Thu, 24 May 2018 17:06:56 +0800
Merge
1 /*
2 * Copyright (c) 1999, 2015, 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 */
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"
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>
105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
107 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
108 // getrusage() is prepared to handle the associated failure.
109 #ifndef RUSAGE_THREAD
110 #define RUSAGE_THREAD (1) /* only the calling thread */
111 #endif
113 #define MAX_PATH (2 * K)
115 #define MAX_SECS 100000000
117 // for timer info max values which include all bits
118 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
120 #define LARGEPAGES_BIT (1 << 6)
121 ////////////////////////////////////////////////////////////////////////////////
122 // global variables
123 julong os::Linux::_physical_memory = 0;
125 address os::Linux::_initial_thread_stack_bottom = NULL;
126 uintptr_t os::Linux::_initial_thread_stack_size = 0;
128 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
129 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
130 Mutex* os::Linux::_createThread_lock = NULL;
131 pthread_t os::Linux::_main_thread;
132 int os::Linux::_page_size = -1;
133 const int os::Linux::_vm_default_page_size = (8 * K);
134 bool os::Linux::_is_floating_stack = false;
135 bool os::Linux::_is_NPTL = false;
136 bool os::Linux::_supports_fast_thread_cpu_time = false;
137 const char * os::Linux::_glibc_version = NULL;
138 const char * os::Linux::_libpthread_version = NULL;
139 pthread_condattr_t os::Linux::_condattr[1];
141 static jlong initial_time_count=0;
143 static int clock_tics_per_sec = 100;
145 // For diagnostics to print a message once. see run_periodic_checks
146 static sigset_t check_signal_done;
147 static bool check_signals = true;
149 static pid_t _initial_pid = 0;
151 /* Signal number used to suspend/resume a thread */
153 /* do not use any signal number less than SIGSEGV, see 4355769 */
154 static int SR_signum = SIGUSR2;
155 sigset_t SR_sigset;
157 /* Used to protect dlsym() calls */
158 static pthread_mutex_t dl_mutex;
160 // Declarations
161 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
163 // utility functions
165 static int SR_initialize();
167 julong os::available_memory() {
168 return Linux::available_memory();
169 }
171 julong os::Linux::available_memory() {
172 // values in struct sysinfo are "unsigned long"
173 struct sysinfo si;
174 sysinfo(&si);
176 return (julong)si.freeram * si.mem_unit;
177 }
179 julong os::physical_memory() {
180 return Linux::physical_memory();
181 }
183 ////////////////////////////////////////////////////////////////////////////////
184 // environment support
186 bool os::getenv(const char* name, char* buf, int len) {
187 const char* val = ::getenv(name);
188 if (val != NULL && strlen(val) < (size_t)len) {
189 strcpy(buf, val);
190 return true;
191 }
192 if (len > 0) buf[0] = 0; // return a null string
193 return false;
194 }
197 // Return true if user is running as root.
199 bool os::have_special_privileges() {
200 static bool init = false;
201 static bool privileges = false;
202 if (!init) {
203 privileges = (getuid() != geteuid()) || (getgid() != getegid());
204 init = true;
205 }
206 return privileges;
207 }
210 #ifndef SYS_gettid
211 // i386: 224, ia64: 1105, amd64: 186, sparc 143
212 #ifdef __ia64__
213 #define SYS_gettid 1105
214 #else
215 #ifdef __i386__
216 #define SYS_gettid 224
217 #else
218 #ifdef __amd64__
219 #define SYS_gettid 186
220 #else
221 #ifdef __sparc__
222 #define SYS_gettid 143
223 #else
224 #error define gettid for the arch
225 #endif
226 #endif
227 #endif
228 #endif
229 #endif
231 // Cpu architecture string
232 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
234 // pid_t gettid()
235 //
236 // Returns the kernel thread id of the currently running thread. Kernel
237 // thread id is used to access /proc.
238 //
239 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
240 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
241 //
242 pid_t os::Linux::gettid() {
243 int rslt = syscall(SYS_gettid);
244 if (rslt == -1) {
245 // old kernel, no NPTL support
246 return getpid();
247 } else {
248 return (pid_t)rslt;
249 }
250 }
252 // Most versions of linux have a bug where the number of processors are
253 // determined by looking at the /proc file system. In a chroot environment,
254 // the system call returns 1. This causes the VM to act as if it is
255 // a single processor and elide locking (see is_MP() call).
256 static bool unsafe_chroot_detected = false;
257 static const char *unstable_chroot_error = "/proc file system not found.\n"
258 "Java may be unstable running multithreaded in a chroot "
259 "environment on Linux when /proc filesystem is not mounted.";
261 void os::Linux::initialize_system_info() {
262 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
263 if (processor_count() == 1) {
264 pid_t pid = os::Linux::gettid();
265 char fname[32];
266 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
267 FILE *fp = fopen(fname, "r");
268 if (fp == NULL) {
269 unsafe_chroot_detected = true;
270 } else {
271 fclose(fp);
272 }
273 }
274 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
275 assert(processor_count() > 0, "linux error");
276 }
278 void os::init_system_properties_values() {
279 // The next steps are taken in the product version:
280 //
281 // Obtain the JAVA_HOME value from the location of libjvm.so.
282 // This library should be located at:
283 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
284 //
285 // If "/jre/lib/" appears at the right place in the path, then we
286 // assume libjvm.so is installed in a JDK and we use this path.
287 //
288 // Otherwise exit with message: "Could not create the Java virtual machine."
289 //
290 // The following extra steps are taken in the debugging version:
291 //
292 // If "/jre/lib/" does NOT appear at the right place in the path
293 // instead of exit check for $JAVA_HOME environment variable.
294 //
295 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
296 // then we append a fake suffix "hotspot/libjvm.so" to this path so
297 // it looks like libjvm.so is installed there
298 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
299 //
300 // Otherwise exit.
301 //
302 // Important note: if the location of libjvm.so changes this
303 // code needs to be changed accordingly.
305 // See ld(1):
306 // The linker uses the following search paths to locate required
307 // shared libraries:
308 // 1: ...
309 // ...
310 // 7: The default directories, normally /lib and /usr/lib.
311 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
312 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
313 #else
314 #define DEFAULT_LIBPATH "/lib:/usr/lib"
315 #endif
317 // Base path of extensions installed on the system.
318 #define SYS_EXT_DIR "/usr/java/packages"
319 #define EXTENSIONS_DIR "/lib/ext"
320 #define ENDORSED_DIR "/lib/endorsed"
322 // Buffer that fits several sprintfs.
323 // Note that the space for the colon and the trailing null are provided
324 // by the nulls included by the sizeof operator.
325 const size_t bufsize =
326 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
327 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
328 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
329 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
331 // sysclasspath, java_home, dll_dir
332 {
333 char *pslash;
334 os::jvm_path(buf, bufsize);
336 // Found the full path to libjvm.so.
337 // Now cut the path to <java_home>/jre if we can.
338 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
339 pslash = strrchr(buf, '/');
340 if (pslash != NULL) {
341 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
342 }
343 Arguments::set_dll_dir(buf);
345 if (pslash != NULL) {
346 pslash = strrchr(buf, '/');
347 if (pslash != NULL) {
348 *pslash = '\0'; // Get rid of /<arch>.
349 pslash = strrchr(buf, '/');
350 if (pslash != NULL) {
351 *pslash = '\0'; // Get rid of /lib.
352 }
353 }
354 }
355 Arguments::set_java_home(buf);
356 set_boot_path('/', ':');
357 }
359 // Where to look for native libraries.
360 //
361 // Note: Due to a legacy implementation, most of the library path
362 // is set in the launcher. This was to accomodate linking restrictions
363 // on legacy Linux implementations (which are no longer supported).
364 // Eventually, all the library path setting will be done here.
365 //
366 // However, to prevent the proliferation of improperly built native
367 // libraries, the new path component /usr/java/packages is added here.
368 // Eventually, all the library path setting will be done here.
369 {
370 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
371 // should always exist (until the legacy problem cited above is
372 // addressed).
373 const char *v = ::getenv("LD_LIBRARY_PATH");
374 const char *v_colon = ":";
375 if (v == NULL) { v = ""; v_colon = ""; }
376 // That's +1 for the colon and +1 for the trailing '\0'.
377 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
378 strlen(v) + 1 +
379 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
380 mtInternal);
381 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
382 Arguments::set_library_path(ld_library_path);
383 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
384 }
386 // Extensions directories.
387 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
388 Arguments::set_ext_dirs(buf);
390 // Endorsed standards default directory.
391 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
392 Arguments::set_endorsed_dirs(buf);
394 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
396 #undef DEFAULT_LIBPATH
397 #undef SYS_EXT_DIR
398 #undef EXTENSIONS_DIR
399 #undef ENDORSED_DIR
400 }
402 ////////////////////////////////////////////////////////////////////////////////
403 // breakpoint support
405 void os::breakpoint() {
406 BREAKPOINT;
407 }
409 extern "C" void breakpoint() {
410 // use debugger to set breakpoint here
411 }
413 ////////////////////////////////////////////////////////////////////////////////
414 // signal support
416 debug_only(static bool signal_sets_initialized = false);
417 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
419 bool os::Linux::is_sig_ignored(int sig) {
420 struct sigaction oact;
421 sigaction(sig, (struct sigaction*)NULL, &oact);
422 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
423 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
424 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
425 return true;
426 else
427 return false;
428 }
430 void os::Linux::signal_sets_init() {
431 // Should also have an assertion stating we are still single-threaded.
432 assert(!signal_sets_initialized, "Already initialized");
433 // Fill in signals that are necessarily unblocked for all threads in
434 // the VM. Currently, we unblock the following signals:
435 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
436 // by -Xrs (=ReduceSignalUsage));
437 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
438 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
439 // the dispositions or masks wrt these signals.
440 // Programs embedding the VM that want to use the above signals for their
441 // own purposes must, at this time, use the "-Xrs" option to prevent
442 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
443 // (See bug 4345157, and other related bugs).
444 // In reality, though, unblocking these signals is really a nop, since
445 // these signals are not blocked by default.
446 sigemptyset(&unblocked_sigs);
447 sigemptyset(&allowdebug_blocked_sigs);
448 sigaddset(&unblocked_sigs, SIGILL);
449 sigaddset(&unblocked_sigs, SIGSEGV);
450 sigaddset(&unblocked_sigs, SIGBUS);
451 sigaddset(&unblocked_sigs, SIGFPE);
452 #if defined(PPC64)
453 sigaddset(&unblocked_sigs, SIGTRAP);
454 #endif
455 sigaddset(&unblocked_sigs, SR_signum);
457 if (!ReduceSignalUsage) {
458 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
459 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
460 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
461 }
462 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
464 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
469 }
470 }
471 // Fill in signals that are blocked by all but the VM thread.
472 sigemptyset(&vm_sigs);
473 if (!ReduceSignalUsage)
474 sigaddset(&vm_sigs, BREAK_SIGNAL);
475 debug_only(signal_sets_initialized = true);
477 }
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
493 // These are signals that are blocked during cond_wait to allow debugger in
494 sigset_t* os::Linux::allowdebug_blocked_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &allowdebug_blocked_sigs;
497 }
499 void os::Linux::hotspot_sigmask(Thread* thread) {
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
526 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
527 // generic name for earlier versions.
528 // Define macros here so we can build HotSpot on old systems.
529 # ifndef _CS_GNU_LIBC_VERSION
530 # define _CS_GNU_LIBC_VERSION 2
531 # endif
532 # ifndef _CS_GNU_LIBPTHREAD_VERSION
533 # define _CS_GNU_LIBPTHREAD_VERSION 3
534 # endif
536 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
537 if (n > 0) {
538 char *str = (char *)malloc(n, mtInternal);
539 confstr(_CS_GNU_LIBC_VERSION, str, n);
540 os::Linux::set_glibc_version(str);
541 } else {
542 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
543 static char _gnu_libc_version[32];
544 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
545 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
546 os::Linux::set_glibc_version(_gnu_libc_version);
547 }
549 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
550 if (n > 0) {
551 char *str = (char *)malloc(n, mtInternal);
552 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
553 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
554 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
555 // is the case. LinuxThreads has a hard limit on max number of threads.
556 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
557 // On the other hand, NPTL does not have such a limit, sysconf()
558 // will return -1 and errno is not changed. Check if it is really NPTL.
559 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
560 strstr(str, "NPTL") &&
561 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
562 free(str);
563 os::Linux::set_libpthread_version("linuxthreads");
564 } else {
565 os::Linux::set_libpthread_version(str);
566 }
567 } else {
568 // glibc before 2.3.2 only has LinuxThreads.
569 os::Linux::set_libpthread_version("linuxthreads");
570 }
572 if (strstr(libpthread_version(), "NPTL")) {
573 os::Linux::set_is_NPTL();
574 } else {
575 os::Linux::set_is_LinuxThreads();
576 }
578 // LinuxThreads have two flavors: floating-stack mode, which allows variable
579 // stack size; and fixed-stack mode. NPTL is always floating-stack.
580 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
581 os::Linux::set_is_floating_stack();
582 }
583 }
585 /////////////////////////////////////////////////////////////////////////////
586 // thread stack
588 // Force Linux kernel to expand current thread stack. If "bottom" is close
589 // to the stack guard, caller should block all signals.
590 //
591 // MAP_GROWSDOWN:
592 // A special mmap() flag that is used to implement thread stacks. It tells
593 // kernel that the memory region should extend downwards when needed. This
594 // allows early versions of LinuxThreads to only mmap the first few pages
595 // when creating a new thread. Linux kernel will automatically expand thread
596 // stack as needed (on page faults).
597 //
598 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
599 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
600 // region, it's hard to tell if the fault is due to a legitimate stack
601 // access or because of reading/writing non-exist memory (e.g. buffer
602 // overrun). As a rule, if the fault happens below current stack pointer,
603 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
604 // application (see Linux kernel fault.c).
605 //
606 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
607 // stack overflow detection.
608 //
609 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
610 // not use this flag. However, the stack of initial thread is not created
611 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
612 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
613 // and then attach the thread to JVM.
614 //
615 // To get around the problem and allow stack banging on Linux, we need to
616 // manually expand thread stack after receiving the SIGSEGV.
617 //
618 // There are two ways to expand thread stack to address "bottom", we used
619 // both of them in JVM before 1.5:
620 // 1. adjust stack pointer first so that it is below "bottom", and then
621 // touch "bottom"
622 // 2. mmap() the page in question
623 //
624 // Now alternate signal stack is gone, it's harder to use 2. For instance,
625 // if current sp is already near the lower end of page 101, and we need to
626 // call mmap() to map page 100, it is possible that part of the mmap() frame
627 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
628 // That will destroy the mmap() frame and cause VM to crash.
629 //
630 // The following code works by adjusting sp first, then accessing the "bottom"
631 // page to force a page fault. Linux kernel will then automatically expand the
632 // stack mapping.
633 //
634 // _expand_stack_to() assumes its frame size is less than page size, which
635 // should always be true if the function is not inlined.
637 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
638 #define NOINLINE
639 #else
640 #define NOINLINE __attribute__ ((noinline))
641 #endif
643 static void _expand_stack_to(address bottom) NOINLINE;
645 static void _expand_stack_to(address bottom) {
646 address sp;
647 size_t size;
648 volatile char *p;
650 // Adjust bottom to point to the largest address within the same page, it
651 // gives us a one-page buffer if alloca() allocates slightly more memory.
652 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
653 bottom += os::Linux::page_size() - 1;
655 // sp might be slightly above current stack pointer; if that's the case, we
656 // will alloca() a little more space than necessary, which is OK. Don't use
657 // os::current_stack_pointer(), as its result can be slightly below current
658 // stack pointer, causing us to not alloca enough to reach "bottom".
659 sp = (address)&sp;
661 if (sp > bottom) {
662 size = sp - bottom;
663 p = (volatile char *)alloca(size);
664 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
665 p[0] = '\0';
666 }
667 }
669 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
670 assert(t!=NULL, "just checking");
671 assert(t->osthread()->expanding_stack(), "expand should be set");
672 assert(t->stack_base() != NULL, "stack_base was not initialized");
674 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
675 sigset_t mask_all, old_sigset;
676 sigfillset(&mask_all);
677 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
678 _expand_stack_to(addr);
679 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
680 return true;
681 }
682 return false;
683 }
685 //////////////////////////////////////////////////////////////////////////////
686 // create new thread
688 static address highest_vm_reserved_address();
690 // check if it's safe to start a new thread
691 static bool _thread_safety_check(Thread* thread) {
692 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
693 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
694 // Heap is mmap'ed at lower end of memory space. Thread stacks are
695 // allocated (MAP_FIXED) from high address space. Every thread stack
696 // occupies a fixed size slot (usually 2Mbytes, but user can change
697 // it to other values if they rebuild LinuxThreads).
698 //
699 // Problem with MAP_FIXED is that mmap() can still succeed even part of
700 // the memory region has already been mmap'ed. That means if we have too
701 // many threads and/or very large heap, eventually thread stack will
702 // collide with heap.
703 //
704 // Here we try to prevent heap/stack collision by comparing current
705 // stack bottom with the highest address that has been mmap'ed by JVM
706 // plus a safety margin for memory maps created by native code.
707 //
708 // This feature can be disabled by setting ThreadSafetyMargin to 0
709 //
710 if (ThreadSafetyMargin > 0) {
711 address stack_bottom = os::current_stack_base() - os::current_stack_size();
713 // not safe if our stack extends below the safety margin
714 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
715 } else {
716 return true;
717 }
718 } else {
719 // Floating stack LinuxThreads or NPTL:
720 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
721 // there's not enough space left, pthread_create() will fail. If we come
722 // here, that means enough space has been reserved for stack.
723 return true;
724 }
725 }
727 // Thread start routine for all newly created threads
728 static void *java_start(Thread *thread) {
729 // Try to randomize the cache line index of hot stack frames.
730 // This helps when threads of the same stack traces evict each other's
731 // cache lines. The threads can be either from the same JVM instance, or
732 // from different JVM instances. The benefit is especially true for
733 // processors with hyperthreading technology.
734 static int counter = 0;
735 int pid = os::current_process_id();
736 alloca(((pid ^ counter++) & 7) * 128);
738 ThreadLocalStorage::set_thread(thread);
740 OSThread* osthread = thread->osthread();
741 Monitor* sync = osthread->startThread_lock();
743 // non floating stack LinuxThreads needs extra check, see above
744 if (!_thread_safety_check(thread)) {
745 // notify parent thread
746 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
747 osthread->set_state(ZOMBIE);
748 sync->notify_all();
749 return NULL;
750 }
752 // thread_id is kernel thread id (similar to Solaris LWP id)
753 osthread->set_thread_id(os::Linux::gettid());
755 if (UseNUMA) {
756 int lgrp_id = os::numa_get_group_id();
757 if (lgrp_id != -1) {
758 thread->set_lgrp_id(lgrp_id);
759 }
760 }
761 // initialize signal mask for this thread
762 os::Linux::hotspot_sigmask(thread);
764 // initialize floating point control register
765 os::Linux::init_thread_fpu_state();
767 // handshaking with parent thread
768 {
769 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
771 // notify parent thread
772 osthread->set_state(INITIALIZED);
773 sync->notify_all();
775 // wait until os::start_thread()
776 while (osthread->get_state() == INITIALIZED) {
777 sync->wait(Mutex::_no_safepoint_check_flag);
778 }
779 }
781 // call one more level start routine
782 thread->run();
784 return 0;
785 }
787 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
788 assert(thread->osthread() == NULL, "caller responsible");
790 // Allocate the OSThread object
791 OSThread* osthread = new OSThread(NULL, NULL);
792 if (osthread == NULL) {
793 return false;
794 }
796 // set the correct thread state
797 osthread->set_thread_type(thr_type);
799 // Initial state is ALLOCATED but not INITIALIZED
800 osthread->set_state(ALLOCATED);
802 thread->set_osthread(osthread);
804 // init thread attributes
805 pthread_attr_t attr;
806 pthread_attr_init(&attr);
807 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
809 // stack size
810 if (os::Linux::supports_variable_stack_size()) {
811 // calculate stack size if it's not specified by caller
812 if (stack_size == 0) {
813 stack_size = os::Linux::default_stack_size(thr_type);
815 switch (thr_type) {
816 case os::java_thread:
817 // Java threads use ThreadStackSize which default value can be
818 // changed with the flag -Xss
819 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
820 stack_size = JavaThread::stack_size_at_create();
821 break;
822 case os::compiler_thread:
823 if (CompilerThreadStackSize > 0) {
824 stack_size = (size_t)(CompilerThreadStackSize * K);
825 break;
826 } // else fall through:
827 // use VMThreadStackSize if CompilerThreadStackSize is not defined
828 case os::vm_thread:
829 case os::pgc_thread:
830 case os::cgc_thread:
831 case os::watcher_thread:
832 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
833 break;
834 }
835 }
837 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
838 pthread_attr_setstacksize(&attr, stack_size);
839 } else {
840 // let pthread_create() pick the default value.
841 }
843 // glibc guard page
844 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
846 ThreadState state;
848 {
849 // Serialize thread creation if we are running with fixed stack LinuxThreads
850 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
851 if (lock) {
852 os::Linux::createThread_lock()->lock_without_safepoint_check();
853 }
855 pthread_t tid;
856 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
858 pthread_attr_destroy(&attr);
860 if (ret != 0) {
861 if (PrintMiscellaneous && (Verbose || WizardMode)) {
862 perror("pthread_create()");
863 }
864 // Need to clean up stuff we've allocated so far
865 thread->set_osthread(NULL);
866 delete osthread;
867 if (lock) os::Linux::createThread_lock()->unlock();
868 return false;
869 }
871 // Store pthread info into the OSThread
872 osthread->set_pthread_id(tid);
874 // Wait until child thread is either initialized or aborted
875 {
876 Monitor* sync_with_child = osthread->startThread_lock();
877 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
878 while ((state = osthread->get_state()) == ALLOCATED) {
879 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
880 }
881 }
883 if (lock) {
884 os::Linux::createThread_lock()->unlock();
885 }
886 }
888 // Aborted due to thread limit being reached
889 if (state == ZOMBIE) {
890 thread->set_osthread(NULL);
891 delete osthread;
892 return false;
893 }
895 // The thread is returned suspended (in state INITIALIZED),
896 // and is started higher up in the call chain
897 assert(state == INITIALIZED, "race condition");
898 return true;
899 }
901 /////////////////////////////////////////////////////////////////////////////
902 // attach existing thread
904 // bootstrap the main thread
905 bool os::create_main_thread(JavaThread* thread) {
906 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
907 return create_attached_thread(thread);
908 }
910 bool os::create_attached_thread(JavaThread* thread) {
911 #ifdef ASSERT
912 thread->verify_not_published();
913 #endif
915 // Allocate the OSThread object
916 OSThread* osthread = new OSThread(NULL, NULL);
918 if (osthread == NULL) {
919 return false;
920 }
922 // Store pthread info into the OSThread
923 osthread->set_thread_id(os::Linux::gettid());
924 osthread->set_pthread_id(::pthread_self());
926 // initialize floating point control register
927 os::Linux::init_thread_fpu_state();
929 // Initial thread state is RUNNABLE
930 osthread->set_state(RUNNABLE);
932 thread->set_osthread(osthread);
934 if (UseNUMA) {
935 int lgrp_id = os::numa_get_group_id();
936 if (lgrp_id != -1) {
937 thread->set_lgrp_id(lgrp_id);
938 }
939 }
941 if (os::Linux::is_initial_thread()) {
942 // If current thread is initial thread, its stack is mapped on demand,
943 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
944 // the entire stack region to avoid SEGV in stack banging.
945 // It is also useful to get around the heap-stack-gap problem on SuSE
946 // kernel (see 4821821 for details). We first expand stack to the top
947 // of yellow zone, then enable stack yellow zone (order is significant,
948 // enabling yellow zone first will crash JVM on SuSE Linux), so there
949 // is no gap between the last two virtual memory regions.
951 JavaThread *jt = (JavaThread *)thread;
952 address addr = jt->stack_yellow_zone_base();
953 assert(addr != NULL, "initialization problem?");
954 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
956 osthread->set_expanding_stack();
957 os::Linux::manually_expand_stack(jt, addr);
958 osthread->clear_expanding_stack();
959 }
961 // initialize signal mask for this thread
962 // and save the caller's signal mask
963 os::Linux::hotspot_sigmask(thread);
965 return true;
966 }
968 void os::pd_start_thread(Thread* thread) {
969 OSThread * osthread = thread->osthread();
970 assert(osthread->get_state() != INITIALIZED, "just checking");
971 Monitor* sync_with_child = osthread->startThread_lock();
972 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
973 sync_with_child->notify();
975 #ifdef MIPS64
976 /* 2013/11/5 Jin: To be accessed in NativeGeneralJump::patch_verified_entry() */
977 if (thread->is_Java_thread())
978 {
979 ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
980 }
981 #endif
982 }
984 // Free Linux resources related to the OSThread
985 void os::free_thread(OSThread* osthread) {
986 assert(osthread != NULL, "osthread not set");
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 }
994 delete osthread;
995 }
997 //////////////////////////////////////////////////////////////////////////////
998 // thread local storage
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 }
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 }
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 }
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 }
1030 extern "C" Thread* get_thread() {
1031 return ThreadLocalStorage::thread();
1032 }
1034 //////////////////////////////////////////////////////////////////////////////
1035 // initial thread
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 }
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 }
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;
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();
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.
1119 uintptr_t stack_start;
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;
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;
1162 fp = fopen("/proc/self/stat", "r");
1163 if (fp) {
1164 statlen = fread(stat, 1, 2047, fp);
1165 stat[statlen] = '\0';
1166 fclose(fp);
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, ')');
1176 i = 0;
1177 if (s) {
1178 // Skip blank chars
1179 do s++; while (isspace(*s));
1181 #define _UFM UINTX_FORMAT
1182 #define _DFM INTX_FORMAT
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 }
1215 #undef _UFM
1216 #undef _DFM
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 }
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.)
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 }
1257 // stack_top could be partially down the page so align it
1258 stack_top = align_size_up(stack_top, page_size());
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 }
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 }
1270 ////////////////////////////////////////////////////////////////////////////////
1271 // time support
1273 // Time since start-up in seconds to a fine granularity.
1274 // Used by VMSelfDestructTimer and the MemProfiler.
1275 double os::elapsedTime() {
1277 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1278 }
1280 jlong os::elapsed_counter() {
1281 return javaTimeNanos() - initial_time_count;
1282 }
1284 jlong os::elapsed_frequency() {
1285 return NANOSECS_PER_SEC; // nanosecond resolution
1286 }
1288 bool os::supports_vtime() { return true; }
1289 bool os::enable_vtime() { return false; }
1290 bool os::vtime_enabled() { return false; }
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 }
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 }
1310 #ifndef CLOCK_MONOTONIC
1311 #define CLOCK_MONOTONIC (1)
1312 #endif
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 }
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 }
1353 #ifndef SYS_clock_getres
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
1363 #else
1364 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1365 #endif
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");
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.
1385 if(pthread_getcpuclockid_func &&
1386 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1387 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1389 _supports_fast_thread_cpu_time = true;
1390 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1391 }
1392 }
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 }
1410 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1411 if (Linux::supports_monotonic_clock()) {
1412 info_ptr->max_value = ALL_64_BITS;
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;
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 }
1426 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1427 }
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);
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;
1445 return true;
1446 }
1447 }
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 }
1461 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1462 return localtime_r(clock, res);
1463 }
1465 ////////////////////////////////////////////////////////////////////////////////
1466 // runtime exit support
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() {
1473 // allow PerfMemory to attempt cleanup of any persistent resources
1474 perfMemory_exit();
1476 // needs to remove object in file system
1477 AttachListener::abort();
1479 // flush buffered output, finish log files
1480 ostream_abort();
1482 // Check for abort hook
1483 abort_hook_t abort_hook = Arguments::abort_hook();
1484 if (abort_hook != NULL) {
1485 abort_hook();
1486 }
1488 }
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 }
1507 ::exit(1);
1508 }
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 }
1517 // This method is a copy of JDK's sysGetLastErrorString
1518 // from src/solaris/hpi/src/system_md.c
1520 size_t os::lasterror(char *buf, size_t len) {
1522 if (errno == 0) return 0;
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 }
1534 intx os::current_thread_id() { return (intx)pthread_self(); }
1535 int os::current_process_id() {
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'.
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.
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
1553 return (int)(_initial_pid ? _initial_pid : getpid());
1554 }
1556 // DLL functions
1558 const char* os::dll_file_extension() { return ".so"; }
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"; }
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 }
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;
1578 // Return error on buffer overflow.
1579 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1580 return retval;
1581 }
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 }
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;
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 }
1631 if (dladdr((void *)addr, &dlinfo) != 0) {
1632 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1633 }
1635 return false;
1636 }
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");
1643 Dl_info dlinfo;
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 }
1663 buf[0] = '\0';
1664 if (offset != NULL) *offset = -1;
1665 return false;
1666 }
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 };
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;
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 }
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 }
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");
1717 Dl_info dlinfo;
1718 struct _address_to_library_name data;
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);
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 }
1746 buf[0] = '\0';
1747 if (offset) *offset = -1;
1748 return false;
1749 }
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
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;
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 };
1781 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1782 {
1783 void * result = NULL;
1784 bool load_attempted = false;
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);
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 }
1826 ThreadInVMfromNative tiv(jt);
1827 debug_only(VMNativeEntryWrapper vew;)
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 }
1840 if (!load_attempted) {
1841 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1842 }
1844 if (result != NULL) {
1845 // Successful loading
1846 return result;
1847 }
1849 Elf32_Ehdr elf_head;
1850 int diag_msg_max_length=ebuflen-strlen(ebuf);
1851 char* diag_msg_buf=ebuf+strlen(ebuf);
1853 if (diag_msg_max_length==0) {
1854 // No more space in ebuf for additional diagnostics message
1855 return NULL;
1856 }
1859 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1861 if (file_descriptor < 0) {
1862 // Can't open library, report dlerror() message
1863 return NULL;
1864 }
1866 bool failed_to_read_elf_head=
1867 (sizeof(elf_head)!=
1868 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1870 ::close(file_descriptor);
1871 if (failed_to_read_elf_head) {
1872 // file i/o error - report dlerror() msg
1873 return NULL;
1874 }
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;
1884 #ifndef EM_486
1885 #define EM_486 6 /* Intel 80486 */
1886 #endif
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_MIPS, EM_MIPS, ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
1908 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1909 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1910 };
1912 #if (defined IA32)
1913 static Elf32_Half running_arch_code=EM_386;
1914 #elif (defined AMD64)
1915 static Elf32_Half running_arch_code=EM_X86_64;
1916 #elif (defined IA64)
1917 static Elf32_Half running_arch_code=EM_IA_64;
1918 #elif (defined __sparc) && (defined _LP64)
1919 static Elf32_Half running_arch_code=EM_SPARCV9;
1920 #elif (defined __sparc) && (!defined _LP64)
1921 static Elf32_Half running_arch_code=EM_SPARC;
1922 #elif (defined MIPS64)
1923 static Elf32_Half running_arch_code=EM_MIPS;
1924 #elif (defined __powerpc64__)
1925 static Elf32_Half running_arch_code=EM_PPC64;
1926 #elif (defined __powerpc__)
1927 static Elf32_Half running_arch_code=EM_PPC;
1928 #elif (defined ARM)
1929 static Elf32_Half running_arch_code=EM_ARM;
1930 #elif (defined S390)
1931 static Elf32_Half running_arch_code=EM_S390;
1932 #elif (defined ALPHA)
1933 static Elf32_Half running_arch_code=EM_ALPHA;
1934 #elif (defined MIPSEL)
1935 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1936 #elif (defined PARISC)
1937 static Elf32_Half running_arch_code=EM_PARISC;
1938 #elif (defined MIPS)
1939 static Elf32_Half running_arch_code=EM_MIPS;
1940 #elif (defined M68K)
1941 static Elf32_Half running_arch_code=EM_68K;
1942 #else
1943 #error Method os::dll_load requires that one of following is defined:\
1944 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, __mips64, PARISC, M68K
1945 #endif
1947 // Identify compatability class for VM's architecture and library's architecture
1948 // Obtain string descriptions for architectures
1950 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1951 int running_arch_index=-1;
1953 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1954 if (running_arch_code == arch_array[i].code) {
1955 running_arch_index = i;
1956 }
1957 if (lib_arch.code == arch_array[i].code) {
1958 lib_arch.compat_class = arch_array[i].compat_class;
1959 lib_arch.name = arch_array[i].name;
1960 }
1961 }
1963 assert(running_arch_index != -1,
1964 "Didn't find running architecture code (running_arch_code) in arch_array");
1965 if (running_arch_index == -1) {
1966 // Even though running architecture detection failed
1967 // we may still continue with reporting dlerror() message
1968 return NULL;
1969 }
1971 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1972 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1973 return NULL;
1974 }
1976 #ifndef S390
1977 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1978 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1979 return NULL;
1980 }
1981 #endif // !S390
1983 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1984 if ( lib_arch.name!=NULL ) {
1985 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1986 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1987 lib_arch.name, arch_array[running_arch_index].name);
1988 } else {
1989 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1990 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1991 lib_arch.code,
1992 arch_array[running_arch_index].name);
1993 }
1994 }
1996 return NULL;
1997 }
1999 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2000 void * result = ::dlopen(filename, RTLD_LAZY);
2001 if (result == NULL) {
2002 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2003 ebuf[ebuflen-1] = '\0';
2004 }
2005 return result;
2006 }
2008 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2009 void * result = NULL;
2010 if (LoadExecStackDllInVMThread) {
2011 result = dlopen_helper(filename, ebuf, ebuflen);
2012 }
2014 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2015 // library that requires an executable stack, or which does not have this
2016 // stack attribute set, dlopen changes the stack attribute to executable. The
2017 // read protection of the guard pages gets lost.
2018 //
2019 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2020 // may have been queued at the same time.
2022 if (!_stack_is_executable) {
2023 JavaThread *jt = Threads::first();
2025 while (jt) {
2026 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2027 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2028 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2029 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2030 warning("Attempt to reguard stack yellow zone failed.");
2031 }
2032 }
2033 jt = jt->next();
2034 }
2035 }
2037 return result;
2038 }
2040 /*
2041 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2042 * chances are you might want to run the generated bits against glibc-2.0
2043 * libdl.so, so always use locking for any version of glibc.
2044 */
2045 void* os::dll_lookup(void* handle, const char* name) {
2046 pthread_mutex_lock(&dl_mutex);
2047 void* res = dlsym(handle, name);
2048 pthread_mutex_unlock(&dl_mutex);
2049 return res;
2050 }
2052 void* os::get_default_process_handle() {
2053 return (void*)::dlopen(NULL, RTLD_LAZY);
2054 }
2056 static bool _print_ascii_file(const char* filename, outputStream* st) {
2057 int fd = ::open(filename, O_RDONLY);
2058 if (fd == -1) {
2059 return false;
2060 }
2062 char buf[32];
2063 int bytes;
2064 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2065 st->print_raw(buf, bytes);
2066 }
2068 ::close(fd);
2070 return true;
2071 }
2073 void os::print_dll_info(outputStream *st) {
2074 st->print_cr("Dynamic libraries:");
2076 char fname[32];
2077 pid_t pid = os::Linux::gettid();
2079 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2081 if (!_print_ascii_file(fname, st)) {
2082 st->print("Can not get library information for pid = %d\n", pid);
2083 }
2084 }
2086 void os::print_os_info_brief(outputStream* st) {
2087 os::Linux::print_distro_info(st);
2089 os::Posix::print_uname_info(st);
2091 os::Linux::print_libversion_info(st);
2093 }
2095 void os::print_os_info(outputStream* st) {
2096 st->print("OS:");
2098 os::Linux::print_distro_info(st);
2100 os::Posix::print_uname_info(st);
2102 // Print warning if unsafe chroot environment detected
2103 if (unsafe_chroot_detected) {
2104 st->print("WARNING!! ");
2105 st->print_cr("%s", unstable_chroot_error);
2106 }
2108 os::Linux::print_libversion_info(st);
2110 os::Posix::print_rlimit_info(st);
2112 os::Posix::print_load_average(st);
2114 os::Linux::print_full_memory_info(st);
2115 }
2117 // Try to identify popular distros.
2118 // Most Linux distributions have a /etc/XXX-release file, which contains
2119 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2120 // file that also contains the OS version string. Some have more than one
2121 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2122 // /etc/redhat-release.), so the order is important.
2123 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2124 // their own specific XXX-release file as well as a redhat-release file.
2125 // Because of this the XXX-release file needs to be searched for before the
2126 // redhat-release file.
2127 // Since Red Hat has a lsb-release file that is not very descriptive the
2128 // search for redhat-release needs to be before lsb-release.
2129 // Since the lsb-release file is the new standard it needs to be searched
2130 // before the older style release files.
2131 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2132 // next to last resort. The os-release file is a new standard that contains
2133 // distribution information and the system-release file seems to be an old
2134 // standard that has been replaced by the lsb-release and os-release files.
2135 // Searching for the debian_version file is the last resort. It contains
2136 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2137 // "Debian " is printed before the contents of the debian_version file.
2138 void os::Linux::print_distro_info(outputStream* st) {
2139 if (!_print_ascii_file("/etc/oracle-release", st) &&
2140 !_print_ascii_file("/etc/mandriva-release", st) &&
2141 !_print_ascii_file("/etc/mandrake-release", st) &&
2142 !_print_ascii_file("/etc/sun-release", st) &&
2143 !_print_ascii_file("/etc/redhat-release", st) &&
2144 !_print_ascii_file("/etc/lsb-release", st) &&
2145 !_print_ascii_file("/etc/SuSE-release", st) &&
2146 !_print_ascii_file("/etc/turbolinux-release", st) &&
2147 !_print_ascii_file("/etc/gentoo-release", st) &&
2148 !_print_ascii_file("/etc/ltib-release", st) &&
2149 !_print_ascii_file("/etc/angstrom-version", st) &&
2150 !_print_ascii_file("/etc/system-release", st) &&
2151 !_print_ascii_file("/etc/os-release", st)) {
2153 if (file_exists("/etc/debian_version")) {
2154 st->print("Debian ");
2155 _print_ascii_file("/etc/debian_version", st);
2156 } else {
2157 st->print("Linux");
2158 }
2159 }
2160 st->cr();
2161 }
2163 void os::Linux::print_libversion_info(outputStream* st) {
2164 // libc, pthread
2165 st->print("libc:");
2166 st->print("%s ", os::Linux::glibc_version());
2167 st->print("%s ", os::Linux::libpthread_version());
2168 if (os::Linux::is_LinuxThreads()) {
2169 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2170 }
2171 st->cr();
2172 }
2174 void os::Linux::print_full_memory_info(outputStream* st) {
2175 st->print("\n/proc/meminfo:\n");
2176 _print_ascii_file("/proc/meminfo", st);
2177 st->cr();
2178 }
2180 void os::print_memory_info(outputStream* st) {
2182 st->print("Memory:");
2183 st->print(" %dk page", os::vm_page_size()>>10);
2185 // values in struct sysinfo are "unsigned long"
2186 struct sysinfo si;
2187 sysinfo(&si);
2189 st->print(", physical " UINT64_FORMAT "k",
2190 os::physical_memory() >> 10);
2191 st->print("(" UINT64_FORMAT "k free)",
2192 os::available_memory() >> 10);
2193 st->print(", swap " UINT64_FORMAT "k",
2194 ((jlong)si.totalswap * si.mem_unit) >> 10);
2195 st->print("(" UINT64_FORMAT "k free)",
2196 ((jlong)si.freeswap * si.mem_unit) >> 10);
2197 st->cr();
2198 }
2200 void os::pd_print_cpu_info(outputStream* st) {
2201 st->print("\n/proc/cpuinfo:\n");
2202 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2203 st->print(" <Not Available>");
2204 }
2205 st->cr();
2206 }
2208 void os::print_siginfo(outputStream* st, void* siginfo) {
2209 const siginfo_t* si = (const siginfo_t*)siginfo;
2211 os::Posix::print_siginfo_brief(st, si);
2212 #if INCLUDE_CDS
2213 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2214 UseSharedSpaces) {
2215 FileMapInfo* mapinfo = FileMapInfo::current_info();
2216 if (mapinfo->is_in_shared_space(si->si_addr)) {
2217 st->print("\n\nError accessing class data sharing archive." \
2218 " Mapped file inaccessible during execution, " \
2219 " possible disk/network problem.");
2220 }
2221 }
2222 #endif
2223 st->cr();
2224 }
2227 static void print_signal_handler(outputStream* st, int sig,
2228 char* buf, size_t buflen);
2230 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2231 st->print_cr("Signal Handlers:");
2232 print_signal_handler(st, SIGSEGV, buf, buflen);
2233 print_signal_handler(st, SIGBUS , buf, buflen);
2234 print_signal_handler(st, SIGFPE , buf, buflen);
2235 print_signal_handler(st, SIGPIPE, buf, buflen);
2236 print_signal_handler(st, SIGXFSZ, buf, buflen);
2237 print_signal_handler(st, SIGILL , buf, buflen);
2238 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2239 print_signal_handler(st, SR_signum, buf, buflen);
2240 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2241 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2242 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2243 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2244 #if defined(PPC64)
2245 print_signal_handler(st, SIGTRAP, buf, buflen);
2246 #endif
2247 }
2249 static char saved_jvm_path[MAXPATHLEN] = {0};
2251 // Find the full path to the current module, libjvm.so
2252 void os::jvm_path(char *buf, jint buflen) {
2253 // Error checking.
2254 if (buflen < MAXPATHLEN) {
2255 assert(false, "must use a large-enough buffer");
2256 buf[0] = '\0';
2257 return;
2258 }
2259 // Lazy resolve the path to current module.
2260 if (saved_jvm_path[0] != 0) {
2261 strcpy(buf, saved_jvm_path);
2262 return;
2263 }
2265 char dli_fname[MAXPATHLEN];
2266 bool ret = dll_address_to_library_name(
2267 CAST_FROM_FN_PTR(address, os::jvm_path),
2268 dli_fname, sizeof(dli_fname), NULL);
2269 assert(ret, "cannot locate libjvm");
2270 char *rp = NULL;
2271 if (ret && dli_fname[0] != '\0') {
2272 rp = realpath(dli_fname, buf);
2273 }
2274 if (rp == NULL)
2275 return;
2277 if (Arguments::created_by_gamma_launcher()) {
2278 // Support for the gamma launcher. Typical value for buf is
2279 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2280 // the right place in the string, then assume we are installed in a JDK and
2281 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2282 // up the path so it looks like libjvm.so is installed there (append a
2283 // fake suffix hotspot/libjvm.so).
2284 const char *p = buf + strlen(buf) - 1;
2285 for (int count = 0; p > buf && count < 5; ++count) {
2286 for (--p; p > buf && *p != '/'; --p)
2287 /* empty */ ;
2288 }
2290 if (strncmp(p, "/jre/lib/", 9) != 0) {
2291 // Look for JAVA_HOME in the environment.
2292 char* java_home_var = ::getenv("JAVA_HOME");
2293 if (java_home_var != NULL && java_home_var[0] != 0) {
2294 char* jrelib_p;
2295 int len;
2297 // Check the current module name "libjvm.so".
2298 p = strrchr(buf, '/');
2299 assert(strstr(p, "/libjvm") == p, "invalid library name");
2301 rp = realpath(java_home_var, buf);
2302 if (rp == NULL)
2303 return;
2305 // determine if this is a legacy image or modules image
2306 // modules image doesn't have "jre" subdirectory
2307 len = strlen(buf);
2308 assert(len < buflen, "Ran out of buffer room");
2309 jrelib_p = buf + len;
2310 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2311 if (0 != access(buf, F_OK)) {
2312 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2313 }
2315 if (0 == access(buf, F_OK)) {
2316 // Use current module name "libjvm.so"
2317 len = strlen(buf);
2318 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2319 } else {
2320 // Go back to path of .so
2321 rp = realpath(dli_fname, buf);
2322 if (rp == NULL)
2323 return;
2324 }
2325 }
2326 }
2327 }
2329 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2330 }
2332 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2333 // no prefix required, not even "_"
2334 }
2336 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2337 // no suffix required
2338 }
2340 ////////////////////////////////////////////////////////////////////////////////
2341 // sun.misc.Signal support
2343 static volatile jint sigint_count = 0;
2345 static void
2346 UserHandler(int sig, void *siginfo, void *context) {
2347 // 4511530 - sem_post is serialized and handled by the manager thread. When
2348 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2349 // don't want to flood the manager thread with sem_post requests.
2350 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2351 return;
2353 // Ctrl-C is pressed during error reporting, likely because the error
2354 // handler fails to abort. Let VM die immediately.
2355 if (sig == SIGINT && is_error_reported()) {
2356 os::die();
2357 }
2359 os::signal_notify(sig);
2360 }
2362 void* os::user_handler() {
2363 return CAST_FROM_FN_PTR(void*, UserHandler);
2364 }
2366 class Semaphore : public StackObj {
2367 public:
2368 Semaphore();
2369 ~Semaphore();
2370 void signal();
2371 void wait();
2372 bool trywait();
2373 bool timedwait(unsigned int sec, int nsec);
2374 private:
2375 sem_t _semaphore;
2376 };
2378 Semaphore::Semaphore() {
2379 sem_init(&_semaphore, 0, 0);
2380 }
2382 Semaphore::~Semaphore() {
2383 sem_destroy(&_semaphore);
2384 }
2386 void Semaphore::signal() {
2387 sem_post(&_semaphore);
2388 }
2390 void Semaphore::wait() {
2391 sem_wait(&_semaphore);
2392 }
2394 bool Semaphore::trywait() {
2395 return sem_trywait(&_semaphore) == 0;
2396 }
2398 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2400 struct timespec ts;
2401 // Semaphore's are always associated with CLOCK_REALTIME
2402 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2403 // see unpackTime for discussion on overflow checking
2404 if (sec >= MAX_SECS) {
2405 ts.tv_sec += MAX_SECS;
2406 ts.tv_nsec = 0;
2407 } else {
2408 ts.tv_sec += sec;
2409 ts.tv_nsec += nsec;
2410 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2411 ts.tv_nsec -= NANOSECS_PER_SEC;
2412 ++ts.tv_sec; // note: this must be <= max_secs
2413 }
2414 }
2416 while (1) {
2417 int result = sem_timedwait(&_semaphore, &ts);
2418 if (result == 0) {
2419 return true;
2420 } else if (errno == EINTR) {
2421 continue;
2422 } else if (errno == ETIMEDOUT) {
2423 return false;
2424 } else {
2425 return false;
2426 }
2427 }
2428 }
2430 extern "C" {
2431 typedef void (*sa_handler_t)(int);
2432 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2433 }
2435 void* os::signal(int signal_number, void* handler) {
2436 struct sigaction sigAct, oldSigAct;
2438 sigfillset(&(sigAct.sa_mask));
2439 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2440 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2442 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2443 // -1 means registration failed
2444 return (void *)-1;
2445 }
2447 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2448 }
2450 void os::signal_raise(int signal_number) {
2451 ::raise(signal_number);
2452 }
2454 /*
2455 * The following code is moved from os.cpp for making this
2456 * code platform specific, which it is by its very nature.
2457 */
2459 // Will be modified when max signal is changed to be dynamic
2460 int os::sigexitnum_pd() {
2461 return NSIG;
2462 }
2464 // a counter for each possible signal value
2465 static volatile jint pending_signals[NSIG+1] = { 0 };
2467 // Linux(POSIX) specific hand shaking semaphore.
2468 static sem_t sig_sem;
2469 static Semaphore sr_semaphore;
2471 void os::signal_init_pd() {
2472 // Initialize signal structures
2473 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2475 // Initialize signal semaphore
2476 ::sem_init(&sig_sem, 0, 0);
2477 }
2479 void os::signal_notify(int sig) {
2480 Atomic::inc(&pending_signals[sig]);
2481 ::sem_post(&sig_sem);
2482 }
2484 static int check_pending_signals(bool wait) {
2485 Atomic::store(0, &sigint_count);
2486 for (;;) {
2487 for (int i = 0; i < NSIG + 1; i++) {
2488 jint n = pending_signals[i];
2489 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2490 return i;
2491 }
2492 }
2493 if (!wait) {
2494 return -1;
2495 }
2496 JavaThread *thread = JavaThread::current();
2497 ThreadBlockInVM tbivm(thread);
2499 bool threadIsSuspended;
2500 do {
2501 thread->set_suspend_equivalent();
2502 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2503 ::sem_wait(&sig_sem);
2505 // were we externally suspended while we were waiting?
2506 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2507 if (threadIsSuspended) {
2508 //
2509 // The semaphore has been incremented, but while we were waiting
2510 // another thread suspended us. We don't want to continue running
2511 // while suspended because that would surprise the thread that
2512 // suspended us.
2513 //
2514 ::sem_post(&sig_sem);
2516 thread->java_suspend_self();
2517 }
2518 } while (threadIsSuspended);
2519 }
2520 }
2522 int os::signal_lookup() {
2523 return check_pending_signals(false);
2524 }
2526 int os::signal_wait() {
2527 return check_pending_signals(true);
2528 }
2530 ////////////////////////////////////////////////////////////////////////////////
2531 // Virtual Memory
2533 int os::vm_page_size() {
2534 // Seems redundant as all get out
2535 assert(os::Linux::page_size() != -1, "must call os::init");
2536 return os::Linux::page_size();
2537 }
2539 // Solaris allocates memory by pages.
2540 int os::vm_allocation_granularity() {
2541 assert(os::Linux::page_size() != -1, "must call os::init");
2542 return os::Linux::page_size();
2543 }
2545 // Rationale behind this function:
2546 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2547 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2548 // samples for JITted code. Here we create private executable mapping over the code cache
2549 // and then we can use standard (well, almost, as mapping can change) way to provide
2550 // info for the reporting script by storing timestamp and location of symbol
2551 void linux_wrap_code(char* base, size_t size) {
2552 static volatile jint cnt = 0;
2554 if (!UseOprofile) {
2555 return;
2556 }
2558 char buf[PATH_MAX+1];
2559 int num = Atomic::add(1, &cnt);
2561 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2562 os::get_temp_directory(), os::current_process_id(), num);
2563 unlink(buf);
2565 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2567 if (fd != -1) {
2568 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2569 if (rv != (off_t)-1) {
2570 if (::write(fd, "", 1) == 1) {
2571 mmap(base, size,
2572 PROT_READ|PROT_WRITE|PROT_EXEC,
2573 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2574 }
2575 }
2576 ::close(fd);
2577 unlink(buf);
2578 }
2579 }
2581 static bool recoverable_mmap_error(int err) {
2582 // See if the error is one we can let the caller handle. This
2583 // list of errno values comes from JBS-6843484. I can't find a
2584 // Linux man page that documents this specific set of errno
2585 // values so while this list currently matches Solaris, it may
2586 // change as we gain experience with this failure mode.
2587 switch (err) {
2588 case EBADF:
2589 case EINVAL:
2590 case ENOTSUP:
2591 // let the caller deal with these errors
2592 return true;
2594 default:
2595 // Any remaining errors on this OS can cause our reserved mapping
2596 // to be lost. That can cause confusion where different data
2597 // structures think they have the same memory mapped. The worst
2598 // scenario is if both the VM and a library think they have the
2599 // same memory mapped.
2600 return false;
2601 }
2602 }
2604 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2605 int err) {
2606 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2607 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2608 strerror(err), err);
2609 }
2611 static void warn_fail_commit_memory(char* addr, size_t size,
2612 size_t alignment_hint, bool exec,
2613 int err) {
2614 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2616 alignment_hint, exec, strerror(err), err);
2617 }
2619 // NOTE: Linux kernel does not really reserve the pages for us.
2620 // All it does is to check if there are enough free pages
2621 // left at the time of mmap(). This could be a potential
2622 // problem.
2623 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2624 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2625 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2626 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2627 if (res != (uintptr_t) MAP_FAILED) {
2628 if (UseNUMAInterleaving) {
2629 numa_make_global(addr, size);
2630 }
2631 return 0;
2632 }
2634 int err = errno; // save errno from mmap() call above
2636 if (!recoverable_mmap_error(err)) {
2637 warn_fail_commit_memory(addr, size, exec, err);
2638 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2639 }
2641 return err;
2642 }
2644 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2645 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2646 }
2648 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2649 const char* mesg) {
2650 assert(mesg != NULL, "mesg must be specified");
2651 int err = os::Linux::commit_memory_impl(addr, size, exec);
2652 if (err != 0) {
2653 // the caller wants all commit errors to exit with the specified mesg:
2654 warn_fail_commit_memory(addr, size, exec, err);
2655 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2656 }
2657 }
2659 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2660 #ifndef MAP_HUGETLB
2661 #define MAP_HUGETLB 0x40000
2662 #endif
2664 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2665 #ifndef MADV_HUGEPAGE
2666 #define MADV_HUGEPAGE 14
2667 #endif
2669 int os::Linux::commit_memory_impl(char* addr, size_t size,
2670 size_t alignment_hint, bool exec) {
2671 int err = os::Linux::commit_memory_impl(addr, size, exec);
2672 if (err == 0) {
2673 realign_memory(addr, size, alignment_hint);
2674 }
2675 return err;
2676 }
2678 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2679 bool exec) {
2680 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2681 }
2683 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2684 size_t alignment_hint, bool exec,
2685 const char* mesg) {
2686 assert(mesg != NULL, "mesg must be specified");
2687 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2688 if (err != 0) {
2689 // the caller wants all commit errors to exit with the specified mesg:
2690 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2691 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2692 }
2693 }
2695 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2696 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2697 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2698 // be supported or the memory may already be backed by huge pages.
2699 ::madvise(addr, bytes, MADV_HUGEPAGE);
2700 }
2701 }
2703 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2704 // This method works by doing an mmap over an existing mmaping and effectively discarding
2705 // the existing pages. However it won't work for SHM-based large pages that cannot be
2706 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2707 // small pages on top of the SHM segment. This method always works for small pages, so we
2708 // allow that in any case.
2709 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2710 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2711 }
2712 }
2714 void os::numa_make_global(char *addr, size_t bytes) {
2715 Linux::numa_interleave_memory(addr, bytes);
2716 }
2718 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2719 // bind policy to MPOL_PREFERRED for the current thread.
2720 #define USE_MPOL_PREFERRED 0
2722 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2723 // To make NUMA and large pages more robust when both enabled, we need to ease
2724 // the requirements on where the memory should be allocated. MPOL_BIND is the
2725 // default policy and it will force memory to be allocated on the specified
2726 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2727 // the specified node, but will not force it. Using this policy will prevent
2728 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2729 // free large pages.
2730 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2731 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2732 }
2734 bool os::numa_topology_changed() { return false; }
2736 size_t os::numa_get_groups_num() {
2737 int max_node = Linux::numa_max_node();
2738 return max_node > 0 ? max_node + 1 : 1;
2739 }
2741 int os::numa_get_group_id() {
2742 int cpu_id = Linux::sched_getcpu();
2743 if (cpu_id != -1) {
2744 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2745 if (lgrp_id != -1) {
2746 return lgrp_id;
2747 }
2748 }
2749 return 0;
2750 }
2752 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2753 for (size_t i = 0; i < size; i++) {
2754 ids[i] = i;
2755 }
2756 return size;
2757 }
2759 bool os::get_page_info(char *start, page_info* info) {
2760 return false;
2761 }
2763 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2764 return end;
2765 }
2768 int os::Linux::sched_getcpu_syscall(void) {
2769 unsigned int cpu = 0;
2770 int retval = -1;
2772 #if defined(IA32)
2773 # ifndef SYS_getcpu
2774 # define SYS_getcpu 318
2775 # endif
2776 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2777 #elif defined(AMD64)
2778 // Unfortunately we have to bring all these macros here from vsyscall.h
2779 // to be able to compile on old linuxes.
2780 # define __NR_vgetcpu 2
2781 # define VSYSCALL_START (-10UL << 20)
2782 # define VSYSCALL_SIZE 1024
2783 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2784 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2785 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2786 retval = vgetcpu(&cpu, NULL, NULL);
2787 #endif
2789 return (retval == -1) ? retval : cpu;
2790 }
2792 // Something to do with the numa-aware allocator needs these symbols
2793 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2794 extern "C" JNIEXPORT void numa_error(char *where) { }
2795 extern "C" JNIEXPORT int fork1() { return fork(); }
2798 // If we are running with libnuma version > 2, then we should
2799 // be trying to use symbols with versions 1.1
2800 // If we are running with earlier version, which did not have symbol versions,
2801 // we should use the base version.
2802 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2803 void *f = dlvsym(handle, name, "libnuma_1.1");
2804 if (f == NULL) {
2805 f = dlsym(handle, name);
2806 }
2807 return f;
2808 }
2810 bool os::Linux::libnuma_init() {
2811 // sched_getcpu() should be in libc.
2812 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2815 // If it's not, try a direct syscall.
2816 if (sched_getcpu() == -1)
2817 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2819 if (sched_getcpu() != -1) { // Does it work?
2820 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2821 if (handle != NULL) {
2822 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2823 libnuma_dlsym(handle, "numa_node_to_cpus")));
2824 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2825 libnuma_dlsym(handle, "numa_max_node")));
2826 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2827 libnuma_dlsym(handle, "numa_available")));
2828 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2829 libnuma_dlsym(handle, "numa_tonode_memory")));
2830 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2831 libnuma_dlsym(handle, "numa_interleave_memory")));
2832 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2833 libnuma_dlsym(handle, "numa_set_bind_policy")));
2836 if (numa_available() != -1) {
2837 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2838 // Create a cpu -> node mapping
2839 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2840 rebuild_cpu_to_node_map();
2841 return true;
2842 }
2843 }
2844 }
2845 return false;
2846 }
2848 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2849 // The table is later used in get_node_by_cpu().
2850 void os::Linux::rebuild_cpu_to_node_map() {
2851 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2852 // in libnuma (possible values are starting from 16,
2853 // and continuing up with every other power of 2, but less
2854 // than the maximum number of CPUs supported by kernel), and
2855 // is a subject to change (in libnuma version 2 the requirements
2856 // are more reasonable) we'll just hardcode the number they use
2857 // in the library.
2858 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2860 size_t cpu_num = os::active_processor_count();
2861 size_t cpu_map_size = NCPUS / BitsPerCLong;
2862 size_t cpu_map_valid_size =
2863 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2865 cpu_to_node()->clear();
2866 cpu_to_node()->at_grow(cpu_num - 1);
2867 size_t node_num = numa_get_groups_num();
2869 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2870 for (size_t i = 0; i < node_num; i++) {
2871 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2872 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2873 if (cpu_map[j] != 0) {
2874 for (size_t k = 0; k < BitsPerCLong; k++) {
2875 if (cpu_map[j] & (1UL << k)) {
2876 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2877 }
2878 }
2879 }
2880 }
2881 }
2882 }
2883 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2884 }
2886 int os::Linux::get_node_by_cpu(int cpu_id) {
2887 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2888 return cpu_to_node()->at(cpu_id);
2889 }
2890 return -1;
2891 }
2893 GrowableArray<int>* os::Linux::_cpu_to_node;
2894 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2895 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2896 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2897 os::Linux::numa_available_func_t os::Linux::_numa_available;
2898 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2899 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2900 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2901 unsigned long* os::Linux::_numa_all_nodes;
2903 bool os::pd_uncommit_memory(char* addr, size_t size) {
2904 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2905 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2906 return res != (uintptr_t) MAP_FAILED;
2907 }
2909 static
2910 address get_stack_commited_bottom(address bottom, size_t size) {
2911 address nbot = bottom;
2912 address ntop = bottom + size;
2914 size_t page_sz = os::vm_page_size();
2915 unsigned pages = size / page_sz;
2917 unsigned char vec[1];
2918 unsigned imin = 1, imax = pages + 1, imid;
2919 int mincore_return_value = 0;
2921 assert(imin <= imax, "Unexpected page size");
2923 while (imin < imax) {
2924 imid = (imax + imin) / 2;
2925 nbot = ntop - (imid * page_sz);
2927 // Use a trick with mincore to check whether the page is mapped or not.
2928 // mincore sets vec to 1 if page resides in memory and to 0 if page
2929 // is swapped output but if page we are asking for is unmapped
2930 // it returns -1,ENOMEM
2931 mincore_return_value = mincore(nbot, page_sz, vec);
2933 if (mincore_return_value == -1) {
2934 // Page is not mapped go up
2935 // to find first mapped page
2936 if (errno != EAGAIN) {
2937 assert(errno == ENOMEM, "Unexpected mincore errno");
2938 imax = imid;
2939 }
2940 } else {
2941 // Page is mapped go down
2942 // to find first not mapped page
2943 imin = imid + 1;
2944 }
2945 }
2947 nbot = nbot + page_sz;
2949 // Adjust stack bottom one page up if last checked page is not mapped
2950 if (mincore_return_value == -1) {
2951 nbot = nbot + page_sz;
2952 }
2954 return nbot;
2955 }
2958 // Linux uses a growable mapping for the stack, and if the mapping for
2959 // the stack guard pages is not removed when we detach a thread the
2960 // stack cannot grow beyond the pages where the stack guard was
2961 // mapped. If at some point later in the process the stack expands to
2962 // that point, the Linux kernel cannot expand the stack any further
2963 // because the guard pages are in the way, and a segfault occurs.
2964 //
2965 // However, it's essential not to split the stack region by unmapping
2966 // a region (leaving a hole) that's already part of the stack mapping,
2967 // so if the stack mapping has already grown beyond the guard pages at
2968 // the time we create them, we have to truncate the stack mapping.
2969 // So, we need to know the extent of the stack mapping when
2970 // create_stack_guard_pages() is called.
2972 // We only need this for stacks that are growable: at the time of
2973 // writing thread stacks don't use growable mappings (i.e. those
2974 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2975 // only applies to the main thread.
2977 // If the (growable) stack mapping already extends beyond the point
2978 // where we're going to put our guard pages, truncate the mapping at
2979 // that point by munmap()ping it. This ensures that when we later
2980 // munmap() the guard pages we don't leave a hole in the stack
2981 // mapping. This only affects the main/initial thread
2983 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2985 if (os::Linux::is_initial_thread()) {
2986 // As we manually grow stack up to bottom inside create_attached_thread(),
2987 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2988 // we don't need to do anything special.
2989 // Check it first, before calling heavy function.
2990 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2991 unsigned char vec[1];
2993 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2994 // Fallback to slow path on all errors, including EAGAIN
2995 stack_extent = (uintptr_t) get_stack_commited_bottom(
2996 os::Linux::initial_thread_stack_bottom(),
2997 (size_t)addr - stack_extent);
2998 }
3000 if (stack_extent < (uintptr_t)addr) {
3001 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3002 }
3003 }
3005 return os::commit_memory(addr, size, !ExecMem);
3006 }
3008 // If this is a growable mapping, remove the guard pages entirely by
3009 // munmap()ping them. If not, just call uncommit_memory(). This only
3010 // affects the main/initial thread, but guard against future OS changes
3011 // It's safe to always unmap guard pages for initial thread because we
3012 // always place it right after end of the mapped region
3014 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3015 uintptr_t stack_extent, stack_base;
3017 if (os::Linux::is_initial_thread()) {
3018 return ::munmap(addr, size) == 0;
3019 }
3021 return os::uncommit_memory(addr, size);
3022 }
3024 static address _highest_vm_reserved_address = NULL;
3026 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3027 // at 'requested_addr'. If there are existing memory mappings at the same
3028 // location, however, they will be overwritten. If 'fixed' is false,
3029 // 'requested_addr' is only treated as a hint, the return value may or
3030 // may not start from the requested address. Unlike Linux mmap(), this
3031 // function returns NULL to indicate failure.
3032 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3033 char * addr;
3034 int flags;
3036 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3037 if (fixed) {
3038 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3039 flags |= MAP_FIXED;
3040 }
3042 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3043 // touch an uncommitted page. Otherwise, the read/write might
3044 // succeed if we have enough swap space to back the physical page.
3045 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3046 flags, -1, 0);
3048 if (addr != MAP_FAILED) {
3049 // anon_mmap() should only get called during VM initialization,
3050 // don't need lock (actually we can skip locking even it can be called
3051 // from multiple threads, because _highest_vm_reserved_address is just a
3052 // hint about the upper limit of non-stack memory regions.)
3053 if ((address)addr + bytes > _highest_vm_reserved_address) {
3054 _highest_vm_reserved_address = (address)addr + bytes;
3055 }
3056 }
3058 return addr == MAP_FAILED ? NULL : addr;
3059 }
3061 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3062 // (req_addr != NULL) or with a given alignment.
3063 // - bytes shall be a multiple of alignment.
3064 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3065 // - alignment sets the alignment at which memory shall be allocated.
3066 // It must be a multiple of allocation granularity.
3067 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3068 // req_addr or NULL.
3069 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3071 size_t extra_size = bytes;
3072 if (req_addr == NULL && alignment > 0) {
3073 extra_size += alignment;
3074 }
3076 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3077 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3078 -1, 0);
3079 if (start == MAP_FAILED) {
3080 start = NULL;
3081 } else {
3082 if (req_addr != NULL) {
3083 if (start != req_addr) {
3084 ::munmap(start, extra_size);
3085 start = NULL;
3086 }
3087 } else {
3088 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3089 char* const end_aligned = start_aligned + bytes;
3090 char* const end = start + extra_size;
3091 if (start_aligned > start) {
3092 ::munmap(start, start_aligned - start);
3093 }
3094 if (end_aligned < end) {
3095 ::munmap(end_aligned, end - end_aligned);
3096 }
3097 start = start_aligned;
3098 }
3099 }
3100 return start;
3101 }
3103 // Don't update _highest_vm_reserved_address, because there might be memory
3104 // regions above addr + size. If so, releasing a memory region only creates
3105 // a hole in the address space, it doesn't help prevent heap-stack collision.
3106 //
3107 static int anon_munmap(char * addr, size_t size) {
3108 return ::munmap(addr, size) == 0;
3109 }
3111 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3112 size_t alignment_hint) {
3113 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3114 }
3116 bool os::pd_release_memory(char* addr, size_t size) {
3117 return anon_munmap(addr, size);
3118 }
3120 static address highest_vm_reserved_address() {
3121 return _highest_vm_reserved_address;
3122 }
3124 static bool linux_mprotect(char* addr, size_t size, int prot) {
3125 // Linux wants the mprotect address argument to be page aligned.
3126 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3128 // According to SUSv3, mprotect() should only be used with mappings
3129 // established by mmap(), and mmap() always maps whole pages. Unaligned
3130 // 'addr' likely indicates problem in the VM (e.g. trying to change
3131 // protection of malloc'ed or statically allocated memory). Check the
3132 // caller if you hit this assert.
3133 assert(addr == bottom, "sanity check");
3135 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3136 return ::mprotect(bottom, size, prot) == 0;
3137 }
3139 // Set protections specified
3140 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3141 bool is_committed) {
3142 unsigned int p = 0;
3143 switch (prot) {
3144 case MEM_PROT_NONE: p = PROT_NONE; break;
3145 case MEM_PROT_READ: p = PROT_READ; break;
3146 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3147 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3148 default:
3149 ShouldNotReachHere();
3150 }
3151 // is_committed is unused.
3152 return linux_mprotect(addr, bytes, p);
3153 }
3155 bool os::guard_memory(char* addr, size_t size) {
3156 return linux_mprotect(addr, size, PROT_NONE);
3157 }
3159 bool os::unguard_memory(char* addr, size_t size) {
3160 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3161 }
3163 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3164 bool result = false;
3165 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3166 MAP_ANONYMOUS|MAP_PRIVATE,
3167 -1, 0);
3168 if (p != MAP_FAILED) {
3169 void *aligned_p = align_ptr_up(p, page_size);
3171 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3173 munmap(p, page_size * 2);
3174 }
3176 if (warn && !result) {
3177 warning("TransparentHugePages is not supported by the operating system.");
3178 }
3180 return result;
3181 }
3183 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3184 bool result = false;
3185 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3186 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3187 -1, 0);
3189 if (p != MAP_FAILED) {
3190 // We don't know if this really is a huge page or not.
3191 FILE *fp = fopen("/proc/self/maps", "r");
3192 if (fp) {
3193 while (!feof(fp)) {
3194 char chars[257];
3195 long x = 0;
3196 if (fgets(chars, sizeof(chars), fp)) {
3197 if (sscanf(chars, "%lx-%*x", &x) == 1
3198 && x == (long)p) {
3199 if (strstr (chars, "hugepage")) {
3200 result = true;
3201 break;
3202 }
3203 }
3204 }
3205 }
3206 fclose(fp);
3207 }
3208 munmap(p, page_size);
3209 }
3211 if (warn && !result) {
3212 warning("HugeTLBFS is not supported by the operating system.");
3213 }
3215 return result;
3216 }
3218 /*
3219 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3220 *
3221 * From the coredump_filter documentation:
3222 *
3223 * - (bit 0) anonymous private memory
3224 * - (bit 1) anonymous shared memory
3225 * - (bit 2) file-backed private memory
3226 * - (bit 3) file-backed shared memory
3227 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3228 * effective only if the bit 2 is cleared)
3229 * - (bit 5) hugetlb private memory
3230 * - (bit 6) hugetlb shared memory
3231 */
3232 static void set_coredump_filter(void) {
3233 FILE *f;
3234 long cdm;
3236 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3237 return;
3238 }
3240 if (fscanf(f, "%lx", &cdm) != 1) {
3241 fclose(f);
3242 return;
3243 }
3245 rewind(f);
3247 if ((cdm & LARGEPAGES_BIT) == 0) {
3248 cdm |= LARGEPAGES_BIT;
3249 fprintf(f, "%#lx", cdm);
3250 }
3252 fclose(f);
3253 }
3255 // Large page support
3257 static size_t _large_page_size = 0;
3259 size_t os::Linux::find_large_page_size() {
3260 size_t large_page_size = 0;
3262 // large_page_size on Linux is used to round up heap size. x86 uses either
3263 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3264 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3265 // page as large as 256M.
3266 //
3267 // Here we try to figure out page size by parsing /proc/meminfo and looking
3268 // for a line with the following format:
3269 // Hugepagesize: 2048 kB
3270 //
3271 // If we can't determine the value (e.g. /proc is not mounted, or the text
3272 // format has been changed), we'll use the largest page size supported by
3273 // the processor.
3275 #ifndef ZERO
3276 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3277 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3278 #endif // ZERO
3280 FILE *fp = fopen("/proc/meminfo", "r");
3281 if (fp) {
3282 while (!feof(fp)) {
3283 int x = 0;
3284 char buf[16];
3285 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3286 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3287 large_page_size = x * K;
3288 break;
3289 }
3290 } else {
3291 // skip to next line
3292 for (;;) {
3293 int ch = fgetc(fp);
3294 if (ch == EOF || ch == (int)'\n') break;
3295 }
3296 }
3297 }
3298 fclose(fp);
3299 }
3301 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3302 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3303 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3304 proper_unit_for_byte_size(large_page_size));
3305 }
3307 return large_page_size;
3308 }
3310 size_t os::Linux::setup_large_page_size() {
3311 _large_page_size = Linux::find_large_page_size();
3312 const size_t default_page_size = (size_t)Linux::page_size();
3313 if (_large_page_size > default_page_size) {
3314 _page_sizes[0] = _large_page_size;
3315 _page_sizes[1] = default_page_size;
3316 _page_sizes[2] = 0;
3317 }
3319 return _large_page_size;
3320 }
3322 bool os::Linux::setup_large_page_type(size_t page_size) {
3323 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3324 FLAG_IS_DEFAULT(UseSHM) &&
3325 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3327 // The type of large pages has not been specified by the user.
3329 // Try UseHugeTLBFS and then UseSHM.
3330 UseHugeTLBFS = UseSHM = true;
3332 // Don't try UseTransparentHugePages since there are known
3333 // performance issues with it turned on. This might change in the future.
3334 UseTransparentHugePages = false;
3335 }
3337 if (UseTransparentHugePages) {
3338 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3339 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3340 UseHugeTLBFS = false;
3341 UseSHM = false;
3342 return true;
3343 }
3344 UseTransparentHugePages = false;
3345 }
3347 if (UseHugeTLBFS) {
3348 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3349 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3350 UseSHM = false;
3351 return true;
3352 }
3353 UseHugeTLBFS = false;
3354 }
3356 return UseSHM;
3357 }
3359 void os::large_page_init() {
3360 if (!UseLargePages &&
3361 !UseTransparentHugePages &&
3362 !UseHugeTLBFS &&
3363 !UseSHM) {
3364 // Not using large pages.
3365 return;
3366 }
3368 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3369 // The user explicitly turned off large pages.
3370 // Ignore the rest of the large pages flags.
3371 UseTransparentHugePages = false;
3372 UseHugeTLBFS = false;
3373 UseSHM = false;
3374 return;
3375 }
3377 size_t large_page_size = Linux::setup_large_page_size();
3378 UseLargePages = Linux::setup_large_page_type(large_page_size);
3380 set_coredump_filter();
3381 }
3383 #ifndef SHM_HUGETLB
3384 #define SHM_HUGETLB 04000
3385 #endif
3387 #define shm_warning_format(format, ...) \
3388 do { \
3389 if (UseLargePages && \
3390 (!FLAG_IS_DEFAULT(UseLargePages) || \
3391 !FLAG_IS_DEFAULT(UseSHM) || \
3392 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3393 warning(format, __VA_ARGS__); \
3394 } \
3395 } while (0)
3397 #define shm_warning(str) shm_warning_format("%s", str)
3399 #define shm_warning_with_errno(str) \
3400 do { \
3401 int err = errno; \
3402 shm_warning_format(str " (error = %d)", err); \
3403 } while (0)
3405 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3406 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3408 if (!is_size_aligned(alignment, SHMLBA)) {
3409 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3410 return NULL;
3411 }
3413 // To ensure that we get 'alignment' aligned memory from shmat,
3414 // we pre-reserve aligned virtual memory and then attach to that.
3416 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3417 if (pre_reserved_addr == NULL) {
3418 // Couldn't pre-reserve aligned memory.
3419 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3420 return NULL;
3421 }
3423 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3424 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3426 if ((intptr_t)addr == -1) {
3427 int err = errno;
3428 shm_warning_with_errno("Failed to attach shared memory.");
3430 assert(err != EACCES, "Unexpected error");
3431 assert(err != EIDRM, "Unexpected error");
3432 assert(err != EINVAL, "Unexpected error");
3434 // Since we don't know if the kernel unmapped the pre-reserved memory area
3435 // we can't unmap it, since that would potentially unmap memory that was
3436 // mapped from other threads.
3437 return NULL;
3438 }
3440 return addr;
3441 }
3443 static char* shmat_at_address(int shmid, char* req_addr) {
3444 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3445 assert(false, "Requested address needs to be SHMLBA aligned");
3446 return NULL;
3447 }
3449 char* addr = (char*)shmat(shmid, req_addr, 0);
3451 if ((intptr_t)addr == -1) {
3452 shm_warning_with_errno("Failed to attach shared memory.");
3453 return NULL;
3454 }
3456 return addr;
3457 }
3459 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3460 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3461 if (req_addr != NULL) {
3462 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3463 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3464 return shmat_at_address(shmid, req_addr);
3465 }
3467 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3468 // return large page size aligned memory addresses when req_addr == NULL.
3469 // However, if the alignment is larger than the large page size, we have
3470 // to manually ensure that the memory returned is 'alignment' aligned.
3471 if (alignment > os::large_page_size()) {
3472 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3473 return shmat_with_alignment(shmid, bytes, alignment);
3474 } else {
3475 return shmat_at_address(shmid, NULL);
3476 }
3477 }
3479 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3480 // "exec" is passed in but not used. Creating the shared image for
3481 // the code cache doesn't have an SHM_X executable permission to check.
3482 assert(UseLargePages && UseSHM, "only for SHM large pages");
3483 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3484 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3486 if (!is_size_aligned(bytes, os::large_page_size())) {
3487 return NULL; // Fallback to small pages.
3488 }
3490 // Create a large shared memory region to attach to based on size.
3491 // Currently, size is the total size of the heap.
3492 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3493 if (shmid == -1) {
3494 // Possible reasons for shmget failure:
3495 // 1. shmmax is too small for Java heap.
3496 // > check shmmax value: cat /proc/sys/kernel/shmmax
3497 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3498 // 2. not enough large page memory.
3499 // > check available large pages: cat /proc/meminfo
3500 // > increase amount of large pages:
3501 // echo new_value > /proc/sys/vm/nr_hugepages
3502 // Note 1: different Linux may use different name for this property,
3503 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3504 // Note 2: it's possible there's enough physical memory available but
3505 // they are so fragmented after a long run that they can't
3506 // coalesce into large pages. Try to reserve large pages when
3507 // the system is still "fresh".
3508 shm_warning_with_errno("Failed to reserve shared memory.");
3509 return NULL;
3510 }
3512 // Attach to the region.
3513 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3515 // Remove shmid. If shmat() is successful, the actual shared memory segment
3516 // will be deleted when it's detached by shmdt() or when the process
3517 // terminates. If shmat() is not successful this will remove the shared
3518 // segment immediately.
3519 shmctl(shmid, IPC_RMID, NULL);
3521 return addr;
3522 }
3524 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3525 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3527 bool warn_on_failure = UseLargePages &&
3528 (!FLAG_IS_DEFAULT(UseLargePages) ||
3529 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3530 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3532 if (warn_on_failure) {
3533 char msg[128];
3534 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3535 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3536 warning("%s", msg);
3537 }
3538 }
3540 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3541 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3542 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3543 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3545 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3546 char* addr = (char*)::mmap(req_addr, bytes, prot,
3547 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3548 -1, 0);
3550 if (addr == MAP_FAILED) {
3551 warn_on_large_pages_failure(req_addr, bytes, errno);
3552 return NULL;
3553 }
3555 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3557 return addr;
3558 }
3560 // Reserve memory using mmap(MAP_HUGETLB).
3561 // - bytes shall be a multiple of alignment.
3562 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3563 // - alignment sets the alignment at which memory shall be allocated.
3564 // It must be a multiple of allocation granularity.
3565 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3566 // req_addr or NULL.
3567 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3568 size_t large_page_size = os::large_page_size();
3569 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3571 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3572 assert(is_size_aligned(bytes, alignment), "Must be");
3574 // First reserve - but not commit - the address range in small pages.
3575 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3577 if (start == NULL) {
3578 return NULL;
3579 }
3581 assert(is_ptr_aligned(start, alignment), "Must be");
3583 char* end = start + bytes;
3585 // Find the regions of the allocated chunk that can be promoted to large pages.
3586 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3587 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3589 size_t lp_bytes = lp_end - lp_start;
3591 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3593 if (lp_bytes == 0) {
3594 // The mapped region doesn't even span the start and the end of a large page.
3595 // Fall back to allocate a non-special area.
3596 ::munmap(start, end - start);
3597 return NULL;
3598 }
3600 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3602 void* result;
3604 // Commit small-paged leading area.
3605 if (start != lp_start) {
3606 result = ::mmap(start, lp_start - start, prot,
3607 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3608 -1, 0);
3609 if (result == MAP_FAILED) {
3610 ::munmap(lp_start, end - lp_start);
3611 return NULL;
3612 }
3613 }
3615 // Commit large-paged area.
3616 result = ::mmap(lp_start, lp_bytes, prot,
3617 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3618 -1, 0);
3619 if (result == MAP_FAILED) {
3620 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3621 // If the mmap above fails, the large pages region will be unmapped and we
3622 // have regions before and after with small pages. Release these regions.
3623 //
3624 // | mapped | unmapped | mapped |
3625 // ^ ^ ^ ^
3626 // start lp_start lp_end end
3627 //
3628 ::munmap(start, lp_start - start);
3629 ::munmap(lp_end, end - lp_end);
3630 return NULL;
3631 }
3633 // Commit small-paged trailing area.
3634 if (lp_end != end) {
3635 result = ::mmap(lp_end, end - lp_end, prot,
3636 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3637 -1, 0);
3638 if (result == MAP_FAILED) {
3639 ::munmap(start, lp_end - start);
3640 return NULL;
3641 }
3642 }
3644 return start;
3645 }
3647 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3648 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3649 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3650 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3651 assert(is_power_of_2(os::large_page_size()), "Must be");
3652 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3654 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3655 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3656 } else {
3657 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3658 }
3659 }
3661 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3662 assert(UseLargePages, "only for large pages");
3664 char* addr;
3665 if (UseSHM) {
3666 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3667 } else {
3668 assert(UseHugeTLBFS, "must be");
3669 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3670 }
3672 if (addr != NULL) {
3673 if (UseNUMAInterleaving) {
3674 numa_make_global(addr, bytes);
3675 }
3677 // The memory is committed
3678 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3679 }
3681 return addr;
3682 }
3684 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3685 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3686 return shmdt(base) == 0;
3687 }
3689 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3690 return pd_release_memory(base, bytes);
3691 }
3693 bool os::release_memory_special(char* base, size_t bytes) {
3694 bool res;
3695 if (MemTracker::tracking_level() > NMT_minimal) {
3696 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3697 res = os::Linux::release_memory_special_impl(base, bytes);
3698 if (res) {
3699 tkr.record((address)base, bytes);
3700 }
3702 } else {
3703 res = os::Linux::release_memory_special_impl(base, bytes);
3704 }
3705 return res;
3706 }
3708 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3709 assert(UseLargePages, "only for large pages");
3710 bool res;
3712 if (UseSHM) {
3713 res = os::Linux::release_memory_special_shm(base, bytes);
3714 } else {
3715 assert(UseHugeTLBFS, "must be");
3716 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3717 }
3718 return res;
3719 }
3721 size_t os::large_page_size() {
3722 return _large_page_size;
3723 }
3725 // With SysV SHM the entire memory region must be allocated as shared
3726 // memory.
3727 // HugeTLBFS allows application to commit large page memory on demand.
3728 // However, when committing memory with HugeTLBFS fails, the region
3729 // that was supposed to be committed will lose the old reservation
3730 // and allow other threads to steal that memory region. Because of this
3731 // behavior we can't commit HugeTLBFS memory.
3732 bool os::can_commit_large_page_memory() {
3733 return UseTransparentHugePages;
3734 }
3736 bool os::can_execute_large_page_memory() {
3737 return UseTransparentHugePages || UseHugeTLBFS;
3738 }
3740 // Reserve memory at an arbitrary address, only if that area is
3741 // available (and not reserved for something else).
3743 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3744 const int max_tries = 10;
3745 char* base[max_tries];
3746 size_t size[max_tries];
3747 const size_t gap = 0x000000;
3749 // Assert only that the size is a multiple of the page size, since
3750 // that's all that mmap requires, and since that's all we really know
3751 // about at this low abstraction level. If we need higher alignment,
3752 // we can either pass an alignment to this method or verify alignment
3753 // in one of the methods further up the call chain. See bug 5044738.
3754 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3756 // Repeatedly allocate blocks until the block is allocated at the
3757 // right spot. Give up after max_tries. Note that reserve_memory() will
3758 // automatically update _highest_vm_reserved_address if the call is
3759 // successful. The variable tracks the highest memory address every reserved
3760 // by JVM. It is used to detect heap-stack collision if running with
3761 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3762 // space than needed, it could confuse the collision detecting code. To
3763 // solve the problem, save current _highest_vm_reserved_address and
3764 // calculate the correct value before return.
3765 address old_highest = _highest_vm_reserved_address;
3767 // Linux mmap allows caller to pass an address as hint; give it a try first,
3768 // if kernel honors the hint then we can return immediately.
3769 char * addr = anon_mmap(requested_addr, bytes, false);
3770 if (addr == requested_addr) {
3771 return requested_addr;
3772 }
3774 if (addr != NULL) {
3775 // mmap() is successful but it fails to reserve at the requested address
3776 anon_munmap(addr, bytes);
3777 }
3779 int i;
3780 for (i = 0; i < max_tries; ++i) {
3781 base[i] = reserve_memory(bytes);
3783 if (base[i] != NULL) {
3784 // Is this the block we wanted?
3785 if (base[i] == requested_addr) {
3786 size[i] = bytes;
3787 break;
3788 }
3790 // Does this overlap the block we wanted? Give back the overlapped
3791 // parts and try again.
3793 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3794 if (top_overlap >= 0 && top_overlap < bytes) {
3795 unmap_memory(base[i], top_overlap);
3796 base[i] += top_overlap;
3797 size[i] = bytes - top_overlap;
3798 } else {
3799 size_t bottom_overlap = base[i] + bytes - requested_addr;
3800 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3801 unmap_memory(requested_addr, bottom_overlap);
3802 size[i] = bytes - bottom_overlap;
3803 } else {
3804 size[i] = bytes;
3805 }
3806 }
3807 }
3808 }
3810 // Give back the unused reserved pieces.
3812 for (int j = 0; j < i; ++j) {
3813 if (base[j] != NULL) {
3814 unmap_memory(base[j], size[j]);
3815 }
3816 }
3818 if (i < max_tries) {
3819 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3820 return requested_addr;
3821 } else {
3822 _highest_vm_reserved_address = old_highest;
3823 return NULL;
3824 }
3825 }
3827 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3828 return ::read(fd, buf, nBytes);
3829 }
3831 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3832 // Solaris uses poll(), linux uses park().
3833 // Poll() is likely a better choice, assuming that Thread.interrupt()
3834 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3835 // SIGSEGV, see 4355769.
3837 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3838 assert(thread == Thread::current(), "thread consistency check");
3840 ParkEvent * const slp = thread->_SleepEvent ;
3841 slp->reset() ;
3842 OrderAccess::fence() ;
3844 if (interruptible) {
3845 jlong prevtime = javaTimeNanos();
3847 for (;;) {
3848 if (os::is_interrupted(thread, true)) {
3849 return OS_INTRPT;
3850 }
3852 jlong newtime = javaTimeNanos();
3854 if (newtime - prevtime < 0) {
3855 // time moving backwards, should only happen if no monotonic clock
3856 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3857 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3858 } else {
3859 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3860 }
3862 if(millis <= 0) {
3863 return OS_OK;
3864 }
3866 prevtime = newtime;
3868 {
3869 assert(thread->is_Java_thread(), "sanity check");
3870 JavaThread *jt = (JavaThread *) thread;
3871 ThreadBlockInVM tbivm(jt);
3872 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3874 jt->set_suspend_equivalent();
3875 // cleared by handle_special_suspend_equivalent_condition() or
3876 // java_suspend_self() via check_and_wait_while_suspended()
3878 slp->park(millis);
3880 // were we externally suspended while we were waiting?
3881 jt->check_and_wait_while_suspended();
3882 }
3883 }
3884 } else {
3885 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3886 jlong prevtime = javaTimeNanos();
3888 for (;;) {
3889 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3890 // the 1st iteration ...
3891 jlong newtime = javaTimeNanos();
3893 if (newtime - prevtime < 0) {
3894 // time moving backwards, should only happen if no monotonic clock
3895 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3896 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3897 } else {
3898 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3899 }
3901 if(millis <= 0) break ;
3903 prevtime = newtime;
3904 slp->park(millis);
3905 }
3906 return OS_OK ;
3907 }
3908 }
3910 //
3911 // Short sleep, direct OS call.
3912 //
3913 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3914 // sched_yield(2) will actually give up the CPU:
3915 //
3916 // * Alone on this pariticular CPU, keeps running.
3917 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3918 // (pre 2.6.39).
3919 //
3920 // So calling this with 0 is an alternative.
3921 //
3922 void os::naked_short_sleep(jlong ms) {
3923 struct timespec req;
3925 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3926 req.tv_sec = 0;
3927 if (ms > 0) {
3928 req.tv_nsec = (ms % 1000) * 1000000;
3929 }
3930 else {
3931 req.tv_nsec = 1;
3932 }
3934 nanosleep(&req, NULL);
3936 return;
3937 }
3939 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3940 void os::infinite_sleep() {
3941 while (true) { // sleep forever ...
3942 ::sleep(100); // ... 100 seconds at a time
3943 }
3944 }
3946 // Used to convert frequent JVM_Yield() to nops
3947 bool os::dont_yield() {
3948 return DontYieldALot;
3949 }
3951 void os::yield() {
3952 sched_yield();
3953 }
3955 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3957 void os::yield_all(int attempts) {
3958 // Yields to all threads, including threads with lower priorities
3959 // Threads on Linux are all with same priority. The Solaris style
3960 // os::yield_all() with nanosleep(1ms) is not necessary.
3961 sched_yield();
3962 }
3964 // Called from the tight loops to possibly influence time-sharing heuristics
3965 void os::loop_breaker(int attempts) {
3966 os::yield_all(attempts);
3967 }
3969 ////////////////////////////////////////////////////////////////////////////////
3970 // thread priority support
3972 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3973 // only supports dynamic priority, static priority must be zero. For real-time
3974 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3975 // However, for large multi-threaded applications, SCHED_RR is not only slower
3976 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3977 // of 5 runs - Sep 2005).
3978 //
3979 // The following code actually changes the niceness of kernel-thread/LWP. It
3980 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3981 // not the entire user process, and user level threads are 1:1 mapped to kernel
3982 // threads. It has always been the case, but could change in the future. For
3983 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3984 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3986 int os::java_to_os_priority[CriticalPriority + 1] = {
3987 19, // 0 Entry should never be used
3989 4, // 1 MinPriority
3990 3, // 2
3991 2, // 3
3993 1, // 4
3994 0, // 5 NormPriority
3995 -1, // 6
3997 -2, // 7
3998 -3, // 8
3999 -4, // 9 NearMaxPriority
4001 -5, // 10 MaxPriority
4003 -5 // 11 CriticalPriority
4004 };
4006 static int prio_init() {
4007 if (ThreadPriorityPolicy == 1) {
4008 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4009 // if effective uid is not root. Perhaps, a more elegant way of doing
4010 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4011 if (geteuid() != 0) {
4012 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4013 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4014 }
4015 ThreadPriorityPolicy = 0;
4016 }
4017 }
4018 if (UseCriticalJavaThreadPriority) {
4019 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4020 }
4021 return 0;
4022 }
4024 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4025 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4027 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4028 return (ret == 0) ? OS_OK : OS_ERR;
4029 }
4031 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4032 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4033 *priority_ptr = java_to_os_priority[NormPriority];
4034 return OS_OK;
4035 }
4037 errno = 0;
4038 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4039 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4040 }
4042 // Hint to the underlying OS that a task switch would not be good.
4043 // Void return because it's a hint and can fail.
4044 void os::hint_no_preempt() {}
4046 ////////////////////////////////////////////////////////////////////////////////
4047 // suspend/resume support
4049 // the low-level signal-based suspend/resume support is a remnant from the
4050 // old VM-suspension that used to be for java-suspension, safepoints etc,
4051 // within hotspot. Now there is a single use-case for this:
4052 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4053 // that runs in the watcher thread.
4054 // The remaining code is greatly simplified from the more general suspension
4055 // code that used to be used.
4056 //
4057 // The protocol is quite simple:
4058 // - suspend:
4059 // - sends a signal to the target thread
4060 // - polls the suspend state of the osthread using a yield loop
4061 // - target thread signal handler (SR_handler) sets suspend state
4062 // and blocks in sigsuspend until continued
4063 // - resume:
4064 // - sets target osthread state to continue
4065 // - sends signal to end the sigsuspend loop in the SR_handler
4066 //
4067 // Note that the SR_lock plays no role in this suspend/resume protocol.
4068 //
4070 static void resume_clear_context(OSThread *osthread) {
4071 osthread->set_ucontext(NULL);
4072 osthread->set_siginfo(NULL);
4073 }
4075 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4076 osthread->set_ucontext(context);
4077 osthread->set_siginfo(siginfo);
4078 }
4080 //
4081 // Handler function invoked when a thread's execution is suspended or
4082 // resumed. We have to be careful that only async-safe functions are
4083 // called here (Note: most pthread functions are not async safe and
4084 // should be avoided.)
4085 //
4086 // Note: sigwait() is a more natural fit than sigsuspend() from an
4087 // interface point of view, but sigwait() prevents the signal hander
4088 // from being run. libpthread would get very confused by not having
4089 // its signal handlers run and prevents sigwait()'s use with the
4090 // mutex granting granting signal.
4091 //
4092 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4093 //
4094 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4095 // Save and restore errno to avoid confusing native code with EINTR
4096 // after sigsuspend.
4097 int old_errno = errno;
4099 Thread* thread = Thread::current();
4100 OSThread* osthread = thread->osthread();
4101 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4103 os::SuspendResume::State current = osthread->sr.state();
4104 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4105 suspend_save_context(osthread, siginfo, context);
4107 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4108 os::SuspendResume::State state = osthread->sr.suspended();
4109 if (state == os::SuspendResume::SR_SUSPENDED) {
4110 sigset_t suspend_set; // signals for sigsuspend()
4112 // get current set of blocked signals and unblock resume signal
4113 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4114 sigdelset(&suspend_set, SR_signum);
4116 sr_semaphore.signal();
4117 // wait here until we are resumed
4118 while (1) {
4119 sigsuspend(&suspend_set);
4121 os::SuspendResume::State result = osthread->sr.running();
4122 if (result == os::SuspendResume::SR_RUNNING) {
4123 sr_semaphore.signal();
4124 break;
4125 }
4126 }
4128 } else if (state == os::SuspendResume::SR_RUNNING) {
4129 // request was cancelled, continue
4130 } else {
4131 ShouldNotReachHere();
4132 }
4134 resume_clear_context(osthread);
4135 } else if (current == os::SuspendResume::SR_RUNNING) {
4136 // request was cancelled, continue
4137 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4138 // ignore
4139 } else {
4140 // ignore
4141 }
4143 errno = old_errno;
4144 }
4147 static int SR_initialize() {
4148 struct sigaction act;
4149 char *s;
4150 /* Get signal number to use for suspend/resume */
4151 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4152 int sig = ::strtol(s, 0, 10);
4153 if (sig > 0 || sig < _NSIG) {
4154 SR_signum = sig;
4155 }
4156 }
4158 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4159 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4161 sigemptyset(&SR_sigset);
4162 sigaddset(&SR_sigset, SR_signum);
4164 /* Set up signal handler for suspend/resume */
4165 act.sa_flags = SA_RESTART|SA_SIGINFO;
4166 act.sa_handler = (void (*)(int)) SR_handler;
4168 // SR_signum is blocked by default.
4169 // 4528190 - We also need to block pthread restart signal (32 on all
4170 // supported Linux platforms). Note that LinuxThreads need to block
4171 // this signal for all threads to work properly. So we don't have
4172 // to use hard-coded signal number when setting up the mask.
4173 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4175 if (sigaction(SR_signum, &act, 0) == -1) {
4176 return -1;
4177 }
4179 // Save signal flag
4180 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4181 return 0;
4182 }
4184 static int sr_notify(OSThread* osthread) {
4185 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4186 assert_status(status == 0, status, "pthread_kill");
4187 return status;
4188 }
4190 // "Randomly" selected value for how long we want to spin
4191 // before bailing out on suspending a thread, also how often
4192 // we send a signal to a thread we want to resume
4193 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4194 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4196 // returns true on success and false on error - really an error is fatal
4197 // but this seems the normal response to library errors
4198 static bool do_suspend(OSThread* osthread) {
4199 assert(osthread->sr.is_running(), "thread should be running");
4200 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4202 // mark as suspended and send signal
4203 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4204 // failed to switch, state wasn't running?
4205 ShouldNotReachHere();
4206 return false;
4207 }
4209 if (sr_notify(osthread) != 0) {
4210 ShouldNotReachHere();
4211 }
4213 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4214 while (true) {
4215 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4216 break;
4217 } else {
4218 // timeout
4219 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4220 if (cancelled == os::SuspendResume::SR_RUNNING) {
4221 return false;
4222 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4223 // make sure that we consume the signal on the semaphore as well
4224 sr_semaphore.wait();
4225 break;
4226 } else {
4227 ShouldNotReachHere();
4228 return false;
4229 }
4230 }
4231 }
4233 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4234 return true;
4235 }
4237 static void do_resume(OSThread* osthread) {
4238 assert(osthread->sr.is_suspended(), "thread should be suspended");
4239 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4241 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4242 // failed to switch to WAKEUP_REQUEST
4243 ShouldNotReachHere();
4244 return;
4245 }
4247 while (true) {
4248 if (sr_notify(osthread) == 0) {
4249 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4250 if (osthread->sr.is_running()) {
4251 return;
4252 }
4253 }
4254 } else {
4255 ShouldNotReachHere();
4256 }
4257 }
4259 guarantee(osthread->sr.is_running(), "Must be running!");
4260 }
4262 ////////////////////////////////////////////////////////////////////////////////
4263 // interrupt support
4265 void os::interrupt(Thread* thread) {
4266 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4267 "possibility of dangling Thread pointer");
4269 OSThread* osthread = thread->osthread();
4271 if (!osthread->interrupted()) {
4272 osthread->set_interrupted(true);
4273 // More than one thread can get here with the same value of osthread,
4274 // resulting in multiple notifications. We do, however, want the store
4275 // to interrupted() to be visible to other threads before we execute unpark().
4276 OrderAccess::fence();
4277 ParkEvent * const slp = thread->_SleepEvent ;
4278 if (slp != NULL) slp->unpark() ;
4279 }
4281 // For JSR166. Unpark even if interrupt status already was set
4282 if (thread->is_Java_thread())
4283 ((JavaThread*)thread)->parker()->unpark();
4285 ParkEvent * ev = thread->_ParkEvent ;
4286 if (ev != NULL) ev->unpark() ;
4288 }
4290 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4291 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4292 "possibility of dangling Thread pointer");
4294 OSThread* osthread = thread->osthread();
4296 bool interrupted = osthread->interrupted();
4298 if (interrupted && clear_interrupted) {
4299 osthread->set_interrupted(false);
4300 // consider thread->_SleepEvent->reset() ... optional optimization
4301 }
4303 return interrupted;
4304 }
4306 ///////////////////////////////////////////////////////////////////////////////////
4307 // signal handling (except suspend/resume)
4309 // This routine may be used by user applications as a "hook" to catch signals.
4310 // The user-defined signal handler must pass unrecognized signals to this
4311 // routine, and if it returns true (non-zero), then the signal handler must
4312 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4313 // routine will never retun false (zero), but instead will execute a VM panic
4314 // routine kill the process.
4315 //
4316 // If this routine returns false, it is OK to call it again. This allows
4317 // the user-defined signal handler to perform checks either before or after
4318 // the VM performs its own checks. Naturally, the user code would be making
4319 // a serious error if it tried to handle an exception (such as a null check
4320 // or breakpoint) that the VM was generating for its own correct operation.
4321 //
4322 // This routine may recognize any of the following kinds of signals:
4323 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4324 // It should be consulted by handlers for any of those signals.
4325 //
4326 // The caller of this routine must pass in the three arguments supplied
4327 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4328 // field of the structure passed to sigaction(). This routine assumes that
4329 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4330 //
4331 // Note that the VM will print warnings if it detects conflicting signal
4332 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4333 //
4334 extern "C" JNIEXPORT int
4335 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4336 void* ucontext, int abort_if_unrecognized);
4338 void signalHandler(int sig, siginfo_t* info, void* uc) {
4339 assert(info != NULL && uc != NULL, "it must be old kernel");
4340 int orig_errno = errno; // Preserve errno value over signal handler.
4341 JVM_handle_linux_signal(sig, info, uc, true);
4342 errno = orig_errno;
4343 }
4346 // This boolean allows users to forward their own non-matching signals
4347 // to JVM_handle_linux_signal, harmlessly.
4348 bool os::Linux::signal_handlers_are_installed = false;
4350 // For signal-chaining
4351 struct sigaction os::Linux::sigact[MAXSIGNUM];
4352 unsigned int os::Linux::sigs = 0;
4353 bool os::Linux::libjsig_is_loaded = false;
4354 typedef struct sigaction *(*get_signal_t)(int);
4355 get_signal_t os::Linux::get_signal_action = NULL;
4357 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4358 struct sigaction *actp = NULL;
4360 if (libjsig_is_loaded) {
4361 // Retrieve the old signal handler from libjsig
4362 actp = (*get_signal_action)(sig);
4363 }
4364 if (actp == NULL) {
4365 // Retrieve the preinstalled signal handler from jvm
4366 actp = get_preinstalled_handler(sig);
4367 }
4369 return actp;
4370 }
4372 static bool call_chained_handler(struct sigaction *actp, int sig,
4373 siginfo_t *siginfo, void *context) {
4374 // Call the old signal handler
4375 if (actp->sa_handler == SIG_DFL) {
4376 // It's more reasonable to let jvm treat it as an unexpected exception
4377 // instead of taking the default action.
4378 return false;
4379 } else if (actp->sa_handler != SIG_IGN) {
4380 if ((actp->sa_flags & SA_NODEFER) == 0) {
4381 // automaticlly block the signal
4382 sigaddset(&(actp->sa_mask), sig);
4383 }
4385 sa_handler_t hand = NULL;
4386 sa_sigaction_t sa = NULL;
4387 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4388 // retrieve the chained handler
4389 if (siginfo_flag_set) {
4390 sa = actp->sa_sigaction;
4391 } else {
4392 hand = actp->sa_handler;
4393 }
4395 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4396 actp->sa_handler = SIG_DFL;
4397 }
4399 // try to honor the signal mask
4400 sigset_t oset;
4401 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4403 // call into the chained handler
4404 if (siginfo_flag_set) {
4405 (*sa)(sig, siginfo, context);
4406 } else {
4407 (*hand)(sig);
4408 }
4410 // restore the signal mask
4411 pthread_sigmask(SIG_SETMASK, &oset, 0);
4412 }
4413 // Tell jvm's signal handler the signal is taken care of.
4414 return true;
4415 }
4417 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4418 bool chained = false;
4419 // signal-chaining
4420 if (UseSignalChaining) {
4421 struct sigaction *actp = get_chained_signal_action(sig);
4422 if (actp != NULL) {
4423 chained = call_chained_handler(actp, sig, siginfo, context);
4424 }
4425 }
4426 return chained;
4427 }
4429 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4430 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4431 return &sigact[sig];
4432 }
4433 return NULL;
4434 }
4436 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4437 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4438 sigact[sig] = oldAct;
4439 sigs |= (unsigned int)1 << sig;
4440 }
4442 // for diagnostic
4443 int os::Linux::sigflags[MAXSIGNUM];
4445 int os::Linux::get_our_sigflags(int sig) {
4446 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4447 return sigflags[sig];
4448 }
4450 void os::Linux::set_our_sigflags(int sig, int flags) {
4451 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4452 sigflags[sig] = flags;
4453 }
4455 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4456 // Check for overwrite.
4457 struct sigaction oldAct;
4458 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4460 void* oldhand = oldAct.sa_sigaction
4461 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4462 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4463 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4464 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4465 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4466 if (AllowUserSignalHandlers || !set_installed) {
4467 // Do not overwrite; user takes responsibility to forward to us.
4468 return;
4469 } else if (UseSignalChaining) {
4470 // save the old handler in jvm
4471 save_preinstalled_handler(sig, oldAct);
4472 // libjsig also interposes the sigaction() call below and saves the
4473 // old sigaction on it own.
4474 } else {
4475 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4476 "%#lx for signal %d.", (long)oldhand, sig));
4477 }
4478 }
4480 struct sigaction sigAct;
4481 sigfillset(&(sigAct.sa_mask));
4482 sigAct.sa_handler = SIG_DFL;
4483 if (!set_installed) {
4484 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4485 } else {
4486 sigAct.sa_sigaction = signalHandler;
4487 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4488 }
4489 // Save flags, which are set by ours
4490 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4491 sigflags[sig] = sigAct.sa_flags;
4493 int ret = sigaction(sig, &sigAct, &oldAct);
4494 assert(ret == 0, "check");
4496 void* oldhand2 = oldAct.sa_sigaction
4497 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4498 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4499 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4500 }
4502 // install signal handlers for signals that HotSpot needs to
4503 // handle in order to support Java-level exception handling.
4505 void os::Linux::install_signal_handlers() {
4506 if (!signal_handlers_are_installed) {
4507 signal_handlers_are_installed = true;
4509 // signal-chaining
4510 typedef void (*signal_setting_t)();
4511 signal_setting_t begin_signal_setting = NULL;
4512 signal_setting_t end_signal_setting = NULL;
4513 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4514 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4515 if (begin_signal_setting != NULL) {
4516 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4517 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4518 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4519 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4520 libjsig_is_loaded = true;
4521 assert(UseSignalChaining, "should enable signal-chaining");
4522 }
4523 if (libjsig_is_loaded) {
4524 // Tell libjsig jvm is setting signal handlers
4525 (*begin_signal_setting)();
4526 }
4528 set_signal_handler(SIGSEGV, true);
4529 set_signal_handler(SIGPIPE, true);
4530 set_signal_handler(SIGBUS, true);
4531 set_signal_handler(SIGILL, true);
4532 set_signal_handler(SIGFPE, true);
4533 #if defined(PPC64)
4534 set_signal_handler(SIGTRAP, true);
4535 #endif
4536 set_signal_handler(SIGXFSZ, true);
4538 if (libjsig_is_loaded) {
4539 // Tell libjsig jvm finishes setting signal handlers
4540 (*end_signal_setting)();
4541 }
4543 // We don't activate signal checker if libjsig is in place, we trust ourselves
4544 // and if UserSignalHandler is installed all bets are off.
4545 // Log that signal checking is off only if -verbose:jni is specified.
4546 if (CheckJNICalls) {
4547 if (libjsig_is_loaded) {
4548 if (PrintJNIResolving) {
4549 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4550 }
4551 check_signals = false;
4552 }
4553 if (AllowUserSignalHandlers) {
4554 if (PrintJNIResolving) {
4555 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4556 }
4557 check_signals = false;
4558 }
4559 }
4560 }
4561 }
4563 // This is the fastest way to get thread cpu time on Linux.
4564 // Returns cpu time (user+sys) for any thread, not only for current.
4565 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4566 // It might work on 2.6.10+ with a special kernel/glibc patch.
4567 // For reference, please, see IEEE Std 1003.1-2004:
4568 // http://www.unix.org/single_unix_specification
4570 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4571 struct timespec tp;
4572 int rc = os::Linux::clock_gettime(clockid, &tp);
4573 assert(rc == 0, "clock_gettime is expected to return 0 code");
4575 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4576 }
4578 /////
4579 // glibc on Linux platform uses non-documented flag
4580 // to indicate, that some special sort of signal
4581 // trampoline is used.
4582 // We will never set this flag, and we should
4583 // ignore this flag in our diagnostic
4584 #ifdef SIGNIFICANT_SIGNAL_MASK
4585 #undef SIGNIFICANT_SIGNAL_MASK
4586 #endif
4587 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4589 static const char* get_signal_handler_name(address handler,
4590 char* buf, int buflen) {
4591 int offset = 0;
4592 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4593 if (found) {
4594 // skip directory names
4595 const char *p1, *p2;
4596 p1 = buf;
4597 size_t len = strlen(os::file_separator());
4598 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4599 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4600 } else {
4601 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4602 }
4603 return buf;
4604 }
4606 static void print_signal_handler(outputStream* st, int sig,
4607 char* buf, size_t buflen) {
4608 struct sigaction sa;
4610 sigaction(sig, NULL, &sa);
4612 // See comment for SIGNIFICANT_SIGNAL_MASK define
4613 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4615 st->print("%s: ", os::exception_name(sig, buf, buflen));
4617 address handler = (sa.sa_flags & SA_SIGINFO)
4618 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4619 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4621 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4622 st->print("SIG_DFL");
4623 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4624 st->print("SIG_IGN");
4625 } else {
4626 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4627 }
4629 st->print(", sa_mask[0]=");
4630 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4632 address rh = VMError::get_resetted_sighandler(sig);
4633 // May be, handler was resetted by VMError?
4634 if(rh != NULL) {
4635 handler = rh;
4636 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4637 }
4639 st->print(", sa_flags=");
4640 os::Posix::print_sa_flags(st, sa.sa_flags);
4642 // Check: is it our handler?
4643 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4644 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4645 // It is our signal handler
4646 // check for flags, reset system-used one!
4647 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4648 st->print(
4649 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4650 os::Linux::get_our_sigflags(sig));
4651 }
4652 }
4653 st->cr();
4654 }
4657 #define DO_SIGNAL_CHECK(sig) \
4658 if (!sigismember(&check_signal_done, sig)) \
4659 os::Linux::check_signal_handler(sig)
4661 // This method is a periodic task to check for misbehaving JNI applications
4662 // under CheckJNI, we can add any periodic checks here
4664 void os::run_periodic_checks() {
4666 if (check_signals == false) return;
4668 // SEGV and BUS if overridden could potentially prevent
4669 // generation of hs*.log in the event of a crash, debugging
4670 // such a case can be very challenging, so we absolutely
4671 // check the following for a good measure:
4672 DO_SIGNAL_CHECK(SIGSEGV);
4673 DO_SIGNAL_CHECK(SIGILL);
4674 DO_SIGNAL_CHECK(SIGFPE);
4675 DO_SIGNAL_CHECK(SIGBUS);
4676 DO_SIGNAL_CHECK(SIGPIPE);
4677 DO_SIGNAL_CHECK(SIGXFSZ);
4678 #if defined(PPC64)
4679 DO_SIGNAL_CHECK(SIGTRAP);
4680 #endif
4682 // ReduceSignalUsage allows the user to override these handlers
4683 // see comments at the very top and jvm_solaris.h
4684 if (!ReduceSignalUsage) {
4685 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4686 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4687 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4688 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4689 }
4691 DO_SIGNAL_CHECK(SR_signum);
4692 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4693 }
4695 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4697 static os_sigaction_t os_sigaction = NULL;
4699 void os::Linux::check_signal_handler(int sig) {
4700 char buf[O_BUFLEN];
4701 address jvmHandler = NULL;
4704 struct sigaction act;
4705 if (os_sigaction == NULL) {
4706 // only trust the default sigaction, in case it has been interposed
4707 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4708 if (os_sigaction == NULL) return;
4709 }
4711 os_sigaction(sig, (struct sigaction*)NULL, &act);
4714 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4716 address thisHandler = (act.sa_flags & SA_SIGINFO)
4717 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4718 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4721 switch(sig) {
4722 case SIGSEGV:
4723 case SIGBUS:
4724 case SIGFPE:
4725 case SIGPIPE:
4726 case SIGILL:
4727 case SIGXFSZ:
4728 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4729 break;
4731 case SHUTDOWN1_SIGNAL:
4732 case SHUTDOWN2_SIGNAL:
4733 case SHUTDOWN3_SIGNAL:
4734 case BREAK_SIGNAL:
4735 jvmHandler = (address)user_handler();
4736 break;
4738 case INTERRUPT_SIGNAL:
4739 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4740 break;
4742 default:
4743 if (sig == SR_signum) {
4744 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4745 } else {
4746 return;
4747 }
4748 break;
4749 }
4751 if (thisHandler != jvmHandler) {
4752 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4753 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4754 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4755 // No need to check this sig any longer
4756 sigaddset(&check_signal_done, sig);
4757 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4758 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4759 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4760 exception_name(sig, buf, O_BUFLEN));
4761 }
4762 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4763 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4764 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4765 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4766 // No need to check this sig any longer
4767 sigaddset(&check_signal_done, sig);
4768 }
4770 // Dump all the signal
4771 if (sigismember(&check_signal_done, sig)) {
4772 print_signal_handlers(tty, buf, O_BUFLEN);
4773 }
4774 }
4776 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4778 extern bool signal_name(int signo, char* buf, size_t len);
4780 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4781 if (0 < exception_code && exception_code <= SIGRTMAX) {
4782 // signal
4783 if (!signal_name(exception_code, buf, size)) {
4784 jio_snprintf(buf, size, "SIG%d", exception_code);
4785 }
4786 return buf;
4787 } else {
4788 return NULL;
4789 }
4790 }
4792 // this is called _before_ the most of global arguments have been parsed
4793 void os::init(void) {
4794 char dummy; /* used to get a guess on initial stack address */
4795 // first_hrtime = gethrtime();
4797 // With LinuxThreads the JavaMain thread pid (primordial thread)
4798 // is different than the pid of the java launcher thread.
4799 // So, on Linux, the launcher thread pid is passed to the VM
4800 // via the sun.java.launcher.pid property.
4801 // Use this property instead of getpid() if it was correctly passed.
4802 // See bug 6351349.
4803 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4805 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4807 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4809 init_random(1234567);
4811 ThreadCritical::initialize();
4813 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4814 if (Linux::page_size() == -1) {
4815 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4816 strerror(errno)));
4817 }
4818 init_page_sizes((size_t) Linux::page_size());
4820 Linux::initialize_system_info();
4822 // main_thread points to the aboriginal thread
4823 Linux::_main_thread = pthread_self();
4825 Linux::clock_init();
4826 initial_time_count = javaTimeNanos();
4828 // pthread_condattr initialization for monotonic clock
4829 int status;
4830 pthread_condattr_t* _condattr = os::Linux::condAttr();
4831 if ((status = pthread_condattr_init(_condattr)) != 0) {
4832 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4833 }
4834 // Only set the clock if CLOCK_MONOTONIC is available
4835 if (Linux::supports_monotonic_clock()) {
4836 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4837 if (status == EINVAL) {
4838 warning("Unable to use monotonic clock with relative timed-waits" \
4839 " - changes to the time-of-day clock may have adverse affects");
4840 } else {
4841 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4842 }
4843 }
4844 }
4845 // else it defaults to CLOCK_REALTIME
4847 pthread_mutex_init(&dl_mutex, NULL);
4849 // If the pagesize of the VM is greater than 8K determine the appropriate
4850 // number of initial guard pages. The user can change this with the
4851 // command line arguments, if needed.
4852 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4853 StackYellowPages = 1;
4854 StackRedPages = 1;
4855 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4856 }
4857 }
4859 // To install functions for atexit system call
4860 extern "C" {
4861 static void perfMemory_exit_helper() {
4862 perfMemory_exit();
4863 }
4864 }
4866 // this is called _after_ the global arguments have been parsed
4867 jint os::init_2(void)
4868 {
4869 Linux::fast_thread_clock_init();
4871 // Allocate a single page and mark it as readable for safepoint polling
4872 #ifdef OPT_SAFEPOINT
4873 void * p = (void *)(0x10000);
4874 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4875 #else
4876 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4877 #endif
4878 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4880 os::set_polling_page( polling_page );
4882 #ifndef PRODUCT
4883 if(Verbose && PrintMiscellaneous)
4884 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4885 #endif
4887 if (!UseMembar) {
4888 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4889 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4890 os::set_memory_serialize_page( mem_serialize_page );
4892 #ifndef PRODUCT
4893 if(Verbose && PrintMiscellaneous)
4894 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4895 #endif
4896 }
4898 // initialize suspend/resume support - must do this before signal_sets_init()
4899 if (SR_initialize() != 0) {
4900 perror("SR_initialize failed");
4901 return JNI_ERR;
4902 }
4904 Linux::signal_sets_init();
4905 Linux::install_signal_handlers();
4907 // Check minimum allowable stack size for thread creation and to initialize
4908 // the java system classes, including StackOverflowError - depends on page
4909 // size. Add a page for compiler2 recursion in main thread.
4910 // Add in 2*BytesPerWord times page size to account for VM stack during
4911 // class initialization depending on 32 or 64 bit VM.
4913 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
4914 * TongWeb cannot run with -Xss192K.
4915 * We are not sure whether this causes errors, so simply print a warning. */
4916 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
4917 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4918 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4919 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4921 size_t threadStackSizeInBytes = ThreadStackSize * K;
4922 if (threadStackSizeInBytes != 0 &&
4923 threadStackSizeInBytes < min_stack_allowed_jdk6) {
4924 tty->print_cr("\nThe stack size specified is too small, "
4925 "Specify at least %dk",
4926 os::Linux::min_stack_allowed/ K);
4927 return JNI_ERR;
4928 }
4930 // Make the stack size a multiple of the page size so that
4931 // the yellow/red zones can be guarded.
4932 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4933 vm_page_size()));
4935 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4937 #if defined(IA32)
4938 workaround_expand_exec_shield_cs_limit();
4939 #endif
4941 Linux::libpthread_init();
4942 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4943 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4944 Linux::glibc_version(), Linux::libpthread_version(),
4945 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4946 }
4948 if (UseNUMA) {
4949 if (!Linux::libnuma_init()) {
4950 UseNUMA = false;
4951 } else {
4952 if ((Linux::numa_max_node() < 1)) {
4953 // There's only one node(they start from 0), disable NUMA.
4954 UseNUMA = false;
4955 }
4956 }
4957 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4958 // we can make the adaptive lgrp chunk resizing work. If the user specified
4959 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4960 // disable adaptive resizing.
4961 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4962 if (FLAG_IS_DEFAULT(UseNUMA)) {
4963 UseNUMA = false;
4964 } else {
4965 if (FLAG_IS_DEFAULT(UseLargePages) &&
4966 FLAG_IS_DEFAULT(UseSHM) &&
4967 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4968 UseLargePages = false;
4969 } else {
4970 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4971 UseAdaptiveSizePolicy = false;
4972 UseAdaptiveNUMAChunkSizing = false;
4973 }
4974 }
4975 }
4976 if (!UseNUMA && ForceNUMA) {
4977 UseNUMA = true;
4978 }
4979 }
4981 if (MaxFDLimit) {
4982 // set the number of file descriptors to max. print out error
4983 // if getrlimit/setrlimit fails but continue regardless.
4984 struct rlimit nbr_files;
4985 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4986 if (status != 0) {
4987 if (PrintMiscellaneous && (Verbose || WizardMode))
4988 perror("os::init_2 getrlimit failed");
4989 } else {
4990 nbr_files.rlim_cur = nbr_files.rlim_max;
4991 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4992 if (status != 0) {
4993 if (PrintMiscellaneous && (Verbose || WizardMode))
4994 perror("os::init_2 setrlimit failed");
4995 }
4996 }
4997 }
4999 // Initialize lock used to serialize thread creation (see os::create_thread)
5000 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5002 // at-exit methods are called in the reverse order of their registration.
5003 // atexit functions are called on return from main or as a result of a
5004 // call to exit(3C). There can be only 32 of these functions registered
5005 // and atexit() does not set errno.
5007 if (PerfAllowAtExitRegistration) {
5008 // only register atexit functions if PerfAllowAtExitRegistration is set.
5009 // atexit functions can be delayed until process exit time, which
5010 // can be problematic for embedded VM situations. Embedded VMs should
5011 // call DestroyJavaVM() to assure that VM resources are released.
5013 // note: perfMemory_exit_helper atexit function may be removed in
5014 // the future if the appropriate cleanup code can be added to the
5015 // VM_Exit VMOperation's doit method.
5016 if (atexit(perfMemory_exit_helper) != 0) {
5017 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5018 }
5019 }
5021 // initialize thread priority policy
5022 prio_init();
5024 return JNI_OK;
5025 }
5027 // Mark the polling page as unreadable
5028 void os::make_polling_page_unreadable(void) {
5029 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5030 fatal("Could not disable polling page");
5031 };
5033 // Mark the polling page as readable
5034 void os::make_polling_page_readable(void) {
5035 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5036 fatal("Could not enable polling page");
5037 }
5038 };
5040 #ifdef MIPS64
5041 /* 2016/9/18 Jin
5042 * Refer to: libnuma_init() */
5043 int get_available_cpus() {
5044 typedef int (*numa_num_task_cpus_t)(void);
5045 static numa_num_task_cpus_t _numa_num_task_cpus;
5047 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
5049 if (handle == NULL)
5050 return sysconf(_SC_NPROCESSORS_ONLN);
5052 _numa_num_task_cpus = CAST_FROM_FN_PTR(numa_num_task_cpus_t, dlsym(handle, "numa_num_task_cpus"));
5054 if (_numa_num_task_cpus == NULL) {
5055 dlclose(handle);
5056 return sysconf(_SC_NPROCESSORS_ONLN);
5057 }
5059 int ret = _numa_num_task_cpus();
5060 dlclose(handle);
5061 return ret;
5062 }
5063 #endif
5065 int os::active_processor_count() {
5066 // Linux doesn't yet have a (official) notion of processor sets,
5067 // so just return the number of online processors.
5068 #ifdef MIPS64
5069 int online_cpus = get_available_cpus();
5070 #else
5071 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5072 #endif
5073 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5074 return online_cpus;
5075 }
5077 void os::set_native_thread_name(const char *name) {
5078 // Not yet implemented.
5079 return;
5080 }
5082 bool os::distribute_processes(uint length, uint* distribution) {
5083 // Not yet implemented.
5084 return false;
5085 }
5087 bool os::bind_to_processor(uint processor_id) {
5088 // Not yet implemented.
5089 return false;
5090 }
5092 ///
5094 void os::SuspendedThreadTask::internal_do_task() {
5095 if (do_suspend(_thread->osthread())) {
5096 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5097 do_task(context);
5098 do_resume(_thread->osthread());
5099 }
5100 }
5102 class PcFetcher : public os::SuspendedThreadTask {
5103 public:
5104 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5105 ExtendedPC result();
5106 protected:
5107 void do_task(const os::SuspendedThreadTaskContext& context);
5108 private:
5109 ExtendedPC _epc;
5110 };
5112 ExtendedPC PcFetcher::result() {
5113 guarantee(is_done(), "task is not done yet.");
5114 return _epc;
5115 }
5117 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5118 Thread* thread = context.thread();
5119 OSThread* osthread = thread->osthread();
5120 if (osthread->ucontext() != NULL) {
5121 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5122 } else {
5123 // NULL context is unexpected, double-check this is the VMThread
5124 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5125 }
5126 }
5128 // Suspends the target using the signal mechanism and then grabs the PC before
5129 // resuming the target. Used by the flat-profiler only
5130 ExtendedPC os::get_thread_pc(Thread* thread) {
5131 // Make sure that it is called by the watcher for the VMThread
5132 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5133 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5135 PcFetcher fetcher(thread);
5136 fetcher.run();
5137 return fetcher.result();
5138 }
5140 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5141 {
5142 if (is_NPTL()) {
5143 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5144 } else {
5145 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5146 // word back to default 64bit precision if condvar is signaled. Java
5147 // wants 53bit precision. Save and restore current value.
5148 int fpu = get_fpu_control_word();
5149 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5150 set_fpu_control_word(fpu);
5151 return status;
5152 }
5153 }
5155 ////////////////////////////////////////////////////////////////////////////////
5156 // debug support
5158 bool os::find(address addr, outputStream* st) {
5159 Dl_info dlinfo;
5160 memset(&dlinfo, 0, sizeof(dlinfo));
5161 if (dladdr(addr, &dlinfo) != 0) {
5162 st->print(PTR_FORMAT ": ", addr);
5163 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5164 st->print("%s+%#x", dlinfo.dli_sname,
5165 addr - (intptr_t)dlinfo.dli_saddr);
5166 } else if (dlinfo.dli_fbase != NULL) {
5167 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5168 } else {
5169 st->print("<absolute address>");
5170 }
5171 if (dlinfo.dli_fname != NULL) {
5172 st->print(" in %s", dlinfo.dli_fname);
5173 }
5174 if (dlinfo.dli_fbase != NULL) {
5175 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5176 }
5177 st->cr();
5179 if (Verbose) {
5180 // decode some bytes around the PC
5181 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5182 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5183 address lowest = (address) dlinfo.dli_sname;
5184 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5185 if (begin < lowest) begin = lowest;
5186 Dl_info dlinfo2;
5187 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5188 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5189 end = (address) dlinfo2.dli_saddr;
5190 Disassembler::decode(begin, end, st);
5191 }
5192 return true;
5193 }
5194 return false;
5195 }
5197 ////////////////////////////////////////////////////////////////////////////////
5198 // misc
5200 // This does not do anything on Linux. This is basically a hook for being
5201 // able to use structured exception handling (thread-local exception filters)
5202 // on, e.g., Win32.
5203 void
5204 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5205 JavaCallArguments* args, Thread* thread) {
5206 f(value, method, args, thread);
5207 }
5209 void os::print_statistics() {
5210 }
5212 int os::message_box(const char* title, const char* message) {
5213 int i;
5214 fdStream err(defaultStream::error_fd());
5215 for (i = 0; i < 78; i++) err.print_raw("=");
5216 err.cr();
5217 err.print_raw_cr(title);
5218 for (i = 0; i < 78; i++) err.print_raw("-");
5219 err.cr();
5220 err.print_raw_cr(message);
5221 for (i = 0; i < 78; i++) err.print_raw("=");
5222 err.cr();
5224 char buf[16];
5225 // Prevent process from exiting upon "read error" without consuming all CPU
5226 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5228 return buf[0] == 'y' || buf[0] == 'Y';
5229 }
5231 int os::stat(const char *path, struct stat *sbuf) {
5232 char pathbuf[MAX_PATH];
5233 if (strlen(path) > MAX_PATH - 1) {
5234 errno = ENAMETOOLONG;
5235 return -1;
5236 }
5237 os::native_path(strcpy(pathbuf, path));
5238 return ::stat(pathbuf, sbuf);
5239 }
5241 bool os::check_heap(bool force) {
5242 return true;
5243 }
5245 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5246 return ::vsnprintf(buf, count, format, args);
5247 }
5249 // Is a (classpath) directory empty?
5250 bool os::dir_is_empty(const char* path) {
5251 DIR *dir = NULL;
5252 struct dirent *ptr;
5254 dir = opendir(path);
5255 if (dir == NULL) return true;
5257 /* Scan the directory */
5258 bool result = true;
5259 char buf[sizeof(struct dirent) + MAX_PATH];
5260 while (result && (ptr = ::readdir(dir)) != NULL) {
5261 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5262 result = false;
5263 }
5264 }
5265 closedir(dir);
5266 return result;
5267 }
5269 // This code originates from JDK's sysOpen and open64_w
5270 // from src/solaris/hpi/src/system_md.c
5272 #ifndef O_DELETE
5273 #define O_DELETE 0x10000
5274 #endif
5276 // Open a file. Unlink the file immediately after open returns
5277 // if the specified oflag has the O_DELETE flag set.
5278 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5280 int os::open(const char *path, int oflag, int mode) {
5282 if (strlen(path) > MAX_PATH - 1) {
5283 errno = ENAMETOOLONG;
5284 return -1;
5285 }
5286 int fd;
5287 int o_delete = (oflag & O_DELETE);
5288 oflag = oflag & ~O_DELETE;
5290 fd = ::open64(path, oflag, mode);
5291 if (fd == -1) return -1;
5293 //If the open succeeded, the file might still be a directory
5294 {
5295 struct stat64 buf64;
5296 int ret = ::fstat64(fd, &buf64);
5297 int st_mode = buf64.st_mode;
5299 if (ret != -1) {
5300 if ((st_mode & S_IFMT) == S_IFDIR) {
5301 errno = EISDIR;
5302 ::close(fd);
5303 return -1;
5304 }
5305 } else {
5306 ::close(fd);
5307 return -1;
5308 }
5309 }
5311 /*
5312 * All file descriptors that are opened in the JVM and not
5313 * specifically destined for a subprocess should have the
5314 * close-on-exec flag set. If we don't set it, then careless 3rd
5315 * party native code might fork and exec without closing all
5316 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5317 * UNIXProcess.c), and this in turn might:
5318 *
5319 * - cause end-of-file to fail to be detected on some file
5320 * descriptors, resulting in mysterious hangs, or
5321 *
5322 * - might cause an fopen in the subprocess to fail on a system
5323 * suffering from bug 1085341.
5324 *
5325 * (Yes, the default setting of the close-on-exec flag is a Unix
5326 * design flaw)
5327 *
5328 * See:
5329 * 1085341: 32-bit stdio routines should support file descriptors >255
5330 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5331 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5332 */
5333 #ifdef FD_CLOEXEC
5334 {
5335 int flags = ::fcntl(fd, F_GETFD);
5336 if (flags != -1)
5337 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5338 }
5339 #endif
5341 if (o_delete != 0) {
5342 ::unlink(path);
5343 }
5344 return fd;
5345 }
5348 // create binary file, rewriting existing file if required
5349 int os::create_binary_file(const char* path, bool rewrite_existing) {
5350 int oflags = O_WRONLY | O_CREAT;
5351 if (!rewrite_existing) {
5352 oflags |= O_EXCL;
5353 }
5354 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5355 }
5357 // return current position of file pointer
5358 jlong os::current_file_offset(int fd) {
5359 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5360 }
5362 // move file pointer to the specified offset
5363 jlong os::seek_to_file_offset(int fd, jlong offset) {
5364 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5365 }
5367 // This code originates from JDK's sysAvailable
5368 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5370 int os::available(int fd, jlong *bytes) {
5371 jlong cur, end;
5372 int mode;
5373 struct stat64 buf64;
5375 if (::fstat64(fd, &buf64) >= 0) {
5376 mode = buf64.st_mode;
5377 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5378 /*
5379 * XXX: is the following call interruptible? If so, this might
5380 * need to go through the INTERRUPT_IO() wrapper as for other
5381 * blocking, interruptible calls in this file.
5382 */
5383 int n;
5384 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5385 *bytes = n;
5386 return 1;
5387 }
5388 }
5389 }
5390 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5391 return 0;
5392 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5393 return 0;
5394 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5395 return 0;
5396 }
5397 *bytes = end - cur;
5398 return 1;
5399 }
5401 int os::socket_available(int fd, jint *pbytes) {
5402 // Linux doc says EINTR not returned, unlike Solaris
5403 int ret = ::ioctl(fd, FIONREAD, pbytes);
5405 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5406 // is expected to return 0 on failure and 1 on success to the jdk.
5407 return (ret < 0) ? 0 : 1;
5408 }
5410 // Map a block of memory.
5411 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5412 char *addr, size_t bytes, bool read_only,
5413 bool allow_exec) {
5414 int prot;
5415 int flags = MAP_PRIVATE;
5417 if (read_only) {
5418 prot = PROT_READ;
5419 } else {
5420 prot = PROT_READ | PROT_WRITE;
5421 }
5423 if (allow_exec) {
5424 prot |= PROT_EXEC;
5425 }
5427 if (addr != NULL) {
5428 flags |= MAP_FIXED;
5429 }
5431 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5432 fd, file_offset);
5433 if (mapped_address == MAP_FAILED) {
5434 return NULL;
5435 }
5436 return mapped_address;
5437 }
5440 // Remap a block of memory.
5441 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5442 char *addr, size_t bytes, bool read_only,
5443 bool allow_exec) {
5444 // same as map_memory() on this OS
5445 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5446 allow_exec);
5447 }
5450 // Unmap a block of memory.
5451 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5452 return munmap(addr, bytes) == 0;
5453 }
5455 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5457 static clockid_t thread_cpu_clockid(Thread* thread) {
5458 pthread_t tid = thread->osthread()->pthread_id();
5459 clockid_t clockid;
5461 // Get thread clockid
5462 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5463 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5464 return clockid;
5465 }
5467 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5468 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5469 // of a thread.
5470 //
5471 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5472 // the fast estimate available on the platform.
5474 jlong os::current_thread_cpu_time() {
5475 if (os::Linux::supports_fast_thread_cpu_time()) {
5476 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5477 } else {
5478 // return user + sys since the cost is the same
5479 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5480 }
5481 }
5483 jlong os::thread_cpu_time(Thread* thread) {
5484 // consistent with what current_thread_cpu_time() returns
5485 if (os::Linux::supports_fast_thread_cpu_time()) {
5486 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5487 } else {
5488 return slow_thread_cpu_time(thread, true /* user + sys */);
5489 }
5490 }
5492 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5493 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5494 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5495 } else {
5496 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5497 }
5498 }
5500 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5501 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5502 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5503 } else {
5504 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5505 }
5506 }
5508 //
5509 // -1 on error.
5510 //
5512 PRAGMA_DIAG_PUSH
5513 PRAGMA_FORMAT_NONLITERAL_IGNORED
5514 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5515 static bool proc_task_unchecked = true;
5516 static const char *proc_stat_path = "/proc/%d/stat";
5517 pid_t tid = thread->osthread()->thread_id();
5518 char *s;
5519 char stat[2048];
5520 int statlen;
5521 char proc_name[64];
5522 int count;
5523 long sys_time, user_time;
5524 char cdummy;
5525 int idummy;
5526 long ldummy;
5527 FILE *fp;
5529 // The /proc/<tid>/stat aggregates per-process usage on
5530 // new Linux kernels 2.6+ where NPTL is supported.
5531 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5532 // See bug 6328462.
5533 // There possibly can be cases where there is no directory
5534 // /proc/self/task, so we check its availability.
5535 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5536 // This is executed only once
5537 proc_task_unchecked = false;
5538 fp = fopen("/proc/self/task", "r");
5539 if (fp != NULL) {
5540 proc_stat_path = "/proc/self/task/%d/stat";
5541 fclose(fp);
5542 }
5543 }
5545 sprintf(proc_name, proc_stat_path, tid);
5546 fp = fopen(proc_name, "r");
5547 if ( fp == NULL ) return -1;
5548 statlen = fread(stat, 1, 2047, fp);
5549 stat[statlen] = '\0';
5550 fclose(fp);
5552 // Skip pid and the command string. Note that we could be dealing with
5553 // weird command names, e.g. user could decide to rename java launcher
5554 // to "java 1.4.2 :)", then the stat file would look like
5555 // 1234 (java 1.4.2 :)) R ... ...
5556 // We don't really need to know the command string, just find the last
5557 // occurrence of ")" and then start parsing from there. See bug 4726580.
5558 s = strrchr(stat, ')');
5559 if (s == NULL ) return -1;
5561 // Skip blank chars
5562 do s++; while (isspace(*s));
5564 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5565 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5566 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5567 &user_time, &sys_time);
5568 if ( count != 13 ) return -1;
5569 if (user_sys_cpu_time) {
5570 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5571 } else {
5572 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5573 }
5574 }
5575 PRAGMA_DIAG_POP
5577 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5578 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5579 info_ptr->may_skip_backward = false; // elapsed time not wall time
5580 info_ptr->may_skip_forward = false; // elapsed time not wall time
5581 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5582 }
5584 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5585 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5586 info_ptr->may_skip_backward = false; // elapsed time not wall time
5587 info_ptr->may_skip_forward = false; // elapsed time not wall time
5588 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5589 }
5591 bool os::is_thread_cpu_time_supported() {
5592 return true;
5593 }
5595 // System loadavg support. Returns -1 if load average cannot be obtained.
5596 // Linux doesn't yet have a (official) notion of processor sets,
5597 // so just return the system wide load average.
5598 int os::loadavg(double loadavg[], int nelem) {
5599 return ::getloadavg(loadavg, nelem);
5600 }
5602 void os::pause() {
5603 char filename[MAX_PATH];
5604 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5605 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5606 } else {
5607 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5608 }
5610 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5611 if (fd != -1) {
5612 struct stat buf;
5613 ::close(fd);
5614 while (::stat(filename, &buf) == 0) {
5615 (void)::poll(NULL, 0, 100);
5616 }
5617 } else {
5618 jio_fprintf(stderr,
5619 "Could not open pause file '%s', continuing immediately.\n", filename);
5620 }
5621 }
5624 // Refer to the comments in os_solaris.cpp park-unpark.
5625 //
5626 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5627 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5628 // For specifics regarding the bug see GLIBC BUGID 261237 :
5629 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5630 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5631 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5632 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5633 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5634 // and monitorenter when we're using 1-0 locking. All those operations may result in
5635 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5636 // of libpthread avoids the problem, but isn't practical.
5637 //
5638 // Possible remedies:
5639 //
5640 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5641 // This is palliative and probabilistic, however. If the thread is preempted
5642 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5643 // than the minimum period may have passed, and the abstime may be stale (in the
5644 // past) resultin in a hang. Using this technique reduces the odds of a hang
5645 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5646 //
5647 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5648 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5649 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5650 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5651 // thread.
5652 //
5653 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5654 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5655 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5656 // This also works well. In fact it avoids kernel-level scalability impediments
5657 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5658 // timers in a graceful fashion.
5659 //
5660 // 4. When the abstime value is in the past it appears that control returns
5661 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5662 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5663 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5664 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5665 // It may be possible to avoid reinitialization by checking the return
5666 // value from pthread_cond_timedwait(). In addition to reinitializing the
5667 // condvar we must establish the invariant that cond_signal() is only called
5668 // within critical sections protected by the adjunct mutex. This prevents
5669 // cond_signal() from "seeing" a condvar that's in the midst of being
5670 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5671 // desirable signal-after-unlock optimization that avoids futile context switching.
5672 //
5673 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5674 // structure when a condvar is used or initialized. cond_destroy() would
5675 // release the helper structure. Our reinitialize-after-timedwait fix
5676 // put excessive stress on malloc/free and locks protecting the c-heap.
5677 //
5678 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5679 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5680 // and only enabling the work-around for vulnerable environments.
5682 // utility to compute the abstime argument to timedwait:
5683 // millis is the relative timeout time
5684 // abstime will be the absolute timeout time
5685 // TODO: replace compute_abstime() with unpackTime()
5687 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5688 if (millis < 0) millis = 0;
5690 jlong seconds = millis / 1000;
5691 millis %= 1000;
5692 if (seconds > 50000000) { // see man cond_timedwait(3T)
5693 seconds = 50000000;
5694 }
5696 if (os::Linux::supports_monotonic_clock()) {
5697 struct timespec now;
5698 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5699 assert_status(status == 0, status, "clock_gettime");
5700 abstime->tv_sec = now.tv_sec + seconds;
5701 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5702 if (nanos >= NANOSECS_PER_SEC) {
5703 abstime->tv_sec += 1;
5704 nanos -= NANOSECS_PER_SEC;
5705 }
5706 abstime->tv_nsec = nanos;
5707 } else {
5708 struct timeval now;
5709 int status = gettimeofday(&now, NULL);
5710 assert(status == 0, "gettimeofday");
5711 abstime->tv_sec = now.tv_sec + seconds;
5712 long usec = now.tv_usec + millis * 1000;
5713 if (usec >= 1000000) {
5714 abstime->tv_sec += 1;
5715 usec -= 1000000;
5716 }
5717 abstime->tv_nsec = usec * 1000;
5718 }
5719 return abstime;
5720 }
5723 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5724 // Conceptually TryPark() should be equivalent to park(0).
5726 int os::PlatformEvent::TryPark() {
5727 for (;;) {
5728 const int v = _Event ;
5729 guarantee ((v == 0) || (v == 1), "invariant") ;
5730 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5731 }
5732 }
5734 void os::PlatformEvent::park() { // AKA "down()"
5735 // Invariant: Only the thread associated with the Event/PlatformEvent
5736 // may call park().
5737 // TODO: assert that _Assoc != NULL or _Assoc == Self
5738 int v ;
5739 for (;;) {
5740 v = _Event ;
5741 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5742 }
5743 guarantee (v >= 0, "invariant") ;
5744 if (v == 0) {
5745 // Do this the hard way by blocking ...
5746 int status = pthread_mutex_lock(_mutex);
5747 assert_status(status == 0, status, "mutex_lock");
5748 guarantee (_nParked == 0, "invariant") ;
5749 ++ _nParked ;
5750 while (_Event < 0) {
5751 status = pthread_cond_wait(_cond, _mutex);
5752 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5753 // Treat this the same as if the wait was interrupted
5754 if (status == ETIME) { status = EINTR; }
5755 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5756 }
5757 -- _nParked ;
5759 _Event = 0 ;
5760 status = pthread_mutex_unlock(_mutex);
5761 assert_status(status == 0, status, "mutex_unlock");
5762 // Paranoia to ensure our locked and lock-free paths interact
5763 // correctly with each other.
5764 OrderAccess::fence();
5765 }
5766 guarantee (_Event >= 0, "invariant") ;
5767 }
5769 int os::PlatformEvent::park(jlong millis) {
5770 guarantee (_nParked == 0, "invariant") ;
5772 int v ;
5773 for (;;) {
5774 v = _Event ;
5775 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5776 }
5777 guarantee (v >= 0, "invariant") ;
5778 if (v != 0) return OS_OK ;
5780 // We do this the hard way, by blocking the thread.
5781 // Consider enforcing a minimum timeout value.
5782 struct timespec abst;
5783 compute_abstime(&abst, millis);
5785 int ret = OS_TIMEOUT;
5786 int status = pthread_mutex_lock(_mutex);
5787 assert_status(status == 0, status, "mutex_lock");
5788 guarantee (_nParked == 0, "invariant") ;
5789 ++_nParked ;
5791 // Object.wait(timo) will return because of
5792 // (a) notification
5793 // (b) timeout
5794 // (c) thread.interrupt
5795 //
5796 // Thread.interrupt and object.notify{All} both call Event::set.
5797 // That is, we treat thread.interrupt as a special case of notification.
5798 // The underlying Solaris implementation, cond_timedwait, admits
5799 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5800 // JVM from making those visible to Java code. As such, we must
5801 // filter out spurious wakeups. We assume all ETIME returns are valid.
5802 //
5803 // TODO: properly differentiate simultaneous notify+interrupt.
5804 // In that case, we should propagate the notify to another waiter.
5806 while (_Event < 0) {
5807 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5808 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5809 pthread_cond_destroy (_cond);
5810 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5811 }
5812 assert_status(status == 0 || status == EINTR ||
5813 status == ETIME || status == ETIMEDOUT,
5814 status, "cond_timedwait");
5815 if (!FilterSpuriousWakeups) break ; // previous semantics
5816 if (status == ETIME || status == ETIMEDOUT) break ;
5817 // We consume and ignore EINTR and spurious wakeups.
5818 }
5819 --_nParked ;
5820 if (_Event >= 0) {
5821 ret = OS_OK;
5822 }
5823 _Event = 0 ;
5824 status = pthread_mutex_unlock(_mutex);
5825 assert_status(status == 0, status, "mutex_unlock");
5826 assert (_nParked == 0, "invariant") ;
5827 // Paranoia to ensure our locked and lock-free paths interact
5828 // correctly with each other.
5829 OrderAccess::fence();
5830 return ret;
5831 }
5833 void os::PlatformEvent::unpark() {
5834 // Transitions for _Event:
5835 // 0 :=> 1
5836 // 1 :=> 1
5837 // -1 :=> either 0 or 1; must signal target thread
5838 // That is, we can safely transition _Event from -1 to either
5839 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5840 // unpark() calls.
5841 // See also: "Semaphores in Plan 9" by Mullender & Cox
5842 //
5843 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5844 // that it will take two back-to-back park() calls for the owning
5845 // thread to block. This has the benefit of forcing a spurious return
5846 // from the first park() call after an unpark() call which will help
5847 // shake out uses of park() and unpark() without condition variables.
5849 if (Atomic::xchg(1, &_Event) >= 0) return;
5851 // Wait for the thread associated with the event to vacate
5852 int status = pthread_mutex_lock(_mutex);
5853 assert_status(status == 0, status, "mutex_lock");
5854 int AnyWaiters = _nParked;
5855 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5856 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5857 AnyWaiters = 0;
5858 pthread_cond_signal(_cond);
5859 }
5860 status = pthread_mutex_unlock(_mutex);
5861 assert_status(status == 0, status, "mutex_unlock");
5862 if (AnyWaiters != 0) {
5863 status = pthread_cond_signal(_cond);
5864 assert_status(status == 0, status, "cond_signal");
5865 }
5867 // Note that we signal() _after dropping the lock for "immortal" Events.
5868 // This is safe and avoids a common class of futile wakeups. In rare
5869 // circumstances this can cause a thread to return prematurely from
5870 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5871 // simply re-test the condition and re-park itself.
5872 }
5875 // JSR166
5876 // -------------------------------------------------------
5878 /*
5879 * The solaris and linux implementations of park/unpark are fairly
5880 * conservative for now, but can be improved. They currently use a
5881 * mutex/condvar pair, plus a a count.
5882 * Park decrements count if > 0, else does a condvar wait. Unpark
5883 * sets count to 1 and signals condvar. Only one thread ever waits
5884 * on the condvar. Contention seen when trying to park implies that someone
5885 * is unparking you, so don't wait. And spurious returns are fine, so there
5886 * is no need to track notifications.
5887 */
5889 /*
5890 * This code is common to linux and solaris and will be moved to a
5891 * common place in dolphin.
5892 *
5893 * The passed in time value is either a relative time in nanoseconds
5894 * or an absolute time in milliseconds. Either way it has to be unpacked
5895 * into suitable seconds and nanoseconds components and stored in the
5896 * given timespec structure.
5897 * Given time is a 64-bit value and the time_t used in the timespec is only
5898 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5899 * overflow if times way in the future are given. Further on Solaris versions
5900 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5901 * number of seconds, in abstime, is less than current_time + 100,000,000.
5902 * As it will be 28 years before "now + 100000000" will overflow we can
5903 * ignore overflow and just impose a hard-limit on seconds using the value
5904 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5905 * years from "now".
5906 */
5908 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5909 assert (time > 0, "convertTime");
5910 time_t max_secs = 0;
5912 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5913 struct timeval now;
5914 int status = gettimeofday(&now, NULL);
5915 assert(status == 0, "gettimeofday");
5917 max_secs = now.tv_sec + MAX_SECS;
5919 if (isAbsolute) {
5920 jlong secs = time / 1000;
5921 if (secs > max_secs) {
5922 absTime->tv_sec = max_secs;
5923 } else {
5924 absTime->tv_sec = secs;
5925 }
5926 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5927 } else {
5928 jlong secs = time / NANOSECS_PER_SEC;
5929 if (secs >= MAX_SECS) {
5930 absTime->tv_sec = max_secs;
5931 absTime->tv_nsec = 0;
5932 } else {
5933 absTime->tv_sec = now.tv_sec + secs;
5934 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5935 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5936 absTime->tv_nsec -= NANOSECS_PER_SEC;
5937 ++absTime->tv_sec; // note: this must be <= max_secs
5938 }
5939 }
5940 }
5941 } else {
5942 // must be relative using monotonic clock
5943 struct timespec now;
5944 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5945 assert_status(status == 0, status, "clock_gettime");
5946 max_secs = now.tv_sec + MAX_SECS;
5947 jlong secs = time / NANOSECS_PER_SEC;
5948 if (secs >= MAX_SECS) {
5949 absTime->tv_sec = max_secs;
5950 absTime->tv_nsec = 0;
5951 } else {
5952 absTime->tv_sec = now.tv_sec + secs;
5953 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5954 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5955 absTime->tv_nsec -= NANOSECS_PER_SEC;
5956 ++absTime->tv_sec; // note: this must be <= max_secs
5957 }
5958 }
5959 }
5960 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5961 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5962 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5963 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5964 }
5966 void Parker::park(bool isAbsolute, jlong time) {
5967 // Ideally we'd do something useful while spinning, such
5968 // as calling unpackTime().
5970 // Optional fast-path check:
5971 // Return immediately if a permit is available.
5972 // We depend on Atomic::xchg() having full barrier semantics
5973 // since we are doing a lock-free update to _counter.
5974 if (Atomic::xchg(0, &_counter) > 0) return;
5976 Thread* thread = Thread::current();
5977 assert(thread->is_Java_thread(), "Must be JavaThread");
5978 JavaThread *jt = (JavaThread *)thread;
5980 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5981 // Check interrupt before trying to wait
5982 if (Thread::is_interrupted(thread, false)) {
5983 return;
5984 }
5986 // Next, demultiplex/decode time arguments
5987 timespec absTime;
5988 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5989 return;
5990 }
5991 if (time > 0) {
5992 unpackTime(&absTime, isAbsolute, time);
5993 }
5996 // Enter safepoint region
5997 // Beware of deadlocks such as 6317397.
5998 // The per-thread Parker:: mutex is a classic leaf-lock.
5999 // In particular a thread must never block on the Threads_lock while
6000 // holding the Parker:: mutex. If safepoints are pending both the
6001 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6002 ThreadBlockInVM tbivm(jt);
6004 // Don't wait if cannot get lock since interference arises from
6005 // unblocking. Also. check interrupt before trying wait
6006 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6007 return;
6008 }
6010 int status ;
6011 if (_counter > 0) { // no wait needed
6012 _counter = 0;
6013 status = pthread_mutex_unlock(_mutex);
6014 assert (status == 0, "invariant") ;
6015 // Paranoia to ensure our locked and lock-free paths interact
6016 // correctly with each other and Java-level accesses.
6017 OrderAccess::fence();
6018 return;
6019 }
6021 #ifdef ASSERT
6022 // Don't catch signals while blocked; let the running threads have the signals.
6023 // (This allows a debugger to break into the running thread.)
6024 sigset_t oldsigs;
6025 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6026 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6027 #endif
6029 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6030 jt->set_suspend_equivalent();
6031 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6033 assert(_cur_index == -1, "invariant");
6034 if (time == 0) {
6035 _cur_index = REL_INDEX; // arbitrary choice when not timed
6036 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6037 } else {
6038 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6039 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6040 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6041 pthread_cond_destroy (&_cond[_cur_index]) ;
6042 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6043 }
6044 }
6045 _cur_index = -1;
6046 assert_status(status == 0 || status == EINTR ||
6047 status == ETIME || status == ETIMEDOUT,
6048 status, "cond_timedwait");
6050 #ifdef ASSERT
6051 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6052 #endif
6054 _counter = 0 ;
6055 status = pthread_mutex_unlock(_mutex) ;
6056 assert_status(status == 0, status, "invariant") ;
6057 // Paranoia to ensure our locked and lock-free paths interact
6058 // correctly with each other and Java-level accesses.
6059 OrderAccess::fence();
6061 // If externally suspended while waiting, re-suspend
6062 if (jt->handle_special_suspend_equivalent_condition()) {
6063 jt->java_suspend_self();
6064 }
6065 }
6067 void Parker::unpark() {
6068 int s, status ;
6069 status = pthread_mutex_lock(_mutex);
6070 assert (status == 0, "invariant") ;
6071 s = _counter;
6072 _counter = 1;
6073 if (s < 1) {
6074 // thread might be parked
6075 if (_cur_index != -1) {
6076 // thread is definitely parked
6077 if (WorkAroundNPTLTimedWaitHang) {
6078 status = pthread_cond_signal (&_cond[_cur_index]);
6079 assert (status == 0, "invariant");
6080 status = pthread_mutex_unlock(_mutex);
6081 assert (status == 0, "invariant");
6082 } else {
6083 // must capture correct index before unlocking
6084 int index = _cur_index;
6085 status = pthread_mutex_unlock(_mutex);
6086 assert (status == 0, "invariant");
6087 status = pthread_cond_signal (&_cond[index]);
6088 assert (status == 0, "invariant");
6089 }
6090 } else {
6091 pthread_mutex_unlock(_mutex);
6092 assert (status == 0, "invariant") ;
6093 }
6094 } else {
6095 pthread_mutex_unlock(_mutex);
6096 assert (status == 0, "invariant") ;
6097 }
6098 }
6101 extern char** environ;
6103 // Run the specified command in a separate process. Return its exit value,
6104 // or -1 on failure (e.g. can't fork a new process).
6105 // Unlike system(), this function can be called from signal handler. It
6106 // doesn't block SIGINT et al.
6107 int os::fork_and_exec(char* cmd) {
6108 const char * argv[4] = {"sh", "-c", cmd, NULL};
6110 pid_t pid = fork();
6112 if (pid < 0) {
6113 // fork failed
6114 return -1;
6116 } else if (pid == 0) {
6117 // child process
6119 execve("/bin/sh", (char* const*)argv, environ);
6121 // execve failed
6122 _exit(-1);
6124 } else {
6125 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6126 // care about the actual exit code, for now.
6128 int status;
6130 // Wait for the child process to exit. This returns immediately if
6131 // the child has already exited. */
6132 while (waitpid(pid, &status, 0) < 0) {
6133 switch (errno) {
6134 case ECHILD: return 0;
6135 case EINTR: break;
6136 default: return -1;
6137 }
6138 }
6140 if (WIFEXITED(status)) {
6141 // The child exited normally; get its exit code.
6142 return WEXITSTATUS(status);
6143 } else if (WIFSIGNALED(status)) {
6144 // The child exited because of a signal
6145 // The best value to return is 0x80 + signal number,
6146 // because that is what all Unix shells do, and because
6147 // it allows callers to distinguish between process exit and
6148 // process death by signal.
6149 return 0x80 + WTERMSIG(status);
6150 } else {
6151 // Unknown exit code; pass it through
6152 return status;
6153 }
6154 }
6155 }
6157 // is_headless_jre()
6158 //
6159 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6160 // in order to report if we are running in a headless jre
6161 //
6162 // Since JDK8 xawt/libmawt.so was moved into the same directory
6163 // as libawt.so, and renamed libawt_xawt.so
6164 //
6165 bool os::is_headless_jre() {
6166 struct stat statbuf;
6167 char buf[MAXPATHLEN];
6168 char libmawtpath[MAXPATHLEN];
6169 const char *xawtstr = "/xawt/libmawt.so";
6170 const char *new_xawtstr = "/libawt_xawt.so";
6171 char *p;
6173 // Get path to libjvm.so
6174 os::jvm_path(buf, sizeof(buf));
6176 // Get rid of libjvm.so
6177 p = strrchr(buf, '/');
6178 if (p == NULL) return false;
6179 else *p = '\0';
6181 // Get rid of client or server
6182 p = strrchr(buf, '/');
6183 if (p == NULL) return false;
6184 else *p = '\0';
6186 // check xawt/libmawt.so
6187 strcpy(libmawtpath, buf);
6188 strcat(libmawtpath, xawtstr);
6189 if (::stat(libmawtpath, &statbuf) == 0) return false;
6191 // check libawt_xawt.so
6192 strcpy(libmawtpath, buf);
6193 strcat(libmawtpath, new_xawtstr);
6194 if (::stat(libmawtpath, &statbuf) == 0) return false;
6196 return true;
6197 }
6199 // Get the default path to the core file
6200 // Returns the length of the string
6201 int os::get_core_path(char* buffer, size_t bufferSize) {
6202 const char* p = get_current_directory(buffer, bufferSize);
6204 if (p == NULL) {
6205 assert(p != NULL, "failed to get current directory");
6206 return 0;
6207 }
6209 return strlen(buffer);
6210 }
6212 /////////////// Unit tests ///////////////
6214 #ifndef PRODUCT
6216 #define test_log(...) \
6217 do {\
6218 if (VerboseInternalVMTests) { \
6219 tty->print_cr(__VA_ARGS__); \
6220 tty->flush(); \
6221 }\
6222 } while (false)
6224 class TestReserveMemorySpecial : AllStatic {
6225 public:
6226 static void small_page_write(void* addr, size_t size) {
6227 size_t page_size = os::vm_page_size();
6229 char* end = (char*)addr + size;
6230 for (char* p = (char*)addr; p < end; p += page_size) {
6231 *p = 1;
6232 }
6233 }
6235 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6236 if (!UseHugeTLBFS) {
6237 return;
6238 }
6240 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6242 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6244 if (addr != NULL) {
6245 small_page_write(addr, size);
6247 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6248 }
6249 }
6251 static void test_reserve_memory_special_huge_tlbfs_only() {
6252 if (!UseHugeTLBFS) {
6253 return;
6254 }
6256 size_t lp = os::large_page_size();
6258 for (size_t size = lp; size <= lp * 10; size += lp) {
6259 test_reserve_memory_special_huge_tlbfs_only(size);
6260 }
6261 }
6263 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6264 size_t lp = os::large_page_size();
6265 size_t ag = os::vm_allocation_granularity();
6267 // sizes to test
6268 const size_t sizes[] = {
6269 lp, lp + ag, lp + lp / 2, lp * 2,
6270 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6271 lp * 10, lp * 10 + lp / 2
6272 };
6273 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6275 // For each size/alignment combination, we test three scenarios:
6276 // 1) with req_addr == NULL
6277 // 2) with a non-null req_addr at which we expect to successfully allocate
6278 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6279 // expect the allocation to either fail or to ignore req_addr
6281 // Pre-allocate two areas; they shall be as large as the largest allocation
6282 // and aligned to the largest alignment we will be testing.
6283 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6284 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6285 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6286 -1, 0);
6287 assert(mapping1 != MAP_FAILED, "should work");
6289 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6290 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6291 -1, 0);
6292 assert(mapping2 != MAP_FAILED, "should work");
6294 // Unmap the first mapping, but leave the second mapping intact: the first
6295 // mapping will serve as a value for a "good" req_addr (case 2). The second
6296 // mapping, still intact, as "bad" req_addr (case 3).
6297 ::munmap(mapping1, mapping_size);
6299 // Case 1
6300 test_log("%s, req_addr NULL:", __FUNCTION__);
6301 test_log("size align result");
6303 for (int i = 0; i < num_sizes; i++) {
6304 const size_t size = sizes[i];
6305 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6306 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6307 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6308 size, alignment, p, (p != NULL ? "" : "(failed)"));
6309 if (p != NULL) {
6310 assert(is_ptr_aligned(p, alignment), "must be");
6311 small_page_write(p, size);
6312 os::Linux::release_memory_special_huge_tlbfs(p, size);
6313 }
6314 }
6315 }
6317 // Case 2
6318 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6319 test_log("size align req_addr result");
6321 for (int i = 0; i < num_sizes; i++) {
6322 const size_t size = sizes[i];
6323 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6324 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6325 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6326 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6327 size, alignment, req_addr, p,
6328 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6329 if (p != NULL) {
6330 assert(p == req_addr, "must be");
6331 small_page_write(p, size);
6332 os::Linux::release_memory_special_huge_tlbfs(p, size);
6333 }
6334 }
6335 }
6337 // Case 3
6338 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6339 test_log("size align req_addr result");
6341 for (int i = 0; i < num_sizes; i++) {
6342 const size_t size = sizes[i];
6343 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6344 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6345 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6346 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6347 size, alignment, req_addr, p,
6348 ((p != NULL ? "" : "(failed)")));
6349 // as the area around req_addr contains already existing mappings, the API should always
6350 // return NULL (as per contract, it cannot return another address)
6351 assert(p == NULL, "must be");
6352 }
6353 }
6355 ::munmap(mapping2, mapping_size);
6357 }
6359 static void test_reserve_memory_special_huge_tlbfs() {
6360 if (!UseHugeTLBFS) {
6361 return;
6362 }
6364 test_reserve_memory_special_huge_tlbfs_only();
6365 test_reserve_memory_special_huge_tlbfs_mixed();
6366 }
6368 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6369 if (!UseSHM) {
6370 return;
6371 }
6373 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6375 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6377 if (addr != NULL) {
6378 assert(is_ptr_aligned(addr, alignment), "Check");
6379 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6381 small_page_write(addr, size);
6383 os::Linux::release_memory_special_shm(addr, size);
6384 }
6385 }
6387 static void test_reserve_memory_special_shm() {
6388 size_t lp = os::large_page_size();
6389 size_t ag = os::vm_allocation_granularity();
6391 for (size_t size = ag; size < lp * 3; size += ag) {
6392 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6393 test_reserve_memory_special_shm(size, alignment);
6394 }
6395 }
6396 }
6398 static void test() {
6399 test_reserve_memory_special_huge_tlbfs();
6400 test_reserve_memory_special_shm();
6401 }
6402 };
6404 void TestReserveMemorySpecial_test() {
6405 TestReserveMemorySpecial::test();
6406 }
6408 #endif