Print this page
Split |
Close |
Expand all |
Collapse all |
--- old/src/os/linux/vm/os_linux.cpp
+++ new/src/os/linux/vm/os_linux.cpp
1 1 /*
2 2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
3 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 4 *
5 5 * This code is free software; you can redistribute it and/or modify it
6 6 * under the terms of the GNU General Public License version 2 only, as
7 7 * published by the Free Software Foundation.
8 8 *
9 9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 12 * version 2 for more details (a copy is included in the LICENSE file that
13 13 * accompanied this code).
14 14 *
15 15 * You should have received a copy of the GNU General Public License version
16 16 * 2 along with this work; if not, write to the Free Software Foundation,
17 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 18 *
19 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 20 * or visit www.oracle.com if you need additional information or have any
21 21 * questions.
22 22 *
23 23 */
24 24
25 25 # define __STDC_FORMAT_MACROS
26 26
27 27 // no precompiled headers
28 28 #include "classfile/classLoader.hpp"
29 29 #include "classfile/systemDictionary.hpp"
30 30 #include "classfile/vmSymbols.hpp"
31 31 #include "code/icBuffer.hpp"
32 32 #include "code/vtableStubs.hpp"
33 33 #include "compiler/compileBroker.hpp"
34 34 #include "interpreter/interpreter.hpp"
35 35 #include "jvm_linux.h"
36 36 #include "memory/allocation.inline.hpp"
37 37 #include "memory/filemap.hpp"
38 38 #include "mutex_linux.inline.hpp"
39 39 #include "oops/oop.inline.hpp"
40 40 #include "os_share_linux.hpp"
41 41 #include "prims/jniFastGetField.hpp"
42 42 #include "prims/jvm.h"
43 43 #include "prims/jvm_misc.hpp"
44 44 #include "runtime/arguments.hpp"
45 45 #include "runtime/extendedPC.hpp"
46 46 #include "runtime/globals.hpp"
47 47 #include "runtime/interfaceSupport.hpp"
48 48 #include "runtime/java.hpp"
49 49 #include "runtime/javaCalls.hpp"
50 50 #include "runtime/mutexLocker.hpp"
51 51 #include "runtime/objectMonitor.hpp"
52 52 #include "runtime/osThread.hpp"
53 53 #include "runtime/perfMemory.hpp"
54 54 #include "runtime/sharedRuntime.hpp"
55 55 #include "runtime/statSampler.hpp"
56 56 #include "runtime/stubRoutines.hpp"
57 57 #include "runtime/threadCritical.hpp"
58 58 #include "runtime/timer.hpp"
59 59 #include "services/attachListener.hpp"
60 60 #include "services/runtimeService.hpp"
61 61 #include "thread_linux.inline.hpp"
62 62 #include "utilities/decoder.hpp"
63 63 #include "utilities/defaultStream.hpp"
64 64 #include "utilities/events.hpp"
65 65 #include "utilities/growableArray.hpp"
66 66 #include "utilities/vmError.hpp"
67 67 #ifdef TARGET_ARCH_x86
68 68 # include "assembler_x86.inline.hpp"
69 69 # include "nativeInst_x86.hpp"
70 70 #endif
71 71 #ifdef TARGET_ARCH_sparc
72 72 # include "assembler_sparc.inline.hpp"
73 73 # include "nativeInst_sparc.hpp"
74 74 #endif
75 75 #ifdef TARGET_ARCH_zero
76 76 # include "assembler_zero.inline.hpp"
77 77 # include "nativeInst_zero.hpp"
78 78 #endif
79 79 #ifdef TARGET_ARCH_arm
80 80 # include "assembler_arm.inline.hpp"
81 81 # include "nativeInst_arm.hpp"
82 82 #endif
83 83 #ifdef TARGET_ARCH_ppc
84 84 # include "assembler_ppc.inline.hpp"
85 85 # include "nativeInst_ppc.hpp"
86 86 #endif
87 87 #ifdef COMPILER1
88 88 #include "c1/c1_Runtime1.hpp"
89 89 #endif
90 90 #ifdef COMPILER2
91 91 #include "opto/runtime.hpp"
92 92 #endif
93 93
94 94 // put OS-includes here
95 95 # include <sys/types.h>
96 96 # include <sys/mman.h>
97 97 # include <sys/stat.h>
98 98 # include <sys/select.h>
99 99 # include <pthread.h>
100 100 # include <signal.h>
101 101 # include <errno.h>
102 102 # include <dlfcn.h>
103 103 # include <stdio.h>
104 104 # include <unistd.h>
105 105 # include <sys/resource.h>
106 106 # include <pthread.h>
107 107 # include <sys/stat.h>
108 108 # include <sys/time.h>
109 109 # include <sys/times.h>
110 110 # include <sys/utsname.h>
111 111 # include <sys/socket.h>
112 112 # include <sys/wait.h>
113 113 # include <pwd.h>
114 114 # include <poll.h>
115 115 # include <semaphore.h>
116 116 # include <fcntl.h>
117 117 # include <string.h>
118 118 # include <syscall.h>
119 119 # include <sys/sysinfo.h>
120 120 # include <gnu/libc-version.h>
121 121 # include <sys/ipc.h>
122 122 # include <sys/shm.h>
123 123 # include <link.h>
124 124 # include <stdint.h>
125 125 # include <inttypes.h>
126 126 # include <sys/ioctl.h>
127 127
128 128 #define MAX_PATH (2 * K)
129 129
130 130 // for timer info max values which include all bits
131 131 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
132 132 #define SEC_IN_NANOSECS 1000000000LL
133 133
134 134 #define LARGEPAGES_BIT (1 << 6)
135 135 ////////////////////////////////////////////////////////////////////////////////
136 136 // global variables
137 137 julong os::Linux::_physical_memory = 0;
138 138
139 139 address os::Linux::_initial_thread_stack_bottom = NULL;
140 140 uintptr_t os::Linux::_initial_thread_stack_size = 0;
141 141
142 142 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
143 143 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
144 144 Mutex* os::Linux::_createThread_lock = NULL;
145 145 pthread_t os::Linux::_main_thread;
146 146 int os::Linux::_page_size = -1;
147 147 bool os::Linux::_is_floating_stack = false;
148 148 bool os::Linux::_is_NPTL = false;
149 149 bool os::Linux::_supports_fast_thread_cpu_time = false;
150 150 const char * os::Linux::_glibc_version = NULL;
151 151 const char * os::Linux::_libpthread_version = NULL;
152 152
153 153 static jlong initial_time_count=0;
154 154
155 155 static int clock_tics_per_sec = 100;
156 156
157 157 // For diagnostics to print a message once. see run_periodic_checks
158 158 static sigset_t check_signal_done;
159 159 static bool check_signals = true;;
160 160
161 161 static pid_t _initial_pid = 0;
162 162
163 163 /* Signal number used to suspend/resume a thread */
164 164
165 165 /* do not use any signal number less than SIGSEGV, see 4355769 */
166 166 static int SR_signum = SIGUSR2;
167 167 sigset_t SR_sigset;
168 168
169 169 /* Used to protect dlsym() calls */
170 170 static pthread_mutex_t dl_mutex;
171 171
172 172 #ifdef JAVASE_EMBEDDED
173 173 class MemNotifyThread: public Thread {
174 174 friend class VMStructs;
175 175 public:
176 176 virtual void run();
177 177
178 178 private:
179 179 static MemNotifyThread* _memnotify_thread;
180 180 int _fd;
181 181
182 182 public:
183 183
184 184 // Constructor
185 185 MemNotifyThread(int fd);
186 186
187 187 // Tester
188 188 bool is_memnotify_thread() const { return true; }
189 189
190 190 // Printing
191 191 char* name() const { return (char*)"Linux MemNotify Thread"; }
192 192
193 193 // Returns the single instance of the MemNotifyThread
194 194 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
195 195
196 196 // Create and start the single instance of MemNotifyThread
197 197 static void start();
198 198 };
199 199 #endif // JAVASE_EMBEDDED
200 200
201 201 // utility functions
202 202
203 203 static int SR_initialize();
204 204 static int SR_finalize();
205 205
206 206 julong os::available_memory() {
207 207 return Linux::available_memory();
208 208 }
209 209
210 210 julong os::Linux::available_memory() {
211 211 // values in struct sysinfo are "unsigned long"
212 212 struct sysinfo si;
213 213 sysinfo(&si);
214 214
215 215 return (julong)si.freeram * si.mem_unit;
216 216 }
217 217
218 218 julong os::physical_memory() {
219 219 return Linux::physical_memory();
220 220 }
221 221
222 222 julong os::allocatable_physical_memory(julong size) {
223 223 #ifdef _LP64
224 224 return size;
225 225 #else
226 226 julong result = MIN2(size, (julong)3800*M);
227 227 if (!is_allocatable(result)) {
228 228 // See comments under solaris for alignment considerations
229 229 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
230 230 result = MIN2(size, reasonable_size);
231 231 }
232 232 return result;
233 233 #endif // _LP64
234 234 }
235 235
236 236 ////////////////////////////////////////////////////////////////////////////////
237 237 // environment support
238 238
239 239 bool os::getenv(const char* name, char* buf, int len) {
240 240 const char* val = ::getenv(name);
241 241 if (val != NULL && strlen(val) < (size_t)len) {
242 242 strcpy(buf, val);
243 243 return true;
244 244 }
245 245 if (len > 0) buf[0] = 0; // return a null string
246 246 return false;
247 247 }
248 248
249 249
250 250 // Return true if user is running as root.
251 251
252 252 bool os::have_special_privileges() {
253 253 static bool init = false;
254 254 static bool privileges = false;
255 255 if (!init) {
256 256 privileges = (getuid() != geteuid()) || (getgid() != getegid());
257 257 init = true;
258 258 }
259 259 return privileges;
260 260 }
261 261
262 262
263 263 #ifndef SYS_gettid
264 264 // i386: 224, ia64: 1105, amd64: 186, sparc 143
265 265 #ifdef __ia64__
266 266 #define SYS_gettid 1105
267 267 #elif __i386__
268 268 #define SYS_gettid 224
269 269 #elif __amd64__
270 270 #define SYS_gettid 186
271 271 #elif __sparc__
272 272 #define SYS_gettid 143
273 273 #else
274 274 #error define gettid for the arch
275 275 #endif
276 276 #endif
277 277
278 278 // Cpu architecture string
279 279 #if defined(ZERO)
280 280 static char cpu_arch[] = ZERO_LIBARCH;
281 281 #elif defined(IA64)
282 282 static char cpu_arch[] = "ia64";
283 283 #elif defined(IA32)
284 284 static char cpu_arch[] = "i386";
285 285 #elif defined(AMD64)
286 286 static char cpu_arch[] = "amd64";
287 287 #elif defined(ARM)
288 288 static char cpu_arch[] = "arm";
289 289 #elif defined(PPC)
290 290 static char cpu_arch[] = "ppc";
291 291 #elif defined(SPARC)
292 292 # ifdef _LP64
293 293 static char cpu_arch[] = "sparcv9";
294 294 # else
295 295 static char cpu_arch[] = "sparc";
296 296 # endif
297 297 #else
298 298 #error Add appropriate cpu_arch setting
299 299 #endif
300 300
301 301
302 302 // pid_t gettid()
303 303 //
304 304 // Returns the kernel thread id of the currently running thread. Kernel
305 305 // thread id is used to access /proc.
306 306 //
307 307 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
308 308 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
309 309 //
310 310 pid_t os::Linux::gettid() {
311 311 int rslt = syscall(SYS_gettid);
312 312 if (rslt == -1) {
313 313 // old kernel, no NPTL support
314 314 return getpid();
315 315 } else {
316 316 return (pid_t)rslt;
317 317 }
318 318 }
319 319
320 320 // Most versions of linux have a bug where the number of processors are
321 321 // determined by looking at the /proc file system. In a chroot environment,
322 322 // the system call returns 1. This causes the VM to act as if it is
323 323 // a single processor and elide locking (see is_MP() call).
324 324 static bool unsafe_chroot_detected = false;
325 325 static const char *unstable_chroot_error = "/proc file system not found.\n"
326 326 "Java may be unstable running multithreaded in a chroot "
327 327 "environment on Linux when /proc filesystem is not mounted.";
328 328
329 329 void os::Linux::initialize_system_info() {
330 330 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
331 331 if (processor_count() == 1) {
332 332 pid_t pid = os::Linux::gettid();
333 333 char fname[32];
334 334 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
335 335 FILE *fp = fopen(fname, "r");
336 336 if (fp == NULL) {
337 337 unsafe_chroot_detected = true;
338 338 } else {
339 339 fclose(fp);
340 340 }
341 341 }
342 342 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
343 343 assert(processor_count() > 0, "linux error");
344 344 }
345 345
346 346 void os::init_system_properties_values() {
347 347 // char arch[12];
348 348 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
349 349
350 350 // The next steps are taken in the product version:
351 351 //
352 352 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
353 353 // This library should be located at:
354 354 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
355 355 //
356 356 // If "/jre/lib/" appears at the right place in the path, then we
357 357 // assume libjvm[_g].so is installed in a JDK and we use this path.
358 358 //
359 359 // Otherwise exit with message: "Could not create the Java virtual machine."
360 360 //
361 361 // The following extra steps are taken in the debugging version:
362 362 //
363 363 // If "/jre/lib/" does NOT appear at the right place in the path
364 364 // instead of exit check for $JAVA_HOME environment variable.
365 365 //
366 366 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
367 367 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
368 368 // it looks like libjvm[_g].so is installed there
369 369 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
370 370 //
371 371 // Otherwise exit.
372 372 //
373 373 // Important note: if the location of libjvm.so changes this
374 374 // code needs to be changed accordingly.
375 375
376 376 // The next few definitions allow the code to be verbatim:
377 377 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
378 378 #define getenv(n) ::getenv(n)
379 379
380 380 /*
381 381 * See ld(1):
382 382 * The linker uses the following search paths to locate required
383 383 * shared libraries:
384 384 * 1: ...
385 385 * ...
386 386 * 7: The default directories, normally /lib and /usr/lib.
387 387 */
388 388 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
389 389 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
390 390 #else
391 391 #define DEFAULT_LIBPATH "/lib:/usr/lib"
392 392 #endif
393 393
394 394 #define EXTENSIONS_DIR "/lib/ext"
395 395 #define ENDORSED_DIR "/lib/endorsed"
396 396 #define REG_DIR "/usr/java/packages"
397 397
398 398 {
399 399 /* sysclasspath, java_home, dll_dir */
400 400 {
401 401 char *home_path;
402 402 char *dll_path;
403 403 char *pslash;
404 404 char buf[MAXPATHLEN];
405 405 os::jvm_path(buf, sizeof(buf));
406 406
407 407 // Found the full path to libjvm.so.
408 408 // Now cut the path to <java_home>/jre if we can.
409 409 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
410 410 pslash = strrchr(buf, '/');
411 411 if (pslash != NULL)
412 412 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
413 413 dll_path = malloc(strlen(buf) + 1);
414 414 if (dll_path == NULL)
415 415 return;
416 416 strcpy(dll_path, buf);
417 417 Arguments::set_dll_dir(dll_path);
418 418
419 419 if (pslash != NULL) {
420 420 pslash = strrchr(buf, '/');
421 421 if (pslash != NULL) {
422 422 *pslash = '\0'; /* get rid of /<arch> */
423 423 pslash = strrchr(buf, '/');
424 424 if (pslash != NULL)
425 425 *pslash = '\0'; /* get rid of /lib */
426 426 }
427 427 }
428 428
429 429 home_path = malloc(strlen(buf) + 1);
430 430 if (home_path == NULL)
431 431 return;
432 432 strcpy(home_path, buf);
433 433 Arguments::set_java_home(home_path);
434 434
435 435 if (!set_boot_path('/', ':'))
436 436 return;
437 437 }
438 438
439 439 /*
440 440 * Where to look for native libraries
441 441 *
442 442 * Note: Due to a legacy implementation, most of the library path
443 443 * is set in the launcher. This was to accomodate linking restrictions
444 444 * on legacy Linux implementations (which are no longer supported).
445 445 * Eventually, all the library path setting will be done here.
446 446 *
447 447 * However, to prevent the proliferation of improperly built native
448 448 * libraries, the new path component /usr/java/packages is added here.
449 449 * Eventually, all the library path setting will be done here.
450 450 */
451 451 {
452 452 char *ld_library_path;
453 453
454 454 /*
455 455 * Construct the invariant part of ld_library_path. Note that the
456 456 * space for the colon and the trailing null are provided by the
457 457 * nulls included by the sizeof operator (so actually we allocate
458 458 * a byte more than necessary).
459 459 */
460 460 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
461 461 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
462 462 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
463 463
464 464 /*
465 465 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
466 466 * should always exist (until the legacy problem cited above is
467 467 * addressed).
468 468 */
469 469 char *v = getenv("LD_LIBRARY_PATH");
470 470 if (v != NULL) {
471 471 char *t = ld_library_path;
472 472 /* That's +1 for the colon and +1 for the trailing '\0' */
473 473 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
474 474 sprintf(ld_library_path, "%s:%s", v, t);
475 475 }
476 476 Arguments::set_library_path(ld_library_path);
477 477 }
478 478
479 479 /*
480 480 * Extensions directories.
481 481 *
482 482 * Note that the space for the colon and the trailing null are provided
483 483 * by the nulls included by the sizeof operator (so actually one byte more
484 484 * than necessary is allocated).
485 485 */
486 486 {
487 487 char *buf = malloc(strlen(Arguments::get_java_home()) +
488 488 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
489 489 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
490 490 Arguments::get_java_home());
491 491 Arguments::set_ext_dirs(buf);
492 492 }
493 493
494 494 /* Endorsed standards default directory. */
495 495 {
496 496 char * buf;
497 497 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
498 498 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
499 499 Arguments::set_endorsed_dirs(buf);
500 500 }
501 501 }
502 502
503 503 #undef malloc
504 504 #undef getenv
505 505 #undef EXTENSIONS_DIR
506 506 #undef ENDORSED_DIR
507 507
508 508 // Done
509 509 return;
510 510 }
511 511
512 512 ////////////////////////////////////////////////////////////////////////////////
513 513 // breakpoint support
514 514
515 515 void os::breakpoint() {
516 516 BREAKPOINT;
517 517 }
518 518
519 519 extern "C" void breakpoint() {
520 520 // use debugger to set breakpoint here
521 521 }
522 522
523 523 ////////////////////////////////////////////////////////////////////////////////
524 524 // signal support
525 525
526 526 debug_only(static bool signal_sets_initialized = false);
527 527 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
528 528
529 529 bool os::Linux::is_sig_ignored(int sig) {
530 530 struct sigaction oact;
531 531 sigaction(sig, (struct sigaction*)NULL, &oact);
532 532 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
533 533 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
534 534 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
535 535 return true;
536 536 else
537 537 return false;
538 538 }
539 539
540 540 void os::Linux::signal_sets_init() {
541 541 // Should also have an assertion stating we are still single-threaded.
542 542 assert(!signal_sets_initialized, "Already initialized");
543 543 // Fill in signals that are necessarily unblocked for all threads in
544 544 // the VM. Currently, we unblock the following signals:
545 545 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
546 546 // by -Xrs (=ReduceSignalUsage));
547 547 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
548 548 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
549 549 // the dispositions or masks wrt these signals.
550 550 // Programs embedding the VM that want to use the above signals for their
551 551 // own purposes must, at this time, use the "-Xrs" option to prevent
552 552 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
553 553 // (See bug 4345157, and other related bugs).
554 554 // In reality, though, unblocking these signals is really a nop, since
555 555 // these signals are not blocked by default.
556 556 sigemptyset(&unblocked_sigs);
557 557 sigemptyset(&allowdebug_blocked_sigs);
558 558 sigaddset(&unblocked_sigs, SIGILL);
559 559 sigaddset(&unblocked_sigs, SIGSEGV);
560 560 sigaddset(&unblocked_sigs, SIGBUS);
561 561 sigaddset(&unblocked_sigs, SIGFPE);
562 562 sigaddset(&unblocked_sigs, SR_signum);
563 563
564 564 if (!ReduceSignalUsage) {
565 565 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
566 566 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
567 567 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
568 568 }
569 569 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
570 570 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
571 571 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
572 572 }
573 573 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
574 574 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
575 575 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
576 576 }
577 577 }
578 578 // Fill in signals that are blocked by all but the VM thread.
579 579 sigemptyset(&vm_sigs);
580 580 if (!ReduceSignalUsage)
581 581 sigaddset(&vm_sigs, BREAK_SIGNAL);
582 582 debug_only(signal_sets_initialized = true);
583 583
584 584 }
585 585
586 586 // These are signals that are unblocked while a thread is running Java.
587 587 // (For some reason, they get blocked by default.)
588 588 sigset_t* os::Linux::unblocked_signals() {
589 589 assert(signal_sets_initialized, "Not initialized");
590 590 return &unblocked_sigs;
591 591 }
592 592
593 593 // These are the signals that are blocked while a (non-VM) thread is
594 594 // running Java. Only the VM thread handles these signals.
595 595 sigset_t* os::Linux::vm_signals() {
596 596 assert(signal_sets_initialized, "Not initialized");
597 597 return &vm_sigs;
598 598 }
599 599
600 600 // These are signals that are blocked during cond_wait to allow debugger in
601 601 sigset_t* os::Linux::allowdebug_blocked_signals() {
602 602 assert(signal_sets_initialized, "Not initialized");
603 603 return &allowdebug_blocked_sigs;
604 604 }
605 605
606 606 void os::Linux::hotspot_sigmask(Thread* thread) {
607 607
608 608 //Save caller's signal mask before setting VM signal mask
609 609 sigset_t caller_sigmask;
610 610 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
611 611
612 612 OSThread* osthread = thread->osthread();
613 613 osthread->set_caller_sigmask(caller_sigmask);
614 614
615 615 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
616 616
617 617 if (!ReduceSignalUsage) {
618 618 if (thread->is_VM_thread()) {
619 619 // Only the VM thread handles BREAK_SIGNAL ...
620 620 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
621 621 } else {
622 622 // ... all other threads block BREAK_SIGNAL
623 623 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
624 624 }
625 625 }
626 626 }
627 627
628 628 //////////////////////////////////////////////////////////////////////////////
629 629 // detecting pthread library
630 630
631 631 void os::Linux::libpthread_init() {
632 632 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
633 633 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
634 634 // generic name for earlier versions.
635 635 // Define macros here so we can build HotSpot on old systems.
636 636 # ifndef _CS_GNU_LIBC_VERSION
637 637 # define _CS_GNU_LIBC_VERSION 2
638 638 # endif
639 639 # ifndef _CS_GNU_LIBPTHREAD_VERSION
640 640 # define _CS_GNU_LIBPTHREAD_VERSION 3
641 641 # endif
642 642
643 643 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
644 644 if (n > 0) {
645 645 char *str = (char *)malloc(n);
646 646 confstr(_CS_GNU_LIBC_VERSION, str, n);
647 647 os::Linux::set_glibc_version(str);
648 648 } else {
649 649 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
650 650 static char _gnu_libc_version[32];
651 651 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
652 652 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
653 653 os::Linux::set_glibc_version(_gnu_libc_version);
654 654 }
655 655
656 656 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
657 657 if (n > 0) {
658 658 char *str = (char *)malloc(n);
659 659 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
660 660 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
661 661 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
662 662 // is the case. LinuxThreads has a hard limit on max number of threads.
663 663 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
664 664 // On the other hand, NPTL does not have such a limit, sysconf()
665 665 // will return -1 and errno is not changed. Check if it is really NPTL.
666 666 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
667 667 strstr(str, "NPTL") &&
668 668 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
669 669 free(str);
670 670 os::Linux::set_libpthread_version("linuxthreads");
671 671 } else {
672 672 os::Linux::set_libpthread_version(str);
673 673 }
674 674 } else {
675 675 // glibc before 2.3.2 only has LinuxThreads.
676 676 os::Linux::set_libpthread_version("linuxthreads");
677 677 }
678 678
679 679 if (strstr(libpthread_version(), "NPTL")) {
680 680 os::Linux::set_is_NPTL();
681 681 } else {
682 682 os::Linux::set_is_LinuxThreads();
683 683 }
684 684
685 685 // LinuxThreads have two flavors: floating-stack mode, which allows variable
686 686 // stack size; and fixed-stack mode. NPTL is always floating-stack.
687 687 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
688 688 os::Linux::set_is_floating_stack();
689 689 }
690 690 }
691 691
692 692 /////////////////////////////////////////////////////////////////////////////
693 693 // thread stack
694 694
695 695 // Force Linux kernel to expand current thread stack. If "bottom" is close
696 696 // to the stack guard, caller should block all signals.
697 697 //
698 698 // MAP_GROWSDOWN:
699 699 // A special mmap() flag that is used to implement thread stacks. It tells
700 700 // kernel that the memory region should extend downwards when needed. This
701 701 // allows early versions of LinuxThreads to only mmap the first few pages
702 702 // when creating a new thread. Linux kernel will automatically expand thread
703 703 // stack as needed (on page faults).
704 704 //
705 705 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
706 706 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
707 707 // region, it's hard to tell if the fault is due to a legitimate stack
708 708 // access or because of reading/writing non-exist memory (e.g. buffer
709 709 // overrun). As a rule, if the fault happens below current stack pointer,
710 710 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
711 711 // application (see Linux kernel fault.c).
712 712 //
713 713 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
714 714 // stack overflow detection.
715 715 //
716 716 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
717 717 // not use this flag. However, the stack of initial thread is not created
718 718 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
719 719 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
720 720 // and then attach the thread to JVM.
721 721 //
722 722 // To get around the problem and allow stack banging on Linux, we need to
723 723 // manually expand thread stack after receiving the SIGSEGV.
724 724 //
725 725 // There are two ways to expand thread stack to address "bottom", we used
726 726 // both of them in JVM before 1.5:
727 727 // 1. adjust stack pointer first so that it is below "bottom", and then
728 728 // touch "bottom"
729 729 // 2. mmap() the page in question
730 730 //
731 731 // Now alternate signal stack is gone, it's harder to use 2. For instance,
732 732 // if current sp is already near the lower end of page 101, and we need to
733 733 // call mmap() to map page 100, it is possible that part of the mmap() frame
734 734 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
735 735 // That will destroy the mmap() frame and cause VM to crash.
736 736 //
737 737 // The following code works by adjusting sp first, then accessing the "bottom"
738 738 // page to force a page fault. Linux kernel will then automatically expand the
739 739 // stack mapping.
740 740 //
741 741 // _expand_stack_to() assumes its frame size is less than page size, which
742 742 // should always be true if the function is not inlined.
743 743
744 744 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
745 745 #define NOINLINE
746 746 #else
747 747 #define NOINLINE __attribute__ ((noinline))
748 748 #endif
749 749
750 750 static void _expand_stack_to(address bottom) NOINLINE;
751 751
752 752 static void _expand_stack_to(address bottom) {
753 753 address sp;
754 754 size_t size;
755 755 volatile char *p;
756 756
757 757 // Adjust bottom to point to the largest address within the same page, it
758 758 // gives us a one-page buffer if alloca() allocates slightly more memory.
759 759 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
760 760 bottom += os::Linux::page_size() - 1;
761 761
762 762 // sp might be slightly above current stack pointer; if that's the case, we
763 763 // will alloca() a little more space than necessary, which is OK. Don't use
764 764 // os::current_stack_pointer(), as its result can be slightly below current
765 765 // stack pointer, causing us to not alloca enough to reach "bottom".
766 766 sp = (address)&sp;
767 767
768 768 if (sp > bottom) {
769 769 size = sp - bottom;
770 770 p = (volatile char *)alloca(size);
771 771 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
772 772 p[0] = '\0';
773 773 }
774 774 }
775 775
776 776 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
777 777 assert(t!=NULL, "just checking");
778 778 assert(t->osthread()->expanding_stack(), "expand should be set");
779 779 assert(t->stack_base() != NULL, "stack_base was not initialized");
780 780
781 781 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
782 782 sigset_t mask_all, old_sigset;
783 783 sigfillset(&mask_all);
784 784 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
785 785 _expand_stack_to(addr);
786 786 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
787 787 return true;
788 788 }
789 789 return false;
790 790 }
791 791
792 792 //////////////////////////////////////////////////////////////////////////////
793 793 // create new thread
794 794
795 795 static address highest_vm_reserved_address();
796 796
797 797 // check if it's safe to start a new thread
798 798 static bool _thread_safety_check(Thread* thread) {
799 799 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
800 800 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
801 801 // Heap is mmap'ed at lower end of memory space. Thread stacks are
802 802 // allocated (MAP_FIXED) from high address space. Every thread stack
803 803 // occupies a fixed size slot (usually 2Mbytes, but user can change
804 804 // it to other values if they rebuild LinuxThreads).
805 805 //
806 806 // Problem with MAP_FIXED is that mmap() can still succeed even part of
807 807 // the memory region has already been mmap'ed. That means if we have too
808 808 // many threads and/or very large heap, eventually thread stack will
809 809 // collide with heap.
810 810 //
811 811 // Here we try to prevent heap/stack collision by comparing current
812 812 // stack bottom with the highest address that has been mmap'ed by JVM
813 813 // plus a safety margin for memory maps created by native code.
814 814 //
815 815 // This feature can be disabled by setting ThreadSafetyMargin to 0
816 816 //
817 817 if (ThreadSafetyMargin > 0) {
818 818 address stack_bottom = os::current_stack_base() - os::current_stack_size();
819 819
820 820 // not safe if our stack extends below the safety margin
821 821 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
822 822 } else {
823 823 return true;
824 824 }
825 825 } else {
826 826 // Floating stack LinuxThreads or NPTL:
827 827 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
828 828 // there's not enough space left, pthread_create() will fail. If we come
829 829 // here, that means enough space has been reserved for stack.
830 830 return true;
831 831 }
832 832 }
833 833
834 834 // Thread start routine for all newly created threads
835 835 static void *java_start(Thread *thread) {
836 836 // Try to randomize the cache line index of hot stack frames.
837 837 // This helps when threads of the same stack traces evict each other's
838 838 // cache lines. The threads can be either from the same JVM instance, or
839 839 // from different JVM instances. The benefit is especially true for
840 840 // processors with hyperthreading technology.
841 841 static int counter = 0;
842 842 int pid = os::current_process_id();
843 843 alloca(((pid ^ counter++) & 7) * 128);
844 844
845 845 ThreadLocalStorage::set_thread(thread);
846 846
847 847 OSThread* osthread = thread->osthread();
848 848 Monitor* sync = osthread->startThread_lock();
849 849
850 850 // non floating stack LinuxThreads needs extra check, see above
851 851 if (!_thread_safety_check(thread)) {
852 852 // notify parent thread
853 853 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
854 854 osthread->set_state(ZOMBIE);
855 855 sync->notify_all();
856 856 return NULL;
857 857 }
858 858
859 859 // thread_id is kernel thread id (similar to Solaris LWP id)
860 860 osthread->set_thread_id(os::Linux::gettid());
861 861
862 862 if (UseNUMA) {
863 863 int lgrp_id = os::numa_get_group_id();
864 864 if (lgrp_id != -1) {
865 865 thread->set_lgrp_id(lgrp_id);
866 866 }
867 867 }
868 868 // initialize signal mask for this thread
869 869 os::Linux::hotspot_sigmask(thread);
870 870
871 871 // initialize floating point control register
872 872 os::Linux::init_thread_fpu_state();
873 873
874 874 // handshaking with parent thread
875 875 {
876 876 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
877 877
878 878 // notify parent thread
879 879 osthread->set_state(INITIALIZED);
880 880 sync->notify_all();
881 881
882 882 // wait until os::start_thread()
883 883 while (osthread->get_state() == INITIALIZED) {
884 884 sync->wait(Mutex::_no_safepoint_check_flag);
885 885 }
886 886 }
887 887
888 888 // call one more level start routine
889 889 thread->run();
890 890
891 891 return 0;
892 892 }
893 893
894 894 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
895 895 assert(thread->osthread() == NULL, "caller responsible");
896 896
897 897 // Allocate the OSThread object
898 898 OSThread* osthread = new OSThread(NULL, NULL);
899 899 if (osthread == NULL) {
900 900 return false;
901 901 }
902 902
903 903 // set the correct thread state
904 904 osthread->set_thread_type(thr_type);
905 905
906 906 // Initial state is ALLOCATED but not INITIALIZED
907 907 osthread->set_state(ALLOCATED);
908 908
909 909 thread->set_osthread(osthread);
910 910
911 911 // init thread attributes
912 912 pthread_attr_t attr;
913 913 pthread_attr_init(&attr);
914 914 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
915 915
916 916 // stack size
917 917 if (os::Linux::supports_variable_stack_size()) {
918 918 // calculate stack size if it's not specified by caller
919 919 if (stack_size == 0) {
920 920 stack_size = os::Linux::default_stack_size(thr_type);
921 921
922 922 switch (thr_type) {
923 923 case os::java_thread:
924 924 // Java threads use ThreadStackSize which default value can be
925 925 // changed with the flag -Xss
926 926 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
927 927 stack_size = JavaThread::stack_size_at_create();
928 928 break;
929 929 case os::compiler_thread:
930 930 if (CompilerThreadStackSize > 0) {
931 931 stack_size = (size_t)(CompilerThreadStackSize * K);
932 932 break;
933 933 } // else fall through:
934 934 // use VMThreadStackSize if CompilerThreadStackSize is not defined
935 935 case os::vm_thread:
936 936 case os::pgc_thread:
937 937 case os::cgc_thread:
938 938 case os::watcher_thread:
939 939 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
940 940 break;
941 941 }
942 942 }
943 943
944 944 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
945 945 pthread_attr_setstacksize(&attr, stack_size);
946 946 } else {
947 947 // let pthread_create() pick the default value.
948 948 }
949 949
950 950 // glibc guard page
951 951 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
952 952
953 953 ThreadState state;
954 954
955 955 {
956 956 // Serialize thread creation if we are running with fixed stack LinuxThreads
957 957 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
958 958 if (lock) {
959 959 os::Linux::createThread_lock()->lock_without_safepoint_check();
960 960 }
961 961
962 962 pthread_t tid;
963 963 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
964 964
965 965 pthread_attr_destroy(&attr);
966 966
967 967 if (ret != 0) {
968 968 if (PrintMiscellaneous && (Verbose || WizardMode)) {
969 969 perror("pthread_create()");
970 970 }
971 971 // Need to clean up stuff we've allocated so far
972 972 thread->set_osthread(NULL);
973 973 delete osthread;
974 974 if (lock) os::Linux::createThread_lock()->unlock();
975 975 return false;
976 976 }
977 977
978 978 // Store pthread info into the OSThread
979 979 osthread->set_pthread_id(tid);
980 980
981 981 // Wait until child thread is either initialized or aborted
982 982 {
983 983 Monitor* sync_with_child = osthread->startThread_lock();
984 984 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
985 985 while ((state = osthread->get_state()) == ALLOCATED) {
986 986 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
987 987 }
988 988 }
989 989
990 990 if (lock) {
991 991 os::Linux::createThread_lock()->unlock();
992 992 }
993 993 }
994 994
995 995 // Aborted due to thread limit being reached
996 996 if (state == ZOMBIE) {
997 997 thread->set_osthread(NULL);
998 998 delete osthread;
999 999 return false;
1000 1000 }
1001 1001
1002 1002 // The thread is returned suspended (in state INITIALIZED),
1003 1003 // and is started higher up in the call chain
1004 1004 assert(state == INITIALIZED, "race condition");
1005 1005 return true;
1006 1006 }
1007 1007
1008 1008 /////////////////////////////////////////////////////////////////////////////
1009 1009 // attach existing thread
1010 1010
1011 1011 // bootstrap the main thread
1012 1012 bool os::create_main_thread(JavaThread* thread) {
1013 1013 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1014 1014 return create_attached_thread(thread);
1015 1015 }
1016 1016
1017 1017 bool os::create_attached_thread(JavaThread* thread) {
1018 1018 #ifdef ASSERT
1019 1019 thread->verify_not_published();
1020 1020 #endif
1021 1021
1022 1022 // Allocate the OSThread object
1023 1023 OSThread* osthread = new OSThread(NULL, NULL);
1024 1024
1025 1025 if (osthread == NULL) {
1026 1026 return false;
1027 1027 }
1028 1028
1029 1029 // Store pthread info into the OSThread
1030 1030 osthread->set_thread_id(os::Linux::gettid());
1031 1031 osthread->set_pthread_id(::pthread_self());
1032 1032
1033 1033 // initialize floating point control register
1034 1034 os::Linux::init_thread_fpu_state();
1035 1035
1036 1036 // Initial thread state is RUNNABLE
1037 1037 osthread->set_state(RUNNABLE);
1038 1038
1039 1039 thread->set_osthread(osthread);
1040 1040
1041 1041 if (UseNUMA) {
1042 1042 int lgrp_id = os::numa_get_group_id();
1043 1043 if (lgrp_id != -1) {
1044 1044 thread->set_lgrp_id(lgrp_id);
1045 1045 }
1046 1046 }
1047 1047
1048 1048 if (os::Linux::is_initial_thread()) {
1049 1049 // If current thread is initial thread, its stack is mapped on demand,
1050 1050 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1051 1051 // the entire stack region to avoid SEGV in stack banging.
1052 1052 // It is also useful to get around the heap-stack-gap problem on SuSE
1053 1053 // kernel (see 4821821 for details). We first expand stack to the top
1054 1054 // of yellow zone, then enable stack yellow zone (order is significant,
1055 1055 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1056 1056 // is no gap between the last two virtual memory regions.
1057 1057
1058 1058 JavaThread *jt = (JavaThread *)thread;
1059 1059 address addr = jt->stack_yellow_zone_base();
1060 1060 assert(addr != NULL, "initialization problem?");
1061 1061 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1062 1062
1063 1063 osthread->set_expanding_stack();
1064 1064 os::Linux::manually_expand_stack(jt, addr);
1065 1065 osthread->clear_expanding_stack();
1066 1066 }
1067 1067
1068 1068 // initialize signal mask for this thread
1069 1069 // and save the caller's signal mask
1070 1070 os::Linux::hotspot_sigmask(thread);
1071 1071
1072 1072 return true;
1073 1073 }
1074 1074
1075 1075 void os::pd_start_thread(Thread* thread) {
1076 1076 OSThread * osthread = thread->osthread();
1077 1077 assert(osthread->get_state() != INITIALIZED, "just checking");
1078 1078 Monitor* sync_with_child = osthread->startThread_lock();
1079 1079 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1080 1080 sync_with_child->notify();
1081 1081 }
1082 1082
1083 1083 // Free Linux resources related to the OSThread
1084 1084 void os::free_thread(OSThread* osthread) {
1085 1085 assert(osthread != NULL, "osthread not set");
1086 1086
1087 1087 if (Thread::current()->osthread() == osthread) {
1088 1088 // Restore caller's signal mask
1089 1089 sigset_t sigmask = osthread->caller_sigmask();
1090 1090 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1091 1091 }
1092 1092
1093 1093 delete osthread;
1094 1094 }
1095 1095
1096 1096 //////////////////////////////////////////////////////////////////////////////
1097 1097 // thread local storage
1098 1098
1099 1099 int os::allocate_thread_local_storage() {
1100 1100 pthread_key_t key;
1101 1101 int rslt = pthread_key_create(&key, NULL);
1102 1102 assert(rslt == 0, "cannot allocate thread local storage");
1103 1103 return (int)key;
1104 1104 }
1105 1105
1106 1106 // Note: This is currently not used by VM, as we don't destroy TLS key
1107 1107 // on VM exit.
1108 1108 void os::free_thread_local_storage(int index) {
1109 1109 int rslt = pthread_key_delete((pthread_key_t)index);
1110 1110 assert(rslt == 0, "invalid index");
1111 1111 }
1112 1112
1113 1113 void os::thread_local_storage_at_put(int index, void* value) {
1114 1114 int rslt = pthread_setspecific((pthread_key_t)index, value);
1115 1115 assert(rslt == 0, "pthread_setspecific failed");
1116 1116 }
1117 1117
1118 1118 extern "C" Thread* get_thread() {
1119 1119 return ThreadLocalStorage::thread();
1120 1120 }
1121 1121
1122 1122 //////////////////////////////////////////////////////////////////////////////
1123 1123 // initial thread
1124 1124
1125 1125 // Check if current thread is the initial thread, similar to Solaris thr_main.
1126 1126 bool os::Linux::is_initial_thread(void) {
1127 1127 char dummy;
1128 1128 // If called before init complete, thread stack bottom will be null.
1129 1129 // Can be called if fatal error occurs before initialization.
1130 1130 if (initial_thread_stack_bottom() == NULL) return false;
1131 1131 assert(initial_thread_stack_bottom() != NULL &&
1132 1132 initial_thread_stack_size() != 0,
1133 1133 "os::init did not locate initial thread's stack region");
1134 1134 if ((address)&dummy >= initial_thread_stack_bottom() &&
1135 1135 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1136 1136 return true;
1137 1137 else return false;
1138 1138 }
1139 1139
1140 1140 // Find the virtual memory area that contains addr
1141 1141 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1142 1142 FILE *fp = fopen("/proc/self/maps", "r");
1143 1143 if (fp) {
1144 1144 address low, high;
1145 1145 while (!feof(fp)) {
1146 1146 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1147 1147 if (low <= addr && addr < high) {
1148 1148 if (vma_low) *vma_low = low;
1149 1149 if (vma_high) *vma_high = high;
1150 1150 fclose (fp);
1151 1151 return true;
1152 1152 }
1153 1153 }
1154 1154 for (;;) {
1155 1155 int ch = fgetc(fp);
1156 1156 if (ch == EOF || ch == (int)'\n') break;
1157 1157 }
1158 1158 }
1159 1159 fclose(fp);
1160 1160 }
1161 1161 return false;
1162 1162 }
1163 1163
1164 1164 // Locate initial thread stack. This special handling of initial thread stack
1165 1165 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1166 1166 // bogus value for initial thread.
1167 1167 void os::Linux::capture_initial_stack(size_t max_size) {
1168 1168 // stack size is the easy part, get it from RLIMIT_STACK
1169 1169 size_t stack_size;
1170 1170 struct rlimit rlim;
1171 1171 getrlimit(RLIMIT_STACK, &rlim);
1172 1172 stack_size = rlim.rlim_cur;
1173 1173
1174 1174 // 6308388: a bug in ld.so will relocate its own .data section to the
1175 1175 // lower end of primordial stack; reduce ulimit -s value a little bit
1176 1176 // so we won't install guard page on ld.so's data section.
1177 1177 stack_size -= 2 * page_size();
1178 1178
1179 1179 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1180 1180 // 7.1, in both cases we will get 2G in return value.
1181 1181 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1182 1182 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1183 1183 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1184 1184 // in case other parts in glibc still assumes 2M max stack size.
1185 1185 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1186 1186 #ifndef IA64
1187 1187 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1188 1188 #else
1189 1189 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1190 1190 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1191 1191 #endif
1192 1192
1193 1193 // Try to figure out where the stack base (top) is. This is harder.
1194 1194 //
1195 1195 // When an application is started, glibc saves the initial stack pointer in
1196 1196 // a global variable "__libc_stack_end", which is then used by system
1197 1197 // libraries. __libc_stack_end should be pretty close to stack top. The
1198 1198 // variable is available since the very early days. However, because it is
1199 1199 // a private interface, it could disappear in the future.
1200 1200 //
1201 1201 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1202 1202 // to __libc_stack_end, it is very close to stack top, but isn't the real
1203 1203 // stack top. Note that /proc may not exist if VM is running as a chroot
1204 1204 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1205 1205 // /proc/<pid>/stat could change in the future (though unlikely).
1206 1206 //
1207 1207 // We try __libc_stack_end first. If that doesn't work, look for
1208 1208 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1209 1209 // as a hint, which should work well in most cases.
1210 1210
1211 1211 uintptr_t stack_start;
1212 1212
1213 1213 // try __libc_stack_end first
1214 1214 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1215 1215 if (p && *p) {
1216 1216 stack_start = *p;
1217 1217 } else {
1218 1218 // see if we can get the start_stack field from /proc/self/stat
1219 1219 FILE *fp;
1220 1220 int pid;
1221 1221 char state;
1222 1222 int ppid;
1223 1223 int pgrp;
1224 1224 int session;
1225 1225 int nr;
1226 1226 int tpgrp;
1227 1227 unsigned long flags;
1228 1228 unsigned long minflt;
1229 1229 unsigned long cminflt;
1230 1230 unsigned long majflt;
1231 1231 unsigned long cmajflt;
1232 1232 unsigned long utime;
1233 1233 unsigned long stime;
1234 1234 long cutime;
1235 1235 long cstime;
1236 1236 long prio;
1237 1237 long nice;
1238 1238 long junk;
1239 1239 long it_real;
1240 1240 uintptr_t start;
1241 1241 uintptr_t vsize;
1242 1242 intptr_t rss;
1243 1243 uintptr_t rsslim;
1244 1244 uintptr_t scodes;
1245 1245 uintptr_t ecode;
1246 1246 int i;
1247 1247
1248 1248 // Figure what the primordial thread stack base is. Code is inspired
1249 1249 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1250 1250 // followed by command name surrounded by parentheses, state, etc.
1251 1251 char stat[2048];
1252 1252 int statlen;
1253 1253
1254 1254 fp = fopen("/proc/self/stat", "r");
1255 1255 if (fp) {
1256 1256 statlen = fread(stat, 1, 2047, fp);
1257 1257 stat[statlen] = '\0';
1258 1258 fclose(fp);
1259 1259
1260 1260 // Skip pid and the command string. Note that we could be dealing with
1261 1261 // weird command names, e.g. user could decide to rename java launcher
1262 1262 // to "java 1.4.2 :)", then the stat file would look like
1263 1263 // 1234 (java 1.4.2 :)) R ... ...
1264 1264 // We don't really need to know the command string, just find the last
1265 1265 // occurrence of ")" and then start parsing from there. See bug 4726580.
1266 1266 char * s = strrchr(stat, ')');
1267 1267
1268 1268 i = 0;
1269 1269 if (s) {
1270 1270 // Skip blank chars
1271 1271 do s++; while (isspace(*s));
1272 1272
1273 1273 #define _UFM UINTX_FORMAT
1274 1274 #define _DFM INTX_FORMAT
1275 1275
1276 1276 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1277 1277 /* 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 */
1278 1278 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1279 1279 &state, /* 3 %c */
1280 1280 &ppid, /* 4 %d */
1281 1281 &pgrp, /* 5 %d */
1282 1282 &session, /* 6 %d */
1283 1283 &nr, /* 7 %d */
1284 1284 &tpgrp, /* 8 %d */
1285 1285 &flags, /* 9 %lu */
1286 1286 &minflt, /* 10 %lu */
1287 1287 &cminflt, /* 11 %lu */
1288 1288 &majflt, /* 12 %lu */
1289 1289 &cmajflt, /* 13 %lu */
1290 1290 &utime, /* 14 %lu */
1291 1291 &stime, /* 15 %lu */
1292 1292 &cutime, /* 16 %ld */
1293 1293 &cstime, /* 17 %ld */
1294 1294 &prio, /* 18 %ld */
1295 1295 &nice, /* 19 %ld */
1296 1296 &junk, /* 20 %ld */
1297 1297 &it_real, /* 21 %ld */
1298 1298 &start, /* 22 UINTX_FORMAT */
1299 1299 &vsize, /* 23 UINTX_FORMAT */
1300 1300 &rss, /* 24 INTX_FORMAT */
1301 1301 &rsslim, /* 25 UINTX_FORMAT */
1302 1302 &scodes, /* 26 UINTX_FORMAT */
1303 1303 &ecode, /* 27 UINTX_FORMAT */
1304 1304 &stack_start); /* 28 UINTX_FORMAT */
1305 1305 }
1306 1306
1307 1307 #undef _UFM
1308 1308 #undef _DFM
1309 1309
1310 1310 if (i != 28 - 2) {
1311 1311 assert(false, "Bad conversion from /proc/self/stat");
1312 1312 // product mode - assume we are the initial thread, good luck in the
1313 1313 // embedded case.
1314 1314 warning("Can't detect initial thread stack location - bad conversion");
1315 1315 stack_start = (uintptr_t) &rlim;
1316 1316 }
1317 1317 } else {
1318 1318 // For some reason we can't open /proc/self/stat (for example, running on
1319 1319 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1320 1320 // most cases, so don't abort:
1321 1321 warning("Can't detect initial thread stack location - no /proc/self/stat");
1322 1322 stack_start = (uintptr_t) &rlim;
1323 1323 }
1324 1324 }
1325 1325
1326 1326 // Now we have a pointer (stack_start) very close to the stack top, the
1327 1327 // next thing to do is to figure out the exact location of stack top. We
1328 1328 // can find out the virtual memory area that contains stack_start by
1329 1329 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1330 1330 // and its upper limit is the real stack top. (again, this would fail if
1331 1331 // running inside chroot, because /proc may not exist.)
1332 1332
1333 1333 uintptr_t stack_top;
1334 1334 address low, high;
1335 1335 if (find_vma((address)stack_start, &low, &high)) {
1336 1336 // success, "high" is the true stack top. (ignore "low", because initial
1337 1337 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1338 1338 stack_top = (uintptr_t)high;
1339 1339 } else {
1340 1340 // failed, likely because /proc/self/maps does not exist
1341 1341 warning("Can't detect initial thread stack location - find_vma failed");
1342 1342 // best effort: stack_start is normally within a few pages below the real
1343 1343 // stack top, use it as stack top, and reduce stack size so we won't put
1344 1344 // guard page outside stack.
1345 1345 stack_top = stack_start;
1346 1346 stack_size -= 16 * page_size();
1347 1347 }
1348 1348
1349 1349 // stack_top could be partially down the page so align it
1350 1350 stack_top = align_size_up(stack_top, page_size());
1351 1351
1352 1352 if (max_size && stack_size > max_size) {
1353 1353 _initial_thread_stack_size = max_size;
1354 1354 } else {
1355 1355 _initial_thread_stack_size = stack_size;
1356 1356 }
1357 1357
1358 1358 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1359 1359 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1360 1360 }
1361 1361
1362 1362 ////////////////////////////////////////////////////////////////////////////////
1363 1363 // time support
1364 1364
1365 1365 // Time since start-up in seconds to a fine granularity.
1366 1366 // Used by VMSelfDestructTimer and the MemProfiler.
1367 1367 double os::elapsedTime() {
1368 1368
1369 1369 return (double)(os::elapsed_counter()) * 0.000001;
1370 1370 }
1371 1371
1372 1372 jlong os::elapsed_counter() {
1373 1373 timeval time;
1374 1374 int status = gettimeofday(&time, NULL);
1375 1375 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1376 1376 }
1377 1377
1378 1378 jlong os::elapsed_frequency() {
1379 1379 return (1000 * 1000);
1380 1380 }
1381 1381
1382 1382 // For now, we say that linux does not support vtime. I have no idea
1383 1383 // whether it can actually be made to (DLD, 9/13/05).
1384 1384
1385 1385 bool os::supports_vtime() { return false; }
1386 1386 bool os::enable_vtime() { return false; }
1387 1387 bool os::vtime_enabled() { return false; }
1388 1388 double os::elapsedVTime() {
1389 1389 // better than nothing, but not much
1390 1390 return elapsedTime();
1391 1391 }
1392 1392
1393 1393 jlong os::javaTimeMillis() {
1394 1394 timeval time;
1395 1395 int status = gettimeofday(&time, NULL);
1396 1396 assert(status != -1, "linux error");
1397 1397 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1398 1398 }
1399 1399
1400 1400 #ifndef CLOCK_MONOTONIC
1401 1401 #define CLOCK_MONOTONIC (1)
1402 1402 #endif
1403 1403
1404 1404 void os::Linux::clock_init() {
1405 1405 // we do dlopen's in this particular order due to bug in linux
1406 1406 // dynamical loader (see 6348968) leading to crash on exit
1407 1407 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1408 1408 if (handle == NULL) {
1409 1409 handle = dlopen("librt.so", RTLD_LAZY);
1410 1410 }
1411 1411
1412 1412 if (handle) {
1413 1413 int (*clock_getres_func)(clockid_t, struct timespec*) =
1414 1414 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1415 1415 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1416 1416 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1417 1417 if (clock_getres_func && clock_gettime_func) {
1418 1418 // See if monotonic clock is supported by the kernel. Note that some
1419 1419 // early implementations simply return kernel jiffies (updated every
1420 1420 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1421 1421 // for nano time (though the monotonic property is still nice to have).
1422 1422 // It's fixed in newer kernels, however clock_getres() still returns
1423 1423 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1424 1424 // resolution for now. Hopefully as people move to new kernels, this
1425 1425 // won't be a problem.
1426 1426 struct timespec res;
1427 1427 struct timespec tp;
1428 1428 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1429 1429 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1430 1430 // yes, monotonic clock is supported
1431 1431 _clock_gettime = clock_gettime_func;
1432 1432 } else {
1433 1433 // close librt if there is no monotonic clock
1434 1434 dlclose(handle);
1435 1435 }
1436 1436 }
1437 1437 }
1438 1438 }
1439 1439
1440 1440 #ifndef SYS_clock_getres
1441 1441
1442 1442 #if defined(IA32) || defined(AMD64)
1443 1443 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1444 1444 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1445 1445 #else
1446 1446 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1447 1447 #define sys_clock_getres(x,y) -1
1448 1448 #endif
1449 1449
1450 1450 #else
1451 1451 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1452 1452 #endif
1453 1453
1454 1454 void os::Linux::fast_thread_clock_init() {
1455 1455 if (!UseLinuxPosixThreadCPUClocks) {
1456 1456 return;
1457 1457 }
1458 1458 clockid_t clockid;
1459 1459 struct timespec tp;
1460 1460 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1461 1461 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1462 1462
1463 1463 // Switch to using fast clocks for thread cpu time if
1464 1464 // the sys_clock_getres() returns 0 error code.
1465 1465 // Note, that some kernels may support the current thread
1466 1466 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1467 1467 // returned by the pthread_getcpuclockid().
1468 1468 // If the fast Posix clocks are supported then the sys_clock_getres()
1469 1469 // must return at least tp.tv_sec == 0 which means a resolution
1470 1470 // better than 1 sec. This is extra check for reliability.
1471 1471
1472 1472 if(pthread_getcpuclockid_func &&
1473 1473 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1474 1474 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1475 1475
1476 1476 _supports_fast_thread_cpu_time = true;
1477 1477 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1478 1478 }
1479 1479 }
1480 1480
1481 1481 jlong os::javaTimeNanos() {
1482 1482 if (Linux::supports_monotonic_clock()) {
1483 1483 struct timespec tp;
1484 1484 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1485 1485 assert(status == 0, "gettime error");
1486 1486 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1487 1487 return result;
1488 1488 } else {
1489 1489 timeval time;
1490 1490 int status = gettimeofday(&time, NULL);
1491 1491 assert(status != -1, "linux error");
1492 1492 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1493 1493 return 1000 * usecs;
1494 1494 }
1495 1495 }
1496 1496
1497 1497 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1498 1498 if (Linux::supports_monotonic_clock()) {
1499 1499 info_ptr->max_value = ALL_64_BITS;
1500 1500
1501 1501 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1502 1502 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1503 1503 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1504 1504 } else {
1505 1505 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1506 1506 info_ptr->max_value = ALL_64_BITS;
1507 1507
1508 1508 // gettimeofday is a real time clock so it skips
1509 1509 info_ptr->may_skip_backward = true;
1510 1510 info_ptr->may_skip_forward = true;
1511 1511 }
1512 1512
1513 1513 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1514 1514 }
1515 1515
1516 1516 // Return the real, user, and system times in seconds from an
1517 1517 // arbitrary fixed point in the past.
1518 1518 bool os::getTimesSecs(double* process_real_time,
1519 1519 double* process_user_time,
1520 1520 double* process_system_time) {
1521 1521 struct tms ticks;
1522 1522 clock_t real_ticks = times(&ticks);
1523 1523
1524 1524 if (real_ticks == (clock_t) (-1)) {
1525 1525 return false;
1526 1526 } else {
1527 1527 double ticks_per_second = (double) clock_tics_per_sec;
1528 1528 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1529 1529 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1530 1530 *process_real_time = ((double) real_ticks) / ticks_per_second;
1531 1531
1532 1532 return true;
1533 1533 }
1534 1534 }
1535 1535
1536 1536
1537 1537 char * os::local_time_string(char *buf, size_t buflen) {
1538 1538 struct tm t;
1539 1539 time_t long_time;
1540 1540 time(&long_time);
1541 1541 localtime_r(&long_time, &t);
1542 1542 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1543 1543 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1544 1544 t.tm_hour, t.tm_min, t.tm_sec);
1545 1545 return buf;
1546 1546 }
1547 1547
1548 1548 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1549 1549 return localtime_r(clock, res);
1550 1550 }
1551 1551
1552 1552 ////////////////////////////////////////////////////////////////////////////////
1553 1553 // runtime exit support
1554 1554
1555 1555 // Note: os::shutdown() might be called very early during initialization, or
1556 1556 // called from signal handler. Before adding something to os::shutdown(), make
1557 1557 // sure it is async-safe and can handle partially initialized VM.
1558 1558 void os::shutdown() {
1559 1559
1560 1560 // allow PerfMemory to attempt cleanup of any persistent resources
1561 1561 perfMemory_exit();
1562 1562
1563 1563 // needs to remove object in file system
1564 1564 AttachListener::abort();
1565 1565
1566 1566 // flush buffered output, finish log files
1567 1567 ostream_abort();
1568 1568
1569 1569 // Check for abort hook
1570 1570 abort_hook_t abort_hook = Arguments::abort_hook();
1571 1571 if (abort_hook != NULL) {
1572 1572 abort_hook();
1573 1573 }
1574 1574
1575 1575 }
1576 1576
1577 1577 // Note: os::abort() might be called very early during initialization, or
1578 1578 // called from signal handler. Before adding something to os::abort(), make
1579 1579 // sure it is async-safe and can handle partially initialized VM.
1580 1580 void os::abort(bool dump_core) {
1581 1581 os::shutdown();
1582 1582 if (dump_core) {
1583 1583 #ifndef PRODUCT
1584 1584 fdStream out(defaultStream::output_fd());
1585 1585 out.print_raw("Current thread is ");
1586 1586 char buf[16];
1587 1587 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1588 1588 out.print_raw_cr(buf);
1589 1589 out.print_raw_cr("Dumping core ...");
1590 1590 #endif
1591 1591 ::abort(); // dump core
1592 1592 }
1593 1593
1594 1594 ::exit(1);
1595 1595 }
1596 1596
1597 1597 // Die immediately, no exit hook, no abort hook, no cleanup.
1598 1598 void os::die() {
1599 1599 // _exit() on LinuxThreads only kills current thread
1600 1600 ::abort();
1601 1601 }
1602 1602
1603 1603 // unused on linux for now.
1604 1604 void os::set_error_file(const char *logfile) {}
1605 1605
1606 1606
1607 1607 // This method is a copy of JDK's sysGetLastErrorString
1608 1608 // from src/solaris/hpi/src/system_md.c
1609 1609
1610 1610 size_t os::lasterror(char *buf, size_t len) {
1611 1611
1612 1612 if (errno == 0) return 0;
1613 1613
1614 1614 const char *s = ::strerror(errno);
1615 1615 size_t n = ::strlen(s);
1616 1616 if (n >= len) {
1617 1617 n = len - 1;
1618 1618 }
1619 1619 ::strncpy(buf, s, n);
1620 1620 buf[n] = '\0';
1621 1621 return n;
1622 1622 }
1623 1623
1624 1624 intx os::current_thread_id() { return (intx)pthread_self(); }
1625 1625 int os::current_process_id() {
1626 1626
1627 1627 // Under the old linux thread library, linux gives each thread
1628 1628 // its own process id. Because of this each thread will return
1629 1629 // a different pid if this method were to return the result
1630 1630 // of getpid(2). Linux provides no api that returns the pid
1631 1631 // of the launcher thread for the vm. This implementation
1632 1632 // returns a unique pid, the pid of the launcher thread
1633 1633 // that starts the vm 'process'.
1634 1634
1635 1635 // Under the NPTL, getpid() returns the same pid as the
1636 1636 // launcher thread rather than a unique pid per thread.
1637 1637 // Use gettid() if you want the old pre NPTL behaviour.
1638 1638
1639 1639 // if you are looking for the result of a call to getpid() that
1640 1640 // returns a unique pid for the calling thread, then look at the
1641 1641 // OSThread::thread_id() method in osThread_linux.hpp file
1642 1642
1643 1643 return (int)(_initial_pid ? _initial_pid : getpid());
1644 1644 }
1645 1645
1646 1646 // DLL functions
1647 1647
1648 1648 const char* os::dll_file_extension() { return ".so"; }
1649 1649
1650 1650 // This must be hard coded because it's the system's temporary
1651 1651 // directory not the java application's temp directory, ala java.io.tmpdir.
1652 1652 const char* os::get_temp_directory() { return "/tmp"; }
1653 1653
1654 1654 static bool file_exists(const char* filename) {
1655 1655 struct stat statbuf;
1656 1656 if (filename == NULL || strlen(filename) == 0) {
1657 1657 return false;
1658 1658 }
1659 1659 return os::stat(filename, &statbuf) == 0;
1660 1660 }
1661 1661
1662 1662 void os::dll_build_name(char* buffer, size_t buflen,
1663 1663 const char* pname, const char* fname) {
1664 1664 // Copied from libhpi
1665 1665 const size_t pnamelen = pname ? strlen(pname) : 0;
1666 1666
1667 1667 // Quietly truncate on buffer overflow. Should be an error.
1668 1668 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1669 1669 *buffer = '\0';
1670 1670 return;
1671 1671 }
1672 1672
1673 1673 if (pnamelen == 0) {
1674 1674 snprintf(buffer, buflen, "lib%s.so", fname);
1675 1675 } else if (strchr(pname, *os::path_separator()) != NULL) {
1676 1676 int n;
1677 1677 char** pelements = split_path(pname, &n);
1678 1678 for (int i = 0 ; i < n ; i++) {
1679 1679 // Really shouldn't be NULL, but check can't hurt
1680 1680 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1681 1681 continue; // skip the empty path values
1682 1682 }
1683 1683 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1684 1684 if (file_exists(buffer)) {
1685 1685 break;
1686 1686 }
1687 1687 }
1688 1688 // release the storage
1689 1689 for (int i = 0 ; i < n ; i++) {
1690 1690 if (pelements[i] != NULL) {
1691 1691 FREE_C_HEAP_ARRAY(char, pelements[i]);
1692 1692 }
1693 1693 }
1694 1694 if (pelements != NULL) {
1695 1695 FREE_C_HEAP_ARRAY(char*, pelements);
1696 1696 }
1697 1697 } else {
1698 1698 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1699 1699 }
1700 1700 }
1701 1701
1702 1702 const char* os::get_current_directory(char *buf, int buflen) {
1703 1703 return getcwd(buf, buflen);
1704 1704 }
1705 1705
1706 1706 // check if addr is inside libjvm[_g].so
1707 1707 bool os::address_is_in_vm(address addr) {
1708 1708 static address libjvm_base_addr;
1709 1709 Dl_info dlinfo;
1710 1710
1711 1711 if (libjvm_base_addr == NULL) {
1712 1712 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1713 1713 libjvm_base_addr = (address)dlinfo.dli_fbase;
1714 1714 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1715 1715 }
1716 1716
1717 1717 if (dladdr((void *)addr, &dlinfo)) {
1718 1718 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1719 1719 }
1720 1720
1721 1721 return false;
1722 1722 }
1723 1723
1724 1724 bool os::dll_address_to_function_name(address addr, char *buf,
1725 1725 int buflen, int *offset) {
1726 1726 Dl_info dlinfo;
1727 1727
1728 1728 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1729 1729 if (buf != NULL) {
1730 1730 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1731 1731 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1732 1732 }
1733 1733 }
1734 1734 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1735 1735 return true;
1736 1736 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1737 1737 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1738 1738 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1739 1739 return true;
1740 1740 }
1741 1741 }
1742 1742
1743 1743 if (buf != NULL) buf[0] = '\0';
1744 1744 if (offset != NULL) *offset = -1;
1745 1745 return false;
1746 1746 }
1747 1747
1748 1748 struct _address_to_library_name {
1749 1749 address addr; // input : memory address
1750 1750 size_t buflen; // size of fname
1751 1751 char* fname; // output: library name
1752 1752 address base; // library base addr
1753 1753 };
1754 1754
1755 1755 static int address_to_library_name_callback(struct dl_phdr_info *info,
1756 1756 size_t size, void *data) {
1757 1757 int i;
1758 1758 bool found = false;
1759 1759 address libbase = NULL;
1760 1760 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1761 1761
1762 1762 // iterate through all loadable segments
1763 1763 for (i = 0; i < info->dlpi_phnum; i++) {
1764 1764 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1765 1765 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1766 1766 // base address of a library is the lowest address of its loaded
1767 1767 // segments.
1768 1768 if (libbase == NULL || libbase > segbase) {
1769 1769 libbase = segbase;
1770 1770 }
1771 1771 // see if 'addr' is within current segment
1772 1772 if (segbase <= d->addr &&
1773 1773 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1774 1774 found = true;
1775 1775 }
1776 1776 }
1777 1777 }
1778 1778
1779 1779 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1780 1780 // so dll_address_to_library_name() can fall through to use dladdr() which
1781 1781 // can figure out executable name from argv[0].
1782 1782 if (found && info->dlpi_name && info->dlpi_name[0]) {
1783 1783 d->base = libbase;
1784 1784 if (d->fname) {
1785 1785 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1786 1786 }
1787 1787 return 1;
1788 1788 }
1789 1789 return 0;
1790 1790 }
1791 1791
1792 1792 bool os::dll_address_to_library_name(address addr, char* buf,
1793 1793 int buflen, int* offset) {
1794 1794 Dl_info dlinfo;
1795 1795 struct _address_to_library_name data;
1796 1796
1797 1797 // There is a bug in old glibc dladdr() implementation that it could resolve
1798 1798 // to wrong library name if the .so file has a base address != NULL. Here
1799 1799 // we iterate through the program headers of all loaded libraries to find
1800 1800 // out which library 'addr' really belongs to. This workaround can be
1801 1801 // removed once the minimum requirement for glibc is moved to 2.3.x.
1802 1802 data.addr = addr;
1803 1803 data.fname = buf;
1804 1804 data.buflen = buflen;
1805 1805 data.base = NULL;
1806 1806 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1807 1807
1808 1808 if (rslt) {
1809 1809 // buf already contains library name
1810 1810 if (offset) *offset = addr - data.base;
1811 1811 return true;
1812 1812 } else if (dladdr((void*)addr, &dlinfo)){
1813 1813 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1814 1814 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1815 1815 return true;
1816 1816 } else {
1817 1817 if (buf) buf[0] = '\0';
1818 1818 if (offset) *offset = -1;
1819 1819 return false;
1820 1820 }
1821 1821 }
1822 1822
1823 1823 // Loads .dll/.so and
1824 1824 // in case of error it checks if .dll/.so was built for the
1825 1825 // same architecture as Hotspot is running on
1826 1826
1827 1827 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1828 1828 {
1829 1829 void * result= ::dlopen(filename, RTLD_LAZY);
1830 1830 if (result != NULL) {
1831 1831 // Successful loading
1832 1832 return result;
1833 1833 }
1834 1834
1835 1835 Elf32_Ehdr elf_head;
1836 1836
1837 1837 // Read system error message into ebuf
1838 1838 // It may or may not be overwritten below
1839 1839 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1840 1840 ebuf[ebuflen-1]='\0';
1841 1841 int diag_msg_max_length=ebuflen-strlen(ebuf);
1842 1842 char* diag_msg_buf=ebuf+strlen(ebuf);
1843 1843
1844 1844 if (diag_msg_max_length==0) {
1845 1845 // No more space in ebuf for additional diagnostics message
1846 1846 return NULL;
1847 1847 }
1848 1848
1849 1849
1850 1850 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1851 1851
1852 1852 if (file_descriptor < 0) {
1853 1853 // Can't open library, report dlerror() message
1854 1854 return NULL;
1855 1855 }
1856 1856
1857 1857 bool failed_to_read_elf_head=
1858 1858 (sizeof(elf_head)!=
1859 1859 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1860 1860
1861 1861 ::close(file_descriptor);
1862 1862 if (failed_to_read_elf_head) {
1863 1863 // file i/o error - report dlerror() msg
1864 1864 return NULL;
1865 1865 }
1866 1866
1867 1867 typedef struct {
1868 1868 Elf32_Half code; // Actual value as defined in elf.h
1869 1869 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1870 1870 char elf_class; // 32 or 64 bit
1871 1871 char endianess; // MSB or LSB
1872 1872 char* name; // String representation
1873 1873 } arch_t;
1874 1874
1875 1875 #ifndef EM_486
1876 1876 #define EM_486 6 /* Intel 80486 */
1877 1877 #endif
1878 1878
1879 1879 static const arch_t arch_array[]={
1880 1880 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1881 1881 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1882 1882 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1883 1883 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1884 1884 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1885 1885 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1886 1886 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1887 1887 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1888 1888 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1889 1889 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1890 1890 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1891 1891 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1892 1892 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1893 1893 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1894 1894 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1895 1895 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1896 1896 };
1897 1897
1898 1898 #if (defined IA32)
1899 1899 static Elf32_Half running_arch_code=EM_386;
1900 1900 #elif (defined AMD64)
1901 1901 static Elf32_Half running_arch_code=EM_X86_64;
1902 1902 #elif (defined IA64)
1903 1903 static Elf32_Half running_arch_code=EM_IA_64;
1904 1904 #elif (defined __sparc) && (defined _LP64)
1905 1905 static Elf32_Half running_arch_code=EM_SPARCV9;
1906 1906 #elif (defined __sparc) && (!defined _LP64)
1907 1907 static Elf32_Half running_arch_code=EM_SPARC;
1908 1908 #elif (defined __powerpc64__)
1909 1909 static Elf32_Half running_arch_code=EM_PPC64;
1910 1910 #elif (defined __powerpc__)
1911 1911 static Elf32_Half running_arch_code=EM_PPC;
1912 1912 #elif (defined ARM)
1913 1913 static Elf32_Half running_arch_code=EM_ARM;
1914 1914 #elif (defined S390)
1915 1915 static Elf32_Half running_arch_code=EM_S390;
1916 1916 #elif (defined ALPHA)
1917 1917 static Elf32_Half running_arch_code=EM_ALPHA;
1918 1918 #elif (defined MIPSEL)
1919 1919 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1920 1920 #elif (defined PARISC)
1921 1921 static Elf32_Half running_arch_code=EM_PARISC;
1922 1922 #elif (defined MIPS)
1923 1923 static Elf32_Half running_arch_code=EM_MIPS;
1924 1924 #elif (defined M68K)
1925 1925 static Elf32_Half running_arch_code=EM_68K;
1926 1926 #else
1927 1927 #error Method os::dll_load requires that one of following is defined:\
1928 1928 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1929 1929 #endif
1930 1930
1931 1931 // Identify compatability class for VM's architecture and library's architecture
1932 1932 // Obtain string descriptions for architectures
1933 1933
1934 1934 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1935 1935 int running_arch_index=-1;
1936 1936
1937 1937 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1938 1938 if (running_arch_code == arch_array[i].code) {
1939 1939 running_arch_index = i;
1940 1940 }
1941 1941 if (lib_arch.code == arch_array[i].code) {
1942 1942 lib_arch.compat_class = arch_array[i].compat_class;
1943 1943 lib_arch.name = arch_array[i].name;
1944 1944 }
1945 1945 }
1946 1946
1947 1947 assert(running_arch_index != -1,
1948 1948 "Didn't find running architecture code (running_arch_code) in arch_array");
1949 1949 if (running_arch_index == -1) {
1950 1950 // Even though running architecture detection failed
1951 1951 // we may still continue with reporting dlerror() message
1952 1952 return NULL;
1953 1953 }
1954 1954
1955 1955 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1956 1956 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1957 1957 return NULL;
1958 1958 }
1959 1959
1960 1960 #ifndef S390
1961 1961 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1962 1962 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1963 1963 return NULL;
1964 1964 }
1965 1965 #endif // !S390
1966 1966
1967 1967 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1968 1968 if ( lib_arch.name!=NULL ) {
1969 1969 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1970 1970 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1971 1971 lib_arch.name, arch_array[running_arch_index].name);
1972 1972 } else {
1973 1973 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1974 1974 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1975 1975 lib_arch.code,
1976 1976 arch_array[running_arch_index].name);
1977 1977 }
1978 1978 }
1979 1979
1980 1980 return NULL;
1981 1981 }
1982 1982
1983 1983 /*
1984 1984 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1985 1985 * chances are you might want to run the generated bits against glibc-2.0
1986 1986 * libdl.so, so always use locking for any version of glibc.
1987 1987 */
1988 1988 void* os::dll_lookup(void* handle, const char* name) {
1989 1989 pthread_mutex_lock(&dl_mutex);
1990 1990 void* res = dlsym(handle, name);
1991 1991 pthread_mutex_unlock(&dl_mutex);
1992 1992 return res;
1993 1993 }
1994 1994
1995 1995
1996 1996 static bool _print_ascii_file(const char* filename, outputStream* st) {
1997 1997 int fd = ::open(filename, O_RDONLY);
1998 1998 if (fd == -1) {
1999 1999 return false;
2000 2000 }
2001 2001
2002 2002 char buf[32];
2003 2003 int bytes;
2004 2004 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2005 2005 st->print_raw(buf, bytes);
2006 2006 }
2007 2007
2008 2008 ::close(fd);
2009 2009
2010 2010 return true;
2011 2011 }
2012 2012
2013 2013 void os::print_dll_info(outputStream *st) {
2014 2014 st->print_cr("Dynamic libraries:");
2015 2015
2016 2016 char fname[32];
2017 2017 pid_t pid = os::Linux::gettid();
2018 2018
2019 2019 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2020 2020
2021 2021 if (!_print_ascii_file(fname, st)) {
2022 2022 st->print("Can not get library information for pid = %d\n", pid);
2023 2023 }
2024 2024 }
2025 2025
2026 2026
2027 2027 void os::print_os_info(outputStream* st) {
2028 2028 st->print("OS:");
2029 2029
2030 2030 // Try to identify popular distros.
2031 2031 // Most Linux distributions have /etc/XXX-release file, which contains
2032 2032 // the OS version string. Some have more than one /etc/XXX-release file
2033 2033 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2034 2034 // so the order is important.
2035 2035 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2036 2036 !_print_ascii_file("/etc/sun-release", st) &&
2037 2037 !_print_ascii_file("/etc/redhat-release", st) &&
2038 2038 !_print_ascii_file("/etc/SuSE-release", st) &&
2039 2039 !_print_ascii_file("/etc/turbolinux-release", st) &&
2040 2040 !_print_ascii_file("/etc/gentoo-release", st) &&
2041 2041 !_print_ascii_file("/etc/debian_version", st) &&
2042 2042 !_print_ascii_file("/etc/ltib-release", st) &&
2043 2043 !_print_ascii_file("/etc/angstrom-version", st)) {
2044 2044 st->print("Linux");
2045 2045 }
2046 2046 st->cr();
2047 2047
2048 2048 // kernel
2049 2049 st->print("uname:");
2050 2050 struct utsname name;
2051 2051 uname(&name);
2052 2052 st->print(name.sysname); st->print(" ");
2053 2053 st->print(name.release); st->print(" ");
2054 2054 st->print(name.version); st->print(" ");
2055 2055 st->print(name.machine);
2056 2056 st->cr();
2057 2057
2058 2058 // Print warning if unsafe chroot environment detected
2059 2059 if (unsafe_chroot_detected) {
2060 2060 st->print("WARNING!! ");
2061 2061 st->print_cr(unstable_chroot_error);
2062 2062 }
2063 2063
2064 2064 // libc, pthread
2065 2065 st->print("libc:");
2066 2066 st->print(os::Linux::glibc_version()); st->print(" ");
2067 2067 st->print(os::Linux::libpthread_version()); st->print(" ");
2068 2068 if (os::Linux::is_LinuxThreads()) {
2069 2069 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2070 2070 }
2071 2071 st->cr();
2072 2072
2073 2073 // rlimit
2074 2074 st->print("rlimit:");
2075 2075 struct rlimit rlim;
2076 2076
2077 2077 st->print(" STACK ");
2078 2078 getrlimit(RLIMIT_STACK, &rlim);
2079 2079 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2080 2080 else st->print("%uk", rlim.rlim_cur >> 10);
2081 2081
2082 2082 st->print(", CORE ");
2083 2083 getrlimit(RLIMIT_CORE, &rlim);
2084 2084 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2085 2085 else st->print("%uk", rlim.rlim_cur >> 10);
2086 2086
2087 2087 st->print(", NPROC ");
2088 2088 getrlimit(RLIMIT_NPROC, &rlim);
2089 2089 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2090 2090 else st->print("%d", rlim.rlim_cur);
2091 2091
2092 2092 st->print(", NOFILE ");
2093 2093 getrlimit(RLIMIT_NOFILE, &rlim);
2094 2094 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2095 2095 else st->print("%d", rlim.rlim_cur);
2096 2096
2097 2097 st->print(", AS ");
2098 2098 getrlimit(RLIMIT_AS, &rlim);
2099 2099 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2100 2100 else st->print("%uk", rlim.rlim_cur >> 10);
2101 2101 st->cr();
2102 2102
2103 2103 // load average
2104 2104 st->print("load average:");
2105 2105 double loadavg[3];
2106 2106 os::loadavg(loadavg, 3);
2107 2107 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2108 2108 st->cr();
2109 2109
2110 2110 // meminfo
2111 2111 st->print("\n/proc/meminfo:\n");
2112 2112 _print_ascii_file("/proc/meminfo", st);
2113 2113 st->cr();
2114 2114 }
2115 2115
2116 2116 void os::pd_print_cpu_info(outputStream* st) {
2117 2117 st->print("\n/proc/cpuinfo:\n");
2118 2118 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2119 2119 st->print(" <Not Available>");
2120 2120 }
2121 2121 st->cr();
2122 2122 }
2123 2123
2124 2124 void os::print_memory_info(outputStream* st) {
2125 2125
2126 2126 st->print("Memory:");
2127 2127 st->print(" %dk page", os::vm_page_size()>>10);
2128 2128
2129 2129 // values in struct sysinfo are "unsigned long"
2130 2130 struct sysinfo si;
2131 2131 sysinfo(&si);
2132 2132
2133 2133 st->print(", physical " UINT64_FORMAT "k",
2134 2134 os::physical_memory() >> 10);
2135 2135 st->print("(" UINT64_FORMAT "k free)",
2136 2136 os::available_memory() >> 10);
2137 2137 st->print(", swap " UINT64_FORMAT "k",
2138 2138 ((jlong)si.totalswap * si.mem_unit) >> 10);
2139 2139 st->print("(" UINT64_FORMAT "k free)",
2140 2140 ((jlong)si.freeswap * si.mem_unit) >> 10);
2141 2141 st->cr();
2142 2142 }
2143 2143
2144 2144 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2145 2145 // but they're the same for all the linux arch that we support
2146 2146 // and they're the same for solaris but there's no common place to put this.
2147 2147 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2148 2148 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2149 2149 "ILL_COPROC", "ILL_BADSTK" };
2150 2150
2151 2151 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2152 2152 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2153 2153 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2154 2154
2155 2155 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2156 2156
2157 2157 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2158 2158
2159 2159 void os::print_siginfo(outputStream* st, void* siginfo) {
2160 2160 st->print("siginfo:");
2161 2161
2162 2162 const int buflen = 100;
2163 2163 char buf[buflen];
2164 2164 siginfo_t *si = (siginfo_t*)siginfo;
2165 2165 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2166 2166 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2167 2167 st->print("si_errno=%s", buf);
2168 2168 } else {
2169 2169 st->print("si_errno=%d", si->si_errno);
2170 2170 }
2171 2171 const int c = si->si_code;
2172 2172 assert(c > 0, "unexpected si_code");
2173 2173 switch (si->si_signo) {
2174 2174 case SIGILL:
2175 2175 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2176 2176 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2177 2177 break;
2178 2178 case SIGFPE:
2179 2179 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2180 2180 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2181 2181 break;
2182 2182 case SIGSEGV:
2183 2183 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2184 2184 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2185 2185 break;
2186 2186 case SIGBUS:
2187 2187 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2188 2188 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2189 2189 break;
2190 2190 default:
2191 2191 st->print(", si_code=%d", si->si_code);
2192 2192 // no si_addr
2193 2193 }
2194 2194
2195 2195 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2196 2196 UseSharedSpaces) {
2197 2197 FileMapInfo* mapinfo = FileMapInfo::current_info();
2198 2198 if (mapinfo->is_in_shared_space(si->si_addr)) {
2199 2199 st->print("\n\nError accessing class data sharing archive." \
2200 2200 " Mapped file inaccessible during execution, " \
2201 2201 " possible disk/network problem.");
2202 2202 }
2203 2203 }
2204 2204 st->cr();
2205 2205 }
2206 2206
2207 2207
2208 2208 static void print_signal_handler(outputStream* st, int sig,
2209 2209 char* buf, size_t buflen);
2210 2210
2211 2211 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2212 2212 st->print_cr("Signal Handlers:");
2213 2213 print_signal_handler(st, SIGSEGV, buf, buflen);
2214 2214 print_signal_handler(st, SIGBUS , buf, buflen);
2215 2215 print_signal_handler(st, SIGFPE , buf, buflen);
2216 2216 print_signal_handler(st, SIGPIPE, buf, buflen);
2217 2217 print_signal_handler(st, SIGXFSZ, buf, buflen);
2218 2218 print_signal_handler(st, SIGILL , buf, buflen);
2219 2219 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2220 2220 print_signal_handler(st, SR_signum, buf, buflen);
2221 2221 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2222 2222 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2223 2223 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2224 2224 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2225 2225 }
2226 2226
2227 2227 static char saved_jvm_path[MAXPATHLEN] = {0};
2228 2228
2229 2229 // Find the full path to the current module, libjvm.so or libjvm_g.so
2230 2230 void os::jvm_path(char *buf, jint buflen) {
2231 2231 // Error checking.
2232 2232 if (buflen < MAXPATHLEN) {
2233 2233 assert(false, "must use a large-enough buffer");
2234 2234 buf[0] = '\0';
2235 2235 return;
2236 2236 }
2237 2237 // Lazy resolve the path to current module.
2238 2238 if (saved_jvm_path[0] != 0) {
2239 2239 strcpy(buf, saved_jvm_path);
2240 2240 return;
2241 2241 }
2242 2242
2243 2243 char dli_fname[MAXPATHLEN];
2244 2244 bool ret = dll_address_to_library_name(
2245 2245 CAST_FROM_FN_PTR(address, os::jvm_path),
2246 2246 dli_fname, sizeof(dli_fname), NULL);
2247 2247 assert(ret != 0, "cannot locate libjvm");
2248 2248 char *rp = realpath(dli_fname, buf);
2249 2249 if (rp == NULL)
2250 2250 return;
2251 2251
2252 2252 if (Arguments::created_by_gamma_launcher()) {
2253 2253 // Support for the gamma launcher. Typical value for buf is
2254 2254 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2255 2255 // the right place in the string, then assume we are installed in a JDK and
2256 2256 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2257 2257 // up the path so it looks like libjvm.so is installed there (append a
2258 2258 // fake suffix hotspot/libjvm.so).
2259 2259 const char *p = buf + strlen(buf) - 1;
2260 2260 for (int count = 0; p > buf && count < 5; ++count) {
2261 2261 for (--p; p > buf && *p != '/'; --p)
2262 2262 /* empty */ ;
2263 2263 }
2264 2264
2265 2265 if (strncmp(p, "/jre/lib/", 9) != 0) {
2266 2266 // Look for JAVA_HOME in the environment.
2267 2267 char* java_home_var = ::getenv("JAVA_HOME");
2268 2268 if (java_home_var != NULL && java_home_var[0] != 0) {
2269 2269 char* jrelib_p;
2270 2270 int len;
2271 2271
2272 2272 // Check the current module name "libjvm.so" or "libjvm_g.so".
2273 2273 p = strrchr(buf, '/');
2274 2274 assert(strstr(p, "/libjvm") == p, "invalid library name");
2275 2275 p = strstr(p, "_g") ? "_g" : "";
2276 2276
2277 2277 rp = realpath(java_home_var, buf);
2278 2278 if (rp == NULL)
2279 2279 return;
2280 2280
2281 2281 // determine if this is a legacy image or modules image
2282 2282 // modules image doesn't have "jre" subdirectory
2283 2283 len = strlen(buf);
2284 2284 jrelib_p = buf + len;
2285 2285 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2286 2286 if (0 != access(buf, F_OK)) {
2287 2287 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2288 2288 }
2289 2289
2290 2290 if (0 == access(buf, F_OK)) {
2291 2291 // Use current module name "libjvm[_g].so" instead of
2292 2292 // "libjvm"debug_only("_g")".so" since for fastdebug version
2293 2293 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2294 2294 len = strlen(buf);
2295 2295 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2296 2296 } else {
2297 2297 // Go back to path of .so
2298 2298 rp = realpath(dli_fname, buf);
2299 2299 if (rp == NULL)
2300 2300 return;
2301 2301 }
2302 2302 }
2303 2303 }
2304 2304 }
2305 2305
2306 2306 strcpy(saved_jvm_path, buf);
2307 2307 }
2308 2308
2309 2309 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2310 2310 // no prefix required, not even "_"
2311 2311 }
2312 2312
2313 2313 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2314 2314 // no suffix required
2315 2315 }
2316 2316
2317 2317 ////////////////////////////////////////////////////////////////////////////////
2318 2318 // sun.misc.Signal support
2319 2319
2320 2320 static volatile jint sigint_count = 0;
2321 2321
2322 2322 static void
2323 2323 UserHandler(int sig, void *siginfo, void *context) {
2324 2324 // 4511530 - sem_post is serialized and handled by the manager thread. When
2325 2325 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2326 2326 // don't want to flood the manager thread with sem_post requests.
2327 2327 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2328 2328 return;
2329 2329
2330 2330 // Ctrl-C is pressed during error reporting, likely because the error
2331 2331 // handler fails to abort. Let VM die immediately.
2332 2332 if (sig == SIGINT && is_error_reported()) {
2333 2333 os::die();
2334 2334 }
2335 2335
2336 2336 os::signal_notify(sig);
2337 2337 }
2338 2338
2339 2339 void* os::user_handler() {
2340 2340 return CAST_FROM_FN_PTR(void*, UserHandler);
2341 2341 }
2342 2342
2343 2343 extern "C" {
2344 2344 typedef void (*sa_handler_t)(int);
2345 2345 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2346 2346 }
2347 2347
2348 2348 void* os::signal(int signal_number, void* handler) {
2349 2349 struct sigaction sigAct, oldSigAct;
2350 2350
2351 2351 sigfillset(&(sigAct.sa_mask));
2352 2352 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2353 2353 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2354 2354
2355 2355 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2356 2356 // -1 means registration failed
2357 2357 return (void *)-1;
2358 2358 }
2359 2359
2360 2360 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2361 2361 }
2362 2362
2363 2363 void os::signal_raise(int signal_number) {
2364 2364 ::raise(signal_number);
2365 2365 }
2366 2366
2367 2367 /*
2368 2368 * The following code is moved from os.cpp for making this
2369 2369 * code platform specific, which it is by its very nature.
2370 2370 */
2371 2371
2372 2372 // Will be modified when max signal is changed to be dynamic
2373 2373 int os::sigexitnum_pd() {
2374 2374 return NSIG;
2375 2375 }
2376 2376
2377 2377 // a counter for each possible signal value
2378 2378 static volatile jint pending_signals[NSIG+1] = { 0 };
2379 2379
2380 2380 // Linux(POSIX) specific hand shaking semaphore.
2381 2381 static sem_t sig_sem;
2382 2382
2383 2383 void os::signal_init_pd() {
2384 2384 // Initialize signal structures
2385 2385 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2386 2386
2387 2387 // Initialize signal semaphore
2388 2388 ::sem_init(&sig_sem, 0, 0);
2389 2389 }
2390 2390
2391 2391 void os::signal_notify(int sig) {
2392 2392 Atomic::inc(&pending_signals[sig]);
2393 2393 ::sem_post(&sig_sem);
2394 2394 }
2395 2395
2396 2396 static int check_pending_signals(bool wait) {
2397 2397 Atomic::store(0, &sigint_count);
2398 2398 for (;;) {
2399 2399 for (int i = 0; i < NSIG + 1; i++) {
2400 2400 jint n = pending_signals[i];
2401 2401 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2402 2402 return i;
2403 2403 }
2404 2404 }
2405 2405 if (!wait) {
2406 2406 return -1;
2407 2407 }
2408 2408 JavaThread *thread = JavaThread::current();
2409 2409 ThreadBlockInVM tbivm(thread);
2410 2410
2411 2411 bool threadIsSuspended;
2412 2412 do {
2413 2413 thread->set_suspend_equivalent();
2414 2414 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2415 2415 ::sem_wait(&sig_sem);
2416 2416
2417 2417 // were we externally suspended while we were waiting?
2418 2418 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2419 2419 if (threadIsSuspended) {
2420 2420 //
2421 2421 // The semaphore has been incremented, but while we were waiting
2422 2422 // another thread suspended us. We don't want to continue running
2423 2423 // while suspended because that would surprise the thread that
2424 2424 // suspended us.
2425 2425 //
2426 2426 ::sem_post(&sig_sem);
2427 2427
2428 2428 thread->java_suspend_self();
2429 2429 }
2430 2430 } while (threadIsSuspended);
2431 2431 }
2432 2432 }
2433 2433
2434 2434 int os::signal_lookup() {
2435 2435 return check_pending_signals(false);
2436 2436 }
2437 2437
2438 2438 int os::signal_wait() {
2439 2439 return check_pending_signals(true);
2440 2440 }
2441 2441
2442 2442 ////////////////////////////////////////////////////////////////////////////////
2443 2443 // Virtual Memory
2444 2444
2445 2445 int os::vm_page_size() {
2446 2446 // Seems redundant as all get out
2447 2447 assert(os::Linux::page_size() != -1, "must call os::init");
2448 2448 return os::Linux::page_size();
2449 2449 }
2450 2450
2451 2451 // Solaris allocates memory by pages.
2452 2452 int os::vm_allocation_granularity() {
2453 2453 assert(os::Linux::page_size() != -1, "must call os::init");
2454 2454 return os::Linux::page_size();
2455 2455 }
2456 2456
2457 2457 // Rationale behind this function:
2458 2458 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2459 2459 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2460 2460 // samples for JITted code. Here we create private executable mapping over the code cache
2461 2461 // and then we can use standard (well, almost, as mapping can change) way to provide
2462 2462 // info for the reporting script by storing timestamp and location of symbol
2463 2463 void linux_wrap_code(char* base, size_t size) {
2464 2464 static volatile jint cnt = 0;
2465 2465
2466 2466 if (!UseOprofile) {
2467 2467 return;
2468 2468 }
2469 2469
2470 2470 char buf[PATH_MAX+1];
2471 2471 int num = Atomic::add(1, &cnt);
2472 2472
2473 2473 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2474 2474 os::get_temp_directory(), os::current_process_id(), num);
2475 2475 unlink(buf);
2476 2476
2477 2477 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2478 2478
2479 2479 if (fd != -1) {
2480 2480 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2481 2481 if (rv != (off_t)-1) {
2482 2482 if (::write(fd, "", 1) == 1) {
2483 2483 mmap(base, size,
2484 2484 PROT_READ|PROT_WRITE|PROT_EXEC,
2485 2485 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2486 2486 }
2487 2487 }
2488 2488 ::close(fd);
2489 2489 unlink(buf);
2490 2490 }
2491 2491 }
2492 2492
2493 2493 // NOTE: Linux kernel does not really reserve the pages for us.
2494 2494 // All it does is to check if there are enough free pages
2495 2495 // left at the time of mmap(). This could be a potential
2496 2496 // problem.
2497 2497 bool os::commit_memory(char* addr, size_t size, bool exec) {
2498 2498 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2499 2499 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2500 2500 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2501 2501 return res != (uintptr_t) MAP_FAILED;
2502 2502 }
2503 2503
2504 2504 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2505 2505 #ifndef MAP_HUGETLB
2506 2506 #define MAP_HUGETLB 0x40000
2507 2507 #endif
2508 2508
2509 2509 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2510 2510 #ifndef MADV_HUGEPAGE
2511 2511 #define MADV_HUGEPAGE 14
2512 2512 #endif
2513 2513
2514 2514 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2515 2515 bool exec) {
2516 2516 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2517 2517 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2518 2518 uintptr_t res =
2519 2519 (uintptr_t) ::mmap(addr, size, prot,
2520 2520 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2521 2521 -1, 0);
2522 2522 return res != (uintptr_t) MAP_FAILED;
2523 2523 }
2524 2524
2525 2525 return commit_memory(addr, size, exec);
2526 2526 }
↓ open down ↓ |
2526 lines elided |
↑ open up ↑ |
2527 2527
2528 2528 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2529 2529 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2530 2530 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2531 2531 // be supported or the memory may already be backed by huge pages.
2532 2532 ::madvise(addr, bytes, MADV_HUGEPAGE);
2533 2533 }
2534 2534 }
2535 2535
2536 2536 void os::free_memory(char *addr, size_t bytes) {
2537 - ::madvise(addr, bytes, MADV_DONTNEED);
2537 + commit_memory(addr, bytes, false);
2538 2538 }
2539 2539
2540 2540 void os::numa_make_global(char *addr, size_t bytes) {
2541 2541 Linux::numa_interleave_memory(addr, bytes);
2542 2542 }
2543 2543
2544 2544 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2545 2545 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2546 2546 }
2547 2547
2548 2548 bool os::numa_topology_changed() { return false; }
2549 2549
2550 2550 size_t os::numa_get_groups_num() {
2551 2551 int max_node = Linux::numa_max_node();
2552 2552 return max_node > 0 ? max_node + 1 : 1;
2553 2553 }
2554 2554
2555 2555 int os::numa_get_group_id() {
2556 2556 int cpu_id = Linux::sched_getcpu();
2557 2557 if (cpu_id != -1) {
2558 2558 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2559 2559 if (lgrp_id != -1) {
2560 2560 return lgrp_id;
2561 2561 }
2562 2562 }
2563 2563 return 0;
2564 2564 }
2565 2565
2566 2566 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2567 2567 for (size_t i = 0; i < size; i++) {
2568 2568 ids[i] = i;
2569 2569 }
2570 2570 return size;
2571 2571 }
2572 2572
2573 2573 bool os::get_page_info(char *start, page_info* info) {
2574 2574 return false;
2575 2575 }
2576 2576
2577 2577 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2578 2578 return end;
2579 2579 }
2580 2580
2581 2581 // Something to do with the numa-aware allocator needs these symbols
2582 2582 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2583 2583 extern "C" JNIEXPORT void numa_error(char *where) { }
2584 2584 extern "C" JNIEXPORT int fork1() { return fork(); }
2585 2585
2586 2586
2587 2587 // If we are running with libnuma version > 2, then we should
2588 2588 // be trying to use symbols with versions 1.1
2589 2589 // If we are running with earlier version, which did not have symbol versions,
2590 2590 // we should use the base version.
2591 2591 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2592 2592 void *f = dlvsym(handle, name, "libnuma_1.1");
2593 2593 if (f == NULL) {
2594 2594 f = dlsym(handle, name);
2595 2595 }
2596 2596 return f;
2597 2597 }
2598 2598
2599 2599 bool os::Linux::libnuma_init() {
2600 2600 // sched_getcpu() should be in libc.
2601 2601 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2602 2602 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2603 2603
2604 2604 if (sched_getcpu() != -1) { // Does it work?
2605 2605 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2606 2606 if (handle != NULL) {
2607 2607 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2608 2608 libnuma_dlsym(handle, "numa_node_to_cpus")));
2609 2609 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2610 2610 libnuma_dlsym(handle, "numa_max_node")));
2611 2611 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2612 2612 libnuma_dlsym(handle, "numa_available")));
2613 2613 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2614 2614 libnuma_dlsym(handle, "numa_tonode_memory")));
2615 2615 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2616 2616 libnuma_dlsym(handle, "numa_interleave_memory")));
2617 2617
2618 2618
2619 2619 if (numa_available() != -1) {
2620 2620 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2621 2621 // Create a cpu -> node mapping
2622 2622 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2623 2623 rebuild_cpu_to_node_map();
2624 2624 return true;
2625 2625 }
2626 2626 }
2627 2627 }
2628 2628 return false;
2629 2629 }
2630 2630
2631 2631 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2632 2632 // The table is later used in get_node_by_cpu().
2633 2633 void os::Linux::rebuild_cpu_to_node_map() {
2634 2634 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2635 2635 // in libnuma (possible values are starting from 16,
2636 2636 // and continuing up with every other power of 2, but less
2637 2637 // than the maximum number of CPUs supported by kernel), and
2638 2638 // is a subject to change (in libnuma version 2 the requirements
2639 2639 // are more reasonable) we'll just hardcode the number they use
2640 2640 // in the library.
2641 2641 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2642 2642
2643 2643 size_t cpu_num = os::active_processor_count();
2644 2644 size_t cpu_map_size = NCPUS / BitsPerCLong;
2645 2645 size_t cpu_map_valid_size =
2646 2646 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2647 2647
2648 2648 cpu_to_node()->clear();
2649 2649 cpu_to_node()->at_grow(cpu_num - 1);
2650 2650 size_t node_num = numa_get_groups_num();
2651 2651
2652 2652 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2653 2653 for (size_t i = 0; i < node_num; i++) {
2654 2654 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2655 2655 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2656 2656 if (cpu_map[j] != 0) {
2657 2657 for (size_t k = 0; k < BitsPerCLong; k++) {
2658 2658 if (cpu_map[j] & (1UL << k)) {
2659 2659 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2660 2660 }
2661 2661 }
2662 2662 }
2663 2663 }
2664 2664 }
2665 2665 }
2666 2666 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2667 2667 }
2668 2668
2669 2669 int os::Linux::get_node_by_cpu(int cpu_id) {
2670 2670 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2671 2671 return cpu_to_node()->at(cpu_id);
2672 2672 }
2673 2673 return -1;
2674 2674 }
2675 2675
2676 2676 GrowableArray<int>* os::Linux::_cpu_to_node;
2677 2677 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2678 2678 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2679 2679 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2680 2680 os::Linux::numa_available_func_t os::Linux::_numa_available;
2681 2681 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2682 2682 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2683 2683 unsigned long* os::Linux::_numa_all_nodes;
2684 2684
2685 2685 bool os::uncommit_memory(char* addr, size_t size) {
2686 2686 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2687 2687 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2688 2688 return res != (uintptr_t) MAP_FAILED;
2689 2689 }
2690 2690
2691 2691 // Linux uses a growable mapping for the stack, and if the mapping for
2692 2692 // the stack guard pages is not removed when we detach a thread the
2693 2693 // stack cannot grow beyond the pages where the stack guard was
2694 2694 // mapped. If at some point later in the process the stack expands to
2695 2695 // that point, the Linux kernel cannot expand the stack any further
2696 2696 // because the guard pages are in the way, and a segfault occurs.
2697 2697 //
2698 2698 // However, it's essential not to split the stack region by unmapping
2699 2699 // a region (leaving a hole) that's already part of the stack mapping,
2700 2700 // so if the stack mapping has already grown beyond the guard pages at
2701 2701 // the time we create them, we have to truncate the stack mapping.
2702 2702 // So, we need to know the extent of the stack mapping when
2703 2703 // create_stack_guard_pages() is called.
2704 2704
2705 2705 // Find the bounds of the stack mapping. Return true for success.
2706 2706 //
2707 2707 // We only need this for stacks that are growable: at the time of
2708 2708 // writing thread stacks don't use growable mappings (i.e. those
2709 2709 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2710 2710 // only applies to the main thread.
2711 2711
2712 2712 static
2713 2713 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2714 2714
2715 2715 char buf[128];
2716 2716 int fd, sz;
2717 2717
2718 2718 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2719 2719 return false;
2720 2720 }
2721 2721
2722 2722 const char kw[] = "[stack]";
2723 2723 const int kwlen = sizeof(kw)-1;
2724 2724
2725 2725 // Address part of /proc/self/maps couldn't be more than 128 bytes
2726 2726 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2727 2727 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2728 2728 // Extract addresses
2729 2729 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2730 2730 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2731 2731 if (sp >= *bottom && sp <= *top) {
2732 2732 ::close(fd);
2733 2733 return true;
2734 2734 }
2735 2735 }
2736 2736 }
2737 2737 }
2738 2738
2739 2739 ::close(fd);
2740 2740 return false;
2741 2741 }
2742 2742
2743 2743
2744 2744 // If the (growable) stack mapping already extends beyond the point
2745 2745 // where we're going to put our guard pages, truncate the mapping at
2746 2746 // that point by munmap()ping it. This ensures that when we later
2747 2747 // munmap() the guard pages we don't leave a hole in the stack
2748 2748 // mapping. This only affects the main/initial thread, but guard
2749 2749 // against future OS changes
2750 2750 bool os::create_stack_guard_pages(char* addr, size_t size) {
2751 2751 uintptr_t stack_extent, stack_base;
2752 2752 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2753 2753 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2754 2754 assert(os::Linux::is_initial_thread(),
2755 2755 "growable stack in non-initial thread");
2756 2756 if (stack_extent < (uintptr_t)addr)
2757 2757 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2758 2758 }
2759 2759
2760 2760 return os::commit_memory(addr, size);
2761 2761 }
2762 2762
2763 2763 // If this is a growable mapping, remove the guard pages entirely by
2764 2764 // munmap()ping them. If not, just call uncommit_memory(). This only
2765 2765 // affects the main/initial thread, but guard against future OS changes
2766 2766 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2767 2767 uintptr_t stack_extent, stack_base;
2768 2768 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2769 2769 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2770 2770 assert(os::Linux::is_initial_thread(),
2771 2771 "growable stack in non-initial thread");
2772 2772
2773 2773 return ::munmap(addr, size) == 0;
2774 2774 }
2775 2775
2776 2776 return os::uncommit_memory(addr, size);
2777 2777 }
2778 2778
2779 2779 static address _highest_vm_reserved_address = NULL;
2780 2780
2781 2781 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2782 2782 // at 'requested_addr'. If there are existing memory mappings at the same
2783 2783 // location, however, they will be overwritten. If 'fixed' is false,
2784 2784 // 'requested_addr' is only treated as a hint, the return value may or
2785 2785 // may not start from the requested address. Unlike Linux mmap(), this
2786 2786 // function returns NULL to indicate failure.
2787 2787 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2788 2788 char * addr;
2789 2789 int flags;
2790 2790
2791 2791 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2792 2792 if (fixed) {
2793 2793 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2794 2794 flags |= MAP_FIXED;
2795 2795 }
2796 2796
2797 2797 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2798 2798 // to PROT_EXEC if executable when we commit the page.
2799 2799 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2800 2800 flags, -1, 0);
2801 2801
2802 2802 if (addr != MAP_FAILED) {
2803 2803 // anon_mmap() should only get called during VM initialization,
2804 2804 // don't need lock (actually we can skip locking even it can be called
2805 2805 // from multiple threads, because _highest_vm_reserved_address is just a
2806 2806 // hint about the upper limit of non-stack memory regions.)
2807 2807 if ((address)addr + bytes > _highest_vm_reserved_address) {
2808 2808 _highest_vm_reserved_address = (address)addr + bytes;
2809 2809 }
2810 2810 }
2811 2811
2812 2812 return addr == MAP_FAILED ? NULL : addr;
2813 2813 }
2814 2814
2815 2815 // Don't update _highest_vm_reserved_address, because there might be memory
2816 2816 // regions above addr + size. If so, releasing a memory region only creates
2817 2817 // a hole in the address space, it doesn't help prevent heap-stack collision.
2818 2818 //
2819 2819 static int anon_munmap(char * addr, size_t size) {
2820 2820 return ::munmap(addr, size) == 0;
2821 2821 }
2822 2822
2823 2823 char* os::reserve_memory(size_t bytes, char* requested_addr,
2824 2824 size_t alignment_hint) {
2825 2825 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2826 2826 }
2827 2827
2828 2828 bool os::release_memory(char* addr, size_t size) {
2829 2829 return anon_munmap(addr, size);
2830 2830 }
2831 2831
2832 2832 static address highest_vm_reserved_address() {
2833 2833 return _highest_vm_reserved_address;
2834 2834 }
2835 2835
2836 2836 static bool linux_mprotect(char* addr, size_t size, int prot) {
2837 2837 // Linux wants the mprotect address argument to be page aligned.
2838 2838 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2839 2839
2840 2840 // According to SUSv3, mprotect() should only be used with mappings
2841 2841 // established by mmap(), and mmap() always maps whole pages. Unaligned
2842 2842 // 'addr' likely indicates problem in the VM (e.g. trying to change
2843 2843 // protection of malloc'ed or statically allocated memory). Check the
2844 2844 // caller if you hit this assert.
2845 2845 assert(addr == bottom, "sanity check");
2846 2846
2847 2847 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2848 2848 return ::mprotect(bottom, size, prot) == 0;
2849 2849 }
2850 2850
2851 2851 // Set protections specified
2852 2852 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2853 2853 bool is_committed) {
2854 2854 unsigned int p = 0;
2855 2855 switch (prot) {
2856 2856 case MEM_PROT_NONE: p = PROT_NONE; break;
2857 2857 case MEM_PROT_READ: p = PROT_READ; break;
2858 2858 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2859 2859 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2860 2860 default:
2861 2861 ShouldNotReachHere();
2862 2862 }
2863 2863 // is_committed is unused.
2864 2864 return linux_mprotect(addr, bytes, p);
2865 2865 }
2866 2866
2867 2867 bool os::guard_memory(char* addr, size_t size) {
2868 2868 return linux_mprotect(addr, size, PROT_NONE);
2869 2869 }
2870 2870
2871 2871 bool os::unguard_memory(char* addr, size_t size) {
2872 2872 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2873 2873 }
2874 2874
2875 2875 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2876 2876 bool result = false;
2877 2877 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2878 2878 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2879 2879 -1, 0);
2880 2880
2881 2881 if (p != (void *) -1) {
2882 2882 // We don't know if this really is a huge page or not.
2883 2883 FILE *fp = fopen("/proc/self/maps", "r");
2884 2884 if (fp) {
2885 2885 while (!feof(fp)) {
2886 2886 char chars[257];
2887 2887 long x = 0;
2888 2888 if (fgets(chars, sizeof(chars), fp)) {
2889 2889 if (sscanf(chars, "%lx-%*x", &x) == 1
2890 2890 && x == (long)p) {
2891 2891 if (strstr (chars, "hugepage")) {
2892 2892 result = true;
2893 2893 break;
2894 2894 }
2895 2895 }
2896 2896 }
2897 2897 }
2898 2898 fclose(fp);
2899 2899 }
2900 2900 munmap (p, page_size);
2901 2901 if (result)
2902 2902 return true;
2903 2903 }
2904 2904
2905 2905 if (warn) {
2906 2906 warning("HugeTLBFS is not supported by the operating system.");
2907 2907 }
2908 2908
2909 2909 return result;
2910 2910 }
2911 2911
2912 2912 /*
2913 2913 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2914 2914 *
2915 2915 * From the coredump_filter documentation:
2916 2916 *
2917 2917 * - (bit 0) anonymous private memory
2918 2918 * - (bit 1) anonymous shared memory
2919 2919 * - (bit 2) file-backed private memory
2920 2920 * - (bit 3) file-backed shared memory
2921 2921 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2922 2922 * effective only if the bit 2 is cleared)
2923 2923 * - (bit 5) hugetlb private memory
2924 2924 * - (bit 6) hugetlb shared memory
2925 2925 */
2926 2926 static void set_coredump_filter(void) {
2927 2927 FILE *f;
2928 2928 long cdm;
2929 2929
2930 2930 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2931 2931 return;
2932 2932 }
2933 2933
2934 2934 if (fscanf(f, "%lx", &cdm) != 1) {
2935 2935 fclose(f);
2936 2936 return;
2937 2937 }
2938 2938
2939 2939 rewind(f);
2940 2940
2941 2941 if ((cdm & LARGEPAGES_BIT) == 0) {
2942 2942 cdm |= LARGEPAGES_BIT;
2943 2943 fprintf(f, "%#lx", cdm);
2944 2944 }
2945 2945
2946 2946 fclose(f);
2947 2947 }
2948 2948
2949 2949 // Large page support
2950 2950
2951 2951 static size_t _large_page_size = 0;
2952 2952
2953 2953 void os::large_page_init() {
2954 2954 if (!UseLargePages) {
2955 2955 UseHugeTLBFS = false;
2956 2956 UseSHM = false;
2957 2957 return;
2958 2958 }
2959 2959
2960 2960 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2961 2961 // If UseLargePages is specified on the command line try both methods,
2962 2962 // if it's default, then try only HugeTLBFS.
2963 2963 if (FLAG_IS_DEFAULT(UseLargePages)) {
2964 2964 UseHugeTLBFS = true;
2965 2965 } else {
2966 2966 UseHugeTLBFS = UseSHM = true;
2967 2967 }
2968 2968 }
2969 2969
2970 2970 if (LargePageSizeInBytes) {
2971 2971 _large_page_size = LargePageSizeInBytes;
2972 2972 } else {
2973 2973 // large_page_size on Linux is used to round up heap size. x86 uses either
2974 2974 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2975 2975 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2976 2976 // page as large as 256M.
2977 2977 //
2978 2978 // Here we try to figure out page size by parsing /proc/meminfo and looking
2979 2979 // for a line with the following format:
2980 2980 // Hugepagesize: 2048 kB
2981 2981 //
2982 2982 // If we can't determine the value (e.g. /proc is not mounted, or the text
2983 2983 // format has been changed), we'll use the largest page size supported by
2984 2984 // the processor.
2985 2985
2986 2986 #ifndef ZERO
2987 2987 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2988 2988 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2989 2989 #endif // ZERO
2990 2990
2991 2991 FILE *fp = fopen("/proc/meminfo", "r");
2992 2992 if (fp) {
2993 2993 while (!feof(fp)) {
2994 2994 int x = 0;
2995 2995 char buf[16];
2996 2996 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2997 2997 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2998 2998 _large_page_size = x * K;
2999 2999 break;
3000 3000 }
3001 3001 } else {
3002 3002 // skip to next line
3003 3003 for (;;) {
3004 3004 int ch = fgetc(fp);
3005 3005 if (ch == EOF || ch == (int)'\n') break;
3006 3006 }
3007 3007 }
3008 3008 }
3009 3009 fclose(fp);
3010 3010 }
3011 3011 }
3012 3012
3013 3013 // print a warning if any large page related flag is specified on command line
3014 3014 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3015 3015
3016 3016 const size_t default_page_size = (size_t)Linux::page_size();
3017 3017 if (_large_page_size > default_page_size) {
3018 3018 _page_sizes[0] = _large_page_size;
3019 3019 _page_sizes[1] = default_page_size;
3020 3020 _page_sizes[2] = 0;
3021 3021 }
3022 3022 UseHugeTLBFS = UseHugeTLBFS &&
3023 3023 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3024 3024
3025 3025 if (UseHugeTLBFS)
3026 3026 UseSHM = false;
3027 3027
3028 3028 UseLargePages = UseHugeTLBFS || UseSHM;
3029 3029
3030 3030 set_coredump_filter();
3031 3031 }
3032 3032
3033 3033 #ifndef SHM_HUGETLB
3034 3034 #define SHM_HUGETLB 04000
3035 3035 #endif
3036 3036
3037 3037 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3038 3038 // "exec" is passed in but not used. Creating the shared image for
3039 3039 // the code cache doesn't have an SHM_X executable permission to check.
3040 3040 assert(UseLargePages && UseSHM, "only for SHM large pages");
3041 3041
3042 3042 key_t key = IPC_PRIVATE;
3043 3043 char *addr;
3044 3044
3045 3045 bool warn_on_failure = UseLargePages &&
3046 3046 (!FLAG_IS_DEFAULT(UseLargePages) ||
3047 3047 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3048 3048 );
3049 3049 char msg[128];
3050 3050
3051 3051 // Create a large shared memory region to attach to based on size.
3052 3052 // Currently, size is the total size of the heap
3053 3053 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3054 3054 if (shmid == -1) {
3055 3055 // Possible reasons for shmget failure:
3056 3056 // 1. shmmax is too small for Java heap.
3057 3057 // > check shmmax value: cat /proc/sys/kernel/shmmax
3058 3058 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3059 3059 // 2. not enough large page memory.
3060 3060 // > check available large pages: cat /proc/meminfo
3061 3061 // > increase amount of large pages:
3062 3062 // echo new_value > /proc/sys/vm/nr_hugepages
3063 3063 // Note 1: different Linux may use different name for this property,
3064 3064 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3065 3065 // Note 2: it's possible there's enough physical memory available but
3066 3066 // they are so fragmented after a long run that they can't
3067 3067 // coalesce into large pages. Try to reserve large pages when
3068 3068 // the system is still "fresh".
3069 3069 if (warn_on_failure) {
3070 3070 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3071 3071 warning(msg);
3072 3072 }
3073 3073 return NULL;
3074 3074 }
3075 3075
3076 3076 // attach to the region
3077 3077 addr = (char*)shmat(shmid, req_addr, 0);
3078 3078 int err = errno;
3079 3079
3080 3080 // Remove shmid. If shmat() is successful, the actual shared memory segment
3081 3081 // will be deleted when it's detached by shmdt() or when the process
3082 3082 // terminates. If shmat() is not successful this will remove the shared
3083 3083 // segment immediately.
3084 3084 shmctl(shmid, IPC_RMID, NULL);
3085 3085
3086 3086 if ((intptr_t)addr == -1) {
3087 3087 if (warn_on_failure) {
3088 3088 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3089 3089 warning(msg);
3090 3090 }
3091 3091 return NULL;
3092 3092 }
3093 3093
3094 3094 return addr;
3095 3095 }
3096 3096
3097 3097 bool os::release_memory_special(char* base, size_t bytes) {
3098 3098 // detaching the SHM segment will also delete it, see reserve_memory_special()
3099 3099 int rslt = shmdt(base);
3100 3100 return rslt == 0;
3101 3101 }
3102 3102
3103 3103 size_t os::large_page_size() {
3104 3104 return _large_page_size;
3105 3105 }
3106 3106
3107 3107 // HugeTLBFS allows application to commit large page memory on demand;
3108 3108 // with SysV SHM the entire memory region must be allocated as shared
3109 3109 // memory.
3110 3110 bool os::can_commit_large_page_memory() {
3111 3111 return UseHugeTLBFS;
3112 3112 }
3113 3113
3114 3114 bool os::can_execute_large_page_memory() {
3115 3115 return UseHugeTLBFS;
3116 3116 }
3117 3117
3118 3118 // Reserve memory at an arbitrary address, only if that area is
3119 3119 // available (and not reserved for something else).
3120 3120
3121 3121 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3122 3122 const int max_tries = 10;
3123 3123 char* base[max_tries];
3124 3124 size_t size[max_tries];
3125 3125 const size_t gap = 0x000000;
3126 3126
3127 3127 // Assert only that the size is a multiple of the page size, since
3128 3128 // that's all that mmap requires, and since that's all we really know
3129 3129 // about at this low abstraction level. If we need higher alignment,
3130 3130 // we can either pass an alignment to this method or verify alignment
3131 3131 // in one of the methods further up the call chain. See bug 5044738.
3132 3132 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3133 3133
3134 3134 // Repeatedly allocate blocks until the block is allocated at the
3135 3135 // right spot. Give up after max_tries. Note that reserve_memory() will
3136 3136 // automatically update _highest_vm_reserved_address if the call is
3137 3137 // successful. The variable tracks the highest memory address every reserved
3138 3138 // by JVM. It is used to detect heap-stack collision if running with
3139 3139 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3140 3140 // space than needed, it could confuse the collision detecting code. To
3141 3141 // solve the problem, save current _highest_vm_reserved_address and
3142 3142 // calculate the correct value before return.
3143 3143 address old_highest = _highest_vm_reserved_address;
3144 3144
3145 3145 // Linux mmap allows caller to pass an address as hint; give it a try first,
3146 3146 // if kernel honors the hint then we can return immediately.
3147 3147 char * addr = anon_mmap(requested_addr, bytes, false);
3148 3148 if (addr == requested_addr) {
3149 3149 return requested_addr;
3150 3150 }
3151 3151
3152 3152 if (addr != NULL) {
3153 3153 // mmap() is successful but it fails to reserve at the requested address
3154 3154 anon_munmap(addr, bytes);
3155 3155 }
3156 3156
3157 3157 int i;
3158 3158 for (i = 0; i < max_tries; ++i) {
3159 3159 base[i] = reserve_memory(bytes);
3160 3160
3161 3161 if (base[i] != NULL) {
3162 3162 // Is this the block we wanted?
3163 3163 if (base[i] == requested_addr) {
3164 3164 size[i] = bytes;
3165 3165 break;
3166 3166 }
3167 3167
3168 3168 // Does this overlap the block we wanted? Give back the overlapped
3169 3169 // parts and try again.
3170 3170
3171 3171 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3172 3172 if (top_overlap >= 0 && top_overlap < bytes) {
3173 3173 unmap_memory(base[i], top_overlap);
3174 3174 base[i] += top_overlap;
3175 3175 size[i] = bytes - top_overlap;
3176 3176 } else {
3177 3177 size_t bottom_overlap = base[i] + bytes - requested_addr;
3178 3178 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3179 3179 unmap_memory(requested_addr, bottom_overlap);
3180 3180 size[i] = bytes - bottom_overlap;
3181 3181 } else {
3182 3182 size[i] = bytes;
3183 3183 }
3184 3184 }
3185 3185 }
3186 3186 }
3187 3187
3188 3188 // Give back the unused reserved pieces.
3189 3189
3190 3190 for (int j = 0; j < i; ++j) {
3191 3191 if (base[j] != NULL) {
3192 3192 unmap_memory(base[j], size[j]);
3193 3193 }
3194 3194 }
3195 3195
3196 3196 if (i < max_tries) {
3197 3197 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3198 3198 return requested_addr;
3199 3199 } else {
3200 3200 _highest_vm_reserved_address = old_highest;
3201 3201 return NULL;
3202 3202 }
3203 3203 }
3204 3204
3205 3205 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3206 3206 return ::read(fd, buf, nBytes);
3207 3207 }
3208 3208
3209 3209 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3210 3210 // Solaris uses poll(), linux uses park().
3211 3211 // Poll() is likely a better choice, assuming that Thread.interrupt()
3212 3212 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3213 3213 // SIGSEGV, see 4355769.
3214 3214
3215 3215 const int NANOSECS_PER_MILLISECS = 1000000;
3216 3216
3217 3217 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3218 3218 assert(thread == Thread::current(), "thread consistency check");
3219 3219
3220 3220 ParkEvent * const slp = thread->_SleepEvent ;
3221 3221 slp->reset() ;
3222 3222 OrderAccess::fence() ;
3223 3223
3224 3224 if (interruptible) {
3225 3225 jlong prevtime = javaTimeNanos();
3226 3226
3227 3227 for (;;) {
3228 3228 if (os::is_interrupted(thread, true)) {
3229 3229 return OS_INTRPT;
3230 3230 }
3231 3231
3232 3232 jlong newtime = javaTimeNanos();
3233 3233
3234 3234 if (newtime - prevtime < 0) {
3235 3235 // time moving backwards, should only happen if no monotonic clock
3236 3236 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3237 3237 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3238 3238 } else {
3239 3239 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3240 3240 }
3241 3241
3242 3242 if(millis <= 0) {
3243 3243 return OS_OK;
3244 3244 }
3245 3245
3246 3246 prevtime = newtime;
3247 3247
3248 3248 {
3249 3249 assert(thread->is_Java_thread(), "sanity check");
3250 3250 JavaThread *jt = (JavaThread *) thread;
3251 3251 ThreadBlockInVM tbivm(jt);
3252 3252 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3253 3253
3254 3254 jt->set_suspend_equivalent();
3255 3255 // cleared by handle_special_suspend_equivalent_condition() or
3256 3256 // java_suspend_self() via check_and_wait_while_suspended()
3257 3257
3258 3258 slp->park(millis);
3259 3259
3260 3260 // were we externally suspended while we were waiting?
3261 3261 jt->check_and_wait_while_suspended();
3262 3262 }
3263 3263 }
3264 3264 } else {
3265 3265 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3266 3266 jlong prevtime = javaTimeNanos();
3267 3267
3268 3268 for (;;) {
3269 3269 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3270 3270 // the 1st iteration ...
3271 3271 jlong newtime = javaTimeNanos();
3272 3272
3273 3273 if (newtime - prevtime < 0) {
3274 3274 // time moving backwards, should only happen if no monotonic clock
3275 3275 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3276 3276 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3277 3277 } else {
3278 3278 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3279 3279 }
3280 3280
3281 3281 if(millis <= 0) break ;
3282 3282
3283 3283 prevtime = newtime;
3284 3284 slp->park(millis);
3285 3285 }
3286 3286 return OS_OK ;
3287 3287 }
3288 3288 }
3289 3289
3290 3290 int os::naked_sleep() {
3291 3291 // %% make the sleep time an integer flag. for now use 1 millisec.
3292 3292 return os::sleep(Thread::current(), 1, false);
3293 3293 }
3294 3294
3295 3295 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3296 3296 void os::infinite_sleep() {
3297 3297 while (true) { // sleep forever ...
3298 3298 ::sleep(100); // ... 100 seconds at a time
3299 3299 }
3300 3300 }
3301 3301
3302 3302 // Used to convert frequent JVM_Yield() to nops
3303 3303 bool os::dont_yield() {
3304 3304 return DontYieldALot;
3305 3305 }
3306 3306
3307 3307 void os::yield() {
3308 3308 sched_yield();
3309 3309 }
3310 3310
3311 3311 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3312 3312
3313 3313 void os::yield_all(int attempts) {
3314 3314 // Yields to all threads, including threads with lower priorities
3315 3315 // Threads on Linux are all with same priority. The Solaris style
3316 3316 // os::yield_all() with nanosleep(1ms) is not necessary.
3317 3317 sched_yield();
3318 3318 }
3319 3319
3320 3320 // Called from the tight loops to possibly influence time-sharing heuristics
3321 3321 void os::loop_breaker(int attempts) {
3322 3322 os::yield_all(attempts);
3323 3323 }
3324 3324
3325 3325 ////////////////////////////////////////////////////////////////////////////////
3326 3326 // thread priority support
3327 3327
3328 3328 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3329 3329 // only supports dynamic priority, static priority must be zero. For real-time
3330 3330 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3331 3331 // However, for large multi-threaded applications, SCHED_RR is not only slower
3332 3332 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3333 3333 // of 5 runs - Sep 2005).
3334 3334 //
3335 3335 // The following code actually changes the niceness of kernel-thread/LWP. It
3336 3336 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3337 3337 // not the entire user process, and user level threads are 1:1 mapped to kernel
3338 3338 // threads. It has always been the case, but could change in the future. For
3339 3339 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3340 3340 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3341 3341
3342 3342 int os::java_to_os_priority[MaxPriority + 1] = {
3343 3343 19, // 0 Entry should never be used
3344 3344
3345 3345 4, // 1 MinPriority
3346 3346 3, // 2
3347 3347 2, // 3
3348 3348
3349 3349 1, // 4
3350 3350 0, // 5 NormPriority
3351 3351 -1, // 6
3352 3352
3353 3353 -2, // 7
3354 3354 -3, // 8
3355 3355 -4, // 9 NearMaxPriority
3356 3356
3357 3357 -5 // 10 MaxPriority
3358 3358 };
3359 3359
3360 3360 static int prio_init() {
3361 3361 if (ThreadPriorityPolicy == 1) {
3362 3362 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3363 3363 // if effective uid is not root. Perhaps, a more elegant way of doing
3364 3364 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3365 3365 if (geteuid() != 0) {
3366 3366 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3367 3367 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3368 3368 }
3369 3369 ThreadPriorityPolicy = 0;
3370 3370 }
3371 3371 }
3372 3372 return 0;
3373 3373 }
3374 3374
3375 3375 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3376 3376 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3377 3377
3378 3378 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3379 3379 return (ret == 0) ? OS_OK : OS_ERR;
3380 3380 }
3381 3381
3382 3382 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3383 3383 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3384 3384 *priority_ptr = java_to_os_priority[NormPriority];
3385 3385 return OS_OK;
3386 3386 }
3387 3387
3388 3388 errno = 0;
3389 3389 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3390 3390 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3391 3391 }
3392 3392
3393 3393 // Hint to the underlying OS that a task switch would not be good.
3394 3394 // Void return because it's a hint and can fail.
3395 3395 void os::hint_no_preempt() {}
3396 3396
3397 3397 ////////////////////////////////////////////////////////////////////////////////
3398 3398 // suspend/resume support
3399 3399
3400 3400 // the low-level signal-based suspend/resume support is a remnant from the
3401 3401 // old VM-suspension that used to be for java-suspension, safepoints etc,
3402 3402 // within hotspot. Now there is a single use-case for this:
3403 3403 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3404 3404 // that runs in the watcher thread.
3405 3405 // The remaining code is greatly simplified from the more general suspension
3406 3406 // code that used to be used.
3407 3407 //
3408 3408 // The protocol is quite simple:
3409 3409 // - suspend:
3410 3410 // - sends a signal to the target thread
3411 3411 // - polls the suspend state of the osthread using a yield loop
3412 3412 // - target thread signal handler (SR_handler) sets suspend state
3413 3413 // and blocks in sigsuspend until continued
3414 3414 // - resume:
3415 3415 // - sets target osthread state to continue
3416 3416 // - sends signal to end the sigsuspend loop in the SR_handler
3417 3417 //
3418 3418 // Note that the SR_lock plays no role in this suspend/resume protocol.
3419 3419 //
3420 3420
3421 3421 static void resume_clear_context(OSThread *osthread) {
3422 3422 osthread->set_ucontext(NULL);
3423 3423 osthread->set_siginfo(NULL);
3424 3424
3425 3425 // notify the suspend action is completed, we have now resumed
3426 3426 osthread->sr.clear_suspended();
3427 3427 }
3428 3428
3429 3429 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3430 3430 osthread->set_ucontext(context);
3431 3431 osthread->set_siginfo(siginfo);
3432 3432 }
3433 3433
3434 3434 //
3435 3435 // Handler function invoked when a thread's execution is suspended or
3436 3436 // resumed. We have to be careful that only async-safe functions are
3437 3437 // called here (Note: most pthread functions are not async safe and
3438 3438 // should be avoided.)
3439 3439 //
3440 3440 // Note: sigwait() is a more natural fit than sigsuspend() from an
3441 3441 // interface point of view, but sigwait() prevents the signal hander
3442 3442 // from being run. libpthread would get very confused by not having
3443 3443 // its signal handlers run and prevents sigwait()'s use with the
3444 3444 // mutex granting granting signal.
3445 3445 //
3446 3446 // Currently only ever called on the VMThread
3447 3447 //
3448 3448 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3449 3449 // Save and restore errno to avoid confusing native code with EINTR
3450 3450 // after sigsuspend.
3451 3451 int old_errno = errno;
3452 3452
3453 3453 Thread* thread = Thread::current();
3454 3454 OSThread* osthread = thread->osthread();
3455 3455 assert(thread->is_VM_thread(), "Must be VMThread");
3456 3456 // read current suspend action
3457 3457 int action = osthread->sr.suspend_action();
3458 3458 if (action == SR_SUSPEND) {
3459 3459 suspend_save_context(osthread, siginfo, context);
3460 3460
3461 3461 // Notify the suspend action is about to be completed. do_suspend()
3462 3462 // waits until SR_SUSPENDED is set and then returns. We will wait
3463 3463 // here for a resume signal and that completes the suspend-other
3464 3464 // action. do_suspend/do_resume is always called as a pair from
3465 3465 // the same thread - so there are no races
3466 3466
3467 3467 // notify the caller
3468 3468 osthread->sr.set_suspended();
3469 3469
3470 3470 sigset_t suspend_set; // signals for sigsuspend()
3471 3471
3472 3472 // get current set of blocked signals and unblock resume signal
3473 3473 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3474 3474 sigdelset(&suspend_set, SR_signum);
3475 3475
3476 3476 // wait here until we are resumed
3477 3477 do {
3478 3478 sigsuspend(&suspend_set);
3479 3479 // ignore all returns until we get a resume signal
3480 3480 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3481 3481
3482 3482 resume_clear_context(osthread);
3483 3483
3484 3484 } else {
3485 3485 assert(action == SR_CONTINUE, "unexpected sr action");
3486 3486 // nothing special to do - just leave the handler
3487 3487 }
3488 3488
3489 3489 errno = old_errno;
3490 3490 }
3491 3491
3492 3492
3493 3493 static int SR_initialize() {
3494 3494 struct sigaction act;
3495 3495 char *s;
3496 3496 /* Get signal number to use for suspend/resume */
3497 3497 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3498 3498 int sig = ::strtol(s, 0, 10);
3499 3499 if (sig > 0 || sig < _NSIG) {
3500 3500 SR_signum = sig;
3501 3501 }
3502 3502 }
3503 3503
3504 3504 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3505 3505 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3506 3506
3507 3507 sigemptyset(&SR_sigset);
3508 3508 sigaddset(&SR_sigset, SR_signum);
3509 3509
3510 3510 /* Set up signal handler for suspend/resume */
3511 3511 act.sa_flags = SA_RESTART|SA_SIGINFO;
3512 3512 act.sa_handler = (void (*)(int)) SR_handler;
3513 3513
3514 3514 // SR_signum is blocked by default.
3515 3515 // 4528190 - We also need to block pthread restart signal (32 on all
3516 3516 // supported Linux platforms). Note that LinuxThreads need to block
3517 3517 // this signal for all threads to work properly. So we don't have
3518 3518 // to use hard-coded signal number when setting up the mask.
3519 3519 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3520 3520
3521 3521 if (sigaction(SR_signum, &act, 0) == -1) {
3522 3522 return -1;
3523 3523 }
3524 3524
3525 3525 // Save signal flag
3526 3526 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3527 3527 return 0;
3528 3528 }
3529 3529
3530 3530 static int SR_finalize() {
3531 3531 return 0;
3532 3532 }
3533 3533
3534 3534
3535 3535 // returns true on success and false on error - really an error is fatal
3536 3536 // but this seems the normal response to library errors
3537 3537 static bool do_suspend(OSThread* osthread) {
3538 3538 // mark as suspended and send signal
3539 3539 osthread->sr.set_suspend_action(SR_SUSPEND);
3540 3540 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3541 3541 assert_status(status == 0, status, "pthread_kill");
3542 3542
3543 3543 // check status and wait until notified of suspension
3544 3544 if (status == 0) {
3545 3545 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3546 3546 os::yield_all(i);
3547 3547 }
3548 3548 osthread->sr.set_suspend_action(SR_NONE);
3549 3549 return true;
3550 3550 }
3551 3551 else {
3552 3552 osthread->sr.set_suspend_action(SR_NONE);
3553 3553 return false;
3554 3554 }
3555 3555 }
3556 3556
3557 3557 static void do_resume(OSThread* osthread) {
3558 3558 assert(osthread->sr.is_suspended(), "thread should be suspended");
3559 3559 osthread->sr.set_suspend_action(SR_CONTINUE);
3560 3560
3561 3561 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3562 3562 assert_status(status == 0, status, "pthread_kill");
3563 3563 // check status and wait unit notified of resumption
3564 3564 if (status == 0) {
3565 3565 for (int i = 0; osthread->sr.is_suspended(); i++) {
3566 3566 os::yield_all(i);
3567 3567 }
3568 3568 }
3569 3569 osthread->sr.set_suspend_action(SR_NONE);
3570 3570 }
3571 3571
3572 3572 ////////////////////////////////////////////////////////////////////////////////
3573 3573 // interrupt support
3574 3574
3575 3575 void os::interrupt(Thread* thread) {
3576 3576 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3577 3577 "possibility of dangling Thread pointer");
3578 3578
3579 3579 OSThread* osthread = thread->osthread();
3580 3580
3581 3581 if (!osthread->interrupted()) {
3582 3582 osthread->set_interrupted(true);
3583 3583 // More than one thread can get here with the same value of osthread,
3584 3584 // resulting in multiple notifications. We do, however, want the store
3585 3585 // to interrupted() to be visible to other threads before we execute unpark().
3586 3586 OrderAccess::fence();
3587 3587 ParkEvent * const slp = thread->_SleepEvent ;
3588 3588 if (slp != NULL) slp->unpark() ;
3589 3589 }
3590 3590
3591 3591 // For JSR166. Unpark even if interrupt status already was set
3592 3592 if (thread->is_Java_thread())
3593 3593 ((JavaThread*)thread)->parker()->unpark();
3594 3594
3595 3595 ParkEvent * ev = thread->_ParkEvent ;
3596 3596 if (ev != NULL) ev->unpark() ;
3597 3597
3598 3598 }
3599 3599
3600 3600 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3601 3601 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3602 3602 "possibility of dangling Thread pointer");
3603 3603
3604 3604 OSThread* osthread = thread->osthread();
3605 3605
3606 3606 bool interrupted = osthread->interrupted();
3607 3607
3608 3608 if (interrupted && clear_interrupted) {
3609 3609 osthread->set_interrupted(false);
3610 3610 // consider thread->_SleepEvent->reset() ... optional optimization
3611 3611 }
3612 3612
3613 3613 return interrupted;
3614 3614 }
3615 3615
3616 3616 ///////////////////////////////////////////////////////////////////////////////////
3617 3617 // signal handling (except suspend/resume)
3618 3618
3619 3619 // This routine may be used by user applications as a "hook" to catch signals.
3620 3620 // The user-defined signal handler must pass unrecognized signals to this
3621 3621 // routine, and if it returns true (non-zero), then the signal handler must
3622 3622 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3623 3623 // routine will never retun false (zero), but instead will execute a VM panic
3624 3624 // routine kill the process.
3625 3625 //
3626 3626 // If this routine returns false, it is OK to call it again. This allows
3627 3627 // the user-defined signal handler to perform checks either before or after
3628 3628 // the VM performs its own checks. Naturally, the user code would be making
3629 3629 // a serious error if it tried to handle an exception (such as a null check
3630 3630 // or breakpoint) that the VM was generating for its own correct operation.
3631 3631 //
3632 3632 // This routine may recognize any of the following kinds of signals:
3633 3633 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3634 3634 // It should be consulted by handlers for any of those signals.
3635 3635 //
3636 3636 // The caller of this routine must pass in the three arguments supplied
3637 3637 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3638 3638 // field of the structure passed to sigaction(). This routine assumes that
3639 3639 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3640 3640 //
3641 3641 // Note that the VM will print warnings if it detects conflicting signal
3642 3642 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3643 3643 //
3644 3644 extern "C" JNIEXPORT int
3645 3645 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3646 3646 void* ucontext, int abort_if_unrecognized);
3647 3647
3648 3648 void signalHandler(int sig, siginfo_t* info, void* uc) {
3649 3649 assert(info != NULL && uc != NULL, "it must be old kernel");
3650 3650 JVM_handle_linux_signal(sig, info, uc, true);
3651 3651 }
3652 3652
3653 3653
3654 3654 // This boolean allows users to forward their own non-matching signals
3655 3655 // to JVM_handle_linux_signal, harmlessly.
3656 3656 bool os::Linux::signal_handlers_are_installed = false;
3657 3657
3658 3658 // For signal-chaining
3659 3659 struct sigaction os::Linux::sigact[MAXSIGNUM];
3660 3660 unsigned int os::Linux::sigs = 0;
3661 3661 bool os::Linux::libjsig_is_loaded = false;
3662 3662 typedef struct sigaction *(*get_signal_t)(int);
3663 3663 get_signal_t os::Linux::get_signal_action = NULL;
3664 3664
3665 3665 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3666 3666 struct sigaction *actp = NULL;
3667 3667
3668 3668 if (libjsig_is_loaded) {
3669 3669 // Retrieve the old signal handler from libjsig
3670 3670 actp = (*get_signal_action)(sig);
3671 3671 }
3672 3672 if (actp == NULL) {
3673 3673 // Retrieve the preinstalled signal handler from jvm
3674 3674 actp = get_preinstalled_handler(sig);
3675 3675 }
3676 3676
3677 3677 return actp;
3678 3678 }
3679 3679
3680 3680 static bool call_chained_handler(struct sigaction *actp, int sig,
3681 3681 siginfo_t *siginfo, void *context) {
3682 3682 // Call the old signal handler
3683 3683 if (actp->sa_handler == SIG_DFL) {
3684 3684 // It's more reasonable to let jvm treat it as an unexpected exception
3685 3685 // instead of taking the default action.
3686 3686 return false;
3687 3687 } else if (actp->sa_handler != SIG_IGN) {
3688 3688 if ((actp->sa_flags & SA_NODEFER) == 0) {
3689 3689 // automaticlly block the signal
3690 3690 sigaddset(&(actp->sa_mask), sig);
3691 3691 }
3692 3692
3693 3693 sa_handler_t hand;
3694 3694 sa_sigaction_t sa;
3695 3695 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3696 3696 // retrieve the chained handler
3697 3697 if (siginfo_flag_set) {
3698 3698 sa = actp->sa_sigaction;
3699 3699 } else {
3700 3700 hand = actp->sa_handler;
3701 3701 }
3702 3702
3703 3703 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3704 3704 actp->sa_handler = SIG_DFL;
3705 3705 }
3706 3706
3707 3707 // try to honor the signal mask
3708 3708 sigset_t oset;
3709 3709 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3710 3710
3711 3711 // call into the chained handler
3712 3712 if (siginfo_flag_set) {
3713 3713 (*sa)(sig, siginfo, context);
3714 3714 } else {
3715 3715 (*hand)(sig);
3716 3716 }
3717 3717
3718 3718 // restore the signal mask
3719 3719 pthread_sigmask(SIG_SETMASK, &oset, 0);
3720 3720 }
3721 3721 // Tell jvm's signal handler the signal is taken care of.
3722 3722 return true;
3723 3723 }
3724 3724
3725 3725 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3726 3726 bool chained = false;
3727 3727 // signal-chaining
3728 3728 if (UseSignalChaining) {
3729 3729 struct sigaction *actp = get_chained_signal_action(sig);
3730 3730 if (actp != NULL) {
3731 3731 chained = call_chained_handler(actp, sig, siginfo, context);
3732 3732 }
3733 3733 }
3734 3734 return chained;
3735 3735 }
3736 3736
3737 3737 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3738 3738 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3739 3739 return &sigact[sig];
3740 3740 }
3741 3741 return NULL;
3742 3742 }
3743 3743
3744 3744 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3745 3745 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3746 3746 sigact[sig] = oldAct;
3747 3747 sigs |= (unsigned int)1 << sig;
3748 3748 }
3749 3749
3750 3750 // for diagnostic
3751 3751 int os::Linux::sigflags[MAXSIGNUM];
3752 3752
3753 3753 int os::Linux::get_our_sigflags(int sig) {
3754 3754 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3755 3755 return sigflags[sig];
3756 3756 }
3757 3757
3758 3758 void os::Linux::set_our_sigflags(int sig, int flags) {
3759 3759 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3760 3760 sigflags[sig] = flags;
3761 3761 }
3762 3762
3763 3763 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3764 3764 // Check for overwrite.
3765 3765 struct sigaction oldAct;
3766 3766 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3767 3767
3768 3768 void* oldhand = oldAct.sa_sigaction
3769 3769 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3770 3770 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3771 3771 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3772 3772 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3773 3773 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3774 3774 if (AllowUserSignalHandlers || !set_installed) {
3775 3775 // Do not overwrite; user takes responsibility to forward to us.
3776 3776 return;
3777 3777 } else if (UseSignalChaining) {
3778 3778 // save the old handler in jvm
3779 3779 save_preinstalled_handler(sig, oldAct);
3780 3780 // libjsig also interposes the sigaction() call below and saves the
3781 3781 // old sigaction on it own.
3782 3782 } else {
3783 3783 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3784 3784 "%#lx for signal %d.", (long)oldhand, sig));
3785 3785 }
3786 3786 }
3787 3787
3788 3788 struct sigaction sigAct;
3789 3789 sigfillset(&(sigAct.sa_mask));
3790 3790 sigAct.sa_handler = SIG_DFL;
3791 3791 if (!set_installed) {
3792 3792 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3793 3793 } else {
3794 3794 sigAct.sa_sigaction = signalHandler;
3795 3795 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3796 3796 }
3797 3797 // Save flags, which are set by ours
3798 3798 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3799 3799 sigflags[sig] = sigAct.sa_flags;
3800 3800
3801 3801 int ret = sigaction(sig, &sigAct, &oldAct);
3802 3802 assert(ret == 0, "check");
3803 3803
3804 3804 void* oldhand2 = oldAct.sa_sigaction
3805 3805 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3806 3806 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3807 3807 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3808 3808 }
3809 3809
3810 3810 // install signal handlers for signals that HotSpot needs to
3811 3811 // handle in order to support Java-level exception handling.
3812 3812
3813 3813 void os::Linux::install_signal_handlers() {
3814 3814 if (!signal_handlers_are_installed) {
3815 3815 signal_handlers_are_installed = true;
3816 3816
3817 3817 // signal-chaining
3818 3818 typedef void (*signal_setting_t)();
3819 3819 signal_setting_t begin_signal_setting = NULL;
3820 3820 signal_setting_t end_signal_setting = NULL;
3821 3821 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3822 3822 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3823 3823 if (begin_signal_setting != NULL) {
3824 3824 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3825 3825 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3826 3826 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3827 3827 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3828 3828 libjsig_is_loaded = true;
3829 3829 assert(UseSignalChaining, "should enable signal-chaining");
3830 3830 }
3831 3831 if (libjsig_is_loaded) {
3832 3832 // Tell libjsig jvm is setting signal handlers
3833 3833 (*begin_signal_setting)();
3834 3834 }
3835 3835
3836 3836 set_signal_handler(SIGSEGV, true);
3837 3837 set_signal_handler(SIGPIPE, true);
3838 3838 set_signal_handler(SIGBUS, true);
3839 3839 set_signal_handler(SIGILL, true);
3840 3840 set_signal_handler(SIGFPE, true);
3841 3841 set_signal_handler(SIGXFSZ, true);
3842 3842
3843 3843 if (libjsig_is_loaded) {
3844 3844 // Tell libjsig jvm finishes setting signal handlers
3845 3845 (*end_signal_setting)();
3846 3846 }
3847 3847
3848 3848 // We don't activate signal checker if libjsig is in place, we trust ourselves
3849 3849 // and if UserSignalHandler is installed all bets are off
3850 3850 if (CheckJNICalls) {
3851 3851 if (libjsig_is_loaded) {
3852 3852 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3853 3853 check_signals = false;
3854 3854 }
3855 3855 if (AllowUserSignalHandlers) {
3856 3856 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3857 3857 check_signals = false;
3858 3858 }
3859 3859 }
3860 3860 }
3861 3861 }
3862 3862
3863 3863 // This is the fastest way to get thread cpu time on Linux.
3864 3864 // Returns cpu time (user+sys) for any thread, not only for current.
3865 3865 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3866 3866 // It might work on 2.6.10+ with a special kernel/glibc patch.
3867 3867 // For reference, please, see IEEE Std 1003.1-2004:
3868 3868 // http://www.unix.org/single_unix_specification
3869 3869
3870 3870 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3871 3871 struct timespec tp;
3872 3872 int rc = os::Linux::clock_gettime(clockid, &tp);
3873 3873 assert(rc == 0, "clock_gettime is expected to return 0 code");
3874 3874
3875 3875 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3876 3876 }
3877 3877
3878 3878 /////
3879 3879 // glibc on Linux platform uses non-documented flag
3880 3880 // to indicate, that some special sort of signal
3881 3881 // trampoline is used.
3882 3882 // We will never set this flag, and we should
3883 3883 // ignore this flag in our diagnostic
3884 3884 #ifdef SIGNIFICANT_SIGNAL_MASK
3885 3885 #undef SIGNIFICANT_SIGNAL_MASK
3886 3886 #endif
3887 3887 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3888 3888
3889 3889 static const char* get_signal_handler_name(address handler,
3890 3890 char* buf, int buflen) {
3891 3891 int offset;
3892 3892 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3893 3893 if (found) {
3894 3894 // skip directory names
3895 3895 const char *p1, *p2;
3896 3896 p1 = buf;
3897 3897 size_t len = strlen(os::file_separator());
3898 3898 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3899 3899 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3900 3900 } else {
3901 3901 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3902 3902 }
3903 3903 return buf;
3904 3904 }
3905 3905
3906 3906 static void print_signal_handler(outputStream* st, int sig,
3907 3907 char* buf, size_t buflen) {
3908 3908 struct sigaction sa;
3909 3909
3910 3910 sigaction(sig, NULL, &sa);
3911 3911
3912 3912 // See comment for SIGNIFICANT_SIGNAL_MASK define
3913 3913 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3914 3914
3915 3915 st->print("%s: ", os::exception_name(sig, buf, buflen));
3916 3916
3917 3917 address handler = (sa.sa_flags & SA_SIGINFO)
3918 3918 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3919 3919 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3920 3920
3921 3921 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3922 3922 st->print("SIG_DFL");
3923 3923 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3924 3924 st->print("SIG_IGN");
3925 3925 } else {
3926 3926 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3927 3927 }
3928 3928
3929 3929 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3930 3930
3931 3931 address rh = VMError::get_resetted_sighandler(sig);
3932 3932 // May be, handler was resetted by VMError?
3933 3933 if(rh != NULL) {
3934 3934 handler = rh;
3935 3935 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3936 3936 }
3937 3937
3938 3938 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3939 3939
3940 3940 // Check: is it our handler?
3941 3941 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3942 3942 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3943 3943 // It is our signal handler
3944 3944 // check for flags, reset system-used one!
3945 3945 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3946 3946 st->print(
3947 3947 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3948 3948 os::Linux::get_our_sigflags(sig));
3949 3949 }
3950 3950 }
3951 3951 st->cr();
3952 3952 }
3953 3953
3954 3954
3955 3955 #define DO_SIGNAL_CHECK(sig) \
3956 3956 if (!sigismember(&check_signal_done, sig)) \
3957 3957 os::Linux::check_signal_handler(sig)
3958 3958
3959 3959 // This method is a periodic task to check for misbehaving JNI applications
3960 3960 // under CheckJNI, we can add any periodic checks here
3961 3961
3962 3962 void os::run_periodic_checks() {
3963 3963
3964 3964 if (check_signals == false) return;
3965 3965
3966 3966 // SEGV and BUS if overridden could potentially prevent
3967 3967 // generation of hs*.log in the event of a crash, debugging
3968 3968 // such a case can be very challenging, so we absolutely
3969 3969 // check the following for a good measure:
3970 3970 DO_SIGNAL_CHECK(SIGSEGV);
3971 3971 DO_SIGNAL_CHECK(SIGILL);
3972 3972 DO_SIGNAL_CHECK(SIGFPE);
3973 3973 DO_SIGNAL_CHECK(SIGBUS);
3974 3974 DO_SIGNAL_CHECK(SIGPIPE);
3975 3975 DO_SIGNAL_CHECK(SIGXFSZ);
3976 3976
3977 3977
3978 3978 // ReduceSignalUsage allows the user to override these handlers
3979 3979 // see comments at the very top and jvm_solaris.h
3980 3980 if (!ReduceSignalUsage) {
3981 3981 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3982 3982 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3983 3983 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3984 3984 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3985 3985 }
3986 3986
3987 3987 DO_SIGNAL_CHECK(SR_signum);
3988 3988 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3989 3989 }
3990 3990
3991 3991 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3992 3992
3993 3993 static os_sigaction_t os_sigaction = NULL;
3994 3994
3995 3995 void os::Linux::check_signal_handler(int sig) {
3996 3996 char buf[O_BUFLEN];
3997 3997 address jvmHandler = NULL;
3998 3998
3999 3999
4000 4000 struct sigaction act;
4001 4001 if (os_sigaction == NULL) {
4002 4002 // only trust the default sigaction, in case it has been interposed
4003 4003 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4004 4004 if (os_sigaction == NULL) return;
4005 4005 }
4006 4006
4007 4007 os_sigaction(sig, (struct sigaction*)NULL, &act);
4008 4008
4009 4009
4010 4010 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4011 4011
4012 4012 address thisHandler = (act.sa_flags & SA_SIGINFO)
4013 4013 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4014 4014 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4015 4015
4016 4016
4017 4017 switch(sig) {
4018 4018 case SIGSEGV:
4019 4019 case SIGBUS:
4020 4020 case SIGFPE:
4021 4021 case SIGPIPE:
4022 4022 case SIGILL:
4023 4023 case SIGXFSZ:
4024 4024 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4025 4025 break;
4026 4026
4027 4027 case SHUTDOWN1_SIGNAL:
4028 4028 case SHUTDOWN2_SIGNAL:
4029 4029 case SHUTDOWN3_SIGNAL:
4030 4030 case BREAK_SIGNAL:
4031 4031 jvmHandler = (address)user_handler();
4032 4032 break;
4033 4033
4034 4034 case INTERRUPT_SIGNAL:
4035 4035 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4036 4036 break;
4037 4037
4038 4038 default:
4039 4039 if (sig == SR_signum) {
4040 4040 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4041 4041 } else {
4042 4042 return;
4043 4043 }
4044 4044 break;
4045 4045 }
4046 4046
4047 4047 if (thisHandler != jvmHandler) {
4048 4048 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4049 4049 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4050 4050 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4051 4051 // No need to check this sig any longer
4052 4052 sigaddset(&check_signal_done, sig);
4053 4053 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4054 4054 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4055 4055 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4056 4056 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4057 4057 // No need to check this sig any longer
4058 4058 sigaddset(&check_signal_done, sig);
4059 4059 }
4060 4060
4061 4061 // Dump all the signal
4062 4062 if (sigismember(&check_signal_done, sig)) {
4063 4063 print_signal_handlers(tty, buf, O_BUFLEN);
4064 4064 }
4065 4065 }
4066 4066
4067 4067 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4068 4068
4069 4069 extern bool signal_name(int signo, char* buf, size_t len);
4070 4070
4071 4071 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4072 4072 if (0 < exception_code && exception_code <= SIGRTMAX) {
4073 4073 // signal
4074 4074 if (!signal_name(exception_code, buf, size)) {
4075 4075 jio_snprintf(buf, size, "SIG%d", exception_code);
4076 4076 }
4077 4077 return buf;
4078 4078 } else {
4079 4079 return NULL;
4080 4080 }
4081 4081 }
4082 4082
4083 4083 // this is called _before_ the most of global arguments have been parsed
4084 4084 void os::init(void) {
4085 4085 char dummy; /* used to get a guess on initial stack address */
4086 4086 // first_hrtime = gethrtime();
4087 4087
4088 4088 // With LinuxThreads the JavaMain thread pid (primordial thread)
4089 4089 // is different than the pid of the java launcher thread.
4090 4090 // So, on Linux, the launcher thread pid is passed to the VM
4091 4091 // via the sun.java.launcher.pid property.
4092 4092 // Use this property instead of getpid() if it was correctly passed.
4093 4093 // See bug 6351349.
4094 4094 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4095 4095
4096 4096 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4097 4097
4098 4098 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4099 4099
4100 4100 init_random(1234567);
4101 4101
4102 4102 ThreadCritical::initialize();
4103 4103
4104 4104 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4105 4105 if (Linux::page_size() == -1) {
4106 4106 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4107 4107 strerror(errno)));
4108 4108 }
4109 4109 init_page_sizes((size_t) Linux::page_size());
4110 4110
4111 4111 Linux::initialize_system_info();
4112 4112
4113 4113 // main_thread points to the aboriginal thread
4114 4114 Linux::_main_thread = pthread_self();
4115 4115
4116 4116 Linux::clock_init();
4117 4117 initial_time_count = os::elapsed_counter();
4118 4118 pthread_mutex_init(&dl_mutex, NULL);
4119 4119 }
4120 4120
4121 4121 // To install functions for atexit system call
4122 4122 extern "C" {
4123 4123 static void perfMemory_exit_helper() {
4124 4124 perfMemory_exit();
4125 4125 }
4126 4126 }
4127 4127
4128 4128 // this is called _after_ the global arguments have been parsed
4129 4129 jint os::init_2(void)
4130 4130 {
4131 4131 Linux::fast_thread_clock_init();
4132 4132
4133 4133 // Allocate a single page and mark it as readable for safepoint polling
4134 4134 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4135 4135 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4136 4136
4137 4137 os::set_polling_page( polling_page );
4138 4138
4139 4139 #ifndef PRODUCT
4140 4140 if(Verbose && PrintMiscellaneous)
4141 4141 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4142 4142 #endif
4143 4143
4144 4144 if (!UseMembar) {
4145 4145 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4146 4146 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4147 4147 os::set_memory_serialize_page( mem_serialize_page );
4148 4148
4149 4149 #ifndef PRODUCT
4150 4150 if(Verbose && PrintMiscellaneous)
4151 4151 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4152 4152 #endif
4153 4153 }
4154 4154
4155 4155 os::large_page_init();
4156 4156
4157 4157 // initialize suspend/resume support - must do this before signal_sets_init()
4158 4158 if (SR_initialize() != 0) {
4159 4159 perror("SR_initialize failed");
4160 4160 return JNI_ERR;
4161 4161 }
4162 4162
4163 4163 Linux::signal_sets_init();
4164 4164 Linux::install_signal_handlers();
4165 4165
4166 4166 // Check minimum allowable stack size for thread creation and to initialize
4167 4167 // the java system classes, including StackOverflowError - depends on page
4168 4168 // size. Add a page for compiler2 recursion in main thread.
4169 4169 // Add in 2*BytesPerWord times page size to account for VM stack during
4170 4170 // class initialization depending on 32 or 64 bit VM.
4171 4171 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4172 4172 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4173 4173 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4174 4174
4175 4175 size_t threadStackSizeInBytes = ThreadStackSize * K;
4176 4176 if (threadStackSizeInBytes != 0 &&
4177 4177 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4178 4178 tty->print_cr("\nThe stack size specified is too small, "
4179 4179 "Specify at least %dk",
4180 4180 os::Linux::min_stack_allowed/ K);
4181 4181 return JNI_ERR;
4182 4182 }
4183 4183
4184 4184 // Make the stack size a multiple of the page size so that
4185 4185 // the yellow/red zones can be guarded.
4186 4186 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4187 4187 vm_page_size()));
4188 4188
4189 4189 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4190 4190
4191 4191 Linux::libpthread_init();
4192 4192 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4193 4193 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4194 4194 Linux::glibc_version(), Linux::libpthread_version(),
4195 4195 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4196 4196 }
4197 4197
4198 4198 if (UseNUMA) {
4199 4199 if (!Linux::libnuma_init()) {
4200 4200 UseNUMA = false;
4201 4201 } else {
4202 4202 if ((Linux::numa_max_node() < 1)) {
4203 4203 // There's only one node(they start from 0), disable NUMA.
4204 4204 UseNUMA = false;
4205 4205 }
4206 4206 }
4207 4207 // With SHM large pages we cannot uncommit a page, so there's not way
4208 4208 // we can make the adaptive lgrp chunk resizing work. If the user specified
4209 4209 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4210 4210 // disable adaptive resizing.
4211 4211 if (UseNUMA && UseLargePages && UseSHM) {
4212 4212 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4213 4213 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4214 4214 UseLargePages = false;
4215 4215 } else {
4216 4216 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4217 4217 UseAdaptiveSizePolicy = false;
4218 4218 UseAdaptiveNUMAChunkSizing = false;
4219 4219 }
4220 4220 } else {
4221 4221 UseNUMA = false;
4222 4222 }
4223 4223 }
4224 4224 if (!UseNUMA && ForceNUMA) {
4225 4225 UseNUMA = true;
4226 4226 }
4227 4227 }
4228 4228
4229 4229 if (MaxFDLimit) {
4230 4230 // set the number of file descriptors to max. print out error
4231 4231 // if getrlimit/setrlimit fails but continue regardless.
4232 4232 struct rlimit nbr_files;
4233 4233 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4234 4234 if (status != 0) {
4235 4235 if (PrintMiscellaneous && (Verbose || WizardMode))
4236 4236 perror("os::init_2 getrlimit failed");
4237 4237 } else {
4238 4238 nbr_files.rlim_cur = nbr_files.rlim_max;
4239 4239 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4240 4240 if (status != 0) {
4241 4241 if (PrintMiscellaneous && (Verbose || WizardMode))
4242 4242 perror("os::init_2 setrlimit failed");
4243 4243 }
4244 4244 }
4245 4245 }
4246 4246
4247 4247 // Initialize lock used to serialize thread creation (see os::create_thread)
4248 4248 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4249 4249
4250 4250 // at-exit methods are called in the reverse order of their registration.
4251 4251 // atexit functions are called on return from main or as a result of a
4252 4252 // call to exit(3C). There can be only 32 of these functions registered
4253 4253 // and atexit() does not set errno.
4254 4254
4255 4255 if (PerfAllowAtExitRegistration) {
4256 4256 // only register atexit functions if PerfAllowAtExitRegistration is set.
4257 4257 // atexit functions can be delayed until process exit time, which
4258 4258 // can be problematic for embedded VM situations. Embedded VMs should
4259 4259 // call DestroyJavaVM() to assure that VM resources are released.
4260 4260
4261 4261 // note: perfMemory_exit_helper atexit function may be removed in
4262 4262 // the future if the appropriate cleanup code can be added to the
4263 4263 // VM_Exit VMOperation's doit method.
4264 4264 if (atexit(perfMemory_exit_helper) != 0) {
4265 4265 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4266 4266 }
4267 4267 }
4268 4268
4269 4269 // initialize thread priority policy
4270 4270 prio_init();
4271 4271
4272 4272 return JNI_OK;
4273 4273 }
4274 4274
4275 4275 // this is called at the end of vm_initialization
4276 4276 void os::init_3(void)
4277 4277 {
4278 4278 #ifdef JAVASE_EMBEDDED
4279 4279 // Start the MemNotifyThread
4280 4280 if (LowMemoryProtection) {
4281 4281 MemNotifyThread::start();
4282 4282 }
4283 4283 return;
4284 4284 #endif
4285 4285 }
4286 4286
4287 4287 // Mark the polling page as unreadable
4288 4288 void os::make_polling_page_unreadable(void) {
4289 4289 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4290 4290 fatal("Could not disable polling page");
4291 4291 };
4292 4292
4293 4293 // Mark the polling page as readable
4294 4294 void os::make_polling_page_readable(void) {
4295 4295 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4296 4296 fatal("Could not enable polling page");
4297 4297 }
4298 4298 };
4299 4299
4300 4300 int os::active_processor_count() {
4301 4301 // Linux doesn't yet have a (official) notion of processor sets,
4302 4302 // so just return the number of online processors.
4303 4303 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4304 4304 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4305 4305 return online_cpus;
4306 4306 }
4307 4307
4308 4308 bool os::distribute_processes(uint length, uint* distribution) {
4309 4309 // Not yet implemented.
4310 4310 return false;
4311 4311 }
4312 4312
4313 4313 bool os::bind_to_processor(uint processor_id) {
4314 4314 // Not yet implemented.
4315 4315 return false;
4316 4316 }
4317 4317
4318 4318 ///
4319 4319
4320 4320 // Suspends the target using the signal mechanism and then grabs the PC before
4321 4321 // resuming the target. Used by the flat-profiler only
4322 4322 ExtendedPC os::get_thread_pc(Thread* thread) {
4323 4323 // Make sure that it is called by the watcher for the VMThread
4324 4324 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4325 4325 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4326 4326
4327 4327 ExtendedPC epc;
4328 4328
4329 4329 OSThread* osthread = thread->osthread();
4330 4330 if (do_suspend(osthread)) {
4331 4331 if (osthread->ucontext() != NULL) {
4332 4332 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4333 4333 } else {
4334 4334 // NULL context is unexpected, double-check this is the VMThread
4335 4335 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4336 4336 }
4337 4337 do_resume(osthread);
4338 4338 }
4339 4339 // failure means pthread_kill failed for some reason - arguably this is
4340 4340 // a fatal problem, but such problems are ignored elsewhere
4341 4341
4342 4342 return epc;
4343 4343 }
4344 4344
4345 4345 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4346 4346 {
4347 4347 if (is_NPTL()) {
4348 4348 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4349 4349 } else {
4350 4350 #ifndef IA64
4351 4351 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4352 4352 // word back to default 64bit precision if condvar is signaled. Java
4353 4353 // wants 53bit precision. Save and restore current value.
4354 4354 int fpu = get_fpu_control_word();
4355 4355 #endif // IA64
4356 4356 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4357 4357 #ifndef IA64
4358 4358 set_fpu_control_word(fpu);
4359 4359 #endif // IA64
4360 4360 return status;
4361 4361 }
4362 4362 }
4363 4363
4364 4364 ////////////////////////////////////////////////////////////////////////////////
4365 4365 // debug support
4366 4366
4367 4367 static address same_page(address x, address y) {
4368 4368 int page_bits = -os::vm_page_size();
4369 4369 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4370 4370 return x;
4371 4371 else if (x > y)
4372 4372 return (address)(intptr_t(y) | ~page_bits) + 1;
4373 4373 else
4374 4374 return (address)(intptr_t(y) & page_bits);
4375 4375 }
4376 4376
4377 4377 bool os::find(address addr, outputStream* st) {
4378 4378 Dl_info dlinfo;
4379 4379 memset(&dlinfo, 0, sizeof(dlinfo));
4380 4380 if (dladdr(addr, &dlinfo)) {
4381 4381 st->print(PTR_FORMAT ": ", addr);
4382 4382 if (dlinfo.dli_sname != NULL) {
4383 4383 st->print("%s+%#x", dlinfo.dli_sname,
4384 4384 addr - (intptr_t)dlinfo.dli_saddr);
4385 4385 } else if (dlinfo.dli_fname) {
4386 4386 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4387 4387 } else {
4388 4388 st->print("<absolute address>");
4389 4389 }
4390 4390 if (dlinfo.dli_fname) {
4391 4391 st->print(" in %s", dlinfo.dli_fname);
4392 4392 }
4393 4393 if (dlinfo.dli_fbase) {
4394 4394 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4395 4395 }
4396 4396 st->cr();
4397 4397
4398 4398 if (Verbose) {
4399 4399 // decode some bytes around the PC
4400 4400 address begin = same_page(addr-40, addr);
4401 4401 address end = same_page(addr+40, addr);
4402 4402 address lowest = (address) dlinfo.dli_sname;
4403 4403 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4404 4404 if (begin < lowest) begin = lowest;
4405 4405 Dl_info dlinfo2;
4406 4406 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4407 4407 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4408 4408 end = (address) dlinfo2.dli_saddr;
4409 4409 Disassembler::decode(begin, end, st);
4410 4410 }
4411 4411 return true;
4412 4412 }
4413 4413 return false;
4414 4414 }
4415 4415
4416 4416 ////////////////////////////////////////////////////////////////////////////////
4417 4417 // misc
4418 4418
4419 4419 // This does not do anything on Linux. This is basically a hook for being
4420 4420 // able to use structured exception handling (thread-local exception filters)
4421 4421 // on, e.g., Win32.
4422 4422 void
4423 4423 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4424 4424 JavaCallArguments* args, Thread* thread) {
4425 4425 f(value, method, args, thread);
4426 4426 }
4427 4427
4428 4428 void os::print_statistics() {
4429 4429 }
4430 4430
4431 4431 int os::message_box(const char* title, const char* message) {
4432 4432 int i;
4433 4433 fdStream err(defaultStream::error_fd());
4434 4434 for (i = 0; i < 78; i++) err.print_raw("=");
4435 4435 err.cr();
4436 4436 err.print_raw_cr(title);
4437 4437 for (i = 0; i < 78; i++) err.print_raw("-");
4438 4438 err.cr();
4439 4439 err.print_raw_cr(message);
4440 4440 for (i = 0; i < 78; i++) err.print_raw("=");
4441 4441 err.cr();
4442 4442
4443 4443 char buf[16];
4444 4444 // Prevent process from exiting upon "read error" without consuming all CPU
4445 4445 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4446 4446
4447 4447 return buf[0] == 'y' || buf[0] == 'Y';
4448 4448 }
4449 4449
4450 4450 int os::stat(const char *path, struct stat *sbuf) {
4451 4451 char pathbuf[MAX_PATH];
4452 4452 if (strlen(path) > MAX_PATH - 1) {
4453 4453 errno = ENAMETOOLONG;
4454 4454 return -1;
4455 4455 }
4456 4456 os::native_path(strcpy(pathbuf, path));
4457 4457 return ::stat(pathbuf, sbuf);
4458 4458 }
4459 4459
4460 4460 bool os::check_heap(bool force) {
4461 4461 return true;
4462 4462 }
4463 4463
4464 4464 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4465 4465 return ::vsnprintf(buf, count, format, args);
4466 4466 }
4467 4467
4468 4468 // Is a (classpath) directory empty?
4469 4469 bool os::dir_is_empty(const char* path) {
4470 4470 DIR *dir = NULL;
4471 4471 struct dirent *ptr;
4472 4472
4473 4473 dir = opendir(path);
4474 4474 if (dir == NULL) return true;
4475 4475
4476 4476 /* Scan the directory */
4477 4477 bool result = true;
4478 4478 char buf[sizeof(struct dirent) + MAX_PATH];
4479 4479 while (result && (ptr = ::readdir(dir)) != NULL) {
4480 4480 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4481 4481 result = false;
4482 4482 }
4483 4483 }
4484 4484 closedir(dir);
4485 4485 return result;
4486 4486 }
4487 4487
4488 4488 // This code originates from JDK's sysOpen and open64_w
4489 4489 // from src/solaris/hpi/src/system_md.c
4490 4490
4491 4491 #ifndef O_DELETE
4492 4492 #define O_DELETE 0x10000
4493 4493 #endif
4494 4494
4495 4495 // Open a file. Unlink the file immediately after open returns
4496 4496 // if the specified oflag has the O_DELETE flag set.
4497 4497 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4498 4498
4499 4499 int os::open(const char *path, int oflag, int mode) {
4500 4500
4501 4501 if (strlen(path) > MAX_PATH - 1) {
4502 4502 errno = ENAMETOOLONG;
4503 4503 return -1;
4504 4504 }
4505 4505 int fd;
4506 4506 int o_delete = (oflag & O_DELETE);
4507 4507 oflag = oflag & ~O_DELETE;
4508 4508
4509 4509 fd = ::open64(path, oflag, mode);
4510 4510 if (fd == -1) return -1;
4511 4511
4512 4512 //If the open succeeded, the file might still be a directory
4513 4513 {
4514 4514 struct stat64 buf64;
4515 4515 int ret = ::fstat64(fd, &buf64);
4516 4516 int st_mode = buf64.st_mode;
4517 4517
4518 4518 if (ret != -1) {
4519 4519 if ((st_mode & S_IFMT) == S_IFDIR) {
4520 4520 errno = EISDIR;
4521 4521 ::close(fd);
4522 4522 return -1;
4523 4523 }
4524 4524 } else {
4525 4525 ::close(fd);
4526 4526 return -1;
4527 4527 }
4528 4528 }
4529 4529
4530 4530 /*
4531 4531 * All file descriptors that are opened in the JVM and not
4532 4532 * specifically destined for a subprocess should have the
4533 4533 * close-on-exec flag set. If we don't set it, then careless 3rd
4534 4534 * party native code might fork and exec without closing all
4535 4535 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4536 4536 * UNIXProcess.c), and this in turn might:
4537 4537 *
4538 4538 * - cause end-of-file to fail to be detected on some file
4539 4539 * descriptors, resulting in mysterious hangs, or
4540 4540 *
4541 4541 * - might cause an fopen in the subprocess to fail on a system
4542 4542 * suffering from bug 1085341.
4543 4543 *
4544 4544 * (Yes, the default setting of the close-on-exec flag is a Unix
4545 4545 * design flaw)
4546 4546 *
4547 4547 * See:
4548 4548 * 1085341: 32-bit stdio routines should support file descriptors >255
4549 4549 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4550 4550 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4551 4551 */
4552 4552 #ifdef FD_CLOEXEC
4553 4553 {
4554 4554 int flags = ::fcntl(fd, F_GETFD);
4555 4555 if (flags != -1)
4556 4556 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4557 4557 }
4558 4558 #endif
4559 4559
4560 4560 if (o_delete != 0) {
4561 4561 ::unlink(path);
4562 4562 }
4563 4563 return fd;
4564 4564 }
4565 4565
4566 4566
4567 4567 // create binary file, rewriting existing file if required
4568 4568 int os::create_binary_file(const char* path, bool rewrite_existing) {
4569 4569 int oflags = O_WRONLY | O_CREAT;
4570 4570 if (!rewrite_existing) {
4571 4571 oflags |= O_EXCL;
4572 4572 }
4573 4573 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4574 4574 }
4575 4575
4576 4576 // return current position of file pointer
4577 4577 jlong os::current_file_offset(int fd) {
4578 4578 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4579 4579 }
4580 4580
4581 4581 // move file pointer to the specified offset
4582 4582 jlong os::seek_to_file_offset(int fd, jlong offset) {
4583 4583 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4584 4584 }
4585 4585
4586 4586 // This code originates from JDK's sysAvailable
4587 4587 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4588 4588
4589 4589 int os::available(int fd, jlong *bytes) {
4590 4590 jlong cur, end;
4591 4591 int mode;
4592 4592 struct stat64 buf64;
4593 4593
4594 4594 if (::fstat64(fd, &buf64) >= 0) {
4595 4595 mode = buf64.st_mode;
4596 4596 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4597 4597 /*
4598 4598 * XXX: is the following call interruptible? If so, this might
4599 4599 * need to go through the INTERRUPT_IO() wrapper as for other
4600 4600 * blocking, interruptible calls in this file.
4601 4601 */
4602 4602 int n;
4603 4603 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4604 4604 *bytes = n;
4605 4605 return 1;
4606 4606 }
4607 4607 }
4608 4608 }
4609 4609 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4610 4610 return 0;
4611 4611 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4612 4612 return 0;
4613 4613 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4614 4614 return 0;
4615 4615 }
4616 4616 *bytes = end - cur;
4617 4617 return 1;
4618 4618 }
4619 4619
4620 4620 int os::socket_available(int fd, jint *pbytes) {
4621 4621 // Linux doc says EINTR not returned, unlike Solaris
4622 4622 int ret = ::ioctl(fd, FIONREAD, pbytes);
4623 4623
4624 4624 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4625 4625 // is expected to return 0 on failure and 1 on success to the jdk.
4626 4626 return (ret < 0) ? 0 : 1;
4627 4627 }
4628 4628
4629 4629 // Map a block of memory.
4630 4630 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4631 4631 char *addr, size_t bytes, bool read_only,
4632 4632 bool allow_exec) {
4633 4633 int prot;
4634 4634 int flags;
4635 4635
4636 4636 if (read_only) {
4637 4637 prot = PROT_READ;
4638 4638 flags = MAP_SHARED;
4639 4639 } else {
4640 4640 prot = PROT_READ | PROT_WRITE;
4641 4641 flags = MAP_PRIVATE;
4642 4642 }
4643 4643
4644 4644 if (allow_exec) {
4645 4645 prot |= PROT_EXEC;
4646 4646 }
4647 4647
4648 4648 if (addr != NULL) {
4649 4649 flags |= MAP_FIXED;
4650 4650 }
4651 4651
4652 4652 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4653 4653 fd, file_offset);
4654 4654 if (mapped_address == MAP_FAILED) {
4655 4655 return NULL;
4656 4656 }
4657 4657 return mapped_address;
4658 4658 }
4659 4659
4660 4660
4661 4661 // Remap a block of memory.
4662 4662 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4663 4663 char *addr, size_t bytes, bool read_only,
4664 4664 bool allow_exec) {
4665 4665 // same as map_memory() on this OS
4666 4666 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4667 4667 allow_exec);
4668 4668 }
4669 4669
4670 4670
4671 4671 // Unmap a block of memory.
4672 4672 bool os::unmap_memory(char* addr, size_t bytes) {
4673 4673 return munmap(addr, bytes) == 0;
4674 4674 }
4675 4675
4676 4676 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4677 4677
4678 4678 static clockid_t thread_cpu_clockid(Thread* thread) {
4679 4679 pthread_t tid = thread->osthread()->pthread_id();
4680 4680 clockid_t clockid;
4681 4681
4682 4682 // Get thread clockid
4683 4683 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4684 4684 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4685 4685 return clockid;
4686 4686 }
4687 4687
4688 4688 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4689 4689 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4690 4690 // of a thread.
4691 4691 //
4692 4692 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4693 4693 // the fast estimate available on the platform.
4694 4694
4695 4695 jlong os::current_thread_cpu_time() {
4696 4696 if (os::Linux::supports_fast_thread_cpu_time()) {
4697 4697 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4698 4698 } else {
4699 4699 // return user + sys since the cost is the same
4700 4700 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4701 4701 }
4702 4702 }
4703 4703
4704 4704 jlong os::thread_cpu_time(Thread* thread) {
4705 4705 // consistent with what current_thread_cpu_time() returns
4706 4706 if (os::Linux::supports_fast_thread_cpu_time()) {
4707 4707 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4708 4708 } else {
4709 4709 return slow_thread_cpu_time(thread, true /* user + sys */);
4710 4710 }
4711 4711 }
4712 4712
4713 4713 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4714 4714 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4715 4715 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4716 4716 } else {
4717 4717 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4718 4718 }
4719 4719 }
4720 4720
4721 4721 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4722 4722 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4723 4723 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4724 4724 } else {
4725 4725 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4726 4726 }
4727 4727 }
4728 4728
4729 4729 //
4730 4730 // -1 on error.
4731 4731 //
4732 4732
4733 4733 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4734 4734 static bool proc_pid_cpu_avail = true;
4735 4735 static bool proc_task_unchecked = true;
4736 4736 static const char *proc_stat_path = "/proc/%d/stat";
4737 4737 pid_t tid = thread->osthread()->thread_id();
4738 4738 int i;
4739 4739 char *s;
4740 4740 char stat[2048];
4741 4741 int statlen;
4742 4742 char proc_name[64];
4743 4743 int count;
4744 4744 long sys_time, user_time;
4745 4745 char string[64];
4746 4746 char cdummy;
4747 4747 int idummy;
4748 4748 long ldummy;
4749 4749 FILE *fp;
4750 4750
4751 4751 // We first try accessing /proc/<pid>/cpu since this is faster to
4752 4752 // process. If this file is not present (linux kernels 2.5 and above)
4753 4753 // then we open /proc/<pid>/stat.
4754 4754 if ( proc_pid_cpu_avail ) {
4755 4755 sprintf(proc_name, "/proc/%d/cpu", tid);
4756 4756 fp = fopen(proc_name, "r");
4757 4757 if ( fp != NULL ) {
4758 4758 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4759 4759 fclose(fp);
4760 4760 if ( count != 3 ) return -1;
4761 4761
4762 4762 if (user_sys_cpu_time) {
4763 4763 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4764 4764 } else {
4765 4765 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4766 4766 }
4767 4767 }
4768 4768 else proc_pid_cpu_avail = false;
4769 4769 }
4770 4770
4771 4771 // The /proc/<tid>/stat aggregates per-process usage on
4772 4772 // new Linux kernels 2.6+ where NPTL is supported.
4773 4773 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4774 4774 // See bug 6328462.
4775 4775 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4776 4776 // and possibly in some other cases, so we check its availability.
4777 4777 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4778 4778 // This is executed only once
4779 4779 proc_task_unchecked = false;
4780 4780 fp = fopen("/proc/self/task", "r");
4781 4781 if (fp != NULL) {
4782 4782 proc_stat_path = "/proc/self/task/%d/stat";
4783 4783 fclose(fp);
4784 4784 }
4785 4785 }
4786 4786
4787 4787 sprintf(proc_name, proc_stat_path, tid);
4788 4788 fp = fopen(proc_name, "r");
4789 4789 if ( fp == NULL ) return -1;
4790 4790 statlen = fread(stat, 1, 2047, fp);
4791 4791 stat[statlen] = '\0';
4792 4792 fclose(fp);
4793 4793
4794 4794 // Skip pid and the command string. Note that we could be dealing with
4795 4795 // weird command names, e.g. user could decide to rename java launcher
4796 4796 // to "java 1.4.2 :)", then the stat file would look like
4797 4797 // 1234 (java 1.4.2 :)) R ... ...
4798 4798 // We don't really need to know the command string, just find the last
4799 4799 // occurrence of ")" and then start parsing from there. See bug 4726580.
4800 4800 s = strrchr(stat, ')');
4801 4801 i = 0;
4802 4802 if (s == NULL ) return -1;
4803 4803
4804 4804 // Skip blank chars
4805 4805 do s++; while (isspace(*s));
4806 4806
4807 4807 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4808 4808 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4809 4809 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4810 4810 &user_time, &sys_time);
4811 4811 if ( count != 13 ) return -1;
4812 4812 if (user_sys_cpu_time) {
4813 4813 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4814 4814 } else {
4815 4815 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4816 4816 }
4817 4817 }
4818 4818
4819 4819 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4820 4820 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4821 4821 info_ptr->may_skip_backward = false; // elapsed time not wall time
4822 4822 info_ptr->may_skip_forward = false; // elapsed time not wall time
4823 4823 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4824 4824 }
4825 4825
4826 4826 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4827 4827 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4828 4828 info_ptr->may_skip_backward = false; // elapsed time not wall time
4829 4829 info_ptr->may_skip_forward = false; // elapsed time not wall time
4830 4830 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4831 4831 }
4832 4832
4833 4833 bool os::is_thread_cpu_time_supported() {
4834 4834 return true;
4835 4835 }
4836 4836
4837 4837 // System loadavg support. Returns -1 if load average cannot be obtained.
4838 4838 // Linux doesn't yet have a (official) notion of processor sets,
4839 4839 // so just return the system wide load average.
4840 4840 int os::loadavg(double loadavg[], int nelem) {
4841 4841 return ::getloadavg(loadavg, nelem);
4842 4842 }
4843 4843
4844 4844 void os::pause() {
4845 4845 char filename[MAX_PATH];
4846 4846 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4847 4847 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4848 4848 } else {
4849 4849 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4850 4850 }
4851 4851
4852 4852 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4853 4853 if (fd != -1) {
4854 4854 struct stat buf;
4855 4855 ::close(fd);
4856 4856 while (::stat(filename, &buf) == 0) {
4857 4857 (void)::poll(NULL, 0, 100);
4858 4858 }
4859 4859 } else {
4860 4860 jio_fprintf(stderr,
4861 4861 "Could not open pause file '%s', continuing immediately.\n", filename);
4862 4862 }
4863 4863 }
4864 4864
4865 4865
4866 4866 // Refer to the comments in os_solaris.cpp park-unpark.
4867 4867 //
4868 4868 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4869 4869 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4870 4870 // For specifics regarding the bug see GLIBC BUGID 261237 :
4871 4871 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4872 4872 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4873 4873 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4874 4874 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4875 4875 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4876 4876 // and monitorenter when we're using 1-0 locking. All those operations may result in
4877 4877 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4878 4878 // of libpthread avoids the problem, but isn't practical.
4879 4879 //
4880 4880 // Possible remedies:
4881 4881 //
4882 4882 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4883 4883 // This is palliative and probabilistic, however. If the thread is preempted
4884 4884 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4885 4885 // than the minimum period may have passed, and the abstime may be stale (in the
4886 4886 // past) resultin in a hang. Using this technique reduces the odds of a hang
4887 4887 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4888 4888 //
4889 4889 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4890 4890 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4891 4891 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4892 4892 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4893 4893 // thread.
4894 4894 //
4895 4895 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4896 4896 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4897 4897 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4898 4898 // This also works well. In fact it avoids kernel-level scalability impediments
4899 4899 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4900 4900 // timers in a graceful fashion.
4901 4901 //
4902 4902 // 4. When the abstime value is in the past it appears that control returns
4903 4903 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4904 4904 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4905 4905 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4906 4906 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4907 4907 // It may be possible to avoid reinitialization by checking the return
4908 4908 // value from pthread_cond_timedwait(). In addition to reinitializing the
4909 4909 // condvar we must establish the invariant that cond_signal() is only called
4910 4910 // within critical sections protected by the adjunct mutex. This prevents
4911 4911 // cond_signal() from "seeing" a condvar that's in the midst of being
4912 4912 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4913 4913 // desirable signal-after-unlock optimization that avoids futile context switching.
4914 4914 //
4915 4915 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4916 4916 // structure when a condvar is used or initialized. cond_destroy() would
4917 4917 // release the helper structure. Our reinitialize-after-timedwait fix
4918 4918 // put excessive stress on malloc/free and locks protecting the c-heap.
4919 4919 //
4920 4920 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4921 4921 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4922 4922 // and only enabling the work-around for vulnerable environments.
4923 4923
4924 4924 // utility to compute the abstime argument to timedwait:
4925 4925 // millis is the relative timeout time
4926 4926 // abstime will be the absolute timeout time
4927 4927 // TODO: replace compute_abstime() with unpackTime()
4928 4928
4929 4929 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4930 4930 if (millis < 0) millis = 0;
4931 4931 struct timeval now;
4932 4932 int status = gettimeofday(&now, NULL);
4933 4933 assert(status == 0, "gettimeofday");
4934 4934 jlong seconds = millis / 1000;
4935 4935 millis %= 1000;
4936 4936 if (seconds > 50000000) { // see man cond_timedwait(3T)
4937 4937 seconds = 50000000;
4938 4938 }
4939 4939 abstime->tv_sec = now.tv_sec + seconds;
4940 4940 long usec = now.tv_usec + millis * 1000;
4941 4941 if (usec >= 1000000) {
4942 4942 abstime->tv_sec += 1;
4943 4943 usec -= 1000000;
4944 4944 }
4945 4945 abstime->tv_nsec = usec * 1000;
4946 4946 return abstime;
4947 4947 }
4948 4948
4949 4949
4950 4950 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4951 4951 // Conceptually TryPark() should be equivalent to park(0).
4952 4952
4953 4953 int os::PlatformEvent::TryPark() {
4954 4954 for (;;) {
4955 4955 const int v = _Event ;
4956 4956 guarantee ((v == 0) || (v == 1), "invariant") ;
4957 4957 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4958 4958 }
4959 4959 }
4960 4960
4961 4961 void os::PlatformEvent::park() { // AKA "down()"
4962 4962 // Invariant: Only the thread associated with the Event/PlatformEvent
4963 4963 // may call park().
4964 4964 // TODO: assert that _Assoc != NULL or _Assoc == Self
4965 4965 int v ;
4966 4966 for (;;) {
4967 4967 v = _Event ;
4968 4968 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4969 4969 }
4970 4970 guarantee (v >= 0, "invariant") ;
4971 4971 if (v == 0) {
4972 4972 // Do this the hard way by blocking ...
4973 4973 int status = pthread_mutex_lock(_mutex);
4974 4974 assert_status(status == 0, status, "mutex_lock");
4975 4975 guarantee (_nParked == 0, "invariant") ;
4976 4976 ++ _nParked ;
4977 4977 while (_Event < 0) {
4978 4978 status = pthread_cond_wait(_cond, _mutex);
4979 4979 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4980 4980 // Treat this the same as if the wait was interrupted
4981 4981 if (status == ETIME) { status = EINTR; }
4982 4982 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4983 4983 }
4984 4984 -- _nParked ;
4985 4985
4986 4986 // In theory we could move the ST of 0 into _Event past the unlock(),
4987 4987 // but then we'd need a MEMBAR after the ST.
4988 4988 _Event = 0 ;
4989 4989 status = pthread_mutex_unlock(_mutex);
4990 4990 assert_status(status == 0, status, "mutex_unlock");
4991 4991 }
4992 4992 guarantee (_Event >= 0, "invariant") ;
4993 4993 }
4994 4994
4995 4995 int os::PlatformEvent::park(jlong millis) {
4996 4996 guarantee (_nParked == 0, "invariant") ;
4997 4997
4998 4998 int v ;
4999 4999 for (;;) {
5000 5000 v = _Event ;
5001 5001 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5002 5002 }
5003 5003 guarantee (v >= 0, "invariant") ;
5004 5004 if (v != 0) return OS_OK ;
5005 5005
5006 5006 // We do this the hard way, by blocking the thread.
5007 5007 // Consider enforcing a minimum timeout value.
5008 5008 struct timespec abst;
5009 5009 compute_abstime(&abst, millis);
5010 5010
5011 5011 int ret = OS_TIMEOUT;
5012 5012 int status = pthread_mutex_lock(_mutex);
5013 5013 assert_status(status == 0, status, "mutex_lock");
5014 5014 guarantee (_nParked == 0, "invariant") ;
5015 5015 ++_nParked ;
5016 5016
5017 5017 // Object.wait(timo) will return because of
5018 5018 // (a) notification
5019 5019 // (b) timeout
5020 5020 // (c) thread.interrupt
5021 5021 //
5022 5022 // Thread.interrupt and object.notify{All} both call Event::set.
5023 5023 // That is, we treat thread.interrupt as a special case of notification.
5024 5024 // The underlying Solaris implementation, cond_timedwait, admits
5025 5025 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5026 5026 // JVM from making those visible to Java code. As such, we must
5027 5027 // filter out spurious wakeups. We assume all ETIME returns are valid.
5028 5028 //
5029 5029 // TODO: properly differentiate simultaneous notify+interrupt.
5030 5030 // In that case, we should propagate the notify to another waiter.
5031 5031
5032 5032 while (_Event < 0) {
5033 5033 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5034 5034 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5035 5035 pthread_cond_destroy (_cond);
5036 5036 pthread_cond_init (_cond, NULL) ;
5037 5037 }
5038 5038 assert_status(status == 0 || status == EINTR ||
5039 5039 status == ETIME || status == ETIMEDOUT,
5040 5040 status, "cond_timedwait");
5041 5041 if (!FilterSpuriousWakeups) break ; // previous semantics
5042 5042 if (status == ETIME || status == ETIMEDOUT) break ;
5043 5043 // We consume and ignore EINTR and spurious wakeups.
5044 5044 }
5045 5045 --_nParked ;
5046 5046 if (_Event >= 0) {
5047 5047 ret = OS_OK;
5048 5048 }
5049 5049 _Event = 0 ;
5050 5050 status = pthread_mutex_unlock(_mutex);
5051 5051 assert_status(status == 0, status, "mutex_unlock");
5052 5052 assert (_nParked == 0, "invariant") ;
5053 5053 return ret;
5054 5054 }
5055 5055
5056 5056 void os::PlatformEvent::unpark() {
5057 5057 int v, AnyWaiters ;
5058 5058 for (;;) {
5059 5059 v = _Event ;
5060 5060 if (v > 0) {
5061 5061 // The LD of _Event could have reordered or be satisfied
5062 5062 // by a read-aside from this processor's write buffer.
5063 5063 // To avoid problems execute a barrier and then
5064 5064 // ratify the value.
5065 5065 OrderAccess::fence() ;
5066 5066 if (_Event == v) return ;
5067 5067 continue ;
5068 5068 }
5069 5069 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5070 5070 }
5071 5071 if (v < 0) {
5072 5072 // Wait for the thread associated with the event to vacate
5073 5073 int status = pthread_mutex_lock(_mutex);
5074 5074 assert_status(status == 0, status, "mutex_lock");
5075 5075 AnyWaiters = _nParked ;
5076 5076 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5077 5077 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5078 5078 AnyWaiters = 0 ;
5079 5079 pthread_cond_signal (_cond);
5080 5080 }
5081 5081 status = pthread_mutex_unlock(_mutex);
5082 5082 assert_status(status == 0, status, "mutex_unlock");
5083 5083 if (AnyWaiters != 0) {
5084 5084 status = pthread_cond_signal(_cond);
5085 5085 assert_status(status == 0, status, "cond_signal");
5086 5086 }
5087 5087 }
5088 5088
5089 5089 // Note that we signal() _after dropping the lock for "immortal" Events.
5090 5090 // This is safe and avoids a common class of futile wakeups. In rare
5091 5091 // circumstances this can cause a thread to return prematurely from
5092 5092 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5093 5093 // simply re-test the condition and re-park itself.
5094 5094 }
5095 5095
5096 5096
5097 5097 // JSR166
5098 5098 // -------------------------------------------------------
5099 5099
5100 5100 /*
5101 5101 * The solaris and linux implementations of park/unpark are fairly
5102 5102 * conservative for now, but can be improved. They currently use a
5103 5103 * mutex/condvar pair, plus a a count.
5104 5104 * Park decrements count if > 0, else does a condvar wait. Unpark
5105 5105 * sets count to 1 and signals condvar. Only one thread ever waits
5106 5106 * on the condvar. Contention seen when trying to park implies that someone
5107 5107 * is unparking you, so don't wait. And spurious returns are fine, so there
5108 5108 * is no need to track notifications.
5109 5109 */
5110 5110
5111 5111
5112 5112 #define NANOSECS_PER_SEC 1000000000
5113 5113 #define NANOSECS_PER_MILLISEC 1000000
5114 5114 #define MAX_SECS 100000000
5115 5115 /*
5116 5116 * This code is common to linux and solaris and will be moved to a
5117 5117 * common place in dolphin.
5118 5118 *
5119 5119 * The passed in time value is either a relative time in nanoseconds
5120 5120 * or an absolute time in milliseconds. Either way it has to be unpacked
5121 5121 * into suitable seconds and nanoseconds components and stored in the
5122 5122 * given timespec structure.
5123 5123 * Given time is a 64-bit value and the time_t used in the timespec is only
5124 5124 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5125 5125 * overflow if times way in the future are given. Further on Solaris versions
5126 5126 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5127 5127 * number of seconds, in abstime, is less than current_time + 100,000,000.
5128 5128 * As it will be 28 years before "now + 100000000" will overflow we can
5129 5129 * ignore overflow and just impose a hard-limit on seconds using the value
5130 5130 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5131 5131 * years from "now".
5132 5132 */
5133 5133
5134 5134 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5135 5135 assert (time > 0, "convertTime");
5136 5136
5137 5137 struct timeval now;
5138 5138 int status = gettimeofday(&now, NULL);
5139 5139 assert(status == 0, "gettimeofday");
5140 5140
5141 5141 time_t max_secs = now.tv_sec + MAX_SECS;
5142 5142
5143 5143 if (isAbsolute) {
5144 5144 jlong secs = time / 1000;
5145 5145 if (secs > max_secs) {
5146 5146 absTime->tv_sec = max_secs;
5147 5147 }
5148 5148 else {
5149 5149 absTime->tv_sec = secs;
5150 5150 }
5151 5151 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5152 5152 }
5153 5153 else {
5154 5154 jlong secs = time / NANOSECS_PER_SEC;
5155 5155 if (secs >= MAX_SECS) {
5156 5156 absTime->tv_sec = max_secs;
5157 5157 absTime->tv_nsec = 0;
5158 5158 }
5159 5159 else {
5160 5160 absTime->tv_sec = now.tv_sec + secs;
5161 5161 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5162 5162 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5163 5163 absTime->tv_nsec -= NANOSECS_PER_SEC;
5164 5164 ++absTime->tv_sec; // note: this must be <= max_secs
5165 5165 }
5166 5166 }
5167 5167 }
5168 5168 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5169 5169 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5170 5170 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5171 5171 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5172 5172 }
5173 5173
5174 5174 void Parker::park(bool isAbsolute, jlong time) {
5175 5175 // Optional fast-path check:
5176 5176 // Return immediately if a permit is available.
5177 5177 if (_counter > 0) {
5178 5178 _counter = 0 ;
5179 5179 OrderAccess::fence();
5180 5180 return ;
5181 5181 }
5182 5182
5183 5183 Thread* thread = Thread::current();
5184 5184 assert(thread->is_Java_thread(), "Must be JavaThread");
5185 5185 JavaThread *jt = (JavaThread *)thread;
5186 5186
5187 5187 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5188 5188 // Check interrupt before trying to wait
5189 5189 if (Thread::is_interrupted(thread, false)) {
5190 5190 return;
5191 5191 }
5192 5192
5193 5193 // Next, demultiplex/decode time arguments
5194 5194 timespec absTime;
5195 5195 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5196 5196 return;
5197 5197 }
5198 5198 if (time > 0) {
5199 5199 unpackTime(&absTime, isAbsolute, time);
5200 5200 }
5201 5201
5202 5202
5203 5203 // Enter safepoint region
5204 5204 // Beware of deadlocks such as 6317397.
5205 5205 // The per-thread Parker:: mutex is a classic leaf-lock.
5206 5206 // In particular a thread must never block on the Threads_lock while
5207 5207 // holding the Parker:: mutex. If safepoints are pending both the
5208 5208 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5209 5209 ThreadBlockInVM tbivm(jt);
5210 5210
5211 5211 // Don't wait if cannot get lock since interference arises from
5212 5212 // unblocking. Also. check interrupt before trying wait
5213 5213 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5214 5214 return;
5215 5215 }
5216 5216
5217 5217 int status ;
5218 5218 if (_counter > 0) { // no wait needed
5219 5219 _counter = 0;
5220 5220 status = pthread_mutex_unlock(_mutex);
5221 5221 assert (status == 0, "invariant") ;
5222 5222 OrderAccess::fence();
5223 5223 return;
5224 5224 }
5225 5225
5226 5226 #ifdef ASSERT
5227 5227 // Don't catch signals while blocked; let the running threads have the signals.
5228 5228 // (This allows a debugger to break into the running thread.)
5229 5229 sigset_t oldsigs;
5230 5230 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5231 5231 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5232 5232 #endif
5233 5233
5234 5234 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5235 5235 jt->set_suspend_equivalent();
5236 5236 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5237 5237
5238 5238 if (time == 0) {
5239 5239 status = pthread_cond_wait (_cond, _mutex) ;
5240 5240 } else {
5241 5241 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5242 5242 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5243 5243 pthread_cond_destroy (_cond) ;
5244 5244 pthread_cond_init (_cond, NULL);
5245 5245 }
5246 5246 }
5247 5247 assert_status(status == 0 || status == EINTR ||
5248 5248 status == ETIME || status == ETIMEDOUT,
5249 5249 status, "cond_timedwait");
5250 5250
5251 5251 #ifdef ASSERT
5252 5252 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5253 5253 #endif
5254 5254
5255 5255 _counter = 0 ;
5256 5256 status = pthread_mutex_unlock(_mutex) ;
5257 5257 assert_status(status == 0, status, "invariant") ;
5258 5258 // If externally suspended while waiting, re-suspend
5259 5259 if (jt->handle_special_suspend_equivalent_condition()) {
5260 5260 jt->java_suspend_self();
5261 5261 }
5262 5262
5263 5263 OrderAccess::fence();
5264 5264 }
5265 5265
5266 5266 void Parker::unpark() {
5267 5267 int s, status ;
5268 5268 status = pthread_mutex_lock(_mutex);
5269 5269 assert (status == 0, "invariant") ;
5270 5270 s = _counter;
5271 5271 _counter = 1;
5272 5272 if (s < 1) {
5273 5273 if (WorkAroundNPTLTimedWaitHang) {
5274 5274 status = pthread_cond_signal (_cond) ;
5275 5275 assert (status == 0, "invariant") ;
5276 5276 status = pthread_mutex_unlock(_mutex);
5277 5277 assert (status == 0, "invariant") ;
5278 5278 } else {
5279 5279 status = pthread_mutex_unlock(_mutex);
5280 5280 assert (status == 0, "invariant") ;
5281 5281 status = pthread_cond_signal (_cond) ;
5282 5282 assert (status == 0, "invariant") ;
5283 5283 }
5284 5284 } else {
5285 5285 pthread_mutex_unlock(_mutex);
5286 5286 assert (status == 0, "invariant") ;
5287 5287 }
5288 5288 }
5289 5289
5290 5290
5291 5291 extern char** environ;
5292 5292
5293 5293 #ifndef __NR_fork
5294 5294 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5295 5295 #endif
5296 5296
5297 5297 #ifndef __NR_execve
5298 5298 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5299 5299 #endif
5300 5300
5301 5301 // Run the specified command in a separate process. Return its exit value,
5302 5302 // or -1 on failure (e.g. can't fork a new process).
5303 5303 // Unlike system(), this function can be called from signal handler. It
5304 5304 // doesn't block SIGINT et al.
5305 5305 int os::fork_and_exec(char* cmd) {
5306 5306 const char * argv[4] = {"sh", "-c", cmd, NULL};
5307 5307
5308 5308 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5309 5309 // pthread_atfork handlers and reset pthread library. All we need is a
5310 5310 // separate process to execve. Make a direct syscall to fork process.
5311 5311 // On IA64 there's no fork syscall, we have to use fork() and hope for
5312 5312 // the best...
5313 5313 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5314 5314 IA64_ONLY(fork();)
5315 5315
5316 5316 if (pid < 0) {
5317 5317 // fork failed
5318 5318 return -1;
5319 5319
5320 5320 } else if (pid == 0) {
5321 5321 // child process
5322 5322
5323 5323 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5324 5324 // first to kill every thread on the thread list. Because this list is
5325 5325 // not reset by fork() (see notes above), execve() will instead kill
5326 5326 // every thread in the parent process. We know this is the only thread
5327 5327 // in the new process, so make a system call directly.
5328 5328 // IA64 should use normal execve() from glibc to match the glibc fork()
5329 5329 // above.
5330 5330 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5331 5331 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5332 5332
5333 5333 // execve failed
5334 5334 _exit(-1);
5335 5335
5336 5336 } else {
5337 5337 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5338 5338 // care about the actual exit code, for now.
5339 5339
5340 5340 int status;
5341 5341
5342 5342 // Wait for the child process to exit. This returns immediately if
5343 5343 // the child has already exited. */
5344 5344 while (waitpid(pid, &status, 0) < 0) {
5345 5345 switch (errno) {
5346 5346 case ECHILD: return 0;
5347 5347 case EINTR: break;
5348 5348 default: return -1;
5349 5349 }
5350 5350 }
5351 5351
5352 5352 if (WIFEXITED(status)) {
5353 5353 // The child exited normally; get its exit code.
5354 5354 return WEXITSTATUS(status);
5355 5355 } else if (WIFSIGNALED(status)) {
5356 5356 // The child exited because of a signal
5357 5357 // The best value to return is 0x80 + signal number,
5358 5358 // because that is what all Unix shells do, and because
5359 5359 // it allows callers to distinguish between process exit and
5360 5360 // process death by signal.
5361 5361 return 0x80 + WTERMSIG(status);
5362 5362 } else {
5363 5363 // Unknown exit code; pass it through
5364 5364 return status;
5365 5365 }
5366 5366 }
5367 5367 }
5368 5368
5369 5369 // is_headless_jre()
5370 5370 //
5371 5371 // Test for the existence of libmawt in motif21 or xawt directories
5372 5372 // in order to report if we are running in a headless jre
5373 5373 //
5374 5374 bool os::is_headless_jre() {
5375 5375 struct stat statbuf;
5376 5376 char buf[MAXPATHLEN];
5377 5377 char libmawtpath[MAXPATHLEN];
5378 5378 const char *xawtstr = "/xawt/libmawt.so";
5379 5379 const char *motifstr = "/motif21/libmawt.so";
5380 5380 char *p;
5381 5381
5382 5382 // Get path to libjvm.so
5383 5383 os::jvm_path(buf, sizeof(buf));
5384 5384
5385 5385 // Get rid of libjvm.so
5386 5386 p = strrchr(buf, '/');
5387 5387 if (p == NULL) return false;
5388 5388 else *p = '\0';
5389 5389
5390 5390 // Get rid of client or server
5391 5391 p = strrchr(buf, '/');
5392 5392 if (p == NULL) return false;
5393 5393 else *p = '\0';
5394 5394
5395 5395 // check xawt/libmawt.so
5396 5396 strcpy(libmawtpath, buf);
5397 5397 strcat(libmawtpath, xawtstr);
5398 5398 if (::stat(libmawtpath, &statbuf) == 0) return false;
5399 5399
5400 5400 // check motif21/libmawt.so
5401 5401 strcpy(libmawtpath, buf);
5402 5402 strcat(libmawtpath, motifstr);
5403 5403 if (::stat(libmawtpath, &statbuf) == 0) return false;
5404 5404
5405 5405 return true;
5406 5406 }
5407 5407
5408 5408
5409 5409 #ifdef JAVASE_EMBEDDED
5410 5410 //
5411 5411 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5412 5412 //
5413 5413 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5414 5414
5415 5415 // ctor
5416 5416 //
5417 5417 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5418 5418 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5419 5419 _fd = fd;
5420 5420
5421 5421 if (os::create_thread(this, os::os_thread)) {
5422 5422 _memnotify_thread = this;
5423 5423 os::set_priority(this, NearMaxPriority);
5424 5424 os::start_thread(this);
5425 5425 }
5426 5426 }
5427 5427
5428 5428 // Where all the work gets done
5429 5429 //
5430 5430 void MemNotifyThread::run() {
5431 5431 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5432 5432
5433 5433 // Set up the select arguments
5434 5434 fd_set rfds;
5435 5435 if (_fd != -1) {
5436 5436 FD_ZERO(&rfds);
5437 5437 FD_SET(_fd, &rfds);
5438 5438 }
5439 5439
5440 5440 // Now wait for the mem_notify device to wake up
5441 5441 while (1) {
5442 5442 // Wait for the mem_notify device to signal us..
5443 5443 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5444 5444 if (rc == -1) {
5445 5445 perror("select!\n");
5446 5446 break;
5447 5447 } else if (rc) {
5448 5448 //ssize_t free_before = os::available_memory();
5449 5449 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5450 5450
5451 5451 // The kernel is telling us there is not much memory left...
5452 5452 // try to do something about that
5453 5453
5454 5454 // If we are not already in a GC, try one.
5455 5455 if (!Universe::heap()->is_gc_active()) {
5456 5456 Universe::heap()->collect(GCCause::_allocation_failure);
5457 5457
5458 5458 //ssize_t free_after = os::available_memory();
5459 5459 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5460 5460 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5461 5461 }
5462 5462 // We might want to do something like the following if we find the GC's are not helping...
5463 5463 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5464 5464 }
5465 5465 }
5466 5466 }
5467 5467
5468 5468 //
5469 5469 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5470 5470 //
5471 5471 void MemNotifyThread::start() {
5472 5472 int fd;
5473 5473 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5474 5474 if (fd < 0) {
5475 5475 return;
5476 5476 }
5477 5477
5478 5478 if (memnotify_thread() == NULL) {
5479 5479 new MemNotifyThread(fd);
5480 5480 }
5481 5481 }
5482 5482 #endif // JAVASE_EMBEDDED
↓ open down ↓ |
2935 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX