1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_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 // Declarations
162 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
163
164 // utility functions
165
166 static int SR_initialize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 sysinfo(&si);
176
177 return (julong)si.freeram * si.mem_unit;
178 }
179
180 julong os::physical_memory() {
181 return Linux::physical_memory();
182 }
183
184 // Return true if user is running as root.
185
186 bool os::have_special_privileges() {
187 static bool init = false;
188 static bool privileges = false;
189 if (!init) {
190 privileges = (getuid() != geteuid()) || (getgid() != getegid());
191 init = true;
192 }
193 return privileges;
194 }
195
196
197 #ifndef SYS_gettid
198 // i386: 224, ia64: 1105, amd64: 186, sparc 143
199 #ifdef __ia64__
200 #define SYS_gettid 1105
201 #elif __i386__
202 #define SYS_gettid 224
203 #elif __amd64__
204 #define SYS_gettid 186
205 #elif __sparc__
206 #define SYS_gettid 143
207 #else
208 #error define gettid for the arch
209 #endif
210 #endif
211
212 // Cpu architecture string
213 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
214
215
216 // pid_t gettid()
217 //
218 // Returns the kernel thread id of the currently running thread. Kernel
219 // thread id is used to access /proc.
220 //
221 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
222 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
223 //
224 pid_t os::Linux::gettid() {
225 int rslt = syscall(SYS_gettid);
226 if (rslt == -1) {
227 // old kernel, no NPTL support
228 return getpid();
229 } else {
230 return (pid_t)rslt;
231 }
232 }
233
234 // Most versions of linux have a bug where the number of processors are
235 // determined by looking at the /proc file system. In a chroot environment,
236 // the system call returns 1. This causes the VM to act as if it is
237 // a single processor and elide locking (see is_MP() call).
238 static bool unsafe_chroot_detected = false;
239 static const char *unstable_chroot_error = "/proc file system not found.\n"
240 "Java may be unstable running multithreaded in a chroot "
241 "environment on Linux when /proc filesystem is not mounted.";
242
243 void os::Linux::initialize_system_info() {
244 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
245 if (processor_count() == 1) {
246 pid_t pid = os::Linux::gettid();
247 char fname[32];
248 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
249 FILE *fp = fopen(fname, "r");
250 if (fp == NULL) {
251 unsafe_chroot_detected = true;
252 } else {
253 fclose(fp);
254 }
255 }
256 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
257 assert(processor_count() > 0, "linux error");
258 }
259
260 void os::init_system_properties_values() {
261 // The next steps are taken in the product version:
262 //
263 // Obtain the JAVA_HOME value from the location of libjvm.so.
264 // This library should be located at:
265 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
266 //
267 // If "/jre/lib/" appears at the right place in the path, then we
268 // assume libjvm.so is installed in a JDK and we use this path.
269 //
270 // Otherwise exit with message: "Could not create the Java virtual machine."
271 //
272 // The following extra steps are taken in the debugging version:
273 //
274 // If "/jre/lib/" does NOT appear at the right place in the path
275 // instead of exit check for $JAVA_HOME environment variable.
276 //
277 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
278 // then we append a fake suffix "hotspot/libjvm.so" to this path so
279 // it looks like libjvm.so is installed there
280 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
281 //
282 // Otherwise exit.
283 //
284 // Important note: if the location of libjvm.so changes this
285 // code needs to be changed accordingly.
286
287 // See ld(1):
288 // The linker uses the following search paths to locate required
289 // shared libraries:
290 // 1: ...
291 // ...
292 // 7: The default directories, normally /lib and /usr/lib.
293 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
294 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
295 #else
296 #define DEFAULT_LIBPATH "/lib:/usr/lib"
297 #endif
298
299 // Base path of extensions installed on the system.
300 #define SYS_EXT_DIR "/usr/java/packages"
301 #define EXTENSIONS_DIR "/lib/ext"
302
303 // Buffer that fits several sprintfs.
304 // Note that the space for the colon and the trailing null are provided
305 // by the nulls included by the sizeof operator.
306 const size_t bufsize =
307 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
308 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
309 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
310
311 // sysclasspath, java_home, dll_dir
312 {
313 char *pslash;
314 os::jvm_path(buf, bufsize);
315
316 // Found the full path to libjvm.so.
317 // Now cut the path to <java_home>/jre if we can.
318 pslash = strrchr(buf, '/');
319 if (pslash != NULL) {
320 *pslash = '\0'; // Get rid of /libjvm.so.
321 }
322 pslash = strrchr(buf, '/');
323 if (pslash != NULL) {
324 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
325 }
326 Arguments::set_dll_dir(buf);
327
328 if (pslash != NULL) {
329 pslash = strrchr(buf, '/');
330 if (pslash != NULL) {
331 *pslash = '\0'; // Get rid of /<arch>.
332 pslash = strrchr(buf, '/');
333 if (pslash != NULL) {
334 *pslash = '\0'; // Get rid of /lib.
335 }
336 }
337 }
338 Arguments::set_java_home(buf);
339 set_boot_path('/', ':');
340 }
341
342 // Where to look for native libraries.
343 //
344 // Note: Due to a legacy implementation, most of the library path
345 // is set in the launcher. This was to accomodate linking restrictions
346 // on legacy Linux implementations (which are no longer supported).
347 // Eventually, all the library path setting will be done here.
348 //
349 // However, to prevent the proliferation of improperly built native
350 // libraries, the new path component /usr/java/packages is added here.
351 // Eventually, all the library path setting will be done here.
352 {
353 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
354 // should always exist (until the legacy problem cited above is
355 // addressed).
356 const char *v = ::getenv("LD_LIBRARY_PATH");
357 const char *v_colon = ":";
358 if (v == NULL) { v = ""; v_colon = ""; }
359 // That's +1 for the colon and +1 for the trailing '\0'.
360 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
361 strlen(v) + 1 +
362 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
363 mtInternal);
364 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
365 Arguments::set_library_path(ld_library_path);
366 FREE_C_HEAP_ARRAY(char, ld_library_path);
367 }
368
369 // Extensions directories.
370 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
371 Arguments::set_ext_dirs(buf);
372
373 FREE_C_HEAP_ARRAY(char, buf);
374
375 #undef DEFAULT_LIBPATH
376 #undef SYS_EXT_DIR
377 #undef EXTENSIONS_DIR
378 }
379
380 ////////////////////////////////////////////////////////////////////////////////
381 // breakpoint support
382
383 void os::breakpoint() {
384 BREAKPOINT;
385 }
386
387 extern "C" void breakpoint() {
388 // use debugger to set breakpoint here
389 }
390
391 ////////////////////////////////////////////////////////////////////////////////
392 // signal support
393
394 debug_only(static bool signal_sets_initialized = false);
395 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
396
397 bool os::Linux::is_sig_ignored(int sig) {
398 struct sigaction oact;
399 sigaction(sig, (struct sigaction*)NULL, &oact);
400 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
401 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
402 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
403 return true;
404 } else {
405 return false;
406 }
407 }
408
409 void os::Linux::signal_sets_init() {
410 // Should also have an assertion stating we are still single-threaded.
411 assert(!signal_sets_initialized, "Already initialized");
412 // Fill in signals that are necessarily unblocked for all threads in
413 // the VM. Currently, we unblock the following signals:
414 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
415 // by -Xrs (=ReduceSignalUsage));
416 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
417 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
418 // the dispositions or masks wrt these signals.
419 // Programs embedding the VM that want to use the above signals for their
420 // own purposes must, at this time, use the "-Xrs" option to prevent
421 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
422 // (See bug 4345157, and other related bugs).
423 // In reality, though, unblocking these signals is really a nop, since
424 // these signals are not blocked by default.
425 sigemptyset(&unblocked_sigs);
426 sigemptyset(&allowdebug_blocked_sigs);
427 sigaddset(&unblocked_sigs, SIGILL);
428 sigaddset(&unblocked_sigs, SIGSEGV);
429 sigaddset(&unblocked_sigs, SIGBUS);
430 sigaddset(&unblocked_sigs, SIGFPE);
431 #if defined(PPC64)
432 sigaddset(&unblocked_sigs, SIGTRAP);
433 #endif
434 sigaddset(&unblocked_sigs, SR_signum);
435
436 if (!ReduceSignalUsage) {
437 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
438 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
439 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
440 }
441 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
442 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
443 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
444 }
445 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
446 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
447 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
448 }
449 }
450 // Fill in signals that are blocked by all but the VM thread.
451 sigemptyset(&vm_sigs);
452 if (!ReduceSignalUsage) {
453 sigaddset(&vm_sigs, BREAK_SIGNAL);
454 }
455 debug_only(signal_sets_initialized = true);
456
457 }
458
459 // These are signals that are unblocked while a thread is running Java.
460 // (For some reason, they get blocked by default.)
461 sigset_t* os::Linux::unblocked_signals() {
462 assert(signal_sets_initialized, "Not initialized");
463 return &unblocked_sigs;
464 }
465
466 // These are the signals that are blocked while a (non-VM) thread is
467 // running Java. Only the VM thread handles these signals.
468 sigset_t* os::Linux::vm_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &vm_sigs;
471 }
472
473 // These are signals that are blocked during cond_wait to allow debugger in
474 sigset_t* os::Linux::allowdebug_blocked_signals() {
475 assert(signal_sets_initialized, "Not initialized");
476 return &allowdebug_blocked_sigs;
477 }
478
479 void os::Linux::hotspot_sigmask(Thread* thread) {
480
481 //Save caller's signal mask before setting VM signal mask
482 sigset_t caller_sigmask;
483 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
484
485 OSThread* osthread = thread->osthread();
486 osthread->set_caller_sigmask(caller_sigmask);
487
488 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
489
490 if (!ReduceSignalUsage) {
491 if (thread->is_VM_thread()) {
492 // Only the VM thread handles BREAK_SIGNAL ...
493 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
494 } else {
495 // ... all other threads block BREAK_SIGNAL
496 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
497 }
498 }
499 }
500
501 //////////////////////////////////////////////////////////////////////////////
502 // detecting pthread library
503
504 void os::Linux::libpthread_init() {
505 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
506 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
507 // generic name for earlier versions.
508 // Define macros here so we can build HotSpot on old systems.
509 #ifndef _CS_GNU_LIBC_VERSION
510 #define _CS_GNU_LIBC_VERSION 2
511 #endif
512 #ifndef _CS_GNU_LIBPTHREAD_VERSION
513 #define _CS_GNU_LIBPTHREAD_VERSION 3
514 #endif
515
516 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
517 if (n > 0) {
518 char *str = (char *)malloc(n, mtInternal);
519 confstr(_CS_GNU_LIBC_VERSION, str, n);
520 os::Linux::set_glibc_version(str);
521 } else {
522 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
523 static char _gnu_libc_version[32];
524 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
525 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
526 os::Linux::set_glibc_version(_gnu_libc_version);
527 }
528
529 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
530 if (n > 0) {
531 char *str = (char *)malloc(n, mtInternal);
532 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
533 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
534 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
535 // is the case. LinuxThreads has a hard limit on max number of threads.
536 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
537 // On the other hand, NPTL does not have such a limit, sysconf()
538 // will return -1 and errno is not changed. Check if it is really NPTL.
539 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
540 strstr(str, "NPTL") &&
541 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
542 free(str);
543 os::Linux::set_libpthread_version("linuxthreads");
544 } else {
545 os::Linux::set_libpthread_version(str);
546 }
547 } else {
548 // glibc before 2.3.2 only has LinuxThreads.
549 os::Linux::set_libpthread_version("linuxthreads");
550 }
551
552 if (strstr(libpthread_version(), "NPTL")) {
553 os::Linux::set_is_NPTL();
554 } else {
555 os::Linux::set_is_LinuxThreads();
556 }
557
558 // LinuxThreads have two flavors: floating-stack mode, which allows variable
559 // stack size; and fixed-stack mode. NPTL is always floating-stack.
560 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
561 os::Linux::set_is_floating_stack();
562 }
563 }
564
565 /////////////////////////////////////////////////////////////////////////////
566 // thread stack
567
568 // Force Linux kernel to expand current thread stack. If "bottom" is close
569 // to the stack guard, caller should block all signals.
570 //
571 // MAP_GROWSDOWN:
572 // A special mmap() flag that is used to implement thread stacks. It tells
573 // kernel that the memory region should extend downwards when needed. This
574 // allows early versions of LinuxThreads to only mmap the first few pages
575 // when creating a new thread. Linux kernel will automatically expand thread
576 // stack as needed (on page faults).
577 //
578 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
579 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
580 // region, it's hard to tell if the fault is due to a legitimate stack
581 // access or because of reading/writing non-exist memory (e.g. buffer
582 // overrun). As a rule, if the fault happens below current stack pointer,
583 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
584 // application (see Linux kernel fault.c).
585 //
586 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
587 // stack overflow detection.
588 //
589 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
590 // not use this flag. However, the stack of initial thread is not created
591 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
592 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
593 // and then attach the thread to JVM.
594 //
595 // To get around the problem and allow stack banging on Linux, we need to
596 // manually expand thread stack after receiving the SIGSEGV.
597 //
598 // There are two ways to expand thread stack to address "bottom", we used
599 // both of them in JVM before 1.5:
600 // 1. adjust stack pointer first so that it is below "bottom", and then
601 // touch "bottom"
602 // 2. mmap() the page in question
603 //
604 // Now alternate signal stack is gone, it's harder to use 2. For instance,
605 // if current sp is already near the lower end of page 101, and we need to
606 // call mmap() to map page 100, it is possible that part of the mmap() frame
607 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
608 // That will destroy the mmap() frame and cause VM to crash.
609 //
610 // The following code works by adjusting sp first, then accessing the "bottom"
611 // page to force a page fault. Linux kernel will then automatically expand the
612 // stack mapping.
613 //
614 // _expand_stack_to() assumes its frame size is less than page size, which
615 // should always be true if the function is not inlined.
616
617 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
618 #define NOINLINE
619 #else
620 #define NOINLINE __attribute__ ((noinline))
621 #endif
622
623 static void _expand_stack_to(address bottom) NOINLINE;
624
625 static void _expand_stack_to(address bottom) {
626 address sp;
627 size_t size;
628 volatile char *p;
629
630 // Adjust bottom to point to the largest address within the same page, it
631 // gives us a one-page buffer if alloca() allocates slightly more memory.
632 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
633 bottom += os::Linux::page_size() - 1;
634
635 // sp might be slightly above current stack pointer; if that's the case, we
636 // will alloca() a little more space than necessary, which is OK. Don't use
637 // os::current_stack_pointer(), as its result can be slightly below current
638 // stack pointer, causing us to not alloca enough to reach "bottom".
639 sp = (address)&sp;
640
641 if (sp > bottom) {
642 size = sp - bottom;
643 p = (volatile char *)alloca(size);
644 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
645 p[0] = '\0';
646 }
647 }
648
649 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
650 assert(t!=NULL, "just checking");
651 assert(t->osthread()->expanding_stack(), "expand should be set");
652 assert(t->stack_base() != NULL, "stack_base was not initialized");
653
654 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
655 sigset_t mask_all, old_sigset;
656 sigfillset(&mask_all);
657 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
658 _expand_stack_to(addr);
659 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
660 return true;
661 }
662 return false;
663 }
664
665 //////////////////////////////////////////////////////////////////////////////
666 // create new thread
667
668 static address highest_vm_reserved_address();
669
670 // check if it's safe to start a new thread
671 static bool _thread_safety_check(Thread* thread) {
672 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
673 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
674 // Heap is mmap'ed at lower end of memory space. Thread stacks are
675 // allocated (MAP_FIXED) from high address space. Every thread stack
676 // occupies a fixed size slot (usually 2Mbytes, but user can change
677 // it to other values if they rebuild LinuxThreads).
678 //
679 // Problem with MAP_FIXED is that mmap() can still succeed even part of
680 // the memory region has already been mmap'ed. That means if we have too
681 // many threads and/or very large heap, eventually thread stack will
682 // collide with heap.
683 //
684 // Here we try to prevent heap/stack collision by comparing current
685 // stack bottom with the highest address that has been mmap'ed by JVM
686 // plus a safety margin for memory maps created by native code.
687 //
688 // This feature can be disabled by setting ThreadSafetyMargin to 0
689 //
690 if (ThreadSafetyMargin > 0) {
691 address stack_bottom = os::current_stack_base() - os::current_stack_size();
692
693 // not safe if our stack extends below the safety margin
694 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
695 } else {
696 return true;
697 }
698 } else {
699 // Floating stack LinuxThreads or NPTL:
700 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
701 // there's not enough space left, pthread_create() will fail. If we come
702 // here, that means enough space has been reserved for stack.
703 return true;
704 }
705 }
706
707 // Thread start routine for all newly created threads
708 static void *java_start(Thread *thread) {
709 // Try to randomize the cache line index of hot stack frames.
710 // This helps when threads of the same stack traces evict each other's
711 // cache lines. The threads can be either from the same JVM instance, or
712 // from different JVM instances. The benefit is especially true for
713 // processors with hyperthreading technology.
714 static int counter = 0;
715 int pid = os::current_process_id();
716 alloca(((pid ^ counter++) & 7) * 128);
717
718 ThreadLocalStorage::set_thread(thread);
719
720 OSThread* osthread = thread->osthread();
721 Monitor* sync = osthread->startThread_lock();
722
723 // non floating stack LinuxThreads needs extra check, see above
724 if (!_thread_safety_check(thread)) {
725 // notify parent thread
726 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
727 osthread->set_state(ZOMBIE);
728 sync->notify_all();
729 return NULL;
730 }
731
732 // thread_id is kernel thread id (similar to Solaris LWP id)
733 osthread->set_thread_id(os::Linux::gettid());
734
735 if (UseNUMA) {
736 int lgrp_id = os::numa_get_group_id();
737 if (lgrp_id != -1) {
738 thread->set_lgrp_id(lgrp_id);
739 }
740 }
741 // initialize signal mask for this thread
742 os::Linux::hotspot_sigmask(thread);
743
744 // initialize floating point control register
745 os::Linux::init_thread_fpu_state();
746
747 // handshaking with parent thread
748 {
749 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
750
751 // notify parent thread
752 osthread->set_state(INITIALIZED);
753 sync->notify_all();
754
755 // wait until os::start_thread()
756 while (osthread->get_state() == INITIALIZED) {
757 sync->wait(Mutex::_no_safepoint_check_flag);
758 }
759 }
760
761 // call one more level start routine
762 thread->run();
763
764 return 0;
765 }
766
767 bool os::create_thread(Thread* thread, ThreadType thr_type,
768 size_t stack_size) {
769 assert(thread->osthread() == NULL, "caller responsible");
770
771 // Allocate the OSThread object
772 OSThread* osthread = new OSThread(NULL, NULL);
773 if (osthread == NULL) {
774 return false;
775 }
776
777 // set the correct thread state
778 osthread->set_thread_type(thr_type);
779
780 // Initial state is ALLOCATED but not INITIALIZED
781 osthread->set_state(ALLOCATED);
782
783 thread->set_osthread(osthread);
784
785 // init thread attributes
786 pthread_attr_t attr;
787 pthread_attr_init(&attr);
788 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
789
790 // stack size
791 if (os::Linux::supports_variable_stack_size()) {
792 // calculate stack size if it's not specified by caller
793 if (stack_size == 0) {
794 stack_size = os::Linux::default_stack_size(thr_type);
795
796 switch (thr_type) {
797 case os::java_thread:
798 // Java threads use ThreadStackSize which default value can be
799 // changed with the flag -Xss
800 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
801 stack_size = JavaThread::stack_size_at_create();
802 break;
803 case os::compiler_thread:
804 if (CompilerThreadStackSize > 0) {
805 stack_size = (size_t)(CompilerThreadStackSize * K);
806 break;
807 } // else fall through:
808 // use VMThreadStackSize if CompilerThreadStackSize is not defined
809 case os::vm_thread:
810 case os::pgc_thread:
811 case os::cgc_thread:
812 case os::watcher_thread:
813 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
814 break;
815 }
816 }
817
818 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
819 pthread_attr_setstacksize(&attr, stack_size);
820 } else {
821 // let pthread_create() pick the default value.
822 }
823
824 // glibc guard page
825 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
826
827 ThreadState state;
828
829 {
830 // Serialize thread creation if we are running with fixed stack LinuxThreads
831 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
832 if (lock) {
833 os::Linux::createThread_lock()->lock_without_safepoint_check();
834 }
835
836 pthread_t tid;
837 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
838
839 pthread_attr_destroy(&attr);
840
841 if (ret != 0) {
842 if (PrintMiscellaneous && (Verbose || WizardMode)) {
843 perror("pthread_create()");
844 }
845 // Need to clean up stuff we've allocated so far
846 thread->set_osthread(NULL);
847 delete osthread;
848 if (lock) os::Linux::createThread_lock()->unlock();
849 return false;
850 }
851
852 // Store pthread info into the OSThread
853 osthread->set_pthread_id(tid);
854
855 // Wait until child thread is either initialized or aborted
856 {
857 Monitor* sync_with_child = osthread->startThread_lock();
858 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
859 while ((state = osthread->get_state()) == ALLOCATED) {
860 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
861 }
862 }
863
864 if (lock) {
865 os::Linux::createThread_lock()->unlock();
866 }
867 }
868
869 // Aborted due to thread limit being reached
870 if (state == ZOMBIE) {
871 thread->set_osthread(NULL);
872 delete osthread;
873 return false;
874 }
875
876 // The thread is returned suspended (in state INITIALIZED),
877 // and is started higher up in the call chain
878 assert(state == INITIALIZED, "race condition");
879 return true;
880 }
881
882 /////////////////////////////////////////////////////////////////////////////
883 // attach existing thread
884
885 // bootstrap the main thread
886 bool os::create_main_thread(JavaThread* thread) {
887 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
888 return create_attached_thread(thread);
889 }
890
891 bool os::create_attached_thread(JavaThread* thread) {
892 #ifdef ASSERT
893 thread->verify_not_published();
894 #endif
895
896 // Allocate the OSThread object
897 OSThread* osthread = new OSThread(NULL, NULL);
898
899 if (osthread == NULL) {
900 return false;
901 }
902
903 // Store pthread info into the OSThread
904 osthread->set_thread_id(os::Linux::gettid());
905 osthread->set_pthread_id(::pthread_self());
906
907 // initialize floating point control register
908 os::Linux::init_thread_fpu_state();
909
910 // Initial thread state is RUNNABLE
911 osthread->set_state(RUNNABLE);
912
913 thread->set_osthread(osthread);
914
915 if (UseNUMA) {
916 int lgrp_id = os::numa_get_group_id();
917 if (lgrp_id != -1) {
918 thread->set_lgrp_id(lgrp_id);
919 }
920 }
921
922 if (os::Linux::is_initial_thread()) {
923 // If current thread is initial thread, its stack is mapped on demand,
924 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
925 // the entire stack region to avoid SEGV in stack banging.
926 // It is also useful to get around the heap-stack-gap problem on SuSE
927 // kernel (see 4821821 for details). We first expand stack to the top
928 // of yellow zone, then enable stack yellow zone (order is significant,
929 // enabling yellow zone first will crash JVM on SuSE Linux), so there
930 // is no gap between the last two virtual memory regions.
931
932 JavaThread *jt = (JavaThread *)thread;
933 address addr = jt->stack_yellow_zone_base();
934 assert(addr != NULL, "initialization problem?");
935 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
936
937 osthread->set_expanding_stack();
938 os::Linux::manually_expand_stack(jt, addr);
939 osthread->clear_expanding_stack();
940 }
941
942 // initialize signal mask for this thread
943 // and save the caller's signal mask
944 os::Linux::hotspot_sigmask(thread);
945
946 return true;
947 }
948
949 void os::pd_start_thread(Thread* thread) {
950 OSThread * osthread = thread->osthread();
951 assert(osthread->get_state() != INITIALIZED, "just checking");
952 Monitor* sync_with_child = osthread->startThread_lock();
953 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
954 sync_with_child->notify();
955 }
956
957 // Free Linux resources related to the OSThread
958 void os::free_thread(OSThread* osthread) {
959 assert(osthread != NULL, "osthread not set");
960
961 if (Thread::current()->osthread() == osthread) {
962 // Restore caller's signal mask
963 sigset_t sigmask = osthread->caller_sigmask();
964 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
965 }
966
967 delete osthread;
968 }
969
970 //////////////////////////////////////////////////////////////////////////////
971 // thread local storage
972
973 // Restore the thread pointer if the destructor is called. This is in case
974 // someone from JNI code sets up a destructor with pthread_key_create to run
975 // detachCurrentThread on thread death. Unless we restore the thread pointer we
976 // will hang or crash. When detachCurrentThread is called the key will be set
977 // to null and we will not be called again. If detachCurrentThread is never
978 // called we could loop forever depending on the pthread implementation.
979 static void restore_thread_pointer(void* p) {
980 Thread* thread = (Thread*) p;
981 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
982 }
983
984 int os::allocate_thread_local_storage() {
985 pthread_key_t key;
986 int rslt = pthread_key_create(&key, restore_thread_pointer);
987 assert(rslt == 0, "cannot allocate thread local storage");
988 return (int)key;
989 }
990
991 // Note: This is currently not used by VM, as we don't destroy TLS key
992 // on VM exit.
993 void os::free_thread_local_storage(int index) {
994 int rslt = pthread_key_delete((pthread_key_t)index);
995 assert(rslt == 0, "invalid index");
996 }
997
998 void os::thread_local_storage_at_put(int index, void* value) {
999 int rslt = pthread_setspecific((pthread_key_t)index, value);
1000 assert(rslt == 0, "pthread_setspecific failed");
1001 }
1002
1003 extern "C" Thread* get_thread() {
1004 return ThreadLocalStorage::thread();
1005 }
1006
1007 //////////////////////////////////////////////////////////////////////////////
1008 // initial thread
1009
1010 // Check if current thread is the initial thread, similar to Solaris thr_main.
1011 bool os::Linux::is_initial_thread(void) {
1012 char dummy;
1013 // If called before init complete, thread stack bottom will be null.
1014 // Can be called if fatal error occurs before initialization.
1015 if (initial_thread_stack_bottom() == NULL) return false;
1016 assert(initial_thread_stack_bottom() != NULL &&
1017 initial_thread_stack_size() != 0,
1018 "os::init did not locate initial thread's stack region");
1019 if ((address)&dummy >= initial_thread_stack_bottom() &&
1020 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1021 return true;
1022 } else {
1023 return false;
1024 }
1025 }
1026
1027 // Find the virtual memory area that contains addr
1028 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1029 FILE *fp = fopen("/proc/self/maps", "r");
1030 if (fp) {
1031 address low, high;
1032 while (!feof(fp)) {
1033 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1034 if (low <= addr && addr < high) {
1035 if (vma_low) *vma_low = low;
1036 if (vma_high) *vma_high = high;
1037 fclose(fp);
1038 return true;
1039 }
1040 }
1041 for (;;) {
1042 int ch = fgetc(fp);
1043 if (ch == EOF || ch == (int)'\n') break;
1044 }
1045 }
1046 fclose(fp);
1047 }
1048 return false;
1049 }
1050
1051 // Locate initial thread stack. This special handling of initial thread stack
1052 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1053 // bogus value for initial thread.
1054 void os::Linux::capture_initial_stack(size_t max_size) {
1055 // stack size is the easy part, get it from RLIMIT_STACK
1056 size_t stack_size;
1057 struct rlimit rlim;
1058 getrlimit(RLIMIT_STACK, &rlim);
1059 stack_size = rlim.rlim_cur;
1060
1061 // 6308388: a bug in ld.so will relocate its own .data section to the
1062 // lower end of primordial stack; reduce ulimit -s value a little bit
1063 // so we won't install guard page on ld.so's data section.
1064 stack_size -= 2 * page_size();
1065
1066 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1067 // 7.1, in both cases we will get 2G in return value.
1068 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1069 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1070 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1071 // in case other parts in glibc still assumes 2M max stack size.
1072 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1073 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1074 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1075 stack_size = 2 * K * K IA64_ONLY(*2);
1076 }
1077 // Try to figure out where the stack base (top) is. This is harder.
1078 //
1079 // When an application is started, glibc saves the initial stack pointer in
1080 // a global variable "__libc_stack_end", which is then used by system
1081 // libraries. __libc_stack_end should be pretty close to stack top. The
1082 // variable is available since the very early days. However, because it is
1083 // a private interface, it could disappear in the future.
1084 //
1085 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1086 // to __libc_stack_end, it is very close to stack top, but isn't the real
1087 // stack top. Note that /proc may not exist if VM is running as a chroot
1088 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1089 // /proc/<pid>/stat could change in the future (though unlikely).
1090 //
1091 // We try __libc_stack_end first. If that doesn't work, look for
1092 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1093 // as a hint, which should work well in most cases.
1094
1095 uintptr_t stack_start;
1096
1097 // try __libc_stack_end first
1098 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1099 if (p && *p) {
1100 stack_start = *p;
1101 } else {
1102 // see if we can get the start_stack field from /proc/self/stat
1103 FILE *fp;
1104 int pid;
1105 char state;
1106 int ppid;
1107 int pgrp;
1108 int session;
1109 int nr;
1110 int tpgrp;
1111 unsigned long flags;
1112 unsigned long minflt;
1113 unsigned long cminflt;
1114 unsigned long majflt;
1115 unsigned long cmajflt;
1116 unsigned long utime;
1117 unsigned long stime;
1118 long cutime;
1119 long cstime;
1120 long prio;
1121 long nice;
1122 long junk;
1123 long it_real;
1124 uintptr_t start;
1125 uintptr_t vsize;
1126 intptr_t rss;
1127 uintptr_t rsslim;
1128 uintptr_t scodes;
1129 uintptr_t ecode;
1130 int i;
1131
1132 // Figure what the primordial thread stack base is. Code is inspired
1133 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1134 // followed by command name surrounded by parentheses, state, etc.
1135 char stat[2048];
1136 int statlen;
1137
1138 fp = fopen("/proc/self/stat", "r");
1139 if (fp) {
1140 statlen = fread(stat, 1, 2047, fp);
1141 stat[statlen] = '\0';
1142 fclose(fp);
1143
1144 // Skip pid and the command string. Note that we could be dealing with
1145 // weird command names, e.g. user could decide to rename java launcher
1146 // to "java 1.4.2 :)", then the stat file would look like
1147 // 1234 (java 1.4.2 :)) R ... ...
1148 // We don't really need to know the command string, just find the last
1149 // occurrence of ")" and then start parsing from there. See bug 4726580.
1150 char * s = strrchr(stat, ')');
1151
1152 i = 0;
1153 if (s) {
1154 // Skip blank chars
1155 do { s++; } while (s && isspace(*s));
1156
1157 #define _UFM UINTX_FORMAT
1158 #define _DFM INTX_FORMAT
1159
1160 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1161 // 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
1162 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,
1163 &state, // 3 %c
1164 &ppid, // 4 %d
1165 &pgrp, // 5 %d
1166 &session, // 6 %d
1167 &nr, // 7 %d
1168 &tpgrp, // 8 %d
1169 &flags, // 9 %lu
1170 &minflt, // 10 %lu
1171 &cminflt, // 11 %lu
1172 &majflt, // 12 %lu
1173 &cmajflt, // 13 %lu
1174 &utime, // 14 %lu
1175 &stime, // 15 %lu
1176 &cutime, // 16 %ld
1177 &cstime, // 17 %ld
1178 &prio, // 18 %ld
1179 &nice, // 19 %ld
1180 &junk, // 20 %ld
1181 &it_real, // 21 %ld
1182 &start, // 22 UINTX_FORMAT
1183 &vsize, // 23 UINTX_FORMAT
1184 &rss, // 24 INTX_FORMAT
1185 &rsslim, // 25 UINTX_FORMAT
1186 &scodes, // 26 UINTX_FORMAT
1187 &ecode, // 27 UINTX_FORMAT
1188 &stack_start); // 28 UINTX_FORMAT
1189 }
1190
1191 #undef _UFM
1192 #undef _DFM
1193
1194 if (i != 28 - 2) {
1195 assert(false, "Bad conversion from /proc/self/stat");
1196 // product mode - assume we are the initial thread, good luck in the
1197 // embedded case.
1198 warning("Can't detect initial thread stack location - bad conversion");
1199 stack_start = (uintptr_t) &rlim;
1200 }
1201 } else {
1202 // For some reason we can't open /proc/self/stat (for example, running on
1203 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1204 // most cases, so don't abort:
1205 warning("Can't detect initial thread stack location - no /proc/self/stat");
1206 stack_start = (uintptr_t) &rlim;
1207 }
1208 }
1209
1210 // Now we have a pointer (stack_start) very close to the stack top, the
1211 // next thing to do is to figure out the exact location of stack top. We
1212 // can find out the virtual memory area that contains stack_start by
1213 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1214 // and its upper limit is the real stack top. (again, this would fail if
1215 // running inside chroot, because /proc may not exist.)
1216
1217 uintptr_t stack_top;
1218 address low, high;
1219 if (find_vma((address)stack_start, &low, &high)) {
1220 // success, "high" is the true stack top. (ignore "low", because initial
1221 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1222 stack_top = (uintptr_t)high;
1223 } else {
1224 // failed, likely because /proc/self/maps does not exist
1225 warning("Can't detect initial thread stack location - find_vma failed");
1226 // best effort: stack_start is normally within a few pages below the real
1227 // stack top, use it as stack top, and reduce stack size so we won't put
1228 // guard page outside stack.
1229 stack_top = stack_start;
1230 stack_size -= 16 * page_size();
1231 }
1232
1233 // stack_top could be partially down the page so align it
1234 stack_top = align_size_up(stack_top, page_size());
1235
1236 if (max_size && stack_size > max_size) {
1237 _initial_thread_stack_size = max_size;
1238 } else {
1239 _initial_thread_stack_size = stack_size;
1240 }
1241
1242 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1243 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1244 }
1245
1246 ////////////////////////////////////////////////////////////////////////////////
1247 // time support
1248
1249 // Time since start-up in seconds to a fine granularity.
1250 // Used by VMSelfDestructTimer and the MemProfiler.
1251 double os::elapsedTime() {
1252
1253 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1254 }
1255
1256 jlong os::elapsed_counter() {
1257 return javaTimeNanos() - initial_time_count;
1258 }
1259
1260 jlong os::elapsed_frequency() {
1261 return NANOSECS_PER_SEC; // nanosecond resolution
1262 }
1263
1264 bool os::supports_vtime() { return true; }
1265 bool os::enable_vtime() { return false; }
1266 bool os::vtime_enabled() { return false; }
1267
1268 double os::elapsedVTime() {
1269 struct rusage usage;
1270 int retval = getrusage(RUSAGE_THREAD, &usage);
1271 if (retval == 0) {
1272 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);
1273 } else {
1274 // better than nothing, but not much
1275 return elapsedTime();
1276 }
1277 }
1278
1279 jlong os::javaTimeMillis() {
1280 timeval time;
1281 int status = gettimeofday(&time, NULL);
1282 assert(status != -1, "linux error");
1283 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1284 }
1285
1286 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1287 timeval time;
1288 int status = gettimeofday(&time, NULL);
1289 assert(status != -1, "linux error");
1290 seconds = jlong(time.tv_sec);
1291 nanos = jlong(time.tv_usec) * 1000;
1292 }
1293
1294
1295 #ifndef CLOCK_MONOTONIC
1296 #define CLOCK_MONOTONIC (1)
1297 #endif
1298
1299 void os::Linux::clock_init() {
1300 // we do dlopen's in this particular order due to bug in linux
1301 // dynamical loader (see 6348968) leading to crash on exit
1302 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1303 if (handle == NULL) {
1304 handle = dlopen("librt.so", RTLD_LAZY);
1305 }
1306
1307 if (handle) {
1308 int (*clock_getres_func)(clockid_t, struct timespec*) =
1309 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1310 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1311 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1312 if (clock_getres_func && clock_gettime_func) {
1313 // See if monotonic clock is supported by the kernel. Note that some
1314 // early implementations simply return kernel jiffies (updated every
1315 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1316 // for nano time (though the monotonic property is still nice to have).
1317 // It's fixed in newer kernels, however clock_getres() still returns
1318 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1319 // resolution for now. Hopefully as people move to new kernels, this
1320 // won't be a problem.
1321 struct timespec res;
1322 struct timespec tp;
1323 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1324 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1325 // yes, monotonic clock is supported
1326 _clock_gettime = clock_gettime_func;
1327 return;
1328 } else {
1329 // close librt if there is no monotonic clock
1330 dlclose(handle);
1331 }
1332 }
1333 }
1334 warning("No monotonic clock was available - timed services may " \
1335 "be adversely affected if the time-of-day clock changes");
1336 }
1337
1338 #ifndef SYS_clock_getres
1339 #if defined(IA32) || defined(AMD64)
1340 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1341 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1342 #else
1343 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1344 #define sys_clock_getres(x,y) -1
1345 #endif
1346 #else
1347 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1348 #endif
1349
1350 void os::Linux::fast_thread_clock_init() {
1351 if (!UseLinuxPosixThreadCPUClocks) {
1352 return;
1353 }
1354 clockid_t clockid;
1355 struct timespec tp;
1356 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1357 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1358
1359 // Switch to using fast clocks for thread cpu time if
1360 // the sys_clock_getres() returns 0 error code.
1361 // Note, that some kernels may support the current thread
1362 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1363 // returned by the pthread_getcpuclockid().
1364 // If the fast Posix clocks are supported then the sys_clock_getres()
1365 // must return at least tp.tv_sec == 0 which means a resolution
1366 // better than 1 sec. This is extra check for reliability.
1367
1368 if (pthread_getcpuclockid_func &&
1369 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1370 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1371 _supports_fast_thread_cpu_time = true;
1372 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1373 }
1374 }
1375
1376 jlong os::javaTimeNanos() {
1377 if (os::supports_monotonic_clock()) {
1378 struct timespec tp;
1379 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1380 assert(status == 0, "gettime error");
1381 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1382 return result;
1383 } else {
1384 timeval time;
1385 int status = gettimeofday(&time, NULL);
1386 assert(status != -1, "linux error");
1387 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1388 return 1000 * usecs;
1389 }
1390 }
1391
1392 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1393 if (os::supports_monotonic_clock()) {
1394 info_ptr->max_value = ALL_64_BITS;
1395
1396 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1397 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1398 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1399 } else {
1400 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1401 info_ptr->max_value = ALL_64_BITS;
1402
1403 // gettimeofday is a real time clock so it skips
1404 info_ptr->may_skip_backward = true;
1405 info_ptr->may_skip_forward = true;
1406 }
1407
1408 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1409 }
1410
1411 // Return the real, user, and system times in seconds from an
1412 // arbitrary fixed point in the past.
1413 bool os::getTimesSecs(double* process_real_time,
1414 double* process_user_time,
1415 double* process_system_time) {
1416 struct tms ticks;
1417 clock_t real_ticks = times(&ticks);
1418
1419 if (real_ticks == (clock_t) (-1)) {
1420 return false;
1421 } else {
1422 double ticks_per_second = (double) clock_tics_per_sec;
1423 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1424 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1425 *process_real_time = ((double) real_ticks) / ticks_per_second;
1426
1427 return true;
1428 }
1429 }
1430
1431
1432 char * os::local_time_string(char *buf, size_t buflen) {
1433 struct tm t;
1434 time_t long_time;
1435 time(&long_time);
1436 localtime_r(&long_time, &t);
1437 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1438 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1439 t.tm_hour, t.tm_min, t.tm_sec);
1440 return buf;
1441 }
1442
1443 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1444 return localtime_r(clock, res);
1445 }
1446
1447 ////////////////////////////////////////////////////////////////////////////////
1448 // runtime exit support
1449
1450 // Note: os::shutdown() might be called very early during initialization, or
1451 // called from signal handler. Before adding something to os::shutdown(), make
1452 // sure it is async-safe and can handle partially initialized VM.
1453 void os::shutdown() {
1454
1455 // allow PerfMemory to attempt cleanup of any persistent resources
1456 perfMemory_exit();
1457
1458 // needs to remove object in file system
1459 AttachListener::abort();
1460
1461 // flush buffered output, finish log files
1462 ostream_abort();
1463
1464 // Check for abort hook
1465 abort_hook_t abort_hook = Arguments::abort_hook();
1466 if (abort_hook != NULL) {
1467 abort_hook();
1468 }
1469
1470 }
1471
1472 // Note: os::abort() might be called very early during initialization, or
1473 // called from signal handler. Before adding something to os::abort(), make
1474 // sure it is async-safe and can handle partially initialized VM.
1475 void os::abort(bool dump_core) {
1476 os::shutdown();
1477 if (dump_core) {
1478 #ifndef PRODUCT
1479 fdStream out(defaultStream::output_fd());
1480 out.print_raw("Current thread is ");
1481 char buf[16];
1482 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1483 out.print_raw_cr(buf);
1484 out.print_raw_cr("Dumping core ...");
1485 #endif
1486 ::abort(); // dump core
1487 }
1488
1489 ::exit(1);
1490 }
1491
1492 // Die immediately, no exit hook, no abort hook, no cleanup.
1493 void os::die() {
1494 // _exit() on LinuxThreads only kills current thread
1495 ::abort();
1496 }
1497
1498
1499 // This method is a copy of JDK's sysGetLastErrorString
1500 // from src/solaris/hpi/src/system_md.c
1501
1502 size_t os::lasterror(char *buf, size_t len) {
1503 if (errno == 0) return 0;
1504
1505 const char *s = ::strerror(errno);
1506 size_t n = ::strlen(s);
1507 if (n >= len) {
1508 n = len - 1;
1509 }
1510 ::strncpy(buf, s, n);
1511 buf[n] = '\0';
1512 return n;
1513 }
1514
1515 intx os::current_thread_id() { return (intx)pthread_self(); }
1516 int os::current_process_id() {
1517
1518 // Under the old linux thread library, linux gives each thread
1519 // its own process id. Because of this each thread will return
1520 // a different pid if this method were to return the result
1521 // of getpid(2). Linux provides no api that returns the pid
1522 // of the launcher thread for the vm. This implementation
1523 // returns a unique pid, the pid of the launcher thread
1524 // that starts the vm 'process'.
1525
1526 // Under the NPTL, getpid() returns the same pid as the
1527 // launcher thread rather than a unique pid per thread.
1528 // Use gettid() if you want the old pre NPTL behaviour.
1529
1530 // if you are looking for the result of a call to getpid() that
1531 // returns a unique pid for the calling thread, then look at the
1532 // OSThread::thread_id() method in osThread_linux.hpp file
1533
1534 return (int)(_initial_pid ? _initial_pid : getpid());
1535 }
1536
1537 // DLL functions
1538
1539 const char* os::dll_file_extension() { return ".so"; }
1540
1541 // This must be hard coded because it's the system's temporary
1542 // directory not the java application's temp directory, ala java.io.tmpdir.
1543 const char* os::get_temp_directory() { return "/tmp"; }
1544
1545 static bool file_exists(const char* filename) {
1546 struct stat statbuf;
1547 if (filename == NULL || strlen(filename) == 0) {
1548 return false;
1549 }
1550 return os::stat(filename, &statbuf) == 0;
1551 }
1552
1553 bool os::dll_build_name(char* buffer, size_t buflen,
1554 const char* pname, const char* fname) {
1555 bool retval = false;
1556 // Copied from libhpi
1557 const size_t pnamelen = pname ? strlen(pname) : 0;
1558
1559 // Return error on buffer overflow.
1560 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1561 return retval;
1562 }
1563
1564 if (pnamelen == 0) {
1565 snprintf(buffer, buflen, "lib%s.so", fname);
1566 retval = true;
1567 } else if (strchr(pname, *os::path_separator()) != NULL) {
1568 int n;
1569 char** pelements = split_path(pname, &n);
1570 if (pelements == NULL) {
1571 return false;
1572 }
1573 for (int i = 0; i < n; i++) {
1574 // Really shouldn't be NULL, but check can't hurt
1575 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1576 continue; // skip the empty path values
1577 }
1578 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1579 if (file_exists(buffer)) {
1580 retval = true;
1581 break;
1582 }
1583 }
1584 // release the storage
1585 for (int i = 0; i < n; i++) {
1586 if (pelements[i] != NULL) {
1587 FREE_C_HEAP_ARRAY(char, pelements[i]);
1588 }
1589 }
1590 if (pelements != NULL) {
1591 FREE_C_HEAP_ARRAY(char*, pelements);
1592 }
1593 } else {
1594 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1595 retval = true;
1596 }
1597 return retval;
1598 }
1599
1600 // check if addr is inside libjvm.so
1601 bool os::address_is_in_vm(address addr) {
1602 static address libjvm_base_addr;
1603 Dl_info dlinfo;
1604
1605 if (libjvm_base_addr == NULL) {
1606 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1607 libjvm_base_addr = (address)dlinfo.dli_fbase;
1608 }
1609 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1610 }
1611
1612 if (dladdr((void *)addr, &dlinfo) != 0) {
1613 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1614 }
1615
1616 return false;
1617 }
1618
1619 bool os::dll_address_to_function_name(address addr, char *buf,
1620 int buflen, int *offset) {
1621 // buf is not optional, but offset is optional
1622 assert(buf != NULL, "sanity check");
1623
1624 Dl_info dlinfo;
1625
1626 if (dladdr((void*)addr, &dlinfo) != 0) {
1627 // see if we have a matching symbol
1628 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1629 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1630 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1631 }
1632 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1633 return true;
1634 }
1635 // no matching symbol so try for just file info
1636 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1637 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1638 buf, buflen, offset, dlinfo.dli_fname)) {
1639 return true;
1640 }
1641 }
1642 }
1643
1644 buf[0] = '\0';
1645 if (offset != NULL) *offset = -1;
1646 return false;
1647 }
1648
1649 struct _address_to_library_name {
1650 address addr; // input : memory address
1651 size_t buflen; // size of fname
1652 char* fname; // output: library name
1653 address base; // library base addr
1654 };
1655
1656 static int address_to_library_name_callback(struct dl_phdr_info *info,
1657 size_t size, void *data) {
1658 int i;
1659 bool found = false;
1660 address libbase = NULL;
1661 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1662
1663 // iterate through all loadable segments
1664 for (i = 0; i < info->dlpi_phnum; i++) {
1665 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1666 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1667 // base address of a library is the lowest address of its loaded
1668 // segments.
1669 if (libbase == NULL || libbase > segbase) {
1670 libbase = segbase;
1671 }
1672 // see if 'addr' is within current segment
1673 if (segbase <= d->addr &&
1674 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1675 found = true;
1676 }
1677 }
1678 }
1679
1680 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1681 // so dll_address_to_library_name() can fall through to use dladdr() which
1682 // can figure out executable name from argv[0].
1683 if (found && info->dlpi_name && info->dlpi_name[0]) {
1684 d->base = libbase;
1685 if (d->fname) {
1686 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1687 }
1688 return 1;
1689 }
1690 return 0;
1691 }
1692
1693 bool os::dll_address_to_library_name(address addr, char* buf,
1694 int buflen, int* offset) {
1695 // buf is not optional, but offset is optional
1696 assert(buf != NULL, "sanity check");
1697
1698 Dl_info dlinfo;
1699 struct _address_to_library_name data;
1700
1701 // There is a bug in old glibc dladdr() implementation that it could resolve
1702 // to wrong library name if the .so file has a base address != NULL. Here
1703 // we iterate through the program headers of all loaded libraries to find
1704 // out which library 'addr' really belongs to. This workaround can be
1705 // removed once the minimum requirement for glibc is moved to 2.3.x.
1706 data.addr = addr;
1707 data.fname = buf;
1708 data.buflen = buflen;
1709 data.base = NULL;
1710 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1711
1712 if (rslt) {
1713 // buf already contains library name
1714 if (offset) *offset = addr - data.base;
1715 return true;
1716 }
1717 if (dladdr((void*)addr, &dlinfo) != 0) {
1718 if (dlinfo.dli_fname != NULL) {
1719 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1720 }
1721 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1722 *offset = addr - (address)dlinfo.dli_fbase;
1723 }
1724 return true;
1725 }
1726
1727 buf[0] = '\0';
1728 if (offset) *offset = -1;
1729 return false;
1730 }
1731
1732 // Loads .dll/.so and
1733 // in case of error it checks if .dll/.so was built for the
1734 // same architecture as Hotspot is running on
1735
1736
1737 // Remember the stack's state. The Linux dynamic linker will change
1738 // the stack to 'executable' at most once, so we must safepoint only once.
1739 bool os::Linux::_stack_is_executable = false;
1740
1741 // VM operation that loads a library. This is necessary if stack protection
1742 // of the Java stacks can be lost during loading the library. If we
1743 // do not stop the Java threads, they can stack overflow before the stacks
1744 // are protected again.
1745 class VM_LinuxDllLoad: public VM_Operation {
1746 private:
1747 const char *_filename;
1748 char *_ebuf;
1749 int _ebuflen;
1750 void *_lib;
1751 public:
1752 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1753 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1754 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1755 void doit() {
1756 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1757 os::Linux::_stack_is_executable = true;
1758 }
1759 void* loaded_library() { return _lib; }
1760 };
1761
1762 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1763 void * result = NULL;
1764 bool load_attempted = false;
1765
1766 // Check whether the library to load might change execution rights
1767 // of the stack. If they are changed, the protection of the stack
1768 // guard pages will be lost. We need a safepoint to fix this.
1769 //
1770 // See Linux man page execstack(8) for more info.
1771 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1772 ElfFile ef(filename);
1773 if (!ef.specifies_noexecstack()) {
1774 if (!is_init_completed()) {
1775 os::Linux::_stack_is_executable = true;
1776 // This is OK - No Java threads have been created yet, and hence no
1777 // stack guard pages to fix.
1778 //
1779 // This should happen only when you are building JDK7 using a very
1780 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1781 //
1782 // Dynamic loader will make all stacks executable after
1783 // this function returns, and will not do that again.
1784 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1785 } else {
1786 warning("You have loaded library %s which might have disabled stack guard. "
1787 "The VM will try to fix the stack guard now.\n"
1788 "It's highly recommended that you fix the library with "
1789 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1790 filename);
1791
1792 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1793 JavaThread *jt = JavaThread::current();
1794 if (jt->thread_state() != _thread_in_native) {
1795 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1796 // that requires ExecStack. Cannot enter safe point. Let's give up.
1797 warning("Unable to fix stack guard. Giving up.");
1798 } else {
1799 if (!LoadExecStackDllInVMThread) {
1800 // This is for the case where the DLL has an static
1801 // constructor function that executes JNI code. We cannot
1802 // load such DLLs in the VMThread.
1803 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1804 }
1805
1806 ThreadInVMfromNative tiv(jt);
1807 debug_only(VMNativeEntryWrapper vew;)
1808
1809 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1810 VMThread::execute(&op);
1811 if (LoadExecStackDllInVMThread) {
1812 result = op.loaded_library();
1813 }
1814 load_attempted = true;
1815 }
1816 }
1817 }
1818 }
1819
1820 if (!load_attempted) {
1821 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1822 }
1823
1824 if (result != NULL) {
1825 // Successful loading
1826 return result;
1827 }
1828
1829 Elf32_Ehdr elf_head;
1830 int diag_msg_max_length=ebuflen-strlen(ebuf);
1831 char* diag_msg_buf=ebuf+strlen(ebuf);
1832
1833 if (diag_msg_max_length==0) {
1834 // No more space in ebuf for additional diagnostics message
1835 return NULL;
1836 }
1837
1838
1839 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1840
1841 if (file_descriptor < 0) {
1842 // Can't open library, report dlerror() message
1843 return NULL;
1844 }
1845
1846 bool failed_to_read_elf_head=
1847 (sizeof(elf_head)!=
1848 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1849
1850 ::close(file_descriptor);
1851 if (failed_to_read_elf_head) {
1852 // file i/o error - report dlerror() msg
1853 return NULL;
1854 }
1855
1856 typedef struct {
1857 Elf32_Half code; // Actual value as defined in elf.h
1858 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1859 char elf_class; // 32 or 64 bit
1860 char endianess; // MSB or LSB
1861 char* name; // String representation
1862 } arch_t;
1863
1864 #ifndef EM_486
1865 #define EM_486 6 /* Intel 80486 */
1866 #endif
1867 #ifndef EM_AARCH64
1868 #define EM_AARCH64 183 /* ARM AARCH64 */
1869 #endif
1870
1871 static const arch_t arch_array[]={
1872 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1873 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1874 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1875 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1876 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1877 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1878 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1879 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1880 #if defined(VM_LITTLE_ENDIAN)
1881 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1882 #else
1883 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1884 #endif
1885 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1886 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1887 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1888 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1889 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1890 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1891 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1892 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1893 };
1894
1895 #if (defined IA32)
1896 static Elf32_Half running_arch_code=EM_386;
1897 #elif (defined AMD64)
1898 static Elf32_Half running_arch_code=EM_X86_64;
1899 #elif (defined IA64)
1900 static Elf32_Half running_arch_code=EM_IA_64;
1901 #elif (defined __sparc) && (defined _LP64)
1902 static Elf32_Half running_arch_code=EM_SPARCV9;
1903 #elif (defined __sparc) && (!defined _LP64)
1904 static Elf32_Half running_arch_code=EM_SPARC;
1905 #elif (defined __powerpc64__)
1906 static Elf32_Half running_arch_code=EM_PPC64;
1907 #elif (defined __powerpc__)
1908 static Elf32_Half running_arch_code=EM_PPC;
1909 #elif (defined ARM)
1910 static Elf32_Half running_arch_code=EM_ARM;
1911 #elif (defined S390)
1912 static Elf32_Half running_arch_code=EM_S390;
1913 #elif (defined ALPHA)
1914 static Elf32_Half running_arch_code=EM_ALPHA;
1915 #elif (defined MIPSEL)
1916 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1917 #elif (defined PARISC)
1918 static Elf32_Half running_arch_code=EM_PARISC;
1919 #elif (defined MIPS)
1920 static Elf32_Half running_arch_code=EM_MIPS;
1921 #elif (defined M68K)
1922 static Elf32_Half running_arch_code=EM_68K;
1923 #elif (defined AARCH64)
1924 static Elf32_Half running_arch_code=EM_AARCH64;
1925 #else
1926 #error Method os::dll_load requires that one of following is defined:\
1927 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1928 #endif
1929
1930 // Identify compatability class for VM's architecture and library's architecture
1931 // Obtain string descriptions for architectures
1932
1933 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1934 int running_arch_index=-1;
1935
1936 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1937 if (running_arch_code == arch_array[i].code) {
1938 running_arch_index = i;
1939 }
1940 if (lib_arch.code == arch_array[i].code) {
1941 lib_arch.compat_class = arch_array[i].compat_class;
1942 lib_arch.name = arch_array[i].name;
1943 }
1944 }
1945
1946 assert(running_arch_index != -1,
1947 "Didn't find running architecture code (running_arch_code) in arch_array");
1948 if (running_arch_index == -1) {
1949 // Even though running architecture detection failed
1950 // we may still continue with reporting dlerror() message
1951 return NULL;
1952 }
1953
1954 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1955 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1956 return NULL;
1957 }
1958
1959 #ifndef S390
1960 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1961 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1962 return NULL;
1963 }
1964 #endif // !S390
1965
1966 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1967 if (lib_arch.name!=NULL) {
1968 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1969 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1970 lib_arch.name, arch_array[running_arch_index].name);
1971 } else {
1972 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1973 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1974 lib_arch.code,
1975 arch_array[running_arch_index].name);
1976 }
1977 }
1978
1979 return NULL;
1980 }
1981
1982 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1983 int ebuflen) {
1984 void * result = ::dlopen(filename, RTLD_LAZY);
1985 if (result == NULL) {
1986 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1987 ebuf[ebuflen-1] = '\0';
1988 }
1989 return result;
1990 }
1991
1992 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1993 int ebuflen) {
1994 void * result = NULL;
1995 if (LoadExecStackDllInVMThread) {
1996 result = dlopen_helper(filename, ebuf, ebuflen);
1997 }
1998
1999 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2000 // library that requires an executable stack, or which does not have this
2001 // stack attribute set, dlopen changes the stack attribute to executable. The
2002 // read protection of the guard pages gets lost.
2003 //
2004 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2005 // may have been queued at the same time.
2006
2007 if (!_stack_is_executable) {
2008 JavaThread *jt = Threads::first();
2009
2010 while (jt) {
2011 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2012 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2013 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2014 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2015 warning("Attempt to reguard stack yellow zone failed.");
2016 }
2017 }
2018 jt = jt->next();
2019 }
2020 }
2021
2022 return result;
2023 }
2024
2025 void* os::dll_lookup(void* handle, const char* name) {
2026 void* res = dlsym(handle, name);
2027 return res;
2028 }
2029
2030 void* os::get_default_process_handle() {
2031 return (void*)::dlopen(NULL, RTLD_LAZY);
2032 }
2033
2034 static bool _print_ascii_file(const char* filename, outputStream* st) {
2035 int fd = ::open(filename, O_RDONLY);
2036 if (fd == -1) {
2037 return false;
2038 }
2039
2040 char buf[32];
2041 int bytes;
2042 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2043 st->print_raw(buf, bytes);
2044 }
2045
2046 ::close(fd);
2047
2048 return true;
2049 }
2050
2051 void os::print_dll_info(outputStream *st) {
2052 st->print_cr("Dynamic libraries:");
2053
2054 char fname[32];
2055 pid_t pid = os::Linux::gettid();
2056
2057 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2058
2059 if (!_print_ascii_file(fname, st)) {
2060 st->print("Can not get library information for pid = %d\n", pid);
2061 }
2062 }
2063
2064 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2065 FILE *procmapsFile = NULL;
2066
2067 // Open the procfs maps file for the current process
2068 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2069 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2070 char line[PATH_MAX + 100];
2071
2072 // Read line by line from 'file'
2073 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2074 u8 base, top, offset, inode;
2075 char permissions[5];
2076 char device[6];
2077 char name[PATH_MAX + 1];
2078
2079 // Parse fields from line
2080 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2081
2082 // Filter by device id '00:00' so that we only get file system mapped files.
2083 if (strcmp(device, "00:00") != 0) {
2084
2085 // Call callback with the fields of interest
2086 if(callback(name, (address)base, (address)top, param)) {
2087 // Oops abort, callback aborted
2088 fclose(procmapsFile);
2089 return 1;
2090 }
2091 }
2092 }
2093 fclose(procmapsFile);
2094 }
2095 return 0;
2096 }
2097
2098 void os::print_os_info_brief(outputStream* st) {
2099 os::Linux::print_distro_info(st);
2100
2101 os::Posix::print_uname_info(st);
2102
2103 os::Linux::print_libversion_info(st);
2104
2105 }
2106
2107 void os::print_os_info(outputStream* st) {
2108 st->print("OS:");
2109
2110 os::Linux::print_distro_info(st);
2111
2112 os::Posix::print_uname_info(st);
2113
2114 // Print warning if unsafe chroot environment detected
2115 if (unsafe_chroot_detected) {
2116 st->print("WARNING!! ");
2117 st->print_cr("%s", unstable_chroot_error);
2118 }
2119
2120 os::Linux::print_libversion_info(st);
2121
2122 os::Posix::print_rlimit_info(st);
2123
2124 os::Posix::print_load_average(st);
2125
2126 os::Linux::print_full_memory_info(st);
2127 }
2128
2129 // Try to identify popular distros.
2130 // Most Linux distributions have a /etc/XXX-release file, which contains
2131 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2132 // file that also contains the OS version string. Some have more than one
2133 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2134 // /etc/redhat-release.), so the order is important.
2135 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2136 // their own specific XXX-release file as well as a redhat-release file.
2137 // Because of this the XXX-release file needs to be searched for before the
2138 // redhat-release file.
2139 // Since Red Hat has a lsb-release file that is not very descriptive the
2140 // search for redhat-release needs to be before lsb-release.
2141 // Since the lsb-release file is the new standard it needs to be searched
2142 // before the older style release files.
2143 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2144 // next to last resort. The os-release file is a new standard that contains
2145 // distribution information and the system-release file seems to be an old
2146 // standard that has been replaced by the lsb-release and os-release files.
2147 // Searching for the debian_version file is the last resort. It contains
2148 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2149 // "Debian " is printed before the contents of the debian_version file.
2150 void os::Linux::print_distro_info(outputStream* st) {
2151 if (!_print_ascii_file("/etc/oracle-release", st) &&
2152 !_print_ascii_file("/etc/mandriva-release", st) &&
2153 !_print_ascii_file("/etc/mandrake-release", st) &&
2154 !_print_ascii_file("/etc/sun-release", st) &&
2155 !_print_ascii_file("/etc/redhat-release", st) &&
2156 !_print_ascii_file("/etc/lsb-release", st) &&
2157 !_print_ascii_file("/etc/SuSE-release", st) &&
2158 !_print_ascii_file("/etc/turbolinux-release", st) &&
2159 !_print_ascii_file("/etc/gentoo-release", st) &&
2160 !_print_ascii_file("/etc/ltib-release", st) &&
2161 !_print_ascii_file("/etc/angstrom-version", st) &&
2162 !_print_ascii_file("/etc/system-release", st) &&
2163 !_print_ascii_file("/etc/os-release", st)) {
2164
2165 if (file_exists("/etc/debian_version")) {
2166 st->print("Debian ");
2167 _print_ascii_file("/etc/debian_version", st);
2168 } else {
2169 st->print("Linux");
2170 }
2171 }
2172 st->cr();
2173 }
2174
2175 void os::Linux::print_libversion_info(outputStream* st) {
2176 // libc, pthread
2177 st->print("libc:");
2178 st->print("%s ", os::Linux::glibc_version());
2179 st->print("%s ", os::Linux::libpthread_version());
2180 if (os::Linux::is_LinuxThreads()) {
2181 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2182 }
2183 st->cr();
2184 }
2185
2186 void os::Linux::print_full_memory_info(outputStream* st) {
2187 st->print("\n/proc/meminfo:\n");
2188 _print_ascii_file("/proc/meminfo", st);
2189 st->cr();
2190 }
2191
2192 void os::print_memory_info(outputStream* st) {
2193
2194 st->print("Memory:");
2195 st->print(" %dk page", os::vm_page_size()>>10);
2196
2197 // values in struct sysinfo are "unsigned long"
2198 struct sysinfo si;
2199 sysinfo(&si);
2200
2201 st->print(", physical " UINT64_FORMAT "k",
2202 os::physical_memory() >> 10);
2203 st->print("(" UINT64_FORMAT "k free)",
2204 os::available_memory() >> 10);
2205 st->print(", swap " UINT64_FORMAT "k",
2206 ((jlong)si.totalswap * si.mem_unit) >> 10);
2207 st->print("(" UINT64_FORMAT "k free)",
2208 ((jlong)si.freeswap * si.mem_unit) >> 10);
2209 st->cr();
2210 }
2211
2212 void os::pd_print_cpu_info(outputStream* st) {
2213 st->print("\n/proc/cpuinfo:\n");
2214 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2215 st->print(" <Not Available>");
2216 }
2217 st->cr();
2218 }
2219
2220 void os::print_siginfo(outputStream* st, void* siginfo) {
2221 const siginfo_t* si = (const siginfo_t*)siginfo;
2222
2223 os::Posix::print_siginfo_brief(st, si);
2224 #if INCLUDE_CDS
2225 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2226 UseSharedSpaces) {
2227 FileMapInfo* mapinfo = FileMapInfo::current_info();
2228 if (mapinfo->is_in_shared_space(si->si_addr)) {
2229 st->print("\n\nError accessing class data sharing archive." \
2230 " Mapped file inaccessible during execution, " \
2231 " possible disk/network problem.");
2232 }
2233 }
2234 #endif
2235 st->cr();
2236 }
2237
2238
2239 static void print_signal_handler(outputStream* st, int sig,
2240 char* buf, size_t buflen);
2241
2242 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2243 st->print_cr("Signal Handlers:");
2244 print_signal_handler(st, SIGSEGV, buf, buflen);
2245 print_signal_handler(st, SIGBUS , buf, buflen);
2246 print_signal_handler(st, SIGFPE , buf, buflen);
2247 print_signal_handler(st, SIGPIPE, buf, buflen);
2248 print_signal_handler(st, SIGXFSZ, buf, buflen);
2249 print_signal_handler(st, SIGILL , buf, buflen);
2250 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2251 print_signal_handler(st, SR_signum, buf, buflen);
2252 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2253 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2254 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2255 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2256 #if defined(PPC64)
2257 print_signal_handler(st, SIGTRAP, buf, buflen);
2258 #endif
2259 }
2260
2261 static char saved_jvm_path[MAXPATHLEN] = {0};
2262
2263 // Find the full path to the current module, libjvm.so
2264 void os::jvm_path(char *buf, jint buflen) {
2265 // Error checking.
2266 if (buflen < MAXPATHLEN) {
2267 assert(false, "must use a large-enough buffer");
2268 buf[0] = '\0';
2269 return;
2270 }
2271 // Lazy resolve the path to current module.
2272 if (saved_jvm_path[0] != 0) {
2273 strcpy(buf, saved_jvm_path);
2274 return;
2275 }
2276
2277 char dli_fname[MAXPATHLEN];
2278 bool ret = dll_address_to_library_name(
2279 CAST_FROM_FN_PTR(address, os::jvm_path),
2280 dli_fname, sizeof(dli_fname), NULL);
2281 assert(ret, "cannot locate libjvm");
2282 char *rp = NULL;
2283 if (ret && dli_fname[0] != '\0') {
2284 rp = realpath(dli_fname, buf);
2285 }
2286 if (rp == NULL) {
2287 return;
2288 }
2289
2290 if (Arguments::sun_java_launcher_is_altjvm()) {
2291 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2292 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2293 // If "/jre/lib/" appears at the right place in the string, then
2294 // assume we are installed in a JDK and we're done. Otherwise, check
2295 // for a JAVA_HOME environment variable and fix up the path so it
2296 // looks like libjvm.so is installed there (append a fake suffix
2297 // hotspot/libjvm.so).
2298 const char *p = buf + strlen(buf) - 1;
2299 for (int count = 0; p > buf && count < 5; ++count) {
2300 for (--p; p > buf && *p != '/'; --p)
2301 /* empty */ ;
2302 }
2303
2304 if (strncmp(p, "/jre/lib/", 9) != 0) {
2305 // Look for JAVA_HOME in the environment.
2306 char* java_home_var = ::getenv("JAVA_HOME");
2307 if (java_home_var != NULL && java_home_var[0] != 0) {
2308 char* jrelib_p;
2309 int len;
2310
2311 // Check the current module name "libjvm.so".
2312 p = strrchr(buf, '/');
2313 if (p == NULL) {
2314 return;
2315 }
2316 assert(strstr(p, "/libjvm") == p, "invalid library name");
2317
2318 rp = realpath(java_home_var, buf);
2319 if (rp == NULL) {
2320 return;
2321 }
2322
2323 // determine if this is a legacy image or modules image
2324 // modules image doesn't have "jre" subdirectory
2325 len = strlen(buf);
2326 assert(len < buflen, "Ran out of buffer room");
2327 jrelib_p = buf + len;
2328 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2329 if (0 != access(buf, F_OK)) {
2330 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2331 }
2332
2333 if (0 == access(buf, F_OK)) {
2334 // Use current module name "libjvm.so"
2335 len = strlen(buf);
2336 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2337 } else {
2338 // Go back to path of .so
2339 rp = realpath(dli_fname, buf);
2340 if (rp == NULL) {
2341 return;
2342 }
2343 }
2344 }
2345 }
2346 }
2347
2348 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2349 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2350 }
2351
2352 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2353 // no prefix required, not even "_"
2354 }
2355
2356 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2357 // no suffix required
2358 }
2359
2360 ////////////////////////////////////////////////////////////////////////////////
2361 // sun.misc.Signal support
2362
2363 static volatile jint sigint_count = 0;
2364
2365 static void UserHandler(int sig, void *siginfo, void *context) {
2366 // 4511530 - sem_post is serialized and handled by the manager thread. When
2367 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2368 // don't want to flood the manager thread with sem_post requests.
2369 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2370 return;
2371 }
2372
2373 // Ctrl-C is pressed during error reporting, likely because the error
2374 // handler fails to abort. Let VM die immediately.
2375 if (sig == SIGINT && is_error_reported()) {
2376 os::die();
2377 }
2378
2379 os::signal_notify(sig);
2380 }
2381
2382 void* os::user_handler() {
2383 return CAST_FROM_FN_PTR(void*, UserHandler);
2384 }
2385
2386 class Semaphore : public StackObj {
2387 public:
2388 Semaphore();
2389 ~Semaphore();
2390 void signal();
2391 void wait();
2392 bool trywait();
2393 bool timedwait(unsigned int sec, int nsec);
2394 private:
2395 sem_t _semaphore;
2396 };
2397
2398 Semaphore::Semaphore() {
2399 sem_init(&_semaphore, 0, 0);
2400 }
2401
2402 Semaphore::~Semaphore() {
2403 sem_destroy(&_semaphore);
2404 }
2405
2406 void Semaphore::signal() {
2407 sem_post(&_semaphore);
2408 }
2409
2410 void Semaphore::wait() {
2411 sem_wait(&_semaphore);
2412 }
2413
2414 bool Semaphore::trywait() {
2415 return sem_trywait(&_semaphore) == 0;
2416 }
2417
2418 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2419
2420 struct timespec ts;
2421 // Semaphore's are always associated with CLOCK_REALTIME
2422 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2423 // see unpackTime for discussion on overflow checking
2424 if (sec >= MAX_SECS) {
2425 ts.tv_sec += MAX_SECS;
2426 ts.tv_nsec = 0;
2427 } else {
2428 ts.tv_sec += sec;
2429 ts.tv_nsec += nsec;
2430 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2431 ts.tv_nsec -= NANOSECS_PER_SEC;
2432 ++ts.tv_sec; // note: this must be <= max_secs
2433 }
2434 }
2435
2436 while (1) {
2437 int result = sem_timedwait(&_semaphore, &ts);
2438 if (result == 0) {
2439 return true;
2440 } else if (errno == EINTR) {
2441 continue;
2442 } else if (errno == ETIMEDOUT) {
2443 return false;
2444 } else {
2445 return false;
2446 }
2447 }
2448 }
2449
2450 extern "C" {
2451 typedef void (*sa_handler_t)(int);
2452 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2453 }
2454
2455 void* os::signal(int signal_number, void* handler) {
2456 struct sigaction sigAct, oldSigAct;
2457
2458 sigfillset(&(sigAct.sa_mask));
2459 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2460 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2461
2462 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2463 // -1 means registration failed
2464 return (void *)-1;
2465 }
2466
2467 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2468 }
2469
2470 void os::signal_raise(int signal_number) {
2471 ::raise(signal_number);
2472 }
2473
2474 // The following code is moved from os.cpp for making this
2475 // code platform specific, which it is by its very nature.
2476
2477 // Will be modified when max signal is changed to be dynamic
2478 int os::sigexitnum_pd() {
2479 return NSIG;
2480 }
2481
2482 // a counter for each possible signal value
2483 static volatile jint pending_signals[NSIG+1] = { 0 };
2484
2485 // Linux(POSIX) specific hand shaking semaphore.
2486 static sem_t sig_sem;
2487 static Semaphore sr_semaphore;
2488
2489 void os::signal_init_pd() {
2490 // Initialize signal structures
2491 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2492
2493 // Initialize signal semaphore
2494 ::sem_init(&sig_sem, 0, 0);
2495 }
2496
2497 void os::signal_notify(int sig) {
2498 Atomic::inc(&pending_signals[sig]);
2499 ::sem_post(&sig_sem);
2500 }
2501
2502 static int check_pending_signals(bool wait) {
2503 Atomic::store(0, &sigint_count);
2504 for (;;) {
2505 for (int i = 0; i < NSIG + 1; i++) {
2506 jint n = pending_signals[i];
2507 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2508 return i;
2509 }
2510 }
2511 if (!wait) {
2512 return -1;
2513 }
2514 JavaThread *thread = JavaThread::current();
2515 ThreadBlockInVM tbivm(thread);
2516
2517 bool threadIsSuspended;
2518 do {
2519 thread->set_suspend_equivalent();
2520 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2521 ::sem_wait(&sig_sem);
2522
2523 // were we externally suspended while we were waiting?
2524 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2525 if (threadIsSuspended) {
2526 // The semaphore has been incremented, but while we were waiting
2527 // another thread suspended us. We don't want to continue running
2528 // while suspended because that would surprise the thread that
2529 // suspended us.
2530 ::sem_post(&sig_sem);
2531
2532 thread->java_suspend_self();
2533 }
2534 } while (threadIsSuspended);
2535 }
2536 }
2537
2538 int os::signal_lookup() {
2539 return check_pending_signals(false);
2540 }
2541
2542 int os::signal_wait() {
2543 return check_pending_signals(true);
2544 }
2545
2546 ////////////////////////////////////////////////////////////////////////////////
2547 // Virtual Memory
2548
2549 int os::vm_page_size() {
2550 // Seems redundant as all get out
2551 assert(os::Linux::page_size() != -1, "must call os::init");
2552 return os::Linux::page_size();
2553 }
2554
2555 // Solaris allocates memory by pages.
2556 int os::vm_allocation_granularity() {
2557 assert(os::Linux::page_size() != -1, "must call os::init");
2558 return os::Linux::page_size();
2559 }
2560
2561 // Rationale behind this function:
2562 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2563 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2564 // samples for JITted code. Here we create private executable mapping over the code cache
2565 // and then we can use standard (well, almost, as mapping can change) way to provide
2566 // info for the reporting script by storing timestamp and location of symbol
2567 void linux_wrap_code(char* base, size_t size) {
2568 static volatile jint cnt = 0;
2569
2570 if (!UseOprofile) {
2571 return;
2572 }
2573
2574 char buf[PATH_MAX+1];
2575 int num = Atomic::add(1, &cnt);
2576
2577 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2578 os::get_temp_directory(), os::current_process_id(), num);
2579 unlink(buf);
2580
2581 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2582
2583 if (fd != -1) {
2584 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2585 if (rv != (off_t)-1) {
2586 if (::write(fd, "", 1) == 1) {
2587 mmap(base, size,
2588 PROT_READ|PROT_WRITE|PROT_EXEC,
2589 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2590 }
2591 }
2592 ::close(fd);
2593 unlink(buf);
2594 }
2595 }
2596
2597 static bool recoverable_mmap_error(int err) {
2598 // See if the error is one we can let the caller handle. This
2599 // list of errno values comes from JBS-6843484. I can't find a
2600 // Linux man page that documents this specific set of errno
2601 // values so while this list currently matches Solaris, it may
2602 // change as we gain experience with this failure mode.
2603 switch (err) {
2604 case EBADF:
2605 case EINVAL:
2606 case ENOTSUP:
2607 // let the caller deal with these errors
2608 return true;
2609
2610 default:
2611 // Any remaining errors on this OS can cause our reserved mapping
2612 // to be lost. That can cause confusion where different data
2613 // structures think they have the same memory mapped. The worst
2614 // scenario is if both the VM and a library think they have the
2615 // same memory mapped.
2616 return false;
2617 }
2618 }
2619
2620 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2621 int err) {
2622 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2624 strerror(err), err);
2625 }
2626
2627 static void warn_fail_commit_memory(char* addr, size_t size,
2628 size_t alignment_hint, bool exec,
2629 int err) {
2630 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2631 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2632 alignment_hint, exec, strerror(err), err);
2633 }
2634
2635 // NOTE: Linux kernel does not really reserve the pages for us.
2636 // All it does is to check if there are enough free pages
2637 // left at the time of mmap(). This could be a potential
2638 // problem.
2639 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2640 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2641 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2642 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2643 if (res != (uintptr_t) MAP_FAILED) {
2644 if (UseNUMAInterleaving) {
2645 numa_make_global(addr, size);
2646 }
2647 return 0;
2648 }
2649
2650 int err = errno; // save errno from mmap() call above
2651
2652 if (!recoverable_mmap_error(err)) {
2653 warn_fail_commit_memory(addr, size, exec, err);
2654 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2655 }
2656
2657 return err;
2658 }
2659
2660 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2661 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2662 }
2663
2664 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2665 const char* mesg) {
2666 assert(mesg != NULL, "mesg must be specified");
2667 int err = os::Linux::commit_memory_impl(addr, size, exec);
2668 if (err != 0) {
2669 // the caller wants all commit errors to exit with the specified mesg:
2670 warn_fail_commit_memory(addr, size, exec, err);
2671 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2672 }
2673 }
2674
2675 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2676 #ifndef MAP_HUGETLB
2677 #define MAP_HUGETLB 0x40000
2678 #endif
2679
2680 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2681 #ifndef MADV_HUGEPAGE
2682 #define MADV_HUGEPAGE 14
2683 #endif
2684
2685 int os::Linux::commit_memory_impl(char* addr, size_t size,
2686 size_t alignment_hint, bool exec) {
2687 int err = os::Linux::commit_memory_impl(addr, size, exec);
2688 if (err == 0) {
2689 realign_memory(addr, size, alignment_hint);
2690 }
2691 return err;
2692 }
2693
2694 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2695 bool exec) {
2696 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2697 }
2698
2699 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2700 size_t alignment_hint, bool exec,
2701 const char* mesg) {
2702 assert(mesg != NULL, "mesg must be specified");
2703 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2704 if (err != 0) {
2705 // the caller wants all commit errors to exit with the specified mesg:
2706 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2708 }
2709 }
2710
2711 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2712 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2713 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2714 // be supported or the memory may already be backed by huge pages.
2715 ::madvise(addr, bytes, MADV_HUGEPAGE);
2716 }
2717 }
2718
2719 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2720 // This method works by doing an mmap over an existing mmaping and effectively discarding
2721 // the existing pages. However it won't work for SHM-based large pages that cannot be
2722 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2723 // small pages on top of the SHM segment. This method always works for small pages, so we
2724 // allow that in any case.
2725 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2726 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2727 }
2728 }
2729
2730 void os::numa_make_global(char *addr, size_t bytes) {
2731 Linux::numa_interleave_memory(addr, bytes);
2732 }
2733
2734 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2735 // bind policy to MPOL_PREFERRED for the current thread.
2736 #define USE_MPOL_PREFERRED 0
2737
2738 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2739 // To make NUMA and large pages more robust when both enabled, we need to ease
2740 // the requirements on where the memory should be allocated. MPOL_BIND is the
2741 // default policy and it will force memory to be allocated on the specified
2742 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2743 // the specified node, but will not force it. Using this policy will prevent
2744 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2745 // free large pages.
2746 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2747 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2748 }
2749
2750 bool os::numa_topology_changed() { return false; }
2751
2752 size_t os::numa_get_groups_num() {
2753 int max_node = Linux::numa_max_node();
2754 return max_node > 0 ? max_node + 1 : 1;
2755 }
2756
2757 int os::numa_get_group_id() {
2758 int cpu_id = Linux::sched_getcpu();
2759 if (cpu_id != -1) {
2760 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2761 if (lgrp_id != -1) {
2762 return lgrp_id;
2763 }
2764 }
2765 return 0;
2766 }
2767
2768 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2769 for (size_t i = 0; i < size; i++) {
2770 ids[i] = i;
2771 }
2772 return size;
2773 }
2774
2775 bool os::get_page_info(char *start, page_info* info) {
2776 return false;
2777 }
2778
2779 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2780 page_info* page_found) {
2781 return end;
2782 }
2783
2784
2785 int os::Linux::sched_getcpu_syscall(void) {
2786 unsigned int cpu;
2787 int retval = -1;
2788
2789 #if defined(IA32)
2790 #ifndef SYS_getcpu
2791 #define SYS_getcpu 318
2792 #endif
2793 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2794 #elif defined(AMD64)
2795 // Unfortunately we have to bring all these macros here from vsyscall.h
2796 // to be able to compile on old linuxes.
2797 #define __NR_vgetcpu 2
2798 #define VSYSCALL_START (-10UL << 20)
2799 #define VSYSCALL_SIZE 1024
2800 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2801 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2802 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2803 retval = vgetcpu(&cpu, NULL, NULL);
2804 #endif
2805
2806 return (retval == -1) ? retval : cpu;
2807 }
2808
2809 // Something to do with the numa-aware allocator needs these symbols
2810 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2811 extern "C" JNIEXPORT void numa_error(char *where) { }
2812 extern "C" JNIEXPORT int fork1() { return fork(); }
2813
2814
2815 // If we are running with libnuma version > 2, then we should
2816 // be trying to use symbols with versions 1.1
2817 // If we are running with earlier version, which did not have symbol versions,
2818 // we should use the base version.
2819 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2820 void *f = dlvsym(handle, name, "libnuma_1.1");
2821 if (f == NULL) {
2822 f = dlsym(handle, name);
2823 }
2824 return f;
2825 }
2826
2827 bool os::Linux::libnuma_init() {
2828 // sched_getcpu() should be in libc.
2829 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2830 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2831
2832 // If it's not, try a direct syscall.
2833 if (sched_getcpu() == -1) {
2834 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2835 (void*)&sched_getcpu_syscall));
2836 }
2837
2838 if (sched_getcpu() != -1) { // Does it work?
2839 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2840 if (handle != NULL) {
2841 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2842 libnuma_dlsym(handle, "numa_node_to_cpus")));
2843 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2844 libnuma_dlsym(handle, "numa_max_node")));
2845 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2846 libnuma_dlsym(handle, "numa_available")));
2847 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2848 libnuma_dlsym(handle, "numa_tonode_memory")));
2849 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2850 libnuma_dlsym(handle, "numa_interleave_memory")));
2851 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2852 libnuma_dlsym(handle, "numa_set_bind_policy")));
2853
2854
2855 if (numa_available() != -1) {
2856 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2857 // Create a cpu -> node mapping
2858 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2859 rebuild_cpu_to_node_map();
2860 return true;
2861 }
2862 }
2863 }
2864 return false;
2865 }
2866
2867 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2868 // The table is later used in get_node_by_cpu().
2869 void os::Linux::rebuild_cpu_to_node_map() {
2870 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2871 // in libnuma (possible values are starting from 16,
2872 // and continuing up with every other power of 2, but less
2873 // than the maximum number of CPUs supported by kernel), and
2874 // is a subject to change (in libnuma version 2 the requirements
2875 // are more reasonable) we'll just hardcode the number they use
2876 // in the library.
2877 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2878
2879 size_t cpu_num = os::active_processor_count();
2880 size_t cpu_map_size = NCPUS / BitsPerCLong;
2881 size_t cpu_map_valid_size =
2882 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2883
2884 cpu_to_node()->clear();
2885 cpu_to_node()->at_grow(cpu_num - 1);
2886 size_t node_num = numa_get_groups_num();
2887
2888 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2889 for (size_t i = 0; i < node_num; i++) {
2890 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2891 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2892 if (cpu_map[j] != 0) {
2893 for (size_t k = 0; k < BitsPerCLong; k++) {
2894 if (cpu_map[j] & (1UL << k)) {
2895 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2896 }
2897 }
2898 }
2899 }
2900 }
2901 }
2902 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2903 }
2904
2905 int os::Linux::get_node_by_cpu(int cpu_id) {
2906 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2907 return cpu_to_node()->at(cpu_id);
2908 }
2909 return -1;
2910 }
2911
2912 GrowableArray<int>* os::Linux::_cpu_to_node;
2913 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2914 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2915 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2916 os::Linux::numa_available_func_t os::Linux::_numa_available;
2917 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2918 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2919 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2920 unsigned long* os::Linux::_numa_all_nodes;
2921
2922 bool os::pd_uncommit_memory(char* addr, size_t size) {
2923 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2924 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2925 return res != (uintptr_t) MAP_FAILED;
2926 }
2927
2928 static address get_stack_commited_bottom(address bottom, size_t size) {
2929 address nbot = bottom;
2930 address ntop = bottom + size;
2931
2932 size_t page_sz = os::vm_page_size();
2933 unsigned pages = size / page_sz;
2934
2935 unsigned char vec[1];
2936 unsigned imin = 1, imax = pages + 1, imid;
2937 int mincore_return_value = 0;
2938
2939 assert(imin <= imax, "Unexpected page size");
2940
2941 while (imin < imax) {
2942 imid = (imax + imin) / 2;
2943 nbot = ntop - (imid * page_sz);
2944
2945 // Use a trick with mincore to check whether the page is mapped or not.
2946 // mincore sets vec to 1 if page resides in memory and to 0 if page
2947 // is swapped output but if page we are asking for is unmapped
2948 // it returns -1,ENOMEM
2949 mincore_return_value = mincore(nbot, page_sz, vec);
2950
2951 if (mincore_return_value == -1) {
2952 // Page is not mapped go up
2953 // to find first mapped page
2954 if (errno != EAGAIN) {
2955 assert(errno == ENOMEM, "Unexpected mincore errno");
2956 imax = imid;
2957 }
2958 } else {
2959 // Page is mapped go down
2960 // to find first not mapped page
2961 imin = imid + 1;
2962 }
2963 }
2964
2965 nbot = nbot + page_sz;
2966
2967 // Adjust stack bottom one page up if last checked page is not mapped
2968 if (mincore_return_value == -1) {
2969 nbot = nbot + page_sz;
2970 }
2971
2972 return nbot;
2973 }
2974
2975
2976 // Linux uses a growable mapping for the stack, and if the mapping for
2977 // the stack guard pages is not removed when we detach a thread the
2978 // stack cannot grow beyond the pages where the stack guard was
2979 // mapped. If at some point later in the process the stack expands to
2980 // that point, the Linux kernel cannot expand the stack any further
2981 // because the guard pages are in the way, and a segfault occurs.
2982 //
2983 // However, it's essential not to split the stack region by unmapping
2984 // a region (leaving a hole) that's already part of the stack mapping,
2985 // so if the stack mapping has already grown beyond the guard pages at
2986 // the time we create them, we have to truncate the stack mapping.
2987 // So, we need to know the extent of the stack mapping when
2988 // create_stack_guard_pages() is called.
2989
2990 // We only need this for stacks that are growable: at the time of
2991 // writing thread stacks don't use growable mappings (i.e. those
2992 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2993 // only applies to the main thread.
2994
2995 // If the (growable) stack mapping already extends beyond the point
2996 // where we're going to put our guard pages, truncate the mapping at
2997 // that point by munmap()ping it. This ensures that when we later
2998 // munmap() the guard pages we don't leave a hole in the stack
2999 // mapping. This only affects the main/initial thread
3000
3001 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3002 if (os::Linux::is_initial_thread()) {
3003 // As we manually grow stack up to bottom inside create_attached_thread(),
3004 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3005 // we don't need to do anything special.
3006 // Check it first, before calling heavy function.
3007 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3008 unsigned char vec[1];
3009
3010 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3011 // Fallback to slow path on all errors, including EAGAIN
3012 stack_extent = (uintptr_t) get_stack_commited_bottom(
3013 os::Linux::initial_thread_stack_bottom(),
3014 (size_t)addr - stack_extent);
3015 }
3016
3017 if (stack_extent < (uintptr_t)addr) {
3018 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3019 }
3020 }
3021
3022 return os::commit_memory(addr, size, !ExecMem);
3023 }
3024
3025 // If this is a growable mapping, remove the guard pages entirely by
3026 // munmap()ping them. If not, just call uncommit_memory(). This only
3027 // affects the main/initial thread, but guard against future OS changes
3028 // It's safe to always unmap guard pages for initial thread because we
3029 // always place it right after end of the mapped region
3030
3031 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3032 uintptr_t stack_extent, stack_base;
3033
3034 if (os::Linux::is_initial_thread()) {
3035 return ::munmap(addr, size) == 0;
3036 }
3037
3038 return os::uncommit_memory(addr, size);
3039 }
3040
3041 static address _highest_vm_reserved_address = NULL;
3042
3043 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3044 // at 'requested_addr'. If there are existing memory mappings at the same
3045 // location, however, they will be overwritten. If 'fixed' is false,
3046 // 'requested_addr' is only treated as a hint, the return value may or
3047 // may not start from the requested address. Unlike Linux mmap(), this
3048 // function returns NULL to indicate failure.
3049 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3050 char * addr;
3051 int flags;
3052
3053 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3054 if (fixed) {
3055 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3056 flags |= MAP_FIXED;
3057 }
3058
3059 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3060 // touch an uncommitted page. Otherwise, the read/write might
3061 // succeed if we have enough swap space to back the physical page.
3062 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3063 flags, -1, 0);
3064
3065 if (addr != MAP_FAILED) {
3066 // anon_mmap() should only get called during VM initialization,
3067 // don't need lock (actually we can skip locking even it can be called
3068 // from multiple threads, because _highest_vm_reserved_address is just a
3069 // hint about the upper limit of non-stack memory regions.)
3070 if ((address)addr + bytes > _highest_vm_reserved_address) {
3071 _highest_vm_reserved_address = (address)addr + bytes;
3072 }
3073 }
3074
3075 return addr == MAP_FAILED ? NULL : addr;
3076 }
3077
3078 // Don't update _highest_vm_reserved_address, because there might be memory
3079 // regions above addr + size. If so, releasing a memory region only creates
3080 // a hole in the address space, it doesn't help prevent heap-stack collision.
3081 //
3082 static int anon_munmap(char * addr, size_t size) {
3083 return ::munmap(addr, size) == 0;
3084 }
3085
3086 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3087 size_t alignment_hint) {
3088 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3089 }
3090
3091 bool os::pd_release_memory(char* addr, size_t size) {
3092 return anon_munmap(addr, size);
3093 }
3094
3095 static address highest_vm_reserved_address() {
3096 return _highest_vm_reserved_address;
3097 }
3098
3099 static bool linux_mprotect(char* addr, size_t size, int prot) {
3100 // Linux wants the mprotect address argument to be page aligned.
3101 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3102
3103 // According to SUSv3, mprotect() should only be used with mappings
3104 // established by mmap(), and mmap() always maps whole pages. Unaligned
3105 // 'addr' likely indicates problem in the VM (e.g. trying to change
3106 // protection of malloc'ed or statically allocated memory). Check the
3107 // caller if you hit this assert.
3108 assert(addr == bottom, "sanity check");
3109
3110 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3111 return ::mprotect(bottom, size, prot) == 0;
3112 }
3113
3114 // Set protections specified
3115 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3116 bool is_committed) {
3117 unsigned int p = 0;
3118 switch (prot) {
3119 case MEM_PROT_NONE: p = PROT_NONE; break;
3120 case MEM_PROT_READ: p = PROT_READ; break;
3121 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3122 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3123 default:
3124 ShouldNotReachHere();
3125 }
3126 // is_committed is unused.
3127 return linux_mprotect(addr, bytes, p);
3128 }
3129
3130 bool os::guard_memory(char* addr, size_t size) {
3131 return linux_mprotect(addr, size, PROT_NONE);
3132 }
3133
3134 bool os::unguard_memory(char* addr, size_t size) {
3135 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3136 }
3137
3138 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3139 size_t page_size) {
3140 bool result = false;
3141 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3142 MAP_ANONYMOUS|MAP_PRIVATE,
3143 -1, 0);
3144 if (p != MAP_FAILED) {
3145 void *aligned_p = align_ptr_up(p, page_size);
3146
3147 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3148
3149 munmap(p, page_size * 2);
3150 }
3151
3152 if (warn && !result) {
3153 warning("TransparentHugePages is not supported by the operating system.");
3154 }
3155
3156 return result;
3157 }
3158
3159 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3160 bool result = false;
3161 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3162 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3163 -1, 0);
3164
3165 if (p != MAP_FAILED) {
3166 // We don't know if this really is a huge page or not.
3167 FILE *fp = fopen("/proc/self/maps", "r");
3168 if (fp) {
3169 while (!feof(fp)) {
3170 char chars[257];
3171 long x = 0;
3172 if (fgets(chars, sizeof(chars), fp)) {
3173 if (sscanf(chars, "%lx-%*x", &x) == 1
3174 && x == (long)p) {
3175 if (strstr (chars, "hugepage")) {
3176 result = true;
3177 break;
3178 }
3179 }
3180 }
3181 }
3182 fclose(fp);
3183 }
3184 munmap(p, page_size);
3185 }
3186
3187 if (warn && !result) {
3188 warning("HugeTLBFS is not supported by the operating system.");
3189 }
3190
3191 return result;
3192 }
3193
3194 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3195 //
3196 // From the coredump_filter documentation:
3197 //
3198 // - (bit 0) anonymous private memory
3199 // - (bit 1) anonymous shared memory
3200 // - (bit 2) file-backed private memory
3201 // - (bit 3) file-backed shared memory
3202 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3203 // effective only if the bit 2 is cleared)
3204 // - (bit 5) hugetlb private memory
3205 // - (bit 6) hugetlb shared memory
3206 //
3207 static void set_coredump_filter(void) {
3208 FILE *f;
3209 long cdm;
3210
3211 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3212 return;
3213 }
3214
3215 if (fscanf(f, "%lx", &cdm) != 1) {
3216 fclose(f);
3217 return;
3218 }
3219
3220 rewind(f);
3221
3222 if ((cdm & LARGEPAGES_BIT) == 0) {
3223 cdm |= LARGEPAGES_BIT;
3224 fprintf(f, "%#lx", cdm);
3225 }
3226
3227 fclose(f);
3228 }
3229
3230 // Large page support
3231
3232 static size_t _large_page_size = 0;
3233
3234 size_t os::Linux::find_large_page_size() {
3235 size_t large_page_size = 0;
3236
3237 // large_page_size on Linux is used to round up heap size. x86 uses either
3238 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3239 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3240 // page as large as 256M.
3241 //
3242 // Here we try to figure out page size by parsing /proc/meminfo and looking
3243 // for a line with the following format:
3244 // Hugepagesize: 2048 kB
3245 //
3246 // If we can't determine the value (e.g. /proc is not mounted, or the text
3247 // format has been changed), we'll use the largest page size supported by
3248 // the processor.
3249
3250 #ifndef ZERO
3251 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3252 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3253 #endif // ZERO
3254
3255 FILE *fp = fopen("/proc/meminfo", "r");
3256 if (fp) {
3257 while (!feof(fp)) {
3258 int x = 0;
3259 char buf[16];
3260 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3261 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3262 large_page_size = x * K;
3263 break;
3264 }
3265 } else {
3266 // skip to next line
3267 for (;;) {
3268 int ch = fgetc(fp);
3269 if (ch == EOF || ch == (int)'\n') break;
3270 }
3271 }
3272 }
3273 fclose(fp);
3274 }
3275
3276 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3277 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3278 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3279 proper_unit_for_byte_size(large_page_size));
3280 }
3281
3282 return large_page_size;
3283 }
3284
3285 size_t os::Linux::setup_large_page_size() {
3286 _large_page_size = Linux::find_large_page_size();
3287 const size_t default_page_size = (size_t)Linux::page_size();
3288 if (_large_page_size > default_page_size) {
3289 _page_sizes[0] = _large_page_size;
3290 _page_sizes[1] = default_page_size;
3291 _page_sizes[2] = 0;
3292 }
3293
3294 return _large_page_size;
3295 }
3296
3297 bool os::Linux::setup_large_page_type(size_t page_size) {
3298 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3299 FLAG_IS_DEFAULT(UseSHM) &&
3300 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3301
3302 // The type of large pages has not been specified by the user.
3303
3304 // Try UseHugeTLBFS and then UseSHM.
3305 UseHugeTLBFS = UseSHM = true;
3306
3307 // Don't try UseTransparentHugePages since there are known
3308 // performance issues with it turned on. This might change in the future.
3309 UseTransparentHugePages = false;
3310 }
3311
3312 if (UseTransparentHugePages) {
3313 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3314 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3315 UseHugeTLBFS = false;
3316 UseSHM = false;
3317 return true;
3318 }
3319 UseTransparentHugePages = false;
3320 }
3321
3322 if (UseHugeTLBFS) {
3323 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3324 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3325 UseSHM = false;
3326 return true;
3327 }
3328 UseHugeTLBFS = false;
3329 }
3330
3331 return UseSHM;
3332 }
3333
3334 void os::large_page_init() {
3335 if (!UseLargePages &&
3336 !UseTransparentHugePages &&
3337 !UseHugeTLBFS &&
3338 !UseSHM) {
3339 // Not using large pages.
3340 return;
3341 }
3342
3343 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3344 // The user explicitly turned off large pages.
3345 // Ignore the rest of the large pages flags.
3346 UseTransparentHugePages = false;
3347 UseHugeTLBFS = false;
3348 UseSHM = false;
3349 return;
3350 }
3351
3352 size_t large_page_size = Linux::setup_large_page_size();
3353 UseLargePages = Linux::setup_large_page_type(large_page_size);
3354
3355 set_coredump_filter();
3356 }
3357
3358 #ifndef SHM_HUGETLB
3359 #define SHM_HUGETLB 04000
3360 #endif
3361
3362 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3363 char* req_addr, bool exec) {
3364 // "exec" is passed in but not used. Creating the shared image for
3365 // the code cache doesn't have an SHM_X executable permission to check.
3366 assert(UseLargePages && UseSHM, "only for SHM large pages");
3367 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3368
3369 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3370 return NULL; // Fallback to small pages.
3371 }
3372
3373 key_t key = IPC_PRIVATE;
3374 char *addr;
3375
3376 bool warn_on_failure = UseLargePages &&
3377 (!FLAG_IS_DEFAULT(UseLargePages) ||
3378 !FLAG_IS_DEFAULT(UseSHM) ||
3379 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3380 char msg[128];
3381
3382 // Create a large shared memory region to attach to based on size.
3383 // Currently, size is the total size of the heap
3384 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3385 if (shmid == -1) {
3386 // Possible reasons for shmget failure:
3387 // 1. shmmax is too small for Java heap.
3388 // > check shmmax value: cat /proc/sys/kernel/shmmax
3389 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3390 // 2. not enough large page memory.
3391 // > check available large pages: cat /proc/meminfo
3392 // > increase amount of large pages:
3393 // echo new_value > /proc/sys/vm/nr_hugepages
3394 // Note 1: different Linux may use different name for this property,
3395 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3396 // Note 2: it's possible there's enough physical memory available but
3397 // they are so fragmented after a long run that they can't
3398 // coalesce into large pages. Try to reserve large pages when
3399 // the system is still "fresh".
3400 if (warn_on_failure) {
3401 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3402 warning("%s", msg);
3403 }
3404 return NULL;
3405 }
3406
3407 // attach to the region
3408 addr = (char*)shmat(shmid, req_addr, 0);
3409 int err = errno;
3410
3411 // Remove shmid. If shmat() is successful, the actual shared memory segment
3412 // will be deleted when it's detached by shmdt() or when the process
3413 // terminates. If shmat() is not successful this will remove the shared
3414 // segment immediately.
3415 shmctl(shmid, IPC_RMID, NULL);
3416
3417 if ((intptr_t)addr == -1) {
3418 if (warn_on_failure) {
3419 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3420 warning("%s", msg);
3421 }
3422 return NULL;
3423 }
3424
3425 return addr;
3426 }
3427
3428 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3429 int error) {
3430 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3431
3432 bool warn_on_failure = UseLargePages &&
3433 (!FLAG_IS_DEFAULT(UseLargePages) ||
3434 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3435 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3436
3437 if (warn_on_failure) {
3438 char msg[128];
3439 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3440 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3441 warning("%s", msg);
3442 }
3443 }
3444
3445 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3446 char* req_addr,
3447 bool exec) {
3448 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3449 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3450 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3451
3452 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3453 char* addr = (char*)::mmap(req_addr, bytes, prot,
3454 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3455 -1, 0);
3456
3457 if (addr == MAP_FAILED) {
3458 warn_on_large_pages_failure(req_addr, bytes, errno);
3459 return NULL;
3460 }
3461
3462 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3463
3464 return addr;
3465 }
3466
3467 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3468 size_t alignment,
3469 char* req_addr,
3470 bool exec) {
3471 size_t large_page_size = os::large_page_size();
3472
3473 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3474
3475 // Allocate small pages.
3476
3477 char* start;
3478 if (req_addr != NULL) {
3479 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3480 assert(is_size_aligned(bytes, alignment), "Must be");
3481 start = os::reserve_memory(bytes, req_addr);
3482 assert(start == NULL || start == req_addr, "Must be");
3483 } else {
3484 start = os::reserve_memory_aligned(bytes, alignment);
3485 }
3486
3487 if (start == NULL) {
3488 return NULL;
3489 }
3490
3491 assert(is_ptr_aligned(start, alignment), "Must be");
3492
3493 if (MemTracker::tracking_level() > NMT_minimal) {
3494 // os::reserve_memory_special will record this memory area.
3495 // Need to release it here to prevent overlapping reservations.
3496 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3497 tkr.record((address)start, bytes);
3498 }
3499
3500 char* end = start + bytes;
3501
3502 // Find the regions of the allocated chunk that can be promoted to large pages.
3503 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3504 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3505
3506 size_t lp_bytes = lp_end - lp_start;
3507
3508 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3509
3510 if (lp_bytes == 0) {
3511 // The mapped region doesn't even span the start and the end of a large page.
3512 // Fall back to allocate a non-special area.
3513 ::munmap(start, end - start);
3514 return NULL;
3515 }
3516
3517 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3518
3519
3520 void* result;
3521
3522 if (start != lp_start) {
3523 result = ::mmap(start, lp_start - start, prot,
3524 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3525 -1, 0);
3526 if (result == MAP_FAILED) {
3527 ::munmap(lp_start, end - lp_start);
3528 return NULL;
3529 }
3530 }
3531
3532 result = ::mmap(lp_start, lp_bytes, prot,
3533 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3534 -1, 0);
3535 if (result == MAP_FAILED) {
3536 warn_on_large_pages_failure(req_addr, bytes, errno);
3537 // If the mmap above fails, the large pages region will be unmapped and we
3538 // have regions before and after with small pages. Release these regions.
3539 //
3540 // | mapped | unmapped | mapped |
3541 // ^ ^ ^ ^
3542 // start lp_start lp_end end
3543 //
3544 ::munmap(start, lp_start - start);
3545 ::munmap(lp_end, end - lp_end);
3546 return NULL;
3547 }
3548
3549 if (lp_end != end) {
3550 result = ::mmap(lp_end, end - lp_end, prot,
3551 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3552 -1, 0);
3553 if (result == MAP_FAILED) {
3554 ::munmap(start, lp_end - start);
3555 return NULL;
3556 }
3557 }
3558
3559 return start;
3560 }
3561
3562 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3563 size_t alignment,
3564 char* req_addr,
3565 bool exec) {
3566 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3567 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3568 assert(is_power_of_2(alignment), "Must be");
3569 assert(is_power_of_2(os::large_page_size()), "Must be");
3570 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3571
3572 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3573 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3574 } else {
3575 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3576 }
3577 }
3578
3579 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3580 char* req_addr, bool exec) {
3581 assert(UseLargePages, "only for large pages");
3582
3583 char* addr;
3584 if (UseSHM) {
3585 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3586 } else {
3587 assert(UseHugeTLBFS, "must be");
3588 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3589 }
3590
3591 if (addr != NULL) {
3592 if (UseNUMAInterleaving) {
3593 numa_make_global(addr, bytes);
3594 }
3595
3596 // The memory is committed
3597 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3598 }
3599
3600 return addr;
3601 }
3602
3603 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3604 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3605 return shmdt(base) == 0;
3606 }
3607
3608 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3609 return pd_release_memory(base, bytes);
3610 }
3611
3612 bool os::release_memory_special(char* base, size_t bytes) {
3613 bool res;
3614 if (MemTracker::tracking_level() > NMT_minimal) {
3615 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3616 res = os::Linux::release_memory_special_impl(base, bytes);
3617 if (res) {
3618 tkr.record((address)base, bytes);
3619 }
3620
3621 } else {
3622 res = os::Linux::release_memory_special_impl(base, bytes);
3623 }
3624 return res;
3625 }
3626
3627 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3628 assert(UseLargePages, "only for large pages");
3629 bool res;
3630
3631 if (UseSHM) {
3632 res = os::Linux::release_memory_special_shm(base, bytes);
3633 } else {
3634 assert(UseHugeTLBFS, "must be");
3635 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3636 }
3637 return res;
3638 }
3639
3640 size_t os::large_page_size() {
3641 return _large_page_size;
3642 }
3643
3644 // With SysV SHM the entire memory region must be allocated as shared
3645 // memory.
3646 // HugeTLBFS allows application to commit large page memory on demand.
3647 // However, when committing memory with HugeTLBFS fails, the region
3648 // that was supposed to be committed will lose the old reservation
3649 // and allow other threads to steal that memory region. Because of this
3650 // behavior we can't commit HugeTLBFS memory.
3651 bool os::can_commit_large_page_memory() {
3652 return UseTransparentHugePages;
3653 }
3654
3655 bool os::can_execute_large_page_memory() {
3656 return UseTransparentHugePages || UseHugeTLBFS;
3657 }
3658
3659 // Reserve memory at an arbitrary address, only if that area is
3660 // available (and not reserved for something else).
3661
3662 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3663 const int max_tries = 10;
3664 char* base[max_tries];
3665 size_t size[max_tries];
3666 const size_t gap = 0x000000;
3667
3668 // Assert only that the size is a multiple of the page size, since
3669 // that's all that mmap requires, and since that's all we really know
3670 // about at this low abstraction level. If we need higher alignment,
3671 // we can either pass an alignment to this method or verify alignment
3672 // in one of the methods further up the call chain. See bug 5044738.
3673 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3674
3675 // Repeatedly allocate blocks until the block is allocated at the
3676 // right spot. Give up after max_tries. Note that reserve_memory() will
3677 // automatically update _highest_vm_reserved_address if the call is
3678 // successful. The variable tracks the highest memory address every reserved
3679 // by JVM. It is used to detect heap-stack collision if running with
3680 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3681 // space than needed, it could confuse the collision detecting code. To
3682 // solve the problem, save current _highest_vm_reserved_address and
3683 // calculate the correct value before return.
3684 address old_highest = _highest_vm_reserved_address;
3685
3686 // Linux mmap allows caller to pass an address as hint; give it a try first,
3687 // if kernel honors the hint then we can return immediately.
3688 char * addr = anon_mmap(requested_addr, bytes, false);
3689 if (addr == requested_addr) {
3690 return requested_addr;
3691 }
3692
3693 if (addr != NULL) {
3694 // mmap() is successful but it fails to reserve at the requested address
3695 anon_munmap(addr, bytes);
3696 }
3697
3698 int i;
3699 for (i = 0; i < max_tries; ++i) {
3700 base[i] = reserve_memory(bytes);
3701
3702 if (base[i] != NULL) {
3703 // Is this the block we wanted?
3704 if (base[i] == requested_addr) {
3705 size[i] = bytes;
3706 break;
3707 }
3708
3709 // Does this overlap the block we wanted? Give back the overlapped
3710 // parts and try again.
3711
3712 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3713 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3714 unmap_memory(base[i], top_overlap);
3715 base[i] += top_overlap;
3716 size[i] = bytes - top_overlap;
3717 } else {
3718 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3719 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3720 unmap_memory(requested_addr, bottom_overlap);
3721 size[i] = bytes - bottom_overlap;
3722 } else {
3723 size[i] = bytes;
3724 }
3725 }
3726 }
3727 }
3728
3729 // Give back the unused reserved pieces.
3730
3731 for (int j = 0; j < i; ++j) {
3732 if (base[j] != NULL) {
3733 unmap_memory(base[j], size[j]);
3734 }
3735 }
3736
3737 if (i < max_tries) {
3738 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3739 return requested_addr;
3740 } else {
3741 _highest_vm_reserved_address = old_highest;
3742 return NULL;
3743 }
3744 }
3745
3746 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3747 return ::read(fd, buf, nBytes);
3748 }
3749
3750 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3751 return ::pread(fd, buf, nBytes, offset);
3752 }
3753
3754 // Short sleep, direct OS call.
3755 //
3756 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3757 // sched_yield(2) will actually give up the CPU:
3758 //
3759 // * Alone on this pariticular CPU, keeps running.
3760 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3761 // (pre 2.6.39).
3762 //
3763 // So calling this with 0 is an alternative.
3764 //
3765 void os::naked_short_sleep(jlong ms) {
3766 struct timespec req;
3767
3768 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3769 req.tv_sec = 0;
3770 if (ms > 0) {
3771 req.tv_nsec = (ms % 1000) * 1000000;
3772 } else {
3773 req.tv_nsec = 1;
3774 }
3775
3776 nanosleep(&req, NULL);
3777
3778 return;
3779 }
3780
3781 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3782 void os::infinite_sleep() {
3783 while (true) { // sleep forever ...
3784 ::sleep(100); // ... 100 seconds at a time
3785 }
3786 }
3787
3788 // Used to convert frequent JVM_Yield() to nops
3789 bool os::dont_yield() {
3790 return DontYieldALot;
3791 }
3792
3793 void os::naked_yield() {
3794 sched_yield();
3795 }
3796
3797 ////////////////////////////////////////////////////////////////////////////////
3798 // thread priority support
3799
3800 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3801 // only supports dynamic priority, static priority must be zero. For real-time
3802 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3803 // However, for large multi-threaded applications, SCHED_RR is not only slower
3804 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3805 // of 5 runs - Sep 2005).
3806 //
3807 // The following code actually changes the niceness of kernel-thread/LWP. It
3808 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3809 // not the entire user process, and user level threads are 1:1 mapped to kernel
3810 // threads. It has always been the case, but could change in the future. For
3811 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3812 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3813
3814 int os::java_to_os_priority[CriticalPriority + 1] = {
3815 19, // 0 Entry should never be used
3816
3817 4, // 1 MinPriority
3818 3, // 2
3819 2, // 3
3820
3821 1, // 4
3822 0, // 5 NormPriority
3823 -1, // 6
3824
3825 -2, // 7
3826 -3, // 8
3827 -4, // 9 NearMaxPriority
3828
3829 -5, // 10 MaxPriority
3830
3831 -5 // 11 CriticalPriority
3832 };
3833
3834 static int prio_init() {
3835 if (ThreadPriorityPolicy == 1) {
3836 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3837 // if effective uid is not root. Perhaps, a more elegant way of doing
3838 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3839 if (geteuid() != 0) {
3840 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3841 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3842 }
3843 ThreadPriorityPolicy = 0;
3844 }
3845 }
3846 if (UseCriticalJavaThreadPriority) {
3847 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3848 }
3849 return 0;
3850 }
3851
3852 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3853 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3854
3855 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3856 return (ret == 0) ? OS_OK : OS_ERR;
3857 }
3858
3859 OSReturn os::get_native_priority(const Thread* const thread,
3860 int *priority_ptr) {
3861 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3862 *priority_ptr = java_to_os_priority[NormPriority];
3863 return OS_OK;
3864 }
3865
3866 errno = 0;
3867 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3868 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3869 }
3870
3871 // Hint to the underlying OS that a task switch would not be good.
3872 // Void return because it's a hint and can fail.
3873 void os::hint_no_preempt() {}
3874
3875 ////////////////////////////////////////////////////////////////////////////////
3876 // suspend/resume support
3877
3878 // the low-level signal-based suspend/resume support is a remnant from the
3879 // old VM-suspension that used to be for java-suspension, safepoints etc,
3880 // within hotspot. Now there is a single use-case for this:
3881 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3882 // that runs in the watcher thread.
3883 // The remaining code is greatly simplified from the more general suspension
3884 // code that used to be used.
3885 //
3886 // The protocol is quite simple:
3887 // - suspend:
3888 // - sends a signal to the target thread
3889 // - polls the suspend state of the osthread using a yield loop
3890 // - target thread signal handler (SR_handler) sets suspend state
3891 // and blocks in sigsuspend until continued
3892 // - resume:
3893 // - sets target osthread state to continue
3894 // - sends signal to end the sigsuspend loop in the SR_handler
3895 //
3896 // Note that the SR_lock plays no role in this suspend/resume protocol.
3897
3898 static void resume_clear_context(OSThread *osthread) {
3899 osthread->set_ucontext(NULL);
3900 osthread->set_siginfo(NULL);
3901 }
3902
3903 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3904 ucontext_t* context) {
3905 osthread->set_ucontext(context);
3906 osthread->set_siginfo(siginfo);
3907 }
3908
3909 // Handler function invoked when a thread's execution is suspended or
3910 // resumed. We have to be careful that only async-safe functions are
3911 // called here (Note: most pthread functions are not async safe and
3912 // should be avoided.)
3913 //
3914 // Note: sigwait() is a more natural fit than sigsuspend() from an
3915 // interface point of view, but sigwait() prevents the signal hander
3916 // from being run. libpthread would get very confused by not having
3917 // its signal handlers run and prevents sigwait()'s use with the
3918 // mutex granting granting signal.
3919 //
3920 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3921 //
3922 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3923 // Save and restore errno to avoid confusing native code with EINTR
3924 // after sigsuspend.
3925 int old_errno = errno;
3926
3927 Thread* thread = Thread::current();
3928 OSThread* osthread = thread->osthread();
3929 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3930
3931 os::SuspendResume::State current = osthread->sr.state();
3932 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3933 suspend_save_context(osthread, siginfo, context);
3934
3935 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3936 os::SuspendResume::State state = osthread->sr.suspended();
3937 if (state == os::SuspendResume::SR_SUSPENDED) {
3938 sigset_t suspend_set; // signals for sigsuspend()
3939
3940 // get current set of blocked signals and unblock resume signal
3941 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3942 sigdelset(&suspend_set, SR_signum);
3943
3944 sr_semaphore.signal();
3945 // wait here until we are resumed
3946 while (1) {
3947 sigsuspend(&suspend_set);
3948
3949 os::SuspendResume::State result = osthread->sr.running();
3950 if (result == os::SuspendResume::SR_RUNNING) {
3951 sr_semaphore.signal();
3952 break;
3953 }
3954 }
3955
3956 } else if (state == os::SuspendResume::SR_RUNNING) {
3957 // request was cancelled, continue
3958 } else {
3959 ShouldNotReachHere();
3960 }
3961
3962 resume_clear_context(osthread);
3963 } else if (current == os::SuspendResume::SR_RUNNING) {
3964 // request was cancelled, continue
3965 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3966 // ignore
3967 } else {
3968 // ignore
3969 }
3970
3971 errno = old_errno;
3972 }
3973
3974
3975 static int SR_initialize() {
3976 struct sigaction act;
3977 char *s;
3978 // Get signal number to use for suspend/resume
3979 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3980 int sig = ::strtol(s, 0, 10);
3981 if (sig > 0 || sig < _NSIG) {
3982 SR_signum = sig;
3983 }
3984 }
3985
3986 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3987 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3988
3989 sigemptyset(&SR_sigset);
3990 sigaddset(&SR_sigset, SR_signum);
3991
3992 // Set up signal handler for suspend/resume
3993 act.sa_flags = SA_RESTART|SA_SIGINFO;
3994 act.sa_handler = (void (*)(int)) SR_handler;
3995
3996 // SR_signum is blocked by default.
3997 // 4528190 - We also need to block pthread restart signal (32 on all
3998 // supported Linux platforms). Note that LinuxThreads need to block
3999 // this signal for all threads to work properly. So we don't have
4000 // to use hard-coded signal number when setting up the mask.
4001 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4002
4003 if (sigaction(SR_signum, &act, 0) == -1) {
4004 return -1;
4005 }
4006
4007 // Save signal flag
4008 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4009 return 0;
4010 }
4011
4012 static int sr_notify(OSThread* osthread) {
4013 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4014 assert_status(status == 0, status, "pthread_kill");
4015 return status;
4016 }
4017
4018 // "Randomly" selected value for how long we want to spin
4019 // before bailing out on suspending a thread, also how often
4020 // we send a signal to a thread we want to resume
4021 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4022 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4023
4024 // returns true on success and false on error - really an error is fatal
4025 // but this seems the normal response to library errors
4026 static bool do_suspend(OSThread* osthread) {
4027 assert(osthread->sr.is_running(), "thread should be running");
4028 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4029
4030 // mark as suspended and send signal
4031 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4032 // failed to switch, state wasn't running?
4033 ShouldNotReachHere();
4034 return false;
4035 }
4036
4037 if (sr_notify(osthread) != 0) {
4038 ShouldNotReachHere();
4039 }
4040
4041 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4042 while (true) {
4043 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4044 break;
4045 } else {
4046 // timeout
4047 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4048 if (cancelled == os::SuspendResume::SR_RUNNING) {
4049 return false;
4050 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4051 // make sure that we consume the signal on the semaphore as well
4052 sr_semaphore.wait();
4053 break;
4054 } else {
4055 ShouldNotReachHere();
4056 return false;
4057 }
4058 }
4059 }
4060
4061 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4062 return true;
4063 }
4064
4065 static void do_resume(OSThread* osthread) {
4066 assert(osthread->sr.is_suspended(), "thread should be suspended");
4067 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4068
4069 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4070 // failed to switch to WAKEUP_REQUEST
4071 ShouldNotReachHere();
4072 return;
4073 }
4074
4075 while (true) {
4076 if (sr_notify(osthread) == 0) {
4077 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4078 if (osthread->sr.is_running()) {
4079 return;
4080 }
4081 }
4082 } else {
4083 ShouldNotReachHere();
4084 }
4085 }
4086
4087 guarantee(osthread->sr.is_running(), "Must be running!");
4088 }
4089
4090 ///////////////////////////////////////////////////////////////////////////////////
4091 // signal handling (except suspend/resume)
4092
4093 // This routine may be used by user applications as a "hook" to catch signals.
4094 // The user-defined signal handler must pass unrecognized signals to this
4095 // routine, and if it returns true (non-zero), then the signal handler must
4096 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4097 // routine will never retun false (zero), but instead will execute a VM panic
4098 // routine kill the process.
4099 //
4100 // If this routine returns false, it is OK to call it again. This allows
4101 // the user-defined signal handler to perform checks either before or after
4102 // the VM performs its own checks. Naturally, the user code would be making
4103 // a serious error if it tried to handle an exception (such as a null check
4104 // or breakpoint) that the VM was generating for its own correct operation.
4105 //
4106 // This routine may recognize any of the following kinds of signals:
4107 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4108 // It should be consulted by handlers for any of those signals.
4109 //
4110 // The caller of this routine must pass in the three arguments supplied
4111 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4112 // field of the structure passed to sigaction(). This routine assumes that
4113 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4114 //
4115 // Note that the VM will print warnings if it detects conflicting signal
4116 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4117 //
4118 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4119 siginfo_t* siginfo,
4120 void* ucontext,
4121 int abort_if_unrecognized);
4122
4123 void signalHandler(int sig, siginfo_t* info, void* uc) {
4124 assert(info != NULL && uc != NULL, "it must be old kernel");
4125 int orig_errno = errno; // Preserve errno value over signal handler.
4126 JVM_handle_linux_signal(sig, info, uc, true);
4127 errno = orig_errno;
4128 }
4129
4130
4131 // This boolean allows users to forward their own non-matching signals
4132 // to JVM_handle_linux_signal, harmlessly.
4133 bool os::Linux::signal_handlers_are_installed = false;
4134
4135 // For signal-chaining
4136 struct sigaction os::Linux::sigact[MAXSIGNUM];
4137 unsigned int os::Linux::sigs = 0;
4138 bool os::Linux::libjsig_is_loaded = false;
4139 typedef struct sigaction *(*get_signal_t)(int);
4140 get_signal_t os::Linux::get_signal_action = NULL;
4141
4142 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4143 struct sigaction *actp = NULL;
4144
4145 if (libjsig_is_loaded) {
4146 // Retrieve the old signal handler from libjsig
4147 actp = (*get_signal_action)(sig);
4148 }
4149 if (actp == NULL) {
4150 // Retrieve the preinstalled signal handler from jvm
4151 actp = get_preinstalled_handler(sig);
4152 }
4153
4154 return actp;
4155 }
4156
4157 static bool call_chained_handler(struct sigaction *actp, int sig,
4158 siginfo_t *siginfo, void *context) {
4159 // Call the old signal handler
4160 if (actp->sa_handler == SIG_DFL) {
4161 // It's more reasonable to let jvm treat it as an unexpected exception
4162 // instead of taking the default action.
4163 return false;
4164 } else if (actp->sa_handler != SIG_IGN) {
4165 if ((actp->sa_flags & SA_NODEFER) == 0) {
4166 // automaticlly block the signal
4167 sigaddset(&(actp->sa_mask), sig);
4168 }
4169
4170 sa_handler_t hand;
4171 sa_sigaction_t sa;
4172 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4173 // retrieve the chained handler
4174 if (siginfo_flag_set) {
4175 sa = actp->sa_sigaction;
4176 } else {
4177 hand = actp->sa_handler;
4178 }
4179
4180 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4181 actp->sa_handler = SIG_DFL;
4182 }
4183
4184 // try to honor the signal mask
4185 sigset_t oset;
4186 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4187
4188 // call into the chained handler
4189 if (siginfo_flag_set) {
4190 (*sa)(sig, siginfo, context);
4191 } else {
4192 (*hand)(sig);
4193 }
4194
4195 // restore the signal mask
4196 pthread_sigmask(SIG_SETMASK, &oset, 0);
4197 }
4198 // Tell jvm's signal handler the signal is taken care of.
4199 return true;
4200 }
4201
4202 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4203 bool chained = false;
4204 // signal-chaining
4205 if (UseSignalChaining) {
4206 struct sigaction *actp = get_chained_signal_action(sig);
4207 if (actp != NULL) {
4208 chained = call_chained_handler(actp, sig, siginfo, context);
4209 }
4210 }
4211 return chained;
4212 }
4213
4214 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4215 if ((((unsigned int)1 << sig) & sigs) != 0) {
4216 return &sigact[sig];
4217 }
4218 return NULL;
4219 }
4220
4221 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4222 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4223 sigact[sig] = oldAct;
4224 sigs |= (unsigned int)1 << sig;
4225 }
4226
4227 // for diagnostic
4228 int os::Linux::sigflags[MAXSIGNUM];
4229
4230 int os::Linux::get_our_sigflags(int sig) {
4231 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4232 return sigflags[sig];
4233 }
4234
4235 void os::Linux::set_our_sigflags(int sig, int flags) {
4236 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4237 sigflags[sig] = flags;
4238 }
4239
4240 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4241 // Check for overwrite.
4242 struct sigaction oldAct;
4243 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4244
4245 void* oldhand = oldAct.sa_sigaction
4246 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4247 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4248 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4249 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4250 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4251 if (AllowUserSignalHandlers || !set_installed) {
4252 // Do not overwrite; user takes responsibility to forward to us.
4253 return;
4254 } else if (UseSignalChaining) {
4255 // save the old handler in jvm
4256 save_preinstalled_handler(sig, oldAct);
4257 // libjsig also interposes the sigaction() call below and saves the
4258 // old sigaction on it own.
4259 } else {
4260 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4261 "%#lx for signal %d.", (long)oldhand, sig));
4262 }
4263 }
4264
4265 struct sigaction sigAct;
4266 sigfillset(&(sigAct.sa_mask));
4267 sigAct.sa_handler = SIG_DFL;
4268 if (!set_installed) {
4269 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4270 } else {
4271 sigAct.sa_sigaction = signalHandler;
4272 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4273 }
4274 // Save flags, which are set by ours
4275 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4276 sigflags[sig] = sigAct.sa_flags;
4277
4278 int ret = sigaction(sig, &sigAct, &oldAct);
4279 assert(ret == 0, "check");
4280
4281 void* oldhand2 = oldAct.sa_sigaction
4282 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4283 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4284 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4285 }
4286
4287 // install signal handlers for signals that HotSpot needs to
4288 // handle in order to support Java-level exception handling.
4289
4290 void os::Linux::install_signal_handlers() {
4291 if (!signal_handlers_are_installed) {
4292 signal_handlers_are_installed = true;
4293
4294 // signal-chaining
4295 typedef void (*signal_setting_t)();
4296 signal_setting_t begin_signal_setting = NULL;
4297 signal_setting_t end_signal_setting = NULL;
4298 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4299 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4300 if (begin_signal_setting != NULL) {
4301 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4302 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4303 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4304 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4305 libjsig_is_loaded = true;
4306 assert(UseSignalChaining, "should enable signal-chaining");
4307 }
4308 if (libjsig_is_loaded) {
4309 // Tell libjsig jvm is setting signal handlers
4310 (*begin_signal_setting)();
4311 }
4312
4313 set_signal_handler(SIGSEGV, true);
4314 set_signal_handler(SIGPIPE, true);
4315 set_signal_handler(SIGBUS, true);
4316 set_signal_handler(SIGILL, true);
4317 set_signal_handler(SIGFPE, true);
4318 #if defined(PPC64)
4319 set_signal_handler(SIGTRAP, true);
4320 #endif
4321 set_signal_handler(SIGXFSZ, true);
4322
4323 if (libjsig_is_loaded) {
4324 // Tell libjsig jvm finishes setting signal handlers
4325 (*end_signal_setting)();
4326 }
4327
4328 // We don't activate signal checker if libjsig is in place, we trust ourselves
4329 // and if UserSignalHandler is installed all bets are off.
4330 // Log that signal checking is off only if -verbose:jni is specified.
4331 if (CheckJNICalls) {
4332 if (libjsig_is_loaded) {
4333 if (PrintJNIResolving) {
4334 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4335 }
4336 check_signals = false;
4337 }
4338 if (AllowUserSignalHandlers) {
4339 if (PrintJNIResolving) {
4340 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4341 }
4342 check_signals = false;
4343 }
4344 }
4345 }
4346 }
4347
4348 // This is the fastest way to get thread cpu time on Linux.
4349 // Returns cpu time (user+sys) for any thread, not only for current.
4350 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4351 // It might work on 2.6.10+ with a special kernel/glibc patch.
4352 // For reference, please, see IEEE Std 1003.1-2004:
4353 // http://www.unix.org/single_unix_specification
4354
4355 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4356 struct timespec tp;
4357 int rc = os::Linux::clock_gettime(clockid, &tp);
4358 assert(rc == 0, "clock_gettime is expected to return 0 code");
4359
4360 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4361 }
4362
4363 /////
4364 // glibc on Linux platform uses non-documented flag
4365 // to indicate, that some special sort of signal
4366 // trampoline is used.
4367 // We will never set this flag, and we should
4368 // ignore this flag in our diagnostic
4369 #ifdef SIGNIFICANT_SIGNAL_MASK
4370 #undef SIGNIFICANT_SIGNAL_MASK
4371 #endif
4372 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4373
4374 static const char* get_signal_handler_name(address handler,
4375 char* buf, int buflen) {
4376 int offset;
4377 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4378 if (found) {
4379 // skip directory names
4380 const char *p1, *p2;
4381 p1 = buf;
4382 size_t len = strlen(os::file_separator());
4383 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4384 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4385 } else {
4386 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4387 }
4388 return buf;
4389 }
4390
4391 static void print_signal_handler(outputStream* st, int sig,
4392 char* buf, size_t buflen) {
4393 struct sigaction sa;
4394
4395 sigaction(sig, NULL, &sa);
4396
4397 // See comment for SIGNIFICANT_SIGNAL_MASK define
4398 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4399
4400 st->print("%s: ", os::exception_name(sig, buf, buflen));
4401
4402 address handler = (sa.sa_flags & SA_SIGINFO)
4403 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4404 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4405
4406 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4407 st->print("SIG_DFL");
4408 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4409 st->print("SIG_IGN");
4410 } else {
4411 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4412 }
4413
4414 st->print(", sa_mask[0]=");
4415 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4416
4417 address rh = VMError::get_resetted_sighandler(sig);
4418 // May be, handler was resetted by VMError?
4419 if (rh != NULL) {
4420 handler = rh;
4421 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4422 }
4423
4424 st->print(", sa_flags=");
4425 os::Posix::print_sa_flags(st, sa.sa_flags);
4426
4427 // Check: is it our handler?
4428 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4429 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4430 // It is our signal handler
4431 // check for flags, reset system-used one!
4432 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4433 st->print(
4434 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4435 os::Linux::get_our_sigflags(sig));
4436 }
4437 }
4438 st->cr();
4439 }
4440
4441
4442 #define DO_SIGNAL_CHECK(sig) \
4443 do { \
4444 if (!sigismember(&check_signal_done, sig)) { \
4445 os::Linux::check_signal_handler(sig); \
4446 } \
4447 } while (0)
4448
4449 // This method is a periodic task to check for misbehaving JNI applications
4450 // under CheckJNI, we can add any periodic checks here
4451
4452 void os::run_periodic_checks() {
4453 if (check_signals == false) return;
4454
4455 // SEGV and BUS if overridden could potentially prevent
4456 // generation of hs*.log in the event of a crash, debugging
4457 // such a case can be very challenging, so we absolutely
4458 // check the following for a good measure:
4459 DO_SIGNAL_CHECK(SIGSEGV);
4460 DO_SIGNAL_CHECK(SIGILL);
4461 DO_SIGNAL_CHECK(SIGFPE);
4462 DO_SIGNAL_CHECK(SIGBUS);
4463 DO_SIGNAL_CHECK(SIGPIPE);
4464 DO_SIGNAL_CHECK(SIGXFSZ);
4465 #if defined(PPC64)
4466 DO_SIGNAL_CHECK(SIGTRAP);
4467 #endif
4468
4469 // ReduceSignalUsage allows the user to override these handlers
4470 // see comments at the very top and jvm_solaris.h
4471 if (!ReduceSignalUsage) {
4472 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4473 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4474 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4475 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4476 }
4477
4478 DO_SIGNAL_CHECK(SR_signum);
4479 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4480 }
4481
4482 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4483
4484 static os_sigaction_t os_sigaction = NULL;
4485
4486 void os::Linux::check_signal_handler(int sig) {
4487 char buf[O_BUFLEN];
4488 address jvmHandler = NULL;
4489
4490
4491 struct sigaction act;
4492 if (os_sigaction == NULL) {
4493 // only trust the default sigaction, in case it has been interposed
4494 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4495 if (os_sigaction == NULL) return;
4496 }
4497
4498 os_sigaction(sig, (struct sigaction*)NULL, &act);
4499
4500
4501 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4502
4503 address thisHandler = (act.sa_flags & SA_SIGINFO)
4504 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4505 : CAST_FROM_FN_PTR(address, act.sa_handler);
4506
4507
4508 switch (sig) {
4509 case SIGSEGV:
4510 case SIGBUS:
4511 case SIGFPE:
4512 case SIGPIPE:
4513 case SIGILL:
4514 case SIGXFSZ:
4515 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4516 break;
4517
4518 case SHUTDOWN1_SIGNAL:
4519 case SHUTDOWN2_SIGNAL:
4520 case SHUTDOWN3_SIGNAL:
4521 case BREAK_SIGNAL:
4522 jvmHandler = (address)user_handler();
4523 break;
4524
4525 case INTERRUPT_SIGNAL:
4526 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4527 break;
4528
4529 default:
4530 if (sig == SR_signum) {
4531 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4532 } else {
4533 return;
4534 }
4535 break;
4536 }
4537
4538 if (thisHandler != jvmHandler) {
4539 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4540 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4541 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4542 // No need to check this sig any longer
4543 sigaddset(&check_signal_done, sig);
4544 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4545 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4546 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4547 exception_name(sig, buf, O_BUFLEN));
4548 }
4549 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4550 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4551 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4552 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4553 // No need to check this sig any longer
4554 sigaddset(&check_signal_done, sig);
4555 }
4556
4557 // Dump all the signal
4558 if (sigismember(&check_signal_done, sig)) {
4559 print_signal_handlers(tty, buf, O_BUFLEN);
4560 }
4561 }
4562
4563 extern void report_error(char* file_name, int line_no, char* title,
4564 char* format, ...);
4565
4566 extern bool signal_name(int signo, char* buf, size_t len);
4567
4568 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4569 if (0 < exception_code && exception_code <= SIGRTMAX) {
4570 // signal
4571 if (!signal_name(exception_code, buf, size)) {
4572 jio_snprintf(buf, size, "SIG%d", exception_code);
4573 }
4574 return buf;
4575 } else {
4576 return NULL;
4577 }
4578 }
4579
4580 // this is called _before_ the most of global arguments have been parsed
4581 void os::init(void) {
4582 char dummy; // used to get a guess on initial stack address
4583 // first_hrtime = gethrtime();
4584
4585 // With LinuxThreads the JavaMain thread pid (primordial thread)
4586 // is different than the pid of the java launcher thread.
4587 // So, on Linux, the launcher thread pid is passed to the VM
4588 // via the sun.java.launcher.pid property.
4589 // Use this property instead of getpid() if it was correctly passed.
4590 // See bug 6351349.
4591 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4592
4593 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4594
4595 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4596
4597 init_random(1234567);
4598
4599 ThreadCritical::initialize();
4600
4601 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4602 if (Linux::page_size() == -1) {
4603 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4604 strerror(errno)));
4605 }
4606 init_page_sizes((size_t) Linux::page_size());
4607
4608 Linux::initialize_system_info();
4609
4610 // main_thread points to the aboriginal thread
4611 Linux::_main_thread = pthread_self();
4612
4613 Linux::clock_init();
4614 initial_time_count = javaTimeNanos();
4615
4616 // pthread_condattr initialization for monotonic clock
4617 int status;
4618 pthread_condattr_t* _condattr = os::Linux::condAttr();
4619 if ((status = pthread_condattr_init(_condattr)) != 0) {
4620 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4621 }
4622 // Only set the clock if CLOCK_MONOTONIC is available
4623 if (os::supports_monotonic_clock()) {
4624 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4625 if (status == EINVAL) {
4626 warning("Unable to use monotonic clock with relative timed-waits" \
4627 " - changes to the time-of-day clock may have adverse affects");
4628 } else {
4629 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4630 }
4631 }
4632 }
4633 // else it defaults to CLOCK_REALTIME
4634
4635 // If the pagesize of the VM is greater than 8K determine the appropriate
4636 // number of initial guard pages. The user can change this with the
4637 // command line arguments, if needed.
4638 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4639 StackYellowPages = 1;
4640 StackRedPages = 1;
4641 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4642 }
4643
4644 // retrieve entry point for pthread_setname_np
4645 Linux::_pthread_setname_np =
4646 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4647
4648 }
4649
4650 // To install functions for atexit system call
4651 extern "C" {
4652 static void perfMemory_exit_helper() {
4653 perfMemory_exit();
4654 }
4655 }
4656
4657 // this is called _after_ the global arguments have been parsed
4658 jint os::init_2(void) {
4659 Linux::fast_thread_clock_init();
4660
4661 // Allocate a single page and mark it as readable for safepoint polling
4662 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4663 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4664
4665 os::set_polling_page(polling_page);
4666
4667 #ifndef PRODUCT
4668 if (Verbose && PrintMiscellaneous) {
4669 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4670 (intptr_t)polling_page);
4671 }
4672 #endif
4673
4674 if (!UseMembar) {
4675 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4676 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4677 os::set_memory_serialize_page(mem_serialize_page);
4678
4679 #ifndef PRODUCT
4680 if (Verbose && PrintMiscellaneous) {
4681 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4682 (intptr_t)mem_serialize_page);
4683 }
4684 #endif
4685 }
4686
4687 // initialize suspend/resume support - must do this before signal_sets_init()
4688 if (SR_initialize() != 0) {
4689 perror("SR_initialize failed");
4690 return JNI_ERR;
4691 }
4692
4693 Linux::signal_sets_init();
4694 Linux::install_signal_handlers();
4695
4696 // Check minimum allowable stack size for thread creation and to initialize
4697 // the java system classes, including StackOverflowError - depends on page
4698 // size. Add a page for compiler2 recursion in main thread.
4699 // Add in 2*BytesPerWord times page size to account for VM stack during
4700 // class initialization depending on 32 or 64 bit VM.
4701 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4702 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4703 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4704
4705 size_t threadStackSizeInBytes = ThreadStackSize * K;
4706 if (threadStackSizeInBytes != 0 &&
4707 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4708 tty->print_cr("\nThe stack size specified is too small, "
4709 "Specify at least %dk",
4710 os::Linux::min_stack_allowed/ K);
4711 return JNI_ERR;
4712 }
4713
4714 // Make the stack size a multiple of the page size so that
4715 // the yellow/red zones can be guarded.
4716 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4717 vm_page_size()));
4718
4719 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4720
4721 #if defined(IA32)
4722 workaround_expand_exec_shield_cs_limit();
4723 #endif
4724
4725 Linux::libpthread_init();
4726 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4727 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4728 Linux::glibc_version(), Linux::libpthread_version(),
4729 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4730 }
4731
4732 if (UseNUMA) {
4733 if (!Linux::libnuma_init()) {
4734 UseNUMA = false;
4735 } else {
4736 if ((Linux::numa_max_node() < 1)) {
4737 // There's only one node(they start from 0), disable NUMA.
4738 UseNUMA = false;
4739 }
4740 }
4741 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4742 // we can make the adaptive lgrp chunk resizing work. If the user specified
4743 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4744 // disable adaptive resizing.
4745 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4746 if (FLAG_IS_DEFAULT(UseNUMA)) {
4747 UseNUMA = false;
4748 } else {
4749 if (FLAG_IS_DEFAULT(UseLargePages) &&
4750 FLAG_IS_DEFAULT(UseSHM) &&
4751 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4752 UseLargePages = false;
4753 } else {
4754 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4755 UseAdaptiveSizePolicy = false;
4756 UseAdaptiveNUMAChunkSizing = false;
4757 }
4758 }
4759 }
4760 if (!UseNUMA && ForceNUMA) {
4761 UseNUMA = true;
4762 }
4763 }
4764
4765 if (MaxFDLimit) {
4766 // set the number of file descriptors to max. print out error
4767 // if getrlimit/setrlimit fails but continue regardless.
4768 struct rlimit nbr_files;
4769 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4770 if (status != 0) {
4771 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4772 perror("os::init_2 getrlimit failed");
4773 }
4774 } else {
4775 nbr_files.rlim_cur = nbr_files.rlim_max;
4776 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4777 if (status != 0) {
4778 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4779 perror("os::init_2 setrlimit failed");
4780 }
4781 }
4782 }
4783 }
4784
4785 // Initialize lock used to serialize thread creation (see os::create_thread)
4786 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4787
4788 // at-exit methods are called in the reverse order of their registration.
4789 // atexit functions are called on return from main or as a result of a
4790 // call to exit(3C). There can be only 32 of these functions registered
4791 // and atexit() does not set errno.
4792
4793 if (PerfAllowAtExitRegistration) {
4794 // only register atexit functions if PerfAllowAtExitRegistration is set.
4795 // atexit functions can be delayed until process exit time, which
4796 // can be problematic for embedded VM situations. Embedded VMs should
4797 // call DestroyJavaVM() to assure that VM resources are released.
4798
4799 // note: perfMemory_exit_helper atexit function may be removed in
4800 // the future if the appropriate cleanup code can be added to the
4801 // VM_Exit VMOperation's doit method.
4802 if (atexit(perfMemory_exit_helper) != 0) {
4803 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4804 }
4805 }
4806
4807 // initialize thread priority policy
4808 prio_init();
4809
4810 return JNI_OK;
4811 }
4812
4813 // Mark the polling page as unreadable
4814 void os::make_polling_page_unreadable(void) {
4815 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4816 fatal("Could not disable polling page");
4817 }
4818 }
4819
4820 // Mark the polling page as readable
4821 void os::make_polling_page_readable(void) {
4822 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4823 fatal("Could not enable polling page");
4824 }
4825 }
4826
4827 int os::active_processor_count() {
4828 // Linux doesn't yet have a (official) notion of processor sets,
4829 // so just return the number of online processors.
4830 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4831 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4832 return online_cpus;
4833 }
4834
4835 void os::set_native_thread_name(const char *name) {
4836 if (Linux::_pthread_setname_np) {
4837 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4838 snprintf(buf, sizeof(buf), "%s", name);
4839 buf[sizeof(buf) - 1] = '\0';
4840 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4841 // ERANGE should not happen; all other errors should just be ignored.
4842 assert(rc != ERANGE, "pthread_setname_np failed");
4843 }
4844 }
4845
4846 bool os::distribute_processes(uint length, uint* distribution) {
4847 // Not yet implemented.
4848 return false;
4849 }
4850
4851 bool os::bind_to_processor(uint processor_id) {
4852 // Not yet implemented.
4853 return false;
4854 }
4855
4856 ///
4857
4858 void os::SuspendedThreadTask::internal_do_task() {
4859 if (do_suspend(_thread->osthread())) {
4860 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4861 do_task(context);
4862 do_resume(_thread->osthread());
4863 }
4864 }
4865
4866 class PcFetcher : public os::SuspendedThreadTask {
4867 public:
4868 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4869 ExtendedPC result();
4870 protected:
4871 void do_task(const os::SuspendedThreadTaskContext& context);
4872 private:
4873 ExtendedPC _epc;
4874 };
4875
4876 ExtendedPC PcFetcher::result() {
4877 guarantee(is_done(), "task is not done yet.");
4878 return _epc;
4879 }
4880
4881 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4882 Thread* thread = context.thread();
4883 OSThread* osthread = thread->osthread();
4884 if (osthread->ucontext() != NULL) {
4885 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4886 } else {
4887 // NULL context is unexpected, double-check this is the VMThread
4888 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4889 }
4890 }
4891
4892 // Suspends the target using the signal mechanism and then grabs the PC before
4893 // resuming the target. Used by the flat-profiler only
4894 ExtendedPC os::get_thread_pc(Thread* thread) {
4895 // Make sure that it is called by the watcher for the VMThread
4896 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4897 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4898
4899 PcFetcher fetcher(thread);
4900 fetcher.run();
4901 return fetcher.result();
4902 }
4903
4904 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4905 pthread_mutex_t *_mutex,
4906 const struct timespec *_abstime) {
4907 if (is_NPTL()) {
4908 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4909 } else {
4910 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4911 // word back to default 64bit precision if condvar is signaled. Java
4912 // wants 53bit precision. Save and restore current value.
4913 int fpu = get_fpu_control_word();
4914 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4915 set_fpu_control_word(fpu);
4916 return status;
4917 }
4918 }
4919
4920 ////////////////////////////////////////////////////////////////////////////////
4921 // debug support
4922
4923 bool os::find(address addr, outputStream* st) {
4924 Dl_info dlinfo;
4925 memset(&dlinfo, 0, sizeof(dlinfo));
4926 if (dladdr(addr, &dlinfo) != 0) {
4927 st->print(PTR_FORMAT ": ", addr);
4928 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4929 st->print("%s+%#x", dlinfo.dli_sname,
4930 addr - (intptr_t)dlinfo.dli_saddr);
4931 } else if (dlinfo.dli_fbase != NULL) {
4932 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4933 } else {
4934 st->print("<absolute address>");
4935 }
4936 if (dlinfo.dli_fname != NULL) {
4937 st->print(" in %s", dlinfo.dli_fname);
4938 }
4939 if (dlinfo.dli_fbase != NULL) {
4940 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4941 }
4942 st->cr();
4943
4944 if (Verbose) {
4945 // decode some bytes around the PC
4946 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4947 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4948 address lowest = (address) dlinfo.dli_sname;
4949 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4950 if (begin < lowest) begin = lowest;
4951 Dl_info dlinfo2;
4952 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4953 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4954 end = (address) dlinfo2.dli_saddr;
4955 }
4956 Disassembler::decode(begin, end, st);
4957 }
4958 return true;
4959 }
4960 return false;
4961 }
4962
4963 ////////////////////////////////////////////////////////////////////////////////
4964 // misc
4965
4966 // This does not do anything on Linux. This is basically a hook for being
4967 // able to use structured exception handling (thread-local exception filters)
4968 // on, e.g., Win32.
4969 void
4970 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4971 JavaCallArguments* args, Thread* thread) {
4972 f(value, method, args, thread);
4973 }
4974
4975 void os::print_statistics() {
4976 }
4977
4978 int os::message_box(const char* title, const char* message) {
4979 int i;
4980 fdStream err(defaultStream::error_fd());
4981 for (i = 0; i < 78; i++) err.print_raw("=");
4982 err.cr();
4983 err.print_raw_cr(title);
4984 for (i = 0; i < 78; i++) err.print_raw("-");
4985 err.cr();
4986 err.print_raw_cr(message);
4987 for (i = 0; i < 78; i++) err.print_raw("=");
4988 err.cr();
4989
4990 char buf[16];
4991 // Prevent process from exiting upon "read error" without consuming all CPU
4992 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4993
4994 return buf[0] == 'y' || buf[0] == 'Y';
4995 }
4996
4997 int os::stat(const char *path, struct stat *sbuf) {
4998 char pathbuf[MAX_PATH];
4999 if (strlen(path) > MAX_PATH - 1) {
5000 errno = ENAMETOOLONG;
5001 return -1;
5002 }
5003 os::native_path(strcpy(pathbuf, path));
5004 return ::stat(pathbuf, sbuf);
5005 }
5006
5007 bool os::check_heap(bool force) {
5008 return true;
5009 }
5010
5011 // Is a (classpath) directory empty?
5012 bool os::dir_is_empty(const char* path) {
5013 DIR *dir = NULL;
5014 struct dirent *ptr;
5015
5016 dir = opendir(path);
5017 if (dir == NULL) return true;
5018
5019 // Scan the directory
5020 bool result = true;
5021 char buf[sizeof(struct dirent) + MAX_PATH];
5022 while (result && (ptr = ::readdir(dir)) != NULL) {
5023 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5024 result = false;
5025 }
5026 }
5027 closedir(dir);
5028 return result;
5029 }
5030
5031 // This code originates from JDK's sysOpen and open64_w
5032 // from src/solaris/hpi/src/system_md.c
5033
5034 int os::open(const char *path, int oflag, int mode) {
5035 if (strlen(path) > MAX_PATH - 1) {
5036 errno = ENAMETOOLONG;
5037 return -1;
5038 }
5039
5040 // All file descriptors that are opened in the Java process and not
5041 // specifically destined for a subprocess should have the close-on-exec
5042 // flag set. If we don't set it, then careless 3rd party native code
5043 // might fork and exec without closing all appropriate file descriptors
5044 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5045 // turn might:
5046 //
5047 // - cause end-of-file to fail to be detected on some file
5048 // descriptors, resulting in mysterious hangs, or
5049 //
5050 // - might cause an fopen in the subprocess to fail on a system
5051 // suffering from bug 1085341.
5052 //
5053 // (Yes, the default setting of the close-on-exec flag is a Unix
5054 // design flaw)
5055 //
5056 // See:
5057 // 1085341: 32-bit stdio routines should support file descriptors >255
5058 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5059 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5060 //
5061 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5062 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5063 // because it saves a system call and removes a small window where the flag
5064 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5065 // and we fall back to using FD_CLOEXEC (see below).
5066 #ifdef O_CLOEXEC
5067 oflag |= O_CLOEXEC;
5068 #endif
5069
5070 int fd = ::open64(path, oflag, mode);
5071 if (fd == -1) return -1;
5072
5073 //If the open succeeded, the file might still be a directory
5074 {
5075 struct stat64 buf64;
5076 int ret = ::fstat64(fd, &buf64);
5077 int st_mode = buf64.st_mode;
5078
5079 if (ret != -1) {
5080 if ((st_mode & S_IFMT) == S_IFDIR) {
5081 errno = EISDIR;
5082 ::close(fd);
5083 return -1;
5084 }
5085 } else {
5086 ::close(fd);
5087 return -1;
5088 }
5089 }
5090
5091 #ifdef FD_CLOEXEC
5092 // Validate that the use of the O_CLOEXEC flag on open above worked.
5093 // With recent kernels, we will perform this check exactly once.
5094 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5095 if (!O_CLOEXEC_is_known_to_work) {
5096 int flags = ::fcntl(fd, F_GETFD);
5097 if (flags != -1) {
5098 if ((flags & FD_CLOEXEC) != 0)
5099 O_CLOEXEC_is_known_to_work = 1;
5100 else
5101 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5102 }
5103 }
5104 #endif
5105
5106 return fd;
5107 }
5108
5109
5110 // create binary file, rewriting existing file if required
5111 int os::create_binary_file(const char* path, bool rewrite_existing) {
5112 int oflags = O_WRONLY | O_CREAT;
5113 if (!rewrite_existing) {
5114 oflags |= O_EXCL;
5115 }
5116 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5117 }
5118
5119 // return current position of file pointer
5120 jlong os::current_file_offset(int fd) {
5121 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5122 }
5123
5124 // move file pointer to the specified offset
5125 jlong os::seek_to_file_offset(int fd, jlong offset) {
5126 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5127 }
5128
5129 // This code originates from JDK's sysAvailable
5130 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5131
5132 int os::available(int fd, jlong *bytes) {
5133 jlong cur, end;
5134 int mode;
5135 struct stat64 buf64;
5136
5137 if (::fstat64(fd, &buf64) >= 0) {
5138 mode = buf64.st_mode;
5139 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5140 // XXX: is the following call interruptible? If so, this might
5141 // need to go through the INTERRUPT_IO() wrapper as for other
5142 // blocking, interruptible calls in this file.
5143 int n;
5144 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5145 *bytes = n;
5146 return 1;
5147 }
5148 }
5149 }
5150 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5151 return 0;
5152 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5153 return 0;
5154 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5155 return 0;
5156 }
5157 *bytes = end - cur;
5158 return 1;
5159 }
5160
5161 // Map a block of memory.
5162 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5163 char *addr, size_t bytes, bool read_only,
5164 bool allow_exec) {
5165 int prot;
5166 int flags = MAP_PRIVATE;
5167
5168 if (read_only) {
5169 prot = PROT_READ;
5170 } else {
5171 prot = PROT_READ | PROT_WRITE;
5172 }
5173
5174 if (allow_exec) {
5175 prot |= PROT_EXEC;
5176 }
5177
5178 if (addr != NULL) {
5179 flags |= MAP_FIXED;
5180 }
5181
5182 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5183 fd, file_offset);
5184 if (mapped_address == MAP_FAILED) {
5185 return NULL;
5186 }
5187 return mapped_address;
5188 }
5189
5190
5191 // Remap a block of memory.
5192 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5193 char *addr, size_t bytes, bool read_only,
5194 bool allow_exec) {
5195 // same as map_memory() on this OS
5196 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5197 allow_exec);
5198 }
5199
5200
5201 // Unmap a block of memory.
5202 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5203 return munmap(addr, bytes) == 0;
5204 }
5205
5206 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5207
5208 static clockid_t thread_cpu_clockid(Thread* thread) {
5209 pthread_t tid = thread->osthread()->pthread_id();
5210 clockid_t clockid;
5211
5212 // Get thread clockid
5213 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5214 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5215 return clockid;
5216 }
5217
5218 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5219 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5220 // of a thread.
5221 //
5222 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5223 // the fast estimate available on the platform.
5224
5225 jlong os::current_thread_cpu_time() {
5226 if (os::Linux::supports_fast_thread_cpu_time()) {
5227 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5228 } else {
5229 // return user + sys since the cost is the same
5230 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5231 }
5232 }
5233
5234 jlong os::thread_cpu_time(Thread* thread) {
5235 // consistent with what current_thread_cpu_time() returns
5236 if (os::Linux::supports_fast_thread_cpu_time()) {
5237 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5238 } else {
5239 return slow_thread_cpu_time(thread, true /* user + sys */);
5240 }
5241 }
5242
5243 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5244 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5245 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5246 } else {
5247 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5248 }
5249 }
5250
5251 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5252 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5253 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5254 } else {
5255 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5256 }
5257 }
5258
5259 // -1 on error.
5260 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5261 pid_t tid = thread->osthread()->thread_id();
5262 char *s;
5263 char stat[2048];
5264 int statlen;
5265 char proc_name[64];
5266 int count;
5267 long sys_time, user_time;
5268 char cdummy;
5269 int idummy;
5270 long ldummy;
5271 FILE *fp;
5272
5273 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5274 fp = fopen(proc_name, "r");
5275 if (fp == NULL) return -1;
5276 statlen = fread(stat, 1, 2047, fp);
5277 stat[statlen] = '\0';
5278 fclose(fp);
5279
5280 // Skip pid and the command string. Note that we could be dealing with
5281 // weird command names, e.g. user could decide to rename java launcher
5282 // to "java 1.4.2 :)", then the stat file would look like
5283 // 1234 (java 1.4.2 :)) R ... ...
5284 // We don't really need to know the command string, just find the last
5285 // occurrence of ")" and then start parsing from there. See bug 4726580.
5286 s = strrchr(stat, ')');
5287 if (s == NULL) return -1;
5288
5289 // Skip blank chars
5290 do { s++; } while (s && isspace(*s));
5291
5292 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5293 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5294 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5295 &user_time, &sys_time);
5296 if (count != 13) return -1;
5297 if (user_sys_cpu_time) {
5298 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5299 } else {
5300 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5301 }
5302 }
5303
5304 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5305 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5306 info_ptr->may_skip_backward = false; // elapsed time not wall time
5307 info_ptr->may_skip_forward = false; // elapsed time not wall time
5308 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5309 }
5310
5311 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5312 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5313 info_ptr->may_skip_backward = false; // elapsed time not wall time
5314 info_ptr->may_skip_forward = false; // elapsed time not wall time
5315 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5316 }
5317
5318 bool os::is_thread_cpu_time_supported() {
5319 return true;
5320 }
5321
5322 // System loadavg support. Returns -1 if load average cannot be obtained.
5323 // Linux doesn't yet have a (official) notion of processor sets,
5324 // so just return the system wide load average.
5325 int os::loadavg(double loadavg[], int nelem) {
5326 return ::getloadavg(loadavg, nelem);
5327 }
5328
5329 void os::pause() {
5330 char filename[MAX_PATH];
5331 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5332 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5333 } else {
5334 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5335 }
5336
5337 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5338 if (fd != -1) {
5339 struct stat buf;
5340 ::close(fd);
5341 while (::stat(filename, &buf) == 0) {
5342 (void)::poll(NULL, 0, 100);
5343 }
5344 } else {
5345 jio_fprintf(stderr,
5346 "Could not open pause file '%s', continuing immediately.\n", filename);
5347 }
5348 }
5349
5350
5351 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5352 // comment paragraphs are worth repeating here:
5353 //
5354 // Assumption:
5355 // Only one parker can exist on an event, which is why we allocate
5356 // them per-thread. Multiple unparkers can coexist.
5357 //
5358 // _Event serves as a restricted-range semaphore.
5359 // -1 : thread is blocked, i.e. there is a waiter
5360 // 0 : neutral: thread is running or ready,
5361 // could have been signaled after a wait started
5362 // 1 : signaled - thread is running or ready
5363 //
5364 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5365 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5366 // For specifics regarding the bug see GLIBC BUGID 261237 :
5367 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5368 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5369 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5370 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5371 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5372 // and monitorenter when we're using 1-0 locking. All those operations may result in
5373 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5374 // of libpthread avoids the problem, but isn't practical.
5375 //
5376 // Possible remedies:
5377 //
5378 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5379 // This is palliative and probabilistic, however. If the thread is preempted
5380 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5381 // than the minimum period may have passed, and the abstime may be stale (in the
5382 // past) resultin in a hang. Using this technique reduces the odds of a hang
5383 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5384 //
5385 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5386 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5387 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5388 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5389 // thread.
5390 //
5391 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5392 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5393 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5394 // This also works well. In fact it avoids kernel-level scalability impediments
5395 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5396 // timers in a graceful fashion.
5397 //
5398 // 4. When the abstime value is in the past it appears that control returns
5399 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5400 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5401 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5402 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5403 // It may be possible to avoid reinitialization by checking the return
5404 // value from pthread_cond_timedwait(). In addition to reinitializing the
5405 // condvar we must establish the invariant that cond_signal() is only called
5406 // within critical sections protected by the adjunct mutex. This prevents
5407 // cond_signal() from "seeing" a condvar that's in the midst of being
5408 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5409 // desirable signal-after-unlock optimization that avoids futile context switching.
5410 //
5411 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5412 // structure when a condvar is used or initialized. cond_destroy() would
5413 // release the helper structure. Our reinitialize-after-timedwait fix
5414 // put excessive stress on malloc/free and locks protecting the c-heap.
5415 //
5416 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5417 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5418 // and only enabling the work-around for vulnerable environments.
5419
5420 // utility to compute the abstime argument to timedwait:
5421 // millis is the relative timeout time
5422 // abstime will be the absolute timeout time
5423 // TODO: replace compute_abstime() with unpackTime()
5424
5425 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5426 if (millis < 0) millis = 0;
5427
5428 jlong seconds = millis / 1000;
5429 millis %= 1000;
5430 if (seconds > 50000000) { // see man cond_timedwait(3T)
5431 seconds = 50000000;
5432 }
5433
5434 if (os::supports_monotonic_clock()) {
5435 struct timespec now;
5436 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5437 assert_status(status == 0, status, "clock_gettime");
5438 abstime->tv_sec = now.tv_sec + seconds;
5439 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5440 if (nanos >= NANOSECS_PER_SEC) {
5441 abstime->tv_sec += 1;
5442 nanos -= NANOSECS_PER_SEC;
5443 }
5444 abstime->tv_nsec = nanos;
5445 } else {
5446 struct timeval now;
5447 int status = gettimeofday(&now, NULL);
5448 assert(status == 0, "gettimeofday");
5449 abstime->tv_sec = now.tv_sec + seconds;
5450 long usec = now.tv_usec + millis * 1000;
5451 if (usec >= 1000000) {
5452 abstime->tv_sec += 1;
5453 usec -= 1000000;
5454 }
5455 abstime->tv_nsec = usec * 1000;
5456 }
5457 return abstime;
5458 }
5459
5460 void os::PlatformEvent::park() { // AKA "down()"
5461 // Transitions for _Event:
5462 // -1 => -1 : illegal
5463 // 1 => 0 : pass - return immediately
5464 // 0 => -1 : block; then set _Event to 0 before returning
5465
5466 // Invariant: Only the thread associated with the Event/PlatformEvent
5467 // may call park().
5468 // TODO: assert that _Assoc != NULL or _Assoc == Self
5469 assert(_nParked == 0, "invariant");
5470
5471 int v;
5472 for (;;) {
5473 v = _Event;
5474 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5475 }
5476 guarantee(v >= 0, "invariant");
5477 if (v == 0) {
5478 // Do this the hard way by blocking ...
5479 int status = pthread_mutex_lock(_mutex);
5480 assert_status(status == 0, status, "mutex_lock");
5481 guarantee(_nParked == 0, "invariant");
5482 ++_nParked;
5483 while (_Event < 0) {
5484 status = pthread_cond_wait(_cond, _mutex);
5485 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5486 // Treat this the same as if the wait was interrupted
5487 if (status == ETIME) { status = EINTR; }
5488 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5489 }
5490 --_nParked;
5491
5492 _Event = 0;
5493 status = pthread_mutex_unlock(_mutex);
5494 assert_status(status == 0, status, "mutex_unlock");
5495 // Paranoia to ensure our locked and lock-free paths interact
5496 // correctly with each other.
5497 OrderAccess::fence();
5498 }
5499 guarantee(_Event >= 0, "invariant");
5500 }
5501
5502 int os::PlatformEvent::park(jlong millis) {
5503 // Transitions for _Event:
5504 // -1 => -1 : illegal
5505 // 1 => 0 : pass - return immediately
5506 // 0 => -1 : block; then set _Event to 0 before returning
5507
5508 guarantee(_nParked == 0, "invariant");
5509
5510 int v;
5511 for (;;) {
5512 v = _Event;
5513 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5514 }
5515 guarantee(v >= 0, "invariant");
5516 if (v != 0) return OS_OK;
5517
5518 // We do this the hard way, by blocking the thread.
5519 // Consider enforcing a minimum timeout value.
5520 struct timespec abst;
5521 compute_abstime(&abst, millis);
5522
5523 int ret = OS_TIMEOUT;
5524 int status = pthread_mutex_lock(_mutex);
5525 assert_status(status == 0, status, "mutex_lock");
5526 guarantee(_nParked == 0, "invariant");
5527 ++_nParked;
5528
5529 // Object.wait(timo) will return because of
5530 // (a) notification
5531 // (b) timeout
5532 // (c) thread.interrupt
5533 //
5534 // Thread.interrupt and object.notify{All} both call Event::set.
5535 // That is, we treat thread.interrupt as a special case of notification.
5536 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5537 // We assume all ETIME returns are valid.
5538 //
5539 // TODO: properly differentiate simultaneous notify+interrupt.
5540 // In that case, we should propagate the notify to another waiter.
5541
5542 while (_Event < 0) {
5543 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5544 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5545 pthread_cond_destroy(_cond);
5546 pthread_cond_init(_cond, os::Linux::condAttr());
5547 }
5548 assert_status(status == 0 || status == EINTR ||
5549 status == ETIME || status == ETIMEDOUT,
5550 status, "cond_timedwait");
5551 if (!FilterSpuriousWakeups) break; // previous semantics
5552 if (status == ETIME || status == ETIMEDOUT) break;
5553 // We consume and ignore EINTR and spurious wakeups.
5554 }
5555 --_nParked;
5556 if (_Event >= 0) {
5557 ret = OS_OK;
5558 }
5559 _Event = 0;
5560 status = pthread_mutex_unlock(_mutex);
5561 assert_status(status == 0, status, "mutex_unlock");
5562 assert(_nParked == 0, "invariant");
5563 // Paranoia to ensure our locked and lock-free paths interact
5564 // correctly with each other.
5565 OrderAccess::fence();
5566 return ret;
5567 }
5568
5569 void os::PlatformEvent::unpark() {
5570 // Transitions for _Event:
5571 // 0 => 1 : just return
5572 // 1 => 1 : just return
5573 // -1 => either 0 or 1; must signal target thread
5574 // That is, we can safely transition _Event from -1 to either
5575 // 0 or 1.
5576 // See also: "Semaphores in Plan 9" by Mullender & Cox
5577 //
5578 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5579 // that it will take two back-to-back park() calls for the owning
5580 // thread to block. This has the benefit of forcing a spurious return
5581 // from the first park() call after an unpark() call which will help
5582 // shake out uses of park() and unpark() without condition variables.
5583
5584 if (Atomic::xchg(1, &_Event) >= 0) return;
5585
5586 // Wait for the thread associated with the event to vacate
5587 int status = pthread_mutex_lock(_mutex);
5588 assert_status(status == 0, status, "mutex_lock");
5589 int AnyWaiters = _nParked;
5590 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5591 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5592 AnyWaiters = 0;
5593 pthread_cond_signal(_cond);
5594 }
5595 status = pthread_mutex_unlock(_mutex);
5596 assert_status(status == 0, status, "mutex_unlock");
5597 if (AnyWaiters != 0) {
5598 // Note that we signal() *after* dropping the lock for "immortal" Events.
5599 // This is safe and avoids a common class of futile wakeups. In rare
5600 // circumstances this can cause a thread to return prematurely from
5601 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5602 // will simply re-test the condition and re-park itself.
5603 // This provides particular benefit if the underlying platform does not
5604 // provide wait morphing.
5605 status = pthread_cond_signal(_cond);
5606 assert_status(status == 0, status, "cond_signal");
5607 }
5608 }
5609
5610
5611 // JSR166
5612 // -------------------------------------------------------
5613
5614 // The solaris and linux implementations of park/unpark are fairly
5615 // conservative for now, but can be improved. They currently use a
5616 // mutex/condvar pair, plus a a count.
5617 // Park decrements count if > 0, else does a condvar wait. Unpark
5618 // sets count to 1 and signals condvar. Only one thread ever waits
5619 // on the condvar. Contention seen when trying to park implies that someone
5620 // is unparking you, so don't wait. And spurious returns are fine, so there
5621 // is no need to track notifications.
5622
5623 // This code is common to linux and solaris and will be moved to a
5624 // common place in dolphin.
5625 //
5626 // The passed in time value is either a relative time in nanoseconds
5627 // or an absolute time in milliseconds. Either way it has to be unpacked
5628 // into suitable seconds and nanoseconds components and stored in the
5629 // given timespec structure.
5630 // Given time is a 64-bit value and the time_t used in the timespec is only
5631 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5632 // overflow if times way in the future are given. Further on Solaris versions
5633 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5634 // number of seconds, in abstime, is less than current_time + 100,000,000.
5635 // As it will be 28 years before "now + 100000000" will overflow we can
5636 // ignore overflow and just impose a hard-limit on seconds using the value
5637 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5638 // years from "now".
5639
5640 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5641 assert(time > 0, "convertTime");
5642 time_t max_secs = 0;
5643
5644 if (!os::supports_monotonic_clock() || isAbsolute) {
5645 struct timeval now;
5646 int status = gettimeofday(&now, NULL);
5647 assert(status == 0, "gettimeofday");
5648
5649 max_secs = now.tv_sec + MAX_SECS;
5650
5651 if (isAbsolute) {
5652 jlong secs = time / 1000;
5653 if (secs > max_secs) {
5654 absTime->tv_sec = max_secs;
5655 } else {
5656 absTime->tv_sec = secs;
5657 }
5658 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5659 } else {
5660 jlong secs = time / NANOSECS_PER_SEC;
5661 if (secs >= MAX_SECS) {
5662 absTime->tv_sec = max_secs;
5663 absTime->tv_nsec = 0;
5664 } else {
5665 absTime->tv_sec = now.tv_sec + secs;
5666 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5667 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5668 absTime->tv_nsec -= NANOSECS_PER_SEC;
5669 ++absTime->tv_sec; // note: this must be <= max_secs
5670 }
5671 }
5672 }
5673 } else {
5674 // must be relative using monotonic clock
5675 struct timespec now;
5676 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5677 assert_status(status == 0, status, "clock_gettime");
5678 max_secs = now.tv_sec + MAX_SECS;
5679 jlong secs = time / NANOSECS_PER_SEC;
5680 if (secs >= MAX_SECS) {
5681 absTime->tv_sec = max_secs;
5682 absTime->tv_nsec = 0;
5683 } else {
5684 absTime->tv_sec = now.tv_sec + secs;
5685 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5686 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5687 absTime->tv_nsec -= NANOSECS_PER_SEC;
5688 ++absTime->tv_sec; // note: this must be <= max_secs
5689 }
5690 }
5691 }
5692 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5693 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5694 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5695 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5696 }
5697
5698 void Parker::park(bool isAbsolute, jlong time) {
5699 // Ideally we'd do something useful while spinning, such
5700 // as calling unpackTime().
5701
5702 // Optional fast-path check:
5703 // Return immediately if a permit is available.
5704 // We depend on Atomic::xchg() having full barrier semantics
5705 // since we are doing a lock-free update to _counter.
5706 if (Atomic::xchg(0, &_counter) > 0) return;
5707
5708 Thread* thread = Thread::current();
5709 assert(thread->is_Java_thread(), "Must be JavaThread");
5710 JavaThread *jt = (JavaThread *)thread;
5711
5712 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5713 // Check interrupt before trying to wait
5714 if (Thread::is_interrupted(thread, false)) {
5715 return;
5716 }
5717
5718 // Next, demultiplex/decode time arguments
5719 timespec absTime;
5720 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5721 return;
5722 }
5723 if (time > 0) {
5724 unpackTime(&absTime, isAbsolute, time);
5725 }
5726
5727
5728 // Enter safepoint region
5729 // Beware of deadlocks such as 6317397.
5730 // The per-thread Parker:: mutex is a classic leaf-lock.
5731 // In particular a thread must never block on the Threads_lock while
5732 // holding the Parker:: mutex. If safepoints are pending both the
5733 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5734 ThreadBlockInVM tbivm(jt);
5735
5736 // Don't wait if cannot get lock since interference arises from
5737 // unblocking. Also. check interrupt before trying wait
5738 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5739 return;
5740 }
5741
5742 int status;
5743 if (_counter > 0) { // no wait needed
5744 _counter = 0;
5745 status = pthread_mutex_unlock(_mutex);
5746 assert(status == 0, "invariant");
5747 // Paranoia to ensure our locked and lock-free paths interact
5748 // correctly with each other and Java-level accesses.
5749 OrderAccess::fence();
5750 return;
5751 }
5752
5753 #ifdef ASSERT
5754 // Don't catch signals while blocked; let the running threads have the signals.
5755 // (This allows a debugger to break into the running thread.)
5756 sigset_t oldsigs;
5757 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5758 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5759 #endif
5760
5761 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5762 jt->set_suspend_equivalent();
5763 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5764
5765 assert(_cur_index == -1, "invariant");
5766 if (time == 0) {
5767 _cur_index = REL_INDEX; // arbitrary choice when not timed
5768 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5769 } else {
5770 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5771 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5772 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5773 pthread_cond_destroy(&_cond[_cur_index]);
5774 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5775 }
5776 }
5777 _cur_index = -1;
5778 assert_status(status == 0 || status == EINTR ||
5779 status == ETIME || status == ETIMEDOUT,
5780 status, "cond_timedwait");
5781
5782 #ifdef ASSERT
5783 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5784 #endif
5785
5786 _counter = 0;
5787 status = pthread_mutex_unlock(_mutex);
5788 assert_status(status == 0, status, "invariant");
5789 // Paranoia to ensure our locked and lock-free paths interact
5790 // correctly with each other and Java-level accesses.
5791 OrderAccess::fence();
5792
5793 // If externally suspended while waiting, re-suspend
5794 if (jt->handle_special_suspend_equivalent_condition()) {
5795 jt->java_suspend_self();
5796 }
5797 }
5798
5799 void Parker::unpark() {
5800 int status = pthread_mutex_lock(_mutex);
5801 assert(status == 0, "invariant");
5802 const int s = _counter;
5803 _counter = 1;
5804 if (s < 1) {
5805 // thread might be parked
5806 if (_cur_index != -1) {
5807 // thread is definitely parked
5808 if (WorkAroundNPTLTimedWaitHang) {
5809 status = pthread_cond_signal(&_cond[_cur_index]);
5810 assert(status == 0, "invariant");
5811 status = pthread_mutex_unlock(_mutex);
5812 assert(status == 0, "invariant");
5813 } else {
5814 status = pthread_mutex_unlock(_mutex);
5815 assert(status == 0, "invariant");
5816 status = pthread_cond_signal(&_cond[_cur_index]);
5817 assert(status == 0, "invariant");
5818 }
5819 } else {
5820 pthread_mutex_unlock(_mutex);
5821 assert(status == 0, "invariant");
5822 }
5823 } else {
5824 pthread_mutex_unlock(_mutex);
5825 assert(status == 0, "invariant");
5826 }
5827 }
5828
5829
5830 extern char** environ;
5831
5832 #ifndef __NR_fork
5833 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5834 #endif
5835
5836 #ifndef __NR_execve
5837 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5838 #endif
5839
5840 // Run the specified command in a separate process. Return its exit value,
5841 // or -1 on failure (e.g. can't fork a new process).
5842 // Unlike system(), this function can be called from signal handler. It
5843 // doesn't block SIGINT et al.
5844 int os::fork_and_exec(char* cmd) {
5845 const char * argv[4] = {"sh", "-c", cmd, NULL};
5846
5847 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5848 // pthread_atfork handlers and reset pthread library. All we need is a
5849 // separate process to execve. Make a direct syscall to fork process.
5850 // On IA64 there's no fork syscall, we have to use fork() and hope for
5851 // the best...
5852 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5853 IA64_ONLY(fork();)
5854
5855 if (pid < 0) {
5856 // fork failed
5857 return -1;
5858
5859 } else if (pid == 0) {
5860 // child process
5861
5862 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5863 // first to kill every thread on the thread list. Because this list is
5864 // not reset by fork() (see notes above), execve() will instead kill
5865 // every thread in the parent process. We know this is the only thread
5866 // in the new process, so make a system call directly.
5867 // IA64 should use normal execve() from glibc to match the glibc fork()
5868 // above.
5869 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5870 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5871
5872 // execve failed
5873 _exit(-1);
5874
5875 } else {
5876 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5877 // care about the actual exit code, for now.
5878
5879 int status;
5880
5881 // Wait for the child process to exit. This returns immediately if
5882 // the child has already exited. */
5883 while (waitpid(pid, &status, 0) < 0) {
5884 switch (errno) {
5885 case ECHILD: return 0;
5886 case EINTR: break;
5887 default: return -1;
5888 }
5889 }
5890
5891 if (WIFEXITED(status)) {
5892 // The child exited normally; get its exit code.
5893 return WEXITSTATUS(status);
5894 } else if (WIFSIGNALED(status)) {
5895 // The child exited because of a signal
5896 // The best value to return is 0x80 + signal number,
5897 // because that is what all Unix shells do, and because
5898 // it allows callers to distinguish between process exit and
5899 // process death by signal.
5900 return 0x80 + WTERMSIG(status);
5901 } else {
5902 // Unknown exit code; pass it through
5903 return status;
5904 }
5905 }
5906 }
5907
5908 // is_headless_jre()
5909 //
5910 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5911 // in order to report if we are running in a headless jre
5912 //
5913 // Since JDK8 xawt/libmawt.so was moved into the same directory
5914 // as libawt.so, and renamed libawt_xawt.so
5915 //
5916 bool os::is_headless_jre() {
5917 struct stat statbuf;
5918 char buf[MAXPATHLEN];
5919 char libmawtpath[MAXPATHLEN];
5920 const char *xawtstr = "/xawt/libmawt.so";
5921 const char *new_xawtstr = "/libawt_xawt.so";
5922 char *p;
5923
5924 // Get path to libjvm.so
5925 os::jvm_path(buf, sizeof(buf));
5926
5927 // Get rid of libjvm.so
5928 p = strrchr(buf, '/');
5929 if (p == NULL) {
5930 return false;
5931 } else {
5932 *p = '\0';
5933 }
5934
5935 // Get rid of client or server
5936 p = strrchr(buf, '/');
5937 if (p == NULL) {
5938 return false;
5939 } else {
5940 *p = '\0';
5941 }
5942
5943 // check xawt/libmawt.so
5944 strcpy(libmawtpath, buf);
5945 strcat(libmawtpath, xawtstr);
5946 if (::stat(libmawtpath, &statbuf) == 0) return false;
5947
5948 // check libawt_xawt.so
5949 strcpy(libmawtpath, buf);
5950 strcat(libmawtpath, new_xawtstr);
5951 if (::stat(libmawtpath, &statbuf) == 0) return false;
5952
5953 return true;
5954 }
5955
5956 // Get the default path to the core file
5957 // Returns the length of the string
5958 int os::get_core_path(char* buffer, size_t bufferSize) {
5959 /*
5960 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5961 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5962 */
5963 const int core_pattern_len = 129;
5964 char core_pattern[core_pattern_len] = {0};
5965
5966 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5967 if (core_pattern_file != -1) {
5968 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5969 ::close(core_pattern_file);
5970
5971 if (ret > 0) {
5972 char *last_char = core_pattern + strlen(core_pattern) - 1;
5973
5974 if (*last_char == '\n') {
5975 *last_char = '\0';
5976 }
5977 }
5978 }
5979
5980 if (strlen(core_pattern) == 0) {
5981 return -1;
5982 }
5983
5984 char *pid_pos = strstr(core_pattern, "%p");
5985 int written;
5986
5987 if (core_pattern[0] == '/') {
5988 written = jio_snprintf(buffer, bufferSize, core_pattern);
5989 } else {
5990 char cwd[PATH_MAX];
5991
5992 const char* p = get_current_directory(cwd, PATH_MAX);
5993 if (p == NULL) {
5994 return -1;
5995 }
5996
5997 if (core_pattern[0] == '|') {
5998 written = jio_snprintf(buffer, bufferSize,
5999 "\"%s\" (or dumping to %s/core.%d)",
6000 &core_pattern[1], p, current_process_id());
6001 } else {
6002 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6003 }
6004 }
6005
6006 if (written < 0) {
6007 return -1;
6008 }
6009
6010 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6011 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6012
6013 if (core_uses_pid_file != -1) {
6014 char core_uses_pid = 0;
6015 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6016 ::close(core_uses_pid_file);
6017
6018 if (core_uses_pid == '1') {
6019 jio_snprintf(buffer + written, bufferSize - written,
6020 ".%d", current_process_id());
6021 }
6022 }
6023 }
6024
6025 return strlen(buffer);
6026 }
6027
6028 /////////////// Unit tests ///////////////
6029
6030 #ifndef PRODUCT
6031
6032 #define test_log(...) \
6033 do { \
6034 if (VerboseInternalVMTests) { \
6035 tty->print_cr(__VA_ARGS__); \
6036 tty->flush(); \
6037 } \
6038 } while (false)
6039
6040 class TestReserveMemorySpecial : AllStatic {
6041 public:
6042 static void small_page_write(void* addr, size_t size) {
6043 size_t page_size = os::vm_page_size();
6044
6045 char* end = (char*)addr + size;
6046 for (char* p = (char*)addr; p < end; p += page_size) {
6047 *p = 1;
6048 }
6049 }
6050
6051 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6052 if (!UseHugeTLBFS) {
6053 return;
6054 }
6055
6056 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6057
6058 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6059
6060 if (addr != NULL) {
6061 small_page_write(addr, size);
6062
6063 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6064 }
6065 }
6066
6067 static void test_reserve_memory_special_huge_tlbfs_only() {
6068 if (!UseHugeTLBFS) {
6069 return;
6070 }
6071
6072 size_t lp = os::large_page_size();
6073
6074 for (size_t size = lp; size <= lp * 10; size += lp) {
6075 test_reserve_memory_special_huge_tlbfs_only(size);
6076 }
6077 }
6078
6079 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6080 if (!UseHugeTLBFS) {
6081 return;
6082 }
6083
6084 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6085 size, alignment);
6086
6087 assert(size >= os::large_page_size(), "Incorrect input to test");
6088
6089 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6090
6091 if (addr != NULL) {
6092 small_page_write(addr, size);
6093
6094 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6095 }
6096 }
6097
6098 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6099 size_t lp = os::large_page_size();
6100 size_t ag = os::vm_allocation_granularity();
6101
6102 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6103 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6104 }
6105 }
6106
6107 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6108 size_t lp = os::large_page_size();
6109 size_t ag = os::vm_allocation_granularity();
6110
6111 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6112 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6113 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6114 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6115 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6116 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6117 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6118 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6119 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6120 }
6121
6122 static void test_reserve_memory_special_huge_tlbfs() {
6123 if (!UseHugeTLBFS) {
6124 return;
6125 }
6126
6127 test_reserve_memory_special_huge_tlbfs_only();
6128 test_reserve_memory_special_huge_tlbfs_mixed();
6129 }
6130
6131 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6132 if (!UseSHM) {
6133 return;
6134 }
6135
6136 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6137
6138 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6139
6140 if (addr != NULL) {
6141 assert(is_ptr_aligned(addr, alignment), "Check");
6142 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6143
6144 small_page_write(addr, size);
6145
6146 os::Linux::release_memory_special_shm(addr, size);
6147 }
6148 }
6149
6150 static void test_reserve_memory_special_shm() {
6151 size_t lp = os::large_page_size();
6152 size_t ag = os::vm_allocation_granularity();
6153
6154 for (size_t size = ag; size < lp * 3; size += ag) {
6155 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6156 test_reserve_memory_special_shm(size, alignment);
6157 }
6158 }
6159 }
6160
6161 static void test() {
6162 test_reserve_memory_special_huge_tlbfs();
6163 test_reserve_memory_special_shm();
6164 }
6165 };
6166
6167 void TestReserveMemorySpecial_test() {
6168 TestReserveMemorySpecial::test();
6169 }
6170
6171 #endif