Mon, 17 May 2010 11:32:56 -0700
6951686: Using large pages on Linux prevents zero based compressed oops
Summary: Use req_addr when attaching shared memory segment.
Reviewed-by: twisti
duke@435 | 1 | /* |
xdono@1014 | 2 | * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
duke@435 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
duke@435 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
duke@435 | 21 | * have any questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
coleenp@1755 | 25 | # define __STDC_FORMAT_MACROS |
coleenp@1755 | 26 | |
duke@435 | 27 | // do not include precompiled header file |
duke@435 | 28 | # include "incls/_os_linux.cpp.incl" |
duke@435 | 29 | |
duke@435 | 30 | // put OS-includes here |
duke@435 | 31 | # include <sys/types.h> |
duke@435 | 32 | # include <sys/mman.h> |
duke@435 | 33 | # include <pthread.h> |
duke@435 | 34 | # include <signal.h> |
duke@435 | 35 | # include <errno.h> |
duke@435 | 36 | # include <dlfcn.h> |
duke@435 | 37 | # include <stdio.h> |
duke@435 | 38 | # include <unistd.h> |
duke@435 | 39 | # include <sys/resource.h> |
duke@435 | 40 | # include <pthread.h> |
duke@435 | 41 | # include <sys/stat.h> |
duke@435 | 42 | # include <sys/time.h> |
duke@435 | 43 | # include <sys/times.h> |
duke@435 | 44 | # include <sys/utsname.h> |
duke@435 | 45 | # include <sys/socket.h> |
duke@435 | 46 | # include <sys/wait.h> |
duke@435 | 47 | # include <pwd.h> |
duke@435 | 48 | # include <poll.h> |
duke@435 | 49 | # include <semaphore.h> |
duke@435 | 50 | # include <fcntl.h> |
duke@435 | 51 | # include <string.h> |
duke@435 | 52 | # include <syscall.h> |
duke@435 | 53 | # include <sys/sysinfo.h> |
duke@435 | 54 | # include <gnu/libc-version.h> |
duke@435 | 55 | # include <sys/ipc.h> |
duke@435 | 56 | # include <sys/shm.h> |
duke@435 | 57 | # include <link.h> |
coleenp@1755 | 58 | # include <stdint.h> |
coleenp@1755 | 59 | # include <inttypes.h> |
duke@435 | 60 | |
duke@435 | 61 | #define MAX_PATH (2 * K) |
duke@435 | 62 | |
duke@435 | 63 | // for timer info max values which include all bits |
duke@435 | 64 | #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) |
duke@435 | 65 | #define SEC_IN_NANOSECS 1000000000LL |
duke@435 | 66 | |
duke@435 | 67 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 68 | // global variables |
duke@435 | 69 | julong os::Linux::_physical_memory = 0; |
duke@435 | 70 | |
duke@435 | 71 | address os::Linux::_initial_thread_stack_bottom = NULL; |
duke@435 | 72 | uintptr_t os::Linux::_initial_thread_stack_size = 0; |
duke@435 | 73 | |
duke@435 | 74 | int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; |
duke@435 | 75 | int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; |
duke@435 | 76 | Mutex* os::Linux::_createThread_lock = NULL; |
duke@435 | 77 | pthread_t os::Linux::_main_thread; |
duke@435 | 78 | int os::Linux::_page_size = -1; |
duke@435 | 79 | bool os::Linux::_is_floating_stack = false; |
duke@435 | 80 | bool os::Linux::_is_NPTL = false; |
duke@435 | 81 | bool os::Linux::_supports_fast_thread_cpu_time = false; |
xlu@634 | 82 | const char * os::Linux::_glibc_version = NULL; |
xlu@634 | 83 | const char * os::Linux::_libpthread_version = NULL; |
duke@435 | 84 | |
duke@435 | 85 | static jlong initial_time_count=0; |
duke@435 | 86 | |
duke@435 | 87 | static int clock_tics_per_sec = 100; |
duke@435 | 88 | |
duke@435 | 89 | // For diagnostics to print a message once. see run_periodic_checks |
duke@435 | 90 | static sigset_t check_signal_done; |
duke@435 | 91 | static bool check_signals = true;; |
duke@435 | 92 | |
duke@435 | 93 | static pid_t _initial_pid = 0; |
duke@435 | 94 | |
duke@435 | 95 | /* Signal number used to suspend/resume a thread */ |
duke@435 | 96 | |
duke@435 | 97 | /* do not use any signal number less than SIGSEGV, see 4355769 */ |
duke@435 | 98 | static int SR_signum = SIGUSR2; |
duke@435 | 99 | sigset_t SR_sigset; |
duke@435 | 100 | |
kamg@677 | 101 | /* Used to protect dlsym() calls */ |
kamg@677 | 102 | static pthread_mutex_t dl_mutex; |
kamg@677 | 103 | |
duke@435 | 104 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 105 | // utility functions |
duke@435 | 106 | |
duke@435 | 107 | static int SR_initialize(); |
duke@435 | 108 | static int SR_finalize(); |
duke@435 | 109 | |
duke@435 | 110 | julong os::available_memory() { |
duke@435 | 111 | return Linux::available_memory(); |
duke@435 | 112 | } |
duke@435 | 113 | |
duke@435 | 114 | julong os::Linux::available_memory() { |
duke@435 | 115 | // values in struct sysinfo are "unsigned long" |
duke@435 | 116 | struct sysinfo si; |
duke@435 | 117 | sysinfo(&si); |
duke@435 | 118 | |
duke@435 | 119 | return (julong)si.freeram * si.mem_unit; |
duke@435 | 120 | } |
duke@435 | 121 | |
duke@435 | 122 | julong os::physical_memory() { |
duke@435 | 123 | return Linux::physical_memory(); |
duke@435 | 124 | } |
duke@435 | 125 | |
phh@455 | 126 | julong os::allocatable_physical_memory(julong size) { |
phh@455 | 127 | #ifdef _LP64 |
phh@455 | 128 | return size; |
phh@455 | 129 | #else |
phh@455 | 130 | julong result = MIN2(size, (julong)3800*M); |
phh@455 | 131 | if (!is_allocatable(result)) { |
phh@455 | 132 | // See comments under solaris for alignment considerations |
phh@455 | 133 | julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); |
phh@455 | 134 | result = MIN2(size, reasonable_size); |
phh@455 | 135 | } |
phh@455 | 136 | return result; |
phh@455 | 137 | #endif // _LP64 |
phh@455 | 138 | } |
phh@455 | 139 | |
duke@435 | 140 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 141 | // environment support |
duke@435 | 142 | |
duke@435 | 143 | bool os::getenv(const char* name, char* buf, int len) { |
duke@435 | 144 | const char* val = ::getenv(name); |
duke@435 | 145 | if (val != NULL && strlen(val) < (size_t)len) { |
duke@435 | 146 | strcpy(buf, val); |
duke@435 | 147 | return true; |
duke@435 | 148 | } |
duke@435 | 149 | if (len > 0) buf[0] = 0; // return a null string |
duke@435 | 150 | return false; |
duke@435 | 151 | } |
duke@435 | 152 | |
duke@435 | 153 | |
duke@435 | 154 | // Return true if user is running as root. |
duke@435 | 155 | |
duke@435 | 156 | bool os::have_special_privileges() { |
duke@435 | 157 | static bool init = false; |
duke@435 | 158 | static bool privileges = false; |
duke@435 | 159 | if (!init) { |
duke@435 | 160 | privileges = (getuid() != geteuid()) || (getgid() != getegid()); |
duke@435 | 161 | init = true; |
duke@435 | 162 | } |
duke@435 | 163 | return privileges; |
duke@435 | 164 | } |
duke@435 | 165 | |
duke@435 | 166 | |
duke@435 | 167 | #ifndef SYS_gettid |
duke@435 | 168 | // i386: 224, ia64: 1105, amd64: 186, sparc 143 |
duke@435 | 169 | #ifdef __ia64__ |
duke@435 | 170 | #define SYS_gettid 1105 |
duke@435 | 171 | #elif __i386__ |
duke@435 | 172 | #define SYS_gettid 224 |
duke@435 | 173 | #elif __amd64__ |
duke@435 | 174 | #define SYS_gettid 186 |
duke@435 | 175 | #elif __sparc__ |
duke@435 | 176 | #define SYS_gettid 143 |
duke@435 | 177 | #else |
duke@435 | 178 | #error define gettid for the arch |
duke@435 | 179 | #endif |
duke@435 | 180 | #endif |
duke@435 | 181 | |
duke@435 | 182 | // Cpu architecture string |
never@1445 | 183 | #if defined(ZERO) |
never@1445 | 184 | static char cpu_arch[] = ZERO_LIBARCH; |
never@1445 | 185 | #elif defined(IA64) |
duke@435 | 186 | static char cpu_arch[] = "ia64"; |
duke@435 | 187 | #elif defined(IA32) |
duke@435 | 188 | static char cpu_arch[] = "i386"; |
duke@435 | 189 | #elif defined(AMD64) |
duke@435 | 190 | static char cpu_arch[] = "amd64"; |
duke@435 | 191 | #elif defined(SPARC) |
duke@435 | 192 | # ifdef _LP64 |
duke@435 | 193 | static char cpu_arch[] = "sparcv9"; |
duke@435 | 194 | # else |
duke@435 | 195 | static char cpu_arch[] = "sparc"; |
duke@435 | 196 | # endif |
duke@435 | 197 | #else |
duke@435 | 198 | #error Add appropriate cpu_arch setting |
duke@435 | 199 | #endif |
duke@435 | 200 | |
duke@435 | 201 | |
duke@435 | 202 | // pid_t gettid() |
duke@435 | 203 | // |
duke@435 | 204 | // Returns the kernel thread id of the currently running thread. Kernel |
duke@435 | 205 | // thread id is used to access /proc. |
duke@435 | 206 | // |
duke@435 | 207 | // (Note that getpid() on LinuxThreads returns kernel thread id too; but |
duke@435 | 208 | // on NPTL, it returns the same pid for all threads, as required by POSIX.) |
duke@435 | 209 | // |
duke@435 | 210 | pid_t os::Linux::gettid() { |
duke@435 | 211 | int rslt = syscall(SYS_gettid); |
duke@435 | 212 | if (rslt == -1) { |
duke@435 | 213 | // old kernel, no NPTL support |
duke@435 | 214 | return getpid(); |
duke@435 | 215 | } else { |
duke@435 | 216 | return (pid_t)rslt; |
duke@435 | 217 | } |
duke@435 | 218 | } |
duke@435 | 219 | |
duke@435 | 220 | // Most versions of linux have a bug where the number of processors are |
duke@435 | 221 | // determined by looking at the /proc file system. In a chroot environment, |
duke@435 | 222 | // the system call returns 1. This causes the VM to act as if it is |
duke@435 | 223 | // a single processor and elide locking (see is_MP() call). |
duke@435 | 224 | static bool unsafe_chroot_detected = false; |
xlu@634 | 225 | static const char *unstable_chroot_error = "/proc file system not found.\n" |
xlu@634 | 226 | "Java may be unstable running multithreaded in a chroot " |
xlu@634 | 227 | "environment on Linux when /proc filesystem is not mounted."; |
duke@435 | 228 | |
duke@435 | 229 | void os::Linux::initialize_system_info() { |
phh@1558 | 230 | set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); |
phh@1558 | 231 | if (processor_count() == 1) { |
duke@435 | 232 | pid_t pid = os::Linux::gettid(); |
duke@435 | 233 | char fname[32]; |
duke@435 | 234 | jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); |
duke@435 | 235 | FILE *fp = fopen(fname, "r"); |
duke@435 | 236 | if (fp == NULL) { |
duke@435 | 237 | unsafe_chroot_detected = true; |
duke@435 | 238 | } else { |
duke@435 | 239 | fclose(fp); |
duke@435 | 240 | } |
duke@435 | 241 | } |
duke@435 | 242 | _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); |
phh@1558 | 243 | assert(processor_count() > 0, "linux error"); |
duke@435 | 244 | } |
duke@435 | 245 | |
duke@435 | 246 | void os::init_system_properties_values() { |
duke@435 | 247 | // char arch[12]; |
duke@435 | 248 | // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); |
duke@435 | 249 | |
duke@435 | 250 | // The next steps are taken in the product version: |
duke@435 | 251 | // |
duke@435 | 252 | // Obtain the JAVA_HOME value from the location of libjvm[_g].so. |
duke@435 | 253 | // This library should be located at: |
duke@435 | 254 | // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. |
duke@435 | 255 | // |
duke@435 | 256 | // If "/jre/lib/" appears at the right place in the path, then we |
duke@435 | 257 | // assume libjvm[_g].so is installed in a JDK and we use this path. |
duke@435 | 258 | // |
duke@435 | 259 | // Otherwise exit with message: "Could not create the Java virtual machine." |
duke@435 | 260 | // |
duke@435 | 261 | // The following extra steps are taken in the debugging version: |
duke@435 | 262 | // |
duke@435 | 263 | // If "/jre/lib/" does NOT appear at the right place in the path |
duke@435 | 264 | // instead of exit check for $JAVA_HOME environment variable. |
duke@435 | 265 | // |
duke@435 | 266 | // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, |
duke@435 | 267 | // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so |
duke@435 | 268 | // it looks like libjvm[_g].so is installed there |
duke@435 | 269 | // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. |
duke@435 | 270 | // |
duke@435 | 271 | // Otherwise exit. |
duke@435 | 272 | // |
duke@435 | 273 | // Important note: if the location of libjvm.so changes this |
duke@435 | 274 | // code needs to be changed accordingly. |
duke@435 | 275 | |
duke@435 | 276 | // The next few definitions allow the code to be verbatim: |
duke@435 | 277 | #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) |
duke@435 | 278 | #define getenv(n) ::getenv(n) |
duke@435 | 279 | |
duke@435 | 280 | /* |
duke@435 | 281 | * See ld(1): |
duke@435 | 282 | * The linker uses the following search paths to locate required |
duke@435 | 283 | * shared libraries: |
duke@435 | 284 | * 1: ... |
duke@435 | 285 | * ... |
duke@435 | 286 | * 7: The default directories, normally /lib and /usr/lib. |
duke@435 | 287 | */ |
kvn@944 | 288 | #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) |
kvn@944 | 289 | #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" |
kvn@944 | 290 | #else |
duke@435 | 291 | #define DEFAULT_LIBPATH "/lib:/usr/lib" |
kvn@944 | 292 | #endif |
duke@435 | 293 | |
duke@435 | 294 | #define EXTENSIONS_DIR "/lib/ext" |
duke@435 | 295 | #define ENDORSED_DIR "/lib/endorsed" |
duke@435 | 296 | #define REG_DIR "/usr/java/packages" |
duke@435 | 297 | |
duke@435 | 298 | { |
duke@435 | 299 | /* sysclasspath, java_home, dll_dir */ |
duke@435 | 300 | { |
duke@435 | 301 | char *home_path; |
duke@435 | 302 | char *dll_path; |
duke@435 | 303 | char *pslash; |
duke@435 | 304 | char buf[MAXPATHLEN]; |
duke@435 | 305 | os::jvm_path(buf, sizeof(buf)); |
duke@435 | 306 | |
duke@435 | 307 | // Found the full path to libjvm.so. |
duke@435 | 308 | // Now cut the path to <java_home>/jre if we can. |
duke@435 | 309 | *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ |
duke@435 | 310 | pslash = strrchr(buf, '/'); |
duke@435 | 311 | if (pslash != NULL) |
duke@435 | 312 | *pslash = '\0'; /* get rid of /{client|server|hotspot} */ |
duke@435 | 313 | dll_path = malloc(strlen(buf) + 1); |
duke@435 | 314 | if (dll_path == NULL) |
duke@435 | 315 | return; |
duke@435 | 316 | strcpy(dll_path, buf); |
duke@435 | 317 | Arguments::set_dll_dir(dll_path); |
duke@435 | 318 | |
duke@435 | 319 | if (pslash != NULL) { |
duke@435 | 320 | pslash = strrchr(buf, '/'); |
duke@435 | 321 | if (pslash != NULL) { |
duke@435 | 322 | *pslash = '\0'; /* get rid of /<arch> */ |
duke@435 | 323 | pslash = strrchr(buf, '/'); |
duke@435 | 324 | if (pslash != NULL) |
duke@435 | 325 | *pslash = '\0'; /* get rid of /lib */ |
duke@435 | 326 | } |
duke@435 | 327 | } |
duke@435 | 328 | |
duke@435 | 329 | home_path = malloc(strlen(buf) + 1); |
duke@435 | 330 | if (home_path == NULL) |
duke@435 | 331 | return; |
duke@435 | 332 | strcpy(home_path, buf); |
duke@435 | 333 | Arguments::set_java_home(home_path); |
duke@435 | 334 | |
duke@435 | 335 | if (!set_boot_path('/', ':')) |
duke@435 | 336 | return; |
duke@435 | 337 | } |
duke@435 | 338 | |
duke@435 | 339 | /* |
duke@435 | 340 | * Where to look for native libraries |
duke@435 | 341 | * |
duke@435 | 342 | * Note: Due to a legacy implementation, most of the library path |
duke@435 | 343 | * is set in the launcher. This was to accomodate linking restrictions |
duke@435 | 344 | * on legacy Linux implementations (which are no longer supported). |
duke@435 | 345 | * Eventually, all the library path setting will be done here. |
duke@435 | 346 | * |
duke@435 | 347 | * However, to prevent the proliferation of improperly built native |
duke@435 | 348 | * libraries, the new path component /usr/java/packages is added here. |
duke@435 | 349 | * Eventually, all the library path setting will be done here. |
duke@435 | 350 | */ |
duke@435 | 351 | { |
duke@435 | 352 | char *ld_library_path; |
duke@435 | 353 | |
duke@435 | 354 | /* |
duke@435 | 355 | * Construct the invariant part of ld_library_path. Note that the |
duke@435 | 356 | * space for the colon and the trailing null are provided by the |
duke@435 | 357 | * nulls included by the sizeof operator (so actually we allocate |
duke@435 | 358 | * a byte more than necessary). |
duke@435 | 359 | */ |
duke@435 | 360 | ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + |
duke@435 | 361 | strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); |
duke@435 | 362 | sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); |
duke@435 | 363 | |
duke@435 | 364 | /* |
duke@435 | 365 | * Get the user setting of LD_LIBRARY_PATH, and prepended it. It |
duke@435 | 366 | * should always exist (until the legacy problem cited above is |
duke@435 | 367 | * addressed). |
duke@435 | 368 | */ |
duke@435 | 369 | char *v = getenv("LD_LIBRARY_PATH"); |
duke@435 | 370 | if (v != NULL) { |
duke@435 | 371 | char *t = ld_library_path; |
duke@435 | 372 | /* That's +1 for the colon and +1 for the trailing '\0' */ |
duke@435 | 373 | ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); |
duke@435 | 374 | sprintf(ld_library_path, "%s:%s", v, t); |
duke@435 | 375 | } |
duke@435 | 376 | Arguments::set_library_path(ld_library_path); |
duke@435 | 377 | } |
duke@435 | 378 | |
duke@435 | 379 | /* |
duke@435 | 380 | * Extensions directories. |
duke@435 | 381 | * |
duke@435 | 382 | * Note that the space for the colon and the trailing null are provided |
duke@435 | 383 | * by the nulls included by the sizeof operator (so actually one byte more |
duke@435 | 384 | * than necessary is allocated). |
duke@435 | 385 | */ |
duke@435 | 386 | { |
duke@435 | 387 | char *buf = malloc(strlen(Arguments::get_java_home()) + |
duke@435 | 388 | sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); |
duke@435 | 389 | sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, |
duke@435 | 390 | Arguments::get_java_home()); |
duke@435 | 391 | Arguments::set_ext_dirs(buf); |
duke@435 | 392 | } |
duke@435 | 393 | |
duke@435 | 394 | /* Endorsed standards default directory. */ |
duke@435 | 395 | { |
duke@435 | 396 | char * buf; |
duke@435 | 397 | buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); |
duke@435 | 398 | sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); |
duke@435 | 399 | Arguments::set_endorsed_dirs(buf); |
duke@435 | 400 | } |
duke@435 | 401 | } |
duke@435 | 402 | |
duke@435 | 403 | #undef malloc |
duke@435 | 404 | #undef getenv |
duke@435 | 405 | #undef EXTENSIONS_DIR |
duke@435 | 406 | #undef ENDORSED_DIR |
duke@435 | 407 | |
duke@435 | 408 | // Done |
duke@435 | 409 | return; |
duke@435 | 410 | } |
duke@435 | 411 | |
duke@435 | 412 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 413 | // breakpoint support |
duke@435 | 414 | |
duke@435 | 415 | void os::breakpoint() { |
duke@435 | 416 | BREAKPOINT; |
duke@435 | 417 | } |
duke@435 | 418 | |
duke@435 | 419 | extern "C" void breakpoint() { |
duke@435 | 420 | // use debugger to set breakpoint here |
duke@435 | 421 | } |
duke@435 | 422 | |
duke@435 | 423 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 424 | // signal support |
duke@435 | 425 | |
duke@435 | 426 | debug_only(static bool signal_sets_initialized = false); |
duke@435 | 427 | static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; |
duke@435 | 428 | |
duke@435 | 429 | bool os::Linux::is_sig_ignored(int sig) { |
duke@435 | 430 | struct sigaction oact; |
duke@435 | 431 | sigaction(sig, (struct sigaction*)NULL, &oact); |
duke@435 | 432 | void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) |
duke@435 | 433 | : CAST_FROM_FN_PTR(void*, oact.sa_handler); |
duke@435 | 434 | if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) |
duke@435 | 435 | return true; |
duke@435 | 436 | else |
duke@435 | 437 | return false; |
duke@435 | 438 | } |
duke@435 | 439 | |
duke@435 | 440 | void os::Linux::signal_sets_init() { |
duke@435 | 441 | // Should also have an assertion stating we are still single-threaded. |
duke@435 | 442 | assert(!signal_sets_initialized, "Already initialized"); |
duke@435 | 443 | // Fill in signals that are necessarily unblocked for all threads in |
duke@435 | 444 | // the VM. Currently, we unblock the following signals: |
duke@435 | 445 | // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden |
duke@435 | 446 | // by -Xrs (=ReduceSignalUsage)); |
duke@435 | 447 | // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all |
duke@435 | 448 | // other threads. The "ReduceSignalUsage" boolean tells us not to alter |
duke@435 | 449 | // the dispositions or masks wrt these signals. |
duke@435 | 450 | // Programs embedding the VM that want to use the above signals for their |
duke@435 | 451 | // own purposes must, at this time, use the "-Xrs" option to prevent |
duke@435 | 452 | // interference with shutdown hooks and BREAK_SIGNAL thread dumping. |
duke@435 | 453 | // (See bug 4345157, and other related bugs). |
duke@435 | 454 | // In reality, though, unblocking these signals is really a nop, since |
duke@435 | 455 | // these signals are not blocked by default. |
duke@435 | 456 | sigemptyset(&unblocked_sigs); |
duke@435 | 457 | sigemptyset(&allowdebug_blocked_sigs); |
duke@435 | 458 | sigaddset(&unblocked_sigs, SIGILL); |
duke@435 | 459 | sigaddset(&unblocked_sigs, SIGSEGV); |
duke@435 | 460 | sigaddset(&unblocked_sigs, SIGBUS); |
duke@435 | 461 | sigaddset(&unblocked_sigs, SIGFPE); |
duke@435 | 462 | sigaddset(&unblocked_sigs, SR_signum); |
duke@435 | 463 | |
duke@435 | 464 | if (!ReduceSignalUsage) { |
duke@435 | 465 | if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { |
duke@435 | 466 | sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); |
duke@435 | 467 | sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); |
duke@435 | 468 | } |
duke@435 | 469 | if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { |
duke@435 | 470 | sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); |
duke@435 | 471 | sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); |
duke@435 | 472 | } |
duke@435 | 473 | if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { |
duke@435 | 474 | sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); |
duke@435 | 475 | sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); |
duke@435 | 476 | } |
duke@435 | 477 | } |
duke@435 | 478 | // Fill in signals that are blocked by all but the VM thread. |
duke@435 | 479 | sigemptyset(&vm_sigs); |
duke@435 | 480 | if (!ReduceSignalUsage) |
duke@435 | 481 | sigaddset(&vm_sigs, BREAK_SIGNAL); |
duke@435 | 482 | debug_only(signal_sets_initialized = true); |
duke@435 | 483 | |
duke@435 | 484 | } |
duke@435 | 485 | |
duke@435 | 486 | // These are signals that are unblocked while a thread is running Java. |
duke@435 | 487 | // (For some reason, they get blocked by default.) |
duke@435 | 488 | sigset_t* os::Linux::unblocked_signals() { |
duke@435 | 489 | assert(signal_sets_initialized, "Not initialized"); |
duke@435 | 490 | return &unblocked_sigs; |
duke@435 | 491 | } |
duke@435 | 492 | |
duke@435 | 493 | // These are the signals that are blocked while a (non-VM) thread is |
duke@435 | 494 | // running Java. Only the VM thread handles these signals. |
duke@435 | 495 | sigset_t* os::Linux::vm_signals() { |
duke@435 | 496 | assert(signal_sets_initialized, "Not initialized"); |
duke@435 | 497 | return &vm_sigs; |
duke@435 | 498 | } |
duke@435 | 499 | |
duke@435 | 500 | // These are signals that are blocked during cond_wait to allow debugger in |
duke@435 | 501 | sigset_t* os::Linux::allowdebug_blocked_signals() { |
duke@435 | 502 | assert(signal_sets_initialized, "Not initialized"); |
duke@435 | 503 | return &allowdebug_blocked_sigs; |
duke@435 | 504 | } |
duke@435 | 505 | |
duke@435 | 506 | void os::Linux::hotspot_sigmask(Thread* thread) { |
duke@435 | 507 | |
duke@435 | 508 | //Save caller's signal mask before setting VM signal mask |
duke@435 | 509 | sigset_t caller_sigmask; |
duke@435 | 510 | pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); |
duke@435 | 511 | |
duke@435 | 512 | OSThread* osthread = thread->osthread(); |
duke@435 | 513 | osthread->set_caller_sigmask(caller_sigmask); |
duke@435 | 514 | |
duke@435 | 515 | pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); |
duke@435 | 516 | |
duke@435 | 517 | if (!ReduceSignalUsage) { |
duke@435 | 518 | if (thread->is_VM_thread()) { |
duke@435 | 519 | // Only the VM thread handles BREAK_SIGNAL ... |
duke@435 | 520 | pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); |
duke@435 | 521 | } else { |
duke@435 | 522 | // ... all other threads block BREAK_SIGNAL |
duke@435 | 523 | pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); |
duke@435 | 524 | } |
duke@435 | 525 | } |
duke@435 | 526 | } |
duke@435 | 527 | |
duke@435 | 528 | ////////////////////////////////////////////////////////////////////////////// |
duke@435 | 529 | // detecting pthread library |
duke@435 | 530 | |
duke@435 | 531 | void os::Linux::libpthread_init() { |
duke@435 | 532 | // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION |
duke@435 | 533 | // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a |
duke@435 | 534 | // generic name for earlier versions. |
duke@435 | 535 | // Define macros here so we can build HotSpot on old systems. |
duke@435 | 536 | # ifndef _CS_GNU_LIBC_VERSION |
duke@435 | 537 | # define _CS_GNU_LIBC_VERSION 2 |
duke@435 | 538 | # endif |
duke@435 | 539 | # ifndef _CS_GNU_LIBPTHREAD_VERSION |
duke@435 | 540 | # define _CS_GNU_LIBPTHREAD_VERSION 3 |
duke@435 | 541 | # endif |
duke@435 | 542 | |
duke@435 | 543 | size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); |
duke@435 | 544 | if (n > 0) { |
duke@435 | 545 | char *str = (char *)malloc(n); |
duke@435 | 546 | confstr(_CS_GNU_LIBC_VERSION, str, n); |
duke@435 | 547 | os::Linux::set_glibc_version(str); |
duke@435 | 548 | } else { |
duke@435 | 549 | // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() |
duke@435 | 550 | static char _gnu_libc_version[32]; |
duke@435 | 551 | jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), |
duke@435 | 552 | "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); |
duke@435 | 553 | os::Linux::set_glibc_version(_gnu_libc_version); |
duke@435 | 554 | } |
duke@435 | 555 | |
duke@435 | 556 | n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); |
duke@435 | 557 | if (n > 0) { |
duke@435 | 558 | char *str = (char *)malloc(n); |
duke@435 | 559 | confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); |
duke@435 | 560 | // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells |
duke@435 | 561 | // us "NPTL-0.29" even we are running with LinuxThreads. Check if this |
xlu@634 | 562 | // is the case. LinuxThreads has a hard limit on max number of threads. |
xlu@634 | 563 | // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. |
xlu@634 | 564 | // On the other hand, NPTL does not have such a limit, sysconf() |
xlu@634 | 565 | // will return -1 and errno is not changed. Check if it is really NPTL. |
duke@435 | 566 | if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && |
xlu@634 | 567 | strstr(str, "NPTL") && |
xlu@634 | 568 | sysconf(_SC_THREAD_THREADS_MAX) > 0) { |
xlu@634 | 569 | free(str); |
xlu@634 | 570 | os::Linux::set_libpthread_version("linuxthreads"); |
xlu@634 | 571 | } else { |
xlu@634 | 572 | os::Linux::set_libpthread_version(str); |
duke@435 | 573 | } |
duke@435 | 574 | } else { |
xlu@634 | 575 | // glibc before 2.3.2 only has LinuxThreads. |
xlu@634 | 576 | os::Linux::set_libpthread_version("linuxthreads"); |
duke@435 | 577 | } |
duke@435 | 578 | |
duke@435 | 579 | if (strstr(libpthread_version(), "NPTL")) { |
duke@435 | 580 | os::Linux::set_is_NPTL(); |
duke@435 | 581 | } else { |
duke@435 | 582 | os::Linux::set_is_LinuxThreads(); |
duke@435 | 583 | } |
duke@435 | 584 | |
duke@435 | 585 | // LinuxThreads have two flavors: floating-stack mode, which allows variable |
duke@435 | 586 | // stack size; and fixed-stack mode. NPTL is always floating-stack. |
duke@435 | 587 | if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { |
duke@435 | 588 | os::Linux::set_is_floating_stack(); |
duke@435 | 589 | } |
duke@435 | 590 | } |
duke@435 | 591 | |
duke@435 | 592 | ///////////////////////////////////////////////////////////////////////////// |
duke@435 | 593 | // thread stack |
duke@435 | 594 | |
duke@435 | 595 | // Force Linux kernel to expand current thread stack. If "bottom" is close |
duke@435 | 596 | // to the stack guard, caller should block all signals. |
duke@435 | 597 | // |
duke@435 | 598 | // MAP_GROWSDOWN: |
duke@435 | 599 | // A special mmap() flag that is used to implement thread stacks. It tells |
duke@435 | 600 | // kernel that the memory region should extend downwards when needed. This |
duke@435 | 601 | // allows early versions of LinuxThreads to only mmap the first few pages |
duke@435 | 602 | // when creating a new thread. Linux kernel will automatically expand thread |
duke@435 | 603 | // stack as needed (on page faults). |
duke@435 | 604 | // |
duke@435 | 605 | // However, because the memory region of a MAP_GROWSDOWN stack can grow on |
duke@435 | 606 | // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN |
duke@435 | 607 | // region, it's hard to tell if the fault is due to a legitimate stack |
duke@435 | 608 | // access or because of reading/writing non-exist memory (e.g. buffer |
duke@435 | 609 | // overrun). As a rule, if the fault happens below current stack pointer, |
duke@435 | 610 | // Linux kernel does not expand stack, instead a SIGSEGV is sent to the |
duke@435 | 611 | // application (see Linux kernel fault.c). |
duke@435 | 612 | // |
duke@435 | 613 | // This Linux feature can cause SIGSEGV when VM bangs thread stack for |
duke@435 | 614 | // stack overflow detection. |
duke@435 | 615 | // |
duke@435 | 616 | // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do |
duke@435 | 617 | // not use this flag. However, the stack of initial thread is not created |
duke@435 | 618 | // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though |
duke@435 | 619 | // unlikely) that user code can create a thread with MAP_GROWSDOWN stack |
duke@435 | 620 | // and then attach the thread to JVM. |
duke@435 | 621 | // |
duke@435 | 622 | // To get around the problem and allow stack banging on Linux, we need to |
duke@435 | 623 | // manually expand thread stack after receiving the SIGSEGV. |
duke@435 | 624 | // |
duke@435 | 625 | // There are two ways to expand thread stack to address "bottom", we used |
duke@435 | 626 | // both of them in JVM before 1.5: |
duke@435 | 627 | // 1. adjust stack pointer first so that it is below "bottom", and then |
duke@435 | 628 | // touch "bottom" |
duke@435 | 629 | // 2. mmap() the page in question |
duke@435 | 630 | // |
duke@435 | 631 | // Now alternate signal stack is gone, it's harder to use 2. For instance, |
duke@435 | 632 | // if current sp is already near the lower end of page 101, and we need to |
duke@435 | 633 | // call mmap() to map page 100, it is possible that part of the mmap() frame |
duke@435 | 634 | // will be placed in page 100. When page 100 is mapped, it is zero-filled. |
duke@435 | 635 | // That will destroy the mmap() frame and cause VM to crash. |
duke@435 | 636 | // |
duke@435 | 637 | // The following code works by adjusting sp first, then accessing the "bottom" |
duke@435 | 638 | // page to force a page fault. Linux kernel will then automatically expand the |
duke@435 | 639 | // stack mapping. |
duke@435 | 640 | // |
duke@435 | 641 | // _expand_stack_to() assumes its frame size is less than page size, which |
duke@435 | 642 | // should always be true if the function is not inlined. |
duke@435 | 643 | |
duke@435 | 644 | #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute |
duke@435 | 645 | #define NOINLINE |
duke@435 | 646 | #else |
duke@435 | 647 | #define NOINLINE __attribute__ ((noinline)) |
duke@435 | 648 | #endif |
duke@435 | 649 | |
duke@435 | 650 | static void _expand_stack_to(address bottom) NOINLINE; |
duke@435 | 651 | |
duke@435 | 652 | static void _expand_stack_to(address bottom) { |
duke@435 | 653 | address sp; |
duke@435 | 654 | size_t size; |
duke@435 | 655 | volatile char *p; |
duke@435 | 656 | |
duke@435 | 657 | // Adjust bottom to point to the largest address within the same page, it |
duke@435 | 658 | // gives us a one-page buffer if alloca() allocates slightly more memory. |
duke@435 | 659 | bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); |
duke@435 | 660 | bottom += os::Linux::page_size() - 1; |
duke@435 | 661 | |
duke@435 | 662 | // sp might be slightly above current stack pointer; if that's the case, we |
duke@435 | 663 | // will alloca() a little more space than necessary, which is OK. Don't use |
duke@435 | 664 | // os::current_stack_pointer(), as its result can be slightly below current |
duke@435 | 665 | // stack pointer, causing us to not alloca enough to reach "bottom". |
duke@435 | 666 | sp = (address)&sp; |
duke@435 | 667 | |
duke@435 | 668 | if (sp > bottom) { |
duke@435 | 669 | size = sp - bottom; |
duke@435 | 670 | p = (volatile char *)alloca(size); |
duke@435 | 671 | assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); |
duke@435 | 672 | p[0] = '\0'; |
duke@435 | 673 | } |
duke@435 | 674 | } |
duke@435 | 675 | |
duke@435 | 676 | bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { |
duke@435 | 677 | assert(t!=NULL, "just checking"); |
duke@435 | 678 | assert(t->osthread()->expanding_stack(), "expand should be set"); |
duke@435 | 679 | assert(t->stack_base() != NULL, "stack_base was not initialized"); |
duke@435 | 680 | |
duke@435 | 681 | if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { |
duke@435 | 682 | sigset_t mask_all, old_sigset; |
duke@435 | 683 | sigfillset(&mask_all); |
duke@435 | 684 | pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); |
duke@435 | 685 | _expand_stack_to(addr); |
duke@435 | 686 | pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); |
duke@435 | 687 | return true; |
duke@435 | 688 | } |
duke@435 | 689 | return false; |
duke@435 | 690 | } |
duke@435 | 691 | |
duke@435 | 692 | ////////////////////////////////////////////////////////////////////////////// |
duke@435 | 693 | // create new thread |
duke@435 | 694 | |
duke@435 | 695 | static address highest_vm_reserved_address(); |
duke@435 | 696 | |
duke@435 | 697 | // check if it's safe to start a new thread |
duke@435 | 698 | static bool _thread_safety_check(Thread* thread) { |
duke@435 | 699 | if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { |
duke@435 | 700 | // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) |
duke@435 | 701 | // Heap is mmap'ed at lower end of memory space. Thread stacks are |
duke@435 | 702 | // allocated (MAP_FIXED) from high address space. Every thread stack |
duke@435 | 703 | // occupies a fixed size slot (usually 2Mbytes, but user can change |
duke@435 | 704 | // it to other values if they rebuild LinuxThreads). |
duke@435 | 705 | // |
duke@435 | 706 | // Problem with MAP_FIXED is that mmap() can still succeed even part of |
duke@435 | 707 | // the memory region has already been mmap'ed. That means if we have too |
duke@435 | 708 | // many threads and/or very large heap, eventually thread stack will |
duke@435 | 709 | // collide with heap. |
duke@435 | 710 | // |
duke@435 | 711 | // Here we try to prevent heap/stack collision by comparing current |
duke@435 | 712 | // stack bottom with the highest address that has been mmap'ed by JVM |
duke@435 | 713 | // plus a safety margin for memory maps created by native code. |
duke@435 | 714 | // |
duke@435 | 715 | // This feature can be disabled by setting ThreadSafetyMargin to 0 |
duke@435 | 716 | // |
duke@435 | 717 | if (ThreadSafetyMargin > 0) { |
duke@435 | 718 | address stack_bottom = os::current_stack_base() - os::current_stack_size(); |
duke@435 | 719 | |
duke@435 | 720 | // not safe if our stack extends below the safety margin |
duke@435 | 721 | return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); |
duke@435 | 722 | } else { |
duke@435 | 723 | return true; |
duke@435 | 724 | } |
duke@435 | 725 | } else { |
duke@435 | 726 | // Floating stack LinuxThreads or NPTL: |
duke@435 | 727 | // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When |
duke@435 | 728 | // there's not enough space left, pthread_create() will fail. If we come |
duke@435 | 729 | // here, that means enough space has been reserved for stack. |
duke@435 | 730 | return true; |
duke@435 | 731 | } |
duke@435 | 732 | } |
duke@435 | 733 | |
duke@435 | 734 | // Thread start routine for all newly created threads |
duke@435 | 735 | static void *java_start(Thread *thread) { |
duke@435 | 736 | // Try to randomize the cache line index of hot stack frames. |
duke@435 | 737 | // This helps when threads of the same stack traces evict each other's |
duke@435 | 738 | // cache lines. The threads can be either from the same JVM instance, or |
duke@435 | 739 | // from different JVM instances. The benefit is especially true for |
duke@435 | 740 | // processors with hyperthreading technology. |
duke@435 | 741 | static int counter = 0; |
duke@435 | 742 | int pid = os::current_process_id(); |
duke@435 | 743 | alloca(((pid ^ counter++) & 7) * 128); |
duke@435 | 744 | |
duke@435 | 745 | ThreadLocalStorage::set_thread(thread); |
duke@435 | 746 | |
duke@435 | 747 | OSThread* osthread = thread->osthread(); |
duke@435 | 748 | Monitor* sync = osthread->startThread_lock(); |
duke@435 | 749 | |
duke@435 | 750 | // non floating stack LinuxThreads needs extra check, see above |
duke@435 | 751 | if (!_thread_safety_check(thread)) { |
duke@435 | 752 | // notify parent thread |
duke@435 | 753 | MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); |
duke@435 | 754 | osthread->set_state(ZOMBIE); |
duke@435 | 755 | sync->notify_all(); |
duke@435 | 756 | return NULL; |
duke@435 | 757 | } |
duke@435 | 758 | |
duke@435 | 759 | // thread_id is kernel thread id (similar to Solaris LWP id) |
duke@435 | 760 | osthread->set_thread_id(os::Linux::gettid()); |
duke@435 | 761 | |
duke@435 | 762 | if (UseNUMA) { |
duke@435 | 763 | int lgrp_id = os::numa_get_group_id(); |
duke@435 | 764 | if (lgrp_id != -1) { |
duke@435 | 765 | thread->set_lgrp_id(lgrp_id); |
duke@435 | 766 | } |
duke@435 | 767 | } |
duke@435 | 768 | // initialize signal mask for this thread |
duke@435 | 769 | os::Linux::hotspot_sigmask(thread); |
duke@435 | 770 | |
duke@435 | 771 | // initialize floating point control register |
duke@435 | 772 | os::Linux::init_thread_fpu_state(); |
duke@435 | 773 | |
duke@435 | 774 | // handshaking with parent thread |
duke@435 | 775 | { |
duke@435 | 776 | MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); |
duke@435 | 777 | |
duke@435 | 778 | // notify parent thread |
duke@435 | 779 | osthread->set_state(INITIALIZED); |
duke@435 | 780 | sync->notify_all(); |
duke@435 | 781 | |
duke@435 | 782 | // wait until os::start_thread() |
duke@435 | 783 | while (osthread->get_state() == INITIALIZED) { |
duke@435 | 784 | sync->wait(Mutex::_no_safepoint_check_flag); |
duke@435 | 785 | } |
duke@435 | 786 | } |
duke@435 | 787 | |
duke@435 | 788 | // call one more level start routine |
duke@435 | 789 | thread->run(); |
duke@435 | 790 | |
duke@435 | 791 | return 0; |
duke@435 | 792 | } |
duke@435 | 793 | |
duke@435 | 794 | bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { |
duke@435 | 795 | assert(thread->osthread() == NULL, "caller responsible"); |
duke@435 | 796 | |
duke@435 | 797 | // Allocate the OSThread object |
duke@435 | 798 | OSThread* osthread = new OSThread(NULL, NULL); |
duke@435 | 799 | if (osthread == NULL) { |
duke@435 | 800 | return false; |
duke@435 | 801 | } |
duke@435 | 802 | |
duke@435 | 803 | // set the correct thread state |
duke@435 | 804 | osthread->set_thread_type(thr_type); |
duke@435 | 805 | |
duke@435 | 806 | // Initial state is ALLOCATED but not INITIALIZED |
duke@435 | 807 | osthread->set_state(ALLOCATED); |
duke@435 | 808 | |
duke@435 | 809 | thread->set_osthread(osthread); |
duke@435 | 810 | |
duke@435 | 811 | // init thread attributes |
duke@435 | 812 | pthread_attr_t attr; |
duke@435 | 813 | pthread_attr_init(&attr); |
duke@435 | 814 | pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); |
duke@435 | 815 | |
duke@435 | 816 | // stack size |
duke@435 | 817 | if (os::Linux::supports_variable_stack_size()) { |
duke@435 | 818 | // calculate stack size if it's not specified by caller |
duke@435 | 819 | if (stack_size == 0) { |
duke@435 | 820 | stack_size = os::Linux::default_stack_size(thr_type); |
duke@435 | 821 | |
duke@435 | 822 | switch (thr_type) { |
duke@435 | 823 | case os::java_thread: |
duke@435 | 824 | // Java threads use ThreadStackSize which default value can be changed with the flag -Xss |
duke@435 | 825 | if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); |
duke@435 | 826 | break; |
duke@435 | 827 | case os::compiler_thread: |
duke@435 | 828 | if (CompilerThreadStackSize > 0) { |
duke@435 | 829 | stack_size = (size_t)(CompilerThreadStackSize * K); |
duke@435 | 830 | break; |
duke@435 | 831 | } // else fall through: |
duke@435 | 832 | // use VMThreadStackSize if CompilerThreadStackSize is not defined |
duke@435 | 833 | case os::vm_thread: |
duke@435 | 834 | case os::pgc_thread: |
duke@435 | 835 | case os::cgc_thread: |
duke@435 | 836 | case os::watcher_thread: |
duke@435 | 837 | if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); |
duke@435 | 838 | break; |
duke@435 | 839 | } |
duke@435 | 840 | } |
duke@435 | 841 | |
duke@435 | 842 | stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); |
duke@435 | 843 | pthread_attr_setstacksize(&attr, stack_size); |
duke@435 | 844 | } else { |
duke@435 | 845 | // let pthread_create() pick the default value. |
duke@435 | 846 | } |
duke@435 | 847 | |
duke@435 | 848 | // glibc guard page |
duke@435 | 849 | pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); |
duke@435 | 850 | |
duke@435 | 851 | ThreadState state; |
duke@435 | 852 | |
duke@435 | 853 | { |
duke@435 | 854 | // Serialize thread creation if we are running with fixed stack LinuxThreads |
duke@435 | 855 | bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); |
duke@435 | 856 | if (lock) { |
duke@435 | 857 | os::Linux::createThread_lock()->lock_without_safepoint_check(); |
duke@435 | 858 | } |
duke@435 | 859 | |
duke@435 | 860 | pthread_t tid; |
duke@435 | 861 | int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); |
duke@435 | 862 | |
duke@435 | 863 | pthread_attr_destroy(&attr); |
duke@435 | 864 | |
duke@435 | 865 | if (ret != 0) { |
duke@435 | 866 | if (PrintMiscellaneous && (Verbose || WizardMode)) { |
duke@435 | 867 | perror("pthread_create()"); |
duke@435 | 868 | } |
duke@435 | 869 | // Need to clean up stuff we've allocated so far |
duke@435 | 870 | thread->set_osthread(NULL); |
duke@435 | 871 | delete osthread; |
duke@435 | 872 | if (lock) os::Linux::createThread_lock()->unlock(); |
duke@435 | 873 | return false; |
duke@435 | 874 | } |
duke@435 | 875 | |
duke@435 | 876 | // Store pthread info into the OSThread |
duke@435 | 877 | osthread->set_pthread_id(tid); |
duke@435 | 878 | |
duke@435 | 879 | // Wait until child thread is either initialized or aborted |
duke@435 | 880 | { |
duke@435 | 881 | Monitor* sync_with_child = osthread->startThread_lock(); |
duke@435 | 882 | MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); |
duke@435 | 883 | while ((state = osthread->get_state()) == ALLOCATED) { |
duke@435 | 884 | sync_with_child->wait(Mutex::_no_safepoint_check_flag); |
duke@435 | 885 | } |
duke@435 | 886 | } |
duke@435 | 887 | |
duke@435 | 888 | if (lock) { |
duke@435 | 889 | os::Linux::createThread_lock()->unlock(); |
duke@435 | 890 | } |
duke@435 | 891 | } |
duke@435 | 892 | |
duke@435 | 893 | // Aborted due to thread limit being reached |
duke@435 | 894 | if (state == ZOMBIE) { |
duke@435 | 895 | thread->set_osthread(NULL); |
duke@435 | 896 | delete osthread; |
duke@435 | 897 | return false; |
duke@435 | 898 | } |
duke@435 | 899 | |
duke@435 | 900 | // The thread is returned suspended (in state INITIALIZED), |
duke@435 | 901 | // and is started higher up in the call chain |
duke@435 | 902 | assert(state == INITIALIZED, "race condition"); |
duke@435 | 903 | return true; |
duke@435 | 904 | } |
duke@435 | 905 | |
duke@435 | 906 | ///////////////////////////////////////////////////////////////////////////// |
duke@435 | 907 | // attach existing thread |
duke@435 | 908 | |
duke@435 | 909 | // bootstrap the main thread |
duke@435 | 910 | bool os::create_main_thread(JavaThread* thread) { |
duke@435 | 911 | assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); |
duke@435 | 912 | return create_attached_thread(thread); |
duke@435 | 913 | } |
duke@435 | 914 | |
duke@435 | 915 | bool os::create_attached_thread(JavaThread* thread) { |
duke@435 | 916 | #ifdef ASSERT |
duke@435 | 917 | thread->verify_not_published(); |
duke@435 | 918 | #endif |
duke@435 | 919 | |
duke@435 | 920 | // Allocate the OSThread object |
duke@435 | 921 | OSThread* osthread = new OSThread(NULL, NULL); |
duke@435 | 922 | |
duke@435 | 923 | if (osthread == NULL) { |
duke@435 | 924 | return false; |
duke@435 | 925 | } |
duke@435 | 926 | |
duke@435 | 927 | // Store pthread info into the OSThread |
duke@435 | 928 | osthread->set_thread_id(os::Linux::gettid()); |
duke@435 | 929 | osthread->set_pthread_id(::pthread_self()); |
duke@435 | 930 | |
duke@435 | 931 | // initialize floating point control register |
duke@435 | 932 | os::Linux::init_thread_fpu_state(); |
duke@435 | 933 | |
duke@435 | 934 | // Initial thread state is RUNNABLE |
duke@435 | 935 | osthread->set_state(RUNNABLE); |
duke@435 | 936 | |
duke@435 | 937 | thread->set_osthread(osthread); |
duke@435 | 938 | |
duke@435 | 939 | if (UseNUMA) { |
duke@435 | 940 | int lgrp_id = os::numa_get_group_id(); |
duke@435 | 941 | if (lgrp_id != -1) { |
duke@435 | 942 | thread->set_lgrp_id(lgrp_id); |
duke@435 | 943 | } |
duke@435 | 944 | } |
duke@435 | 945 | |
duke@435 | 946 | if (os::Linux::is_initial_thread()) { |
duke@435 | 947 | // If current thread is initial thread, its stack is mapped on demand, |
duke@435 | 948 | // see notes about MAP_GROWSDOWN. Here we try to force kernel to map |
duke@435 | 949 | // the entire stack region to avoid SEGV in stack banging. |
duke@435 | 950 | // It is also useful to get around the heap-stack-gap problem on SuSE |
duke@435 | 951 | // kernel (see 4821821 for details). We first expand stack to the top |
duke@435 | 952 | // of yellow zone, then enable stack yellow zone (order is significant, |
duke@435 | 953 | // enabling yellow zone first will crash JVM on SuSE Linux), so there |
duke@435 | 954 | // is no gap between the last two virtual memory regions. |
duke@435 | 955 | |
duke@435 | 956 | JavaThread *jt = (JavaThread *)thread; |
duke@435 | 957 | address addr = jt->stack_yellow_zone_base(); |
duke@435 | 958 | assert(addr != NULL, "initialization problem?"); |
duke@435 | 959 | assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); |
duke@435 | 960 | |
duke@435 | 961 | osthread->set_expanding_stack(); |
duke@435 | 962 | os::Linux::manually_expand_stack(jt, addr); |
duke@435 | 963 | osthread->clear_expanding_stack(); |
duke@435 | 964 | } |
duke@435 | 965 | |
duke@435 | 966 | // initialize signal mask for this thread |
duke@435 | 967 | // and save the caller's signal mask |
duke@435 | 968 | os::Linux::hotspot_sigmask(thread); |
duke@435 | 969 | |
duke@435 | 970 | return true; |
duke@435 | 971 | } |
duke@435 | 972 | |
duke@435 | 973 | void os::pd_start_thread(Thread* thread) { |
duke@435 | 974 | OSThread * osthread = thread->osthread(); |
duke@435 | 975 | assert(osthread->get_state() != INITIALIZED, "just checking"); |
duke@435 | 976 | Monitor* sync_with_child = osthread->startThread_lock(); |
duke@435 | 977 | MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); |
duke@435 | 978 | sync_with_child->notify(); |
duke@435 | 979 | } |
duke@435 | 980 | |
duke@435 | 981 | // Free Linux resources related to the OSThread |
duke@435 | 982 | void os::free_thread(OSThread* osthread) { |
duke@435 | 983 | assert(osthread != NULL, "osthread not set"); |
duke@435 | 984 | |
duke@435 | 985 | if (Thread::current()->osthread() == osthread) { |
duke@435 | 986 | // Restore caller's signal mask |
duke@435 | 987 | sigset_t sigmask = osthread->caller_sigmask(); |
duke@435 | 988 | pthread_sigmask(SIG_SETMASK, &sigmask, NULL); |
duke@435 | 989 | } |
duke@435 | 990 | |
duke@435 | 991 | delete osthread; |
duke@435 | 992 | } |
duke@435 | 993 | |
duke@435 | 994 | ////////////////////////////////////////////////////////////////////////////// |
duke@435 | 995 | // thread local storage |
duke@435 | 996 | |
duke@435 | 997 | int os::allocate_thread_local_storage() { |
duke@435 | 998 | pthread_key_t key; |
duke@435 | 999 | int rslt = pthread_key_create(&key, NULL); |
duke@435 | 1000 | assert(rslt == 0, "cannot allocate thread local storage"); |
duke@435 | 1001 | return (int)key; |
duke@435 | 1002 | } |
duke@435 | 1003 | |
duke@435 | 1004 | // Note: This is currently not used by VM, as we don't destroy TLS key |
duke@435 | 1005 | // on VM exit. |
duke@435 | 1006 | void os::free_thread_local_storage(int index) { |
duke@435 | 1007 | int rslt = pthread_key_delete((pthread_key_t)index); |
duke@435 | 1008 | assert(rslt == 0, "invalid index"); |
duke@435 | 1009 | } |
duke@435 | 1010 | |
duke@435 | 1011 | void os::thread_local_storage_at_put(int index, void* value) { |
duke@435 | 1012 | int rslt = pthread_setspecific((pthread_key_t)index, value); |
duke@435 | 1013 | assert(rslt == 0, "pthread_setspecific failed"); |
duke@435 | 1014 | } |
duke@435 | 1015 | |
duke@435 | 1016 | extern "C" Thread* get_thread() { |
duke@435 | 1017 | return ThreadLocalStorage::thread(); |
duke@435 | 1018 | } |
duke@435 | 1019 | |
duke@435 | 1020 | ////////////////////////////////////////////////////////////////////////////// |
duke@435 | 1021 | // initial thread |
duke@435 | 1022 | |
duke@435 | 1023 | // Check if current thread is the initial thread, similar to Solaris thr_main. |
duke@435 | 1024 | bool os::Linux::is_initial_thread(void) { |
duke@435 | 1025 | char dummy; |
duke@435 | 1026 | // If called before init complete, thread stack bottom will be null. |
duke@435 | 1027 | // Can be called if fatal error occurs before initialization. |
duke@435 | 1028 | if (initial_thread_stack_bottom() == NULL) return false; |
duke@435 | 1029 | assert(initial_thread_stack_bottom() != NULL && |
duke@435 | 1030 | initial_thread_stack_size() != 0, |
duke@435 | 1031 | "os::init did not locate initial thread's stack region"); |
duke@435 | 1032 | if ((address)&dummy >= initial_thread_stack_bottom() && |
duke@435 | 1033 | (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) |
duke@435 | 1034 | return true; |
duke@435 | 1035 | else return false; |
duke@435 | 1036 | } |
duke@435 | 1037 | |
duke@435 | 1038 | // Find the virtual memory area that contains addr |
duke@435 | 1039 | static bool find_vma(address addr, address* vma_low, address* vma_high) { |
duke@435 | 1040 | FILE *fp = fopen("/proc/self/maps", "r"); |
duke@435 | 1041 | if (fp) { |
duke@435 | 1042 | address low, high; |
duke@435 | 1043 | while (!feof(fp)) { |
duke@435 | 1044 | if (fscanf(fp, "%p-%p", &low, &high) == 2) { |
duke@435 | 1045 | if (low <= addr && addr < high) { |
duke@435 | 1046 | if (vma_low) *vma_low = low; |
duke@435 | 1047 | if (vma_high) *vma_high = high; |
duke@435 | 1048 | fclose (fp); |
duke@435 | 1049 | return true; |
duke@435 | 1050 | } |
duke@435 | 1051 | } |
duke@435 | 1052 | for (;;) { |
duke@435 | 1053 | int ch = fgetc(fp); |
duke@435 | 1054 | if (ch == EOF || ch == (int)'\n') break; |
duke@435 | 1055 | } |
duke@435 | 1056 | } |
duke@435 | 1057 | fclose(fp); |
duke@435 | 1058 | } |
duke@435 | 1059 | return false; |
duke@435 | 1060 | } |
duke@435 | 1061 | |
duke@435 | 1062 | // Locate initial thread stack. This special handling of initial thread stack |
duke@435 | 1063 | // is needed because pthread_getattr_np() on most (all?) Linux distros returns |
duke@435 | 1064 | // bogus value for initial thread. |
duke@435 | 1065 | void os::Linux::capture_initial_stack(size_t max_size) { |
duke@435 | 1066 | // stack size is the easy part, get it from RLIMIT_STACK |
duke@435 | 1067 | size_t stack_size; |
duke@435 | 1068 | struct rlimit rlim; |
duke@435 | 1069 | getrlimit(RLIMIT_STACK, &rlim); |
duke@435 | 1070 | stack_size = rlim.rlim_cur; |
duke@435 | 1071 | |
duke@435 | 1072 | // 6308388: a bug in ld.so will relocate its own .data section to the |
duke@435 | 1073 | // lower end of primordial stack; reduce ulimit -s value a little bit |
duke@435 | 1074 | // so we won't install guard page on ld.so's data section. |
duke@435 | 1075 | stack_size -= 2 * page_size(); |
duke@435 | 1076 | |
duke@435 | 1077 | // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat |
duke@435 | 1078 | // 7.1, in both cases we will get 2G in return value. |
duke@435 | 1079 | // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, |
duke@435 | 1080 | // SuSE 7.2, Debian) can not handle alternate signal stack correctly |
duke@435 | 1081 | // for initial thread if its stack size exceeds 6M. Cap it at 2M, |
duke@435 | 1082 | // in case other parts in glibc still assumes 2M max stack size. |
duke@435 | 1083 | // FIXME: alt signal stack is gone, maybe we can relax this constraint? |
duke@435 | 1084 | #ifndef IA64 |
duke@435 | 1085 | if (stack_size > 2 * K * K) stack_size = 2 * K * K; |
duke@435 | 1086 | #else |
duke@435 | 1087 | // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small |
duke@435 | 1088 | if (stack_size > 4 * K * K) stack_size = 4 * K * K; |
duke@435 | 1089 | #endif |
duke@435 | 1090 | |
duke@435 | 1091 | // Try to figure out where the stack base (top) is. This is harder. |
duke@435 | 1092 | // |
duke@435 | 1093 | // When an application is started, glibc saves the initial stack pointer in |
duke@435 | 1094 | // a global variable "__libc_stack_end", which is then used by system |
duke@435 | 1095 | // libraries. __libc_stack_end should be pretty close to stack top. The |
duke@435 | 1096 | // variable is available since the very early days. However, because it is |
duke@435 | 1097 | // a private interface, it could disappear in the future. |
duke@435 | 1098 | // |
duke@435 | 1099 | // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar |
duke@435 | 1100 | // to __libc_stack_end, it is very close to stack top, but isn't the real |
duke@435 | 1101 | // stack top. Note that /proc may not exist if VM is running as a chroot |
duke@435 | 1102 | // program, so reading /proc/<pid>/stat could fail. Also the contents of |
duke@435 | 1103 | // /proc/<pid>/stat could change in the future (though unlikely). |
duke@435 | 1104 | // |
duke@435 | 1105 | // We try __libc_stack_end first. If that doesn't work, look for |
duke@435 | 1106 | // /proc/<pid>/stat. If neither of them works, we use current stack pointer |
duke@435 | 1107 | // as a hint, which should work well in most cases. |
duke@435 | 1108 | |
duke@435 | 1109 | uintptr_t stack_start; |
duke@435 | 1110 | |
duke@435 | 1111 | // try __libc_stack_end first |
duke@435 | 1112 | uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); |
duke@435 | 1113 | if (p && *p) { |
duke@435 | 1114 | stack_start = *p; |
duke@435 | 1115 | } else { |
duke@435 | 1116 | // see if we can get the start_stack field from /proc/self/stat |
duke@435 | 1117 | FILE *fp; |
duke@435 | 1118 | int pid; |
duke@435 | 1119 | char state; |
duke@435 | 1120 | int ppid; |
duke@435 | 1121 | int pgrp; |
duke@435 | 1122 | int session; |
duke@435 | 1123 | int nr; |
duke@435 | 1124 | int tpgrp; |
duke@435 | 1125 | unsigned long flags; |
duke@435 | 1126 | unsigned long minflt; |
duke@435 | 1127 | unsigned long cminflt; |
duke@435 | 1128 | unsigned long majflt; |
duke@435 | 1129 | unsigned long cmajflt; |
duke@435 | 1130 | unsigned long utime; |
duke@435 | 1131 | unsigned long stime; |
duke@435 | 1132 | long cutime; |
duke@435 | 1133 | long cstime; |
duke@435 | 1134 | long prio; |
duke@435 | 1135 | long nice; |
duke@435 | 1136 | long junk; |
duke@435 | 1137 | long it_real; |
duke@435 | 1138 | uintptr_t start; |
duke@435 | 1139 | uintptr_t vsize; |
duke@435 | 1140 | uintptr_t rss; |
duke@435 | 1141 | unsigned long rsslim; |
duke@435 | 1142 | uintptr_t scodes; |
duke@435 | 1143 | uintptr_t ecode; |
duke@435 | 1144 | int i; |
duke@435 | 1145 | |
duke@435 | 1146 | // Figure what the primordial thread stack base is. Code is inspired |
duke@435 | 1147 | // by email from Hans Boehm. /proc/self/stat begins with current pid, |
duke@435 | 1148 | // followed by command name surrounded by parentheses, state, etc. |
duke@435 | 1149 | char stat[2048]; |
duke@435 | 1150 | int statlen; |
duke@435 | 1151 | |
duke@435 | 1152 | fp = fopen("/proc/self/stat", "r"); |
duke@435 | 1153 | if (fp) { |
duke@435 | 1154 | statlen = fread(stat, 1, 2047, fp); |
duke@435 | 1155 | stat[statlen] = '\0'; |
duke@435 | 1156 | fclose(fp); |
duke@435 | 1157 | |
duke@435 | 1158 | // Skip pid and the command string. Note that we could be dealing with |
duke@435 | 1159 | // weird command names, e.g. user could decide to rename java launcher |
duke@435 | 1160 | // to "java 1.4.2 :)", then the stat file would look like |
duke@435 | 1161 | // 1234 (java 1.4.2 :)) R ... ... |
duke@435 | 1162 | // We don't really need to know the command string, just find the last |
duke@435 | 1163 | // occurrence of ")" and then start parsing from there. See bug 4726580. |
duke@435 | 1164 | char * s = strrchr(stat, ')'); |
duke@435 | 1165 | |
duke@435 | 1166 | i = 0; |
duke@435 | 1167 | if (s) { |
duke@435 | 1168 | // Skip blank chars |
duke@435 | 1169 | do s++; while (isspace(*s)); |
duke@435 | 1170 | |
duke@435 | 1171 | /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ |
duke@435 | 1172 | /* 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 */ |
xlu@948 | 1173 | i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " |
xlu@948 | 1174 | UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT |
xlu@948 | 1175 | " %lu " |
xlu@948 | 1176 | UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT, |
duke@435 | 1177 | &state, /* 3 %c */ |
duke@435 | 1178 | &ppid, /* 4 %d */ |
duke@435 | 1179 | &pgrp, /* 5 %d */ |
duke@435 | 1180 | &session, /* 6 %d */ |
duke@435 | 1181 | &nr, /* 7 %d */ |
duke@435 | 1182 | &tpgrp, /* 8 %d */ |
duke@435 | 1183 | &flags, /* 9 %lu */ |
duke@435 | 1184 | &minflt, /* 10 %lu */ |
duke@435 | 1185 | &cminflt, /* 11 %lu */ |
duke@435 | 1186 | &majflt, /* 12 %lu */ |
duke@435 | 1187 | &cmajflt, /* 13 %lu */ |
duke@435 | 1188 | &utime, /* 14 %lu */ |
duke@435 | 1189 | &stime, /* 15 %lu */ |
duke@435 | 1190 | &cutime, /* 16 %ld */ |
duke@435 | 1191 | &cstime, /* 17 %ld */ |
duke@435 | 1192 | &prio, /* 18 %ld */ |
duke@435 | 1193 | &nice, /* 19 %ld */ |
duke@435 | 1194 | &junk, /* 20 %ld */ |
duke@435 | 1195 | &it_real, /* 21 %ld */ |
xlu@948 | 1196 | &start, /* 22 UINTX_FORMAT */ |
xlu@948 | 1197 | &vsize, /* 23 UINTX_FORMAT */ |
xlu@948 | 1198 | &rss, /* 24 UINTX_FORMAT */ |
duke@435 | 1199 | &rsslim, /* 25 %lu */ |
xlu@948 | 1200 | &scodes, /* 26 UINTX_FORMAT */ |
xlu@948 | 1201 | &ecode, /* 27 UINTX_FORMAT */ |
xlu@948 | 1202 | &stack_start); /* 28 UINTX_FORMAT */ |
duke@435 | 1203 | } |
duke@435 | 1204 | |
duke@435 | 1205 | if (i != 28 - 2) { |
duke@435 | 1206 | assert(false, "Bad conversion from /proc/self/stat"); |
duke@435 | 1207 | // product mode - assume we are the initial thread, good luck in the |
duke@435 | 1208 | // embedded case. |
duke@435 | 1209 | warning("Can't detect initial thread stack location - bad conversion"); |
duke@435 | 1210 | stack_start = (uintptr_t) &rlim; |
duke@435 | 1211 | } |
duke@435 | 1212 | } else { |
duke@435 | 1213 | // For some reason we can't open /proc/self/stat (for example, running on |
duke@435 | 1214 | // FreeBSD with a Linux emulator, or inside chroot), this should work for |
duke@435 | 1215 | // most cases, so don't abort: |
duke@435 | 1216 | warning("Can't detect initial thread stack location - no /proc/self/stat"); |
duke@435 | 1217 | stack_start = (uintptr_t) &rlim; |
duke@435 | 1218 | } |
duke@435 | 1219 | } |
duke@435 | 1220 | |
duke@435 | 1221 | // Now we have a pointer (stack_start) very close to the stack top, the |
duke@435 | 1222 | // next thing to do is to figure out the exact location of stack top. We |
duke@435 | 1223 | // can find out the virtual memory area that contains stack_start by |
duke@435 | 1224 | // reading /proc/self/maps, it should be the last vma in /proc/self/maps, |
duke@435 | 1225 | // and its upper limit is the real stack top. (again, this would fail if |
duke@435 | 1226 | // running inside chroot, because /proc may not exist.) |
duke@435 | 1227 | |
duke@435 | 1228 | uintptr_t stack_top; |
duke@435 | 1229 | address low, high; |
duke@435 | 1230 | if (find_vma((address)stack_start, &low, &high)) { |
duke@435 | 1231 | // success, "high" is the true stack top. (ignore "low", because initial |
duke@435 | 1232 | // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) |
duke@435 | 1233 | stack_top = (uintptr_t)high; |
duke@435 | 1234 | } else { |
duke@435 | 1235 | // failed, likely because /proc/self/maps does not exist |
duke@435 | 1236 | warning("Can't detect initial thread stack location - find_vma failed"); |
duke@435 | 1237 | // best effort: stack_start is normally within a few pages below the real |
duke@435 | 1238 | // stack top, use it as stack top, and reduce stack size so we won't put |
duke@435 | 1239 | // guard page outside stack. |
duke@435 | 1240 | stack_top = stack_start; |
duke@435 | 1241 | stack_size -= 16 * page_size(); |
duke@435 | 1242 | } |
duke@435 | 1243 | |
duke@435 | 1244 | // stack_top could be partially down the page so align it |
duke@435 | 1245 | stack_top = align_size_up(stack_top, page_size()); |
duke@435 | 1246 | |
duke@435 | 1247 | if (max_size && stack_size > max_size) { |
duke@435 | 1248 | _initial_thread_stack_size = max_size; |
duke@435 | 1249 | } else { |
duke@435 | 1250 | _initial_thread_stack_size = stack_size; |
duke@435 | 1251 | } |
duke@435 | 1252 | |
duke@435 | 1253 | _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); |
duke@435 | 1254 | _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; |
duke@435 | 1255 | } |
duke@435 | 1256 | |
duke@435 | 1257 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 1258 | // time support |
duke@435 | 1259 | |
duke@435 | 1260 | // Time since start-up in seconds to a fine granularity. |
duke@435 | 1261 | // Used by VMSelfDestructTimer and the MemProfiler. |
duke@435 | 1262 | double os::elapsedTime() { |
duke@435 | 1263 | |
duke@435 | 1264 | return (double)(os::elapsed_counter()) * 0.000001; |
duke@435 | 1265 | } |
duke@435 | 1266 | |
duke@435 | 1267 | jlong os::elapsed_counter() { |
duke@435 | 1268 | timeval time; |
duke@435 | 1269 | int status = gettimeofday(&time, NULL); |
duke@435 | 1270 | return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; |
duke@435 | 1271 | } |
duke@435 | 1272 | |
duke@435 | 1273 | jlong os::elapsed_frequency() { |
duke@435 | 1274 | return (1000 * 1000); |
duke@435 | 1275 | } |
duke@435 | 1276 | |
ysr@777 | 1277 | // For now, we say that linux does not support vtime. I have no idea |
ysr@777 | 1278 | // whether it can actually be made to (DLD, 9/13/05). |
ysr@777 | 1279 | |
ysr@777 | 1280 | bool os::supports_vtime() { return false; } |
ysr@777 | 1281 | bool os::enable_vtime() { return false; } |
ysr@777 | 1282 | bool os::vtime_enabled() { return false; } |
ysr@777 | 1283 | double os::elapsedVTime() { |
ysr@777 | 1284 | // better than nothing, but not much |
ysr@777 | 1285 | return elapsedTime(); |
ysr@777 | 1286 | } |
ysr@777 | 1287 | |
sbohne@496 | 1288 | jlong os::javaTimeMillis() { |
duke@435 | 1289 | timeval time; |
duke@435 | 1290 | int status = gettimeofday(&time, NULL); |
duke@435 | 1291 | assert(status != -1, "linux error"); |
duke@435 | 1292 | return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); |
duke@435 | 1293 | } |
duke@435 | 1294 | |
duke@435 | 1295 | #ifndef CLOCK_MONOTONIC |
duke@435 | 1296 | #define CLOCK_MONOTONIC (1) |
duke@435 | 1297 | #endif |
duke@435 | 1298 | |
duke@435 | 1299 | void os::Linux::clock_init() { |
duke@435 | 1300 | // we do dlopen's in this particular order due to bug in linux |
duke@435 | 1301 | // dynamical loader (see 6348968) leading to crash on exit |
duke@435 | 1302 | void* handle = dlopen("librt.so.1", RTLD_LAZY); |
duke@435 | 1303 | if (handle == NULL) { |
duke@435 | 1304 | handle = dlopen("librt.so", RTLD_LAZY); |
duke@435 | 1305 | } |
duke@435 | 1306 | |
duke@435 | 1307 | if (handle) { |
duke@435 | 1308 | int (*clock_getres_func)(clockid_t, struct timespec*) = |
duke@435 | 1309 | (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); |
duke@435 | 1310 | int (*clock_gettime_func)(clockid_t, struct timespec*) = |
duke@435 | 1311 | (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); |
duke@435 | 1312 | if (clock_getres_func && clock_gettime_func) { |
duke@435 | 1313 | // See if monotonic clock is supported by the kernel. Note that some |
duke@435 | 1314 | // early implementations simply return kernel jiffies (updated every |
duke@435 | 1315 | // 1/100 or 1/1000 second). It would be bad to use such a low res clock |
duke@435 | 1316 | // for nano time (though the monotonic property is still nice to have). |
duke@435 | 1317 | // It's fixed in newer kernels, however clock_getres() still returns |
duke@435 | 1318 | // 1/HZ. We check if clock_getres() works, but will ignore its reported |
duke@435 | 1319 | // resolution for now. Hopefully as people move to new kernels, this |
duke@435 | 1320 | // won't be a problem. |
duke@435 | 1321 | struct timespec res; |
duke@435 | 1322 | struct timespec tp; |
duke@435 | 1323 | if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && |
duke@435 | 1324 | clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { |
duke@435 | 1325 | // yes, monotonic clock is supported |
duke@435 | 1326 | _clock_gettime = clock_gettime_func; |
duke@435 | 1327 | } else { |
duke@435 | 1328 | // close librt if there is no monotonic clock |
duke@435 | 1329 | dlclose(handle); |
duke@435 | 1330 | } |
duke@435 | 1331 | } |
duke@435 | 1332 | } |
duke@435 | 1333 | } |
duke@435 | 1334 | |
duke@435 | 1335 | #ifndef SYS_clock_getres |
duke@435 | 1336 | |
duke@435 | 1337 | #if defined(IA32) || defined(AMD64) |
duke@435 | 1338 | #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) |
duke@435 | 1339 | #else |
duke@435 | 1340 | #error Value of SYS_clock_getres not known on this platform |
duke@435 | 1341 | #endif |
duke@435 | 1342 | |
duke@435 | 1343 | #endif |
duke@435 | 1344 | |
duke@435 | 1345 | #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) |
duke@435 | 1346 | |
duke@435 | 1347 | void os::Linux::fast_thread_clock_init() { |
duke@435 | 1348 | if (!UseLinuxPosixThreadCPUClocks) { |
duke@435 | 1349 | return; |
duke@435 | 1350 | } |
duke@435 | 1351 | clockid_t clockid; |
duke@435 | 1352 | struct timespec tp; |
duke@435 | 1353 | int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = |
duke@435 | 1354 | (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); |
duke@435 | 1355 | |
duke@435 | 1356 | // Switch to using fast clocks for thread cpu time if |
duke@435 | 1357 | // the sys_clock_getres() returns 0 error code. |
duke@435 | 1358 | // Note, that some kernels may support the current thread |
duke@435 | 1359 | // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks |
duke@435 | 1360 | // returned by the pthread_getcpuclockid(). |
duke@435 | 1361 | // If the fast Posix clocks are supported then the sys_clock_getres() |
duke@435 | 1362 | // must return at least tp.tv_sec == 0 which means a resolution |
duke@435 | 1363 | // better than 1 sec. This is extra check for reliability. |
duke@435 | 1364 | |
duke@435 | 1365 | if(pthread_getcpuclockid_func && |
duke@435 | 1366 | pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && |
duke@435 | 1367 | sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { |
duke@435 | 1368 | |
duke@435 | 1369 | _supports_fast_thread_cpu_time = true; |
duke@435 | 1370 | _pthread_getcpuclockid = pthread_getcpuclockid_func; |
duke@435 | 1371 | } |
duke@435 | 1372 | } |
duke@435 | 1373 | |
duke@435 | 1374 | jlong os::javaTimeNanos() { |
duke@435 | 1375 | if (Linux::supports_monotonic_clock()) { |
duke@435 | 1376 | struct timespec tp; |
duke@435 | 1377 | int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); |
duke@435 | 1378 | assert(status == 0, "gettime error"); |
duke@435 | 1379 | jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); |
duke@435 | 1380 | return result; |
duke@435 | 1381 | } else { |
duke@435 | 1382 | timeval time; |
duke@435 | 1383 | int status = gettimeofday(&time, NULL); |
duke@435 | 1384 | assert(status != -1, "linux error"); |
duke@435 | 1385 | jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); |
duke@435 | 1386 | return 1000 * usecs; |
duke@435 | 1387 | } |
duke@435 | 1388 | } |
duke@435 | 1389 | |
duke@435 | 1390 | void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { |
duke@435 | 1391 | if (Linux::supports_monotonic_clock()) { |
duke@435 | 1392 | info_ptr->max_value = ALL_64_BITS; |
duke@435 | 1393 | |
duke@435 | 1394 | // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past |
duke@435 | 1395 | info_ptr->may_skip_backward = false; // not subject to resetting or drifting |
duke@435 | 1396 | info_ptr->may_skip_forward = false; // not subject to resetting or drifting |
duke@435 | 1397 | } else { |
duke@435 | 1398 | // gettimeofday - based on time in seconds since the Epoch thus does not wrap |
duke@435 | 1399 | info_ptr->max_value = ALL_64_BITS; |
duke@435 | 1400 | |
duke@435 | 1401 | // gettimeofday is a real time clock so it skips |
duke@435 | 1402 | info_ptr->may_skip_backward = true; |
duke@435 | 1403 | info_ptr->may_skip_forward = true; |
duke@435 | 1404 | } |
duke@435 | 1405 | |
duke@435 | 1406 | info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time |
duke@435 | 1407 | } |
duke@435 | 1408 | |
duke@435 | 1409 | // Return the real, user, and system times in seconds from an |
duke@435 | 1410 | // arbitrary fixed point in the past. |
duke@435 | 1411 | bool os::getTimesSecs(double* process_real_time, |
duke@435 | 1412 | double* process_user_time, |
duke@435 | 1413 | double* process_system_time) { |
duke@435 | 1414 | struct tms ticks; |
duke@435 | 1415 | clock_t real_ticks = times(&ticks); |
duke@435 | 1416 | |
duke@435 | 1417 | if (real_ticks == (clock_t) (-1)) { |
duke@435 | 1418 | return false; |
duke@435 | 1419 | } else { |
duke@435 | 1420 | double ticks_per_second = (double) clock_tics_per_sec; |
duke@435 | 1421 | *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; |
duke@435 | 1422 | *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; |
duke@435 | 1423 | *process_real_time = ((double) real_ticks) / ticks_per_second; |
duke@435 | 1424 | |
duke@435 | 1425 | return true; |
duke@435 | 1426 | } |
duke@435 | 1427 | } |
duke@435 | 1428 | |
duke@435 | 1429 | |
duke@435 | 1430 | char * os::local_time_string(char *buf, size_t buflen) { |
duke@435 | 1431 | struct tm t; |
duke@435 | 1432 | time_t long_time; |
duke@435 | 1433 | time(&long_time); |
duke@435 | 1434 | localtime_r(&long_time, &t); |
duke@435 | 1435 | jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", |
duke@435 | 1436 | t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, |
duke@435 | 1437 | t.tm_hour, t.tm_min, t.tm_sec); |
duke@435 | 1438 | return buf; |
duke@435 | 1439 | } |
duke@435 | 1440 | |
ysr@983 | 1441 | struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { |
ysr@983 | 1442 | return localtime_r(clock, res); |
ysr@983 | 1443 | } |
ysr@983 | 1444 | |
duke@435 | 1445 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 1446 | // runtime exit support |
duke@435 | 1447 | |
duke@435 | 1448 | // Note: os::shutdown() might be called very early during initialization, or |
duke@435 | 1449 | // called from signal handler. Before adding something to os::shutdown(), make |
duke@435 | 1450 | // sure it is async-safe and can handle partially initialized VM. |
duke@435 | 1451 | void os::shutdown() { |
duke@435 | 1452 | |
duke@435 | 1453 | // allow PerfMemory to attempt cleanup of any persistent resources |
duke@435 | 1454 | perfMemory_exit(); |
duke@435 | 1455 | |
duke@435 | 1456 | // needs to remove object in file system |
duke@435 | 1457 | AttachListener::abort(); |
duke@435 | 1458 | |
duke@435 | 1459 | // flush buffered output, finish log files |
duke@435 | 1460 | ostream_abort(); |
duke@435 | 1461 | |
duke@435 | 1462 | // Check for abort hook |
duke@435 | 1463 | abort_hook_t abort_hook = Arguments::abort_hook(); |
duke@435 | 1464 | if (abort_hook != NULL) { |
duke@435 | 1465 | abort_hook(); |
duke@435 | 1466 | } |
duke@435 | 1467 | |
duke@435 | 1468 | } |
duke@435 | 1469 | |
duke@435 | 1470 | // Note: os::abort() might be called very early during initialization, or |
duke@435 | 1471 | // called from signal handler. Before adding something to os::abort(), make |
duke@435 | 1472 | // sure it is async-safe and can handle partially initialized VM. |
duke@435 | 1473 | void os::abort(bool dump_core) { |
duke@435 | 1474 | os::shutdown(); |
duke@435 | 1475 | if (dump_core) { |
duke@435 | 1476 | #ifndef PRODUCT |
duke@435 | 1477 | fdStream out(defaultStream::output_fd()); |
duke@435 | 1478 | out.print_raw("Current thread is "); |
duke@435 | 1479 | char buf[16]; |
duke@435 | 1480 | jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); |
duke@435 | 1481 | out.print_raw_cr(buf); |
duke@435 | 1482 | out.print_raw_cr("Dumping core ..."); |
duke@435 | 1483 | #endif |
duke@435 | 1484 | ::abort(); // dump core |
duke@435 | 1485 | } |
duke@435 | 1486 | |
duke@435 | 1487 | ::exit(1); |
duke@435 | 1488 | } |
duke@435 | 1489 | |
duke@435 | 1490 | // Die immediately, no exit hook, no abort hook, no cleanup. |
duke@435 | 1491 | void os::die() { |
duke@435 | 1492 | // _exit() on LinuxThreads only kills current thread |
duke@435 | 1493 | ::abort(); |
duke@435 | 1494 | } |
duke@435 | 1495 | |
duke@435 | 1496 | // unused on linux for now. |
duke@435 | 1497 | void os::set_error_file(const char *logfile) {} |
duke@435 | 1498 | |
duke@435 | 1499 | intx os::current_thread_id() { return (intx)pthread_self(); } |
duke@435 | 1500 | int os::current_process_id() { |
duke@435 | 1501 | |
duke@435 | 1502 | // Under the old linux thread library, linux gives each thread |
duke@435 | 1503 | // its own process id. Because of this each thread will return |
duke@435 | 1504 | // a different pid if this method were to return the result |
duke@435 | 1505 | // of getpid(2). Linux provides no api that returns the pid |
duke@435 | 1506 | // of the launcher thread for the vm. This implementation |
duke@435 | 1507 | // returns a unique pid, the pid of the launcher thread |
duke@435 | 1508 | // that starts the vm 'process'. |
duke@435 | 1509 | |
duke@435 | 1510 | // Under the NPTL, getpid() returns the same pid as the |
duke@435 | 1511 | // launcher thread rather than a unique pid per thread. |
duke@435 | 1512 | // Use gettid() if you want the old pre NPTL behaviour. |
duke@435 | 1513 | |
duke@435 | 1514 | // if you are looking for the result of a call to getpid() that |
duke@435 | 1515 | // returns a unique pid for the calling thread, then look at the |
duke@435 | 1516 | // OSThread::thread_id() method in osThread_linux.hpp file |
duke@435 | 1517 | |
duke@435 | 1518 | return (int)(_initial_pid ? _initial_pid : getpid()); |
duke@435 | 1519 | } |
duke@435 | 1520 | |
duke@435 | 1521 | // DLL functions |
duke@435 | 1522 | |
duke@435 | 1523 | const char* os::dll_file_extension() { return ".so"; } |
duke@435 | 1524 | |
coleenp@1788 | 1525 | const char* os::get_temp_directory() { |
coleenp@1788 | 1526 | const char *prop = Arguments::get_property("java.io.tmpdir"); |
coleenp@1788 | 1527 | return prop == NULL ? "/tmp" : prop; |
coleenp@1788 | 1528 | } |
duke@435 | 1529 | |
phh@1126 | 1530 | static bool file_exists(const char* filename) { |
phh@1126 | 1531 | struct stat statbuf; |
phh@1126 | 1532 | if (filename == NULL || strlen(filename) == 0) { |
phh@1126 | 1533 | return false; |
phh@1126 | 1534 | } |
phh@1126 | 1535 | return os::stat(filename, &statbuf) == 0; |
phh@1126 | 1536 | } |
phh@1126 | 1537 | |
phh@1126 | 1538 | void os::dll_build_name(char* buffer, size_t buflen, |
phh@1126 | 1539 | const char* pname, const char* fname) { |
phh@1126 | 1540 | // Copied from libhpi |
kamg@677 | 1541 | const size_t pnamelen = pname ? strlen(pname) : 0; |
kamg@677 | 1542 | |
phh@1126 | 1543 | // Quietly truncate on buffer overflow. Should be an error. |
kamg@677 | 1544 | if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { |
kamg@677 | 1545 | *buffer = '\0'; |
kamg@677 | 1546 | return; |
kamg@677 | 1547 | } |
kamg@677 | 1548 | |
kamg@677 | 1549 | if (pnamelen == 0) { |
phh@1126 | 1550 | snprintf(buffer, buflen, "lib%s.so", fname); |
phh@1126 | 1551 | } else if (strchr(pname, *os::path_separator()) != NULL) { |
phh@1126 | 1552 | int n; |
phh@1126 | 1553 | char** pelements = split_path(pname, &n); |
phh@1126 | 1554 | for (int i = 0 ; i < n ; i++) { |
phh@1126 | 1555 | // Really shouldn't be NULL, but check can't hurt |
phh@1126 | 1556 | if (pelements[i] == NULL || strlen(pelements[i]) == 0) { |
phh@1126 | 1557 | continue; // skip the empty path values |
phh@1126 | 1558 | } |
phh@1126 | 1559 | snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); |
phh@1126 | 1560 | if (file_exists(buffer)) { |
phh@1126 | 1561 | break; |
phh@1126 | 1562 | } |
phh@1126 | 1563 | } |
phh@1126 | 1564 | // release the storage |
phh@1126 | 1565 | for (int i = 0 ; i < n ; i++) { |
phh@1126 | 1566 | if (pelements[i] != NULL) { |
phh@1126 | 1567 | FREE_C_HEAP_ARRAY(char, pelements[i]); |
phh@1126 | 1568 | } |
phh@1126 | 1569 | } |
phh@1126 | 1570 | if (pelements != NULL) { |
phh@1126 | 1571 | FREE_C_HEAP_ARRAY(char*, pelements); |
phh@1126 | 1572 | } |
kamg@677 | 1573 | } else { |
phh@1126 | 1574 | snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); |
kamg@677 | 1575 | } |
kamg@677 | 1576 | } |
kamg@677 | 1577 | |
duke@435 | 1578 | const char* os::get_current_directory(char *buf, int buflen) { |
duke@435 | 1579 | return getcwd(buf, buflen); |
duke@435 | 1580 | } |
duke@435 | 1581 | |
duke@435 | 1582 | // check if addr is inside libjvm[_g].so |
duke@435 | 1583 | bool os::address_is_in_vm(address addr) { |
duke@435 | 1584 | static address libjvm_base_addr; |
duke@435 | 1585 | Dl_info dlinfo; |
duke@435 | 1586 | |
duke@435 | 1587 | if (libjvm_base_addr == NULL) { |
duke@435 | 1588 | dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); |
duke@435 | 1589 | libjvm_base_addr = (address)dlinfo.dli_fbase; |
duke@435 | 1590 | assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); |
duke@435 | 1591 | } |
duke@435 | 1592 | |
duke@435 | 1593 | if (dladdr((void *)addr, &dlinfo)) { |
duke@435 | 1594 | if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; |
duke@435 | 1595 | } |
duke@435 | 1596 | |
duke@435 | 1597 | return false; |
duke@435 | 1598 | } |
duke@435 | 1599 | |
duke@435 | 1600 | bool os::dll_address_to_function_name(address addr, char *buf, |
duke@435 | 1601 | int buflen, int *offset) { |
duke@435 | 1602 | Dl_info dlinfo; |
duke@435 | 1603 | |
duke@435 | 1604 | if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { |
duke@435 | 1605 | if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); |
duke@435 | 1606 | if (offset) *offset = addr - (address)dlinfo.dli_saddr; |
duke@435 | 1607 | return true; |
duke@435 | 1608 | } else { |
duke@435 | 1609 | if (buf) buf[0] = '\0'; |
duke@435 | 1610 | if (offset) *offset = -1; |
duke@435 | 1611 | return false; |
duke@435 | 1612 | } |
duke@435 | 1613 | } |
duke@435 | 1614 | |
duke@435 | 1615 | struct _address_to_library_name { |
duke@435 | 1616 | address addr; // input : memory address |
duke@435 | 1617 | size_t buflen; // size of fname |
duke@435 | 1618 | char* fname; // output: library name |
duke@435 | 1619 | address base; // library base addr |
duke@435 | 1620 | }; |
duke@435 | 1621 | |
duke@435 | 1622 | static int address_to_library_name_callback(struct dl_phdr_info *info, |
duke@435 | 1623 | size_t size, void *data) { |
duke@435 | 1624 | int i; |
duke@435 | 1625 | bool found = false; |
duke@435 | 1626 | address libbase = NULL; |
duke@435 | 1627 | struct _address_to_library_name * d = (struct _address_to_library_name *)data; |
duke@435 | 1628 | |
duke@435 | 1629 | // iterate through all loadable segments |
duke@435 | 1630 | for (i = 0; i < info->dlpi_phnum; i++) { |
duke@435 | 1631 | address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); |
duke@435 | 1632 | if (info->dlpi_phdr[i].p_type == PT_LOAD) { |
duke@435 | 1633 | // base address of a library is the lowest address of its loaded |
duke@435 | 1634 | // segments. |
duke@435 | 1635 | if (libbase == NULL || libbase > segbase) { |
duke@435 | 1636 | libbase = segbase; |
duke@435 | 1637 | } |
duke@435 | 1638 | // see if 'addr' is within current segment |
duke@435 | 1639 | if (segbase <= d->addr && |
duke@435 | 1640 | d->addr < segbase + info->dlpi_phdr[i].p_memsz) { |
duke@435 | 1641 | found = true; |
duke@435 | 1642 | } |
duke@435 | 1643 | } |
duke@435 | 1644 | } |
duke@435 | 1645 | |
duke@435 | 1646 | // dlpi_name is NULL or empty if the ELF file is executable, return 0 |
duke@435 | 1647 | // so dll_address_to_library_name() can fall through to use dladdr() which |
duke@435 | 1648 | // can figure out executable name from argv[0]. |
duke@435 | 1649 | if (found && info->dlpi_name && info->dlpi_name[0]) { |
duke@435 | 1650 | d->base = libbase; |
duke@435 | 1651 | if (d->fname) { |
duke@435 | 1652 | jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); |
duke@435 | 1653 | } |
duke@435 | 1654 | return 1; |
duke@435 | 1655 | } |
duke@435 | 1656 | return 0; |
duke@435 | 1657 | } |
duke@435 | 1658 | |
duke@435 | 1659 | bool os::dll_address_to_library_name(address addr, char* buf, |
duke@435 | 1660 | int buflen, int* offset) { |
duke@435 | 1661 | Dl_info dlinfo; |
duke@435 | 1662 | struct _address_to_library_name data; |
duke@435 | 1663 | |
duke@435 | 1664 | // There is a bug in old glibc dladdr() implementation that it could resolve |
duke@435 | 1665 | // to wrong library name if the .so file has a base address != NULL. Here |
duke@435 | 1666 | // we iterate through the program headers of all loaded libraries to find |
duke@435 | 1667 | // out which library 'addr' really belongs to. This workaround can be |
duke@435 | 1668 | // removed once the minimum requirement for glibc is moved to 2.3.x. |
duke@435 | 1669 | data.addr = addr; |
duke@435 | 1670 | data.fname = buf; |
duke@435 | 1671 | data.buflen = buflen; |
duke@435 | 1672 | data.base = NULL; |
duke@435 | 1673 | int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); |
duke@435 | 1674 | |
duke@435 | 1675 | if (rslt) { |
duke@435 | 1676 | // buf already contains library name |
duke@435 | 1677 | if (offset) *offset = addr - data.base; |
duke@435 | 1678 | return true; |
duke@435 | 1679 | } else if (dladdr((void*)addr, &dlinfo)){ |
duke@435 | 1680 | if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); |
duke@435 | 1681 | if (offset) *offset = addr - (address)dlinfo.dli_fbase; |
duke@435 | 1682 | return true; |
duke@435 | 1683 | } else { |
duke@435 | 1684 | if (buf) buf[0] = '\0'; |
duke@435 | 1685 | if (offset) *offset = -1; |
duke@435 | 1686 | return false; |
duke@435 | 1687 | } |
duke@435 | 1688 | } |
duke@435 | 1689 | |
duke@435 | 1690 | // Loads .dll/.so and |
duke@435 | 1691 | // in case of error it checks if .dll/.so was built for the |
duke@435 | 1692 | // same architecture as Hotspot is running on |
duke@435 | 1693 | |
duke@435 | 1694 | void * os::dll_load(const char *filename, char *ebuf, int ebuflen) |
duke@435 | 1695 | { |
duke@435 | 1696 | void * result= ::dlopen(filename, RTLD_LAZY); |
duke@435 | 1697 | if (result != NULL) { |
duke@435 | 1698 | // Successful loading |
duke@435 | 1699 | return result; |
duke@435 | 1700 | } |
duke@435 | 1701 | |
duke@435 | 1702 | Elf32_Ehdr elf_head; |
duke@435 | 1703 | |
duke@435 | 1704 | // Read system error message into ebuf |
duke@435 | 1705 | // It may or may not be overwritten below |
duke@435 | 1706 | ::strncpy(ebuf, ::dlerror(), ebuflen-1); |
duke@435 | 1707 | ebuf[ebuflen-1]='\0'; |
duke@435 | 1708 | int diag_msg_max_length=ebuflen-strlen(ebuf); |
duke@435 | 1709 | char* diag_msg_buf=ebuf+strlen(ebuf); |
duke@435 | 1710 | |
duke@435 | 1711 | if (diag_msg_max_length==0) { |
duke@435 | 1712 | // No more space in ebuf for additional diagnostics message |
duke@435 | 1713 | return NULL; |
duke@435 | 1714 | } |
duke@435 | 1715 | |
duke@435 | 1716 | |
duke@435 | 1717 | int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); |
duke@435 | 1718 | |
duke@435 | 1719 | if (file_descriptor < 0) { |
duke@435 | 1720 | // Can't open library, report dlerror() message |
duke@435 | 1721 | return NULL; |
duke@435 | 1722 | } |
duke@435 | 1723 | |
duke@435 | 1724 | bool failed_to_read_elf_head= |
duke@435 | 1725 | (sizeof(elf_head)!= |
duke@435 | 1726 | (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; |
duke@435 | 1727 | |
duke@435 | 1728 | ::close(file_descriptor); |
duke@435 | 1729 | if (failed_to_read_elf_head) { |
duke@435 | 1730 | // file i/o error - report dlerror() msg |
duke@435 | 1731 | return NULL; |
duke@435 | 1732 | } |
duke@435 | 1733 | |
duke@435 | 1734 | typedef struct { |
duke@435 | 1735 | Elf32_Half code; // Actual value as defined in elf.h |
duke@435 | 1736 | Elf32_Half compat_class; // Compatibility of archs at VM's sense |
duke@435 | 1737 | char elf_class; // 32 or 64 bit |
duke@435 | 1738 | char endianess; // MSB or LSB |
duke@435 | 1739 | char* name; // String representation |
duke@435 | 1740 | } arch_t; |
duke@435 | 1741 | |
duke@435 | 1742 | #ifndef EM_486 |
duke@435 | 1743 | #define EM_486 6 /* Intel 80486 */ |
duke@435 | 1744 | #endif |
duke@435 | 1745 | |
duke@435 | 1746 | static const arch_t arch_array[]={ |
duke@435 | 1747 | {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, |
duke@435 | 1748 | {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, |
duke@435 | 1749 | {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, |
duke@435 | 1750 | {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, |
duke@435 | 1751 | {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, |
duke@435 | 1752 | {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, |
duke@435 | 1753 | {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, |
duke@435 | 1754 | {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, |
never@1445 | 1755 | {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, |
never@1445 | 1756 | {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, |
never@1445 | 1757 | {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, |
never@1445 | 1758 | {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, |
never@1445 | 1759 | {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, |
never@1445 | 1760 | {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, |
never@1445 | 1761 | {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, |
never@1445 | 1762 | {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} |
duke@435 | 1763 | }; |
duke@435 | 1764 | |
duke@435 | 1765 | #if (defined IA32) |
duke@435 | 1766 | static Elf32_Half running_arch_code=EM_386; |
duke@435 | 1767 | #elif (defined AMD64) |
duke@435 | 1768 | static Elf32_Half running_arch_code=EM_X86_64; |
duke@435 | 1769 | #elif (defined IA64) |
duke@435 | 1770 | static Elf32_Half running_arch_code=EM_IA_64; |
duke@435 | 1771 | #elif (defined __sparc) && (defined _LP64) |
duke@435 | 1772 | static Elf32_Half running_arch_code=EM_SPARCV9; |
duke@435 | 1773 | #elif (defined __sparc) && (!defined _LP64) |
duke@435 | 1774 | static Elf32_Half running_arch_code=EM_SPARC; |
duke@435 | 1775 | #elif (defined __powerpc64__) |
duke@435 | 1776 | static Elf32_Half running_arch_code=EM_PPC64; |
duke@435 | 1777 | #elif (defined __powerpc__) |
duke@435 | 1778 | static Elf32_Half running_arch_code=EM_PPC; |
never@1445 | 1779 | #elif (defined ARM) |
never@1445 | 1780 | static Elf32_Half running_arch_code=EM_ARM; |
never@1445 | 1781 | #elif (defined S390) |
never@1445 | 1782 | static Elf32_Half running_arch_code=EM_S390; |
never@1445 | 1783 | #elif (defined ALPHA) |
never@1445 | 1784 | static Elf32_Half running_arch_code=EM_ALPHA; |
never@1445 | 1785 | #elif (defined MIPSEL) |
never@1445 | 1786 | static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; |
never@1445 | 1787 | #elif (defined PARISC) |
never@1445 | 1788 | static Elf32_Half running_arch_code=EM_PARISC; |
never@1445 | 1789 | #elif (defined MIPS) |
never@1445 | 1790 | static Elf32_Half running_arch_code=EM_MIPS; |
never@1445 | 1791 | #elif (defined M68K) |
never@1445 | 1792 | static Elf32_Half running_arch_code=EM_68K; |
duke@435 | 1793 | #else |
duke@435 | 1794 | #error Method os::dll_load requires that one of following is defined:\ |
never@1445 | 1795 | IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K |
duke@435 | 1796 | #endif |
duke@435 | 1797 | |
duke@435 | 1798 | // Identify compatability class for VM's architecture and library's architecture |
duke@435 | 1799 | // Obtain string descriptions for architectures |
duke@435 | 1800 | |
duke@435 | 1801 | arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; |
duke@435 | 1802 | int running_arch_index=-1; |
duke@435 | 1803 | |
duke@435 | 1804 | for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { |
duke@435 | 1805 | if (running_arch_code == arch_array[i].code) { |
duke@435 | 1806 | running_arch_index = i; |
duke@435 | 1807 | } |
duke@435 | 1808 | if (lib_arch.code == arch_array[i].code) { |
duke@435 | 1809 | lib_arch.compat_class = arch_array[i].compat_class; |
duke@435 | 1810 | lib_arch.name = arch_array[i].name; |
duke@435 | 1811 | } |
duke@435 | 1812 | } |
duke@435 | 1813 | |
duke@435 | 1814 | assert(running_arch_index != -1, |
duke@435 | 1815 | "Didn't find running architecture code (running_arch_code) in arch_array"); |
duke@435 | 1816 | if (running_arch_index == -1) { |
duke@435 | 1817 | // Even though running architecture detection failed |
duke@435 | 1818 | // we may still continue with reporting dlerror() message |
duke@435 | 1819 | return NULL; |
duke@435 | 1820 | } |
duke@435 | 1821 | |
duke@435 | 1822 | if (lib_arch.endianess != arch_array[running_arch_index].endianess) { |
duke@435 | 1823 | ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); |
duke@435 | 1824 | return NULL; |
duke@435 | 1825 | } |
duke@435 | 1826 | |
never@1445 | 1827 | #ifndef S390 |
duke@435 | 1828 | if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { |
duke@435 | 1829 | ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); |
duke@435 | 1830 | return NULL; |
duke@435 | 1831 | } |
never@1445 | 1832 | #endif // !S390 |
duke@435 | 1833 | |
duke@435 | 1834 | if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { |
duke@435 | 1835 | if ( lib_arch.name!=NULL ) { |
duke@435 | 1836 | ::snprintf(diag_msg_buf, diag_msg_max_length-1, |
duke@435 | 1837 | " (Possible cause: can't load %s-bit .so on a %s-bit platform)", |
duke@435 | 1838 | lib_arch.name, arch_array[running_arch_index].name); |
duke@435 | 1839 | } else { |
duke@435 | 1840 | ::snprintf(diag_msg_buf, diag_msg_max_length-1, |
duke@435 | 1841 | " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", |
duke@435 | 1842 | lib_arch.code, |
duke@435 | 1843 | arch_array[running_arch_index].name); |
duke@435 | 1844 | } |
duke@435 | 1845 | } |
duke@435 | 1846 | |
duke@435 | 1847 | return NULL; |
duke@435 | 1848 | } |
duke@435 | 1849 | |
kamg@677 | 1850 | /* |
kamg@677 | 1851 | * glibc-2.0 libdl is not MT safe. If you are building with any glibc, |
kamg@677 | 1852 | * chances are you might want to run the generated bits against glibc-2.0 |
kamg@677 | 1853 | * libdl.so, so always use locking for any version of glibc. |
kamg@677 | 1854 | */ |
kamg@677 | 1855 | void* os::dll_lookup(void* handle, const char* name) { |
kamg@677 | 1856 | pthread_mutex_lock(&dl_mutex); |
kamg@677 | 1857 | void* res = dlsym(handle, name); |
kamg@677 | 1858 | pthread_mutex_unlock(&dl_mutex); |
kamg@677 | 1859 | return res; |
kamg@677 | 1860 | } |
duke@435 | 1861 | |
duke@435 | 1862 | |
duke@435 | 1863 | bool _print_ascii_file(const char* filename, outputStream* st) { |
duke@435 | 1864 | int fd = open(filename, O_RDONLY); |
duke@435 | 1865 | if (fd == -1) { |
duke@435 | 1866 | return false; |
duke@435 | 1867 | } |
duke@435 | 1868 | |
duke@435 | 1869 | char buf[32]; |
duke@435 | 1870 | int bytes; |
duke@435 | 1871 | while ((bytes = read(fd, buf, sizeof(buf))) > 0) { |
duke@435 | 1872 | st->print_raw(buf, bytes); |
duke@435 | 1873 | } |
duke@435 | 1874 | |
duke@435 | 1875 | close(fd); |
duke@435 | 1876 | |
duke@435 | 1877 | return true; |
duke@435 | 1878 | } |
duke@435 | 1879 | |
duke@435 | 1880 | void os::print_dll_info(outputStream *st) { |
duke@435 | 1881 | st->print_cr("Dynamic libraries:"); |
duke@435 | 1882 | |
duke@435 | 1883 | char fname[32]; |
duke@435 | 1884 | pid_t pid = os::Linux::gettid(); |
duke@435 | 1885 | |
duke@435 | 1886 | jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); |
duke@435 | 1887 | |
duke@435 | 1888 | if (!_print_ascii_file(fname, st)) { |
duke@435 | 1889 | st->print("Can not get library information for pid = %d\n", pid); |
duke@435 | 1890 | } |
duke@435 | 1891 | } |
duke@435 | 1892 | |
duke@435 | 1893 | |
duke@435 | 1894 | void os::print_os_info(outputStream* st) { |
duke@435 | 1895 | st->print("OS:"); |
duke@435 | 1896 | |
duke@435 | 1897 | // Try to identify popular distros. |
duke@435 | 1898 | // Most Linux distributions have /etc/XXX-release file, which contains |
duke@435 | 1899 | // the OS version string. Some have more than one /etc/XXX-release file |
duke@435 | 1900 | // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), |
duke@435 | 1901 | // so the order is important. |
duke@435 | 1902 | if (!_print_ascii_file("/etc/mandrake-release", st) && |
duke@435 | 1903 | !_print_ascii_file("/etc/sun-release", st) && |
duke@435 | 1904 | !_print_ascii_file("/etc/redhat-release", st) && |
duke@435 | 1905 | !_print_ascii_file("/etc/SuSE-release", st) && |
duke@435 | 1906 | !_print_ascii_file("/etc/turbolinux-release", st) && |
duke@435 | 1907 | !_print_ascii_file("/etc/gentoo-release", st) && |
duke@435 | 1908 | !_print_ascii_file("/etc/debian_version", st)) { |
duke@435 | 1909 | st->print("Linux"); |
duke@435 | 1910 | } |
duke@435 | 1911 | st->cr(); |
duke@435 | 1912 | |
duke@435 | 1913 | // kernel |
duke@435 | 1914 | st->print("uname:"); |
duke@435 | 1915 | struct utsname name; |
duke@435 | 1916 | uname(&name); |
duke@435 | 1917 | st->print(name.sysname); st->print(" "); |
duke@435 | 1918 | st->print(name.release); st->print(" "); |
duke@435 | 1919 | st->print(name.version); st->print(" "); |
duke@435 | 1920 | st->print(name.machine); |
duke@435 | 1921 | st->cr(); |
duke@435 | 1922 | |
duke@435 | 1923 | // Print warning if unsafe chroot environment detected |
duke@435 | 1924 | if (unsafe_chroot_detected) { |
duke@435 | 1925 | st->print("WARNING!! "); |
duke@435 | 1926 | st->print_cr(unstable_chroot_error); |
duke@435 | 1927 | } |
duke@435 | 1928 | |
duke@435 | 1929 | // libc, pthread |
duke@435 | 1930 | st->print("libc:"); |
duke@435 | 1931 | st->print(os::Linux::glibc_version()); st->print(" "); |
duke@435 | 1932 | st->print(os::Linux::libpthread_version()); st->print(" "); |
duke@435 | 1933 | if (os::Linux::is_LinuxThreads()) { |
duke@435 | 1934 | st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); |
duke@435 | 1935 | } |
duke@435 | 1936 | st->cr(); |
duke@435 | 1937 | |
duke@435 | 1938 | // rlimit |
duke@435 | 1939 | st->print("rlimit:"); |
duke@435 | 1940 | struct rlimit rlim; |
duke@435 | 1941 | |
duke@435 | 1942 | st->print(" STACK "); |
duke@435 | 1943 | getrlimit(RLIMIT_STACK, &rlim); |
duke@435 | 1944 | if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); |
duke@435 | 1945 | else st->print("%uk", rlim.rlim_cur >> 10); |
duke@435 | 1946 | |
duke@435 | 1947 | st->print(", CORE "); |
duke@435 | 1948 | getrlimit(RLIMIT_CORE, &rlim); |
duke@435 | 1949 | if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); |
duke@435 | 1950 | else st->print("%uk", rlim.rlim_cur >> 10); |
duke@435 | 1951 | |
duke@435 | 1952 | st->print(", NPROC "); |
duke@435 | 1953 | getrlimit(RLIMIT_NPROC, &rlim); |
duke@435 | 1954 | if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); |
duke@435 | 1955 | else st->print("%d", rlim.rlim_cur); |
duke@435 | 1956 | |
duke@435 | 1957 | st->print(", NOFILE "); |
duke@435 | 1958 | getrlimit(RLIMIT_NOFILE, &rlim); |
duke@435 | 1959 | if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); |
duke@435 | 1960 | else st->print("%d", rlim.rlim_cur); |
duke@435 | 1961 | |
duke@435 | 1962 | st->print(", AS "); |
duke@435 | 1963 | getrlimit(RLIMIT_AS, &rlim); |
duke@435 | 1964 | if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); |
duke@435 | 1965 | else st->print("%uk", rlim.rlim_cur >> 10); |
duke@435 | 1966 | st->cr(); |
duke@435 | 1967 | |
duke@435 | 1968 | // load average |
duke@435 | 1969 | st->print("load average:"); |
duke@435 | 1970 | double loadavg[3]; |
duke@435 | 1971 | os::loadavg(loadavg, 3); |
duke@435 | 1972 | st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); |
duke@435 | 1973 | st->cr(); |
duke@435 | 1974 | } |
duke@435 | 1975 | |
duke@435 | 1976 | void os::print_memory_info(outputStream* st) { |
duke@435 | 1977 | |
duke@435 | 1978 | st->print("Memory:"); |
duke@435 | 1979 | st->print(" %dk page", os::vm_page_size()>>10); |
duke@435 | 1980 | |
duke@435 | 1981 | // values in struct sysinfo are "unsigned long" |
duke@435 | 1982 | struct sysinfo si; |
duke@435 | 1983 | sysinfo(&si); |
duke@435 | 1984 | |
duke@435 | 1985 | st->print(", physical " UINT64_FORMAT "k", |
duke@435 | 1986 | os::physical_memory() >> 10); |
duke@435 | 1987 | st->print("(" UINT64_FORMAT "k free)", |
duke@435 | 1988 | os::available_memory() >> 10); |
duke@435 | 1989 | st->print(", swap " UINT64_FORMAT "k", |
duke@435 | 1990 | ((jlong)si.totalswap * si.mem_unit) >> 10); |
duke@435 | 1991 | st->print("(" UINT64_FORMAT "k free)", |
duke@435 | 1992 | ((jlong)si.freeswap * si.mem_unit) >> 10); |
duke@435 | 1993 | st->cr(); |
duke@435 | 1994 | } |
duke@435 | 1995 | |
duke@435 | 1996 | // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific |
duke@435 | 1997 | // but they're the same for all the linux arch that we support |
duke@435 | 1998 | // and they're the same for solaris but there's no common place to put this. |
duke@435 | 1999 | const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", |
duke@435 | 2000 | "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", |
duke@435 | 2001 | "ILL_COPROC", "ILL_BADSTK" }; |
duke@435 | 2002 | |
duke@435 | 2003 | const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", |
duke@435 | 2004 | "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", |
duke@435 | 2005 | "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; |
duke@435 | 2006 | |
duke@435 | 2007 | const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; |
duke@435 | 2008 | |
duke@435 | 2009 | const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; |
duke@435 | 2010 | |
duke@435 | 2011 | void os::print_siginfo(outputStream* st, void* siginfo) { |
duke@435 | 2012 | st->print("siginfo:"); |
duke@435 | 2013 | |
duke@435 | 2014 | const int buflen = 100; |
duke@435 | 2015 | char buf[buflen]; |
duke@435 | 2016 | siginfo_t *si = (siginfo_t*)siginfo; |
duke@435 | 2017 | st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); |
duke@435 | 2018 | if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { |
duke@435 | 2019 | st->print("si_errno=%s", buf); |
duke@435 | 2020 | } else { |
duke@435 | 2021 | st->print("si_errno=%d", si->si_errno); |
duke@435 | 2022 | } |
duke@435 | 2023 | const int c = si->si_code; |
duke@435 | 2024 | assert(c > 0, "unexpected si_code"); |
duke@435 | 2025 | switch (si->si_signo) { |
duke@435 | 2026 | case SIGILL: |
duke@435 | 2027 | st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); |
duke@435 | 2028 | st->print(", si_addr=" PTR_FORMAT, si->si_addr); |
duke@435 | 2029 | break; |
duke@435 | 2030 | case SIGFPE: |
duke@435 | 2031 | st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); |
duke@435 | 2032 | st->print(", si_addr=" PTR_FORMAT, si->si_addr); |
duke@435 | 2033 | break; |
duke@435 | 2034 | case SIGSEGV: |
duke@435 | 2035 | st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); |
duke@435 | 2036 | st->print(", si_addr=" PTR_FORMAT, si->si_addr); |
duke@435 | 2037 | break; |
duke@435 | 2038 | case SIGBUS: |
duke@435 | 2039 | st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); |
duke@435 | 2040 | st->print(", si_addr=" PTR_FORMAT, si->si_addr); |
duke@435 | 2041 | break; |
duke@435 | 2042 | default: |
duke@435 | 2043 | st->print(", si_code=%d", si->si_code); |
duke@435 | 2044 | // no si_addr |
duke@435 | 2045 | } |
duke@435 | 2046 | |
duke@435 | 2047 | if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && |
duke@435 | 2048 | UseSharedSpaces) { |
duke@435 | 2049 | FileMapInfo* mapinfo = FileMapInfo::current_info(); |
duke@435 | 2050 | if (mapinfo->is_in_shared_space(si->si_addr)) { |
duke@435 | 2051 | st->print("\n\nError accessing class data sharing archive." \ |
duke@435 | 2052 | " Mapped file inaccessible during execution, " \ |
duke@435 | 2053 | " possible disk/network problem."); |
duke@435 | 2054 | } |
duke@435 | 2055 | } |
duke@435 | 2056 | st->cr(); |
duke@435 | 2057 | } |
duke@435 | 2058 | |
duke@435 | 2059 | |
duke@435 | 2060 | static void print_signal_handler(outputStream* st, int sig, |
duke@435 | 2061 | char* buf, size_t buflen); |
duke@435 | 2062 | |
duke@435 | 2063 | void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { |
duke@435 | 2064 | st->print_cr("Signal Handlers:"); |
duke@435 | 2065 | print_signal_handler(st, SIGSEGV, buf, buflen); |
duke@435 | 2066 | print_signal_handler(st, SIGBUS , buf, buflen); |
duke@435 | 2067 | print_signal_handler(st, SIGFPE , buf, buflen); |
duke@435 | 2068 | print_signal_handler(st, SIGPIPE, buf, buflen); |
duke@435 | 2069 | print_signal_handler(st, SIGXFSZ, buf, buflen); |
duke@435 | 2070 | print_signal_handler(st, SIGILL , buf, buflen); |
duke@435 | 2071 | print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); |
duke@435 | 2072 | print_signal_handler(st, SR_signum, buf, buflen); |
duke@435 | 2073 | print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); |
duke@435 | 2074 | print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); |
duke@435 | 2075 | print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); |
duke@435 | 2076 | print_signal_handler(st, BREAK_SIGNAL, buf, buflen); |
duke@435 | 2077 | } |
duke@435 | 2078 | |
duke@435 | 2079 | static char saved_jvm_path[MAXPATHLEN] = {0}; |
duke@435 | 2080 | |
duke@435 | 2081 | // Find the full path to the current module, libjvm.so or libjvm_g.so |
duke@435 | 2082 | void os::jvm_path(char *buf, jint len) { |
duke@435 | 2083 | // Error checking. |
duke@435 | 2084 | if (len < MAXPATHLEN) { |
duke@435 | 2085 | assert(false, "must use a large-enough buffer"); |
duke@435 | 2086 | buf[0] = '\0'; |
duke@435 | 2087 | return; |
duke@435 | 2088 | } |
duke@435 | 2089 | // Lazy resolve the path to current module. |
duke@435 | 2090 | if (saved_jvm_path[0] != 0) { |
duke@435 | 2091 | strcpy(buf, saved_jvm_path); |
duke@435 | 2092 | return; |
duke@435 | 2093 | } |
duke@435 | 2094 | |
duke@435 | 2095 | char dli_fname[MAXPATHLEN]; |
duke@435 | 2096 | bool ret = dll_address_to_library_name( |
duke@435 | 2097 | CAST_FROM_FN_PTR(address, os::jvm_path), |
duke@435 | 2098 | dli_fname, sizeof(dli_fname), NULL); |
duke@435 | 2099 | assert(ret != 0, "cannot locate libjvm"); |
xlu@948 | 2100 | if (realpath(dli_fname, buf) == NULL) |
xlu@948 | 2101 | return; |
duke@435 | 2102 | |
duke@435 | 2103 | if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { |
duke@435 | 2104 | // Support for the gamma launcher. Typical value for buf is |
duke@435 | 2105 | // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at |
duke@435 | 2106 | // the right place in the string, then assume we are installed in a JDK and |
duke@435 | 2107 | // we're done. Otherwise, check for a JAVA_HOME environment variable and fix |
duke@435 | 2108 | // up the path so it looks like libjvm.so is installed there (append a |
duke@435 | 2109 | // fake suffix hotspot/libjvm.so). |
duke@435 | 2110 | const char *p = buf + strlen(buf) - 1; |
duke@435 | 2111 | for (int count = 0; p > buf && count < 5; ++count) { |
duke@435 | 2112 | for (--p; p > buf && *p != '/'; --p) |
duke@435 | 2113 | /* empty */ ; |
duke@435 | 2114 | } |
duke@435 | 2115 | |
duke@435 | 2116 | if (strncmp(p, "/jre/lib/", 9) != 0) { |
duke@435 | 2117 | // Look for JAVA_HOME in the environment. |
duke@435 | 2118 | char* java_home_var = ::getenv("JAVA_HOME"); |
duke@435 | 2119 | if (java_home_var != NULL && java_home_var[0] != 0) { |
duke@435 | 2120 | // Check the current module name "libjvm.so" or "libjvm_g.so". |
duke@435 | 2121 | p = strrchr(buf, '/'); |
duke@435 | 2122 | assert(strstr(p, "/libjvm") == p, "invalid library name"); |
duke@435 | 2123 | p = strstr(p, "_g") ? "_g" : ""; |
duke@435 | 2124 | |
xlu@948 | 2125 | if (realpath(java_home_var, buf) == NULL) |
xlu@948 | 2126 | return; |
duke@435 | 2127 | sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); |
duke@435 | 2128 | if (0 == access(buf, F_OK)) { |
duke@435 | 2129 | // Use current module name "libjvm[_g].so" instead of |
duke@435 | 2130 | // "libjvm"debug_only("_g")".so" since for fastdebug version |
duke@435 | 2131 | // we should have "libjvm.so" but debug_only("_g") adds "_g"! |
duke@435 | 2132 | // It is used when we are choosing the HPI library's name |
duke@435 | 2133 | // "libhpi[_g].so" in hpi::initialize_get_interface(). |
duke@435 | 2134 | sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); |
duke@435 | 2135 | } else { |
duke@435 | 2136 | // Go back to path of .so |
xlu@948 | 2137 | if (realpath(dli_fname, buf) == NULL) |
xlu@948 | 2138 | return; |
duke@435 | 2139 | } |
duke@435 | 2140 | } |
duke@435 | 2141 | } |
duke@435 | 2142 | } |
duke@435 | 2143 | |
duke@435 | 2144 | strcpy(saved_jvm_path, buf); |
duke@435 | 2145 | } |
duke@435 | 2146 | |
duke@435 | 2147 | void os::print_jni_name_prefix_on(outputStream* st, int args_size) { |
duke@435 | 2148 | // no prefix required, not even "_" |
duke@435 | 2149 | } |
duke@435 | 2150 | |
duke@435 | 2151 | void os::print_jni_name_suffix_on(outputStream* st, int args_size) { |
duke@435 | 2152 | // no suffix required |
duke@435 | 2153 | } |
duke@435 | 2154 | |
duke@435 | 2155 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 2156 | // sun.misc.Signal support |
duke@435 | 2157 | |
duke@435 | 2158 | static volatile jint sigint_count = 0; |
duke@435 | 2159 | |
duke@435 | 2160 | static void |
duke@435 | 2161 | UserHandler(int sig, void *siginfo, void *context) { |
duke@435 | 2162 | // 4511530 - sem_post is serialized and handled by the manager thread. When |
duke@435 | 2163 | // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We |
duke@435 | 2164 | // don't want to flood the manager thread with sem_post requests. |
duke@435 | 2165 | if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) |
duke@435 | 2166 | return; |
duke@435 | 2167 | |
duke@435 | 2168 | // Ctrl-C is pressed during error reporting, likely because the error |
duke@435 | 2169 | // handler fails to abort. Let VM die immediately. |
duke@435 | 2170 | if (sig == SIGINT && is_error_reported()) { |
duke@435 | 2171 | os::die(); |
duke@435 | 2172 | } |
duke@435 | 2173 | |
duke@435 | 2174 | os::signal_notify(sig); |
duke@435 | 2175 | } |
duke@435 | 2176 | |
duke@435 | 2177 | void* os::user_handler() { |
duke@435 | 2178 | return CAST_FROM_FN_PTR(void*, UserHandler); |
duke@435 | 2179 | } |
duke@435 | 2180 | |
duke@435 | 2181 | extern "C" { |
duke@435 | 2182 | typedef void (*sa_handler_t)(int); |
duke@435 | 2183 | typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); |
duke@435 | 2184 | } |
duke@435 | 2185 | |
duke@435 | 2186 | void* os::signal(int signal_number, void* handler) { |
duke@435 | 2187 | struct sigaction sigAct, oldSigAct; |
duke@435 | 2188 | |
duke@435 | 2189 | sigfillset(&(sigAct.sa_mask)); |
duke@435 | 2190 | sigAct.sa_flags = SA_RESTART|SA_SIGINFO; |
duke@435 | 2191 | sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); |
duke@435 | 2192 | |
duke@435 | 2193 | if (sigaction(signal_number, &sigAct, &oldSigAct)) { |
duke@435 | 2194 | // -1 means registration failed |
duke@435 | 2195 | return (void *)-1; |
duke@435 | 2196 | } |
duke@435 | 2197 | |
duke@435 | 2198 | return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); |
duke@435 | 2199 | } |
duke@435 | 2200 | |
duke@435 | 2201 | void os::signal_raise(int signal_number) { |
duke@435 | 2202 | ::raise(signal_number); |
duke@435 | 2203 | } |
duke@435 | 2204 | |
duke@435 | 2205 | /* |
duke@435 | 2206 | * The following code is moved from os.cpp for making this |
duke@435 | 2207 | * code platform specific, which it is by its very nature. |
duke@435 | 2208 | */ |
duke@435 | 2209 | |
duke@435 | 2210 | // Will be modified when max signal is changed to be dynamic |
duke@435 | 2211 | int os::sigexitnum_pd() { |
duke@435 | 2212 | return NSIG; |
duke@435 | 2213 | } |
duke@435 | 2214 | |
duke@435 | 2215 | // a counter for each possible signal value |
duke@435 | 2216 | static volatile jint pending_signals[NSIG+1] = { 0 }; |
duke@435 | 2217 | |
duke@435 | 2218 | // Linux(POSIX) specific hand shaking semaphore. |
duke@435 | 2219 | static sem_t sig_sem; |
duke@435 | 2220 | |
duke@435 | 2221 | void os::signal_init_pd() { |
duke@435 | 2222 | // Initialize signal structures |
duke@435 | 2223 | ::memset((void*)pending_signals, 0, sizeof(pending_signals)); |
duke@435 | 2224 | |
duke@435 | 2225 | // Initialize signal semaphore |
duke@435 | 2226 | ::sem_init(&sig_sem, 0, 0); |
duke@435 | 2227 | } |
duke@435 | 2228 | |
duke@435 | 2229 | void os::signal_notify(int sig) { |
duke@435 | 2230 | Atomic::inc(&pending_signals[sig]); |
duke@435 | 2231 | ::sem_post(&sig_sem); |
duke@435 | 2232 | } |
duke@435 | 2233 | |
duke@435 | 2234 | static int check_pending_signals(bool wait) { |
duke@435 | 2235 | Atomic::store(0, &sigint_count); |
duke@435 | 2236 | for (;;) { |
duke@435 | 2237 | for (int i = 0; i < NSIG + 1; i++) { |
duke@435 | 2238 | jint n = pending_signals[i]; |
duke@435 | 2239 | if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { |
duke@435 | 2240 | return i; |
duke@435 | 2241 | } |
duke@435 | 2242 | } |
duke@435 | 2243 | if (!wait) { |
duke@435 | 2244 | return -1; |
duke@435 | 2245 | } |
duke@435 | 2246 | JavaThread *thread = JavaThread::current(); |
duke@435 | 2247 | ThreadBlockInVM tbivm(thread); |
duke@435 | 2248 | |
duke@435 | 2249 | bool threadIsSuspended; |
duke@435 | 2250 | do { |
duke@435 | 2251 | thread->set_suspend_equivalent(); |
duke@435 | 2252 | // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() |
duke@435 | 2253 | ::sem_wait(&sig_sem); |
duke@435 | 2254 | |
duke@435 | 2255 | // were we externally suspended while we were waiting? |
duke@435 | 2256 | threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); |
duke@435 | 2257 | if (threadIsSuspended) { |
duke@435 | 2258 | // |
duke@435 | 2259 | // The semaphore has been incremented, but while we were waiting |
duke@435 | 2260 | // another thread suspended us. We don't want to continue running |
duke@435 | 2261 | // while suspended because that would surprise the thread that |
duke@435 | 2262 | // suspended us. |
duke@435 | 2263 | // |
duke@435 | 2264 | ::sem_post(&sig_sem); |
duke@435 | 2265 | |
duke@435 | 2266 | thread->java_suspend_self(); |
duke@435 | 2267 | } |
duke@435 | 2268 | } while (threadIsSuspended); |
duke@435 | 2269 | } |
duke@435 | 2270 | } |
duke@435 | 2271 | |
duke@435 | 2272 | int os::signal_lookup() { |
duke@435 | 2273 | return check_pending_signals(false); |
duke@435 | 2274 | } |
duke@435 | 2275 | |
duke@435 | 2276 | int os::signal_wait() { |
duke@435 | 2277 | return check_pending_signals(true); |
duke@435 | 2278 | } |
duke@435 | 2279 | |
duke@435 | 2280 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 2281 | // Virtual Memory |
duke@435 | 2282 | |
duke@435 | 2283 | int os::vm_page_size() { |
duke@435 | 2284 | // Seems redundant as all get out |
duke@435 | 2285 | assert(os::Linux::page_size() != -1, "must call os::init"); |
duke@435 | 2286 | return os::Linux::page_size(); |
duke@435 | 2287 | } |
duke@435 | 2288 | |
duke@435 | 2289 | // Solaris allocates memory by pages. |
duke@435 | 2290 | int os::vm_allocation_granularity() { |
duke@435 | 2291 | assert(os::Linux::page_size() != -1, "must call os::init"); |
duke@435 | 2292 | return os::Linux::page_size(); |
duke@435 | 2293 | } |
duke@435 | 2294 | |
duke@435 | 2295 | // Rationale behind this function: |
duke@435 | 2296 | // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable |
duke@435 | 2297 | // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get |
duke@435 | 2298 | // samples for JITted code. Here we create private executable mapping over the code cache |
duke@435 | 2299 | // and then we can use standard (well, almost, as mapping can change) way to provide |
duke@435 | 2300 | // info for the reporting script by storing timestamp and location of symbol |
duke@435 | 2301 | void linux_wrap_code(char* base, size_t size) { |
duke@435 | 2302 | static volatile jint cnt = 0; |
duke@435 | 2303 | |
duke@435 | 2304 | if (!UseOprofile) { |
duke@435 | 2305 | return; |
duke@435 | 2306 | } |
duke@435 | 2307 | |
coleenp@1852 | 2308 | char buf[PATH_MAX+1]; |
duke@435 | 2309 | int num = Atomic::add(1, &cnt); |
duke@435 | 2310 | |
coleenp@1788 | 2311 | snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", |
coleenp@1788 | 2312 | os::get_temp_directory(), os::current_process_id(), num); |
duke@435 | 2313 | unlink(buf); |
duke@435 | 2314 | |
duke@435 | 2315 | int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); |
duke@435 | 2316 | |
duke@435 | 2317 | if (fd != -1) { |
duke@435 | 2318 | off_t rv = lseek(fd, size-2, SEEK_SET); |
duke@435 | 2319 | if (rv != (off_t)-1) { |
duke@435 | 2320 | if (write(fd, "", 1) == 1) { |
duke@435 | 2321 | mmap(base, size, |
duke@435 | 2322 | PROT_READ|PROT_WRITE|PROT_EXEC, |
duke@435 | 2323 | MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); |
duke@435 | 2324 | } |
duke@435 | 2325 | } |
duke@435 | 2326 | close(fd); |
duke@435 | 2327 | unlink(buf); |
duke@435 | 2328 | } |
duke@435 | 2329 | } |
duke@435 | 2330 | |
duke@435 | 2331 | // NOTE: Linux kernel does not really reserve the pages for us. |
duke@435 | 2332 | // All it does is to check if there are enough free pages |
duke@435 | 2333 | // left at the time of mmap(). This could be a potential |
duke@435 | 2334 | // problem. |
coleenp@1091 | 2335 | bool os::commit_memory(char* addr, size_t size, bool exec) { |
coleenp@1091 | 2336 | int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; |
coleenp@1091 | 2337 | uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, |
duke@435 | 2338 | MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); |
duke@435 | 2339 | return res != (uintptr_t) MAP_FAILED; |
duke@435 | 2340 | } |
duke@435 | 2341 | |
coleenp@1091 | 2342 | bool os::commit_memory(char* addr, size_t size, size_t alignment_hint, |
coleenp@1091 | 2343 | bool exec) { |
coleenp@1091 | 2344 | return commit_memory(addr, size, exec); |
duke@435 | 2345 | } |
duke@435 | 2346 | |
duke@435 | 2347 | void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } |
iveresov@576 | 2348 | |
iveresov@576 | 2349 | void os::free_memory(char *addr, size_t bytes) { |
iveresov@1196 | 2350 | ::mmap(addr, bytes, PROT_READ | PROT_WRITE, |
iveresov@1196 | 2351 | MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); |
iveresov@576 | 2352 | } |
iveresov@576 | 2353 | |
iveresov@897 | 2354 | void os::numa_make_global(char *addr, size_t bytes) { |
iveresov@897 | 2355 | Linux::numa_interleave_memory(addr, bytes); |
iveresov@897 | 2356 | } |
iveresov@576 | 2357 | |
iveresov@576 | 2358 | void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { |
iveresov@576 | 2359 | Linux::numa_tonode_memory(addr, bytes, lgrp_hint); |
iveresov@576 | 2360 | } |
iveresov@576 | 2361 | |
iveresov@576 | 2362 | bool os::numa_topology_changed() { return false; } |
iveresov@576 | 2363 | |
iveresov@576 | 2364 | size_t os::numa_get_groups_num() { |
iveresov@576 | 2365 | int max_node = Linux::numa_max_node(); |
iveresov@576 | 2366 | return max_node > 0 ? max_node + 1 : 1; |
iveresov@576 | 2367 | } |
iveresov@576 | 2368 | |
iveresov@576 | 2369 | int os::numa_get_group_id() { |
iveresov@576 | 2370 | int cpu_id = Linux::sched_getcpu(); |
iveresov@576 | 2371 | if (cpu_id != -1) { |
iveresov@576 | 2372 | int lgrp_id = Linux::get_node_by_cpu(cpu_id); |
iveresov@576 | 2373 | if (lgrp_id != -1) { |
iveresov@576 | 2374 | return lgrp_id; |
iveresov@576 | 2375 | } |
duke@435 | 2376 | } |
duke@435 | 2377 | return 0; |
duke@435 | 2378 | } |
duke@435 | 2379 | |
iveresov@576 | 2380 | size_t os::numa_get_leaf_groups(int *ids, size_t size) { |
iveresov@576 | 2381 | for (size_t i = 0; i < size; i++) { |
iveresov@576 | 2382 | ids[i] = i; |
iveresov@576 | 2383 | } |
iveresov@576 | 2384 | return size; |
iveresov@576 | 2385 | } |
iveresov@576 | 2386 | |
duke@435 | 2387 | bool os::get_page_info(char *start, page_info* info) { |
duke@435 | 2388 | return false; |
duke@435 | 2389 | } |
duke@435 | 2390 | |
duke@435 | 2391 | char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { |
duke@435 | 2392 | return end; |
duke@435 | 2393 | } |
duke@435 | 2394 | |
iveresov@576 | 2395 | extern "C" void numa_warn(int number, char *where, ...) { } |
iveresov@576 | 2396 | extern "C" void numa_error(char *where) { } |
iveresov@576 | 2397 | |
iveresov@1198 | 2398 | |
iveresov@1198 | 2399 | // If we are running with libnuma version > 2, then we should |
iveresov@1198 | 2400 | // be trying to use symbols with versions 1.1 |
iveresov@1198 | 2401 | // If we are running with earlier version, which did not have symbol versions, |
iveresov@1198 | 2402 | // we should use the base version. |
iveresov@1198 | 2403 | void* os::Linux::libnuma_dlsym(void* handle, const char *name) { |
iveresov@1198 | 2404 | void *f = dlvsym(handle, name, "libnuma_1.1"); |
iveresov@1198 | 2405 | if (f == NULL) { |
iveresov@1198 | 2406 | f = dlsym(handle, name); |
iveresov@1198 | 2407 | } |
iveresov@1198 | 2408 | return f; |
iveresov@1198 | 2409 | } |
iveresov@1198 | 2410 | |
iveresov@897 | 2411 | bool os::Linux::libnuma_init() { |
iveresov@576 | 2412 | // sched_getcpu() should be in libc. |
iveresov@576 | 2413 | set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, |
iveresov@576 | 2414 | dlsym(RTLD_DEFAULT, "sched_getcpu"))); |
iveresov@576 | 2415 | |
iveresov@576 | 2416 | if (sched_getcpu() != -1) { // Does it work? |
iveresov@702 | 2417 | void *handle = dlopen("libnuma.so.1", RTLD_LAZY); |
iveresov@576 | 2418 | if (handle != NULL) { |
iveresov@576 | 2419 | set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, |
iveresov@1198 | 2420 | libnuma_dlsym(handle, "numa_node_to_cpus"))); |
iveresov@576 | 2421 | set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, |
iveresov@1198 | 2422 | libnuma_dlsym(handle, "numa_max_node"))); |
iveresov@576 | 2423 | set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, |
iveresov@1198 | 2424 | libnuma_dlsym(handle, "numa_available"))); |
iveresov@576 | 2425 | set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, |
iveresov@1198 | 2426 | libnuma_dlsym(handle, "numa_tonode_memory"))); |
iveresov@897 | 2427 | set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, |
iveresov@1198 | 2428 | libnuma_dlsym(handle, "numa_interleave_memory"))); |
iveresov@897 | 2429 | |
iveresov@897 | 2430 | |
iveresov@576 | 2431 | if (numa_available() != -1) { |
iveresov@1198 | 2432 | set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); |
iveresov@576 | 2433 | // Create a cpu -> node mapping |
iveresov@576 | 2434 | _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); |
iveresov@576 | 2435 | rebuild_cpu_to_node_map(); |
iveresov@897 | 2436 | return true; |
iveresov@576 | 2437 | } |
iveresov@576 | 2438 | } |
iveresov@576 | 2439 | } |
iveresov@897 | 2440 | return false; |
iveresov@576 | 2441 | } |
iveresov@576 | 2442 | |
iveresov@576 | 2443 | // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. |
iveresov@576 | 2444 | // The table is later used in get_node_by_cpu(). |
iveresov@576 | 2445 | void os::Linux::rebuild_cpu_to_node_map() { |
iveresov@897 | 2446 | const size_t NCPUS = 32768; // Since the buffer size computation is very obscure |
iveresov@897 | 2447 | // in libnuma (possible values are starting from 16, |
iveresov@897 | 2448 | // and continuing up with every other power of 2, but less |
iveresov@897 | 2449 | // than the maximum number of CPUs supported by kernel), and |
iveresov@897 | 2450 | // is a subject to change (in libnuma version 2 the requirements |
iveresov@897 | 2451 | // are more reasonable) we'll just hardcode the number they use |
iveresov@897 | 2452 | // in the library. |
iveresov@897 | 2453 | const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; |
iveresov@897 | 2454 | |
iveresov@897 | 2455 | size_t cpu_num = os::active_processor_count(); |
iveresov@897 | 2456 | size_t cpu_map_size = NCPUS / BitsPerCLong; |
iveresov@897 | 2457 | size_t cpu_map_valid_size = |
iveresov@897 | 2458 | MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); |
iveresov@897 | 2459 | |
iveresov@576 | 2460 | cpu_to_node()->clear(); |
iveresov@576 | 2461 | cpu_to_node()->at_grow(cpu_num - 1); |
iveresov@897 | 2462 | size_t node_num = numa_get_groups_num(); |
iveresov@897 | 2463 | |
iveresov@576 | 2464 | unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); |
iveresov@897 | 2465 | for (size_t i = 0; i < node_num; i++) { |
iveresov@576 | 2466 | if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { |
iveresov@897 | 2467 | for (size_t j = 0; j < cpu_map_valid_size; j++) { |
iveresov@576 | 2468 | if (cpu_map[j] != 0) { |
iveresov@897 | 2469 | for (size_t k = 0; k < BitsPerCLong; k++) { |
iveresov@576 | 2470 | if (cpu_map[j] & (1UL << k)) { |
iveresov@897 | 2471 | cpu_to_node()->at_put(j * BitsPerCLong + k, i); |
iveresov@576 | 2472 | } |
iveresov@576 | 2473 | } |
iveresov@576 | 2474 | } |
iveresov@576 | 2475 | } |
iveresov@576 | 2476 | } |
iveresov@576 | 2477 | } |
iveresov@576 | 2478 | FREE_C_HEAP_ARRAY(unsigned long, cpu_map); |
iveresov@576 | 2479 | } |
iveresov@576 | 2480 | |
iveresov@576 | 2481 | int os::Linux::get_node_by_cpu(int cpu_id) { |
iveresov@576 | 2482 | if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { |
iveresov@576 | 2483 | return cpu_to_node()->at(cpu_id); |
iveresov@576 | 2484 | } |
iveresov@576 | 2485 | return -1; |
iveresov@576 | 2486 | } |
iveresov@576 | 2487 | |
iveresov@576 | 2488 | GrowableArray<int>* os::Linux::_cpu_to_node; |
iveresov@576 | 2489 | os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; |
iveresov@576 | 2490 | os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; |
iveresov@576 | 2491 | os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; |
iveresov@576 | 2492 | os::Linux::numa_available_func_t os::Linux::_numa_available; |
iveresov@576 | 2493 | os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; |
iveresov@897 | 2494 | os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; |
iveresov@897 | 2495 | unsigned long* os::Linux::_numa_all_nodes; |
iveresov@576 | 2496 | |
duke@435 | 2497 | bool os::uncommit_memory(char* addr, size_t size) { |
coleenp@1091 | 2498 | return ::mmap(addr, size, PROT_NONE, |
duke@435 | 2499 | MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) |
duke@435 | 2500 | != MAP_FAILED; |
duke@435 | 2501 | } |
duke@435 | 2502 | |
coleenp@1755 | 2503 | // Linux uses a growable mapping for the stack, and if the mapping for |
coleenp@1755 | 2504 | // the stack guard pages is not removed when we detach a thread the |
coleenp@1755 | 2505 | // stack cannot grow beyond the pages where the stack guard was |
coleenp@1755 | 2506 | // mapped. If at some point later in the process the stack expands to |
coleenp@1755 | 2507 | // that point, the Linux kernel cannot expand the stack any further |
coleenp@1755 | 2508 | // because the guard pages are in the way, and a segfault occurs. |
coleenp@1755 | 2509 | // |
coleenp@1755 | 2510 | // However, it's essential not to split the stack region by unmapping |
coleenp@1755 | 2511 | // a region (leaving a hole) that's already part of the stack mapping, |
coleenp@1755 | 2512 | // so if the stack mapping has already grown beyond the guard pages at |
coleenp@1755 | 2513 | // the time we create them, we have to truncate the stack mapping. |
coleenp@1755 | 2514 | // So, we need to know the extent of the stack mapping when |
coleenp@1755 | 2515 | // create_stack_guard_pages() is called. |
coleenp@1755 | 2516 | |
coleenp@1755 | 2517 | // Find the bounds of the stack mapping. Return true for success. |
coleenp@1755 | 2518 | // |
coleenp@1755 | 2519 | // We only need this for stacks that are growable: at the time of |
coleenp@1755 | 2520 | // writing thread stacks don't use growable mappings (i.e. those |
coleenp@1755 | 2521 | // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this |
coleenp@1755 | 2522 | // only applies to the main thread. |
coleenp@1755 | 2523 | static bool |
coleenp@1755 | 2524 | get_stack_bounds(uintptr_t *bottom, uintptr_t *top) |
coleenp@1755 | 2525 | { |
coleenp@1755 | 2526 | FILE *f = fopen("/proc/self/maps", "r"); |
coleenp@1755 | 2527 | if (f == NULL) |
coleenp@1755 | 2528 | return false; |
coleenp@1755 | 2529 | |
coleenp@1755 | 2530 | while (!feof(f)) { |
coleenp@1755 | 2531 | size_t dummy; |
coleenp@1755 | 2532 | char *str = NULL; |
coleenp@1755 | 2533 | ssize_t len = getline(&str, &dummy, f); |
coleenp@1755 | 2534 | if (len == -1) { |
coleenp@1760 | 2535 | fclose(f); |
coleenp@1755 | 2536 | return false; |
coleenp@1755 | 2537 | } |
coleenp@1755 | 2538 | |
coleenp@1755 | 2539 | if (len > 0 && str[len-1] == '\n') { |
coleenp@1755 | 2540 | str[len-1] = 0; |
coleenp@1755 | 2541 | len--; |
coleenp@1755 | 2542 | } |
coleenp@1755 | 2543 | |
coleenp@1755 | 2544 | static const char *stack_str = "[stack]"; |
coleenp@1755 | 2545 | if (len > (ssize_t)strlen(stack_str) |
coleenp@1755 | 2546 | && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) { |
coleenp@1755 | 2547 | if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { |
coleenp@1755 | 2548 | uintptr_t sp = (uintptr_t)__builtin_frame_address(0); |
coleenp@1755 | 2549 | if (sp >= *bottom && sp <= *top) { |
coleenp@1755 | 2550 | free(str); |
coleenp@1760 | 2551 | fclose(f); |
coleenp@1755 | 2552 | return true; |
coleenp@1755 | 2553 | } |
coleenp@1755 | 2554 | } |
coleenp@1755 | 2555 | } |
coleenp@1755 | 2556 | free(str); |
coleenp@1755 | 2557 | } |
coleenp@1760 | 2558 | fclose(f); |
coleenp@1755 | 2559 | return false; |
coleenp@1755 | 2560 | } |
coleenp@1755 | 2561 | |
coleenp@1755 | 2562 | // If the (growable) stack mapping already extends beyond the point |
coleenp@1755 | 2563 | // where we're going to put our guard pages, truncate the mapping at |
coleenp@1755 | 2564 | // that point by munmap()ping it. This ensures that when we later |
coleenp@1755 | 2565 | // munmap() the guard pages we don't leave a hole in the stack |
coleenp@1755 | 2566 | // mapping. |
coleenp@1755 | 2567 | bool os::create_stack_guard_pages(char* addr, size_t size) { |
coleenp@1755 | 2568 | uintptr_t stack_extent, stack_base; |
coleenp@1755 | 2569 | if (get_stack_bounds(&stack_extent, &stack_base)) { |
coleenp@1755 | 2570 | if (stack_extent < (uintptr_t)addr) |
coleenp@1755 | 2571 | ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); |
coleenp@1755 | 2572 | } |
coleenp@1755 | 2573 | |
coleenp@1755 | 2574 | return os::commit_memory(addr, size); |
coleenp@1755 | 2575 | } |
coleenp@1755 | 2576 | |
coleenp@1755 | 2577 | // If this is a growable mapping, remove the guard pages entirely by |
coleenp@1755 | 2578 | // munmap()ping them. If not, just call uncommit_memory(). |
coleenp@1755 | 2579 | bool os::remove_stack_guard_pages(char* addr, size_t size) { |
coleenp@1755 | 2580 | uintptr_t stack_extent, stack_base; |
coleenp@1755 | 2581 | if (get_stack_bounds(&stack_extent, &stack_base)) { |
coleenp@1755 | 2582 | return ::munmap(addr, size) == 0; |
coleenp@1755 | 2583 | } |
coleenp@1755 | 2584 | |
coleenp@1755 | 2585 | return os::uncommit_memory(addr, size); |
coleenp@1755 | 2586 | } |
coleenp@1755 | 2587 | |
duke@435 | 2588 | static address _highest_vm_reserved_address = NULL; |
duke@435 | 2589 | |
duke@435 | 2590 | // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory |
duke@435 | 2591 | // at 'requested_addr'. If there are existing memory mappings at the same |
duke@435 | 2592 | // location, however, they will be overwritten. If 'fixed' is false, |
duke@435 | 2593 | // 'requested_addr' is only treated as a hint, the return value may or |
duke@435 | 2594 | // may not start from the requested address. Unlike Linux mmap(), this |
duke@435 | 2595 | // function returns NULL to indicate failure. |
duke@435 | 2596 | static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { |
duke@435 | 2597 | char * addr; |
duke@435 | 2598 | int flags; |
duke@435 | 2599 | |
duke@435 | 2600 | flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; |
duke@435 | 2601 | if (fixed) { |
duke@435 | 2602 | assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); |
duke@435 | 2603 | flags |= MAP_FIXED; |
duke@435 | 2604 | } |
duke@435 | 2605 | |
coleenp@1091 | 2606 | // Map uncommitted pages PROT_READ and PROT_WRITE, change access |
coleenp@1091 | 2607 | // to PROT_EXEC if executable when we commit the page. |
coleenp@1091 | 2608 | addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, |
duke@435 | 2609 | flags, -1, 0); |
duke@435 | 2610 | |
duke@435 | 2611 | if (addr != MAP_FAILED) { |
duke@435 | 2612 | // anon_mmap() should only get called during VM initialization, |
duke@435 | 2613 | // don't need lock (actually we can skip locking even it can be called |
duke@435 | 2614 | // from multiple threads, because _highest_vm_reserved_address is just a |
duke@435 | 2615 | // hint about the upper limit of non-stack memory regions.) |
duke@435 | 2616 | if ((address)addr + bytes > _highest_vm_reserved_address) { |
duke@435 | 2617 | _highest_vm_reserved_address = (address)addr + bytes; |
duke@435 | 2618 | } |
duke@435 | 2619 | } |
duke@435 | 2620 | |
duke@435 | 2621 | return addr == MAP_FAILED ? NULL : addr; |
duke@435 | 2622 | } |
duke@435 | 2623 | |
duke@435 | 2624 | // Don't update _highest_vm_reserved_address, because there might be memory |
duke@435 | 2625 | // regions above addr + size. If so, releasing a memory region only creates |
duke@435 | 2626 | // a hole in the address space, it doesn't help prevent heap-stack collision. |
duke@435 | 2627 | // |
duke@435 | 2628 | static int anon_munmap(char * addr, size_t size) { |
duke@435 | 2629 | return ::munmap(addr, size) == 0; |
duke@435 | 2630 | } |
duke@435 | 2631 | |
duke@435 | 2632 | char* os::reserve_memory(size_t bytes, char* requested_addr, |
duke@435 | 2633 | size_t alignment_hint) { |
duke@435 | 2634 | return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); |
duke@435 | 2635 | } |
duke@435 | 2636 | |
duke@435 | 2637 | bool os::release_memory(char* addr, size_t size) { |
duke@435 | 2638 | return anon_munmap(addr, size); |
duke@435 | 2639 | } |
duke@435 | 2640 | |
duke@435 | 2641 | static address highest_vm_reserved_address() { |
duke@435 | 2642 | return _highest_vm_reserved_address; |
duke@435 | 2643 | } |
duke@435 | 2644 | |
duke@435 | 2645 | static bool linux_mprotect(char* addr, size_t size, int prot) { |
duke@435 | 2646 | // Linux wants the mprotect address argument to be page aligned. |
duke@435 | 2647 | char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); |
duke@435 | 2648 | |
duke@435 | 2649 | // According to SUSv3, mprotect() should only be used with mappings |
duke@435 | 2650 | // established by mmap(), and mmap() always maps whole pages. Unaligned |
duke@435 | 2651 | // 'addr' likely indicates problem in the VM (e.g. trying to change |
duke@435 | 2652 | // protection of malloc'ed or statically allocated memory). Check the |
duke@435 | 2653 | // caller if you hit this assert. |
duke@435 | 2654 | assert(addr == bottom, "sanity check"); |
duke@435 | 2655 | |
duke@435 | 2656 | size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); |
duke@435 | 2657 | return ::mprotect(bottom, size, prot) == 0; |
duke@435 | 2658 | } |
duke@435 | 2659 | |
coleenp@672 | 2660 | // Set protections specified |
coleenp@672 | 2661 | bool os::protect_memory(char* addr, size_t bytes, ProtType prot, |
coleenp@672 | 2662 | bool is_committed) { |
coleenp@672 | 2663 | unsigned int p = 0; |
coleenp@672 | 2664 | switch (prot) { |
coleenp@672 | 2665 | case MEM_PROT_NONE: p = PROT_NONE; break; |
coleenp@672 | 2666 | case MEM_PROT_READ: p = PROT_READ; break; |
coleenp@672 | 2667 | case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; |
coleenp@672 | 2668 | case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; |
coleenp@672 | 2669 | default: |
coleenp@672 | 2670 | ShouldNotReachHere(); |
coleenp@672 | 2671 | } |
coleenp@672 | 2672 | // is_committed is unused. |
coleenp@672 | 2673 | return linux_mprotect(addr, bytes, p); |
duke@435 | 2674 | } |
duke@435 | 2675 | |
duke@435 | 2676 | bool os::guard_memory(char* addr, size_t size) { |
duke@435 | 2677 | return linux_mprotect(addr, size, PROT_NONE); |
duke@435 | 2678 | } |
duke@435 | 2679 | |
duke@435 | 2680 | bool os::unguard_memory(char* addr, size_t size) { |
coleenp@912 | 2681 | return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); |
duke@435 | 2682 | } |
duke@435 | 2683 | |
duke@435 | 2684 | // Large page support |
duke@435 | 2685 | |
duke@435 | 2686 | static size_t _large_page_size = 0; |
duke@435 | 2687 | |
duke@435 | 2688 | bool os::large_page_init() { |
duke@435 | 2689 | if (!UseLargePages) return false; |
duke@435 | 2690 | |
duke@435 | 2691 | if (LargePageSizeInBytes) { |
duke@435 | 2692 | _large_page_size = LargePageSizeInBytes; |
duke@435 | 2693 | } else { |
duke@435 | 2694 | // large_page_size on Linux is used to round up heap size. x86 uses either |
duke@435 | 2695 | // 2M or 4M page, depending on whether PAE (Physical Address Extensions) |
duke@435 | 2696 | // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use |
duke@435 | 2697 | // page as large as 256M. |
duke@435 | 2698 | // |
duke@435 | 2699 | // Here we try to figure out page size by parsing /proc/meminfo and looking |
duke@435 | 2700 | // for a line with the following format: |
duke@435 | 2701 | // Hugepagesize: 2048 kB |
duke@435 | 2702 | // |
duke@435 | 2703 | // If we can't determine the value (e.g. /proc is not mounted, or the text |
duke@435 | 2704 | // format has been changed), we'll use the largest page size supported by |
duke@435 | 2705 | // the processor. |
duke@435 | 2706 | |
never@1445 | 2707 | #ifndef ZERO |
duke@435 | 2708 | _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); |
never@1445 | 2709 | #endif // ZERO |
duke@435 | 2710 | |
duke@435 | 2711 | FILE *fp = fopen("/proc/meminfo", "r"); |
duke@435 | 2712 | if (fp) { |
duke@435 | 2713 | while (!feof(fp)) { |
duke@435 | 2714 | int x = 0; |
duke@435 | 2715 | char buf[16]; |
duke@435 | 2716 | if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { |
duke@435 | 2717 | if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { |
duke@435 | 2718 | _large_page_size = x * K; |
duke@435 | 2719 | break; |
duke@435 | 2720 | } |
duke@435 | 2721 | } else { |
duke@435 | 2722 | // skip to next line |
duke@435 | 2723 | for (;;) { |
duke@435 | 2724 | int ch = fgetc(fp); |
duke@435 | 2725 | if (ch == EOF || ch == (int)'\n') break; |
duke@435 | 2726 | } |
duke@435 | 2727 | } |
duke@435 | 2728 | } |
duke@435 | 2729 | fclose(fp); |
duke@435 | 2730 | } |
duke@435 | 2731 | } |
duke@435 | 2732 | |
duke@435 | 2733 | const size_t default_page_size = (size_t)Linux::page_size(); |
duke@435 | 2734 | if (_large_page_size > default_page_size) { |
duke@435 | 2735 | _page_sizes[0] = _large_page_size; |
duke@435 | 2736 | _page_sizes[1] = default_page_size; |
duke@435 | 2737 | _page_sizes[2] = 0; |
duke@435 | 2738 | } |
duke@435 | 2739 | |
duke@435 | 2740 | // Large page support is available on 2.6 or newer kernel, some vendors |
duke@435 | 2741 | // (e.g. Redhat) have backported it to their 2.4 based distributions. |
duke@435 | 2742 | // We optimistically assume the support is available. If later it turns out |
duke@435 | 2743 | // not true, VM will automatically switch to use regular page size. |
duke@435 | 2744 | return true; |
duke@435 | 2745 | } |
duke@435 | 2746 | |
duke@435 | 2747 | #ifndef SHM_HUGETLB |
duke@435 | 2748 | #define SHM_HUGETLB 04000 |
duke@435 | 2749 | #endif |
duke@435 | 2750 | |
coleenp@1091 | 2751 | char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { |
coleenp@1091 | 2752 | // "exec" is passed in but not used. Creating the shared image for |
coleenp@1091 | 2753 | // the code cache doesn't have an SHM_X executable permission to check. |
duke@435 | 2754 | assert(UseLargePages, "only for large pages"); |
duke@435 | 2755 | |
duke@435 | 2756 | key_t key = IPC_PRIVATE; |
duke@435 | 2757 | char *addr; |
duke@435 | 2758 | |
duke@435 | 2759 | bool warn_on_failure = UseLargePages && |
duke@435 | 2760 | (!FLAG_IS_DEFAULT(UseLargePages) || |
duke@435 | 2761 | !FLAG_IS_DEFAULT(LargePageSizeInBytes) |
duke@435 | 2762 | ); |
duke@435 | 2763 | char msg[128]; |
duke@435 | 2764 | |
duke@435 | 2765 | // Create a large shared memory region to attach to based on size. |
duke@435 | 2766 | // Currently, size is the total size of the heap |
duke@435 | 2767 | int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); |
duke@435 | 2768 | if (shmid == -1) { |
duke@435 | 2769 | // Possible reasons for shmget failure: |
duke@435 | 2770 | // 1. shmmax is too small for Java heap. |
duke@435 | 2771 | // > check shmmax value: cat /proc/sys/kernel/shmmax |
duke@435 | 2772 | // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax |
duke@435 | 2773 | // 2. not enough large page memory. |
duke@435 | 2774 | // > check available large pages: cat /proc/meminfo |
duke@435 | 2775 | // > increase amount of large pages: |
duke@435 | 2776 | // echo new_value > /proc/sys/vm/nr_hugepages |
duke@435 | 2777 | // Note 1: different Linux may use different name for this property, |
duke@435 | 2778 | // e.g. on Redhat AS-3 it is "hugetlb_pool". |
duke@435 | 2779 | // Note 2: it's possible there's enough physical memory available but |
duke@435 | 2780 | // they are so fragmented after a long run that they can't |
duke@435 | 2781 | // coalesce into large pages. Try to reserve large pages when |
duke@435 | 2782 | // the system is still "fresh". |
duke@435 | 2783 | if (warn_on_failure) { |
duke@435 | 2784 | jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); |
duke@435 | 2785 | warning(msg); |
duke@435 | 2786 | } |
duke@435 | 2787 | return NULL; |
duke@435 | 2788 | } |
duke@435 | 2789 | |
duke@435 | 2790 | // attach to the region |
kvn@1892 | 2791 | addr = (char*)shmat(shmid, req_addr, 0); |
duke@435 | 2792 | int err = errno; |
duke@435 | 2793 | |
duke@435 | 2794 | // Remove shmid. If shmat() is successful, the actual shared memory segment |
duke@435 | 2795 | // will be deleted when it's detached by shmdt() or when the process |
duke@435 | 2796 | // terminates. If shmat() is not successful this will remove the shared |
duke@435 | 2797 | // segment immediately. |
duke@435 | 2798 | shmctl(shmid, IPC_RMID, NULL); |
duke@435 | 2799 | |
duke@435 | 2800 | if ((intptr_t)addr == -1) { |
duke@435 | 2801 | if (warn_on_failure) { |
duke@435 | 2802 | jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); |
duke@435 | 2803 | warning(msg); |
duke@435 | 2804 | } |
duke@435 | 2805 | return NULL; |
duke@435 | 2806 | } |
duke@435 | 2807 | |
duke@435 | 2808 | return addr; |
duke@435 | 2809 | } |
duke@435 | 2810 | |
duke@435 | 2811 | bool os::release_memory_special(char* base, size_t bytes) { |
duke@435 | 2812 | // detaching the SHM segment will also delete it, see reserve_memory_special() |
duke@435 | 2813 | int rslt = shmdt(base); |
duke@435 | 2814 | return rslt == 0; |
duke@435 | 2815 | } |
duke@435 | 2816 | |
duke@435 | 2817 | size_t os::large_page_size() { |
duke@435 | 2818 | return _large_page_size; |
duke@435 | 2819 | } |
duke@435 | 2820 | |
duke@435 | 2821 | // Linux does not support anonymous mmap with large page memory. The only way |
duke@435 | 2822 | // to reserve large page memory without file backing is through SysV shared |
duke@435 | 2823 | // memory API. The entire memory region is committed and pinned upfront. |
duke@435 | 2824 | // Hopefully this will change in the future... |
duke@435 | 2825 | bool os::can_commit_large_page_memory() { |
duke@435 | 2826 | return false; |
duke@435 | 2827 | } |
duke@435 | 2828 | |
jcoomes@514 | 2829 | bool os::can_execute_large_page_memory() { |
jcoomes@514 | 2830 | return false; |
jcoomes@514 | 2831 | } |
jcoomes@514 | 2832 | |
duke@435 | 2833 | // Reserve memory at an arbitrary address, only if that area is |
duke@435 | 2834 | // available (and not reserved for something else). |
duke@435 | 2835 | |
duke@435 | 2836 | char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { |
duke@435 | 2837 | const int max_tries = 10; |
duke@435 | 2838 | char* base[max_tries]; |
duke@435 | 2839 | size_t size[max_tries]; |
duke@435 | 2840 | const size_t gap = 0x000000; |
duke@435 | 2841 | |
duke@435 | 2842 | // Assert only that the size is a multiple of the page size, since |
duke@435 | 2843 | // that's all that mmap requires, and since that's all we really know |
duke@435 | 2844 | // about at this low abstraction level. If we need higher alignment, |
duke@435 | 2845 | // we can either pass an alignment to this method or verify alignment |
duke@435 | 2846 | // in one of the methods further up the call chain. See bug 5044738. |
duke@435 | 2847 | assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); |
duke@435 | 2848 | |
duke@435 | 2849 | // Repeatedly allocate blocks until the block is allocated at the |
duke@435 | 2850 | // right spot. Give up after max_tries. Note that reserve_memory() will |
duke@435 | 2851 | // automatically update _highest_vm_reserved_address if the call is |
duke@435 | 2852 | // successful. The variable tracks the highest memory address every reserved |
duke@435 | 2853 | // by JVM. It is used to detect heap-stack collision if running with |
duke@435 | 2854 | // fixed-stack LinuxThreads. Because here we may attempt to reserve more |
duke@435 | 2855 | // space than needed, it could confuse the collision detecting code. To |
duke@435 | 2856 | // solve the problem, save current _highest_vm_reserved_address and |
duke@435 | 2857 | // calculate the correct value before return. |
duke@435 | 2858 | address old_highest = _highest_vm_reserved_address; |
duke@435 | 2859 | |
duke@435 | 2860 | // Linux mmap allows caller to pass an address as hint; give it a try first, |
duke@435 | 2861 | // if kernel honors the hint then we can return immediately. |
duke@435 | 2862 | char * addr = anon_mmap(requested_addr, bytes, false); |
duke@435 | 2863 | if (addr == requested_addr) { |
duke@435 | 2864 | return requested_addr; |
duke@435 | 2865 | } |
duke@435 | 2866 | |
duke@435 | 2867 | if (addr != NULL) { |
duke@435 | 2868 | // mmap() is successful but it fails to reserve at the requested address |
duke@435 | 2869 | anon_munmap(addr, bytes); |
duke@435 | 2870 | } |
duke@435 | 2871 | |
duke@435 | 2872 | int i; |
duke@435 | 2873 | for (i = 0; i < max_tries; ++i) { |
duke@435 | 2874 | base[i] = reserve_memory(bytes); |
duke@435 | 2875 | |
duke@435 | 2876 | if (base[i] != NULL) { |
duke@435 | 2877 | // Is this the block we wanted? |
duke@435 | 2878 | if (base[i] == requested_addr) { |
duke@435 | 2879 | size[i] = bytes; |
duke@435 | 2880 | break; |
duke@435 | 2881 | } |
duke@435 | 2882 | |
duke@435 | 2883 | // Does this overlap the block we wanted? Give back the overlapped |
duke@435 | 2884 | // parts and try again. |
duke@435 | 2885 | |
duke@435 | 2886 | size_t top_overlap = requested_addr + (bytes + gap) - base[i]; |
duke@435 | 2887 | if (top_overlap >= 0 && top_overlap < bytes) { |
duke@435 | 2888 | unmap_memory(base[i], top_overlap); |
duke@435 | 2889 | base[i] += top_overlap; |
duke@435 | 2890 | size[i] = bytes - top_overlap; |
duke@435 | 2891 | } else { |
duke@435 | 2892 | size_t bottom_overlap = base[i] + bytes - requested_addr; |
duke@435 | 2893 | if (bottom_overlap >= 0 && bottom_overlap < bytes) { |
duke@435 | 2894 | unmap_memory(requested_addr, bottom_overlap); |
duke@435 | 2895 | size[i] = bytes - bottom_overlap; |
duke@435 | 2896 | } else { |
duke@435 | 2897 | size[i] = bytes; |
duke@435 | 2898 | } |
duke@435 | 2899 | } |
duke@435 | 2900 | } |
duke@435 | 2901 | } |
duke@435 | 2902 | |
duke@435 | 2903 | // Give back the unused reserved pieces. |
duke@435 | 2904 | |
duke@435 | 2905 | for (int j = 0; j < i; ++j) { |
duke@435 | 2906 | if (base[j] != NULL) { |
duke@435 | 2907 | unmap_memory(base[j], size[j]); |
duke@435 | 2908 | } |
duke@435 | 2909 | } |
duke@435 | 2910 | |
duke@435 | 2911 | if (i < max_tries) { |
duke@435 | 2912 | _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); |
duke@435 | 2913 | return requested_addr; |
duke@435 | 2914 | } else { |
duke@435 | 2915 | _highest_vm_reserved_address = old_highest; |
duke@435 | 2916 | return NULL; |
duke@435 | 2917 | } |
duke@435 | 2918 | } |
duke@435 | 2919 | |
duke@435 | 2920 | size_t os::read(int fd, void *buf, unsigned int nBytes) { |
duke@435 | 2921 | return ::read(fd, buf, nBytes); |
duke@435 | 2922 | } |
duke@435 | 2923 | |
duke@435 | 2924 | // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. |
duke@435 | 2925 | // Solaris uses poll(), linux uses park(). |
duke@435 | 2926 | // Poll() is likely a better choice, assuming that Thread.interrupt() |
duke@435 | 2927 | // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with |
duke@435 | 2928 | // SIGSEGV, see 4355769. |
duke@435 | 2929 | |
duke@435 | 2930 | const int NANOSECS_PER_MILLISECS = 1000000; |
duke@435 | 2931 | |
duke@435 | 2932 | int os::sleep(Thread* thread, jlong millis, bool interruptible) { |
duke@435 | 2933 | assert(thread == Thread::current(), "thread consistency check"); |
duke@435 | 2934 | |
duke@435 | 2935 | ParkEvent * const slp = thread->_SleepEvent ; |
duke@435 | 2936 | slp->reset() ; |
duke@435 | 2937 | OrderAccess::fence() ; |
duke@435 | 2938 | |
duke@435 | 2939 | if (interruptible) { |
duke@435 | 2940 | jlong prevtime = javaTimeNanos(); |
duke@435 | 2941 | |
duke@435 | 2942 | for (;;) { |
duke@435 | 2943 | if (os::is_interrupted(thread, true)) { |
duke@435 | 2944 | return OS_INTRPT; |
duke@435 | 2945 | } |
duke@435 | 2946 | |
duke@435 | 2947 | jlong newtime = javaTimeNanos(); |
duke@435 | 2948 | |
duke@435 | 2949 | if (newtime - prevtime < 0) { |
duke@435 | 2950 | // time moving backwards, should only happen if no monotonic clock |
duke@435 | 2951 | // not a guarantee() because JVM should not abort on kernel/glibc bugs |
duke@435 | 2952 | assert(!Linux::supports_monotonic_clock(), "time moving backwards"); |
duke@435 | 2953 | } else { |
duke@435 | 2954 | millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; |
duke@435 | 2955 | } |
duke@435 | 2956 | |
duke@435 | 2957 | if(millis <= 0) { |
duke@435 | 2958 | return OS_OK; |
duke@435 | 2959 | } |
duke@435 | 2960 | |
duke@435 | 2961 | prevtime = newtime; |
duke@435 | 2962 | |
duke@435 | 2963 | { |
duke@435 | 2964 | assert(thread->is_Java_thread(), "sanity check"); |
duke@435 | 2965 | JavaThread *jt = (JavaThread *) thread; |
duke@435 | 2966 | ThreadBlockInVM tbivm(jt); |
duke@435 | 2967 | OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); |
duke@435 | 2968 | |
duke@435 | 2969 | jt->set_suspend_equivalent(); |
duke@435 | 2970 | // cleared by handle_special_suspend_equivalent_condition() or |
duke@435 | 2971 | // java_suspend_self() via check_and_wait_while_suspended() |
duke@435 | 2972 | |
duke@435 | 2973 | slp->park(millis); |
duke@435 | 2974 | |
duke@435 | 2975 | // were we externally suspended while we were waiting? |
duke@435 | 2976 | jt->check_and_wait_while_suspended(); |
duke@435 | 2977 | } |
duke@435 | 2978 | } |
duke@435 | 2979 | } else { |
duke@435 | 2980 | OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); |
duke@435 | 2981 | jlong prevtime = javaTimeNanos(); |
duke@435 | 2982 | |
duke@435 | 2983 | for (;;) { |
duke@435 | 2984 | // It'd be nice to avoid the back-to-back javaTimeNanos() calls on |
duke@435 | 2985 | // the 1st iteration ... |
duke@435 | 2986 | jlong newtime = javaTimeNanos(); |
duke@435 | 2987 | |
duke@435 | 2988 | if (newtime - prevtime < 0) { |
duke@435 | 2989 | // time moving backwards, should only happen if no monotonic clock |
duke@435 | 2990 | // not a guarantee() because JVM should not abort on kernel/glibc bugs |
duke@435 | 2991 | assert(!Linux::supports_monotonic_clock(), "time moving backwards"); |
duke@435 | 2992 | } else { |
duke@435 | 2993 | millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; |
duke@435 | 2994 | } |
duke@435 | 2995 | |
duke@435 | 2996 | if(millis <= 0) break ; |
duke@435 | 2997 | |
duke@435 | 2998 | prevtime = newtime; |
duke@435 | 2999 | slp->park(millis); |
duke@435 | 3000 | } |
duke@435 | 3001 | return OS_OK ; |
duke@435 | 3002 | } |
duke@435 | 3003 | } |
duke@435 | 3004 | |
duke@435 | 3005 | int os::naked_sleep() { |
duke@435 | 3006 | // %% make the sleep time an integer flag. for now use 1 millisec. |
duke@435 | 3007 | return os::sleep(Thread::current(), 1, false); |
duke@435 | 3008 | } |
duke@435 | 3009 | |
duke@435 | 3010 | // Sleep forever; naked call to OS-specific sleep; use with CAUTION |
duke@435 | 3011 | void os::infinite_sleep() { |
duke@435 | 3012 | while (true) { // sleep forever ... |
duke@435 | 3013 | ::sleep(100); // ... 100 seconds at a time |
duke@435 | 3014 | } |
duke@435 | 3015 | } |
duke@435 | 3016 | |
duke@435 | 3017 | // Used to convert frequent JVM_Yield() to nops |
duke@435 | 3018 | bool os::dont_yield() { |
duke@435 | 3019 | return DontYieldALot; |
duke@435 | 3020 | } |
duke@435 | 3021 | |
duke@435 | 3022 | void os::yield() { |
duke@435 | 3023 | sched_yield(); |
duke@435 | 3024 | } |
duke@435 | 3025 | |
duke@435 | 3026 | os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} |
duke@435 | 3027 | |
duke@435 | 3028 | void os::yield_all(int attempts) { |
duke@435 | 3029 | // Yields to all threads, including threads with lower priorities |
duke@435 | 3030 | // Threads on Linux are all with same priority. The Solaris style |
duke@435 | 3031 | // os::yield_all() with nanosleep(1ms) is not necessary. |
duke@435 | 3032 | sched_yield(); |
duke@435 | 3033 | } |
duke@435 | 3034 | |
duke@435 | 3035 | // Called from the tight loops to possibly influence time-sharing heuristics |
duke@435 | 3036 | void os::loop_breaker(int attempts) { |
duke@435 | 3037 | os::yield_all(attempts); |
duke@435 | 3038 | } |
duke@435 | 3039 | |
duke@435 | 3040 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 3041 | // thread priority support |
duke@435 | 3042 | |
duke@435 | 3043 | // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER |
duke@435 | 3044 | // only supports dynamic priority, static priority must be zero. For real-time |
duke@435 | 3045 | // applications, Linux supports SCHED_RR which allows static priority (1-99). |
duke@435 | 3046 | // However, for large multi-threaded applications, SCHED_RR is not only slower |
duke@435 | 3047 | // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out |
duke@435 | 3048 | // of 5 runs - Sep 2005). |
duke@435 | 3049 | // |
duke@435 | 3050 | // The following code actually changes the niceness of kernel-thread/LWP. It |
duke@435 | 3051 | // has an assumption that setpriority() only modifies one kernel-thread/LWP, |
duke@435 | 3052 | // not the entire user process, and user level threads are 1:1 mapped to kernel |
duke@435 | 3053 | // threads. It has always been the case, but could change in the future. For |
duke@435 | 3054 | // this reason, the code should not be used as default (ThreadPriorityPolicy=0). |
duke@435 | 3055 | // It is only used when ThreadPriorityPolicy=1 and requires root privilege. |
duke@435 | 3056 | |
duke@435 | 3057 | int os::java_to_os_priority[MaxPriority + 1] = { |
duke@435 | 3058 | 19, // 0 Entry should never be used |
duke@435 | 3059 | |
duke@435 | 3060 | 4, // 1 MinPriority |
duke@435 | 3061 | 3, // 2 |
duke@435 | 3062 | 2, // 3 |
duke@435 | 3063 | |
duke@435 | 3064 | 1, // 4 |
duke@435 | 3065 | 0, // 5 NormPriority |
duke@435 | 3066 | -1, // 6 |
duke@435 | 3067 | |
duke@435 | 3068 | -2, // 7 |
duke@435 | 3069 | -3, // 8 |
duke@435 | 3070 | -4, // 9 NearMaxPriority |
duke@435 | 3071 | |
duke@435 | 3072 | -5 // 10 MaxPriority |
duke@435 | 3073 | }; |
duke@435 | 3074 | |
duke@435 | 3075 | static int prio_init() { |
duke@435 | 3076 | if (ThreadPriorityPolicy == 1) { |
duke@435 | 3077 | // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 |
duke@435 | 3078 | // if effective uid is not root. Perhaps, a more elegant way of doing |
duke@435 | 3079 | // this is to test CAP_SYS_NICE capability, but that will require libcap.so |
duke@435 | 3080 | if (geteuid() != 0) { |
duke@435 | 3081 | if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { |
duke@435 | 3082 | warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); |
duke@435 | 3083 | } |
duke@435 | 3084 | ThreadPriorityPolicy = 0; |
duke@435 | 3085 | } |
duke@435 | 3086 | } |
duke@435 | 3087 | return 0; |
duke@435 | 3088 | } |
duke@435 | 3089 | |
duke@435 | 3090 | OSReturn os::set_native_priority(Thread* thread, int newpri) { |
duke@435 | 3091 | if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; |
duke@435 | 3092 | |
duke@435 | 3093 | int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); |
duke@435 | 3094 | return (ret == 0) ? OS_OK : OS_ERR; |
duke@435 | 3095 | } |
duke@435 | 3096 | |
duke@435 | 3097 | OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { |
duke@435 | 3098 | if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { |
duke@435 | 3099 | *priority_ptr = java_to_os_priority[NormPriority]; |
duke@435 | 3100 | return OS_OK; |
duke@435 | 3101 | } |
duke@435 | 3102 | |
duke@435 | 3103 | errno = 0; |
duke@435 | 3104 | *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); |
duke@435 | 3105 | return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); |
duke@435 | 3106 | } |
duke@435 | 3107 | |
duke@435 | 3108 | // Hint to the underlying OS that a task switch would not be good. |
duke@435 | 3109 | // Void return because it's a hint and can fail. |
duke@435 | 3110 | void os::hint_no_preempt() {} |
duke@435 | 3111 | |
duke@435 | 3112 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 3113 | // suspend/resume support |
duke@435 | 3114 | |
duke@435 | 3115 | // the low-level signal-based suspend/resume support is a remnant from the |
duke@435 | 3116 | // old VM-suspension that used to be for java-suspension, safepoints etc, |
duke@435 | 3117 | // within hotspot. Now there is a single use-case for this: |
duke@435 | 3118 | // - calling get_thread_pc() on the VMThread by the flat-profiler task |
duke@435 | 3119 | // that runs in the watcher thread. |
duke@435 | 3120 | // The remaining code is greatly simplified from the more general suspension |
duke@435 | 3121 | // code that used to be used. |
duke@435 | 3122 | // |
duke@435 | 3123 | // The protocol is quite simple: |
duke@435 | 3124 | // - suspend: |
duke@435 | 3125 | // - sends a signal to the target thread |
duke@435 | 3126 | // - polls the suspend state of the osthread using a yield loop |
duke@435 | 3127 | // - target thread signal handler (SR_handler) sets suspend state |
duke@435 | 3128 | // and blocks in sigsuspend until continued |
duke@435 | 3129 | // - resume: |
duke@435 | 3130 | // - sets target osthread state to continue |
duke@435 | 3131 | // - sends signal to end the sigsuspend loop in the SR_handler |
duke@435 | 3132 | // |
duke@435 | 3133 | // Note that the SR_lock plays no role in this suspend/resume protocol. |
duke@435 | 3134 | // |
duke@435 | 3135 | |
duke@435 | 3136 | static void resume_clear_context(OSThread *osthread) { |
duke@435 | 3137 | osthread->set_ucontext(NULL); |
duke@435 | 3138 | osthread->set_siginfo(NULL); |
duke@435 | 3139 | |
duke@435 | 3140 | // notify the suspend action is completed, we have now resumed |
duke@435 | 3141 | osthread->sr.clear_suspended(); |
duke@435 | 3142 | } |
duke@435 | 3143 | |
duke@435 | 3144 | static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { |
duke@435 | 3145 | osthread->set_ucontext(context); |
duke@435 | 3146 | osthread->set_siginfo(siginfo); |
duke@435 | 3147 | } |
duke@435 | 3148 | |
duke@435 | 3149 | // |
duke@435 | 3150 | // Handler function invoked when a thread's execution is suspended or |
duke@435 | 3151 | // resumed. We have to be careful that only async-safe functions are |
duke@435 | 3152 | // called here (Note: most pthread functions are not async safe and |
duke@435 | 3153 | // should be avoided.) |
duke@435 | 3154 | // |
duke@435 | 3155 | // Note: sigwait() is a more natural fit than sigsuspend() from an |
duke@435 | 3156 | // interface point of view, but sigwait() prevents the signal hander |
duke@435 | 3157 | // from being run. libpthread would get very confused by not having |
duke@435 | 3158 | // its signal handlers run and prevents sigwait()'s use with the |
duke@435 | 3159 | // mutex granting granting signal. |
duke@435 | 3160 | // |
duke@435 | 3161 | // Currently only ever called on the VMThread |
duke@435 | 3162 | // |
duke@435 | 3163 | static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { |
duke@435 | 3164 | // Save and restore errno to avoid confusing native code with EINTR |
duke@435 | 3165 | // after sigsuspend. |
duke@435 | 3166 | int old_errno = errno; |
duke@435 | 3167 | |
duke@435 | 3168 | Thread* thread = Thread::current(); |
duke@435 | 3169 | OSThread* osthread = thread->osthread(); |
duke@435 | 3170 | assert(thread->is_VM_thread(), "Must be VMThread"); |
duke@435 | 3171 | // read current suspend action |
duke@435 | 3172 | int action = osthread->sr.suspend_action(); |
duke@435 | 3173 | if (action == SR_SUSPEND) { |
duke@435 | 3174 | suspend_save_context(osthread, siginfo, context); |
duke@435 | 3175 | |
duke@435 | 3176 | // Notify the suspend action is about to be completed. do_suspend() |
duke@435 | 3177 | // waits until SR_SUSPENDED is set and then returns. We will wait |
duke@435 | 3178 | // here for a resume signal and that completes the suspend-other |
duke@435 | 3179 | // action. do_suspend/do_resume is always called as a pair from |
duke@435 | 3180 | // the same thread - so there are no races |
duke@435 | 3181 | |
duke@435 | 3182 | // notify the caller |
duke@435 | 3183 | osthread->sr.set_suspended(); |
duke@435 | 3184 | |
duke@435 | 3185 | sigset_t suspend_set; // signals for sigsuspend() |
duke@435 | 3186 | |
duke@435 | 3187 | // get current set of blocked signals and unblock resume signal |
duke@435 | 3188 | pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); |
duke@435 | 3189 | sigdelset(&suspend_set, SR_signum); |
duke@435 | 3190 | |
duke@435 | 3191 | // wait here until we are resumed |
duke@435 | 3192 | do { |
duke@435 | 3193 | sigsuspend(&suspend_set); |
duke@435 | 3194 | // ignore all returns until we get a resume signal |
duke@435 | 3195 | } while (osthread->sr.suspend_action() != SR_CONTINUE); |
duke@435 | 3196 | |
duke@435 | 3197 | resume_clear_context(osthread); |
duke@435 | 3198 | |
duke@435 | 3199 | } else { |
duke@435 | 3200 | assert(action == SR_CONTINUE, "unexpected sr action"); |
duke@435 | 3201 | // nothing special to do - just leave the handler |
duke@435 | 3202 | } |
duke@435 | 3203 | |
duke@435 | 3204 | errno = old_errno; |
duke@435 | 3205 | } |
duke@435 | 3206 | |
duke@435 | 3207 | |
duke@435 | 3208 | static int SR_initialize() { |
duke@435 | 3209 | struct sigaction act; |
duke@435 | 3210 | char *s; |
duke@435 | 3211 | /* Get signal number to use for suspend/resume */ |
duke@435 | 3212 | if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { |
duke@435 | 3213 | int sig = ::strtol(s, 0, 10); |
duke@435 | 3214 | if (sig > 0 || sig < _NSIG) { |
duke@435 | 3215 | SR_signum = sig; |
duke@435 | 3216 | } |
duke@435 | 3217 | } |
duke@435 | 3218 | |
duke@435 | 3219 | assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, |
duke@435 | 3220 | "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); |
duke@435 | 3221 | |
duke@435 | 3222 | sigemptyset(&SR_sigset); |
duke@435 | 3223 | sigaddset(&SR_sigset, SR_signum); |
duke@435 | 3224 | |
duke@435 | 3225 | /* Set up signal handler for suspend/resume */ |
duke@435 | 3226 | act.sa_flags = SA_RESTART|SA_SIGINFO; |
duke@435 | 3227 | act.sa_handler = (void (*)(int)) SR_handler; |
duke@435 | 3228 | |
duke@435 | 3229 | // SR_signum is blocked by default. |
duke@435 | 3230 | // 4528190 - We also need to block pthread restart signal (32 on all |
duke@435 | 3231 | // supported Linux platforms). Note that LinuxThreads need to block |
duke@435 | 3232 | // this signal for all threads to work properly. So we don't have |
duke@435 | 3233 | // to use hard-coded signal number when setting up the mask. |
duke@435 | 3234 | pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); |
duke@435 | 3235 | |
duke@435 | 3236 | if (sigaction(SR_signum, &act, 0) == -1) { |
duke@435 | 3237 | return -1; |
duke@435 | 3238 | } |
duke@435 | 3239 | |
duke@435 | 3240 | // Save signal flag |
duke@435 | 3241 | os::Linux::set_our_sigflags(SR_signum, act.sa_flags); |
duke@435 | 3242 | return 0; |
duke@435 | 3243 | } |
duke@435 | 3244 | |
duke@435 | 3245 | static int SR_finalize() { |
duke@435 | 3246 | return 0; |
duke@435 | 3247 | } |
duke@435 | 3248 | |
duke@435 | 3249 | |
duke@435 | 3250 | // returns true on success and false on error - really an error is fatal |
duke@435 | 3251 | // but this seems the normal response to library errors |
duke@435 | 3252 | static bool do_suspend(OSThread* osthread) { |
duke@435 | 3253 | // mark as suspended and send signal |
duke@435 | 3254 | osthread->sr.set_suspend_action(SR_SUSPEND); |
duke@435 | 3255 | int status = pthread_kill(osthread->pthread_id(), SR_signum); |
duke@435 | 3256 | assert_status(status == 0, status, "pthread_kill"); |
duke@435 | 3257 | |
duke@435 | 3258 | // check status and wait until notified of suspension |
duke@435 | 3259 | if (status == 0) { |
duke@435 | 3260 | for (int i = 0; !osthread->sr.is_suspended(); i++) { |
duke@435 | 3261 | os::yield_all(i); |
duke@435 | 3262 | } |
duke@435 | 3263 | osthread->sr.set_suspend_action(SR_NONE); |
duke@435 | 3264 | return true; |
duke@435 | 3265 | } |
duke@435 | 3266 | else { |
duke@435 | 3267 | osthread->sr.set_suspend_action(SR_NONE); |
duke@435 | 3268 | return false; |
duke@435 | 3269 | } |
duke@435 | 3270 | } |
duke@435 | 3271 | |
duke@435 | 3272 | static void do_resume(OSThread* osthread) { |
duke@435 | 3273 | assert(osthread->sr.is_suspended(), "thread should be suspended"); |
duke@435 | 3274 | osthread->sr.set_suspend_action(SR_CONTINUE); |
duke@435 | 3275 | |
duke@435 | 3276 | int status = pthread_kill(osthread->pthread_id(), SR_signum); |
duke@435 | 3277 | assert_status(status == 0, status, "pthread_kill"); |
duke@435 | 3278 | // check status and wait unit notified of resumption |
duke@435 | 3279 | if (status == 0) { |
duke@435 | 3280 | for (int i = 0; osthread->sr.is_suspended(); i++) { |
duke@435 | 3281 | os::yield_all(i); |
duke@435 | 3282 | } |
duke@435 | 3283 | } |
duke@435 | 3284 | osthread->sr.set_suspend_action(SR_NONE); |
duke@435 | 3285 | } |
duke@435 | 3286 | |
duke@435 | 3287 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 3288 | // interrupt support |
duke@435 | 3289 | |
duke@435 | 3290 | void os::interrupt(Thread* thread) { |
duke@435 | 3291 | assert(Thread::current() == thread || Threads_lock->owned_by_self(), |
duke@435 | 3292 | "possibility of dangling Thread pointer"); |
duke@435 | 3293 | |
duke@435 | 3294 | OSThread* osthread = thread->osthread(); |
duke@435 | 3295 | |
duke@435 | 3296 | if (!osthread->interrupted()) { |
duke@435 | 3297 | osthread->set_interrupted(true); |
duke@435 | 3298 | // More than one thread can get here with the same value of osthread, |
duke@435 | 3299 | // resulting in multiple notifications. We do, however, want the store |
duke@435 | 3300 | // to interrupted() to be visible to other threads before we execute unpark(). |
duke@435 | 3301 | OrderAccess::fence(); |
duke@435 | 3302 | ParkEvent * const slp = thread->_SleepEvent ; |
duke@435 | 3303 | if (slp != NULL) slp->unpark() ; |
duke@435 | 3304 | } |
duke@435 | 3305 | |
duke@435 | 3306 | // For JSR166. Unpark even if interrupt status already was set |
duke@435 | 3307 | if (thread->is_Java_thread()) |
duke@435 | 3308 | ((JavaThread*)thread)->parker()->unpark(); |
duke@435 | 3309 | |
duke@435 | 3310 | ParkEvent * ev = thread->_ParkEvent ; |
duke@435 | 3311 | if (ev != NULL) ev->unpark() ; |
duke@435 | 3312 | |
duke@435 | 3313 | } |
duke@435 | 3314 | |
duke@435 | 3315 | bool os::is_interrupted(Thread* thread, bool clear_interrupted) { |
duke@435 | 3316 | assert(Thread::current() == thread || Threads_lock->owned_by_self(), |
duke@435 | 3317 | "possibility of dangling Thread pointer"); |
duke@435 | 3318 | |
duke@435 | 3319 | OSThread* osthread = thread->osthread(); |
duke@435 | 3320 | |
duke@435 | 3321 | bool interrupted = osthread->interrupted(); |
duke@435 | 3322 | |
duke@435 | 3323 | if (interrupted && clear_interrupted) { |
duke@435 | 3324 | osthread->set_interrupted(false); |
duke@435 | 3325 | // consider thread->_SleepEvent->reset() ... optional optimization |
duke@435 | 3326 | } |
duke@435 | 3327 | |
duke@435 | 3328 | return interrupted; |
duke@435 | 3329 | } |
duke@435 | 3330 | |
duke@435 | 3331 | /////////////////////////////////////////////////////////////////////////////////// |
duke@435 | 3332 | // signal handling (except suspend/resume) |
duke@435 | 3333 | |
duke@435 | 3334 | // This routine may be used by user applications as a "hook" to catch signals. |
duke@435 | 3335 | // The user-defined signal handler must pass unrecognized signals to this |
duke@435 | 3336 | // routine, and if it returns true (non-zero), then the signal handler must |
duke@435 | 3337 | // return immediately. If the flag "abort_if_unrecognized" is true, then this |
duke@435 | 3338 | // routine will never retun false (zero), but instead will execute a VM panic |
duke@435 | 3339 | // routine kill the process. |
duke@435 | 3340 | // |
duke@435 | 3341 | // If this routine returns false, it is OK to call it again. This allows |
duke@435 | 3342 | // the user-defined signal handler to perform checks either before or after |
duke@435 | 3343 | // the VM performs its own checks. Naturally, the user code would be making |
duke@435 | 3344 | // a serious error if it tried to handle an exception (such as a null check |
duke@435 | 3345 | // or breakpoint) that the VM was generating for its own correct operation. |
duke@435 | 3346 | // |
duke@435 | 3347 | // This routine may recognize any of the following kinds of signals: |
duke@435 | 3348 | // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. |
duke@435 | 3349 | // It should be consulted by handlers for any of those signals. |
duke@435 | 3350 | // |
duke@435 | 3351 | // The caller of this routine must pass in the three arguments supplied |
duke@435 | 3352 | // to the function referred to in the "sa_sigaction" (not the "sa_handler") |
duke@435 | 3353 | // field of the structure passed to sigaction(). This routine assumes that |
duke@435 | 3354 | // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. |
duke@435 | 3355 | // |
duke@435 | 3356 | // Note that the VM will print warnings if it detects conflicting signal |
duke@435 | 3357 | // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". |
duke@435 | 3358 | // |
duke@435 | 3359 | extern "C" int |
duke@435 | 3360 | JVM_handle_linux_signal(int signo, siginfo_t* siginfo, |
duke@435 | 3361 | void* ucontext, int abort_if_unrecognized); |
duke@435 | 3362 | |
duke@435 | 3363 | void signalHandler(int sig, siginfo_t* info, void* uc) { |
duke@435 | 3364 | assert(info != NULL && uc != NULL, "it must be old kernel"); |
duke@435 | 3365 | JVM_handle_linux_signal(sig, info, uc, true); |
duke@435 | 3366 | } |
duke@435 | 3367 | |
duke@435 | 3368 | |
duke@435 | 3369 | // This boolean allows users to forward their own non-matching signals |
duke@435 | 3370 | // to JVM_handle_linux_signal, harmlessly. |
duke@435 | 3371 | bool os::Linux::signal_handlers_are_installed = false; |
duke@435 | 3372 | |
duke@435 | 3373 | // For signal-chaining |
duke@435 | 3374 | struct sigaction os::Linux::sigact[MAXSIGNUM]; |
duke@435 | 3375 | unsigned int os::Linux::sigs = 0; |
duke@435 | 3376 | bool os::Linux::libjsig_is_loaded = false; |
duke@435 | 3377 | typedef struct sigaction *(*get_signal_t)(int); |
duke@435 | 3378 | get_signal_t os::Linux::get_signal_action = NULL; |
duke@435 | 3379 | |
duke@435 | 3380 | struct sigaction* os::Linux::get_chained_signal_action(int sig) { |
duke@435 | 3381 | struct sigaction *actp = NULL; |
duke@435 | 3382 | |
duke@435 | 3383 | if (libjsig_is_loaded) { |
duke@435 | 3384 | // Retrieve the old signal handler from libjsig |
duke@435 | 3385 | actp = (*get_signal_action)(sig); |
duke@435 | 3386 | } |
duke@435 | 3387 | if (actp == NULL) { |
duke@435 | 3388 | // Retrieve the preinstalled signal handler from jvm |
duke@435 | 3389 | actp = get_preinstalled_handler(sig); |
duke@435 | 3390 | } |
duke@435 | 3391 | |
duke@435 | 3392 | return actp; |
duke@435 | 3393 | } |
duke@435 | 3394 | |
duke@435 | 3395 | static bool call_chained_handler(struct sigaction *actp, int sig, |
duke@435 | 3396 | siginfo_t *siginfo, void *context) { |
duke@435 | 3397 | // Call the old signal handler |
duke@435 | 3398 | if (actp->sa_handler == SIG_DFL) { |
duke@435 | 3399 | // It's more reasonable to let jvm treat it as an unexpected exception |
duke@435 | 3400 | // instead of taking the default action. |
duke@435 | 3401 | return false; |
duke@435 | 3402 | } else if (actp->sa_handler != SIG_IGN) { |
duke@435 | 3403 | if ((actp->sa_flags & SA_NODEFER) == 0) { |
duke@435 | 3404 | // automaticlly block the signal |
duke@435 | 3405 | sigaddset(&(actp->sa_mask), sig); |
duke@435 | 3406 | } |
duke@435 | 3407 | |
duke@435 | 3408 | sa_handler_t hand; |
duke@435 | 3409 | sa_sigaction_t sa; |
duke@435 | 3410 | bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; |
duke@435 | 3411 | // retrieve the chained handler |
duke@435 | 3412 | if (siginfo_flag_set) { |
duke@435 | 3413 | sa = actp->sa_sigaction; |
duke@435 | 3414 | } else { |
duke@435 | 3415 | hand = actp->sa_handler; |
duke@435 | 3416 | } |
duke@435 | 3417 | |
duke@435 | 3418 | if ((actp->sa_flags & SA_RESETHAND) != 0) { |
duke@435 | 3419 | actp->sa_handler = SIG_DFL; |
duke@435 | 3420 | } |
duke@435 | 3421 | |
duke@435 | 3422 | // try to honor the signal mask |
duke@435 | 3423 | sigset_t oset; |
duke@435 | 3424 | pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); |
duke@435 | 3425 | |
duke@435 | 3426 | // call into the chained handler |
duke@435 | 3427 | if (siginfo_flag_set) { |
duke@435 | 3428 | (*sa)(sig, siginfo, context); |
duke@435 | 3429 | } else { |
duke@435 | 3430 | (*hand)(sig); |
duke@435 | 3431 | } |
duke@435 | 3432 | |
duke@435 | 3433 | // restore the signal mask |
duke@435 | 3434 | pthread_sigmask(SIG_SETMASK, &oset, 0); |
duke@435 | 3435 | } |
duke@435 | 3436 | // Tell jvm's signal handler the signal is taken care of. |
duke@435 | 3437 | return true; |
duke@435 | 3438 | } |
duke@435 | 3439 | |
duke@435 | 3440 | bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { |
duke@435 | 3441 | bool chained = false; |
duke@435 | 3442 | // signal-chaining |
duke@435 | 3443 | if (UseSignalChaining) { |
duke@435 | 3444 | struct sigaction *actp = get_chained_signal_action(sig); |
duke@435 | 3445 | if (actp != NULL) { |
duke@435 | 3446 | chained = call_chained_handler(actp, sig, siginfo, context); |
duke@435 | 3447 | } |
duke@435 | 3448 | } |
duke@435 | 3449 | return chained; |
duke@435 | 3450 | } |
duke@435 | 3451 | |
duke@435 | 3452 | struct sigaction* os::Linux::get_preinstalled_handler(int sig) { |
duke@435 | 3453 | if ((( (unsigned int)1 << sig ) & sigs) != 0) { |
duke@435 | 3454 | return &sigact[sig]; |
duke@435 | 3455 | } |
duke@435 | 3456 | return NULL; |
duke@435 | 3457 | } |
duke@435 | 3458 | |
duke@435 | 3459 | void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { |
duke@435 | 3460 | assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); |
duke@435 | 3461 | sigact[sig] = oldAct; |
duke@435 | 3462 | sigs |= (unsigned int)1 << sig; |
duke@435 | 3463 | } |
duke@435 | 3464 | |
duke@435 | 3465 | // for diagnostic |
duke@435 | 3466 | int os::Linux::sigflags[MAXSIGNUM]; |
duke@435 | 3467 | |
duke@435 | 3468 | int os::Linux::get_our_sigflags(int sig) { |
duke@435 | 3469 | assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); |
duke@435 | 3470 | return sigflags[sig]; |
duke@435 | 3471 | } |
duke@435 | 3472 | |
duke@435 | 3473 | void os::Linux::set_our_sigflags(int sig, int flags) { |
duke@435 | 3474 | assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); |
duke@435 | 3475 | sigflags[sig] = flags; |
duke@435 | 3476 | } |
duke@435 | 3477 | |
duke@435 | 3478 | void os::Linux::set_signal_handler(int sig, bool set_installed) { |
duke@435 | 3479 | // Check for overwrite. |
duke@435 | 3480 | struct sigaction oldAct; |
duke@435 | 3481 | sigaction(sig, (struct sigaction*)NULL, &oldAct); |
duke@435 | 3482 | |
duke@435 | 3483 | void* oldhand = oldAct.sa_sigaction |
duke@435 | 3484 | ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) |
duke@435 | 3485 | : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); |
duke@435 | 3486 | if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && |
duke@435 | 3487 | oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && |
duke@435 | 3488 | oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { |
duke@435 | 3489 | if (AllowUserSignalHandlers || !set_installed) { |
duke@435 | 3490 | // Do not overwrite; user takes responsibility to forward to us. |
duke@435 | 3491 | return; |
duke@435 | 3492 | } else if (UseSignalChaining) { |
duke@435 | 3493 | // save the old handler in jvm |
duke@435 | 3494 | save_preinstalled_handler(sig, oldAct); |
duke@435 | 3495 | // libjsig also interposes the sigaction() call below and saves the |
duke@435 | 3496 | // old sigaction on it own. |
duke@435 | 3497 | } else { |
jcoomes@1845 | 3498 | fatal(err_msg("Encountered unexpected pre-existing sigaction handler " |
jcoomes@1845 | 3499 | "%#lx for signal %d.", (long)oldhand, sig)); |
duke@435 | 3500 | } |
duke@435 | 3501 | } |
duke@435 | 3502 | |
duke@435 | 3503 | struct sigaction sigAct; |
duke@435 | 3504 | sigfillset(&(sigAct.sa_mask)); |
duke@435 | 3505 | sigAct.sa_handler = SIG_DFL; |
duke@435 | 3506 | if (!set_installed) { |
duke@435 | 3507 | sigAct.sa_flags = SA_SIGINFO|SA_RESTART; |
duke@435 | 3508 | } else { |
duke@435 | 3509 | sigAct.sa_sigaction = signalHandler; |
duke@435 | 3510 | sigAct.sa_flags = SA_SIGINFO|SA_RESTART; |
duke@435 | 3511 | } |
duke@435 | 3512 | // Save flags, which are set by ours |
duke@435 | 3513 | assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); |
duke@435 | 3514 | sigflags[sig] = sigAct.sa_flags; |
duke@435 | 3515 | |
duke@435 | 3516 | int ret = sigaction(sig, &sigAct, &oldAct); |
duke@435 | 3517 | assert(ret == 0, "check"); |
duke@435 | 3518 | |
duke@435 | 3519 | void* oldhand2 = oldAct.sa_sigaction |
duke@435 | 3520 | ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) |
duke@435 | 3521 | : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); |
duke@435 | 3522 | assert(oldhand2 == oldhand, "no concurrent signal handler installation"); |
duke@435 | 3523 | } |
duke@435 | 3524 | |
duke@435 | 3525 | // install signal handlers for signals that HotSpot needs to |
duke@435 | 3526 | // handle in order to support Java-level exception handling. |
duke@435 | 3527 | |
duke@435 | 3528 | void os::Linux::install_signal_handlers() { |
duke@435 | 3529 | if (!signal_handlers_are_installed) { |
duke@435 | 3530 | signal_handlers_are_installed = true; |
duke@435 | 3531 | |
duke@435 | 3532 | // signal-chaining |
duke@435 | 3533 | typedef void (*signal_setting_t)(); |
duke@435 | 3534 | signal_setting_t begin_signal_setting = NULL; |
duke@435 | 3535 | signal_setting_t end_signal_setting = NULL; |
duke@435 | 3536 | begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, |
duke@435 | 3537 | dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); |
duke@435 | 3538 | if (begin_signal_setting != NULL) { |
duke@435 | 3539 | end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, |
duke@435 | 3540 | dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); |
duke@435 | 3541 | get_signal_action = CAST_TO_FN_PTR(get_signal_t, |
duke@435 | 3542 | dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); |
duke@435 | 3543 | libjsig_is_loaded = true; |
duke@435 | 3544 | assert(UseSignalChaining, "should enable signal-chaining"); |
duke@435 | 3545 | } |
duke@435 | 3546 | if (libjsig_is_loaded) { |
duke@435 | 3547 | // Tell libjsig jvm is setting signal handlers |
duke@435 | 3548 | (*begin_signal_setting)(); |
duke@435 | 3549 | } |
duke@435 | 3550 | |
duke@435 | 3551 | set_signal_handler(SIGSEGV, true); |
duke@435 | 3552 | set_signal_handler(SIGPIPE, true); |
duke@435 | 3553 | set_signal_handler(SIGBUS, true); |
duke@435 | 3554 | set_signal_handler(SIGILL, true); |
duke@435 | 3555 | set_signal_handler(SIGFPE, true); |
duke@435 | 3556 | set_signal_handler(SIGXFSZ, true); |
duke@435 | 3557 | |
duke@435 | 3558 | if (libjsig_is_loaded) { |
duke@435 | 3559 | // Tell libjsig jvm finishes setting signal handlers |
duke@435 | 3560 | (*end_signal_setting)(); |
duke@435 | 3561 | } |
duke@435 | 3562 | |
duke@435 | 3563 | // We don't activate signal checker if libjsig is in place, we trust ourselves |
duke@435 | 3564 | // and if UserSignalHandler is installed all bets are off |
duke@435 | 3565 | if (CheckJNICalls) { |
duke@435 | 3566 | if (libjsig_is_loaded) { |
duke@435 | 3567 | tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); |
duke@435 | 3568 | check_signals = false; |
duke@435 | 3569 | } |
duke@435 | 3570 | if (AllowUserSignalHandlers) { |
duke@435 | 3571 | tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); |
duke@435 | 3572 | check_signals = false; |
duke@435 | 3573 | } |
duke@435 | 3574 | } |
duke@435 | 3575 | } |
duke@435 | 3576 | } |
duke@435 | 3577 | |
duke@435 | 3578 | // This is the fastest way to get thread cpu time on Linux. |
duke@435 | 3579 | // Returns cpu time (user+sys) for any thread, not only for current. |
duke@435 | 3580 | // POSIX compliant clocks are implemented in the kernels 2.6.16+. |
duke@435 | 3581 | // It might work on 2.6.10+ with a special kernel/glibc patch. |
duke@435 | 3582 | // For reference, please, see IEEE Std 1003.1-2004: |
duke@435 | 3583 | // http://www.unix.org/single_unix_specification |
duke@435 | 3584 | |
duke@435 | 3585 | jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { |
duke@435 | 3586 | struct timespec tp; |
duke@435 | 3587 | int rc = os::Linux::clock_gettime(clockid, &tp); |
duke@435 | 3588 | assert(rc == 0, "clock_gettime is expected to return 0 code"); |
duke@435 | 3589 | |
duke@435 | 3590 | return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; |
duke@435 | 3591 | } |
duke@435 | 3592 | |
duke@435 | 3593 | ///// |
duke@435 | 3594 | // glibc on Linux platform uses non-documented flag |
duke@435 | 3595 | // to indicate, that some special sort of signal |
duke@435 | 3596 | // trampoline is used. |
duke@435 | 3597 | // We will never set this flag, and we should |
duke@435 | 3598 | // ignore this flag in our diagnostic |
duke@435 | 3599 | #ifdef SIGNIFICANT_SIGNAL_MASK |
duke@435 | 3600 | #undef SIGNIFICANT_SIGNAL_MASK |
duke@435 | 3601 | #endif |
duke@435 | 3602 | #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) |
duke@435 | 3603 | |
duke@435 | 3604 | static const char* get_signal_handler_name(address handler, |
duke@435 | 3605 | char* buf, int buflen) { |
duke@435 | 3606 | int offset; |
duke@435 | 3607 | bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); |
duke@435 | 3608 | if (found) { |
duke@435 | 3609 | // skip directory names |
duke@435 | 3610 | const char *p1, *p2; |
duke@435 | 3611 | p1 = buf; |
duke@435 | 3612 | size_t len = strlen(os::file_separator()); |
duke@435 | 3613 | while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; |
duke@435 | 3614 | jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); |
duke@435 | 3615 | } else { |
duke@435 | 3616 | jio_snprintf(buf, buflen, PTR_FORMAT, handler); |
duke@435 | 3617 | } |
duke@435 | 3618 | return buf; |
duke@435 | 3619 | } |
duke@435 | 3620 | |
duke@435 | 3621 | static void print_signal_handler(outputStream* st, int sig, |
duke@435 | 3622 | char* buf, size_t buflen) { |
duke@435 | 3623 | struct sigaction sa; |
duke@435 | 3624 | |
duke@435 | 3625 | sigaction(sig, NULL, &sa); |
duke@435 | 3626 | |
duke@435 | 3627 | // See comment for SIGNIFICANT_SIGNAL_MASK define |
duke@435 | 3628 | sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; |
duke@435 | 3629 | |
duke@435 | 3630 | st->print("%s: ", os::exception_name(sig, buf, buflen)); |
duke@435 | 3631 | |
duke@435 | 3632 | address handler = (sa.sa_flags & SA_SIGINFO) |
duke@435 | 3633 | ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) |
duke@435 | 3634 | : CAST_FROM_FN_PTR(address, sa.sa_handler); |
duke@435 | 3635 | |
duke@435 | 3636 | if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { |
duke@435 | 3637 | st->print("SIG_DFL"); |
duke@435 | 3638 | } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { |
duke@435 | 3639 | st->print("SIG_IGN"); |
duke@435 | 3640 | } else { |
duke@435 | 3641 | st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); |
duke@435 | 3642 | } |
duke@435 | 3643 | |
duke@435 | 3644 | st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); |
duke@435 | 3645 | |
duke@435 | 3646 | address rh = VMError::get_resetted_sighandler(sig); |
duke@435 | 3647 | // May be, handler was resetted by VMError? |
duke@435 | 3648 | if(rh != NULL) { |
duke@435 | 3649 | handler = rh; |
duke@435 | 3650 | sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; |
duke@435 | 3651 | } |
duke@435 | 3652 | |
duke@435 | 3653 | st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); |
duke@435 | 3654 | |
duke@435 | 3655 | // Check: is it our handler? |
duke@435 | 3656 | if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || |
duke@435 | 3657 | handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { |
duke@435 | 3658 | // It is our signal handler |
duke@435 | 3659 | // check for flags, reset system-used one! |
duke@435 | 3660 | if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { |
duke@435 | 3661 | st->print( |
duke@435 | 3662 | ", flags was changed from " PTR32_FORMAT ", consider using jsig library", |
duke@435 | 3663 | os::Linux::get_our_sigflags(sig)); |
duke@435 | 3664 | } |
duke@435 | 3665 | } |
duke@435 | 3666 | st->cr(); |
duke@435 | 3667 | } |
duke@435 | 3668 | |
duke@435 | 3669 | |
duke@435 | 3670 | #define DO_SIGNAL_CHECK(sig) \ |
duke@435 | 3671 | if (!sigismember(&check_signal_done, sig)) \ |
duke@435 | 3672 | os::Linux::check_signal_handler(sig) |
duke@435 | 3673 | |
duke@435 | 3674 | // This method is a periodic task to check for misbehaving JNI applications |
duke@435 | 3675 | // under CheckJNI, we can add any periodic checks here |
duke@435 | 3676 | |
duke@435 | 3677 | void os::run_periodic_checks() { |
duke@435 | 3678 | |
duke@435 | 3679 | if (check_signals == false) return; |
duke@435 | 3680 | |
duke@435 | 3681 | // SEGV and BUS if overridden could potentially prevent |
duke@435 | 3682 | // generation of hs*.log in the event of a crash, debugging |
duke@435 | 3683 | // such a case can be very challenging, so we absolutely |
duke@435 | 3684 | // check the following for a good measure: |
duke@435 | 3685 | DO_SIGNAL_CHECK(SIGSEGV); |
duke@435 | 3686 | DO_SIGNAL_CHECK(SIGILL); |
duke@435 | 3687 | DO_SIGNAL_CHECK(SIGFPE); |
duke@435 | 3688 | DO_SIGNAL_CHECK(SIGBUS); |
duke@435 | 3689 | DO_SIGNAL_CHECK(SIGPIPE); |
duke@435 | 3690 | DO_SIGNAL_CHECK(SIGXFSZ); |
duke@435 | 3691 | |
duke@435 | 3692 | |
duke@435 | 3693 | // ReduceSignalUsage allows the user to override these handlers |
duke@435 | 3694 | // see comments at the very top and jvm_solaris.h |
duke@435 | 3695 | if (!ReduceSignalUsage) { |
duke@435 | 3696 | DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); |
duke@435 | 3697 | DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); |
duke@435 | 3698 | DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); |
duke@435 | 3699 | DO_SIGNAL_CHECK(BREAK_SIGNAL); |
duke@435 | 3700 | } |
duke@435 | 3701 | |
duke@435 | 3702 | DO_SIGNAL_CHECK(SR_signum); |
duke@435 | 3703 | DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); |
duke@435 | 3704 | } |
duke@435 | 3705 | |
duke@435 | 3706 | typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); |
duke@435 | 3707 | |
duke@435 | 3708 | static os_sigaction_t os_sigaction = NULL; |
duke@435 | 3709 | |
duke@435 | 3710 | void os::Linux::check_signal_handler(int sig) { |
duke@435 | 3711 | char buf[O_BUFLEN]; |
duke@435 | 3712 | address jvmHandler = NULL; |
duke@435 | 3713 | |
duke@435 | 3714 | |
duke@435 | 3715 | struct sigaction act; |
duke@435 | 3716 | if (os_sigaction == NULL) { |
duke@435 | 3717 | // only trust the default sigaction, in case it has been interposed |
duke@435 | 3718 | os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); |
duke@435 | 3719 | if (os_sigaction == NULL) return; |
duke@435 | 3720 | } |
duke@435 | 3721 | |
duke@435 | 3722 | os_sigaction(sig, (struct sigaction*)NULL, &act); |
duke@435 | 3723 | |
duke@435 | 3724 | |
duke@435 | 3725 | act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; |
duke@435 | 3726 | |
duke@435 | 3727 | address thisHandler = (act.sa_flags & SA_SIGINFO) |
duke@435 | 3728 | ? CAST_FROM_FN_PTR(address, act.sa_sigaction) |
duke@435 | 3729 | : CAST_FROM_FN_PTR(address, act.sa_handler) ; |
duke@435 | 3730 | |
duke@435 | 3731 | |
duke@435 | 3732 | switch(sig) { |
duke@435 | 3733 | case SIGSEGV: |
duke@435 | 3734 | case SIGBUS: |
duke@435 | 3735 | case SIGFPE: |
duke@435 | 3736 | case SIGPIPE: |
duke@435 | 3737 | case SIGILL: |
duke@435 | 3738 | case SIGXFSZ: |
duke@435 | 3739 | jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); |
duke@435 | 3740 | break; |
duke@435 | 3741 | |
duke@435 | 3742 | case SHUTDOWN1_SIGNAL: |
duke@435 | 3743 | case SHUTDOWN2_SIGNAL: |
duke@435 | 3744 | case SHUTDOWN3_SIGNAL: |
duke@435 | 3745 | case BREAK_SIGNAL: |
duke@435 | 3746 | jvmHandler = (address)user_handler(); |
duke@435 | 3747 | break; |
duke@435 | 3748 | |
duke@435 | 3749 | case INTERRUPT_SIGNAL: |
duke@435 | 3750 | jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); |
duke@435 | 3751 | break; |
duke@435 | 3752 | |
duke@435 | 3753 | default: |
duke@435 | 3754 | if (sig == SR_signum) { |
duke@435 | 3755 | jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); |
duke@435 | 3756 | } else { |
duke@435 | 3757 | return; |
duke@435 | 3758 | } |
duke@435 | 3759 | break; |
duke@435 | 3760 | } |
duke@435 | 3761 | |
duke@435 | 3762 | if (thisHandler != jvmHandler) { |
duke@435 | 3763 | tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); |
duke@435 | 3764 | tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); |
duke@435 | 3765 | tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); |
duke@435 | 3766 | // No need to check this sig any longer |
duke@435 | 3767 | sigaddset(&check_signal_done, sig); |
duke@435 | 3768 | } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { |
duke@435 | 3769 | tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); |
duke@435 | 3770 | tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); |
duke@435 | 3771 | tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); |
duke@435 | 3772 | // No need to check this sig any longer |
duke@435 | 3773 | sigaddset(&check_signal_done, sig); |
duke@435 | 3774 | } |
duke@435 | 3775 | |
duke@435 | 3776 | // Dump all the signal |
duke@435 | 3777 | if (sigismember(&check_signal_done, sig)) { |
duke@435 | 3778 | print_signal_handlers(tty, buf, O_BUFLEN); |
duke@435 | 3779 | } |
duke@435 | 3780 | } |
duke@435 | 3781 | |
duke@435 | 3782 | extern void report_error(char* file_name, int line_no, char* title, char* format, ...); |
duke@435 | 3783 | |
duke@435 | 3784 | extern bool signal_name(int signo, char* buf, size_t len); |
duke@435 | 3785 | |
duke@435 | 3786 | const char* os::exception_name(int exception_code, char* buf, size_t size) { |
duke@435 | 3787 | if (0 < exception_code && exception_code <= SIGRTMAX) { |
duke@435 | 3788 | // signal |
duke@435 | 3789 | if (!signal_name(exception_code, buf, size)) { |
duke@435 | 3790 | jio_snprintf(buf, size, "SIG%d", exception_code); |
duke@435 | 3791 | } |
duke@435 | 3792 | return buf; |
duke@435 | 3793 | } else { |
duke@435 | 3794 | return NULL; |
duke@435 | 3795 | } |
duke@435 | 3796 | } |
duke@435 | 3797 | |
duke@435 | 3798 | // this is called _before_ the most of global arguments have been parsed |
duke@435 | 3799 | void os::init(void) { |
duke@435 | 3800 | char dummy; /* used to get a guess on initial stack address */ |
duke@435 | 3801 | // first_hrtime = gethrtime(); |
duke@435 | 3802 | |
duke@435 | 3803 | // With LinuxThreads the JavaMain thread pid (primordial thread) |
duke@435 | 3804 | // is different than the pid of the java launcher thread. |
duke@435 | 3805 | // So, on Linux, the launcher thread pid is passed to the VM |
duke@435 | 3806 | // via the sun.java.launcher.pid property. |
duke@435 | 3807 | // Use this property instead of getpid() if it was correctly passed. |
duke@435 | 3808 | // See bug 6351349. |
duke@435 | 3809 | pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); |
duke@435 | 3810 | |
duke@435 | 3811 | _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); |
duke@435 | 3812 | |
duke@435 | 3813 | clock_tics_per_sec = sysconf(_SC_CLK_TCK); |
duke@435 | 3814 | |
duke@435 | 3815 | init_random(1234567); |
duke@435 | 3816 | |
duke@435 | 3817 | ThreadCritical::initialize(); |
duke@435 | 3818 | |
duke@435 | 3819 | Linux::set_page_size(sysconf(_SC_PAGESIZE)); |
duke@435 | 3820 | if (Linux::page_size() == -1) { |
jcoomes@1845 | 3821 | fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", |
jcoomes@1845 | 3822 | strerror(errno))); |
duke@435 | 3823 | } |
duke@435 | 3824 | init_page_sizes((size_t) Linux::page_size()); |
duke@435 | 3825 | |
duke@435 | 3826 | Linux::initialize_system_info(); |
duke@435 | 3827 | |
duke@435 | 3828 | // main_thread points to the aboriginal thread |
duke@435 | 3829 | Linux::_main_thread = pthread_self(); |
duke@435 | 3830 | |
duke@435 | 3831 | Linux::clock_init(); |
duke@435 | 3832 | initial_time_count = os::elapsed_counter(); |
kamg@677 | 3833 | pthread_mutex_init(&dl_mutex, NULL); |
duke@435 | 3834 | } |
duke@435 | 3835 | |
duke@435 | 3836 | // To install functions for atexit system call |
duke@435 | 3837 | extern "C" { |
duke@435 | 3838 | static void perfMemory_exit_helper() { |
duke@435 | 3839 | perfMemory_exit(); |
duke@435 | 3840 | } |
duke@435 | 3841 | } |
duke@435 | 3842 | |
duke@435 | 3843 | // this is called _after_ the global arguments have been parsed |
duke@435 | 3844 | jint os::init_2(void) |
duke@435 | 3845 | { |
duke@435 | 3846 | Linux::fast_thread_clock_init(); |
duke@435 | 3847 | |
duke@435 | 3848 | // Allocate a single page and mark it as readable for safepoint polling |
duke@435 | 3849 | address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); |
duke@435 | 3850 | guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); |
duke@435 | 3851 | |
duke@435 | 3852 | os::set_polling_page( polling_page ); |
duke@435 | 3853 | |
duke@435 | 3854 | #ifndef PRODUCT |
duke@435 | 3855 | if(Verbose && PrintMiscellaneous) |
duke@435 | 3856 | tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); |
duke@435 | 3857 | #endif |
duke@435 | 3858 | |
duke@435 | 3859 | if (!UseMembar) { |
duke@435 | 3860 | address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); |
duke@435 | 3861 | guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); |
duke@435 | 3862 | os::set_memory_serialize_page( mem_serialize_page ); |
duke@435 | 3863 | |
duke@435 | 3864 | #ifndef PRODUCT |
duke@435 | 3865 | if(Verbose && PrintMiscellaneous) |
duke@435 | 3866 | tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); |
duke@435 | 3867 | #endif |
duke@435 | 3868 | } |
duke@435 | 3869 | |
duke@435 | 3870 | FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); |
duke@435 | 3871 | |
duke@435 | 3872 | // initialize suspend/resume support - must do this before signal_sets_init() |
duke@435 | 3873 | if (SR_initialize() != 0) { |
duke@435 | 3874 | perror("SR_initialize failed"); |
duke@435 | 3875 | return JNI_ERR; |
duke@435 | 3876 | } |
duke@435 | 3877 | |
duke@435 | 3878 | Linux::signal_sets_init(); |
duke@435 | 3879 | Linux::install_signal_handlers(); |
duke@435 | 3880 | |
duke@435 | 3881 | size_t threadStackSizeInBytes = ThreadStackSize * K; |
duke@435 | 3882 | if (threadStackSizeInBytes != 0 && |
duke@435 | 3883 | threadStackSizeInBytes < Linux::min_stack_allowed) { |
duke@435 | 3884 | tty->print_cr("\nThe stack size specified is too small, " |
duke@435 | 3885 | "Specify at least %dk", |
duke@435 | 3886 | Linux::min_stack_allowed / K); |
duke@435 | 3887 | return JNI_ERR; |
duke@435 | 3888 | } |
duke@435 | 3889 | |
duke@435 | 3890 | // Make the stack size a multiple of the page size so that |
duke@435 | 3891 | // the yellow/red zones can be guarded. |
duke@435 | 3892 | JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, |
duke@435 | 3893 | vm_page_size())); |
duke@435 | 3894 | |
duke@435 | 3895 | Linux::capture_initial_stack(JavaThread::stack_size_at_create()); |
duke@435 | 3896 | |
duke@435 | 3897 | Linux::libpthread_init(); |
duke@435 | 3898 | if (PrintMiscellaneous && (Verbose || WizardMode)) { |
duke@435 | 3899 | tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", |
duke@435 | 3900 | Linux::glibc_version(), Linux::libpthread_version(), |
duke@435 | 3901 | Linux::is_floating_stack() ? "floating stack" : "fixed stack"); |
duke@435 | 3902 | } |
duke@435 | 3903 | |
iveresov@576 | 3904 | if (UseNUMA) { |
iveresov@897 | 3905 | if (!Linux::libnuma_init()) { |
iveresov@897 | 3906 | UseNUMA = false; |
iveresov@897 | 3907 | } else { |
iveresov@897 | 3908 | if ((Linux::numa_max_node() < 1)) { |
iveresov@897 | 3909 | // There's only one node(they start from 0), disable NUMA. |
iveresov@897 | 3910 | UseNUMA = false; |
iveresov@897 | 3911 | } |
iveresov@897 | 3912 | } |
iveresov@897 | 3913 | if (!UseNUMA && ForceNUMA) { |
iveresov@897 | 3914 | UseNUMA = true; |
iveresov@897 | 3915 | } |
iveresov@576 | 3916 | } |
iveresov@576 | 3917 | |
duke@435 | 3918 | if (MaxFDLimit) { |
duke@435 | 3919 | // set the number of file descriptors to max. print out error |
duke@435 | 3920 | // if getrlimit/setrlimit fails but continue regardless. |
duke@435 | 3921 | struct rlimit nbr_files; |
duke@435 | 3922 | int status = getrlimit(RLIMIT_NOFILE, &nbr_files); |
duke@435 | 3923 | if (status != 0) { |
duke@435 | 3924 | if (PrintMiscellaneous && (Verbose || WizardMode)) |
duke@435 | 3925 | perror("os::init_2 getrlimit failed"); |
duke@435 | 3926 | } else { |
duke@435 | 3927 | nbr_files.rlim_cur = nbr_files.rlim_max; |
duke@435 | 3928 | status = setrlimit(RLIMIT_NOFILE, &nbr_files); |
duke@435 | 3929 | if (status != 0) { |
duke@435 | 3930 | if (PrintMiscellaneous && (Verbose || WizardMode)) |
duke@435 | 3931 | perror("os::init_2 setrlimit failed"); |
duke@435 | 3932 | } |
duke@435 | 3933 | } |
duke@435 | 3934 | } |
duke@435 | 3935 | |
duke@435 | 3936 | // Initialize lock used to serialize thread creation (see os::create_thread) |
duke@435 | 3937 | Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); |
duke@435 | 3938 | |
duke@435 | 3939 | // Initialize HPI. |
duke@435 | 3940 | jint hpi_result = hpi::initialize(); |
duke@435 | 3941 | if (hpi_result != JNI_OK) { |
duke@435 | 3942 | tty->print_cr("There was an error trying to initialize the HPI library."); |
duke@435 | 3943 | return hpi_result; |
duke@435 | 3944 | } |
duke@435 | 3945 | |
duke@435 | 3946 | // at-exit methods are called in the reverse order of their registration. |
duke@435 | 3947 | // atexit functions are called on return from main or as a result of a |
duke@435 | 3948 | // call to exit(3C). There can be only 32 of these functions registered |
duke@435 | 3949 | // and atexit() does not set errno. |
duke@435 | 3950 | |
duke@435 | 3951 | if (PerfAllowAtExitRegistration) { |
duke@435 | 3952 | // only register atexit functions if PerfAllowAtExitRegistration is set. |
duke@435 | 3953 | // atexit functions can be delayed until process exit time, which |
duke@435 | 3954 | // can be problematic for embedded VM situations. Embedded VMs should |
duke@435 | 3955 | // call DestroyJavaVM() to assure that VM resources are released. |
duke@435 | 3956 | |
duke@435 | 3957 | // note: perfMemory_exit_helper atexit function may be removed in |
duke@435 | 3958 | // the future if the appropriate cleanup code can be added to the |
duke@435 | 3959 | // VM_Exit VMOperation's doit method. |
duke@435 | 3960 | if (atexit(perfMemory_exit_helper) != 0) { |
duke@435 | 3961 | warning("os::init2 atexit(perfMemory_exit_helper) failed"); |
duke@435 | 3962 | } |
duke@435 | 3963 | } |
duke@435 | 3964 | |
duke@435 | 3965 | // initialize thread priority policy |
duke@435 | 3966 | prio_init(); |
duke@435 | 3967 | |
duke@435 | 3968 | return JNI_OK; |
duke@435 | 3969 | } |
duke@435 | 3970 | |
duke@435 | 3971 | // Mark the polling page as unreadable |
duke@435 | 3972 | void os::make_polling_page_unreadable(void) { |
duke@435 | 3973 | if( !guard_memory((char*)_polling_page, Linux::page_size()) ) |
duke@435 | 3974 | fatal("Could not disable polling page"); |
duke@435 | 3975 | }; |
duke@435 | 3976 | |
duke@435 | 3977 | // Mark the polling page as readable |
duke@435 | 3978 | void os::make_polling_page_readable(void) { |
coleenp@672 | 3979 | if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { |
duke@435 | 3980 | fatal("Could not enable polling page"); |
coleenp@672 | 3981 | } |
duke@435 | 3982 | }; |
duke@435 | 3983 | |
duke@435 | 3984 | int os::active_processor_count() { |
duke@435 | 3985 | // Linux doesn't yet have a (official) notion of processor sets, |
duke@435 | 3986 | // so just return the number of online processors. |
duke@435 | 3987 | int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); |
duke@435 | 3988 | assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); |
duke@435 | 3989 | return online_cpus; |
duke@435 | 3990 | } |
duke@435 | 3991 | |
duke@435 | 3992 | bool os::distribute_processes(uint length, uint* distribution) { |
duke@435 | 3993 | // Not yet implemented. |
duke@435 | 3994 | return false; |
duke@435 | 3995 | } |
duke@435 | 3996 | |
duke@435 | 3997 | bool os::bind_to_processor(uint processor_id) { |
duke@435 | 3998 | // Not yet implemented. |
duke@435 | 3999 | return false; |
duke@435 | 4000 | } |
duke@435 | 4001 | |
duke@435 | 4002 | /// |
duke@435 | 4003 | |
duke@435 | 4004 | // Suspends the target using the signal mechanism and then grabs the PC before |
duke@435 | 4005 | // resuming the target. Used by the flat-profiler only |
duke@435 | 4006 | ExtendedPC os::get_thread_pc(Thread* thread) { |
duke@435 | 4007 | // Make sure that it is called by the watcher for the VMThread |
duke@435 | 4008 | assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); |
duke@435 | 4009 | assert(thread->is_VM_thread(), "Can only be called for VMThread"); |
duke@435 | 4010 | |
duke@435 | 4011 | ExtendedPC epc; |
duke@435 | 4012 | |
duke@435 | 4013 | OSThread* osthread = thread->osthread(); |
duke@435 | 4014 | if (do_suspend(osthread)) { |
duke@435 | 4015 | if (osthread->ucontext() != NULL) { |
duke@435 | 4016 | epc = os::Linux::ucontext_get_pc(osthread->ucontext()); |
duke@435 | 4017 | } else { |
duke@435 | 4018 | // NULL context is unexpected, double-check this is the VMThread |
duke@435 | 4019 | guarantee(thread->is_VM_thread(), "can only be called for VMThread"); |
duke@435 | 4020 | } |
duke@435 | 4021 | do_resume(osthread); |
duke@435 | 4022 | } |
duke@435 | 4023 | // failure means pthread_kill failed for some reason - arguably this is |
duke@435 | 4024 | // a fatal problem, but such problems are ignored elsewhere |
duke@435 | 4025 | |
duke@435 | 4026 | return epc; |
duke@435 | 4027 | } |
duke@435 | 4028 | |
duke@435 | 4029 | int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) |
duke@435 | 4030 | { |
duke@435 | 4031 | if (is_NPTL()) { |
duke@435 | 4032 | return pthread_cond_timedwait(_cond, _mutex, _abstime); |
duke@435 | 4033 | } else { |
duke@435 | 4034 | #ifndef IA64 |
duke@435 | 4035 | // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control |
duke@435 | 4036 | // word back to default 64bit precision if condvar is signaled. Java |
duke@435 | 4037 | // wants 53bit precision. Save and restore current value. |
duke@435 | 4038 | int fpu = get_fpu_control_word(); |
duke@435 | 4039 | #endif // IA64 |
duke@435 | 4040 | int status = pthread_cond_timedwait(_cond, _mutex, _abstime); |
duke@435 | 4041 | #ifndef IA64 |
duke@435 | 4042 | set_fpu_control_word(fpu); |
duke@435 | 4043 | #endif // IA64 |
duke@435 | 4044 | return status; |
duke@435 | 4045 | } |
duke@435 | 4046 | } |
duke@435 | 4047 | |
duke@435 | 4048 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 4049 | // debug support |
duke@435 | 4050 | |
duke@435 | 4051 | #ifndef PRODUCT |
duke@435 | 4052 | static address same_page(address x, address y) { |
duke@435 | 4053 | int page_bits = -os::vm_page_size(); |
duke@435 | 4054 | if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) |
duke@435 | 4055 | return x; |
duke@435 | 4056 | else if (x > y) |
duke@435 | 4057 | return (address)(intptr_t(y) | ~page_bits) + 1; |
duke@435 | 4058 | else |
duke@435 | 4059 | return (address)(intptr_t(y) & page_bits); |
duke@435 | 4060 | } |
duke@435 | 4061 | |
duke@435 | 4062 | bool os::find(address addr) { |
duke@435 | 4063 | Dl_info dlinfo; |
duke@435 | 4064 | memset(&dlinfo, 0, sizeof(dlinfo)); |
duke@435 | 4065 | if (dladdr(addr, &dlinfo)) { |
duke@435 | 4066 | tty->print(PTR_FORMAT ": ", addr); |
duke@435 | 4067 | if (dlinfo.dli_sname != NULL) { |
duke@435 | 4068 | tty->print("%s+%#x", dlinfo.dli_sname, |
duke@435 | 4069 | addr - (intptr_t)dlinfo.dli_saddr); |
duke@435 | 4070 | } else if (dlinfo.dli_fname) { |
duke@435 | 4071 | tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); |
duke@435 | 4072 | } else { |
duke@435 | 4073 | tty->print("<absolute address>"); |
duke@435 | 4074 | } |
duke@435 | 4075 | if (dlinfo.dli_fname) { |
duke@435 | 4076 | tty->print(" in %s", dlinfo.dli_fname); |
duke@435 | 4077 | } |
duke@435 | 4078 | if (dlinfo.dli_fbase) { |
duke@435 | 4079 | tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); |
duke@435 | 4080 | } |
duke@435 | 4081 | tty->cr(); |
duke@435 | 4082 | |
duke@435 | 4083 | if (Verbose) { |
duke@435 | 4084 | // decode some bytes around the PC |
duke@435 | 4085 | address begin = same_page(addr-40, addr); |
duke@435 | 4086 | address end = same_page(addr+40, addr); |
duke@435 | 4087 | address lowest = (address) dlinfo.dli_sname; |
duke@435 | 4088 | if (!lowest) lowest = (address) dlinfo.dli_fbase; |
duke@435 | 4089 | if (begin < lowest) begin = lowest; |
duke@435 | 4090 | Dl_info dlinfo2; |
duke@435 | 4091 | if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr |
duke@435 | 4092 | && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) |
duke@435 | 4093 | end = (address) dlinfo2.dli_saddr; |
duke@435 | 4094 | Disassembler::decode(begin, end); |
duke@435 | 4095 | } |
duke@435 | 4096 | return true; |
duke@435 | 4097 | } |
duke@435 | 4098 | return false; |
duke@435 | 4099 | } |
duke@435 | 4100 | |
duke@435 | 4101 | #endif |
duke@435 | 4102 | |
duke@435 | 4103 | //////////////////////////////////////////////////////////////////////////////// |
duke@435 | 4104 | // misc |
duke@435 | 4105 | |
duke@435 | 4106 | // This does not do anything on Linux. This is basically a hook for being |
duke@435 | 4107 | // able to use structured exception handling (thread-local exception filters) |
duke@435 | 4108 | // on, e.g., Win32. |
duke@435 | 4109 | void |
duke@435 | 4110 | os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, |
duke@435 | 4111 | JavaCallArguments* args, Thread* thread) { |
duke@435 | 4112 | f(value, method, args, thread); |
duke@435 | 4113 | } |
duke@435 | 4114 | |
duke@435 | 4115 | void os::print_statistics() { |
duke@435 | 4116 | } |
duke@435 | 4117 | |
duke@435 | 4118 | int os::message_box(const char* title, const char* message) { |
duke@435 | 4119 | int i; |
duke@435 | 4120 | fdStream err(defaultStream::error_fd()); |
duke@435 | 4121 | for (i = 0; i < 78; i++) err.print_raw("="); |
duke@435 | 4122 | err.cr(); |
duke@435 | 4123 | err.print_raw_cr(title); |
duke@435 | 4124 | for (i = 0; i < 78; i++) err.print_raw("-"); |
duke@435 | 4125 | err.cr(); |
duke@435 | 4126 | err.print_raw_cr(message); |
duke@435 | 4127 | for (i = 0; i < 78; i++) err.print_raw("="); |
duke@435 | 4128 | err.cr(); |
duke@435 | 4129 | |
duke@435 | 4130 | char buf[16]; |
duke@435 | 4131 | // Prevent process from exiting upon "read error" without consuming all CPU |
duke@435 | 4132 | while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } |
duke@435 | 4133 | |
duke@435 | 4134 | return buf[0] == 'y' || buf[0] == 'Y'; |
duke@435 | 4135 | } |
duke@435 | 4136 | |
duke@435 | 4137 | int os::stat(const char *path, struct stat *sbuf) { |
duke@435 | 4138 | char pathbuf[MAX_PATH]; |
duke@435 | 4139 | if (strlen(path) > MAX_PATH - 1) { |
duke@435 | 4140 | errno = ENAMETOOLONG; |
duke@435 | 4141 | return -1; |
duke@435 | 4142 | } |
duke@435 | 4143 | hpi::native_path(strcpy(pathbuf, path)); |
duke@435 | 4144 | return ::stat(pathbuf, sbuf); |
duke@435 | 4145 | } |
duke@435 | 4146 | |
duke@435 | 4147 | bool os::check_heap(bool force) { |
duke@435 | 4148 | return true; |
duke@435 | 4149 | } |
duke@435 | 4150 | |
duke@435 | 4151 | int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { |
duke@435 | 4152 | return ::vsnprintf(buf, count, format, args); |
duke@435 | 4153 | } |
duke@435 | 4154 | |
duke@435 | 4155 | // Is a (classpath) directory empty? |
duke@435 | 4156 | bool os::dir_is_empty(const char* path) { |
duke@435 | 4157 | DIR *dir = NULL; |
duke@435 | 4158 | struct dirent *ptr; |
duke@435 | 4159 | |
duke@435 | 4160 | dir = opendir(path); |
duke@435 | 4161 | if (dir == NULL) return true; |
duke@435 | 4162 | |
duke@435 | 4163 | /* Scan the directory */ |
duke@435 | 4164 | bool result = true; |
duke@435 | 4165 | char buf[sizeof(struct dirent) + MAX_PATH]; |
duke@435 | 4166 | while (result && (ptr = ::readdir(dir)) != NULL) { |
duke@435 | 4167 | if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { |
duke@435 | 4168 | result = false; |
duke@435 | 4169 | } |
duke@435 | 4170 | } |
duke@435 | 4171 | closedir(dir); |
duke@435 | 4172 | return result; |
duke@435 | 4173 | } |
duke@435 | 4174 | |
duke@435 | 4175 | // create binary file, rewriting existing file if required |
duke@435 | 4176 | int os::create_binary_file(const char* path, bool rewrite_existing) { |
duke@435 | 4177 | int oflags = O_WRONLY | O_CREAT; |
duke@435 | 4178 | if (!rewrite_existing) { |
duke@435 | 4179 | oflags |= O_EXCL; |
duke@435 | 4180 | } |
duke@435 | 4181 | return ::open64(path, oflags, S_IREAD | S_IWRITE); |
duke@435 | 4182 | } |
duke@435 | 4183 | |
duke@435 | 4184 | // return current position of file pointer |
duke@435 | 4185 | jlong os::current_file_offset(int fd) { |
duke@435 | 4186 | return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); |
duke@435 | 4187 | } |
duke@435 | 4188 | |
duke@435 | 4189 | // move file pointer to the specified offset |
duke@435 | 4190 | jlong os::seek_to_file_offset(int fd, jlong offset) { |
duke@435 | 4191 | return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); |
duke@435 | 4192 | } |
duke@435 | 4193 | |
duke@435 | 4194 | // Map a block of memory. |
duke@435 | 4195 | char* os::map_memory(int fd, const char* file_name, size_t file_offset, |
duke@435 | 4196 | char *addr, size_t bytes, bool read_only, |
duke@435 | 4197 | bool allow_exec) { |
duke@435 | 4198 | int prot; |
duke@435 | 4199 | int flags; |
duke@435 | 4200 | |
duke@435 | 4201 | if (read_only) { |
duke@435 | 4202 | prot = PROT_READ; |
duke@435 | 4203 | flags = MAP_SHARED; |
duke@435 | 4204 | } else { |
duke@435 | 4205 | prot = PROT_READ | PROT_WRITE; |
duke@435 | 4206 | flags = MAP_PRIVATE; |
duke@435 | 4207 | } |
duke@435 | 4208 | |
duke@435 | 4209 | if (allow_exec) { |
duke@435 | 4210 | prot |= PROT_EXEC; |
duke@435 | 4211 | } |
duke@435 | 4212 | |
duke@435 | 4213 | if (addr != NULL) { |
duke@435 | 4214 | flags |= MAP_FIXED; |
duke@435 | 4215 | } |
duke@435 | 4216 | |
duke@435 | 4217 | char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, |
duke@435 | 4218 | fd, file_offset); |
duke@435 | 4219 | if (mapped_address == MAP_FAILED) { |
duke@435 | 4220 | return NULL; |
duke@435 | 4221 | } |
duke@435 | 4222 | return mapped_address; |
duke@435 | 4223 | } |
duke@435 | 4224 | |
duke@435 | 4225 | |
duke@435 | 4226 | // Remap a block of memory. |
duke@435 | 4227 | char* os::remap_memory(int fd, const char* file_name, size_t file_offset, |
duke@435 | 4228 | char *addr, size_t bytes, bool read_only, |
duke@435 | 4229 | bool allow_exec) { |
duke@435 | 4230 | // same as map_memory() on this OS |
duke@435 | 4231 | return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, |
duke@435 | 4232 | allow_exec); |
duke@435 | 4233 | } |
duke@435 | 4234 | |
duke@435 | 4235 | |
duke@435 | 4236 | // Unmap a block of memory. |
duke@435 | 4237 | bool os::unmap_memory(char* addr, size_t bytes) { |
duke@435 | 4238 | return munmap(addr, bytes) == 0; |
duke@435 | 4239 | } |
duke@435 | 4240 | |
duke@435 | 4241 | static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); |
duke@435 | 4242 | |
duke@435 | 4243 | static clockid_t thread_cpu_clockid(Thread* thread) { |
duke@435 | 4244 | pthread_t tid = thread->osthread()->pthread_id(); |
duke@435 | 4245 | clockid_t clockid; |
duke@435 | 4246 | |
duke@435 | 4247 | // Get thread clockid |
duke@435 | 4248 | int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); |
duke@435 | 4249 | assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); |
duke@435 | 4250 | return clockid; |
duke@435 | 4251 | } |
duke@435 | 4252 | |
duke@435 | 4253 | // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) |
duke@435 | 4254 | // are used by JVM M&M and JVMTI to get user+sys or user CPU time |
duke@435 | 4255 | // of a thread. |
duke@435 | 4256 | // |
duke@435 | 4257 | // current_thread_cpu_time() and thread_cpu_time(Thread*) returns |
duke@435 | 4258 | // the fast estimate available on the platform. |
duke@435 | 4259 | |
duke@435 | 4260 | jlong os::current_thread_cpu_time() { |
duke@435 | 4261 | if (os::Linux::supports_fast_thread_cpu_time()) { |
duke@435 | 4262 | return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); |
duke@435 | 4263 | } else { |
duke@435 | 4264 | // return user + sys since the cost is the same |
duke@435 | 4265 | return slow_thread_cpu_time(Thread::current(), true /* user + sys */); |
duke@435 | 4266 | } |
duke@435 | 4267 | } |
duke@435 | 4268 | |
duke@435 | 4269 | jlong os::thread_cpu_time(Thread* thread) { |
duke@435 | 4270 | // consistent with what current_thread_cpu_time() returns |
duke@435 | 4271 | if (os::Linux::supports_fast_thread_cpu_time()) { |
duke@435 | 4272 | return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); |
duke@435 | 4273 | } else { |
duke@435 | 4274 | return slow_thread_cpu_time(thread, true /* user + sys */); |
duke@435 | 4275 | } |
duke@435 | 4276 | } |
duke@435 | 4277 | |
duke@435 | 4278 | jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { |
duke@435 | 4279 | if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { |
duke@435 | 4280 | return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); |
duke@435 | 4281 | } else { |
duke@435 | 4282 | return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); |
duke@435 | 4283 | } |
duke@435 | 4284 | } |
duke@435 | 4285 | |
duke@435 | 4286 | jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { |
duke@435 | 4287 | if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { |
duke@435 | 4288 | return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); |
duke@435 | 4289 | } else { |
duke@435 | 4290 | return slow_thread_cpu_time(thread, user_sys_cpu_time); |
duke@435 | 4291 | } |
duke@435 | 4292 | } |
duke@435 | 4293 | |
duke@435 | 4294 | // |
duke@435 | 4295 | // -1 on error. |
duke@435 | 4296 | // |
duke@435 | 4297 | |
duke@435 | 4298 | static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { |
duke@435 | 4299 | static bool proc_pid_cpu_avail = true; |
duke@435 | 4300 | static bool proc_task_unchecked = true; |
duke@435 | 4301 | static const char *proc_stat_path = "/proc/%d/stat"; |
duke@435 | 4302 | pid_t tid = thread->osthread()->thread_id(); |
duke@435 | 4303 | int i; |
duke@435 | 4304 | char *s; |
duke@435 | 4305 | char stat[2048]; |
duke@435 | 4306 | int statlen; |
duke@435 | 4307 | char proc_name[64]; |
duke@435 | 4308 | int count; |
duke@435 | 4309 | long sys_time, user_time; |
duke@435 | 4310 | char string[64]; |
duke@435 | 4311 | int idummy; |
duke@435 | 4312 | long ldummy; |
duke@435 | 4313 | FILE *fp; |
duke@435 | 4314 | |
duke@435 | 4315 | // We first try accessing /proc/<pid>/cpu since this is faster to |
duke@435 | 4316 | // process. If this file is not present (linux kernels 2.5 and above) |
duke@435 | 4317 | // then we open /proc/<pid>/stat. |
duke@435 | 4318 | if ( proc_pid_cpu_avail ) { |
duke@435 | 4319 | sprintf(proc_name, "/proc/%d/cpu", tid); |
duke@435 | 4320 | fp = fopen(proc_name, "r"); |
duke@435 | 4321 | if ( fp != NULL ) { |
duke@435 | 4322 | count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); |
duke@435 | 4323 | fclose(fp); |
duke@435 | 4324 | if ( count != 3 ) return -1; |
duke@435 | 4325 | |
duke@435 | 4326 | if (user_sys_cpu_time) { |
duke@435 | 4327 | return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); |
duke@435 | 4328 | } else { |
duke@435 | 4329 | return (jlong)user_time * (1000000000 / clock_tics_per_sec); |
duke@435 | 4330 | } |
duke@435 | 4331 | } |
duke@435 | 4332 | else proc_pid_cpu_avail = false; |
duke@435 | 4333 | } |
duke@435 | 4334 | |
duke@435 | 4335 | // The /proc/<tid>/stat aggregates per-process usage on |
duke@435 | 4336 | // new Linux kernels 2.6+ where NPTL is supported. |
duke@435 | 4337 | // The /proc/self/task/<tid>/stat still has the per-thread usage. |
duke@435 | 4338 | // See bug 6328462. |
duke@435 | 4339 | // There can be no directory /proc/self/task on kernels 2.4 with NPTL |
duke@435 | 4340 | // and possibly in some other cases, so we check its availability. |
duke@435 | 4341 | if (proc_task_unchecked && os::Linux::is_NPTL()) { |
duke@435 | 4342 | // This is executed only once |
duke@435 | 4343 | proc_task_unchecked = false; |
duke@435 | 4344 | fp = fopen("/proc/self/task", "r"); |
duke@435 | 4345 | if (fp != NULL) { |
duke@435 | 4346 | proc_stat_path = "/proc/self/task/%d/stat"; |
duke@435 | 4347 | fclose(fp); |
duke@435 | 4348 | } |
duke@435 | 4349 | } |
duke@435 | 4350 | |
duke@435 | 4351 | sprintf(proc_name, proc_stat_path, tid); |
duke@435 | 4352 | fp = fopen(proc_name, "r"); |
duke@435 | 4353 | if ( fp == NULL ) return -1; |
duke@435 | 4354 | statlen = fread(stat, 1, 2047, fp); |
duke@435 | 4355 | stat[statlen] = '\0'; |
duke@435 | 4356 | fclose(fp); |
duke@435 | 4357 | |
duke@435 | 4358 | // Skip pid and the command string. Note that we could be dealing with |
duke@435 | 4359 | // weird command names, e.g. user could decide to rename java launcher |
duke@435 | 4360 | // to "java 1.4.2 :)", then the stat file would look like |
duke@435 | 4361 | // 1234 (java 1.4.2 :)) R ... ... |
duke@435 | 4362 | // We don't really need to know the command string, just find the last |
duke@435 | 4363 | // occurrence of ")" and then start parsing from there. See bug 4726580. |
duke@435 | 4364 | s = strrchr(stat, ')'); |
duke@435 | 4365 | i = 0; |
duke@435 | 4366 | if (s == NULL ) return -1; |
duke@435 | 4367 | |
duke@435 | 4368 | // Skip blank chars |
duke@435 | 4369 | do s++; while (isspace(*s)); |
duke@435 | 4370 | |
xlu@948 | 4371 | count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", |
xlu@948 | 4372 | &idummy, &idummy, &idummy, &idummy, &idummy, |
duke@435 | 4373 | &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, |
duke@435 | 4374 | &user_time, &sys_time); |
xlu@948 | 4375 | if ( count != 12 ) return -1; |
duke@435 | 4376 | if (user_sys_cpu_time) { |
duke@435 | 4377 | return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); |
duke@435 | 4378 | } else { |
duke@435 | 4379 | return (jlong)user_time * (1000000000 / clock_tics_per_sec); |
duke@435 | 4380 | } |
duke@435 | 4381 | } |
duke@435 | 4382 | |
duke@435 | 4383 | void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { |
duke@435 | 4384 | info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits |
duke@435 | 4385 | info_ptr->may_skip_backward = false; // elapsed time not wall time |
duke@435 | 4386 | info_ptr->may_skip_forward = false; // elapsed time not wall time |
duke@435 | 4387 | info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned |
duke@435 | 4388 | } |
duke@435 | 4389 | |
duke@435 | 4390 | void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { |
duke@435 | 4391 | info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits |
duke@435 | 4392 | info_ptr->may_skip_backward = false; // elapsed time not wall time |
duke@435 | 4393 | info_ptr->may_skip_forward = false; // elapsed time not wall time |
duke@435 | 4394 | info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned |
duke@435 | 4395 | } |
duke@435 | 4396 | |
duke@435 | 4397 | bool os::is_thread_cpu_time_supported() { |
duke@435 | 4398 | return true; |
duke@435 | 4399 | } |
duke@435 | 4400 | |
duke@435 | 4401 | // System loadavg support. Returns -1 if load average cannot be obtained. |
duke@435 | 4402 | // Linux doesn't yet have a (official) notion of processor sets, |
duke@435 | 4403 | // so just return the system wide load average. |
duke@435 | 4404 | int os::loadavg(double loadavg[], int nelem) { |
duke@435 | 4405 | return ::getloadavg(loadavg, nelem); |
duke@435 | 4406 | } |
duke@435 | 4407 | |
duke@435 | 4408 | void os::pause() { |
duke@435 | 4409 | char filename[MAX_PATH]; |
duke@435 | 4410 | if (PauseAtStartupFile && PauseAtStartupFile[0]) { |
duke@435 | 4411 | jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); |
duke@435 | 4412 | } else { |
duke@435 | 4413 | jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); |
duke@435 | 4414 | } |
duke@435 | 4415 | |
duke@435 | 4416 | int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); |
duke@435 | 4417 | if (fd != -1) { |
duke@435 | 4418 | struct stat buf; |
duke@435 | 4419 | close(fd); |
duke@435 | 4420 | while (::stat(filename, &buf) == 0) { |
duke@435 | 4421 | (void)::poll(NULL, 0, 100); |
duke@435 | 4422 | } |
duke@435 | 4423 | } else { |
duke@435 | 4424 | jio_fprintf(stderr, |
duke@435 | 4425 | "Could not open pause file '%s', continuing immediately.\n", filename); |
duke@435 | 4426 | } |
duke@435 | 4427 | } |
duke@435 | 4428 | |
duke@435 | 4429 | extern "C" { |
duke@435 | 4430 | |
duke@435 | 4431 | /** |
duke@435 | 4432 | * NOTE: the following code is to keep the green threads code |
duke@435 | 4433 | * in the libjava.so happy. Once the green threads is removed, |
duke@435 | 4434 | * these code will no longer be needed. |
duke@435 | 4435 | */ |
duke@435 | 4436 | int |
duke@435 | 4437 | jdk_waitpid(pid_t pid, int* status, int options) { |
duke@435 | 4438 | return waitpid(pid, status, options); |
duke@435 | 4439 | } |
duke@435 | 4440 | |
duke@435 | 4441 | int |
duke@435 | 4442 | fork1() { |
duke@435 | 4443 | return fork(); |
duke@435 | 4444 | } |
duke@435 | 4445 | |
duke@435 | 4446 | int |
duke@435 | 4447 | jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { |
duke@435 | 4448 | return sem_init(sem, pshared, value); |
duke@435 | 4449 | } |
duke@435 | 4450 | |
duke@435 | 4451 | int |
duke@435 | 4452 | jdk_sem_post(sem_t *sem) { |
duke@435 | 4453 | return sem_post(sem); |
duke@435 | 4454 | } |
duke@435 | 4455 | |
duke@435 | 4456 | int |
duke@435 | 4457 | jdk_sem_wait(sem_t *sem) { |
duke@435 | 4458 | return sem_wait(sem); |
duke@435 | 4459 | } |
duke@435 | 4460 | |
duke@435 | 4461 | int |
duke@435 | 4462 | jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { |
duke@435 | 4463 | return pthread_sigmask(how , newmask, oldmask); |
duke@435 | 4464 | } |
duke@435 | 4465 | |
duke@435 | 4466 | } |
duke@435 | 4467 | |
duke@435 | 4468 | // Refer to the comments in os_solaris.cpp park-unpark. |
duke@435 | 4469 | // |
duke@435 | 4470 | // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can |
duke@435 | 4471 | // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. |
duke@435 | 4472 | // For specifics regarding the bug see GLIBC BUGID 261237 : |
duke@435 | 4473 | // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. |
duke@435 | 4474 | // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future |
duke@435 | 4475 | // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar |
duke@435 | 4476 | // is used. (The simple C test-case provided in the GLIBC bug report manifests the |
duke@435 | 4477 | // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() |
duke@435 | 4478 | // and monitorenter when we're using 1-0 locking. All those operations may result in |
duke@435 | 4479 | // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version |
duke@435 | 4480 | // of libpthread avoids the problem, but isn't practical. |
duke@435 | 4481 | // |
duke@435 | 4482 | // Possible remedies: |
duke@435 | 4483 | // |
duke@435 | 4484 | // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. |
duke@435 | 4485 | // This is palliative and probabilistic, however. If the thread is preempted |
duke@435 | 4486 | // between the call to compute_abstime() and pthread_cond_timedwait(), more |
duke@435 | 4487 | // than the minimum period may have passed, and the abstime may be stale (in the |
duke@435 | 4488 | // past) resultin in a hang. Using this technique reduces the odds of a hang |
duke@435 | 4489 | // but the JVM is still vulnerable, particularly on heavily loaded systems. |
duke@435 | 4490 | // |
duke@435 | 4491 | // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead |
duke@435 | 4492 | // of the usual flag-condvar-mutex idiom. The write side of the pipe is set |
duke@435 | 4493 | // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) |
duke@435 | 4494 | // reduces to poll()+read(). This works well, but consumes 2 FDs per extant |
duke@435 | 4495 | // thread. |
duke@435 | 4496 | // |
duke@435 | 4497 | // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread |
duke@435 | 4498 | // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing |
duke@435 | 4499 | // a timeout request to the chron thread and then blocking via pthread_cond_wait(). |
duke@435 | 4500 | // This also works well. In fact it avoids kernel-level scalability impediments |
duke@435 | 4501 | // on certain platforms that don't handle lots of active pthread_cond_timedwait() |
duke@435 | 4502 | // timers in a graceful fashion. |
duke@435 | 4503 | // |
duke@435 | 4504 | // 4. When the abstime value is in the past it appears that control returns |
duke@435 | 4505 | // correctly from pthread_cond_timedwait(), but the condvar is left corrupt. |
duke@435 | 4506 | // Subsequent timedwait/wait calls may hang indefinitely. Given that, we |
duke@435 | 4507 | // can avoid the problem by reinitializing the condvar -- by cond_destroy() |
duke@435 | 4508 | // followed by cond_init() -- after all calls to pthread_cond_timedwait(). |
duke@435 | 4509 | // It may be possible to avoid reinitialization by checking the return |
duke@435 | 4510 | // value from pthread_cond_timedwait(). In addition to reinitializing the |
duke@435 | 4511 | // condvar we must establish the invariant that cond_signal() is only called |
duke@435 | 4512 | // within critical sections protected by the adjunct mutex. This prevents |
duke@435 | 4513 | // cond_signal() from "seeing" a condvar that's in the midst of being |
duke@435 | 4514 | // reinitialized or that is corrupt. Sadly, this invariant obviates the |
duke@435 | 4515 | // desirable signal-after-unlock optimization that avoids futile context switching. |
duke@435 | 4516 | // |
duke@435 | 4517 | // I'm also concerned that some versions of NTPL might allocate an auxilliary |
duke@435 | 4518 | // structure when a condvar is used or initialized. cond_destroy() would |
duke@435 | 4519 | // release the helper structure. Our reinitialize-after-timedwait fix |
duke@435 | 4520 | // put excessive stress on malloc/free and locks protecting the c-heap. |
duke@435 | 4521 | // |
duke@435 | 4522 | // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. |
duke@435 | 4523 | // It may be possible to refine (4) by checking the kernel and NTPL verisons |
duke@435 | 4524 | // and only enabling the work-around for vulnerable environments. |
duke@435 | 4525 | |
duke@435 | 4526 | // utility to compute the abstime argument to timedwait: |
duke@435 | 4527 | // millis is the relative timeout time |
duke@435 | 4528 | // abstime will be the absolute timeout time |
duke@435 | 4529 | // TODO: replace compute_abstime() with unpackTime() |
duke@435 | 4530 | |
duke@435 | 4531 | static struct timespec* compute_abstime(timespec* abstime, jlong millis) { |
duke@435 | 4532 | if (millis < 0) millis = 0; |
duke@435 | 4533 | struct timeval now; |
duke@435 | 4534 | int status = gettimeofday(&now, NULL); |
duke@435 | 4535 | assert(status == 0, "gettimeofday"); |
duke@435 | 4536 | jlong seconds = millis / 1000; |
duke@435 | 4537 | millis %= 1000; |
duke@435 | 4538 | if (seconds > 50000000) { // see man cond_timedwait(3T) |
duke@435 | 4539 | seconds = 50000000; |
duke@435 | 4540 | } |
duke@435 | 4541 | abstime->tv_sec = now.tv_sec + seconds; |
duke@435 | 4542 | long usec = now.tv_usec + millis * 1000; |
duke@435 | 4543 | if (usec >= 1000000) { |
duke@435 | 4544 | abstime->tv_sec += 1; |
duke@435 | 4545 | usec -= 1000000; |
duke@435 | 4546 | } |
duke@435 | 4547 | abstime->tv_nsec = usec * 1000; |
duke@435 | 4548 | return abstime; |
duke@435 | 4549 | } |
duke@435 | 4550 | |
duke@435 | 4551 | |
duke@435 | 4552 | // Test-and-clear _Event, always leaves _Event set to 0, returns immediately. |
duke@435 | 4553 | // Conceptually TryPark() should be equivalent to park(0). |
duke@435 | 4554 | |
duke@435 | 4555 | int os::PlatformEvent::TryPark() { |
duke@435 | 4556 | for (;;) { |
duke@435 | 4557 | const int v = _Event ; |
duke@435 | 4558 | guarantee ((v == 0) || (v == 1), "invariant") ; |
duke@435 | 4559 | if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; |
duke@435 | 4560 | } |
duke@435 | 4561 | } |
duke@435 | 4562 | |
duke@435 | 4563 | void os::PlatformEvent::park() { // AKA "down()" |
duke@435 | 4564 | // Invariant: Only the thread associated with the Event/PlatformEvent |
duke@435 | 4565 | // may call park(). |
duke@435 | 4566 | // TODO: assert that _Assoc != NULL or _Assoc == Self |
duke@435 | 4567 | int v ; |
duke@435 | 4568 | for (;;) { |
duke@435 | 4569 | v = _Event ; |
duke@435 | 4570 | if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; |
duke@435 | 4571 | } |
duke@435 | 4572 | guarantee (v >= 0, "invariant") ; |
duke@435 | 4573 | if (v == 0) { |
duke@435 | 4574 | // Do this the hard way by blocking ... |
duke@435 | 4575 | int status = pthread_mutex_lock(_mutex); |
duke@435 | 4576 | assert_status(status == 0, status, "mutex_lock"); |
duke@435 | 4577 | guarantee (_nParked == 0, "invariant") ; |
duke@435 | 4578 | ++ _nParked ; |
duke@435 | 4579 | while (_Event < 0) { |
duke@435 | 4580 | status = pthread_cond_wait(_cond, _mutex); |
duke@435 | 4581 | // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... |
duke@435 | 4582 | // Treat this the same as if the wait was interrupted |
duke@435 | 4583 | if (status == ETIME) { status = EINTR; } |
duke@435 | 4584 | assert_status(status == 0 || status == EINTR, status, "cond_wait"); |
duke@435 | 4585 | } |
duke@435 | 4586 | -- _nParked ; |
duke@435 | 4587 | |
duke@435 | 4588 | // In theory we could move the ST of 0 into _Event past the unlock(), |
duke@435 | 4589 | // but then we'd need a MEMBAR after the ST. |
duke@435 | 4590 | _Event = 0 ; |
duke@435 | 4591 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4592 | assert_status(status == 0, status, "mutex_unlock"); |
duke@435 | 4593 | } |
duke@435 | 4594 | guarantee (_Event >= 0, "invariant") ; |
duke@435 | 4595 | } |
duke@435 | 4596 | |
duke@435 | 4597 | int os::PlatformEvent::park(jlong millis) { |
duke@435 | 4598 | guarantee (_nParked == 0, "invariant") ; |
duke@435 | 4599 | |
duke@435 | 4600 | int v ; |
duke@435 | 4601 | for (;;) { |
duke@435 | 4602 | v = _Event ; |
duke@435 | 4603 | if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; |
duke@435 | 4604 | } |
duke@435 | 4605 | guarantee (v >= 0, "invariant") ; |
duke@435 | 4606 | if (v != 0) return OS_OK ; |
duke@435 | 4607 | |
duke@435 | 4608 | // We do this the hard way, by blocking the thread. |
duke@435 | 4609 | // Consider enforcing a minimum timeout value. |
duke@435 | 4610 | struct timespec abst; |
duke@435 | 4611 | compute_abstime(&abst, millis); |
duke@435 | 4612 | |
duke@435 | 4613 | int ret = OS_TIMEOUT; |
duke@435 | 4614 | int status = pthread_mutex_lock(_mutex); |
duke@435 | 4615 | assert_status(status == 0, status, "mutex_lock"); |
duke@435 | 4616 | guarantee (_nParked == 0, "invariant") ; |
duke@435 | 4617 | ++_nParked ; |
duke@435 | 4618 | |
duke@435 | 4619 | // Object.wait(timo) will return because of |
duke@435 | 4620 | // (a) notification |
duke@435 | 4621 | // (b) timeout |
duke@435 | 4622 | // (c) thread.interrupt |
duke@435 | 4623 | // |
duke@435 | 4624 | // Thread.interrupt and object.notify{All} both call Event::set. |
duke@435 | 4625 | // That is, we treat thread.interrupt as a special case of notification. |
duke@435 | 4626 | // The underlying Solaris implementation, cond_timedwait, admits |
duke@435 | 4627 | // spurious/premature wakeups, but the JLS/JVM spec prevents the |
duke@435 | 4628 | // JVM from making those visible to Java code. As such, we must |
duke@435 | 4629 | // filter out spurious wakeups. We assume all ETIME returns are valid. |
duke@435 | 4630 | // |
duke@435 | 4631 | // TODO: properly differentiate simultaneous notify+interrupt. |
duke@435 | 4632 | // In that case, we should propagate the notify to another waiter. |
duke@435 | 4633 | |
duke@435 | 4634 | while (_Event < 0) { |
duke@435 | 4635 | status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); |
duke@435 | 4636 | if (status != 0 && WorkAroundNPTLTimedWaitHang) { |
duke@435 | 4637 | pthread_cond_destroy (_cond); |
duke@435 | 4638 | pthread_cond_init (_cond, NULL) ; |
duke@435 | 4639 | } |
duke@435 | 4640 | assert_status(status == 0 || status == EINTR || |
duke@435 | 4641 | status == ETIME || status == ETIMEDOUT, |
duke@435 | 4642 | status, "cond_timedwait"); |
duke@435 | 4643 | if (!FilterSpuriousWakeups) break ; // previous semantics |
duke@435 | 4644 | if (status == ETIME || status == ETIMEDOUT) break ; |
duke@435 | 4645 | // We consume and ignore EINTR and spurious wakeups. |
duke@435 | 4646 | } |
duke@435 | 4647 | --_nParked ; |
duke@435 | 4648 | if (_Event >= 0) { |
duke@435 | 4649 | ret = OS_OK; |
duke@435 | 4650 | } |
duke@435 | 4651 | _Event = 0 ; |
duke@435 | 4652 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4653 | assert_status(status == 0, status, "mutex_unlock"); |
duke@435 | 4654 | assert (_nParked == 0, "invariant") ; |
duke@435 | 4655 | return ret; |
duke@435 | 4656 | } |
duke@435 | 4657 | |
duke@435 | 4658 | void os::PlatformEvent::unpark() { |
duke@435 | 4659 | int v, AnyWaiters ; |
duke@435 | 4660 | for (;;) { |
duke@435 | 4661 | v = _Event ; |
duke@435 | 4662 | if (v > 0) { |
duke@435 | 4663 | // The LD of _Event could have reordered or be satisfied |
duke@435 | 4664 | // by a read-aside from this processor's write buffer. |
duke@435 | 4665 | // To avoid problems execute a barrier and then |
duke@435 | 4666 | // ratify the value. |
duke@435 | 4667 | OrderAccess::fence() ; |
duke@435 | 4668 | if (_Event == v) return ; |
duke@435 | 4669 | continue ; |
duke@435 | 4670 | } |
duke@435 | 4671 | if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; |
duke@435 | 4672 | } |
duke@435 | 4673 | if (v < 0) { |
duke@435 | 4674 | // Wait for the thread associated with the event to vacate |
duke@435 | 4675 | int status = pthread_mutex_lock(_mutex); |
duke@435 | 4676 | assert_status(status == 0, status, "mutex_lock"); |
duke@435 | 4677 | AnyWaiters = _nParked ; |
duke@435 | 4678 | assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; |
duke@435 | 4679 | if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { |
duke@435 | 4680 | AnyWaiters = 0 ; |
duke@435 | 4681 | pthread_cond_signal (_cond); |
duke@435 | 4682 | } |
duke@435 | 4683 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4684 | assert_status(status == 0, status, "mutex_unlock"); |
duke@435 | 4685 | if (AnyWaiters != 0) { |
duke@435 | 4686 | status = pthread_cond_signal(_cond); |
duke@435 | 4687 | assert_status(status == 0, status, "cond_signal"); |
duke@435 | 4688 | } |
duke@435 | 4689 | } |
duke@435 | 4690 | |
duke@435 | 4691 | // Note that we signal() _after dropping the lock for "immortal" Events. |
duke@435 | 4692 | // This is safe and avoids a common class of futile wakeups. In rare |
duke@435 | 4693 | // circumstances this can cause a thread to return prematurely from |
duke@435 | 4694 | // cond_{timed}wait() but the spurious wakeup is benign and the victim will |
duke@435 | 4695 | // simply re-test the condition and re-park itself. |
duke@435 | 4696 | } |
duke@435 | 4697 | |
duke@435 | 4698 | |
duke@435 | 4699 | // JSR166 |
duke@435 | 4700 | // ------------------------------------------------------- |
duke@435 | 4701 | |
duke@435 | 4702 | /* |
duke@435 | 4703 | * The solaris and linux implementations of park/unpark are fairly |
duke@435 | 4704 | * conservative for now, but can be improved. They currently use a |
duke@435 | 4705 | * mutex/condvar pair, plus a a count. |
duke@435 | 4706 | * Park decrements count if > 0, else does a condvar wait. Unpark |
duke@435 | 4707 | * sets count to 1 and signals condvar. Only one thread ever waits |
duke@435 | 4708 | * on the condvar. Contention seen when trying to park implies that someone |
duke@435 | 4709 | * is unparking you, so don't wait. And spurious returns are fine, so there |
duke@435 | 4710 | * is no need to track notifications. |
duke@435 | 4711 | */ |
duke@435 | 4712 | |
duke@435 | 4713 | |
duke@435 | 4714 | #define NANOSECS_PER_SEC 1000000000 |
duke@435 | 4715 | #define NANOSECS_PER_MILLISEC 1000000 |
duke@435 | 4716 | #define MAX_SECS 100000000 |
duke@435 | 4717 | /* |
duke@435 | 4718 | * This code is common to linux and solaris and will be moved to a |
duke@435 | 4719 | * common place in dolphin. |
duke@435 | 4720 | * |
duke@435 | 4721 | * The passed in time value is either a relative time in nanoseconds |
duke@435 | 4722 | * or an absolute time in milliseconds. Either way it has to be unpacked |
duke@435 | 4723 | * into suitable seconds and nanoseconds components and stored in the |
duke@435 | 4724 | * given timespec structure. |
duke@435 | 4725 | * Given time is a 64-bit value and the time_t used in the timespec is only |
duke@435 | 4726 | * a signed-32-bit value (except on 64-bit Linux) we have to watch for |
duke@435 | 4727 | * overflow if times way in the future are given. Further on Solaris versions |
duke@435 | 4728 | * prior to 10 there is a restriction (see cond_timedwait) that the specified |
duke@435 | 4729 | * number of seconds, in abstime, is less than current_time + 100,000,000. |
duke@435 | 4730 | * As it will be 28 years before "now + 100000000" will overflow we can |
duke@435 | 4731 | * ignore overflow and just impose a hard-limit on seconds using the value |
duke@435 | 4732 | * of "now + 100,000,000". This places a limit on the timeout of about 3.17 |
duke@435 | 4733 | * years from "now". |
duke@435 | 4734 | */ |
duke@435 | 4735 | |
duke@435 | 4736 | static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { |
duke@435 | 4737 | assert (time > 0, "convertTime"); |
duke@435 | 4738 | |
duke@435 | 4739 | struct timeval now; |
duke@435 | 4740 | int status = gettimeofday(&now, NULL); |
duke@435 | 4741 | assert(status == 0, "gettimeofday"); |
duke@435 | 4742 | |
duke@435 | 4743 | time_t max_secs = now.tv_sec + MAX_SECS; |
duke@435 | 4744 | |
duke@435 | 4745 | if (isAbsolute) { |
duke@435 | 4746 | jlong secs = time / 1000; |
duke@435 | 4747 | if (secs > max_secs) { |
duke@435 | 4748 | absTime->tv_sec = max_secs; |
duke@435 | 4749 | } |
duke@435 | 4750 | else { |
duke@435 | 4751 | absTime->tv_sec = secs; |
duke@435 | 4752 | } |
duke@435 | 4753 | absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; |
duke@435 | 4754 | } |
duke@435 | 4755 | else { |
duke@435 | 4756 | jlong secs = time / NANOSECS_PER_SEC; |
duke@435 | 4757 | if (secs >= MAX_SECS) { |
duke@435 | 4758 | absTime->tv_sec = max_secs; |
duke@435 | 4759 | absTime->tv_nsec = 0; |
duke@435 | 4760 | } |
duke@435 | 4761 | else { |
duke@435 | 4762 | absTime->tv_sec = now.tv_sec + secs; |
duke@435 | 4763 | absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; |
duke@435 | 4764 | if (absTime->tv_nsec >= NANOSECS_PER_SEC) { |
duke@435 | 4765 | absTime->tv_nsec -= NANOSECS_PER_SEC; |
duke@435 | 4766 | ++absTime->tv_sec; // note: this must be <= max_secs |
duke@435 | 4767 | } |
duke@435 | 4768 | } |
duke@435 | 4769 | } |
duke@435 | 4770 | assert(absTime->tv_sec >= 0, "tv_sec < 0"); |
duke@435 | 4771 | assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); |
duke@435 | 4772 | assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); |
duke@435 | 4773 | assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); |
duke@435 | 4774 | } |
duke@435 | 4775 | |
duke@435 | 4776 | void Parker::park(bool isAbsolute, jlong time) { |
duke@435 | 4777 | // Optional fast-path check: |
duke@435 | 4778 | // Return immediately if a permit is available. |
duke@435 | 4779 | if (_counter > 0) { |
duke@435 | 4780 | _counter = 0 ; |
dholmes@1552 | 4781 | OrderAccess::fence(); |
duke@435 | 4782 | return ; |
duke@435 | 4783 | } |
duke@435 | 4784 | |
duke@435 | 4785 | Thread* thread = Thread::current(); |
duke@435 | 4786 | assert(thread->is_Java_thread(), "Must be JavaThread"); |
duke@435 | 4787 | JavaThread *jt = (JavaThread *)thread; |
duke@435 | 4788 | |
duke@435 | 4789 | // Optional optimization -- avoid state transitions if there's an interrupt pending. |
duke@435 | 4790 | // Check interrupt before trying to wait |
duke@435 | 4791 | if (Thread::is_interrupted(thread, false)) { |
duke@435 | 4792 | return; |
duke@435 | 4793 | } |
duke@435 | 4794 | |
duke@435 | 4795 | // Next, demultiplex/decode time arguments |
duke@435 | 4796 | timespec absTime; |
duke@435 | 4797 | if (time < 0) { // don't wait at all |
duke@435 | 4798 | return; |
duke@435 | 4799 | } |
duke@435 | 4800 | if (time > 0) { |
duke@435 | 4801 | unpackTime(&absTime, isAbsolute, time); |
duke@435 | 4802 | } |
duke@435 | 4803 | |
duke@435 | 4804 | |
duke@435 | 4805 | // Enter safepoint region |
duke@435 | 4806 | // Beware of deadlocks such as 6317397. |
duke@435 | 4807 | // The per-thread Parker:: mutex is a classic leaf-lock. |
duke@435 | 4808 | // In particular a thread must never block on the Threads_lock while |
duke@435 | 4809 | // holding the Parker:: mutex. If safepoints are pending both the |
duke@435 | 4810 | // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. |
duke@435 | 4811 | ThreadBlockInVM tbivm(jt); |
duke@435 | 4812 | |
duke@435 | 4813 | // Don't wait if cannot get lock since interference arises from |
duke@435 | 4814 | // unblocking. Also. check interrupt before trying wait |
duke@435 | 4815 | if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { |
duke@435 | 4816 | return; |
duke@435 | 4817 | } |
duke@435 | 4818 | |
duke@435 | 4819 | int status ; |
duke@435 | 4820 | if (_counter > 0) { // no wait needed |
duke@435 | 4821 | _counter = 0; |
duke@435 | 4822 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4823 | assert (status == 0, "invariant") ; |
dholmes@1552 | 4824 | OrderAccess::fence(); |
duke@435 | 4825 | return; |
duke@435 | 4826 | } |
duke@435 | 4827 | |
duke@435 | 4828 | #ifdef ASSERT |
duke@435 | 4829 | // Don't catch signals while blocked; let the running threads have the signals. |
duke@435 | 4830 | // (This allows a debugger to break into the running thread.) |
duke@435 | 4831 | sigset_t oldsigs; |
duke@435 | 4832 | sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); |
duke@435 | 4833 | pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); |
duke@435 | 4834 | #endif |
duke@435 | 4835 | |
duke@435 | 4836 | OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); |
duke@435 | 4837 | jt->set_suspend_equivalent(); |
duke@435 | 4838 | // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() |
duke@435 | 4839 | |
duke@435 | 4840 | if (time == 0) { |
duke@435 | 4841 | status = pthread_cond_wait (_cond, _mutex) ; |
duke@435 | 4842 | } else { |
duke@435 | 4843 | status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; |
duke@435 | 4844 | if (status != 0 && WorkAroundNPTLTimedWaitHang) { |
duke@435 | 4845 | pthread_cond_destroy (_cond) ; |
duke@435 | 4846 | pthread_cond_init (_cond, NULL); |
duke@435 | 4847 | } |
duke@435 | 4848 | } |
duke@435 | 4849 | assert_status(status == 0 || status == EINTR || |
duke@435 | 4850 | status == ETIME || status == ETIMEDOUT, |
duke@435 | 4851 | status, "cond_timedwait"); |
duke@435 | 4852 | |
duke@435 | 4853 | #ifdef ASSERT |
duke@435 | 4854 | pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); |
duke@435 | 4855 | #endif |
duke@435 | 4856 | |
duke@435 | 4857 | _counter = 0 ; |
duke@435 | 4858 | status = pthread_mutex_unlock(_mutex) ; |
duke@435 | 4859 | assert_status(status == 0, status, "invariant") ; |
duke@435 | 4860 | // If externally suspended while waiting, re-suspend |
duke@435 | 4861 | if (jt->handle_special_suspend_equivalent_condition()) { |
duke@435 | 4862 | jt->java_suspend_self(); |
duke@435 | 4863 | } |
duke@435 | 4864 | |
dholmes@1552 | 4865 | OrderAccess::fence(); |
duke@435 | 4866 | } |
duke@435 | 4867 | |
duke@435 | 4868 | void Parker::unpark() { |
duke@435 | 4869 | int s, status ; |
duke@435 | 4870 | status = pthread_mutex_lock(_mutex); |
duke@435 | 4871 | assert (status == 0, "invariant") ; |
duke@435 | 4872 | s = _counter; |
duke@435 | 4873 | _counter = 1; |
duke@435 | 4874 | if (s < 1) { |
duke@435 | 4875 | if (WorkAroundNPTLTimedWaitHang) { |
duke@435 | 4876 | status = pthread_cond_signal (_cond) ; |
duke@435 | 4877 | assert (status == 0, "invariant") ; |
duke@435 | 4878 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4879 | assert (status == 0, "invariant") ; |
duke@435 | 4880 | } else { |
duke@435 | 4881 | status = pthread_mutex_unlock(_mutex); |
duke@435 | 4882 | assert (status == 0, "invariant") ; |
duke@435 | 4883 | status = pthread_cond_signal (_cond) ; |
duke@435 | 4884 | assert (status == 0, "invariant") ; |
duke@435 | 4885 | } |
duke@435 | 4886 | } else { |
duke@435 | 4887 | pthread_mutex_unlock(_mutex); |
duke@435 | 4888 | assert (status == 0, "invariant") ; |
duke@435 | 4889 | } |
duke@435 | 4890 | } |
duke@435 | 4891 | |
duke@435 | 4892 | |
duke@435 | 4893 | extern char** environ; |
duke@435 | 4894 | |
duke@435 | 4895 | #ifndef __NR_fork |
duke@435 | 4896 | #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) |
duke@435 | 4897 | #endif |
duke@435 | 4898 | |
duke@435 | 4899 | #ifndef __NR_execve |
duke@435 | 4900 | #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) |
duke@435 | 4901 | #endif |
duke@435 | 4902 | |
duke@435 | 4903 | // Run the specified command in a separate process. Return its exit value, |
duke@435 | 4904 | // or -1 on failure (e.g. can't fork a new process). |
duke@435 | 4905 | // Unlike system(), this function can be called from signal handler. It |
duke@435 | 4906 | // doesn't block SIGINT et al. |
duke@435 | 4907 | int os::fork_and_exec(char* cmd) { |
xlu@634 | 4908 | const char * argv[4] = {"sh", "-c", cmd, NULL}; |
duke@435 | 4909 | |
duke@435 | 4910 | // fork() in LinuxThreads/NPTL is not async-safe. It needs to run |
duke@435 | 4911 | // pthread_atfork handlers and reset pthread library. All we need is a |
duke@435 | 4912 | // separate process to execve. Make a direct syscall to fork process. |
duke@435 | 4913 | // On IA64 there's no fork syscall, we have to use fork() and hope for |
duke@435 | 4914 | // the best... |
duke@435 | 4915 | pid_t pid = NOT_IA64(syscall(__NR_fork);) |
duke@435 | 4916 | IA64_ONLY(fork();) |
duke@435 | 4917 | |
duke@435 | 4918 | if (pid < 0) { |
duke@435 | 4919 | // fork failed |
duke@435 | 4920 | return -1; |
duke@435 | 4921 | |
duke@435 | 4922 | } else if (pid == 0) { |
duke@435 | 4923 | // child process |
duke@435 | 4924 | |
duke@435 | 4925 | // execve() in LinuxThreads will call pthread_kill_other_threads_np() |
duke@435 | 4926 | // first to kill every thread on the thread list. Because this list is |
duke@435 | 4927 | // not reset by fork() (see notes above), execve() will instead kill |
duke@435 | 4928 | // every thread in the parent process. We know this is the only thread |
duke@435 | 4929 | // in the new process, so make a system call directly. |
duke@435 | 4930 | // IA64 should use normal execve() from glibc to match the glibc fork() |
duke@435 | 4931 | // above. |
duke@435 | 4932 | NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) |
xlu@634 | 4933 | IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) |
duke@435 | 4934 | |
duke@435 | 4935 | // execve failed |
duke@435 | 4936 | _exit(-1); |
duke@435 | 4937 | |
duke@435 | 4938 | } else { |
duke@435 | 4939 | // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't |
duke@435 | 4940 | // care about the actual exit code, for now. |
duke@435 | 4941 | |
duke@435 | 4942 | int status; |
duke@435 | 4943 | |
duke@435 | 4944 | // Wait for the child process to exit. This returns immediately if |
duke@435 | 4945 | // the child has already exited. */ |
duke@435 | 4946 | while (waitpid(pid, &status, 0) < 0) { |
duke@435 | 4947 | switch (errno) { |
duke@435 | 4948 | case ECHILD: return 0; |
duke@435 | 4949 | case EINTR: break; |
duke@435 | 4950 | default: return -1; |
duke@435 | 4951 | } |
duke@435 | 4952 | } |
duke@435 | 4953 | |
duke@435 | 4954 | if (WIFEXITED(status)) { |
duke@435 | 4955 | // The child exited normally; get its exit code. |
duke@435 | 4956 | return WEXITSTATUS(status); |
duke@435 | 4957 | } else if (WIFSIGNALED(status)) { |
duke@435 | 4958 | // The child exited because of a signal |
duke@435 | 4959 | // The best value to return is 0x80 + signal number, |
duke@435 | 4960 | // because that is what all Unix shells do, and because |
duke@435 | 4961 | // it allows callers to distinguish between process exit and |
duke@435 | 4962 | // process death by signal. |
duke@435 | 4963 | return 0x80 + WTERMSIG(status); |
duke@435 | 4964 | } else { |
duke@435 | 4965 | // Unknown exit code; pass it through |
duke@435 | 4966 | return status; |
duke@435 | 4967 | } |
duke@435 | 4968 | } |
duke@435 | 4969 | } |