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