1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/orderAccess.inline.hpp"
53 #include "runtime/osThread.hpp"
54 #include "runtime/perfMemory.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/statSampler.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "runtime/thread.inline.hpp"
59 #include "runtime/threadCritical.hpp"
60 #include "runtime/timer.hpp"
61 #include "services/attachListener.hpp"
62 #include "services/memTracker.hpp"
63 #include "services/runtimeService.hpp"
64 #include "utilities/decoder.hpp"
65 #include "utilities/defaultStream.hpp"
66 #include "utilities/events.hpp"
67 #include "utilities/elfFile.hpp"
68 #include "utilities/growableArray.hpp"
69 #include "utilities/vmError.hpp"
70
71 // put OS-includes here
72 # include <sys/types.h>
73 # include <sys/mman.h>
74 # include <sys/stat.h>
75 # include <sys/select.h>
76 # include <pthread.h>
77 # include <signal.h>
78 # include <errno.h>
79 # include <dlfcn.h>
80 # include <stdio.h>
81 # include <unistd.h>
82 # include <sys/resource.h>
83 # include <pthread.h>
84 # include <sys/stat.h>
85 # include <sys/time.h>
86 # include <sys/times.h>
87 # include <sys/utsname.h>
88 # include <sys/socket.h>
89 # include <sys/wait.h>
90 # include <pwd.h>
91 # include <poll.h>
92 # include <semaphore.h>
93 # include <fcntl.h>
94 # include <string.h>
95 # include <syscall.h>
96 # include <sys/sysinfo.h>
97 # include <gnu/libc-version.h>
98 # include <sys/ipc.h>
99 # include <sys/shm.h>
100 # include <link.h>
101 # include <stdint.h>
102 # include <inttypes.h>
103 # include <sys/ioctl.h>
104
105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
106
107 #ifndef _GNU_SOURCE
108 #define _GNU_SOURCE
109 #include <sched.h>
110 #undef _GNU_SOURCE
111 #else
112 #include <sched.h>
113 #endif
114
115 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
116 // getrusage() is prepared to handle the associated failure.
117 #ifndef RUSAGE_THREAD
118 #define RUSAGE_THREAD (1) /* only the calling thread */
119 #endif
120
121 #define MAX_PATH (2 * K)
122
123 #define MAX_SECS 100000000
124
125 // for timer info max values which include all bits
126 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
127
128 #define LARGEPAGES_BIT (1 << 6)
129 ////////////////////////////////////////////////////////////////////////////////
130 // global variables
131 julong os::Linux::_physical_memory = 0;
132
133 address os::Linux::_initial_thread_stack_bottom = NULL;
134 uintptr_t os::Linux::_initial_thread_stack_size = 0;
135
136 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
137 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
138 Mutex* os::Linux::_createThread_lock = NULL;
139 pthread_t os::Linux::_main_thread;
140 int os::Linux::_page_size = -1;
141 const int os::Linux::_vm_default_page_size = (8 * K);
142 bool os::Linux::_is_floating_stack = false;
143 bool os::Linux::_is_NPTL = false;
144 bool os::Linux::_supports_fast_thread_cpu_time = false;
145 const char * os::Linux::_glibc_version = NULL;
146 const char * os::Linux::_libpthread_version = NULL;
147 pthread_condattr_t os::Linux::_condattr[1];
148
149 static jlong initial_time_count=0;
150
151 static int clock_tics_per_sec = 100;
152
153 // For diagnostics to print a message once. see run_periodic_checks
154 static sigset_t check_signal_done;
155 static bool check_signals = true;
156
157 static pid_t _initial_pid = 0;
158
159 /* Signal number used to suspend/resume a thread */
160
161 /* do not use any signal number less than SIGSEGV, see 4355769 */
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 /* Used to protect dlsym() calls */
166 static pthread_mutex_t dl_mutex;
167
168 // Declarations
169 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
170
171 // utility functions
172
173 static int SR_initialize();
174
175 julong os::available_memory() {
176 return Linux::available_memory();
177 }
178
179 julong os::Linux::available_memory() {
180 // values in struct sysinfo are "unsigned long"
181 struct sysinfo si;
182 sysinfo(&si);
183
184 return (julong)si.freeram * si.mem_unit;
185 }
186
187 julong os::physical_memory() {
188 return Linux::physical_memory();
189 }
190
191 ////////////////////////////////////////////////////////////////////////////////
192 // environment support
193
194 bool os::getenv(const char* name, char* buf, int len) {
195 const char* val = ::getenv(name);
196 if (val != NULL && strlen(val) < (size_t)len) {
197 strcpy(buf, val);
198 return true;
199 }
200 if (len > 0) buf[0] = 0; // return a null string
201 return false;
202 }
203
204
205 // Return true if user is running as root.
206
207 bool os::have_special_privileges() {
208 static bool init = false;
209 static bool privileges = false;
210 if (!init) {
211 privileges = (getuid() != geteuid()) || (getgid() != getegid());
212 init = true;
213 }
214 return privileges;
215 }
216
217
218 #ifndef SYS_gettid
219 // i386: 224, ia64: 1105, amd64: 186, sparc 143
220 #ifdef __ia64__
221 #define SYS_gettid 1105
222 #else
223 #ifdef __i386__
224 #define SYS_gettid 224
225 #else
226 #ifdef __amd64__
227 #define SYS_gettid 186
228 #else
229 #ifdef __sparc__
230 #define SYS_gettid 143
231 #else
232 #error define gettid for the arch
233 #endif
234 #endif
235 #endif
236 #endif
237 #endif
238
239 // Cpu architecture string
240 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
241
242 // pid_t gettid()
243 //
244 // Returns the kernel thread id of the currently running thread. Kernel
245 // thread id is used to access /proc.
246 //
247 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
248 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
249 //
250 pid_t os::Linux::gettid() {
251 int rslt = syscall(SYS_gettid);
252 if (rslt == -1) {
253 // old kernel, no NPTL support
254 return getpid();
255 } else {
256 return (pid_t)rslt;
257 }
258 }
259
260 // Most versions of linux have a bug where the number of processors are
261 // determined by looking at the /proc file system. In a chroot environment,
262 // the system call returns 1. This causes the VM to act as if it is
263 // a single processor and elide locking (see is_MP() call).
264 static bool unsafe_chroot_detected = false;
265 static const char *unstable_chroot_error = "/proc file system not found.\n"
266 "Java may be unstable running multithreaded in a chroot "
267 "environment on Linux when /proc filesystem is not mounted.";
268
269 void os::Linux::initialize_system_info() {
270 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
271 if (processor_count() == 1) {
272 pid_t pid = os::Linux::gettid();
273 char fname[32];
274 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
275 FILE *fp = fopen(fname, "r");
276 if (fp == NULL) {
277 unsafe_chroot_detected = true;
278 } else {
279 fclose(fp);
280 }
281 }
282 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
283 assert(processor_count() > 0, "linux error");
284 }
285
286 void os::init_system_properties_values() {
287 // The next steps are taken in the product version:
288 //
289 // Obtain the JAVA_HOME value from the location of libjvm.so.
290 // This library should be located at:
291 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
292 //
293 // If "/jre/lib/" appears at the right place in the path, then we
294 // assume libjvm.so is installed in a JDK and we use this path.
295 //
296 // Otherwise exit with message: "Could not create the Java virtual machine."
297 //
298 // The following extra steps are taken in the debugging version:
299 //
300 // If "/jre/lib/" does NOT appear at the right place in the path
301 // instead of exit check for $JAVA_HOME environment variable.
302 //
303 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
304 // then we append a fake suffix "hotspot/libjvm.so" to this path so
305 // it looks like libjvm.so is installed there
306 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
307 //
308 // Otherwise exit.
309 //
310 // Important note: if the location of libjvm.so changes this
311 // code needs to be changed accordingly.
312
313 // See ld(1):
314 // The linker uses the following search paths to locate required
315 // shared libraries:
316 // 1: ...
317 // ...
318 // 7: The default directories, normally /lib and /usr/lib.
319 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
320 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
321 #else
322 #define DEFAULT_LIBPATH "/lib:/usr/lib"
323 #endif
324
325 // Base path of extensions installed on the system.
326 #define SYS_EXT_DIR "/usr/java/packages"
327 #define EXTENSIONS_DIR "/lib/ext"
328 #define ENDORSED_DIR "/lib/endorsed"
329
330 // Buffer that fits several sprintfs.
331 // Note that the space for the colon and the trailing null are provided
332 // by the nulls included by the sizeof operator.
333 const size_t bufsize =
334 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
335 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
336 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
337 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
338
339 // sysclasspath, java_home, dll_dir
340 {
341 char *pslash;
342 os::jvm_path(buf, bufsize);
343
344 // Found the full path to libjvm.so.
345 // Now cut the path to <java_home>/jre if we can.
346 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
347 pslash = strrchr(buf, '/');
348 if (pslash != NULL) {
349 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
350 }
351 Arguments::set_dll_dir(buf);
352
353 if (pslash != NULL) {
354 pslash = strrchr(buf, '/');
355 if (pslash != NULL) {
356 *pslash = '\0'; // Get rid of /<arch>.
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /lib.
360 }
361 }
362 }
363 Arguments::set_java_home(buf);
364 set_boot_path('/', ':');
365 }
366
367 // Where to look for native libraries.
368 //
369 // Note: Due to a legacy implementation, most of the library path
370 // is set in the launcher. This was to accomodate linking restrictions
371 // on legacy Linux implementations (which are no longer supported).
372 // Eventually, all the library path setting will be done here.
373 //
374 // However, to prevent the proliferation of improperly built native
375 // libraries, the new path component /usr/java/packages is added here.
376 // Eventually, all the library path setting will be done here.
377 {
378 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
379 // should always exist (until the legacy problem cited above is
380 // addressed).
381 const char *v = ::getenv("LD_LIBRARY_PATH");
382 const char *v_colon = ":";
383 if (v == NULL) { v = ""; v_colon = ""; }
384 // That's +1 for the colon and +1 for the trailing '\0'.
385 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
386 strlen(v) + 1 +
387 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
388 mtInternal);
389 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
390 Arguments::set_library_path(ld_library_path);
391 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
392 }
393
394 // Extensions directories.
395 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
396 Arguments::set_ext_dirs(buf);
397
398 // Endorsed standards default directory.
399 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
400 Arguments::set_endorsed_dirs(buf);
401
402 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
403
404 #undef DEFAULT_LIBPATH
405 #undef SYS_EXT_DIR
406 #undef EXTENSIONS_DIR
407 #undef ENDORSED_DIR
408 }
409
410 ////////////////////////////////////////////////////////////////////////////////
411 // breakpoint support
412
413 void os::breakpoint() {
414 BREAKPOINT;
415 }
416
417 extern "C" void breakpoint() {
418 // use debugger to set breakpoint here
419 }
420
421 ////////////////////////////////////////////////////////////////////////////////
422 // signal support
423
424 debug_only(static bool signal_sets_initialized = false);
425 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
426
427 bool os::Linux::is_sig_ignored(int sig) {
428 struct sigaction oact;
429 sigaction(sig, (struct sigaction*)NULL, &oact);
430 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
431 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
432 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
433 return true;
434 else
435 return false;
436 }
437
438 void os::Linux::signal_sets_init() {
439 // Should also have an assertion stating we are still single-threaded.
440 assert(!signal_sets_initialized, "Already initialized");
441 // Fill in signals that are necessarily unblocked for all threads in
442 // the VM. Currently, we unblock the following signals:
443 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
444 // by -Xrs (=ReduceSignalUsage));
445 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
446 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
447 // the dispositions or masks wrt these signals.
448 // Programs embedding the VM that want to use the above signals for their
449 // own purposes must, at this time, use the "-Xrs" option to prevent
450 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
451 // (See bug 4345157, and other related bugs).
452 // In reality, though, unblocking these signals is really a nop, since
453 // these signals are not blocked by default.
454 sigemptyset(&unblocked_sigs);
455 sigemptyset(&allowdebug_blocked_sigs);
456 sigaddset(&unblocked_sigs, SIGILL);
457 sigaddset(&unblocked_sigs, SIGSEGV);
458 sigaddset(&unblocked_sigs, SIGBUS);
459 sigaddset(&unblocked_sigs, SIGFPE);
460 #if defined(PPC64)
461 sigaddset(&unblocked_sigs, SIGTRAP);
462 #endif
463 sigaddset(&unblocked_sigs, SR_signum);
464
465 if (!ReduceSignalUsage) {
466 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
468 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
469 }
470 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
471 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
472 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
473 }
474 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
475 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
476 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
477 }
478 }
479 // Fill in signals that are blocked by all but the VM thread.
480 sigemptyset(&vm_sigs);
481 if (!ReduceSignalUsage)
482 sigaddset(&vm_sigs, BREAK_SIGNAL);
483 debug_only(signal_sets_initialized = true);
484
485 }
486
487 // These are signals that are unblocked while a thread is running Java.
488 // (For some reason, they get blocked by default.)
489 sigset_t* os::Linux::unblocked_signals() {
490 assert(signal_sets_initialized, "Not initialized");
491 return &unblocked_sigs;
492 }
493
494 // These are the signals that are blocked while a (non-VM) thread is
495 // running Java. Only the VM thread handles these signals.
496 sigset_t* os::Linux::vm_signals() {
497 assert(signal_sets_initialized, "Not initialized");
498 return &vm_sigs;
499 }
500
501 // These are signals that are blocked during cond_wait to allow debugger in
502 sigset_t* os::Linux::allowdebug_blocked_signals() {
503 assert(signal_sets_initialized, "Not initialized");
504 return &allowdebug_blocked_sigs;
505 }
506
507 void os::Linux::hotspot_sigmask(Thread* thread) {
508
509 //Save caller's signal mask before setting VM signal mask
510 sigset_t caller_sigmask;
511 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
512
513 OSThread* osthread = thread->osthread();
514 osthread->set_caller_sigmask(caller_sigmask);
515
516 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
517
518 if (!ReduceSignalUsage) {
519 if (thread->is_VM_thread()) {
520 // Only the VM thread handles BREAK_SIGNAL ...
521 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
522 } else {
523 // ... all other threads block BREAK_SIGNAL
524 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
525 }
526 }
527 }
528
529 //////////////////////////////////////////////////////////////////////////////
530 // detecting pthread library
531
532 void os::Linux::libpthread_init() {
533 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
534 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
535 // generic name for earlier versions.
536 // Define macros here so we can build HotSpot on old systems.
537 # ifndef _CS_GNU_LIBC_VERSION
538 # define _CS_GNU_LIBC_VERSION 2
539 # endif
540 # ifndef _CS_GNU_LIBPTHREAD_VERSION
541 # define _CS_GNU_LIBPTHREAD_VERSION 3
542 # endif
543
544 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
545 if (n > 0) {
546 char *str = (char *)malloc(n, mtInternal);
547 confstr(_CS_GNU_LIBC_VERSION, str, n);
548 os::Linux::set_glibc_version(str);
549 } else {
550 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
551 static char _gnu_libc_version[32];
552 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
553 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
554 os::Linux::set_glibc_version(_gnu_libc_version);
555 }
556
557 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
558 if (n > 0) {
559 char *str = (char *)malloc(n, mtInternal);
560 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
561 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
562 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
563 // is the case. LinuxThreads has a hard limit on max number of threads.
564 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
565 // On the other hand, NPTL does not have such a limit, sysconf()
566 // will return -1 and errno is not changed. Check if it is really NPTL.
567 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
568 strstr(str, "NPTL") &&
569 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
570 free(str);
571 os::Linux::set_libpthread_version("linuxthreads");
572 } else {
573 os::Linux::set_libpthread_version(str);
574 }
575 } else {
576 // glibc before 2.3.2 only has LinuxThreads.
577 os::Linux::set_libpthread_version("linuxthreads");
578 }
579
580 if (strstr(libpthread_version(), "NPTL")) {
581 os::Linux::set_is_NPTL();
582 } else {
583 os::Linux::set_is_LinuxThreads();
584 }
585
586 // LinuxThreads have two flavors: floating-stack mode, which allows variable
587 // stack size; and fixed-stack mode. NPTL is always floating-stack.
588 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
589 os::Linux::set_is_floating_stack();
590 }
591 }
592
593 /////////////////////////////////////////////////////////////////////////////
594 // thread stack
595
596 // Force Linux kernel to expand current thread stack. If "bottom" is close
597 // to the stack guard, caller should block all signals.
598 //
599 // MAP_GROWSDOWN:
600 // A special mmap() flag that is used to implement thread stacks. It tells
601 // kernel that the memory region should extend downwards when needed. This
602 // allows early versions of LinuxThreads to only mmap the first few pages
603 // when creating a new thread. Linux kernel will automatically expand thread
604 // stack as needed (on page faults).
605 //
606 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
607 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
608 // region, it's hard to tell if the fault is due to a legitimate stack
609 // access or because of reading/writing non-exist memory (e.g. buffer
610 // overrun). As a rule, if the fault happens below current stack pointer,
611 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
612 // application (see Linux kernel fault.c).
613 //
614 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
615 // stack overflow detection.
616 //
617 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
618 // not use this flag. However, the stack of initial thread is not created
619 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
620 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
621 // and then attach the thread to JVM.
622 //
623 // To get around the problem and allow stack banging on Linux, we need to
624 // manually expand thread stack after receiving the SIGSEGV.
625 //
626 // There are two ways to expand thread stack to address "bottom", we used
627 // both of them in JVM before 1.5:
628 // 1. adjust stack pointer first so that it is below "bottom", and then
629 // touch "bottom"
630 // 2. mmap() the page in question
631 //
632 // Now alternate signal stack is gone, it's harder to use 2. For instance,
633 // if current sp is already near the lower end of page 101, and we need to
634 // call mmap() to map page 100, it is possible that part of the mmap() frame
635 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
636 // That will destroy the mmap() frame and cause VM to crash.
637 //
638 // The following code works by adjusting sp first, then accessing the "bottom"
639 // page to force a page fault. Linux kernel will then automatically expand the
640 // stack mapping.
641 //
642 // _expand_stack_to() assumes its frame size is less than page size, which
643 // should always be true if the function is not inlined.
644
645 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
646 #define NOINLINE
647 #else
648 #define NOINLINE __attribute__ ((noinline))
649 #endif
650
651 static void _expand_stack_to(address bottom) NOINLINE;
652
653 static void _expand_stack_to(address bottom) {
654 address sp;
655 size_t size;
656 volatile char *p;
657
658 // Adjust bottom to point to the largest address within the same page, it
659 // gives us a one-page buffer if alloca() allocates slightly more memory.
660 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
661 bottom += os::Linux::page_size() - 1;
662
663 // sp might be slightly above current stack pointer; if that's the case, we
664 // will alloca() a little more space than necessary, which is OK. Don't use
665 // os::current_stack_pointer(), as its result can be slightly below current
666 // stack pointer, causing us to not alloca enough to reach "bottom".
667 sp = (address)&sp;
668
669 if (sp > bottom) {
670 size = sp - bottom;
671 p = (volatile char *)alloca(size);
672 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
673 p[0] = '\0';
674 }
675 }
676
677 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
678 assert(t!=NULL, "just checking");
679 assert(t->osthread()->expanding_stack(), "expand should be set");
680 assert(t->stack_base() != NULL, "stack_base was not initialized");
681
682 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
683 sigset_t mask_all, old_sigset;
684 sigfillset(&mask_all);
685 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
686 _expand_stack_to(addr);
687 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
688 return true;
689 }
690 return false;
691 }
692
693 //////////////////////////////////////////////////////////////////////////////
694 // create new thread
695
696 static address highest_vm_reserved_address();
697
698 // check if it's safe to start a new thread
699 static bool _thread_safety_check(Thread* thread) {
700 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
701 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
702 // Heap is mmap'ed at lower end of memory space. Thread stacks are
703 // allocated (MAP_FIXED) from high address space. Every thread stack
704 // occupies a fixed size slot (usually 2Mbytes, but user can change
705 // it to other values if they rebuild LinuxThreads).
706 //
707 // Problem with MAP_FIXED is that mmap() can still succeed even part of
708 // the memory region has already been mmap'ed. That means if we have too
709 // many threads and/or very large heap, eventually thread stack will
710 // collide with heap.
711 //
712 // Here we try to prevent heap/stack collision by comparing current
713 // stack bottom with the highest address that has been mmap'ed by JVM
714 // plus a safety margin for memory maps created by native code.
715 //
716 // This feature can be disabled by setting ThreadSafetyMargin to 0
717 //
718 if (ThreadSafetyMargin > 0) {
719 address stack_bottom = os::current_stack_base() - os::current_stack_size();
720
721 // not safe if our stack extends below the safety margin
722 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
723 } else {
724 return true;
725 }
726 } else {
727 // Floating stack LinuxThreads or NPTL:
728 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
729 // there's not enough space left, pthread_create() will fail. If we come
730 // here, that means enough space has been reserved for stack.
731 return true;
732 }
733 }
734
735 // Thread start routine for all newly created threads
736 static void *java_start(Thread *thread) {
737 // Try to randomize the cache line index of hot stack frames.
738 // This helps when threads of the same stack traces evict each other's
739 // cache lines. The threads can be either from the same JVM instance, or
740 // from different JVM instances. The benefit is especially true for
741 // processors with hyperthreading technology.
742 static int counter = 0;
743 int pid = os::current_process_id();
744 alloca(((pid ^ counter++) & 7) * 128);
745
746 ThreadLocalStorage::set_thread(thread);
747
748 OSThread* osthread = thread->osthread();
749 Monitor* sync = osthread->startThread_lock();
750
751 // non floating stack LinuxThreads needs extra check, see above
752 if (!_thread_safety_check(thread)) {
753 // notify parent thread
754 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
755 osthread->set_state(ZOMBIE);
756 sync->notify_all();
757 return NULL;
758 }
759
760 // thread_id is kernel thread id (similar to Solaris LWP id)
761 osthread->set_thread_id(os::Linux::gettid());
762
763 if (UseNUMA) {
764 int lgrp_id = os::numa_get_group_id();
765 if (lgrp_id != -1) {
766 thread->set_lgrp_id(lgrp_id);
767 }
768 }
769 // initialize signal mask for this thread
770 os::Linux::hotspot_sigmask(thread);
771
772 // initialize floating point control register
773 os::Linux::init_thread_fpu_state();
774
775 // handshaking with parent thread
776 {
777 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
778
779 // notify parent thread
780 osthread->set_state(INITIALIZED);
781 sync->notify_all();
782
783 // wait until os::start_thread()
784 while (osthread->get_state() == INITIALIZED) {
785 sync->wait(Mutex::_no_safepoint_check_flag);
786 }
787 }
788
789 // call one more level start routine
790 thread->run();
791
792 return 0;
793 }
794
795 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
796 assert(thread->osthread() == NULL, "caller responsible");
797
798 // Allocate the OSThread object
799 OSThread* osthread = new OSThread(NULL, NULL);
800 if (osthread == NULL) {
801 return false;
802 }
803
804 // set the correct thread state
805 osthread->set_thread_type(thr_type);
806
807 // Initial state is ALLOCATED but not INITIALIZED
808 osthread->set_state(ALLOCATED);
809
810 thread->set_osthread(osthread);
811
812 // init thread attributes
813 pthread_attr_t attr;
814 pthread_attr_init(&attr);
815 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
816
817 // stack size
818 if (os::Linux::supports_variable_stack_size()) {
819 // calculate stack size if it's not specified by caller
820 if (stack_size == 0) {
821 stack_size = os::Linux::default_stack_size(thr_type);
822
823 switch (thr_type) {
824 case os::java_thread:
825 // Java threads use ThreadStackSize which default value can be
826 // changed with the flag -Xss
827 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
828 stack_size = JavaThread::stack_size_at_create();
829 break;
830 case os::compiler_thread:
831 if (CompilerThreadStackSize > 0) {
832 stack_size = (size_t)(CompilerThreadStackSize * K);
833 break;
834 } // else fall through:
835 // use VMThreadStackSize if CompilerThreadStackSize is not defined
836 case os::vm_thread:
837 case os::pgc_thread:
838 case os::cgc_thread:
839 case os::watcher_thread:
840 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
841 break;
842 }
843 }
844
845 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
846 pthread_attr_setstacksize(&attr, stack_size);
847 } else {
848 // let pthread_create() pick the default value.
849 }
850
851 // glibc guard page
852 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
853
854 ThreadState state;
855
856 {
857 // Serialize thread creation if we are running with fixed stack LinuxThreads
858 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
859 if (lock) {
860 os::Linux::createThread_lock()->lock_without_safepoint_check();
861 }
862
863 pthread_t tid;
864 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
865
866 pthread_attr_destroy(&attr);
867
868 if (ret != 0) {
869 if (PrintMiscellaneous && (Verbose || WizardMode)) {
870 perror("pthread_create()");
871 }
872 // Need to clean up stuff we've allocated so far
873 thread->set_osthread(NULL);
874 delete osthread;
875 if (lock) os::Linux::createThread_lock()->unlock();
876 return false;
877 }
878
879 // Store pthread info into the OSThread
880 osthread->set_pthread_id(tid);
881
882 // Wait until child thread is either initialized or aborted
883 {
884 Monitor* sync_with_child = osthread->startThread_lock();
885 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
886 while ((state = osthread->get_state()) == ALLOCATED) {
887 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
888 }
889 }
890
891 if (lock) {
892 os::Linux::createThread_lock()->unlock();
893 }
894 }
895
896 // Aborted due to thread limit being reached
897 if (state == ZOMBIE) {
898 thread->set_osthread(NULL);
899 delete osthread;
900 return false;
901 }
902
903 // The thread is returned suspended (in state INITIALIZED),
904 // and is started higher up in the call chain
905 assert(state == INITIALIZED, "race condition");
906 return true;
907 }
908
909 /////////////////////////////////////////////////////////////////////////////
910 // attach existing thread
911
912 // bootstrap the main thread
913 bool os::create_main_thread(JavaThread* thread) {
914 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
915 return create_attached_thread(thread);
916 }
917
918 bool os::create_attached_thread(JavaThread* thread) {
919 #ifdef ASSERT
920 thread->verify_not_published();
921 #endif
922
923 // Allocate the OSThread object
924 OSThread* osthread = new OSThread(NULL, NULL);
925
926 if (osthread == NULL) {
927 return false;
928 }
929
930 // Store pthread info into the OSThread
931 osthread->set_thread_id(os::Linux::gettid());
932 osthread->set_pthread_id(::pthread_self());
933
934 // initialize floating point control register
935 os::Linux::init_thread_fpu_state();
936
937 // Initial thread state is RUNNABLE
938 osthread->set_state(RUNNABLE);
939
940 thread->set_osthread(osthread);
941
942 if (UseNUMA) {
943 int lgrp_id = os::numa_get_group_id();
944 if (lgrp_id != -1) {
945 thread->set_lgrp_id(lgrp_id);
946 }
947 }
948
949 if (os::Linux::is_initial_thread()) {
950 // If current thread is initial thread, its stack is mapped on demand,
951 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
952 // the entire stack region to avoid SEGV in stack banging.
953 // It is also useful to get around the heap-stack-gap problem on SuSE
954 // kernel (see 4821821 for details). We first expand stack to the top
955 // of yellow zone, then enable stack yellow zone (order is significant,
956 // enabling yellow zone first will crash JVM on SuSE Linux), so there
957 // is no gap between the last two virtual memory regions.
958
959 JavaThread *jt = (JavaThread *)thread;
960 address addr = jt->stack_yellow_zone_base();
961 assert(addr != NULL, "initialization problem?");
962 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
963
964 osthread->set_expanding_stack();
965 os::Linux::manually_expand_stack(jt, addr);
966 osthread->clear_expanding_stack();
967 }
968
969 // initialize signal mask for this thread
970 // and save the caller's signal mask
971 os::Linux::hotspot_sigmask(thread);
972
973 return true;
974 }
975
976 void os::pd_start_thread(Thread* thread) {
977 OSThread * osthread = thread->osthread();
978 assert(osthread->get_state() != INITIALIZED, "just checking");
979 Monitor* sync_with_child = osthread->startThread_lock();
980 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
981 sync_with_child->notify();
982 }
983
984 // Free Linux resources related to the OSThread
985 void os::free_thread(OSThread* osthread) {
986 assert(osthread != NULL, "osthread not set");
987
988 if (Thread::current()->osthread() == osthread) {
989 // Restore caller's signal mask
990 sigset_t sigmask = osthread->caller_sigmask();
991 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
992 }
993
994 delete osthread;
995 }
996
997 //////////////////////////////////////////////////////////////////////////////
998 // thread local storage
999
1000 // Restore the thread pointer if the destructor is called. This is in case
1001 // someone from JNI code sets up a destructor with pthread_key_create to run
1002 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1003 // will hang or crash. When detachCurrentThread is called the key will be set
1004 // to null and we will not be called again. If detachCurrentThread is never
1005 // called we could loop forever depending on the pthread implementation.
1006 static void restore_thread_pointer(void* p) {
1007 Thread* thread = (Thread*) p;
1008 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1009 }
1010
1011 int os::allocate_thread_local_storage() {
1012 pthread_key_t key;
1013 int rslt = pthread_key_create(&key, restore_thread_pointer);
1014 assert(rslt == 0, "cannot allocate thread local storage");
1015 return (int)key;
1016 }
1017
1018 // Note: This is currently not used by VM, as we don't destroy TLS key
1019 // on VM exit.
1020 void os::free_thread_local_storage(int index) {
1021 int rslt = pthread_key_delete((pthread_key_t)index);
1022 assert(rslt == 0, "invalid index");
1023 }
1024
1025 void os::thread_local_storage_at_put(int index, void* value) {
1026 int rslt = pthread_setspecific((pthread_key_t)index, value);
1027 assert(rslt == 0, "pthread_setspecific failed");
1028 }
1029
1030 extern "C" Thread* get_thread() {
1031 return ThreadLocalStorage::thread();
1032 }
1033
1034 //////////////////////////////////////////////////////////////////////////////
1035 // initial thread
1036
1037 // Check if current thread is the initial thread, similar to Solaris thr_main.
1038 bool os::Linux::is_initial_thread(void) {
1039 char dummy;
1040 // If called before init complete, thread stack bottom will be null.
1041 // Can be called if fatal error occurs before initialization.
1042 if (initial_thread_stack_bottom() == NULL) return false;
1043 assert(initial_thread_stack_bottom() != NULL &&
1044 initial_thread_stack_size() != 0,
1045 "os::init did not locate initial thread's stack region");
1046 if ((address)&dummy >= initial_thread_stack_bottom() &&
1047 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1048 return true;
1049 else return false;
1050 }
1051
1052 // Find the virtual memory area that contains addr
1053 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1054 FILE *fp = fopen("/proc/self/maps", "r");
1055 if (fp) {
1056 address low, high;
1057 while (!feof(fp)) {
1058 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1059 if (low <= addr && addr < high) {
1060 if (vma_low) *vma_low = low;
1061 if (vma_high) *vma_high = high;
1062 fclose (fp);
1063 return true;
1064 }
1065 }
1066 for (;;) {
1067 int ch = fgetc(fp);
1068 if (ch == EOF || ch == (int)'\n') break;
1069 }
1070 }
1071 fclose(fp);
1072 }
1073 return false;
1074 }
1075
1076 // Locate initial thread stack. This special handling of initial thread stack
1077 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1078 // bogus value for initial thread.
1079 void os::Linux::capture_initial_stack(size_t max_size) {
1080 // stack size is the easy part, get it from RLIMIT_STACK
1081 size_t stack_size;
1082 struct rlimit rlim;
1083 getrlimit(RLIMIT_STACK, &rlim);
1084 stack_size = rlim.rlim_cur;
1085
1086 // 6308388: a bug in ld.so will relocate its own .data section to the
1087 // lower end of primordial stack; reduce ulimit -s value a little bit
1088 // so we won't install guard page on ld.so's data section.
1089 stack_size -= 2 * page_size();
1090
1091 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1092 // 7.1, in both cases we will get 2G in return value.
1093 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1094 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1095 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1096 // in case other parts in glibc still assumes 2M max stack size.
1097 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1098 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1099 if (stack_size > 2 * K * K IA64_ONLY(*2))
1100 stack_size = 2 * K * K IA64_ONLY(*2);
1101 // Try to figure out where the stack base (top) is. This is harder.
1102 //
1103 // When an application is started, glibc saves the initial stack pointer in
1104 // a global variable "__libc_stack_end", which is then used by system
1105 // libraries. __libc_stack_end should be pretty close to stack top. The
1106 // variable is available since the very early days. However, because it is
1107 // a private interface, it could disappear in the future.
1108 //
1109 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1110 // to __libc_stack_end, it is very close to stack top, but isn't the real
1111 // stack top. Note that /proc may not exist if VM is running as a chroot
1112 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1113 // /proc/<pid>/stat could change in the future (though unlikely).
1114 //
1115 // We try __libc_stack_end first. If that doesn't work, look for
1116 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1117 // as a hint, which should work well in most cases.
1118
1119 uintptr_t stack_start;
1120
1121 // try __libc_stack_end first
1122 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1123 if (p && *p) {
1124 stack_start = *p;
1125 } else {
1126 // see if we can get the start_stack field from /proc/self/stat
1127 FILE *fp;
1128 int pid;
1129 char state;
1130 int ppid;
1131 int pgrp;
1132 int session;
1133 int nr;
1134 int tpgrp;
1135 unsigned long flags;
1136 unsigned long minflt;
1137 unsigned long cminflt;
1138 unsigned long majflt;
1139 unsigned long cmajflt;
1140 unsigned long utime;
1141 unsigned long stime;
1142 long cutime;
1143 long cstime;
1144 long prio;
1145 long nice;
1146 long junk;
1147 long it_real;
1148 uintptr_t start;
1149 uintptr_t vsize;
1150 intptr_t rss;
1151 uintptr_t rsslim;
1152 uintptr_t scodes;
1153 uintptr_t ecode;
1154 int i;
1155
1156 // Figure what the primordial thread stack base is. Code is inspired
1157 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1158 // followed by command name surrounded by parentheses, state, etc.
1159 char stat[2048];
1160 int statlen;
1161
1162 fp = fopen("/proc/self/stat", "r");
1163 if (fp) {
1164 statlen = fread(stat, 1, 2047, fp);
1165 stat[statlen] = '\0';
1166 fclose(fp);
1167
1168 // Skip pid and the command string. Note that we could be dealing with
1169 // weird command names, e.g. user could decide to rename java launcher
1170 // to "java 1.4.2 :)", then the stat file would look like
1171 // 1234 (java 1.4.2 :)) R ... ...
1172 // We don't really need to know the command string, just find the last
1173 // occurrence of ")" and then start parsing from there. See bug 4726580.
1174 char * s = strrchr(stat, ')');
1175
1176 i = 0;
1177 if (s) {
1178 // Skip blank chars
1179 do s++; while (isspace(*s));
1180
1181 #define _UFM UINTX_FORMAT
1182 #define _DFM INTX_FORMAT
1183
1184 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1185 /* 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 */
1186 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,
1187 &state, /* 3 %c */
1188 &ppid, /* 4 %d */
1189 &pgrp, /* 5 %d */
1190 &session, /* 6 %d */
1191 &nr, /* 7 %d */
1192 &tpgrp, /* 8 %d */
1193 &flags, /* 9 %lu */
1194 &minflt, /* 10 %lu */
1195 &cminflt, /* 11 %lu */
1196 &majflt, /* 12 %lu */
1197 &cmajflt, /* 13 %lu */
1198 &utime, /* 14 %lu */
1199 &stime, /* 15 %lu */
1200 &cutime, /* 16 %ld */
1201 &cstime, /* 17 %ld */
1202 &prio, /* 18 %ld */
1203 &nice, /* 19 %ld */
1204 &junk, /* 20 %ld */
1205 &it_real, /* 21 %ld */
1206 &start, /* 22 UINTX_FORMAT */
1207 &vsize, /* 23 UINTX_FORMAT */
1208 &rss, /* 24 INTX_FORMAT */
1209 &rsslim, /* 25 UINTX_FORMAT */
1210 &scodes, /* 26 UINTX_FORMAT */
1211 &ecode, /* 27 UINTX_FORMAT */
1212 &stack_start); /* 28 UINTX_FORMAT */
1213 }
1214
1215 #undef _UFM
1216 #undef _DFM
1217
1218 if (i != 28 - 2) {
1219 assert(false, "Bad conversion from /proc/self/stat");
1220 // product mode - assume we are the initial thread, good luck in the
1221 // embedded case.
1222 warning("Can't detect initial thread stack location - bad conversion");
1223 stack_start = (uintptr_t) &rlim;
1224 }
1225 } else {
1226 // For some reason we can't open /proc/self/stat (for example, running on
1227 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1228 // most cases, so don't abort:
1229 warning("Can't detect initial thread stack location - no /proc/self/stat");
1230 stack_start = (uintptr_t) &rlim;
1231 }
1232 }
1233
1234 // Now we have a pointer (stack_start) very close to the stack top, the
1235 // next thing to do is to figure out the exact location of stack top. We
1236 // can find out the virtual memory area that contains stack_start by
1237 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1238 // and its upper limit is the real stack top. (again, this would fail if
1239 // running inside chroot, because /proc may not exist.)
1240
1241 uintptr_t stack_top;
1242 address low, high;
1243 if (find_vma((address)stack_start, &low, &high)) {
1244 // success, "high" is the true stack top. (ignore "low", because initial
1245 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1246 stack_top = (uintptr_t)high;
1247 } else {
1248 // failed, likely because /proc/self/maps does not exist
1249 warning("Can't detect initial thread stack location - find_vma failed");
1250 // best effort: stack_start is normally within a few pages below the real
1251 // stack top, use it as stack top, and reduce stack size so we won't put
1252 // guard page outside stack.
1253 stack_top = stack_start;
1254 stack_size -= 16 * page_size();
1255 }
1256
1257 // stack_top could be partially down the page so align it
1258 stack_top = align_size_up(stack_top, page_size());
1259
1260 if (max_size && stack_size > max_size) {
1261 _initial_thread_stack_size = max_size;
1262 } else {
1263 _initial_thread_stack_size = stack_size;
1264 }
1265
1266 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1267 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1268 }
1269
1270 ////////////////////////////////////////////////////////////////////////////////
1271 // time support
1272
1273 // Time since start-up in seconds to a fine granularity.
1274 // Used by VMSelfDestructTimer and the MemProfiler.
1275 double os::elapsedTime() {
1276
1277 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1278 }
1279
1280 jlong os::elapsed_counter() {
1281 return javaTimeNanos() - initial_time_count;
1282 }
1283
1284 jlong os::elapsed_frequency() {
1285 return NANOSECS_PER_SEC; // nanosecond resolution
1286 }
1287
1288 bool os::supports_vtime() { return true; }
1289 bool os::enable_vtime() { return false; }
1290 bool os::vtime_enabled() { return false; }
1291
1292 double os::elapsedVTime() {
1293 struct rusage usage;
1294 int retval = getrusage(RUSAGE_THREAD, &usage);
1295 if (retval == 0) {
1296 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1297 } else {
1298 // better than nothing, but not much
1299 return elapsedTime();
1300 }
1301 }
1302
1303 jlong os::javaTimeMillis() {
1304 timeval time;
1305 int status = gettimeofday(&time, NULL);
1306 assert(status != -1, "linux error");
1307 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1308 }
1309
1310 #ifndef CLOCK_MONOTONIC
1311 #define CLOCK_MONOTONIC (1)
1312 #endif
1313
1314 void os::Linux::clock_init() {
1315 // we do dlopen's in this particular order due to bug in linux
1316 // dynamical loader (see 6348968) leading to crash on exit
1317 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1318 if (handle == NULL) {
1319 handle = dlopen("librt.so", RTLD_LAZY);
1320 }
1321
1322 if (handle) {
1323 int (*clock_getres_func)(clockid_t, struct timespec*) =
1324 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1325 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1326 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1327 if (clock_getres_func && clock_gettime_func) {
1328 // See if monotonic clock is supported by the kernel. Note that some
1329 // early implementations simply return kernel jiffies (updated every
1330 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1331 // for nano time (though the monotonic property is still nice to have).
1332 // It's fixed in newer kernels, however clock_getres() still returns
1333 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1334 // resolution for now. Hopefully as people move to new kernels, this
1335 // won't be a problem.
1336 struct timespec res;
1337 struct timespec tp;
1338 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1339 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1340 // yes, monotonic clock is supported
1341 _clock_gettime = clock_gettime_func;
1342 return;
1343 } else {
1344 // close librt if there is no monotonic clock
1345 dlclose(handle);
1346 }
1347 }
1348 }
1349 warning("No monotonic clock was available - timed services may " \
1350 "be adversely affected if the time-of-day clock changes");
1351 }
1352
1353 #ifndef SYS_clock_getres
1354
1355 #if defined(IA32) || defined(AMD64)
1356 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1357 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1358 #else
1359 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1360 #define sys_clock_getres(x,y) -1
1361 #endif
1362
1363 #else
1364 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1365 #endif
1366
1367 void os::Linux::fast_thread_clock_init() {
1368 if (!UseLinuxPosixThreadCPUClocks) {
1369 return;
1370 }
1371 clockid_t clockid;
1372 struct timespec tp;
1373 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1374 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1375
1376 // Switch to using fast clocks for thread cpu time if
1377 // the sys_clock_getres() returns 0 error code.
1378 // Note, that some kernels may support the current thread
1379 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1380 // returned by the pthread_getcpuclockid().
1381 // If the fast Posix clocks are supported then the sys_clock_getres()
1382 // must return at least tp.tv_sec == 0 which means a resolution
1383 // better than 1 sec. This is extra check for reliability.
1384
1385 if(pthread_getcpuclockid_func &&
1386 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1387 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1388
1389 _supports_fast_thread_cpu_time = true;
1390 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1391 }
1392 }
1393
1394 jlong os::javaTimeNanos() {
1395 if (Linux::supports_monotonic_clock()) {
1396 struct timespec tp;
1397 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1398 assert(status == 0, "gettime error");
1399 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1400 return result;
1401 } else {
1402 timeval time;
1403 int status = gettimeofday(&time, NULL);
1404 assert(status != -1, "linux error");
1405 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1406 return 1000 * usecs;
1407 }
1408 }
1409
1410 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1411 if (Linux::supports_monotonic_clock()) {
1412 info_ptr->max_value = ALL_64_BITS;
1413
1414 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1415 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1416 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1417 } else {
1418 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1419 info_ptr->max_value = ALL_64_BITS;
1420
1421 // gettimeofday is a real time clock so it skips
1422 info_ptr->may_skip_backward = true;
1423 info_ptr->may_skip_forward = true;
1424 }
1425
1426 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1427 }
1428
1429 // Return the real, user, and system times in seconds from an
1430 // arbitrary fixed point in the past.
1431 bool os::getTimesSecs(double* process_real_time,
1432 double* process_user_time,
1433 double* process_system_time) {
1434 struct tms ticks;
1435 clock_t real_ticks = times(&ticks);
1436
1437 if (real_ticks == (clock_t) (-1)) {
1438 return false;
1439 } else {
1440 double ticks_per_second = (double) clock_tics_per_sec;
1441 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1442 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1443 *process_real_time = ((double) real_ticks) / ticks_per_second;
1444
1445 return true;
1446 }
1447 }
1448
1449
1450 char * os::local_time_string(char *buf, size_t buflen) {
1451 struct tm t;
1452 time_t long_time;
1453 time(&long_time);
1454 localtime_r(&long_time, &t);
1455 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1456 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1457 t.tm_hour, t.tm_min, t.tm_sec);
1458 return buf;
1459 }
1460
1461 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1462 return localtime_r(clock, res);
1463 }
1464
1465 ////////////////////////////////////////////////////////////////////////////////
1466 // runtime exit support
1467
1468 // Note: os::shutdown() might be called very early during initialization, or
1469 // called from signal handler. Before adding something to os::shutdown(), make
1470 // sure it is async-safe and can handle partially initialized VM.
1471 void os::shutdown() {
1472
1473 // allow PerfMemory to attempt cleanup of any persistent resources
1474 perfMemory_exit();
1475
1476 // needs to remove object in file system
1477 AttachListener::abort();
1478
1479 // flush buffered output, finish log files
1480 ostream_abort();
1481
1482 // Check for abort hook
1483 abort_hook_t abort_hook = Arguments::abort_hook();
1484 if (abort_hook != NULL) {
1485 abort_hook();
1486 }
1487
1488 }
1489
1490 // Note: os::abort() might be called very early during initialization, or
1491 // called from signal handler. Before adding something to os::abort(), make
1492 // sure it is async-safe and can handle partially initialized VM.
1493 void os::abort(bool dump_core) {
1494 os::shutdown();
1495 if (dump_core) {
1496 #ifndef PRODUCT
1497 fdStream out(defaultStream::output_fd());
1498 out.print_raw("Current thread is ");
1499 char buf[16];
1500 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1501 out.print_raw_cr(buf);
1502 out.print_raw_cr("Dumping core ...");
1503 #endif
1504 ::abort(); // dump core
1505 }
1506
1507 ::exit(1);
1508 }
1509
1510 // Die immediately, no exit hook, no abort hook, no cleanup.
1511 void os::die() {
1512 // _exit() on LinuxThreads only kills current thread
1513 ::abort();
1514 }
1515
1516
1517 // This method is a copy of JDK's sysGetLastErrorString
1518 // from src/solaris/hpi/src/system_md.c
1519
1520 size_t os::lasterror(char *buf, size_t len) {
1521
1522 if (errno == 0) return 0;
1523
1524 const char *s = ::strerror(errno);
1525 size_t n = ::strlen(s);
1526 if (n >= len) {
1527 n = len - 1;
1528 }
1529 ::strncpy(buf, s, n);
1530 buf[n] = '\0';
1531 return n;
1532 }
1533
1534 intx os::current_thread_id() { return (intx)pthread_self(); }
1535 int os::current_process_id() {
1536
1537 // Under the old linux thread library, linux gives each thread
1538 // its own process id. Because of this each thread will return
1539 // a different pid if this method were to return the result
1540 // of getpid(2). Linux provides no api that returns the pid
1541 // of the launcher thread for the vm. This implementation
1542 // returns a unique pid, the pid of the launcher thread
1543 // that starts the vm 'process'.
1544
1545 // Under the NPTL, getpid() returns the same pid as the
1546 // launcher thread rather than a unique pid per thread.
1547 // Use gettid() if you want the old pre NPTL behaviour.
1548
1549 // if you are looking for the result of a call to getpid() that
1550 // returns a unique pid for the calling thread, then look at the
1551 // OSThread::thread_id() method in osThread_linux.hpp file
1552
1553 return (int)(_initial_pid ? _initial_pid : getpid());
1554 }
1555
1556 // DLL functions
1557
1558 const char* os::dll_file_extension() { return ".so"; }
1559
1560 // This must be hard coded because it's the system's temporary
1561 // directory not the java application's temp directory, ala java.io.tmpdir.
1562 const char* os::get_temp_directory() { return "/tmp"; }
1563
1564 static bool file_exists(const char* filename) {
1565 struct stat statbuf;
1566 if (filename == NULL || strlen(filename) == 0) {
1567 return false;
1568 }
1569 return os::stat(filename, &statbuf) == 0;
1570 }
1571
1572 bool os::dll_build_name(char* buffer, size_t buflen,
1573 const char* pname, const char* fname) {
1574 bool retval = false;
1575 // Copied from libhpi
1576 const size_t pnamelen = pname ? strlen(pname) : 0;
1577
1578 // Return error on buffer overflow.
1579 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1580 return retval;
1581 }
1582
1583 if (pnamelen == 0) {
1584 snprintf(buffer, buflen, "lib%s.so", fname);
1585 retval = true;
1586 } else if (strchr(pname, *os::path_separator()) != NULL) {
1587 int n;
1588 char** pelements = split_path(pname, &n);
1589 if (pelements == NULL) {
1590 return false;
1591 }
1592 for (int i = 0 ; i < n ; i++) {
1593 // Really shouldn't be NULL, but check can't hurt
1594 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1595 continue; // skip the empty path values
1596 }
1597 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1598 if (file_exists(buffer)) {
1599 retval = true;
1600 break;
1601 }
1602 }
1603 // release the storage
1604 for (int i = 0 ; i < n ; i++) {
1605 if (pelements[i] != NULL) {
1606 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1607 }
1608 }
1609 if (pelements != NULL) {
1610 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1611 }
1612 } else {
1613 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1614 retval = true;
1615 }
1616 return retval;
1617 }
1618
1619 // check if addr is inside libjvm.so
1620 bool os::address_is_in_vm(address addr) {
1621 static address libjvm_base_addr;
1622 Dl_info dlinfo;
1623
1624 if (libjvm_base_addr == NULL) {
1625 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1626 libjvm_base_addr = (address)dlinfo.dli_fbase;
1627 }
1628 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1629 }
1630
1631 if (dladdr((void *)addr, &dlinfo) != 0) {
1632 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1633 }
1634
1635 return false;
1636 }
1637
1638 bool os::dll_address_to_function_name(address addr, char *buf,
1639 int buflen, int *offset) {
1640 // buf is not optional, but offset is optional
1641 assert(buf != NULL, "sanity check");
1642
1643 Dl_info dlinfo;
1644
1645 if (dladdr((void*)addr, &dlinfo) != 0) {
1646 // see if we have a matching symbol
1647 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1648 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1649 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1650 }
1651 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1652 return true;
1653 }
1654 // no matching symbol so try for just file info
1655 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1656 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1657 buf, buflen, offset, dlinfo.dli_fname)) {
1658 return true;
1659 }
1660 }
1661 }
1662
1663 buf[0] = '\0';
1664 if (offset != NULL) *offset = -1;
1665 return false;
1666 }
1667
1668 struct _address_to_library_name {
1669 address addr; // input : memory address
1670 size_t buflen; // size of fname
1671 char* fname; // output: library name
1672 address base; // library base addr
1673 };
1674
1675 static int address_to_library_name_callback(struct dl_phdr_info *info,
1676 size_t size, void *data) {
1677 int i;
1678 bool found = false;
1679 address libbase = NULL;
1680 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1681
1682 // iterate through all loadable segments
1683 for (i = 0; i < info->dlpi_phnum; i++) {
1684 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1685 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1686 // base address of a library is the lowest address of its loaded
1687 // segments.
1688 if (libbase == NULL || libbase > segbase) {
1689 libbase = segbase;
1690 }
1691 // see if 'addr' is within current segment
1692 if (segbase <= d->addr &&
1693 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1694 found = true;
1695 }
1696 }
1697 }
1698
1699 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1700 // so dll_address_to_library_name() can fall through to use dladdr() which
1701 // can figure out executable name from argv[0].
1702 if (found && info->dlpi_name && info->dlpi_name[0]) {
1703 d->base = libbase;
1704 if (d->fname) {
1705 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1706 }
1707 return 1;
1708 }
1709 return 0;
1710 }
1711
1712 bool os::dll_address_to_library_name(address addr, char* buf,
1713 int buflen, int* offset) {
1714 // buf is not optional, but offset is optional
1715 assert(buf != NULL, "sanity check");
1716
1717 Dl_info dlinfo;
1718 struct _address_to_library_name data;
1719
1720 // There is a bug in old glibc dladdr() implementation that it could resolve
1721 // to wrong library name if the .so file has a base address != NULL. Here
1722 // we iterate through the program headers of all loaded libraries to find
1723 // out which library 'addr' really belongs to. This workaround can be
1724 // removed once the minimum requirement for glibc is moved to 2.3.x.
1725 data.addr = addr;
1726 data.fname = buf;
1727 data.buflen = buflen;
1728 data.base = NULL;
1729 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1730
1731 if (rslt) {
1732 // buf already contains library name
1733 if (offset) *offset = addr - data.base;
1734 return true;
1735 }
1736 if (dladdr((void*)addr, &dlinfo) != 0) {
1737 if (dlinfo.dli_fname != NULL) {
1738 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1739 }
1740 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1741 *offset = addr - (address)dlinfo.dli_fbase;
1742 }
1743 return true;
1744 }
1745
1746 buf[0] = '\0';
1747 if (offset) *offset = -1;
1748 return false;
1749 }
1750
1751 // Loads .dll/.so and
1752 // in case of error it checks if .dll/.so was built for the
1753 // same architecture as Hotspot is running on
1754
1755
1756 // Remember the stack's state. The Linux dynamic linker will change
1757 // the stack to 'executable' at most once, so we must safepoint only once.
1758 bool os::Linux::_stack_is_executable = false;
1759
1760 // VM operation that loads a library. This is necessary if stack protection
1761 // of the Java stacks can be lost during loading the library. If we
1762 // do not stop the Java threads, they can stack overflow before the stacks
1763 // are protected again.
1764 class VM_LinuxDllLoad: public VM_Operation {
1765 private:
1766 const char *_filename;
1767 char *_ebuf;
1768 int _ebuflen;
1769 void *_lib;
1770 public:
1771 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1772 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1773 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1774 void doit() {
1775 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1776 os::Linux::_stack_is_executable = true;
1777 }
1778 void* loaded_library() { return _lib; }
1779 };
1780
1781 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1782 {
1783 void * result = NULL;
1784 bool load_attempted = false;
1785
1786 // Check whether the library to load might change execution rights
1787 // of the stack. If they are changed, the protection of the stack
1788 // guard pages will be lost. We need a safepoint to fix this.
1789 //
1790 // See Linux man page execstack(8) for more info.
1791 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1792 ElfFile ef(filename);
1793 if (!ef.specifies_noexecstack()) {
1794 if (!is_init_completed()) {
1795 os::Linux::_stack_is_executable = true;
1796 // This is OK - No Java threads have been created yet, and hence no
1797 // stack guard pages to fix.
1798 //
1799 // This should happen only when you are building JDK7 using a very
1800 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1801 //
1802 // Dynamic loader will make all stacks executable after
1803 // this function returns, and will not do that again.
1804 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1805 } else {
1806 warning("You have loaded library %s which might have disabled stack guard. "
1807 "The VM will try to fix the stack guard now.\n"
1808 "It's highly recommended that you fix the library with "
1809 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1810 filename);
1811
1812 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1813 JavaThread *jt = JavaThread::current();
1814 if (jt->thread_state() != _thread_in_native) {
1815 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1816 // that requires ExecStack. Cannot enter safe point. Let's give up.
1817 warning("Unable to fix stack guard. Giving up.");
1818 } else {
1819 if (!LoadExecStackDllInVMThread) {
1820 // This is for the case where the DLL has an static
1821 // constructor function that executes JNI code. We cannot
1822 // load such DLLs in the VMThread.
1823 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1824 }
1825
1826 ThreadInVMfromNative tiv(jt);
1827 debug_only(VMNativeEntryWrapper vew;)
1828
1829 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1830 VMThread::execute(&op);
1831 if (LoadExecStackDllInVMThread) {
1832 result = op.loaded_library();
1833 }
1834 load_attempted = true;
1835 }
1836 }
1837 }
1838 }
1839
1840 if (!load_attempted) {
1841 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1842 }
1843
1844 if (result != NULL) {
1845 // Successful loading
1846 return result;
1847 }
1848
1849 Elf32_Ehdr elf_head;
1850 int diag_msg_max_length=ebuflen-strlen(ebuf);
1851 char* diag_msg_buf=ebuf+strlen(ebuf);
1852
1853 if (diag_msg_max_length==0) {
1854 // No more space in ebuf for additional diagnostics message
1855 return NULL;
1856 }
1857
1858
1859 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1860
1861 if (file_descriptor < 0) {
1862 // Can't open library, report dlerror() message
1863 return NULL;
1864 }
1865
1866 bool failed_to_read_elf_head=
1867 (sizeof(elf_head)!=
1868 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1869
1870 ::close(file_descriptor);
1871 if (failed_to_read_elf_head) {
1872 // file i/o error - report dlerror() msg
1873 return NULL;
1874 }
1875
1876 typedef struct {
1877 Elf32_Half code; // Actual value as defined in elf.h
1878 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1879 char elf_class; // 32 or 64 bit
1880 char endianess; // MSB or LSB
1881 char* name; // String representation
1882 } arch_t;
1883
1884 #ifndef EM_486
1885 #define EM_486 6 /* Intel 80486 */
1886 #endif
1887
1888 static const arch_t arch_array[]={
1889 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1890 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1891 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1892 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1893 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1894 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1895 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1896 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1897 #if defined(VM_LITTLE_ENDIAN)
1898 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1899 #else
1900 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1901 #endif
1902 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1903 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1904 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1905 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1906 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1907 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1908 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1909 };
1910
1911 #if (defined IA32)
1912 static Elf32_Half running_arch_code=EM_386;
1913 #elif (defined AMD64)
1914 static Elf32_Half running_arch_code=EM_X86_64;
1915 #elif (defined IA64)
1916 static Elf32_Half running_arch_code=EM_IA_64;
1917 #elif (defined __sparc) && (defined _LP64)
1918 static Elf32_Half running_arch_code=EM_SPARCV9;
1919 #elif (defined __sparc) && (!defined _LP64)
1920 static Elf32_Half running_arch_code=EM_SPARC;
1921 #elif (defined __powerpc64__)
1922 static Elf32_Half running_arch_code=EM_PPC64;
1923 #elif (defined __powerpc__)
1924 static Elf32_Half running_arch_code=EM_PPC;
1925 #elif (defined ARM)
1926 static Elf32_Half running_arch_code=EM_ARM;
1927 #elif (defined S390)
1928 static Elf32_Half running_arch_code=EM_S390;
1929 #elif (defined ALPHA)
1930 static Elf32_Half running_arch_code=EM_ALPHA;
1931 #elif (defined MIPSEL)
1932 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1933 #elif (defined PARISC)
1934 static Elf32_Half running_arch_code=EM_PARISC;
1935 #elif (defined MIPS)
1936 static Elf32_Half running_arch_code=EM_MIPS;
1937 #elif (defined M68K)
1938 static Elf32_Half running_arch_code=EM_68K;
1939 #else
1940 #error Method os::dll_load requires that one of following is defined:\
1941 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1942 #endif
1943
1944 // Identify compatability class for VM's architecture and library's architecture
1945 // Obtain string descriptions for architectures
1946
1947 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1948 int running_arch_index=-1;
1949
1950 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1951 if (running_arch_code == arch_array[i].code) {
1952 running_arch_index = i;
1953 }
1954 if (lib_arch.code == arch_array[i].code) {
1955 lib_arch.compat_class = arch_array[i].compat_class;
1956 lib_arch.name = arch_array[i].name;
1957 }
1958 }
1959
1960 assert(running_arch_index != -1,
1961 "Didn't find running architecture code (running_arch_code) in arch_array");
1962 if (running_arch_index == -1) {
1963 // Even though running architecture detection failed
1964 // we may still continue with reporting dlerror() message
1965 return NULL;
1966 }
1967
1968 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1969 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1970 return NULL;
1971 }
1972
1973 #ifndef S390
1974 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1975 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1976 return NULL;
1977 }
1978 #endif // !S390
1979
1980 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1981 if ( lib_arch.name!=NULL ) {
1982 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1983 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1984 lib_arch.name, arch_array[running_arch_index].name);
1985 } else {
1986 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1987 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1988 lib_arch.code,
1989 arch_array[running_arch_index].name);
1990 }
1991 }
1992
1993 return NULL;
1994 }
1995
1996 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
1997 void * result = ::dlopen(filename, RTLD_LAZY);
1998 if (result == NULL) {
1999 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2000 ebuf[ebuflen-1] = '\0';
2001 }
2002 return result;
2003 }
2004
2005 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2006 void * result = NULL;
2007 if (LoadExecStackDllInVMThread) {
2008 result = dlopen_helper(filename, ebuf, ebuflen);
2009 }
2010
2011 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2012 // library that requires an executable stack, or which does not have this
2013 // stack attribute set, dlopen changes the stack attribute to executable. The
2014 // read protection of the guard pages gets lost.
2015 //
2016 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2017 // may have been queued at the same time.
2018
2019 if (!_stack_is_executable) {
2020 JavaThread *jt = Threads::first();
2021
2022 while (jt) {
2023 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2024 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2025 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2026 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2027 warning("Attempt to reguard stack yellow zone failed.");
2028 }
2029 }
2030 jt = jt->next();
2031 }
2032 }
2033
2034 return result;
2035 }
2036
2037 /*
2038 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2039 * chances are you might want to run the generated bits against glibc-2.0
2040 * libdl.so, so always use locking for any version of glibc.
2041 */
2042 void* os::dll_lookup(void* handle, const char* name) {
2043 pthread_mutex_lock(&dl_mutex);
2044 void* res = dlsym(handle, name);
2045 pthread_mutex_unlock(&dl_mutex);
2046 return res;
2047 }
2048
2049 void* os::get_default_process_handle() {
2050 return (void*)::dlopen(NULL, RTLD_LAZY);
2051 }
2052
2053 static bool _print_ascii_file(const char* filename, outputStream* st) {
2054 int fd = ::open(filename, O_RDONLY);
2055 if (fd == -1) {
2056 return false;
2057 }
2058
2059 char buf[32];
2060 int bytes;
2061 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2062 st->print_raw(buf, bytes);
2063 }
2064
2065 ::close(fd);
2066
2067 return true;
2068 }
2069
2070 void os::print_dll_info(outputStream *st) {
2071 st->print_cr("Dynamic libraries:");
2072
2073 char fname[32];
2074 pid_t pid = os::Linux::gettid();
2075
2076 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2077
2078 if (!_print_ascii_file(fname, st)) {
2079 st->print("Can not get library information for pid = %d\n", pid);
2080 }
2081 }
2082
2083 void os::print_os_info_brief(outputStream* st) {
2084 os::Linux::print_distro_info(st);
2085
2086 os::Posix::print_uname_info(st);
2087
2088 os::Linux::print_libversion_info(st);
2089
2090 }
2091
2092 void os::print_os_info(outputStream* st) {
2093 st->print("OS:");
2094
2095 os::Linux::print_distro_info(st);
2096
2097 os::Posix::print_uname_info(st);
2098
2099 // Print warning if unsafe chroot environment detected
2100 if (unsafe_chroot_detected) {
2101 st->print("WARNING!! ");
2102 st->print_cr("%s", unstable_chroot_error);
2103 }
2104
2105 os::Linux::print_libversion_info(st);
2106
2107 os::Posix::print_rlimit_info(st);
2108
2109 os::Posix::print_load_average(st);
2110
2111 os::Linux::print_full_memory_info(st);
2112 }
2113
2114 // Try to identify popular distros.
2115 // Most Linux distributions have a /etc/XXX-release file, which contains
2116 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2117 // file that also contains the OS version string. Some have more than one
2118 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2119 // /etc/redhat-release.), so the order is important.
2120 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2121 // their own specific XXX-release file as well as a redhat-release file.
2122 // Because of this the XXX-release file needs to be searched for before the
2123 // redhat-release file.
2124 // Since Red Hat has a lsb-release file that is not very descriptive the
2125 // search for redhat-release needs to be before lsb-release.
2126 // Since the lsb-release file is the new standard it needs to be searched
2127 // before the older style release files.
2128 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2129 // next to last resort. The os-release file is a new standard that contains
2130 // distribution information and the system-release file seems to be an old
2131 // standard that has been replaced by the lsb-release and os-release files.
2132 // Searching for the debian_version file is the last resort. It contains
2133 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2134 // "Debian " is printed before the contents of the debian_version file.
2135 void os::Linux::print_distro_info(outputStream* st) {
2136 if (!_print_ascii_file("/etc/oracle-release", st) &&
2137 !_print_ascii_file("/etc/mandriva-release", st) &&
2138 !_print_ascii_file("/etc/mandrake-release", st) &&
2139 !_print_ascii_file("/etc/sun-release", st) &&
2140 !_print_ascii_file("/etc/redhat-release", st) &&
2141 !_print_ascii_file("/etc/lsb-release", st) &&
2142 !_print_ascii_file("/etc/SuSE-release", st) &&
2143 !_print_ascii_file("/etc/turbolinux-release", st) &&
2144 !_print_ascii_file("/etc/gentoo-release", st) &&
2145 !_print_ascii_file("/etc/ltib-release", st) &&
2146 !_print_ascii_file("/etc/angstrom-version", st) &&
2147 !_print_ascii_file("/etc/system-release", st) &&
2148 !_print_ascii_file("/etc/os-release", st)) {
2149
2150 if (file_exists("/etc/debian_version")) {
2151 st->print("Debian ");
2152 _print_ascii_file("/etc/debian_version", st);
2153 } else {
2154 st->print("Linux");
2155 }
2156 }
2157 st->cr();
2158 }
2159
2160 void os::Linux::print_libversion_info(outputStream* st) {
2161 // libc, pthread
2162 st->print("libc:");
2163 st->print("%s ", os::Linux::glibc_version());
2164 st->print("%s ", os::Linux::libpthread_version());
2165 if (os::Linux::is_LinuxThreads()) {
2166 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2167 }
2168 st->cr();
2169 }
2170
2171 void os::Linux::print_full_memory_info(outputStream* st) {
2172 st->print("\n/proc/meminfo:\n");
2173 _print_ascii_file("/proc/meminfo", st);
2174 st->cr();
2175 }
2176
2177 void os::print_memory_info(outputStream* st) {
2178
2179 st->print("Memory:");
2180 st->print(" %dk page", os::vm_page_size()>>10);
2181
2182 // values in struct sysinfo are "unsigned long"
2183 struct sysinfo si;
2184 sysinfo(&si);
2185
2186 st->print(", physical " UINT64_FORMAT "k",
2187 os::physical_memory() >> 10);
2188 st->print("(" UINT64_FORMAT "k free)",
2189 os::available_memory() >> 10);
2190 st->print(", swap " UINT64_FORMAT "k",
2191 ((jlong)si.totalswap * si.mem_unit) >> 10);
2192 st->print("(" UINT64_FORMAT "k free)",
2193 ((jlong)si.freeswap * si.mem_unit) >> 10);
2194 st->cr();
2195 }
2196
2197 void os::pd_print_cpu_info(outputStream* st) {
2198 st->print("\n/proc/cpuinfo:\n");
2199 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2200 st->print(" <Not Available>");
2201 }
2202 st->cr();
2203 }
2204
2205 void os::print_siginfo(outputStream* st, void* siginfo) {
2206 const siginfo_t* si = (const siginfo_t*)siginfo;
2207
2208 os::Posix::print_siginfo_brief(st, si);
2209 #if INCLUDE_CDS
2210 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2211 UseSharedSpaces) {
2212 FileMapInfo* mapinfo = FileMapInfo::current_info();
2213 if (mapinfo->is_in_shared_space(si->si_addr)) {
2214 st->print("\n\nError accessing class data sharing archive." \
2215 " Mapped file inaccessible during execution, " \
2216 " possible disk/network problem.");
2217 }
2218 }
2219 #endif
2220 st->cr();
2221 }
2222
2223
2224 static void print_signal_handler(outputStream* st, int sig,
2225 char* buf, size_t buflen);
2226
2227 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2228 st->print_cr("Signal Handlers:");
2229 print_signal_handler(st, SIGSEGV, buf, buflen);
2230 print_signal_handler(st, SIGBUS , buf, buflen);
2231 print_signal_handler(st, SIGFPE , buf, buflen);
2232 print_signal_handler(st, SIGPIPE, buf, buflen);
2233 print_signal_handler(st, SIGXFSZ, buf, buflen);
2234 print_signal_handler(st, SIGILL , buf, buflen);
2235 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2236 print_signal_handler(st, SR_signum, buf, buflen);
2237 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2238 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2239 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2240 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2241 #if defined(PPC64)
2242 print_signal_handler(st, SIGTRAP, buf, buflen);
2243 #endif
2244 }
2245
2246 static char saved_jvm_path[MAXPATHLEN] = {0};
2247
2248 // Find the full path to the current module, libjvm.so
2249 void os::jvm_path(char *buf, jint buflen) {
2250 // Error checking.
2251 if (buflen < MAXPATHLEN) {
2252 assert(false, "must use a large-enough buffer");
2253 buf[0] = '\0';
2254 return;
2255 }
2256 // Lazy resolve the path to current module.
2257 if (saved_jvm_path[0] != 0) {
2258 strcpy(buf, saved_jvm_path);
2259 return;
2260 }
2261
2262 char dli_fname[MAXPATHLEN];
2263 bool ret = dll_address_to_library_name(
2264 CAST_FROM_FN_PTR(address, os::jvm_path),
2265 dli_fname, sizeof(dli_fname), NULL);
2266 assert(ret, "cannot locate libjvm");
2267 char *rp = NULL;
2268 if (ret && dli_fname[0] != '\0') {
2269 rp = realpath(dli_fname, buf);
2270 }
2271 if (rp == NULL)
2272 return;
2273
2274 if (Arguments::created_by_gamma_launcher()) {
2275 // Support for the gamma launcher. Typical value for buf is
2276 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2277 // the right place in the string, then assume we are installed in a JDK and
2278 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2279 // up the path so it looks like libjvm.so is installed there (append a
2280 // fake suffix hotspot/libjvm.so).
2281 const char *p = buf + strlen(buf) - 1;
2282 for (int count = 0; p > buf && count < 5; ++count) {
2283 for (--p; p > buf && *p != '/'; --p)
2284 /* empty */ ;
2285 }
2286
2287 if (strncmp(p, "/jre/lib/", 9) != 0) {
2288 // Look for JAVA_HOME in the environment.
2289 char* java_home_var = ::getenv("JAVA_HOME");
2290 if (java_home_var != NULL && java_home_var[0] != 0) {
2291 char* jrelib_p;
2292 int len;
2293
2294 // Check the current module name "libjvm.so".
2295 p = strrchr(buf, '/');
2296 assert(strstr(p, "/libjvm") == p, "invalid library name");
2297
2298 rp = realpath(java_home_var, buf);
2299 if (rp == NULL)
2300 return;
2301
2302 // determine if this is a legacy image or modules image
2303 // modules image doesn't have "jre" subdirectory
2304 len = strlen(buf);
2305 assert(len < buflen, "Ran out of buffer room");
2306 jrelib_p = buf + len;
2307 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2308 if (0 != access(buf, F_OK)) {
2309 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2310 }
2311
2312 if (0 == access(buf, F_OK)) {
2313 // Use current module name "libjvm.so"
2314 len = strlen(buf);
2315 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2316 } else {
2317 // Go back to path of .so
2318 rp = realpath(dli_fname, buf);
2319 if (rp == NULL)
2320 return;
2321 }
2322 }
2323 }
2324 }
2325
2326 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2327 }
2328
2329 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2330 // no prefix required, not even "_"
2331 }
2332
2333 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2334 // no suffix required
2335 }
2336
2337 ////////////////////////////////////////////////////////////////////////////////
2338 // sun.misc.Signal support
2339
2340 static volatile jint sigint_count = 0;
2341
2342 static void
2343 UserHandler(int sig, void *siginfo, void *context) {
2344 // 4511530 - sem_post is serialized and handled by the manager thread. When
2345 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2346 // don't want to flood the manager thread with sem_post requests.
2347 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2348 return;
2349
2350 // Ctrl-C is pressed during error reporting, likely because the error
2351 // handler fails to abort. Let VM die immediately.
2352 if (sig == SIGINT && is_error_reported()) {
2353 os::die();
2354 }
2355
2356 os::signal_notify(sig);
2357 }
2358
2359 void* os::user_handler() {
2360 return CAST_FROM_FN_PTR(void*, UserHandler);
2361 }
2362
2363 class Semaphore : public StackObj {
2364 public:
2365 Semaphore();
2366 ~Semaphore();
2367 void signal();
2368 void wait();
2369 bool trywait();
2370 bool timedwait(unsigned int sec, int nsec);
2371 private:
2372 sem_t _semaphore;
2373 };
2374
2375 Semaphore::Semaphore() {
2376 sem_init(&_semaphore, 0, 0);
2377 }
2378
2379 Semaphore::~Semaphore() {
2380 sem_destroy(&_semaphore);
2381 }
2382
2383 void Semaphore::signal() {
2384 sem_post(&_semaphore);
2385 }
2386
2387 void Semaphore::wait() {
2388 sem_wait(&_semaphore);
2389 }
2390
2391 bool Semaphore::trywait() {
2392 return sem_trywait(&_semaphore) == 0;
2393 }
2394
2395 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2396
2397 struct timespec ts;
2398 // Semaphore's are always associated with CLOCK_REALTIME
2399 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2400 // see unpackTime for discussion on overflow checking
2401 if (sec >= MAX_SECS) {
2402 ts.tv_sec += MAX_SECS;
2403 ts.tv_nsec = 0;
2404 } else {
2405 ts.tv_sec += sec;
2406 ts.tv_nsec += nsec;
2407 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2408 ts.tv_nsec -= NANOSECS_PER_SEC;
2409 ++ts.tv_sec; // note: this must be <= max_secs
2410 }
2411 }
2412
2413 while (1) {
2414 int result = sem_timedwait(&_semaphore, &ts);
2415 if (result == 0) {
2416 return true;
2417 } else if (errno == EINTR) {
2418 continue;
2419 } else if (errno == ETIMEDOUT) {
2420 return false;
2421 } else {
2422 return false;
2423 }
2424 }
2425 }
2426
2427 extern "C" {
2428 typedef void (*sa_handler_t)(int);
2429 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2430 }
2431
2432 void* os::signal(int signal_number, void* handler) {
2433 struct sigaction sigAct, oldSigAct;
2434
2435 sigfillset(&(sigAct.sa_mask));
2436 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2437 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2438
2439 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2440 // -1 means registration failed
2441 return (void *)-1;
2442 }
2443
2444 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2445 }
2446
2447 void os::signal_raise(int signal_number) {
2448 ::raise(signal_number);
2449 }
2450
2451 /*
2452 * The following code is moved from os.cpp for making this
2453 * code platform specific, which it is by its very nature.
2454 */
2455
2456 // Will be modified when max signal is changed to be dynamic
2457 int os::sigexitnum_pd() {
2458 return NSIG;
2459 }
2460
2461 // a counter for each possible signal value
2462 static volatile jint pending_signals[NSIG+1] = { 0 };
2463
2464 // Linux(POSIX) specific hand shaking semaphore.
2465 static sem_t sig_sem;
2466 static Semaphore sr_semaphore;
2467
2468 void os::signal_init_pd() {
2469 // Initialize signal structures
2470 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2471
2472 // Initialize signal semaphore
2473 ::sem_init(&sig_sem, 0, 0);
2474 }
2475
2476 void os::signal_notify(int sig) {
2477 Atomic::inc(&pending_signals[sig]);
2478 ::sem_post(&sig_sem);
2479 }
2480
2481 static int check_pending_signals(bool wait) {
2482 Atomic::store(0, &sigint_count);
2483 for (;;) {
2484 for (int i = 0; i < NSIG + 1; i++) {
2485 jint n = pending_signals[i];
2486 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2487 return i;
2488 }
2489 }
2490 if (!wait) {
2491 return -1;
2492 }
2493 JavaThread *thread = JavaThread::current();
2494 ThreadBlockInVM tbivm(thread);
2495
2496 bool threadIsSuspended;
2497 do {
2498 thread->set_suspend_equivalent();
2499 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2500 ::sem_wait(&sig_sem);
2501
2502 // were we externally suspended while we were waiting?
2503 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2504 if (threadIsSuspended) {
2505 //
2506 // The semaphore has been incremented, but while we were waiting
2507 // another thread suspended us. We don't want to continue running
2508 // while suspended because that would surprise the thread that
2509 // suspended us.
2510 //
2511 ::sem_post(&sig_sem);
2512
2513 thread->java_suspend_self();
2514 }
2515 } while (threadIsSuspended);
2516 }
2517 }
2518
2519 int os::signal_lookup() {
2520 return check_pending_signals(false);
2521 }
2522
2523 int os::signal_wait() {
2524 return check_pending_signals(true);
2525 }
2526
2527 ////////////////////////////////////////////////////////////////////////////////
2528 // Virtual Memory
2529
2530 int os::vm_page_size() {
2531 // Seems redundant as all get out
2532 assert(os::Linux::page_size() != -1, "must call os::init");
2533 return os::Linux::page_size();
2534 }
2535
2536 // Solaris allocates memory by pages.
2537 int os::vm_allocation_granularity() {
2538 assert(os::Linux::page_size() != -1, "must call os::init");
2539 return os::Linux::page_size();
2540 }
2541
2542 // Rationale behind this function:
2543 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2544 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2545 // samples for JITted code. Here we create private executable mapping over the code cache
2546 // and then we can use standard (well, almost, as mapping can change) way to provide
2547 // info for the reporting script by storing timestamp and location of symbol
2548 void linux_wrap_code(char* base, size_t size) {
2549 static volatile jint cnt = 0;
2550
2551 if (!UseOprofile) {
2552 return;
2553 }
2554
2555 char buf[PATH_MAX+1];
2556 int num = Atomic::add(1, &cnt);
2557
2558 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2559 os::get_temp_directory(), os::current_process_id(), num);
2560 unlink(buf);
2561
2562 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2563
2564 if (fd != -1) {
2565 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2566 if (rv != (off_t)-1) {
2567 if (::write(fd, "", 1) == 1) {
2568 mmap(base, size,
2569 PROT_READ|PROT_WRITE|PROT_EXEC,
2570 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2571 }
2572 }
2573 ::close(fd);
2574 unlink(buf);
2575 }
2576 }
2577
2578 static bool recoverable_mmap_error(int err) {
2579 // See if the error is one we can let the caller handle. This
2580 // list of errno values comes from JBS-6843484. I can't find a
2581 // Linux man page that documents this specific set of errno
2582 // values so while this list currently matches Solaris, it may
2583 // change as we gain experience with this failure mode.
2584 switch (err) {
2585 case EBADF:
2586 case EINVAL:
2587 case ENOTSUP:
2588 // let the caller deal with these errors
2589 return true;
2590
2591 default:
2592 // Any remaining errors on this OS can cause our reserved mapping
2593 // to be lost. That can cause confusion where different data
2594 // structures think they have the same memory mapped. The worst
2595 // scenario is if both the VM and a library think they have the
2596 // same memory mapped.
2597 return false;
2598 }
2599 }
2600
2601 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2602 int err) {
2603 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2604 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2605 strerror(err), err);
2606 }
2607
2608 static void warn_fail_commit_memory(char* addr, size_t size,
2609 size_t alignment_hint, bool exec,
2610 int err) {
2611 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2612 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2613 alignment_hint, exec, strerror(err), err);
2614 }
2615
2616 // NOTE: Linux kernel does not really reserve the pages for us.
2617 // All it does is to check if there are enough free pages
2618 // left at the time of mmap(). This could be a potential
2619 // problem.
2620 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2621 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2622 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2623 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2624 if (res != (uintptr_t) MAP_FAILED) {
2625 if (UseNUMAInterleaving) {
2626 numa_make_global(addr, size);
2627 }
2628 return 0;
2629 }
2630
2631 int err = errno; // save errno from mmap() call above
2632
2633 if (!recoverable_mmap_error(err)) {
2634 warn_fail_commit_memory(addr, size, exec, err);
2635 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2636 }
2637
2638 return err;
2639 }
2640
2641 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2642 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2643 }
2644
2645 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2646 const char* mesg) {
2647 assert(mesg != NULL, "mesg must be specified");
2648 int err = os::Linux::commit_memory_impl(addr, size, exec);
2649 if (err != 0) {
2650 // the caller wants all commit errors to exit with the specified mesg:
2651 warn_fail_commit_memory(addr, size, exec, err);
2652 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2653 }
2654 }
2655
2656 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2657 #ifndef MAP_HUGETLB
2658 #define MAP_HUGETLB 0x40000
2659 #endif
2660
2661 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2662 #ifndef MADV_HUGEPAGE
2663 #define MADV_HUGEPAGE 14
2664 #endif
2665
2666 int os::Linux::commit_memory_impl(char* addr, size_t size,
2667 size_t alignment_hint, bool exec) {
2668 int err = os::Linux::commit_memory_impl(addr, size, exec);
2669 if (err == 0) {
2670 realign_memory(addr, size, alignment_hint);
2671 }
2672 return err;
2673 }
2674
2675 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2676 bool exec) {
2677 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2678 }
2679
2680 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2681 size_t alignment_hint, bool exec,
2682 const char* mesg) {
2683 assert(mesg != NULL, "mesg must be specified");
2684 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2685 if (err != 0) {
2686 // the caller wants all commit errors to exit with the specified mesg:
2687 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2688 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2689 }
2690 }
2691
2692 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2693 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2694 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2695 // be supported or the memory may already be backed by huge pages.
2696 ::madvise(addr, bytes, MADV_HUGEPAGE);
2697 }
2698 }
2699
2700 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2701 // This method works by doing an mmap over an existing mmaping and effectively discarding
2702 // the existing pages. However it won't work for SHM-based large pages that cannot be
2703 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2704 // small pages on top of the SHM segment. This method always works for small pages, so we
2705 // allow that in any case.
2706 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2707 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2708 }
2709 }
2710
2711 void os::numa_make_global(char *addr, size_t bytes) {
2712 Linux::numa_interleave_memory(addr, bytes);
2713 }
2714
2715 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2716 // bind policy to MPOL_PREFERRED for the current thread.
2717 #define USE_MPOL_PREFERRED 0
2718
2719 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2720 // To make NUMA and large pages more robust when both enabled, we need to ease
2721 // the requirements on where the memory should be allocated. MPOL_BIND is the
2722 // default policy and it will force memory to be allocated on the specified
2723 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2724 // the specified node, but will not force it. Using this policy will prevent
2725 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2726 // free large pages.
2727 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2728 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2729 }
2730
2731 bool os::numa_topology_changed() { return false; }
2732
2733 size_t os::numa_get_groups_num() {
2734 int max_node = Linux::numa_max_node();
2735 return max_node > 0 ? max_node + 1 : 1;
2736 }
2737
2738 int os::numa_get_group_id() {
2739 int cpu_id = Linux::sched_getcpu();
2740 if (cpu_id != -1) {
2741 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2742 if (lgrp_id != -1) {
2743 return lgrp_id;
2744 }
2745 }
2746 return 0;
2747 }
2748
2749 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2750 for (size_t i = 0; i < size; i++) {
2751 ids[i] = i;
2752 }
2753 return size;
2754 }
2755
2756 bool os::get_page_info(char *start, page_info* info) {
2757 return false;
2758 }
2759
2760 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2761 return end;
2762 }
2763
2764
2765 int os::Linux::sched_getcpu_syscall(void) {
2766 unsigned int cpu = 0;
2767 int retval = -1;
2768
2769 #if defined(IA32)
2770 # ifndef SYS_getcpu
2771 # define SYS_getcpu 318
2772 # endif
2773 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2774 #elif defined(AMD64)
2775 // Unfortunately we have to bring all these macros here from vsyscall.h
2776 // to be able to compile on old linuxes.
2777 # define __NR_vgetcpu 2
2778 # define VSYSCALL_START (-10UL << 20)
2779 # define VSYSCALL_SIZE 1024
2780 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2781 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2782 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2783 retval = vgetcpu(&cpu, NULL, NULL);
2784 #endif
2785
2786 return (retval == -1) ? retval : cpu;
2787 }
2788
2789 // Something to do with the numa-aware allocator needs these symbols
2790 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2791 extern "C" JNIEXPORT void numa_error(char *where) { }
2792 extern "C" JNIEXPORT int fork1() { return fork(); }
2793
2794
2795 // If we are running with libnuma version > 2, then we should
2796 // be trying to use symbols with versions 1.1
2797 // If we are running with earlier version, which did not have symbol versions,
2798 // we should use the base version.
2799 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2800 void *f = dlvsym(handle, name, "libnuma_1.1");
2801 if (f == NULL) {
2802 f = dlsym(handle, name);
2803 }
2804 return f;
2805 }
2806
2807 bool os::Linux::libnuma_init() {
2808 // sched_getcpu() should be in libc.
2809 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2810 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2811
2812 // If it's not, try a direct syscall.
2813 if (sched_getcpu() == -1)
2814 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2815
2816 if (sched_getcpu() != -1) { // Does it work?
2817 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2818 if (handle != NULL) {
2819 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2820 libnuma_dlsym(handle, "numa_node_to_cpus")));
2821 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2822 libnuma_dlsym(handle, "numa_max_node")));
2823 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2824 libnuma_dlsym(handle, "numa_available")));
2825 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2826 libnuma_dlsym(handle, "numa_tonode_memory")));
2827 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2828 libnuma_dlsym(handle, "numa_interleave_memory")));
2829 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2830 libnuma_dlsym(handle, "numa_set_bind_policy")));
2831
2832
2833 if (numa_available() != -1) {
2834 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2835 // Create a cpu -> node mapping
2836 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2837 rebuild_cpu_to_node_map();
2838 return true;
2839 }
2840 }
2841 }
2842 return false;
2843 }
2844
2845 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2846 // The table is later used in get_node_by_cpu().
2847 void os::Linux::rebuild_cpu_to_node_map() {
2848 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2849 // in libnuma (possible values are starting from 16,
2850 // and continuing up with every other power of 2, but less
2851 // than the maximum number of CPUs supported by kernel), and
2852 // is a subject to change (in libnuma version 2 the requirements
2853 // are more reasonable) we'll just hardcode the number they use
2854 // in the library.
2855 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2856
2857 size_t cpu_num = os::active_processor_count();
2858 size_t cpu_map_size = NCPUS / BitsPerCLong;
2859 size_t cpu_map_valid_size =
2860 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2861
2862 cpu_to_node()->clear();
2863 cpu_to_node()->at_grow(cpu_num - 1);
2864 size_t node_num = numa_get_groups_num();
2865
2866 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2867 for (size_t i = 0; i < node_num; i++) {
2868 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2869 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2870 if (cpu_map[j] != 0) {
2871 for (size_t k = 0; k < BitsPerCLong; k++) {
2872 if (cpu_map[j] & (1UL << k)) {
2873 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2874 }
2875 }
2876 }
2877 }
2878 }
2879 }
2880 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2881 }
2882
2883 int os::Linux::get_node_by_cpu(int cpu_id) {
2884 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2885 return cpu_to_node()->at(cpu_id);
2886 }
2887 return -1;
2888 }
2889
2890 GrowableArray<int>* os::Linux::_cpu_to_node;
2891 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2892 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2893 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2894 os::Linux::numa_available_func_t os::Linux::_numa_available;
2895 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2896 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2897 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2898 unsigned long* os::Linux::_numa_all_nodes;
2899
2900 bool os::pd_uncommit_memory(char* addr, size_t size) {
2901 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2902 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2903 return res != (uintptr_t) MAP_FAILED;
2904 }
2905
2906 static
2907 address get_stack_commited_bottom(address bottom, size_t size) {
2908 address nbot = bottom;
2909 address ntop = bottom + size;
2910
2911 size_t page_sz = os::vm_page_size();
2912 unsigned pages = size / page_sz;
2913
2914 unsigned char vec[1];
2915 unsigned imin = 1, imax = pages + 1, imid;
2916 int mincore_return_value = 0;
2917
2918 assert(imin <= imax, "Unexpected page size");
2919
2920 while (imin < imax) {
2921 imid = (imax + imin) / 2;
2922 nbot = ntop - (imid * page_sz);
2923
2924 // Use a trick with mincore to check whether the page is mapped or not.
2925 // mincore sets vec to 1 if page resides in memory and to 0 if page
2926 // is swapped output but if page we are asking for is unmapped
2927 // it returns -1,ENOMEM
2928 mincore_return_value = mincore(nbot, page_sz, vec);
2929
2930 if (mincore_return_value == -1) {
2931 // Page is not mapped go up
2932 // to find first mapped page
2933 if (errno != EAGAIN) {
2934 assert(errno == ENOMEM, "Unexpected mincore errno");
2935 imax = imid;
2936 }
2937 } else {
2938 // Page is mapped go down
2939 // to find first not mapped page
2940 imin = imid + 1;
2941 }
2942 }
2943
2944 nbot = nbot + page_sz;
2945
2946 // Adjust stack bottom one page up if last checked page is not mapped
2947 if (mincore_return_value == -1) {
2948 nbot = nbot + page_sz;
2949 }
2950
2951 return nbot;
2952 }
2953
2954
2955 // Linux uses a growable mapping for the stack, and if the mapping for
2956 // the stack guard pages is not removed when we detach a thread the
2957 // stack cannot grow beyond the pages where the stack guard was
2958 // mapped. If at some point later in the process the stack expands to
2959 // that point, the Linux kernel cannot expand the stack any further
2960 // because the guard pages are in the way, and a segfault occurs.
2961 //
2962 // However, it's essential not to split the stack region by unmapping
2963 // a region (leaving a hole) that's already part of the stack mapping,
2964 // so if the stack mapping has already grown beyond the guard pages at
2965 // the time we create them, we have to truncate the stack mapping.
2966 // So, we need to know the extent of the stack mapping when
2967 // create_stack_guard_pages() is called.
2968
2969 // We only need this for stacks that are growable: at the time of
2970 // writing thread stacks don't use growable mappings (i.e. those
2971 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2972 // only applies to the main thread.
2973
2974 // If the (growable) stack mapping already extends beyond the point
2975 // where we're going to put our guard pages, truncate the mapping at
2976 // that point by munmap()ping it. This ensures that when we later
2977 // munmap() the guard pages we don't leave a hole in the stack
2978 // mapping. This only affects the main/initial thread
2979
2980 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2981
2982 if (os::Linux::is_initial_thread()) {
2983 // As we manually grow stack up to bottom inside create_attached_thread(),
2984 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2985 // we don't need to do anything special.
2986 // Check it first, before calling heavy function.
2987 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2988 unsigned char vec[1];
2989
2990 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2991 // Fallback to slow path on all errors, including EAGAIN
2992 stack_extent = (uintptr_t) get_stack_commited_bottom(
2993 os::Linux::initial_thread_stack_bottom(),
2994 (size_t)addr - stack_extent);
2995 }
2996
2997 if (stack_extent < (uintptr_t)addr) {
2998 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2999 }
3000 }
3001
3002 return os::commit_memory(addr, size, !ExecMem);
3003 }
3004
3005 // If this is a growable mapping, remove the guard pages entirely by
3006 // munmap()ping them. If not, just call uncommit_memory(). This only
3007 // affects the main/initial thread, but guard against future OS changes
3008 // It's safe to always unmap guard pages for initial thread because we
3009 // always place it right after end of the mapped region
3010
3011 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3012 uintptr_t stack_extent, stack_base;
3013
3014 if (os::Linux::is_initial_thread()) {
3015 return ::munmap(addr, size) == 0;
3016 }
3017
3018 return os::uncommit_memory(addr, size);
3019 }
3020
3021 static address _highest_vm_reserved_address = NULL;
3022
3023 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3024 // at 'requested_addr'. If there are existing memory mappings at the same
3025 // location, however, they will be overwritten. If 'fixed' is false,
3026 // 'requested_addr' is only treated as a hint, the return value may or
3027 // may not start from the requested address. Unlike Linux mmap(), this
3028 // function returns NULL to indicate failure.
3029 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3030 char * addr;
3031 int flags;
3032
3033 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3034 if (fixed) {
3035 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3036 flags |= MAP_FIXED;
3037 }
3038
3039 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3040 // touch an uncommitted page. Otherwise, the read/write might
3041 // succeed if we have enough swap space to back the physical page.
3042 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3043 flags, -1, 0);
3044
3045 if (addr != MAP_FAILED) {
3046 // anon_mmap() should only get called during VM initialization,
3047 // don't need lock (actually we can skip locking even it can be called
3048 // from multiple threads, because _highest_vm_reserved_address is just a
3049 // hint about the upper limit of non-stack memory regions.)
3050 if ((address)addr + bytes > _highest_vm_reserved_address) {
3051 _highest_vm_reserved_address = (address)addr + bytes;
3052 }
3053 }
3054
3055 return addr == MAP_FAILED ? NULL : addr;
3056 }
3057
3058 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3059 // (req_addr != NULL) or with a given alignment.
3060 // - bytes shall be a multiple of alignment.
3061 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3062 // - alignment sets the alignment at which memory shall be allocated.
3063 // It must be a multiple of allocation granularity.
3064 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3065 // req_addr or NULL.
3066 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3067
3068 size_t extra_size = bytes;
3069 if (req_addr == NULL && alignment > 0) {
3070 extra_size += alignment;
3071 }
3072
3073 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3074 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3075 -1, 0);
3076 if (start == MAP_FAILED) {
3077 start = NULL;
3078 } else {
3079 if (req_addr != NULL) {
3080 if (start != req_addr) {
3081 ::munmap(start, extra_size);
3082 start = NULL;
3083 }
3084 } else {
3085 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3086 char* const end_aligned = start_aligned + bytes;
3087 char* const end = start + extra_size;
3088 if (start_aligned > start) {
3089 ::munmap(start, start_aligned - start);
3090 }
3091 if (end_aligned < end) {
3092 ::munmap(end_aligned, end - end_aligned);
3093 }
3094 start = start_aligned;
3095 }
3096 }
3097 return start;
3098 }
3099
3100 // Don't update _highest_vm_reserved_address, because there might be memory
3101 // regions above addr + size. If so, releasing a memory region only creates
3102 // a hole in the address space, it doesn't help prevent heap-stack collision.
3103 //
3104 static int anon_munmap(char * addr, size_t size) {
3105 return ::munmap(addr, size) == 0;
3106 }
3107
3108 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3109 size_t alignment_hint) {
3110 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3111 }
3112
3113 bool os::pd_release_memory(char* addr, size_t size) {
3114 return anon_munmap(addr, size);
3115 }
3116
3117 static address highest_vm_reserved_address() {
3118 return _highest_vm_reserved_address;
3119 }
3120
3121 static bool linux_mprotect(char* addr, size_t size, int prot) {
3122 // Linux wants the mprotect address argument to be page aligned.
3123 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3124
3125 // According to SUSv3, mprotect() should only be used with mappings
3126 // established by mmap(), and mmap() always maps whole pages. Unaligned
3127 // 'addr' likely indicates problem in the VM (e.g. trying to change
3128 // protection of malloc'ed or statically allocated memory). Check the
3129 // caller if you hit this assert.
3130 assert(addr == bottom, "sanity check");
3131
3132 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3133 return ::mprotect(bottom, size, prot) == 0;
3134 }
3135
3136 // Set protections specified
3137 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3138 bool is_committed) {
3139 unsigned int p = 0;
3140 switch (prot) {
3141 case MEM_PROT_NONE: p = PROT_NONE; break;
3142 case MEM_PROT_READ: p = PROT_READ; break;
3143 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3144 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3145 default:
3146 ShouldNotReachHere();
3147 }
3148 // is_committed is unused.
3149 return linux_mprotect(addr, bytes, p);
3150 }
3151
3152 bool os::guard_memory(char* addr, size_t size) {
3153 return linux_mprotect(addr, size, PROT_NONE);
3154 }
3155
3156 bool os::unguard_memory(char* addr, size_t size) {
3157 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3158 }
3159
3160 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3161 bool result = false;
3162 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3163 MAP_ANONYMOUS|MAP_PRIVATE,
3164 -1, 0);
3165 if (p != MAP_FAILED) {
3166 void *aligned_p = align_ptr_up(p, page_size);
3167
3168 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3169
3170 munmap(p, page_size * 2);
3171 }
3172
3173 if (warn && !result) {
3174 warning("TransparentHugePages is not supported by the operating system.");
3175 }
3176
3177 return result;
3178 }
3179
3180 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3181 bool result = false;
3182 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3183 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3184 -1, 0);
3185
3186 if (p != MAP_FAILED) {
3187 // We don't know if this really is a huge page or not.
3188 FILE *fp = fopen("/proc/self/maps", "r");
3189 if (fp) {
3190 while (!feof(fp)) {
3191 char chars[257];
3192 long x = 0;
3193 if (fgets(chars, sizeof(chars), fp)) {
3194 if (sscanf(chars, "%lx-%*x", &x) == 1
3195 && x == (long)p) {
3196 if (strstr (chars, "hugepage")) {
3197 result = true;
3198 break;
3199 }
3200 }
3201 }
3202 }
3203 fclose(fp);
3204 }
3205 munmap(p, page_size);
3206 }
3207
3208 if (warn && !result) {
3209 warning("HugeTLBFS is not supported by the operating system.");
3210 }
3211
3212 return result;
3213 }
3214
3215 /*
3216 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3217 *
3218 * From the coredump_filter documentation:
3219 *
3220 * - (bit 0) anonymous private memory
3221 * - (bit 1) anonymous shared memory
3222 * - (bit 2) file-backed private memory
3223 * - (bit 3) file-backed shared memory
3224 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3225 * effective only if the bit 2 is cleared)
3226 * - (bit 5) hugetlb private memory
3227 * - (bit 6) hugetlb shared memory
3228 */
3229 static void set_coredump_filter(void) {
3230 FILE *f;
3231 long cdm;
3232
3233 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3234 return;
3235 }
3236
3237 if (fscanf(f, "%lx", &cdm) != 1) {
3238 fclose(f);
3239 return;
3240 }
3241
3242 rewind(f);
3243
3244 if ((cdm & LARGEPAGES_BIT) == 0) {
3245 cdm |= LARGEPAGES_BIT;
3246 fprintf(f, "%#lx", cdm);
3247 }
3248
3249 fclose(f);
3250 }
3251
3252 // Large page support
3253
3254 static size_t _large_page_size = 0;
3255
3256 size_t os::Linux::find_large_page_size() {
3257 size_t large_page_size = 0;
3258
3259 // large_page_size on Linux is used to round up heap size. x86 uses either
3260 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3261 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3262 // page as large as 256M.
3263 //
3264 // Here we try to figure out page size by parsing /proc/meminfo and looking
3265 // for a line with the following format:
3266 // Hugepagesize: 2048 kB
3267 //
3268 // If we can't determine the value (e.g. /proc is not mounted, or the text
3269 // format has been changed), we'll use the largest page size supported by
3270 // the processor.
3271
3272 #ifndef ZERO
3273 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3274 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3275 #endif // ZERO
3276
3277 FILE *fp = fopen("/proc/meminfo", "r");
3278 if (fp) {
3279 while (!feof(fp)) {
3280 int x = 0;
3281 char buf[16];
3282 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3283 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3284 large_page_size = x * K;
3285 break;
3286 }
3287 } else {
3288 // skip to next line
3289 for (;;) {
3290 int ch = fgetc(fp);
3291 if (ch == EOF || ch == (int)'\n') break;
3292 }
3293 }
3294 }
3295 fclose(fp);
3296 }
3297
3298 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3299 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3300 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3301 proper_unit_for_byte_size(large_page_size));
3302 }
3303
3304 return large_page_size;
3305 }
3306
3307 size_t os::Linux::setup_large_page_size() {
3308 _large_page_size = Linux::find_large_page_size();
3309 const size_t default_page_size = (size_t)Linux::page_size();
3310 if (_large_page_size > default_page_size) {
3311 _page_sizes[0] = _large_page_size;
3312 _page_sizes[1] = default_page_size;
3313 _page_sizes[2] = 0;
3314 }
3315
3316 return _large_page_size;
3317 }
3318
3319 bool os::Linux::setup_large_page_type(size_t page_size) {
3320 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3321 FLAG_IS_DEFAULT(UseSHM) &&
3322 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3323
3324 // The type of large pages has not been specified by the user.
3325
3326 // Try UseHugeTLBFS and then UseSHM.
3327 UseHugeTLBFS = UseSHM = true;
3328
3329 // Don't try UseTransparentHugePages since there are known
3330 // performance issues with it turned on. This might change in the future.
3331 UseTransparentHugePages = false;
3332 }
3333
3334 if (UseTransparentHugePages) {
3335 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3336 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3337 UseHugeTLBFS = false;
3338 UseSHM = false;
3339 return true;
3340 }
3341 UseTransparentHugePages = false;
3342 }
3343
3344 if (UseHugeTLBFS) {
3345 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3346 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3347 UseSHM = false;
3348 return true;
3349 }
3350 UseHugeTLBFS = false;
3351 }
3352
3353 return UseSHM;
3354 }
3355
3356 void os::large_page_init() {
3357 if (!UseLargePages &&
3358 !UseTransparentHugePages &&
3359 !UseHugeTLBFS &&
3360 !UseSHM) {
3361 // Not using large pages.
3362 return;
3363 }
3364
3365 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3366 // The user explicitly turned off large pages.
3367 // Ignore the rest of the large pages flags.
3368 UseTransparentHugePages = false;
3369 UseHugeTLBFS = false;
3370 UseSHM = false;
3371 return;
3372 }
3373
3374 size_t large_page_size = Linux::setup_large_page_size();
3375 UseLargePages = Linux::setup_large_page_type(large_page_size);
3376
3377 set_coredump_filter();
3378 }
3379
3380 #ifndef SHM_HUGETLB
3381 #define SHM_HUGETLB 04000
3382 #endif
3383
3384 #define shm_warning_format(format, ...) \
3385 do { \
3386 if (UseLargePages && \
3387 (!FLAG_IS_DEFAULT(UseLargePages) || \
3388 !FLAG_IS_DEFAULT(UseSHM) || \
3389 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3390 warning(format, __VA_ARGS__); \
3391 } \
3392 } while (0)
3393
3394 #define shm_warning(str) shm_warning_format("%s", str)
3395
3396 #define shm_warning_with_errno(str) \
3397 do { \
3398 int err = errno; \
3399 shm_warning_format(str " (error = %d)", err); \
3400 } while (0)
3401
3402 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3403 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3404
3405 if (!is_size_aligned(alignment, SHMLBA)) {
3406 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3407 return NULL;
3408 }
3409
3410 // To ensure that we get 'alignment' aligned memory from shmat,
3411 // we pre-reserve aligned virtual memory and then attach to that.
3412
3413 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3414 if (pre_reserved_addr == NULL) {
3415 // Couldn't pre-reserve aligned memory.
3416 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3417 return NULL;
3418 }
3419
3420 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3421 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3422
3423 if ((intptr_t)addr == -1) {
3424 int err = errno;
3425 shm_warning_with_errno("Failed to attach shared memory.");
3426
3427 assert(err != EACCES, "Unexpected error");
3428 assert(err != EIDRM, "Unexpected error");
3429 assert(err != EINVAL, "Unexpected error");
3430
3431 // Since we don't know if the kernel unmapped the pre-reserved memory area
3432 // we can't unmap it, since that would potentially unmap memory that was
3433 // mapped from other threads.
3434 return NULL;
3435 }
3436
3437 return addr;
3438 }
3439
3440 static char* shmat_at_address(int shmid, char* req_addr) {
3441 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3442 assert(false, "Requested address needs to be SHMLBA aligned");
3443 return NULL;
3444 }
3445
3446 char* addr = (char*)shmat(shmid, req_addr, 0);
3447
3448 if ((intptr_t)addr == -1) {
3449 shm_warning_with_errno("Failed to attach shared memory.");
3450 return NULL;
3451 }
3452
3453 return addr;
3454 }
3455
3456 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3457 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3458 if (req_addr != NULL) {
3459 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3460 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3461 return shmat_at_address(shmid, req_addr);
3462 }
3463
3464 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3465 // return large page size aligned memory addresses when req_addr == NULL.
3466 // However, if the alignment is larger than the large page size, we have
3467 // to manually ensure that the memory returned is 'alignment' aligned.
3468 if (alignment > os::large_page_size()) {
3469 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3470 return shmat_with_alignment(shmid, bytes, alignment);
3471 } else {
3472 return shmat_at_address(shmid, NULL);
3473 }
3474 }
3475
3476 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3477 // "exec" is passed in but not used. Creating the shared image for
3478 // the code cache doesn't have an SHM_X executable permission to check.
3479 assert(UseLargePages && UseSHM, "only for SHM large pages");
3480 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3481 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3482
3483 if (!is_size_aligned(bytes, os::large_page_size())) {
3484 return NULL; // Fallback to small pages.
3485 }
3486
3487 // Create a large shared memory region to attach to based on size.
3488 // Currently, size is the total size of the heap.
3489 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3490 if (shmid == -1) {
3491 // Possible reasons for shmget failure:
3492 // 1. shmmax is too small for Java heap.
3493 // > check shmmax value: cat /proc/sys/kernel/shmmax
3494 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3495 // 2. not enough large page memory.
3496 // > check available large pages: cat /proc/meminfo
3497 // > increase amount of large pages:
3498 // echo new_value > /proc/sys/vm/nr_hugepages
3499 // Note 1: different Linux may use different name for this property,
3500 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3501 // Note 2: it's possible there's enough physical memory available but
3502 // they are so fragmented after a long run that they can't
3503 // coalesce into large pages. Try to reserve large pages when
3504 // the system is still "fresh".
3505 shm_warning_with_errno("Failed to reserve shared memory.");
3506 return NULL;
3507 }
3508
3509 // Attach to the region.
3510 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3511
3512 // Remove shmid. If shmat() is successful, the actual shared memory segment
3513 // will be deleted when it's detached by shmdt() or when the process
3514 // terminates. If shmat() is not successful this will remove the shared
3515 // segment immediately.
3516 shmctl(shmid, IPC_RMID, NULL);
3517
3518 return addr;
3519 }
3520
3521 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3522 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3523
3524 bool warn_on_failure = UseLargePages &&
3525 (!FLAG_IS_DEFAULT(UseLargePages) ||
3526 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3527 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3528
3529 if (warn_on_failure) {
3530 char msg[128];
3531 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3532 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3533 warning("%s", msg);
3534 }
3535 }
3536
3537 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3538 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3539 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3540 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3541
3542 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3543 char* addr = (char*)::mmap(req_addr, bytes, prot,
3544 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3545 -1, 0);
3546
3547 if (addr == MAP_FAILED) {
3548 warn_on_large_pages_failure(req_addr, bytes, errno);
3549 return NULL;
3550 }
3551
3552 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3553
3554 return addr;
3555 }
3556
3557 // Reserve memory using mmap(MAP_HUGETLB).
3558 // - bytes shall be a multiple of alignment.
3559 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3560 // - alignment sets the alignment at which memory shall be allocated.
3561 // It must be a multiple of allocation granularity.
3562 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3563 // req_addr or NULL.
3564 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3565 size_t large_page_size = os::large_page_size();
3566 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3567
3568 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3569 assert(is_size_aligned(bytes, alignment), "Must be");
3570
3571 // First reserve - but not commit - the address range in small pages.
3572 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3573
3574 if (start == NULL) {
3575 return NULL;
3576 }
3577
3578 assert(is_ptr_aligned(start, alignment), "Must be");
3579
3580 char* end = start + bytes;
3581
3582 // Find the regions of the allocated chunk that can be promoted to large pages.
3583 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3584 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3585
3586 size_t lp_bytes = lp_end - lp_start;
3587
3588 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3589
3590 if (lp_bytes == 0) {
3591 // The mapped region doesn't even span the start and the end of a large page.
3592 // Fall back to allocate a non-special area.
3593 ::munmap(start, end - start);
3594 return NULL;
3595 }
3596
3597 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3598
3599 void* result;
3600
3601 // Commit small-paged leading area.
3602 if (start != lp_start) {
3603 result = ::mmap(start, lp_start - start, prot,
3604 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3605 -1, 0);
3606 if (result == MAP_FAILED) {
3607 ::munmap(lp_start, end - lp_start);
3608 return NULL;
3609 }
3610 }
3611
3612 // Commit large-paged area.
3613 result = ::mmap(lp_start, lp_bytes, prot,
3614 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3615 -1, 0);
3616 if (result == MAP_FAILED) {
3617 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3618 // If the mmap above fails, the large pages region will be unmapped and we
3619 // have regions before and after with small pages. Release these regions.
3620 //
3621 // | mapped | unmapped | mapped |
3622 // ^ ^ ^ ^
3623 // start lp_start lp_end end
3624 //
3625 ::munmap(start, lp_start - start);
3626 ::munmap(lp_end, end - lp_end);
3627 return NULL;
3628 }
3629
3630 // Commit small-paged trailing area.
3631 if (lp_end != end) {
3632 result = ::mmap(lp_end, end - lp_end, prot,
3633 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3634 -1, 0);
3635 if (result == MAP_FAILED) {
3636 ::munmap(start, lp_end - start);
3637 return NULL;
3638 }
3639 }
3640
3641 return start;
3642 }
3643
3644 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3645 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3646 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3647 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3648 assert(is_power_of_2(os::large_page_size()), "Must be");
3649 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3650
3651 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3652 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3653 } else {
3654 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3655 }
3656 }
3657
3658 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3659 assert(UseLargePages, "only for large pages");
3660
3661 char* addr;
3662 if (UseSHM) {
3663 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3664 } else {
3665 assert(UseHugeTLBFS, "must be");
3666 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3667 }
3668
3669 if (addr != NULL) {
3670 if (UseNUMAInterleaving) {
3671 numa_make_global(addr, bytes);
3672 }
3673
3674 // The memory is committed
3675 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3676 }
3677
3678 return addr;
3679 }
3680
3681 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3682 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3683 return shmdt(base) == 0;
3684 }
3685
3686 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3687 return pd_release_memory(base, bytes);
3688 }
3689
3690 bool os::release_memory_special(char* base, size_t bytes) {
3691 bool res;
3692 if (MemTracker::tracking_level() > NMT_minimal) {
3693 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3694 res = os::Linux::release_memory_special_impl(base, bytes);
3695 if (res) {
3696 tkr.record((address)base, bytes);
3697 }
3698
3699 } else {
3700 res = os::Linux::release_memory_special_impl(base, bytes);
3701 }
3702 return res;
3703 }
3704
3705 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3706 assert(UseLargePages, "only for large pages");
3707 bool res;
3708
3709 if (UseSHM) {
3710 res = os::Linux::release_memory_special_shm(base, bytes);
3711 } else {
3712 assert(UseHugeTLBFS, "must be");
3713 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3714 }
3715 return res;
3716 }
3717
3718 size_t os::large_page_size() {
3719 return _large_page_size;
3720 }
3721
3722 // With SysV SHM the entire memory region must be allocated as shared
3723 // memory.
3724 // HugeTLBFS allows application to commit large page memory on demand.
3725 // However, when committing memory with HugeTLBFS fails, the region
3726 // that was supposed to be committed will lose the old reservation
3727 // and allow other threads to steal that memory region. Because of this
3728 // behavior we can't commit HugeTLBFS memory.
3729 bool os::can_commit_large_page_memory() {
3730 return UseTransparentHugePages;
3731 }
3732
3733 bool os::can_execute_large_page_memory() {
3734 return UseTransparentHugePages || UseHugeTLBFS;
3735 }
3736
3737 // Reserve memory at an arbitrary address, only if that area is
3738 // available (and not reserved for something else).
3739
3740 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3741 const int max_tries = 10;
3742 char* base[max_tries];
3743 size_t size[max_tries];
3744 const size_t gap = 0x000000;
3745
3746 // Assert only that the size is a multiple of the page size, since
3747 // that's all that mmap requires, and since that's all we really know
3748 // about at this low abstraction level. If we need higher alignment,
3749 // we can either pass an alignment to this method or verify alignment
3750 // in one of the methods further up the call chain. See bug 5044738.
3751 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3752
3753 // Repeatedly allocate blocks until the block is allocated at the
3754 // right spot. Give up after max_tries. Note that reserve_memory() will
3755 // automatically update _highest_vm_reserved_address if the call is
3756 // successful. The variable tracks the highest memory address every reserved
3757 // by JVM. It is used to detect heap-stack collision if running with
3758 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3759 // space than needed, it could confuse the collision detecting code. To
3760 // solve the problem, save current _highest_vm_reserved_address and
3761 // calculate the correct value before return.
3762 address old_highest = _highest_vm_reserved_address;
3763
3764 // Linux mmap allows caller to pass an address as hint; give it a try first,
3765 // if kernel honors the hint then we can return immediately.
3766 char * addr = anon_mmap(requested_addr, bytes, false);
3767 if (addr == requested_addr) {
3768 return requested_addr;
3769 }
3770
3771 if (addr != NULL) {
3772 // mmap() is successful but it fails to reserve at the requested address
3773 anon_munmap(addr, bytes);
3774 }
3775
3776 int i;
3777 for (i = 0; i < max_tries; ++i) {
3778 base[i] = reserve_memory(bytes);
3779
3780 if (base[i] != NULL) {
3781 // Is this the block we wanted?
3782 if (base[i] == requested_addr) {
3783 size[i] = bytes;
3784 break;
3785 }
3786
3787 // Does this overlap the block we wanted? Give back the overlapped
3788 // parts and try again.
3789
3790 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3791 if (top_overlap >= 0 && top_overlap < bytes) {
3792 unmap_memory(base[i], top_overlap);
3793 base[i] += top_overlap;
3794 size[i] = bytes - top_overlap;
3795 } else {
3796 size_t bottom_overlap = base[i] + bytes - requested_addr;
3797 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3798 unmap_memory(requested_addr, bottom_overlap);
3799 size[i] = bytes - bottom_overlap;
3800 } else {
3801 size[i] = bytes;
3802 }
3803 }
3804 }
3805 }
3806
3807 // Give back the unused reserved pieces.
3808
3809 for (int j = 0; j < i; ++j) {
3810 if (base[j] != NULL) {
3811 unmap_memory(base[j], size[j]);
3812 }
3813 }
3814
3815 if (i < max_tries) {
3816 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3817 return requested_addr;
3818 } else {
3819 _highest_vm_reserved_address = old_highest;
3820 return NULL;
3821 }
3822 }
3823
3824 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3825 return ::read(fd, buf, nBytes);
3826 }
3827
3828 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3829 // Solaris uses poll(), linux uses park().
3830 // Poll() is likely a better choice, assuming that Thread.interrupt()
3831 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3832 // SIGSEGV, see 4355769.
3833
3834 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3835 assert(thread == Thread::current(), "thread consistency check");
3836
3837 ParkEvent * const slp = thread->_SleepEvent ;
3838 slp->reset() ;
3839 OrderAccess::fence() ;
3840
3841 if (interruptible) {
3842 jlong prevtime = javaTimeNanos();
3843
3844 for (;;) {
3845 if (os::is_interrupted(thread, true)) {
3846 return OS_INTRPT;
3847 }
3848
3849 jlong newtime = javaTimeNanos();
3850
3851 if (newtime - prevtime < 0) {
3852 // time moving backwards, should only happen if no monotonic clock
3853 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3854 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3855 } else {
3856 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3857 }
3858
3859 if(millis <= 0) {
3860 return OS_OK;
3861 }
3862
3863 prevtime = newtime;
3864
3865 {
3866 assert(thread->is_Java_thread(), "sanity check");
3867 JavaThread *jt = (JavaThread *) thread;
3868 ThreadBlockInVM tbivm(jt);
3869 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3870
3871 jt->set_suspend_equivalent();
3872 // cleared by handle_special_suspend_equivalent_condition() or
3873 // java_suspend_self() via check_and_wait_while_suspended()
3874
3875 slp->park(millis);
3876
3877 // were we externally suspended while we were waiting?
3878 jt->check_and_wait_while_suspended();
3879 }
3880 }
3881 } else {
3882 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3883 jlong prevtime = javaTimeNanos();
3884
3885 for (;;) {
3886 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3887 // the 1st iteration ...
3888 jlong newtime = javaTimeNanos();
3889
3890 if (newtime - prevtime < 0) {
3891 // time moving backwards, should only happen if no monotonic clock
3892 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3893 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3894 } else {
3895 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3896 }
3897
3898 if(millis <= 0) break ;
3899
3900 prevtime = newtime;
3901 slp->park(millis);
3902 }
3903 return OS_OK ;
3904 }
3905 }
3906
3907 //
3908 // Short sleep, direct OS call.
3909 //
3910 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3911 // sched_yield(2) will actually give up the CPU:
3912 //
3913 // * Alone on this pariticular CPU, keeps running.
3914 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3915 // (pre 2.6.39).
3916 //
3917 // So calling this with 0 is an alternative.
3918 //
3919 void os::naked_short_sleep(jlong ms) {
3920 struct timespec req;
3921
3922 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3923 req.tv_sec = 0;
3924 if (ms > 0) {
3925 req.tv_nsec = (ms % 1000) * 1000000;
3926 }
3927 else {
3928 req.tv_nsec = 1;
3929 }
3930
3931 nanosleep(&req, NULL);
3932
3933 return;
3934 }
3935
3936 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3937 void os::infinite_sleep() {
3938 while (true) { // sleep forever ...
3939 ::sleep(100); // ... 100 seconds at a time
3940 }
3941 }
3942
3943 // Used to convert frequent JVM_Yield() to nops
3944 bool os::dont_yield() {
3945 return DontYieldALot;
3946 }
3947
3948 void os::yield() {
3949 sched_yield();
3950 }
3951
3952 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3953
3954 void os::yield_all(int attempts) {
3955 // Yields to all threads, including threads with lower priorities
3956 // Threads on Linux are all with same priority. The Solaris style
3957 // os::yield_all() with nanosleep(1ms) is not necessary.
3958 sched_yield();
3959 }
3960
3961 // Called from the tight loops to possibly influence time-sharing heuristics
3962 void os::loop_breaker(int attempts) {
3963 os::yield_all(attempts);
3964 }
3965
3966 ////////////////////////////////////////////////////////////////////////////////
3967 // thread priority support
3968
3969 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3970 // only supports dynamic priority, static priority must be zero. For real-time
3971 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3972 // However, for large multi-threaded applications, SCHED_RR is not only slower
3973 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3974 // of 5 runs - Sep 2005).
3975 //
3976 // The following code actually changes the niceness of kernel-thread/LWP. It
3977 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3978 // not the entire user process, and user level threads are 1:1 mapped to kernel
3979 // threads. It has always been the case, but could change in the future. For
3980 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3981 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3982
3983 int os::java_to_os_priority[CriticalPriority + 1] = {
3984 19, // 0 Entry should never be used
3985
3986 4, // 1 MinPriority
3987 3, // 2
3988 2, // 3
3989
3990 1, // 4
3991 0, // 5 NormPriority
3992 -1, // 6
3993
3994 -2, // 7
3995 -3, // 8
3996 -4, // 9 NearMaxPriority
3997
3998 -5, // 10 MaxPriority
3999
4000 -5 // 11 CriticalPriority
4001 };
4002
4003 static int prio_init() {
4004 if (ThreadPriorityPolicy == 1) {
4005 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4006 // if effective uid is not root. Perhaps, a more elegant way of doing
4007 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4008 if (geteuid() != 0) {
4009 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4010 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4011 }
4012 ThreadPriorityPolicy = 0;
4013 }
4014 }
4015 if (UseCriticalJavaThreadPriority) {
4016 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4017 }
4018 return 0;
4019 }
4020
4021 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4022 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4023
4024 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4025 return (ret == 0) ? OS_OK : OS_ERR;
4026 }
4027
4028 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4029 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4030 *priority_ptr = java_to_os_priority[NormPriority];
4031 return OS_OK;
4032 }
4033
4034 errno = 0;
4035 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4036 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4037 }
4038
4039 // Hint to the underlying OS that a task switch would not be good.
4040 // Void return because it's a hint and can fail.
4041 void os::hint_no_preempt() {}
4042
4043 ////////////////////////////////////////////////////////////////////////////////
4044 // suspend/resume support
4045
4046 // the low-level signal-based suspend/resume support is a remnant from the
4047 // old VM-suspension that used to be for java-suspension, safepoints etc,
4048 // within hotspot. Now there is a single use-case for this:
4049 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4050 // that runs in the watcher thread.
4051 // The remaining code is greatly simplified from the more general suspension
4052 // code that used to be used.
4053 //
4054 // The protocol is quite simple:
4055 // - suspend:
4056 // - sends a signal to the target thread
4057 // - polls the suspend state of the osthread using a yield loop
4058 // - target thread signal handler (SR_handler) sets suspend state
4059 // and blocks in sigsuspend until continued
4060 // - resume:
4061 // - sets target osthread state to continue
4062 // - sends signal to end the sigsuspend loop in the SR_handler
4063 //
4064 // Note that the SR_lock plays no role in this suspend/resume protocol.
4065 //
4066
4067 static void resume_clear_context(OSThread *osthread) {
4068 osthread->set_ucontext(NULL);
4069 osthread->set_siginfo(NULL);
4070 }
4071
4072 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4073 osthread->set_ucontext(context);
4074 osthread->set_siginfo(siginfo);
4075 }
4076
4077 //
4078 // Handler function invoked when a thread's execution is suspended or
4079 // resumed. We have to be careful that only async-safe functions are
4080 // called here (Note: most pthread functions are not async safe and
4081 // should be avoided.)
4082 //
4083 // Note: sigwait() is a more natural fit than sigsuspend() from an
4084 // interface point of view, but sigwait() prevents the signal hander
4085 // from being run. libpthread would get very confused by not having
4086 // its signal handlers run and prevents sigwait()'s use with the
4087 // mutex granting granting signal.
4088 //
4089 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4090 //
4091 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4092 // Save and restore errno to avoid confusing native code with EINTR
4093 // after sigsuspend.
4094 int old_errno = errno;
4095
4096 Thread* thread = Thread::current();
4097 OSThread* osthread = thread->osthread();
4098 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4099
4100 os::SuspendResume::State current = osthread->sr.state();
4101 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4102 suspend_save_context(osthread, siginfo, context);
4103
4104 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4105 os::SuspendResume::State state = osthread->sr.suspended();
4106 if (state == os::SuspendResume::SR_SUSPENDED) {
4107 sigset_t suspend_set; // signals for sigsuspend()
4108
4109 // get current set of blocked signals and unblock resume signal
4110 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4111 sigdelset(&suspend_set, SR_signum);
4112
4113 sr_semaphore.signal();
4114 // wait here until we are resumed
4115 while (1) {
4116 sigsuspend(&suspend_set);
4117
4118 os::SuspendResume::State result = osthread->sr.running();
4119 if (result == os::SuspendResume::SR_RUNNING) {
4120 sr_semaphore.signal();
4121 break;
4122 }
4123 }
4124
4125 } else if (state == os::SuspendResume::SR_RUNNING) {
4126 // request was cancelled, continue
4127 } else {
4128 ShouldNotReachHere();
4129 }
4130
4131 resume_clear_context(osthread);
4132 } else if (current == os::SuspendResume::SR_RUNNING) {
4133 // request was cancelled, continue
4134 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4135 // ignore
4136 } else {
4137 // ignore
4138 }
4139
4140 errno = old_errno;
4141 }
4142
4143
4144 static int SR_initialize() {
4145 struct sigaction act;
4146 char *s;
4147 /* Get signal number to use for suspend/resume */
4148 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4149 int sig = ::strtol(s, 0, 10);
4150 if (sig > 0 || sig < _NSIG) {
4151 SR_signum = sig;
4152 }
4153 }
4154
4155 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4156 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4157
4158 sigemptyset(&SR_sigset);
4159 sigaddset(&SR_sigset, SR_signum);
4160
4161 /* Set up signal handler for suspend/resume */
4162 act.sa_flags = SA_RESTART|SA_SIGINFO;
4163 act.sa_handler = (void (*)(int)) SR_handler;
4164
4165 // SR_signum is blocked by default.
4166 // 4528190 - We also need to block pthread restart signal (32 on all
4167 // supported Linux platforms). Note that LinuxThreads need to block
4168 // this signal for all threads to work properly. So we don't have
4169 // to use hard-coded signal number when setting up the mask.
4170 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4171
4172 if (sigaction(SR_signum, &act, 0) == -1) {
4173 return -1;
4174 }
4175
4176 // Save signal flag
4177 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4178 return 0;
4179 }
4180
4181 static int sr_notify(OSThread* osthread) {
4182 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4183 assert_status(status == 0, status, "pthread_kill");
4184 return status;
4185 }
4186
4187 // "Randomly" selected value for how long we want to spin
4188 // before bailing out on suspending a thread, also how often
4189 // we send a signal to a thread we want to resume
4190 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4191 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4192
4193 // returns true on success and false on error - really an error is fatal
4194 // but this seems the normal response to library errors
4195 static bool do_suspend(OSThread* osthread) {
4196 assert(osthread->sr.is_running(), "thread should be running");
4197 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4198
4199 // mark as suspended and send signal
4200 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4201 // failed to switch, state wasn't running?
4202 ShouldNotReachHere();
4203 return false;
4204 }
4205
4206 if (sr_notify(osthread) != 0) {
4207 ShouldNotReachHere();
4208 }
4209
4210 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4211 while (true) {
4212 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4213 break;
4214 } else {
4215 // timeout
4216 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4217 if (cancelled == os::SuspendResume::SR_RUNNING) {
4218 return false;
4219 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4220 // make sure that we consume the signal on the semaphore as well
4221 sr_semaphore.wait();
4222 break;
4223 } else {
4224 ShouldNotReachHere();
4225 return false;
4226 }
4227 }
4228 }
4229
4230 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4231 return true;
4232 }
4233
4234 static void do_resume(OSThread* osthread) {
4235 assert(osthread->sr.is_suspended(), "thread should be suspended");
4236 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4237
4238 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4239 // failed to switch to WAKEUP_REQUEST
4240 ShouldNotReachHere();
4241 return;
4242 }
4243
4244 while (true) {
4245 if (sr_notify(osthread) == 0) {
4246 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4247 if (osthread->sr.is_running()) {
4248 return;
4249 }
4250 }
4251 } else {
4252 ShouldNotReachHere();
4253 }
4254 }
4255
4256 guarantee(osthread->sr.is_running(), "Must be running!");
4257 }
4258
4259 ////////////////////////////////////////////////////////////////////////////////
4260 // interrupt support
4261
4262 void os::interrupt(Thread* thread) {
4263 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4264 "possibility of dangling Thread pointer");
4265
4266 OSThread* osthread = thread->osthread();
4267
4268 if (!osthread->interrupted()) {
4269 osthread->set_interrupted(true);
4270 // More than one thread can get here with the same value of osthread,
4271 // resulting in multiple notifications. We do, however, want the store
4272 // to interrupted() to be visible to other threads before we execute unpark().
4273 OrderAccess::fence();
4274 ParkEvent * const slp = thread->_SleepEvent ;
4275 if (slp != NULL) slp->unpark() ;
4276 }
4277
4278 // For JSR166. Unpark even if interrupt status already was set
4279 if (thread->is_Java_thread())
4280 ((JavaThread*)thread)->parker()->unpark();
4281
4282 ParkEvent * ev = thread->_ParkEvent ;
4283 if (ev != NULL) ev->unpark() ;
4284
4285 }
4286
4287 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4288 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4289 "possibility of dangling Thread pointer");
4290
4291 OSThread* osthread = thread->osthread();
4292
4293 bool interrupted = osthread->interrupted();
4294
4295 if (interrupted && clear_interrupted) {
4296 osthread->set_interrupted(false);
4297 // consider thread->_SleepEvent->reset() ... optional optimization
4298 }
4299
4300 return interrupted;
4301 }
4302
4303 ///////////////////////////////////////////////////////////////////////////////////
4304 // signal handling (except suspend/resume)
4305
4306 // This routine may be used by user applications as a "hook" to catch signals.
4307 // The user-defined signal handler must pass unrecognized signals to this
4308 // routine, and if it returns true (non-zero), then the signal handler must
4309 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4310 // routine will never retun false (zero), but instead will execute a VM panic
4311 // routine kill the process.
4312 //
4313 // If this routine returns false, it is OK to call it again. This allows
4314 // the user-defined signal handler to perform checks either before or after
4315 // the VM performs its own checks. Naturally, the user code would be making
4316 // a serious error if it tried to handle an exception (such as a null check
4317 // or breakpoint) that the VM was generating for its own correct operation.
4318 //
4319 // This routine may recognize any of the following kinds of signals:
4320 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4321 // It should be consulted by handlers for any of those signals.
4322 //
4323 // The caller of this routine must pass in the three arguments supplied
4324 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4325 // field of the structure passed to sigaction(). This routine assumes that
4326 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4327 //
4328 // Note that the VM will print warnings if it detects conflicting signal
4329 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4330 //
4331 extern "C" JNIEXPORT int
4332 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4333 void* ucontext, int abort_if_unrecognized);
4334
4335 void signalHandler(int sig, siginfo_t* info, void* uc) {
4336 assert(info != NULL && uc != NULL, "it must be old kernel");
4337 int orig_errno = errno; // Preserve errno value over signal handler.
4338 JVM_handle_linux_signal(sig, info, uc, true);
4339 errno = orig_errno;
4340 }
4341
4342
4343 // This boolean allows users to forward their own non-matching signals
4344 // to JVM_handle_linux_signal, harmlessly.
4345 bool os::Linux::signal_handlers_are_installed = false;
4346
4347 // For signal-chaining
4348 struct sigaction os::Linux::sigact[MAXSIGNUM];
4349 unsigned int os::Linux::sigs = 0;
4350 bool os::Linux::libjsig_is_loaded = false;
4351 typedef struct sigaction *(*get_signal_t)(int);
4352 get_signal_t os::Linux::get_signal_action = NULL;
4353
4354 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4355 struct sigaction *actp = NULL;
4356
4357 if (libjsig_is_loaded) {
4358 // Retrieve the old signal handler from libjsig
4359 actp = (*get_signal_action)(sig);
4360 }
4361 if (actp == NULL) {
4362 // Retrieve the preinstalled signal handler from jvm
4363 actp = get_preinstalled_handler(sig);
4364 }
4365
4366 return actp;
4367 }
4368
4369 static bool call_chained_handler(struct sigaction *actp, int sig,
4370 siginfo_t *siginfo, void *context) {
4371 // Call the old signal handler
4372 if (actp->sa_handler == SIG_DFL) {
4373 // It's more reasonable to let jvm treat it as an unexpected exception
4374 // instead of taking the default action.
4375 return false;
4376 } else if (actp->sa_handler != SIG_IGN) {
4377 if ((actp->sa_flags & SA_NODEFER) == 0) {
4378 // automaticlly block the signal
4379 sigaddset(&(actp->sa_mask), sig);
4380 }
4381
4382 sa_handler_t hand = NULL;
4383 sa_sigaction_t sa = NULL;
4384 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4385 // retrieve the chained handler
4386 if (siginfo_flag_set) {
4387 sa = actp->sa_sigaction;
4388 } else {
4389 hand = actp->sa_handler;
4390 }
4391
4392 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4393 actp->sa_handler = SIG_DFL;
4394 }
4395
4396 // try to honor the signal mask
4397 sigset_t oset;
4398 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4399
4400 // call into the chained handler
4401 if (siginfo_flag_set) {
4402 (*sa)(sig, siginfo, context);
4403 } else {
4404 (*hand)(sig);
4405 }
4406
4407 // restore the signal mask
4408 pthread_sigmask(SIG_SETMASK, &oset, 0);
4409 }
4410 // Tell jvm's signal handler the signal is taken care of.
4411 return true;
4412 }
4413
4414 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4415 bool chained = false;
4416 // signal-chaining
4417 if (UseSignalChaining) {
4418 struct sigaction *actp = get_chained_signal_action(sig);
4419 if (actp != NULL) {
4420 chained = call_chained_handler(actp, sig, siginfo, context);
4421 }
4422 }
4423 return chained;
4424 }
4425
4426 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4427 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4428 return &sigact[sig];
4429 }
4430 return NULL;
4431 }
4432
4433 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4434 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4435 sigact[sig] = oldAct;
4436 sigs |= (unsigned int)1 << sig;
4437 }
4438
4439 // for diagnostic
4440 int os::Linux::sigflags[MAXSIGNUM];
4441
4442 int os::Linux::get_our_sigflags(int sig) {
4443 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4444 return sigflags[sig];
4445 }
4446
4447 void os::Linux::set_our_sigflags(int sig, int flags) {
4448 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4449 sigflags[sig] = flags;
4450 }
4451
4452 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4453 // Check for overwrite.
4454 struct sigaction oldAct;
4455 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4456
4457 void* oldhand = oldAct.sa_sigaction
4458 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4459 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4460 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4461 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4462 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4463 if (AllowUserSignalHandlers || !set_installed) {
4464 // Do not overwrite; user takes responsibility to forward to us.
4465 return;
4466 } else if (UseSignalChaining) {
4467 // save the old handler in jvm
4468 save_preinstalled_handler(sig, oldAct);
4469 // libjsig also interposes the sigaction() call below and saves the
4470 // old sigaction on it own.
4471 } else {
4472 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4473 "%#lx for signal %d.", (long)oldhand, sig));
4474 }
4475 }
4476
4477 struct sigaction sigAct;
4478 sigfillset(&(sigAct.sa_mask));
4479 sigAct.sa_handler = SIG_DFL;
4480 if (!set_installed) {
4481 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4482 } else {
4483 sigAct.sa_sigaction = signalHandler;
4484 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4485 }
4486 // Save flags, which are set by ours
4487 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4488 sigflags[sig] = sigAct.sa_flags;
4489
4490 int ret = sigaction(sig, &sigAct, &oldAct);
4491 assert(ret == 0, "check");
4492
4493 void* oldhand2 = oldAct.sa_sigaction
4494 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4495 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4496 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4497 }
4498
4499 // install signal handlers for signals that HotSpot needs to
4500 // handle in order to support Java-level exception handling.
4501
4502 void os::Linux::install_signal_handlers() {
4503 if (!signal_handlers_are_installed) {
4504 signal_handlers_are_installed = true;
4505
4506 // signal-chaining
4507 typedef void (*signal_setting_t)();
4508 signal_setting_t begin_signal_setting = NULL;
4509 signal_setting_t end_signal_setting = NULL;
4510 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4511 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4512 if (begin_signal_setting != NULL) {
4513 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4514 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4515 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4516 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4517 libjsig_is_loaded = true;
4518 assert(UseSignalChaining, "should enable signal-chaining");
4519 }
4520 if (libjsig_is_loaded) {
4521 // Tell libjsig jvm is setting signal handlers
4522 (*begin_signal_setting)();
4523 }
4524
4525 set_signal_handler(SIGSEGV, true);
4526 set_signal_handler(SIGPIPE, true);
4527 set_signal_handler(SIGBUS, true);
4528 set_signal_handler(SIGILL, true);
4529 set_signal_handler(SIGFPE, true);
4530 #if defined(PPC64)
4531 set_signal_handler(SIGTRAP, true);
4532 #endif
4533 set_signal_handler(SIGXFSZ, true);
4534
4535 if (libjsig_is_loaded) {
4536 // Tell libjsig jvm finishes setting signal handlers
4537 (*end_signal_setting)();
4538 }
4539
4540 // We don't activate signal checker if libjsig is in place, we trust ourselves
4541 // and if UserSignalHandler is installed all bets are off.
4542 // Log that signal checking is off only if -verbose:jni is specified.
4543 if (CheckJNICalls) {
4544 if (libjsig_is_loaded) {
4545 if (PrintJNIResolving) {
4546 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4547 }
4548 check_signals = false;
4549 }
4550 if (AllowUserSignalHandlers) {
4551 if (PrintJNIResolving) {
4552 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4553 }
4554 check_signals = false;
4555 }
4556 }
4557 }
4558 }
4559
4560 // This is the fastest way to get thread cpu time on Linux.
4561 // Returns cpu time (user+sys) for any thread, not only for current.
4562 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4563 // It might work on 2.6.10+ with a special kernel/glibc patch.
4564 // For reference, please, see IEEE Std 1003.1-2004:
4565 // http://www.unix.org/single_unix_specification
4566
4567 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4568 struct timespec tp;
4569 int rc = os::Linux::clock_gettime(clockid, &tp);
4570 assert(rc == 0, "clock_gettime is expected to return 0 code");
4571
4572 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4573 }
4574
4575 /////
4576 // glibc on Linux platform uses non-documented flag
4577 // to indicate, that some special sort of signal
4578 // trampoline is used.
4579 // We will never set this flag, and we should
4580 // ignore this flag in our diagnostic
4581 #ifdef SIGNIFICANT_SIGNAL_MASK
4582 #undef SIGNIFICANT_SIGNAL_MASK
4583 #endif
4584 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4585
4586 static const char* get_signal_handler_name(address handler,
4587 char* buf, int buflen) {
4588 int offset = 0;
4589 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4590 if (found) {
4591 // skip directory names
4592 const char *p1, *p2;
4593 p1 = buf;
4594 size_t len = strlen(os::file_separator());
4595 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4596 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4597 } else {
4598 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4599 }
4600 return buf;
4601 }
4602
4603 static void print_signal_handler(outputStream* st, int sig,
4604 char* buf, size_t buflen) {
4605 struct sigaction sa;
4606
4607 sigaction(sig, NULL, &sa);
4608
4609 // See comment for SIGNIFICANT_SIGNAL_MASK define
4610 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4611
4612 st->print("%s: ", os::exception_name(sig, buf, buflen));
4613
4614 address handler = (sa.sa_flags & SA_SIGINFO)
4615 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4616 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4617
4618 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4619 st->print("SIG_DFL");
4620 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4621 st->print("SIG_IGN");
4622 } else {
4623 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4624 }
4625
4626 st->print(", sa_mask[0]=");
4627 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4628
4629 address rh = VMError::get_resetted_sighandler(sig);
4630 // May be, handler was resetted by VMError?
4631 if(rh != NULL) {
4632 handler = rh;
4633 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4634 }
4635
4636 st->print(", sa_flags=");
4637 os::Posix::print_sa_flags(st, sa.sa_flags);
4638
4639 // Check: is it our handler?
4640 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4641 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4642 // It is our signal handler
4643 // check for flags, reset system-used one!
4644 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4645 st->print(
4646 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4647 os::Linux::get_our_sigflags(sig));
4648 }
4649 }
4650 st->cr();
4651 }
4652
4653
4654 #define DO_SIGNAL_CHECK(sig) \
4655 if (!sigismember(&check_signal_done, sig)) \
4656 os::Linux::check_signal_handler(sig)
4657
4658 // This method is a periodic task to check for misbehaving JNI applications
4659 // under CheckJNI, we can add any periodic checks here
4660
4661 void os::run_periodic_checks() {
4662
4663 if (check_signals == false) return;
4664
4665 // SEGV and BUS if overridden could potentially prevent
4666 // generation of hs*.log in the event of a crash, debugging
4667 // such a case can be very challenging, so we absolutely
4668 // check the following for a good measure:
4669 DO_SIGNAL_CHECK(SIGSEGV);
4670 DO_SIGNAL_CHECK(SIGILL);
4671 DO_SIGNAL_CHECK(SIGFPE);
4672 DO_SIGNAL_CHECK(SIGBUS);
4673 DO_SIGNAL_CHECK(SIGPIPE);
4674 DO_SIGNAL_CHECK(SIGXFSZ);
4675 #if defined(PPC64)
4676 DO_SIGNAL_CHECK(SIGTRAP);
4677 #endif
4678
4679 // ReduceSignalUsage allows the user to override these handlers
4680 // see comments at the very top and jvm_solaris.h
4681 if (!ReduceSignalUsage) {
4682 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4683 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4684 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4685 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4686 }
4687
4688 DO_SIGNAL_CHECK(SR_signum);
4689 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4690 }
4691
4692 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4693
4694 static os_sigaction_t os_sigaction = NULL;
4695
4696 void os::Linux::check_signal_handler(int sig) {
4697 char buf[O_BUFLEN];
4698 address jvmHandler = NULL;
4699
4700
4701 struct sigaction act;
4702 if (os_sigaction == NULL) {
4703 // only trust the default sigaction, in case it has been interposed
4704 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4705 if (os_sigaction == NULL) return;
4706 }
4707
4708 os_sigaction(sig, (struct sigaction*)NULL, &act);
4709
4710
4711 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4712
4713 address thisHandler = (act.sa_flags & SA_SIGINFO)
4714 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4715 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4716
4717
4718 switch(sig) {
4719 case SIGSEGV:
4720 case SIGBUS:
4721 case SIGFPE:
4722 case SIGPIPE:
4723 case SIGILL:
4724 case SIGXFSZ:
4725 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4726 break;
4727
4728 case SHUTDOWN1_SIGNAL:
4729 case SHUTDOWN2_SIGNAL:
4730 case SHUTDOWN3_SIGNAL:
4731 case BREAK_SIGNAL:
4732 jvmHandler = (address)user_handler();
4733 break;
4734
4735 case INTERRUPT_SIGNAL:
4736 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4737 break;
4738
4739 default:
4740 if (sig == SR_signum) {
4741 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4742 } else {
4743 return;
4744 }
4745 break;
4746 }
4747
4748 if (thisHandler != jvmHandler) {
4749 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4750 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4751 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4752 // No need to check this sig any longer
4753 sigaddset(&check_signal_done, sig);
4754 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4755 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4756 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4757 exception_name(sig, buf, O_BUFLEN));
4758 }
4759 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4760 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4761 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4762 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4763 // No need to check this sig any longer
4764 sigaddset(&check_signal_done, sig);
4765 }
4766
4767 // Dump all the signal
4768 if (sigismember(&check_signal_done, sig)) {
4769 print_signal_handlers(tty, buf, O_BUFLEN);
4770 }
4771 }
4772
4773 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4774
4775 extern bool signal_name(int signo, char* buf, size_t len);
4776
4777 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4778 if (0 < exception_code && exception_code <= SIGRTMAX) {
4779 // signal
4780 if (!signal_name(exception_code, buf, size)) {
4781 jio_snprintf(buf, size, "SIG%d", exception_code);
4782 }
4783 return buf;
4784 } else {
4785 return NULL;
4786 }
4787 }
4788
4789 // this is called _before_ the most of global arguments have been parsed
4790 void os::init(void) {
4791 char dummy; /* used to get a guess on initial stack address */
4792 // first_hrtime = gethrtime();
4793
4794 // With LinuxThreads the JavaMain thread pid (primordial thread)
4795 // is different than the pid of the java launcher thread.
4796 // So, on Linux, the launcher thread pid is passed to the VM
4797 // via the sun.java.launcher.pid property.
4798 // Use this property instead of getpid() if it was correctly passed.
4799 // See bug 6351349.
4800 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4801
4802 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4803
4804 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4805
4806 init_random(1234567);
4807
4808 ThreadCritical::initialize();
4809
4810 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4811 if (Linux::page_size() == -1) {
4812 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4813 strerror(errno)));
4814 }
4815 init_page_sizes((size_t) Linux::page_size());
4816
4817 Linux::initialize_system_info();
4818
4819 // main_thread points to the aboriginal thread
4820 Linux::_main_thread = pthread_self();
4821
4822 Linux::clock_init();
4823 initial_time_count = javaTimeNanos();
4824
4825 // pthread_condattr initialization for monotonic clock
4826 int status;
4827 pthread_condattr_t* _condattr = os::Linux::condAttr();
4828 if ((status = pthread_condattr_init(_condattr)) != 0) {
4829 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4830 }
4831 // Only set the clock if CLOCK_MONOTONIC is available
4832 if (Linux::supports_monotonic_clock()) {
4833 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4834 if (status == EINVAL) {
4835 warning("Unable to use monotonic clock with relative timed-waits" \
4836 " - changes to the time-of-day clock may have adverse affects");
4837 } else {
4838 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4839 }
4840 }
4841 }
4842 // else it defaults to CLOCK_REALTIME
4843
4844 pthread_mutex_init(&dl_mutex, NULL);
4845
4846 // If the pagesize of the VM is greater than 8K determine the appropriate
4847 // number of initial guard pages. The user can change this with the
4848 // command line arguments, if needed.
4849 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4850 StackYellowPages = 1;
4851 StackRedPages = 1;
4852 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4853 }
4854 }
4855
4856 // To install functions for atexit system call
4857 extern "C" {
4858 static void perfMemory_exit_helper() {
4859 perfMemory_exit();
4860 }
4861 }
4862
4863 // this is called _after_ the global arguments have been parsed
4864 jint os::init_2(void)
4865 {
4866 Linux::fast_thread_clock_init();
4867
4868 // Allocate a single page and mark it as readable for safepoint polling
4869 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4870 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4871
4872 os::set_polling_page( polling_page );
4873
4874 #ifndef PRODUCT
4875 if(Verbose && PrintMiscellaneous)
4876 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4877 #endif
4878
4879 if (!UseMembar) {
4880 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4881 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4882 os::set_memory_serialize_page( mem_serialize_page );
4883
4884 #ifndef PRODUCT
4885 if(Verbose && PrintMiscellaneous)
4886 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4887 #endif
4888 }
4889
4890 // initialize suspend/resume support - must do this before signal_sets_init()
4891 if (SR_initialize() != 0) {
4892 perror("SR_initialize failed");
4893 return JNI_ERR;
4894 }
4895
4896 Linux::signal_sets_init();
4897 Linux::install_signal_handlers();
4898
4899 // Check minimum allowable stack size for thread creation and to initialize
4900 // the java system classes, including StackOverflowError - depends on page
4901 // size. Add a page for compiler2 recursion in main thread.
4902 // Add in 2*BytesPerWord times page size to account for VM stack during
4903 // class initialization depending on 32 or 64 bit VM.
4904 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4905 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4906 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4907
4908 size_t threadStackSizeInBytes = ThreadStackSize * K;
4909 if (threadStackSizeInBytes != 0 &&
4910 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4911 tty->print_cr("\nThe stack size specified is too small, "
4912 "Specify at least %dk",
4913 os::Linux::min_stack_allowed/ K);
4914 return JNI_ERR;
4915 }
4916
4917 // Make the stack size a multiple of the page size so that
4918 // the yellow/red zones can be guarded.
4919 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4920 vm_page_size()));
4921
4922 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4923
4924 #if defined(IA32)
4925 workaround_expand_exec_shield_cs_limit();
4926 #endif
4927
4928 Linux::libpthread_init();
4929 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4930 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4931 Linux::glibc_version(), Linux::libpthread_version(),
4932 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4933 }
4934
4935 if (UseNUMA) {
4936 if (!Linux::libnuma_init()) {
4937 UseNUMA = false;
4938 } else {
4939 if ((Linux::numa_max_node() < 1)) {
4940 // There's only one node(they start from 0), disable NUMA.
4941 UseNUMA = false;
4942 }
4943 }
4944 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4945 // we can make the adaptive lgrp chunk resizing work. If the user specified
4946 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4947 // disable adaptive resizing.
4948 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4949 if (FLAG_IS_DEFAULT(UseNUMA)) {
4950 UseNUMA = false;
4951 } else {
4952 if (FLAG_IS_DEFAULT(UseLargePages) &&
4953 FLAG_IS_DEFAULT(UseSHM) &&
4954 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4955 UseLargePages = false;
4956 } else {
4957 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4958 UseAdaptiveSizePolicy = false;
4959 UseAdaptiveNUMAChunkSizing = false;
4960 }
4961 }
4962 }
4963 if (!UseNUMA && ForceNUMA) {
4964 UseNUMA = true;
4965 }
4966 }
4967
4968 if (MaxFDLimit) {
4969 // set the number of file descriptors to max. print out error
4970 // if getrlimit/setrlimit fails but continue regardless.
4971 struct rlimit nbr_files;
4972 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4973 if (status != 0) {
4974 if (PrintMiscellaneous && (Verbose || WizardMode))
4975 perror("os::init_2 getrlimit failed");
4976 } else {
4977 nbr_files.rlim_cur = nbr_files.rlim_max;
4978 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4979 if (status != 0) {
4980 if (PrintMiscellaneous && (Verbose || WizardMode))
4981 perror("os::init_2 setrlimit failed");
4982 }
4983 }
4984 }
4985
4986 // Initialize lock used to serialize thread creation (see os::create_thread)
4987 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4988
4989 // at-exit methods are called in the reverse order of their registration.
4990 // atexit functions are called on return from main or as a result of a
4991 // call to exit(3C). There can be only 32 of these functions registered
4992 // and atexit() does not set errno.
4993
4994 if (PerfAllowAtExitRegistration) {
4995 // only register atexit functions if PerfAllowAtExitRegistration is set.
4996 // atexit functions can be delayed until process exit time, which
4997 // can be problematic for embedded VM situations. Embedded VMs should
4998 // call DestroyJavaVM() to assure that VM resources are released.
4999
5000 // note: perfMemory_exit_helper atexit function may be removed in
5001 // the future if the appropriate cleanup code can be added to the
5002 // VM_Exit VMOperation's doit method.
5003 if (atexit(perfMemory_exit_helper) != 0) {
5004 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5005 }
5006 }
5007
5008 // initialize thread priority policy
5009 prio_init();
5010
5011 return JNI_OK;
5012 }
5013
5014 // Mark the polling page as unreadable
5015 void os::make_polling_page_unreadable(void) {
5016 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5017 fatal("Could not disable polling page");
5018 };
5019
5020 // Mark the polling page as readable
5021 void os::make_polling_page_readable(void) {
5022 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5023 fatal("Could not enable polling page");
5024 }
5025 };
5026
5027 static int os_cpu_count(const cpu_set_t* cpus) {
5028 int count = 0;
5029 // only look up to the number of configured processors
5030 for (int i = 0; i < os::processor_count(); i++) {
5031 if (CPU_ISSET(i, cpus)) {
5032 count++;
5033 }
5034 }
5035 return count;
5036 }
5037
5038 // Get the current number of available processors for this process.
5039 // This value can change at any time during a process's lifetime.
5040 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5041 // If anything goes wrong we fallback to returning the number of online
5042 // processors - which can be greater than the number available to the process.
5043 int os::active_processor_count() {
5044 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5045 int cpus_size = sizeof(cpu_set_t);
5046 int cpu_count = 0;
5047
5048 // pid 0 means the current thread - which we have to assume represents the process
5049 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5050 cpu_count = os_cpu_count(&cpus);
5051 if(PrintActiveCpus) {
5052 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5053 }
5054 }
5055 else {
5056 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5057 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5058 "which may exceed available processors", strerror(errno), cpu_count);
5059 }
5060
5061 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5062 return cpu_count;
5063 }
5064
5065 void os::set_native_thread_name(const char *name) {
5066 // Not yet implemented.
5067 return;
5068 }
5069
5070 bool os::distribute_processes(uint length, uint* distribution) {
5071 // Not yet implemented.
5072 return false;
5073 }
5074
5075 bool os::bind_to_processor(uint processor_id) {
5076 // Not yet implemented.
5077 return false;
5078 }
5079
5080 ///
5081
5082 void os::SuspendedThreadTask::internal_do_task() {
5083 if (do_suspend(_thread->osthread())) {
5084 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5085 do_task(context);
5086 do_resume(_thread->osthread());
5087 }
5088 }
5089
5090 class PcFetcher : public os::SuspendedThreadTask {
5091 public:
5092 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5093 ExtendedPC result();
5094 protected:
5095 void do_task(const os::SuspendedThreadTaskContext& context);
5096 private:
5097 ExtendedPC _epc;
5098 };
5099
5100 ExtendedPC PcFetcher::result() {
5101 guarantee(is_done(), "task is not done yet.");
5102 return _epc;
5103 }
5104
5105 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5106 Thread* thread = context.thread();
5107 OSThread* osthread = thread->osthread();
5108 if (osthread->ucontext() != NULL) {
5109 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5110 } else {
5111 // NULL context is unexpected, double-check this is the VMThread
5112 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5113 }
5114 }
5115
5116 // Suspends the target using the signal mechanism and then grabs the PC before
5117 // resuming the target. Used by the flat-profiler only
5118 ExtendedPC os::get_thread_pc(Thread* thread) {
5119 // Make sure that it is called by the watcher for the VMThread
5120 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5121 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5122
5123 PcFetcher fetcher(thread);
5124 fetcher.run();
5125 return fetcher.result();
5126 }
5127
5128 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5129 {
5130 if (is_NPTL()) {
5131 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5132 } else {
5133 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5134 // word back to default 64bit precision if condvar is signaled. Java
5135 // wants 53bit precision. Save and restore current value.
5136 int fpu = get_fpu_control_word();
5137 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5138 set_fpu_control_word(fpu);
5139 return status;
5140 }
5141 }
5142
5143 ////////////////////////////////////////////////////////////////////////////////
5144 // debug support
5145
5146 bool os::find(address addr, outputStream* st) {
5147 Dl_info dlinfo;
5148 memset(&dlinfo, 0, sizeof(dlinfo));
5149 if (dladdr(addr, &dlinfo) != 0) {
5150 st->print(PTR_FORMAT ": ", addr);
5151 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5152 st->print("%s+%#x", dlinfo.dli_sname,
5153 addr - (intptr_t)dlinfo.dli_saddr);
5154 } else if (dlinfo.dli_fbase != NULL) {
5155 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5156 } else {
5157 st->print("<absolute address>");
5158 }
5159 if (dlinfo.dli_fname != NULL) {
5160 st->print(" in %s", dlinfo.dli_fname);
5161 }
5162 if (dlinfo.dli_fbase != NULL) {
5163 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5164 }
5165 st->cr();
5166
5167 if (Verbose) {
5168 // decode some bytes around the PC
5169 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5170 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5171 address lowest = (address) dlinfo.dli_sname;
5172 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5173 if (begin < lowest) begin = lowest;
5174 Dl_info dlinfo2;
5175 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5176 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5177 end = (address) dlinfo2.dli_saddr;
5178 Disassembler::decode(begin, end, st);
5179 }
5180 return true;
5181 }
5182 return false;
5183 }
5184
5185 ////////////////////////////////////////////////////////////////////////////////
5186 // misc
5187
5188 // This does not do anything on Linux. This is basically a hook for being
5189 // able to use structured exception handling (thread-local exception filters)
5190 // on, e.g., Win32.
5191 void
5192 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5193 JavaCallArguments* args, Thread* thread) {
5194 f(value, method, args, thread);
5195 }
5196
5197 void os::print_statistics() {
5198 }
5199
5200 int os::message_box(const char* title, const char* message) {
5201 int i;
5202 fdStream err(defaultStream::error_fd());
5203 for (i = 0; i < 78; i++) err.print_raw("=");
5204 err.cr();
5205 err.print_raw_cr(title);
5206 for (i = 0; i < 78; i++) err.print_raw("-");
5207 err.cr();
5208 err.print_raw_cr(message);
5209 for (i = 0; i < 78; i++) err.print_raw("=");
5210 err.cr();
5211
5212 char buf[16];
5213 // Prevent process from exiting upon "read error" without consuming all CPU
5214 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5215
5216 return buf[0] == 'y' || buf[0] == 'Y';
5217 }
5218
5219 int os::stat(const char *path, struct stat *sbuf) {
5220 char pathbuf[MAX_PATH];
5221 if (strlen(path) > MAX_PATH - 1) {
5222 errno = ENAMETOOLONG;
5223 return -1;
5224 }
5225 os::native_path(strcpy(pathbuf, path));
5226 return ::stat(pathbuf, sbuf);
5227 }
5228
5229 bool os::check_heap(bool force) {
5230 return true;
5231 }
5232
5233 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5234 return ::vsnprintf(buf, count, format, args);
5235 }
5236
5237 // Is a (classpath) directory empty?
5238 bool os::dir_is_empty(const char* path) {
5239 DIR *dir = NULL;
5240 struct dirent *ptr;
5241
5242 dir = opendir(path);
5243 if (dir == NULL) return true;
5244
5245 /* Scan the directory */
5246 bool result = true;
5247 char buf[sizeof(struct dirent) + MAX_PATH];
5248 while (result && (ptr = ::readdir(dir)) != NULL) {
5249 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5250 result = false;
5251 }
5252 }
5253 closedir(dir);
5254 return result;
5255 }
5256
5257 // This code originates from JDK's sysOpen and open64_w
5258 // from src/solaris/hpi/src/system_md.c
5259
5260 #ifndef O_DELETE
5261 #define O_DELETE 0x10000
5262 #endif
5263
5264 // Open a file. Unlink the file immediately after open returns
5265 // if the specified oflag has the O_DELETE flag set.
5266 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5267
5268 int os::open(const char *path, int oflag, int mode) {
5269
5270 if (strlen(path) > MAX_PATH - 1) {
5271 errno = ENAMETOOLONG;
5272 return -1;
5273 }
5274 int fd;
5275 int o_delete = (oflag & O_DELETE);
5276 oflag = oflag & ~O_DELETE;
5277
5278 fd = ::open64(path, oflag, mode);
5279 if (fd == -1) return -1;
5280
5281 //If the open succeeded, the file might still be a directory
5282 {
5283 struct stat64 buf64;
5284 int ret = ::fstat64(fd, &buf64);
5285 int st_mode = buf64.st_mode;
5286
5287 if (ret != -1) {
5288 if ((st_mode & S_IFMT) == S_IFDIR) {
5289 errno = EISDIR;
5290 ::close(fd);
5291 return -1;
5292 }
5293 } else {
5294 ::close(fd);
5295 return -1;
5296 }
5297 }
5298
5299 /*
5300 * All file descriptors that are opened in the JVM and not
5301 * specifically destined for a subprocess should have the
5302 * close-on-exec flag set. If we don't set it, then careless 3rd
5303 * party native code might fork and exec without closing all
5304 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5305 * UNIXProcess.c), and this in turn might:
5306 *
5307 * - cause end-of-file to fail to be detected on some file
5308 * descriptors, resulting in mysterious hangs, or
5309 *
5310 * - might cause an fopen in the subprocess to fail on a system
5311 * suffering from bug 1085341.
5312 *
5313 * (Yes, the default setting of the close-on-exec flag is a Unix
5314 * design flaw)
5315 *
5316 * See:
5317 * 1085341: 32-bit stdio routines should support file descriptors >255
5318 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5319 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5320 */
5321 #ifdef FD_CLOEXEC
5322 {
5323 int flags = ::fcntl(fd, F_GETFD);
5324 if (flags != -1)
5325 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5326 }
5327 #endif
5328
5329 if (o_delete != 0) {
5330 ::unlink(path);
5331 }
5332 return fd;
5333 }
5334
5335
5336 // create binary file, rewriting existing file if required
5337 int os::create_binary_file(const char* path, bool rewrite_existing) {
5338 int oflags = O_WRONLY | O_CREAT;
5339 if (!rewrite_existing) {
5340 oflags |= O_EXCL;
5341 }
5342 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5343 }
5344
5345 // return current position of file pointer
5346 jlong os::current_file_offset(int fd) {
5347 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5348 }
5349
5350 // move file pointer to the specified offset
5351 jlong os::seek_to_file_offset(int fd, jlong offset) {
5352 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5353 }
5354
5355 // This code originates from JDK's sysAvailable
5356 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5357
5358 int os::available(int fd, jlong *bytes) {
5359 jlong cur, end;
5360 int mode;
5361 struct stat64 buf64;
5362
5363 if (::fstat64(fd, &buf64) >= 0) {
5364 mode = buf64.st_mode;
5365 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5366 /*
5367 * XXX: is the following call interruptible? If so, this might
5368 * need to go through the INTERRUPT_IO() wrapper as for other
5369 * blocking, interruptible calls in this file.
5370 */
5371 int n;
5372 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5373 *bytes = n;
5374 return 1;
5375 }
5376 }
5377 }
5378 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5379 return 0;
5380 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5381 return 0;
5382 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5383 return 0;
5384 }
5385 *bytes = end - cur;
5386 return 1;
5387 }
5388
5389 int os::socket_available(int fd, jint *pbytes) {
5390 // Linux doc says EINTR not returned, unlike Solaris
5391 int ret = ::ioctl(fd, FIONREAD, pbytes);
5392
5393 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5394 // is expected to return 0 on failure and 1 on success to the jdk.
5395 return (ret < 0) ? 0 : 1;
5396 }
5397
5398 // Map a block of memory.
5399 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5400 char *addr, size_t bytes, bool read_only,
5401 bool allow_exec) {
5402 int prot;
5403 int flags = MAP_PRIVATE;
5404
5405 if (read_only) {
5406 prot = PROT_READ;
5407 } else {
5408 prot = PROT_READ | PROT_WRITE;
5409 }
5410
5411 if (allow_exec) {
5412 prot |= PROT_EXEC;
5413 }
5414
5415 if (addr != NULL) {
5416 flags |= MAP_FIXED;
5417 }
5418
5419 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5420 fd, file_offset);
5421 if (mapped_address == MAP_FAILED) {
5422 return NULL;
5423 }
5424 return mapped_address;
5425 }
5426
5427
5428 // Remap a block of memory.
5429 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5430 char *addr, size_t bytes, bool read_only,
5431 bool allow_exec) {
5432 // same as map_memory() on this OS
5433 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5434 allow_exec);
5435 }
5436
5437
5438 // Unmap a block of memory.
5439 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5440 return munmap(addr, bytes) == 0;
5441 }
5442
5443 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5444
5445 static clockid_t thread_cpu_clockid(Thread* thread) {
5446 pthread_t tid = thread->osthread()->pthread_id();
5447 clockid_t clockid;
5448
5449 // Get thread clockid
5450 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5451 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5452 return clockid;
5453 }
5454
5455 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5456 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5457 // of a thread.
5458 //
5459 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5460 // the fast estimate available on the platform.
5461
5462 jlong os::current_thread_cpu_time() {
5463 if (os::Linux::supports_fast_thread_cpu_time()) {
5464 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5465 } else {
5466 // return user + sys since the cost is the same
5467 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5468 }
5469 }
5470
5471 jlong os::thread_cpu_time(Thread* thread) {
5472 // consistent with what current_thread_cpu_time() returns
5473 if (os::Linux::supports_fast_thread_cpu_time()) {
5474 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5475 } else {
5476 return slow_thread_cpu_time(thread, true /* user + sys */);
5477 }
5478 }
5479
5480 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5481 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5482 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5483 } else {
5484 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5485 }
5486 }
5487
5488 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5489 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5490 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5491 } else {
5492 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5493 }
5494 }
5495
5496 //
5497 // -1 on error.
5498 //
5499
5500 PRAGMA_DIAG_PUSH
5501 PRAGMA_FORMAT_NONLITERAL_IGNORED
5502 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5503 static bool proc_task_unchecked = true;
5504 static const char *proc_stat_path = "/proc/%d/stat";
5505 pid_t tid = thread->osthread()->thread_id();
5506 char *s;
5507 char stat[2048];
5508 int statlen;
5509 char proc_name[64];
5510 int count;
5511 long sys_time, user_time;
5512 char cdummy;
5513 int idummy;
5514 long ldummy;
5515 FILE *fp;
5516
5517 // The /proc/<tid>/stat aggregates per-process usage on
5518 // new Linux kernels 2.6+ where NPTL is supported.
5519 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5520 // See bug 6328462.
5521 // There possibly can be cases where there is no directory
5522 // /proc/self/task, so we check its availability.
5523 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5524 // This is executed only once
5525 proc_task_unchecked = false;
5526 fp = fopen("/proc/self/task", "r");
5527 if (fp != NULL) {
5528 proc_stat_path = "/proc/self/task/%d/stat";
5529 fclose(fp);
5530 }
5531 }
5532
5533 sprintf(proc_name, proc_stat_path, tid);
5534 fp = fopen(proc_name, "r");
5535 if ( fp == NULL ) return -1;
5536 statlen = fread(stat, 1, 2047, fp);
5537 stat[statlen] = '\0';
5538 fclose(fp);
5539
5540 // Skip pid and the command string. Note that we could be dealing with
5541 // weird command names, e.g. user could decide to rename java launcher
5542 // to "java 1.4.2 :)", then the stat file would look like
5543 // 1234 (java 1.4.2 :)) R ... ...
5544 // We don't really need to know the command string, just find the last
5545 // occurrence of ")" and then start parsing from there. See bug 4726580.
5546 s = strrchr(stat, ')');
5547 if (s == NULL ) return -1;
5548
5549 // Skip blank chars
5550 do s++; while (isspace(*s));
5551
5552 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5553 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5554 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5555 &user_time, &sys_time);
5556 if ( count != 13 ) return -1;
5557 if (user_sys_cpu_time) {
5558 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5559 } else {
5560 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5561 }
5562 }
5563 PRAGMA_DIAG_POP
5564
5565 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5566 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5567 info_ptr->may_skip_backward = false; // elapsed time not wall time
5568 info_ptr->may_skip_forward = false; // elapsed time not wall time
5569 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5570 }
5571
5572 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5573 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5574 info_ptr->may_skip_backward = false; // elapsed time not wall time
5575 info_ptr->may_skip_forward = false; // elapsed time not wall time
5576 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5577 }
5578
5579 bool os::is_thread_cpu_time_supported() {
5580 return true;
5581 }
5582
5583 // System loadavg support. Returns -1 if load average cannot be obtained.
5584 // Linux doesn't yet have a (official) notion of processor sets,
5585 // so just return the system wide load average.
5586 int os::loadavg(double loadavg[], int nelem) {
5587 return ::getloadavg(loadavg, nelem);
5588 }
5589
5590 void os::pause() {
5591 char filename[MAX_PATH];
5592 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5593 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5594 } else {
5595 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5596 }
5597
5598 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5599 if (fd != -1) {
5600 struct stat buf;
5601 ::close(fd);
5602 while (::stat(filename, &buf) == 0) {
5603 (void)::poll(NULL, 0, 100);
5604 }
5605 } else {
5606 jio_fprintf(stderr,
5607 "Could not open pause file '%s', continuing immediately.\n", filename);
5608 }
5609 }
5610
5611
5612 // Refer to the comments in os_solaris.cpp park-unpark.
5613 //
5614 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5615 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5616 // For specifics regarding the bug see GLIBC BUGID 261237 :
5617 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5618 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5619 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5620 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5621 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5622 // and monitorenter when we're using 1-0 locking. All those operations may result in
5623 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5624 // of libpthread avoids the problem, but isn't practical.
5625 //
5626 // Possible remedies:
5627 //
5628 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5629 // This is palliative and probabilistic, however. If the thread is preempted
5630 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5631 // than the minimum period may have passed, and the abstime may be stale (in the
5632 // past) resultin in a hang. Using this technique reduces the odds of a hang
5633 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5634 //
5635 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5636 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5637 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5638 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5639 // thread.
5640 //
5641 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5642 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5643 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5644 // This also works well. In fact it avoids kernel-level scalability impediments
5645 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5646 // timers in a graceful fashion.
5647 //
5648 // 4. When the abstime value is in the past it appears that control returns
5649 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5650 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5651 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5652 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5653 // It may be possible to avoid reinitialization by checking the return
5654 // value from pthread_cond_timedwait(). In addition to reinitializing the
5655 // condvar we must establish the invariant that cond_signal() is only called
5656 // within critical sections protected by the adjunct mutex. This prevents
5657 // cond_signal() from "seeing" a condvar that's in the midst of being
5658 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5659 // desirable signal-after-unlock optimization that avoids futile context switching.
5660 //
5661 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5662 // structure when a condvar is used or initialized. cond_destroy() would
5663 // release the helper structure. Our reinitialize-after-timedwait fix
5664 // put excessive stress on malloc/free and locks protecting the c-heap.
5665 //
5666 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5667 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5668 // and only enabling the work-around for vulnerable environments.
5669
5670 // utility to compute the abstime argument to timedwait:
5671 // millis is the relative timeout time
5672 // abstime will be the absolute timeout time
5673 // TODO: replace compute_abstime() with unpackTime()
5674
5675 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5676 if (millis < 0) millis = 0;
5677
5678 jlong seconds = millis / 1000;
5679 millis %= 1000;
5680 if (seconds > 50000000) { // see man cond_timedwait(3T)
5681 seconds = 50000000;
5682 }
5683
5684 if (os::Linux::supports_monotonic_clock()) {
5685 struct timespec now;
5686 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5687 assert_status(status == 0, status, "clock_gettime");
5688 abstime->tv_sec = now.tv_sec + seconds;
5689 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5690 if (nanos >= NANOSECS_PER_SEC) {
5691 abstime->tv_sec += 1;
5692 nanos -= NANOSECS_PER_SEC;
5693 }
5694 abstime->tv_nsec = nanos;
5695 } else {
5696 struct timeval now;
5697 int status = gettimeofday(&now, NULL);
5698 assert(status == 0, "gettimeofday");
5699 abstime->tv_sec = now.tv_sec + seconds;
5700 long usec = now.tv_usec + millis * 1000;
5701 if (usec >= 1000000) {
5702 abstime->tv_sec += 1;
5703 usec -= 1000000;
5704 }
5705 abstime->tv_nsec = usec * 1000;
5706 }
5707 return abstime;
5708 }
5709
5710
5711 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5712 // Conceptually TryPark() should be equivalent to park(0).
5713
5714 int os::PlatformEvent::TryPark() {
5715 for (;;) {
5716 const int v = _Event ;
5717 guarantee ((v == 0) || (v == 1), "invariant") ;
5718 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5719 }
5720 }
5721
5722 void os::PlatformEvent::park() { // AKA "down()"
5723 // Invariant: Only the thread associated with the Event/PlatformEvent
5724 // may call park().
5725 // TODO: assert that _Assoc != NULL or _Assoc == Self
5726 int v ;
5727 for (;;) {
5728 v = _Event ;
5729 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5730 }
5731 guarantee (v >= 0, "invariant") ;
5732 if (v == 0) {
5733 // Do this the hard way by blocking ...
5734 int status = pthread_mutex_lock(_mutex);
5735 assert_status(status == 0, status, "mutex_lock");
5736 guarantee (_nParked == 0, "invariant") ;
5737 ++ _nParked ;
5738 while (_Event < 0) {
5739 status = pthread_cond_wait(_cond, _mutex);
5740 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5741 // Treat this the same as if the wait was interrupted
5742 if (status == ETIME) { status = EINTR; }
5743 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5744 }
5745 -- _nParked ;
5746
5747 _Event = 0 ;
5748 status = pthread_mutex_unlock(_mutex);
5749 assert_status(status == 0, status, "mutex_unlock");
5750 // Paranoia to ensure our locked and lock-free paths interact
5751 // correctly with each other.
5752 OrderAccess::fence();
5753 }
5754 guarantee (_Event >= 0, "invariant") ;
5755 }
5756
5757 int os::PlatformEvent::park(jlong millis) {
5758 guarantee (_nParked == 0, "invariant") ;
5759
5760 int v ;
5761 for (;;) {
5762 v = _Event ;
5763 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5764 }
5765 guarantee (v >= 0, "invariant") ;
5766 if (v != 0) return OS_OK ;
5767
5768 // We do this the hard way, by blocking the thread.
5769 // Consider enforcing a minimum timeout value.
5770 struct timespec abst;
5771 compute_abstime(&abst, millis);
5772
5773 int ret = OS_TIMEOUT;
5774 int status = pthread_mutex_lock(_mutex);
5775 assert_status(status == 0, status, "mutex_lock");
5776 guarantee (_nParked == 0, "invariant") ;
5777 ++_nParked ;
5778
5779 // Object.wait(timo) will return because of
5780 // (a) notification
5781 // (b) timeout
5782 // (c) thread.interrupt
5783 //
5784 // Thread.interrupt and object.notify{All} both call Event::set.
5785 // That is, we treat thread.interrupt as a special case of notification.
5786 // The underlying Solaris implementation, cond_timedwait, admits
5787 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5788 // JVM from making those visible to Java code. As such, we must
5789 // filter out spurious wakeups. We assume all ETIME returns are valid.
5790 //
5791 // TODO: properly differentiate simultaneous notify+interrupt.
5792 // In that case, we should propagate the notify to another waiter.
5793
5794 while (_Event < 0) {
5795 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5796 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5797 pthread_cond_destroy (_cond);
5798 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5799 }
5800 assert_status(status == 0 || status == EINTR ||
5801 status == ETIME || status == ETIMEDOUT,
5802 status, "cond_timedwait");
5803 if (!FilterSpuriousWakeups) break ; // previous semantics
5804 if (status == ETIME || status == ETIMEDOUT) break ;
5805 // We consume and ignore EINTR and spurious wakeups.
5806 }
5807 --_nParked ;
5808 if (_Event >= 0) {
5809 ret = OS_OK;
5810 }
5811 _Event = 0 ;
5812 status = pthread_mutex_unlock(_mutex);
5813 assert_status(status == 0, status, "mutex_unlock");
5814 assert (_nParked == 0, "invariant") ;
5815 // Paranoia to ensure our locked and lock-free paths interact
5816 // correctly with each other.
5817 OrderAccess::fence();
5818 return ret;
5819 }
5820
5821 void os::PlatformEvent::unpark() {
5822 // Transitions for _Event:
5823 // 0 :=> 1
5824 // 1 :=> 1
5825 // -1 :=> either 0 or 1; must signal target thread
5826 // That is, we can safely transition _Event from -1 to either
5827 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5828 // unpark() calls.
5829 // See also: "Semaphores in Plan 9" by Mullender & Cox
5830 //
5831 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5832 // that it will take two back-to-back park() calls for the owning
5833 // thread to block. This has the benefit of forcing a spurious return
5834 // from the first park() call after an unpark() call which will help
5835 // shake out uses of park() and unpark() without condition variables.
5836
5837 if (Atomic::xchg(1, &_Event) >= 0) return;
5838
5839 // Wait for the thread associated with the event to vacate
5840 int status = pthread_mutex_lock(_mutex);
5841 assert_status(status == 0, status, "mutex_lock");
5842 int AnyWaiters = _nParked;
5843 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5844 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5845 AnyWaiters = 0;
5846 pthread_cond_signal(_cond);
5847 }
5848 status = pthread_mutex_unlock(_mutex);
5849 assert_status(status == 0, status, "mutex_unlock");
5850 if (AnyWaiters != 0) {
5851 status = pthread_cond_signal(_cond);
5852 assert_status(status == 0, status, "cond_signal");
5853 }
5854
5855 // Note that we signal() _after dropping the lock for "immortal" Events.
5856 // This is safe and avoids a common class of futile wakeups. In rare
5857 // circumstances this can cause a thread to return prematurely from
5858 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5859 // simply re-test the condition and re-park itself.
5860 }
5861
5862
5863 // JSR166
5864 // -------------------------------------------------------
5865
5866 /*
5867 * The solaris and linux implementations of park/unpark are fairly
5868 * conservative for now, but can be improved. They currently use a
5869 * mutex/condvar pair, plus a a count.
5870 * Park decrements count if > 0, else does a condvar wait. Unpark
5871 * sets count to 1 and signals condvar. Only one thread ever waits
5872 * on the condvar. Contention seen when trying to park implies that someone
5873 * is unparking you, so don't wait. And spurious returns are fine, so there
5874 * is no need to track notifications.
5875 */
5876
5877 /*
5878 * This code is common to linux and solaris and will be moved to a
5879 * common place in dolphin.
5880 *
5881 * The passed in time value is either a relative time in nanoseconds
5882 * or an absolute time in milliseconds. Either way it has to be unpacked
5883 * into suitable seconds and nanoseconds components and stored in the
5884 * given timespec structure.
5885 * Given time is a 64-bit value and the time_t used in the timespec is only
5886 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5887 * overflow if times way in the future are given. Further on Solaris versions
5888 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5889 * number of seconds, in abstime, is less than current_time + 100,000,000.
5890 * As it will be 28 years before "now + 100000000" will overflow we can
5891 * ignore overflow and just impose a hard-limit on seconds using the value
5892 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5893 * years from "now".
5894 */
5895
5896 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5897 assert (time > 0, "convertTime");
5898 time_t max_secs = 0;
5899
5900 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5901 struct timeval now;
5902 int status = gettimeofday(&now, NULL);
5903 assert(status == 0, "gettimeofday");
5904
5905 max_secs = now.tv_sec + MAX_SECS;
5906
5907 if (isAbsolute) {
5908 jlong secs = time / 1000;
5909 if (secs > max_secs) {
5910 absTime->tv_sec = max_secs;
5911 } else {
5912 absTime->tv_sec = secs;
5913 }
5914 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5915 } else {
5916 jlong secs = time / NANOSECS_PER_SEC;
5917 if (secs >= MAX_SECS) {
5918 absTime->tv_sec = max_secs;
5919 absTime->tv_nsec = 0;
5920 } else {
5921 absTime->tv_sec = now.tv_sec + secs;
5922 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5923 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5924 absTime->tv_nsec -= NANOSECS_PER_SEC;
5925 ++absTime->tv_sec; // note: this must be <= max_secs
5926 }
5927 }
5928 }
5929 } else {
5930 // must be relative using monotonic clock
5931 struct timespec now;
5932 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5933 assert_status(status == 0, status, "clock_gettime");
5934 max_secs = now.tv_sec + MAX_SECS;
5935 jlong secs = time / NANOSECS_PER_SEC;
5936 if (secs >= MAX_SECS) {
5937 absTime->tv_sec = max_secs;
5938 absTime->tv_nsec = 0;
5939 } else {
5940 absTime->tv_sec = now.tv_sec + secs;
5941 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5942 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5943 absTime->tv_nsec -= NANOSECS_PER_SEC;
5944 ++absTime->tv_sec; // note: this must be <= max_secs
5945 }
5946 }
5947 }
5948 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5949 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5950 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5951 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5952 }
5953
5954 void Parker::park(bool isAbsolute, jlong time) {
5955 // Ideally we'd do something useful while spinning, such
5956 // as calling unpackTime().
5957
5958 // Optional fast-path check:
5959 // Return immediately if a permit is available.
5960 // We depend on Atomic::xchg() having full barrier semantics
5961 // since we are doing a lock-free update to _counter.
5962 if (Atomic::xchg(0, &_counter) > 0) return;
5963
5964 Thread* thread = Thread::current();
5965 assert(thread->is_Java_thread(), "Must be JavaThread");
5966 JavaThread *jt = (JavaThread *)thread;
5967
5968 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5969 // Check interrupt before trying to wait
5970 if (Thread::is_interrupted(thread, false)) {
5971 return;
5972 }
5973
5974 // Next, demultiplex/decode time arguments
5975 timespec absTime;
5976 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5977 return;
5978 }
5979 if (time > 0) {
5980 unpackTime(&absTime, isAbsolute, time);
5981 }
5982
5983
5984 // Enter safepoint region
5985 // Beware of deadlocks such as 6317397.
5986 // The per-thread Parker:: mutex is a classic leaf-lock.
5987 // In particular a thread must never block on the Threads_lock while
5988 // holding the Parker:: mutex. If safepoints are pending both the
5989 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5990 ThreadBlockInVM tbivm(jt);
5991
5992 // Don't wait if cannot get lock since interference arises from
5993 // unblocking. Also. check interrupt before trying wait
5994 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5995 return;
5996 }
5997
5998 int status ;
5999 if (_counter > 0) { // no wait needed
6000 _counter = 0;
6001 status = pthread_mutex_unlock(_mutex);
6002 assert (status == 0, "invariant") ;
6003 // Paranoia to ensure our locked and lock-free paths interact
6004 // correctly with each other and Java-level accesses.
6005 OrderAccess::fence();
6006 return;
6007 }
6008
6009 #ifdef ASSERT
6010 // Don't catch signals while blocked; let the running threads have the signals.
6011 // (This allows a debugger to break into the running thread.)
6012 sigset_t oldsigs;
6013 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6014 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6015 #endif
6016
6017 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6018 jt->set_suspend_equivalent();
6019 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6020
6021 assert(_cur_index == -1, "invariant");
6022 if (time == 0) {
6023 _cur_index = REL_INDEX; // arbitrary choice when not timed
6024 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6025 } else {
6026 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6027 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6028 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6029 pthread_cond_destroy (&_cond[_cur_index]) ;
6030 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6031 }
6032 }
6033 _cur_index = -1;
6034 assert_status(status == 0 || status == EINTR ||
6035 status == ETIME || status == ETIMEDOUT,
6036 status, "cond_timedwait");
6037
6038 #ifdef ASSERT
6039 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6040 #endif
6041
6042 _counter = 0 ;
6043 status = pthread_mutex_unlock(_mutex) ;
6044 assert_status(status == 0, status, "invariant") ;
6045 // Paranoia to ensure our locked and lock-free paths interact
6046 // correctly with each other and Java-level accesses.
6047 OrderAccess::fence();
6048
6049 // If externally suspended while waiting, re-suspend
6050 if (jt->handle_special_suspend_equivalent_condition()) {
6051 jt->java_suspend_self();
6052 }
6053 }
6054
6055 void Parker::unpark() {
6056 int s, status ;
6057 status = pthread_mutex_lock(_mutex);
6058 assert (status == 0, "invariant") ;
6059 s = _counter;
6060 _counter = 1;
6061 if (s < 1) {
6062 // thread might be parked
6063 if (_cur_index != -1) {
6064 // thread is definitely parked
6065 if (WorkAroundNPTLTimedWaitHang) {
6066 status = pthread_cond_signal (&_cond[_cur_index]);
6067 assert (status == 0, "invariant");
6068 status = pthread_mutex_unlock(_mutex);
6069 assert (status == 0, "invariant");
6070 } else {
6071 // must capture correct index before unlocking
6072 int index = _cur_index;
6073 status = pthread_mutex_unlock(_mutex);
6074 assert (status == 0, "invariant");
6075 status = pthread_cond_signal (&_cond[index]);
6076 assert (status == 0, "invariant");
6077 }
6078 } else {
6079 pthread_mutex_unlock(_mutex);
6080 assert (status == 0, "invariant") ;
6081 }
6082 } else {
6083 pthread_mutex_unlock(_mutex);
6084 assert (status == 0, "invariant") ;
6085 }
6086 }
6087
6088
6089 extern char** environ;
6090
6091 // Run the specified command in a separate process. Return its exit value,
6092 // or -1 on failure (e.g. can't fork a new process).
6093 // Unlike system(), this function can be called from signal handler. It
6094 // doesn't block SIGINT et al.
6095 int os::fork_and_exec(char* cmd) {
6096 const char * argv[4] = {"sh", "-c", cmd, NULL};
6097
6098 pid_t pid = fork();
6099
6100 if (pid < 0) {
6101 // fork failed
6102 return -1;
6103
6104 } else if (pid == 0) {
6105 // child process
6106
6107 execve("/bin/sh", (char* const*)argv, environ);
6108
6109 // execve failed
6110 _exit(-1);
6111
6112 } else {
6113 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6114 // care about the actual exit code, for now.
6115
6116 int status;
6117
6118 // Wait for the child process to exit. This returns immediately if
6119 // the child has already exited. */
6120 while (waitpid(pid, &status, 0) < 0) {
6121 switch (errno) {
6122 case ECHILD: return 0;
6123 case EINTR: break;
6124 default: return -1;
6125 }
6126 }
6127
6128 if (WIFEXITED(status)) {
6129 // The child exited normally; get its exit code.
6130 return WEXITSTATUS(status);
6131 } else if (WIFSIGNALED(status)) {
6132 // The child exited because of a signal
6133 // The best value to return is 0x80 + signal number,
6134 // because that is what all Unix shells do, and because
6135 // it allows callers to distinguish between process exit and
6136 // process death by signal.
6137 return 0x80 + WTERMSIG(status);
6138 } else {
6139 // Unknown exit code; pass it through
6140 return status;
6141 }
6142 }
6143 }
6144
6145 // is_headless_jre()
6146 //
6147 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6148 // in order to report if we are running in a headless jre
6149 //
6150 // Since JDK8 xawt/libmawt.so was moved into the same directory
6151 // as libawt.so, and renamed libawt_xawt.so
6152 //
6153 bool os::is_headless_jre() {
6154 struct stat statbuf;
6155 char buf[MAXPATHLEN];
6156 char libmawtpath[MAXPATHLEN];
6157 const char *xawtstr = "/xawt/libmawt.so";
6158 const char *new_xawtstr = "/libawt_xawt.so";
6159 char *p;
6160
6161 // Get path to libjvm.so
6162 os::jvm_path(buf, sizeof(buf));
6163
6164 // Get rid of libjvm.so
6165 p = strrchr(buf, '/');
6166 if (p == NULL) return false;
6167 else *p = '\0';
6168
6169 // Get rid of client or server
6170 p = strrchr(buf, '/');
6171 if (p == NULL) return false;
6172 else *p = '\0';
6173
6174 // check xawt/libmawt.so
6175 strcpy(libmawtpath, buf);
6176 strcat(libmawtpath, xawtstr);
6177 if (::stat(libmawtpath, &statbuf) == 0) return false;
6178
6179 // check libawt_xawt.so
6180 strcpy(libmawtpath, buf);
6181 strcat(libmawtpath, new_xawtstr);
6182 if (::stat(libmawtpath, &statbuf) == 0) return false;
6183
6184 return true;
6185 }
6186
6187 // Get the default path to the core file
6188 // Returns the length of the string
6189 int os::get_core_path(char* buffer, size_t bufferSize) {
6190 const char* p = get_current_directory(buffer, bufferSize);
6191
6192 if (p == NULL) {
6193 assert(p != NULL, "failed to get current directory");
6194 return 0;
6195 }
6196
6197 return strlen(buffer);
6198 }
6199
6200 /////////////// Unit tests ///////////////
6201
6202 #ifndef PRODUCT
6203
6204 #define test_log(...) \
6205 do {\
6206 if (VerboseInternalVMTests) { \
6207 tty->print_cr(__VA_ARGS__); \
6208 tty->flush(); \
6209 }\
6210 } while (false)
6211
6212 class TestReserveMemorySpecial : AllStatic {
6213 public:
6214 static void small_page_write(void* addr, size_t size) {
6215 size_t page_size = os::vm_page_size();
6216
6217 char* end = (char*)addr + size;
6218 for (char* p = (char*)addr; p < end; p += page_size) {
6219 *p = 1;
6220 }
6221 }
6222
6223 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6224 if (!UseHugeTLBFS) {
6225 return;
6226 }
6227
6228 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6229
6230 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6231
6232 if (addr != NULL) {
6233 small_page_write(addr, size);
6234
6235 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6236 }
6237 }
6238
6239 static void test_reserve_memory_special_huge_tlbfs_only() {
6240 if (!UseHugeTLBFS) {
6241 return;
6242 }
6243
6244 size_t lp = os::large_page_size();
6245
6246 for (size_t size = lp; size <= lp * 10; size += lp) {
6247 test_reserve_memory_special_huge_tlbfs_only(size);
6248 }
6249 }
6250
6251 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6252 size_t lp = os::large_page_size();
6253 size_t ag = os::vm_allocation_granularity();
6254
6255 // sizes to test
6256 const size_t sizes[] = {
6257 lp, lp + ag, lp + lp / 2, lp * 2,
6258 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6259 lp * 10, lp * 10 + lp / 2
6260 };
6261 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6262
6263 // For each size/alignment combination, we test three scenarios:
6264 // 1) with req_addr == NULL
6265 // 2) with a non-null req_addr at which we expect to successfully allocate
6266 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6267 // expect the allocation to either fail or to ignore req_addr
6268
6269 // Pre-allocate two areas; they shall be as large as the largest allocation
6270 // and aligned to the largest alignment we will be testing.
6271 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6272 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6273 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6274 -1, 0);
6275 assert(mapping1 != MAP_FAILED, "should work");
6276
6277 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6278 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6279 -1, 0);
6280 assert(mapping2 != MAP_FAILED, "should work");
6281
6282 // Unmap the first mapping, but leave the second mapping intact: the first
6283 // mapping will serve as a value for a "good" req_addr (case 2). The second
6284 // mapping, still intact, as "bad" req_addr (case 3).
6285 ::munmap(mapping1, mapping_size);
6286
6287 // Case 1
6288 test_log("%s, req_addr NULL:", __FUNCTION__);
6289 test_log("size align result");
6290
6291 for (int i = 0; i < num_sizes; i++) {
6292 const size_t size = sizes[i];
6293 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6294 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6295 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6296 size, alignment, p, (p != NULL ? "" : "(failed)"));
6297 if (p != NULL) {
6298 assert(is_ptr_aligned(p, alignment), "must be");
6299 small_page_write(p, size);
6300 os::Linux::release_memory_special_huge_tlbfs(p, size);
6301 }
6302 }
6303 }
6304
6305 // Case 2
6306 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6307 test_log("size align req_addr result");
6308
6309 for (int i = 0; i < num_sizes; i++) {
6310 const size_t size = sizes[i];
6311 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6312 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6313 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6314 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6315 size, alignment, req_addr, p,
6316 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6317 if (p != NULL) {
6318 assert(p == req_addr, "must be");
6319 small_page_write(p, size);
6320 os::Linux::release_memory_special_huge_tlbfs(p, size);
6321 }
6322 }
6323 }
6324
6325 // Case 3
6326 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6327 test_log("size align req_addr result");
6328
6329 for (int i = 0; i < num_sizes; i++) {
6330 const size_t size = sizes[i];
6331 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6332 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6333 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6334 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6335 size, alignment, req_addr, p,
6336 ((p != NULL ? "" : "(failed)")));
6337 // as the area around req_addr contains already existing mappings, the API should always
6338 // return NULL (as per contract, it cannot return another address)
6339 assert(p == NULL, "must be");
6340 }
6341 }
6342
6343 ::munmap(mapping2, mapping_size);
6344
6345 }
6346
6347 static void test_reserve_memory_special_huge_tlbfs() {
6348 if (!UseHugeTLBFS) {
6349 return;
6350 }
6351
6352 test_reserve_memory_special_huge_tlbfs_only();
6353 test_reserve_memory_special_huge_tlbfs_mixed();
6354 }
6355
6356 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6357 if (!UseSHM) {
6358 return;
6359 }
6360
6361 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6362
6363 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6364
6365 if (addr != NULL) {
6366 assert(is_ptr_aligned(addr, alignment), "Check");
6367 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6368
6369 small_page_write(addr, size);
6370
6371 os::Linux::release_memory_special_shm(addr, size);
6372 }
6373 }
6374
6375 static void test_reserve_memory_special_shm() {
6376 size_t lp = os::large_page_size();
6377 size_t ag = os::vm_allocation_granularity();
6378
6379 for (size_t size = ag; size < lp * 3; size += ag) {
6380 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6381 test_reserve_memory_special_shm(size, alignment);
6382 }
6383 }
6384 }
6385
6386 static void test() {
6387 test_reserve_memory_special_huge_tlbfs();
6388 test_reserve_memory_special_shm();
6389 }
6390 };
6391
6392 void TestReserveMemorySpecial_test() {
6393 TestReserveMemorySpecial::test();
6394 }
6395
6396 #endif