775 } | |
| 776 } | |
| 777 | |
| 778 // call one more level start routine | |
| 779 thread->run(); | |
| 780 | |
| 781 return 0; | |
| 782 } | |
| 783 | |
| 784 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { | |
| 785 assert(thread->osthread() == NULL, "caller responsible"); | |
| 786 | |
| 787 // Allocate the OSThread object | |
| 788 OSThread* osthread = new OSThread(NULL, NULL); | |
| 789 if (osthread == NULL) { | |
| 790 return false; | |
| 791 } | |
| 792 | |
| 793 // set the correct thread state | |
| 794 osthread->set_thread_type(thr_type); | |
| 795 | |
| 796 // Initial state is ALLOCATED but not INITIALIZED | |
| 797 osthread->set_state(ALLOCATED); | |
| 798 | |
| 799 thread->set_osthread(osthread); | |
| 800 | |
| 801 // init thread attributes | |
| 802 pthread_attr_t attr; | |
| 803 pthread_attr_init(&attr); | |
| 804 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); | |
| 805 | |
| 806 // stack size | |
| 807 if (os::Linux::supports_variable_stack_size()) { | |
| 808 // calculate stack size if it's not specified by caller | |
| 809 if (stack_size == 0) { | |
| 810 stack_size = os::Linux::default_stack_size(thr_type); | |
| 811 | |
| 812 switch (thr_type) { | |
| 813 case os::java_thread: | |
| 814 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss | |
| 815 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); | |
| 816 break; | |
| 817 case os::compiler_thread: | |
| 818 if (CompilerThreadStackSize > 0) { | |
| 819 stack_size = (size_t)(CompilerThreadStackSize * K); | |
| 820 break; | |
| 821 } // else fall through: | |
| 822 // use VMThreadStackSize if CompilerThreadStackSize is not defined | |
| 823 case os::vm_thread: | |
| 824 case os::pgc_thread: | |
| 825 case os::cgc_thread: | |
| 826 case os::watcher_thread: | |
| 827 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); | |
| 828 break; | |
| 829 } | |
| 830 } | |
| 831 | |
| 832 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); | |
| 833 pthread_attr_setstacksize(&attr, stack_size); | |
| 834 } else { | |
| 835 // let pthread_create() pick the default value. | |
| 836 } | |
| 837 | |
| 838 // glibc guard page | |
| 839 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); | |
| 840 | |
| 841 ThreadState state; | |
| 842 | |
| 843 { | |
| 844 // Serialize thread creation if we are running with fixed stack LinuxThreads | |
| 845 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); | |
| 846 if (lock) { | |
| 847 os::Linux::createThread_lock()->lock_without_safepoint_check(); | |
| 848 } | |
| 849 | |
| 850 pthread_t tid; | |
| 851 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); | |
| 852 | |
| 853 pthread_attr_destroy(&attr); | |
| 854 | |
| 855 if (ret != 0) { | |
| 856 if (PrintMiscellaneous && (Verbose || WizardMode)) { | |
| 857 perror("pthread_create()"); | |
| 858 } | |
| 859 // Need to clean up stuff we've allocated so far | |
| 860 thread->set_osthread(NULL); | |
| 861 delete osthread; | |
| 862 if (lock) os::Linux::createThread_lock()->unlock(); | |
| 863 return false; | |
| 864 } | |
| 865 | |
| 866 // Store pthread info into the OSThread | |
| 867 osthread->set_pthread_id(tid); | |
| 868 | |
| 869 // Wait until child thread is either initialized or aborted | |
| 870 { | |
| 871 Monitor* sync_with_child = osthread->startThread_lock(); | |
| 872 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); | |
| 873 while ((state = osthread->get_state()) == ALLOCATED) { | |
| 874 sync_with_child->wait(Mutex::_no_safepoint_check_flag); | |
| 875 } | |
| 876 } | |
| 877 | |
| 878 if (lock) { | |
| 879 os::Linux::createThread_lock()->unlock(); | |
| 880 } | |
| 881 } | |
| 882 | |
| 883 // Aborted due to thread limit being reached | |
| 884 if (state == ZOMBIE) { | |
| 885 thread->set_osthread(NULL); | |
| 886 delete osthread; | |
| 887 return false; | |
| 888 } | |
| 889 | |
| 890 // The thread is returned suspended (in state INITIALIZED), | |
| 891 // and is started higher up in the call chain | |
| 892 assert(state == INITIALIZED, "race condition"); | |
| 893 return true; | |
| 894 } | |
| 895 | |
| 896 ///////////////////////////////////////////////////////////////////////////// | |
| 897 // attach existing thread | |
| 898 | |
| 899 // bootstrap the main thread | |
| 900 bool os::create_main_thread(JavaThread* thread) { | |
| 901 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); | |
| 902 return create_attached_thread(thread); | |
| 903 } | |
| 904 | |
| 905 bool os::create_attached_thread(JavaThread* thread) { | |
| 906 #ifdef ASSERT | |
| 907 thread->verify_not_published(); | |
| 908 #endif | |
| 909 | |
| 910 // Allocate the OSThread object | |
| 911 OSThread* osthread = new OSThread(NULL, NULL); | |
| 912 | |
| 913 if (osthread == NULL) { | |
| 914 return false; | |
| 915 } | |
| 916 | |
| 917 // Store pthread info into the OSThread | |
| 918 osthread->set_thread_id(os::Linux::gettid()); | |
| 919 osthread->set_pthread_id(::pthread_self()); | |
| 920 | |
| 921 // initialize floating point control register | |
| 922 os::Linux::init_thread_fpu_state(); | |
| 923 | |
| 924 // Initial thread state is RUNNABLE | |
| 925 osthread->set_state(RUNNABLE); | |
| 926 | |
| 927 thread->set_osthread(osthread); | |
| 928 | |
| 929 if (UseNUMA) { | |
| 930 int lgrp_id = os::numa_get_group_id(); | |
| 931 if (lgrp_id != -1) { | |
| 932 thread->set_lgrp_id(lgrp_id); | |
| 933 } | |
| 934 } | |
| 935 | |
| 936 if (os::Linux::is_initial_thread()) { | |
| 937 // If current thread is initial thread, its stack is mapped on demand, | |
| 938 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map | |
| 939 // the entire stack region to avoid SEGV in stack banging. | |
| 940 // It is also useful to get around the heap-stack-gap problem on SuSE | |
| 941 // kernel (see 4821821 for details). We first expand stack to the top | |
| 942 // of yellow zone, then enable stack yellow zone (order is significant, | |
| 943 // enabling yellow zone first will crash JVM on SuSE Linux), so there | |
| 944 // is no gap between the last two virtual memory regions. | |
| 945 | |
| 946 JavaThread *jt = (JavaThread *)thread; | |
| 947 address addr = jt->stack_yellow_zone_base(); | |
| 948 assert(addr != NULL, "initialization problem?"); | |
| 949 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); | |
| 950 | |
| 951 osthread->set_expanding_stack(); | |
| 952 os::Linux::manually_expand_stack(jt, addr); | |
| 953 osthread->clear_expanding_stack(); | |
| 954 } | |
| 955 | |
| 956 // initialize signal mask for this thread | |
| 957 // and save the caller's signal mask | |
| 958 os::Linux::hotspot_sigmask(thread); | |
| 959 | |
| 960 return true; | |
| 961 } | |
| 962 | |
| 963 void os::pd_start_thread(Thread* thread) { | |
| 964 OSThread * osthread = thread->osthread(); | |
| 965 assert(osthread->get_state() != INITIALIZED, "just checking"); | |
| 966 Monitor* sync_with_child = osthread->startThread_lock(); | |
| 967 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); | |
| 968 sync_with_child->notify(); | |
| 969 } | |
| 970 | |
| 971 // Free Linux resources related to the OSThread | |
| 972 void os::free_thread(OSThread* osthread) { | |
| 973 assert(osthread != NULL, "osthread not set"); | |
| 974 | |
| 975 if (Thread::current()->osthread() == osthread) { | |
| 976 // Restore caller's signal mask | |
| 977 sigset_t sigmask = osthread->caller_sigmask(); | |
| 978 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); | |
| 979 } | |
| 980 | |
| 981 delete osthread; | |
| 982 } | |
| 983 | |
| 984 ////////////////////////////////////////////////////////////////////////////// | |
| 985 // thread local storage | |
| 986 | |
| 987 int os::allocate_thread_local_storage() { | |
| 988 pthread_key_t key; | |
| 989 int rslt = pthread_key_create(&key, NULL); | |
| 990 assert(rslt == 0, "cannot allocate thread local storage"); | |
| 991 return (int)key; | |
| 992 } | |
| 993 | |
| 994 // Note: This is currently not used by VM, as we don't destroy TLS key | |
| 995 // on VM exit. | |
| 996 void os::free_thread_local_storage(int index) { | |
| 997 int rslt = pthread_key_delete((pthread_key_t)index); | |
| 998 assert(rslt == 0, "invalid index"); | |
| 999 } | |
| 1000 | |
| 1001 void os::thread_local_storage_at_put(int index, void* value) { | |
| 1002 int rslt = pthread_setspecific((pthread_key_t)index, value); | |
| 1003 assert(rslt == 0, "pthread_setspecific failed"); | |
| 1004 } | |
| 1005 | |
| 1006 extern "C" Thread* get_thread() { | |
| 1007 return ThreadLocalStorage::thread(); | |
| 1008 } | |
| 1009 | |
| 1010 ////////////////////////////////////////////////////////////////////////////// | |
| 1011 // initial thread | |
| 1012 | |
| 1013 // Check if current thread is the initial thread, similar to Solaris thr_main. | |
| 1014 bool os::Linux::is_initial_thread(void) { | |
| 1015 char dummy; | |
| 1016 // If called before init complete, thread stack bottom will be null. | |
| 1017 // Can be called if fatal error occurs before initialization. | |
| 1018 if (initial_thread_stack_bottom() == NULL) return false; | |
| 1019 assert(initial_thread_stack_bottom() != NULL && | |
| 1020 initial_thread_stack_size() != 0, | |
| 1021 "os::init did not locate initial thread's stack region"); | |
| 1022 if ((address)&dummy >= initial_thread_stack_bottom() && | |
| 1023 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) | |
| 1024 return true; | |
| 1025 else return false; | |
| 1026 } | |
| 1027 | |
| 1028 // Find the virtual memory area that contains addr | |
| 1029 static bool find_vma(address addr, address* vma_low, address* vma_high) { | |
| 1030 FILE *fp = fopen("/proc/self/maps", "r"); | |
| 1031 if (fp) { | |
| 1032 address low, high; | |
| 1033 while (!feof(fp)) { | |
| 1034 if (fscanf(fp, "%p-%p", &low, &high) == 2) { | |
| 1035 if (low <= addr && addr < high) { | |
| 1036 if (vma_low) *vma_low = low; | |
| 1037 if (vma_high) *vma_high = high; | |
| 1038 fclose (fp); | |
| 1039 return true; | |
| 1040 } | |
| 1041 } | |
| 1042 for (;;) { | |
| 1043 int ch = fgetc(fp); | |
| 1044 if (ch == EOF || ch == (int)'\n') break; | |
| 1045 } | |
| 1046 } | |
| 1047 fclose(fp); | |
| 1048 } | |
| 1049 return false; | |
| 1050 } | |
| 1051 | |
| 1052 // Locate initial thread stack. This special handling of initial thread stack | |
| 1053 // is needed because pthread_getattr_np() on most (all?) Linux distros returns | |
| 1054 // bogus value for initial thread. | |
| 1055 void os::Linux::capture_initial_stack(size_t max_size) { | |
| 1056 // stack size is the easy part, get it from RLIMIT_STACK | |
| 1057 size_t stack_size; | |
| 1058 struct rlimit rlim; | |
| 1059 getrlimit(RLIMIT_STACK, &rlim); | |
| 1060 stack_size = rlim.rlim_cur; | |
| 1061 | |
| 1062 // 6308388: a bug in ld.so will relocate its own .data section to the | |
| 1063 // lower end of primordial stack; reduce ulimit -s value a little bit | |
| 1064 // so we won't install guard page on ld.so's data section. | |
| 1065 stack_size -= 2 * page_size(); | |
| 1066 | |
| 1067 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat | |
| 1068 // 7.1, in both cases we will get 2G in return value. | |
| 1069 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, | |
| 1070 // SuSE 7.2, Debian) can not handle alternate signal stack correctly | |
| 1071 // for initial thread if its stack size exceeds 6M. Cap it at 2M, | |
| 1072 // in case other parts in glibc still assumes 2M max stack size. | |
| 1073 // FIXME: alt signal stack is gone, maybe we can relax this constraint? | |
| 1074 #ifndef IA64 | |
| 1075 if (stack_size > 2 * K * K) stack_size = 2 * K * K; | |
| 1076 #else | |
| 1077 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small | |
| 1078 if (stack_size > 4 * K * K) stack_size = 4 * K * K; | |
| 1079 #endif | |
| 1080 | |
| 1081 // Try to figure out where the stack base (top) is. This is harder. | |
| 1082 // | |
| 1083 // When an application is started, glibc saves the initial stack pointer in | |
| 1084 // a global variable "__libc_stack_end", which is then used by system | |
| 1085 // libraries. __libc_stack_end should be pretty close to stack top. The | |
| 1086 // variable is available since the very early days. However, because it is | |
| 1087 // a private interface, it could disappear in the future. | |
| 1088 // | |
| 1089 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar | |
| 1090 // to __libc_stack_end, it is very close to stack top, but isn't the real | |
| 1091 // stack top. Note that /proc may not exist if VM is running as a chroot | |
| 1092 // program, so reading /proc/<pid>/stat could fail. Also the contents of | |
| 1093 // /proc/<pid>/stat could change in the future (though unlikely). | |
| 1094 // | |
| 1095 // We try __libc_stack_end first. If that doesn't work, look for | |
| 1096 // /proc/<pid>/stat. If neither of them works, we use current stack pointer | |
| 1097 // as a hint, which should work well in most cases. | |
| 1098 | |
| 1099 uintptr_t stack_start; | |
| 1100 | |
| 1101 // try __libc_stack_end first | |
| 1102 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); | |
| 1103 if (p && *p) { | |
| 1104 stack_start = *p; | |
| 1105 } else { | |
| 1106 // see if we can get the start_stack field from /proc/self/stat | |
| 1107 FILE *fp; | |
| 1108 int pid; | |
| 1109 char state; | |
| 1110 int ppid; | |
| 1111 int pgrp; | |
| 1112 int session; | |
| 1113 int nr; | |
| 1114 int tpgrp; | |
| 1115 unsigned long flags; | |
| 1116 unsigned long minflt; | |
| 1117 unsigned long cminflt; | |
| 1118 unsigned long majflt; | |
| 1119 unsigned long cmajflt; | |
| 1120 unsigned long utime; | |
| 1121 unsigned long stime; | |
| 1122 long cutime; | |
| 1123 long cstime; | |
| 1124 long prio; | |
| 1125 long nice; | |
| 1126 long junk; | |
| 1127 long it_real; | |
| 1128 uintptr_t start; | |
| 1129 uintptr_t vsize; | |
| 1130 uintptr_t rss; | |
| 1131 unsigned long rsslim; | |
| 1132 uintptr_t scodes; | |
| 1133 uintptr_t ecode; | |
| 1134 int i; | |
| 1135 | |
| 1136 // Figure what the primordial thread stack base is. Code is inspired | |
| 1137 // by email from Hans Boehm. /proc/self/stat begins with current pid, | |
| 1138 // followed by command name surrounded by parentheses, state, etc. | |
| 1139 char stat[2048]; | |
| 1140 int statlen; | |
| 1141 | |
| 1142 fp = fopen("/proc/self/stat", "r"); | |
| 1143 if (fp) { | |
| 1144 statlen = fread(stat, 1, 2047, fp); | |
| 1145 stat[statlen] = '\0'; | |
| 1146 fclose(fp); | |
| 1147 | |
| 1148 // Skip pid and the command string. Note that we could be dealing with | |
| 1149 // weird command names, e.g. user could decide to rename java launcher | |
| 1150 // to "java 1.4.2 :)", then the stat file would look like | |
| 1151 // 1234 (java 1.4.2 :)) R ... ... | |
| 1152 // We don't really need to know the command string, just find the last | |
| 1153 // occurrence of ")" and then start parsing from there. See bug 4726580. | |
| 1154 char * s = strrchr(stat, ')'); | |
| 1155 | |
| 1156 i = 0; | |
| 1157 if (s) { | |
| 1158 // Skip blank chars | |
| 1159 do s++; while (isspace(*s)); | |
| 1160 | |
| 1161 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ | |
| 1162 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */ | |
| 1163 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", | |
| 1164 &state, /* 3 %c */ | |
| 1165 &ppid, /* 4 %d */ | |
| 1166 &pgrp, /* 5 %d */ | |
| 1167 &session, /* 6 %d */ | |
| 1168 &nr, /* 7 %d */ | |
| 1169 &tpgrp, /* 8 %d */ | |
| 1170 &flags, /* 9 %lu */ | |
| 1171 &minflt, /* 10 %lu */ | |
| 1172 &cminflt, /* 11 %lu */ | |
| 1173 &majflt, /* 12 %lu */ | |
| 1174 &cmajflt, /* 13 %lu */ | |
| 1175 &utime, /* 14 %lu */ | |
| 1176 &stime, /* 15 %lu */ | |
| 1177 &cutime, /* 16 %ld */ | |
| 1178 &cstime, /* 17 %ld */ | |
| 1179 &prio, /* 18 %ld */ | |
| 1180 &nice, /* 19 %ld */ | |
| 1181 &junk, /* 20 %ld */ | |
| 1182 &it_real, /* 21 %ld */ | |
| 1183 &start, /* 22 %lu */ | |
| 1184 &vsize, /* 23 %lu */ | |
| 1185 &rss, /* 24 %ld */ | |
| 1186 &rsslim, /* 25 %lu */ | |
| 1187 &scodes, /* 26 %lu */ | |
| 1188 &ecode, /* 27 %lu */ | |
| 1189 &stack_start); /* 28 %lu */ | |
| 1190 } | |
| 1191 | |
| 1192 if (i != 28 - 2) { | |
| 1193 assert(false, "Bad conversion from /proc/self/stat"); | |
| 1194 // product mode - assume we are the initial thread, good luck in the | |
| 1195 // embedded case. | |
| 1196 warning("Can't detect initial thread stack location - bad conversion"); | |
| 1197 stack_start = (uintptr_t) &rlim; | |
| 1198 } | |
| 1199 } else { | |
| 1200 // For some reason we can't open /proc/self/stat (for example, running on | |
| 1201 // FreeBSD with a Linux emulator, or inside chroot), this should work for | |
| 1202 // most cases, so don't abort: | |
| 1203 warning("Can't detect initial thread stack location - no /proc/self/stat"); | |
| 1204 stack_start = (uintptr_t) &rlim; | |
| 1205 } | |
| 1206 } | |
| 1207 | |
| 1208 // Now we have a pointer (stack_start) very close to the stack top, the | |
| 1209 // next thing to do is to figure out the exact location of stack top. We | |
| 1210 // can find out the virtual memory area that contains stack_start by | |
| 1211 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, | |
| 1212 // and its upper limit is the real stack top. (again, this would fail if | |
| 1213 // running inside chroot, because /proc may not exist.) | |
| 1214 | |
| 1215 uintptr_t stack_top; | |
| 1216 address low, high; | |
| 1217 if (find_vma((address)stack_start, &low, &high)) { | |
| 1218 // success, "high" is the true stack top. (ignore "low", because initial | |
| 1219 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) | |
| 1220 stack_top = (uintptr_t)high; | |
| 1221 } else { | |
| 1222 // failed, likely because /proc/self/maps does not exist | |
| 1223 warning("Can't detect initial thread stack location - find_vma failed"); | |
| 1224 // best effort: stack_start is normally within a few pages below the real | |
| 1225 // stack top, use it as stack top, and reduce stack size so we won't put | |
| 1226 // guard page outside stack. | |
| 1227 stack_top = stack_start; | |
| 1228 stack_size -= 16 * page_size(); | |
| 1229 } | |
| 1230 | |
| 1231 // stack_top could be partially down the page so align it | |
| 1232 stack_top = align_size_up(stack_top, page_size()); | |
| 1233 | |
| 1234 if (max_size && stack_size > max_size) { | |
| 1235 _initial_thread_stack_size = max_size; | |
| 1236 } else { | |
| 1237 _initial_thread_stack_size = stack_size; | |
| 1238 } | |
| 1239 | |
| 1240 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); | |
| 1241 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; | |
| 1242 } | |
| 1243 | |
| 1244 //////////////////////////////////////////////////////////////////////////////// | |
| 1245 // time support | |
| 1246 | |
| 1247 // Time since start-up in seconds to a fine granularity. | |
| 1248 // Used by VMSelfDestructTimer and the MemProfiler. | |
| 1249 double os::elapsedTime() { | |
| 1250 | |
| 1251 return (double)(os::elapsed_counter()) * 0.000001; | |
| 1252 } | |
| 1253 | |
| 1254 jlong os::elapsed_counter() { | |
| 1255 timeval time; | |
| 1256 int status = gettimeofday(&time, NULL); | |
| 1257 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; | |
| 1258 } | |
| 1259 | |
| 1260 jlong os::elapsed_frequency() { | |
| 1261 return (1000 * 1000); | |
| 1262 } | |
| 1263 | |
| 61 | 1264 jlong os::javaTimeMillis() { |
| 0 | 1265 timeval time; |
| 1266 int status = gettimeofday(&time, NULL); | |
| 1267 assert(status != -1, "linux error"); | |
| 1268 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); | |
| 1269 } | |
| 1270 | |
| 1271 #ifndef CLOCK_MONOTONIC | |
| 1272 #define CLOCK_MONOTONIC (1) | |
| 1273 #endif | |
| 1274 | |
| 1275 void os::Linux::clock_init() { | |
| 1276 // we do dlopen's in this particular order due to bug in linux | |
| 1277 // dynamical loader (see 6348968) leading to crash on exit | |
| 1278 void* handle = dlopen("librt.so.1", RTLD_LAZY); | |
| 1279 if (handle == NULL) { | |
| 1280 handle = dlopen("librt.so", RTLD_LAZY); | |
| 1281 } | |
| 1282 | |
| 1283 if (handle) { | |
| 1284 int (*clock_getres_func)(clockid_t, struct timespec*) = | |
| 1285 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); | |
| 1286 int (*clock_gettime_func)(clockid_t, struct timespec*) = | |
| 1287 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); | |
| 1288 if (clock_getres_func && clock_gettime_func) { | |
| 1289 // See if monotonic clock is supported by the kernel. Note that some | |
| 1290 // early implementations simply return kernel jiffies (updated every | |
| 1291 // 1/100 or 1/1000 second). It would be bad to use such a low res clock | |
| 1292 // for nano time (though the monotonic property is still nice to have). | |
| 1293 // It's fixed in newer kernels, however clock_getres() still returns | |
| 1294 // 1/HZ. We check if clock_getres() works, but will ignore its reported | |
| 1295 // resolution for now. Hopefully as people move to new kernels, this | |
| 1296 // won't be a problem. | |
| 1297 struct timespec res; | |
| 1298 struct timespec tp; | |
| 1299 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && | |
| 1300 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { | |
| 1301 // yes, monotonic clock is supported | |
| 1302 _clock_gettime = clock_gettime_func; | |
| 1303 } else { | |
| 1304 // close librt if there is no monotonic clock | |
| 1305 dlclose(handle); | |
| 1306 } | |
| 1307 } | |
| 1308 } | |
| 1309 } | |
| 1310 | |
| 1311 #ifndef SYS_clock_getres | |
| 1312 | |
| 1313 #if defined(IA32) || defined(AMD64) | |
| 1314 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) | |
| 1315 #else | |
| 1316 #error Value of SYS_clock_getres not known on this platform | |
| 1317 #endif | |
| 1318 | |
| 1319 #endif | |
| 1320 | |
| 1321 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) | |
| 1322 | |
| 1323 void os::Linux::fast_thread_clock_init() { | |
| 1324 if (!UseLinuxPosixThreadCPUClocks) { | |
| 1325 return; | |
| 1326 } | |
| 1327 clockid_t clockid; | |
| 1328 struct timespec tp; | |
| 1329 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = | |
| 1330 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); | |
| 1331 | |
| 1332 // Switch to using fast clocks for thread cpu time if | |
| 1333 // the sys_clock_getres() returns 0 error code. | |
| 1334 // Note, that some kernels may support the current thread | |
| 1335 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks | |
| 1336 // returned by the pthread_getcpuclockid(). | |
| 1337 // If the fast Posix clocks are supported then the sys_clock_getres() | |
| 1338 // must return at least tp.tv_sec == 0 which means a resolution | |
| 1339 // better than 1 sec. This is extra check for reliability. | |
| 1340 | |
| 1341 if(pthread_getcpuclockid_func && | |
| 1342 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && | |
| 1343 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { | |
| 1344 | |
| 1345 _supports_fast_thread_cpu_time = true; | |
| 1346 _pthread_getcpuclockid = pthread_getcpuclockid_func; | |
| 1347 } | |
| 1348 } | |
| 1349 | |
| 1350 jlong os::javaTimeNanos() { | |
| 1351 if (Linux::supports_monotonic_clock()) { | |
| 1352 struct timespec tp; | |
| 1353 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); | |
| 1354 assert(status == 0, "gettime error"); | |
| 1355 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); | |
| 1356 return result; | |
| 1357 } else { | |
| 1358 timeval time; | |
| 1359 int status = gettimeofday(&time, NULL); | |
| 1360 assert(status != -1, "linux error"); | |
| 1361 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); | |
| 1362 return 1000 * usecs; | |
| 1363 } | |
| 1364 } | |
| 1365 | |
| 1366 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { | |
| 1367 if (Linux::supports_monotonic_clock()) { | |
| 1368 info_ptr->max_value = ALL_64_BITS; | |
| 1369 | |
| 1370 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past | |
| 1371 info_ptr->may_skip_backward = false; // not subject to resetting or drifting | |
| 1372 info_ptr->may_skip_forward = false; // not subject to resetting or drifting | |
| 1373 } else { | |
| 1374 // gettimeofday - based on time in seconds since the Epoch thus does not wrap | |
| 1375 info_ptr->max_value = ALL_64_BITS; | |
| 1376 | |
| 1377 // gettimeofday is a real time clock so it skips | |
| 1378 info_ptr->may_skip_backward = true; | |
| 1379 info_ptr->may_skip_forward = true; | |
| 1380 } | |
| 1381 | |
| 1382 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time | |
| 1383 } | |
| 1384 | |
| 1385 // Return the real, user, and system times in seconds from an | |
| 1386 // arbitrary fixed point in the past. | |
| 1387 bool os::getTimesSecs(double* process_real_time, | |
| 1388 double* process_user_time, | |
| 1389 double* process_system_time) { | |
| 1390 struct tms ticks; | |
| 1391 clock_t real_ticks = times(&ticks); | |
| 1392 | |
| 1393 if (real_ticks == (clock_t) (-1)) { | |
| 1394 return false; | |
| 1395 } else { | |
| 1396 double ticks_per_second = (double) clock_tics_per_sec; | |
| 1397 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; | |
| 1398 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; | |
| 1399 *process_real_time = ((double) real_ticks) / ticks_per_second; | |
| 1400 | |
| 1401 return true; | |
| 1402 } | |
| 1403 } | |
| 1404 | |
| 1405 | |
| 1406 char * os::local_time_string(char *buf, size_t buflen) { | |
| 1407 struct tm t; | |
| 1408 time_t long_time; | |
| 1409 time(&long_time); | |
| 1410 localtime_r(&long_time, &t); | |
| 1411 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", | |
| 1412 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, | |
| 1413 t.tm_hour, t.tm_min, t.tm_sec); | |
| 1414 return buf; | |
| 1415 } | |
| 1416 | |
| 1417 //////////////////////////////////////////////////////////////////////////////// | |
| 1418 // runtime exit support | |
| 1419 | |
| 1420 // Note: os::shutdown() might be called very early during initialization, or | |
| 1421 // called from signal handler. Before adding something to os::shutdown(), make | |
| 1422 // sure it is async-safe and can handle partially initialized VM. | |
| 1423 void os::shutdown() { | |
| 1424 | |
| 1425 // allow PerfMemory to attempt cleanup of any persistent resources | |
| 1426 perfMemory_exit(); | |
| 1427 | |
| 1428 // needs to remove object in file system | |
| 1429 AttachListener::abort(); | |
| 1430 | |
| 1431 // flush buffered output, finish log files | |
| 1432 ostream_abort(); | |
| 1433 | |
| 1434 // Check for abort hook | |
| 1435 abort_hook_t abort_hook = Arguments::abort_hook(); | |
| 1436 if (abort_hook != NULL) { | |
| 1437 abort_hook(); | |
| 1438 } | |
| 1439 | |
| 1440 } | |
| 1441 | |
| 1442 // Note: os::abort() might be called very early during initialization, or | |
| 1443 // called from signal handler. Before adding something to os::abort(), make | |
| 1444 // sure it is async-safe and can handle partially initialized VM. | |
| 1445 void os::abort(bool dump_core) { | |
| 1446 os::shutdown(); | |
| 1447 if (dump_core) { | |
| 1448 #ifndef PRODUCT | |
| 1449 fdStream out(defaultStream::output_fd()); | |
| 1450 out.print_raw("Current thread is "); | |
| 1451 char buf[16]; | |
| 1452 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); | |
| 1453 out.print_raw_cr(buf); | |
| 1454 out.print_raw_cr("Dumping core ..."); | |
| 1455 #endif | |
| 1456 ::abort(); // dump core | |
| 1457 } | |
| 1458 | |
| 1459 ::exit(1); | |
| 1460 } | |
| 1461 | |
| 1462 // Die immediately, no exit hook, no abort hook, no cleanup. | |
| 1463 void os::die() { | |
| 1464 // _exit() on LinuxThreads only kills current thread | |
| 1465 ::abort(); | |
| 1466 } | |
| 1467 | |
| 1468 // unused on linux for now. | |
| 1469 void os::set_error_file(const char *logfile) {} | |
| 1470 | |
| 1471 intx os::current_thread_id() { return (intx)pthread_self(); } | |
| 1472 int os::current_process_id() { | |
| 1473 | |
| 1474 // Under the old linux thread library, linux gives each thread | |
| 1475 // its own process id. Because of this each thread will return | |
| 1476 // a different pid if this method were to return the result | |
| 1477 // of getpid(2). Linux provides no api that returns the pid | |
| 1478 // of the launcher thread for the vm. This implementation | |
| 1479 // returns a unique pid, the pid of the launcher thread | |
| 1480 // that starts the vm 'process'. | |
| 1481 | |
| 1482 // Under the NPTL, getpid() returns the same pid as the | |
| 1483 // launcher thread rather than a unique pid per thread. | |
| 1484 // Use gettid() if you want the old pre NPTL behaviour. | |
| 1485 | |
| 1486 // if you are looking for the result of a call to getpid() that | |
| 1487 // returns a unique pid for the calling thread, then look at the | |
| 1488 // OSThread::thread_id() method in osThread_linux.hpp file | |
| 1489 | |
| 1490 return (int)(_initial_pid ? _initial_pid : getpid()); | |
| 1491 } | |
| 1492 | |
| 1493 // DLL functions | |
| 1494 | |
| 1495 const char* os::dll_file_extension() { return ".so"; } | |
| 1496 | |
| 1497 const char* os::get_temp_directory() { return "/tmp/"; } | |
| 1498 | |
| 1499 const char* os::get_current_directory(char *buf, int buflen) { | |
| 1500 return getcwd(buf, buflen); | |
| 1501 } | |
| 1502 | |
| 1503 // check if addr is inside libjvm[_g].so | |
| 1504 bool os::address_is_in_vm(address addr) { | |
| 1505 static address libjvm_base_addr; | |
| 1506 Dl_info dlinfo; | |
| 1507 | |
| 1508 if (libjvm_base_addr == NULL) { | |
| 1509 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); | |
| 1510 libjvm_base_addr = (address)dlinfo.dli_fbase; | |
| 1511 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); | |
| 1512 } | |
| 1513 | |
| 1514 if (dladdr((void *)addr, &dlinfo)) { | |
| 1515 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; | |
| 1516 } | |
| 1517 | |
| 1518 return false; | |
| 1519 } | |
| 1520 | |
| 1521 bool os::dll_address_to_function_name(address addr, char *buf, | |
| 1522 int buflen, int *offset) { | |
| 1523 Dl_info dlinfo; | |
| 1524 | |
| 1525 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { | |
| 1526 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); | |
| 1527 if (offset) *offset = addr - (address)dlinfo.dli_saddr; | |
| 1528 return true; | |
| 1529 } else { | |
| 1530 if (buf) buf[0] = '\0'; | |
| 1531 if (offset) *offset = -1; | |
| 1532 return false; | |
| 1533 } | |
| 1534 } | |
| 1535 | |
| 1536 struct _address_to_library_name { | |
| 1537 address addr; // input : memory address | |
| 1538 size_t buflen; // size of fname | |
| 1539 char* fname; // output: library name | |
| 1540 address base; // library base addr | |
| 1541 }; | |
| 1542 | |
| 1543 static int address_to_library_name_callback(struct dl_phdr_info *info, | |
| 1544 size_t size, void *data) { | |
| 1545 int i; | |
| 1546 bool found = false; | |
| 1547 address libbase = NULL; | |
| 1548 struct _address_to_library_name * d = (struct _address_to_library_name *)data; | |
| 1549 | |
| 1550 // iterate through all loadable segments | |
| 1551 for (i = 0; i < info->dlpi_phnum; i++) { | |
| 1552 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); | |
| 1553 if (info->dlpi_phdr[i].p_type == PT_LOAD) { | |
| 1554 // base address of a library is the lowest address of its loaded | |
| 1555 // segments. | |
| 1556 if (libbase == NULL || libbase > segbase) { | |
| 1557 libbase = segbase; | |
| 1558 } | |
| 1559 // see if 'addr' is within current segment | |
| 1560 if (segbase <= d->addr && | |
| 1561 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { | |
| 1562 found = true; | |
| 1563 } | |
| 1564 } | |
| 1565 } | |
| 1566 | |
| 1567 // dlpi_name is NULL or empty if the ELF file is executable, return 0 | |
| 1568 // so dll_address_to_library_name() can fall through to use dladdr() which | |
| 1569 // can figure out executable name from argv[0]. | |
| 1570 if (found && info->dlpi_name && info->dlpi_name[0]) { | |
| 1571 d->base = libbase; | |
| 1572 if (d->fname) { | |
| 1573 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); | |
| 1574 } | |
| 1575 return 1; | |
| 1576 } | |
| 1577 return 0; | |
| 1578 } | |
| 1579 | |
| 1580 bool os::dll_address_to_library_name(address addr, char* buf, | |
| 1581 int buflen, int* offset) { | |
| 1582 Dl_info dlinfo; | |
| 1583 struct _address_to_library_name data; | |
| 1584 | |
| 1585 // There is a bug in old glibc dladdr() implementation that it could resolve | |
| 1586 // to wrong library name if the .so file has a base address != NULL. Here | |
| 1587 // we iterate through the program headers of all loaded libraries to find | |
| 1588 // out which library 'addr' really belongs to. This workaround can be | |
| 1589 // removed once the minimum requirement for glibc is moved to 2.3.x. | |
| 1590 data.addr = addr; | |
| 1591 data.fname = buf; | |
| 1592 data.buflen = buflen; | |
| 1593 data.base = NULL; | |
| 1594 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); | |
| 1595 | |
| 1596 if (rslt) { | |
| 1597 // buf already contains library name | |
| 1598 if (offset) *offset = addr - data.base; | |
| 1599 return true; | |
| 1600 } else if (dladdr((void*)addr, &dlinfo)){ | |
| 1601 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); | |
| 1602 if (offset) *offset = addr - (address)dlinfo.dli_fbase; | |
| 1603 return true; | |
| 1604 } else { | |
| 1605 if (buf) buf[0] = '\0'; | |
| 1606 if (offset) *offset = -1; | |
| 1607 return false; | |
| 1608 } | |
| 1609 } | |
| 1610 | |
| 1611 // Loads .dll/.so and | |
| 1612 // in case of error it checks if .dll/.so was built for the | |
| 1613 // same architecture as Hotspot is running on | |
| 1614 | |
| 1615 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) | |
| 1616 { | |
| 1617 void * result= ::dlopen(filename, RTLD_LAZY); | |
| 1618 if (result != NULL) { | |
| 1619 // Successful loading | |
| 1620 return result; | |
| 1621 } | |
| 1622 | |
| 1623 Elf32_Ehdr elf_head; | |
| 1624 | |
| 1625 // Read system error message into ebuf | |
| 1626 // It may or may not be overwritten below | |
| 1627 ::strncpy(ebuf, ::dlerror(), ebuflen-1); | |
| 1628 ebuf[ebuflen-1]='\0'; | |
| 1629 int diag_msg_max_length=ebuflen-strlen(ebuf); | |
| 1630 char* diag_msg_buf=ebuf+strlen(ebuf); | |
| 1631 | |
| 1632 if (diag_msg_max_length==0) { | |
| 1633 // No more space in ebuf for additional diagnostics message | |
| 1634 return NULL; | |
| 1635 } | |
| 1636 | |
| 1637 | |
| 1638 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); | |
| 1639 | |
| 1640 if (file_descriptor < 0) { | |
| 1641 // Can't open library, report dlerror() message | |
| 1642 return NULL; | |
| 1643 } | |
| 1644 | |
| 1645 bool failed_to_read_elf_head= | |
| 1646 (sizeof(elf_head)!= | |
| 1647 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; | |
| 1648 | |
| 1649 ::close(file_descriptor); | |
| 1650 if (failed_to_read_elf_head) { | |
| 1651 // file i/o error - report dlerror() msg | |
| 1652 return NULL; | |
| 1653 } | |
| 1654 | |
| 1655 typedef struct { | |
| 1656 Elf32_Half code; // Actual value as defined in elf.h | |
| 1657 Elf32_Half compat_class; // Compatibility of archs at VM's sense | |
| 1658 char elf_class; // 32 or 64 bit | |
| 1659 char endianess; // MSB or LSB | |
| 1660 char* name; // String representation | |
| 1661 } arch_t; | |
| 1662 | |
| 1663 #ifndef EM_486 | |
| 1664 #define EM_486 6 /* Intel 80486 */ | |
| 1665 #endif | |
| 1666 | |
| 1667 static const arch_t arch_array[]={ | |
| 1668 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, | |
| 1669 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, | |
| 1670 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, | |
| 1671 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, | |
| 1672 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, | |
| 1673 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, | |
| 1674 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, | |
| 1675 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, | |
| 1676 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} | |
| 1677 }; | |
| 1678 | |
| 1679 #if (defined IA32) | |
| 1680 static Elf32_Half running_arch_code=EM_386; | |
| 1681 #elif (defined AMD64) | |
| 1682 static Elf32_Half running_arch_code=EM_X86_64; | |
| 1683 #elif (defined IA64) | |
| 1684 static Elf32_Half running_arch_code=EM_IA_64; | |
| 1685 #elif (defined __sparc) && (defined _LP64) | |
| 1686 static Elf32_Half running_arch_code=EM_SPARCV9; | |
| 1687 #elif (defined __sparc) && (!defined _LP64) | |
| 1688 static Elf32_Half running_arch_code=EM_SPARC; | |
| 1689 #elif (defined __powerpc64__) | |
| 1690 static Elf32_Half running_arch_code=EM_PPC64; | |
| 1691 #elif (defined __powerpc__) | |
| 1692 static Elf32_Half running_arch_code=EM_PPC; | |
| 1693 #else | |
| 1694 #error Method os::dll_load requires that one of following is defined:\ | |
| 1695 IA32, AMD64, IA64, __sparc, __powerpc__ | |
| 1696 #endif | |
| 1697 | |
| 1698 // Identify compatability class for VM's architecture and library's architecture | |
| 1699 // Obtain string descriptions for architectures | |
| 1700 | |
| 1701 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; | |
| 1702 int running_arch_index=-1; | |
| 1703 | |
| 1704 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { | |
| 1705 if (running_arch_code == arch_array[i].code) { | |
| 1706 running_arch_index = i; | |
| 1707 } | |
| 1708 if (lib_arch.code == arch_array[i].code) { | |
| 1709 lib_arch.compat_class = arch_array[i].compat_class; | |
| 1710 lib_arch.name = arch_array[i].name; | |
| 1711 } | |
| 1712 } | |
| 1713 | |
| 1714 assert(running_arch_index != -1, | |
| 1715 "Didn't find running architecture code (running_arch_code) in arch_array"); | |
| 1716 if (running_arch_index == -1) { | |
| 1717 // Even though running architecture detection failed | |
| 1718 // we may still continue with reporting dlerror() message | |
| 1719 return NULL; | |
| 1720 } | |
| 1721 | |
| 1722 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { | |
| 1723 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); | |
| 1724 return NULL; | |
| 1725 } | |
| 1726 | |
| 1727 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { | |
| 1728 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); | |
| 1729 return NULL; | |
| 1730 } | |
| 1731 | |
| 1732 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { | |
| 1733 if ( lib_arch.name!=NULL ) { | |
| 1734 ::snprintf(diag_msg_buf, diag_msg_max_length-1, | |
| 1735 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", | |
| 1736 lib_arch.name, arch_array[running_arch_index].name); | |
| 1737 } else { | |
| 1738 ::snprintf(diag_msg_buf, diag_msg_max_length-1, | |
| 1739 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", | |
| 1740 lib_arch.code, | |
| 1741 arch_array[running_arch_index].name); | |
| 1742 } | |
| 1743 } | |
| 1744 | |
| 1745 return NULL; | |
| 1746 } | |
| 1747 | |
| 1748 | |
| 1749 | |
| 1750 | |
| 1751 bool _print_ascii_file(const char* filename, outputStream* st) { | |
| 1752 int fd = open(filename, O_RDONLY); | |
| 1753 if (fd == -1) { | |
| 1754 return false; | |
| 1755 } | |
| 1756 | |
| 1757 char buf[32]; | |
| 1758 int bytes; | |
| 1759 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { | |
| 1760 st->print_raw(buf, bytes); | |
| 1761 } | |
| 1762 | |
| 1763 close(fd); | |
| 1764 | |
| 1765 return true; | |
| 1766 } | |
| 1767 | |
| 1768 void os::print_dll_info(outputStream *st) { | |
| 1769 st->print_cr("Dynamic libraries:"); | |
| 1770 | |
| 1771 char fname[32]; | |
| 1772 pid_t pid = os::Linux::gettid(); | |
| 1773 | |
| 1774 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); | |
| 1775 | |
| 1776 if (!_print_ascii_file(fname, st)) { | |
| 1777 st->print("Can not get library information for pid = %d\n", pid); | |
| 1778 } | |
| 1779 } | |
| 1780 | |
| 1781 | |
| 1782 void os::print_os_info(outputStream* st) { | |
| 1783 st->print("OS:"); | |
| 1784 | |
| 1785 // Try to identify popular distros. | |
| 1786 // Most Linux distributions have /etc/XXX-release file, which contains | |
| 1787 // the OS version string. Some have more than one /etc/XXX-release file | |
| 1788 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), | |
| 1789 // so the order is important. | |
| 1790 if (!_print_ascii_file("/etc/mandrake-release", st) && | |
| 1791 !_print_ascii_file("/etc/sun-release", st) && | |
| 1792 !_print_ascii_file("/etc/redhat-release", st) && | |
| 1793 !_print_ascii_file("/etc/SuSE-release", st) && | |
| 1794 !_print_ascii_file("/etc/turbolinux-release", st) && | |
| 1795 !_print_ascii_file("/etc/gentoo-release", st) && | |
| 1796 !_print_ascii_file("/etc/debian_version", st)) { | |
| 1797 st->print("Linux"); | |
| 1798 } | |
| 1799 st->cr(); | |
| 1800 | |
| 1801 // kernel | |
| 1802 st->print("uname:"); | |
| 1803 struct utsname name; | |
| 1804 uname(&name); | |
| 1805 st->print(name.sysname); st->print(" "); | |
| 1806 st->print(name.release); st->print(" "); | |
| 1807 st->print(name.version); st->print(" "); | |
| 1808 st->print(name.machine); | |
| 1809 st->cr(); | |
| 1810 | |
| 1811 // Print warning if unsafe chroot environment detected | |
| 1812 if (unsafe_chroot_detected) { | |
| 1813 st->print("WARNING!! "); | |
| 1814 st->print_cr(unstable_chroot_error); | |
| 1815 } | |
| 1816 | |
| 1817 // libc, pthread | |
| 1818 st->print("libc:"); | |
| 1819 st->print(os::Linux::glibc_version()); st->print(" "); | |
| 1820 st->print(os::Linux::libpthread_version()); st->print(" "); | |
| 1821 if (os::Linux::is_LinuxThreads()) { | |
| 1822 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); | |
| 1823 } | |
| 1824 st->cr(); | |
| 1825 | |
| 1826 // rlimit | |
| 1827 st->print("rlimit:"); | |
| 1828 struct rlimit rlim; | |
| 1829 | |
| 1830 st->print(" STACK "); | |
| 1831 getrlimit(RLIMIT_STACK, &rlim); | |
| 1832 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); | |
| 1833 else st->print("%uk", rlim.rlim_cur >> 10); | |
| 1834 | |
| 1835 st->print(", CORE "); | |
| 1836 getrlimit(RLIMIT_CORE, &rlim); | |
| 1837 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); | |
| 1838 else st->print("%uk", rlim.rlim_cur >> 10); | |
| 1839 | |
| 1840 st->print(", NPROC "); | |
| 1841 getrlimit(RLIMIT_NPROC, &rlim); | |
| 1842 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); | |
| 1843 else st->print("%d", rlim.rlim_cur); | |
| 1844 | |
| 1845 st->print(", NOFILE "); | |
| 1846 getrlimit(RLIMIT_NOFILE, &rlim); | |
| 1847 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); | |
| 1848 else st->print("%d", rlim.rlim_cur); | |
| 1849 | |
| 1850 st->print(", AS "); | |
| 1851 getrlimit(RLIMIT_AS, &rlim); | |
| 1852 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); | |
| 1853 else st->print("%uk", rlim.rlim_cur >> 10); | |
| 1854 st->cr(); | |
| 1855 | |
| 1856 // load average | |
| 1857 st->print("load average:"); | |
| 1858 double loadavg[3]; | |
| 1859 os::loadavg(loadavg, 3); | |
| 1860 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); | |
| 1861 st->cr(); | |
| 1862 } | |
| 1863 | |
| 1864 void os::print_memory_info(outputStream* st) { | |
| 1865 | |
| 1866 st->print("Memory:"); | |
| 1867 st->print(" %dk page", os::vm_page_size()>>10); | |
| 1868 | |
| 1869 // values in struct sysinfo are "unsigned long" | |
| 1870 struct sysinfo si; | |
| 1871 sysinfo(&si); | |
| 1872 | |
| 1873 st->print(", physical " UINT64_FORMAT "k", | |
| 1874 os::physical_memory() >> 10); | |
| 1875 st->print("(" UINT64_FORMAT "k free)", | |
| 1876 os::available_memory() >> 10); | |
| 1877 st->print(", swap " UINT64_FORMAT "k", | |
| 1878 ((jlong)si.totalswap * si.mem_unit) >> 10); | |
| 1879 st->print("(" UINT64_FORMAT "k free)", | |
| 1880 ((jlong)si.freeswap * si.mem_unit) >> 10); | |
| 1881 st->cr(); | |
| 1882 } | |
| 1883 | |
| 1884 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific | |
| 1885 // but they're the same for all the linux arch that we support | |
| 1886 // and they're the same for solaris but there's no common place to put this. | |
| 1887 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", | |
| 1888 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", | |
| 1889 "ILL_COPROC", "ILL_BADSTK" }; | |
| 1890 | |
| 1891 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", | |
| 1892 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", | |
| 1893 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; | |
| 1894 | |
| 1895 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; | |
| 1896 | |
| 1897 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; | |
| 1898 | |
| 1899 void os::print_siginfo(outputStream* st, void* siginfo) { | |
| 1900 st->print("siginfo:"); | |
| 1901 | |
| 1902 const int buflen = 100; | |
| 1903 char buf[buflen]; | |
| 1904 siginfo_t *si = (siginfo_t*)siginfo; | |
| 1905 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); | |
| 1906 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { | |
| 1907 st->print("si_errno=%s", buf); | |
| 1908 } else { | |
| 1909 st->print("si_errno=%d", si->si_errno); | |
| 1910 } | |
| 1911 const int c = si->si_code; | |
| 1912 assert(c > 0, "unexpected si_code"); | |
| 1913 switch (si->si_signo) { | |
| 1914 case SIGILL: | |
| 1915 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); | |
| 1916 st->print(", si_addr=" PTR_FORMAT, si->si_addr); | |
| 1917 break; | |
| 1918 case SIGFPE: | |
| 1919 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); | |
| 1920 st->print(", si_addr=" PTR_FORMAT, si->si_addr); | |
| 1921 break; | |
| 1922 case SIGSEGV: | |
| 1923 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); | |
| 1924 st->print(", si_addr=" PTR_FORMAT, si->si_addr); | |
| 1925 break; | |
| 1926 case SIGBUS: | |
| 1927 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); | |
| 1928 st->print(", si_addr=" PTR_FORMAT, si->si_addr); | |
| 1929 break; | |
| 1930 default: | |
| 1931 st->print(", si_code=%d", si->si_code); | |
| 1932 // no si_addr | |
| 1933 } | |
| 1934 | |
| 1935 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && | |
| 1936 UseSharedSpaces) { | |
| 1937 FileMapInfo* mapinfo = FileMapInfo::current_info(); | |
| 1938 if (mapinfo->is_in_shared_space(si->si_addr)) { | |
| 1939 st->print("\n\nError accessing class data sharing archive." \ | |
| 1940 " Mapped file inaccessible during execution, " \ | |
| 1941 " possible disk/network problem."); | |
| 1942 } | |
| 1943 } | |
| 1944 st->cr(); | |
| 1945 } | |
| 1946 | |
| 1947 | |
| 1948 static void print_signal_handler(outputStream* st, int sig, | |
| 1949 char* buf, size_t buflen); | |
| 1950 | |
| 1951 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { | |
| 1952 st->print_cr("Signal Handlers:"); | |
| 1953 print_signal_handler(st, SIGSEGV, buf, buflen); | |
| 1954 print_signal_handler(st, SIGBUS , buf, buflen); | |
| 1955 print_signal_handler(st, SIGFPE , buf, buflen); | |
| 1956 print_signal_handler(st, SIGPIPE, buf, buflen); | |
| 1957 print_signal_handler(st, SIGXFSZ, buf, buflen); | |
| 1958 print_signal_handler(st, SIGILL , buf, buflen); | |
| 1959 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); | |
| 1960 print_signal_handler(st, SR_signum, buf, buflen); | |
| 1961 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); | |
| 1962 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); | |
| 1963 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); | |
| 1964 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); | |
| 1965 } | |
| 1966 | |
| 1967 static char saved_jvm_path[MAXPATHLEN] = {0}; | |
| 1968 | |
| 1969 // Find the full path to the current module, libjvm.so or libjvm_g.so | |
| 1970 void os::jvm_path(char *buf, jint len) { | |
| 1971 // Error checking. | |
| 1972 if (len < MAXPATHLEN) { | |
| 1973 assert(false, "must use a large-enough buffer"); | |
| 1974 buf[0] = '\0'; | |
| 1975 return; | |
| 1976 } | |
| 1977 // Lazy resolve the path to current module. | |
| 1978 if (saved_jvm_path[0] != 0) { | |
| 1979 strcpy(buf, saved_jvm_path); | |
| 1980 return; | |
| 1981 } | |
| 1982 | |
| 1983 char dli_fname[MAXPATHLEN]; | |
| 1984 bool ret = dll_address_to_library_name( | |
| 1985 CAST_FROM_FN_PTR(address, os::jvm_path), | |
| 1986 dli_fname, sizeof(dli_fname), NULL); | |
| 1987 assert(ret != 0, "cannot locate libjvm"); | |
| 1988 realpath(dli_fname, buf); | |
| 1989 | |
| 1990 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { | |
| 1991 // Support for the gamma launcher. Typical value for buf is | |
| 1992 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at | |
| 1993 // the right place in the string, then assume we are installed in a JDK and | |
| 1994 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix | |
| 1995 // up the path so it looks like libjvm.so is installed there (append a | |
| 1996 // fake suffix hotspot/libjvm.so). | |
| 1997 const char *p = buf + strlen(buf) - 1; | |
| 1998 for (int count = 0; p > buf && count < 5; ++count) { | |
| 1999 for (--p; p > buf && *p != '/'; --p) | |
| 2000 /* empty */ ; | |
| 2001 } | |
| 2002 | |
| 2003 if (strncmp(p, "/jre/lib/", 9) != 0) { | |
| 2004 // Look for JAVA_HOME in the environment. | |
| 2005 char* java_home_var = ::getenv("JAVA_HOME"); | |
| 2006 if (java_home_var != NULL && java_home_var[0] != 0) { | |
| 2007 // Check the current module name "libjvm.so" or "libjvm_g.so". | |
| 2008 p = strrchr(buf, '/'); | |
| 2009 assert(strstr(p, "/libjvm") == p, "invalid library name"); | |
| 2010 p = strstr(p, "_g") ? "_g" : ""; | |
| 2011 | |
| 2012 realpath(java_home_var, buf); | |
| 2013 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); | |
| 2014 if (0 == access(buf, F_OK)) { | |
| 2015 // Use current module name "libjvm[_g].so" instead of | |
| 2016 // "libjvm"debug_only("_g")".so" since for fastdebug version | |
| 2017 // we should have "libjvm.so" but debug_only("_g") adds "_g"! | |
| 2018 // It is used when we are choosing the HPI library's name | |
| 2019 // "libhpi[_g].so" in hpi::initialize_get_interface(). | |
| 2020 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); | |
| 2021 } else { | |
| 2022 // Go back to path of .so | |
| 2023 realpath(dli_fname, buf); | |
| 2024 } | |
| 2025 } | |
| 2026 } | |
| 2027 } | |
| 2028 | |
| 2029 strcpy(saved_jvm_path, buf); | |
| 2030 } | |
| 2031 | |
| 2032 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { | |
| 2033 // no prefix required, not even "_" | |
| 2034 } | |
| 2035 | |
| 2036 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { | |
| 2037 // no suffix required | |
| 2038 } | |
| 2039 | |
| 2040 //////////////////////////////////////////////////////////////////////////////// | |
| 2041 // sun.misc.Signal support | |
| 2042 | |
| 2043 static volatile jint sigint_count = 0; | |
| 2044 | |
| 2045 static void | |
| 2046 UserHandler(int sig, void *siginfo, void *context) { | |
| 2047 // 4511530 - sem_post is serialized and handled by the manager thread. When | |
| 2048 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We | |
| 2049 // don't want to flood the manager thread with sem_post requests. | |
| 2050 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) | |
| 2051 return; | |
| 2052 | |
| 2053 // Ctrl-C is pressed during error reporting, likely because the error | |
| 2054 // handler fails to abort. Let VM die immediately. | |
| 2055 if (sig == SIGINT && is_error_reported()) { | |
| 2056 os::die(); | |
| 2057 } | |
| 2058 | |
| 2059 os::signal_notify(sig); | |
| 2060 } | |
| 2061 | |
| 2062 void* os::user_handler() { | |
| 2063 return CAST_FROM_FN_PTR(void*, UserHandler); | |
| 2064 } | |
| 2065 | |
| 2066 extern "C" { | |
| 2067 typedef void (*sa_handler_t)(int); | |
| 2068 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); | |
| 2069 } | |
| 2070 | |
| 2071 void* os::signal(int signal_number, void* handler) { | |
| 2072 struct sigaction sigAct, oldSigAct; | |
| 2073 | |
| 2074 sigfillset(&(sigAct.sa_mask)); | |
| 2075 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; | |
| 2076 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); | |
| 2077 | |
| 2078 if (sigaction(signal_number, &sigAct, &oldSigAct)) { | |
| 2079 // -1 means registration failed | |
| 2080 return (void *)-1; | |
| 2081 } | |
| 2082 | |
| 2083 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); | |
| 2084 } | |
| 2085 | |
| 2086 void os::signal_raise(int signal_number) { | |
| 2087 ::raise(signal_number); | |
| 2088 } | |
| 2089 | |
| 2090 /* | |
| 2091 * The following code is moved from os.cpp for making this | |
| 2092 * code platform specific, which it is by its very nature. | |
| 2093 */ | |
| 2094 | |
| 2095 // Will be modified when max signal is changed to be dynamic | |
| 2096 int os::sigexitnum_pd() { | |
| 2097 return NSIG; | |
| 2098 } | |
| 2099 | |
| 2100 // a counter for each possible signal value | |
| 2101 static volatile jint pending_signals[NSIG+1] = { 0 }; | |
| 2102 | |
| 2103 // Linux(POSIX) specific hand shaking semaphore. | |
| 2104 static sem_t sig_sem; | |
| 2105 | |
| 2106 void os::signal_init_pd() { | |
| 2107 // Initialize signal structures | |
| 2108 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); | |
| 2109 | |
| 2110 // Initialize signal semaphore | |
| 2111 ::sem_init(&sig_sem, 0, 0); | |
| 2112 } | |
| 2113 | |
| 2114 void os::signal_notify(int sig) { | |
| 2115 Atomic::inc(&pending_signals[sig]); | |
| 2116 ::sem_post(&sig_sem); | |
| 2117 } | |
| 2118 | |
| 2119 static int check_pending_signals(bool wait) { | |
| 2120 Atomic::store(0, &sigint_count); | |
| 2121 for (;;) { | |
| 2122 for (int i = 0; i < NSIG + 1; i++) { | |
| 2123 jint n = pending_signals[i]; | |
| 2124 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { | |
| 2125 return i; | |
| 2126 } | |
| 2127 } | |
| 2128 if (!wait) { | |
| 2129 return -1; | |
| 2130 } | |
| 2131 JavaThread *thread = JavaThread::current(); | |
| 2132 ThreadBlockInVM tbivm(thread); | |
| 2133 | |
| 2134 bool threadIsSuspended; | |
| 2135 do { | |
| 2136 thread->set_suspend_equivalent(); | |
| 2137 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() | |
| 2138 ::sem_wait(&sig_sem); | |
| 2139 | |
| 2140 // were we externally suspended while we were waiting? | |
| 2141 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); | |
| 2142 if (threadIsSuspended) { | |
| 2143 // | |
| 2144 // The semaphore has been incremented, but while we were waiting | |
| 2145 // another thread suspended us. We don't want to continue running | |
| 2146 // while suspended because that would surprise the thread that | |
| 2147 // suspended us. | |
| 2148 // | |
| 2149 ::sem_post(&sig_sem); | |
| 2150 | |
| 2151 thread->java_suspend_self(); | |
| 2152 } | |
| 2153 } while (threadIsSuspended); | |
| 2154 } | |
| 2155 } | |
| 2156 | |
| 2157 int os::signal_lookup() { | |
| 2158 return check_pending_signals(false); | |
| 2159 } | |
| 2160 | |
| 2161 int os::signal_wait() { | |
| 2162 return check_pending_signals(true); | |
| 2163 } | |
| 2164 | |
| 2165 //////////////////////////////////////////////////////////////////////////////// | |
| 2166 // Virtual Memory | |
| 2167 | |
| 2168 int os::vm_page_size() { | |
| 2169 // Seems redundant as all get out | |
| 2170 assert(os::Linux::page_size() != -1, "must call os::init"); | |
| 2171 return os::Linux::page_size(); | |
| 2172 } | |
| 2173 | |
| 2174 // Solaris allocates memory by pages. | |
| 2175 int os::vm_allocation_granularity() { | |
| 2176 assert(os::Linux::page_size() != -1, "must call os::init"); | |
| 2177 return os::Linux::page_size(); | |
| 2178 } | |
| 2179 | |
| 2180 // Rationale behind this function: | |
| 2181 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable | |
| 2182 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get | |
| 2183 // samples for JITted code. Here we create private executable mapping over the code cache | |
| 2184 // and then we can use standard (well, almost, as mapping can change) way to provide | |
| 2185 // info for the reporting script by storing timestamp and location of symbol | |
| 2186 void linux_wrap_code(char* base, size_t size) { | |
| 2187 static volatile jint cnt = 0; | |
| 2188 | |
| 2189 if (!UseOprofile) { | |
| 2190 return; | |
| 2191 } | |
| 2192 | |
| 2193 char buf[40]; | |
| 2194 int num = Atomic::add(1, &cnt); | |
| 2195 | |
| 2196 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); | |
| 2197 unlink(buf); | |
| 2198 | |
| 2199 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); | |
| 2200 | |
| 2201 if (fd != -1) { | |
| 2202 off_t rv = lseek(fd, size-2, SEEK_SET); | |
| 2203 if (rv != (off_t)-1) { | |
| 2204 if (write(fd, "", 1) == 1) { | |
| 2205 mmap(base, size, | |
| 2206 PROT_READ|PROT_WRITE|PROT_EXEC, | |
| 2207 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); | |
| 2208 } | |
| 2209 } | |
| 2210 close(fd); | |
| 2211 unlink(buf); | |
| 2212 } | |
| 2213 } | |
| 2214 | |
| 2215 // NOTE: Linux kernel does not really reserve the pages for us. | |
| 2216 // All it does is to check if there are enough free pages | |
| 2217 // left at the time of mmap(). This could be a potential | |
| 2218 // problem. | |
| 2219 bool os::commit_memory(char* addr, size_t size) { | |
| 2220 uintptr_t res = (uintptr_t) ::mmap(addr, size, | |
| 2221 PROT_READ|PROT_WRITE|PROT_EXEC, | |
| 2222 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); | |
| 2223 return res != (uintptr_t) MAP_FAILED; | |
| 2224 } | |
| 2225 | |
| 2226 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { | |
| 2227 return commit_memory(addr, size); | |
| 2228 } | |
| 2229 | |
| 2230 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } | |
| 2231 void os::free_memory(char *addr, size_t bytes) { } | |
| 2232 void os::numa_make_global(char *addr, size_t bytes) { } | |
| 2233 void os::numa_make_local(char *addr, size_t bytes) { } | |
| 2234 bool os::numa_topology_changed() { return false; } | |
| 2235 size_t os::numa_get_groups_num() { return 1; } | |
| 2236 int os::numa_get_group_id() { return 0; } | |
| 2237 size_t os::numa_get_leaf_groups(int *ids, size_t size) { | |
| 2238 if (size > 0) { | |
| 2239 ids[0] = 0; | |
| 2240 return 1; | |
| 2241 } | |
| 2242 return 0; | |
| 2243 } | |
| 2244 | |
| 2245 bool os::get_page_info(char *start, page_info* info) { | |
| 2246 return false; | |
| 2247 } | |
| 2248 | |
| 2249 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { | |
| 2250 return end; | |
| 2251 } | |
| 2252 | |
| 2253 bool os::uncommit_memory(char* addr, size_t size) { | |
| 2254 return ::mmap(addr, size, | |
| 2255 PROT_READ|PROT_WRITE|PROT_EXEC, | |
| 2256 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) | |
| 2257 != MAP_FAILED; | |
| 2258 } | |
| 2259 | |
| 2260 static address _highest_vm_reserved_address = NULL; | |
| 2261 | |
| 2262 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory | |
| 2263 // at 'requested_addr'. If there are existing memory mappings at the same | |
| 2264 // location, however, they will be overwritten. If 'fixed' is false, | |
| 2265 // 'requested_addr' is only treated as a hint, the return value may or | |
| 2266 // may not start from the requested address. Unlike Linux mmap(), this | |
| 2267 // function returns NULL to indicate failure. | |
| 2268 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { | |
| 2269 char * addr; | |
| 2270 int flags; | |
| 2271 | |
| 2272 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; | |
| 2273 if (fixed) { | |
| 2274 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); | |
| 2275 flags |= MAP_FIXED; | |
| 2276 } | |
| 2277 | |
| 2278 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, | |
| 2279 flags, -1, 0); | |
| 2280 | |
| 2281 if (addr != MAP_FAILED) { | |
| 2282 // anon_mmap() should only get called during VM initialization, | |
| 2283 // don't need lock (actually we can skip locking even it can be called | |
| 2284 // from multiple threads, because _highest_vm_reserved_address is just a | |
| 2285 // hint about the upper limit of non-stack memory regions.) | |
| 2286 if ((address)addr + bytes > _highest_vm_reserved_address) { | |
| 2287 _highest_vm_reserved_address = (address)addr + bytes; | |
| 2288 } | |
| 2289 } | |
| 2290 | |
| 2291 return addr == MAP_FAILED ? NULL : addr; | |
| 2292 } | |
| 2293 | |
| 2294 // Don't update _highest_vm_reserved_address, because there might be memory | |
| 2295 // regions above addr + size. If so, releasing a memory region only creates | |
| 2296 // a hole in the address space, it doesn't help prevent heap-stack collision. | |
| 2297 // | |
| 2298 static int anon_munmap(char * addr, size_t size) { | |
| 2299 return ::munmap(addr, size) == 0; | |
| 2300 } | |
| 2301 | |
| 2302 char* os::reserve_memory(size_t bytes, char* requested_addr, | |
| 2303 size_t alignment_hint) { | |
| 2304 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); | |
| 2305 } | |
| 2306 | |
| 2307 bool os::release_memory(char* addr, size_t size) { | |
| 2308 return anon_munmap(addr, size); | |
| 2309 } | |
| 2310 | |
| 2311 static address highest_vm_reserved_address() { | |
| 2312 return _highest_vm_reserved_address; | |
| 2313 } | |
| 2314 | |
| 2315 static bool linux_mprotect(char* addr, size_t size, int prot) { | |
| 2316 // Linux wants the mprotect address argument to be page aligned. | |
| 2317 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); | |
| 2318 | |
| 2319 // According to SUSv3, mprotect() should only be used with mappings | |
| 2320 // established by mmap(), and mmap() always maps whole pages. Unaligned | |
| 2321 // 'addr' likely indicates problem in the VM (e.g. trying to change | |
| 2322 // protection of malloc'ed or statically allocated memory). Check the | |
| 2323 // caller if you hit this assert. | |
| 2324 assert(addr == bottom, "sanity check"); | |
| 2325 | |
| 2326 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); | |
| 2327 return ::mprotect(bottom, size, prot) == 0; | |
| 2328 } | |
| 2329 | |
| 2330 bool os::protect_memory(char* addr, size_t size) { | |
| 2331 return linux_mprotect(addr, size, PROT_READ); | |
| 2332 } | |
| 2333 | |
| 2334 bool os::guard_memory(char* addr, size_t size) { | |
| 2335 return linux_mprotect(addr, size, PROT_NONE); | |
| 2336 } | |
| 2337 | |
| 2338 bool os::unguard_memory(char* addr, size_t size) { | |
| 2339 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC); | |
| 2340 } | |
| 2341 | |
| 2342 // Large page support | |
| 2343 | |
| 2344 static size_t _large_page_size = 0; | |
| 2345 | |
| 2346 bool os::large_page_init() { | |
| 2347 if (!UseLargePages) return false; | |
| 2348 | |
| 2349 if (LargePageSizeInBytes) { | |
| 2350 _large_page_size = LargePageSizeInBytes; | |
| 2351 } else { | |
| 2352 // large_page_size on Linux is used to round up heap size. x86 uses either | |
| 2353 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) | |
| 2354 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use | |
| 2355 // page as large as 256M. | |
| 2356 // | |
| 2357 // Here we try to figure out page size by parsing /proc/meminfo and looking | |
| 2358 // for a line with the following format: | |
| 2359 // Hugepagesize: 2048 kB | |
| 2360 // | |
| 2361 // If we can't determine the value (e.g. /proc is not mounted, or the text | |
| 2362 // format has been changed), we'll use the largest page size supported by | |
| 2363 // the processor. | |
| 2364 | |
| 2365 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); | |
| 2366 | |
| 2367 FILE *fp = fopen("/proc/meminfo", "r"); | |
| 2368 if (fp) { | |
| 2369 while (!feof(fp)) { | |
| 2370 int x = 0; | |
| 2371 char buf[16]; | |
| 2372 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { | |
| 2373 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { | |
| 2374 _large_page_size = x * K; | |
| 2375 break; | |
| 2376 } | |
| 2377 } else { | |
| 2378 // skip to next line | |
| 2379 for (;;) { | |
| 2380 int ch = fgetc(fp); | |
| 2381 if (ch == EOF || ch == (int)'\n') break; | |
| 2382 } | |
| 2383 } | |
| 2384 } | |
| 2385 fclose(fp); | |
| 2386 } | |
| 2387 } | |
| 2388 | |
| 2389 const size_t default_page_size = (size_t)Linux::page_size(); | |
| 2390 if (_large_page_size > default_page_size) { | |
| 2391 _page_sizes[0] = _large_page_size; | |
| 2392 _page_sizes[1] = default_page_size; | |
| 2393 _page_sizes[2] = 0; | |
| 2394 } | |
| 2395 | |
| 2396 // Large page support is available on 2.6 or newer kernel, some vendors | |
| 2397 // (e.g. Redhat) have backported it to their 2.4 based distributions. | |
| 2398 // We optimistically assume the support is available. If later it turns out | |
| 2399 // not true, VM will automatically switch to use regular page size. | |
| 2400 return true; | |
| 2401 } | |
| 2402 | |
| 2403 #ifndef SHM_HUGETLB | |
| 2404 #define SHM_HUGETLB 04000 | |
| 2405 #endif | |
| 2406 | |
| 2407 char* os::reserve_memory_special(size_t bytes) { | |
| 2408 assert(UseLargePages, "only for large pages"); | |
| 2409 | |
| 2410 key_t key = IPC_PRIVATE; | |
| 2411 char *addr; | |
| 2412 | |
| 2413 bool warn_on_failure = UseLargePages && | |
| 2414 (!FLAG_IS_DEFAULT(UseLargePages) || | |
| 2415 !FLAG_IS_DEFAULT(LargePageSizeInBytes) | |
| 2416 ); | |
| 2417 char msg[128]; | |
| 2418 | |
| 2419 // Create a large shared memory region to attach to based on size. | |
| 2420 // Currently, size is the total size of the heap | |
| 2421 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); | |
| 2422 if (shmid == -1) { | |
| 2423 // Possible reasons for shmget failure: | |
| 2424 // 1. shmmax is too small for Java heap. | |
| 2425 // > check shmmax value: cat /proc/sys/kernel/shmmax | |
| 2426 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax | |
| 2427 // 2. not enough large page memory. | |
| 2428 // > check available large pages: cat /proc/meminfo | |
| 2429 // > increase amount of large pages: | |
| 2430 // echo new_value > /proc/sys/vm/nr_hugepages | |
| 2431 // Note 1: different Linux may use different name for this property, | |
| 2432 // e.g. on Redhat AS-3 it is "hugetlb_pool". | |
| 2433 // Note 2: it's possible there's enough physical memory available but | |
| 2434 // they are so fragmented after a long run that they can't | |
| 2435 // coalesce into large pages. Try to reserve large pages when | |
| 2436 // the system is still "fresh". | |
| 2437 if (warn_on_failure) { | |
| 2438 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); | |
| 2439 warning(msg); | |
| 2440 } | |
| 2441 return NULL; | |
| 2442 } | |
| 2443 | |
| 2444 // attach to the region | |
| 2445 addr = (char*)shmat(shmid, NULL, 0); | |
| 2446 int err = errno; | |
| 2447 | |
| 2448 // Remove shmid. If shmat() is successful, the actual shared memory segment | |
| 2449 // will be deleted when it's detached by shmdt() or when the process | |
| 2450 // terminates. If shmat() is not successful this will remove the shared | |
| 2451 // segment immediately. | |
| 2452 shmctl(shmid, IPC_RMID, NULL); | |
| 2453 | |
| 2454 if ((intptr_t)addr == -1) { | |
| 2455 if (warn_on_failure) { | |
| 2456 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); | |
| 2457 warning(msg); | |
| 2458 } | |
| 2459 return NULL; | |
| 2460 } | |
| 2461 | |
| 2462 return addr; | |
| 2463 } | |
| 2464 | |
| 2465 bool os::release_memory_special(char* base, size_t bytes) { | |
| 2466 // detaching the SHM segment will also delete it, see reserve_memory_special() | |
| 2467 int rslt = shmdt(base); | |
| 2468 return rslt == 0; | |
| 2469 } | |
| 2470 | |
| 2471 size_t os::large_page_size() { | |
| 2472 return _large_page_size; | |
| 2473 } | |
| 2474 | |
| 2475 // Linux does not support anonymous mmap with large page memory. The only way | |
| 2476 // to reserve large page memory without file backing is through SysV shared | |
| 2477 // memory API. The entire memory region is committed and pinned upfront. | |
| 2478 // Hopefully this will change in the future... | |
| 2479 bool os::can_commit_large_page_memory() { | |
| 2480 return false; | |
| 2481 } | |
| 2482 | |
|
79
82db0859acbe
6642862: Code cache allocation fails with large pages after 6588638
jcoomes
parents:
62
diff
changeset
|
2483 bool os::can_execute_large_page_memory() { |
|
82db0859acbe
6642862: Code cache allocation fails with large pages after 6588638
jcoomes
parents:
62
diff
changeset
|
2484 return false; |
|
82db0859acbe
6642862: Code cache allocation fails with large pages after 6588638
jcoomes
parents:
62
diff
changeset
|
2485 } |
|
82db0859acbe
6642862: Code cache allocation fails with large pages after 6588638
jcoomes
parents:
62
diff
changeset
|
2486 |
| 0 | 2487 // Reserve memory at an arbitrary address, only if that area is |
| 2488 // available (and not reserved for something else). | |
| 2489 | |
| 2490 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { | |
| 2491 const int max_tries = 10; | |
| 2492 char* base[max_tries]; | |
| 2493 size_t size[max_tries]; | |
| 2494 const size_t gap = 0x000000; | |
| 2495 | |
| 2496 // Assert only that the size is a multiple of the page size, since | |
| 2497 // that's all that mmap requires, and since that's all we really know | |
| 2498 // about at this low abstraction level. If we need higher alignment, | |
| 2499 // we can either pass an alignment to this method or verify alignment | |
| 2500 // in one of the methods further up the call chain. See bug 5044738. | |
| 2501 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); | |
| 2502 | |
| 2503 // Repeatedly allocate blocks until the block is allocated at the | |
| 2504 // right spot. Give up after max_tries. Note that reserve_memory() will | |
| 2505 // automatically update _highest_vm_reserved_address if the call is | |
| 2506 // successful. The variable tracks the highest memory address every reserved | |
| 2507 // by JVM. It is used to detect heap-stack collision if running with | |
| 2508 // fixed-stack LinuxThreads. Because here we may attempt to reserve more | |
| 2509 // space than needed, it could confuse the collision detecting code. To | |
| 2510 // solve the problem, save current _highest_vm_reserved_address and | |
| 2511 // calculate the correct value before return. | |
| 2512 address old_highest = _highest_vm_reserved_address; | |
| 2513 | |
| 2514 // Linux mmap allows caller to pass an address as hint; give it a try first, | |
| 2515 // if kernel honors the hint then we can return immediately. | |
| 2516 char * addr = anon_mmap(requested_addr, bytes, false); | |
| 2517 if (addr == requested_addr) { | |
| 2518 return requested_addr; | |
| 2519 } | |
| 2520 | |
| 2521 if (addr != NULL) { | |
| 2522 // mmap() is successful but it fails to reserve at the requested address | |
| 2523 anon_munmap(addr, bytes); | |
| 2524 } | |
| 2525 | |
| 2526 int i; | |
| 2527 for (i = 0; i < max_tries; ++i) { | |
| 2528 base[i] = reserve_memory(bytes); | |
| 2529 | |
| 2530 if (base[i] != NULL) { | |
| 2531 // Is this the block we wanted? | |
| 2532 if (base[i] == requested_addr) { | |
| 2533 size[i] = bytes; | |
| 2534 break; | |
| 2535 } | |
| 2536 | |
| 2537 // Does this overlap the block we wanted? Give back the overlapped | |
| 2538 // parts and try again. | |
| 2539 | |
| 2540 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; | |
| 2541 if (top_overlap >= 0 && top_overlap < bytes) { | |
| 2542 unmap_memory(base[i], top_overlap); | |
| 2543 base[i] += top_overlap; | |
| 2544 size[i] = bytes - top_overlap; | |
| 2545 } else { | |
| 2546 size_t bottom_overlap = base[i] + bytes - requested_addr; | |
| 2547 if (bottom_overlap >= 0 && bottom_overlap < bytes) { | |
| 2548 unmap_memory(requested_addr, bottom_overlap); | |
| 2549 size[i] = bytes - bottom_overlap; | |
| 2550 } else { | |
| 2551 size[i] = bytes; | |
| 2552 } | |
| 2553 } | |
| 2554 } | |
| 2555 } | |
| 2556 | |
| 2557 // Give back the unused reserved pieces. | |
| 2558 | |
| 2559 for (int j = 0; j < i; ++j) { | |
| 2560 if (base[j] != NULL) { | |
| 2561 unmap_memory(base[j], size[j]); | |
| 2562 } | |
| 2563 } | |
| 2564 | |
| 2565 if (i < max_tries) { | |
| 2566 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); | |
| 2567 return requested_addr; | |
| 2568 } else { | |
| 2569 _highest_vm_reserved_address = old_highest; | |
| 2570 return NULL; | |
| 2571 } | |
| 2572 } | |
| 2573 | |
| 2574 size_t os::read(int fd, void *buf, unsigned int nBytes) { | |
| 2575 return ::read(fd, buf, nBytes); | |
| 2576 } | |
| 2577 | |
| 2578 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. | |
| 2579 // Solaris uses poll(), linux uses park(). | |
| 2580 // Poll() is likely a better choice, assuming that Thread.interrupt() | |
| 2581 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with | |
| 2582 // SIGSEGV, see 4355769. | |
| 2583 | |
| 2584 const int NANOSECS_PER_MILLISECS = 1000000; | |
| 2585 | |
| 2586 int os::sleep(Thread* thread, jlong millis, bool interruptible) { | |
| 2587 assert(thread == Thread::current(), "thread consistency check"); | |
| 2588 | |
| 2589 ParkEvent * const slp = thread->_SleepEvent ; | |
| 2590 slp->reset() ; | |
| 2591 OrderAccess::fence() ; | |
| 2592 | |
| 2593 if (interruptible) { | |
| 2594 jlong prevtime = javaTimeNanos(); | |
| 2595 | |
| 2596 for (;;) { | |
| 2597 if (os::is_interrupted(thread, true)) { | |
| 2598 return OS_INTRPT; | |
| 2599 } | |
| 2600 | |
| 2601 jlong newtime = javaTimeNanos(); | |
| 2602 | |
| 2603 if (newtime - prevtime < 0) { | |
| 2604 // time moving backwards, should only happen if no monotonic clock | |
| 2605 // not a guarantee() because JVM should not abort on kernel/glibc bugs | |
| 2606 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); | |
| 2607 } else { | |
| 2608 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; | |
| 2609 } | |
| 2610 | |
| 2611 if(millis <= 0) { | |
| 2612 return OS_OK; | |
| 2613 } | |
| 2614 | |
| 2615 prevtime = newtime; | |
| 2616 | |
| 2617 { | |
| 2618 assert(thread->is_Java_thread(), "sanity check"); | |
| 2619 JavaThread *jt = (JavaThread *) thread; | |
| 2620 ThreadBlockInVM tbivm(jt); | |
| 2621 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); | |
| 2622 | |
| 2623 jt->set_suspend_equivalent(); | |
| 2624 // cleared by handle_special_suspend_equivalent_condition() or | |
| 2625 // java_suspend_self() via check_and_wait_while_suspended() | |
| 2626 | |
| 2627 slp->park(millis); | |
| 2628 | |
| 2629 // were we externally suspended while we were waiting? | |
| 2630 jt->check_and_wait_while_suspended(); | |
| 2631 } | |
| 2632 } | |
| 2633 } else { | |
| 2634 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); | |
| 2635 jlong prevtime = javaTimeNanos(); | |
| 2636 | |
| 2637 for (;;) { | |
| 2638 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on | |
| 2639 // the 1st iteration ... | |
| 2640 jlong newtime = javaTimeNanos(); | |
| 2641 | |
| 2642 if (newtime - prevtime < 0) { | |
| 2643 // time moving backwards, should only happen if no monotonic clock | |
| 2644 // not a guarantee() because JVM should not abort on kernel/glibc bugs | |
| 2645 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); | |
| 2646 } else { | |
| 2647 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; | |
| 2648 } | |
| 2649 | |
| 2650 if(millis <= 0) break ; | |
| 2651 | |
| 2652 prevtime = newtime; | |
| 2653 slp->park(millis); | |
| 2654 } | |
| 2655 return OS_OK ; | |
| 2656 } | |
| 2657 } | |
| 2658 | |
| 2659 int os::naked_sleep() { | |
| 2660 // %% make the sleep time an integer flag. for now use 1 millisec. | |
| 2661 return os::sleep(Thread::current(), 1, false); | |
| 2662 } | |
| 2663 | |
| 2664 // Sleep forever; naked call to OS-specific sleep; use with CAUTION | |
| 2665 void os::infinite_sleep() { | |
| 2666 while (true) { // sleep forever ... | |
| 2667 ::sleep(100); // ... 100 seconds at a time | |
| 2668 } | |
| 2669 } | |
| 2670 | |
| 2671 // Used to convert frequent JVM_Yield() to nops | |
| 2672 bool os::dont_yield() { | |
| 2673 return DontYieldALot; | |
| 2674 } | |
| 2675 | |
| 2676 void os::yield() { | |
| 2677 sched_yield(); | |
| 2678 } | |
| 2679 | |
| 2680 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} | |
| 2681 | |
| 2682 void os::yield_all(int attempts) { | |
| 2683 // Yields to all threads, including threads with lower priorities | |
| 2684 // Threads on Linux are all with same priority. The Solaris style | |
| 2685 // os::yield_all() with nanosleep(1ms) is not necessary. | |
| 2686 sched_yield(); | |
| 2687 } | |
| 2688 | |
| 2689 // Called from the tight loops to possibly influence time-sharing heuristics | |
| 2690 void os::loop_breaker(int attempts) { | |
| 2691 os::yield_all(attempts); | |
| 2692 } | |
| 2693 | |
| 2694 //////////////////////////////////////////////////////////////////////////////// | |
| 2695 // thread priority support | |
| 2696 | |
| 2697 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER | |
| 2698 // only supports dynamic priority, static priority must be zero. For real-time | |
| 2699 // applications, Linux supports SCHED_RR which allows static priority (1-99). | |
| 2700 // However, for large multi-threaded applications, SCHED_RR is not only slower | |
| 2701 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out | |
| 2702 // of 5 runs - Sep 2005). | |
| 2703 // | |
| 2704 // The following code actually changes the niceness of kernel-thread/LWP. It | |
| 2705 // has an assumption that setpriority() only modifies one kernel-thread/LWP, | |
| 2706 // not the entire user process, and user level threads are 1:1 mapped to kernel | |
| 2707 // threads. It has always been the case, but could change in the future. For | |
| 2708 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). | |
| 2709 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. | |
| 2710 | |
| 2711 int os::java_to_os_priority[MaxPriority + 1] = { | |
| 2712 19, // 0 Entry should never be used | |
| 2713 | |
| 2714 4, // 1 MinPriority | |
| 2715 3, // 2 | |
| 2716 2, // 3 | |
| 2717 | |
| 2718 1, // 4 | |
| 2719 0, // 5 NormPriority | |
| 2720 -1, // 6 | |
| 2721 | |
| 2722 -2, // 7 | |
| 2723 -3, // 8 | |
| 2724 -4, // 9 NearMaxPriority | |
| 2725 | |
| 2726 -5 // 10 MaxPriority | |
| 2727 }; | |
| 2728 | |
| 2729 static int prio_init() { | |
| 2730 if (ThreadPriorityPolicy == 1) { | |
| 2731 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 | |
| 2732 // if effective uid is not root. Perhaps, a more elegant way of doing | |
| 2733 // this is to test CAP_SYS_NICE capability, but that will require libcap.so | |
| 2734 if (geteuid() != 0) { | |
| 2735 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { | |
| 2736 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); | |
| 2737 } | |
| 2738 ThreadPriorityPolicy = 0; | |
| 2739 } | |
| 2740 } | |
| 2741 return 0; | |
| 2742 } | |
| 2743 | |
| 2744 OSReturn os::set_native_priority(Thread* thread, int newpri) { | |
| 2745 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; | |
| 2746 | |
| 2747 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); | |
| 2748 return (ret == 0) ? OS_OK : OS_ERR; | |
| 2749 } | |
| 2750 | |
| 2751 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { | |
| 2752 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { | |
| 2753 *priority_ptr = java_to_os_priority[NormPriority]; | |
| 2754 return OS_OK; | |
| 2755 } | |
| 2756 | |
| 2757 errno = 0; | |
| 2758 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); | |
| 2759 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); | |
| 2760 } | |
| 2761 | |
| 2762 // Hint to the underlying OS that a task switch would not be good. | |
| 2763 // Void return because it's a hint and can fail. | |
| 2764 void os::hint_no_preempt() {} | |
| 2765 | |
| 2766 //////////////////////////////////////////////////////////////////////////////// | |
| 2767 // suspend/resume support | |
| 2768 | |
| 2769 // the low-level signal-based suspend/resume support is a remnant from the | |
| 2770 // old VM-suspension that used to be for java-suspension, safepoints etc, | |
| 2771 // within hotspot. Now there is a single use-case for this: | |
| 2772 // - calling get_thread_pc() on the VMThread by the flat-profiler task | |
| 2773 // that runs in the watcher thread. | |
| 2774 // The remaining code is greatly simplified from the more general suspension | |
| 2775 // code that used to be used. | |
| 2776 // | |
| 2777 // The protocol is quite simple: | |
| 2778 // - suspend: | |
| 2779 // - sends a signal to the target thread | |
| 2780 // - polls the suspend state of the osthread using a yield loop | |
| 2781 // - target thread signal handler (SR_handler) sets suspend state | |
| 2782 // and blocks in sigsuspend until continued | |
| 2783 // - resume: | |
| 2784 // - sets target osthread state to continue | |
| 2785 // - sends signal to end the sigsuspend loop in the SR_handler | |
| 2786 // | |
| 2787 // Note that the SR_lock plays no role in this suspend/resume protocol. | |
| 2788 // | |
| 2789 | |
| 2790 static void resume_clear_context(OSThread *osthread) { | |
| 2791 osthread->set_ucontext(NULL); | |
| 2792 osthread->set_siginfo(NULL); | |
| 2793 | |
| 2794 // notify the suspend action is completed, we have now resumed | |
| 2795 osthread->sr.clear_suspended(); | |
| 2796 } | |
| 2797 | |
| 2798 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { | |
| 2799 osthread->set_ucontext(context); | |
| 2800 osthread->set_siginfo(siginfo); | |
| 2801 } | |
| 2802 | |
| 2803 // | |
| 2804 // Handler function invoked when a thread's execution is suspended or | |
| 2805 // resumed. We have to be careful that only async-safe functions are | |
| 2806 // called here (Note: most pthread functions are not async safe and | |
| 2807 // should be avoided.) | |
| 2808 // | |
| 2809 // Note: sigwait() is a more natural fit than sigsuspend() from an | |
| 2810 // interface point of view, but sigwait() prevents the signal hander | |
| 2811 // from being run. libpthread would get very confused by not having | |
| 2812 // its signal handlers run and prevents sigwait()'s use with the | |
| 2813 // mutex granting granting signal. | |
| 2814 // | |
| 2815 // Currently only ever called on the VMThread | |
| 2816 // | |
| 2817 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { | |
| 2818 // Save and restore errno to avoid confusing native code with EINTR | |
| 2819 // after sigsuspend. | |
| 2820 int old_errno = errno; | |
| 2821 | |
| 2822 Thread* thread = Thread::current(); | |
| 2823 OSThread* osthread = thread->osthread(); | |
| 2824 assert(thread->is_VM_thread(), "Must be VMThread"); | |
| 2825 // read current suspend action | |
| 2826 int action = osthread->sr.suspend_action(); | |
| 2827 if (action == SR_SUSPEND) { | |
| 2828 suspend_save_context(osthread, siginfo, context); | |
| 2829 | |
| 2830 // Notify the suspend action is about to be completed. do_suspend() | |
| 2831 // waits until SR_SUSPENDED is set and then returns. We will wait | |
| 2832 // here for a resume signal and that completes the suspend-other | |
| 2833 // action. do_suspend/do_resume is always called as a pair from | |
| 2834 // the same thread - so there are no races | |
| 2835 | |
| 2836 // notify the caller | |
| 2837 osthread->sr.set_suspended(); | |
| 2838 | |
| 2839 sigset_t suspend_set; // signals for sigsuspend() | |
| 2840 | |
| 2841 // get current set of blocked signals and unblock resume signal | |
| 2842 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); | |
| 2843 sigdelset(&suspend_set, SR_signum); | |
| 2844 | |
| 2845 // wait here until we are resumed | |
| 2846 do { | |
| 2847 sigsuspend(&suspend_set); | |
| 2848 // ignore all returns until we get a resume signal | |
| 2849 } while (osthread->sr.suspend_action() != SR_CONTINUE); | |
| 2850 | |
| 2851 resume_clear_context(osthread); | |
| 2852 | |
| 2853 } else { | |
| 2854 assert(action == SR_CONTINUE, "unexpected sr action"); | |
| 2855 // nothing special to do - just leave the handler | |
| 2856 } | |
| 2857 | |
| 2858 errno = old_errno; | |
| 2859 } | |
| 2860 | |
| 2861 | |
| 2862 static int SR_initialize() { | |
| 2863 struct sigaction act; | |
| 2864 char *s; | |
| 2865 /* Get signal number to use for suspend/resume */ | |
| 2866 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { | |
| 2867 int sig = ::strtol(s, 0, 10); | |
| 2868 if (sig > 0 || sig < _NSIG) { | |
| 2869 SR_signum = sig; | |
| 2870 } | |
| 2871 } | |
| 2872 | |
| 2873 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, | |
| 2874 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); | |
| 2875 | |
| 2876 sigemptyset(&SR_sigset); | |
| 2877 sigaddset(&SR_sigset, SR_signum); | |
| 2878 | |
| 2879 /* Set up signal handler for suspend/resume */ | |
| 2880 act.sa_flags = SA_RESTART|SA_SIGINFO; | |
| 2881 act.sa_handler = (void (*)(int)) SR_handler; | |
| 2882 | |
| 2883 // SR_signum is blocked by default. | |
| 2884 // 4528190 - We also need to block pthread restart signal (32 on all | |
| 2885 // supported Linux platforms). Note that LinuxThreads need to block | |
| 2886 // this signal for all threads to work properly. So we don't have | |
| 2887 // to use hard-coded signal number when setting up the mask. | |
| 2888 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); | |
| 2889 | |
| 2890 if (sigaction(SR_signum, &act, 0) == -1) { | |
| 2891 return -1; | |
| 2892 } | |
| 2893 | |
| 2894 // Save signal flag | |
| 2895 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); | |
| 2896 return 0; | |
| 2897 } | |
| 2898 | |
| 2899 static int SR_finalize() { | |
| 2900 return 0; | |
| 2901 } | |
| 2902 | |
| 2903 | |
| 2904 // returns true on success and false on error - really an error is fatal | |
| 2905 // but this seems the normal response to library errors | |
| 2906 static bool do_suspend(OSThread* osthread) { | |
| 2907 // mark as suspended and send signal | |
| 2908 osthread->sr.set_suspend_action(SR_SUSPEND); | |
| 2909 int status = pthread_kill(osthread->pthread_id(), SR_signum); | |
| 2910 assert_status(status == 0, status, "pthread_kill"); | |
| 2911 | |
| 2912 // check status and wait until notified of suspension | |
| 2913 if (status == 0) { | |
| 2914 for (int i = 0; !osthread->sr.is_suspended(); i++) { | |
| 2915 os::yield_all(i); | |
| 2916 } | |
| 2917 osthread->sr.set_suspend_action(SR_NONE); | |
| 2918 return true; | |
| 2919 } | |
| 2920 else { | |
| 2921 osthread->sr.set_suspend_action(SR_NONE); | |
| 2922 return false; | |
| 2923 } | |
| 2924 } | |
| 2925 | |
| 2926 static void do_resume(OSThread* osthread) { | |
| 2927 assert(osthread->sr.is_suspended(), "thread should be suspended"); | |
| 2928 osthread->sr.set_suspend_action(SR_CONTINUE); | |
| 2929 | |
| 2930 int status = pthread_kill(osthread->pthread_id(), SR_signum); | |
| 2931 assert_status(status == 0, status, "pthread_kill"); | |
| 2932 // check status and wait unit notified of resumption | |
| 2933 if (status == 0) { | |
| 2934 for (int i = 0; osthread->sr.is_suspended(); i++) { | |
| 2935 os::yield_all(i); | |
| 2936 } | |
| 2937 } | |
| 2938 osthread->sr.set_suspend_action(SR_NONE); | |
| 2939 } | |
| 2940 | |
| 2941 //////////////////////////////////////////////////////////////////////////////// | |
| 2942 // interrupt support | |
| 2943 | |
| 2944 void os::interrupt(Thread* thread) { | |
| 2945 assert(Thread::current() == thread || Threads_lock->owned_by_self(), | |
| 2946 "possibility of dangling Thread pointer"); | |
| 2947 | |
| 2948 OSThread* osthread = thread->osthread(); | |
| 2949 | |
| 2950 if (!osthread->interrupted()) { | |
| 2951 osthread->set_interrupted(true); | |
| 2952 // More than one thread can get here with the same value of osthread, | |
| 2953 // resulting in multiple notifications. We do, however, want the store | |
| 2954 // to interrupted() to be visible to other threads before we execute unpark(). | |
| 2955 OrderAccess::fence(); | |
| 2956 ParkEvent * const slp = thread->_SleepEvent ; | |
| 2957 if (slp != NULL) slp->unpark() ; | |
| 2958 } | |
| 2959 | |
| 2960 // For JSR166. Unpark even if interrupt status already was set | |
| 2961 if (thread->is_Java_thread()) | |
| 2962 ((JavaThread*)thread)->parker()->unpark(); | |
| 2963 | |
| 2964 ParkEvent * ev = thread->_ParkEvent ; | |
| 2965 if (ev != NULL) ev->unpark() ; | |
| 2966 | |
| 2967 } | |
| 2968 | |
| 2969 bool os::is_interrupted(Thread* thread, bool clear_interrupted) { | |
| 2970 assert(Thread::current() == thread || Threads_lock->owned_by_self(), | |
| 2971 "possibility of dangling Thread pointer"); | |
| 2972 | |
| 2973 OSThread* osthread = thread->osthread(); | |
| 2974 | |
| 2975 bool interrupted = osthread->interrupted(); | |
| 2976 | |
| 2977 if (interrupted && clear_interrupted) { | |
| 2978 osthread->set_interrupted(false); | |
| 2979 // consider thread->_SleepEvent->reset() ... optional optimization | |
| 2980 } | |
| 2981 | |
| 2982 return interrupted; | |
| 2983 } | |
| 2984 | |
| 2985 /////////////////////////////////////////////////////////////////////////////////// | |
| 2986 // signal handling (except suspend/resume) | |
| 2987 | |
| 2988 // This routine may be used by user applications as a "hook" to catch signals. | |
| 2989 // The user-defined signal handler must pass unrecognized signals to this | |
| 2990 // routine, and if it returns true (non-zero), then the signal handler must | |
| 2991 // return immediately. If the flag "abort_if_unrecognized" is true, then this | |
| 2992 // routine will never retun false (zero), but instead will execute a VM panic | |
| 2993 // routine kill the process. | |
| 2994 // | |
| 2995 // If this routine returns false, it is OK to call it again. This allows | |
| 2996 // the user-defined signal handler to perform checks either before or after | |
| 2997 // the VM performs its own checks. Naturally, the user code would be making | |
| 2998 // a serious error if it tried to handle an exception (such as a null check | |
| 2999 // or breakpoint) that the VM was generating for its own correct operation. | |
| 3000 // | |
| 3001 // This routine may recognize any of the following kinds of signals: | |
| 3002 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. | |
| 3003 // It should be consulted by handlers for any of those signals. | |
| 3004 // | |
| 3005 // The caller of this routine must pass in the three arguments supplied | |
| 3006 // to the function referred to in the "sa_sigaction" (not the "sa_handler") | |
| 3007 // field of the structure passed to sigaction(). This routine assumes that | |
| 3008 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. | |
| 3009 // | |
| 3010 // Note that the VM will print warnings if it detects conflicting signal | |
| 3011 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". | |
| 3012 // | |
| 3013 extern "C" int | |
| 3014 JVM_handle_linux_signal(int signo, siginfo_t* siginfo, | |
| 3015 void* ucontext, int abort_if_unrecognized); | |
| 3016 | |
| 3017 void signalHandler(int sig, siginfo_t* info, void* uc) { | |
| 3018 assert(info != NULL && uc != NULL, "it must be old kernel"); | |
| 3019 JVM_handle_linux_signal(sig, info, uc, true); | |
| 3020 } | |
| 3021 | |
| 3022 | |
| 3023 // This boolean allows users to forward their own non-matching signals | |
| 3024 // to JVM_handle_linux_signal, harmlessly. | |
| 3025 bool os::Linux::signal_handlers_are_installed = false; | |
| 3026 | |
| 3027 // For signal-chaining | |
| 3028 struct sigaction os::Linux::sigact[MAXSIGNUM]; | |
| 3029 unsigned int os::Linux::sigs = 0; | |
| 3030 bool os::Linux::libjsig_is_loaded = false; | |
| 3031 typedef struct sigaction *(*get_signal_t)(int); | |
| 3032 get_signal_t os::Linux::get_signal_action = NULL; | |
| 3033 | |
| 3034 struct sigaction* os::Linux::get_chained_signal_action(int sig) { | |
| 3035 struct sigaction *actp = NULL; | |
| 3036 | |
| 3037 if (libjsig_is_loaded) { | |
| 3038 // Retrieve the old signal handler from libjsig | |
| 3039 actp = (*get_signal_action)(sig); | |
| 3040 } | |
| 3041 if (actp == NULL) { | |
| 3042 // Retrieve the preinstalled signal handler from jvm | |
| 3043 actp = get_preinstalled_handler(sig); | |
| 3044 } | |
| 3045 | |
| 3046 return actp; | |
| 3047 } | |
| 3048 | |
| 3049 static bool call_chained_handler(struct sigaction *actp, int sig, | |
| 3050 siginfo_t *siginfo, void *context) { | |
| 3051 // Call the old signal handler | |
| 3052 if (actp->sa_handler == SIG_DFL) { | |
| 3053 // It's more reasonable to let jvm treat it as an unexpected exception | |
| 3054 // instead of taking the default action. | |
| 3055 return false; | |
| 3056 } else if (actp->sa_handler != SIG_IGN) { | |
| 3057 if ((actp->sa_flags & SA_NODEFER) == 0) { | |
| 3058 // automaticlly block the signal | |
| 3059 sigaddset(&(actp->sa_mask), sig); | |
| 3060 } | |
| 3061 | |
| 3062 sa_handler_t hand; | |
| 3063 sa_sigaction_t sa; | |
| 3064 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; | |
| 3065 // retrieve the chained handler | |
| 3066 if (siginfo_flag_set) { | |
| 3067 sa = actp->sa_sigaction; | |
| 3068 } else { | |
| 3069 hand = actp->sa_handler; | |
| 3070 } | |
| 3071 | |
| 3072 if ((actp->sa_flags & SA_RESETHAND) != 0) { | |
| 3073 actp->sa_handler = SIG_DFL; | |
| 3074 } | |
| 3075 | |
| 3076 // try to honor the signal mask | |
| 3077 sigset_t oset; | |
| 3078 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); | |
| 3079 | |
| 3080 // call into the chained handler | |
| 3081 if (siginfo_flag_set) { | |
| 3082 (*sa)(sig, siginfo, context); | |
| 3083 } else { | |
| 3084 (*hand)(sig); | |
| 3085 } | |
| 3086 | |
| 3087 // restore the signal mask | |
| 3088 pthread_sigmask(SIG_SETMASK, &oset, 0); | |
| 3089 } | |
| 3090 // Tell jvm's signal handler the signal is taken care of. | |
| 3091 return true; | |
| 3092 } | |
| 3093 | |
| 3094 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { | |
| 3095 bool chained = false; | |
| 3096 // signal-chaining | |
| 3097 if (UseSignalChaining) { | |
| 3098 struct sigaction *actp = get_chained_signal_action(sig); | |
| 3099 if (actp != NULL) { | |
| 3100 chained = call_chained_handler(actp, sig, siginfo, context); | |
| 3101 } | |
| 3102 } | |
| 3103 return chained; | |
| 3104 } | |
| 3105 | |
| 3106 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { | |
| 3107 if ((( (unsigned int)1 << sig ) & sigs) != 0) { | |
| 3108 return &sigact[sig]; | |
| 3109 } | |
| 3110 return NULL; | |
| 3111 } | |
| 3112 | |
| 3113 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { | |
| 3114 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); | |
| 3115 sigact[sig] = oldAct; | |
| 3116 sigs |= (unsigned int)1 << sig; | |
| 3117 } | |
| 3118 | |
| 3119 // for diagnostic | |
| 3120 int os::Linux::sigflags[MAXSIGNUM]; | |
| 3121 | |
| 3122 int os::Linux::get_our_sigflags(int sig) { | |
| 3123 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); | |
| 3124 return sigflags[sig]; | |
| 3125 } | |
| 3126 | |
| 3127 void os::Linux::set_our_sigflags(int sig, int flags) { | |
| 3128 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); | |
| 3129 sigflags[sig] = flags; | |
| 3130 } | |
| 3131 | |
| 3132 void os::Linux::set_signal_handler(int sig, bool set_installed) { | |
| 3133 // Check for overwrite. | |
| 3134 struct sigaction oldAct; | |
| 3135 sigaction(sig, (struct sigaction*)NULL, &oldAct); | |
| 3136 | |
| 3137 void* oldhand = oldAct.sa_sigaction | |
| 3138 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) | |
| 3139 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); | |
| 3140 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && | |
| 3141 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && | |
| 3142 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { | |
| 3143 if (AllowUserSignalHandlers || !set_installed) { | |
| 3144 // Do not overwrite; user takes responsibility to forward to us. | |
| 3145 return; | |
| 3146 } else if (UseSignalChaining) { | |
| 3147 // save the old handler in jvm | |
| 3148 save_preinstalled_handler(sig, oldAct); | |
| 3149 // libjsig also interposes the sigaction() call below and saves the | |
| 3150 // old sigaction on it own. | |
| 3151 } else { | |
| 3152 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); | |
| 3153 } | |
| 3154 } | |
| 3155 | |
| 3156 struct sigaction sigAct; | |
| 3157 sigfillset(&(sigAct.sa_mask)); | |
| 3158 sigAct.sa_handler = SIG_DFL; | |
| 3159 if (!set_installed) { | |
| 3160 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; | |
| 3161 } else { | |
| 3162 sigAct.sa_sigaction = signalHandler; | |
| 3163 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; | |
| 3164 } | |
| 3165 // Save flags, which are set by ours | |
| 3166 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); | |
| 3167 sigflags[sig] = sigAct.sa_flags; | |
| 3168 | |
| 3169 int ret = sigaction(sig, &sigAct, &oldAct); | |
| 3170 assert(ret == 0, "check"); | |
| 3171 | |
| 3172 void* oldhand2 = oldAct.sa_sigaction | |
| 3173 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) | |
| 3174 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); | |
| 3175 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); | |
| 3176 } | |
| 3177 | |
| 3178 // install signal handlers for signals that HotSpot needs to | |
| 3179 // handle in order to support Java-level exception handling. | |
| 3180 | |
| 3181 void os::Linux::install_signal_handlers() { | |
| 3182 if (!signal_handlers_are_installed) { | |
| 3183 signal_handlers_are_installed = true; | |
| 3184 | |
| 3185 // signal-chaining | |
| 3186 typedef void (*signal_setting_t)(); | |
| 3187 signal_setting_t begin_signal_setting = NULL; | |
| 3188 signal_setting_t end_signal_setting = NULL; | |
| 3189 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, | |
| 3190 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); | |
| 3191 if (begin_signal_setting != NULL) { | |
| 3192 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, | |
| 3193 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); | |
| 3194 get_signal_action = CAST_TO_FN_PTR(get_signal_t, | |
| 3195 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); | |
| 3196 libjsig_is_loaded = true; | |
| 3197 assert(UseSignalChaining, "should enable signal-chaining"); | |
| 3198 } | |
| 3199 if (libjsig_is_loaded) { | |
| 3200 // Tell libjsig jvm is setting signal handlers | |
| 3201 (*begin_signal_setting)(); | |
| 3202 } | |
| 3203 | |
| 3204 set_signal_handler(SIGSEGV, true); | |
| 3205 set_signal_handler(SIGPIPE, true); | |
| 3206 set_signal_handler(SIGBUS, true); | |
| 3207 set_signal_handler(SIGILL, true); | |
| 3208 set_signal_handler(SIGFPE, true); | |
| 3209 set_signal_handler(SIGXFSZ, true); | |
| 3210 | |
| 3211 if (libjsig_is_loaded) { | |
| 3212 // Tell libjsig jvm finishes setting signal handlers | |
| 3213 (*end_signal_setting)(); | |
| 3214 } | |
| 3215 | |
| 3216 // We don't activate signal checker if libjsig is in place, we trust ourselves | |
| 3217 // and if UserSignalHandler is installed all bets are off | |
| 3218 if (CheckJNICalls) { | |
| 3219 if (libjsig_is_loaded) { | |
| 3220 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); | |
| 3221 check_signals = false; | |
| 3222 } | |
| 3223 if (AllowUserSignalHandlers) { | |
| 3224 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); | |
| 3225 check_signals = false; | |
| 3226 } | |
| 3227 } | |
| 3228 } | |
| 3229 } | |
| 3230 | |
| 3231 // This is the fastest way to get thread cpu time on Linux. | |
| 3232 // Returns cpu time (user+sys) for any thread, not only for current. | |
| 3233 // POSIX compliant clocks are implemented in the kernels 2.6.16+. | |
| 3234 // It might work on 2.6.10+ with a special kernel/glibc patch. | |
| 3235 // For reference, please, see IEEE Std 1003.1-2004: | |
| 3236 // http://www.unix.org/single_unix_specification | |
| 3237 | |
| 3238 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { | |
| 3239 struct timespec tp; | |
| 3240 int rc = os::Linux::clock_gettime(clockid, &tp); | |
| 3241 assert(rc == 0, "clock_gettime is expected to return 0 code"); | |
| 3242 | |
| 3243 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; | |
| 3244 } | |
| 3245 | |
| 3246 ///// | |
| 3247 // glibc on Linux platform uses non-documented flag | |
| 3248 // to indicate, that some special sort of signal | |
| 3249 // trampoline is used. | |
| 3250 // We will never set this flag, and we should | |
| 3251 // ignore this flag in our diagnostic | |
| 3252 #ifdef SIGNIFICANT_SIGNAL_MASK | |
| 3253 #undef SIGNIFICANT_SIGNAL_MASK | |
| 3254 #endif | |
| 3255 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) | |
| 3256 | |
| 3257 static const char* get_signal_handler_name(address handler, | |
| 3258 char* buf, int buflen) { | |
| 3259 int offset; | |
| 3260 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); | |
| 3261 if (found) { | |
| 3262 // skip directory names | |
| 3263 const char *p1, *p2; | |
| 3264 p1 = buf; | |
| 3265 size_t len = strlen(os::file_separator()); | |
| 3266 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; | |
| 3267 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); | |
| 3268 } else { | |
| 3269 jio_snprintf(buf, buflen, PTR_FORMAT, handler); | |
| 3270 } | |
| 3271 return buf; | |
| 3272 } | |
| 3273 | |
| 3274 static void print_signal_handler(outputStream* st, int sig, | |
| 3275 char* buf, size_t buflen) { | |
| 3276 struct sigaction sa; | |
| 3277 | |
| 3278 sigaction(sig, NULL, &sa); | |
| 3279 | |
| 3280 // See comment for SIGNIFICANT_SIGNAL_MASK define | |
| 3281 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; | |
| 3282 | |
| 3283 st->print("%s: ", os::exception_name(sig, buf, buflen)); | |
| 3284 | |
| 3285 address handler = (sa.sa_flags & SA_SIGINFO) | |
| 3286 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) | |
| 3287 : CAST_FROM_FN_PTR(address, sa.sa_handler); | |
| 3288 | |
| 3289 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { | |
| 3290 st->print("SIG_DFL"); | |
| 3291 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { | |
| 3292 st->print("SIG_IGN"); | |
| 3293 } else { | |
| 3294 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); | |
| 3295 } | |
| 3296 | |
| 3297 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); | |
| 3298 | |
| 3299 address rh = VMError::get_resetted_sighandler(sig); | |
| 3300 // May be, handler was resetted by VMError? | |
| 3301 if(rh != NULL) { | |
| 3302 handler = rh; | |
| 3303 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; | |
| 3304 } | |
| 3305 | |
| 3306 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); | |
| 3307 | |
| 3308 // Check: is it our handler? | |
| 3309 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || | |
| 3310 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { | |
| 3311 // It is our signal handler | |
| 3312 // check for flags, reset system-used one! | |
| 3313 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { | |
| 3314 st->print( | |
| 3315 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", | |
| 3316 os::Linux::get_our_sigflags(sig)); | |
| 3317 } | |
| 3318 } | |
| 3319 st->cr(); | |
| 3320 } | |
| 3321 | |
| 3322 | |
| 3323 #define DO_SIGNAL_CHECK(sig) \ | |
| 3324 if (!sigismember(&check_signal_done, sig)) \ | |
| 3325 os::Linux::check_signal_handler(sig) | |
| 3326 | |
| 3327 // This method is a periodic task to check for misbehaving JNI applications | |
| 3328 // under CheckJNI, we can add any periodic checks here | |
| 3329 | |
| 3330 void os::run_periodic_checks() { | |
| 3331 | |
| 3332 if (check_signals == false) return; | |
| 3333 | |
| 3334 // SEGV and BUS if overridden could potentially prevent | |
| 3335 // generation of hs*.log in the event of a crash, debugging | |
| 3336 // such a case can be very challenging, so we absolutely | |
| 3337 // check the following for a good measure: | |
| 3338 DO_SIGNAL_CHECK(SIGSEGV); | |
| 3339 DO_SIGNAL_CHECK(SIGILL); | |
| 3340 DO_SIGNAL_CHECK(SIGFPE); | |
| 3341 DO_SIGNAL_CHECK(SIGBUS); | |
| 3342 DO_SIGNAL_CHECK(SIGPIPE); | |
| 3343 DO_SIGNAL_CHECK(SIGXFSZ); | |
| 3344 | |
| 3345 | |
| 3346 // ReduceSignalUsage allows the user to override these handlers | |
| 3347 // see comments at the very top and jvm_solaris.h | |
| 3348 if (!ReduceSignalUsage) { | |
| 3349 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); | |
| 3350 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); | |
| 3351 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); | |
| 3352 DO_SIGNAL_CHECK(BREAK_SIGNAL); | |
| 3353 } | |
| 3354 | |
| 3355 DO_SIGNAL_CHECK(SR_signum); | |
| 3356 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); | |
| 3357 } | |
| 3358 | |
| 3359 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); | |
| 3360 | |
| 3361 static os_sigaction_t os_sigaction = NULL; | |
| 3362 | |
| 3363 void os::Linux::check_signal_handler(int sig) { | |
| 3364 char buf[O_BUFLEN]; | |
| 3365 address jvmHandler = NULL; | |
| 3366 | |
| 3367 | |
| 3368 struct sigaction act; | |
| 3369 if (os_sigaction == NULL) { | |
| 3370 // only trust the default sigaction, in case it has been interposed | |
| 3371 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); | |
| 3372 if (os_sigaction == NULL) return; | |
| 3373 } | |
| 3374 | |
| 3375 os_sigaction(sig, (struct sigaction*)NULL, &act); | |
| 3376 | |
| 3377 | |
| 3378 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; | |
| 3379 | |
| 3380 address thisHandler = (act.sa_flags & SA_SIGINFO) | |
| 3381 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) | |
| 3382 : CAST_FROM_FN_PTR(address, act.sa_handler) ; | |
| 3383 | |
| 3384 | |
| 3385 switch(sig) { | |
| 3386 case SIGSEGV: | |
| 3387 case SIGBUS: | |
| 3388 case SIGFPE: | |
| 3389 case SIGPIPE: | |
| 3390 case SIGILL: | |
| 3391 case SIGXFSZ: | |
| 3392 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); | |
| 3393 break; | |
| 3394 | |
| 3395 case SHUTDOWN1_SIGNAL: | |
| 3396 case SHUTDOWN2_SIGNAL: | |
| 3397 case SHUTDOWN3_SIGNAL: | |
| 3398 case BREAK_SIGNAL: | |
| 3399 jvmHandler = (address)user_handler(); | |
| 3400 break; | |
| 3401 | |
| 3402 case INTERRUPT_SIGNAL: | |
| 3403 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); | |
| 3404 break; | |
| 3405 | |
| 3406 default: | |
| 3407 if (sig == SR_signum) { | |
| 3408 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); | |
| 3409 } else { | |
| 3410 return; | |
| 3411 } | |
| 3412 break; | |
| 3413 } | |
| 3414 | |
| 3415 if (thisHandler != jvmHandler) { | |
| 3416 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); | |
| 3417 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); | |
| 3418 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); | |
| 3419 // No need to check this sig any longer | |
| 3420 sigaddset(&check_signal_done, sig); | |
| 3421 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { | |
| 3422 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); | |
| 3423 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); | |
| 3424 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); | |
| 3425 // No need to check this sig any longer | |
| 3426 sigaddset(&check_signal_done, sig); | |
| 3427 } | |
| 3428 | |
| 3429 // Dump all the signal | |
| 3430 if (sigismember(&check_signal_done, sig)) { | |
| 3431 print_signal_handlers(tty, buf, O_BUFLEN); | |
| 3432 } | |
| 3433 } | |
| 3434 | |
| 3435 extern void report_error(char* file_name, int line_no, char* title, char* format, ...); | |
| 3436 | |
| 3437 extern bool signal_name(int signo, char* buf, size_t len); | |
| 3438 | |
| 3439 const char* os::exception_name(int exception_code, char* buf, size_t size) { | |
| 3440 if (0 < exception_code && exception_code <= SIGRTMAX) { | |
| 3441 // signal | |
| 3442 if (!signal_name(exception_code, buf, size)) { | |
| 3443 jio_snprintf(buf, size, "SIG%d", exception_code); | |
| 3444 } | |
| 3445 return buf; | |
| 3446 } else { | |
| 3447 return NULL; | |
| 3448 } | |
| 3449 } | |
| 3450 | |
| 3451 // this is called _before_ the most of global arguments have been parsed | |
| 3452 void os::init(void) { | |
| 3453 char dummy; /* used to get a guess on initial stack address */ | |
| 3454 // first_hrtime = gethrtime(); | |
| 3455 | |
| 3456 // With LinuxThreads the JavaMain thread pid (primordial thread) | |
| 3457 // is different than the pid of the java launcher thread. | |
| 3458 // So, on Linux, the launcher thread pid is passed to the VM | |
| 3459 // via the sun.java.launcher.pid property. | |
| 3460 // Use this property instead of getpid() if it was correctly passed. | |
| 3461 // See bug 6351349. | |
| 3462 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); | |
| 3463 | |
| 3464 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); | |
| 3465 | |
| 3466 clock_tics_per_sec = sysconf(_SC_CLK_TCK); | |
| 3467 | |
| 3468 init_random(1234567); | |
| 3469 | |
| 3470 ThreadCritical::initialize(); | |
| 3471 | |
| 3472 Linux::set_page_size(sysconf(_SC_PAGESIZE)); | |
| 3473 if (Linux::page_size() == -1) { | |
| 3474 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); | |
| 3475 } | |
| 3476 init_page_sizes((size_t) Linux::page_size()); | |
| 3477 | |
| 3478 Linux::initialize_system_info(); | |
| 3479 | |
| 3480 // main_thread points to the aboriginal thread | |
| 3481 Linux::_main_thread = pthread_self(); | |
| 3482 | |
| 3483 Linux::clock_init(); | |
| 3484 initial_time_count = os::elapsed_counter(); | |
| 3485 } | |
| 3486 | |
| 3487 // To install functions for atexit system call | |
| 3488 extern "C" { | |
| 3489 static void perfMemory_exit_helper() { | |
| 3490 perfMemory_exit(); | |
| 3491 } | |
| 3492 } | |
| 3493 | |
| 3494 // this is called _after_ the global arguments have been parsed | |
| 3495 jint os::init_2(void) | |
| 3496 { | |
| 3497 Linux::fast_thread_clock_init(); | |
| 3498 | |
| 3499 // Allocate a single page and mark it as readable for safepoint polling | |
| 3500 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
| 3501 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); | |
| 3502 | |
| 3503 os::set_polling_page( polling_page ); | |
| 3504 | |
| 3505 #ifndef PRODUCT | |
| 3506 if(Verbose && PrintMiscellaneous) | |
| 3507 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); | |
| 3508 #endif | |
| 3509 | |
| 3510 if (!UseMembar) { | |
| 3511 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
| 3512 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); | |
| 3513 os::set_memory_serialize_page( mem_serialize_page ); | |
| 3514 | |
| 3515 #ifndef PRODUCT | |
| 3516 if(Verbose && PrintMiscellaneous) | |
| 3517 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); | |
| 3518 #endif | |
| 3519 } | |
| 3520 | |
| 3521 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); | |
| 3522 | |
| 3523 // initialize suspend/resume support - must do this before signal_sets_init() | |
| 3524 if (SR_initialize() != 0) { | |
| 3525 perror("SR_initialize failed"); | |
| 3526 return JNI_ERR; | |
| 3527 } | |
| 3528 | |
| 3529 Linux::signal_sets_init(); | |
| 3530 Linux::install_signal_handlers(); | |
| 3531 | |
| 3532 size_t threadStackSizeInBytes = ThreadStackSize * K; | |
| 3533 if (threadStackSizeInBytes != 0 && | |
| 3534 threadStackSizeInBytes < Linux::min_stack_allowed) { | |
| 3535 tty->print_cr("\nThe stack size specified is too small, " | |
| 3536 "Specify at least %dk", | |
| 3537 Linux::min_stack_allowed / K); | |
| 3538 return JNI_ERR; | |
| 3539 } | |
| 3540 | |
| 3541 // Make the stack size a multiple of the page size so that | |
| 3542 // the yellow/red zones can be guarded. | |
| 3543 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, | |
| 3544 vm_page_size())); | |
| 3545 | |
| 3546 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); | |
| 3547 | |
| 3548 Linux::libpthread_init(); | |
| 3549 if (PrintMiscellaneous && (Verbose || WizardMode)) { | |
| 3550 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", | |
| 3551 Linux::glibc_version(), Linux::libpthread_version(), | |
| 3552 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); | |
| 3553 } | |
| 3554 | |
| 3555 if (MaxFDLimit) { | |
| 3556 // set the number of file descriptors to max. print out error | |
| 3557 // if getrlimit/setrlimit fails but continue regardless. | |
| 3558 struct rlimit nbr_files; | |
| 3559 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); | |
| 3560 if (status != 0) { | |
| 3561 if (PrintMiscellaneous && (Verbose || WizardMode)) | |
| 3562 perror("os::init_2 getrlimit failed"); | |
| 3563 } else { | |
| 3564 nbr_files.rlim_cur = nbr_files.rlim_max; | |
| 3565 status = setrlimit(RLIMIT_NOFILE, &nbr_files); | |
| 3566 if (status != 0) { | |
| 3567 if (PrintMiscellaneous && (Verbose || WizardMode)) | |
| 3568 perror("os::init_2 setrlimit failed"); | |
| 3569 } | |
| 3570 } | |
| 3571 } | |
| 3572 | |
| 3573 // Initialize lock used to serialize thread creation (see os::create_thread) | |
| 3574 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); | |
| 3575 | |
| 3576 // Initialize HPI. | |
| 3577 jint hpi_result = hpi::initialize(); | |
| 3578 if (hpi_result != JNI_OK) { | |
| 3579 tty->print_cr("There was an error trying to initialize the HPI library."); | |
| 3580 return hpi_result; | |
| 3581 } | |
| 3582 | |
| 3583 // at-exit methods are called in the reverse order of their registration. | |
| 3584 // atexit functions are called on return from main or as a result of a | |
| 3585 // call to exit(3C). There can be only 32 of these functions registered | |
| 3586 // and atexit() does not set errno. | |
| 3587 | |
| 3588 if (PerfAllowAtExitRegistration) { | |
| 3589 // only register atexit functions if PerfAllowAtExitRegistration is set. | |
| 3590 // atexit functions can be delayed until process exit time, which | |
| 3591 // can be problematic for embedded VM situations. Embedded VMs should | |
| 3592 // call DestroyJavaVM() to assure that VM resources are released. | |
| 3593 | |
| 3594 // note: perfMemory_exit_helper atexit function may be removed in | |
| 3595 // the future if the appropriate cleanup code can be added to the | |
| 3596 // VM_Exit VMOperation's doit method. | |
| 3597 if (atexit(perfMemory_exit_helper) != 0) { | |
| 3598 warning("os::init2 atexit(perfMemory_exit_helper) failed"); | |
| 3599 } | |
| 3600 } | |
| 3601 | |
| 3602 // initialize thread priority policy | |
| 3603 prio_init(); | |
| 3604 | |
| 3605 return JNI_OK; | |
| 3606 } | |
| 3607 | |
| 3608 // Mark the polling page as unreadable | |
| 3609 void os::make_polling_page_unreadable(void) { | |
| 3610 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) | |
| 3611 fatal("Could not disable polling page"); | |
| 3612 }; | |
| 3613 | |
| 3614 // Mark the polling page as readable | |
| 3615 void os::make_polling_page_readable(void) { | |
| 3616 if( !protect_memory((char *)_polling_page, Linux::page_size()) ) | |
| 3617 fatal("Could not enable polling page"); | |
| 3618 }; | |
| 3619 | |
| 3620 int os::active_processor_count() { | |
| 3621 // Linux doesn't yet have a (official) notion of processor sets, | |
| 3622 // so just return the number of online processors. | |
| 3623 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); | |
| 3624 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); | |
| 3625 return online_cpus; | |
| 3626 } | |
| 3627 | |
| 3628 bool os::distribute_processes(uint length, uint* distribution) { | |
| 3629 // Not yet implemented. | |
| 3630 return false; | |
| 3631 } | |
| 3632 | |
| 3633 bool os::bind_to_processor(uint processor_id) { | |
| 3634 // Not yet implemented. | |
| 3635 return false; | |
| 3636 } | |
| 3637 | |
| 3638 /// | |
| 3639 | |
| 3640 // Suspends the target using the signal mechanism and then grabs the PC before | |
| 3641 // resuming the target. Used by the flat-profiler only | |
| 3642 ExtendedPC os::get_thread_pc(Thread* thread) { | |
| 3643 // Make sure that it is called by the watcher for the VMThread | |
| 3644 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); | |
| 3645 assert(thread->is_VM_thread(), "Can only be called for VMThread"); | |
| 3646 | |
| 3647 ExtendedPC epc; | |
| 3648 | |
| 3649 OSThread* osthread = thread->osthread(); | |
| 3650 if (do_suspend(osthread)) { | |
| 3651 if (osthread->ucontext() != NULL) { | |
| 3652 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); | |
| 3653 } else { | |
| 3654 // NULL context is unexpected, double-check this is the VMThread | |
| 3655 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); | |
| 3656 } | |
| 3657 do_resume(osthread); | |
| 3658 } | |
| 3659 // failure means pthread_kill failed for some reason - arguably this is | |
| 3660 // a fatal problem, but such problems are ignored elsewhere | |
| 3661 | |
| 3662 return epc; | |
| 3663 } | |
| 3664 | |
| 3665 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) | |
| 3666 { | |
| 3667 if (is_NPTL()) { | |
| 3668 return pthread_cond_timedwait(_cond, _mutex, _abstime); | |
| 3669 } else { | |
| 3670 #ifndef IA64 | |
| 3671 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control | |
| 3672 // word back to default 64bit precision if condvar is signaled. Java | |
| 3673 // wants 53bit precision. Save and restore current value. | |
| 3674 int fpu = get_fpu_control_word(); | |
| 3675 #endif // IA64 | |
| 3676 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); | |
| 3677 #ifndef IA64 | |
| 3678 set_fpu_control_word(fpu); | |
| 3679 #endif // IA64 | |
| 3680 return status; | |
| 3681 } | |
| 3682 } | |
| 3683 | |
| 3684 //////////////////////////////////////////////////////////////////////////////// | |
| 3685 // debug support | |
| 3686 | |
| 3687 #ifndef PRODUCT | |
| 3688 static address same_page(address x, address y) { | |
| 3689 int page_bits = -os::vm_page_size(); | |
| 3690 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) | |
| 3691 return x; | |
| 3692 else if (x > y) | |
| 3693 return (address)(intptr_t(y) | ~page_bits) + 1; | |
| 3694 else | |
| 3695 return (address)(intptr_t(y) & page_bits); | |
| 3696 } | |
| 3697 | |
| 3698 bool os::find(address addr) { | |
| 3699 Dl_info dlinfo; | |
| 3700 memset(&dlinfo, 0, sizeof(dlinfo)); | |
| 3701 if (dladdr(addr, &dlinfo)) { | |
| 3702 tty->print(PTR_FORMAT ": ", addr); | |
| 3703 if (dlinfo.dli_sname != NULL) { | |
| 3704 tty->print("%s+%#x", dlinfo.dli_sname, | |
| 3705 addr - (intptr_t)dlinfo.dli_saddr); | |
| 3706 } else if (dlinfo.dli_fname) { | |
| 3707 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); | |
| 3708 } else { | |
| 3709 tty->print("<absolute address>"); | |
| 3710 } | |
| 3711 if (dlinfo.dli_fname) { | |
| 3712 tty->print(" in %s", dlinfo.dli_fname); | |
| 3713 } | |
| 3714 if (dlinfo.dli_fbase) { | |
| 3715 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); | |
| 3716 } | |
| 3717 tty->cr(); | |
| 3718 | |
| 3719 if (Verbose) { | |
| 3720 // decode some bytes around the PC | |
| 3721 address begin = same_page(addr-40, addr); | |
| 3722 address end = same_page(addr+40, addr); | |
| 3723 address lowest = (address) dlinfo.dli_sname; | |
| 3724 if (!lowest) lowest = (address) dlinfo.dli_fbase; | |
| 3725 if (begin < lowest) begin = lowest; | |
| 3726 Dl_info dlinfo2; | |
| 3727 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr | |
| 3728 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) | |
| 3729 end = (address) dlinfo2.dli_saddr; | |
| 3730 Disassembler::decode(begin, end); | |
| 3731 } | |
| 3732 return true; | |
| 3733 } | |
| 3734 return false; | |
| 3735 } | |
| 3736 | |
| 3737 #endif | |
| 3738 | |
| 3739 //////////////////////////////////////////////////////////////////////////////// | |
| 3740 // misc | |
| 3741 | |
| 3742 // This does not do anything on Linux. This is basically a hook for being | |
| 3743 // able to use structured exception handling (thread-local exception filters) | |
| 3744 // on, e.g., Win32. | |
| 3745 void | |
| 3746 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, | |
| 3747 JavaCallArguments* args, Thread* thread) { | |
| 3748 f(value, method, args, thread); | |
| 3749 } | |
| 3750 | |
| 3751 void os::print_statistics() { | |
| 3752 } | |
| 3753 | |
| 3754 int os::message_box(const char* title, const char* message) { | |
| 3755 int i; | |
| 3756 fdStream err(defaultStream::error_fd()); | |
| 3757 for (i = 0; i < 78; i++) err.print_raw("="); | |
| 3758 err.cr(); | |
| 3759 err.print_raw_cr(title); | |
| 3760 for (i = 0; i < 78; i++) err.print_raw("-"); | |
| 3761 err.cr(); | |
| 3762 err.print_raw_cr(message); | |
| 3763 for (i = 0; i < 78; i++) err.print_raw("="); | |
| 3764 err.cr(); | |
| 3765 | |
| 3766 char buf[16]; | |
| 3767 // Prevent process from exiting upon "read error" without consuming all CPU | |
| 3768 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } | |
| 3769 | |
| 3770 return buf[0] == 'y' || buf[0] == 'Y'; | |
| 3771 } | |
| 3772 | |
| 3773 int os::stat(const char *path, struct stat *sbuf) { | |
| 3774 char pathbuf[MAX_PATH]; | |
| 3775 if (strlen(path) > MAX_PATH - 1) { | |
| 3776 errno = ENAMETOOLONG; | |
| 3777 return -1; | |
| 3778 } | |
| 3779 hpi::native_path(strcpy(pathbuf, path)); | |
| 3780 return ::stat(pathbuf, sbuf); | |
| 3781 } | |
| 3782 | |
| 3783 bool os::check_heap(bool force) { | |
| 3784 return true; | |
| 3785 } | |
| 3786 | |
| 3787 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { | |
| 3788 return ::vsnprintf(buf, count, format, args); | |
| 3789 } | |
| 3790 | |
| 3791 // Is a (classpath) directory empty? | |
| 3792 bool os::dir_is_empty(const char* path) { | |
| 3793 DIR *dir = NULL; | |
| 3794 struct dirent *ptr; | |
| 3795 | |
| 3796 dir = opendir(path); | |
| 3797 if (dir == NULL) return true; | |
| 3798 | |
| 3799 /* Scan the directory */ | |
| 3800 bool result = true; | |
| 3801 char buf[sizeof(struct dirent) + MAX_PATH]; | |
| 3802 while (result && (ptr = ::readdir(dir)) != NULL) { | |
| 3803 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { | |
| 3804 result = false; | |
| 3805 } | |
| 3806 } | |
| 3807 closedir(dir); | |
| 3808 return result; | |
| 3809 } | |
| 3810 | |
| 3811 // create binary file, rewriting existing file if required | |
| 3812 int os::create_binary_file(const char* path, bool rewrite_existing) { | |
| 3813 int oflags = O_WRONLY | O_CREAT; | |
| 3814 if (!rewrite_existing) { | |
| 3815 oflags |= O_EXCL; | |
| 3816 } | |
| 3817 return ::open64(path, oflags, S_IREAD | S_IWRITE); | |
| 3818 } | |
| 3819 | |
| 3820 // return current position of file pointer | |
| 3821 jlong os::current_file_offset(int fd) { | |
| 3822 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); | |
| 3823 } | |
| 3824 | |
| 3825 // move file pointer to the specified offset | |
| 3826 jlong os::seek_to_file_offset(int fd, jlong offset) { | |
| 3827 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); | |
| 3828 } | |
| 3829 | |
| 3830 // Map a block of memory. | |
| 3831 char* os::map_memory(int fd, const char* file_name, size_t file_offset, | |
| 3832 char *addr, size_t bytes, bool read_only, | |
| 3833 bool allow_exec) { | |
| 3834 int prot; | |
| 3835 int flags; | |
| 3836 | |
| 3837 if (read_only) { | |
| 3838 prot = PROT_READ; | |
| 3839 flags = MAP_SHARED; | |
| 3840 } else { | |
| 3841 prot = PROT_READ | PROT_WRITE; | |
| 3842 flags = MAP_PRIVATE; | |
| 3843 } | |
| 3844 | |
| 3845 if (allow_exec) { | |
| 3846 prot |= PROT_EXEC; | |
| 3847 } | |
| 3848 | |
| 3849 if (addr != NULL) { | |
| 3850 flags |= MAP_FIXED; | |
| 3851 } | |
| 3852 | |
| 3853 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, | |
| 3854 fd, file_offset); | |
| 3855 if (mapped_address == MAP_FAILED) { | |
| 3856 return NULL; | |
| 3857 } | |
| 3858 return mapped_address; | |
| 3859 } | |
| 3860 | |
| 3861 | |
| 3862 // Remap a block of memory. | |
| 3863 char* os::remap_memory(int fd, const char* file_name, size_t file_offset, | |
| 3864 char *addr, size_t bytes, bool read_only, | |
| 3865 bool allow_exec) { | |
| 3866 // same as map_memory() on this OS | |
| 3867 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, | |
| 3868 allow_exec); | |
| 3869 } | |
| 3870 | |
| 3871 | |
| 3872 // Unmap a block of memory. | |
| 3873 bool os::unmap_memory(char* addr, size_t bytes) { | |
| 3874 return munmap(addr, bytes) == 0; | |
| 3875 } | |
| 3876 | |
| 3877 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); | |
| 3878 | |
| 3879 static clockid_t thread_cpu_clockid(Thread* thread) { | |
| 3880 pthread_t tid = thread->osthread()->pthread_id(); | |
| 3881 clockid_t clockid; | |
| 3882 | |
| 3883 // Get thread clockid | |
| 3884 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); | |
| 3885 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); | |
| 3886 return clockid; | |
| 3887 } | |
| 3888 | |
| 3889 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) | |
| 3890 // are used by JVM M&M and JVMTI to get user+sys or user CPU time | |
| 3891 // of a thread. | |
| 3892 // | |
| 3893 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns | |
| 3894 // the fast estimate available on the platform. | |
| 3895 | |
| 3896 jlong os::current_thread_cpu_time() { | |
| 3897 if (os::Linux::supports_fast_thread_cpu_time()) { | |
| 3898 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); | |
| 3899 } else { | |
| 3900 // return user + sys since the cost is the same | |
| 3901 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); | |
| 3902 } | |
| 3903 } | |
| 3904 | |
| 3905 jlong os::thread_cpu_time(Thread* thread) { | |
| 3906 // consistent with what current_thread_cpu_time() returns | |
| 3907 if (os::Linux::supports_fast_thread_cpu_time()) { | |
| 3908 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); | |
| 3909 } else { | |
| 3910 return slow_thread_cpu_time(thread, true /* user + sys */); | |
| 3911 } | |
| 3912 } | |
| 3913 | |
| 3914 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { | |
| 3915 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { | |
| 3916 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); | |
| 3917 } else { | |
| 3918 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); | |
| 3919 } | |
| 3920 } | |
| 3921 | |
| 3922 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { | |
| 3923 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { | |
| 3924 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); | |
| 3925 } else { | |
| 3926 return slow_thread_cpu_time(thread, user_sys_cpu_time); | |
| 3927 } | |
| 3928 } | |
| 3929 | |
| 3930 // | |
| 3931 // -1 on error. | |
| 3932 // | |
| 3933 | |
| 3934 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { | |
| 3935 static bool proc_pid_cpu_avail = true; | |
| 3936 static bool proc_task_unchecked = true; | |
| 3937 static const char *proc_stat_path = "/proc/%d/stat"; | |
| 3938 pid_t tid = thread->osthread()->thread_id(); | |
| 3939 int i; | |
| 3940 char *s; | |
| 3941 char stat[2048]; | |
| 3942 int statlen; | |
| 3943 char proc_name[64]; | |
| 3944 int count; | |
| 3945 long sys_time, user_time; | |
| 3946 char string[64]; | |
| 3947 int idummy; | |
| 3948 long ldummy; | |
| 3949 FILE *fp; | |
| 3950 | |
| 3951 // We first try accessing /proc/<pid>/cpu since this is faster to | |
| 3952 // process. If this file is not present (linux kernels 2.5 and above) | |
| 3953 // then we open /proc/<pid>/stat. | |
| 3954 if ( proc_pid_cpu_avail ) { | |
| 3955 sprintf(proc_name, "/proc/%d/cpu", tid); | |
| 3956 fp = fopen(proc_name, "r"); | |
| 3957 if ( fp != NULL ) { | |
| 3958 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); | |
| 3959 fclose(fp); | |
| 3960 if ( count != 3 ) return -1; | |
| 3961 | |
| 3962 if (user_sys_cpu_time) { | |
| 3963 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); | |
| 3964 } else { | |
| 3965 return (jlong)user_time * (1000000000 / clock_tics_per_sec); | |
| 3966 } | |
| 3967 } | |
| 3968 else proc_pid_cpu_avail = false; | |
| 3969 } | |
| 3970 | |
| 3971 // The /proc/<tid>/stat aggregates per-process usage on | |
| 3972 // new Linux kernels 2.6+ where NPTL is supported. | |
| 3973 // The /proc/self/task/<tid>/stat still has the per-thread usage. | |
| 3974 // See bug 6328462. | |
| 3975 // There can be no directory /proc/self/task on kernels 2.4 with NPTL | |
| 3976 // and possibly in some other cases, so we check its availability. | |
| 3977 if (proc_task_unchecked && os::Linux::is_NPTL()) { | |
| 3978 // This is executed only once | |
| 3979 proc_task_unchecked = false; | |
| 3980 fp = fopen("/proc/self/task", "r"); | |
| 3981 if (fp != NULL) { | |
| 3982 proc_stat_path = "/proc/self/task/%d/stat"; | |
| 3983 fclose(fp); | |
| 3984 } | |
| 3985 } | |
| 3986 | |
| 3987 sprintf(proc_name, proc_stat_path, tid); | |
| 3988 fp = fopen(proc_name, "r"); | |
| 3989 if ( fp == NULL ) return -1; | |
| 3990 statlen = fread(stat, 1, 2047, fp); | |
| 3991 stat[statlen] = '\0'; | |
| 3992 fclose(fp); | |
| 3993 | |
| 3994 // Skip pid and the command string. Note that we could be dealing with | |
| 3995 // weird command names, e.g. user could decide to rename java launcher | |
| 3996 // to "java 1.4.2 :)", then the stat file would look like | |
| 3997 // 1234 (java 1.4.2 :)) R ... ... | |
| 3998 // We don't really need to know the command string, just find the last | |
| 3999 // occurrence of ")" and then start parsing from there. See bug 4726580. | |
| 4000 s = strrchr(stat, ')'); | |
| 4001 i = 0; | |
| 4002 if (s == NULL ) return -1; | |
| 4003 | |
| 4004 // Skip blank chars | |
| 4005 do s++; while (isspace(*s)); | |
| 4006 | |
| 4007 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", | |
| 4008 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, | |
| 4009 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, | |
| 4010 &user_time, &sys_time); | |
| 4011 if ( count != 13 ) return -1; | |
| 4012 if (user_sys_cpu_time) { | |
| 4013 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); | |
| 4014 } else { | |
| 4015 return (jlong)user_time * (1000000000 / clock_tics_per_sec); | |
| 4016 } | |
| 4017 } | |
| 4018 | |
| 4019 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { | |
| 4020 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits | |
| 4021 info_ptr->may_skip_backward = false; // elapsed time not wall time | |
| 4022 info_ptr->may_skip_forward = false; // elapsed time not wall time | |
| 4023 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned | |
| 4024 } | |
| 4025 | |
| 4026 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { | |
| 4027 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits | |
| 4028 info_ptr->may_skip_backward = false; // elapsed time not wall time | |
| 4029 info_ptr->may_skip_forward = false; // elapsed time not wall time | |
| 4030 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned | |
| 4031 } | |
| 4032 | |
| 4033 bool os::is_thread_cpu_time_supported() { | |
| 4034 return true; | |
| 4035 } | |
| 4036 | |
| 4037 // System loadavg support. Returns -1 if load average cannot be obtained. | |
| 4038 // Linux doesn't yet have a (official) notion of processor sets, | |
| 4039 // so just return the system wide load average. | |
| 4040 int os::loadavg(double loadavg[], int nelem) { | |
| 4041 return ::getloadavg(loadavg, nelem); | |
| 4042 } | |
| 4043 | |
| 4044 void os::pause() { | |
| 4045 char filename[MAX_PATH]; | |
| 4046 if (PauseAtStartupFile && PauseAtStartupFile[0]) { | |
| 4047 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); | |
| 4048 } else { | |
| 4049 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); | |
| 4050 } | |
| 4051 | |
| 4052 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); | |
| 4053 if (fd != -1) { | |
| 4054 struct stat buf; | |
| 4055 close(fd); | |
| 4056 while (::stat(filename, &buf) == 0) { | |
| 4057 (void)::poll(NULL, 0, 100); | |
| 4058 } | |
| 4059 } else { | |
| 4060 jio_fprintf(stderr, | |
| 4061 "Could not open pause file '%s', continuing immediately.\n", filename); | |
| 4062 } | |
| 4063 } | |
| 4064 | |
| 4065 extern "C" { | |
| 4066 | |
| 4067 /** | |
| 4068 * NOTE: the following code is to keep the green threads code | |
| 4069 * in the libjava.so happy. Once the green threads is removed, | |
| 4070 * these code will no longer be needed. | |
| 4071 */ | |
| 4072 int | |
| 4073 jdk_waitpid(pid_t pid, int* status, int options) { | |
| 4074 return waitpid(pid, status, options); | |
| 4075 } | |
| 4076 | |
| 4077 int | |
| 4078 fork1() { | |
| 4079 return fork(); | |
| 4080 } | |
| 4081 | |
| 4082 int | |
| 4083 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { | |
| 4084 return sem_init(sem, pshared, value); | |
| 4085 } | |
| 4086 | |
| 4087 int | |
| 4088 jdk_sem_post(sem_t *sem) { | |
| 4089 return sem_post(sem); | |
| 4090 } | |
| 4091 | |
| 4092 int | |
| 4093 jdk_sem_wait(sem_t *sem) { | |
| 4094 return sem_wait(sem); | |
| 4095 } | |
| 4096 | |
| 4097 int | |
| 4098 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { | |
| 4099 return pthread_sigmask(how , newmask, oldmask); | |
| 4100 } | |
| 4101 | |
| 4102 } | |
| 4103 | |
| 4104 // Refer to the comments in os_solaris.cpp park-unpark. | |
| 4105 // | |
| 4106 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can | |
| 4107 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. | |
| 4108 // For specifics regarding the bug see GLIBC BUGID 261237 : | |
| 4109 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. | |
| 4110 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future | |
| 4111 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar | |
| 4112 // is used. (The simple C test-case provided in the GLIBC bug report manifests the | |
| 4113 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() | |
| 4114 // and monitorenter when we're using 1-0 locking. All those operations may result in | |
| 4115 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version | |
| 4116 // of libpthread avoids the problem, but isn't practical. | |
| 4117 // | |
| 4118 // Possible remedies: | |
| 4119 // | |
| 4120 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. | |
| 4121 // This is palliative and probabilistic, however. If the thread is preempted | |
| 4122 // between the call to compute_abstime() and pthread_cond_timedwait(), more | |
| 4123 // than the minimum period may have passed, and the abstime may be stale (in the | |
| 4124 // past) resultin in a hang. Using this technique reduces the odds of a hang | |
| 4125 // but the JVM is still vulnerable, particularly on heavily loaded systems. | |
| 4126 // | |
| 4127 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead | |
| 4128 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set | |
| 4129 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) | |
| 4130 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant | |
| 4131 // thread. | |
| 4132 // | |
| 4133 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread | |
| 4134 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing | |
| 4135 // a timeout request to the chron thread and then blocking via pthread_cond_wait(). | |
| 4136 // This also works well. In fact it avoids kernel-level scalability impediments | |
| 4137 // on certain platforms that don't handle lots of active pthread_cond_timedwait() | |
| 4138 // timers in a graceful fashion. | |
| 4139 // | |
| 4140 // 4. When the abstime value is in the past it appears that control returns | |
| 4141 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt. | |
| 4142 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we | |
| 4143 // can avoid the problem by reinitializing the condvar -- by cond_destroy() | |
| 4144 // followed by cond_init() -- after all calls to pthread_cond_timedwait(). | |
| 4145 // It may be possible to avoid reinitialization by checking the return | |
| 4146 // value from pthread_cond_timedwait(). In addition to reinitializing the | |
| 4147 // condvar we must establish the invariant that cond_signal() is only called | |
| 4148 // within critical sections protected by the adjunct mutex. This prevents | |
| 4149 // cond_signal() from "seeing" a condvar that's in the midst of being | |
| 4150 // reinitialized or that is corrupt. Sadly, this invariant obviates the | |
| 4151 // desirable signal-after-unlock optimization that avoids futile context switching. | |
| 4152 // | |
| 4153 // I'm also concerned that some versions of NTPL might allocate an auxilliary | |
| 4154 // structure when a condvar is used or initialized. cond_destroy() would | |
| 4155 // release the helper structure. Our reinitialize-after-timedwait fix | |
| 4156 // put excessive stress on malloc/free and locks protecting the c-heap. | |
| 4157 // | |
| 4158 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. | |
| 4159 // It may be possible to refine (4) by checking the kernel and NTPL verisons | |
| 4160 // and only enabling the work-around for vulnerable environments. | |
| 4161 | |
| 4162 // utility to compute the abstime argument to timedwait: | |
| 4163 // millis is the relative timeout time | |
| 4164 // abstime will be the absolute timeout time | |
| 4165 // TODO: replace compute_abstime() with unpackTime() | |
| 4166 | |
| 4167 static struct timespec* compute_abstime(timespec* abstime, jlong millis) { | |
| 4168 if (millis < 0) millis = 0; | |
| 4169 struct timeval now; | |
| 4170 int status = gettimeofday(&now, NULL); | |
| 4171 assert(status == 0, "gettimeofday"); | |
| 4172 jlong seconds = millis / 1000; | |
| 4173 millis %= 1000; | |
| 4174 if (seconds > 50000000) { // see man cond_timedwait(3T) | |
| 4175 seconds = 50000000; | |
| 4176 } | |
| 4177 abstime->tv_sec = now.tv_sec + seconds; | |
| 4178 long usec = now.tv_usec + millis * 1000; | |
| 4179 if (usec >= 1000000) { | |
| 4180 abstime->tv_sec += 1; | |
| 4181 usec -= 1000000; | |
| 4182 } | |
| 4183 abstime->tv_nsec = usec * 1000; | |
| 4184 return abstime; | |
| 4185 } | |
| 4186 | |
| 4187 | |
| 4188 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately. | |
| 4189 // Conceptually TryPark() should be equivalent to park(0). | |
| 4190 | |
| 4191 int os::PlatformEvent::TryPark() { | |
| 4192 for (;;) { | |
| 4193 const int v = _Event ; | |
| 4194 guarantee ((v == 0) || (v == 1), "invariant") ; | |
| 4195 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; | |
| 4196 } | |
| 4197 } | |
| 4198 | |
| 4199 void os::PlatformEvent::park() { // AKA "down()" | |
| 4200 // Invariant: Only the thread associated with the Event/PlatformEvent | |
| 4201 // may call park(). | |
| 4202 // TODO: assert that _Assoc != NULL or _Assoc == Self | |
| 4203 int v ; | |
| 4204 for (;;) { | |
| 4205 v = _Event ; | |
| 4206 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; | |
| 4207 } | |
| 4208 guarantee (v >= 0, "invariant") ; | |
| 4209 if (v == 0) { | |
| 4210 // Do this the hard way by blocking ... | |
| 4211 int status = pthread_mutex_lock(_mutex); | |
| 4212 assert_status(status == 0, status, "mutex_lock"); | |
| 4213 guarantee (_nParked == 0, "invariant") ; | |
| 4214 ++ _nParked ; | |
| 4215 while (_Event < 0) { | |
| 4216 status = pthread_cond_wait(_cond, _mutex); | |
| 4217 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... | |
| 4218 // Treat this the same as if the wait was interrupted | |
| 4219 if (status == ETIME) { status = EINTR; } | |
| 4220 assert_status(status == 0 || status == EINTR, status, "cond_wait"); | |
| 4221 } | |
| 4222 -- _nParked ; | |
| 4223 | |
| 4224 // In theory we could move the ST of 0 into _Event past the unlock(), | |
| 4225 // but then we'd need a MEMBAR after the ST. | |
| 4226 _Event = 0 ; | |
| 4227 status = pthread_mutex_unlock(_mutex); | |
| 4228 assert_status(status == 0, status, "mutex_unlock"); | |
| 4229 } | |
| 4230 guarantee (_Event >= 0, "invariant") ; | |
| 4231 } | |
| 4232 | |
| 4233 int os::PlatformEvent::park(jlong millis) { | |
| 4234 guarantee (_nParked == 0, "invariant") ; | |
| 4235 | |
| 4236 int v ; | |
| 4237 for (;;) { | |
| 4238 v = _Event ; | |
| 4239 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; | |
| 4240 } | |
| 4241 guarantee (v >= 0, "invariant") ; | |
| 4242 if (v != 0) return OS_OK ; | |
| 4243 | |
| 4244 // We do this the hard way, by blocking the thread. | |
| 4245 // Consider enforcing a minimum timeout value. | |
| 4246 struct timespec abst; | |
| 4247 compute_abstime(&abst, millis); | |
| 4248 | |
| 4249 int ret = OS_TIMEOUT; | |
| 4250 int status = pthread_mutex_lock(_mutex); | |
| 4251 assert_status(status == 0, status, "mutex_lock"); | |
| 4252 guarantee (_nParked == 0, "invariant") ; | |
| 4253 ++_nParked ; | |
| 4254 | |
| 4255 // Object.wait(timo) will return because of | |
| 4256 // (a) notification | |
| 4257 // (b) timeout | |
| 4258 // (c) thread.interrupt | |
| 4259 // | |
| 4260 // Thread.interrupt and object.notify{All} both call Event::set. | |
| 4261 // That is, we treat thread.interrupt as a special case of notification. | |
| 4262 // The underlying Solaris implementation, cond_timedwait, admits | |
| 4263 // spurious/premature wakeups, but the JLS/JVM spec prevents the | |
| 4264 // JVM from making those visible to Java code. As such, we must | |
| 4265 // filter out spurious wakeups. We assume all ETIME returns are valid. | |
| 4266 // | |
| 4267 // TODO: properly differentiate simultaneous notify+interrupt. | |
| 4268 // In that case, we should propagate the notify to another waiter. | |
| 4269 | |
| 4270 while (_Event < 0) { | |
| 4271 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); | |
| 4272 if (status != 0 && WorkAroundNPTLTimedWaitHang) { | |
| 4273 pthread_cond_destroy (_cond); | |
| 4274 pthread_cond_init (_cond, NULL) ; | |
| 4275 } | |
| 4276 assert_status(status == 0 || status == EINTR || | |
| 4277 status == ETIME || status == ETIMEDOUT, | |
| 4278 status, "cond_timedwait"); | |
| 4279 if (!FilterSpuriousWakeups) break ; // previous semantics | |
| 4280 if (status == ETIME || status == ETIMEDOUT) break ; | |
| 4281 // We consume and ignore EINTR and spurious wakeups. | |
| 4282 } | |
| 4283 --_nParked ; | |
| 4284 if (_Event >= 0) { | |
| 4285 ret = OS_OK; | |
| 4286 } | |
| 4287 _Event = 0 ; | |
| 4288 status = pthread_mutex_unlock(_mutex); | |
| 4289 assert_status(status == 0, status, "mutex_unlock"); | |
| 4290 assert (_nParked == 0, "invariant") ; | |
| 4291 return ret; | |
| 4292 } | |
| 4293 | |
| 4294 void os::PlatformEvent::unpark() { | |
| 4295 int v, AnyWaiters ; | |
| 4296 for (;;) { | |
| 4297 v = _Event ; | |
| 4298 if (v > 0) { | |
| 4299 // The LD of _Event could have reordered or be satisfied | |
| 4300 // by a read-aside from this processor's write buffer. | |
| 4301 // To avoid problems execute a barrier and then | |
| 4302 // ratify the value. | |
| 4303 OrderAccess::fence() ; | |
| 4304 if (_Event == v) return ; | |
| 4305 continue ; | |
| 4306 } | |
| 4307 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; | |
| 4308 } | |
| 4309 if (v < 0) { | |
| 4310 // Wait for the thread associated with the event to vacate | |
| 4311 int status = pthread_mutex_lock(_mutex); | |
| 4312 assert_status(status == 0, status, "mutex_lock"); | |
| 4313 AnyWaiters = _nParked ; | |
| 4314 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; | |
| 4315 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { | |
| 4316 AnyWaiters = 0 ; | |
| 4317 pthread_cond_signal (_cond); | |
| 4318 } | |
| 4319 status = pthread_mutex_unlock(_mutex); | |
| 4320 assert_status(status == 0, status, "mutex_unlock"); | |
| 4321 if (AnyWaiters != 0) { | |
| 4322 status = pthread_cond_signal(_cond); | |
| 4323 assert_status(status == 0, status, "cond_signal"); | |
| 4324 } | |
| 4325 } | |
| 4326 | |
| 4327 // Note that we signal() _after dropping the lock for "immortal" Events. | |
| 4328 // This is safe and avoids a common class of futile wakeups. In rare | |
| 4329 // circumstances this can cause a thread to return prematurely from | |
| 4330 // cond_{timed}wait() but the spurious wakeup is benign and the victim will | |
| 4331 // simply re-test the condition and re-park itself. | |
| 4332 } | |
| 4333 | |
| 4334 | |
| 4335 // JSR166 | |
| 4336 // ------------------------------------------------------- | |
| 4337 | |
| 4338 /* | |
| 4339 * The solaris and linux implementations of park/unpark are fairly | |
| 4340 * conservative for now, but can be improved. They currently use a | |
| 4341 * mutex/condvar pair, plus a a count. | |
| 4342 * Park decrements count if > 0, else does a condvar wait. Unpark | |
| 4343 * sets count to 1 and signals condvar. Only one thread ever waits | |
| 4344 * on the condvar. Contention seen when trying to park implies that someone | |
| 4345 * is unparking you, so don't wait. And spurious returns are fine, so there | |
| 4346 * is no need to track notifications. | |
| 4347 */ | |
| 4348 | |
| 4349 | |
| 4350 #define NANOSECS_PER_SEC 1000000000 | |
| 4351 #define NANOSECS_PER_MILLISEC 1000000 | |
| 4352 #define MAX_SECS 100000000 | |
| 4353 /* | |
| 4354 * This code is common to linux and solaris and will be moved to a | |
| 4355 * common place in dolphin. | |
| 4356 * | |
| 4357 * The passed in time value is either a relative time in nanoseconds | |
| 4358 * or an absolute time in milliseconds. Either way it has to be unpacked | |
| 4359 * into suitable seconds and nanoseconds components and stored in the | |
| 4360 * given timespec structure. | |
| 4361 * Given time is a 64-bit value and the time_t used in the timespec is only | |
| 4362 * a signed-32-bit value (except on 64-bit Linux) we have to watch for | |
| 4363 * overflow if times way in the future are given. Further on Solaris versions | |
| 4364 * prior to 10 there is a restriction (see cond_timedwait) that the specified | |
| 4365 * number of seconds, in abstime, is less than current_time + 100,000,000. | |
| 4366 * As it will be 28 years before "now + 100000000" will overflow we can | |
| 4367 * ignore overflow and just impose a hard-limit on seconds using the value | |
| 4368 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 | |
| 4369 * years from "now". | |
| 4370 */ | |
| 4371 | |
| 4372 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { | |
| 4373 assert (time > 0, "convertTime"); | |
| 4374 | |
| 4375 struct timeval now; | |
| 4376 int status = gettimeofday(&now, NULL); | |
| 4377 assert(status == 0, "gettimeofday"); | |
| 4378 | |
| 4379 time_t max_secs = now.tv_sec + MAX_SECS; | |
| 4380 | |
| 4381 if (isAbsolute) { | |
| 4382 jlong secs = time / 1000; | |
| 4383 if (secs > max_secs) { | |
| 4384 absTime->tv_sec = max_secs; | |
| 4385 } | |
| 4386 else { | |
| 4387 absTime->tv_sec = secs; | |
| 4388 } | |
| 4389 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; | |
| 4390 } | |
| 4391 else { | |
| 4392 jlong secs = time / NANOSECS_PER_SEC; | |
| 4393 if (secs >= MAX_SECS) { | |
| 4394 absTime->tv_sec = max_secs; | |
| 4395 absTime->tv_nsec = 0; | |
| 4396 } | |
| 4397 else { | |
| 4398 absTime->tv_sec = now.tv_sec + secs; | |
| 4399 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; | |
| 4400 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { | |
| 4401 absTime->tv_nsec -= NANOSECS_PER_SEC; | |
| 4402 ++absTime->tv_sec; // note: this must be <= max_secs | |
| 4403 } | |
| 4404 } | |
| 4405 } | |
| 4406 assert(absTime->tv_sec >= 0, "tv_sec < 0"); | |
| 4407 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); | |
| 4408 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); | |
| 4409 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); | |
| 4410 } | |
| 4411 | |
| 4412 void Parker::park(bool isAbsolute, jlong time) { | |
| 4413 // Optional fast-path check: | |
| 4414 // Return immediately if a permit is available. | |
| 4415 if (_counter > 0) { | |
| 4416 _counter = 0 ; | |
| 4417 return ; | |
| 4418 } | |
| 4419 | |
| 4420 Thread* thread = Thread::current(); | |
| 4421 assert(thread->is_Java_thread(), "Must be JavaThread"); | |
| 4422 JavaThread *jt = (JavaThread *)thread; | |
| 4423 | |
| 4424 // Optional optimization -- avoid state transitions if there's an interrupt pending. | |
| 4425 // Check interrupt before trying to wait | |
| 4426 if (Thread::is_interrupted(thread, false)) { | |
| 4427 return; | |
| 4428 } | |
| 4429 | |
| 4430 // Next, demultiplex/decode time arguments | |
| 4431 timespec absTime; | |
| 4432 if (time < 0) { // don't wait at all | |
| 4433 return; | |
| 4434 } | |
| 4435 if (time > 0) { | |
| 4436 unpackTime(&absTime, isAbsolute, time); | |
| 4437 } | |
| 4438 | |
| 4439 | |
| 4440 // Enter safepoint region | |
| 4441 // Beware of deadlocks such as 6317397. | |
| 4442 // The per-thread Parker:: mutex is a classic leaf-lock. | |
| 4443 // In particular a thread must never block on the Threads_lock while | |
| 4444 // holding the Parker:: mutex. If safepoints are pending both the | |
| 4445 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. | |
| 4446 ThreadBlockInVM tbivm(jt); | |
| 4447 | |
| 4448 // Don't wait if cannot get lock since interference arises from | |
| 4449 // unblocking. Also. check interrupt before trying wait | |
| 4450 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { | |
| 4451 return; | |
| 4452 } | |
| 4453 | |
| 4454 int status ; | |
| 4455 if (_counter > 0) { // no wait needed | |
| 4456 _counter = 0; | |
| 4457 status = pthread_mutex_unlock(_mutex); | |
| 4458 assert (status == 0, "invariant") ; | |
| 4459 return; | |
| 4460 } | |
| 4461 | |
| 4462 #ifdef ASSERT | |
| 4463 // Don't catch signals while blocked; let the running threads have the signals. | |
| 4464 // (This allows a debugger to break into the running thread.) | |
| 4465 sigset_t oldsigs; | |
| 4466 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); | |
| 4467 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); | |
| 4468 #endif | |
| 4469 | |
| 4470 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); | |
| 4471 jt->set_suspend_equivalent(); | |
| 4472 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() | |
| 4473 | |
| 4474 if (time == 0) { | |
| 4475 status = pthread_cond_wait (_cond, _mutex) ; | |
| 4476 } else { | |
| 4477 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; | |
| 4478 if (status != 0 && WorkAroundNPTLTimedWaitHang) { | |
| 4479 pthread_cond_destroy (_cond) ; | |
| 4480 pthread_cond_init (_cond, NULL); | |
| 4481 } | |
| 4482 } | |
| 4483 assert_status(status == 0 || status == EINTR || | |
| 4484 status == ETIME || status == ETIMEDOUT, | |
| 4485 status, "cond_timedwait"); | |
| 4486 | |
| 4487 #ifdef ASSERT | |
| 4488 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); | |
| 4489 #endif | |
| 4490 | |
| 4491 _counter = 0 ; | |
| 4492 status = pthread_mutex_unlock(_mutex) ; | |
| 4493 assert_status(status == 0, status, "invariant") ; | |
| 4494 // If externally suspended while waiting, re-suspend | |
| 4495 if (jt->handle_special_suspend_equivalent_condition()) { | |
| 4496 jt->java_suspend_self(); | |
| 4497 } | |
| 4498 | |
| 4499 } | |
| 4500 | |
| 4501 void Parker::unpark() { | |
| 4502 int s, status ; | |
| 4503 status = pthread_mutex_lock(_mutex); | |
| 4504 assert (status == 0, "invariant") ; | |
| 4505 s = _counter; | |
| 4506 _counter = 1; | |
| 4507 if (s < 1) { | |
| 4508 if (WorkAroundNPTLTimedWaitHang) { | |
| 4509 status = pthread_cond_signal (_cond) ; | |
| 4510 assert (status == 0, "invariant") ; | |
| 4511 status = pthread_mutex_unlock(_mutex); | |
| 4512 assert (status == 0, "invariant") ; | |
| 4513 } else { | |
| 4514 status = pthread_mutex_unlock(_mutex); | |
| 4515 assert (status == 0, "invariant") ; | |
| 4516 status = pthread_cond_signal (_cond) ; | |
| 4517 assert (status == 0, "invariant") ; | |
| 4518 } | |
| 4519 } else { | |
| 4520 pthread_mutex_unlock(_mutex); | |
| 4521 assert (status == 0, "invariant") ; | |
| 4522 } | |
| 4523 } | |
| 4524 | |
| 4525 | |
| 4526 extern char** environ; | |
| 4527 | |
| 4528 #ifndef __NR_fork | |
| 4529 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) | |
| 4530 #endif | |
| 4531 | |
| 4532 #ifndef __NR_execve | |
| 4533 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) | |
| 4534 #endif | |
| 4535 | |
| 4536 // Run the specified command in a separate process. Return its exit value, | |
| 4537 // or -1 on failure (e.g. can't fork a new process). | |
| 4538 // Unlike system(), this function can be called from signal handler. It | |
| 4539 // doesn't block SIGINT et al. | |
| 4540 int os::fork_and_exec(char* cmd) { | |
| 4541 char * argv[4]; | |
| 4542 argv[0] = "sh"; | |
| 4543 argv[1] = "-c"; | |
| 4544 argv[2] = cmd; | |
| 4545 argv[3] = NULL; | |
| 4546 | |
| 4547 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run | |
| 4548 // pthread_atfork handlers and reset pthread library. All we need is a | |
| 4549 // separate process to execve. Make a direct syscall to fork process. | |
| 4550 // On IA64 there's no fork syscall, we have to use fork() and hope for | |
| 4551 // the best... | |
| 4552 pid_t pid = NOT_IA64(syscall(__NR_fork);) | |
| 4553 IA64_ONLY(fork();) | |
| 4554 | |
| 4555 if (pid < 0) { | |
| 4556 // fork failed | |
| 4557 return -1; | |
| 4558 | |
| 4559 } else if (pid == 0) { | |
| 4560 // child process | |
| 4561 | |
| 4562 // execve() in LinuxThreads will call pthread_kill_other_threads_np() | |
| 4563 // first to kill every thread on the thread list. Because this list is | |
| 4564 // not reset by fork() (see notes above), execve() will instead kill | |
| 4565 // every thread in the parent process. We know this is the only thread | |
| 4566 // in the new process, so make a system call directly. | |
| 4567 // IA64 should use normal execve() from glibc to match the glibc fork() | |
| 4568 // above. | |
| 4569 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) | |
| 4570 IA64_ONLY(execve("/bin/sh", argv, environ);) | |
| 4571 | |
| 4572 // execve failed | |
| 4573 _exit(-1); | |
| 4574 | |
| 4575 } else { | |
| 4576 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't | |
| 4577 // care about the actual exit code, for now. | |
| 4578 | |
| 4579 int status; | |
| 4580 | |
| 4581 // Wait for the child process to exit. This returns immediately if | |
| 4582 // the child has already exited. */ | |
| 4583 while (waitpid(pid, &status, 0) < 0) { | |
| 4584 switch (errno) { | |
| 4585 case ECHILD: return 0; | |
| 4586 case EINTR: break; | |
| 4587 default: return -1; | |
| 4588 } | |
| 4589 } | |
| 4590 | |
| 4591 if (WIFEXITED(status)) { | |
| 4592 // The child exited normally; get its exit code. | |
| 4593 return WEXITSTATUS(status); | |
| 4594 } else if (WIFSIGNALED(status)) { | |
| 4595 // The child exited because of a signal | |
| 4596 // The best value to return is 0x80 + signal number, | |
| 4597 // because that is what all Unix shells do, and because | |
| 4598 // it allows callers to distinguish between process exit and | |
| 4599 // process death by signal. | |
| 4600 return 0x80 + WTERMSIG(status); | |
| 4601 } else { | |
| 4602 // Unknown exit code; pass it through | |
| 4603 return status; | |
| 4604 } | |
| 4605 } | |
| 4606 } |