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 #else
202 #ifdef __i386__
203 #define SYS_gettid 224
204 #else
205 #ifdef __amd64__
206 #define SYS_gettid 186
207 #else
208 #ifdef __sparc__
209 #define SYS_gettid 143
210 #else
211 #error define gettid for the arch
212 #endif
213 #endif
214 #endif
215 #endif
216 #endif
217
218 // Cpu architecture string
219 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
220
221
222 // pid_t gettid()
223 //
224 // Returns the kernel thread id of the currently running thread. Kernel
225 // thread id is used to access /proc.
226 //
227 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
228 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
229 //
230 pid_t os::Linux::gettid() {
231 int rslt = syscall(SYS_gettid);
232 if (rslt == -1) {
233 // old kernel, no NPTL support
234 return getpid();
235 } else {
236 return (pid_t)rslt;
237 }
238 }
239
240 // Most versions of linux have a bug where the number of processors are
241 // determined by looking at the /proc file system. In a chroot environment,
242 // the system call returns 1. This causes the VM to act as if it is
243 // a single processor and elide locking (see is_MP() call).
244 static bool unsafe_chroot_detected = false;
245 static const char *unstable_chroot_error = "/proc file system not found.\n"
246 "Java may be unstable running multithreaded in a chroot "
247 "environment on Linux when /proc filesystem is not mounted.";
248
249 void os::Linux::initialize_system_info() {
250 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
251 if (processor_count() == 1) {
252 pid_t pid = os::Linux::gettid();
253 char fname[32];
254 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
255 FILE *fp = fopen(fname, "r");
256 if (fp == NULL) {
257 unsafe_chroot_detected = true;
258 } else {
259 fclose(fp);
260 }
261 }
262 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
263 assert(processor_count() > 0, "linux error");
264 }
265
266 void os::init_system_properties_values() {
267 // The next steps are taken in the product version:
268 //
269 // Obtain the JAVA_HOME value from the location of libjvm.so.
270 // This library should be located at:
271 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
272 //
273 // If "/jre/lib/" appears at the right place in the path, then we
274 // assume libjvm.so is installed in a JDK and we use this path.
275 //
276 // Otherwise exit with message: "Could not create the Java virtual machine."
277 //
278 // The following extra steps are taken in the debugging version:
279 //
280 // If "/jre/lib/" does NOT appear at the right place in the path
281 // instead of exit check for $JAVA_HOME environment variable.
282 //
283 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
284 // then we append a fake suffix "hotspot/libjvm.so" to this path so
285 // it looks like libjvm.so is installed there
286 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
287 //
288 // Otherwise exit.
289 //
290 // Important note: if the location of libjvm.so changes this
291 // code needs to be changed accordingly.
292
293 // See ld(1):
294 // The linker uses the following search paths to locate required
295 // shared libraries:
296 // 1: ...
297 // ...
298 // 7: The default directories, normally /lib and /usr/lib.
299 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
300 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
301 #else
302 #define DEFAULT_LIBPATH "/lib:/usr/lib"
303 #endif
304
305 // Base path of extensions installed on the system.
306 #define SYS_EXT_DIR "/usr/java/packages"
307 #define EXTENSIONS_DIR "/lib/ext"
308
309 // Buffer that fits several sprintfs.
310 // Note that the space for the colon and the trailing null are provided
311 // by the nulls included by the sizeof operator.
312 const size_t bufsize =
313 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
314 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
315 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
316
317 // sysclasspath, java_home, dll_dir
318 {
319 char *pslash;
320 os::jvm_path(buf, bufsize);
321
322 // Found the full path to libjvm.so.
323 // Now cut the path to <java_home>/jre if we can.
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /libjvm.so.
327 }
328 pslash = strrchr(buf, '/');
329 if (pslash != NULL) {
330 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
331 }
332 Arguments::set_dll_dir(buf);
333
334 if (pslash != NULL) {
335 pslash = strrchr(buf, '/');
336 if (pslash != NULL) {
337 *pslash = '\0'; // Get rid of /<arch>.
338 pslash = strrchr(buf, '/');
339 if (pslash != NULL) {
340 *pslash = '\0'; // Get rid of /lib.
341 }
342 }
343 }
344 Arguments::set_java_home(buf);
345 set_boot_path('/', ':');
346 }
347
348 // Where to look for native libraries.
349 //
350 // Note: Due to a legacy implementation, most of the library path
351 // is set in the launcher. This was to accomodate linking restrictions
352 // on legacy Linux implementations (which are no longer supported).
353 // Eventually, all the library path setting will be done here.
354 //
355 // However, to prevent the proliferation of improperly built native
356 // libraries, the new path component /usr/java/packages is added here.
357 // Eventually, all the library path setting will be done here.
358 {
359 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
360 // should always exist (until the legacy problem cited above is
361 // addressed).
362 const char *v = ::getenv("LD_LIBRARY_PATH");
363 const char *v_colon = ":";
364 if (v == NULL) { v = ""; v_colon = ""; }
365 // That's +1 for the colon and +1 for the trailing '\0'.
366 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
367 strlen(v) + 1 +
368 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
369 mtInternal);
370 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
371 Arguments::set_library_path(ld_library_path);
372 FREE_C_HEAP_ARRAY(char, ld_library_path);
373 }
374
375 // Extensions directories.
376 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
377 Arguments::set_ext_dirs(buf);
378
379 FREE_C_HEAP_ARRAY(char, buf);
380
381 #undef DEFAULT_LIBPATH
382 #undef SYS_EXT_DIR
383 #undef EXTENSIONS_DIR
384 }
385
386 ////////////////////////////////////////////////////////////////////////////////
387 // breakpoint support
388
389 void os::breakpoint() {
390 BREAKPOINT;
391 }
392
393 extern "C" void breakpoint() {
394 // use debugger to set breakpoint here
395 }
396
397 ////////////////////////////////////////////////////////////////////////////////
398 // signal support
399
400 debug_only(static bool signal_sets_initialized = false);
401 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
402
403 bool os::Linux::is_sig_ignored(int sig) {
404 struct sigaction oact;
405 sigaction(sig, (struct sigaction*)NULL, &oact);
406 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
407 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
408 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
409 return true;
410 } else {
411 return false;
412 }
413 }
414
415 void os::Linux::signal_sets_init() {
416 // Should also have an assertion stating we are still single-threaded.
417 assert(!signal_sets_initialized, "Already initialized");
418 // Fill in signals that are necessarily unblocked for all threads in
419 // the VM. Currently, we unblock the following signals:
420 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
421 // by -Xrs (=ReduceSignalUsage));
422 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
423 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
424 // the dispositions or masks wrt these signals.
425 // Programs embedding the VM that want to use the above signals for their
426 // own purposes must, at this time, use the "-Xrs" option to prevent
427 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
428 // (See bug 4345157, and other related bugs).
429 // In reality, though, unblocking these signals is really a nop, since
430 // these signals are not blocked by default.
431 sigemptyset(&unblocked_sigs);
432 sigemptyset(&allowdebug_blocked_sigs);
433 sigaddset(&unblocked_sigs, SIGILL);
434 sigaddset(&unblocked_sigs, SIGSEGV);
435 sigaddset(&unblocked_sigs, SIGBUS);
436 sigaddset(&unblocked_sigs, SIGFPE);
437 #if defined(PPC64)
438 sigaddset(&unblocked_sigs, SIGTRAP);
439 #endif
440 sigaddset(&unblocked_sigs, SR_signum);
441
442 if (!ReduceSignalUsage) {
443 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
444 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
445 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
446 }
447 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
448 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
449 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
450 }
451 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
452 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
453 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
454 }
455 }
456 // Fill in signals that are blocked by all but the VM thread.
457 sigemptyset(&vm_sigs);
458 if (!ReduceSignalUsage) {
459 sigaddset(&vm_sigs, BREAK_SIGNAL);
460 }
461 debug_only(signal_sets_initialized = true);
462
463 }
464
465 // These are signals that are unblocked while a thread is running Java.
466 // (For some reason, they get blocked by default.)
467 sigset_t* os::Linux::unblocked_signals() {
468 assert(signal_sets_initialized, "Not initialized");
469 return &unblocked_sigs;
470 }
471
472 // These are the signals that are blocked while a (non-VM) thread is
473 // running Java. Only the VM thread handles these signals.
474 sigset_t* os::Linux::vm_signals() {
475 assert(signal_sets_initialized, "Not initialized");
476 return &vm_sigs;
477 }
478
479 // These are signals that are blocked during cond_wait to allow debugger in
480 sigset_t* os::Linux::allowdebug_blocked_signals() {
481 assert(signal_sets_initialized, "Not initialized");
482 return &allowdebug_blocked_sigs;
483 }
484
485 void os::Linux::hotspot_sigmask(Thread* thread) {
486
487 //Save caller's signal mask before setting VM signal mask
488 sigset_t caller_sigmask;
489 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
490
491 OSThread* osthread = thread->osthread();
492 osthread->set_caller_sigmask(caller_sigmask);
493
494 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
495
496 if (!ReduceSignalUsage) {
497 if (thread->is_VM_thread()) {
498 // Only the VM thread handles BREAK_SIGNAL ...
499 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
500 } else {
501 // ... all other threads block BREAK_SIGNAL
502 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
503 }
504 }
505 }
506
507 //////////////////////////////////////////////////////////////////////////////
508 // detecting pthread library
509
510 void os::Linux::libpthread_init() {
511 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
512 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
513 // generic name for earlier versions.
514 // Define macros here so we can build HotSpot on old systems.
515 #ifndef _CS_GNU_LIBC_VERSION
516 #define _CS_GNU_LIBC_VERSION 2
517 #endif
518 #ifndef _CS_GNU_LIBPTHREAD_VERSION
519 #define _CS_GNU_LIBPTHREAD_VERSION 3
520 #endif
521
522 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
523 if (n > 0) {
524 char *str = (char *)malloc(n, mtInternal);
525 confstr(_CS_GNU_LIBC_VERSION, str, n);
526 os::Linux::set_glibc_version(str);
527 } else {
528 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
529 static char _gnu_libc_version[32];
530 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
531 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
532 os::Linux::set_glibc_version(_gnu_libc_version);
533 }
534
535 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
536 if (n > 0) {
537 char *str = (char *)malloc(n, mtInternal);
538 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
539 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
540 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
541 // is the case. LinuxThreads has a hard limit on max number of threads.
542 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
543 // On the other hand, NPTL does not have such a limit, sysconf()
544 // will return -1 and errno is not changed. Check if it is really NPTL.
545 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
546 strstr(str, "NPTL") &&
547 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
548 free(str);
549 os::Linux::set_libpthread_version("linuxthreads");
550 } else {
551 os::Linux::set_libpthread_version(str);
552 }
553 } else {
554 // glibc before 2.3.2 only has LinuxThreads.
555 os::Linux::set_libpthread_version("linuxthreads");
556 }
557
558 if (strstr(libpthread_version(), "NPTL")) {
559 os::Linux::set_is_NPTL();
560 } else {
561 os::Linux::set_is_LinuxThreads();
562 }
563
564 // LinuxThreads have two flavors: floating-stack mode, which allows variable
565 // stack size; and fixed-stack mode. NPTL is always floating-stack.
566 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
567 os::Linux::set_is_floating_stack();
568 }
569 }
570
571 /////////////////////////////////////////////////////////////////////////////
572 // thread stack
573
574 // Force Linux kernel to expand current thread stack. If "bottom" is close
575 // to the stack guard, caller should block all signals.
576 //
577 // MAP_GROWSDOWN:
578 // A special mmap() flag that is used to implement thread stacks. It tells
579 // kernel that the memory region should extend downwards when needed. This
580 // allows early versions of LinuxThreads to only mmap the first few pages
581 // when creating a new thread. Linux kernel will automatically expand thread
582 // stack as needed (on page faults).
583 //
584 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
585 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
586 // region, it's hard to tell if the fault is due to a legitimate stack
587 // access or because of reading/writing non-exist memory (e.g. buffer
588 // overrun). As a rule, if the fault happens below current stack pointer,
589 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
590 // application (see Linux kernel fault.c).
591 //
592 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
593 // stack overflow detection.
594 //
595 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
596 // not use this flag. However, the stack of initial thread is not created
597 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
598 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
599 // and then attach the thread to JVM.
600 //
601 // To get around the problem and allow stack banging on Linux, we need to
602 // manually expand thread stack after receiving the SIGSEGV.
603 //
604 // There are two ways to expand thread stack to address "bottom", we used
605 // both of them in JVM before 1.5:
606 // 1. adjust stack pointer first so that it is below "bottom", and then
607 // touch "bottom"
608 // 2. mmap() the page in question
609 //
610 // Now alternate signal stack is gone, it's harder to use 2. For instance,
611 // if current sp is already near the lower end of page 101, and we need to
612 // call mmap() to map page 100, it is possible that part of the mmap() frame
613 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
614 // That will destroy the mmap() frame and cause VM to crash.
615 //
616 // The following code works by adjusting sp first, then accessing the "bottom"
617 // page to force a page fault. Linux kernel will then automatically expand the
618 // stack mapping.
619 //
620 // _expand_stack_to() assumes its frame size is less than page size, which
621 // should always be true if the function is not inlined.
622
623 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
624 #define NOINLINE
625 #else
626 #define NOINLINE __attribute__ ((noinline))
627 #endif
628
629 static void _expand_stack_to(address bottom) NOINLINE;
630
631 static void _expand_stack_to(address bottom) {
632 address sp;
633 size_t size;
634 volatile char *p;
635
636 // Adjust bottom to point to the largest address within the same page, it
637 // gives us a one-page buffer if alloca() allocates slightly more memory.
638 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
639 bottom += os::Linux::page_size() - 1;
640
641 // sp might be slightly above current stack pointer; if that's the case, we
642 // will alloca() a little more space than necessary, which is OK. Don't use
643 // os::current_stack_pointer(), as its result can be slightly below current
644 // stack pointer, causing us to not alloca enough to reach "bottom".
645 sp = (address)&sp;
646
647 if (sp > bottom) {
648 size = sp - bottom;
649 p = (volatile char *)alloca(size);
650 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
651 p[0] = '\0';
652 }
653 }
654
655 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
656 assert(t!=NULL, "just checking");
657 assert(t->osthread()->expanding_stack(), "expand should be set");
658 assert(t->stack_base() != NULL, "stack_base was not initialized");
659
660 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
661 sigset_t mask_all, old_sigset;
662 sigfillset(&mask_all);
663 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
664 _expand_stack_to(addr);
665 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
666 return true;
667 }
668 return false;
669 }
670
671 //////////////////////////////////////////////////////////////////////////////
672 // create new thread
673
674 static address highest_vm_reserved_address();
675
676 // check if it's safe to start a new thread
677 static bool _thread_safety_check(Thread* thread) {
678 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
679 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
680 // Heap is mmap'ed at lower end of memory space. Thread stacks are
681 // allocated (MAP_FIXED) from high address space. Every thread stack
682 // occupies a fixed size slot (usually 2Mbytes, but user can change
683 // it to other values if they rebuild LinuxThreads).
684 //
685 // Problem with MAP_FIXED is that mmap() can still succeed even part of
686 // the memory region has already been mmap'ed. That means if we have too
687 // many threads and/or very large heap, eventually thread stack will
688 // collide with heap.
689 //
690 // Here we try to prevent heap/stack collision by comparing current
691 // stack bottom with the highest address that has been mmap'ed by JVM
692 // plus a safety margin for memory maps created by native code.
693 //
694 // This feature can be disabled by setting ThreadSafetyMargin to 0
695 //
696 if (ThreadSafetyMargin > 0) {
697 address stack_bottom = os::current_stack_base() - os::current_stack_size();
698
699 // not safe if our stack extends below the safety margin
700 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
701 } else {
702 return true;
703 }
704 } else {
705 // Floating stack LinuxThreads or NPTL:
706 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
707 // there's not enough space left, pthread_create() will fail. If we come
708 // here, that means enough space has been reserved for stack.
709 return true;
710 }
711 }
712
713 // Thread start routine for all newly created threads
714 static void *java_start(Thread *thread) {
715 // Try to randomize the cache line index of hot stack frames.
716 // This helps when threads of the same stack traces evict each other's
717 // cache lines. The threads can be either from the same JVM instance, or
718 // from different JVM instances. The benefit is especially true for
719 // processors with hyperthreading technology.
720 static int counter = 0;
721 int pid = os::current_process_id();
722 alloca(((pid ^ counter++) & 7) * 128);
723
724 ThreadLocalStorage::set_thread(thread);
725
726 OSThread* osthread = thread->osthread();
727 Monitor* sync = osthread->startThread_lock();
728
729 // non floating stack LinuxThreads needs extra check, see above
730 if (!_thread_safety_check(thread)) {
731 // notify parent thread
732 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
733 osthread->set_state(ZOMBIE);
734 sync->notify_all();
735 return NULL;
736 }
737
738 // thread_id is kernel thread id (similar to Solaris LWP id)
739 osthread->set_thread_id(os::Linux::gettid());
740
741 if (UseNUMA) {
742 int lgrp_id = os::numa_get_group_id();
743 if (lgrp_id != -1) {
744 thread->set_lgrp_id(lgrp_id);
745 }
746 }
747 // initialize signal mask for this thread
748 os::Linux::hotspot_sigmask(thread);
749
750 // initialize floating point control register
751 os::Linux::init_thread_fpu_state();
752
753 // handshaking with parent thread
754 {
755 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
756
757 // notify parent thread
758 osthread->set_state(INITIALIZED);
759 sync->notify_all();
760
761 // wait until os::start_thread()
762 while (osthread->get_state() == INITIALIZED) {
763 sync->wait(Mutex::_no_safepoint_check_flag);
764 }
765 }
766
767 // call one more level start routine
768 thread->run();
769
770 return 0;
771 }
772
773 bool os::create_thread(Thread* thread, ThreadType thr_type,
774 size_t stack_size) {
775 assert(thread->osthread() == NULL, "caller responsible");
776
777 // Allocate the OSThread object
778 OSThread* osthread = new OSThread(NULL, NULL);
779 if (osthread == NULL) {
780 return false;
781 }
782
783 // set the correct thread state
784 osthread->set_thread_type(thr_type);
785
786 // Initial state is ALLOCATED but not INITIALIZED
787 osthread->set_state(ALLOCATED);
788
789 thread->set_osthread(osthread);
790
791 // init thread attributes
792 pthread_attr_t attr;
793 pthread_attr_init(&attr);
794 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
795
796 // stack size
797 if (os::Linux::supports_variable_stack_size()) {
798 // calculate stack size if it's not specified by caller
799 if (stack_size == 0) {
800 stack_size = os::Linux::default_stack_size(thr_type);
801
802 switch (thr_type) {
803 case os::java_thread:
804 // Java threads use ThreadStackSize which default value can be
805 // changed with the flag -Xss
806 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
807 stack_size = JavaThread::stack_size_at_create();
808 break;
809 case os::compiler_thread:
810 if (CompilerThreadStackSize > 0) {
811 stack_size = (size_t)(CompilerThreadStackSize * K);
812 break;
813 } // else fall through:
814 // use VMThreadStackSize if CompilerThreadStackSize is not defined
815 case os::vm_thread:
816 case os::pgc_thread:
817 case os::cgc_thread:
818 case os::watcher_thread:
819 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
820 break;
821 }
822 }
823
824 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
825 pthread_attr_setstacksize(&attr, stack_size);
826 } else {
827 // let pthread_create() pick the default value.
828 }
829
830 // glibc guard page
831 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
832
833 ThreadState state;
834
835 {
836 // Serialize thread creation if we are running with fixed stack LinuxThreads
837 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
838 if (lock) {
839 os::Linux::createThread_lock()->lock_without_safepoint_check();
840 }
841
842 pthread_t tid;
843 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
844
845 pthread_attr_destroy(&attr);
846
847 if (ret != 0) {
848 if (PrintMiscellaneous && (Verbose || WizardMode)) {
849 perror("pthread_create()");
850 }
851 // Need to clean up stuff we've allocated so far
852 thread->set_osthread(NULL);
853 delete osthread;
854 if (lock) os::Linux::createThread_lock()->unlock();
855 return false;
856 }
857
858 // Store pthread info into the OSThread
859 osthread->set_pthread_id(tid);
860
861 // Wait until child thread is either initialized or aborted
862 {
863 Monitor* sync_with_child = osthread->startThread_lock();
864 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
865 while ((state = osthread->get_state()) == ALLOCATED) {
866 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
867 }
868 }
869
870 if (lock) {
871 os::Linux::createThread_lock()->unlock();
872 }
873 }
874
875 // Aborted due to thread limit being reached
876 if (state == ZOMBIE) {
877 thread->set_osthread(NULL);
878 delete osthread;
879 return false;
880 }
881
882 // The thread is returned suspended (in state INITIALIZED),
883 // and is started higher up in the call chain
884 assert(state == INITIALIZED, "race condition");
885 return true;
886 }
887
888 /////////////////////////////////////////////////////////////////////////////
889 // attach existing thread
890
891 // bootstrap the main thread
892 bool os::create_main_thread(JavaThread* thread) {
893 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
894 return create_attached_thread(thread);
895 }
896
897 bool os::create_attached_thread(JavaThread* thread) {
898 #ifdef ASSERT
899 thread->verify_not_published();
900 #endif
901
902 // Allocate the OSThread object
903 OSThread* osthread = new OSThread(NULL, NULL);
904
905 if (osthread == NULL) {
906 return false;
907 }
908
909 // Store pthread info into the OSThread
910 osthread->set_thread_id(os::Linux::gettid());
911 osthread->set_pthread_id(::pthread_self());
912
913 // initialize floating point control register
914 os::Linux::init_thread_fpu_state();
915
916 // Initial thread state is RUNNABLE
917 osthread->set_state(RUNNABLE);
918
919 thread->set_osthread(osthread);
920
921 if (UseNUMA) {
922 int lgrp_id = os::numa_get_group_id();
923 if (lgrp_id != -1) {
924 thread->set_lgrp_id(lgrp_id);
925 }
926 }
927
928 if (os::Linux::is_initial_thread()) {
929 // If current thread is initial thread, its stack is mapped on demand,
930 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
931 // the entire stack region to avoid SEGV in stack banging.
932 // It is also useful to get around the heap-stack-gap problem on SuSE
933 // kernel (see 4821821 for details). We first expand stack to the top
934 // of yellow zone, then enable stack yellow zone (order is significant,
935 // enabling yellow zone first will crash JVM on SuSE Linux), so there
936 // is no gap between the last two virtual memory regions.
937
938 JavaThread *jt = (JavaThread *)thread;
939 address addr = jt->stack_yellow_zone_base();
940 assert(addr != NULL, "initialization problem?");
941 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
942
943 osthread->set_expanding_stack();
944 os::Linux::manually_expand_stack(jt, addr);
945 osthread->clear_expanding_stack();
946 }
947
948 // initialize signal mask for this thread
949 // and save the caller's signal mask
950 os::Linux::hotspot_sigmask(thread);
951
952 return true;
953 }
954
955 void os::pd_start_thread(Thread* thread) {
956 OSThread * osthread = thread->osthread();
957 assert(osthread->get_state() != INITIALIZED, "just checking");
958 Monitor* sync_with_child = osthread->startThread_lock();
959 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
960 sync_with_child->notify();
961 }
962
963 // Free Linux resources related to the OSThread
964 void os::free_thread(OSThread* osthread) {
965 assert(osthread != NULL, "osthread not set");
966
967 if (Thread::current()->osthread() == osthread) {
968 // Restore caller's signal mask
969 sigset_t sigmask = osthread->caller_sigmask();
970 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
971 }
972
973 delete osthread;
974 }
975
976 //////////////////////////////////////////////////////////////////////////////
977 // thread local storage
978
979 // Restore the thread pointer if the destructor is called. This is in case
980 // someone from JNI code sets up a destructor with pthread_key_create to run
981 // detachCurrentThread on thread death. Unless we restore the thread pointer we
982 // will hang or crash. When detachCurrentThread is called the key will be set
983 // to null and we will not be called again. If detachCurrentThread is never
984 // called we could loop forever depending on the pthread implementation.
985 static void restore_thread_pointer(void* p) {
986 Thread* thread = (Thread*) p;
987 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
988 }
989
990 int os::allocate_thread_local_storage() {
991 pthread_key_t key;
992 int rslt = pthread_key_create(&key, restore_thread_pointer);
993 assert(rslt == 0, "cannot allocate thread local storage");
994 return (int)key;
995 }
996
997 // Note: This is currently not used by VM, as we don't destroy TLS key
998 // on VM exit.
999 void os::free_thread_local_storage(int index) {
1000 int rslt = pthread_key_delete((pthread_key_t)index);
1001 assert(rslt == 0, "invalid index");
1002 }
1003
1004 void os::thread_local_storage_at_put(int index, void* value) {
1005 int rslt = pthread_setspecific((pthread_key_t)index, value);
1006 assert(rslt == 0, "pthread_setspecific failed");
1007 }
1008
1009 extern "C" Thread* get_thread() {
1010 return ThreadLocalStorage::thread();
1011 }
1012
1013 //////////////////////////////////////////////////////////////////////////////
1014 // initial thread
1015
1016 // Check if current thread is the initial thread, similar to Solaris thr_main.
1017 bool os::Linux::is_initial_thread(void) {
1018 char dummy;
1019 // If called before init complete, thread stack bottom will be null.
1020 // Can be called if fatal error occurs before initialization.
1021 if (initial_thread_stack_bottom() == NULL) return false;
1022 assert(initial_thread_stack_bottom() != NULL &&
1023 initial_thread_stack_size() != 0,
1024 "os::init did not locate initial thread's stack region");
1025 if ((address)&dummy >= initial_thread_stack_bottom() &&
1026 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1027 return true;
1028 } else {
1029 return false;
1030 }
1031 }
1032
1033 // Find the virtual memory area that contains addr
1034 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1035 FILE *fp = fopen("/proc/self/maps", "r");
1036 if (fp) {
1037 address low, high;
1038 while (!feof(fp)) {
1039 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1040 if (low <= addr && addr < high) {
1041 if (vma_low) *vma_low = low;
1042 if (vma_high) *vma_high = high;
1043 fclose(fp);
1044 return true;
1045 }
1046 }
1047 for (;;) {
1048 int ch = fgetc(fp);
1049 if (ch == EOF || ch == (int)'\n') break;
1050 }
1051 }
1052 fclose(fp);
1053 }
1054 return false;
1055 }
1056
1057 // Locate initial thread stack. This special handling of initial thread stack
1058 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1059 // bogus value for initial thread.
1060 void os::Linux::capture_initial_stack(size_t max_size) {
1061 // stack size is the easy part, get it from RLIMIT_STACK
1062 size_t stack_size;
1063 struct rlimit rlim;
1064 getrlimit(RLIMIT_STACK, &rlim);
1065 stack_size = rlim.rlim_cur;
1066
1067 // 6308388: a bug in ld.so will relocate its own .data section to the
1068 // lower end of primordial stack; reduce ulimit -s value a little bit
1069 // so we won't install guard page on ld.so's data section.
1070 stack_size -= 2 * page_size();
1071
1072 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1073 // 7.1, in both cases we will get 2G in return value.
1074 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1075 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1076 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1077 // in case other parts in glibc still assumes 2M max stack size.
1078 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1079 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1080 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1081 stack_size = 2 * K * K IA64_ONLY(*2);
1082 }
1083 // Try to figure out where the stack base (top) is. This is harder.
1084 //
1085 // When an application is started, glibc saves the initial stack pointer in
1086 // a global variable "__libc_stack_end", which is then used by system
1087 // libraries. __libc_stack_end should be pretty close to stack top. The
1088 // variable is available since the very early days. However, because it is
1089 // a private interface, it could disappear in the future.
1090 //
1091 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1092 // to __libc_stack_end, it is very close to stack top, but isn't the real
1093 // stack top. Note that /proc may not exist if VM is running as a chroot
1094 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1095 // /proc/<pid>/stat could change in the future (though unlikely).
1096 //
1097 // We try __libc_stack_end first. If that doesn't work, look for
1098 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1099 // as a hint, which should work well in most cases.
1100
1101 uintptr_t stack_start;
1102
1103 // try __libc_stack_end first
1104 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1105 if (p && *p) {
1106 stack_start = *p;
1107 } else {
1108 // see if we can get the start_stack field from /proc/self/stat
1109 FILE *fp;
1110 int pid;
1111 char state;
1112 int ppid;
1113 int pgrp;
1114 int session;
1115 int nr;
1116 int tpgrp;
1117 unsigned long flags;
1118 unsigned long minflt;
1119 unsigned long cminflt;
1120 unsigned long majflt;
1121 unsigned long cmajflt;
1122 unsigned long utime;
1123 unsigned long stime;
1124 long cutime;
1125 long cstime;
1126 long prio;
1127 long nice;
1128 long junk;
1129 long it_real;
1130 uintptr_t start;
1131 uintptr_t vsize;
1132 intptr_t rss;
1133 uintptr_t rsslim;
1134 uintptr_t scodes;
1135 uintptr_t ecode;
1136 int i;
1137
1138 // Figure what the primordial thread stack base is. Code is inspired
1139 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1140 // followed by command name surrounded by parentheses, state, etc.
1141 char stat[2048];
1142 int statlen;
1143
1144 fp = fopen("/proc/self/stat", "r");
1145 if (fp) {
1146 statlen = fread(stat, 1, 2047, fp);
1147 stat[statlen] = '\0';
1148 fclose(fp);
1149
1150 // Skip pid and the command string. Note that we could be dealing with
1151 // weird command names, e.g. user could decide to rename java launcher
1152 // to "java 1.4.2 :)", then the stat file would look like
1153 // 1234 (java 1.4.2 :)) R ... ...
1154 // We don't really need to know the command string, just find the last
1155 // occurrence of ")" and then start parsing from there. See bug 4726580.
1156 char * s = strrchr(stat, ')');
1157
1158 i = 0;
1159 if (s) {
1160 // Skip blank chars
1161 do { s++; } while (s && isspace(*s));
1162
1163 #define _UFM UINTX_FORMAT
1164 #define _DFM INTX_FORMAT
1165
1166 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1167 // 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
1168 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,
1169 &state, // 3 %c
1170 &ppid, // 4 %d
1171 &pgrp, // 5 %d
1172 &session, // 6 %d
1173 &nr, // 7 %d
1174 &tpgrp, // 8 %d
1175 &flags, // 9 %lu
1176 &minflt, // 10 %lu
1177 &cminflt, // 11 %lu
1178 &majflt, // 12 %lu
1179 &cmajflt, // 13 %lu
1180 &utime, // 14 %lu
1181 &stime, // 15 %lu
1182 &cutime, // 16 %ld
1183 &cstime, // 17 %ld
1184 &prio, // 18 %ld
1185 &nice, // 19 %ld
1186 &junk, // 20 %ld
1187 &it_real, // 21 %ld
1188 &start, // 22 UINTX_FORMAT
1189 &vsize, // 23 UINTX_FORMAT
1190 &rss, // 24 INTX_FORMAT
1191 &rsslim, // 25 UINTX_FORMAT
1192 &scodes, // 26 UINTX_FORMAT
1193 &ecode, // 27 UINTX_FORMAT
1194 &stack_start); // 28 UINTX_FORMAT
1195 }
1196
1197 #undef _UFM
1198 #undef _DFM
1199
1200 if (i != 28 - 2) {
1201 assert(false, "Bad conversion from /proc/self/stat");
1202 // product mode - assume we are the initial thread, good luck in the
1203 // embedded case.
1204 warning("Can't detect initial thread stack location - bad conversion");
1205 stack_start = (uintptr_t) &rlim;
1206 }
1207 } else {
1208 // For some reason we can't open /proc/self/stat (for example, running on
1209 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1210 // most cases, so don't abort:
1211 warning("Can't detect initial thread stack location - no /proc/self/stat");
1212 stack_start = (uintptr_t) &rlim;
1213 }
1214 }
1215
1216 // Now we have a pointer (stack_start) very close to the stack top, the
1217 // next thing to do is to figure out the exact location of stack top. We
1218 // can find out the virtual memory area that contains stack_start by
1219 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1220 // and its upper limit is the real stack top. (again, this would fail if
1221 // running inside chroot, because /proc may not exist.)
1222
1223 uintptr_t stack_top;
1224 address low, high;
1225 if (find_vma((address)stack_start, &low, &high)) {
1226 // success, "high" is the true stack top. (ignore "low", because initial
1227 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1228 stack_top = (uintptr_t)high;
1229 } else {
1230 // failed, likely because /proc/self/maps does not exist
1231 warning("Can't detect initial thread stack location - find_vma failed");
1232 // best effort: stack_start is normally within a few pages below the real
1233 // stack top, use it as stack top, and reduce stack size so we won't put
1234 // guard page outside stack.
1235 stack_top = stack_start;
1236 stack_size -= 16 * page_size();
1237 }
1238
1239 // stack_top could be partially down the page so align it
1240 stack_top = align_size_up(stack_top, page_size());
1241
1242 if (max_size && stack_size > max_size) {
1243 _initial_thread_stack_size = max_size;
1244 } else {
1245 _initial_thread_stack_size = stack_size;
1246 }
1247
1248 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1249 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1250 }
1251
1252 ////////////////////////////////////////////////////////////////////////////////
1253 // time support
1254
1255 // Time since start-up in seconds to a fine granularity.
1256 // Used by VMSelfDestructTimer and the MemProfiler.
1257 double os::elapsedTime() {
1258
1259 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1260 }
1261
1262 jlong os::elapsed_counter() {
1263 return javaTimeNanos() - initial_time_count;
1264 }
1265
1266 jlong os::elapsed_frequency() {
1267 return NANOSECS_PER_SEC; // nanosecond resolution
1268 }
1269
1270 bool os::supports_vtime() { return true; }
1271 bool os::enable_vtime() { return false; }
1272 bool os::vtime_enabled() { return false; }
1273
1274 double os::elapsedVTime() {
1275 struct rusage usage;
1276 int retval = getrusage(RUSAGE_THREAD, &usage);
1277 if (retval == 0) {
1278 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);
1279 } else {
1280 // better than nothing, but not much
1281 return elapsedTime();
1282 }
1283 }
1284
1285 jlong os::javaTimeMillis() {
1286 timeval time;
1287 int status = gettimeofday(&time, NULL);
1288 assert(status != -1, "linux error");
1289 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1290 }
1291
1292 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1293 timeval time;
1294 int status = gettimeofday(&time, NULL);
1295 assert(status != -1, "linux error");
1296 seconds = jlong(time.tv_sec);
1297 nanos = jlong(time.tv_usec) * 1000;
1298 }
1299
1300
1301 #ifndef CLOCK_MONOTONIC
1302 #define CLOCK_MONOTONIC (1)
1303 #endif
1304
1305 void os::Linux::clock_init() {
1306 // we do dlopen's in this particular order due to bug in linux
1307 // dynamical loader (see 6348968) leading to crash on exit
1308 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1309 if (handle == NULL) {
1310 handle = dlopen("librt.so", RTLD_LAZY);
1311 }
1312
1313 if (handle) {
1314 int (*clock_getres_func)(clockid_t, struct timespec*) =
1315 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1316 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1317 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1318 if (clock_getres_func && clock_gettime_func) {
1319 // See if monotonic clock is supported by the kernel. Note that some
1320 // early implementations simply return kernel jiffies (updated every
1321 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1322 // for nano time (though the monotonic property is still nice to have).
1323 // It's fixed in newer kernels, however clock_getres() still returns
1324 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1325 // resolution for now. Hopefully as people move to new kernels, this
1326 // won't be a problem.
1327 struct timespec res;
1328 struct timespec tp;
1329 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1330 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1331 // yes, monotonic clock is supported
1332 _clock_gettime = clock_gettime_func;
1333 return;
1334 } else {
1335 // close librt if there is no monotonic clock
1336 dlclose(handle);
1337 }
1338 }
1339 }
1340 warning("No monotonic clock was available - timed services may " \
1341 "be adversely affected if the time-of-day clock changes");
1342 }
1343
1344 #ifndef SYS_clock_getres
1345 #if defined(IA32) || defined(AMD64)
1346 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1347 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1348 #else
1349 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1350 #define sys_clock_getres(x,y) -1
1351 #endif
1352 #else
1353 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1354 #endif
1355
1356 void os::Linux::fast_thread_clock_init() {
1357 if (!UseLinuxPosixThreadCPUClocks) {
1358 return;
1359 }
1360 clockid_t clockid;
1361 struct timespec tp;
1362 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1363 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1364
1365 // Switch to using fast clocks for thread cpu time if
1366 // the sys_clock_getres() returns 0 error code.
1367 // Note, that some kernels may support the current thread
1368 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1369 // returned by the pthread_getcpuclockid().
1370 // If the fast Posix clocks are supported then the sys_clock_getres()
1371 // must return at least tp.tv_sec == 0 which means a resolution
1372 // better than 1 sec. This is extra check for reliability.
1373
1374 if (pthread_getcpuclockid_func &&
1375 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1376 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1377 _supports_fast_thread_cpu_time = true;
1378 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1379 }
1380 }
1381
1382 jlong os::javaTimeNanos() {
1383 if (os::supports_monotonic_clock()) {
1384 struct timespec tp;
1385 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1386 assert(status == 0, "gettime error");
1387 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1388 return result;
1389 } else {
1390 timeval time;
1391 int status = gettimeofday(&time, NULL);
1392 assert(status != -1, "linux error");
1393 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1394 return 1000 * usecs;
1395 }
1396 }
1397
1398 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1399 if (os::supports_monotonic_clock()) {
1400 info_ptr->max_value = ALL_64_BITS;
1401
1402 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1403 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1404 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1405 } else {
1406 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1407 info_ptr->max_value = ALL_64_BITS;
1408
1409 // gettimeofday is a real time clock so it skips
1410 info_ptr->may_skip_backward = true;
1411 info_ptr->may_skip_forward = true;
1412 }
1413
1414 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1415 }
1416
1417 // Return the real, user, and system times in seconds from an
1418 // arbitrary fixed point in the past.
1419 bool os::getTimesSecs(double* process_real_time,
1420 double* process_user_time,
1421 double* process_system_time) {
1422 struct tms ticks;
1423 clock_t real_ticks = times(&ticks);
1424
1425 if (real_ticks == (clock_t) (-1)) {
1426 return false;
1427 } else {
1428 double ticks_per_second = (double) clock_tics_per_sec;
1429 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1430 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1431 *process_real_time = ((double) real_ticks) / ticks_per_second;
1432
1433 return true;
1434 }
1435 }
1436
1437
1438 char * os::local_time_string(char *buf, size_t buflen) {
1439 struct tm t;
1440 time_t long_time;
1441 time(&long_time);
1442 localtime_r(&long_time, &t);
1443 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1444 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1445 t.tm_hour, t.tm_min, t.tm_sec);
1446 return buf;
1447 }
1448
1449 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1450 return localtime_r(clock, res);
1451 }
1452
1453 ////////////////////////////////////////////////////////////////////////////////
1454 // runtime exit support
1455
1456 // Note: os::shutdown() might be called very early during initialization, or
1457 // called from signal handler. Before adding something to os::shutdown(), make
1458 // sure it is async-safe and can handle partially initialized VM.
1459 void os::shutdown() {
1460
1461 // allow PerfMemory to attempt cleanup of any persistent resources
1462 perfMemory_exit();
1463
1464 // needs to remove object in file system
1465 AttachListener::abort();
1466
1467 // flush buffered output, finish log files
1468 ostream_abort();
1469
1470 // Check for abort hook
1471 abort_hook_t abort_hook = Arguments::abort_hook();
1472 if (abort_hook != NULL) {
1473 abort_hook();
1474 }
1475
1476 }
1477
1478 // Note: os::abort() might be called very early during initialization, or
1479 // called from signal handler. Before adding something to os::abort(), make
1480 // sure it is async-safe and can handle partially initialized VM.
1481 void os::abort(bool dump_core) {
1482 abort(dump_core, NULL, NULL);
1483 }
1484
1485 void os::abort(bool dump_core, void* siginfo, void* context) {
1486 os::shutdown();
1487 if (dump_core) {
1488 #ifndef PRODUCT
1489 fdStream out(defaultStream::output_fd());
1490 out.print_raw("Current thread is ");
1491 char buf[16];
1492 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1493 out.print_raw_cr(buf);
1494 out.print_raw_cr("Dumping core ...");
1495 #endif
1496 ::abort(); // dump core
1497 }
1498
1499 ::exit(1);
1500 }
1501
1502 // Die immediately, no exit hook, no abort hook, no cleanup.
1503 void os::die() {
1504 // _exit() on LinuxThreads only kills current thread
1505 ::abort();
1506 }
1507
1508
1509 // This method is a copy of JDK's sysGetLastErrorString
1510 // from src/solaris/hpi/src/system_md.c
1511
1512 size_t os::lasterror(char *buf, size_t len) {
1513 if (errno == 0) return 0;
1514
1515 const char *s = ::strerror(errno);
1516 size_t n = ::strlen(s);
1517 if (n >= len) {
1518 n = len - 1;
1519 }
1520 ::strncpy(buf, s, n);
1521 buf[n] = '\0';
1522 return n;
1523 }
1524
1525 intx os::current_thread_id() { return (intx)pthread_self(); }
1526 int os::current_process_id() {
1527
1528 // Under the old linux thread library, linux gives each thread
1529 // its own process id. Because of this each thread will return
1530 // a different pid if this method were to return the result
1531 // of getpid(2). Linux provides no api that returns the pid
1532 // of the launcher thread for the vm. This implementation
1533 // returns a unique pid, the pid of the launcher thread
1534 // that starts the vm 'process'.
1535
1536 // Under the NPTL, getpid() returns the same pid as the
1537 // launcher thread rather than a unique pid per thread.
1538 // Use gettid() if you want the old pre NPTL behaviour.
1539
1540 // if you are looking for the result of a call to getpid() that
1541 // returns a unique pid for the calling thread, then look at the
1542 // OSThread::thread_id() method in osThread_linux.hpp file
1543
1544 return (int)(_initial_pid ? _initial_pid : getpid());
1545 }
1546
1547 // DLL functions
1548
1549 const char* os::dll_file_extension() { return ".so"; }
1550
1551 // This must be hard coded because it's the system's temporary
1552 // directory not the java application's temp directory, ala java.io.tmpdir.
1553 const char* os::get_temp_directory() { return "/tmp"; }
1554
1555 static bool file_exists(const char* filename) {
1556 struct stat statbuf;
1557 if (filename == NULL || strlen(filename) == 0) {
1558 return false;
1559 }
1560 return os::stat(filename, &statbuf) == 0;
1561 }
1562
1563 bool os::dll_build_name(char* buffer, size_t buflen,
1564 const char* pname, const char* fname) {
1565 bool retval = false;
1566 // Copied from libhpi
1567 const size_t pnamelen = pname ? strlen(pname) : 0;
1568
1569 // Return error on buffer overflow.
1570 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1571 return retval;
1572 }
1573
1574 if (pnamelen == 0) {
1575 snprintf(buffer, buflen, "lib%s.so", fname);
1576 retval = true;
1577 } else if (strchr(pname, *os::path_separator()) != NULL) {
1578 int n;
1579 char** pelements = split_path(pname, &n);
1580 if (pelements == NULL) {
1581 return false;
1582 }
1583 for (int i = 0; i < n; i++) {
1584 // Really shouldn't be NULL, but check can't hurt
1585 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1586 continue; // skip the empty path values
1587 }
1588 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1589 if (file_exists(buffer)) {
1590 retval = true;
1591 break;
1592 }
1593 }
1594 // release the storage
1595 for (int i = 0; i < n; i++) {
1596 if (pelements[i] != NULL) {
1597 FREE_C_HEAP_ARRAY(char, pelements[i]);
1598 }
1599 }
1600 if (pelements != NULL) {
1601 FREE_C_HEAP_ARRAY(char*, pelements);
1602 }
1603 } else {
1604 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1605 retval = true;
1606 }
1607 return retval;
1608 }
1609
1610 // check if addr is inside libjvm.so
1611 bool os::address_is_in_vm(address addr) {
1612 static address libjvm_base_addr;
1613 Dl_info dlinfo;
1614
1615 if (libjvm_base_addr == NULL) {
1616 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1617 libjvm_base_addr = (address)dlinfo.dli_fbase;
1618 }
1619 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1620 }
1621
1622 if (dladdr((void *)addr, &dlinfo) != 0) {
1623 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1624 }
1625
1626 return false;
1627 }
1628
1629 bool os::dll_address_to_function_name(address addr, char *buf,
1630 int buflen, int *offset) {
1631 // buf is not optional, but offset is optional
1632 assert(buf != NULL, "sanity check");
1633
1634 Dl_info dlinfo;
1635
1636 if (dladdr((void*)addr, &dlinfo) != 0) {
1637 // see if we have a matching symbol
1638 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1639 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1640 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1641 }
1642 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1643 return true;
1644 }
1645 // no matching symbol so try for just file info
1646 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1647 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1648 buf, buflen, offset, dlinfo.dli_fname)) {
1649 return true;
1650 }
1651 }
1652 }
1653
1654 buf[0] = '\0';
1655 if (offset != NULL) *offset = -1;
1656 return false;
1657 }
1658
1659 struct _address_to_library_name {
1660 address addr; // input : memory address
1661 size_t buflen; // size of fname
1662 char* fname; // output: library name
1663 address base; // library base addr
1664 };
1665
1666 static int address_to_library_name_callback(struct dl_phdr_info *info,
1667 size_t size, void *data) {
1668 int i;
1669 bool found = false;
1670 address libbase = NULL;
1671 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1672
1673 // iterate through all loadable segments
1674 for (i = 0; i < info->dlpi_phnum; i++) {
1675 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1676 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1677 // base address of a library is the lowest address of its loaded
1678 // segments.
1679 if (libbase == NULL || libbase > segbase) {
1680 libbase = segbase;
1681 }
1682 // see if 'addr' is within current segment
1683 if (segbase <= d->addr &&
1684 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1685 found = true;
1686 }
1687 }
1688 }
1689
1690 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1691 // so dll_address_to_library_name() can fall through to use dladdr() which
1692 // can figure out executable name from argv[0].
1693 if (found && info->dlpi_name && info->dlpi_name[0]) {
1694 d->base = libbase;
1695 if (d->fname) {
1696 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1697 }
1698 return 1;
1699 }
1700 return 0;
1701 }
1702
1703 bool os::dll_address_to_library_name(address addr, char* buf,
1704 int buflen, int* offset) {
1705 // buf is not optional, but offset is optional
1706 assert(buf != NULL, "sanity check");
1707
1708 Dl_info dlinfo;
1709 struct _address_to_library_name data;
1710
1711 // There is a bug in old glibc dladdr() implementation that it could resolve
1712 // to wrong library name if the .so file has a base address != NULL. Here
1713 // we iterate through the program headers of all loaded libraries to find
1714 // out which library 'addr' really belongs to. This workaround can be
1715 // removed once the minimum requirement for glibc is moved to 2.3.x.
1716 data.addr = addr;
1717 data.fname = buf;
1718 data.buflen = buflen;
1719 data.base = NULL;
1720 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1721
1722 if (rslt) {
1723 // buf already contains library name
1724 if (offset) *offset = addr - data.base;
1725 return true;
1726 }
1727 if (dladdr((void*)addr, &dlinfo) != 0) {
1728 if (dlinfo.dli_fname != NULL) {
1729 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1730 }
1731 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1732 *offset = addr - (address)dlinfo.dli_fbase;
1733 }
1734 return true;
1735 }
1736
1737 buf[0] = '\0';
1738 if (offset) *offset = -1;
1739 return false;
1740 }
1741
1742 // Loads .dll/.so and
1743 // in case of error it checks if .dll/.so was built for the
1744 // same architecture as Hotspot is running on
1745
1746
1747 // Remember the stack's state. The Linux dynamic linker will change
1748 // the stack to 'executable' at most once, so we must safepoint only once.
1749 bool os::Linux::_stack_is_executable = false;
1750
1751 // VM operation that loads a library. This is necessary if stack protection
1752 // of the Java stacks can be lost during loading the library. If we
1753 // do not stop the Java threads, they can stack overflow before the stacks
1754 // are protected again.
1755 class VM_LinuxDllLoad: public VM_Operation {
1756 private:
1757 const char *_filename;
1758 char *_ebuf;
1759 int _ebuflen;
1760 void *_lib;
1761 public:
1762 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1763 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1764 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1765 void doit() {
1766 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1767 os::Linux::_stack_is_executable = true;
1768 }
1769 void* loaded_library() { return _lib; }
1770 };
1771
1772 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1773 void * result = NULL;
1774 bool load_attempted = false;
1775
1776 // Check whether the library to load might change execution rights
1777 // of the stack. If they are changed, the protection of the stack
1778 // guard pages will be lost. We need a safepoint to fix this.
1779 //
1780 // See Linux man page execstack(8) for more info.
1781 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1782 ElfFile ef(filename);
1783 if (!ef.specifies_noexecstack()) {
1784 if (!is_init_completed()) {
1785 os::Linux::_stack_is_executable = true;
1786 // This is OK - No Java threads have been created yet, and hence no
1787 // stack guard pages to fix.
1788 //
1789 // This should happen only when you are building JDK7 using a very
1790 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1791 //
1792 // Dynamic loader will make all stacks executable after
1793 // this function returns, and will not do that again.
1794 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1795 } else {
1796 warning("You have loaded library %s which might have disabled stack guard. "
1797 "The VM will try to fix the stack guard now.\n"
1798 "It's highly recommended that you fix the library with "
1799 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1800 filename);
1801
1802 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1803 JavaThread *jt = JavaThread::current();
1804 if (jt->thread_state() != _thread_in_native) {
1805 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1806 // that requires ExecStack. Cannot enter safe point. Let's give up.
1807 warning("Unable to fix stack guard. Giving up.");
1808 } else {
1809 if (!LoadExecStackDllInVMThread) {
1810 // This is for the case where the DLL has an static
1811 // constructor function that executes JNI code. We cannot
1812 // load such DLLs in the VMThread.
1813 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1814 }
1815
1816 ThreadInVMfromNative tiv(jt);
1817 debug_only(VMNativeEntryWrapper vew;)
1818
1819 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1820 VMThread::execute(&op);
1821 if (LoadExecStackDllInVMThread) {
1822 result = op.loaded_library();
1823 }
1824 load_attempted = true;
1825 }
1826 }
1827 }
1828 }
1829
1830 if (!load_attempted) {
1831 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1832 }
1833
1834 if (result != NULL) {
1835 // Successful loading
1836 return result;
1837 }
1838
1839 Elf32_Ehdr elf_head;
1840 int diag_msg_max_length=ebuflen-strlen(ebuf);
1841 char* diag_msg_buf=ebuf+strlen(ebuf);
1842
1843 if (diag_msg_max_length==0) {
1844 // No more space in ebuf for additional diagnostics message
1845 return NULL;
1846 }
1847
1848
1849 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1850
1851 if (file_descriptor < 0) {
1852 // Can't open library, report dlerror() message
1853 return NULL;
1854 }
1855
1856 bool failed_to_read_elf_head=
1857 (sizeof(elf_head)!=
1858 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1859
1860 ::close(file_descriptor);
1861 if (failed_to_read_elf_head) {
1862 // file i/o error - report dlerror() msg
1863 return NULL;
1864 }
1865
1866 typedef struct {
1867 Elf32_Half code; // Actual value as defined in elf.h
1868 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1869 char elf_class; // 32 or 64 bit
1870 char endianess; // MSB or LSB
1871 char* name; // String representation
1872 } arch_t;
1873
1874 #ifndef EM_486
1875 #define EM_486 6 /* Intel 80486 */
1876 #endif
1877 #ifndef EM_AARCH64
1878 #define EM_AARCH64 183 /* ARM AARCH64 */
1879 #endif
1880
1881 static const arch_t arch_array[]={
1882 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1883 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1884 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1885 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1886 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1887 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1888 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1889 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1890 #if defined(VM_LITTLE_ENDIAN)
1891 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1892 #else
1893 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1894 #endif
1895 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1896 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1897 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1898 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1899 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1900 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1901 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1902 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1903 };
1904
1905 #if (defined IA32)
1906 static Elf32_Half running_arch_code=EM_386;
1907 #elif (defined AMD64)
1908 static Elf32_Half running_arch_code=EM_X86_64;
1909 #elif (defined IA64)
1910 static Elf32_Half running_arch_code=EM_IA_64;
1911 #elif (defined __sparc) && (defined _LP64)
1912 static Elf32_Half running_arch_code=EM_SPARCV9;
1913 #elif (defined __sparc) && (!defined _LP64)
1914 static Elf32_Half running_arch_code=EM_SPARC;
1915 #elif (defined __powerpc64__)
1916 static Elf32_Half running_arch_code=EM_PPC64;
1917 #elif (defined __powerpc__)
1918 static Elf32_Half running_arch_code=EM_PPC;
1919 #elif (defined ARM)
1920 static Elf32_Half running_arch_code=EM_ARM;
1921 #elif (defined S390)
1922 static Elf32_Half running_arch_code=EM_S390;
1923 #elif (defined ALPHA)
1924 static Elf32_Half running_arch_code=EM_ALPHA;
1925 #elif (defined MIPSEL)
1926 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1927 #elif (defined PARISC)
1928 static Elf32_Half running_arch_code=EM_PARISC;
1929 #elif (defined MIPS)
1930 static Elf32_Half running_arch_code=EM_MIPS;
1931 #elif (defined M68K)
1932 static Elf32_Half running_arch_code=EM_68K;
1933 #elif (defined AARCH64)
1934 static Elf32_Half running_arch_code=EM_AARCH64;
1935 #else
1936 #error Method os::dll_load requires that one of following is defined:\
1937 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1938 #endif
1939
1940 // Identify compatability class for VM's architecture and library's architecture
1941 // Obtain string descriptions for architectures
1942
1943 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1944 int running_arch_index=-1;
1945
1946 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1947 if (running_arch_code == arch_array[i].code) {
1948 running_arch_index = i;
1949 }
1950 if (lib_arch.code == arch_array[i].code) {
1951 lib_arch.compat_class = arch_array[i].compat_class;
1952 lib_arch.name = arch_array[i].name;
1953 }
1954 }
1955
1956 assert(running_arch_index != -1,
1957 "Didn't find running architecture code (running_arch_code) in arch_array");
1958 if (running_arch_index == -1) {
1959 // Even though running architecture detection failed
1960 // we may still continue with reporting dlerror() message
1961 return NULL;
1962 }
1963
1964 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1965 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1966 return NULL;
1967 }
1968
1969 #ifndef S390
1970 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1971 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1972 return NULL;
1973 }
1974 #endif // !S390
1975
1976 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1977 if (lib_arch.name!=NULL) {
1978 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1979 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1980 lib_arch.name, arch_array[running_arch_index].name);
1981 } else {
1982 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1983 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1984 lib_arch.code,
1985 arch_array[running_arch_index].name);
1986 }
1987 }
1988
1989 return NULL;
1990 }
1991
1992 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1993 int ebuflen) {
1994 void * result = ::dlopen(filename, RTLD_LAZY);
1995 if (result == NULL) {
1996 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1997 ebuf[ebuflen-1] = '\0';
1998 }
1999 return result;
2000 }
2001
2002 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2003 int ebuflen) {
2004 void * result = NULL;
2005 if (LoadExecStackDllInVMThread) {
2006 result = dlopen_helper(filename, ebuf, ebuflen);
2007 }
2008
2009 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2010 // library that requires an executable stack, or which does not have this
2011 // stack attribute set, dlopen changes the stack attribute to executable. The
2012 // read protection of the guard pages gets lost.
2013 //
2014 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2015 // may have been queued at the same time.
2016
2017 if (!_stack_is_executable) {
2018 JavaThread *jt = Threads::first();
2019
2020 while (jt) {
2021 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2022 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2023 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2024 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2025 warning("Attempt to reguard stack yellow zone failed.");
2026 }
2027 }
2028 jt = jt->next();
2029 }
2030 }
2031
2032 return result;
2033 }
2034
2035 void* os::dll_lookup(void* handle, const char* name) {
2036 void* res = dlsym(handle, name);
2037 return res;
2038 }
2039
2040 void* os::get_default_process_handle() {
2041 return (void*)::dlopen(NULL, RTLD_LAZY);
2042 }
2043
2044 static bool _print_ascii_file(const char* filename, outputStream* st) {
2045 int fd = ::open(filename, O_RDONLY);
2046 if (fd == -1) {
2047 return false;
2048 }
2049
2050 char buf[32];
2051 int bytes;
2052 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2053 st->print_raw(buf, bytes);
2054 }
2055
2056 ::close(fd);
2057
2058 return true;
2059 }
2060
2061 void os::print_dll_info(outputStream *st) {
2062 st->print_cr("Dynamic libraries:");
2063
2064 char fname[32];
2065 pid_t pid = os::Linux::gettid();
2066
2067 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2068
2069 if (!_print_ascii_file(fname, st)) {
2070 st->print("Can not get library information for pid = %d\n", pid);
2071 }
2072 }
2073
2074 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2075 FILE *procmapsFile = NULL;
2076
2077 // Open the procfs maps file for the current process
2078 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2079 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2080 char line[PATH_MAX + 100];
2081
2082 // Read line by line from 'file'
2083 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2084 u8 base, top, offset, inode;
2085 char permissions[5];
2086 char device[6];
2087 char name[PATH_MAX + 1];
2088
2089 // Parse fields from line
2090 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2091
2092 // Filter by device id '00:00' so that we only get file system mapped files.
2093 if (strcmp(device, "00:00") != 0) {
2094
2095 // Call callback with the fields of interest
2096 if(callback(name, (address)base, (address)top, param)) {
2097 // Oops abort, callback aborted
2098 fclose(procmapsFile);
2099 return 1;
2100 }
2101 }
2102 }
2103 fclose(procmapsFile);
2104 }
2105 return 0;
2106 }
2107
2108 void os::print_os_info_brief(outputStream* st) {
2109 os::Linux::print_distro_info(st);
2110
2111 os::Posix::print_uname_info(st);
2112
2113 os::Linux::print_libversion_info(st);
2114
2115 }
2116
2117 void os::print_os_info(outputStream* st) {
2118 st->print("OS:");
2119
2120 os::Linux::print_distro_info(st);
2121
2122 os::Posix::print_uname_info(st);
2123
2124 // Print warning if unsafe chroot environment detected
2125 if (unsafe_chroot_detected) {
2126 st->print("WARNING!! ");
2127 st->print_cr("%s", unstable_chroot_error);
2128 }
2129
2130 os::Linux::print_libversion_info(st);
2131
2132 os::Posix::print_rlimit_info(st);
2133
2134 os::Posix::print_load_average(st);
2135
2136 os::Linux::print_full_memory_info(st);
2137 }
2138
2139 // Try to identify popular distros.
2140 // Most Linux distributions have a /etc/XXX-release file, which contains
2141 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2142 // file that also contains the OS version string. Some have more than one
2143 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2144 // /etc/redhat-release.), so the order is important.
2145 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2146 // their own specific XXX-release file as well as a redhat-release file.
2147 // Because of this the XXX-release file needs to be searched for before the
2148 // redhat-release file.
2149 // Since Red Hat has a lsb-release file that is not very descriptive the
2150 // search for redhat-release needs to be before lsb-release.
2151 // Since the lsb-release file is the new standard it needs to be searched
2152 // before the older style release files.
2153 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2154 // next to last resort. The os-release file is a new standard that contains
2155 // distribution information and the system-release file seems to be an old
2156 // standard that has been replaced by the lsb-release and os-release files.
2157 // Searching for the debian_version file is the last resort. It contains
2158 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2159 // "Debian " is printed before the contents of the debian_version file.
2160 void os::Linux::print_distro_info(outputStream* st) {
2161 if (!_print_ascii_file("/etc/oracle-release", st) &&
2162 !_print_ascii_file("/etc/mandriva-release", st) &&
2163 !_print_ascii_file("/etc/mandrake-release", st) &&
2164 !_print_ascii_file("/etc/sun-release", st) &&
2165 !_print_ascii_file("/etc/redhat-release", st) &&
2166 !_print_ascii_file("/etc/lsb-release", st) &&
2167 !_print_ascii_file("/etc/SuSE-release", st) &&
2168 !_print_ascii_file("/etc/turbolinux-release", st) &&
2169 !_print_ascii_file("/etc/gentoo-release", st) &&
2170 !_print_ascii_file("/etc/ltib-release", st) &&
2171 !_print_ascii_file("/etc/angstrom-version", st) &&
2172 !_print_ascii_file("/etc/system-release", st) &&
2173 !_print_ascii_file("/etc/os-release", st)) {
2174
2175 if (file_exists("/etc/debian_version")) {
2176 st->print("Debian ");
2177 _print_ascii_file("/etc/debian_version", st);
2178 } else {
2179 st->print("Linux");
2180 }
2181 }
2182 st->cr();
2183 }
2184
2185 void os::Linux::print_libversion_info(outputStream* st) {
2186 // libc, pthread
2187 st->print("libc:");
2188 st->print("%s ", os::Linux::glibc_version());
2189 st->print("%s ", os::Linux::libpthread_version());
2190 if (os::Linux::is_LinuxThreads()) {
2191 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2192 }
2193 st->cr();
2194 }
2195
2196 void os::Linux::print_full_memory_info(outputStream* st) {
2197 st->print("\n/proc/meminfo:\n");
2198 _print_ascii_file("/proc/meminfo", st);
2199 st->cr();
2200 }
2201
2202 void os::print_memory_info(outputStream* st) {
2203
2204 st->print("Memory:");
2205 st->print(" %dk page", os::vm_page_size()>>10);
2206
2207 // values in struct sysinfo are "unsigned long"
2208 struct sysinfo si;
2209 sysinfo(&si);
2210
2211 st->print(", physical " UINT64_FORMAT "k",
2212 os::physical_memory() >> 10);
2213 st->print("(" UINT64_FORMAT "k free)",
2214 os::available_memory() >> 10);
2215 st->print(", swap " UINT64_FORMAT "k",
2216 ((jlong)si.totalswap * si.mem_unit) >> 10);
2217 st->print("(" UINT64_FORMAT "k free)",
2218 ((jlong)si.freeswap * si.mem_unit) >> 10);
2219 st->cr();
2220 }
2221
2222 void os::pd_print_cpu_info(outputStream* st) {
2223 st->print("\n/proc/cpuinfo:\n");
2224 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2225 st->print(" <Not Available>");
2226 }
2227 st->cr();
2228 }
2229
2230 void os::print_siginfo(outputStream* st, void* siginfo) {
2231 const siginfo_t* si = (const siginfo_t*)siginfo;
2232
2233 os::Posix::print_siginfo_brief(st, si);
2234 #if INCLUDE_CDS
2235 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2236 UseSharedSpaces) {
2237 FileMapInfo* mapinfo = FileMapInfo::current_info();
2238 if (mapinfo->is_in_shared_space(si->si_addr)) {
2239 st->print("\n\nError accessing class data sharing archive." \
2240 " Mapped file inaccessible during execution, " \
2241 " possible disk/network problem.");
2242 }
2243 }
2244 #endif
2245 st->cr();
2246 }
2247
2248
2249 static void print_signal_handler(outputStream* st, int sig,
2250 char* buf, size_t buflen);
2251
2252 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2253 st->print_cr("Signal Handlers:");
2254 print_signal_handler(st, SIGSEGV, buf, buflen);
2255 print_signal_handler(st, SIGBUS , buf, buflen);
2256 print_signal_handler(st, SIGFPE , buf, buflen);
2257 print_signal_handler(st, SIGPIPE, buf, buflen);
2258 print_signal_handler(st, SIGXFSZ, buf, buflen);
2259 print_signal_handler(st, SIGILL , buf, buflen);
2260 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2261 print_signal_handler(st, SR_signum, buf, buflen);
2262 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2263 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2264 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2265 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2266 #if defined(PPC64)
2267 print_signal_handler(st, SIGTRAP, buf, buflen);
2268 #endif
2269 }
2270
2271 static char saved_jvm_path[MAXPATHLEN] = {0};
2272
2273 // Find the full path to the current module, libjvm.so
2274 void os::jvm_path(char *buf, jint buflen) {
2275 // Error checking.
2276 if (buflen < MAXPATHLEN) {
2277 assert(false, "must use a large-enough buffer");
2278 buf[0] = '\0';
2279 return;
2280 }
2281 // Lazy resolve the path to current module.
2282 if (saved_jvm_path[0] != 0) {
2283 strcpy(buf, saved_jvm_path);
2284 return;
2285 }
2286
2287 char dli_fname[MAXPATHLEN];
2288 bool ret = dll_address_to_library_name(
2289 CAST_FROM_FN_PTR(address, os::jvm_path),
2290 dli_fname, sizeof(dli_fname), NULL);
2291 assert(ret, "cannot locate libjvm");
2292 char *rp = NULL;
2293 if (ret && dli_fname[0] != '\0') {
2294 rp = realpath(dli_fname, buf);
2295 }
2296 if (rp == NULL) {
2297 return;
2298 }
2299
2300 if (Arguments::sun_java_launcher_is_altjvm()) {
2301 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2302 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2303 // If "/jre/lib/" appears at the right place in the string, then
2304 // assume we are installed in a JDK and we're done. Otherwise, check
2305 // for a JAVA_HOME environment variable and fix up the path so it
2306 // looks like libjvm.so is installed there (append a fake suffix
2307 // hotspot/libjvm.so).
2308 const char *p = buf + strlen(buf) - 1;
2309 for (int count = 0; p > buf && count < 5; ++count) {
2310 for (--p; p > buf && *p != '/'; --p)
2311 /* empty */ ;
2312 }
2313
2314 if (strncmp(p, "/jre/lib/", 9) != 0) {
2315 // Look for JAVA_HOME in the environment.
2316 char* java_home_var = ::getenv("JAVA_HOME");
2317 if (java_home_var != NULL && java_home_var[0] != 0) {
2318 char* jrelib_p;
2319 int len;
2320
2321 // Check the current module name "libjvm.so".
2322 p = strrchr(buf, '/');
2323 if (p == NULL) {
2324 return;
2325 }
2326 assert(strstr(p, "/libjvm") == p, "invalid library name");
2327
2328 rp = realpath(java_home_var, buf);
2329 if (rp == NULL) {
2330 return;
2331 }
2332
2333 // determine if this is a legacy image or modules image
2334 // modules image doesn't have "jre" subdirectory
2335 len = strlen(buf);
2336 assert(len < buflen, "Ran out of buffer room");
2337 jrelib_p = buf + len;
2338 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2339 if (0 != access(buf, F_OK)) {
2340 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2341 }
2342
2343 if (0 == access(buf, F_OK)) {
2344 // Use current module name "libjvm.so"
2345 len = strlen(buf);
2346 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2347 } else {
2348 // Go back to path of .so
2349 rp = realpath(dli_fname, buf);
2350 if (rp == NULL) {
2351 return;
2352 }
2353 }
2354 }
2355 }
2356 }
2357
2358 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2359 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2360 }
2361
2362 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2363 // no prefix required, not even "_"
2364 }
2365
2366 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2367 // no suffix required
2368 }
2369
2370 ////////////////////////////////////////////////////////////////////////////////
2371 // sun.misc.Signal support
2372
2373 static volatile jint sigint_count = 0;
2374
2375 static void UserHandler(int sig, void *siginfo, void *context) {
2376 // 4511530 - sem_post is serialized and handled by the manager thread. When
2377 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2378 // don't want to flood the manager thread with sem_post requests.
2379 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2380 return;
2381 }
2382
2383 // Ctrl-C is pressed during error reporting, likely because the error
2384 // handler fails to abort. Let VM die immediately.
2385 if (sig == SIGINT && is_error_reported()) {
2386 os::die();
2387 }
2388
2389 os::signal_notify(sig);
2390 }
2391
2392 void* os::user_handler() {
2393 return CAST_FROM_FN_PTR(void*, UserHandler);
2394 }
2395
2396 struct timespec os::PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2397 struct timespec ts;
2398 // Semaphore's are always associated with CLOCK_REALTIME
2399 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2400 // see unpackTime for discussion on overflow checking
2401 if (sec >= MAX_SECS) {
2402 ts.tv_sec += MAX_SECS;
2403 ts.tv_nsec = 0;
2404 } else {
2405 ts.tv_sec += sec;
2406 ts.tv_nsec += nsec;
2407 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2408 ts.tv_nsec -= NANOSECS_PER_SEC;
2409 ++ts.tv_sec; // note: this must be <= max_secs
2410 }
2411 }
2412
2413 return ts;
2414 }
2415
2416 extern "C" {
2417 typedef void (*sa_handler_t)(int);
2418 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2419 }
2420
2421 void* os::signal(int signal_number, void* handler) {
2422 struct sigaction sigAct, oldSigAct;
2423
2424 sigfillset(&(sigAct.sa_mask));
2425 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2426 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2427
2428 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2429 // -1 means registration failed
2430 return (void *)-1;
2431 }
2432
2433 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2434 }
2435
2436 void os::signal_raise(int signal_number) {
2437 ::raise(signal_number);
2438 }
2439
2440 // The following code is moved from os.cpp for making this
2441 // code platform specific, which it is by its very nature.
2442
2443 // Will be modified when max signal is changed to be dynamic
2444 int os::sigexitnum_pd() {
2445 return NSIG;
2446 }
2447
2448 // a counter for each possible signal value
2449 static volatile jint pending_signals[NSIG+1] = { 0 };
2450
2451 // Linux(POSIX) specific hand shaking semaphore.
2452 static sem_t sig_sem;
2453 static os::PosixSemaphore sr_semaphore;
2454
2455 void os::signal_init_pd() {
2456 // Initialize signal structures
2457 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2458
2459 // Initialize signal semaphore
2460 ::sem_init(&sig_sem, 0, 0);
2461 }
2462
2463 void os::signal_notify(int sig) {
2464 Atomic::inc(&pending_signals[sig]);
2465 ::sem_post(&sig_sem);
2466 }
2467
2468 static int check_pending_signals(bool wait) {
2469 Atomic::store(0, &sigint_count);
2470 for (;;) {
2471 for (int i = 0; i < NSIG + 1; i++) {
2472 jint n = pending_signals[i];
2473 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2474 return i;
2475 }
2476 }
2477 if (!wait) {
2478 return -1;
2479 }
2480 JavaThread *thread = JavaThread::current();
2481 ThreadBlockInVM tbivm(thread);
2482
2483 bool threadIsSuspended;
2484 do {
2485 thread->set_suspend_equivalent();
2486 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2487 ::sem_wait(&sig_sem);
2488
2489 // were we externally suspended while we were waiting?
2490 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2491 if (threadIsSuspended) {
2492 // The semaphore has been incremented, but while we were waiting
2493 // another thread suspended us. We don't want to continue running
2494 // while suspended because that would surprise the thread that
2495 // suspended us.
2496 ::sem_post(&sig_sem);
2497
2498 thread->java_suspend_self();
2499 }
2500 } while (threadIsSuspended);
2501 }
2502 }
2503
2504 int os::signal_lookup() {
2505 return check_pending_signals(false);
2506 }
2507
2508 int os::signal_wait() {
2509 return check_pending_signals(true);
2510 }
2511
2512 ////////////////////////////////////////////////////////////////////////////////
2513 // Virtual Memory
2514
2515 int os::vm_page_size() {
2516 // Seems redundant as all get out
2517 assert(os::Linux::page_size() != -1, "must call os::init");
2518 return os::Linux::page_size();
2519 }
2520
2521 // Solaris allocates memory by pages.
2522 int os::vm_allocation_granularity() {
2523 assert(os::Linux::page_size() != -1, "must call os::init");
2524 return os::Linux::page_size();
2525 }
2526
2527 // Rationale behind this function:
2528 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2529 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2530 // samples for JITted code. Here we create private executable mapping over the code cache
2531 // and then we can use standard (well, almost, as mapping can change) way to provide
2532 // info for the reporting script by storing timestamp and location of symbol
2533 void linux_wrap_code(char* base, size_t size) {
2534 static volatile jint cnt = 0;
2535
2536 if (!UseOprofile) {
2537 return;
2538 }
2539
2540 char buf[PATH_MAX+1];
2541 int num = Atomic::add(1, &cnt);
2542
2543 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2544 os::get_temp_directory(), os::current_process_id(), num);
2545 unlink(buf);
2546
2547 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2548
2549 if (fd != -1) {
2550 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2551 if (rv != (off_t)-1) {
2552 if (::write(fd, "", 1) == 1) {
2553 mmap(base, size,
2554 PROT_READ|PROT_WRITE|PROT_EXEC,
2555 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2556 }
2557 }
2558 ::close(fd);
2559 unlink(buf);
2560 }
2561 }
2562
2563 static bool recoverable_mmap_error(int err) {
2564 // See if the error is one we can let the caller handle. This
2565 // list of errno values comes from JBS-6843484. I can't find a
2566 // Linux man page that documents this specific set of errno
2567 // values so while this list currently matches Solaris, it may
2568 // change as we gain experience with this failure mode.
2569 switch (err) {
2570 case EBADF:
2571 case EINVAL:
2572 case ENOTSUP:
2573 // let the caller deal with these errors
2574 return true;
2575
2576 default:
2577 // Any remaining errors on this OS can cause our reserved mapping
2578 // to be lost. That can cause confusion where different data
2579 // structures think they have the same memory mapped. The worst
2580 // scenario is if both the VM and a library think they have the
2581 // same memory mapped.
2582 return false;
2583 }
2584 }
2585
2586 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2587 int err) {
2588 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2589 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2590 strerror(err), err);
2591 }
2592
2593 static void warn_fail_commit_memory(char* addr, size_t size,
2594 size_t alignment_hint, bool exec,
2595 int err) {
2596 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2597 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2598 alignment_hint, exec, strerror(err), err);
2599 }
2600
2601 // NOTE: Linux kernel does not really reserve the pages for us.
2602 // All it does is to check if there are enough free pages
2603 // left at the time of mmap(). This could be a potential
2604 // problem.
2605 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2606 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2607 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2608 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2609 if (res != (uintptr_t) MAP_FAILED) {
2610 if (UseNUMAInterleaving) {
2611 numa_make_global(addr, size);
2612 }
2613 return 0;
2614 }
2615
2616 int err = errno; // save errno from mmap() call above
2617
2618 if (!recoverable_mmap_error(err)) {
2619 warn_fail_commit_memory(addr, size, exec, err);
2620 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2621 }
2622
2623 return err;
2624 }
2625
2626 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2627 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2628 }
2629
2630 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2631 const char* mesg) {
2632 assert(mesg != NULL, "mesg must be specified");
2633 int err = os::Linux::commit_memory_impl(addr, size, exec);
2634 if (err != 0) {
2635 // the caller wants all commit errors to exit with the specified mesg:
2636 warn_fail_commit_memory(addr, size, exec, err);
2637 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2638 }
2639 }
2640
2641 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2642 #ifndef MAP_HUGETLB
2643 #define MAP_HUGETLB 0x40000
2644 #endif
2645
2646 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2647 #ifndef MADV_HUGEPAGE
2648 #define MADV_HUGEPAGE 14
2649 #endif
2650
2651 int os::Linux::commit_memory_impl(char* addr, size_t size,
2652 size_t alignment_hint, bool exec) {
2653 int err = os::Linux::commit_memory_impl(addr, size, exec);
2654 if (err == 0) {
2655 realign_memory(addr, size, alignment_hint);
2656 }
2657 return err;
2658 }
2659
2660 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2661 bool exec) {
2662 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2663 }
2664
2665 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2666 size_t alignment_hint, bool exec,
2667 const char* mesg) {
2668 assert(mesg != NULL, "mesg must be specified");
2669 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2670 if (err != 0) {
2671 // the caller wants all commit errors to exit with the specified mesg:
2672 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2673 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2674 }
2675 }
2676
2677 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2678 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2679 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2680 // be supported or the memory may already be backed by huge pages.
2681 ::madvise(addr, bytes, MADV_HUGEPAGE);
2682 }
2683 }
2684
2685 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2686 // This method works by doing an mmap over an existing mmaping and effectively discarding
2687 // the existing pages. However it won't work for SHM-based large pages that cannot be
2688 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2689 // small pages on top of the SHM segment. This method always works for small pages, so we
2690 // allow that in any case.
2691 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2692 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2693 }
2694 }
2695
2696 void os::numa_make_global(char *addr, size_t bytes) {
2697 Linux::numa_interleave_memory(addr, bytes);
2698 }
2699
2700 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2701 // bind policy to MPOL_PREFERRED for the current thread.
2702 #define USE_MPOL_PREFERRED 0
2703
2704 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2705 // To make NUMA and large pages more robust when both enabled, we need to ease
2706 // the requirements on where the memory should be allocated. MPOL_BIND is the
2707 // default policy and it will force memory to be allocated on the specified
2708 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2709 // the specified node, but will not force it. Using this policy will prevent
2710 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2711 // free large pages.
2712 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2713 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2714 }
2715
2716 bool os::numa_topology_changed() { return false; }
2717
2718 size_t os::numa_get_groups_num() {
2719 int max_node = Linux::numa_max_node();
2720 return max_node > 0 ? max_node + 1 : 1;
2721 }
2722
2723 int os::numa_get_group_id() {
2724 int cpu_id = Linux::sched_getcpu();
2725 if (cpu_id != -1) {
2726 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2727 if (lgrp_id != -1) {
2728 return lgrp_id;
2729 }
2730 }
2731 return 0;
2732 }
2733
2734 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2735 for (size_t i = 0; i < size; i++) {
2736 ids[i] = i;
2737 }
2738 return size;
2739 }
2740
2741 bool os::get_page_info(char *start, page_info* info) {
2742 return false;
2743 }
2744
2745 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2746 page_info* page_found) {
2747 return end;
2748 }
2749
2750
2751 int os::Linux::sched_getcpu_syscall(void) {
2752 unsigned int cpu;
2753 int retval = -1;
2754
2755 #if defined(IA32)
2756 #ifndef SYS_getcpu
2757 #define SYS_getcpu 318
2758 #endif
2759 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2760 #elif defined(AMD64)
2761 // Unfortunately we have to bring all these macros here from vsyscall.h
2762 // to be able to compile on old linuxes.
2763 #define __NR_vgetcpu 2
2764 #define VSYSCALL_START (-10UL << 20)
2765 #define VSYSCALL_SIZE 1024
2766 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2767 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2768 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2769 retval = vgetcpu(&cpu, NULL, NULL);
2770 #endif
2771
2772 return (retval == -1) ? retval : cpu;
2773 }
2774
2775 // Something to do with the numa-aware allocator needs these symbols
2776 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2777 extern "C" JNIEXPORT void numa_error(char *where) { }
2778 extern "C" JNIEXPORT int fork1() { return fork(); }
2779
2780
2781 // If we are running with libnuma version > 2, then we should
2782 // be trying to use symbols with versions 1.1
2783 // If we are running with earlier version, which did not have symbol versions,
2784 // we should use the base version.
2785 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2786 void *f = dlvsym(handle, name, "libnuma_1.1");
2787 if (f == NULL) {
2788 f = dlsym(handle, name);
2789 }
2790 return f;
2791 }
2792
2793 bool os::Linux::libnuma_init() {
2794 // sched_getcpu() should be in libc.
2795 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2796 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2797
2798 // If it's not, try a direct syscall.
2799 if (sched_getcpu() == -1) {
2800 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2801 (void*)&sched_getcpu_syscall));
2802 }
2803
2804 if (sched_getcpu() != -1) { // Does it work?
2805 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2806 if (handle != NULL) {
2807 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2808 libnuma_dlsym(handle, "numa_node_to_cpus")));
2809 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2810 libnuma_dlsym(handle, "numa_max_node")));
2811 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2812 libnuma_dlsym(handle, "numa_available")));
2813 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2814 libnuma_dlsym(handle, "numa_tonode_memory")));
2815 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2816 libnuma_dlsym(handle, "numa_interleave_memory")));
2817 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2818 libnuma_dlsym(handle, "numa_set_bind_policy")));
2819
2820
2821 if (numa_available() != -1) {
2822 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2823 // Create a cpu -> node mapping
2824 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2825 rebuild_cpu_to_node_map();
2826 return true;
2827 }
2828 }
2829 }
2830 return false;
2831 }
2832
2833 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2834 // The table is later used in get_node_by_cpu().
2835 void os::Linux::rebuild_cpu_to_node_map() {
2836 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2837 // in libnuma (possible values are starting from 16,
2838 // and continuing up with every other power of 2, but less
2839 // than the maximum number of CPUs supported by kernel), and
2840 // is a subject to change (in libnuma version 2 the requirements
2841 // are more reasonable) we'll just hardcode the number they use
2842 // in the library.
2843 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2844
2845 size_t cpu_num = os::active_processor_count();
2846 size_t cpu_map_size = NCPUS / BitsPerCLong;
2847 size_t cpu_map_valid_size =
2848 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2849
2850 cpu_to_node()->clear();
2851 cpu_to_node()->at_grow(cpu_num - 1);
2852 size_t node_num = numa_get_groups_num();
2853
2854 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2855 for (size_t i = 0; i < node_num; i++) {
2856 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2857 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2858 if (cpu_map[j] != 0) {
2859 for (size_t k = 0; k < BitsPerCLong; k++) {
2860 if (cpu_map[j] & (1UL << k)) {
2861 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2862 }
2863 }
2864 }
2865 }
2866 }
2867 }
2868 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2869 }
2870
2871 int os::Linux::get_node_by_cpu(int cpu_id) {
2872 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2873 return cpu_to_node()->at(cpu_id);
2874 }
2875 return -1;
2876 }
2877
2878 GrowableArray<int>* os::Linux::_cpu_to_node;
2879 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2880 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2881 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2882 os::Linux::numa_available_func_t os::Linux::_numa_available;
2883 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2884 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2885 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2886 unsigned long* os::Linux::_numa_all_nodes;
2887
2888 bool os::pd_uncommit_memory(char* addr, size_t size) {
2889 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2890 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2891 return res != (uintptr_t) MAP_FAILED;
2892 }
2893
2894 static address get_stack_commited_bottom(address bottom, size_t size) {
2895 address nbot = bottom;
2896 address ntop = bottom + size;
2897
2898 size_t page_sz = os::vm_page_size();
2899 unsigned pages = size / page_sz;
2900
2901 unsigned char vec[1];
2902 unsigned imin = 1, imax = pages + 1, imid;
2903 int mincore_return_value = 0;
2904
2905 assert(imin <= imax, "Unexpected page size");
2906
2907 while (imin < imax) {
2908 imid = (imax + imin) / 2;
2909 nbot = ntop - (imid * page_sz);
2910
2911 // Use a trick with mincore to check whether the page is mapped or not.
2912 // mincore sets vec to 1 if page resides in memory and to 0 if page
2913 // is swapped output but if page we are asking for is unmapped
2914 // it returns -1,ENOMEM
2915 mincore_return_value = mincore(nbot, page_sz, vec);
2916
2917 if (mincore_return_value == -1) {
2918 // Page is not mapped go up
2919 // to find first mapped page
2920 if (errno != EAGAIN) {
2921 assert(errno == ENOMEM, "Unexpected mincore errno");
2922 imax = imid;
2923 }
2924 } else {
2925 // Page is mapped go down
2926 // to find first not mapped page
2927 imin = imid + 1;
2928 }
2929 }
2930
2931 nbot = nbot + page_sz;
2932
2933 // Adjust stack bottom one page up if last checked page is not mapped
2934 if (mincore_return_value == -1) {
2935 nbot = nbot + page_sz;
2936 }
2937
2938 return nbot;
2939 }
2940
2941
2942 // Linux uses a growable mapping for the stack, and if the mapping for
2943 // the stack guard pages is not removed when we detach a thread the
2944 // stack cannot grow beyond the pages where the stack guard was
2945 // mapped. If at some point later in the process the stack expands to
2946 // that point, the Linux kernel cannot expand the stack any further
2947 // because the guard pages are in the way, and a segfault occurs.
2948 //
2949 // However, it's essential not to split the stack region by unmapping
2950 // a region (leaving a hole) that's already part of the stack mapping,
2951 // so if the stack mapping has already grown beyond the guard pages at
2952 // the time we create them, we have to truncate the stack mapping.
2953 // So, we need to know the extent of the stack mapping when
2954 // create_stack_guard_pages() is called.
2955
2956 // We only need this for stacks that are growable: at the time of
2957 // writing thread stacks don't use growable mappings (i.e. those
2958 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2959 // only applies to the main thread.
2960
2961 // If the (growable) stack mapping already extends beyond the point
2962 // where we're going to put our guard pages, truncate the mapping at
2963 // that point by munmap()ping it. This ensures that when we later
2964 // munmap() the guard pages we don't leave a hole in the stack
2965 // mapping. This only affects the main/initial thread
2966
2967 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2968 if (os::Linux::is_initial_thread()) {
2969 // As we manually grow stack up to bottom inside create_attached_thread(),
2970 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2971 // we don't need to do anything special.
2972 // Check it first, before calling heavy function.
2973 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2974 unsigned char vec[1];
2975
2976 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2977 // Fallback to slow path on all errors, including EAGAIN
2978 stack_extent = (uintptr_t) get_stack_commited_bottom(
2979 os::Linux::initial_thread_stack_bottom(),
2980 (size_t)addr - stack_extent);
2981 }
2982
2983 if (stack_extent < (uintptr_t)addr) {
2984 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2985 }
2986 }
2987
2988 return os::commit_memory(addr, size, !ExecMem);
2989 }
2990
2991 // If this is a growable mapping, remove the guard pages entirely by
2992 // munmap()ping them. If not, just call uncommit_memory(). This only
2993 // affects the main/initial thread, but guard against future OS changes
2994 // It's safe to always unmap guard pages for initial thread because we
2995 // always place it right after end of the mapped region
2996
2997 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2998 uintptr_t stack_extent, stack_base;
2999
3000 if (os::Linux::is_initial_thread()) {
3001 return ::munmap(addr, size) == 0;
3002 }
3003
3004 return os::uncommit_memory(addr, size);
3005 }
3006
3007 static address _highest_vm_reserved_address = NULL;
3008
3009 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3010 // at 'requested_addr'. If there are existing memory mappings at the same
3011 // location, however, they will be overwritten. If 'fixed' is false,
3012 // 'requested_addr' is only treated as a hint, the return value may or
3013 // may not start from the requested address. Unlike Linux mmap(), this
3014 // function returns NULL to indicate failure.
3015 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3016 char * addr;
3017 int flags;
3018
3019 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3020 if (fixed) {
3021 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3022 flags |= MAP_FIXED;
3023 }
3024
3025 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3026 // touch an uncommitted page. Otherwise, the read/write might
3027 // succeed if we have enough swap space to back the physical page.
3028 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3029 flags, -1, 0);
3030
3031 if (addr != MAP_FAILED) {
3032 // anon_mmap() should only get called during VM initialization,
3033 // don't need lock (actually we can skip locking even it can be called
3034 // from multiple threads, because _highest_vm_reserved_address is just a
3035 // hint about the upper limit of non-stack memory regions.)
3036 if ((address)addr + bytes > _highest_vm_reserved_address) {
3037 _highest_vm_reserved_address = (address)addr + bytes;
3038 }
3039 }
3040
3041 return addr == MAP_FAILED ? NULL : addr;
3042 }
3043
3044 // Don't update _highest_vm_reserved_address, because there might be memory
3045 // regions above addr + size. If so, releasing a memory region only creates
3046 // a hole in the address space, it doesn't help prevent heap-stack collision.
3047 //
3048 static int anon_munmap(char * addr, size_t size) {
3049 return ::munmap(addr, size) == 0;
3050 }
3051
3052 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3053 size_t alignment_hint) {
3054 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3055 }
3056
3057 bool os::pd_release_memory(char* addr, size_t size) {
3058 return anon_munmap(addr, size);
3059 }
3060
3061 static address highest_vm_reserved_address() {
3062 return _highest_vm_reserved_address;
3063 }
3064
3065 static bool linux_mprotect(char* addr, size_t size, int prot) {
3066 // Linux wants the mprotect address argument to be page aligned.
3067 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3068
3069 // According to SUSv3, mprotect() should only be used with mappings
3070 // established by mmap(), and mmap() always maps whole pages. Unaligned
3071 // 'addr' likely indicates problem in the VM (e.g. trying to change
3072 // protection of malloc'ed or statically allocated memory). Check the
3073 // caller if you hit this assert.
3074 assert(addr == bottom, "sanity check");
3075
3076 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3077 return ::mprotect(bottom, size, prot) == 0;
3078 }
3079
3080 // Set protections specified
3081 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3082 bool is_committed) {
3083 unsigned int p = 0;
3084 switch (prot) {
3085 case MEM_PROT_NONE: p = PROT_NONE; break;
3086 case MEM_PROT_READ: p = PROT_READ; break;
3087 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3088 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3089 default:
3090 ShouldNotReachHere();
3091 }
3092 // is_committed is unused.
3093 return linux_mprotect(addr, bytes, p);
3094 }
3095
3096 bool os::guard_memory(char* addr, size_t size) {
3097 return linux_mprotect(addr, size, PROT_NONE);
3098 }
3099
3100 bool os::unguard_memory(char* addr, size_t size) {
3101 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3102 }
3103
3104 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3105 size_t page_size) {
3106 bool result = false;
3107 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3108 MAP_ANONYMOUS|MAP_PRIVATE,
3109 -1, 0);
3110 if (p != MAP_FAILED) {
3111 void *aligned_p = align_ptr_up(p, page_size);
3112
3113 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3114
3115 munmap(p, page_size * 2);
3116 }
3117
3118 if (warn && !result) {
3119 warning("TransparentHugePages is not supported by the operating system.");
3120 }
3121
3122 return result;
3123 }
3124
3125 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3126 bool result = false;
3127 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3128 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3129 -1, 0);
3130
3131 if (p != MAP_FAILED) {
3132 // We don't know if this really is a huge page or not.
3133 FILE *fp = fopen("/proc/self/maps", "r");
3134 if (fp) {
3135 while (!feof(fp)) {
3136 char chars[257];
3137 long x = 0;
3138 if (fgets(chars, sizeof(chars), fp)) {
3139 if (sscanf(chars, "%lx-%*x", &x) == 1
3140 && x == (long)p) {
3141 if (strstr (chars, "hugepage")) {
3142 result = true;
3143 break;
3144 }
3145 }
3146 }
3147 }
3148 fclose(fp);
3149 }
3150 munmap(p, page_size);
3151 }
3152
3153 if (warn && !result) {
3154 warning("HugeTLBFS is not supported by the operating system.");
3155 }
3156
3157 return result;
3158 }
3159
3160 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3161 //
3162 // From the coredump_filter documentation:
3163 //
3164 // - (bit 0) anonymous private memory
3165 // - (bit 1) anonymous shared memory
3166 // - (bit 2) file-backed private memory
3167 // - (bit 3) file-backed shared memory
3168 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3169 // effective only if the bit 2 is cleared)
3170 // - (bit 5) hugetlb private memory
3171 // - (bit 6) hugetlb shared memory
3172 //
3173 static void set_coredump_filter(void) {
3174 FILE *f;
3175 long cdm;
3176
3177 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3178 return;
3179 }
3180
3181 if (fscanf(f, "%lx", &cdm) != 1) {
3182 fclose(f);
3183 return;
3184 }
3185
3186 rewind(f);
3187
3188 if ((cdm & LARGEPAGES_BIT) == 0) {
3189 cdm |= LARGEPAGES_BIT;
3190 fprintf(f, "%#lx", cdm);
3191 }
3192
3193 fclose(f);
3194 }
3195
3196 // Large page support
3197
3198 static size_t _large_page_size = 0;
3199
3200 size_t os::Linux::find_large_page_size() {
3201 size_t large_page_size = 0;
3202
3203 // large_page_size on Linux is used to round up heap size. x86 uses either
3204 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3205 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3206 // page as large as 256M.
3207 //
3208 // Here we try to figure out page size by parsing /proc/meminfo and looking
3209 // for a line with the following format:
3210 // Hugepagesize: 2048 kB
3211 //
3212 // If we can't determine the value (e.g. /proc is not mounted, or the text
3213 // format has been changed), we'll use the largest page size supported by
3214 // the processor.
3215
3216 #ifndef ZERO
3217 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3218 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3219 #endif // ZERO
3220
3221 FILE *fp = fopen("/proc/meminfo", "r");
3222 if (fp) {
3223 while (!feof(fp)) {
3224 int x = 0;
3225 char buf[16];
3226 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3227 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3228 large_page_size = x * K;
3229 break;
3230 }
3231 } else {
3232 // skip to next line
3233 for (;;) {
3234 int ch = fgetc(fp);
3235 if (ch == EOF || ch == (int)'\n') break;
3236 }
3237 }
3238 }
3239 fclose(fp);
3240 }
3241
3242 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3243 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3244 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3245 proper_unit_for_byte_size(large_page_size));
3246 }
3247
3248 return large_page_size;
3249 }
3250
3251 size_t os::Linux::setup_large_page_size() {
3252 _large_page_size = Linux::find_large_page_size();
3253 const size_t default_page_size = (size_t)Linux::page_size();
3254 if (_large_page_size > default_page_size) {
3255 _page_sizes[0] = _large_page_size;
3256 _page_sizes[1] = default_page_size;
3257 _page_sizes[2] = 0;
3258 }
3259
3260 return _large_page_size;
3261 }
3262
3263 bool os::Linux::setup_large_page_type(size_t page_size) {
3264 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3265 FLAG_IS_DEFAULT(UseSHM) &&
3266 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3267
3268 // The type of large pages has not been specified by the user.
3269
3270 // Try UseHugeTLBFS and then UseSHM.
3271 UseHugeTLBFS = UseSHM = true;
3272
3273 // Don't try UseTransparentHugePages since there are known
3274 // performance issues with it turned on. This might change in the future.
3275 UseTransparentHugePages = false;
3276 }
3277
3278 if (UseTransparentHugePages) {
3279 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3280 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3281 UseHugeTLBFS = false;
3282 UseSHM = false;
3283 return true;
3284 }
3285 UseTransparentHugePages = false;
3286 }
3287
3288 if (UseHugeTLBFS) {
3289 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3290 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3291 UseSHM = false;
3292 return true;
3293 }
3294 UseHugeTLBFS = false;
3295 }
3296
3297 return UseSHM;
3298 }
3299
3300 void os::large_page_init() {
3301 if (!UseLargePages &&
3302 !UseTransparentHugePages &&
3303 !UseHugeTLBFS &&
3304 !UseSHM) {
3305 // Not using large pages.
3306 return;
3307 }
3308
3309 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3310 // The user explicitly turned off large pages.
3311 // Ignore the rest of the large pages flags.
3312 UseTransparentHugePages = false;
3313 UseHugeTLBFS = false;
3314 UseSHM = false;
3315 return;
3316 }
3317
3318 size_t large_page_size = Linux::setup_large_page_size();
3319 UseLargePages = Linux::setup_large_page_type(large_page_size);
3320
3321 set_coredump_filter();
3322 }
3323
3324 #ifndef SHM_HUGETLB
3325 #define SHM_HUGETLB 04000
3326 #endif
3327
3328 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3329 char* req_addr, bool exec) {
3330 // "exec" is passed in but not used. Creating the shared image for
3331 // the code cache doesn't have an SHM_X executable permission to check.
3332 assert(UseLargePages && UseSHM, "only for SHM large pages");
3333 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3334
3335 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3336 return NULL; // Fallback to small pages.
3337 }
3338
3339 key_t key = IPC_PRIVATE;
3340 char *addr;
3341
3342 bool warn_on_failure = UseLargePages &&
3343 (!FLAG_IS_DEFAULT(UseLargePages) ||
3344 !FLAG_IS_DEFAULT(UseSHM) ||
3345 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3346 char msg[128];
3347
3348 // Create a large shared memory region to attach to based on size.
3349 // Currently, size is the total size of the heap
3350 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3351 if (shmid == -1) {
3352 // Possible reasons for shmget failure:
3353 // 1. shmmax is too small for Java heap.
3354 // > check shmmax value: cat /proc/sys/kernel/shmmax
3355 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3356 // 2. not enough large page memory.
3357 // > check available large pages: cat /proc/meminfo
3358 // > increase amount of large pages:
3359 // echo new_value > /proc/sys/vm/nr_hugepages
3360 // Note 1: different Linux may use different name for this property,
3361 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3362 // Note 2: it's possible there's enough physical memory available but
3363 // they are so fragmented after a long run that they can't
3364 // coalesce into large pages. Try to reserve large pages when
3365 // the system is still "fresh".
3366 if (warn_on_failure) {
3367 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3368 warning("%s", msg);
3369 }
3370 return NULL;
3371 }
3372
3373 // attach to the region
3374 addr = (char*)shmat(shmid, req_addr, 0);
3375 int err = errno;
3376
3377 // Remove shmid. If shmat() is successful, the actual shared memory segment
3378 // will be deleted when it's detached by shmdt() or when the process
3379 // terminates. If shmat() is not successful this will remove the shared
3380 // segment immediately.
3381 shmctl(shmid, IPC_RMID, NULL);
3382
3383 if ((intptr_t)addr == -1) {
3384 if (warn_on_failure) {
3385 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3386 warning("%s", msg);
3387 }
3388 return NULL;
3389 }
3390
3391 return addr;
3392 }
3393
3394 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3395 int error) {
3396 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3397
3398 bool warn_on_failure = UseLargePages &&
3399 (!FLAG_IS_DEFAULT(UseLargePages) ||
3400 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3401 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3402
3403 if (warn_on_failure) {
3404 char msg[128];
3405 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3406 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3407 warning("%s", msg);
3408 }
3409 }
3410
3411 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3412 char* req_addr,
3413 bool exec) {
3414 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3415 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3416 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3417
3418 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3419 char* addr = (char*)::mmap(req_addr, bytes, prot,
3420 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3421 -1, 0);
3422
3423 if (addr == MAP_FAILED) {
3424 warn_on_large_pages_failure(req_addr, bytes, errno);
3425 return NULL;
3426 }
3427
3428 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3429
3430 return addr;
3431 }
3432
3433 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3434 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3435 // (req_addr != NULL) or with a given alignment.
3436 // - bytes shall be a multiple of alignment.
3437 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3438 // - alignment sets the alignment at which memory shall be allocated.
3439 // It must be a multiple of allocation granularity.
3440 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3441 // req_addr or NULL.
3442 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3443
3444 size_t extra_size = bytes;
3445 if (req_addr == NULL && alignment > 0) {
3446 extra_size += alignment;
3447 }
3448
3449 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3450 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3451 -1, 0);
3452 if (start == MAP_FAILED) {
3453 start = NULL;
3454 } else {
3455 if (req_addr != NULL) {
3456 if (start != req_addr) {
3457 ::munmap(start, extra_size);
3458 start = NULL;
3459 }
3460 } else {
3461 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3462 char* const end_aligned = start_aligned + bytes;
3463 char* const end = start + extra_size;
3464 if (start_aligned > start) {
3465 ::munmap(start, start_aligned - start);
3466 }
3467 if (end_aligned < end) {
3468 ::munmap(end_aligned, end - end_aligned);
3469 }
3470 start = start_aligned;
3471 }
3472 }
3473 return start;
3474
3475 }
3476
3477 // Reserve memory using mmap(MAP_HUGETLB).
3478 // - bytes shall be a multiple of alignment.
3479 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3480 // - alignment sets the alignment at which memory shall be allocated.
3481 // It must be a multiple of allocation granularity.
3482 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3483 // req_addr or NULL.
3484 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3485 size_t alignment,
3486 char* req_addr,
3487 bool exec) {
3488 size_t large_page_size = os::large_page_size();
3489 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3490
3491 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3492 assert(is_size_aligned(bytes, alignment), "Must be");
3493
3494 // First reserve - but not commit - the address range in small pages.
3495 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3496
3497 if (start == NULL) {
3498 return NULL;
3499 }
3500
3501 assert(is_ptr_aligned(start, alignment), "Must be");
3502
3503 char* end = start + bytes;
3504
3505 // Find the regions of the allocated chunk that can be promoted to large pages.
3506 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3507 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3508
3509 size_t lp_bytes = lp_end - lp_start;
3510
3511 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3512
3513 if (lp_bytes == 0) {
3514 // The mapped region doesn't even span the start and the end of a large page.
3515 // Fall back to allocate a non-special area.
3516 ::munmap(start, end - start);
3517 return NULL;
3518 }
3519
3520 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3521
3522 void* result;
3523
3524 // Commit small-paged leading area.
3525 if (start != lp_start) {
3526 result = ::mmap(start, lp_start - start, prot,
3527 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3528 -1, 0);
3529 if (result == MAP_FAILED) {
3530 ::munmap(lp_start, end - lp_start);
3531 return NULL;
3532 }
3533 }
3534
3535 // Commit large-paged area.
3536 result = ::mmap(lp_start, lp_bytes, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3538 -1, 0);
3539 if (result == MAP_FAILED) {
3540 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3541 // If the mmap above fails, the large pages region will be unmapped and we
3542 // have regions before and after with small pages. Release these regions.
3543 //
3544 // | mapped | unmapped | mapped |
3545 // ^ ^ ^ ^
3546 // start lp_start lp_end end
3547 //
3548 ::munmap(start, lp_start - start);
3549 ::munmap(lp_end, end - lp_end);
3550 return NULL;
3551 }
3552
3553 // Commit small-paged trailing area.
3554 if (lp_end != end) {
3555 result = ::mmap(lp_end, end - lp_end, prot,
3556 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3557 -1, 0);
3558 if (result == MAP_FAILED) {
3559 ::munmap(start, lp_end - start);
3560 return NULL;
3561 }
3562 }
3563
3564 return start;
3565 }
3566
3567 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3568 size_t alignment,
3569 char* req_addr,
3570 bool exec) {
3571 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3572 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3573 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3574 assert(is_power_of_2(os::large_page_size()), "Must be");
3575 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3576
3577 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3578 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3579 } else {
3580 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3581 }
3582 }
3583
3584 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3585 char* req_addr, bool exec) {
3586 assert(UseLargePages, "only for large pages");
3587
3588 char* addr;
3589 if (UseSHM) {
3590 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3591 } else {
3592 assert(UseHugeTLBFS, "must be");
3593 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3594 }
3595
3596 if (addr != NULL) {
3597 if (UseNUMAInterleaving) {
3598 numa_make_global(addr, bytes);
3599 }
3600
3601 // The memory is committed
3602 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3603 }
3604
3605 return addr;
3606 }
3607
3608 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3609 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3610 return shmdt(base) == 0;
3611 }
3612
3613 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3614 return pd_release_memory(base, bytes);
3615 }
3616
3617 bool os::release_memory_special(char* base, size_t bytes) {
3618 bool res;
3619 if (MemTracker::tracking_level() > NMT_minimal) {
3620 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3621 res = os::Linux::release_memory_special_impl(base, bytes);
3622 if (res) {
3623 tkr.record((address)base, bytes);
3624 }
3625
3626 } else {
3627 res = os::Linux::release_memory_special_impl(base, bytes);
3628 }
3629 return res;
3630 }
3631
3632 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3633 assert(UseLargePages, "only for large pages");
3634 bool res;
3635
3636 if (UseSHM) {
3637 res = os::Linux::release_memory_special_shm(base, bytes);
3638 } else {
3639 assert(UseHugeTLBFS, "must be");
3640 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3641 }
3642 return res;
3643 }
3644
3645 size_t os::large_page_size() {
3646 return _large_page_size;
3647 }
3648
3649 // With SysV SHM the entire memory region must be allocated as shared
3650 // memory.
3651 // HugeTLBFS allows application to commit large page memory on demand.
3652 // However, when committing memory with HugeTLBFS fails, the region
3653 // that was supposed to be committed will lose the old reservation
3654 // and allow other threads to steal that memory region. Because of this
3655 // behavior we can't commit HugeTLBFS memory.
3656 bool os::can_commit_large_page_memory() {
3657 return UseTransparentHugePages;
3658 }
3659
3660 bool os::can_execute_large_page_memory() {
3661 return UseTransparentHugePages || UseHugeTLBFS;
3662 }
3663
3664 // Reserve memory at an arbitrary address, only if that area is
3665 // available (and not reserved for something else).
3666
3667 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3668 const int max_tries = 10;
3669 char* base[max_tries];
3670 size_t size[max_tries];
3671 const size_t gap = 0x000000;
3672
3673 // Assert only that the size is a multiple of the page size, since
3674 // that's all that mmap requires, and since that's all we really know
3675 // about at this low abstraction level. If we need higher alignment,
3676 // we can either pass an alignment to this method or verify alignment
3677 // in one of the methods further up the call chain. See bug 5044738.
3678 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3679
3680 // Repeatedly allocate blocks until the block is allocated at the
3681 // right spot. Give up after max_tries. Note that reserve_memory() will
3682 // automatically update _highest_vm_reserved_address if the call is
3683 // successful. The variable tracks the highest memory address every reserved
3684 // by JVM. It is used to detect heap-stack collision if running with
3685 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3686 // space than needed, it could confuse the collision detecting code. To
3687 // solve the problem, save current _highest_vm_reserved_address and
3688 // calculate the correct value before return.
3689 address old_highest = _highest_vm_reserved_address;
3690
3691 // Linux mmap allows caller to pass an address as hint; give it a try first,
3692 // if kernel honors the hint then we can return immediately.
3693 char * addr = anon_mmap(requested_addr, bytes, false);
3694 if (addr == requested_addr) {
3695 return requested_addr;
3696 }
3697
3698 if (addr != NULL) {
3699 // mmap() is successful but it fails to reserve at the requested address
3700 anon_munmap(addr, bytes);
3701 }
3702
3703 int i;
3704 for (i = 0; i < max_tries; ++i) {
3705 base[i] = reserve_memory(bytes);
3706
3707 if (base[i] != NULL) {
3708 // Is this the block we wanted?
3709 if (base[i] == requested_addr) {
3710 size[i] = bytes;
3711 break;
3712 }
3713
3714 // Does this overlap the block we wanted? Give back the overlapped
3715 // parts and try again.
3716
3717 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3718 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3719 unmap_memory(base[i], top_overlap);
3720 base[i] += top_overlap;
3721 size[i] = bytes - top_overlap;
3722 } else {
3723 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3724 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3725 unmap_memory(requested_addr, bottom_overlap);
3726 size[i] = bytes - bottom_overlap;
3727 } else {
3728 size[i] = bytes;
3729 }
3730 }
3731 }
3732 }
3733
3734 // Give back the unused reserved pieces.
3735
3736 for (int j = 0; j < i; ++j) {
3737 if (base[j] != NULL) {
3738 unmap_memory(base[j], size[j]);
3739 }
3740 }
3741
3742 if (i < max_tries) {
3743 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3744 return requested_addr;
3745 } else {
3746 _highest_vm_reserved_address = old_highest;
3747 return NULL;
3748 }
3749 }
3750
3751 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3752 return ::read(fd, buf, nBytes);
3753 }
3754
3755 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3756 return ::pread(fd, buf, nBytes, offset);
3757 }
3758
3759 // Short sleep, direct OS call.
3760 //
3761 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3762 // sched_yield(2) will actually give up the CPU:
3763 //
3764 // * Alone on this pariticular CPU, keeps running.
3765 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3766 // (pre 2.6.39).
3767 //
3768 // So calling this with 0 is an alternative.
3769 //
3770 void os::naked_short_sleep(jlong ms) {
3771 struct timespec req;
3772
3773 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3774 req.tv_sec = 0;
3775 if (ms > 0) {
3776 req.tv_nsec = (ms % 1000) * 1000000;
3777 } else {
3778 req.tv_nsec = 1;
3779 }
3780
3781 nanosleep(&req, NULL);
3782
3783 return;
3784 }
3785
3786 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3787 void os::infinite_sleep() {
3788 while (true) { // sleep forever ...
3789 ::sleep(100); // ... 100 seconds at a time
3790 }
3791 }
3792
3793 // Used to convert frequent JVM_Yield() to nops
3794 bool os::dont_yield() {
3795 return DontYieldALot;
3796 }
3797
3798 void os::naked_yield() {
3799 sched_yield();
3800 }
3801
3802 ////////////////////////////////////////////////////////////////////////////////
3803 // thread priority support
3804
3805 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3806 // only supports dynamic priority, static priority must be zero. For real-time
3807 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3808 // However, for large multi-threaded applications, SCHED_RR is not only slower
3809 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3810 // of 5 runs - Sep 2005).
3811 //
3812 // The following code actually changes the niceness of kernel-thread/LWP. It
3813 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3814 // not the entire user process, and user level threads are 1:1 mapped to kernel
3815 // threads. It has always been the case, but could change in the future. For
3816 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3817 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3818
3819 int os::java_to_os_priority[CriticalPriority + 1] = {
3820 19, // 0 Entry should never be used
3821
3822 4, // 1 MinPriority
3823 3, // 2
3824 2, // 3
3825
3826 1, // 4
3827 0, // 5 NormPriority
3828 -1, // 6
3829
3830 -2, // 7
3831 -3, // 8
3832 -4, // 9 NearMaxPriority
3833
3834 -5, // 10 MaxPriority
3835
3836 -5 // 11 CriticalPriority
3837 };
3838
3839 static int prio_init() {
3840 if (ThreadPriorityPolicy == 1) {
3841 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3842 // if effective uid is not root. Perhaps, a more elegant way of doing
3843 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3844 if (geteuid() != 0) {
3845 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3846 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3847 }
3848 ThreadPriorityPolicy = 0;
3849 }
3850 }
3851 if (UseCriticalJavaThreadPriority) {
3852 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3853 }
3854 return 0;
3855 }
3856
3857 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3858 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3859
3860 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3861 return (ret == 0) ? OS_OK : OS_ERR;
3862 }
3863
3864 OSReturn os::get_native_priority(const Thread* const thread,
3865 int *priority_ptr) {
3866 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3867 *priority_ptr = java_to_os_priority[NormPriority];
3868 return OS_OK;
3869 }
3870
3871 errno = 0;
3872 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3873 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3874 }
3875
3876 // Hint to the underlying OS that a task switch would not be good.
3877 // Void return because it's a hint and can fail.
3878 void os::hint_no_preempt() {}
3879
3880 ////////////////////////////////////////////////////////////////////////////////
3881 // suspend/resume support
3882
3883 // the low-level signal-based suspend/resume support is a remnant from the
3884 // old VM-suspension that used to be for java-suspension, safepoints etc,
3885 // within hotspot. Now there is a single use-case for this:
3886 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3887 // that runs in the watcher thread.
3888 // The remaining code is greatly simplified from the more general suspension
3889 // code that used to be used.
3890 //
3891 // The protocol is quite simple:
3892 // - suspend:
3893 // - sends a signal to the target thread
3894 // - polls the suspend state of the osthread using a yield loop
3895 // - target thread signal handler (SR_handler) sets suspend state
3896 // and blocks in sigsuspend until continued
3897 // - resume:
3898 // - sets target osthread state to continue
3899 // - sends signal to end the sigsuspend loop in the SR_handler
3900 //
3901 // Note that the SR_lock plays no role in this suspend/resume protocol.
3902
3903 static void resume_clear_context(OSThread *osthread) {
3904 osthread->set_ucontext(NULL);
3905 osthread->set_siginfo(NULL);
3906 }
3907
3908 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3909 ucontext_t* context) {
3910 osthread->set_ucontext(context);
3911 osthread->set_siginfo(siginfo);
3912 }
3913
3914 // Handler function invoked when a thread's execution is suspended or
3915 // resumed. We have to be careful that only async-safe functions are
3916 // called here (Note: most pthread functions are not async safe and
3917 // should be avoided.)
3918 //
3919 // Note: sigwait() is a more natural fit than sigsuspend() from an
3920 // interface point of view, but sigwait() prevents the signal hander
3921 // from being run. libpthread would get very confused by not having
3922 // its signal handlers run and prevents sigwait()'s use with the
3923 // mutex granting granting signal.
3924 //
3925 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3926 //
3927 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3928 // Save and restore errno to avoid confusing native code with EINTR
3929 // after sigsuspend.
3930 int old_errno = errno;
3931
3932 Thread* thread = Thread::current();
3933 OSThread* osthread = thread->osthread();
3934 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3935
3936 os::SuspendResume::State current = osthread->sr.state();
3937 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3938 suspend_save_context(osthread, siginfo, context);
3939
3940 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3941 os::SuspendResume::State state = osthread->sr.suspended();
3942 if (state == os::SuspendResume::SR_SUSPENDED) {
3943 sigset_t suspend_set; // signals for sigsuspend()
3944
3945 // get current set of blocked signals and unblock resume signal
3946 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3947 sigdelset(&suspend_set, SR_signum);
3948
3949 sr_semaphore.signal();
3950 // wait here until we are resumed
3951 while (1) {
3952 sigsuspend(&suspend_set);
3953
3954 os::SuspendResume::State result = osthread->sr.running();
3955 if (result == os::SuspendResume::SR_RUNNING) {
3956 sr_semaphore.signal();
3957 break;
3958 }
3959 }
3960
3961 } else if (state == os::SuspendResume::SR_RUNNING) {
3962 // request was cancelled, continue
3963 } else {
3964 ShouldNotReachHere();
3965 }
3966
3967 resume_clear_context(osthread);
3968 } else if (current == os::SuspendResume::SR_RUNNING) {
3969 // request was cancelled, continue
3970 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3971 // ignore
3972 } else {
3973 // ignore
3974 }
3975
3976 errno = old_errno;
3977 }
3978
3979
3980 static int SR_initialize() {
3981 struct sigaction act;
3982 char *s;
3983 // Get signal number to use for suspend/resume
3984 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3985 int sig = ::strtol(s, 0, 10);
3986 if (sig > 0 || sig < _NSIG) {
3987 SR_signum = sig;
3988 }
3989 }
3990
3991 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3992 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3993
3994 sigemptyset(&SR_sigset);
3995 sigaddset(&SR_sigset, SR_signum);
3996
3997 // Set up signal handler for suspend/resume
3998 act.sa_flags = SA_RESTART|SA_SIGINFO;
3999 act.sa_handler = (void (*)(int)) SR_handler;
4000
4001 // SR_signum is blocked by default.
4002 // 4528190 - We also need to block pthread restart signal (32 on all
4003 // supported Linux platforms). Note that LinuxThreads need to block
4004 // this signal for all threads to work properly. So we don't have
4005 // to use hard-coded signal number when setting up the mask.
4006 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4007
4008 if (sigaction(SR_signum, &act, 0) == -1) {
4009 return -1;
4010 }
4011
4012 // Save signal flag
4013 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4014 return 0;
4015 }
4016
4017 static int sr_notify(OSThread* osthread) {
4018 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4019 assert_status(status == 0, status, "pthread_kill");
4020 return status;
4021 }
4022
4023 // "Randomly" selected value for how long we want to spin
4024 // before bailing out on suspending a thread, also how often
4025 // we send a signal to a thread we want to resume
4026 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4027 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4028
4029 // returns true on success and false on error - really an error is fatal
4030 // but this seems the normal response to library errors
4031 static bool do_suspend(OSThread* osthread) {
4032 assert(osthread->sr.is_running(), "thread should be running");
4033 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4034
4035 // mark as suspended and send signal
4036 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4037 // failed to switch, state wasn't running?
4038 ShouldNotReachHere();
4039 return false;
4040 }
4041
4042 if (sr_notify(osthread) != 0) {
4043 ShouldNotReachHere();
4044 }
4045
4046 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4047 while (true) {
4048 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4049 break;
4050 } else {
4051 // timeout
4052 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4053 if (cancelled == os::SuspendResume::SR_RUNNING) {
4054 return false;
4055 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4056 // make sure that we consume the signal on the semaphore as well
4057 sr_semaphore.wait();
4058 break;
4059 } else {
4060 ShouldNotReachHere();
4061 return false;
4062 }
4063 }
4064 }
4065
4066 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4067 return true;
4068 }
4069
4070 static void do_resume(OSThread* osthread) {
4071 assert(osthread->sr.is_suspended(), "thread should be suspended");
4072 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4073
4074 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4075 // failed to switch to WAKEUP_REQUEST
4076 ShouldNotReachHere();
4077 return;
4078 }
4079
4080 while (true) {
4081 if (sr_notify(osthread) == 0) {
4082 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4083 if (osthread->sr.is_running()) {
4084 return;
4085 }
4086 }
4087 } else {
4088 ShouldNotReachHere();
4089 }
4090 }
4091
4092 guarantee(osthread->sr.is_running(), "Must be running!");
4093 }
4094
4095 ///////////////////////////////////////////////////////////////////////////////////
4096 // signal handling (except suspend/resume)
4097
4098 // This routine may be used by user applications as a "hook" to catch signals.
4099 // The user-defined signal handler must pass unrecognized signals to this
4100 // routine, and if it returns true (non-zero), then the signal handler must
4101 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4102 // routine will never retun false (zero), but instead will execute a VM panic
4103 // routine kill the process.
4104 //
4105 // If this routine returns false, it is OK to call it again. This allows
4106 // the user-defined signal handler to perform checks either before or after
4107 // the VM performs its own checks. Naturally, the user code would be making
4108 // a serious error if it tried to handle an exception (such as a null check
4109 // or breakpoint) that the VM was generating for its own correct operation.
4110 //
4111 // This routine may recognize any of the following kinds of signals:
4112 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4113 // It should be consulted by handlers for any of those signals.
4114 //
4115 // The caller of this routine must pass in the three arguments supplied
4116 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4117 // field of the structure passed to sigaction(). This routine assumes that
4118 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4119 //
4120 // Note that the VM will print warnings if it detects conflicting signal
4121 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4122 //
4123 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4124 siginfo_t* siginfo,
4125 void* ucontext,
4126 int abort_if_unrecognized);
4127
4128 void signalHandler(int sig, siginfo_t* info, void* uc) {
4129 assert(info != NULL && uc != NULL, "it must be old kernel");
4130 int orig_errno = errno; // Preserve errno value over signal handler.
4131 JVM_handle_linux_signal(sig, info, uc, true);
4132 errno = orig_errno;
4133 }
4134
4135
4136 // This boolean allows users to forward their own non-matching signals
4137 // to JVM_handle_linux_signal, harmlessly.
4138 bool os::Linux::signal_handlers_are_installed = false;
4139
4140 // For signal-chaining
4141 struct sigaction os::Linux::sigact[MAXSIGNUM];
4142 unsigned int os::Linux::sigs = 0;
4143 bool os::Linux::libjsig_is_loaded = false;
4144 typedef struct sigaction *(*get_signal_t)(int);
4145 get_signal_t os::Linux::get_signal_action = NULL;
4146
4147 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4148 struct sigaction *actp = NULL;
4149
4150 if (libjsig_is_loaded) {
4151 // Retrieve the old signal handler from libjsig
4152 actp = (*get_signal_action)(sig);
4153 }
4154 if (actp == NULL) {
4155 // Retrieve the preinstalled signal handler from jvm
4156 actp = get_preinstalled_handler(sig);
4157 }
4158
4159 return actp;
4160 }
4161
4162 static bool call_chained_handler(struct sigaction *actp, int sig,
4163 siginfo_t *siginfo, void *context) {
4164 // Call the old signal handler
4165 if (actp->sa_handler == SIG_DFL) {
4166 // It's more reasonable to let jvm treat it as an unexpected exception
4167 // instead of taking the default action.
4168 return false;
4169 } else if (actp->sa_handler != SIG_IGN) {
4170 if ((actp->sa_flags & SA_NODEFER) == 0) {
4171 // automaticlly block the signal
4172 sigaddset(&(actp->sa_mask), sig);
4173 }
4174
4175 sa_handler_t hand;
4176 sa_sigaction_t sa;
4177 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4178 // retrieve the chained handler
4179 if (siginfo_flag_set) {
4180 sa = actp->sa_sigaction;
4181 } else {
4182 hand = actp->sa_handler;
4183 }
4184
4185 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4186 actp->sa_handler = SIG_DFL;
4187 }
4188
4189 // try to honor the signal mask
4190 sigset_t oset;
4191 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4192
4193 // call into the chained handler
4194 if (siginfo_flag_set) {
4195 (*sa)(sig, siginfo, context);
4196 } else {
4197 (*hand)(sig);
4198 }
4199
4200 // restore the signal mask
4201 pthread_sigmask(SIG_SETMASK, &oset, 0);
4202 }
4203 // Tell jvm's signal handler the signal is taken care of.
4204 return true;
4205 }
4206
4207 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4208 bool chained = false;
4209 // signal-chaining
4210 if (UseSignalChaining) {
4211 struct sigaction *actp = get_chained_signal_action(sig);
4212 if (actp != NULL) {
4213 chained = call_chained_handler(actp, sig, siginfo, context);
4214 }
4215 }
4216 return chained;
4217 }
4218
4219 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4220 if ((((unsigned int)1 << sig) & sigs) != 0) {
4221 return &sigact[sig];
4222 }
4223 return NULL;
4224 }
4225
4226 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4227 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4228 sigact[sig] = oldAct;
4229 sigs |= (unsigned int)1 << sig;
4230 }
4231
4232 // for diagnostic
4233 int os::Linux::sigflags[MAXSIGNUM];
4234
4235 int os::Linux::get_our_sigflags(int sig) {
4236 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4237 return sigflags[sig];
4238 }
4239
4240 void os::Linux::set_our_sigflags(int sig, int flags) {
4241 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4242 sigflags[sig] = flags;
4243 }
4244
4245 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4246 // Check for overwrite.
4247 struct sigaction oldAct;
4248 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4249
4250 void* oldhand = oldAct.sa_sigaction
4251 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4252 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4253 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4254 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4255 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4256 if (AllowUserSignalHandlers || !set_installed) {
4257 // Do not overwrite; user takes responsibility to forward to us.
4258 return;
4259 } else if (UseSignalChaining) {
4260 // save the old handler in jvm
4261 save_preinstalled_handler(sig, oldAct);
4262 // libjsig also interposes the sigaction() call below and saves the
4263 // old sigaction on it own.
4264 } else {
4265 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4266 "%#lx for signal %d.", (long)oldhand, sig));
4267 }
4268 }
4269
4270 struct sigaction sigAct;
4271 sigfillset(&(sigAct.sa_mask));
4272 sigAct.sa_handler = SIG_DFL;
4273 if (!set_installed) {
4274 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4275 } else {
4276 sigAct.sa_sigaction = signalHandler;
4277 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4278 }
4279 // Save flags, which are set by ours
4280 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4281 sigflags[sig] = sigAct.sa_flags;
4282
4283 int ret = sigaction(sig, &sigAct, &oldAct);
4284 assert(ret == 0, "check");
4285
4286 void* oldhand2 = oldAct.sa_sigaction
4287 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4288 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4289 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4290 }
4291
4292 // install signal handlers for signals that HotSpot needs to
4293 // handle in order to support Java-level exception handling.
4294
4295 void os::Linux::install_signal_handlers() {
4296 if (!signal_handlers_are_installed) {
4297 signal_handlers_are_installed = true;
4298
4299 // signal-chaining
4300 typedef void (*signal_setting_t)();
4301 signal_setting_t begin_signal_setting = NULL;
4302 signal_setting_t end_signal_setting = NULL;
4303 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4304 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4305 if (begin_signal_setting != NULL) {
4306 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4307 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4308 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4309 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4310 libjsig_is_loaded = true;
4311 assert(UseSignalChaining, "should enable signal-chaining");
4312 }
4313 if (libjsig_is_loaded) {
4314 // Tell libjsig jvm is setting signal handlers
4315 (*begin_signal_setting)();
4316 }
4317
4318 set_signal_handler(SIGSEGV, true);
4319 set_signal_handler(SIGPIPE, true);
4320 set_signal_handler(SIGBUS, true);
4321 set_signal_handler(SIGILL, true);
4322 set_signal_handler(SIGFPE, true);
4323 #if defined(PPC64)
4324 set_signal_handler(SIGTRAP, true);
4325 #endif
4326 set_signal_handler(SIGXFSZ, true);
4327
4328 if (libjsig_is_loaded) {
4329 // Tell libjsig jvm finishes setting signal handlers
4330 (*end_signal_setting)();
4331 }
4332
4333 // We don't activate signal checker if libjsig is in place, we trust ourselves
4334 // and if UserSignalHandler is installed all bets are off.
4335 // Log that signal checking is off only if -verbose:jni is specified.
4336 if (CheckJNICalls) {
4337 if (libjsig_is_loaded) {
4338 if (PrintJNIResolving) {
4339 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4340 }
4341 check_signals = false;
4342 }
4343 if (AllowUserSignalHandlers) {
4344 if (PrintJNIResolving) {
4345 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4346 }
4347 check_signals = false;
4348 }
4349 }
4350 }
4351 }
4352
4353 // This is the fastest way to get thread cpu time on Linux.
4354 // Returns cpu time (user+sys) for any thread, not only for current.
4355 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4356 // It might work on 2.6.10+ with a special kernel/glibc patch.
4357 // For reference, please, see IEEE Std 1003.1-2004:
4358 // http://www.unix.org/single_unix_specification
4359
4360 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4361 struct timespec tp;
4362 int rc = os::Linux::clock_gettime(clockid, &tp);
4363 assert(rc == 0, "clock_gettime is expected to return 0 code");
4364
4365 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4366 }
4367
4368 /////
4369 // glibc on Linux platform uses non-documented flag
4370 // to indicate, that some special sort of signal
4371 // trampoline is used.
4372 // We will never set this flag, and we should
4373 // ignore this flag in our diagnostic
4374 #ifdef SIGNIFICANT_SIGNAL_MASK
4375 #undef SIGNIFICANT_SIGNAL_MASK
4376 #endif
4377 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4378
4379 static const char* get_signal_handler_name(address handler,
4380 char* buf, int buflen) {
4381 int offset;
4382 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4383 if (found) {
4384 // skip directory names
4385 const char *p1, *p2;
4386 p1 = buf;
4387 size_t len = strlen(os::file_separator());
4388 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4389 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4390 } else {
4391 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4392 }
4393 return buf;
4394 }
4395
4396 static void print_signal_handler(outputStream* st, int sig,
4397 char* buf, size_t buflen) {
4398 struct sigaction sa;
4399
4400 sigaction(sig, NULL, &sa);
4401
4402 // See comment for SIGNIFICANT_SIGNAL_MASK define
4403 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4404
4405 st->print("%s: ", os::exception_name(sig, buf, buflen));
4406
4407 address handler = (sa.sa_flags & SA_SIGINFO)
4408 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4409 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4410
4411 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4412 st->print("SIG_DFL");
4413 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4414 st->print("SIG_IGN");
4415 } else {
4416 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4417 }
4418
4419 st->print(", sa_mask[0]=");
4420 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4421
4422 address rh = VMError::get_resetted_sighandler(sig);
4423 // May be, handler was resetted by VMError?
4424 if (rh != NULL) {
4425 handler = rh;
4426 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4427 }
4428
4429 st->print(", sa_flags=");
4430 os::Posix::print_sa_flags(st, sa.sa_flags);
4431
4432 // Check: is it our handler?
4433 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4434 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4435 // It is our signal handler
4436 // check for flags, reset system-used one!
4437 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4438 st->print(
4439 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4440 os::Linux::get_our_sigflags(sig));
4441 }
4442 }
4443 st->cr();
4444 }
4445
4446
4447 #define DO_SIGNAL_CHECK(sig) \
4448 do { \
4449 if (!sigismember(&check_signal_done, sig)) { \
4450 os::Linux::check_signal_handler(sig); \
4451 } \
4452 } while (0)
4453
4454 // This method is a periodic task to check for misbehaving JNI applications
4455 // under CheckJNI, we can add any periodic checks here
4456
4457 void os::run_periodic_checks() {
4458 if (check_signals == false) return;
4459
4460 // SEGV and BUS if overridden could potentially prevent
4461 // generation of hs*.log in the event of a crash, debugging
4462 // such a case can be very challenging, so we absolutely
4463 // check the following for a good measure:
4464 DO_SIGNAL_CHECK(SIGSEGV);
4465 DO_SIGNAL_CHECK(SIGILL);
4466 DO_SIGNAL_CHECK(SIGFPE);
4467 DO_SIGNAL_CHECK(SIGBUS);
4468 DO_SIGNAL_CHECK(SIGPIPE);
4469 DO_SIGNAL_CHECK(SIGXFSZ);
4470 #if defined(PPC64)
4471 DO_SIGNAL_CHECK(SIGTRAP);
4472 #endif
4473
4474 // ReduceSignalUsage allows the user to override these handlers
4475 // see comments at the very top and jvm_solaris.h
4476 if (!ReduceSignalUsage) {
4477 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4478 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4479 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4480 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4481 }
4482
4483 DO_SIGNAL_CHECK(SR_signum);
4484 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4485 }
4486
4487 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4488
4489 static os_sigaction_t os_sigaction = NULL;
4490
4491 void os::Linux::check_signal_handler(int sig) {
4492 char buf[O_BUFLEN];
4493 address jvmHandler = NULL;
4494
4495
4496 struct sigaction act;
4497 if (os_sigaction == NULL) {
4498 // only trust the default sigaction, in case it has been interposed
4499 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4500 if (os_sigaction == NULL) return;
4501 }
4502
4503 os_sigaction(sig, (struct sigaction*)NULL, &act);
4504
4505
4506 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4507
4508 address thisHandler = (act.sa_flags & SA_SIGINFO)
4509 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4510 : CAST_FROM_FN_PTR(address, act.sa_handler);
4511
4512
4513 switch (sig) {
4514 case SIGSEGV:
4515 case SIGBUS:
4516 case SIGFPE:
4517 case SIGPIPE:
4518 case SIGILL:
4519 case SIGXFSZ:
4520 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4521 break;
4522
4523 case SHUTDOWN1_SIGNAL:
4524 case SHUTDOWN2_SIGNAL:
4525 case SHUTDOWN3_SIGNAL:
4526 case BREAK_SIGNAL:
4527 jvmHandler = (address)user_handler();
4528 break;
4529
4530 case INTERRUPT_SIGNAL:
4531 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4532 break;
4533
4534 default:
4535 if (sig == SR_signum) {
4536 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4537 } else {
4538 return;
4539 }
4540 break;
4541 }
4542
4543 if (thisHandler != jvmHandler) {
4544 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4545 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4546 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4547 // No need to check this sig any longer
4548 sigaddset(&check_signal_done, sig);
4549 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4550 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4551 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4552 exception_name(sig, buf, O_BUFLEN));
4553 }
4554 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4555 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4556 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4557 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4558 // No need to check this sig any longer
4559 sigaddset(&check_signal_done, sig);
4560 }
4561
4562 // Dump all the signal
4563 if (sigismember(&check_signal_done, sig)) {
4564 print_signal_handlers(tty, buf, O_BUFLEN);
4565 }
4566 }
4567
4568 extern void report_error(char* file_name, int line_no, char* title,
4569 char* format, ...);
4570
4571 extern bool signal_name(int signo, char* buf, size_t len);
4572
4573 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4574 if (0 < exception_code && exception_code <= SIGRTMAX) {
4575 // signal
4576 if (!signal_name(exception_code, buf, size)) {
4577 jio_snprintf(buf, size, "SIG%d", exception_code);
4578 }
4579 return buf;
4580 } else {
4581 return NULL;
4582 }
4583 }
4584
4585 // this is called _before_ the most of global arguments have been parsed
4586 void os::init(void) {
4587 char dummy; // used to get a guess on initial stack address
4588 // first_hrtime = gethrtime();
4589
4590 // With LinuxThreads the JavaMain thread pid (primordial thread)
4591 // is different than the pid of the java launcher thread.
4592 // So, on Linux, the launcher thread pid is passed to the VM
4593 // via the sun.java.launcher.pid property.
4594 // Use this property instead of getpid() if it was correctly passed.
4595 // See bug 6351349.
4596 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4597
4598 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4599
4600 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4601
4602 init_random(1234567);
4603
4604 ThreadCritical::initialize();
4605
4606 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4607 if (Linux::page_size() == -1) {
4608 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4609 strerror(errno)));
4610 }
4611 init_page_sizes((size_t) Linux::page_size());
4612
4613 Linux::initialize_system_info();
4614
4615 // main_thread points to the aboriginal thread
4616 Linux::_main_thread = pthread_self();
4617
4618 Linux::clock_init();
4619 initial_time_count = javaTimeNanos();
4620
4621 // pthread_condattr initialization for monotonic clock
4622 int status;
4623 pthread_condattr_t* _condattr = os::Linux::condAttr();
4624 if ((status = pthread_condattr_init(_condattr)) != 0) {
4625 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4626 }
4627 // Only set the clock if CLOCK_MONOTONIC is available
4628 if (os::supports_monotonic_clock()) {
4629 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4630 if (status == EINVAL) {
4631 warning("Unable to use monotonic clock with relative timed-waits" \
4632 " - changes to the time-of-day clock may have adverse affects");
4633 } else {
4634 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4635 }
4636 }
4637 }
4638 // else it defaults to CLOCK_REALTIME
4639
4640 // If the pagesize of the VM is greater than 8K determine the appropriate
4641 // number of initial guard pages. The user can change this with the
4642 // command line arguments, if needed.
4643 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4644 StackYellowPages = 1;
4645 StackRedPages = 1;
4646 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4647 }
4648
4649 // retrieve entry point for pthread_setname_np
4650 Linux::_pthread_setname_np =
4651 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4652
4653 }
4654
4655 // To install functions for atexit system call
4656 extern "C" {
4657 static void perfMemory_exit_helper() {
4658 perfMemory_exit();
4659 }
4660 }
4661
4662 // this is called _after_ the global arguments have been parsed
4663 jint os::init_2(void) {
4664 Linux::fast_thread_clock_init();
4665
4666 // Allocate a single page and mark it as readable for safepoint polling
4667 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4668 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4669
4670 os::set_polling_page(polling_page);
4671
4672 #ifndef PRODUCT
4673 if (Verbose && PrintMiscellaneous) {
4674 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4675 (intptr_t)polling_page);
4676 }
4677 #endif
4678
4679 if (!UseMembar) {
4680 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4681 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4682 os::set_memory_serialize_page(mem_serialize_page);
4683
4684 #ifndef PRODUCT
4685 if (Verbose && PrintMiscellaneous) {
4686 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4687 (intptr_t)mem_serialize_page);
4688 }
4689 #endif
4690 }
4691
4692 // initialize suspend/resume support - must do this before signal_sets_init()
4693 if (SR_initialize() != 0) {
4694 perror("SR_initialize failed");
4695 return JNI_ERR;
4696 }
4697
4698 Linux::signal_sets_init();
4699 Linux::install_signal_handlers();
4700
4701 // Check minimum allowable stack size for thread creation and to initialize
4702 // the java system classes, including StackOverflowError - depends on page
4703 // size. Add a page for compiler2 recursion in main thread.
4704 // Add in 2*BytesPerWord times page size to account for VM stack during
4705 // class initialization depending on 32 or 64 bit VM.
4706 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4707 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4708 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4709
4710 size_t threadStackSizeInBytes = ThreadStackSize * K;
4711 if (threadStackSizeInBytes != 0 &&
4712 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4713 tty->print_cr("\nThe stack size specified is too small, "
4714 "Specify at least %dk",
4715 os::Linux::min_stack_allowed/ K);
4716 return JNI_ERR;
4717 }
4718
4719 // Make the stack size a multiple of the page size so that
4720 // the yellow/red zones can be guarded.
4721 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4722 vm_page_size()));
4723
4724 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4725
4726 #if defined(IA32)
4727 workaround_expand_exec_shield_cs_limit();
4728 #endif
4729
4730 Linux::libpthread_init();
4731 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4732 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4733 Linux::glibc_version(), Linux::libpthread_version(),
4734 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4735 }
4736
4737 if (UseNUMA) {
4738 if (!Linux::libnuma_init()) {
4739 UseNUMA = false;
4740 } else {
4741 if ((Linux::numa_max_node() < 1)) {
4742 // There's only one node(they start from 0), disable NUMA.
4743 UseNUMA = false;
4744 }
4745 }
4746 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4747 // we can make the adaptive lgrp chunk resizing work. If the user specified
4748 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4749 // disable adaptive resizing.
4750 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4751 if (FLAG_IS_DEFAULT(UseNUMA)) {
4752 UseNUMA = false;
4753 } else {
4754 if (FLAG_IS_DEFAULT(UseLargePages) &&
4755 FLAG_IS_DEFAULT(UseSHM) &&
4756 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4757 UseLargePages = false;
4758 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4759 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4760 UseAdaptiveSizePolicy = false;
4761 UseAdaptiveNUMAChunkSizing = false;
4762 }
4763 }
4764 }
4765 if (!UseNUMA && ForceNUMA) {
4766 UseNUMA = true;
4767 }
4768 }
4769
4770 if (MaxFDLimit) {
4771 // set the number of file descriptors to max. print out error
4772 // if getrlimit/setrlimit fails but continue regardless.
4773 struct rlimit nbr_files;
4774 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4775 if (status != 0) {
4776 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4777 perror("os::init_2 getrlimit failed");
4778 }
4779 } else {
4780 nbr_files.rlim_cur = nbr_files.rlim_max;
4781 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4782 if (status != 0) {
4783 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4784 perror("os::init_2 setrlimit failed");
4785 }
4786 }
4787 }
4788 }
4789
4790 // Initialize lock used to serialize thread creation (see os::create_thread)
4791 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4792
4793 // at-exit methods are called in the reverse order of their registration.
4794 // atexit functions are called on return from main or as a result of a
4795 // call to exit(3C). There can be only 32 of these functions registered
4796 // and atexit() does not set errno.
4797
4798 if (PerfAllowAtExitRegistration) {
4799 // only register atexit functions if PerfAllowAtExitRegistration is set.
4800 // atexit functions can be delayed until process exit time, which
4801 // can be problematic for embedded VM situations. Embedded VMs should
4802 // call DestroyJavaVM() to assure that VM resources are released.
4803
4804 // note: perfMemory_exit_helper atexit function may be removed in
4805 // the future if the appropriate cleanup code can be added to the
4806 // VM_Exit VMOperation's doit method.
4807 if (atexit(perfMemory_exit_helper) != 0) {
4808 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4809 }
4810 }
4811
4812 // initialize thread priority policy
4813 prio_init();
4814
4815 return JNI_OK;
4816 }
4817
4818 // Mark the polling page as unreadable
4819 void os::make_polling_page_unreadable(void) {
4820 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4821 fatal("Could not disable polling page");
4822 }
4823 }
4824
4825 // Mark the polling page as readable
4826 void os::make_polling_page_readable(void) {
4827 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4828 fatal("Could not enable polling page");
4829 }
4830 }
4831
4832 int os::active_processor_count() {
4833 // Linux doesn't yet have a (official) notion of processor sets,
4834 // so just return the number of online processors.
4835 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4836 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4837 return online_cpus;
4838 }
4839
4840 void os::set_native_thread_name(const char *name) {
4841 if (Linux::_pthread_setname_np) {
4842 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4843 snprintf(buf, sizeof(buf), "%s", name);
4844 buf[sizeof(buf) - 1] = '\0';
4845 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4846 // ERANGE should not happen; all other errors should just be ignored.
4847 assert(rc != ERANGE, "pthread_setname_np failed");
4848 }
4849 }
4850
4851 bool os::distribute_processes(uint length, uint* distribution) {
4852 // Not yet implemented.
4853 return false;
4854 }
4855
4856 bool os::bind_to_processor(uint processor_id) {
4857 // Not yet implemented.
4858 return false;
4859 }
4860
4861 ///
4862
4863 void os::SuspendedThreadTask::internal_do_task() {
4864 if (do_suspend(_thread->osthread())) {
4865 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4866 do_task(context);
4867 do_resume(_thread->osthread());
4868 }
4869 }
4870
4871 class PcFetcher : public os::SuspendedThreadTask {
4872 public:
4873 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4874 ExtendedPC result();
4875 protected:
4876 void do_task(const os::SuspendedThreadTaskContext& context);
4877 private:
4878 ExtendedPC _epc;
4879 };
4880
4881 ExtendedPC PcFetcher::result() {
4882 guarantee(is_done(), "task is not done yet.");
4883 return _epc;
4884 }
4885
4886 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4887 Thread* thread = context.thread();
4888 OSThread* osthread = thread->osthread();
4889 if (osthread->ucontext() != NULL) {
4890 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4891 } else {
4892 // NULL context is unexpected, double-check this is the VMThread
4893 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4894 }
4895 }
4896
4897 // Suspends the target using the signal mechanism and then grabs the PC before
4898 // resuming the target. Used by the flat-profiler only
4899 ExtendedPC os::get_thread_pc(Thread* thread) {
4900 // Make sure that it is called by the watcher for the VMThread
4901 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4902 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4903
4904 PcFetcher fetcher(thread);
4905 fetcher.run();
4906 return fetcher.result();
4907 }
4908
4909 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4910 pthread_mutex_t *_mutex,
4911 const struct timespec *_abstime) {
4912 if (is_NPTL()) {
4913 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4914 } else {
4915 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4916 // word back to default 64bit precision if condvar is signaled. Java
4917 // wants 53bit precision. Save and restore current value.
4918 int fpu = get_fpu_control_word();
4919 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4920 set_fpu_control_word(fpu);
4921 return status;
4922 }
4923 }
4924
4925 ////////////////////////////////////////////////////////////////////////////////
4926 // debug support
4927
4928 bool os::find(address addr, outputStream* st) {
4929 Dl_info dlinfo;
4930 memset(&dlinfo, 0, sizeof(dlinfo));
4931 if (dladdr(addr, &dlinfo) != 0) {
4932 st->print(PTR_FORMAT ": ", addr);
4933 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4934 st->print("%s+%#x", dlinfo.dli_sname,
4935 addr - (intptr_t)dlinfo.dli_saddr);
4936 } else if (dlinfo.dli_fbase != NULL) {
4937 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4938 } else {
4939 st->print("<absolute address>");
4940 }
4941 if (dlinfo.dli_fname != NULL) {
4942 st->print(" in %s", dlinfo.dli_fname);
4943 }
4944 if (dlinfo.dli_fbase != NULL) {
4945 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4946 }
4947 st->cr();
4948
4949 if (Verbose) {
4950 // decode some bytes around the PC
4951 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4952 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4953 address lowest = (address) dlinfo.dli_sname;
4954 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4955 if (begin < lowest) begin = lowest;
4956 Dl_info dlinfo2;
4957 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4958 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4959 end = (address) dlinfo2.dli_saddr;
4960 }
4961 Disassembler::decode(begin, end, st);
4962 }
4963 return true;
4964 }
4965 return false;
4966 }
4967
4968 ////////////////////////////////////////////////////////////////////////////////
4969 // misc
4970
4971 // This does not do anything on Linux. This is basically a hook for being
4972 // able to use structured exception handling (thread-local exception filters)
4973 // on, e.g., Win32.
4974 void
4975 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4976 JavaCallArguments* args, Thread* thread) {
4977 f(value, method, args, thread);
4978 }
4979
4980 void os::print_statistics() {
4981 }
4982
4983 int os::message_box(const char* title, const char* message) {
4984 int i;
4985 fdStream err(defaultStream::error_fd());
4986 for (i = 0; i < 78; i++) err.print_raw("=");
4987 err.cr();
4988 err.print_raw_cr(title);
4989 for (i = 0; i < 78; i++) err.print_raw("-");
4990 err.cr();
4991 err.print_raw_cr(message);
4992 for (i = 0; i < 78; i++) err.print_raw("=");
4993 err.cr();
4994
4995 char buf[16];
4996 // Prevent process from exiting upon "read error" without consuming all CPU
4997 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4998
4999 return buf[0] == 'y' || buf[0] == 'Y';
5000 }
5001
5002 int os::stat(const char *path, struct stat *sbuf) {
5003 char pathbuf[MAX_PATH];
5004 if (strlen(path) > MAX_PATH - 1) {
5005 errno = ENAMETOOLONG;
5006 return -1;
5007 }
5008 os::native_path(strcpy(pathbuf, path));
5009 return ::stat(pathbuf, sbuf);
5010 }
5011
5012 bool os::check_heap(bool force) {
5013 return true;
5014 }
5015
5016 // Is a (classpath) directory empty?
5017 bool os::dir_is_empty(const char* path) {
5018 DIR *dir = NULL;
5019 struct dirent *ptr;
5020
5021 dir = opendir(path);
5022 if (dir == NULL) return true;
5023
5024 // Scan the directory
5025 bool result = true;
5026 char buf[sizeof(struct dirent) + MAX_PATH];
5027 while (result && (ptr = ::readdir(dir)) != NULL) {
5028 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5029 result = false;
5030 }
5031 }
5032 closedir(dir);
5033 return result;
5034 }
5035
5036 // This code originates from JDK's sysOpen and open64_w
5037 // from src/solaris/hpi/src/system_md.c
5038
5039 int os::open(const char *path, int oflag, int mode) {
5040 if (strlen(path) > MAX_PATH - 1) {
5041 errno = ENAMETOOLONG;
5042 return -1;
5043 }
5044
5045 // All file descriptors that are opened in the Java process and not
5046 // specifically destined for a subprocess should have the close-on-exec
5047 // flag set. If we don't set it, then careless 3rd party native code
5048 // might fork and exec without closing all appropriate file descriptors
5049 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5050 // turn might:
5051 //
5052 // - cause end-of-file to fail to be detected on some file
5053 // descriptors, resulting in mysterious hangs, or
5054 //
5055 // - might cause an fopen in the subprocess to fail on a system
5056 // suffering from bug 1085341.
5057 //
5058 // (Yes, the default setting of the close-on-exec flag is a Unix
5059 // design flaw)
5060 //
5061 // See:
5062 // 1085341: 32-bit stdio routines should support file descriptors >255
5063 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5064 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5065 //
5066 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5067 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5068 // because it saves a system call and removes a small window where the flag
5069 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5070 // and we fall back to using FD_CLOEXEC (see below).
5071 #ifdef O_CLOEXEC
5072 oflag |= O_CLOEXEC;
5073 #endif
5074
5075 int fd = ::open64(path, oflag, mode);
5076 if (fd == -1) return -1;
5077
5078 //If the open succeeded, the file might still be a directory
5079 {
5080 struct stat64 buf64;
5081 int ret = ::fstat64(fd, &buf64);
5082 int st_mode = buf64.st_mode;
5083
5084 if (ret != -1) {
5085 if ((st_mode & S_IFMT) == S_IFDIR) {
5086 errno = EISDIR;
5087 ::close(fd);
5088 return -1;
5089 }
5090 } else {
5091 ::close(fd);
5092 return -1;
5093 }
5094 }
5095
5096 #ifdef FD_CLOEXEC
5097 // Validate that the use of the O_CLOEXEC flag on open above worked.
5098 // With recent kernels, we will perform this check exactly once.
5099 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5100 if (!O_CLOEXEC_is_known_to_work) {
5101 int flags = ::fcntl(fd, F_GETFD);
5102 if (flags != -1) {
5103 if ((flags & FD_CLOEXEC) != 0)
5104 O_CLOEXEC_is_known_to_work = 1;
5105 else
5106 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5107 }
5108 }
5109 #endif
5110
5111 return fd;
5112 }
5113
5114
5115 // create binary file, rewriting existing file if required
5116 int os::create_binary_file(const char* path, bool rewrite_existing) {
5117 int oflags = O_WRONLY | O_CREAT;
5118 if (!rewrite_existing) {
5119 oflags |= O_EXCL;
5120 }
5121 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5122 }
5123
5124 // return current position of file pointer
5125 jlong os::current_file_offset(int fd) {
5126 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5127 }
5128
5129 // move file pointer to the specified offset
5130 jlong os::seek_to_file_offset(int fd, jlong offset) {
5131 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5132 }
5133
5134 // This code originates from JDK's sysAvailable
5135 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5136
5137 int os::available(int fd, jlong *bytes) {
5138 jlong cur, end;
5139 int mode;
5140 struct stat64 buf64;
5141
5142 if (::fstat64(fd, &buf64) >= 0) {
5143 mode = buf64.st_mode;
5144 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5145 // XXX: is the following call interruptible? If so, this might
5146 // need to go through the INTERRUPT_IO() wrapper as for other
5147 // blocking, interruptible calls in this file.
5148 int n;
5149 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5150 *bytes = n;
5151 return 1;
5152 }
5153 }
5154 }
5155 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5156 return 0;
5157 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5158 return 0;
5159 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5160 return 0;
5161 }
5162 *bytes = end - cur;
5163 return 1;
5164 }
5165
5166 // Map a block of memory.
5167 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5168 char *addr, size_t bytes, bool read_only,
5169 bool allow_exec) {
5170 int prot;
5171 int flags = MAP_PRIVATE;
5172
5173 if (read_only) {
5174 prot = PROT_READ;
5175 } else {
5176 prot = PROT_READ | PROT_WRITE;
5177 }
5178
5179 if (allow_exec) {
5180 prot |= PROT_EXEC;
5181 }
5182
5183 if (addr != NULL) {
5184 flags |= MAP_FIXED;
5185 }
5186
5187 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5188 fd, file_offset);
5189 if (mapped_address == MAP_FAILED) {
5190 return NULL;
5191 }
5192 return mapped_address;
5193 }
5194
5195
5196 // Remap a block of memory.
5197 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5198 char *addr, size_t bytes, bool read_only,
5199 bool allow_exec) {
5200 // same as map_memory() on this OS
5201 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5202 allow_exec);
5203 }
5204
5205
5206 // Unmap a block of memory.
5207 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5208 return munmap(addr, bytes) == 0;
5209 }
5210
5211 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5212
5213 static clockid_t thread_cpu_clockid(Thread* thread) {
5214 pthread_t tid = thread->osthread()->pthread_id();
5215 clockid_t clockid;
5216
5217 // Get thread clockid
5218 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5219 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5220 return clockid;
5221 }
5222
5223 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5224 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5225 // of a thread.
5226 //
5227 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5228 // the fast estimate available on the platform.
5229
5230 jlong os::current_thread_cpu_time() {
5231 if (os::Linux::supports_fast_thread_cpu_time()) {
5232 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5233 } else {
5234 // return user + sys since the cost is the same
5235 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5236 }
5237 }
5238
5239 jlong os::thread_cpu_time(Thread* thread) {
5240 // consistent with what current_thread_cpu_time() returns
5241 if (os::Linux::supports_fast_thread_cpu_time()) {
5242 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5243 } else {
5244 return slow_thread_cpu_time(thread, true /* user + sys */);
5245 }
5246 }
5247
5248 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5249 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5250 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5251 } else {
5252 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5253 }
5254 }
5255
5256 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5257 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5258 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5259 } else {
5260 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5261 }
5262 }
5263
5264 // -1 on error.
5265 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5266 pid_t tid = thread->osthread()->thread_id();
5267 char *s;
5268 char stat[2048];
5269 int statlen;
5270 char proc_name[64];
5271 int count;
5272 long sys_time, user_time;
5273 char cdummy;
5274 int idummy;
5275 long ldummy;
5276 FILE *fp;
5277
5278 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5279 fp = fopen(proc_name, "r");
5280 if (fp == NULL) return -1;
5281 statlen = fread(stat, 1, 2047, fp);
5282 stat[statlen] = '\0';
5283 fclose(fp);
5284
5285 // Skip pid and the command string. Note that we could be dealing with
5286 // weird command names, e.g. user could decide to rename java launcher
5287 // to "java 1.4.2 :)", then the stat file would look like
5288 // 1234 (java 1.4.2 :)) R ... ...
5289 // We don't really need to know the command string, just find the last
5290 // occurrence of ")" and then start parsing from there. See bug 4726580.
5291 s = strrchr(stat, ')');
5292 if (s == NULL) return -1;
5293
5294 // Skip blank chars
5295 do { s++; } while (s && isspace(*s));
5296
5297 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5298 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5299 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5300 &user_time, &sys_time);
5301 if (count != 13) return -1;
5302 if (user_sys_cpu_time) {
5303 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5304 } else {
5305 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5306 }
5307 }
5308
5309 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5310 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5311 info_ptr->may_skip_backward = false; // elapsed time not wall time
5312 info_ptr->may_skip_forward = false; // elapsed time not wall time
5313 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5314 }
5315
5316 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5317 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5318 info_ptr->may_skip_backward = false; // elapsed time not wall time
5319 info_ptr->may_skip_forward = false; // elapsed time not wall time
5320 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5321 }
5322
5323 bool os::is_thread_cpu_time_supported() {
5324 return true;
5325 }
5326
5327 // System loadavg support. Returns -1 if load average cannot be obtained.
5328 // Linux doesn't yet have a (official) notion of processor sets,
5329 // so just return the system wide load average.
5330 int os::loadavg(double loadavg[], int nelem) {
5331 return ::getloadavg(loadavg, nelem);
5332 }
5333
5334 void os::pause() {
5335 char filename[MAX_PATH];
5336 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5337 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5338 } else {
5339 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5340 }
5341
5342 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5343 if (fd != -1) {
5344 struct stat buf;
5345 ::close(fd);
5346 while (::stat(filename, &buf) == 0) {
5347 (void)::poll(NULL, 0, 100);
5348 }
5349 } else {
5350 jio_fprintf(stderr,
5351 "Could not open pause file '%s', continuing immediately.\n", filename);
5352 }
5353 }
5354
5355
5356 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5357 // comment paragraphs are worth repeating here:
5358 //
5359 // Assumption:
5360 // Only one parker can exist on an event, which is why we allocate
5361 // them per-thread. Multiple unparkers can coexist.
5362 //
5363 // _Event serves as a restricted-range semaphore.
5364 // -1 : thread is blocked, i.e. there is a waiter
5365 // 0 : neutral: thread is running or ready,
5366 // could have been signaled after a wait started
5367 // 1 : signaled - thread is running or ready
5368 //
5369 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5370 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5371 // For specifics regarding the bug see GLIBC BUGID 261237 :
5372 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5373 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5374 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5375 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5376 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5377 // and monitorenter when we're using 1-0 locking. All those operations may result in
5378 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5379 // of libpthread avoids the problem, but isn't practical.
5380 //
5381 // Possible remedies:
5382 //
5383 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5384 // This is palliative and probabilistic, however. If the thread is preempted
5385 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5386 // than the minimum period may have passed, and the abstime may be stale (in the
5387 // past) resultin in a hang. Using this technique reduces the odds of a hang
5388 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5389 //
5390 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5391 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5392 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5393 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5394 // thread.
5395 //
5396 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5397 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5398 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5399 // This also works well. In fact it avoids kernel-level scalability impediments
5400 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5401 // timers in a graceful fashion.
5402 //
5403 // 4. When the abstime value is in the past it appears that control returns
5404 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5405 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5406 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5407 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5408 // It may be possible to avoid reinitialization by checking the return
5409 // value from pthread_cond_timedwait(). In addition to reinitializing the
5410 // condvar we must establish the invariant that cond_signal() is only called
5411 // within critical sections protected by the adjunct mutex. This prevents
5412 // cond_signal() from "seeing" a condvar that's in the midst of being
5413 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5414 // desirable signal-after-unlock optimization that avoids futile context switching.
5415 //
5416 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5417 // structure when a condvar is used or initialized. cond_destroy() would
5418 // release the helper structure. Our reinitialize-after-timedwait fix
5419 // put excessive stress on malloc/free and locks protecting the c-heap.
5420 //
5421 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5422 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5423 // and only enabling the work-around for vulnerable environments.
5424
5425 // utility to compute the abstime argument to timedwait:
5426 // millis is the relative timeout time
5427 // abstime will be the absolute timeout time
5428 // TODO: replace compute_abstime() with unpackTime()
5429
5430 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5431 if (millis < 0) millis = 0;
5432
5433 jlong seconds = millis / 1000;
5434 millis %= 1000;
5435 if (seconds > 50000000) { // see man cond_timedwait(3T)
5436 seconds = 50000000;
5437 }
5438
5439 if (os::supports_monotonic_clock()) {
5440 struct timespec now;
5441 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5442 assert_status(status == 0, status, "clock_gettime");
5443 abstime->tv_sec = now.tv_sec + seconds;
5444 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5445 if (nanos >= NANOSECS_PER_SEC) {
5446 abstime->tv_sec += 1;
5447 nanos -= NANOSECS_PER_SEC;
5448 }
5449 abstime->tv_nsec = nanos;
5450 } else {
5451 struct timeval now;
5452 int status = gettimeofday(&now, NULL);
5453 assert(status == 0, "gettimeofday");
5454 abstime->tv_sec = now.tv_sec + seconds;
5455 long usec = now.tv_usec + millis * 1000;
5456 if (usec >= 1000000) {
5457 abstime->tv_sec += 1;
5458 usec -= 1000000;
5459 }
5460 abstime->tv_nsec = usec * 1000;
5461 }
5462 return abstime;
5463 }
5464
5465 void os::PlatformEvent::park() { // AKA "down()"
5466 // Transitions for _Event:
5467 // -1 => -1 : illegal
5468 // 1 => 0 : pass - return immediately
5469 // 0 => -1 : block; then set _Event to 0 before returning
5470
5471 // Invariant: Only the thread associated with the Event/PlatformEvent
5472 // may call park().
5473 // TODO: assert that _Assoc != NULL or _Assoc == Self
5474 assert(_nParked == 0, "invariant");
5475
5476 int v;
5477 for (;;) {
5478 v = _Event;
5479 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5480 }
5481 guarantee(v >= 0, "invariant");
5482 if (v == 0) {
5483 // Do this the hard way by blocking ...
5484 int status = pthread_mutex_lock(_mutex);
5485 assert_status(status == 0, status, "mutex_lock");
5486 guarantee(_nParked == 0, "invariant");
5487 ++_nParked;
5488 while (_Event < 0) {
5489 status = pthread_cond_wait(_cond, _mutex);
5490 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5491 // Treat this the same as if the wait was interrupted
5492 if (status == ETIME) { status = EINTR; }
5493 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5494 }
5495 --_nParked;
5496
5497 _Event = 0;
5498 status = pthread_mutex_unlock(_mutex);
5499 assert_status(status == 0, status, "mutex_unlock");
5500 // Paranoia to ensure our locked and lock-free paths interact
5501 // correctly with each other.
5502 OrderAccess::fence();
5503 }
5504 guarantee(_Event >= 0, "invariant");
5505 }
5506
5507 int os::PlatformEvent::park(jlong millis) {
5508 // Transitions for _Event:
5509 // -1 => -1 : illegal
5510 // 1 => 0 : pass - return immediately
5511 // 0 => -1 : block; then set _Event to 0 before returning
5512
5513 guarantee(_nParked == 0, "invariant");
5514
5515 int v;
5516 for (;;) {
5517 v = _Event;
5518 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5519 }
5520 guarantee(v >= 0, "invariant");
5521 if (v != 0) return OS_OK;
5522
5523 // We do this the hard way, by blocking the thread.
5524 // Consider enforcing a minimum timeout value.
5525 struct timespec abst;
5526 compute_abstime(&abst, millis);
5527
5528 int ret = OS_TIMEOUT;
5529 int status = pthread_mutex_lock(_mutex);
5530 assert_status(status == 0, status, "mutex_lock");
5531 guarantee(_nParked == 0, "invariant");
5532 ++_nParked;
5533
5534 // Object.wait(timo) will return because of
5535 // (a) notification
5536 // (b) timeout
5537 // (c) thread.interrupt
5538 //
5539 // Thread.interrupt and object.notify{All} both call Event::set.
5540 // That is, we treat thread.interrupt as a special case of notification.
5541 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5542 // We assume all ETIME returns are valid.
5543 //
5544 // TODO: properly differentiate simultaneous notify+interrupt.
5545 // In that case, we should propagate the notify to another waiter.
5546
5547 while (_Event < 0) {
5548 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5549 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5550 pthread_cond_destroy(_cond);
5551 pthread_cond_init(_cond, os::Linux::condAttr());
5552 }
5553 assert_status(status == 0 || status == EINTR ||
5554 status == ETIME || status == ETIMEDOUT,
5555 status, "cond_timedwait");
5556 if (!FilterSpuriousWakeups) break; // previous semantics
5557 if (status == ETIME || status == ETIMEDOUT) break;
5558 // We consume and ignore EINTR and spurious wakeups.
5559 }
5560 --_nParked;
5561 if (_Event >= 0) {
5562 ret = OS_OK;
5563 }
5564 _Event = 0;
5565 status = pthread_mutex_unlock(_mutex);
5566 assert_status(status == 0, status, "mutex_unlock");
5567 assert(_nParked == 0, "invariant");
5568 // Paranoia to ensure our locked and lock-free paths interact
5569 // correctly with each other.
5570 OrderAccess::fence();
5571 return ret;
5572 }
5573
5574 void os::PlatformEvent::unpark() {
5575 // Transitions for _Event:
5576 // 0 => 1 : just return
5577 // 1 => 1 : just return
5578 // -1 => either 0 or 1; must signal target thread
5579 // That is, we can safely transition _Event from -1 to either
5580 // 0 or 1.
5581 // See also: "Semaphores in Plan 9" by Mullender & Cox
5582 //
5583 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5584 // that it will take two back-to-back park() calls for the owning
5585 // thread to block. This has the benefit of forcing a spurious return
5586 // from the first park() call after an unpark() call which will help
5587 // shake out uses of park() and unpark() without condition variables.
5588
5589 if (Atomic::xchg(1, &_Event) >= 0) return;
5590
5591 // Wait for the thread associated with the event to vacate
5592 int status = pthread_mutex_lock(_mutex);
5593 assert_status(status == 0, status, "mutex_lock");
5594 int AnyWaiters = _nParked;
5595 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5596 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5597 AnyWaiters = 0;
5598 pthread_cond_signal(_cond);
5599 }
5600 status = pthread_mutex_unlock(_mutex);
5601 assert_status(status == 0, status, "mutex_unlock");
5602 if (AnyWaiters != 0) {
5603 // Note that we signal() *after* dropping the lock for "immortal" Events.
5604 // This is safe and avoids a common class of futile wakeups. In rare
5605 // circumstances this can cause a thread to return prematurely from
5606 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5607 // will simply re-test the condition and re-park itself.
5608 // This provides particular benefit if the underlying platform does not
5609 // provide wait morphing.
5610 status = pthread_cond_signal(_cond);
5611 assert_status(status == 0, status, "cond_signal");
5612 }
5613 }
5614
5615
5616 // JSR166
5617 // -------------------------------------------------------
5618
5619 // The solaris and linux implementations of park/unpark are fairly
5620 // conservative for now, but can be improved. They currently use a
5621 // mutex/condvar pair, plus a a count.
5622 // Park decrements count if > 0, else does a condvar wait. Unpark
5623 // sets count to 1 and signals condvar. Only one thread ever waits
5624 // on the condvar. Contention seen when trying to park implies that someone
5625 // is unparking you, so don't wait. And spurious returns are fine, so there
5626 // is no need to track notifications.
5627
5628 // This code is common to linux and solaris and will be moved to a
5629 // common place in dolphin.
5630 //
5631 // The passed in time value is either a relative time in nanoseconds
5632 // or an absolute time in milliseconds. Either way it has to be unpacked
5633 // into suitable seconds and nanoseconds components and stored in the
5634 // given timespec structure.
5635 // Given time is a 64-bit value and the time_t used in the timespec is only
5636 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5637 // overflow if times way in the future are given. Further on Solaris versions
5638 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5639 // number of seconds, in abstime, is less than current_time + 100,000,000.
5640 // As it will be 28 years before "now + 100000000" will overflow we can
5641 // ignore overflow and just impose a hard-limit on seconds using the value
5642 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5643 // years from "now".
5644
5645 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5646 assert(time > 0, "convertTime");
5647 time_t max_secs = 0;
5648
5649 if (!os::supports_monotonic_clock() || isAbsolute) {
5650 struct timeval now;
5651 int status = gettimeofday(&now, NULL);
5652 assert(status == 0, "gettimeofday");
5653
5654 max_secs = now.tv_sec + MAX_SECS;
5655
5656 if (isAbsolute) {
5657 jlong secs = time / 1000;
5658 if (secs > max_secs) {
5659 absTime->tv_sec = max_secs;
5660 } else {
5661 absTime->tv_sec = secs;
5662 }
5663 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5664 } else {
5665 jlong secs = time / NANOSECS_PER_SEC;
5666 if (secs >= MAX_SECS) {
5667 absTime->tv_sec = max_secs;
5668 absTime->tv_nsec = 0;
5669 } else {
5670 absTime->tv_sec = now.tv_sec + secs;
5671 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5672 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5673 absTime->tv_nsec -= NANOSECS_PER_SEC;
5674 ++absTime->tv_sec; // note: this must be <= max_secs
5675 }
5676 }
5677 }
5678 } else {
5679 // must be relative using monotonic clock
5680 struct timespec now;
5681 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5682 assert_status(status == 0, status, "clock_gettime");
5683 max_secs = now.tv_sec + MAX_SECS;
5684 jlong secs = time / NANOSECS_PER_SEC;
5685 if (secs >= MAX_SECS) {
5686 absTime->tv_sec = max_secs;
5687 absTime->tv_nsec = 0;
5688 } else {
5689 absTime->tv_sec = now.tv_sec + secs;
5690 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5691 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5692 absTime->tv_nsec -= NANOSECS_PER_SEC;
5693 ++absTime->tv_sec; // note: this must be <= max_secs
5694 }
5695 }
5696 }
5697 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5698 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5699 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5700 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5701 }
5702
5703 void Parker::park(bool isAbsolute, jlong time) {
5704 // Ideally we'd do something useful while spinning, such
5705 // as calling unpackTime().
5706
5707 // Optional fast-path check:
5708 // Return immediately if a permit is available.
5709 // We depend on Atomic::xchg() having full barrier semantics
5710 // since we are doing a lock-free update to _counter.
5711 if (Atomic::xchg(0, &_counter) > 0) return;
5712
5713 Thread* thread = Thread::current();
5714 assert(thread->is_Java_thread(), "Must be JavaThread");
5715 JavaThread *jt = (JavaThread *)thread;
5716
5717 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5718 // Check interrupt before trying to wait
5719 if (Thread::is_interrupted(thread, false)) {
5720 return;
5721 }
5722
5723 // Next, demultiplex/decode time arguments
5724 timespec absTime;
5725 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5726 return;
5727 }
5728 if (time > 0) {
5729 unpackTime(&absTime, isAbsolute, time);
5730 }
5731
5732
5733 // Enter safepoint region
5734 // Beware of deadlocks such as 6317397.
5735 // The per-thread Parker:: mutex is a classic leaf-lock.
5736 // In particular a thread must never block on the Threads_lock while
5737 // holding the Parker:: mutex. If safepoints are pending both the
5738 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5739 ThreadBlockInVM tbivm(jt);
5740
5741 // Don't wait if cannot get lock since interference arises from
5742 // unblocking. Also. check interrupt before trying wait
5743 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5744 return;
5745 }
5746
5747 int status;
5748 if (_counter > 0) { // no wait needed
5749 _counter = 0;
5750 status = pthread_mutex_unlock(_mutex);
5751 assert(status == 0, "invariant");
5752 // Paranoia to ensure our locked and lock-free paths interact
5753 // correctly with each other and Java-level accesses.
5754 OrderAccess::fence();
5755 return;
5756 }
5757
5758 #ifdef ASSERT
5759 // Don't catch signals while blocked; let the running threads have the signals.
5760 // (This allows a debugger to break into the running thread.)
5761 sigset_t oldsigs;
5762 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5763 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5764 #endif
5765
5766 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5767 jt->set_suspend_equivalent();
5768 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5769
5770 assert(_cur_index == -1, "invariant");
5771 if (time == 0) {
5772 _cur_index = REL_INDEX; // arbitrary choice when not timed
5773 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5774 } else {
5775 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5776 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5777 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5778 pthread_cond_destroy(&_cond[_cur_index]);
5779 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5780 }
5781 }
5782 _cur_index = -1;
5783 assert_status(status == 0 || status == EINTR ||
5784 status == ETIME || status == ETIMEDOUT,
5785 status, "cond_timedwait");
5786
5787 #ifdef ASSERT
5788 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5789 #endif
5790
5791 _counter = 0;
5792 status = pthread_mutex_unlock(_mutex);
5793 assert_status(status == 0, status, "invariant");
5794 // Paranoia to ensure our locked and lock-free paths interact
5795 // correctly with each other and Java-level accesses.
5796 OrderAccess::fence();
5797
5798 // If externally suspended while waiting, re-suspend
5799 if (jt->handle_special_suspend_equivalent_condition()) {
5800 jt->java_suspend_self();
5801 }
5802 }
5803
5804 void Parker::unpark() {
5805 int status = pthread_mutex_lock(_mutex);
5806 assert(status == 0, "invariant");
5807 const int s = _counter;
5808 _counter = 1;
5809 if (s < 1) {
5810 // thread might be parked
5811 if (_cur_index != -1) {
5812 // thread is definitely parked
5813 if (WorkAroundNPTLTimedWaitHang) {
5814 status = pthread_cond_signal(&_cond[_cur_index]);
5815 assert(status == 0, "invariant");
5816 status = pthread_mutex_unlock(_mutex);
5817 assert(status == 0, "invariant");
5818 } else {
5819 status = pthread_mutex_unlock(_mutex);
5820 assert(status == 0, "invariant");
5821 status = pthread_cond_signal(&_cond[_cur_index]);
5822 assert(status == 0, "invariant");
5823 }
5824 } else {
5825 pthread_mutex_unlock(_mutex);
5826 assert(status == 0, "invariant");
5827 }
5828 } else {
5829 pthread_mutex_unlock(_mutex);
5830 assert(status == 0, "invariant");
5831 }
5832 }
5833
5834
5835 extern char** environ;
5836
5837 // Run the specified command in a separate process. Return its exit value,
5838 // or -1 on failure (e.g. can't fork a new process).
5839 // Unlike system(), this function can be called from signal handler. It
5840 // doesn't block SIGINT et al.
5841 int os::fork_and_exec(char* cmd) {
5842 const char * argv[4] = {"sh", "-c", cmd, NULL};
5843
5844 pid_t pid = fork();
5845
5846 if (pid < 0) {
5847 // fork failed
5848 return -1;
5849
5850 } else if (pid == 0) {
5851 // child process
5852
5853 execve("/bin/sh", (char* const*)argv, environ);
5854
5855 // execve failed
5856 _exit(-1);
5857
5858 } else {
5859 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5860 // care about the actual exit code, for now.
5861
5862 int status;
5863
5864 // Wait for the child process to exit. This returns immediately if
5865 // the child has already exited. */
5866 while (waitpid(pid, &status, 0) < 0) {
5867 switch (errno) {
5868 case ECHILD: return 0;
5869 case EINTR: break;
5870 default: return -1;
5871 }
5872 }
5873
5874 if (WIFEXITED(status)) {
5875 // The child exited normally; get its exit code.
5876 return WEXITSTATUS(status);
5877 } else if (WIFSIGNALED(status)) {
5878 // The child exited because of a signal
5879 // The best value to return is 0x80 + signal number,
5880 // because that is what all Unix shells do, and because
5881 // it allows callers to distinguish between process exit and
5882 // process death by signal.
5883 return 0x80 + WTERMSIG(status);
5884 } else {
5885 // Unknown exit code; pass it through
5886 return status;
5887 }
5888 }
5889 }
5890
5891 // is_headless_jre()
5892 //
5893 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5894 // in order to report if we are running in a headless jre
5895 //
5896 // Since JDK8 xawt/libmawt.so was moved into the same directory
5897 // as libawt.so, and renamed libawt_xawt.so
5898 //
5899 bool os::is_headless_jre() {
5900 struct stat statbuf;
5901 char buf[MAXPATHLEN];
5902 char libmawtpath[MAXPATHLEN];
5903 const char *xawtstr = "/xawt/libmawt.so";
5904 const char *new_xawtstr = "/libawt_xawt.so";
5905 char *p;
5906
5907 // Get path to libjvm.so
5908 os::jvm_path(buf, sizeof(buf));
5909
5910 // Get rid of libjvm.so
5911 p = strrchr(buf, '/');
5912 if (p == NULL) {
5913 return false;
5914 } else {
5915 *p = '\0';
5916 }
5917
5918 // Get rid of client or server
5919 p = strrchr(buf, '/');
5920 if (p == NULL) {
5921 return false;
5922 } else {
5923 *p = '\0';
5924 }
5925
5926 // check xawt/libmawt.so
5927 strcpy(libmawtpath, buf);
5928 strcat(libmawtpath, xawtstr);
5929 if (::stat(libmawtpath, &statbuf) == 0) return false;
5930
5931 // check libawt_xawt.so
5932 strcpy(libmawtpath, buf);
5933 strcat(libmawtpath, new_xawtstr);
5934 if (::stat(libmawtpath, &statbuf) == 0) return false;
5935
5936 return true;
5937 }
5938
5939 // Get the default path to the core file
5940 // Returns the length of the string
5941 int os::get_core_path(char* buffer, size_t bufferSize) {
5942 /*
5943 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5944 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5945 */
5946 const int core_pattern_len = 129;
5947 char core_pattern[core_pattern_len] = {0};
5948
5949 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5950 if (core_pattern_file != -1) {
5951 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5952 ::close(core_pattern_file);
5953
5954 if (ret > 0) {
5955 char *last_char = core_pattern + strlen(core_pattern) - 1;
5956
5957 if (*last_char == '\n') {
5958 *last_char = '\0';
5959 }
5960 }
5961 }
5962
5963 if (strlen(core_pattern) == 0) {
5964 return -1;
5965 }
5966
5967 char *pid_pos = strstr(core_pattern, "%p");
5968 int written;
5969
5970 if (core_pattern[0] == '/') {
5971 written = jio_snprintf(buffer, bufferSize, core_pattern);
5972 } else {
5973 char cwd[PATH_MAX];
5974
5975 const char* p = get_current_directory(cwd, PATH_MAX);
5976 if (p == NULL) {
5977 return -1;
5978 }
5979
5980 if (core_pattern[0] == '|') {
5981 written = jio_snprintf(buffer, bufferSize,
5982 "\"%s\" (or dumping to %s/core.%d)",
5983 &core_pattern[1], p, current_process_id());
5984 } else {
5985 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5986 }
5987 }
5988
5989 if (written < 0) {
5990 return -1;
5991 }
5992
5993 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5994 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5995
5996 if (core_uses_pid_file != -1) {
5997 char core_uses_pid = 0;
5998 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5999 ::close(core_uses_pid_file);
6000
6001 if (core_uses_pid == '1') {
6002 jio_snprintf(buffer + written, bufferSize - written,
6003 ".%d", current_process_id());
6004 }
6005 }
6006 }
6007
6008 return strlen(buffer);
6009 }
6010
6011 /////////////// Unit tests ///////////////
6012
6013 #ifndef PRODUCT
6014
6015 #define test_log(...) \
6016 do { \
6017 if (VerboseInternalVMTests) { \
6018 tty->print_cr(__VA_ARGS__); \
6019 tty->flush(); \
6020 } \
6021 } while (false)
6022
6023 class TestReserveMemorySpecial : AllStatic {
6024 public:
6025 static void small_page_write(void* addr, size_t size) {
6026 size_t page_size = os::vm_page_size();
6027
6028 char* end = (char*)addr + size;
6029 for (char* p = (char*)addr; p < end; p += page_size) {
6030 *p = 1;
6031 }
6032 }
6033
6034 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6035 if (!UseHugeTLBFS) {
6036 return;
6037 }
6038
6039 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6040
6041 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6042
6043 if (addr != NULL) {
6044 small_page_write(addr, size);
6045
6046 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6047 }
6048 }
6049
6050 static void test_reserve_memory_special_huge_tlbfs_only() {
6051 if (!UseHugeTLBFS) {
6052 return;
6053 }
6054
6055 size_t lp = os::large_page_size();
6056
6057 for (size_t size = lp; size <= lp * 10; size += lp) {
6058 test_reserve_memory_special_huge_tlbfs_only(size);
6059 }
6060 }
6061
6062 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6063 size_t lp = os::large_page_size();
6064 size_t ag = os::vm_allocation_granularity();
6065
6066 // sizes to test
6067 const size_t sizes[] = {
6068 lp, lp + ag, lp + lp / 2, lp * 2,
6069 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6070 lp * 10, lp * 10 + lp / 2
6071 };
6072 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6073
6074 // For each size/alignment combination, we test three scenarios:
6075 // 1) with req_addr == NULL
6076 // 2) with a non-null req_addr at which we expect to successfully allocate
6077 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6078 // expect the allocation to either fail or to ignore req_addr
6079
6080 // Pre-allocate two areas; they shall be as large as the largest allocation
6081 // and aligned to the largest alignment we will be testing.
6082 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6083 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6084 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6085 -1, 0);
6086 assert(mapping1 != MAP_FAILED, "should work");
6087
6088 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6089 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6090 -1, 0);
6091 assert(mapping2 != MAP_FAILED, "should work");
6092
6093 // Unmap the first mapping, but leave the second mapping intact: the first
6094 // mapping will serve as a value for a "good" req_addr (case 2). The second
6095 // mapping, still intact, as "bad" req_addr (case 3).
6096 ::munmap(mapping1, mapping_size);
6097
6098 // Case 1
6099 test_log("%s, req_addr NULL:", __FUNCTION__);
6100 test_log("size align result");
6101
6102 for (int i = 0; i < num_sizes; i++) {
6103 const size_t size = sizes[i];
6104 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6105 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6106 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6107 size, alignment, p, (p != NULL ? "" : "(failed)"));
6108 if (p != NULL) {
6109 assert(is_ptr_aligned(p, alignment), "must be");
6110 small_page_write(p, size);
6111 os::Linux::release_memory_special_huge_tlbfs(p, size);
6112 }
6113 }
6114 }
6115
6116 // Case 2
6117 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6118 test_log("size align req_addr result");
6119
6120 for (int i = 0; i < num_sizes; i++) {
6121 const size_t size = sizes[i];
6122 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6123 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6124 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6125 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6126 size, alignment, req_addr, p,
6127 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6128 if (p != NULL) {
6129 assert(p == req_addr, "must be");
6130 small_page_write(p, size);
6131 os::Linux::release_memory_special_huge_tlbfs(p, size);
6132 }
6133 }
6134 }
6135
6136 // Case 3
6137 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6138 test_log("size align req_addr result");
6139
6140 for (int i = 0; i < num_sizes; i++) {
6141 const size_t size = sizes[i];
6142 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6143 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6144 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6145 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6146 size, alignment, req_addr, p,
6147 ((p != NULL ? "" : "(failed)")));
6148 // as the area around req_addr contains already existing mappings, the API should always
6149 // return NULL (as per contract, it cannot return another address)
6150 assert(p == NULL, "must be");
6151 }
6152 }
6153
6154 ::munmap(mapping2, mapping_size);
6155
6156 }
6157
6158 static void test_reserve_memory_special_huge_tlbfs() {
6159 if (!UseHugeTLBFS) {
6160 return;
6161 }
6162
6163 test_reserve_memory_special_huge_tlbfs_only();
6164 test_reserve_memory_special_huge_tlbfs_mixed();
6165 }
6166
6167 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6168 if (!UseSHM) {
6169 return;
6170 }
6171
6172 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6173
6174 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6175
6176 if (addr != NULL) {
6177 assert(is_ptr_aligned(addr, alignment), "Check");
6178 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6179
6180 small_page_write(addr, size);
6181
6182 os::Linux::release_memory_special_shm(addr, size);
6183 }
6184 }
6185
6186 static void test_reserve_memory_special_shm() {
6187 size_t lp = os::large_page_size();
6188 size_t ag = os::vm_allocation_granularity();
6189
6190 for (size_t size = ag; size < lp * 3; size += ag) {
6191 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6192 test_reserve_memory_special_shm(size, alignment);
6193 }
6194 }
6195 }
6196
6197 static void test() {
6198 test_reserve_memory_special_huge_tlbfs();
6199 test_reserve_memory_special_shm();
6200 }
6201 };
6202
6203 void TestReserveMemorySpecial_test() {
6204 TestReserveMemorySpecial::test();
6205 }
6206
6207 #endif