1 /*
2 * Copyright (c) 1999, 2014, 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_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/macros.hpp"
72 #include "utilities/vmError.hpp"
73
74 // put OS-includes here
75 # include <sys/types.h>
76 # include <sys/mman.h>
77 # include <sys/stat.h>
78 # include <sys/select.h>
79 # include <pthread.h>
80 # include <signal.h>
81 # include <errno.h>
82 # include <dlfcn.h>
83 # include <stdio.h>
84 # include <unistd.h>
85 # include <sys/resource.h>
86 # include <pthread.h>
87 # include <sys/stat.h>
88 # include <sys/time.h>
89 # include <sys/times.h>
90 # include <sys/utsname.h>
91 # include <sys/socket.h>
92 # include <sys/wait.h>
93 # include <pwd.h>
94 # include <poll.h>
95 # include <semaphore.h>
96 # include <fcntl.h>
97 # include <string.h>
98 # include <syscall.h>
99 # include <sys/sysinfo.h>
100 # include <gnu/libc-version.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
109
110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
111 // getrusage() is prepared to handle the associated failure.
112 #ifndef RUSAGE_THREAD
113 #define RUSAGE_THREAD (1) /* only the calling thread */
114 #endif
115
116 #define MAX_PATH (2 * K)
117
118 #define MAX_SECS 100000000
119
120 // for timer info max values which include all bits
121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123 #define LARGEPAGES_BIT (1 << 6)
124 ////////////////////////////////////////////////////////////////////////////////
125 // global variables
126 julong os::Linux::_physical_memory = 0;
127
128 address os::Linux::_initial_thread_stack_bottom = NULL;
129 uintptr_t os::Linux::_initial_thread_stack_size = 0;
130
131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 const int os::Linux::_vm_default_page_size = (8 * K);
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
143 pthread_condattr_t os::Linux::_condattr[1];
144
145 static jlong initial_time_count=0;
146
147 static int clock_tics_per_sec = 100;
148
149 // For diagnostics to print a message once. see run_periodic_checks
150 static sigset_t check_signal_done;
151 static bool check_signals = true;
152
153 static pid_t _initial_pid = 0;
154
155 // Signal number used to suspend/resume a thread
156
157 // do not use any signal number less than SIGSEGV, see 4355769
158 static int SR_signum = SIGUSR2;
159 sigset_t SR_sigset;
160
161 // Used to protect dlsym() calls
162 static pthread_mutex_t dl_mutex;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 ////////////////////////////////////////////////////////////////////////////////
188 // environment support
189
190 bool os::getenv(const char* name, char* buf, int len) {
191 const char* val = ::getenv(name);
192 if (val != NULL && strlen(val) < (size_t)len) {
193 strcpy(buf, val);
194 return true;
195 }
196 if (len > 0) buf[0] = 0; // return a null string
197 return false;
198 }
199
200
201 // Return true if user is running as root.
202
203 bool os::have_special_privileges() {
204 static bool init = false;
205 static bool privileges = false;
206 if (!init) {
207 privileges = (getuid() != geteuid()) || (getgid() != getegid());
208 init = true;
209 }
210 return privileges;
211 }
212
213
214 #ifndef SYS_gettid
215 // i386: 224, ia64: 1105, amd64: 186, sparc 143
216 #ifdef __ia64__
217 #define SYS_gettid 1105
218 #elif __i386__
219 #define SYS_gettid 224
220 #elif __amd64__
221 #define SYS_gettid 186
222 #elif __sparc__
223 #define SYS_gettid 143
224 #else
225 #error define gettid for the arch
226 #endif
227 #endif
228
229 // Cpu architecture string
230 #if defined(ZERO)
231 static char cpu_arch[] = ZERO_LIBARCH;
232 #elif defined(IA64)
233 static char cpu_arch[] = "ia64";
234 #elif defined(IA32)
235 static char cpu_arch[] = "i386";
236 #elif defined(AMD64)
237 static char cpu_arch[] = "amd64";
238 #elif defined(ARM)
239 static char cpu_arch[] = "arm";
240 #elif defined(PPC32)
241 static char cpu_arch[] = "ppc";
242 #elif defined(PPC64)
243 static char cpu_arch[] = "ppc64";
244 #elif defined(SPARC)
245 #ifdef _LP64
246 static char cpu_arch[] = "sparcv9";
247 #else
248 static char cpu_arch[] = "sparc";
249 #endif
250 #else
251 #error Add appropriate cpu_arch setting
252 #endif
253
254
255 // pid_t gettid()
256 //
257 // Returns the kernel thread id of the currently running thread. Kernel
258 // thread id is used to access /proc.
259 //
260 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
261 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
262 //
263 pid_t os::Linux::gettid() {
264 int rslt = syscall(SYS_gettid);
265 if (rslt == -1) {
266 // old kernel, no NPTL support
267 return getpid();
268 } else {
269 return (pid_t)rslt;
270 }
271 }
272
273 // Most versions of linux have a bug where the number of processors are
274 // determined by looking at the /proc file system. In a chroot environment,
275 // the system call returns 1. This causes the VM to act as if it is
276 // a single processor and elide locking (see is_MP() call).
277 static bool unsafe_chroot_detected = false;
278 static const char *unstable_chroot_error = "/proc file system not found.\n"
279 "Java may be unstable running multithreaded in a chroot "
280 "environment on Linux when /proc filesystem is not mounted.";
281
282 void os::Linux::initialize_system_info() {
283 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
284 if (processor_count() == 1) {
285 pid_t pid = os::Linux::gettid();
286 char fname[32];
287 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
288 FILE *fp = fopen(fname, "r");
289 if (fp == NULL) {
290 unsafe_chroot_detected = true;
291 } else {
292 fclose(fp);
293 }
294 }
295 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
296 assert(processor_count() > 0, "linux error");
297 }
298
299 void os::init_system_properties_values() {
300 // The next steps are taken in the product version:
301 //
302 // Obtain the JAVA_HOME value from the location of libjvm.so.
303 // This library should be located at:
304 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
305 //
306 // If "/jre/lib/" appears at the right place in the path, then we
307 // assume libjvm.so is installed in a JDK and we use this path.
308 //
309 // Otherwise exit with message: "Could not create the Java virtual machine."
310 //
311 // The following extra steps are taken in the debugging version:
312 //
313 // If "/jre/lib/" does NOT appear at the right place in the path
314 // instead of exit check for $JAVA_HOME environment variable.
315 //
316 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
317 // then we append a fake suffix "hotspot/libjvm.so" to this path so
318 // it looks like libjvm.so is installed there
319 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
320 //
321 // Otherwise exit.
322 //
323 // Important note: if the location of libjvm.so changes this
324 // code needs to be changed accordingly.
325
326 // See ld(1):
327 // The linker uses the following search paths to locate required
328 // shared libraries:
329 // 1: ...
330 // ...
331 // 7: The default directories, normally /lib and /usr/lib.
332 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
333 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
334 #else
335 #define DEFAULT_LIBPATH "/lib:/usr/lib"
336 #endif
337
338 // Base path of extensions installed on the system.
339 #define SYS_EXT_DIR "/usr/java/packages"
340 #define EXTENSIONS_DIR "/lib/ext"
341
342 // Buffer that fits several sprintfs.
343 // Note that the space for the colon and the trailing null are provided
344 // by the nulls included by the sizeof operator.
345 const size_t bufsize =
346 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
347 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
348 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
349
350 // sysclasspath, java_home, dll_dir
351 {
352 char *pslash;
353 os::jvm_path(buf, bufsize);
354
355 // Found the full path to libjvm.so.
356 // Now cut the path to <java_home>/jre if we can.
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /libjvm.so.
360 }
361 pslash = strrchr(buf, '/');
362 if (pslash != NULL) {
363 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
364 }
365 Arguments::set_dll_dir(buf);
366
367 if (pslash != NULL) {
368 pslash = strrchr(buf, '/');
369 if (pslash != NULL) {
370 *pslash = '\0'; // Get rid of /<arch>.
371 pslash = strrchr(buf, '/');
372 if (pslash != NULL) {
373 *pslash = '\0'; // Get rid of /lib.
374 }
375 }
376 }
377 Arguments::set_java_home(buf);
378 set_boot_path('/', ':');
379 }
380
381 // Where to look for native libraries.
382 //
383 // Note: Due to a legacy implementation, most of the library path
384 // is set in the launcher. This was to accomodate linking restrictions
385 // on legacy Linux implementations (which are no longer supported).
386 // Eventually, all the library path setting will be done here.
387 //
388 // However, to prevent the proliferation of improperly built native
389 // libraries, the new path component /usr/java/packages is added here.
390 // Eventually, all the library path setting will be done here.
391 {
392 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
393 // should always exist (until the legacy problem cited above is
394 // addressed).
395 const char *v = ::getenv("LD_LIBRARY_PATH");
396 const char *v_colon = ":";
397 if (v == NULL) { v = ""; v_colon = ""; }
398 // That's +1 for the colon and +1 for the trailing '\0'.
399 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
400 strlen(v) + 1 +
401 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
402 mtInternal);
403 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
404 Arguments::set_library_path(ld_library_path);
405 FREE_C_HEAP_ARRAY(char, ld_library_path);
406 }
407
408 // Extensions directories.
409 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
410 Arguments::set_ext_dirs(buf);
411
412 FREE_C_HEAP_ARRAY(char, buf);
413
414 #undef DEFAULT_LIBPATH
415 #undef SYS_EXT_DIR
416 #undef EXTENSIONS_DIR
417 }
418
419 ////////////////////////////////////////////////////////////////////////////////
420 // breakpoint support
421
422 void os::breakpoint() {
423 BREAKPOINT;
424 }
425
426 extern "C" void breakpoint() {
427 // use debugger to set breakpoint here
428 }
429
430 ////////////////////////////////////////////////////////////////////////////////
431 // signal support
432
433 debug_only(static bool signal_sets_initialized = false);
434 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
435
436 bool os::Linux::is_sig_ignored(int sig) {
437 struct sigaction oact;
438 sigaction(sig, (struct sigaction*)NULL, &oact);
439 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
440 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
441 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
442 return true;
443 } else {
444 return false;
445 }
446 }
447
448 void os::Linux::signal_sets_init() {
449 // Should also have an assertion stating we are still single-threaded.
450 assert(!signal_sets_initialized, "Already initialized");
451 // Fill in signals that are necessarily unblocked for all threads in
452 // the VM. Currently, we unblock the following signals:
453 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
454 // by -Xrs (=ReduceSignalUsage));
455 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
456 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
457 // the dispositions or masks wrt these signals.
458 // Programs embedding the VM that want to use the above signals for their
459 // own purposes must, at this time, use the "-Xrs" option to prevent
460 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
461 // (See bug 4345157, and other related bugs).
462 // In reality, though, unblocking these signals is really a nop, since
463 // these signals are not blocked by default.
464 sigemptyset(&unblocked_sigs);
465 sigemptyset(&allowdebug_blocked_sigs);
466 sigaddset(&unblocked_sigs, SIGILL);
467 sigaddset(&unblocked_sigs, SIGSEGV);
468 sigaddset(&unblocked_sigs, SIGBUS);
469 sigaddset(&unblocked_sigs, SIGFPE);
470 #if defined(PPC64)
471 sigaddset(&unblocked_sigs, SIGTRAP);
472 #endif
473 sigaddset(&unblocked_sigs, SR_signum);
474
475 if (!ReduceSignalUsage) {
476 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
477 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
478 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
479 }
480 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
481 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
482 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
483 }
484 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
485 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
486 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
487 }
488 }
489 // Fill in signals that are blocked by all but the VM thread.
490 sigemptyset(&vm_sigs);
491 if (!ReduceSignalUsage) {
492 sigaddset(&vm_sigs, BREAK_SIGNAL);
493 }
494 debug_only(signal_sets_initialized = true);
495
496 }
497
498 // These are signals that are unblocked while a thread is running Java.
499 // (For some reason, they get blocked by default.)
500 sigset_t* os::Linux::unblocked_signals() {
501 assert(signal_sets_initialized, "Not initialized");
502 return &unblocked_sigs;
503 }
504
505 // These are the signals that are blocked while a (non-VM) thread is
506 // running Java. Only the VM thread handles these signals.
507 sigset_t* os::Linux::vm_signals() {
508 assert(signal_sets_initialized, "Not initialized");
509 return &vm_sigs;
510 }
511
512 // These are signals that are blocked during cond_wait to allow debugger in
513 sigset_t* os::Linux::allowdebug_blocked_signals() {
514 assert(signal_sets_initialized, "Not initialized");
515 return &allowdebug_blocked_sigs;
516 }
517
518 void os::Linux::hotspot_sigmask(Thread* thread) {
519
520 //Save caller's signal mask before setting VM signal mask
521 sigset_t caller_sigmask;
522 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
523
524 OSThread* osthread = thread->osthread();
525 osthread->set_caller_sigmask(caller_sigmask);
526
527 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
528
529 if (!ReduceSignalUsage) {
530 if (thread->is_VM_thread()) {
531 // Only the VM thread handles BREAK_SIGNAL ...
532 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
533 } else {
534 // ... all other threads block BREAK_SIGNAL
535 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
536 }
537 }
538 }
539
540 //////////////////////////////////////////////////////////////////////////////
541 // detecting pthread library
542
543 void os::Linux::libpthread_init() {
544 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
545 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
546 // generic name for earlier versions.
547 // Define macros here so we can build HotSpot on old systems.
548 #ifndef _CS_GNU_LIBC_VERSION
549 #define _CS_GNU_LIBC_VERSION 2
550 #endif
551 #ifndef _CS_GNU_LIBPTHREAD_VERSION
552 #define _CS_GNU_LIBPTHREAD_VERSION 3
553 #endif
554
555 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
556 if (n > 0) {
557 char *str = (char *)malloc(n, mtInternal);
558 confstr(_CS_GNU_LIBC_VERSION, str, n);
559 os::Linux::set_glibc_version(str);
560 } else {
561 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
562 static char _gnu_libc_version[32];
563 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
564 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
565 os::Linux::set_glibc_version(_gnu_libc_version);
566 }
567
568 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
569 if (n > 0) {
570 char *str = (char *)malloc(n, mtInternal);
571 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
572 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
573 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
574 // is the case. LinuxThreads has a hard limit on max number of threads.
575 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
576 // On the other hand, NPTL does not have such a limit, sysconf()
577 // will return -1 and errno is not changed. Check if it is really NPTL.
578 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
579 strstr(str, "NPTL") &&
580 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
581 free(str);
582 os::Linux::set_libpthread_version("linuxthreads");
583 } else {
584 os::Linux::set_libpthread_version(str);
585 }
586 } else {
587 // glibc before 2.3.2 only has LinuxThreads.
588 os::Linux::set_libpthread_version("linuxthreads");
589 }
590
591 if (strstr(libpthread_version(), "NPTL")) {
592 os::Linux::set_is_NPTL();
593 } else {
594 os::Linux::set_is_LinuxThreads();
595 }
596
597 // LinuxThreads have two flavors: floating-stack mode, which allows variable
598 // stack size; and fixed-stack mode. NPTL is always floating-stack.
599 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
600 os::Linux::set_is_floating_stack();
601 }
602 }
603
604 /////////////////////////////////////////////////////////////////////////////
605 // thread stack
606
607 // Force Linux kernel to expand current thread stack. If "bottom" is close
608 // to the stack guard, caller should block all signals.
609 //
610 // MAP_GROWSDOWN:
611 // A special mmap() flag that is used to implement thread stacks. It tells
612 // kernel that the memory region should extend downwards when needed. This
613 // allows early versions of LinuxThreads to only mmap the first few pages
614 // when creating a new thread. Linux kernel will automatically expand thread
615 // stack as needed (on page faults).
616 //
617 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
618 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
619 // region, it's hard to tell if the fault is due to a legitimate stack
620 // access or because of reading/writing non-exist memory (e.g. buffer
621 // overrun). As a rule, if the fault happens below current stack pointer,
622 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
623 // application (see Linux kernel fault.c).
624 //
625 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
626 // stack overflow detection.
627 //
628 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
629 // not use this flag. However, the stack of initial thread is not created
630 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
631 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
632 // and then attach the thread to JVM.
633 //
634 // To get around the problem and allow stack banging on Linux, we need to
635 // manually expand thread stack after receiving the SIGSEGV.
636 //
637 // There are two ways to expand thread stack to address "bottom", we used
638 // both of them in JVM before 1.5:
639 // 1. adjust stack pointer first so that it is below "bottom", and then
640 // touch "bottom"
641 // 2. mmap() the page in question
642 //
643 // Now alternate signal stack is gone, it's harder to use 2. For instance,
644 // if current sp is already near the lower end of page 101, and we need to
645 // call mmap() to map page 100, it is possible that part of the mmap() frame
646 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
647 // That will destroy the mmap() frame and cause VM to crash.
648 //
649 // The following code works by adjusting sp first, then accessing the "bottom"
650 // page to force a page fault. Linux kernel will then automatically expand the
651 // stack mapping.
652 //
653 // _expand_stack_to() assumes its frame size is less than page size, which
654 // should always be true if the function is not inlined.
655
656 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
657 #define NOINLINE
658 #else
659 #define NOINLINE __attribute__ ((noinline))
660 #endif
661
662 static void _expand_stack_to(address bottom) NOINLINE;
663
664 static void _expand_stack_to(address bottom) {
665 address sp;
666 size_t size;
667 volatile char *p;
668
669 // Adjust bottom to point to the largest address within the same page, it
670 // gives us a one-page buffer if alloca() allocates slightly more memory.
671 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
672 bottom += os::Linux::page_size() - 1;
673
674 // sp might be slightly above current stack pointer; if that's the case, we
675 // will alloca() a little more space than necessary, which is OK. Don't use
676 // os::current_stack_pointer(), as its result can be slightly below current
677 // stack pointer, causing us to not alloca enough to reach "bottom".
678 sp = (address)&sp;
679
680 if (sp > bottom) {
681 size = sp - bottom;
682 p = (volatile char *)alloca(size);
683 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
684 p[0] = '\0';
685 }
686 }
687
688 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
689 assert(t!=NULL, "just checking");
690 assert(t->osthread()->expanding_stack(), "expand should be set");
691 assert(t->stack_base() != NULL, "stack_base was not initialized");
692
693 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
694 sigset_t mask_all, old_sigset;
695 sigfillset(&mask_all);
696 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
697 _expand_stack_to(addr);
698 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
699 return true;
700 }
701 return false;
702 }
703
704 //////////////////////////////////////////////////////////////////////////////
705 // create new thread
706
707 static address highest_vm_reserved_address();
708
709 // check if it's safe to start a new thread
710 static bool _thread_safety_check(Thread* thread) {
711 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
712 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
713 // Heap is mmap'ed at lower end of memory space. Thread stacks are
714 // allocated (MAP_FIXED) from high address space. Every thread stack
715 // occupies a fixed size slot (usually 2Mbytes, but user can change
716 // it to other values if they rebuild LinuxThreads).
717 //
718 // Problem with MAP_FIXED is that mmap() can still succeed even part of
719 // the memory region has already been mmap'ed. That means if we have too
720 // many threads and/or very large heap, eventually thread stack will
721 // collide with heap.
722 //
723 // Here we try to prevent heap/stack collision by comparing current
724 // stack bottom with the highest address that has been mmap'ed by JVM
725 // plus a safety margin for memory maps created by native code.
726 //
727 // This feature can be disabled by setting ThreadSafetyMargin to 0
728 //
729 if (ThreadSafetyMargin > 0) {
730 address stack_bottom = os::current_stack_base() - os::current_stack_size();
731
732 // not safe if our stack extends below the safety margin
733 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
734 } else {
735 return true;
736 }
737 } else {
738 // Floating stack LinuxThreads or NPTL:
739 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
740 // there's not enough space left, pthread_create() will fail. If we come
741 // here, that means enough space has been reserved for stack.
742 return true;
743 }
744 }
745
746 // Thread start routine for all newly created threads
747 static void *java_start(Thread *thread) {
748 // Try to randomize the cache line index of hot stack frames.
749 // This helps when threads of the same stack traces evict each other's
750 // cache lines. The threads can be either from the same JVM instance, or
751 // from different JVM instances. The benefit is especially true for
752 // processors with hyperthreading technology.
753 static int counter = 0;
754 int pid = os::current_process_id();
755 alloca(((pid ^ counter++) & 7) * 128);
756
757 ThreadLocalStorage::set_thread(thread);
758
759 OSThread* osthread = thread->osthread();
760 Monitor* sync = osthread->startThread_lock();
761
762 // non floating stack LinuxThreads needs extra check, see above
763 if (!_thread_safety_check(thread)) {
764 // notify parent thread
765 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
766 osthread->set_state(ZOMBIE);
767 sync->notify_all();
768 return NULL;
769 }
770
771 // thread_id is kernel thread id (similar to Solaris LWP id)
772 osthread->set_thread_id(os::Linux::gettid());
773
774 if (UseNUMA) {
775 int lgrp_id = os::numa_get_group_id();
776 if (lgrp_id != -1) {
777 thread->set_lgrp_id(lgrp_id);
778 }
779 }
780 // initialize signal mask for this thread
781 os::Linux::hotspot_sigmask(thread);
782
783 // initialize floating point control register
784 os::Linux::init_thread_fpu_state();
785
786 // handshaking with parent thread
787 {
788 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
789
790 // notify parent thread
791 osthread->set_state(INITIALIZED);
792 sync->notify_all();
793
794 // wait until os::start_thread()
795 while (osthread->get_state() == INITIALIZED) {
796 sync->wait(Mutex::_no_safepoint_check_flag);
797 }
798 }
799
800 // call one more level start routine
801 thread->run();
802
803 return 0;
804 }
805
806 bool os::create_thread(Thread* thread, ThreadType thr_type,
807 size_t stack_size) {
808 assert(thread->osthread() == NULL, "caller responsible");
809
810 // Allocate the OSThread object
811 OSThread* osthread = new OSThread(NULL, NULL);
812 if (osthread == NULL) {
813 return false;
814 }
815
816 // set the correct thread state
817 osthread->set_thread_type(thr_type);
818
819 // Initial state is ALLOCATED but not INITIALIZED
820 osthread->set_state(ALLOCATED);
821
822 thread->set_osthread(osthread);
823
824 // init thread attributes
825 pthread_attr_t attr;
826 pthread_attr_init(&attr);
827 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
828
829 // stack size
830 if (os::Linux::supports_variable_stack_size()) {
831 // calculate stack size if it's not specified by caller
832 if (stack_size == 0) {
833 stack_size = os::Linux::default_stack_size(thr_type);
834
835 switch (thr_type) {
836 case os::java_thread:
837 // Java threads use ThreadStackSize which default value can be
838 // changed with the flag -Xss
839 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
840 stack_size = JavaThread::stack_size_at_create();
841 break;
842 case os::compiler_thread:
843 if (CompilerThreadStackSize > 0) {
844 stack_size = (size_t)(CompilerThreadStackSize * K);
845 break;
846 } // else fall through:
847 // use VMThreadStackSize if CompilerThreadStackSize is not defined
848 case os::vm_thread:
849 case os::pgc_thread:
850 case os::cgc_thread:
851 case os::watcher_thread:
852 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
853 break;
854 }
855 }
856
857 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
858 pthread_attr_setstacksize(&attr, stack_size);
859 } else {
860 // let pthread_create() pick the default value.
861 }
862
863 // glibc guard page
864 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
865
866 ThreadState state;
867
868 {
869 // Serialize thread creation if we are running with fixed stack LinuxThreads
870 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
871 if (lock) {
872 os::Linux::createThread_lock()->lock_without_safepoint_check();
873 }
874
875 pthread_t tid;
876 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
877
878 pthread_attr_destroy(&attr);
879
880 if (ret != 0) {
881 if (PrintMiscellaneous && (Verbose || WizardMode)) {
882 perror("pthread_create()");
883 }
884 // Need to clean up stuff we've allocated so far
885 thread->set_osthread(NULL);
886 delete osthread;
887 if (lock) os::Linux::createThread_lock()->unlock();
888 return false;
889 }
890
891 // Store pthread info into the OSThread
892 osthread->set_pthread_id(tid);
893
894 // Wait until child thread is either initialized or aborted
895 {
896 Monitor* sync_with_child = osthread->startThread_lock();
897 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
898 while ((state = osthread->get_state()) == ALLOCATED) {
899 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
900 }
901 }
902
903 if (lock) {
904 os::Linux::createThread_lock()->unlock();
905 }
906 }
907
908 // Aborted due to thread limit being reached
909 if (state == ZOMBIE) {
910 thread->set_osthread(NULL);
911 delete osthread;
912 return false;
913 }
914
915 // The thread is returned suspended (in state INITIALIZED),
916 // and is started higher up in the call chain
917 assert(state == INITIALIZED, "race condition");
918 return true;
919 }
920
921 /////////////////////////////////////////////////////////////////////////////
922 // attach existing thread
923
924 // bootstrap the main thread
925 bool os::create_main_thread(JavaThread* thread) {
926 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
927 return create_attached_thread(thread);
928 }
929
930 bool os::create_attached_thread(JavaThread* thread) {
931 #ifdef ASSERT
932 thread->verify_not_published();
933 #endif
934
935 // Allocate the OSThread object
936 OSThread* osthread = new OSThread(NULL, NULL);
937
938 if (osthread == NULL) {
939 return false;
940 }
941
942 // Store pthread info into the OSThread
943 osthread->set_thread_id(os::Linux::gettid());
944 osthread->set_pthread_id(::pthread_self());
945
946 // initialize floating point control register
947 os::Linux::init_thread_fpu_state();
948
949 // Initial thread state is RUNNABLE
950 osthread->set_state(RUNNABLE);
951
952 thread->set_osthread(osthread);
953
954 if (UseNUMA) {
955 int lgrp_id = os::numa_get_group_id();
956 if (lgrp_id != -1) {
957 thread->set_lgrp_id(lgrp_id);
958 }
959 }
960
961 if (os::Linux::is_initial_thread()) {
962 // If current thread is initial thread, its stack is mapped on demand,
963 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
964 // the entire stack region to avoid SEGV in stack banging.
965 // It is also useful to get around the heap-stack-gap problem on SuSE
966 // kernel (see 4821821 for details). We first expand stack to the top
967 // of yellow zone, then enable stack yellow zone (order is significant,
968 // enabling yellow zone first will crash JVM on SuSE Linux), so there
969 // is no gap between the last two virtual memory regions.
970
971 JavaThread *jt = (JavaThread *)thread;
972 address addr = jt->stack_yellow_zone_base();
973 assert(addr != NULL, "initialization problem?");
974 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
975
976 osthread->set_expanding_stack();
977 os::Linux::manually_expand_stack(jt, addr);
978 osthread->clear_expanding_stack();
979 }
980
981 // initialize signal mask for this thread
982 // and save the caller's signal mask
983 os::Linux::hotspot_sigmask(thread);
984
985 return true;
986 }
987
988 void os::pd_start_thread(Thread* thread) {
989 OSThread * osthread = thread->osthread();
990 assert(osthread->get_state() != INITIALIZED, "just checking");
991 Monitor* sync_with_child = osthread->startThread_lock();
992 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
993 sync_with_child->notify();
994 }
995
996 // Free Linux resources related to the OSThread
997 void os::free_thread(OSThread* osthread) {
998 assert(osthread != NULL, "osthread not set");
999
1000 if (Thread::current()->osthread() == osthread) {
1001 // Restore caller's signal mask
1002 sigset_t sigmask = osthread->caller_sigmask();
1003 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1004 }
1005
1006 delete osthread;
1007 }
1008
1009 //////////////////////////////////////////////////////////////////////////////
1010 // thread local storage
1011
1012 // Restore the thread pointer if the destructor is called. This is in case
1013 // someone from JNI code sets up a destructor with pthread_key_create to run
1014 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1015 // will hang or crash. When detachCurrentThread is called the key will be set
1016 // to null and we will not be called again. If detachCurrentThread is never
1017 // called we could loop forever depending on the pthread implementation.
1018 static void restore_thread_pointer(void* p) {
1019 Thread* thread = (Thread*) p;
1020 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1021 }
1022
1023 int os::allocate_thread_local_storage() {
1024 pthread_key_t key;
1025 int rslt = pthread_key_create(&key, restore_thread_pointer);
1026 assert(rslt == 0, "cannot allocate thread local storage");
1027 return (int)key;
1028 }
1029
1030 // Note: This is currently not used by VM, as we don't destroy TLS key
1031 // on VM exit.
1032 void os::free_thread_local_storage(int index) {
1033 int rslt = pthread_key_delete((pthread_key_t)index);
1034 assert(rslt == 0, "invalid index");
1035 }
1036
1037 void os::thread_local_storage_at_put(int index, void* value) {
1038 int rslt = pthread_setspecific((pthread_key_t)index, value);
1039 assert(rslt == 0, "pthread_setspecific failed");
1040 }
1041
1042 extern "C" Thread* get_thread() {
1043 return ThreadLocalStorage::thread();
1044 }
1045
1046 //////////////////////////////////////////////////////////////////////////////
1047 // initial thread
1048
1049 // Check if current thread is the initial thread, similar to Solaris thr_main.
1050 bool os::Linux::is_initial_thread(void) {
1051 char dummy;
1052 // If called before init complete, thread stack bottom will be null.
1053 // Can be called if fatal error occurs before initialization.
1054 if (initial_thread_stack_bottom() == NULL) return false;
1055 assert(initial_thread_stack_bottom() != NULL &&
1056 initial_thread_stack_size() != 0,
1057 "os::init did not locate initial thread's stack region");
1058 if ((address)&dummy >= initial_thread_stack_bottom() &&
1059 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1060 return true;
1061 } else {
1062 return false;
1063 }
1064 }
1065
1066 // Find the virtual memory area that contains addr
1067 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1068 FILE *fp = fopen("/proc/self/maps", "r");
1069 if (fp) {
1070 address low, high;
1071 while (!feof(fp)) {
1072 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1073 if (low <= addr && addr < high) {
1074 if (vma_low) *vma_low = low;
1075 if (vma_high) *vma_high = high;
1076 fclose(fp);
1077 return true;
1078 }
1079 }
1080 for (;;) {
1081 int ch = fgetc(fp);
1082 if (ch == EOF || ch == (int)'\n') break;
1083 }
1084 }
1085 fclose(fp);
1086 }
1087 return false;
1088 }
1089
1090 // Locate initial thread stack. This special handling of initial thread stack
1091 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1092 // bogus value for initial thread.
1093 void os::Linux::capture_initial_stack(size_t max_size) {
1094 // stack size is the easy part, get it from RLIMIT_STACK
1095 size_t stack_size;
1096 struct rlimit rlim;
1097 getrlimit(RLIMIT_STACK, &rlim);
1098 stack_size = rlim.rlim_cur;
1099
1100 // 6308388: a bug in ld.so will relocate its own .data section to the
1101 // lower end of primordial stack; reduce ulimit -s value a little bit
1102 // so we won't install guard page on ld.so's data section.
1103 stack_size -= 2 * page_size();
1104
1105 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1106 // 7.1, in both cases we will get 2G in return value.
1107 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1108 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1109 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1110 // in case other parts in glibc still assumes 2M max stack size.
1111 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1112 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1113 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1114 stack_size = 2 * K * K IA64_ONLY(*2);
1115 }
1116 // Try to figure out where the stack base (top) is. This is harder.
1117 //
1118 // When an application is started, glibc saves the initial stack pointer in
1119 // a global variable "__libc_stack_end", which is then used by system
1120 // libraries. __libc_stack_end should be pretty close to stack top. The
1121 // variable is available since the very early days. However, because it is
1122 // a private interface, it could disappear in the future.
1123 //
1124 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1125 // to __libc_stack_end, it is very close to stack top, but isn't the real
1126 // stack top. Note that /proc may not exist if VM is running as a chroot
1127 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1128 // /proc/<pid>/stat could change in the future (though unlikely).
1129 //
1130 // We try __libc_stack_end first. If that doesn't work, look for
1131 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1132 // as a hint, which should work well in most cases.
1133
1134 uintptr_t stack_start;
1135
1136 // try __libc_stack_end first
1137 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1138 if (p && *p) {
1139 stack_start = *p;
1140 } else {
1141 // see if we can get the start_stack field from /proc/self/stat
1142 FILE *fp;
1143 int pid;
1144 char state;
1145 int ppid;
1146 int pgrp;
1147 int session;
1148 int nr;
1149 int tpgrp;
1150 unsigned long flags;
1151 unsigned long minflt;
1152 unsigned long cminflt;
1153 unsigned long majflt;
1154 unsigned long cmajflt;
1155 unsigned long utime;
1156 unsigned long stime;
1157 long cutime;
1158 long cstime;
1159 long prio;
1160 long nice;
1161 long junk;
1162 long it_real;
1163 uintptr_t start;
1164 uintptr_t vsize;
1165 intptr_t rss;
1166 uintptr_t rsslim;
1167 uintptr_t scodes;
1168 uintptr_t ecode;
1169 int i;
1170
1171 // Figure what the primordial thread stack base is. Code is inspired
1172 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1173 // followed by command name surrounded by parentheses, state, etc.
1174 char stat[2048];
1175 int statlen;
1176
1177 fp = fopen("/proc/self/stat", "r");
1178 if (fp) {
1179 statlen = fread(stat, 1, 2047, fp);
1180 stat[statlen] = '\0';
1181 fclose(fp);
1182
1183 // Skip pid and the command string. Note that we could be dealing with
1184 // weird command names, e.g. user could decide to rename java launcher
1185 // to "java 1.4.2 :)", then the stat file would look like
1186 // 1234 (java 1.4.2 :)) R ... ...
1187 // We don't really need to know the command string, just find the last
1188 // occurrence of ")" and then start parsing from there. See bug 4726580.
1189 char * s = strrchr(stat, ')');
1190
1191 i = 0;
1192 if (s) {
1193 // Skip blank chars
1194 do { s++; } while (s && isspace(*s));
1195
1196 #define _UFM UINTX_FORMAT
1197 #define _DFM INTX_FORMAT
1198
1199 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1200 // 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
1201 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,
1202 &state, // 3 %c
1203 &ppid, // 4 %d
1204 &pgrp, // 5 %d
1205 &session, // 6 %d
1206 &nr, // 7 %d
1207 &tpgrp, // 8 %d
1208 &flags, // 9 %lu
1209 &minflt, // 10 %lu
1210 &cminflt, // 11 %lu
1211 &majflt, // 12 %lu
1212 &cmajflt, // 13 %lu
1213 &utime, // 14 %lu
1214 &stime, // 15 %lu
1215 &cutime, // 16 %ld
1216 &cstime, // 17 %ld
1217 &prio, // 18 %ld
1218 &nice, // 19 %ld
1219 &junk, // 20 %ld
1220 &it_real, // 21 %ld
1221 &start, // 22 UINTX_FORMAT
1222 &vsize, // 23 UINTX_FORMAT
1223 &rss, // 24 INTX_FORMAT
1224 &rsslim, // 25 UINTX_FORMAT
1225 &scodes, // 26 UINTX_FORMAT
1226 &ecode, // 27 UINTX_FORMAT
1227 &stack_start); // 28 UINTX_FORMAT
1228 }
1229
1230 #undef _UFM
1231 #undef _DFM
1232
1233 if (i != 28 - 2) {
1234 assert(false, "Bad conversion from /proc/self/stat");
1235 // product mode - assume we are the initial thread, good luck in the
1236 // embedded case.
1237 warning("Can't detect initial thread stack location - bad conversion");
1238 stack_start = (uintptr_t) &rlim;
1239 }
1240 } else {
1241 // For some reason we can't open /proc/self/stat (for example, running on
1242 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1243 // most cases, so don't abort:
1244 warning("Can't detect initial thread stack location - no /proc/self/stat");
1245 stack_start = (uintptr_t) &rlim;
1246 }
1247 }
1248
1249 // Now we have a pointer (stack_start) very close to the stack top, the
1250 // next thing to do is to figure out the exact location of stack top. We
1251 // can find out the virtual memory area that contains stack_start by
1252 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1253 // and its upper limit is the real stack top. (again, this would fail if
1254 // running inside chroot, because /proc may not exist.)
1255
1256 uintptr_t stack_top;
1257 address low, high;
1258 if (find_vma((address)stack_start, &low, &high)) {
1259 // success, "high" is the true stack top. (ignore "low", because initial
1260 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1261 stack_top = (uintptr_t)high;
1262 } else {
1263 // failed, likely because /proc/self/maps does not exist
1264 warning("Can't detect initial thread stack location - find_vma failed");
1265 // best effort: stack_start is normally within a few pages below the real
1266 // stack top, use it as stack top, and reduce stack size so we won't put
1267 // guard page outside stack.
1268 stack_top = stack_start;
1269 stack_size -= 16 * page_size();
1270 }
1271
1272 // stack_top could be partially down the page so align it
1273 stack_top = align_size_up(stack_top, page_size());
1274
1275 if (max_size && stack_size > max_size) {
1276 _initial_thread_stack_size = max_size;
1277 } else {
1278 _initial_thread_stack_size = stack_size;
1279 }
1280
1281 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1282 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1283 }
1284
1285 ////////////////////////////////////////////////////////////////////////////////
1286 // time support
1287
1288 // Time since start-up in seconds to a fine granularity.
1289 // Used by VMSelfDestructTimer and the MemProfiler.
1290 double os::elapsedTime() {
1291
1292 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1293 }
1294
1295 jlong os::elapsed_counter() {
1296 return javaTimeNanos() - initial_time_count;
1297 }
1298
1299 jlong os::elapsed_frequency() {
1300 return NANOSECS_PER_SEC; // nanosecond resolution
1301 }
1302
1303 bool os::supports_vtime() { return true; }
1304 bool os::enable_vtime() { return false; }
1305 bool os::vtime_enabled() { return false; }
1306
1307 double os::elapsedVTime() {
1308 struct rusage usage;
1309 int retval = getrusage(RUSAGE_THREAD, &usage);
1310 if (retval == 0) {
1311 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);
1312 } else {
1313 // better than nothing, but not much
1314 return elapsedTime();
1315 }
1316 }
1317
1318 jlong os::javaTimeMillis() {
1319 timeval time;
1320 int status = gettimeofday(&time, NULL);
1321 assert(status != -1, "linux error");
1322 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1323 }
1324
1325 #ifndef CLOCK_MONOTONIC
1326 #define CLOCK_MONOTONIC (1)
1327 #endif
1328
1329 void os::Linux::clock_init() {
1330 // we do dlopen's in this particular order due to bug in linux
1331 // dynamical loader (see 6348968) leading to crash on exit
1332 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1333 if (handle == NULL) {
1334 handle = dlopen("librt.so", RTLD_LAZY);
1335 }
1336
1337 if (handle) {
1338 int (*clock_getres_func)(clockid_t, struct timespec*) =
1339 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1340 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1341 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1342 if (clock_getres_func && clock_gettime_func) {
1343 // See if monotonic clock is supported by the kernel. Note that some
1344 // early implementations simply return kernel jiffies (updated every
1345 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1346 // for nano time (though the monotonic property is still nice to have).
1347 // It's fixed in newer kernels, however clock_getres() still returns
1348 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1349 // resolution for now. Hopefully as people move to new kernels, this
1350 // won't be a problem.
1351 struct timespec res;
1352 struct timespec tp;
1353 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1354 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1355 // yes, monotonic clock is supported
1356 _clock_gettime = clock_gettime_func;
1357 return;
1358 } else {
1359 // close librt if there is no monotonic clock
1360 dlclose(handle);
1361 }
1362 }
1363 }
1364 warning("No monotonic clock was available - timed services may " \
1365 "be adversely affected if the time-of-day clock changes");
1366 }
1367
1368 #ifndef SYS_clock_getres
1369 #if defined(IA32) || defined(AMD64)
1370 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1371 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1372 #else
1373 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1374 #define sys_clock_getres(x,y) -1
1375 #endif
1376 #else
1377 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1378 #endif
1379
1380 void os::Linux::fast_thread_clock_init() {
1381 if (!UseLinuxPosixThreadCPUClocks) {
1382 return;
1383 }
1384 clockid_t clockid;
1385 struct timespec tp;
1386 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1387 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1388
1389 // Switch to using fast clocks for thread cpu time if
1390 // the sys_clock_getres() returns 0 error code.
1391 // Note, that some kernels may support the current thread
1392 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1393 // returned by the pthread_getcpuclockid().
1394 // If the fast Posix clocks are supported then the sys_clock_getres()
1395 // must return at least tp.tv_sec == 0 which means a resolution
1396 // better than 1 sec. This is extra check for reliability.
1397
1398 if (pthread_getcpuclockid_func &&
1399 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1400 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1401 _supports_fast_thread_cpu_time = true;
1402 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1403 }
1404 }
1405
1406 jlong os::javaTimeNanos() {
1407 if (os::supports_monotonic_clock()) {
1408 struct timespec tp;
1409 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1410 assert(status == 0, "gettime error");
1411 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1412 return result;
1413 } else {
1414 timeval time;
1415 int status = gettimeofday(&time, NULL);
1416 assert(status != -1, "linux error");
1417 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1418 return 1000 * usecs;
1419 }
1420 }
1421
1422 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1423 if (os::supports_monotonic_clock()) {
1424 info_ptr->max_value = ALL_64_BITS;
1425
1426 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1427 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1428 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1429 } else {
1430 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1431 info_ptr->max_value = ALL_64_BITS;
1432
1433 // gettimeofday is a real time clock so it skips
1434 info_ptr->may_skip_backward = true;
1435 info_ptr->may_skip_forward = true;
1436 }
1437
1438 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1439 }
1440
1441 // Return the real, user, and system times in seconds from an
1442 // arbitrary fixed point in the past.
1443 bool os::getTimesSecs(double* process_real_time,
1444 double* process_user_time,
1445 double* process_system_time) {
1446 struct tms ticks;
1447 clock_t real_ticks = times(&ticks);
1448
1449 if (real_ticks == (clock_t) (-1)) {
1450 return false;
1451 } else {
1452 double ticks_per_second = (double) clock_tics_per_sec;
1453 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1454 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1455 *process_real_time = ((double) real_ticks) / ticks_per_second;
1456
1457 return true;
1458 }
1459 }
1460
1461
1462 char * os::local_time_string(char *buf, size_t buflen) {
1463 struct tm t;
1464 time_t long_time;
1465 time(&long_time);
1466 localtime_r(&long_time, &t);
1467 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1468 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1469 t.tm_hour, t.tm_min, t.tm_sec);
1470 return buf;
1471 }
1472
1473 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1474 return localtime_r(clock, res);
1475 }
1476
1477 ////////////////////////////////////////////////////////////////////////////////
1478 // runtime exit support
1479
1480 // Note: os::shutdown() might be called very early during initialization, or
1481 // called from signal handler. Before adding something to os::shutdown(), make
1482 // sure it is async-safe and can handle partially initialized VM.
1483 void os::shutdown() {
1484
1485 // allow PerfMemory to attempt cleanup of any persistent resources
1486 perfMemory_exit();
1487
1488 // needs to remove object in file system
1489 AttachListener::abort();
1490
1491 // flush buffered output, finish log files
1492 ostream_abort();
1493
1494 // Check for abort hook
1495 abort_hook_t abort_hook = Arguments::abort_hook();
1496 if (abort_hook != NULL) {
1497 abort_hook();
1498 }
1499
1500 }
1501
1502 // Note: os::abort() might be called very early during initialization, or
1503 // called from signal handler. Before adding something to os::abort(), make
1504 // sure it is async-safe and can handle partially initialized VM.
1505 void os::abort(bool dump_core) {
1506 os::shutdown();
1507 if (dump_core) {
1508 #ifndef PRODUCT
1509 fdStream out(defaultStream::output_fd());
1510 out.print_raw("Current thread is ");
1511 char buf[16];
1512 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1513 out.print_raw_cr(buf);
1514 out.print_raw_cr("Dumping core ...");
1515 #endif
1516 ::abort(); // dump core
1517 }
1518
1519 ::exit(1);
1520 }
1521
1522 // Die immediately, no exit hook, no abort hook, no cleanup.
1523 void os::die() {
1524 // _exit() on LinuxThreads only kills current thread
1525 ::abort();
1526 }
1527
1528
1529 // This method is a copy of JDK's sysGetLastErrorString
1530 // from src/solaris/hpi/src/system_md.c
1531
1532 size_t os::lasterror(char *buf, size_t len) {
1533 if (errno == 0) return 0;
1534
1535 const char *s = ::strerror(errno);
1536 size_t n = ::strlen(s);
1537 if (n >= len) {
1538 n = len - 1;
1539 }
1540 ::strncpy(buf, s, n);
1541 buf[n] = '\0';
1542 return n;
1543 }
1544
1545 intx os::current_thread_id() { return (intx)pthread_self(); }
1546 int os::current_process_id() {
1547
1548 // Under the old linux thread library, linux gives each thread
1549 // its own process id. Because of this each thread will return
1550 // a different pid if this method were to return the result
1551 // of getpid(2). Linux provides no api that returns the pid
1552 // of the launcher thread for the vm. This implementation
1553 // returns a unique pid, the pid of the launcher thread
1554 // that starts the vm 'process'.
1555
1556 // Under the NPTL, getpid() returns the same pid as the
1557 // launcher thread rather than a unique pid per thread.
1558 // Use gettid() if you want the old pre NPTL behaviour.
1559
1560 // if you are looking for the result of a call to getpid() that
1561 // returns a unique pid for the calling thread, then look at the
1562 // OSThread::thread_id() method in osThread_linux.hpp file
1563
1564 return (int)(_initial_pid ? _initial_pid : getpid());
1565 }
1566
1567 // DLL functions
1568
1569 const char* os::dll_file_extension() { return ".so"; }
1570
1571 // This must be hard coded because it's the system's temporary
1572 // directory not the java application's temp directory, ala java.io.tmpdir.
1573 const char* os::get_temp_directory() { return "/tmp"; }
1574
1575 static bool file_exists(const char* filename) {
1576 struct stat statbuf;
1577 if (filename == NULL || strlen(filename) == 0) {
1578 return false;
1579 }
1580 return os::stat(filename, &statbuf) == 0;
1581 }
1582
1583 bool os::dll_build_name(char* buffer, size_t buflen,
1584 const char* pname, const char* fname) {
1585 bool retval = false;
1586 // Copied from libhpi
1587 const size_t pnamelen = pname ? strlen(pname) : 0;
1588
1589 // Return error on buffer overflow.
1590 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1591 return retval;
1592 }
1593
1594 if (pnamelen == 0) {
1595 snprintf(buffer, buflen, "lib%s.so", fname);
1596 retval = true;
1597 } else if (strchr(pname, *os::path_separator()) != NULL) {
1598 int n;
1599 char** pelements = split_path(pname, &n);
1600 if (pelements == NULL) {
1601 return false;
1602 }
1603 for (int i = 0; i < n; i++) {
1604 // Really shouldn't be NULL, but check can't hurt
1605 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1606 continue; // skip the empty path values
1607 }
1608 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1609 if (file_exists(buffer)) {
1610 retval = true;
1611 break;
1612 }
1613 }
1614 // release the storage
1615 for (int i = 0; i < n; i++) {
1616 if (pelements[i] != NULL) {
1617 FREE_C_HEAP_ARRAY(char, pelements[i]);
1618 }
1619 }
1620 if (pelements != NULL) {
1621 FREE_C_HEAP_ARRAY(char*, pelements);
1622 }
1623 } else {
1624 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1625 retval = true;
1626 }
1627 return retval;
1628 }
1629
1630 // check if addr is inside libjvm.so
1631 bool os::address_is_in_vm(address addr) {
1632 static address libjvm_base_addr;
1633 Dl_info dlinfo;
1634
1635 if (libjvm_base_addr == NULL) {
1636 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1637 libjvm_base_addr = (address)dlinfo.dli_fbase;
1638 }
1639 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1640 }
1641
1642 if (dladdr((void *)addr, &dlinfo) != 0) {
1643 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1644 }
1645
1646 return false;
1647 }
1648
1649 bool os::dll_address_to_function_name(address addr, char *buf,
1650 int buflen, int *offset) {
1651 // buf is not optional, but offset is optional
1652 assert(buf != NULL, "sanity check");
1653
1654 Dl_info dlinfo;
1655
1656 if (dladdr((void*)addr, &dlinfo) != 0) {
1657 // see if we have a matching symbol
1658 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1659 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1660 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1661 }
1662 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1663 return true;
1664 }
1665 // no matching symbol so try for just file info
1666 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1667 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1668 buf, buflen, offset, dlinfo.dli_fname)) {
1669 return true;
1670 }
1671 }
1672 }
1673
1674 buf[0] = '\0';
1675 if (offset != NULL) *offset = -1;
1676 return false;
1677 }
1678
1679 struct _address_to_library_name {
1680 address addr; // input : memory address
1681 size_t buflen; // size of fname
1682 char* fname; // output: library name
1683 address base; // library base addr
1684 };
1685
1686 static int address_to_library_name_callback(struct dl_phdr_info *info,
1687 size_t size, void *data) {
1688 int i;
1689 bool found = false;
1690 address libbase = NULL;
1691 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1692
1693 // iterate through all loadable segments
1694 for (i = 0; i < info->dlpi_phnum; i++) {
1695 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1696 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1697 // base address of a library is the lowest address of its loaded
1698 // segments.
1699 if (libbase == NULL || libbase > segbase) {
1700 libbase = segbase;
1701 }
1702 // see if 'addr' is within current segment
1703 if (segbase <= d->addr &&
1704 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1705 found = true;
1706 }
1707 }
1708 }
1709
1710 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1711 // so dll_address_to_library_name() can fall through to use dladdr() which
1712 // can figure out executable name from argv[0].
1713 if (found && info->dlpi_name && info->dlpi_name[0]) {
1714 d->base = libbase;
1715 if (d->fname) {
1716 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1717 }
1718 return 1;
1719 }
1720 return 0;
1721 }
1722
1723 bool os::dll_address_to_library_name(address addr, char* buf,
1724 int buflen, int* offset) {
1725 // buf is not optional, but offset is optional
1726 assert(buf != NULL, "sanity check");
1727
1728 Dl_info dlinfo;
1729 struct _address_to_library_name data;
1730
1731 // There is a bug in old glibc dladdr() implementation that it could resolve
1732 // to wrong library name if the .so file has a base address != NULL. Here
1733 // we iterate through the program headers of all loaded libraries to find
1734 // out which library 'addr' really belongs to. This workaround can be
1735 // removed once the minimum requirement for glibc is moved to 2.3.x.
1736 data.addr = addr;
1737 data.fname = buf;
1738 data.buflen = buflen;
1739 data.base = NULL;
1740 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1741
1742 if (rslt) {
1743 // buf already contains library name
1744 if (offset) *offset = addr - data.base;
1745 return true;
1746 }
1747 if (dladdr((void*)addr, &dlinfo) != 0) {
1748 if (dlinfo.dli_fname != NULL) {
1749 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1750 }
1751 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1752 *offset = addr - (address)dlinfo.dli_fbase;
1753 }
1754 return true;
1755 }
1756
1757 buf[0] = '\0';
1758 if (offset) *offset = -1;
1759 return false;
1760 }
1761
1762 // Loads .dll/.so and
1763 // in case of error it checks if .dll/.so was built for the
1764 // same architecture as Hotspot is running on
1765
1766
1767 // Remember the stack's state. The Linux dynamic linker will change
1768 // the stack to 'executable' at most once, so we must safepoint only once.
1769 bool os::Linux::_stack_is_executable = false;
1770
1771 // VM operation that loads a library. This is necessary if stack protection
1772 // of the Java stacks can be lost during loading the library. If we
1773 // do not stop the Java threads, they can stack overflow before the stacks
1774 // are protected again.
1775 class VM_LinuxDllLoad: public VM_Operation {
1776 private:
1777 const char *_filename;
1778 char *_ebuf;
1779 int _ebuflen;
1780 void *_lib;
1781 public:
1782 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1783 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1784 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1785 void doit() {
1786 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1787 os::Linux::_stack_is_executable = true;
1788 }
1789 void* loaded_library() { return _lib; }
1790 };
1791
1792 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1793 void * result = NULL;
1794 bool load_attempted = false;
1795
1796 // Check whether the library to load might change execution rights
1797 // of the stack. If they are changed, the protection of the stack
1798 // guard pages will be lost. We need a safepoint to fix this.
1799 //
1800 // See Linux man page execstack(8) for more info.
1801 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1802 ElfFile ef(filename);
1803 if (!ef.specifies_noexecstack()) {
1804 if (!is_init_completed()) {
1805 os::Linux::_stack_is_executable = true;
1806 // This is OK - No Java threads have been created yet, and hence no
1807 // stack guard pages to fix.
1808 //
1809 // This should happen only when you are building JDK7 using a very
1810 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1811 //
1812 // Dynamic loader will make all stacks executable after
1813 // this function returns, and will not do that again.
1814 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1815 } else {
1816 warning("You have loaded library %s which might have disabled stack guard. "
1817 "The VM will try to fix the stack guard now.\n"
1818 "It's highly recommended that you fix the library with "
1819 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1820 filename);
1821
1822 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1823 JavaThread *jt = JavaThread::current();
1824 if (jt->thread_state() != _thread_in_native) {
1825 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1826 // that requires ExecStack. Cannot enter safe point. Let's give up.
1827 warning("Unable to fix stack guard. Giving up.");
1828 } else {
1829 if (!LoadExecStackDllInVMThread) {
1830 // This is for the case where the DLL has an static
1831 // constructor function that executes JNI code. We cannot
1832 // load such DLLs in the VMThread.
1833 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1834 }
1835
1836 ThreadInVMfromNative tiv(jt);
1837 debug_only(VMNativeEntryWrapper vew;)
1838
1839 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1840 VMThread::execute(&op);
1841 if (LoadExecStackDllInVMThread) {
1842 result = op.loaded_library();
1843 }
1844 load_attempted = true;
1845 }
1846 }
1847 }
1848 }
1849
1850 if (!load_attempted) {
1851 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1852 }
1853
1854 if (result != NULL) {
1855 // Successful loading
1856 return result;
1857 }
1858
1859 Elf32_Ehdr elf_head;
1860 int diag_msg_max_length=ebuflen-strlen(ebuf);
1861 char* diag_msg_buf=ebuf+strlen(ebuf);
1862
1863 if (diag_msg_max_length==0) {
1864 // No more space in ebuf for additional diagnostics message
1865 return NULL;
1866 }
1867
1868
1869 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1870
1871 if (file_descriptor < 0) {
1872 // Can't open library, report dlerror() message
1873 return NULL;
1874 }
1875
1876 bool failed_to_read_elf_head=
1877 (sizeof(elf_head)!=
1878 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1879
1880 ::close(file_descriptor);
1881 if (failed_to_read_elf_head) {
1882 // file i/o error - report dlerror() msg
1883 return NULL;
1884 }
1885
1886 typedef struct {
1887 Elf32_Half code; // Actual value as defined in elf.h
1888 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1889 char elf_class; // 32 or 64 bit
1890 char endianess; // MSB or LSB
1891 char* name; // String representation
1892 } arch_t;
1893
1894 #ifndef EM_486
1895 #define EM_486 6 /* Intel 80486 */
1896 #endif
1897
1898 static const arch_t arch_array[]={
1899 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1900 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1901 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1902 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1903 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1904 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1905 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1906 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1907 #if defined(VM_LITTLE_ENDIAN)
1908 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1909 #else
1910 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1911 #endif
1912 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1913 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1914 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1915 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1916 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1917 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1918 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1919 };
1920
1921 #if (defined IA32)
1922 static Elf32_Half running_arch_code=EM_386;
1923 #elif (defined AMD64)
1924 static Elf32_Half running_arch_code=EM_X86_64;
1925 #elif (defined IA64)
1926 static Elf32_Half running_arch_code=EM_IA_64;
1927 #elif (defined __sparc) && (defined _LP64)
1928 static Elf32_Half running_arch_code=EM_SPARCV9;
1929 #elif (defined __sparc) && (!defined _LP64)
1930 static Elf32_Half running_arch_code=EM_SPARC;
1931 #elif (defined __powerpc64__)
1932 static Elf32_Half running_arch_code=EM_PPC64;
1933 #elif (defined __powerpc__)
1934 static Elf32_Half running_arch_code=EM_PPC;
1935 #elif (defined ARM)
1936 static Elf32_Half running_arch_code=EM_ARM;
1937 #elif (defined S390)
1938 static Elf32_Half running_arch_code=EM_S390;
1939 #elif (defined ALPHA)
1940 static Elf32_Half running_arch_code=EM_ALPHA;
1941 #elif (defined MIPSEL)
1942 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1943 #elif (defined PARISC)
1944 static Elf32_Half running_arch_code=EM_PARISC;
1945 #elif (defined MIPS)
1946 static Elf32_Half running_arch_code=EM_MIPS;
1947 #elif (defined M68K)
1948 static Elf32_Half running_arch_code=EM_68K;
1949 #else
1950 #error Method os::dll_load requires that one of following is defined:\
1951 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1952 #endif
1953
1954 // Identify compatability class for VM's architecture and library's architecture
1955 // Obtain string descriptions for architectures
1956
1957 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1958 int running_arch_index=-1;
1959
1960 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1961 if (running_arch_code == arch_array[i].code) {
1962 running_arch_index = i;
1963 }
1964 if (lib_arch.code == arch_array[i].code) {
1965 lib_arch.compat_class = arch_array[i].compat_class;
1966 lib_arch.name = arch_array[i].name;
1967 }
1968 }
1969
1970 assert(running_arch_index != -1,
1971 "Didn't find running architecture code (running_arch_code) in arch_array");
1972 if (running_arch_index == -1) {
1973 // Even though running architecture detection failed
1974 // we may still continue with reporting dlerror() message
1975 return NULL;
1976 }
1977
1978 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1979 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1980 return NULL;
1981 }
1982
1983 #ifndef S390
1984 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1985 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1986 return NULL;
1987 }
1988 #endif // !S390
1989
1990 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1991 if (lib_arch.name!=NULL) {
1992 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1993 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1994 lib_arch.name, arch_array[running_arch_index].name);
1995 } else {
1996 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1997 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1998 lib_arch.code,
1999 arch_array[running_arch_index].name);
2000 }
2001 }
2002
2003 return NULL;
2004 }
2005
2006 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2007 int ebuflen) {
2008 void * result = ::dlopen(filename, RTLD_LAZY);
2009 if (result == NULL) {
2010 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2011 ebuf[ebuflen-1] = '\0';
2012 }
2013 return result;
2014 }
2015
2016 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2017 int ebuflen) {
2018 void * result = NULL;
2019 if (LoadExecStackDllInVMThread) {
2020 result = dlopen_helper(filename, ebuf, ebuflen);
2021 }
2022
2023 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2024 // library that requires an executable stack, or which does not have this
2025 // stack attribute set, dlopen changes the stack attribute to executable. The
2026 // read protection of the guard pages gets lost.
2027 //
2028 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2029 // may have been queued at the same time.
2030
2031 if (!_stack_is_executable) {
2032 JavaThread *jt = Threads::first();
2033
2034 while (jt) {
2035 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2036 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2037 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2038 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2039 warning("Attempt to reguard stack yellow zone failed.");
2040 }
2041 }
2042 jt = jt->next();
2043 }
2044 }
2045
2046 return result;
2047 }
2048
2049 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2050 // chances are you might want to run the generated bits against glibc-2.0
2051 // libdl.so, so always use locking for any version of glibc.
2052 //
2053 void* os::dll_lookup(void* handle, const char* name) {
2054 pthread_mutex_lock(&dl_mutex);
2055 void* res = dlsym(handle, name);
2056 pthread_mutex_unlock(&dl_mutex);
2057 return res;
2058 }
2059
2060 void* os::get_default_process_handle() {
2061 return (void*)::dlopen(NULL, RTLD_LAZY);
2062 }
2063
2064 static bool _print_ascii_file(const char* filename, outputStream* st) {
2065 int fd = ::open(filename, O_RDONLY);
2066 if (fd == -1) {
2067 return false;
2068 }
2069
2070 char buf[32];
2071 int bytes;
2072 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2073 st->print_raw(buf, bytes);
2074 }
2075
2076 ::close(fd);
2077
2078 return true;
2079 }
2080
2081 void os::print_dll_info(outputStream *st) {
2082 st->print_cr("Dynamic libraries:");
2083
2084 char fname[32];
2085 pid_t pid = os::Linux::gettid();
2086
2087 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2088
2089 if (!_print_ascii_file(fname, st)) {
2090 st->print("Can not get library information for pid = %d\n", pid);
2091 }
2092 }
2093
2094 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2095 FILE *procmapsFile = NULL;
2096
2097 // Open the procfs maps file for the current process
2098 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2099 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2100 char line[PATH_MAX + 100];
2101
2102 // Read line by line from 'file'
2103 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2104 u8 base, top, offset, inode;
2105 char permissions[5];
2106 char device[6];
2107 char name[PATH_MAX + 1];
2108
2109 // Parse fields from line
2110 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2111
2112 // Filter by device id '00:00' so that we only get file system mapped files.
2113 if (strcmp(device, "00:00") != 0) {
2114
2115 // Call callback with the fields of interest
2116 if(callback(name, (address)base, (address)top, param)) {
2117 // Oops abort, callback aborted
2118 fclose(procmapsFile);
2119 return 1;
2120 }
2121 }
2122 }
2123 fclose(procmapsFile);
2124 }
2125 return 0;
2126 }
2127
2128 void os::print_os_info_brief(outputStream* st) {
2129 os::Linux::print_distro_info(st);
2130
2131 os::Posix::print_uname_info(st);
2132
2133 os::Linux::print_libversion_info(st);
2134
2135 }
2136
2137 void os::print_os_info(outputStream* st) {
2138 st->print("OS:");
2139
2140 os::Linux::print_distro_info(st);
2141
2142 os::Posix::print_uname_info(st);
2143
2144 // Print warning if unsafe chroot environment detected
2145 if (unsafe_chroot_detected) {
2146 st->print("WARNING!! ");
2147 st->print_cr("%s", unstable_chroot_error);
2148 }
2149
2150 os::Linux::print_libversion_info(st);
2151
2152 os::Posix::print_rlimit_info(st);
2153
2154 os::Posix::print_load_average(st);
2155
2156 os::Linux::print_full_memory_info(st);
2157 }
2158
2159 // Try to identify popular distros.
2160 // Most Linux distributions have a /etc/XXX-release file, which contains
2161 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2162 // file that also contains the OS version string. Some have more than one
2163 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2164 // /etc/redhat-release.), so the order is important.
2165 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2166 // their own specific XXX-release file as well as a redhat-release file.
2167 // Because of this the XXX-release file needs to be searched for before the
2168 // redhat-release file.
2169 // Since Red Hat has a lsb-release file that is not very descriptive the
2170 // search for redhat-release needs to be before lsb-release.
2171 // Since the lsb-release file is the new standard it needs to be searched
2172 // before the older style release files.
2173 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2174 // next to last resort. The os-release file is a new standard that contains
2175 // distribution information and the system-release file seems to be an old
2176 // standard that has been replaced by the lsb-release and os-release files.
2177 // Searching for the debian_version file is the last resort. It contains
2178 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2179 // "Debian " is printed before the contents of the debian_version file.
2180 void os::Linux::print_distro_info(outputStream* st) {
2181 if (!_print_ascii_file("/etc/oracle-release", st) &&
2182 !_print_ascii_file("/etc/mandriva-release", st) &&
2183 !_print_ascii_file("/etc/mandrake-release", st) &&
2184 !_print_ascii_file("/etc/sun-release", st) &&
2185 !_print_ascii_file("/etc/redhat-release", st) &&
2186 !_print_ascii_file("/etc/lsb-release", st) &&
2187 !_print_ascii_file("/etc/SuSE-release", st) &&
2188 !_print_ascii_file("/etc/turbolinux-release", st) &&
2189 !_print_ascii_file("/etc/gentoo-release", st) &&
2190 !_print_ascii_file("/etc/ltib-release", st) &&
2191 !_print_ascii_file("/etc/angstrom-version", st) &&
2192 !_print_ascii_file("/etc/system-release", st) &&
2193 !_print_ascii_file("/etc/os-release", st)) {
2194
2195 if (file_exists("/etc/debian_version")) {
2196 st->print("Debian ");
2197 _print_ascii_file("/etc/debian_version", st);
2198 } else {
2199 st->print("Linux");
2200 }
2201 }
2202 st->cr();
2203 }
2204
2205 void os::Linux::print_libversion_info(outputStream* st) {
2206 // libc, pthread
2207 st->print("libc:");
2208 st->print("%s ", os::Linux::glibc_version());
2209 st->print("%s ", os::Linux::libpthread_version());
2210 if (os::Linux::is_LinuxThreads()) {
2211 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2212 }
2213 st->cr();
2214 }
2215
2216 void os::Linux::print_full_memory_info(outputStream* st) {
2217 st->print("\n/proc/meminfo:\n");
2218 _print_ascii_file("/proc/meminfo", st);
2219 st->cr();
2220 }
2221
2222 void os::print_memory_info(outputStream* st) {
2223
2224 st->print("Memory:");
2225 st->print(" %dk page", os::vm_page_size()>>10);
2226
2227 // values in struct sysinfo are "unsigned long"
2228 struct sysinfo si;
2229 sysinfo(&si);
2230
2231 st->print(", physical " UINT64_FORMAT "k",
2232 os::physical_memory() >> 10);
2233 st->print("(" UINT64_FORMAT "k free)",
2234 os::available_memory() >> 10);
2235 st->print(", swap " UINT64_FORMAT "k",
2236 ((jlong)si.totalswap * si.mem_unit) >> 10);
2237 st->print("(" UINT64_FORMAT "k free)",
2238 ((jlong)si.freeswap * si.mem_unit) >> 10);
2239 st->cr();
2240 }
2241
2242 void os::pd_print_cpu_info(outputStream* st) {
2243 st->print("\n/proc/cpuinfo:\n");
2244 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2245 st->print(" <Not Available>");
2246 }
2247 st->cr();
2248 }
2249
2250 void os::print_siginfo(outputStream* st, void* siginfo) {
2251 const siginfo_t* si = (const siginfo_t*)siginfo;
2252
2253 os::Posix::print_siginfo_brief(st, si);
2254 #if INCLUDE_CDS
2255 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2256 UseSharedSpaces) {
2257 FileMapInfo* mapinfo = FileMapInfo::current_info();
2258 if (mapinfo->is_in_shared_space(si->si_addr)) {
2259 st->print("\n\nError accessing class data sharing archive." \
2260 " Mapped file inaccessible during execution, " \
2261 " possible disk/network problem.");
2262 }
2263 }
2264 #endif
2265 st->cr();
2266 }
2267
2268
2269 static void print_signal_handler(outputStream* st, int sig,
2270 char* buf, size_t buflen);
2271
2272 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2273 st->print_cr("Signal Handlers:");
2274 print_signal_handler(st, SIGSEGV, buf, buflen);
2275 print_signal_handler(st, SIGBUS , buf, buflen);
2276 print_signal_handler(st, SIGFPE , buf, buflen);
2277 print_signal_handler(st, SIGPIPE, buf, buflen);
2278 print_signal_handler(st, SIGXFSZ, buf, buflen);
2279 print_signal_handler(st, SIGILL , buf, buflen);
2280 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2281 print_signal_handler(st, SR_signum, buf, buflen);
2282 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2283 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2284 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2285 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2286 #if defined(PPC64)
2287 print_signal_handler(st, SIGTRAP, buf, buflen);
2288 #endif
2289 }
2290
2291 static char saved_jvm_path[MAXPATHLEN] = {0};
2292
2293 // Find the full path to the current module, libjvm.so
2294 void os::jvm_path(char *buf, jint buflen) {
2295 // Error checking.
2296 if (buflen < MAXPATHLEN) {
2297 assert(false, "must use a large-enough buffer");
2298 buf[0] = '\0';
2299 return;
2300 }
2301 // Lazy resolve the path to current module.
2302 if (saved_jvm_path[0] != 0) {
2303 strcpy(buf, saved_jvm_path);
2304 return;
2305 }
2306
2307 char dli_fname[MAXPATHLEN];
2308 bool ret = dll_address_to_library_name(
2309 CAST_FROM_FN_PTR(address, os::jvm_path),
2310 dli_fname, sizeof(dli_fname), NULL);
2311 assert(ret, "cannot locate libjvm");
2312 char *rp = NULL;
2313 if (ret && dli_fname[0] != '\0') {
2314 rp = realpath(dli_fname, buf);
2315 }
2316 if (rp == NULL) {
2317 return;
2318 }
2319
2320 if (Arguments::sun_java_launcher_is_altjvm()) {
2321 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2322 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2323 // If "/jre/lib/" appears at the right place in the string, then
2324 // assume we are installed in a JDK and we're done. Otherwise, check
2325 // for a JAVA_HOME environment variable and fix up the path so it
2326 // looks like libjvm.so is installed there (append a fake suffix
2327 // hotspot/libjvm.so).
2328 const char *p = buf + strlen(buf) - 1;
2329 for (int count = 0; p > buf && count < 5; ++count) {
2330 for (--p; p > buf && *p != '/'; --p)
2331 /* empty */ ;
2332 }
2333
2334 if (strncmp(p, "/jre/lib/", 9) != 0) {
2335 // Look for JAVA_HOME in the environment.
2336 char* java_home_var = ::getenv("JAVA_HOME");
2337 if (java_home_var != NULL && java_home_var[0] != 0) {
2338 char* jrelib_p;
2339 int len;
2340
2341 // Check the current module name "libjvm.so".
2342 p = strrchr(buf, '/');
2343 if (p == NULL) {
2344 return;
2345 }
2346 assert(strstr(p, "/libjvm") == p, "invalid library name");
2347
2348 rp = realpath(java_home_var, buf);
2349 if (rp == NULL) {
2350 return;
2351 }
2352
2353 // determine if this is a legacy image or modules image
2354 // modules image doesn't have "jre" subdirectory
2355 len = strlen(buf);
2356 assert(len < buflen, "Ran out of buffer room");
2357 jrelib_p = buf + len;
2358 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2359 if (0 != access(buf, F_OK)) {
2360 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2361 }
2362
2363 if (0 == access(buf, F_OK)) {
2364 // Use current module name "libjvm.so"
2365 len = strlen(buf);
2366 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2367 } else {
2368 // Go back to path of .so
2369 rp = realpath(dli_fname, buf);
2370 if (rp == NULL) {
2371 return;
2372 }
2373 }
2374 }
2375 }
2376 }
2377
2378 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2379 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2380 }
2381
2382 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2383 // no prefix required, not even "_"
2384 }
2385
2386 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2387 // no suffix required
2388 }
2389
2390 ////////////////////////////////////////////////////////////////////////////////
2391 // sun.misc.Signal support
2392
2393 static volatile jint sigint_count = 0;
2394
2395 static void UserHandler(int sig, void *siginfo, void *context) {
2396 // 4511530 - sem_post is serialized and handled by the manager thread. When
2397 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2398 // don't want to flood the manager thread with sem_post requests.
2399 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2400 return;
2401 }
2402
2403 // Ctrl-C is pressed during error reporting, likely because the error
2404 // handler fails to abort. Let VM die immediately.
2405 if (sig == SIGINT && is_error_reported()) {
2406 os::die();
2407 }
2408
2409 os::signal_notify(sig);
2410 }
2411
2412 void* os::user_handler() {
2413 return CAST_FROM_FN_PTR(void*, UserHandler);
2414 }
2415
2416 class Semaphore : public StackObj {
2417 public:
2418 Semaphore();
2419 ~Semaphore();
2420 void signal();
2421 void wait();
2422 bool trywait();
2423 bool timedwait(unsigned int sec, int nsec);
2424 private:
2425 sem_t _semaphore;
2426 };
2427
2428 Semaphore::Semaphore() {
2429 sem_init(&_semaphore, 0, 0);
2430 }
2431
2432 Semaphore::~Semaphore() {
2433 sem_destroy(&_semaphore);
2434 }
2435
2436 void Semaphore::signal() {
2437 sem_post(&_semaphore);
2438 }
2439
2440 void Semaphore::wait() {
2441 sem_wait(&_semaphore);
2442 }
2443
2444 bool Semaphore::trywait() {
2445 return sem_trywait(&_semaphore) == 0;
2446 }
2447
2448 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2449
2450 struct timespec ts;
2451 // Semaphore's are always associated with CLOCK_REALTIME
2452 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2453 // see unpackTime for discussion on overflow checking
2454 if (sec >= MAX_SECS) {
2455 ts.tv_sec += MAX_SECS;
2456 ts.tv_nsec = 0;
2457 } else {
2458 ts.tv_sec += sec;
2459 ts.tv_nsec += nsec;
2460 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2461 ts.tv_nsec -= NANOSECS_PER_SEC;
2462 ++ts.tv_sec; // note: this must be <= max_secs
2463 }
2464 }
2465
2466 while (1) {
2467 int result = sem_timedwait(&_semaphore, &ts);
2468 if (result == 0) {
2469 return true;
2470 } else if (errno == EINTR) {
2471 continue;
2472 } else if (errno == ETIMEDOUT) {
2473 return false;
2474 } else {
2475 return false;
2476 }
2477 }
2478 }
2479
2480 extern "C" {
2481 typedef void (*sa_handler_t)(int);
2482 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2483 }
2484
2485 void* os::signal(int signal_number, void* handler) {
2486 struct sigaction sigAct, oldSigAct;
2487
2488 sigfillset(&(sigAct.sa_mask));
2489 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2490 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2491
2492 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2493 // -1 means registration failed
2494 return (void *)-1;
2495 }
2496
2497 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2498 }
2499
2500 void os::signal_raise(int signal_number) {
2501 ::raise(signal_number);
2502 }
2503
2504 // The following code is moved from os.cpp for making this
2505 // code platform specific, which it is by its very nature.
2506
2507 // Will be modified when max signal is changed to be dynamic
2508 int os::sigexitnum_pd() {
2509 return NSIG;
2510 }
2511
2512 // a counter for each possible signal value
2513 static volatile jint pending_signals[NSIG+1] = { 0 };
2514
2515 // Linux(POSIX) specific hand shaking semaphore.
2516 static sem_t sig_sem;
2517 static Semaphore sr_semaphore;
2518
2519 void os::signal_init_pd() {
2520 // Initialize signal structures
2521 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2522
2523 // Initialize signal semaphore
2524 ::sem_init(&sig_sem, 0, 0);
2525 }
2526
2527 void os::signal_notify(int sig) {
2528 Atomic::inc(&pending_signals[sig]);
2529 ::sem_post(&sig_sem);
2530 }
2531
2532 static int check_pending_signals(bool wait) {
2533 Atomic::store(0, &sigint_count);
2534 for (;;) {
2535 for (int i = 0; i < NSIG + 1; i++) {
2536 jint n = pending_signals[i];
2537 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2538 return i;
2539 }
2540 }
2541 if (!wait) {
2542 return -1;
2543 }
2544 JavaThread *thread = JavaThread::current();
2545 ThreadBlockInVM tbivm(thread);
2546
2547 bool threadIsSuspended;
2548 do {
2549 thread->set_suspend_equivalent();
2550 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2551 ::sem_wait(&sig_sem);
2552
2553 // were we externally suspended while we were waiting?
2554 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2555 if (threadIsSuspended) {
2556 // The semaphore has been incremented, but while we were waiting
2557 // another thread suspended us. We don't want to continue running
2558 // while suspended because that would surprise the thread that
2559 // suspended us.
2560 ::sem_post(&sig_sem);
2561
2562 thread->java_suspend_self();
2563 }
2564 } while (threadIsSuspended);
2565 }
2566 }
2567
2568 int os::signal_lookup() {
2569 return check_pending_signals(false);
2570 }
2571
2572 int os::signal_wait() {
2573 return check_pending_signals(true);
2574 }
2575
2576 ////////////////////////////////////////////////////////////////////////////////
2577 // Virtual Memory
2578
2579 int os::vm_page_size() {
2580 // Seems redundant as all get out
2581 assert(os::Linux::page_size() != -1, "must call os::init");
2582 return os::Linux::page_size();
2583 }
2584
2585 // Solaris allocates memory by pages.
2586 int os::vm_allocation_granularity() {
2587 assert(os::Linux::page_size() != -1, "must call os::init");
2588 return os::Linux::page_size();
2589 }
2590
2591 // Rationale behind this function:
2592 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2593 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2594 // samples for JITted code. Here we create private executable mapping over the code cache
2595 // and then we can use standard (well, almost, as mapping can change) way to provide
2596 // info for the reporting script by storing timestamp and location of symbol
2597 void linux_wrap_code(char* base, size_t size) {
2598 static volatile jint cnt = 0;
2599
2600 if (!UseOprofile) {
2601 return;
2602 }
2603
2604 char buf[PATH_MAX+1];
2605 int num = Atomic::add(1, &cnt);
2606
2607 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2608 os::get_temp_directory(), os::current_process_id(), num);
2609 unlink(buf);
2610
2611 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2612
2613 if (fd != -1) {
2614 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2615 if (rv != (off_t)-1) {
2616 if (::write(fd, "", 1) == 1) {
2617 mmap(base, size,
2618 PROT_READ|PROT_WRITE|PROT_EXEC,
2619 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2620 }
2621 }
2622 ::close(fd);
2623 unlink(buf);
2624 }
2625 }
2626
2627 static bool recoverable_mmap_error(int err) {
2628 // See if the error is one we can let the caller handle. This
2629 // list of errno values comes from JBS-6843484. I can't find a
2630 // Linux man page that documents this specific set of errno
2631 // values so while this list currently matches Solaris, it may
2632 // change as we gain experience with this failure mode.
2633 switch (err) {
2634 case EBADF:
2635 case EINVAL:
2636 case ENOTSUP:
2637 // let the caller deal with these errors
2638 return true;
2639
2640 default:
2641 // Any remaining errors on this OS can cause our reserved mapping
2642 // to be lost. That can cause confusion where different data
2643 // structures think they have the same memory mapped. The worst
2644 // scenario is if both the VM and a library think they have the
2645 // same memory mapped.
2646 return false;
2647 }
2648 }
2649
2650 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2651 int err) {
2652 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2653 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2654 strerror(err), err);
2655 }
2656
2657 static void warn_fail_commit_memory(char* addr, size_t size,
2658 size_t alignment_hint, bool exec,
2659 int err) {
2660 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2661 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2662 alignment_hint, exec, strerror(err), err);
2663 }
2664
2665 // NOTE: Linux kernel does not really reserve the pages for us.
2666 // All it does is to check if there are enough free pages
2667 // left at the time of mmap(). This could be a potential
2668 // problem.
2669 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2670 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2671 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2672 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2673 if (res != (uintptr_t) MAP_FAILED) {
2674 if (UseNUMAInterleaving) {
2675 numa_make_global(addr, size);
2676 }
2677 return 0;
2678 }
2679
2680 int err = errno; // save errno from mmap() call above
2681
2682 if (!recoverable_mmap_error(err)) {
2683 warn_fail_commit_memory(addr, size, exec, err);
2684 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2685 }
2686
2687 return err;
2688 }
2689
2690 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2691 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2692 }
2693
2694 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2695 const char* mesg) {
2696 assert(mesg != NULL, "mesg must be specified");
2697 int err = os::Linux::commit_memory_impl(addr, size, exec);
2698 if (err != 0) {
2699 // the caller wants all commit errors to exit with the specified mesg:
2700 warn_fail_commit_memory(addr, size, exec, err);
2701 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2702 }
2703 }
2704
2705 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2706 #ifndef MAP_HUGETLB
2707 #define MAP_HUGETLB 0x40000
2708 #endif
2709
2710 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2711 #ifndef MADV_HUGEPAGE
2712 #define MADV_HUGEPAGE 14
2713 #endif
2714
2715 int os::Linux::commit_memory_impl(char* addr, size_t size,
2716 size_t alignment_hint, bool exec) {
2717 int err = os::Linux::commit_memory_impl(addr, size, exec);
2718 if (err == 0) {
2719 realign_memory(addr, size, alignment_hint);
2720 }
2721 return err;
2722 }
2723
2724 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2725 bool exec) {
2726 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2727 }
2728
2729 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2730 size_t alignment_hint, bool exec,
2731 const char* mesg) {
2732 assert(mesg != NULL, "mesg must be specified");
2733 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2734 if (err != 0) {
2735 // the caller wants all commit errors to exit with the specified mesg:
2736 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2737 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2738 }
2739 }
2740
2741 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2742 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2743 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2744 // be supported or the memory may already be backed by huge pages.
2745 ::madvise(addr, bytes, MADV_HUGEPAGE);
2746 }
2747 }
2748
2749 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2750 // This method works by doing an mmap over an existing mmaping and effectively discarding
2751 // the existing pages. However it won't work for SHM-based large pages that cannot be
2752 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2753 // small pages on top of the SHM segment. This method always works for small pages, so we
2754 // allow that in any case.
2755 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2756 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2757 }
2758 }
2759
2760 void os::numa_make_global(char *addr, size_t bytes) {
2761 Linux::numa_interleave_memory(addr, bytes);
2762 }
2763
2764 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2765 // bind policy to MPOL_PREFERRED for the current thread.
2766 #define USE_MPOL_PREFERRED 0
2767
2768 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2769 // To make NUMA and large pages more robust when both enabled, we need to ease
2770 // the requirements on where the memory should be allocated. MPOL_BIND is the
2771 // default policy and it will force memory to be allocated on the specified
2772 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2773 // the specified node, but will not force it. Using this policy will prevent
2774 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2775 // free large pages.
2776 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2777 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2778 }
2779
2780 bool os::numa_topology_changed() { return false; }
2781
2782 size_t os::numa_get_groups_num() {
2783 int max_node = Linux::numa_max_node();
2784 return max_node > 0 ? max_node + 1 : 1;
2785 }
2786
2787 int os::numa_get_group_id() {
2788 int cpu_id = Linux::sched_getcpu();
2789 if (cpu_id != -1) {
2790 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2791 if (lgrp_id != -1) {
2792 return lgrp_id;
2793 }
2794 }
2795 return 0;
2796 }
2797
2798 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2799 for (size_t i = 0; i < size; i++) {
2800 ids[i] = i;
2801 }
2802 return size;
2803 }
2804
2805 bool os::get_page_info(char *start, page_info* info) {
2806 return false;
2807 }
2808
2809 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2810 page_info* page_found) {
2811 return end;
2812 }
2813
2814
2815 int os::Linux::sched_getcpu_syscall(void) {
2816 unsigned int cpu;
2817 int retval = -1;
2818
2819 #if defined(IA32)
2820 #ifndef SYS_getcpu
2821 #define SYS_getcpu 318
2822 #endif
2823 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2824 #elif defined(AMD64)
2825 // Unfortunately we have to bring all these macros here from vsyscall.h
2826 // to be able to compile on old linuxes.
2827 #define __NR_vgetcpu 2
2828 #define VSYSCALL_START (-10UL << 20)
2829 #define VSYSCALL_SIZE 1024
2830 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2831 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2832 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2833 retval = vgetcpu(&cpu, NULL, NULL);
2834 #endif
2835
2836 return (retval == -1) ? retval : cpu;
2837 }
2838
2839 // Something to do with the numa-aware allocator needs these symbols
2840 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2841 extern "C" JNIEXPORT void numa_error(char *where) { }
2842 extern "C" JNIEXPORT int fork1() { return fork(); }
2843
2844
2845 // If we are running with libnuma version > 2, then we should
2846 // be trying to use symbols with versions 1.1
2847 // If we are running with earlier version, which did not have symbol versions,
2848 // we should use the base version.
2849 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2850 void *f = dlvsym(handle, name, "libnuma_1.1");
2851 if (f == NULL) {
2852 f = dlsym(handle, name);
2853 }
2854 return f;
2855 }
2856
2857 bool os::Linux::libnuma_init() {
2858 // sched_getcpu() should be in libc.
2859 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2860 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2861
2862 // If it's not, try a direct syscall.
2863 if (sched_getcpu() == -1) {
2864 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2865 (void*)&sched_getcpu_syscall));
2866 }
2867
2868 if (sched_getcpu() != -1) { // Does it work?
2869 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2870 if (handle != NULL) {
2871 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2872 libnuma_dlsym(handle, "numa_node_to_cpus")));
2873 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2874 libnuma_dlsym(handle, "numa_max_node")));
2875 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2876 libnuma_dlsym(handle, "numa_available")));
2877 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2878 libnuma_dlsym(handle, "numa_tonode_memory")));
2879 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2880 libnuma_dlsym(handle, "numa_interleave_memory")));
2881 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2882 libnuma_dlsym(handle, "numa_set_bind_policy")));
2883
2884
2885 if (numa_available() != -1) {
2886 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2887 // Create a cpu -> node mapping
2888 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2889 rebuild_cpu_to_node_map();
2890 return true;
2891 }
2892 }
2893 }
2894 return false;
2895 }
2896
2897 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2898 // The table is later used in get_node_by_cpu().
2899 void os::Linux::rebuild_cpu_to_node_map() {
2900 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2901 // in libnuma (possible values are starting from 16,
2902 // and continuing up with every other power of 2, but less
2903 // than the maximum number of CPUs supported by kernel), and
2904 // is a subject to change (in libnuma version 2 the requirements
2905 // are more reasonable) we'll just hardcode the number they use
2906 // in the library.
2907 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2908
2909 size_t cpu_num = os::active_processor_count();
2910 size_t cpu_map_size = NCPUS / BitsPerCLong;
2911 size_t cpu_map_valid_size =
2912 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2913
2914 cpu_to_node()->clear();
2915 cpu_to_node()->at_grow(cpu_num - 1);
2916 size_t node_num = numa_get_groups_num();
2917
2918 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2919 for (size_t i = 0; i < node_num; i++) {
2920 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2921 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2922 if (cpu_map[j] != 0) {
2923 for (size_t k = 0; k < BitsPerCLong; k++) {
2924 if (cpu_map[j] & (1UL << k)) {
2925 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2926 }
2927 }
2928 }
2929 }
2930 }
2931 }
2932 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2933 }
2934
2935 int os::Linux::get_node_by_cpu(int cpu_id) {
2936 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2937 return cpu_to_node()->at(cpu_id);
2938 }
2939 return -1;
2940 }
2941
2942 GrowableArray<int>* os::Linux::_cpu_to_node;
2943 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2944 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2945 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2946 os::Linux::numa_available_func_t os::Linux::_numa_available;
2947 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2948 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2949 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2950 unsigned long* os::Linux::_numa_all_nodes;
2951
2952 bool os::pd_uncommit_memory(char* addr, size_t size) {
2953 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2954 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2955 return res != (uintptr_t) MAP_FAILED;
2956 }
2957
2958 static address get_stack_commited_bottom(address bottom, size_t size) {
2959 address nbot = bottom;
2960 address ntop = bottom + size;
2961
2962 size_t page_sz = os::vm_page_size();
2963 unsigned pages = size / page_sz;
2964
2965 unsigned char vec[1];
2966 unsigned imin = 1, imax = pages + 1, imid;
2967 int mincore_return_value = 0;
2968
2969 assert(imin <= imax, "Unexpected page size");
2970
2971 while (imin < imax) {
2972 imid = (imax + imin) / 2;
2973 nbot = ntop - (imid * page_sz);
2974
2975 // Use a trick with mincore to check whether the page is mapped or not.
2976 // mincore sets vec to 1 if page resides in memory and to 0 if page
2977 // is swapped output but if page we are asking for is unmapped
2978 // it returns -1,ENOMEM
2979 mincore_return_value = mincore(nbot, page_sz, vec);
2980
2981 if (mincore_return_value == -1) {
2982 // Page is not mapped go up
2983 // to find first mapped page
2984 if (errno != EAGAIN) {
2985 assert(errno == ENOMEM, "Unexpected mincore errno");
2986 imax = imid;
2987 }
2988 } else {
2989 // Page is mapped go down
2990 // to find first not mapped page
2991 imin = imid + 1;
2992 }
2993 }
2994
2995 nbot = nbot + page_sz;
2996
2997 // Adjust stack bottom one page up if last checked page is not mapped
2998 if (mincore_return_value == -1) {
2999 nbot = nbot + page_sz;
3000 }
3001
3002 return nbot;
3003 }
3004
3005
3006 // Linux uses a growable mapping for the stack, and if the mapping for
3007 // the stack guard pages is not removed when we detach a thread the
3008 // stack cannot grow beyond the pages where the stack guard was
3009 // mapped. If at some point later in the process the stack expands to
3010 // that point, the Linux kernel cannot expand the stack any further
3011 // because the guard pages are in the way, and a segfault occurs.
3012 //
3013 // However, it's essential not to split the stack region by unmapping
3014 // a region (leaving a hole) that's already part of the stack mapping,
3015 // so if the stack mapping has already grown beyond the guard pages at
3016 // the time we create them, we have to truncate the stack mapping.
3017 // So, we need to know the extent of the stack mapping when
3018 // create_stack_guard_pages() is called.
3019
3020 // We only need this for stacks that are growable: at the time of
3021 // writing thread stacks don't use growable mappings (i.e. those
3022 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3023 // only applies to the main thread.
3024
3025 // If the (growable) stack mapping already extends beyond the point
3026 // where we're going to put our guard pages, truncate the mapping at
3027 // that point by munmap()ping it. This ensures that when we later
3028 // munmap() the guard pages we don't leave a hole in the stack
3029 // mapping. This only affects the main/initial thread
3030
3031 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3032 if (os::Linux::is_initial_thread()) {
3033 // As we manually grow stack up to bottom inside create_attached_thread(),
3034 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3035 // we don't need to do anything special.
3036 // Check it first, before calling heavy function.
3037 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3038 unsigned char vec[1];
3039
3040 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3041 // Fallback to slow path on all errors, including EAGAIN
3042 stack_extent = (uintptr_t) get_stack_commited_bottom(
3043 os::Linux::initial_thread_stack_bottom(),
3044 (size_t)addr - stack_extent);
3045 }
3046
3047 if (stack_extent < (uintptr_t)addr) {
3048 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3049 }
3050 }
3051
3052 return os::commit_memory(addr, size, !ExecMem);
3053 }
3054
3055 // If this is a growable mapping, remove the guard pages entirely by
3056 // munmap()ping them. If not, just call uncommit_memory(). This only
3057 // affects the main/initial thread, but guard against future OS changes
3058 // It's safe to always unmap guard pages for initial thread because we
3059 // always place it right after end of the mapped region
3060
3061 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3062 uintptr_t stack_extent, stack_base;
3063
3064 if (os::Linux::is_initial_thread()) {
3065 return ::munmap(addr, size) == 0;
3066 }
3067
3068 return os::uncommit_memory(addr, size);
3069 }
3070
3071 static address _highest_vm_reserved_address = NULL;
3072
3073 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3074 // at 'requested_addr'. If there are existing memory mappings at the same
3075 // location, however, they will be overwritten. If 'fixed' is false,
3076 // 'requested_addr' is only treated as a hint, the return value may or
3077 // may not start from the requested address. Unlike Linux mmap(), this
3078 // function returns NULL to indicate failure.
3079 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3080 char * addr;
3081 int flags;
3082
3083 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3084 if (fixed) {
3085 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3086 flags |= MAP_FIXED;
3087 }
3088
3089 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3090 // touch an uncommitted page. Otherwise, the read/write might
3091 // succeed if we have enough swap space to back the physical page.
3092 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3093 flags, -1, 0);
3094
3095 if (addr != MAP_FAILED) {
3096 // anon_mmap() should only get called during VM initialization,
3097 // don't need lock (actually we can skip locking even it can be called
3098 // from multiple threads, because _highest_vm_reserved_address is just a
3099 // hint about the upper limit of non-stack memory regions.)
3100 if ((address)addr + bytes > _highest_vm_reserved_address) {
3101 _highest_vm_reserved_address = (address)addr + bytes;
3102 }
3103 }
3104
3105 return addr == MAP_FAILED ? NULL : addr;
3106 }
3107
3108 // Don't update _highest_vm_reserved_address, because there might be memory
3109 // regions above addr + size. If so, releasing a memory region only creates
3110 // a hole in the address space, it doesn't help prevent heap-stack collision.
3111 //
3112 static int anon_munmap(char * addr, size_t size) {
3113 return ::munmap(addr, size) == 0;
3114 }
3115
3116 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3117 size_t alignment_hint) {
3118 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3119 }
3120
3121 bool os::pd_release_memory(char* addr, size_t size) {
3122 return anon_munmap(addr, size);
3123 }
3124
3125 static address highest_vm_reserved_address() {
3126 return _highest_vm_reserved_address;
3127 }
3128
3129 static bool linux_mprotect(char* addr, size_t size, int prot) {
3130 // Linux wants the mprotect address argument to be page aligned.
3131 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3132
3133 // According to SUSv3, mprotect() should only be used with mappings
3134 // established by mmap(), and mmap() always maps whole pages. Unaligned
3135 // 'addr' likely indicates problem in the VM (e.g. trying to change
3136 // protection of malloc'ed or statically allocated memory). Check the
3137 // caller if you hit this assert.
3138 assert(addr == bottom, "sanity check");
3139
3140 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3141 return ::mprotect(bottom, size, prot) == 0;
3142 }
3143
3144 // Set protections specified
3145 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3146 bool is_committed) {
3147 unsigned int p = 0;
3148 switch (prot) {
3149 case MEM_PROT_NONE: p = PROT_NONE; break;
3150 case MEM_PROT_READ: p = PROT_READ; break;
3151 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3152 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3153 default:
3154 ShouldNotReachHere();
3155 }
3156 // is_committed is unused.
3157 return linux_mprotect(addr, bytes, p);
3158 }
3159
3160 bool os::guard_memory(char* addr, size_t size) {
3161 return linux_mprotect(addr, size, PROT_NONE);
3162 }
3163
3164 bool os::unguard_memory(char* addr, size_t size) {
3165 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3166 }
3167
3168 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3169 size_t page_size) {
3170 bool result = false;
3171 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3172 MAP_ANONYMOUS|MAP_PRIVATE,
3173 -1, 0);
3174 if (p != MAP_FAILED) {
3175 void *aligned_p = align_ptr_up(p, page_size);
3176
3177 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3178
3179 munmap(p, page_size * 2);
3180 }
3181
3182 if (warn && !result) {
3183 warning("TransparentHugePages is not supported by the operating system.");
3184 }
3185
3186 return result;
3187 }
3188
3189 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3190 bool result = false;
3191 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3192 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3193 -1, 0);
3194
3195 if (p != MAP_FAILED) {
3196 // We don't know if this really is a huge page or not.
3197 FILE *fp = fopen("/proc/self/maps", "r");
3198 if (fp) {
3199 while (!feof(fp)) {
3200 char chars[257];
3201 long x = 0;
3202 if (fgets(chars, sizeof(chars), fp)) {
3203 if (sscanf(chars, "%lx-%*x", &x) == 1
3204 && x == (long)p) {
3205 if (strstr (chars, "hugepage")) {
3206 result = true;
3207 break;
3208 }
3209 }
3210 }
3211 }
3212 fclose(fp);
3213 }
3214 munmap(p, page_size);
3215 }
3216
3217 if (warn && !result) {
3218 warning("HugeTLBFS is not supported by the operating system.");
3219 }
3220
3221 return result;
3222 }
3223
3224 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3225 //
3226 // From the coredump_filter documentation:
3227 //
3228 // - (bit 0) anonymous private memory
3229 // - (bit 1) anonymous shared memory
3230 // - (bit 2) file-backed private memory
3231 // - (bit 3) file-backed shared memory
3232 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3233 // effective only if the bit 2 is cleared)
3234 // - (bit 5) hugetlb private memory
3235 // - (bit 6) hugetlb shared memory
3236 //
3237 static void set_coredump_filter(void) {
3238 FILE *f;
3239 long cdm;
3240
3241 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3242 return;
3243 }
3244
3245 if (fscanf(f, "%lx", &cdm) != 1) {
3246 fclose(f);
3247 return;
3248 }
3249
3250 rewind(f);
3251
3252 if ((cdm & LARGEPAGES_BIT) == 0) {
3253 cdm |= LARGEPAGES_BIT;
3254 fprintf(f, "%#lx", cdm);
3255 }
3256
3257 fclose(f);
3258 }
3259
3260 // Large page support
3261
3262 static size_t _large_page_size = 0;
3263
3264 size_t os::Linux::find_large_page_size() {
3265 size_t large_page_size = 0;
3266
3267 // large_page_size on Linux is used to round up heap size. x86 uses either
3268 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3269 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3270 // page as large as 256M.
3271 //
3272 // Here we try to figure out page size by parsing /proc/meminfo and looking
3273 // for a line with the following format:
3274 // Hugepagesize: 2048 kB
3275 //
3276 // If we can't determine the value (e.g. /proc is not mounted, or the text
3277 // format has been changed), we'll use the largest page size supported by
3278 // the processor.
3279
3280 #ifndef ZERO
3281 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3282 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3283 #endif // ZERO
3284
3285 FILE *fp = fopen("/proc/meminfo", "r");
3286 if (fp) {
3287 while (!feof(fp)) {
3288 int x = 0;
3289 char buf[16];
3290 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3291 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3292 large_page_size = x * K;
3293 break;
3294 }
3295 } else {
3296 // skip to next line
3297 for (;;) {
3298 int ch = fgetc(fp);
3299 if (ch == EOF || ch == (int)'\n') break;
3300 }
3301 }
3302 }
3303 fclose(fp);
3304 }
3305
3306 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3307 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3308 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3309 proper_unit_for_byte_size(large_page_size));
3310 }
3311
3312 return large_page_size;
3313 }
3314
3315 size_t os::Linux::setup_large_page_size() {
3316 _large_page_size = Linux::find_large_page_size();
3317 const size_t default_page_size = (size_t)Linux::page_size();
3318 if (_large_page_size > default_page_size) {
3319 _page_sizes[0] = _large_page_size;
3320 _page_sizes[1] = default_page_size;
3321 _page_sizes[2] = 0;
3322 }
3323
3324 return _large_page_size;
3325 }
3326
3327 bool os::Linux::setup_large_page_type(size_t page_size) {
3328 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3329 FLAG_IS_DEFAULT(UseSHM) &&
3330 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3331
3332 // The type of large pages has not been specified by the user.
3333
3334 // Try UseHugeTLBFS and then UseSHM.
3335 UseHugeTLBFS = UseSHM = true;
3336
3337 // Don't try UseTransparentHugePages since there are known
3338 // performance issues with it turned on. This might change in the future.
3339 UseTransparentHugePages = false;
3340 }
3341
3342 if (UseTransparentHugePages) {
3343 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3344 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3345 UseHugeTLBFS = false;
3346 UseSHM = false;
3347 return true;
3348 }
3349 UseTransparentHugePages = false;
3350 }
3351
3352 if (UseHugeTLBFS) {
3353 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3354 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3355 UseSHM = false;
3356 return true;
3357 }
3358 UseHugeTLBFS = false;
3359 }
3360
3361 return UseSHM;
3362 }
3363
3364 void os::large_page_init() {
3365 if (!UseLargePages &&
3366 !UseTransparentHugePages &&
3367 !UseHugeTLBFS &&
3368 !UseSHM) {
3369 // Not using large pages.
3370 return;
3371 }
3372
3373 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3374 // The user explicitly turned off large pages.
3375 // Ignore the rest of the large pages flags.
3376 UseTransparentHugePages = false;
3377 UseHugeTLBFS = false;
3378 UseSHM = false;
3379 return;
3380 }
3381
3382 size_t large_page_size = Linux::setup_large_page_size();
3383 UseLargePages = Linux::setup_large_page_type(large_page_size);
3384
3385 set_coredump_filter();
3386 }
3387
3388 #ifndef SHM_HUGETLB
3389 #define SHM_HUGETLB 04000
3390 #endif
3391
3392 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3393 char* req_addr, bool exec) {
3394 // "exec" is passed in but not used. Creating the shared image for
3395 // the code cache doesn't have an SHM_X executable permission to check.
3396 assert(UseLargePages && UseSHM, "only for SHM large pages");
3397 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3398
3399 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3400 return NULL; // Fallback to small pages.
3401 }
3402
3403 key_t key = IPC_PRIVATE;
3404 char *addr;
3405
3406 bool warn_on_failure = UseLargePages &&
3407 (!FLAG_IS_DEFAULT(UseLargePages) ||
3408 !FLAG_IS_DEFAULT(UseSHM) ||
3409 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3410 char msg[128];
3411
3412 // Create a large shared memory region to attach to based on size.
3413 // Currently, size is the total size of the heap
3414 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3415 if (shmid == -1) {
3416 // Possible reasons for shmget failure:
3417 // 1. shmmax is too small for Java heap.
3418 // > check shmmax value: cat /proc/sys/kernel/shmmax
3419 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3420 // 2. not enough large page memory.
3421 // > check available large pages: cat /proc/meminfo
3422 // > increase amount of large pages:
3423 // echo new_value > /proc/sys/vm/nr_hugepages
3424 // Note 1: different Linux may use different name for this property,
3425 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3426 // Note 2: it's possible there's enough physical memory available but
3427 // they are so fragmented after a long run that they can't
3428 // coalesce into large pages. Try to reserve large pages when
3429 // the system is still "fresh".
3430 if (warn_on_failure) {
3431 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3432 warning("%s", msg);
3433 }
3434 return NULL;
3435 }
3436
3437 // attach to the region
3438 addr = (char*)shmat(shmid, req_addr, 0);
3439 int err = errno;
3440
3441 // Remove shmid. If shmat() is successful, the actual shared memory segment
3442 // will be deleted when it's detached by shmdt() or when the process
3443 // terminates. If shmat() is not successful this will remove the shared
3444 // segment immediately.
3445 shmctl(shmid, IPC_RMID, NULL);
3446
3447 if ((intptr_t)addr == -1) {
3448 if (warn_on_failure) {
3449 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3450 warning("%s", msg);
3451 }
3452 return NULL;
3453 }
3454
3455 return addr;
3456 }
3457
3458 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3459 int error) {
3460 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3461
3462 bool warn_on_failure = UseLargePages &&
3463 (!FLAG_IS_DEFAULT(UseLargePages) ||
3464 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3465 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3466
3467 if (warn_on_failure) {
3468 char msg[128];
3469 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3470 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3471 warning("%s", msg);
3472 }
3473 }
3474
3475 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3476 char* req_addr,
3477 bool exec) {
3478 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3479 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3480 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3481
3482 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3483 char* addr = (char*)::mmap(req_addr, bytes, prot,
3484 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3485 -1, 0);
3486
3487 if (addr == MAP_FAILED) {
3488 warn_on_large_pages_failure(req_addr, bytes, errno);
3489 return NULL;
3490 }
3491
3492 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3493
3494 return addr;
3495 }
3496
3497 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3498 size_t alignment,
3499 char* req_addr,
3500 bool exec) {
3501 size_t large_page_size = os::large_page_size();
3502
3503 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3504
3505 // Allocate small pages.
3506
3507 char* start;
3508 if (req_addr != NULL) {
3509 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3510 assert(is_size_aligned(bytes, alignment), "Must be");
3511 start = os::reserve_memory(bytes, req_addr);
3512 assert(start == NULL || start == req_addr, "Must be");
3513 } else {
3514 start = os::reserve_memory_aligned(bytes, alignment);
3515 }
3516
3517 if (start == NULL) {
3518 return NULL;
3519 }
3520
3521 assert(is_ptr_aligned(start, alignment), "Must be");
3522
3523 if (MemTracker::tracking_level() > NMT_minimal) {
3524 // os::reserve_memory_special will record this memory area.
3525 // Need to release it here to prevent overlapping reservations.
3526 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3527 tkr.record((address)start, bytes);
3528 }
3529
3530 char* end = start + bytes;
3531
3532 // Find the regions of the allocated chunk that can be promoted to large pages.
3533 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3534 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3535
3536 size_t lp_bytes = lp_end - lp_start;
3537
3538 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3539
3540 if (lp_bytes == 0) {
3541 // The mapped region doesn't even span the start and the end of a large page.
3542 // Fall back to allocate a non-special area.
3543 ::munmap(start, end - start);
3544 return NULL;
3545 }
3546
3547 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3548
3549
3550 void* result;
3551
3552 if (start != lp_start) {
3553 result = ::mmap(start, lp_start - start, prot,
3554 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3555 -1, 0);
3556 if (result == MAP_FAILED) {
3557 ::munmap(lp_start, end - lp_start);
3558 return NULL;
3559 }
3560 }
3561
3562 result = ::mmap(lp_start, lp_bytes, prot,
3563 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3564 -1, 0);
3565 if (result == MAP_FAILED) {
3566 warn_on_large_pages_failure(req_addr, bytes, errno);
3567 // If the mmap above fails, the large pages region will be unmapped and we
3568 // have regions before and after with small pages. Release these regions.
3569 //
3570 // | mapped | unmapped | mapped |
3571 // ^ ^ ^ ^
3572 // start lp_start lp_end end
3573 //
3574 ::munmap(start, lp_start - start);
3575 ::munmap(lp_end, end - lp_end);
3576 return NULL;
3577 }
3578
3579 if (lp_end != end) {
3580 result = ::mmap(lp_end, end - lp_end, prot,
3581 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3582 -1, 0);
3583 if (result == MAP_FAILED) {
3584 ::munmap(start, lp_end - start);
3585 return NULL;
3586 }
3587 }
3588
3589 return start;
3590 }
3591
3592 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3593 size_t alignment,
3594 char* req_addr,
3595 bool exec) {
3596 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3597 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3598 assert(is_power_of_2(alignment), "Must be");
3599 assert(is_power_of_2(os::large_page_size()), "Must be");
3600 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3601
3602 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3603 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3604 } else {
3605 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3606 }
3607 }
3608
3609 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3610 char* req_addr, bool exec) {
3611 assert(UseLargePages, "only for large pages");
3612
3613 char* addr;
3614 if (UseSHM) {
3615 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3616 } else {
3617 assert(UseHugeTLBFS, "must be");
3618 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3619 }
3620
3621 if (addr != NULL) {
3622 if (UseNUMAInterleaving) {
3623 numa_make_global(addr, bytes);
3624 }
3625
3626 // The memory is committed
3627 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3628 }
3629
3630 return addr;
3631 }
3632
3633 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3634 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3635 return shmdt(base) == 0;
3636 }
3637
3638 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3639 return pd_release_memory(base, bytes);
3640 }
3641
3642 bool os::release_memory_special(char* base, size_t bytes) {
3643 bool res;
3644 if (MemTracker::tracking_level() > NMT_minimal) {
3645 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3646 res = os::Linux::release_memory_special_impl(base, bytes);
3647 if (res) {
3648 tkr.record((address)base, bytes);
3649 }
3650
3651 } else {
3652 res = os::Linux::release_memory_special_impl(base, bytes);
3653 }
3654 return res;
3655 }
3656
3657 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3658 assert(UseLargePages, "only for large pages");
3659 bool res;
3660
3661 if (UseSHM) {
3662 res = os::Linux::release_memory_special_shm(base, bytes);
3663 } else {
3664 assert(UseHugeTLBFS, "must be");
3665 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3666 }
3667 return res;
3668 }
3669
3670 size_t os::large_page_size() {
3671 return _large_page_size;
3672 }
3673
3674 // With SysV SHM the entire memory region must be allocated as shared
3675 // memory.
3676 // HugeTLBFS allows application to commit large page memory on demand.
3677 // However, when committing memory with HugeTLBFS fails, the region
3678 // that was supposed to be committed will lose the old reservation
3679 // and allow other threads to steal that memory region. Because of this
3680 // behavior we can't commit HugeTLBFS memory.
3681 bool os::can_commit_large_page_memory() {
3682 return UseTransparentHugePages;
3683 }
3684
3685 bool os::can_execute_large_page_memory() {
3686 return UseTransparentHugePages || UseHugeTLBFS;
3687 }
3688
3689 // Reserve memory at an arbitrary address, only if that area is
3690 // available (and not reserved for something else).
3691
3692 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3693 const int max_tries = 10;
3694 char* base[max_tries];
3695 size_t size[max_tries];
3696 const size_t gap = 0x000000;
3697
3698 // Assert only that the size is a multiple of the page size, since
3699 // that's all that mmap requires, and since that's all we really know
3700 // about at this low abstraction level. If we need higher alignment,
3701 // we can either pass an alignment to this method or verify alignment
3702 // in one of the methods further up the call chain. See bug 5044738.
3703 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3704
3705 // Repeatedly allocate blocks until the block is allocated at the
3706 // right spot. Give up after max_tries. Note that reserve_memory() will
3707 // automatically update _highest_vm_reserved_address if the call is
3708 // successful. The variable tracks the highest memory address every reserved
3709 // by JVM. It is used to detect heap-stack collision if running with
3710 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3711 // space than needed, it could confuse the collision detecting code. To
3712 // solve the problem, save current _highest_vm_reserved_address and
3713 // calculate the correct value before return.
3714 address old_highest = _highest_vm_reserved_address;
3715
3716 // Linux mmap allows caller to pass an address as hint; give it a try first,
3717 // if kernel honors the hint then we can return immediately.
3718 char * addr = anon_mmap(requested_addr, bytes, false);
3719 if (addr == requested_addr) {
3720 return requested_addr;
3721 }
3722
3723 if (addr != NULL) {
3724 // mmap() is successful but it fails to reserve at the requested address
3725 anon_munmap(addr, bytes);
3726 }
3727
3728 int i;
3729 for (i = 0; i < max_tries; ++i) {
3730 base[i] = reserve_memory(bytes);
3731
3732 if (base[i] != NULL) {
3733 // Is this the block we wanted?
3734 if (base[i] == requested_addr) {
3735 size[i] = bytes;
3736 break;
3737 }
3738
3739 // Does this overlap the block we wanted? Give back the overlapped
3740 // parts and try again.
3741
3742 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3743 if (top_overlap >= 0 && top_overlap < bytes) {
3744 unmap_memory(base[i], top_overlap);
3745 base[i] += top_overlap;
3746 size[i] = bytes - top_overlap;
3747 } else {
3748 size_t bottom_overlap = base[i] + bytes - requested_addr;
3749 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3750 unmap_memory(requested_addr, bottom_overlap);
3751 size[i] = bytes - bottom_overlap;
3752 } else {
3753 size[i] = bytes;
3754 }
3755 }
3756 }
3757 }
3758
3759 // Give back the unused reserved pieces.
3760
3761 for (int j = 0; j < i; ++j) {
3762 if (base[j] != NULL) {
3763 unmap_memory(base[j], size[j]);
3764 }
3765 }
3766
3767 if (i < max_tries) {
3768 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3769 return requested_addr;
3770 } else {
3771 _highest_vm_reserved_address = old_highest;
3772 return NULL;
3773 }
3774 }
3775
3776 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3777 return ::read(fd, buf, nBytes);
3778 }
3779
3780 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3781 return ::pread(fd, buf, nBytes, offset);
3782 }
3783
3784 // Short sleep, direct OS call.
3785 //
3786 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3787 // sched_yield(2) will actually give up the CPU:
3788 //
3789 // * Alone on this pariticular CPU, keeps running.
3790 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3791 // (pre 2.6.39).
3792 //
3793 // So calling this with 0 is an alternative.
3794 //
3795 void os::naked_short_sleep(jlong ms) {
3796 struct timespec req;
3797
3798 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3799 req.tv_sec = 0;
3800 if (ms > 0) {
3801 req.tv_nsec = (ms % 1000) * 1000000;
3802 } else {
3803 req.tv_nsec = 1;
3804 }
3805
3806 nanosleep(&req, NULL);
3807
3808 return;
3809 }
3810
3811 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3812 void os::infinite_sleep() {
3813 while (true) { // sleep forever ...
3814 ::sleep(100); // ... 100 seconds at a time
3815 }
3816 }
3817
3818 // Used to convert frequent JVM_Yield() to nops
3819 bool os::dont_yield() {
3820 return DontYieldALot;
3821 }
3822
3823 void os::naked_yield() {
3824 sched_yield();
3825 }
3826
3827 ////////////////////////////////////////////////////////////////////////////////
3828 // thread priority support
3829
3830 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3831 // only supports dynamic priority, static priority must be zero. For real-time
3832 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3833 // However, for large multi-threaded applications, SCHED_RR is not only slower
3834 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3835 // of 5 runs - Sep 2005).
3836 //
3837 // The following code actually changes the niceness of kernel-thread/LWP. It
3838 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3839 // not the entire user process, and user level threads are 1:1 mapped to kernel
3840 // threads. It has always been the case, but could change in the future. For
3841 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3842 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3843
3844 int os::java_to_os_priority[CriticalPriority + 1] = {
3845 19, // 0 Entry should never be used
3846
3847 4, // 1 MinPriority
3848 3, // 2
3849 2, // 3
3850
3851 1, // 4
3852 0, // 5 NormPriority
3853 -1, // 6
3854
3855 -2, // 7
3856 -3, // 8
3857 -4, // 9 NearMaxPriority
3858
3859 -5, // 10 MaxPriority
3860
3861 -5 // 11 CriticalPriority
3862 };
3863
3864 static int prio_init() {
3865 if (ThreadPriorityPolicy == 1) {
3866 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3867 // if effective uid is not root. Perhaps, a more elegant way of doing
3868 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3869 if (geteuid() != 0) {
3870 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3871 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3872 }
3873 ThreadPriorityPolicy = 0;
3874 }
3875 }
3876 if (UseCriticalJavaThreadPriority) {
3877 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3878 }
3879 return 0;
3880 }
3881
3882 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3883 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3884
3885 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3886 return (ret == 0) ? OS_OK : OS_ERR;
3887 }
3888
3889 OSReturn os::get_native_priority(const Thread* const thread,
3890 int *priority_ptr) {
3891 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3892 *priority_ptr = java_to_os_priority[NormPriority];
3893 return OS_OK;
3894 }
3895
3896 errno = 0;
3897 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3898 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3899 }
3900
3901 // Hint to the underlying OS that a task switch would not be good.
3902 // Void return because it's a hint and can fail.
3903 void os::hint_no_preempt() {}
3904
3905 ////////////////////////////////////////////////////////////////////////////////
3906 // suspend/resume support
3907
3908 // the low-level signal-based suspend/resume support is a remnant from the
3909 // old VM-suspension that used to be for java-suspension, safepoints etc,
3910 // within hotspot. Now there is a single use-case for this:
3911 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3912 // that runs in the watcher thread.
3913 // The remaining code is greatly simplified from the more general suspension
3914 // code that used to be used.
3915 //
3916 // The protocol is quite simple:
3917 // - suspend:
3918 // - sends a signal to the target thread
3919 // - polls the suspend state of the osthread using a yield loop
3920 // - target thread signal handler (SR_handler) sets suspend state
3921 // and blocks in sigsuspend until continued
3922 // - resume:
3923 // - sets target osthread state to continue
3924 // - sends signal to end the sigsuspend loop in the SR_handler
3925 //
3926 // Note that the SR_lock plays no role in this suspend/resume protocol.
3927
3928 static void resume_clear_context(OSThread *osthread) {
3929 osthread->set_ucontext(NULL);
3930 osthread->set_siginfo(NULL);
3931 }
3932
3933 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3934 ucontext_t* context) {
3935 osthread->set_ucontext(context);
3936 osthread->set_siginfo(siginfo);
3937 }
3938
3939 // Handler function invoked when a thread's execution is suspended or
3940 // resumed. We have to be careful that only async-safe functions are
3941 // called here (Note: most pthread functions are not async safe and
3942 // should be avoided.)
3943 //
3944 // Note: sigwait() is a more natural fit than sigsuspend() from an
3945 // interface point of view, but sigwait() prevents the signal hander
3946 // from being run. libpthread would get very confused by not having
3947 // its signal handlers run and prevents sigwait()'s use with the
3948 // mutex granting granting signal.
3949 //
3950 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3951 //
3952 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3953 // Save and restore errno to avoid confusing native code with EINTR
3954 // after sigsuspend.
3955 int old_errno = errno;
3956
3957 Thread* thread = Thread::current();
3958 OSThread* osthread = thread->osthread();
3959 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3960
3961 os::SuspendResume::State current = osthread->sr.state();
3962 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3963 suspend_save_context(osthread, siginfo, context);
3964
3965 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3966 os::SuspendResume::State state = osthread->sr.suspended();
3967 if (state == os::SuspendResume::SR_SUSPENDED) {
3968 sigset_t suspend_set; // signals for sigsuspend()
3969
3970 // get current set of blocked signals and unblock resume signal
3971 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3972 sigdelset(&suspend_set, SR_signum);
3973
3974 sr_semaphore.signal();
3975 // wait here until we are resumed
3976 while (1) {
3977 sigsuspend(&suspend_set);
3978
3979 os::SuspendResume::State result = osthread->sr.running();
3980 if (result == os::SuspendResume::SR_RUNNING) {
3981 sr_semaphore.signal();
3982 break;
3983 }
3984 }
3985
3986 } else if (state == os::SuspendResume::SR_RUNNING) {
3987 // request was cancelled, continue
3988 } else {
3989 ShouldNotReachHere();
3990 }
3991
3992 resume_clear_context(osthread);
3993 } else if (current == os::SuspendResume::SR_RUNNING) {
3994 // request was cancelled, continue
3995 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3996 // ignore
3997 } else {
3998 // ignore
3999 }
4000
4001 errno = old_errno;
4002 }
4003
4004
4005 static int SR_initialize() {
4006 struct sigaction act;
4007 char *s;
4008 // Get signal number to use for suspend/resume
4009 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4010 int sig = ::strtol(s, 0, 10);
4011 if (sig > 0 || sig < _NSIG) {
4012 SR_signum = sig;
4013 }
4014 }
4015
4016 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4017 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4018
4019 sigemptyset(&SR_sigset);
4020 sigaddset(&SR_sigset, SR_signum);
4021
4022 // Set up signal handler for suspend/resume
4023 act.sa_flags = SA_RESTART|SA_SIGINFO;
4024 act.sa_handler = (void (*)(int)) SR_handler;
4025
4026 // SR_signum is blocked by default.
4027 // 4528190 - We also need to block pthread restart signal (32 on all
4028 // supported Linux platforms). Note that LinuxThreads need to block
4029 // this signal for all threads to work properly. So we don't have
4030 // to use hard-coded signal number when setting up the mask.
4031 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4032
4033 if (sigaction(SR_signum, &act, 0) == -1) {
4034 return -1;
4035 }
4036
4037 // Save signal flag
4038 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4039 return 0;
4040 }
4041
4042 static int sr_notify(OSThread* osthread) {
4043 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4044 assert_status(status == 0, status, "pthread_kill");
4045 return status;
4046 }
4047
4048 // "Randomly" selected value for how long we want to spin
4049 // before bailing out on suspending a thread, also how often
4050 // we send a signal to a thread we want to resume
4051 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4052 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4053
4054 // returns true on success and false on error - really an error is fatal
4055 // but this seems the normal response to library errors
4056 static bool do_suspend(OSThread* osthread) {
4057 assert(osthread->sr.is_running(), "thread should be running");
4058 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4059
4060 // mark as suspended and send signal
4061 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4062 // failed to switch, state wasn't running?
4063 ShouldNotReachHere();
4064 return false;
4065 }
4066
4067 if (sr_notify(osthread) != 0) {
4068 ShouldNotReachHere();
4069 }
4070
4071 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4072 while (true) {
4073 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4074 break;
4075 } else {
4076 // timeout
4077 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4078 if (cancelled == os::SuspendResume::SR_RUNNING) {
4079 return false;
4080 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4081 // make sure that we consume the signal on the semaphore as well
4082 sr_semaphore.wait();
4083 break;
4084 } else {
4085 ShouldNotReachHere();
4086 return false;
4087 }
4088 }
4089 }
4090
4091 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4092 return true;
4093 }
4094
4095 static void do_resume(OSThread* osthread) {
4096 assert(osthread->sr.is_suspended(), "thread should be suspended");
4097 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4098
4099 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4100 // failed to switch to WAKEUP_REQUEST
4101 ShouldNotReachHere();
4102 return;
4103 }
4104
4105 while (true) {
4106 if (sr_notify(osthread) == 0) {
4107 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4108 if (osthread->sr.is_running()) {
4109 return;
4110 }
4111 }
4112 } else {
4113 ShouldNotReachHere();
4114 }
4115 }
4116
4117 guarantee(osthread->sr.is_running(), "Must be running!");
4118 }
4119
4120 ///////////////////////////////////////////////////////////////////////////////////
4121 // signal handling (except suspend/resume)
4122
4123 // This routine may be used by user applications as a "hook" to catch signals.
4124 // The user-defined signal handler must pass unrecognized signals to this
4125 // routine, and if it returns true (non-zero), then the signal handler must
4126 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4127 // routine will never retun false (zero), but instead will execute a VM panic
4128 // routine kill the process.
4129 //
4130 // If this routine returns false, it is OK to call it again. This allows
4131 // the user-defined signal handler to perform checks either before or after
4132 // the VM performs its own checks. Naturally, the user code would be making
4133 // a serious error if it tried to handle an exception (such as a null check
4134 // or breakpoint) that the VM was generating for its own correct operation.
4135 //
4136 // This routine may recognize any of the following kinds of signals:
4137 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4138 // It should be consulted by handlers for any of those signals.
4139 //
4140 // The caller of this routine must pass in the three arguments supplied
4141 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4142 // field of the structure passed to sigaction(). This routine assumes that
4143 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4144 //
4145 // Note that the VM will print warnings if it detects conflicting signal
4146 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4147 //
4148 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4149 siginfo_t* siginfo,
4150 void* ucontext,
4151 int abort_if_unrecognized);
4152
4153 void signalHandler(int sig, siginfo_t* info, void* uc) {
4154 assert(info != NULL && uc != NULL, "it must be old kernel");
4155 int orig_errno = errno; // Preserve errno value over signal handler.
4156 JVM_handle_linux_signal(sig, info, uc, true);
4157 errno = orig_errno;
4158 }
4159
4160
4161 // This boolean allows users to forward their own non-matching signals
4162 // to JVM_handle_linux_signal, harmlessly.
4163 bool os::Linux::signal_handlers_are_installed = false;
4164
4165 // For signal-chaining
4166 struct sigaction os::Linux::sigact[MAXSIGNUM];
4167 unsigned int os::Linux::sigs = 0;
4168 bool os::Linux::libjsig_is_loaded = false;
4169 typedef struct sigaction *(*get_signal_t)(int);
4170 get_signal_t os::Linux::get_signal_action = NULL;
4171
4172 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4173 struct sigaction *actp = NULL;
4174
4175 if (libjsig_is_loaded) {
4176 // Retrieve the old signal handler from libjsig
4177 actp = (*get_signal_action)(sig);
4178 }
4179 if (actp == NULL) {
4180 // Retrieve the preinstalled signal handler from jvm
4181 actp = get_preinstalled_handler(sig);
4182 }
4183
4184 return actp;
4185 }
4186
4187 static bool call_chained_handler(struct sigaction *actp, int sig,
4188 siginfo_t *siginfo, void *context) {
4189 // Call the old signal handler
4190 if (actp->sa_handler == SIG_DFL) {
4191 // It's more reasonable to let jvm treat it as an unexpected exception
4192 // instead of taking the default action.
4193 return false;
4194 } else if (actp->sa_handler != SIG_IGN) {
4195 if ((actp->sa_flags & SA_NODEFER) == 0) {
4196 // automaticlly block the signal
4197 sigaddset(&(actp->sa_mask), sig);
4198 }
4199
4200 sa_handler_t hand;
4201 sa_sigaction_t sa;
4202 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4203 // retrieve the chained handler
4204 if (siginfo_flag_set) {
4205 sa = actp->sa_sigaction;
4206 } else {
4207 hand = actp->sa_handler;
4208 }
4209
4210 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4211 actp->sa_handler = SIG_DFL;
4212 }
4213
4214 // try to honor the signal mask
4215 sigset_t oset;
4216 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4217
4218 // call into the chained handler
4219 if (siginfo_flag_set) {
4220 (*sa)(sig, siginfo, context);
4221 } else {
4222 (*hand)(sig);
4223 }
4224
4225 // restore the signal mask
4226 pthread_sigmask(SIG_SETMASK, &oset, 0);
4227 }
4228 // Tell jvm's signal handler the signal is taken care of.
4229 return true;
4230 }
4231
4232 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4233 bool chained = false;
4234 // signal-chaining
4235 if (UseSignalChaining) {
4236 struct sigaction *actp = get_chained_signal_action(sig);
4237 if (actp != NULL) {
4238 chained = call_chained_handler(actp, sig, siginfo, context);
4239 }
4240 }
4241 return chained;
4242 }
4243
4244 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4245 if ((((unsigned int)1 << sig) & sigs) != 0) {
4246 return &sigact[sig];
4247 }
4248 return NULL;
4249 }
4250
4251 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4252 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4253 sigact[sig] = oldAct;
4254 sigs |= (unsigned int)1 << sig;
4255 }
4256
4257 // for diagnostic
4258 int os::Linux::sigflags[MAXSIGNUM];
4259
4260 int os::Linux::get_our_sigflags(int sig) {
4261 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4262 return sigflags[sig];
4263 }
4264
4265 void os::Linux::set_our_sigflags(int sig, int flags) {
4266 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4267 sigflags[sig] = flags;
4268 }
4269
4270 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4271 // Check for overwrite.
4272 struct sigaction oldAct;
4273 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4274
4275 void* oldhand = oldAct.sa_sigaction
4276 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4277 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4278 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4279 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4280 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4281 if (AllowUserSignalHandlers || !set_installed) {
4282 // Do not overwrite; user takes responsibility to forward to us.
4283 return;
4284 } else if (UseSignalChaining) {
4285 // save the old handler in jvm
4286 save_preinstalled_handler(sig, oldAct);
4287 // libjsig also interposes the sigaction() call below and saves the
4288 // old sigaction on it own.
4289 } else {
4290 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4291 "%#lx for signal %d.", (long)oldhand, sig));
4292 }
4293 }
4294
4295 struct sigaction sigAct;
4296 sigfillset(&(sigAct.sa_mask));
4297 sigAct.sa_handler = SIG_DFL;
4298 if (!set_installed) {
4299 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4300 } else {
4301 sigAct.sa_sigaction = signalHandler;
4302 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4303 }
4304 // Save flags, which are set by ours
4305 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4306 sigflags[sig] = sigAct.sa_flags;
4307
4308 int ret = sigaction(sig, &sigAct, &oldAct);
4309 assert(ret == 0, "check");
4310
4311 void* oldhand2 = oldAct.sa_sigaction
4312 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4313 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4314 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4315 }
4316
4317 // install signal handlers for signals that HotSpot needs to
4318 // handle in order to support Java-level exception handling.
4319
4320 void os::Linux::install_signal_handlers() {
4321 if (!signal_handlers_are_installed) {
4322 signal_handlers_are_installed = true;
4323
4324 // signal-chaining
4325 typedef void (*signal_setting_t)();
4326 signal_setting_t begin_signal_setting = NULL;
4327 signal_setting_t end_signal_setting = NULL;
4328 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4329 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4330 if (begin_signal_setting != NULL) {
4331 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4332 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4333 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4334 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4335 libjsig_is_loaded = true;
4336 assert(UseSignalChaining, "should enable signal-chaining");
4337 }
4338 if (libjsig_is_loaded) {
4339 // Tell libjsig jvm is setting signal handlers
4340 (*begin_signal_setting)();
4341 }
4342
4343 set_signal_handler(SIGSEGV, true);
4344 set_signal_handler(SIGPIPE, true);
4345 set_signal_handler(SIGBUS, true);
4346 set_signal_handler(SIGILL, true);
4347 set_signal_handler(SIGFPE, true);
4348 #if defined(PPC64)
4349 set_signal_handler(SIGTRAP, true);
4350 #endif
4351 set_signal_handler(SIGXFSZ, true);
4352
4353 if (libjsig_is_loaded) {
4354 // Tell libjsig jvm finishes setting signal handlers
4355 (*end_signal_setting)();
4356 }
4357
4358 // We don't activate signal checker if libjsig is in place, we trust ourselves
4359 // and if UserSignalHandler is installed all bets are off.
4360 // Log that signal checking is off only if -verbose:jni is specified.
4361 if (CheckJNICalls) {
4362 if (libjsig_is_loaded) {
4363 if (PrintJNIResolving) {
4364 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4365 }
4366 check_signals = false;
4367 }
4368 if (AllowUserSignalHandlers) {
4369 if (PrintJNIResolving) {
4370 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4371 }
4372 check_signals = false;
4373 }
4374 }
4375 }
4376 }
4377
4378 // This is the fastest way to get thread cpu time on Linux.
4379 // Returns cpu time (user+sys) for any thread, not only for current.
4380 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4381 // It might work on 2.6.10+ with a special kernel/glibc patch.
4382 // For reference, please, see IEEE Std 1003.1-2004:
4383 // http://www.unix.org/single_unix_specification
4384
4385 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4386 struct timespec tp;
4387 int rc = os::Linux::clock_gettime(clockid, &tp);
4388 assert(rc == 0, "clock_gettime is expected to return 0 code");
4389
4390 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4391 }
4392
4393 /////
4394 // glibc on Linux platform uses non-documented flag
4395 // to indicate, that some special sort of signal
4396 // trampoline is used.
4397 // We will never set this flag, and we should
4398 // ignore this flag in our diagnostic
4399 #ifdef SIGNIFICANT_SIGNAL_MASK
4400 #undef SIGNIFICANT_SIGNAL_MASK
4401 #endif
4402 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4403
4404 static const char* get_signal_handler_name(address handler,
4405 char* buf, int buflen) {
4406 int offset;
4407 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4408 if (found) {
4409 // skip directory names
4410 const char *p1, *p2;
4411 p1 = buf;
4412 size_t len = strlen(os::file_separator());
4413 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4414 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4415 } else {
4416 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4417 }
4418 return buf;
4419 }
4420
4421 static void print_signal_handler(outputStream* st, int sig,
4422 char* buf, size_t buflen) {
4423 struct sigaction sa;
4424
4425 sigaction(sig, NULL, &sa);
4426
4427 // See comment for SIGNIFICANT_SIGNAL_MASK define
4428 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4429
4430 st->print("%s: ", os::exception_name(sig, buf, buflen));
4431
4432 address handler = (sa.sa_flags & SA_SIGINFO)
4433 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4434 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4435
4436 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4437 st->print("SIG_DFL");
4438 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4439 st->print("SIG_IGN");
4440 } else {
4441 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4442 }
4443
4444 st->print(", sa_mask[0]=");
4445 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4446
4447 address rh = VMError::get_resetted_sighandler(sig);
4448 // May be, handler was resetted by VMError?
4449 if (rh != NULL) {
4450 handler = rh;
4451 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4452 }
4453
4454 st->print(", sa_flags=");
4455 os::Posix::print_sa_flags(st, sa.sa_flags);
4456
4457 // Check: is it our handler?
4458 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4459 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4460 // It is our signal handler
4461 // check for flags, reset system-used one!
4462 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4463 st->print(
4464 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4465 os::Linux::get_our_sigflags(sig));
4466 }
4467 }
4468 st->cr();
4469 }
4470
4471
4472 #define DO_SIGNAL_CHECK(sig) \
4473 do { \
4474 if (!sigismember(&check_signal_done, sig)) { \
4475 os::Linux::check_signal_handler(sig); \
4476 } \
4477 } while (0)
4478
4479 // This method is a periodic task to check for misbehaving JNI applications
4480 // under CheckJNI, we can add any periodic checks here
4481
4482 void os::run_periodic_checks() {
4483 if (check_signals == false) return;
4484
4485 // SEGV and BUS if overridden could potentially prevent
4486 // generation of hs*.log in the event of a crash, debugging
4487 // such a case can be very challenging, so we absolutely
4488 // check the following for a good measure:
4489 DO_SIGNAL_CHECK(SIGSEGV);
4490 DO_SIGNAL_CHECK(SIGILL);
4491 DO_SIGNAL_CHECK(SIGFPE);
4492 DO_SIGNAL_CHECK(SIGBUS);
4493 DO_SIGNAL_CHECK(SIGPIPE);
4494 DO_SIGNAL_CHECK(SIGXFSZ);
4495 #if defined(PPC64)
4496 DO_SIGNAL_CHECK(SIGTRAP);
4497 #endif
4498
4499 // ReduceSignalUsage allows the user to override these handlers
4500 // see comments at the very top and jvm_solaris.h
4501 if (!ReduceSignalUsage) {
4502 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4503 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4504 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4505 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4506 }
4507
4508 DO_SIGNAL_CHECK(SR_signum);
4509 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4510 }
4511
4512 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4513
4514 static os_sigaction_t os_sigaction = NULL;
4515
4516 void os::Linux::check_signal_handler(int sig) {
4517 char buf[O_BUFLEN];
4518 address jvmHandler = NULL;
4519
4520
4521 struct sigaction act;
4522 if (os_sigaction == NULL) {
4523 // only trust the default sigaction, in case it has been interposed
4524 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4525 if (os_sigaction == NULL) return;
4526 }
4527
4528 os_sigaction(sig, (struct sigaction*)NULL, &act);
4529
4530
4531 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4532
4533 address thisHandler = (act.sa_flags & SA_SIGINFO)
4534 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4535 : CAST_FROM_FN_PTR(address, act.sa_handler);
4536
4537
4538 switch (sig) {
4539 case SIGSEGV:
4540 case SIGBUS:
4541 case SIGFPE:
4542 case SIGPIPE:
4543 case SIGILL:
4544 case SIGXFSZ:
4545 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4546 break;
4547
4548 case SHUTDOWN1_SIGNAL:
4549 case SHUTDOWN2_SIGNAL:
4550 case SHUTDOWN3_SIGNAL:
4551 case BREAK_SIGNAL:
4552 jvmHandler = (address)user_handler();
4553 break;
4554
4555 case INTERRUPT_SIGNAL:
4556 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4557 break;
4558
4559 default:
4560 if (sig == SR_signum) {
4561 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4562 } else {
4563 return;
4564 }
4565 break;
4566 }
4567
4568 if (thisHandler != jvmHandler) {
4569 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4570 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4571 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4572 // No need to check this sig any longer
4573 sigaddset(&check_signal_done, sig);
4574 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4575 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4576 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4577 exception_name(sig, buf, O_BUFLEN));
4578 }
4579 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4580 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4581 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4582 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4583 // No need to check this sig any longer
4584 sigaddset(&check_signal_done, sig);
4585 }
4586
4587 // Dump all the signal
4588 if (sigismember(&check_signal_done, sig)) {
4589 print_signal_handlers(tty, buf, O_BUFLEN);
4590 }
4591 }
4592
4593 extern void report_error(char* file_name, int line_no, char* title,
4594 char* format, ...);
4595
4596 extern bool signal_name(int signo, char* buf, size_t len);
4597
4598 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4599 if (0 < exception_code && exception_code <= SIGRTMAX) {
4600 // signal
4601 if (!signal_name(exception_code, buf, size)) {
4602 jio_snprintf(buf, size, "SIG%d", exception_code);
4603 }
4604 return buf;
4605 } else {
4606 return NULL;
4607 }
4608 }
4609
4610 // this is called _before_ the most of global arguments have been parsed
4611 void os::init(void) {
4612 char dummy; // used to get a guess on initial stack address
4613 // first_hrtime = gethrtime();
4614
4615 // With LinuxThreads the JavaMain thread pid (primordial thread)
4616 // is different than the pid of the java launcher thread.
4617 // So, on Linux, the launcher thread pid is passed to the VM
4618 // via the sun.java.launcher.pid property.
4619 // Use this property instead of getpid() if it was correctly passed.
4620 // See bug 6351349.
4621 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4622
4623 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4624
4625 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4626
4627 init_random(1234567);
4628
4629 ThreadCritical::initialize();
4630
4631 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4632 if (Linux::page_size() == -1) {
4633 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4634 strerror(errno)));
4635 }
4636 init_page_sizes((size_t) Linux::page_size());
4637
4638 Linux::initialize_system_info();
4639
4640 // main_thread points to the aboriginal thread
4641 Linux::_main_thread = pthread_self();
4642
4643 Linux::clock_init();
4644 initial_time_count = javaTimeNanos();
4645
4646 // pthread_condattr initialization for monotonic clock
4647 int status;
4648 pthread_condattr_t* _condattr = os::Linux::condAttr();
4649 if ((status = pthread_condattr_init(_condattr)) != 0) {
4650 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4651 }
4652 // Only set the clock if CLOCK_MONOTONIC is available
4653 if (os::supports_monotonic_clock()) {
4654 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4655 if (status == EINVAL) {
4656 warning("Unable to use monotonic clock with relative timed-waits" \
4657 " - changes to the time-of-day clock may have adverse affects");
4658 } else {
4659 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4660 }
4661 }
4662 }
4663 // else it defaults to CLOCK_REALTIME
4664
4665 pthread_mutex_init(&dl_mutex, NULL);
4666
4667 // If the pagesize of the VM is greater than 8K determine the appropriate
4668 // number of initial guard pages. The user can change this with the
4669 // command line arguments, if needed.
4670 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4671 StackYellowPages = 1;
4672 StackRedPages = 1;
4673 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4674 }
4675
4676 // retrieve entry point for pthread_setname_np
4677 Linux::_pthread_setname_np =
4678 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4679
4680 }
4681
4682 // To install functions for atexit system call
4683 extern "C" {
4684 static void perfMemory_exit_helper() {
4685 perfMemory_exit();
4686 }
4687 }
4688
4689 // this is called _after_ the global arguments have been parsed
4690 jint os::init_2(void) {
4691 Linux::fast_thread_clock_init();
4692
4693 // Allocate a single page and mark it as readable for safepoint polling
4694 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4695 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4696
4697 os::set_polling_page(polling_page);
4698
4699 #ifndef PRODUCT
4700 if (Verbose && PrintMiscellaneous) {
4701 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4702 (intptr_t)polling_page);
4703 }
4704 #endif
4705
4706 if (!UseMembar) {
4707 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4708 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4709 os::set_memory_serialize_page(mem_serialize_page);
4710
4711 #ifndef PRODUCT
4712 if (Verbose && PrintMiscellaneous) {
4713 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4714 (intptr_t)mem_serialize_page);
4715 }
4716 #endif
4717 }
4718
4719 // initialize suspend/resume support - must do this before signal_sets_init()
4720 if (SR_initialize() != 0) {
4721 perror("SR_initialize failed");
4722 return JNI_ERR;
4723 }
4724
4725 Linux::signal_sets_init();
4726 Linux::install_signal_handlers();
4727
4728 // Check minimum allowable stack size for thread creation and to initialize
4729 // the java system classes, including StackOverflowError - depends on page
4730 // size. Add a page for compiler2 recursion in main thread.
4731 // Add in 2*BytesPerWord times page size to account for VM stack during
4732 // class initialization depending on 32 or 64 bit VM.
4733 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4734 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4735 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4736
4737 size_t threadStackSizeInBytes = ThreadStackSize * K;
4738 if (threadStackSizeInBytes != 0 &&
4739 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4740 tty->print_cr("\nThe stack size specified is too small, "
4741 "Specify at least %dk",
4742 os::Linux::min_stack_allowed/ K);
4743 return JNI_ERR;
4744 }
4745
4746 // Make the stack size a multiple of the page size so that
4747 // the yellow/red zones can be guarded.
4748 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4749 vm_page_size()));
4750
4751 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4752
4753 #if defined(IA32)
4754 workaround_expand_exec_shield_cs_limit();
4755 #endif
4756
4757 Linux::libpthread_init();
4758 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4759 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4760 Linux::glibc_version(), Linux::libpthread_version(),
4761 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4762 }
4763
4764 if (UseNUMA) {
4765 if (!Linux::libnuma_init()) {
4766 UseNUMA = false;
4767 } else {
4768 if ((Linux::numa_max_node() < 1)) {
4769 // There's only one node(they start from 0), disable NUMA.
4770 UseNUMA = false;
4771 }
4772 }
4773 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4774 // we can make the adaptive lgrp chunk resizing work. If the user specified
4775 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4776 // disable adaptive resizing.
4777 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4778 if (FLAG_IS_DEFAULT(UseNUMA)) {
4779 UseNUMA = false;
4780 } else {
4781 if (FLAG_IS_DEFAULT(UseLargePages) &&
4782 FLAG_IS_DEFAULT(UseSHM) &&
4783 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4784 UseLargePages = false;
4785 } else {
4786 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4787 UseAdaptiveSizePolicy = false;
4788 UseAdaptiveNUMAChunkSizing = false;
4789 }
4790 }
4791 }
4792 if (!UseNUMA && ForceNUMA) {
4793 UseNUMA = true;
4794 }
4795 }
4796
4797 if (MaxFDLimit) {
4798 // set the number of file descriptors to max. print out error
4799 // if getrlimit/setrlimit fails but continue regardless.
4800 struct rlimit nbr_files;
4801 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4802 if (status != 0) {
4803 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4804 perror("os::init_2 getrlimit failed");
4805 }
4806 } else {
4807 nbr_files.rlim_cur = nbr_files.rlim_max;
4808 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4809 if (status != 0) {
4810 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4811 perror("os::init_2 setrlimit failed");
4812 }
4813 }
4814 }
4815 }
4816
4817 // Initialize lock used to serialize thread creation (see os::create_thread)
4818 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4819
4820 // at-exit methods are called in the reverse order of their registration.
4821 // atexit functions are called on return from main or as a result of a
4822 // call to exit(3C). There can be only 32 of these functions registered
4823 // and atexit() does not set errno.
4824
4825 if (PerfAllowAtExitRegistration) {
4826 // only register atexit functions if PerfAllowAtExitRegistration is set.
4827 // atexit functions can be delayed until process exit time, which
4828 // can be problematic for embedded VM situations. Embedded VMs should
4829 // call DestroyJavaVM() to assure that VM resources are released.
4830
4831 // note: perfMemory_exit_helper atexit function may be removed in
4832 // the future if the appropriate cleanup code can be added to the
4833 // VM_Exit VMOperation's doit method.
4834 if (atexit(perfMemory_exit_helper) != 0) {
4835 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4836 }
4837 }
4838
4839 // initialize thread priority policy
4840 prio_init();
4841
4842 return JNI_OK;
4843 }
4844
4845 // Mark the polling page as unreadable
4846 void os::make_polling_page_unreadable(void) {
4847 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4848 fatal("Could not disable polling page");
4849 }
4850 }
4851
4852 // Mark the polling page as readable
4853 void os::make_polling_page_readable(void) {
4854 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4855 fatal("Could not enable polling page");
4856 }
4857 }
4858
4859 int os::active_processor_count() {
4860 // Linux doesn't yet have a (official) notion of processor sets,
4861 // so just return the number of online processors.
4862 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4863 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4864 return online_cpus;
4865 }
4866
4867 void os::set_native_thread_name(const char *name) {
4868 if (Linux::_pthread_setname_np) {
4869 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4870 snprintf(buf, sizeof(buf), "%s", name);
4871 buf[sizeof(buf) - 1] = '\0';
4872 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4873 // ERANGE should not happen; all other errors should just be ignored.
4874 assert(rc != ERANGE, "pthread_setname_np failed");
4875 }
4876 }
4877
4878 bool os::distribute_processes(uint length, uint* distribution) {
4879 // Not yet implemented.
4880 return false;
4881 }
4882
4883 bool os::bind_to_processor(uint processor_id) {
4884 // Not yet implemented.
4885 return false;
4886 }
4887
4888 ///
4889
4890 void os::SuspendedThreadTask::internal_do_task() {
4891 if (do_suspend(_thread->osthread())) {
4892 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4893 do_task(context);
4894 do_resume(_thread->osthread());
4895 }
4896 }
4897
4898 class PcFetcher : public os::SuspendedThreadTask {
4899 public:
4900 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4901 ExtendedPC result();
4902 protected:
4903 void do_task(const os::SuspendedThreadTaskContext& context);
4904 private:
4905 ExtendedPC _epc;
4906 };
4907
4908 ExtendedPC PcFetcher::result() {
4909 guarantee(is_done(), "task is not done yet.");
4910 return _epc;
4911 }
4912
4913 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4914 Thread* thread = context.thread();
4915 OSThread* osthread = thread->osthread();
4916 if (osthread->ucontext() != NULL) {
4917 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4918 } else {
4919 // NULL context is unexpected, double-check this is the VMThread
4920 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4921 }
4922 }
4923
4924 // Suspends the target using the signal mechanism and then grabs the PC before
4925 // resuming the target. Used by the flat-profiler only
4926 ExtendedPC os::get_thread_pc(Thread* thread) {
4927 // Make sure that it is called by the watcher for the VMThread
4928 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4929 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4930
4931 PcFetcher fetcher(thread);
4932 fetcher.run();
4933 return fetcher.result();
4934 }
4935
4936 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4937 pthread_mutex_t *_mutex,
4938 const struct timespec *_abstime) {
4939 if (is_NPTL()) {
4940 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4941 } else {
4942 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4943 // word back to default 64bit precision if condvar is signaled. Java
4944 // wants 53bit precision. Save and restore current value.
4945 int fpu = get_fpu_control_word();
4946 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4947 set_fpu_control_word(fpu);
4948 return status;
4949 }
4950 }
4951
4952 ////////////////////////////////////////////////////////////////////////////////
4953 // debug support
4954
4955 bool os::find(address addr, outputStream* st) {
4956 Dl_info dlinfo;
4957 memset(&dlinfo, 0, sizeof(dlinfo));
4958 if (dladdr(addr, &dlinfo) != 0) {
4959 st->print(PTR_FORMAT ": ", addr);
4960 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4961 st->print("%s+%#x", dlinfo.dli_sname,
4962 addr - (intptr_t)dlinfo.dli_saddr);
4963 } else if (dlinfo.dli_fbase != NULL) {
4964 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4965 } else {
4966 st->print("<absolute address>");
4967 }
4968 if (dlinfo.dli_fname != NULL) {
4969 st->print(" in %s", dlinfo.dli_fname);
4970 }
4971 if (dlinfo.dli_fbase != NULL) {
4972 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4973 }
4974 st->cr();
4975
4976 if (Verbose) {
4977 // decode some bytes around the PC
4978 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4979 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4980 address lowest = (address) dlinfo.dli_sname;
4981 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4982 if (begin < lowest) begin = lowest;
4983 Dl_info dlinfo2;
4984 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4985 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4986 end = (address) dlinfo2.dli_saddr;
4987 }
4988 Disassembler::decode(begin, end, st);
4989 }
4990 return true;
4991 }
4992 return false;
4993 }
4994
4995 ////////////////////////////////////////////////////////////////////////////////
4996 // misc
4997
4998 // This does not do anything on Linux. This is basically a hook for being
4999 // able to use structured exception handling (thread-local exception filters)
5000 // on, e.g., Win32.
5001 void
5002 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5003 JavaCallArguments* args, Thread* thread) {
5004 f(value, method, args, thread);
5005 }
5006
5007 void os::print_statistics() {
5008 }
5009
5010 int os::message_box(const char* title, const char* message) {
5011 int i;
5012 fdStream err(defaultStream::error_fd());
5013 for (i = 0; i < 78; i++) err.print_raw("=");
5014 err.cr();
5015 err.print_raw_cr(title);
5016 for (i = 0; i < 78; i++) err.print_raw("-");
5017 err.cr();
5018 err.print_raw_cr(message);
5019 for (i = 0; i < 78; i++) err.print_raw("=");
5020 err.cr();
5021
5022 char buf[16];
5023 // Prevent process from exiting upon "read error" without consuming all CPU
5024 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5025
5026 return buf[0] == 'y' || buf[0] == 'Y';
5027 }
5028
5029 int os::stat(const char *path, struct stat *sbuf) {
5030 char pathbuf[MAX_PATH];
5031 if (strlen(path) > MAX_PATH - 1) {
5032 errno = ENAMETOOLONG;
5033 return -1;
5034 }
5035 os::native_path(strcpy(pathbuf, path));
5036 return ::stat(pathbuf, sbuf);
5037 }
5038
5039 bool os::check_heap(bool force) {
5040 return true;
5041 }
5042
5043 // Is a (classpath) directory empty?
5044 bool os::dir_is_empty(const char* path) {
5045 DIR *dir = NULL;
5046 struct dirent *ptr;
5047
5048 dir = opendir(path);
5049 if (dir == NULL) return true;
5050
5051 // Scan the directory
5052 bool result = true;
5053 char buf[sizeof(struct dirent) + MAX_PATH];
5054 while (result && (ptr = ::readdir(dir)) != NULL) {
5055 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5056 result = false;
5057 }
5058 }
5059 closedir(dir);
5060 return result;
5061 }
5062
5063 // This code originates from JDK's sysOpen and open64_w
5064 // from src/solaris/hpi/src/system_md.c
5065
5066 int os::open(const char *path, int oflag, int mode) {
5067 if (strlen(path) > MAX_PATH - 1) {
5068 errno = ENAMETOOLONG;
5069 return -1;
5070 }
5071
5072 // All file descriptors that are opened in the Java process and not
5073 // specifically destined for a subprocess should have the close-on-exec
5074 // flag set. If we don't set it, then careless 3rd party native code
5075 // might fork and exec without closing all appropriate file descriptors
5076 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5077 // turn might:
5078 //
5079 // - cause end-of-file to fail to be detected on some file
5080 // descriptors, resulting in mysterious hangs, or
5081 //
5082 // - might cause an fopen in the subprocess to fail on a system
5083 // suffering from bug 1085341.
5084 //
5085 // (Yes, the default setting of the close-on-exec flag is a Unix
5086 // design flaw)
5087 //
5088 // See:
5089 // 1085341: 32-bit stdio routines should support file descriptors >255
5090 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5091 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5092 //
5093 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5094 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5095 // because it saves a system call and removes a small window where the flag
5096 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5097 // and we fall back to using FD_CLOEXEC (see below).
5098 #ifdef O_CLOEXEC
5099 oflag |= O_CLOEXEC;
5100 #endif
5101
5102 int fd = ::open64(path, oflag, mode);
5103 if (fd == -1) return -1;
5104
5105 //If the open succeeded, the file might still be a directory
5106 {
5107 struct stat64 buf64;
5108 int ret = ::fstat64(fd, &buf64);
5109 int st_mode = buf64.st_mode;
5110
5111 if (ret != -1) {
5112 if ((st_mode & S_IFMT) == S_IFDIR) {
5113 errno = EISDIR;
5114 ::close(fd);
5115 return -1;
5116 }
5117 } else {
5118 ::close(fd);
5119 return -1;
5120 }
5121 }
5122
5123 #ifdef FD_CLOEXEC
5124 // Validate that the use of the O_CLOEXEC flag on open above worked.
5125 // With recent kernels, we will perform this check exactly once.
5126 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5127 if (!O_CLOEXEC_is_known_to_work) {
5128 int flags = ::fcntl(fd, F_GETFD);
5129 if (flags != -1) {
5130 if ((flags & FD_CLOEXEC) != 0)
5131 O_CLOEXEC_is_known_to_work = 1;
5132 else
5133 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5134 }
5135 }
5136 #endif
5137
5138 return fd;
5139 }
5140
5141
5142 // create binary file, rewriting existing file if required
5143 int os::create_binary_file(const char* path, bool rewrite_existing) {
5144 int oflags = O_WRONLY | O_CREAT;
5145 if (!rewrite_existing) {
5146 oflags |= O_EXCL;
5147 }
5148 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5149 }
5150
5151 // return current position of file pointer
5152 jlong os::current_file_offset(int fd) {
5153 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5154 }
5155
5156 // move file pointer to the specified offset
5157 jlong os::seek_to_file_offset(int fd, jlong offset) {
5158 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5159 }
5160
5161 // This code originates from JDK's sysAvailable
5162 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5163
5164 int os::available(int fd, jlong *bytes) {
5165 jlong cur, end;
5166 int mode;
5167 struct stat64 buf64;
5168
5169 if (::fstat64(fd, &buf64) >= 0) {
5170 mode = buf64.st_mode;
5171 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5172 // XXX: is the following call interruptible? If so, this might
5173 // need to go through the INTERRUPT_IO() wrapper as for other
5174 // blocking, interruptible calls in this file.
5175 int n;
5176 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5177 *bytes = n;
5178 return 1;
5179 }
5180 }
5181 }
5182 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5183 return 0;
5184 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5185 return 0;
5186 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5187 return 0;
5188 }
5189 *bytes = end - cur;
5190 return 1;
5191 }
5192
5193 // Map a block of memory.
5194 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5195 char *addr, size_t bytes, bool read_only,
5196 bool allow_exec) {
5197 int prot;
5198 int flags = MAP_PRIVATE;
5199
5200 if (read_only) {
5201 prot = PROT_READ;
5202 } else {
5203 prot = PROT_READ | PROT_WRITE;
5204 }
5205
5206 if (allow_exec) {
5207 prot |= PROT_EXEC;
5208 }
5209
5210 if (addr != NULL) {
5211 flags |= MAP_FIXED;
5212 }
5213
5214 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5215 fd, file_offset);
5216 if (mapped_address == MAP_FAILED) {
5217 return NULL;
5218 }
5219 return mapped_address;
5220 }
5221
5222
5223 // Remap a block of memory.
5224 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5225 char *addr, size_t bytes, bool read_only,
5226 bool allow_exec) {
5227 // same as map_memory() on this OS
5228 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5229 allow_exec);
5230 }
5231
5232
5233 // Unmap a block of memory.
5234 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5235 return munmap(addr, bytes) == 0;
5236 }
5237
5238 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5239
5240 static clockid_t thread_cpu_clockid(Thread* thread) {
5241 pthread_t tid = thread->osthread()->pthread_id();
5242 clockid_t clockid;
5243
5244 // Get thread clockid
5245 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5246 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5247 return clockid;
5248 }
5249
5250 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5251 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5252 // of a thread.
5253 //
5254 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5255 // the fast estimate available on the platform.
5256
5257 jlong os::current_thread_cpu_time() {
5258 if (os::Linux::supports_fast_thread_cpu_time()) {
5259 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5260 } else {
5261 // return user + sys since the cost is the same
5262 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5263 }
5264 }
5265
5266 jlong os::thread_cpu_time(Thread* thread) {
5267 // consistent with what current_thread_cpu_time() returns
5268 if (os::Linux::supports_fast_thread_cpu_time()) {
5269 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5270 } else {
5271 return slow_thread_cpu_time(thread, true /* user + sys */);
5272 }
5273 }
5274
5275 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5276 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5277 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5278 } else {
5279 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5280 }
5281 }
5282
5283 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5284 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5285 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5286 } else {
5287 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5288 }
5289 }
5290
5291 // -1 on error.
5292 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5293 pid_t tid = thread->osthread()->thread_id();
5294 char *s;
5295 char stat[2048];
5296 int statlen;
5297 char proc_name[64];
5298 int count;
5299 long sys_time, user_time;
5300 char cdummy;
5301 int idummy;
5302 long ldummy;
5303 FILE *fp;
5304
5305 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5306 fp = fopen(proc_name, "r");
5307 if (fp == NULL) return -1;
5308 statlen = fread(stat, 1, 2047, fp);
5309 stat[statlen] = '\0';
5310 fclose(fp);
5311
5312 // Skip pid and the command string. Note that we could be dealing with
5313 // weird command names, e.g. user could decide to rename java launcher
5314 // to "java 1.4.2 :)", then the stat file would look like
5315 // 1234 (java 1.4.2 :)) R ... ...
5316 // We don't really need to know the command string, just find the last
5317 // occurrence of ")" and then start parsing from there. See bug 4726580.
5318 s = strrchr(stat, ')');
5319 if (s == NULL) return -1;
5320
5321 // Skip blank chars
5322 do { s++; } while (s && isspace(*s));
5323
5324 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5325 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5326 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5327 &user_time, &sys_time);
5328 if (count != 13) return -1;
5329 if (user_sys_cpu_time) {
5330 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5331 } else {
5332 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5333 }
5334 }
5335
5336 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5337 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5338 info_ptr->may_skip_backward = false; // elapsed time not wall time
5339 info_ptr->may_skip_forward = false; // elapsed time not wall time
5340 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5341 }
5342
5343 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5344 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5345 info_ptr->may_skip_backward = false; // elapsed time not wall time
5346 info_ptr->may_skip_forward = false; // elapsed time not wall time
5347 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5348 }
5349
5350 bool os::is_thread_cpu_time_supported() {
5351 return true;
5352 }
5353
5354 // System loadavg support. Returns -1 if load average cannot be obtained.
5355 // Linux doesn't yet have a (official) notion of processor sets,
5356 // so just return the system wide load average.
5357 int os::loadavg(double loadavg[], int nelem) {
5358 return ::getloadavg(loadavg, nelem);
5359 }
5360
5361 void os::pause() {
5362 char filename[MAX_PATH];
5363 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5364 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5365 } else {
5366 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5367 }
5368
5369 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5370 if (fd != -1) {
5371 struct stat buf;
5372 ::close(fd);
5373 while (::stat(filename, &buf) == 0) {
5374 (void)::poll(NULL, 0, 100);
5375 }
5376 } else {
5377 jio_fprintf(stderr,
5378 "Could not open pause file '%s', continuing immediately.\n", filename);
5379 }
5380 }
5381
5382
5383 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5384 // comment paragraphs are worth repeating here:
5385 //
5386 // Assumption:
5387 // Only one parker can exist on an event, which is why we allocate
5388 // them per-thread. Multiple unparkers can coexist.
5389 //
5390 // _Event serves as a restricted-range semaphore.
5391 // -1 : thread is blocked, i.e. there is a waiter
5392 // 0 : neutral: thread is running or ready,
5393 // could have been signaled after a wait started
5394 // 1 : signaled - thread is running or ready
5395 //
5396 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5397 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5398 // For specifics regarding the bug see GLIBC BUGID 261237 :
5399 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5400 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5401 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5402 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5403 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5404 // and monitorenter when we're using 1-0 locking. All those operations may result in
5405 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5406 // of libpthread avoids the problem, but isn't practical.
5407 //
5408 // Possible remedies:
5409 //
5410 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5411 // This is palliative and probabilistic, however. If the thread is preempted
5412 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5413 // than the minimum period may have passed, and the abstime may be stale (in the
5414 // past) resultin in a hang. Using this technique reduces the odds of a hang
5415 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5416 //
5417 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5418 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5419 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5420 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5421 // thread.
5422 //
5423 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5424 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5425 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5426 // This also works well. In fact it avoids kernel-level scalability impediments
5427 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5428 // timers in a graceful fashion.
5429 //
5430 // 4. When the abstime value is in the past it appears that control returns
5431 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5432 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5433 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5434 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5435 // It may be possible to avoid reinitialization by checking the return
5436 // value from pthread_cond_timedwait(). In addition to reinitializing the
5437 // condvar we must establish the invariant that cond_signal() is only called
5438 // within critical sections protected by the adjunct mutex. This prevents
5439 // cond_signal() from "seeing" a condvar that's in the midst of being
5440 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5441 // desirable signal-after-unlock optimization that avoids futile context switching.
5442 //
5443 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5444 // structure when a condvar is used or initialized. cond_destroy() would
5445 // release the helper structure. Our reinitialize-after-timedwait fix
5446 // put excessive stress on malloc/free and locks protecting the c-heap.
5447 //
5448 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5449 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5450 // and only enabling the work-around for vulnerable environments.
5451
5452 // utility to compute the abstime argument to timedwait:
5453 // millis is the relative timeout time
5454 // abstime will be the absolute timeout time
5455 // TODO: replace compute_abstime() with unpackTime()
5456
5457 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5458 if (millis < 0) millis = 0;
5459
5460 jlong seconds = millis / 1000;
5461 millis %= 1000;
5462 if (seconds > 50000000) { // see man cond_timedwait(3T)
5463 seconds = 50000000;
5464 }
5465
5466 if (os::supports_monotonic_clock()) {
5467 struct timespec now;
5468 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5469 assert_status(status == 0, status, "clock_gettime");
5470 abstime->tv_sec = now.tv_sec + seconds;
5471 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5472 if (nanos >= NANOSECS_PER_SEC) {
5473 abstime->tv_sec += 1;
5474 nanos -= NANOSECS_PER_SEC;
5475 }
5476 abstime->tv_nsec = nanos;
5477 } else {
5478 struct timeval now;
5479 int status = gettimeofday(&now, NULL);
5480 assert(status == 0, "gettimeofday");
5481 abstime->tv_sec = now.tv_sec + seconds;
5482 long usec = now.tv_usec + millis * 1000;
5483 if (usec >= 1000000) {
5484 abstime->tv_sec += 1;
5485 usec -= 1000000;
5486 }
5487 abstime->tv_nsec = usec * 1000;
5488 }
5489 return abstime;
5490 }
5491
5492 void os::PlatformEvent::park() { // AKA "down()"
5493 // Transitions for _Event:
5494 // -1 => -1 : illegal
5495 // 1 => 0 : pass - return immediately
5496 // 0 => -1 : block; then set _Event to 0 before returning
5497
5498 // Invariant: Only the thread associated with the Event/PlatformEvent
5499 // may call park().
5500 // TODO: assert that _Assoc != NULL or _Assoc == Self
5501 assert(_nParked == 0, "invariant");
5502
5503 int v;
5504 for (;;) {
5505 v = _Event;
5506 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5507 }
5508 guarantee(v >= 0, "invariant");
5509 if (v == 0) {
5510 // Do this the hard way by blocking ...
5511 int status = pthread_mutex_lock(_mutex);
5512 assert_status(status == 0, status, "mutex_lock");
5513 guarantee(_nParked == 0, "invariant");
5514 ++_nParked;
5515 while (_Event < 0) {
5516 status = pthread_cond_wait(_cond, _mutex);
5517 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5518 // Treat this the same as if the wait was interrupted
5519 if (status == ETIME) { status = EINTR; }
5520 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5521 }
5522 --_nParked;
5523
5524 _Event = 0;
5525 status = pthread_mutex_unlock(_mutex);
5526 assert_status(status == 0, status, "mutex_unlock");
5527 // Paranoia to ensure our locked and lock-free paths interact
5528 // correctly with each other.
5529 OrderAccess::fence();
5530 }
5531 guarantee(_Event >= 0, "invariant");
5532 }
5533
5534 int os::PlatformEvent::park(jlong millis) {
5535 // Transitions for _Event:
5536 // -1 => -1 : illegal
5537 // 1 => 0 : pass - return immediately
5538 // 0 => -1 : block; then set _Event to 0 before returning
5539
5540 guarantee(_nParked == 0, "invariant");
5541
5542 int v;
5543 for (;;) {
5544 v = _Event;
5545 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5546 }
5547 guarantee(v >= 0, "invariant");
5548 if (v != 0) return OS_OK;
5549
5550 // We do this the hard way, by blocking the thread.
5551 // Consider enforcing a minimum timeout value.
5552 struct timespec abst;
5553 compute_abstime(&abst, millis);
5554
5555 int ret = OS_TIMEOUT;
5556 int status = pthread_mutex_lock(_mutex);
5557 assert_status(status == 0, status, "mutex_lock");
5558 guarantee(_nParked == 0, "invariant");
5559 ++_nParked;
5560
5561 // Object.wait(timo) will return because of
5562 // (a) notification
5563 // (b) timeout
5564 // (c) thread.interrupt
5565 //
5566 // Thread.interrupt and object.notify{All} both call Event::set.
5567 // That is, we treat thread.interrupt as a special case of notification.
5568 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5569 // We assume all ETIME returns are valid.
5570 //
5571 // TODO: properly differentiate simultaneous notify+interrupt.
5572 // In that case, we should propagate the notify to another waiter.
5573
5574 while (_Event < 0) {
5575 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5576 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5577 pthread_cond_destroy(_cond);
5578 pthread_cond_init(_cond, os::Linux::condAttr());
5579 }
5580 assert_status(status == 0 || status == EINTR ||
5581 status == ETIME || status == ETIMEDOUT,
5582 status, "cond_timedwait");
5583 if (!FilterSpuriousWakeups) break; // previous semantics
5584 if (status == ETIME || status == ETIMEDOUT) break;
5585 // We consume and ignore EINTR and spurious wakeups.
5586 }
5587 --_nParked;
5588 if (_Event >= 0) {
5589 ret = OS_OK;
5590 }
5591 _Event = 0;
5592 status = pthread_mutex_unlock(_mutex);
5593 assert_status(status == 0, status, "mutex_unlock");
5594 assert(_nParked == 0, "invariant");
5595 // Paranoia to ensure our locked and lock-free paths interact
5596 // correctly with each other.
5597 OrderAccess::fence();
5598 return ret;
5599 }
5600
5601 void os::PlatformEvent::unpark() {
5602 // Transitions for _Event:
5603 // 0 => 1 : just return
5604 // 1 => 1 : just return
5605 // -1 => either 0 or 1; must signal target thread
5606 // That is, we can safely transition _Event from -1 to either
5607 // 0 or 1.
5608 // See also: "Semaphores in Plan 9" by Mullender & Cox
5609 //
5610 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5611 // that it will take two back-to-back park() calls for the owning
5612 // thread to block. This has the benefit of forcing a spurious return
5613 // from the first park() call after an unpark() call which will help
5614 // shake out uses of park() and unpark() without condition variables.
5615
5616 if (Atomic::xchg(1, &_Event) >= 0) return;
5617
5618 // Wait for the thread associated with the event to vacate
5619 int status = pthread_mutex_lock(_mutex);
5620 assert_status(status == 0, status, "mutex_lock");
5621 int AnyWaiters = _nParked;
5622 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5623 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5624 AnyWaiters = 0;
5625 pthread_cond_signal(_cond);
5626 }
5627 status = pthread_mutex_unlock(_mutex);
5628 assert_status(status == 0, status, "mutex_unlock");
5629 if (AnyWaiters != 0) {
5630 // Note that we signal() *after* dropping the lock for "immortal" Events.
5631 // This is safe and avoids a common class of futile wakeups. In rare
5632 // circumstances this can cause a thread to return prematurely from
5633 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5634 // will simply re-test the condition and re-park itself.
5635 // This provides particular benefit if the underlying platform does not
5636 // provide wait morphing.
5637 status = pthread_cond_signal(_cond);
5638 assert_status(status == 0, status, "cond_signal");
5639 }
5640 }
5641
5642
5643 // JSR166
5644 // -------------------------------------------------------
5645
5646 // The solaris and linux implementations of park/unpark are fairly
5647 // conservative for now, but can be improved. They currently use a
5648 // mutex/condvar pair, plus a a count.
5649 // Park decrements count if > 0, else does a condvar wait. Unpark
5650 // sets count to 1 and signals condvar. Only one thread ever waits
5651 // on the condvar. Contention seen when trying to park implies that someone
5652 // is unparking you, so don't wait. And spurious returns are fine, so there
5653 // is no need to track notifications.
5654
5655 // This code is common to linux and solaris and will be moved to a
5656 // common place in dolphin.
5657 //
5658 // The passed in time value is either a relative time in nanoseconds
5659 // or an absolute time in milliseconds. Either way it has to be unpacked
5660 // into suitable seconds and nanoseconds components and stored in the
5661 // given timespec structure.
5662 // Given time is a 64-bit value and the time_t used in the timespec is only
5663 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5664 // overflow if times way in the future are given. Further on Solaris versions
5665 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5666 // number of seconds, in abstime, is less than current_time + 100,000,000.
5667 // As it will be 28 years before "now + 100000000" will overflow we can
5668 // ignore overflow and just impose a hard-limit on seconds using the value
5669 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5670 // years from "now".
5671
5672 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5673 assert(time > 0, "convertTime");
5674 time_t max_secs = 0;
5675
5676 if (!os::supports_monotonic_clock() || isAbsolute) {
5677 struct timeval now;
5678 int status = gettimeofday(&now, NULL);
5679 assert(status == 0, "gettimeofday");
5680
5681 max_secs = now.tv_sec + MAX_SECS;
5682
5683 if (isAbsolute) {
5684 jlong secs = time / 1000;
5685 if (secs > max_secs) {
5686 absTime->tv_sec = max_secs;
5687 } else {
5688 absTime->tv_sec = secs;
5689 }
5690 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5691 } else {
5692 jlong secs = time / NANOSECS_PER_SEC;
5693 if (secs >= MAX_SECS) {
5694 absTime->tv_sec = max_secs;
5695 absTime->tv_nsec = 0;
5696 } else {
5697 absTime->tv_sec = now.tv_sec + secs;
5698 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5699 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5700 absTime->tv_nsec -= NANOSECS_PER_SEC;
5701 ++absTime->tv_sec; // note: this must be <= max_secs
5702 }
5703 }
5704 }
5705 } else {
5706 // must be relative using monotonic clock
5707 struct timespec now;
5708 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5709 assert_status(status == 0, status, "clock_gettime");
5710 max_secs = now.tv_sec + MAX_SECS;
5711 jlong secs = time / NANOSECS_PER_SEC;
5712 if (secs >= MAX_SECS) {
5713 absTime->tv_sec = max_secs;
5714 absTime->tv_nsec = 0;
5715 } else {
5716 absTime->tv_sec = now.tv_sec + secs;
5717 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5718 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5719 absTime->tv_nsec -= NANOSECS_PER_SEC;
5720 ++absTime->tv_sec; // note: this must be <= max_secs
5721 }
5722 }
5723 }
5724 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5725 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5726 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5727 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5728 }
5729
5730 void Parker::park(bool isAbsolute, jlong time) {
5731 // Ideally we'd do something useful while spinning, such
5732 // as calling unpackTime().
5733
5734 // Optional fast-path check:
5735 // Return immediately if a permit is available.
5736 // We depend on Atomic::xchg() having full barrier semantics
5737 // since we are doing a lock-free update to _counter.
5738 if (Atomic::xchg(0, &_counter) > 0) return;
5739
5740 Thread* thread = Thread::current();
5741 assert(thread->is_Java_thread(), "Must be JavaThread");
5742 JavaThread *jt = (JavaThread *)thread;
5743
5744 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5745 // Check interrupt before trying to wait
5746 if (Thread::is_interrupted(thread, false)) {
5747 return;
5748 }
5749
5750 // Next, demultiplex/decode time arguments
5751 timespec absTime;
5752 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5753 return;
5754 }
5755 if (time > 0) {
5756 unpackTime(&absTime, isAbsolute, time);
5757 }
5758
5759
5760 // Enter safepoint region
5761 // Beware of deadlocks such as 6317397.
5762 // The per-thread Parker:: mutex is a classic leaf-lock.
5763 // In particular a thread must never block on the Threads_lock while
5764 // holding the Parker:: mutex. If safepoints are pending both the
5765 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5766 ThreadBlockInVM tbivm(jt);
5767
5768 // Don't wait if cannot get lock since interference arises from
5769 // unblocking. Also. check interrupt before trying wait
5770 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5771 return;
5772 }
5773
5774 int status;
5775 if (_counter > 0) { // no wait needed
5776 _counter = 0;
5777 status = pthread_mutex_unlock(_mutex);
5778 assert(status == 0, "invariant");
5779 // Paranoia to ensure our locked and lock-free paths interact
5780 // correctly with each other and Java-level accesses.
5781 OrderAccess::fence();
5782 return;
5783 }
5784
5785 #ifdef ASSERT
5786 // Don't catch signals while blocked; let the running threads have the signals.
5787 // (This allows a debugger to break into the running thread.)
5788 sigset_t oldsigs;
5789 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5790 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5791 #endif
5792
5793 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5794 jt->set_suspend_equivalent();
5795 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5796
5797 assert(_cur_index == -1, "invariant");
5798 if (time == 0) {
5799 _cur_index = REL_INDEX; // arbitrary choice when not timed
5800 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5801 } else {
5802 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5803 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5804 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5805 pthread_cond_destroy(&_cond[_cur_index]);
5806 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5807 }
5808 }
5809 _cur_index = -1;
5810 assert_status(status == 0 || status == EINTR ||
5811 status == ETIME || status == ETIMEDOUT,
5812 status, "cond_timedwait");
5813
5814 #ifdef ASSERT
5815 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5816 #endif
5817
5818 _counter = 0;
5819 status = pthread_mutex_unlock(_mutex);
5820 assert_status(status == 0, status, "invariant");
5821 // Paranoia to ensure our locked and lock-free paths interact
5822 // correctly with each other and Java-level accesses.
5823 OrderAccess::fence();
5824
5825 // If externally suspended while waiting, re-suspend
5826 if (jt->handle_special_suspend_equivalent_condition()) {
5827 jt->java_suspend_self();
5828 }
5829 }
5830
5831 void Parker::unpark() {
5832 int status = pthread_mutex_lock(_mutex);
5833 assert(status == 0, "invariant");
5834 const int s = _counter;
5835 _counter = 1;
5836 if (s < 1) {
5837 // thread might be parked
5838 if (_cur_index != -1) {
5839 // thread is definitely parked
5840 if (WorkAroundNPTLTimedWaitHang) {
5841 status = pthread_cond_signal(&_cond[_cur_index]);
5842 assert(status == 0, "invariant");
5843 status = pthread_mutex_unlock(_mutex);
5844 assert(status == 0, "invariant");
5845 } else {
5846 status = pthread_mutex_unlock(_mutex);
5847 assert(status == 0, "invariant");
5848 status = pthread_cond_signal(&_cond[_cur_index]);
5849 assert(status == 0, "invariant");
5850 }
5851 } else {
5852 pthread_mutex_unlock(_mutex);
5853 assert(status == 0, "invariant");
5854 }
5855 } else {
5856 pthread_mutex_unlock(_mutex);
5857 assert(status == 0, "invariant");
5858 }
5859 }
5860
5861
5862 extern char** environ;
5863
5864 #ifndef __NR_fork
5865 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5866 #endif
5867
5868 #ifndef __NR_execve
5869 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5870 #endif
5871
5872 // Run the specified command in a separate process. Return its exit value,
5873 // or -1 on failure (e.g. can't fork a new process).
5874 // Unlike system(), this function can be called from signal handler. It
5875 // doesn't block SIGINT et al.
5876 int os::fork_and_exec(char* cmd) {
5877 const char * argv[4] = {"sh", "-c", cmd, NULL};
5878
5879 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5880 // pthread_atfork handlers and reset pthread library. All we need is a
5881 // separate process to execve. Make a direct syscall to fork process.
5882 // On IA64 there's no fork syscall, we have to use fork() and hope for
5883 // the best...
5884 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5885 IA64_ONLY(fork();)
5886
5887 if (pid < 0) {
5888 // fork failed
5889 return -1;
5890
5891 } else if (pid == 0) {
5892 // child process
5893
5894 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5895 // first to kill every thread on the thread list. Because this list is
5896 // not reset by fork() (see notes above), execve() will instead kill
5897 // every thread in the parent process. We know this is the only thread
5898 // in the new process, so make a system call directly.
5899 // IA64 should use normal execve() from glibc to match the glibc fork()
5900 // above.
5901 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5902 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5903
5904 // execve failed
5905 _exit(-1);
5906
5907 } else {
5908 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5909 // care about the actual exit code, for now.
5910
5911 int status;
5912
5913 // Wait for the child process to exit. This returns immediately if
5914 // the child has already exited. */
5915 while (waitpid(pid, &status, 0) < 0) {
5916 switch (errno) {
5917 case ECHILD: return 0;
5918 case EINTR: break;
5919 default: return -1;
5920 }
5921 }
5922
5923 if (WIFEXITED(status)) {
5924 // The child exited normally; get its exit code.
5925 return WEXITSTATUS(status);
5926 } else if (WIFSIGNALED(status)) {
5927 // The child exited because of a signal
5928 // The best value to return is 0x80 + signal number,
5929 // because that is what all Unix shells do, and because
5930 // it allows callers to distinguish between process exit and
5931 // process death by signal.
5932 return 0x80 + WTERMSIG(status);
5933 } else {
5934 // Unknown exit code; pass it through
5935 return status;
5936 }
5937 }
5938 }
5939
5940 // is_headless_jre()
5941 //
5942 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5943 // in order to report if we are running in a headless jre
5944 //
5945 // Since JDK8 xawt/libmawt.so was moved into the same directory
5946 // as libawt.so, and renamed libawt_xawt.so
5947 //
5948 bool os::is_headless_jre() {
5949 struct stat statbuf;
5950 char buf[MAXPATHLEN];
5951 char libmawtpath[MAXPATHLEN];
5952 const char *xawtstr = "/xawt/libmawt.so";
5953 const char *new_xawtstr = "/libawt_xawt.so";
5954 char *p;
5955
5956 // Get path to libjvm.so
5957 os::jvm_path(buf, sizeof(buf));
5958
5959 // Get rid of libjvm.so
5960 p = strrchr(buf, '/');
5961 if (p == NULL) {
5962 return false;
5963 } else {
5964 *p = '\0';
5965 }
5966
5967 // Get rid of client or server
5968 p = strrchr(buf, '/');
5969 if (p == NULL) {
5970 return false;
5971 } else {
5972 *p = '\0';
5973 }
5974
5975 // check xawt/libmawt.so
5976 strcpy(libmawtpath, buf);
5977 strcat(libmawtpath, xawtstr);
5978 if (::stat(libmawtpath, &statbuf) == 0) return false;
5979
5980 // check libawt_xawt.so
5981 strcpy(libmawtpath, buf);
5982 strcat(libmawtpath, new_xawtstr);
5983 if (::stat(libmawtpath, &statbuf) == 0) return false;
5984
5985 return true;
5986 }
5987
5988 // Get the default path to the core file
5989 // Returns the length of the string
5990 int os::get_core_path(char* buffer, size_t bufferSize) {
5991 /*
5992 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5993 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5994 */
5995 const int core_pattern_len = 129;
5996 char core_pattern[core_pattern_len] = {0};
5997
5998 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5999 if (core_pattern_file != -1) {
6000 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6001 ::close(core_pattern_file);
6002
6003 if (ret > 0) {
6004 char *last_char = core_pattern + strlen(core_pattern) - 1;
6005
6006 if (*last_char == '\n') {
6007 *last_char = '\0';
6008 }
6009 }
6010 }
6011
6012 if (strlen(core_pattern) == 0) {
6013 return 0;
6014 }
6015
6016 char *pid_pos = strstr(core_pattern, "%p");
6017 size_t written;
6018
6019 if (core_pattern[0] == '/') {
6020 written = jio_snprintf(buffer, bufferSize, core_pattern);
6021 } else {
6022 char cwd[PATH_MAX];
6023
6024 const char* p = get_current_directory(cwd, PATH_MAX);
6025 if (p == NULL) {
6026 assert(p != NULL, "failed to get current directory");
6027 return 0;
6028 }
6029
6030 if (core_pattern[0] == '|') {
6031 written = jio_snprintf(buffer, bufferSize,
6032 "\"%s\" (or dumping to %s/core.%d)",
6033 &core_pattern[1], p, current_process_id());
6034 } else {
6035 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6036 }
6037 }
6038
6039 if ((written >= 0) && (written < bufferSize)
6040 && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6041 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6042
6043 if (core_uses_pid_file != -1) {
6044 char core_uses_pid = 0;
6045 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6046 ::close(core_uses_pid_file);
6047
6048 if (core_uses_pid == '1'){
6049 jio_snprintf(buffer + written, bufferSize - written,
6050 ".%d", current_process_id());
6051 }
6052 }
6053 }
6054
6055 return strlen(buffer);
6056 }
6057
6058 /////////////// Unit tests ///////////////
6059
6060 #ifndef PRODUCT
6061
6062 #define test_log(...) \
6063 do { \
6064 if (VerboseInternalVMTests) { \
6065 tty->print_cr(__VA_ARGS__); \
6066 tty->flush(); \
6067 } \
6068 } while (false)
6069
6070 class TestReserveMemorySpecial : AllStatic {
6071 public:
6072 static void small_page_write(void* addr, size_t size) {
6073 size_t page_size = os::vm_page_size();
6074
6075 char* end = (char*)addr + size;
6076 for (char* p = (char*)addr; p < end; p += page_size) {
6077 *p = 1;
6078 }
6079 }
6080
6081 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6082 if (!UseHugeTLBFS) {
6083 return;
6084 }
6085
6086 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6087
6088 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6089
6090 if (addr != NULL) {
6091 small_page_write(addr, size);
6092
6093 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6094 }
6095 }
6096
6097 static void test_reserve_memory_special_huge_tlbfs_only() {
6098 if (!UseHugeTLBFS) {
6099 return;
6100 }
6101
6102 size_t lp = os::large_page_size();
6103
6104 for (size_t size = lp; size <= lp * 10; size += lp) {
6105 test_reserve_memory_special_huge_tlbfs_only(size);
6106 }
6107 }
6108
6109 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6110 if (!UseHugeTLBFS) {
6111 return;
6112 }
6113
6114 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6115 size, alignment);
6116
6117 assert(size >= os::large_page_size(), "Incorrect input to test");
6118
6119 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6120
6121 if (addr != NULL) {
6122 small_page_write(addr, size);
6123
6124 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6125 }
6126 }
6127
6128 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6129 size_t lp = os::large_page_size();
6130 size_t ag = os::vm_allocation_granularity();
6131
6132 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6133 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6134 }
6135 }
6136
6137 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6138 size_t lp = os::large_page_size();
6139 size_t ag = os::vm_allocation_granularity();
6140
6141 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6142 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6143 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6144 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6145 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6146 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6147 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6148 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6149 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6150 }
6151
6152 static void test_reserve_memory_special_huge_tlbfs() {
6153 if (!UseHugeTLBFS) {
6154 return;
6155 }
6156
6157 test_reserve_memory_special_huge_tlbfs_only();
6158 test_reserve_memory_special_huge_tlbfs_mixed();
6159 }
6160
6161 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6162 if (!UseSHM) {
6163 return;
6164 }
6165
6166 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6167
6168 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6169
6170 if (addr != NULL) {
6171 assert(is_ptr_aligned(addr, alignment), "Check");
6172 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6173
6174 small_page_write(addr, size);
6175
6176 os::Linux::release_memory_special_shm(addr, size);
6177 }
6178 }
6179
6180 static void test_reserve_memory_special_shm() {
6181 size_t lp = os::large_page_size();
6182 size_t ag = os::vm_allocation_granularity();
6183
6184 for (size_t size = ag; size < lp * 3; size += ag) {
6185 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6186 test_reserve_memory_special_shm(size, alignment);
6187 }
6188 }
6189 }
6190
6191 static void test() {
6192 test_reserve_memory_special_huge_tlbfs();
6193 test_reserve_memory_special_shm();
6194 }
6195 };
6196
6197 void TestReserveMemorySpecial_test() {
6198 TestReserveMemorySpecial::test();
6199 }
6200
6201 #endif