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 "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
110
111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
112 // getrusage() is prepared to handle the associated failure.
113 #ifndef RUSAGE_THREAD
114 #define RUSAGE_THREAD (1) /* only the calling thread */
115 #endif
116
117 #define MAX_PATH (2 * K)
118
119 #define MAX_SECS 100000000
120
121 // for timer info max values which include all bits
122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123
124 #define LARGEPAGES_BIT (1 << 6)
125 ////////////////////////////////////////////////////////////////////////////////
126 // global variables
127 julong os::Linux::_physical_memory = 0;
128
129 address os::Linux::_initial_thread_stack_bottom = NULL;
130 uintptr_t os::Linux::_initial_thread_stack_size = 0;
131
132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 const int os::Linux::_vm_default_page_size = (8 * K);
139 bool os::Linux::_is_floating_stack = false;
140 bool os::Linux::_is_NPTL = false;
141 bool os::Linux::_supports_fast_thread_cpu_time = false;
142 const char * os::Linux::_glibc_version = NULL;
143 const char * os::Linux::_libpthread_version = NULL;
144 pthread_condattr_t os::Linux::_condattr[1];
145
146 static jlong initial_time_count=0;
147
148 static int clock_tics_per_sec = 100;
149
150 // For diagnostics to print a message once. see run_periodic_checks
151 static sigset_t check_signal_done;
152 static bool check_signals = true;
153
154 static pid_t _initial_pid = 0;
155
156 // Signal number used to suspend/resume a thread
157
158 // do not use any signal number less than SIGSEGV, see 4355769
159 static int SR_signum = SIGUSR2;
160 sigset_t SR_sigset;
161
162 // Declarations
163 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 sysinfo(&si);
177
178 return (julong)si.freeram * si.mem_unit;
179 }
180
181 julong os::physical_memory() {
182 return Linux::physical_memory();
183 }
184
185 // Return true if user is running as root.
186
187 bool os::have_special_privileges() {
188 static bool init = false;
189 static bool privileges = false;
190 if (!init) {
191 privileges = (getuid() != geteuid()) || (getgid() != getegid());
192 init = true;
193 }
194 return privileges;
195 }
196
197
198 #ifndef SYS_gettid
199 // i386: 224, ia64: 1105, amd64: 186, sparc 143
200 #ifdef __ia64__
201 #define SYS_gettid 1105
202 #else
203 #ifdef __i386__
204 #define SYS_gettid 224
205 #else
206 #ifdef __amd64__
207 #define SYS_gettid 186
208 #else
209 #ifdef __sparc__
210 #define SYS_gettid 143
211 #else
212 #error define gettid for the arch
213 #endif
214 #endif
215 #endif
216 #endif
217 #endif
218
219 // Cpu architecture string
220 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
221
222
223 // pid_t gettid()
224 //
225 // Returns the kernel thread id of the currently running thread. Kernel
226 // thread id is used to access /proc.
227 //
228 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
229 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
230 //
231 pid_t os::Linux::gettid() {
232 int rslt = syscall(SYS_gettid);
233 if (rslt == -1) {
234 // old kernel, no NPTL support
235 return getpid();
236 } else {
237 return (pid_t)rslt;
238 }
239 }
240
241 // Most versions of linux have a bug where the number of processors are
242 // determined by looking at the /proc file system. In a chroot environment,
243 // the system call returns 1. This causes the VM to act as if it is
244 // a single processor and elide locking (see is_MP() call).
245 static bool unsafe_chroot_detected = false;
246 static const char *unstable_chroot_error = "/proc file system not found.\n"
247 "Java may be unstable running multithreaded in a chroot "
248 "environment on Linux when /proc filesystem is not mounted.";
249
250 void os::Linux::initialize_system_info() {
251 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
252 if (processor_count() == 1) {
253 pid_t pid = os::Linux::gettid();
254 char fname[32];
255 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
256 FILE *fp = fopen(fname, "r");
257 if (fp == NULL) {
258 unsafe_chroot_detected = true;
259 } else {
260 fclose(fp);
261 }
262 }
263 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
264 assert(processor_count() > 0, "linux error");
265 }
266
267 void os::init_system_properties_values() {
268 // The next steps are taken in the product version:
269 //
270 // Obtain the JAVA_HOME value from the location of libjvm.so.
271 // This library should be located at:
272 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
273 //
274 // If "/jre/lib/" appears at the right place in the path, then we
275 // assume libjvm.so is installed in a JDK and we use this path.
276 //
277 // Otherwise exit with message: "Could not create the Java virtual machine."
278 //
279 // The following extra steps are taken in the debugging version:
280 //
281 // If "/jre/lib/" does NOT appear at the right place in the path
282 // instead of exit check for $JAVA_HOME environment variable.
283 //
284 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
285 // then we append a fake suffix "hotspot/libjvm.so" to this path so
286 // it looks like libjvm.so is installed there
287 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
288 //
289 // Otherwise exit.
290 //
291 // Important note: if the location of libjvm.so changes this
292 // code needs to be changed accordingly.
293
294 // See ld(1):
295 // The linker uses the following search paths to locate required
296 // shared libraries:
297 // 1: ...
298 // ...
299 // 7: The default directories, normally /lib and /usr/lib.
300 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
301 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
302 #else
303 #define DEFAULT_LIBPATH "/lib:/usr/lib"
304 #endif
305
306 // Base path of extensions installed on the system.
307 #define SYS_EXT_DIR "/usr/java/packages"
308 #define EXTENSIONS_DIR "/lib/ext"
309
310 // Buffer that fits several sprintfs.
311 // Note that the space for the colon and the trailing null are provided
312 // by the nulls included by the sizeof operator.
313 const size_t bufsize =
314 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
315 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
316 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
317
318 // sysclasspath, java_home, dll_dir
319 {
320 char *pslash;
321 os::jvm_path(buf, bufsize);
322
323 // Found the full path to libjvm.so.
324 // Now cut the path to <java_home>/jre if we can.
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /libjvm.so.
328 }
329 pslash = strrchr(buf, '/');
330 if (pslash != NULL) {
331 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
332 }
333 Arguments::set_dll_dir(buf);
334
335 if (pslash != NULL) {
336 pslash = strrchr(buf, '/');
337 if (pslash != NULL) {
338 *pslash = '\0'; // Get rid of /<arch>.
339 pslash = strrchr(buf, '/');
340 if (pslash != NULL) {
341 *pslash = '\0'; // Get rid of /lib.
342 }
343 }
344 }
345 Arguments::set_java_home(buf);
346 set_boot_path('/', ':');
347 }
348
349 // Where to look for native libraries.
350 //
351 // Note: Due to a legacy implementation, most of the library path
352 // is set in the launcher. This was to accomodate linking restrictions
353 // on legacy Linux implementations (which are no longer supported).
354 // Eventually, all the library path setting will be done here.
355 //
356 // However, to prevent the proliferation of improperly built native
357 // libraries, the new path component /usr/java/packages is added here.
358 // Eventually, all the library path setting will be done here.
359 {
360 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
361 // should always exist (until the legacy problem cited above is
362 // addressed).
363 const char *v = ::getenv("LD_LIBRARY_PATH");
364 const char *v_colon = ":";
365 if (v == NULL) { v = ""; v_colon = ""; }
366 // That's +1 for the colon and +1 for the trailing '\0'.
367 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
368 strlen(v) + 1 +
369 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
370 mtInternal);
371 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
372 Arguments::set_library_path(ld_library_path);
373 FREE_C_HEAP_ARRAY(char, ld_library_path);
374 }
375
376 // Extensions directories.
377 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
378 Arguments::set_ext_dirs(buf);
379
380 FREE_C_HEAP_ARRAY(char, buf);
381
382 #undef DEFAULT_LIBPATH
383 #undef SYS_EXT_DIR
384 #undef EXTENSIONS_DIR
385 }
386
387 ////////////////////////////////////////////////////////////////////////////////
388 // breakpoint support
389
390 void os::breakpoint() {
391 BREAKPOINT;
392 }
393
394 extern "C" void breakpoint() {
395 // use debugger to set breakpoint here
396 }
397
398 ////////////////////////////////////////////////////////////////////////////////
399 // signal support
400
401 debug_only(static bool signal_sets_initialized = false);
402 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
403
404 bool os::Linux::is_sig_ignored(int sig) {
405 struct sigaction oact;
406 sigaction(sig, (struct sigaction*)NULL, &oact);
407 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
408 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
409 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
410 return true;
411 } else {
412 return false;
413 }
414 }
415
416 void os::Linux::signal_sets_init() {
417 // Should also have an assertion stating we are still single-threaded.
418 assert(!signal_sets_initialized, "Already initialized");
419 // Fill in signals that are necessarily unblocked for all threads in
420 // the VM. Currently, we unblock the following signals:
421 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
422 // by -Xrs (=ReduceSignalUsage));
423 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
424 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
425 // the dispositions or masks wrt these signals.
426 // Programs embedding the VM that want to use the above signals for their
427 // own purposes must, at this time, use the "-Xrs" option to prevent
428 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
429 // (See bug 4345157, and other related bugs).
430 // In reality, though, unblocking these signals is really a nop, since
431 // these signals are not blocked by default.
432 sigemptyset(&unblocked_sigs);
433 sigemptyset(&allowdebug_blocked_sigs);
434 sigaddset(&unblocked_sigs, SIGILL);
435 sigaddset(&unblocked_sigs, SIGSEGV);
436 sigaddset(&unblocked_sigs, SIGBUS);
437 sigaddset(&unblocked_sigs, SIGFPE);
438 #if defined(PPC64)
439 sigaddset(&unblocked_sigs, SIGTRAP);
440 #endif
441 sigaddset(&unblocked_sigs, SR_signum);
442
443 if (!ReduceSignalUsage) {
444 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
445 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
447 }
448 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
449 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
450 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
451 }
452 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
453 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
454 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
455 }
456 }
457 // Fill in signals that are blocked by all but the VM thread.
458 sigemptyset(&vm_sigs);
459 if (!ReduceSignalUsage) {
460 sigaddset(&vm_sigs, BREAK_SIGNAL);
461 }
462 debug_only(signal_sets_initialized = true);
463
464 }
465
466 // These are signals that are unblocked while a thread is running Java.
467 // (For some reason, they get blocked by default.)
468 sigset_t* os::Linux::unblocked_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &unblocked_sigs;
471 }
472
473 // These are the signals that are blocked while a (non-VM) thread is
474 // running Java. Only the VM thread handles these signals.
475 sigset_t* os::Linux::vm_signals() {
476 assert(signal_sets_initialized, "Not initialized");
477 return &vm_sigs;
478 }
479
480 // These are signals that are blocked during cond_wait to allow debugger in
481 sigset_t* os::Linux::allowdebug_blocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &allowdebug_blocked_sigs;
484 }
485
486 void os::Linux::hotspot_sigmask(Thread* thread) {
487
488 //Save caller's signal mask before setting VM signal mask
489 sigset_t caller_sigmask;
490 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
491
492 OSThread* osthread = thread->osthread();
493 osthread->set_caller_sigmask(caller_sigmask);
494
495 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
496
497 if (!ReduceSignalUsage) {
498 if (thread->is_VM_thread()) {
499 // Only the VM thread handles BREAK_SIGNAL ...
500 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
501 } else {
502 // ... all other threads block BREAK_SIGNAL
503 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
504 }
505 }
506 }
507
508 //////////////////////////////////////////////////////////////////////////////
509 // detecting pthread library
510
511 void os::Linux::libpthread_init() {
512 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
513 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
514 // generic name for earlier versions.
515 // Define macros here so we can build HotSpot on old systems.
516 #ifndef _CS_GNU_LIBC_VERSION
517 #define _CS_GNU_LIBC_VERSION 2
518 #endif
519 #ifndef _CS_GNU_LIBPTHREAD_VERSION
520 #define _CS_GNU_LIBPTHREAD_VERSION 3
521 #endif
522
523 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
524 if (n > 0) {
525 char *str = (char *)malloc(n, mtInternal);
526 confstr(_CS_GNU_LIBC_VERSION, str, n);
527 os::Linux::set_glibc_version(str);
528 } else {
529 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
530 static char _gnu_libc_version[32];
531 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
532 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
533 os::Linux::set_glibc_version(_gnu_libc_version);
534 }
535
536 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
537 if (n > 0) {
538 char *str = (char *)malloc(n, mtInternal);
539 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
540 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
541 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
542 // is the case. LinuxThreads has a hard limit on max number of threads.
543 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
544 // On the other hand, NPTL does not have such a limit, sysconf()
545 // will return -1 and errno is not changed. Check if it is really NPTL.
546 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
547 strstr(str, "NPTL") &&
548 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
549 free(str);
550 os::Linux::set_libpthread_version("linuxthreads");
551 } else {
552 os::Linux::set_libpthread_version(str);
553 }
554 } else {
555 // glibc before 2.3.2 only has LinuxThreads.
556 os::Linux::set_libpthread_version("linuxthreads");
557 }
558
559 if (strstr(libpthread_version(), "NPTL")) {
560 os::Linux::set_is_NPTL();
561 } else {
562 os::Linux::set_is_LinuxThreads();
563 }
564
565 // LinuxThreads have two flavors: floating-stack mode, which allows variable
566 // stack size; and fixed-stack mode. NPTL is always floating-stack.
567 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
568 os::Linux::set_is_floating_stack();
569 }
570 }
571
572 /////////////////////////////////////////////////////////////////////////////
573 // thread stack
574
575 // Force Linux kernel to expand current thread stack. If "bottom" is close
576 // to the stack guard, caller should block all signals.
577 //
578 // MAP_GROWSDOWN:
579 // A special mmap() flag that is used to implement thread stacks. It tells
580 // kernel that the memory region should extend downwards when needed. This
581 // allows early versions of LinuxThreads to only mmap the first few pages
582 // when creating a new thread. Linux kernel will automatically expand thread
583 // stack as needed (on page faults).
584 //
585 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
586 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
587 // region, it's hard to tell if the fault is due to a legitimate stack
588 // access or because of reading/writing non-exist memory (e.g. buffer
589 // overrun). As a rule, if the fault happens below current stack pointer,
590 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
591 // application (see Linux kernel fault.c).
592 //
593 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
594 // stack overflow detection.
595 //
596 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
597 // not use this flag. However, the stack of initial thread is not created
598 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
599 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
600 // and then attach the thread to JVM.
601 //
602 // To get around the problem and allow stack banging on Linux, we need to
603 // manually expand thread stack after receiving the SIGSEGV.
604 //
605 // There are two ways to expand thread stack to address "bottom", we used
606 // both of them in JVM before 1.5:
607 // 1. adjust stack pointer first so that it is below "bottom", and then
608 // touch "bottom"
609 // 2. mmap() the page in question
610 //
611 // Now alternate signal stack is gone, it's harder to use 2. For instance,
612 // if current sp is already near the lower end of page 101, and we need to
613 // call mmap() to map page 100, it is possible that part of the mmap() frame
614 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
615 // That will destroy the mmap() frame and cause VM to crash.
616 //
617 // The following code works by adjusting sp first, then accessing the "bottom"
618 // page to force a page fault. Linux kernel will then automatically expand the
619 // stack mapping.
620 //
621 // _expand_stack_to() assumes its frame size is less than page size, which
622 // should always be true if the function is not inlined.
623
624 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
625 #define NOINLINE
626 #else
627 #define NOINLINE __attribute__ ((noinline))
628 #endif
629
630 static void _expand_stack_to(address bottom) NOINLINE;
631
632 static void _expand_stack_to(address bottom) {
633 address sp;
634 size_t size;
635 volatile char *p;
636
637 // Adjust bottom to point to the largest address within the same page, it
638 // gives us a one-page buffer if alloca() allocates slightly more memory.
639 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
640 bottom += os::Linux::page_size() - 1;
641
642 // sp might be slightly above current stack pointer; if that's the case, we
643 // will alloca() a little more space than necessary, which is OK. Don't use
644 // os::current_stack_pointer(), as its result can be slightly below current
645 // stack pointer, causing us to not alloca enough to reach "bottom".
646 sp = (address)&sp;
647
648 if (sp > bottom) {
649 size = sp - bottom;
650 p = (volatile char *)alloca(size);
651 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
652 p[0] = '\0';
653 }
654 }
655
656 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
657 assert(t!=NULL, "just checking");
658 assert(t->osthread()->expanding_stack(), "expand should be set");
659 assert(t->stack_base() != NULL, "stack_base was not initialized");
660
661 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
662 sigset_t mask_all, old_sigset;
663 sigfillset(&mask_all);
664 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
665 _expand_stack_to(addr);
666 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
667 return true;
668 }
669 return false;
670 }
671
672 //////////////////////////////////////////////////////////////////////////////
673 // create new thread
674
675 static address highest_vm_reserved_address();
676
677 // check if it's safe to start a new thread
678 static bool _thread_safety_check(Thread* thread) {
679 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
680 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
681 // Heap is mmap'ed at lower end of memory space. Thread stacks are
682 // allocated (MAP_FIXED) from high address space. Every thread stack
683 // occupies a fixed size slot (usually 2Mbytes, but user can change
684 // it to other values if they rebuild LinuxThreads).
685 //
686 // Problem with MAP_FIXED is that mmap() can still succeed even part of
687 // the memory region has already been mmap'ed. That means if we have too
688 // many threads and/or very large heap, eventually thread stack will
689 // collide with heap.
690 //
691 // Here we try to prevent heap/stack collision by comparing current
692 // stack bottom with the highest address that has been mmap'ed by JVM
693 // plus a safety margin for memory maps created by native code.
694 //
695 // This feature can be disabled by setting ThreadSafetyMargin to 0
696 //
697 if (ThreadSafetyMargin > 0) {
698 address stack_bottom = os::current_stack_base() - os::current_stack_size();
699
700 // not safe if our stack extends below the safety margin
701 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
702 } else {
703 return true;
704 }
705 } else {
706 // Floating stack LinuxThreads or NPTL:
707 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
708 // there's not enough space left, pthread_create() will fail. If we come
709 // here, that means enough space has been reserved for stack.
710 return true;
711 }
712 }
713
714 // Thread start routine for all newly created threads
715 static void *java_start(Thread *thread) {
716 // Try to randomize the cache line index of hot stack frames.
717 // This helps when threads of the same stack traces evict each other's
718 // cache lines. The threads can be either from the same JVM instance, or
719 // from different JVM instances. The benefit is especially true for
720 // processors with hyperthreading technology.
721 static int counter = 0;
722 int pid = os::current_process_id();
723 alloca(((pid ^ counter++) & 7) * 128);
724
725 ThreadLocalStorage::set_thread(thread);
726
727 OSThread* osthread = thread->osthread();
728 Monitor* sync = osthread->startThread_lock();
729
730 // non floating stack LinuxThreads needs extra check, see above
731 if (!_thread_safety_check(thread)) {
732 // notify parent thread
733 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
734 osthread->set_state(ZOMBIE);
735 sync->notify_all();
736 return NULL;
737 }
738
739 // thread_id is kernel thread id (similar to Solaris LWP id)
740 osthread->set_thread_id(os::Linux::gettid());
741
742 if (UseNUMA) {
743 int lgrp_id = os::numa_get_group_id();
744 if (lgrp_id != -1) {
745 thread->set_lgrp_id(lgrp_id);
746 }
747 }
748 // initialize signal mask for this thread
749 os::Linux::hotspot_sigmask(thread);
750
751 // initialize floating point control register
752 os::Linux::init_thread_fpu_state();
753
754 // handshaking with parent thread
755 {
756 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
757
758 // notify parent thread
759 osthread->set_state(INITIALIZED);
760 sync->notify_all();
761
762 // wait until os::start_thread()
763 while (osthread->get_state() == INITIALIZED) {
764 sync->wait(Mutex::_no_safepoint_check_flag);
765 }
766 }
767
768 // call one more level start routine
769 thread->run();
770
771 return 0;
772 }
773
774 bool os::create_thread(Thread* thread, ThreadType thr_type,
775 size_t stack_size) {
776 assert(thread->osthread() == NULL, "caller responsible");
777
778 // Allocate the OSThread object
779 OSThread* osthread = new OSThread(NULL, NULL);
780 if (osthread == NULL) {
781 return false;
782 }
783
784 // set the correct thread state
785 osthread->set_thread_type(thr_type);
786
787 // Initial state is ALLOCATED but not INITIALIZED
788 osthread->set_state(ALLOCATED);
789
790 thread->set_osthread(osthread);
791
792 // init thread attributes
793 pthread_attr_t attr;
794 pthread_attr_init(&attr);
795 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
796
797 // stack size
798 if (os::Linux::supports_variable_stack_size()) {
799 // calculate stack size if it's not specified by caller
800 if (stack_size == 0) {
801 stack_size = os::Linux::default_stack_size(thr_type);
802
803 switch (thr_type) {
804 case os::java_thread:
805 // Java threads use ThreadStackSize which default value can be
806 // changed with the flag -Xss
807 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
808 stack_size = JavaThread::stack_size_at_create();
809 break;
810 case os::compiler_thread:
811 if (CompilerThreadStackSize > 0) {
812 stack_size = (size_t)(CompilerThreadStackSize * K);
813 break;
814 } // else fall through:
815 // use VMThreadStackSize if CompilerThreadStackSize is not defined
816 case os::vm_thread:
817 case os::pgc_thread:
818 case os::cgc_thread:
819 case os::watcher_thread:
820 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
821 break;
822 }
823 }
824
825 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
826 pthread_attr_setstacksize(&attr, stack_size);
827 } else {
828 // let pthread_create() pick the default value.
829 }
830
831 // glibc guard page
832 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
833
834 ThreadState state;
835
836 {
837 // Serialize thread creation if we are running with fixed stack LinuxThreads
838 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
839 if (lock) {
840 os::Linux::createThread_lock()->lock_without_safepoint_check();
841 }
842
843 pthread_t tid;
844 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
845
846 pthread_attr_destroy(&attr);
847
848 if (ret != 0) {
849 if (PrintMiscellaneous && (Verbose || WizardMode)) {
850 perror("pthread_create()");
851 }
852 // Need to clean up stuff we've allocated so far
853 thread->set_osthread(NULL);
854 delete osthread;
855 if (lock) os::Linux::createThread_lock()->unlock();
856 return false;
857 }
858
859 // Store pthread info into the OSThread
860 osthread->set_pthread_id(tid);
861
862 // Wait until child thread is either initialized or aborted
863 {
864 Monitor* sync_with_child = osthread->startThread_lock();
865 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
866 while ((state = osthread->get_state()) == ALLOCATED) {
867 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
868 }
869 }
870
871 if (lock) {
872 os::Linux::createThread_lock()->unlock();
873 }
874 }
875
876 // Aborted due to thread limit being reached
877 if (state == ZOMBIE) {
878 thread->set_osthread(NULL);
879 delete osthread;
880 return false;
881 }
882
883 // The thread is returned suspended (in state INITIALIZED),
884 // and is started higher up in the call chain
885 assert(state == INITIALIZED, "race condition");
886 return true;
887 }
888
889 /////////////////////////////////////////////////////////////////////////////
890 // attach existing thread
891
892 // bootstrap the main thread
893 bool os::create_main_thread(JavaThread* thread) {
894 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
895 return create_attached_thread(thread);
896 }
897
898 bool os::create_attached_thread(JavaThread* thread) {
899 #ifdef ASSERT
900 thread->verify_not_published();
901 #endif
902
903 // Allocate the OSThread object
904 OSThread* osthread = new OSThread(NULL, NULL);
905
906 if (osthread == NULL) {
907 return false;
908 }
909
910 // Store pthread info into the OSThread
911 osthread->set_thread_id(os::Linux::gettid());
912 osthread->set_pthread_id(::pthread_self());
913
914 // initialize floating point control register
915 os::Linux::init_thread_fpu_state();
916
917 // Initial thread state is RUNNABLE
918 osthread->set_state(RUNNABLE);
919
920 thread->set_osthread(osthread);
921
922 if (UseNUMA) {
923 int lgrp_id = os::numa_get_group_id();
924 if (lgrp_id != -1) {
925 thread->set_lgrp_id(lgrp_id);
926 }
927 }
928
929 if (os::Linux::is_initial_thread()) {
930 // If current thread is initial thread, its stack is mapped on demand,
931 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
932 // the entire stack region to avoid SEGV in stack banging.
933 // It is also useful to get around the heap-stack-gap problem on SuSE
934 // kernel (see 4821821 for details). We first expand stack to the top
935 // of yellow zone, then enable stack yellow zone (order is significant,
936 // enabling yellow zone first will crash JVM on SuSE Linux), so there
937 // is no gap between the last two virtual memory regions.
938
939 JavaThread *jt = (JavaThread *)thread;
940 address addr = jt->stack_yellow_zone_base();
941 assert(addr != NULL, "initialization problem?");
942 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
943
944 osthread->set_expanding_stack();
945 os::Linux::manually_expand_stack(jt, addr);
946 osthread->clear_expanding_stack();
947 }
948
949 // initialize signal mask for this thread
950 // and save the caller's signal mask
951 os::Linux::hotspot_sigmask(thread);
952
953 return true;
954 }
955
956 void os::pd_start_thread(Thread* thread) {
957 OSThread * osthread = thread->osthread();
958 assert(osthread->get_state() != INITIALIZED, "just checking");
959 Monitor* sync_with_child = osthread->startThread_lock();
960 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
961 sync_with_child->notify();
962 }
963
964 // Free Linux resources related to the OSThread
965 void os::free_thread(OSThread* osthread) {
966 assert(osthread != NULL, "osthread not set");
967
968 if (Thread::current()->osthread() == osthread) {
969 // Restore caller's signal mask
970 sigset_t sigmask = osthread->caller_sigmask();
971 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
972 }
973
974 delete osthread;
975 }
976
977 //////////////////////////////////////////////////////////////////////////////
978 // thread local storage
979
980 // Restore the thread pointer if the destructor is called. This is in case
981 // someone from JNI code sets up a destructor with pthread_key_create to run
982 // detachCurrentThread on thread death. Unless we restore the thread pointer we
983 // will hang or crash. When detachCurrentThread is called the key will be set
984 // to null and we will not be called again. If detachCurrentThread is never
985 // called we could loop forever depending on the pthread implementation.
986 static void restore_thread_pointer(void* p) {
987 Thread* thread = (Thread*) p;
988 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
989 }
990
991 int os::allocate_thread_local_storage() {
992 pthread_key_t key;
993 int rslt = pthread_key_create(&key, restore_thread_pointer);
994 assert(rslt == 0, "cannot allocate thread local storage");
995 return (int)key;
996 }
997
998 // Note: This is currently not used by VM, as we don't destroy TLS key
999 // on VM exit.
1000 void os::free_thread_local_storage(int index) {
1001 int rslt = pthread_key_delete((pthread_key_t)index);
1002 assert(rslt == 0, "invalid index");
1003 }
1004
1005 void os::thread_local_storage_at_put(int index, void* value) {
1006 int rslt = pthread_setspecific((pthread_key_t)index, value);
1007 assert(rslt == 0, "pthread_setspecific failed");
1008 }
1009
1010 extern "C" Thread* get_thread() {
1011 return ThreadLocalStorage::thread();
1012 }
1013
1014 //////////////////////////////////////////////////////////////////////////////
1015 // initial thread
1016
1017 // Check if current thread is the initial thread, similar to Solaris thr_main.
1018 bool os::Linux::is_initial_thread(void) {
1019 char dummy;
1020 // If called before init complete, thread stack bottom will be null.
1021 // Can be called if fatal error occurs before initialization.
1022 if (initial_thread_stack_bottom() == NULL) return false;
1023 assert(initial_thread_stack_bottom() != NULL &&
1024 initial_thread_stack_size() != 0,
1025 "os::init did not locate initial thread's stack region");
1026 if ((address)&dummy >= initial_thread_stack_bottom() &&
1027 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1028 return true;
1029 } else {
1030 return false;
1031 }
1032 }
1033
1034 // Find the virtual memory area that contains addr
1035 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1036 FILE *fp = fopen("/proc/self/maps", "r");
1037 if (fp) {
1038 address low, high;
1039 while (!feof(fp)) {
1040 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1041 if (low <= addr && addr < high) {
1042 if (vma_low) *vma_low = low;
1043 if (vma_high) *vma_high = high;
1044 fclose(fp);
1045 return true;
1046 }
1047 }
1048 for (;;) {
1049 int ch = fgetc(fp);
1050 if (ch == EOF || ch == (int)'\n') break;
1051 }
1052 }
1053 fclose(fp);
1054 }
1055 return false;
1056 }
1057
1058 // Locate initial thread stack. This special handling of initial thread stack
1059 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1060 // bogus value for initial thread.
1061 void os::Linux::capture_initial_stack(size_t max_size) {
1062 // stack size is the easy part, get it from RLIMIT_STACK
1063 size_t stack_size;
1064 struct rlimit rlim;
1065 getrlimit(RLIMIT_STACK, &rlim);
1066 stack_size = rlim.rlim_cur;
1067
1068 // 6308388: a bug in ld.so will relocate its own .data section to the
1069 // lower end of primordial stack; reduce ulimit -s value a little bit
1070 // so we won't install guard page on ld.so's data section.
1071 stack_size -= 2 * page_size();
1072
1073 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1074 // 7.1, in both cases we will get 2G in return value.
1075 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1076 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1077 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1078 // in case other parts in glibc still assumes 2M max stack size.
1079 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1080 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1081 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1082 stack_size = 2 * K * K IA64_ONLY(*2);
1083 }
1084 // Try to figure out where the stack base (top) is. This is harder.
1085 //
1086 // When an application is started, glibc saves the initial stack pointer in
1087 // a global variable "__libc_stack_end", which is then used by system
1088 // libraries. __libc_stack_end should be pretty close to stack top. The
1089 // variable is available since the very early days. However, because it is
1090 // a private interface, it could disappear in the future.
1091 //
1092 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1093 // to __libc_stack_end, it is very close to stack top, but isn't the real
1094 // stack top. Note that /proc may not exist if VM is running as a chroot
1095 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1096 // /proc/<pid>/stat could change in the future (though unlikely).
1097 //
1098 // We try __libc_stack_end first. If that doesn't work, look for
1099 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1100 // as a hint, which should work well in most cases.
1101
1102 uintptr_t stack_start;
1103
1104 // try __libc_stack_end first
1105 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1106 if (p && *p) {
1107 stack_start = *p;
1108 } else {
1109 // see if we can get the start_stack field from /proc/self/stat
1110 FILE *fp;
1111 int pid;
1112 char state;
1113 int ppid;
1114 int pgrp;
1115 int session;
1116 int nr;
1117 int tpgrp;
1118 unsigned long flags;
1119 unsigned long minflt;
1120 unsigned long cminflt;
1121 unsigned long majflt;
1122 unsigned long cmajflt;
1123 unsigned long utime;
1124 unsigned long stime;
1125 long cutime;
1126 long cstime;
1127 long prio;
1128 long nice;
1129 long junk;
1130 long it_real;
1131 uintptr_t start;
1132 uintptr_t vsize;
1133 intptr_t rss;
1134 uintptr_t rsslim;
1135 uintptr_t scodes;
1136 uintptr_t ecode;
1137 int i;
1138
1139 // Figure what the primordial thread stack base is. Code is inspired
1140 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1141 // followed by command name surrounded by parentheses, state, etc.
1142 char stat[2048];
1143 int statlen;
1144
1145 fp = fopen("/proc/self/stat", "r");
1146 if (fp) {
1147 statlen = fread(stat, 1, 2047, fp);
1148 stat[statlen] = '\0';
1149 fclose(fp);
1150
1151 // Skip pid and the command string. Note that we could be dealing with
1152 // weird command names, e.g. user could decide to rename java launcher
1153 // to "java 1.4.2 :)", then the stat file would look like
1154 // 1234 (java 1.4.2 :)) R ... ...
1155 // We don't really need to know the command string, just find the last
1156 // occurrence of ")" and then start parsing from there. See bug 4726580.
1157 char * s = strrchr(stat, ')');
1158
1159 i = 0;
1160 if (s) {
1161 // Skip blank chars
1162 do { s++; } while (s && isspace(*s));
1163
1164 #define _UFM UINTX_FORMAT
1165 #define _DFM INTX_FORMAT
1166
1167 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1168 // 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
1169 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,
1170 &state, // 3 %c
1171 &ppid, // 4 %d
1172 &pgrp, // 5 %d
1173 &session, // 6 %d
1174 &nr, // 7 %d
1175 &tpgrp, // 8 %d
1176 &flags, // 9 %lu
1177 &minflt, // 10 %lu
1178 &cminflt, // 11 %lu
1179 &majflt, // 12 %lu
1180 &cmajflt, // 13 %lu
1181 &utime, // 14 %lu
1182 &stime, // 15 %lu
1183 &cutime, // 16 %ld
1184 &cstime, // 17 %ld
1185 &prio, // 18 %ld
1186 &nice, // 19 %ld
1187 &junk, // 20 %ld
1188 &it_real, // 21 %ld
1189 &start, // 22 UINTX_FORMAT
1190 &vsize, // 23 UINTX_FORMAT
1191 &rss, // 24 INTX_FORMAT
1192 &rsslim, // 25 UINTX_FORMAT
1193 &scodes, // 26 UINTX_FORMAT
1194 &ecode, // 27 UINTX_FORMAT
1195 &stack_start); // 28 UINTX_FORMAT
1196 }
1197
1198 #undef _UFM
1199 #undef _DFM
1200
1201 if (i != 28 - 2) {
1202 assert(false, "Bad conversion from /proc/self/stat");
1203 // product mode - assume we are the initial thread, good luck in the
1204 // embedded case.
1205 warning("Can't detect initial thread stack location - bad conversion");
1206 stack_start = (uintptr_t) &rlim;
1207 }
1208 } else {
1209 // For some reason we can't open /proc/self/stat (for example, running on
1210 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1211 // most cases, so don't abort:
1212 warning("Can't detect initial thread stack location - no /proc/self/stat");
1213 stack_start = (uintptr_t) &rlim;
1214 }
1215 }
1216
1217 // Now we have a pointer (stack_start) very close to the stack top, the
1218 // next thing to do is to figure out the exact location of stack top. We
1219 // can find out the virtual memory area that contains stack_start by
1220 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1221 // and its upper limit is the real stack top. (again, this would fail if
1222 // running inside chroot, because /proc may not exist.)
1223
1224 uintptr_t stack_top;
1225 address low, high;
1226 if (find_vma((address)stack_start, &low, &high)) {
1227 // success, "high" is the true stack top. (ignore "low", because initial
1228 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1229 stack_top = (uintptr_t)high;
1230 } else {
1231 // failed, likely because /proc/self/maps does not exist
1232 warning("Can't detect initial thread stack location - find_vma failed");
1233 // best effort: stack_start is normally within a few pages below the real
1234 // stack top, use it as stack top, and reduce stack size so we won't put
1235 // guard page outside stack.
1236 stack_top = stack_start;
1237 stack_size -= 16 * page_size();
1238 }
1239
1240 // stack_top could be partially down the page so align it
1241 stack_top = align_size_up(stack_top, page_size());
1242
1243 if (max_size && stack_size > max_size) {
1244 _initial_thread_stack_size = max_size;
1245 } else {
1246 _initial_thread_stack_size = stack_size;
1247 }
1248
1249 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1250 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1251 }
1252
1253 ////////////////////////////////////////////////////////////////////////////////
1254 // time support
1255
1256 // Time since start-up in seconds to a fine granularity.
1257 // Used by VMSelfDestructTimer and the MemProfiler.
1258 double os::elapsedTime() {
1259
1260 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1261 }
1262
1263 jlong os::elapsed_counter() {
1264 return javaTimeNanos() - initial_time_count;
1265 }
1266
1267 jlong os::elapsed_frequency() {
1268 return NANOSECS_PER_SEC; // nanosecond resolution
1269 }
1270
1271 bool os::supports_vtime() { return true; }
1272 bool os::enable_vtime() { return false; }
1273 bool os::vtime_enabled() { return false; }
1274
1275 double os::elapsedVTime() {
1276 struct rusage usage;
1277 int retval = getrusage(RUSAGE_THREAD, &usage);
1278 if (retval == 0) {
1279 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);
1280 } else {
1281 // better than nothing, but not much
1282 return elapsedTime();
1283 }
1284 }
1285
1286 jlong os::javaTimeMillis() {
1287 timeval time;
1288 int status = gettimeofday(&time, NULL);
1289 assert(status != -1, "linux error");
1290 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1291 }
1292
1293 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1294 timeval time;
1295 int status = gettimeofday(&time, NULL);
1296 assert(status != -1, "linux error");
1297 seconds = jlong(time.tv_sec);
1298 nanos = jlong(time.tv_usec) * 1000;
1299 }
1300
1301
1302 #ifndef CLOCK_MONOTONIC
1303 #define CLOCK_MONOTONIC (1)
1304 #endif
1305
1306 void os::Linux::clock_init() {
1307 // we do dlopen's in this particular order due to bug in linux
1308 // dynamical loader (see 6348968) leading to crash on exit
1309 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1310 if (handle == NULL) {
1311 handle = dlopen("librt.so", RTLD_LAZY);
1312 }
1313
1314 if (handle) {
1315 int (*clock_getres_func)(clockid_t, struct timespec*) =
1316 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1317 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1318 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1319 if (clock_getres_func && clock_gettime_func) {
1320 // See if monotonic clock is supported by the kernel. Note that some
1321 // early implementations simply return kernel jiffies (updated every
1322 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1323 // for nano time (though the monotonic property is still nice to have).
1324 // It's fixed in newer kernels, however clock_getres() still returns
1325 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1326 // resolution for now. Hopefully as people move to new kernels, this
1327 // won't be a problem.
1328 struct timespec res;
1329 struct timespec tp;
1330 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1331 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1332 // yes, monotonic clock is supported
1333 _clock_gettime = clock_gettime_func;
1334 return;
1335 } else {
1336 // close librt if there is no monotonic clock
1337 dlclose(handle);
1338 }
1339 }
1340 }
1341 warning("No monotonic clock was available - timed services may " \
1342 "be adversely affected if the time-of-day clock changes");
1343 }
1344
1345 #ifndef SYS_clock_getres
1346 #if defined(IA32) || defined(AMD64)
1347 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1348 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1349 #else
1350 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1351 #define sys_clock_getres(x,y) -1
1352 #endif
1353 #else
1354 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1355 #endif
1356
1357 void os::Linux::fast_thread_clock_init() {
1358 if (!UseLinuxPosixThreadCPUClocks) {
1359 return;
1360 }
1361 clockid_t clockid;
1362 struct timespec tp;
1363 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1364 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1365
1366 // Switch to using fast clocks for thread cpu time if
1367 // the sys_clock_getres() returns 0 error code.
1368 // Note, that some kernels may support the current thread
1369 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1370 // returned by the pthread_getcpuclockid().
1371 // If the fast Posix clocks are supported then the sys_clock_getres()
1372 // must return at least tp.tv_sec == 0 which means a resolution
1373 // better than 1 sec. This is extra check for reliability.
1374
1375 if (pthread_getcpuclockid_func &&
1376 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1377 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1378 _supports_fast_thread_cpu_time = true;
1379 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1380 }
1381 }
1382
1383 jlong os::javaTimeNanos() {
1384 if (os::supports_monotonic_clock()) {
1385 struct timespec tp;
1386 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1387 assert(status == 0, "gettime error");
1388 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1389 return result;
1390 } else {
1391 timeval time;
1392 int status = gettimeofday(&time, NULL);
1393 assert(status != -1, "linux error");
1394 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1395 return 1000 * usecs;
1396 }
1397 }
1398
1399 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1400 if (os::supports_monotonic_clock()) {
1401 info_ptr->max_value = ALL_64_BITS;
1402
1403 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1404 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1405 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1406 } else {
1407 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1408 info_ptr->max_value = ALL_64_BITS;
1409
1410 // gettimeofday is a real time clock so it skips
1411 info_ptr->may_skip_backward = true;
1412 info_ptr->may_skip_forward = true;
1413 }
1414
1415 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1416 }
1417
1418 // Return the real, user, and system times in seconds from an
1419 // arbitrary fixed point in the past.
1420 bool os::getTimesSecs(double* process_real_time,
1421 double* process_user_time,
1422 double* process_system_time) {
1423 struct tms ticks;
1424 clock_t real_ticks = times(&ticks);
1425
1426 if (real_ticks == (clock_t) (-1)) {
1427 return false;
1428 } else {
1429 double ticks_per_second = (double) clock_tics_per_sec;
1430 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1431 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1432 *process_real_time = ((double) real_ticks) / ticks_per_second;
1433
1434 return true;
1435 }
1436 }
1437
1438
1439 char * os::local_time_string(char *buf, size_t buflen) {
1440 struct tm t;
1441 time_t long_time;
1442 time(&long_time);
1443 localtime_r(&long_time, &t);
1444 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1445 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1446 t.tm_hour, t.tm_min, t.tm_sec);
1447 return buf;
1448 }
1449
1450 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1451 return localtime_r(clock, res);
1452 }
1453
1454 ////////////////////////////////////////////////////////////////////////////////
1455 // runtime exit support
1456
1457 // Note: os::shutdown() might be called very early during initialization, or
1458 // called from signal handler. Before adding something to os::shutdown(), make
1459 // sure it is async-safe and can handle partially initialized VM.
1460 void os::shutdown() {
1461
1462 // allow PerfMemory to attempt cleanup of any persistent resources
1463 perfMemory_exit();
1464
1465 // needs to remove object in file system
1466 AttachListener::abort();
1467
1468 // flush buffered output, finish log files
1469 ostream_abort();
1470
1471 // Check for abort hook
1472 abort_hook_t abort_hook = Arguments::abort_hook();
1473 if (abort_hook != NULL) {
1474 abort_hook();
1475 }
1476
1477 }
1478
1479 // Note: os::abort() might be called very early during initialization, or
1480 // called from signal handler. Before adding something to os::abort(), make
1481 // sure it is async-safe and can handle partially initialized VM.
1482 void os::abort(bool dump_core) {
1483 abort(dump_core, NULL, NULL);
1484 }
1485
1486 void os::abort(bool dump_core, void* siginfo, void* context) {
1487 os::shutdown();
1488 if (dump_core) {
1489 #ifndef PRODUCT
1490 fdStream out(defaultStream::output_fd());
1491 out.print_raw("Current thread is ");
1492 char buf[16];
1493 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1494 out.print_raw_cr(buf);
1495 out.print_raw_cr("Dumping core ...");
1496 #endif
1497 ::abort(); // dump core
1498 }
1499
1500 ::exit(1);
1501 }
1502
1503 // Die immediately, no exit hook, no abort hook, no cleanup.
1504 void os::die() {
1505 // _exit() on LinuxThreads only kills current thread
1506 ::abort();
1507 }
1508
1509
1510 // This method is a copy of JDK's sysGetLastErrorString
1511 // from src/solaris/hpi/src/system_md.c
1512
1513 size_t os::lasterror(char *buf, size_t len) {
1514 if (errno == 0) return 0;
1515
1516 const char *s = ::strerror(errno);
1517 size_t n = ::strlen(s);
1518 if (n >= len) {
1519 n = len - 1;
1520 }
1521 ::strncpy(buf, s, n);
1522 buf[n] = '\0';
1523 return n;
1524 }
1525
1526 intx os::current_thread_id() { return (intx)pthread_self(); }
1527 int os::current_process_id() {
1528
1529 // Under the old linux thread library, linux gives each thread
1530 // its own process id. Because of this each thread will return
1531 // a different pid if this method were to return the result
1532 // of getpid(2). Linux provides no api that returns the pid
1533 // of the launcher thread for the vm. This implementation
1534 // returns a unique pid, the pid of the launcher thread
1535 // that starts the vm 'process'.
1536
1537 // Under the NPTL, getpid() returns the same pid as the
1538 // launcher thread rather than a unique pid per thread.
1539 // Use gettid() if you want the old pre NPTL behaviour.
1540
1541 // if you are looking for the result of a call to getpid() that
1542 // returns a unique pid for the calling thread, then look at the
1543 // OSThread::thread_id() method in osThread_linux.hpp file
1544
1545 return (int)(_initial_pid ? _initial_pid : getpid());
1546 }
1547
1548 // DLL functions
1549
1550 const char* os::dll_file_extension() { return ".so"; }
1551
1552 // This must be hard coded because it's the system's temporary
1553 // directory not the java application's temp directory, ala java.io.tmpdir.
1554 const char* os::get_temp_directory() { return "/tmp"; }
1555
1556 static bool file_exists(const char* filename) {
1557 struct stat statbuf;
1558 if (filename == NULL || strlen(filename) == 0) {
1559 return false;
1560 }
1561 return os::stat(filename, &statbuf) == 0;
1562 }
1563
1564 bool os::dll_build_name(char* buffer, size_t buflen,
1565 const char* pname, const char* fname) {
1566 bool retval = false;
1567 // Copied from libhpi
1568 const size_t pnamelen = pname ? strlen(pname) : 0;
1569
1570 // Return error on buffer overflow.
1571 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1572 return retval;
1573 }
1574
1575 if (pnamelen == 0) {
1576 snprintf(buffer, buflen, "lib%s.so", fname);
1577 retval = true;
1578 } else if (strchr(pname, *os::path_separator()) != NULL) {
1579 int n;
1580 char** pelements = split_path(pname, &n);
1581 if (pelements == NULL) {
1582 return false;
1583 }
1584 for (int i = 0; i < n; i++) {
1585 // Really shouldn't be NULL, but check can't hurt
1586 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1587 continue; // skip the empty path values
1588 }
1589 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1590 if (file_exists(buffer)) {
1591 retval = true;
1592 break;
1593 }
1594 }
1595 // release the storage
1596 for (int i = 0; i < n; i++) {
1597 if (pelements[i] != NULL) {
1598 FREE_C_HEAP_ARRAY(char, pelements[i]);
1599 }
1600 }
1601 if (pelements != NULL) {
1602 FREE_C_HEAP_ARRAY(char*, pelements);
1603 }
1604 } else {
1605 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1606 retval = true;
1607 }
1608 return retval;
1609 }
1610
1611 // check if addr is inside libjvm.so
1612 bool os::address_is_in_vm(address addr) {
1613 static address libjvm_base_addr;
1614 Dl_info dlinfo;
1615
1616 if (libjvm_base_addr == NULL) {
1617 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1618 libjvm_base_addr = (address)dlinfo.dli_fbase;
1619 }
1620 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1621 }
1622
1623 if (dladdr((void *)addr, &dlinfo) != 0) {
1624 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1625 }
1626
1627 return false;
1628 }
1629
1630 bool os::dll_address_to_function_name(address addr, char *buf,
1631 int buflen, int *offset) {
1632 // buf is not optional, but offset is optional
1633 assert(buf != NULL, "sanity check");
1634
1635 Dl_info dlinfo;
1636
1637 if (dladdr((void*)addr, &dlinfo) != 0) {
1638 // see if we have a matching symbol
1639 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1640 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1641 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1642 }
1643 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1644 return true;
1645 }
1646 // no matching symbol so try for just file info
1647 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1648 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1649 buf, buflen, offset, dlinfo.dli_fname)) {
1650 return true;
1651 }
1652 }
1653 }
1654
1655 buf[0] = '\0';
1656 if (offset != NULL) *offset = -1;
1657 return false;
1658 }
1659
1660 struct _address_to_library_name {
1661 address addr; // input : memory address
1662 size_t buflen; // size of fname
1663 char* fname; // output: library name
1664 address base; // library base addr
1665 };
1666
1667 static int address_to_library_name_callback(struct dl_phdr_info *info,
1668 size_t size, void *data) {
1669 int i;
1670 bool found = false;
1671 address libbase = NULL;
1672 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1673
1674 // iterate through all loadable segments
1675 for (i = 0; i < info->dlpi_phnum; i++) {
1676 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1677 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1678 // base address of a library is the lowest address of its loaded
1679 // segments.
1680 if (libbase == NULL || libbase > segbase) {
1681 libbase = segbase;
1682 }
1683 // see if 'addr' is within current segment
1684 if (segbase <= d->addr &&
1685 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1686 found = true;
1687 }
1688 }
1689 }
1690
1691 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1692 // so dll_address_to_library_name() can fall through to use dladdr() which
1693 // can figure out executable name from argv[0].
1694 if (found && info->dlpi_name && info->dlpi_name[0]) {
1695 d->base = libbase;
1696 if (d->fname) {
1697 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1698 }
1699 return 1;
1700 }
1701 return 0;
1702 }
1703
1704 bool os::dll_address_to_library_name(address addr, char* buf,
1705 int buflen, int* offset) {
1706 // buf is not optional, but offset is optional
1707 assert(buf != NULL, "sanity check");
1708
1709 Dl_info dlinfo;
1710 struct _address_to_library_name data;
1711
1712 // There is a bug in old glibc dladdr() implementation that it could resolve
1713 // to wrong library name if the .so file has a base address != NULL. Here
1714 // we iterate through the program headers of all loaded libraries to find
1715 // out which library 'addr' really belongs to. This workaround can be
1716 // removed once the minimum requirement for glibc is moved to 2.3.x.
1717 data.addr = addr;
1718 data.fname = buf;
1719 data.buflen = buflen;
1720 data.base = NULL;
1721 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1722
1723 if (rslt) {
1724 // buf already contains library name
1725 if (offset) *offset = addr - data.base;
1726 return true;
1727 }
1728 if (dladdr((void*)addr, &dlinfo) != 0) {
1729 if (dlinfo.dli_fname != NULL) {
1730 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1731 }
1732 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1733 *offset = addr - (address)dlinfo.dli_fbase;
1734 }
1735 return true;
1736 }
1737
1738 buf[0] = '\0';
1739 if (offset) *offset = -1;
1740 return false;
1741 }
1742
1743 // Loads .dll/.so and
1744 // in case of error it checks if .dll/.so was built for the
1745 // same architecture as Hotspot is running on
1746
1747
1748 // Remember the stack's state. The Linux dynamic linker will change
1749 // the stack to 'executable' at most once, so we must safepoint only once.
1750 bool os::Linux::_stack_is_executable = false;
1751
1752 // VM operation that loads a library. This is necessary if stack protection
1753 // of the Java stacks can be lost during loading the library. If we
1754 // do not stop the Java threads, they can stack overflow before the stacks
1755 // are protected again.
1756 class VM_LinuxDllLoad: public VM_Operation {
1757 private:
1758 const char *_filename;
1759 char *_ebuf;
1760 int _ebuflen;
1761 void *_lib;
1762 public:
1763 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1764 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1765 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1766 void doit() {
1767 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1768 os::Linux::_stack_is_executable = true;
1769 }
1770 void* loaded_library() { return _lib; }
1771 };
1772
1773 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1774 void * result = NULL;
1775 bool load_attempted = false;
1776
1777 // Check whether the library to load might change execution rights
1778 // of the stack. If they are changed, the protection of the stack
1779 // guard pages will be lost. We need a safepoint to fix this.
1780 //
1781 // See Linux man page execstack(8) for more info.
1782 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1783 ElfFile ef(filename);
1784 if (!ef.specifies_noexecstack()) {
1785 if (!is_init_completed()) {
1786 os::Linux::_stack_is_executable = true;
1787 // This is OK - No Java threads have been created yet, and hence no
1788 // stack guard pages to fix.
1789 //
1790 // This should happen only when you are building JDK7 using a very
1791 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1792 //
1793 // Dynamic loader will make all stacks executable after
1794 // this function returns, and will not do that again.
1795 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1796 } else {
1797 warning("You have loaded library %s which might have disabled stack guard. "
1798 "The VM will try to fix the stack guard now.\n"
1799 "It's highly recommended that you fix the library with "
1800 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1801 filename);
1802
1803 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1804 JavaThread *jt = JavaThread::current();
1805 if (jt->thread_state() != _thread_in_native) {
1806 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1807 // that requires ExecStack. Cannot enter safe point. Let's give up.
1808 warning("Unable to fix stack guard. Giving up.");
1809 } else {
1810 if (!LoadExecStackDllInVMThread) {
1811 // This is for the case where the DLL has an static
1812 // constructor function that executes JNI code. We cannot
1813 // load such DLLs in the VMThread.
1814 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1815 }
1816
1817 ThreadInVMfromNative tiv(jt);
1818 debug_only(VMNativeEntryWrapper vew;)
1819
1820 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1821 VMThread::execute(&op);
1822 if (LoadExecStackDllInVMThread) {
1823 result = op.loaded_library();
1824 }
1825 load_attempted = true;
1826 }
1827 }
1828 }
1829 }
1830
1831 if (!load_attempted) {
1832 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1833 }
1834
1835 if (result != NULL) {
1836 // Successful loading
1837 return result;
1838 }
1839
1840 Elf32_Ehdr elf_head;
1841 int diag_msg_max_length=ebuflen-strlen(ebuf);
1842 char* diag_msg_buf=ebuf+strlen(ebuf);
1843
1844 if (diag_msg_max_length==0) {
1845 // No more space in ebuf for additional diagnostics message
1846 return NULL;
1847 }
1848
1849
1850 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1851
1852 if (file_descriptor < 0) {
1853 // Can't open library, report dlerror() message
1854 return NULL;
1855 }
1856
1857 bool failed_to_read_elf_head=
1858 (sizeof(elf_head)!=
1859 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1860
1861 ::close(file_descriptor);
1862 if (failed_to_read_elf_head) {
1863 // file i/o error - report dlerror() msg
1864 return NULL;
1865 }
1866
1867 typedef struct {
1868 Elf32_Half code; // Actual value as defined in elf.h
1869 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1870 char elf_class; // 32 or 64 bit
1871 char endianess; // MSB or LSB
1872 char* name; // String representation
1873 } arch_t;
1874
1875 #ifndef EM_486
1876 #define EM_486 6 /* Intel 80486 */
1877 #endif
1878 #ifndef EM_AARCH64
1879 #define EM_AARCH64 183 /* ARM AARCH64 */
1880 #endif
1881
1882 static const arch_t arch_array[]={
1883 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1884 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1885 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1886 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1887 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1888 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1889 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1890 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1891 #if defined(VM_LITTLE_ENDIAN)
1892 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1893 #else
1894 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1895 #endif
1896 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1897 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1898 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1899 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1900 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1901 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1902 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1903 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1904 };
1905
1906 #if (defined IA32)
1907 static Elf32_Half running_arch_code=EM_386;
1908 #elif (defined AMD64)
1909 static Elf32_Half running_arch_code=EM_X86_64;
1910 #elif (defined IA64)
1911 static Elf32_Half running_arch_code=EM_IA_64;
1912 #elif (defined __sparc) && (defined _LP64)
1913 static Elf32_Half running_arch_code=EM_SPARCV9;
1914 #elif (defined __sparc) && (!defined _LP64)
1915 static Elf32_Half running_arch_code=EM_SPARC;
1916 #elif (defined __powerpc64__)
1917 static Elf32_Half running_arch_code=EM_PPC64;
1918 #elif (defined __powerpc__)
1919 static Elf32_Half running_arch_code=EM_PPC;
1920 #elif (defined ARM)
1921 static Elf32_Half running_arch_code=EM_ARM;
1922 #elif (defined S390)
1923 static Elf32_Half running_arch_code=EM_S390;
1924 #elif (defined ALPHA)
1925 static Elf32_Half running_arch_code=EM_ALPHA;
1926 #elif (defined MIPSEL)
1927 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1928 #elif (defined PARISC)
1929 static Elf32_Half running_arch_code=EM_PARISC;
1930 #elif (defined MIPS)
1931 static Elf32_Half running_arch_code=EM_MIPS;
1932 #elif (defined M68K)
1933 static Elf32_Half running_arch_code=EM_68K;
1934 #elif (defined AARCH64)
1935 static Elf32_Half running_arch_code=EM_AARCH64;
1936 #else
1937 #error Method os::dll_load requires that one of following is defined:\
1938 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1939 #endif
1940
1941 // Identify compatability class for VM's architecture and library's architecture
1942 // Obtain string descriptions for architectures
1943
1944 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1945 int running_arch_index=-1;
1946
1947 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1948 if (running_arch_code == arch_array[i].code) {
1949 running_arch_index = i;
1950 }
1951 if (lib_arch.code == arch_array[i].code) {
1952 lib_arch.compat_class = arch_array[i].compat_class;
1953 lib_arch.name = arch_array[i].name;
1954 }
1955 }
1956
1957 assert(running_arch_index != -1,
1958 "Didn't find running architecture code (running_arch_code) in arch_array");
1959 if (running_arch_index == -1) {
1960 // Even though running architecture detection failed
1961 // we may still continue with reporting dlerror() message
1962 return NULL;
1963 }
1964
1965 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1966 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1967 return NULL;
1968 }
1969
1970 #ifndef S390
1971 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1972 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1973 return NULL;
1974 }
1975 #endif // !S390
1976
1977 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1978 if (lib_arch.name!=NULL) {
1979 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1980 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1981 lib_arch.name, arch_array[running_arch_index].name);
1982 } else {
1983 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1984 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1985 lib_arch.code,
1986 arch_array[running_arch_index].name);
1987 }
1988 }
1989
1990 return NULL;
1991 }
1992
1993 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1994 int ebuflen) {
1995 void * result = ::dlopen(filename, RTLD_LAZY);
1996 if (result == NULL) {
1997 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1998 ebuf[ebuflen-1] = '\0';
1999 }
2000 return result;
2001 }
2002
2003 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2004 int ebuflen) {
2005 void * result = NULL;
2006 if (LoadExecStackDllInVMThread) {
2007 result = dlopen_helper(filename, ebuf, ebuflen);
2008 }
2009
2010 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2011 // library that requires an executable stack, or which does not have this
2012 // stack attribute set, dlopen changes the stack attribute to executable. The
2013 // read protection of the guard pages gets lost.
2014 //
2015 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2016 // may have been queued at the same time.
2017
2018 if (!_stack_is_executable) {
2019 JavaThread *jt = Threads::first();
2020
2021 while (jt) {
2022 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2023 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2024 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2025 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2026 warning("Attempt to reguard stack yellow zone failed.");
2027 }
2028 }
2029 jt = jt->next();
2030 }
2031 }
2032
2033 return result;
2034 }
2035
2036 void* os::dll_lookup(void* handle, const char* name) {
2037 void* res = dlsym(handle, name);
2038 return res;
2039 }
2040
2041 void* os::get_default_process_handle() {
2042 return (void*)::dlopen(NULL, RTLD_LAZY);
2043 }
2044
2045 static bool _print_ascii_file(const char* filename, outputStream* st) {
2046 int fd = ::open(filename, O_RDONLY);
2047 if (fd == -1) {
2048 return false;
2049 }
2050
2051 char buf[32];
2052 int bytes;
2053 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2054 st->print_raw(buf, bytes);
2055 }
2056
2057 ::close(fd);
2058
2059 return true;
2060 }
2061
2062 void os::print_dll_info(outputStream *st) {
2063 st->print_cr("Dynamic libraries:");
2064
2065 char fname[32];
2066 pid_t pid = os::Linux::gettid();
2067
2068 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2069
2070 if (!_print_ascii_file(fname, st)) {
2071 st->print("Can not get library information for pid = %d\n", pid);
2072 }
2073 }
2074
2075 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2076 FILE *procmapsFile = NULL;
2077
2078 // Open the procfs maps file for the current process
2079 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2080 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2081 char line[PATH_MAX + 100];
2082
2083 // Read line by line from 'file'
2084 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2085 u8 base, top, offset, inode;
2086 char permissions[5];
2087 char device[6];
2088 char name[PATH_MAX + 1];
2089
2090 // Parse fields from line
2091 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2092
2093 // Filter by device id '00:00' so that we only get file system mapped files.
2094 if (strcmp(device, "00:00") != 0) {
2095
2096 // Call callback with the fields of interest
2097 if(callback(name, (address)base, (address)top, param)) {
2098 // Oops abort, callback aborted
2099 fclose(procmapsFile);
2100 return 1;
2101 }
2102 }
2103 }
2104 fclose(procmapsFile);
2105 }
2106 return 0;
2107 }
2108
2109 void os::print_os_info_brief(outputStream* st) {
2110 os::Linux::print_distro_info(st);
2111
2112 os::Posix::print_uname_info(st);
2113
2114 os::Linux::print_libversion_info(st);
2115
2116 }
2117
2118 void os::print_os_info(outputStream* st) {
2119 st->print("OS:");
2120
2121 os::Linux::print_distro_info(st);
2122
2123 os::Posix::print_uname_info(st);
2124
2125 // Print warning if unsafe chroot environment detected
2126 if (unsafe_chroot_detected) {
2127 st->print("WARNING!! ");
2128 st->print_cr("%s", unstable_chroot_error);
2129 }
2130
2131 os::Linux::print_libversion_info(st);
2132
2133 os::Posix::print_rlimit_info(st);
2134
2135 os::Posix::print_load_average(st);
2136
2137 os::Linux::print_full_memory_info(st);
2138 }
2139
2140 // Try to identify popular distros.
2141 // Most Linux distributions have a /etc/XXX-release file, which contains
2142 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2143 // file that also contains the OS version string. Some have more than one
2144 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2145 // /etc/redhat-release.), so the order is important.
2146 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2147 // their own specific XXX-release file as well as a redhat-release file.
2148 // Because of this the XXX-release file needs to be searched for before the
2149 // redhat-release file.
2150 // Since Red Hat has a lsb-release file that is not very descriptive the
2151 // search for redhat-release needs to be before lsb-release.
2152 // Since the lsb-release file is the new standard it needs to be searched
2153 // before the older style release files.
2154 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2155 // next to last resort. The os-release file is a new standard that contains
2156 // distribution information and the system-release file seems to be an old
2157 // standard that has been replaced by the lsb-release and os-release files.
2158 // Searching for the debian_version file is the last resort. It contains
2159 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2160 // "Debian " is printed before the contents of the debian_version file.
2161 void os::Linux::print_distro_info(outputStream* st) {
2162 if (!_print_ascii_file("/etc/oracle-release", st) &&
2163 !_print_ascii_file("/etc/mandriva-release", st) &&
2164 !_print_ascii_file("/etc/mandrake-release", st) &&
2165 !_print_ascii_file("/etc/sun-release", st) &&
2166 !_print_ascii_file("/etc/redhat-release", st) &&
2167 !_print_ascii_file("/etc/lsb-release", st) &&
2168 !_print_ascii_file("/etc/SuSE-release", st) &&
2169 !_print_ascii_file("/etc/turbolinux-release", st) &&
2170 !_print_ascii_file("/etc/gentoo-release", st) &&
2171 !_print_ascii_file("/etc/ltib-release", st) &&
2172 !_print_ascii_file("/etc/angstrom-version", st) &&
2173 !_print_ascii_file("/etc/system-release", st) &&
2174 !_print_ascii_file("/etc/os-release", st)) {
2175
2176 if (file_exists("/etc/debian_version")) {
2177 st->print("Debian ");
2178 _print_ascii_file("/etc/debian_version", st);
2179 } else {
2180 st->print("Linux");
2181 }
2182 }
2183 st->cr();
2184 }
2185
2186 void os::Linux::print_libversion_info(outputStream* st) {
2187 // libc, pthread
2188 st->print("libc:");
2189 st->print("%s ", os::Linux::glibc_version());
2190 st->print("%s ", os::Linux::libpthread_version());
2191 if (os::Linux::is_LinuxThreads()) {
2192 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2193 }
2194 st->cr();
2195 }
2196
2197 void os::Linux::print_full_memory_info(outputStream* st) {
2198 st->print("\n/proc/meminfo:\n");
2199 _print_ascii_file("/proc/meminfo", st);
2200 st->cr();
2201 }
2202
2203 void os::print_memory_info(outputStream* st) {
2204
2205 st->print("Memory:");
2206 st->print(" %dk page", os::vm_page_size()>>10);
2207
2208 // values in struct sysinfo are "unsigned long"
2209 struct sysinfo si;
2210 sysinfo(&si);
2211
2212 st->print(", physical " UINT64_FORMAT "k",
2213 os::physical_memory() >> 10);
2214 st->print("(" UINT64_FORMAT "k free)",
2215 os::available_memory() >> 10);
2216 st->print(", swap " UINT64_FORMAT "k",
2217 ((jlong)si.totalswap * si.mem_unit) >> 10);
2218 st->print("(" UINT64_FORMAT "k free)",
2219 ((jlong)si.freeswap * si.mem_unit) >> 10);
2220 st->cr();
2221 }
2222
2223 void os::pd_print_cpu_info(outputStream* st) {
2224 st->print("\n/proc/cpuinfo:\n");
2225 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2226 st->print(" <Not Available>");
2227 }
2228 st->cr();
2229 }
2230
2231 void os::print_siginfo(outputStream* st, void* siginfo) {
2232 const siginfo_t* si = (const siginfo_t*)siginfo;
2233
2234 os::Posix::print_siginfo_brief(st, si);
2235 #if INCLUDE_CDS
2236 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2237 UseSharedSpaces) {
2238 FileMapInfo* mapinfo = FileMapInfo::current_info();
2239 if (mapinfo->is_in_shared_space(si->si_addr)) {
2240 st->print("\n\nError accessing class data sharing archive." \
2241 " Mapped file inaccessible during execution, " \
2242 " possible disk/network problem.");
2243 }
2244 }
2245 #endif
2246 st->cr();
2247 }
2248
2249
2250 static void print_signal_handler(outputStream* st, int sig,
2251 char* buf, size_t buflen);
2252
2253 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2254 st->print_cr("Signal Handlers:");
2255 print_signal_handler(st, SIGSEGV, buf, buflen);
2256 print_signal_handler(st, SIGBUS , buf, buflen);
2257 print_signal_handler(st, SIGFPE , buf, buflen);
2258 print_signal_handler(st, SIGPIPE, buf, buflen);
2259 print_signal_handler(st, SIGXFSZ, buf, buflen);
2260 print_signal_handler(st, SIGILL , buf, buflen);
2261 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2262 print_signal_handler(st, SR_signum, buf, buflen);
2263 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2264 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2265 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2266 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2267 #if defined(PPC64)
2268 print_signal_handler(st, SIGTRAP, buf, buflen);
2269 #endif
2270 }
2271
2272 static char saved_jvm_path[MAXPATHLEN] = {0};
2273
2274 // Find the full path to the current module, libjvm.so
2275 void os::jvm_path(char *buf, jint buflen) {
2276 // Error checking.
2277 if (buflen < MAXPATHLEN) {
2278 assert(false, "must use a large-enough buffer");
2279 buf[0] = '\0';
2280 return;
2281 }
2282 // Lazy resolve the path to current module.
2283 if (saved_jvm_path[0] != 0) {
2284 strcpy(buf, saved_jvm_path);
2285 return;
2286 }
2287
2288 char dli_fname[MAXPATHLEN];
2289 bool ret = dll_address_to_library_name(
2290 CAST_FROM_FN_PTR(address, os::jvm_path),
2291 dli_fname, sizeof(dli_fname), NULL);
2292 assert(ret, "cannot locate libjvm");
2293 char *rp = NULL;
2294 if (ret && dli_fname[0] != '\0') {
2295 rp = realpath(dli_fname, buf);
2296 }
2297 if (rp == NULL) {
2298 return;
2299 }
2300
2301 if (Arguments::sun_java_launcher_is_altjvm()) {
2302 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2303 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2304 // If "/jre/lib/" appears at the right place in the string, then
2305 // assume we are installed in a JDK and we're done. Otherwise, check
2306 // for a JAVA_HOME environment variable and fix up the path so it
2307 // looks like libjvm.so is installed there (append a fake suffix
2308 // hotspot/libjvm.so).
2309 const char *p = buf + strlen(buf) - 1;
2310 for (int count = 0; p > buf && count < 5; ++count) {
2311 for (--p; p > buf && *p != '/'; --p)
2312 /* empty */ ;
2313 }
2314
2315 if (strncmp(p, "/jre/lib/", 9) != 0) {
2316 // Look for JAVA_HOME in the environment.
2317 char* java_home_var = ::getenv("JAVA_HOME");
2318 if (java_home_var != NULL && java_home_var[0] != 0) {
2319 char* jrelib_p;
2320 int len;
2321
2322 // Check the current module name "libjvm.so".
2323 p = strrchr(buf, '/');
2324 if (p == NULL) {
2325 return;
2326 }
2327 assert(strstr(p, "/libjvm") == p, "invalid library name");
2328
2329 rp = realpath(java_home_var, buf);
2330 if (rp == NULL) {
2331 return;
2332 }
2333
2334 // determine if this is a legacy image or modules image
2335 // modules image doesn't have "jre" subdirectory
2336 len = strlen(buf);
2337 assert(len < buflen, "Ran out of buffer room");
2338 jrelib_p = buf + len;
2339 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2340 if (0 != access(buf, F_OK)) {
2341 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2342 }
2343
2344 if (0 == access(buf, F_OK)) {
2345 // Use current module name "libjvm.so"
2346 len = strlen(buf);
2347 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2348 } else {
2349 // Go back to path of .so
2350 rp = realpath(dli_fname, buf);
2351 if (rp == NULL) {
2352 return;
2353 }
2354 }
2355 }
2356 }
2357 }
2358
2359 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2360 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2361 }
2362
2363 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2364 // no prefix required, not even "_"
2365 }
2366
2367 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2368 // no suffix required
2369 }
2370
2371 ////////////////////////////////////////////////////////////////////////////////
2372 // sun.misc.Signal support
2373
2374 static volatile jint sigint_count = 0;
2375
2376 static void UserHandler(int sig, void *siginfo, void *context) {
2377 // 4511530 - sem_post is serialized and handled by the manager thread. When
2378 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2379 // don't want to flood the manager thread with sem_post requests.
2380 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2381 return;
2382 }
2383
2384 // Ctrl-C is pressed during error reporting, likely because the error
2385 // handler fails to abort. Let VM die immediately.
2386 if (sig == SIGINT && is_error_reported()) {
2387 os::die();
2388 }
2389
2390 os::signal_notify(sig);
2391 }
2392
2393 void* os::user_handler() {
2394 return CAST_FROM_FN_PTR(void*, UserHandler);
2395 }
2396
2397 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2398 struct timespec ts;
2399 // Semaphore's are always associated with CLOCK_REALTIME
2400 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2401 // see unpackTime for discussion on overflow checking
2402 if (sec >= MAX_SECS) {
2403 ts.tv_sec += MAX_SECS;
2404 ts.tv_nsec = 0;
2405 } else {
2406 ts.tv_sec += sec;
2407 ts.tv_nsec += nsec;
2408 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2409 ts.tv_nsec -= NANOSECS_PER_SEC;
2410 ++ts.tv_sec; // note: this must be <= max_secs
2411 }
2412 }
2413
2414 return ts;
2415 }
2416
2417 extern "C" {
2418 typedef void (*sa_handler_t)(int);
2419 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2420 }
2421
2422 void* os::signal(int signal_number, void* handler) {
2423 struct sigaction sigAct, oldSigAct;
2424
2425 sigfillset(&(sigAct.sa_mask));
2426 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2427 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2428
2429 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2430 // -1 means registration failed
2431 return (void *)-1;
2432 }
2433
2434 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2435 }
2436
2437 void os::signal_raise(int signal_number) {
2438 ::raise(signal_number);
2439 }
2440
2441 // The following code is moved from os.cpp for making this
2442 // code platform specific, which it is by its very nature.
2443
2444 // Will be modified when max signal is changed to be dynamic
2445 int os::sigexitnum_pd() {
2446 return NSIG;
2447 }
2448
2449 // a counter for each possible signal value
2450 static volatile jint pending_signals[NSIG+1] = { 0 };
2451
2452 // Linux(POSIX) specific hand shaking semaphore.
2453 static sem_t sig_sem;
2454 static PosixSemaphore sr_semaphore;
2455
2456 void os::signal_init_pd() {
2457 // Initialize signal structures
2458 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2459
2460 // Initialize signal semaphore
2461 ::sem_init(&sig_sem, 0, 0);
2462 }
2463
2464 void os::signal_notify(int sig) {
2465 Atomic::inc(&pending_signals[sig]);
2466 ::sem_post(&sig_sem);
2467 }
2468
2469 static int check_pending_signals(bool wait) {
2470 Atomic::store(0, &sigint_count);
2471 for (;;) {
2472 for (int i = 0; i < NSIG + 1; i++) {
2473 jint n = pending_signals[i];
2474 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2475 return i;
2476 }
2477 }
2478 if (!wait) {
2479 return -1;
2480 }
2481 JavaThread *thread = JavaThread::current();
2482 ThreadBlockInVM tbivm(thread);
2483
2484 bool threadIsSuspended;
2485 do {
2486 thread->set_suspend_equivalent();
2487 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2488 ::sem_wait(&sig_sem);
2489
2490 // were we externally suspended while we were waiting?
2491 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2492 if (threadIsSuspended) {
2493 // The semaphore has been incremented, but while we were waiting
2494 // another thread suspended us. We don't want to continue running
2495 // while suspended because that would surprise the thread that
2496 // suspended us.
2497 ::sem_post(&sig_sem);
2498
2499 thread->java_suspend_self();
2500 }
2501 } while (threadIsSuspended);
2502 }
2503 }
2504
2505 int os::signal_lookup() {
2506 return check_pending_signals(false);
2507 }
2508
2509 int os::signal_wait() {
2510 return check_pending_signals(true);
2511 }
2512
2513 ////////////////////////////////////////////////////////////////////////////////
2514 // Virtual Memory
2515
2516 int os::vm_page_size() {
2517 // Seems redundant as all get out
2518 assert(os::Linux::page_size() != -1, "must call os::init");
2519 return os::Linux::page_size();
2520 }
2521
2522 // Solaris allocates memory by pages.
2523 int os::vm_allocation_granularity() {
2524 assert(os::Linux::page_size() != -1, "must call os::init");
2525 return os::Linux::page_size();
2526 }
2527
2528 // Rationale behind this function:
2529 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2530 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2531 // samples for JITted code. Here we create private executable mapping over the code cache
2532 // and then we can use standard (well, almost, as mapping can change) way to provide
2533 // info for the reporting script by storing timestamp and location of symbol
2534 void linux_wrap_code(char* base, size_t size) {
2535 static volatile jint cnt = 0;
2536
2537 if (!UseOprofile) {
2538 return;
2539 }
2540
2541 char buf[PATH_MAX+1];
2542 int num = Atomic::add(1, &cnt);
2543
2544 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2545 os::get_temp_directory(), os::current_process_id(), num);
2546 unlink(buf);
2547
2548 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2549
2550 if (fd != -1) {
2551 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2552 if (rv != (off_t)-1) {
2553 if (::write(fd, "", 1) == 1) {
2554 mmap(base, size,
2555 PROT_READ|PROT_WRITE|PROT_EXEC,
2556 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2557 }
2558 }
2559 ::close(fd);
2560 unlink(buf);
2561 }
2562 }
2563
2564 static bool recoverable_mmap_error(int err) {
2565 // See if the error is one we can let the caller handle. This
2566 // list of errno values comes from JBS-6843484. I can't find a
2567 // Linux man page that documents this specific set of errno
2568 // values so while this list currently matches Solaris, it may
2569 // change as we gain experience with this failure mode.
2570 switch (err) {
2571 case EBADF:
2572 case EINVAL:
2573 case ENOTSUP:
2574 // let the caller deal with these errors
2575 return true;
2576
2577 default:
2578 // Any remaining errors on this OS can cause our reserved mapping
2579 // to be lost. That can cause confusion where different data
2580 // structures think they have the same memory mapped. The worst
2581 // scenario is if both the VM and a library think they have the
2582 // same memory mapped.
2583 return false;
2584 }
2585 }
2586
2587 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2588 int err) {
2589 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2590 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2591 strerror(err), err);
2592 }
2593
2594 static void warn_fail_commit_memory(char* addr, size_t size,
2595 size_t alignment_hint, bool exec,
2596 int err) {
2597 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2598 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2599 alignment_hint, exec, strerror(err), err);
2600 }
2601
2602 // NOTE: Linux kernel does not really reserve the pages for us.
2603 // All it does is to check if there are enough free pages
2604 // left at the time of mmap(). This could be a potential
2605 // problem.
2606 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2607 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2608 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2609 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2610 if (res != (uintptr_t) MAP_FAILED) {
2611 if (UseNUMAInterleaving) {
2612 numa_make_global(addr, size);
2613 }
2614 return 0;
2615 }
2616
2617 int err = errno; // save errno from mmap() call above
2618
2619 if (!recoverable_mmap_error(err)) {
2620 warn_fail_commit_memory(addr, size, exec, err);
2621 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2622 }
2623
2624 return err;
2625 }
2626
2627 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2628 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2629 }
2630
2631 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2632 const char* mesg) {
2633 assert(mesg != NULL, "mesg must be specified");
2634 int err = os::Linux::commit_memory_impl(addr, size, exec);
2635 if (err != 0) {
2636 // the caller wants all commit errors to exit with the specified mesg:
2637 warn_fail_commit_memory(addr, size, exec, err);
2638 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2639 }
2640 }
2641
2642 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2643 #ifndef MAP_HUGETLB
2644 #define MAP_HUGETLB 0x40000
2645 #endif
2646
2647 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2648 #ifndef MADV_HUGEPAGE
2649 #define MADV_HUGEPAGE 14
2650 #endif
2651
2652 int os::Linux::commit_memory_impl(char* addr, size_t size,
2653 size_t alignment_hint, bool exec) {
2654 int err = os::Linux::commit_memory_impl(addr, size, exec);
2655 if (err == 0) {
2656 realign_memory(addr, size, alignment_hint);
2657 }
2658 return err;
2659 }
2660
2661 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2662 bool exec) {
2663 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2664 }
2665
2666 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2667 size_t alignment_hint, bool exec,
2668 const char* mesg) {
2669 assert(mesg != NULL, "mesg must be specified");
2670 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2671 if (err != 0) {
2672 // the caller wants all commit errors to exit with the specified mesg:
2673 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2674 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2675 }
2676 }
2677
2678 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2679 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2680 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2681 // be supported or the memory may already be backed by huge pages.
2682 ::madvise(addr, bytes, MADV_HUGEPAGE);
2683 }
2684 }
2685
2686 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2687 // This method works by doing an mmap over an existing mmaping and effectively discarding
2688 // the existing pages. However it won't work for SHM-based large pages that cannot be
2689 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2690 // small pages on top of the SHM segment. This method always works for small pages, so we
2691 // allow that in any case.
2692 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2693 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2694 }
2695 }
2696
2697 void os::numa_make_global(char *addr, size_t bytes) {
2698 Linux::numa_interleave_memory(addr, bytes);
2699 }
2700
2701 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2702 // bind policy to MPOL_PREFERRED for the current thread.
2703 #define USE_MPOL_PREFERRED 0
2704
2705 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2706 // To make NUMA and large pages more robust when both enabled, we need to ease
2707 // the requirements on where the memory should be allocated. MPOL_BIND is the
2708 // default policy and it will force memory to be allocated on the specified
2709 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2710 // the specified node, but will not force it. Using this policy will prevent
2711 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2712 // free large pages.
2713 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2714 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2715 }
2716
2717 bool os::numa_topology_changed() { return false; }
2718
2719 size_t os::numa_get_groups_num() {
2720 int max_node = Linux::numa_max_node();
2721 return max_node > 0 ? max_node + 1 : 1;
2722 }
2723
2724 int os::numa_get_group_id() {
2725 int cpu_id = Linux::sched_getcpu();
2726 if (cpu_id != -1) {
2727 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2728 if (lgrp_id != -1) {
2729 return lgrp_id;
2730 }
2731 }
2732 return 0;
2733 }
2734
2735 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2736 for (size_t i = 0; i < size; i++) {
2737 ids[i] = i;
2738 }
2739 return size;
2740 }
2741
2742 bool os::get_page_info(char *start, page_info* info) {
2743 return false;
2744 }
2745
2746 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2747 page_info* page_found) {
2748 return end;
2749 }
2750
2751
2752 int os::Linux::sched_getcpu_syscall(void) {
2753 unsigned int cpu;
2754 int retval = -1;
2755
2756 #if defined(IA32)
2757 #ifndef SYS_getcpu
2758 #define SYS_getcpu 318
2759 #endif
2760 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2761 #elif defined(AMD64)
2762 // Unfortunately we have to bring all these macros here from vsyscall.h
2763 // to be able to compile on old linuxes.
2764 #define __NR_vgetcpu 2
2765 #define VSYSCALL_START (-10UL << 20)
2766 #define VSYSCALL_SIZE 1024
2767 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2768 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2769 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2770 retval = vgetcpu(&cpu, NULL, NULL);
2771 #endif
2772
2773 return (retval == -1) ? retval : cpu;
2774 }
2775
2776 // Something to do with the numa-aware allocator needs these symbols
2777 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2778 extern "C" JNIEXPORT void numa_error(char *where) { }
2779 extern "C" JNIEXPORT int fork1() { return fork(); }
2780
2781
2782 // If we are running with libnuma version > 2, then we should
2783 // be trying to use symbols with versions 1.1
2784 // If we are running with earlier version, which did not have symbol versions,
2785 // we should use the base version.
2786 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2787 void *f = dlvsym(handle, name, "libnuma_1.1");
2788 if (f == NULL) {
2789 f = dlsym(handle, name);
2790 }
2791 return f;
2792 }
2793
2794 bool os::Linux::libnuma_init() {
2795 // sched_getcpu() should be in libc.
2796 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2797 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2798
2799 // If it's not, try a direct syscall.
2800 if (sched_getcpu() == -1) {
2801 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2802 (void*)&sched_getcpu_syscall));
2803 }
2804
2805 if (sched_getcpu() != -1) { // Does it work?
2806 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2807 if (handle != NULL) {
2808 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2809 libnuma_dlsym(handle, "numa_node_to_cpus")));
2810 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2811 libnuma_dlsym(handle, "numa_max_node")));
2812 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2813 libnuma_dlsym(handle, "numa_available")));
2814 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2815 libnuma_dlsym(handle, "numa_tonode_memory")));
2816 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2817 libnuma_dlsym(handle, "numa_interleave_memory")));
2818 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2819 libnuma_dlsym(handle, "numa_set_bind_policy")));
2820
2821
2822 if (numa_available() != -1) {
2823 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2824 // Create a cpu -> node mapping
2825 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2826 rebuild_cpu_to_node_map();
2827 return true;
2828 }
2829 }
2830 }
2831 return false;
2832 }
2833
2834 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2835 // The table is later used in get_node_by_cpu().
2836 void os::Linux::rebuild_cpu_to_node_map() {
2837 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2838 // in libnuma (possible values are starting from 16,
2839 // and continuing up with every other power of 2, but less
2840 // than the maximum number of CPUs supported by kernel), and
2841 // is a subject to change (in libnuma version 2 the requirements
2842 // are more reasonable) we'll just hardcode the number they use
2843 // in the library.
2844 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2845
2846 size_t cpu_num = os::active_processor_count();
2847 size_t cpu_map_size = NCPUS / BitsPerCLong;
2848 size_t cpu_map_valid_size =
2849 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2850
2851 cpu_to_node()->clear();
2852 cpu_to_node()->at_grow(cpu_num - 1);
2853 size_t node_num = numa_get_groups_num();
2854
2855 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2856 for (size_t i = 0; i < node_num; i++) {
2857 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2858 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2859 if (cpu_map[j] != 0) {
2860 for (size_t k = 0; k < BitsPerCLong; k++) {
2861 if (cpu_map[j] & (1UL << k)) {
2862 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2863 }
2864 }
2865 }
2866 }
2867 }
2868 }
2869 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2870 }
2871
2872 int os::Linux::get_node_by_cpu(int cpu_id) {
2873 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2874 return cpu_to_node()->at(cpu_id);
2875 }
2876 return -1;
2877 }
2878
2879 GrowableArray<int>* os::Linux::_cpu_to_node;
2880 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2881 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2882 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2883 os::Linux::numa_available_func_t os::Linux::_numa_available;
2884 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2885 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2886 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2887 unsigned long* os::Linux::_numa_all_nodes;
2888
2889 bool os::pd_uncommit_memory(char* addr, size_t size) {
2890 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2891 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2892 return res != (uintptr_t) MAP_FAILED;
2893 }
2894
2895 static address get_stack_commited_bottom(address bottom, size_t size) {
2896 address nbot = bottom;
2897 address ntop = bottom + size;
2898
2899 size_t page_sz = os::vm_page_size();
2900 unsigned pages = size / page_sz;
2901
2902 unsigned char vec[1];
2903 unsigned imin = 1, imax = pages + 1, imid;
2904 int mincore_return_value = 0;
2905
2906 assert(imin <= imax, "Unexpected page size");
2907
2908 while (imin < imax) {
2909 imid = (imax + imin) / 2;
2910 nbot = ntop - (imid * page_sz);
2911
2912 // Use a trick with mincore to check whether the page is mapped or not.
2913 // mincore sets vec to 1 if page resides in memory and to 0 if page
2914 // is swapped output but if page we are asking for is unmapped
2915 // it returns -1,ENOMEM
2916 mincore_return_value = mincore(nbot, page_sz, vec);
2917
2918 if (mincore_return_value == -1) {
2919 // Page is not mapped go up
2920 // to find first mapped page
2921 if (errno != EAGAIN) {
2922 assert(errno == ENOMEM, "Unexpected mincore errno");
2923 imax = imid;
2924 }
2925 } else {
2926 // Page is mapped go down
2927 // to find first not mapped page
2928 imin = imid + 1;
2929 }
2930 }
2931
2932 nbot = nbot + page_sz;
2933
2934 // Adjust stack bottom one page up if last checked page is not mapped
2935 if (mincore_return_value == -1) {
2936 nbot = nbot + page_sz;
2937 }
2938
2939 return nbot;
2940 }
2941
2942
2943 // Linux uses a growable mapping for the stack, and if the mapping for
2944 // the stack guard pages is not removed when we detach a thread the
2945 // stack cannot grow beyond the pages where the stack guard was
2946 // mapped. If at some point later in the process the stack expands to
2947 // that point, the Linux kernel cannot expand the stack any further
2948 // because the guard pages are in the way, and a segfault occurs.
2949 //
2950 // However, it's essential not to split the stack region by unmapping
2951 // a region (leaving a hole) that's already part of the stack mapping,
2952 // so if the stack mapping has already grown beyond the guard pages at
2953 // the time we create them, we have to truncate the stack mapping.
2954 // So, we need to know the extent of the stack mapping when
2955 // create_stack_guard_pages() is called.
2956
2957 // We only need this for stacks that are growable: at the time of
2958 // writing thread stacks don't use growable mappings (i.e. those
2959 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2960 // only applies to the main thread.
2961
2962 // If the (growable) stack mapping already extends beyond the point
2963 // where we're going to put our guard pages, truncate the mapping at
2964 // that point by munmap()ping it. This ensures that when we later
2965 // munmap() the guard pages we don't leave a hole in the stack
2966 // mapping. This only affects the main/initial thread
2967
2968 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2969 if (os::Linux::is_initial_thread()) {
2970 // As we manually grow stack up to bottom inside create_attached_thread(),
2971 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2972 // we don't need to do anything special.
2973 // Check it first, before calling heavy function.
2974 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2975 unsigned char vec[1];
2976
2977 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2978 // Fallback to slow path on all errors, including EAGAIN
2979 stack_extent = (uintptr_t) get_stack_commited_bottom(
2980 os::Linux::initial_thread_stack_bottom(),
2981 (size_t)addr - stack_extent);
2982 }
2983
2984 if (stack_extent < (uintptr_t)addr) {
2985 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2986 }
2987 }
2988
2989 return os::commit_memory(addr, size, !ExecMem);
2990 }
2991
2992 // If this is a growable mapping, remove the guard pages entirely by
2993 // munmap()ping them. If not, just call uncommit_memory(). This only
2994 // affects the main/initial thread, but guard against future OS changes
2995 // It's safe to always unmap guard pages for initial thread because we
2996 // always place it right after end of the mapped region
2997
2998 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2999 uintptr_t stack_extent, stack_base;
3000
3001 if (os::Linux::is_initial_thread()) {
3002 return ::munmap(addr, size) == 0;
3003 }
3004
3005 return os::uncommit_memory(addr, size);
3006 }
3007
3008 static address _highest_vm_reserved_address = NULL;
3009
3010 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3011 // at 'requested_addr'. If there are existing memory mappings at the same
3012 // location, however, they will be overwritten. If 'fixed' is false,
3013 // 'requested_addr' is only treated as a hint, the return value may or
3014 // may not start from the requested address. Unlike Linux mmap(), this
3015 // function returns NULL to indicate failure.
3016 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3017 char * addr;
3018 int flags;
3019
3020 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3021 if (fixed) {
3022 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3023 flags |= MAP_FIXED;
3024 }
3025
3026 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3027 // touch an uncommitted page. Otherwise, the read/write might
3028 // succeed if we have enough swap space to back the physical page.
3029 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3030 flags, -1, 0);
3031
3032 if (addr != MAP_FAILED) {
3033 // anon_mmap() should only get called during VM initialization,
3034 // don't need lock (actually we can skip locking even it can be called
3035 // from multiple threads, because _highest_vm_reserved_address is just a
3036 // hint about the upper limit of non-stack memory regions.)
3037 if ((address)addr + bytes > _highest_vm_reserved_address) {
3038 _highest_vm_reserved_address = (address)addr + bytes;
3039 }
3040 }
3041
3042 return addr == MAP_FAILED ? NULL : addr;
3043 }
3044
3045 // Don't update _highest_vm_reserved_address, because there might be memory
3046 // regions above addr + size. If so, releasing a memory region only creates
3047 // a hole in the address space, it doesn't help prevent heap-stack collision.
3048 //
3049 static int anon_munmap(char * addr, size_t size) {
3050 return ::munmap(addr, size) == 0;
3051 }
3052
3053 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3054 size_t alignment_hint) {
3055 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3056 }
3057
3058 bool os::pd_release_memory(char* addr, size_t size) {
3059 return anon_munmap(addr, size);
3060 }
3061
3062 static address highest_vm_reserved_address() {
3063 return _highest_vm_reserved_address;
3064 }
3065
3066 static bool linux_mprotect(char* addr, size_t size, int prot) {
3067 // Linux wants the mprotect address argument to be page aligned.
3068 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3069
3070 // According to SUSv3, mprotect() should only be used with mappings
3071 // established by mmap(), and mmap() always maps whole pages. Unaligned
3072 // 'addr' likely indicates problem in the VM (e.g. trying to change
3073 // protection of malloc'ed or statically allocated memory). Check the
3074 // caller if you hit this assert.
3075 assert(addr == bottom, "sanity check");
3076
3077 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3078 return ::mprotect(bottom, size, prot) == 0;
3079 }
3080
3081 // Set protections specified
3082 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3083 bool is_committed) {
3084 unsigned int p = 0;
3085 switch (prot) {
3086 case MEM_PROT_NONE: p = PROT_NONE; break;
3087 case MEM_PROT_READ: p = PROT_READ; break;
3088 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3089 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3090 default:
3091 ShouldNotReachHere();
3092 }
3093 // is_committed is unused.
3094 return linux_mprotect(addr, bytes, p);
3095 }
3096
3097 bool os::guard_memory(char* addr, size_t size) {
3098 return linux_mprotect(addr, size, PROT_NONE);
3099 }
3100
3101 bool os::unguard_memory(char* addr, size_t size) {
3102 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3103 }
3104
3105 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3106 size_t page_size) {
3107 bool result = false;
3108 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3109 MAP_ANONYMOUS|MAP_PRIVATE,
3110 -1, 0);
3111 if (p != MAP_FAILED) {
3112 void *aligned_p = align_ptr_up(p, page_size);
3113
3114 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3115
3116 munmap(p, page_size * 2);
3117 }
3118
3119 if (warn && !result) {
3120 warning("TransparentHugePages is not supported by the operating system.");
3121 }
3122
3123 return result;
3124 }
3125
3126 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3127 bool result = false;
3128 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3129 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3130 -1, 0);
3131
3132 if (p != MAP_FAILED) {
3133 // We don't know if this really is a huge page or not.
3134 FILE *fp = fopen("/proc/self/maps", "r");
3135 if (fp) {
3136 while (!feof(fp)) {
3137 char chars[257];
3138 long x = 0;
3139 if (fgets(chars, sizeof(chars), fp)) {
3140 if (sscanf(chars, "%lx-%*x", &x) == 1
3141 && x == (long)p) {
3142 if (strstr (chars, "hugepage")) {
3143 result = true;
3144 break;
3145 }
3146 }
3147 }
3148 }
3149 fclose(fp);
3150 }
3151 munmap(p, page_size);
3152 }
3153
3154 if (warn && !result) {
3155 warning("HugeTLBFS is not supported by the operating system.");
3156 }
3157
3158 return result;
3159 }
3160
3161 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3162 //
3163 // From the coredump_filter documentation:
3164 //
3165 // - (bit 0) anonymous private memory
3166 // - (bit 1) anonymous shared memory
3167 // - (bit 2) file-backed private memory
3168 // - (bit 3) file-backed shared memory
3169 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3170 // effective only if the bit 2 is cleared)
3171 // - (bit 5) hugetlb private memory
3172 // - (bit 6) hugetlb shared memory
3173 //
3174 static void set_coredump_filter(void) {
3175 FILE *f;
3176 long cdm;
3177
3178 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3179 return;
3180 }
3181
3182 if (fscanf(f, "%lx", &cdm) != 1) {
3183 fclose(f);
3184 return;
3185 }
3186
3187 rewind(f);
3188
3189 if ((cdm & LARGEPAGES_BIT) == 0) {
3190 cdm |= LARGEPAGES_BIT;
3191 fprintf(f, "%#lx", cdm);
3192 }
3193
3194 fclose(f);
3195 }
3196
3197 // Large page support
3198
3199 static size_t _large_page_size = 0;
3200
3201 size_t os::Linux::find_large_page_size() {
3202 size_t large_page_size = 0;
3203
3204 // large_page_size on Linux is used to round up heap size. x86 uses either
3205 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3206 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3207 // page as large as 256M.
3208 //
3209 // Here we try to figure out page size by parsing /proc/meminfo and looking
3210 // for a line with the following format:
3211 // Hugepagesize: 2048 kB
3212 //
3213 // If we can't determine the value (e.g. /proc is not mounted, or the text
3214 // format has been changed), we'll use the largest page size supported by
3215 // the processor.
3216
3217 #ifndef ZERO
3218 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3219 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3220 #endif // ZERO
3221
3222 FILE *fp = fopen("/proc/meminfo", "r");
3223 if (fp) {
3224 while (!feof(fp)) {
3225 int x = 0;
3226 char buf[16];
3227 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3228 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3229 large_page_size = x * K;
3230 break;
3231 }
3232 } else {
3233 // skip to next line
3234 for (;;) {
3235 int ch = fgetc(fp);
3236 if (ch == EOF || ch == (int)'\n') break;
3237 }
3238 }
3239 }
3240 fclose(fp);
3241 }
3242
3243 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3244 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3245 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3246 proper_unit_for_byte_size(large_page_size));
3247 }
3248
3249 return large_page_size;
3250 }
3251
3252 size_t os::Linux::setup_large_page_size() {
3253 _large_page_size = Linux::find_large_page_size();
3254 const size_t default_page_size = (size_t)Linux::page_size();
3255 if (_large_page_size > default_page_size) {
3256 _page_sizes[0] = _large_page_size;
3257 _page_sizes[1] = default_page_size;
3258 _page_sizes[2] = 0;
3259 }
3260
3261 return _large_page_size;
3262 }
3263
3264 bool os::Linux::setup_large_page_type(size_t page_size) {
3265 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3266 FLAG_IS_DEFAULT(UseSHM) &&
3267 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3268
3269 // The type of large pages has not been specified by the user.
3270
3271 // Try UseHugeTLBFS and then UseSHM.
3272 UseHugeTLBFS = UseSHM = true;
3273
3274 // Don't try UseTransparentHugePages since there are known
3275 // performance issues with it turned on. This might change in the future.
3276 UseTransparentHugePages = false;
3277 }
3278
3279 if (UseTransparentHugePages) {
3280 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3281 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3282 UseHugeTLBFS = false;
3283 UseSHM = false;
3284 return true;
3285 }
3286 UseTransparentHugePages = false;
3287 }
3288
3289 if (UseHugeTLBFS) {
3290 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3291 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3292 UseSHM = false;
3293 return true;
3294 }
3295 UseHugeTLBFS = false;
3296 }
3297
3298 return UseSHM;
3299 }
3300
3301 void os::large_page_init() {
3302 if (!UseLargePages &&
3303 !UseTransparentHugePages &&
3304 !UseHugeTLBFS &&
3305 !UseSHM) {
3306 // Not using large pages.
3307 return;
3308 }
3309
3310 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3311 // The user explicitly turned off large pages.
3312 // Ignore the rest of the large pages flags.
3313 UseTransparentHugePages = false;
3314 UseHugeTLBFS = false;
3315 UseSHM = false;
3316 return;
3317 }
3318
3319 size_t large_page_size = Linux::setup_large_page_size();
3320 UseLargePages = Linux::setup_large_page_type(large_page_size);
3321
3322 set_coredump_filter();
3323 }
3324
3325 #ifndef SHM_HUGETLB
3326 #define SHM_HUGETLB 04000
3327 #endif
3328
3329 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3330 char* req_addr, bool exec) {
3331 // "exec" is passed in but not used. Creating the shared image for
3332 // the code cache doesn't have an SHM_X executable permission to check.
3333 assert(UseLargePages && UseSHM, "only for SHM large pages");
3334 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3335
3336 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3337 return NULL; // Fallback to small pages.
3338 }
3339
3340 key_t key = IPC_PRIVATE;
3341 char *addr;
3342
3343 bool warn_on_failure = UseLargePages &&
3344 (!FLAG_IS_DEFAULT(UseLargePages) ||
3345 !FLAG_IS_DEFAULT(UseSHM) ||
3346 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3347 char msg[128];
3348
3349 // Create a large shared memory region to attach to based on size.
3350 // Currently, size is the total size of the heap
3351 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3352 if (shmid == -1) {
3353 // Possible reasons for shmget failure:
3354 // 1. shmmax is too small for Java heap.
3355 // > check shmmax value: cat /proc/sys/kernel/shmmax
3356 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3357 // 2. not enough large page memory.
3358 // > check available large pages: cat /proc/meminfo
3359 // > increase amount of large pages:
3360 // echo new_value > /proc/sys/vm/nr_hugepages
3361 // Note 1: different Linux may use different name for this property,
3362 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3363 // Note 2: it's possible there's enough physical memory available but
3364 // they are so fragmented after a long run that they can't
3365 // coalesce into large pages. Try to reserve large pages when
3366 // the system is still "fresh".
3367 if (warn_on_failure) {
3368 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3369 warning("%s", msg);
3370 }
3371 return NULL;
3372 }
3373
3374 // attach to the region
3375 addr = (char*)shmat(shmid, req_addr, 0);
3376 int err = errno;
3377
3378 // Remove shmid. If shmat() is successful, the actual shared memory segment
3379 // will be deleted when it's detached by shmdt() or when the process
3380 // terminates. If shmat() is not successful this will remove the shared
3381 // segment immediately.
3382 shmctl(shmid, IPC_RMID, NULL);
3383
3384 if ((intptr_t)addr == -1) {
3385 if (warn_on_failure) {
3386 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3387 warning("%s", msg);
3388 }
3389 return NULL;
3390 }
3391
3392 return addr;
3393 }
3394
3395 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3396 int error) {
3397 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3398
3399 bool warn_on_failure = UseLargePages &&
3400 (!FLAG_IS_DEFAULT(UseLargePages) ||
3401 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3402 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3403
3404 if (warn_on_failure) {
3405 char msg[128];
3406 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3407 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3408 warning("%s", msg);
3409 }
3410 }
3411
3412 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3413 char* req_addr,
3414 bool exec) {
3415 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3416 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3417 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3418
3419 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3420 char* addr = (char*)::mmap(req_addr, bytes, prot,
3421 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3422 -1, 0);
3423
3424 if (addr == MAP_FAILED) {
3425 warn_on_large_pages_failure(req_addr, bytes, errno);
3426 return NULL;
3427 }
3428
3429 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3430
3431 return addr;
3432 }
3433
3434 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3435 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3436 // (req_addr != NULL) or with a given alignment.
3437 // - bytes shall be a multiple of alignment.
3438 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3439 // - alignment sets the alignment at which memory shall be allocated.
3440 // It must be a multiple of allocation granularity.
3441 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3442 // req_addr or NULL.
3443 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3444
3445 size_t extra_size = bytes;
3446 if (req_addr == NULL && alignment > 0) {
3447 extra_size += alignment;
3448 }
3449
3450 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3451 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3452 -1, 0);
3453 if (start == MAP_FAILED) {
3454 start = NULL;
3455 } else {
3456 if (req_addr != NULL) {
3457 if (start != req_addr) {
3458 ::munmap(start, extra_size);
3459 start = NULL;
3460 }
3461 } else {
3462 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3463 char* const end_aligned = start_aligned + bytes;
3464 char* const end = start + extra_size;
3465 if (start_aligned > start) {
3466 ::munmap(start, start_aligned - start);
3467 }
3468 if (end_aligned < end) {
3469 ::munmap(end_aligned, end - end_aligned);
3470 }
3471 start = start_aligned;
3472 }
3473 }
3474 return start;
3475
3476 }
3477
3478 // Reserve memory using mmap(MAP_HUGETLB).
3479 // - bytes shall be a multiple of alignment.
3480 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3481 // - alignment sets the alignment at which memory shall be allocated.
3482 // It must be a multiple of allocation granularity.
3483 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3484 // req_addr or NULL.
3485 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3486 size_t alignment,
3487 char* req_addr,
3488 bool exec) {
3489 size_t large_page_size = os::large_page_size();
3490 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3491
3492 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3493 assert(is_size_aligned(bytes, alignment), "Must be");
3494
3495 // First reserve - but not commit - the address range in small pages.
3496 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3497
3498 if (start == NULL) {
3499 return NULL;
3500 }
3501
3502 assert(is_ptr_aligned(start, alignment), "Must be");
3503
3504 char* end = start + bytes;
3505
3506 // Find the regions of the allocated chunk that can be promoted to large pages.
3507 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3508 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3509
3510 size_t lp_bytes = lp_end - lp_start;
3511
3512 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3513
3514 if (lp_bytes == 0) {
3515 // The mapped region doesn't even span the start and the end of a large page.
3516 // Fall back to allocate a non-special area.
3517 ::munmap(start, end - start);
3518 return NULL;
3519 }
3520
3521 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3522
3523 void* result;
3524
3525 // Commit small-paged leading area.
3526 if (start != lp_start) {
3527 result = ::mmap(start, lp_start - start, prot,
3528 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3529 -1, 0);
3530 if (result == MAP_FAILED) {
3531 ::munmap(lp_start, end - lp_start);
3532 return NULL;
3533 }
3534 }
3535
3536 // Commit large-paged area.
3537 result = ::mmap(lp_start, lp_bytes, prot,
3538 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3539 -1, 0);
3540 if (result == MAP_FAILED) {
3541 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3542 // If the mmap above fails, the large pages region will be unmapped and we
3543 // have regions before and after with small pages. Release these regions.
3544 //
3545 // | mapped | unmapped | mapped |
3546 // ^ ^ ^ ^
3547 // start lp_start lp_end end
3548 //
3549 ::munmap(start, lp_start - start);
3550 ::munmap(lp_end, end - lp_end);
3551 return NULL;
3552 }
3553
3554 // Commit small-paged trailing area.
3555 if (lp_end != end) {
3556 result = ::mmap(lp_end, end - lp_end, prot,
3557 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3558 -1, 0);
3559 if (result == MAP_FAILED) {
3560 ::munmap(start, lp_end - start);
3561 return NULL;
3562 }
3563 }
3564
3565 return start;
3566 }
3567
3568 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3569 size_t alignment,
3570 char* req_addr,
3571 bool exec) {
3572 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3573 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3574 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3575 assert(is_power_of_2(os::large_page_size()), "Must be");
3576 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3577
3578 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3579 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3580 } else {
3581 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3582 }
3583 }
3584
3585 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3586 char* req_addr, bool exec) {
3587 assert(UseLargePages, "only for large pages");
3588
3589 char* addr;
3590 if (UseSHM) {
3591 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3592 } else {
3593 assert(UseHugeTLBFS, "must be");
3594 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3595 }
3596
3597 if (addr != NULL) {
3598 if (UseNUMAInterleaving) {
3599 numa_make_global(addr, bytes);
3600 }
3601
3602 // The memory is committed
3603 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3604 }
3605
3606 return addr;
3607 }
3608
3609 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3610 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3611 return shmdt(base) == 0;
3612 }
3613
3614 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3615 return pd_release_memory(base, bytes);
3616 }
3617
3618 bool os::release_memory_special(char* base, size_t bytes) {
3619 bool res;
3620 if (MemTracker::tracking_level() > NMT_minimal) {
3621 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3622 res = os::Linux::release_memory_special_impl(base, bytes);
3623 if (res) {
3624 tkr.record((address)base, bytes);
3625 }
3626
3627 } else {
3628 res = os::Linux::release_memory_special_impl(base, bytes);
3629 }
3630 return res;
3631 }
3632
3633 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3634 assert(UseLargePages, "only for large pages");
3635 bool res;
3636
3637 if (UseSHM) {
3638 res = os::Linux::release_memory_special_shm(base, bytes);
3639 } else {
3640 assert(UseHugeTLBFS, "must be");
3641 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3642 }
3643 return res;
3644 }
3645
3646 size_t os::large_page_size() {
3647 return _large_page_size;
3648 }
3649
3650 // With SysV SHM the entire memory region must be allocated as shared
3651 // memory.
3652 // HugeTLBFS allows application to commit large page memory on demand.
3653 // However, when committing memory with HugeTLBFS fails, the region
3654 // that was supposed to be committed will lose the old reservation
3655 // and allow other threads to steal that memory region. Because of this
3656 // behavior we can't commit HugeTLBFS memory.
3657 bool os::can_commit_large_page_memory() {
3658 return UseTransparentHugePages;
3659 }
3660
3661 bool os::can_execute_large_page_memory() {
3662 return UseTransparentHugePages || UseHugeTLBFS;
3663 }
3664
3665 // Reserve memory at an arbitrary address, only if that area is
3666 // available (and not reserved for something else).
3667
3668 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3669 const int max_tries = 10;
3670 char* base[max_tries];
3671 size_t size[max_tries];
3672 const size_t gap = 0x000000;
3673
3674 // Assert only that the size is a multiple of the page size, since
3675 // that's all that mmap requires, and since that's all we really know
3676 // about at this low abstraction level. If we need higher alignment,
3677 // we can either pass an alignment to this method or verify alignment
3678 // in one of the methods further up the call chain. See bug 5044738.
3679 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3680
3681 // Repeatedly allocate blocks until the block is allocated at the
3682 // right spot. Give up after max_tries. Note that reserve_memory() will
3683 // automatically update _highest_vm_reserved_address if the call is
3684 // successful. The variable tracks the highest memory address every reserved
3685 // by JVM. It is used to detect heap-stack collision if running with
3686 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3687 // space than needed, it could confuse the collision detecting code. To
3688 // solve the problem, save current _highest_vm_reserved_address and
3689 // calculate the correct value before return.
3690 address old_highest = _highest_vm_reserved_address;
3691
3692 // Linux mmap allows caller to pass an address as hint; give it a try first,
3693 // if kernel honors the hint then we can return immediately.
3694 char * addr = anon_mmap(requested_addr, bytes, false);
3695 if (addr == requested_addr) {
3696 return requested_addr;
3697 }
3698
3699 if (addr != NULL) {
3700 // mmap() is successful but it fails to reserve at the requested address
3701 anon_munmap(addr, bytes);
3702 }
3703
3704 int i;
3705 for (i = 0; i < max_tries; ++i) {
3706 base[i] = reserve_memory(bytes);
3707
3708 if (base[i] != NULL) {
3709 // Is this the block we wanted?
3710 if (base[i] == requested_addr) {
3711 size[i] = bytes;
3712 break;
3713 }
3714
3715 // Does this overlap the block we wanted? Give back the overlapped
3716 // parts and try again.
3717
3718 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3719 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3720 unmap_memory(base[i], top_overlap);
3721 base[i] += top_overlap;
3722 size[i] = bytes - top_overlap;
3723 } else {
3724 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3725 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3726 unmap_memory(requested_addr, bottom_overlap);
3727 size[i] = bytes - bottom_overlap;
3728 } else {
3729 size[i] = bytes;
3730 }
3731 }
3732 }
3733 }
3734
3735 // Give back the unused reserved pieces.
3736
3737 for (int j = 0; j < i; ++j) {
3738 if (base[j] != NULL) {
3739 unmap_memory(base[j], size[j]);
3740 }
3741 }
3742
3743 if (i < max_tries) {
3744 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3745 return requested_addr;
3746 } else {
3747 _highest_vm_reserved_address = old_highest;
3748 return NULL;
3749 }
3750 }
3751
3752 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3753 return ::read(fd, buf, nBytes);
3754 }
3755
3756 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3757 return ::pread(fd, buf, nBytes, offset);
3758 }
3759
3760 // Short sleep, direct OS call.
3761 //
3762 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3763 // sched_yield(2) will actually give up the CPU:
3764 //
3765 // * Alone on this pariticular CPU, keeps running.
3766 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3767 // (pre 2.6.39).
3768 //
3769 // So calling this with 0 is an alternative.
3770 //
3771 void os::naked_short_sleep(jlong ms) {
3772 struct timespec req;
3773
3774 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3775 req.tv_sec = 0;
3776 if (ms > 0) {
3777 req.tv_nsec = (ms % 1000) * 1000000;
3778 } else {
3779 req.tv_nsec = 1;
3780 }
3781
3782 nanosleep(&req, NULL);
3783
3784 return;
3785 }
3786
3787 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3788 void os::infinite_sleep() {
3789 while (true) { // sleep forever ...
3790 ::sleep(100); // ... 100 seconds at a time
3791 }
3792 }
3793
3794 // Used to convert frequent JVM_Yield() to nops
3795 bool os::dont_yield() {
3796 return DontYieldALot;
3797 }
3798
3799 void os::naked_yield() {
3800 sched_yield();
3801 }
3802
3803 ////////////////////////////////////////////////////////////////////////////////
3804 // thread priority support
3805
3806 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3807 // only supports dynamic priority, static priority must be zero. For real-time
3808 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3809 // However, for large multi-threaded applications, SCHED_RR is not only slower
3810 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3811 // of 5 runs - Sep 2005).
3812 //
3813 // The following code actually changes the niceness of kernel-thread/LWP. It
3814 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3815 // not the entire user process, and user level threads are 1:1 mapped to kernel
3816 // threads. It has always been the case, but could change in the future. For
3817 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3818 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3819
3820 int os::java_to_os_priority[CriticalPriority + 1] = {
3821 19, // 0 Entry should never be used
3822
3823 4, // 1 MinPriority
3824 3, // 2
3825 2, // 3
3826
3827 1, // 4
3828 0, // 5 NormPriority
3829 -1, // 6
3830
3831 -2, // 7
3832 -3, // 8
3833 -4, // 9 NearMaxPriority
3834
3835 -5, // 10 MaxPriority
3836
3837 -5 // 11 CriticalPriority
3838 };
3839
3840 static int prio_init() {
3841 if (ThreadPriorityPolicy == 1) {
3842 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3843 // if effective uid is not root. Perhaps, a more elegant way of doing
3844 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3845 if (geteuid() != 0) {
3846 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3847 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3848 }
3849 ThreadPriorityPolicy = 0;
3850 }
3851 }
3852 if (UseCriticalJavaThreadPriority) {
3853 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3854 }
3855 return 0;
3856 }
3857
3858 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3859 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3860
3861 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3862 return (ret == 0) ? OS_OK : OS_ERR;
3863 }
3864
3865 OSReturn os::get_native_priority(const Thread* const thread,
3866 int *priority_ptr) {
3867 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3868 *priority_ptr = java_to_os_priority[NormPriority];
3869 return OS_OK;
3870 }
3871
3872 errno = 0;
3873 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3874 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3875 }
3876
3877 // Hint to the underlying OS that a task switch would not be good.
3878 // Void return because it's a hint and can fail.
3879 void os::hint_no_preempt() {}
3880
3881 ////////////////////////////////////////////////////////////////////////////////
3882 // suspend/resume support
3883
3884 // the low-level signal-based suspend/resume support is a remnant from the
3885 // old VM-suspension that used to be for java-suspension, safepoints etc,
3886 // within hotspot. Now there is a single use-case for this:
3887 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3888 // that runs in the watcher thread.
3889 // The remaining code is greatly simplified from the more general suspension
3890 // code that used to be used.
3891 //
3892 // The protocol is quite simple:
3893 // - suspend:
3894 // - sends a signal to the target thread
3895 // - polls the suspend state of the osthread using a yield loop
3896 // - target thread signal handler (SR_handler) sets suspend state
3897 // and blocks in sigsuspend until continued
3898 // - resume:
3899 // - sets target osthread state to continue
3900 // - sends signal to end the sigsuspend loop in the SR_handler
3901 //
3902 // Note that the SR_lock plays no role in this suspend/resume protocol.
3903
3904 static void resume_clear_context(OSThread *osthread) {
3905 osthread->set_ucontext(NULL);
3906 osthread->set_siginfo(NULL);
3907 }
3908
3909 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3910 ucontext_t* context) {
3911 osthread->set_ucontext(context);
3912 osthread->set_siginfo(siginfo);
3913 }
3914
3915 // Handler function invoked when a thread's execution is suspended or
3916 // resumed. We have to be careful that only async-safe functions are
3917 // called here (Note: most pthread functions are not async safe and
3918 // should be avoided.)
3919 //
3920 // Note: sigwait() is a more natural fit than sigsuspend() from an
3921 // interface point of view, but sigwait() prevents the signal hander
3922 // from being run. libpthread would get very confused by not having
3923 // its signal handlers run and prevents sigwait()'s use with the
3924 // mutex granting granting signal.
3925 //
3926 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3927 //
3928 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3929 // Save and restore errno to avoid confusing native code with EINTR
3930 // after sigsuspend.
3931 int old_errno = errno;
3932
3933 Thread* thread = Thread::current();
3934 OSThread* osthread = thread->osthread();
3935 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3936
3937 os::SuspendResume::State current = osthread->sr.state();
3938 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3939 suspend_save_context(osthread, siginfo, context);
3940
3941 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3942 os::SuspendResume::State state = osthread->sr.suspended();
3943 if (state == os::SuspendResume::SR_SUSPENDED) {
3944 sigset_t suspend_set; // signals for sigsuspend()
3945
3946 // get current set of blocked signals and unblock resume signal
3947 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3948 sigdelset(&suspend_set, SR_signum);
3949
3950 sr_semaphore.signal();
3951 // wait here until we are resumed
3952 while (1) {
3953 sigsuspend(&suspend_set);
3954
3955 os::SuspendResume::State result = osthread->sr.running();
3956 if (result == os::SuspendResume::SR_RUNNING) {
3957 sr_semaphore.signal();
3958 break;
3959 }
3960 }
3961
3962 } else if (state == os::SuspendResume::SR_RUNNING) {
3963 // request was cancelled, continue
3964 } else {
3965 ShouldNotReachHere();
3966 }
3967
3968 resume_clear_context(osthread);
3969 } else if (current == os::SuspendResume::SR_RUNNING) {
3970 // request was cancelled, continue
3971 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3972 // ignore
3973 } else {
3974 // ignore
3975 }
3976
3977 errno = old_errno;
3978 }
3979
3980
3981 static int SR_initialize() {
3982 struct sigaction act;
3983 char *s;
3984 // Get signal number to use for suspend/resume
3985 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3986 int sig = ::strtol(s, 0, 10);
3987 if (sig > 0 || sig < _NSIG) {
3988 SR_signum = sig;
3989 }
3990 }
3991
3992 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3993 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3994
3995 sigemptyset(&SR_sigset);
3996 sigaddset(&SR_sigset, SR_signum);
3997
3998 // Set up signal handler for suspend/resume
3999 act.sa_flags = SA_RESTART|SA_SIGINFO;
4000 act.sa_handler = (void (*)(int)) SR_handler;
4001
4002 // SR_signum is blocked by default.
4003 // 4528190 - We also need to block pthread restart signal (32 on all
4004 // supported Linux platforms). Note that LinuxThreads need to block
4005 // this signal for all threads to work properly. So we don't have
4006 // to use hard-coded signal number when setting up the mask.
4007 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4008
4009 if (sigaction(SR_signum, &act, 0) == -1) {
4010 return -1;
4011 }
4012
4013 // Save signal flag
4014 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4015 return 0;
4016 }
4017
4018 static int sr_notify(OSThread* osthread) {
4019 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4020 assert_status(status == 0, status, "pthread_kill");
4021 return status;
4022 }
4023
4024 // "Randomly" selected value for how long we want to spin
4025 // before bailing out on suspending a thread, also how often
4026 // we send a signal to a thread we want to resume
4027 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4028 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4029
4030 // returns true on success and false on error - really an error is fatal
4031 // but this seems the normal response to library errors
4032 static bool do_suspend(OSThread* osthread) {
4033 assert(osthread->sr.is_running(), "thread should be running");
4034 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4035
4036 // mark as suspended and send signal
4037 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4038 // failed to switch, state wasn't running?
4039 ShouldNotReachHere();
4040 return false;
4041 }
4042
4043 if (sr_notify(osthread) != 0) {
4044 ShouldNotReachHere();
4045 }
4046
4047 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4048 while (true) {
4049 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4050 break;
4051 } else {
4052 // timeout
4053 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4054 if (cancelled == os::SuspendResume::SR_RUNNING) {
4055 return false;
4056 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4057 // make sure that we consume the signal on the semaphore as well
4058 sr_semaphore.wait();
4059 break;
4060 } else {
4061 ShouldNotReachHere();
4062 return false;
4063 }
4064 }
4065 }
4066
4067 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4068 return true;
4069 }
4070
4071 static void do_resume(OSThread* osthread) {
4072 assert(osthread->sr.is_suspended(), "thread should be suspended");
4073 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4074
4075 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4076 // failed to switch to WAKEUP_REQUEST
4077 ShouldNotReachHere();
4078 return;
4079 }
4080
4081 while (true) {
4082 if (sr_notify(osthread) == 0) {
4083 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4084 if (osthread->sr.is_running()) {
4085 return;
4086 }
4087 }
4088 } else {
4089 ShouldNotReachHere();
4090 }
4091 }
4092
4093 guarantee(osthread->sr.is_running(), "Must be running!");
4094 }
4095
4096 ///////////////////////////////////////////////////////////////////////////////////
4097 // signal handling (except suspend/resume)
4098
4099 // This routine may be used by user applications as a "hook" to catch signals.
4100 // The user-defined signal handler must pass unrecognized signals to this
4101 // routine, and if it returns true (non-zero), then the signal handler must
4102 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4103 // routine will never retun false (zero), but instead will execute a VM panic
4104 // routine kill the process.
4105 //
4106 // If this routine returns false, it is OK to call it again. This allows
4107 // the user-defined signal handler to perform checks either before or after
4108 // the VM performs its own checks. Naturally, the user code would be making
4109 // a serious error if it tried to handle an exception (such as a null check
4110 // or breakpoint) that the VM was generating for its own correct operation.
4111 //
4112 // This routine may recognize any of the following kinds of signals:
4113 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4114 // It should be consulted by handlers for any of those signals.
4115 //
4116 // The caller of this routine must pass in the three arguments supplied
4117 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4118 // field of the structure passed to sigaction(). This routine assumes that
4119 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4120 //
4121 // Note that the VM will print warnings if it detects conflicting signal
4122 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4123 //
4124 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4125 siginfo_t* siginfo,
4126 void* ucontext,
4127 int abort_if_unrecognized);
4128
4129 void signalHandler(int sig, siginfo_t* info, void* uc) {
4130 assert(info != NULL && uc != NULL, "it must be old kernel");
4131 int orig_errno = errno; // Preserve errno value over signal handler.
4132 JVM_handle_linux_signal(sig, info, uc, true);
4133 errno = orig_errno;
4134 }
4135
4136
4137 // This boolean allows users to forward their own non-matching signals
4138 // to JVM_handle_linux_signal, harmlessly.
4139 bool os::Linux::signal_handlers_are_installed = false;
4140
4141 // For signal-chaining
4142 struct sigaction os::Linux::sigact[MAXSIGNUM];
4143 unsigned int os::Linux::sigs = 0;
4144 bool os::Linux::libjsig_is_loaded = false;
4145 typedef struct sigaction *(*get_signal_t)(int);
4146 get_signal_t os::Linux::get_signal_action = NULL;
4147
4148 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4149 struct sigaction *actp = NULL;
4150
4151 if (libjsig_is_loaded) {
4152 // Retrieve the old signal handler from libjsig
4153 actp = (*get_signal_action)(sig);
4154 }
4155 if (actp == NULL) {
4156 // Retrieve the preinstalled signal handler from jvm
4157 actp = get_preinstalled_handler(sig);
4158 }
4159
4160 return actp;
4161 }
4162
4163 static bool call_chained_handler(struct sigaction *actp, int sig,
4164 siginfo_t *siginfo, void *context) {
4165 // Call the old signal handler
4166 if (actp->sa_handler == SIG_DFL) {
4167 // It's more reasonable to let jvm treat it as an unexpected exception
4168 // instead of taking the default action.
4169 return false;
4170 } else if (actp->sa_handler != SIG_IGN) {
4171 if ((actp->sa_flags & SA_NODEFER) == 0) {
4172 // automaticlly block the signal
4173 sigaddset(&(actp->sa_mask), sig);
4174 }
4175
4176 sa_handler_t hand;
4177 sa_sigaction_t sa;
4178 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4179 // retrieve the chained handler
4180 if (siginfo_flag_set) {
4181 sa = actp->sa_sigaction;
4182 } else {
4183 hand = actp->sa_handler;
4184 }
4185
4186 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4187 actp->sa_handler = SIG_DFL;
4188 }
4189
4190 // try to honor the signal mask
4191 sigset_t oset;
4192 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4193
4194 // call into the chained handler
4195 if (siginfo_flag_set) {
4196 (*sa)(sig, siginfo, context);
4197 } else {
4198 (*hand)(sig);
4199 }
4200
4201 // restore the signal mask
4202 pthread_sigmask(SIG_SETMASK, &oset, 0);
4203 }
4204 // Tell jvm's signal handler the signal is taken care of.
4205 return true;
4206 }
4207
4208 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4209 bool chained = false;
4210 // signal-chaining
4211 if (UseSignalChaining) {
4212 struct sigaction *actp = get_chained_signal_action(sig);
4213 if (actp != NULL) {
4214 chained = call_chained_handler(actp, sig, siginfo, context);
4215 }
4216 }
4217 return chained;
4218 }
4219
4220 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4221 if ((((unsigned int)1 << sig) & sigs) != 0) {
4222 return &sigact[sig];
4223 }
4224 return NULL;
4225 }
4226
4227 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4228 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4229 sigact[sig] = oldAct;
4230 sigs |= (unsigned int)1 << sig;
4231 }
4232
4233 // for diagnostic
4234 int os::Linux::sigflags[MAXSIGNUM];
4235
4236 int os::Linux::get_our_sigflags(int sig) {
4237 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4238 return sigflags[sig];
4239 }
4240
4241 void os::Linux::set_our_sigflags(int sig, int flags) {
4242 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4243 sigflags[sig] = flags;
4244 }
4245
4246 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4247 // Check for overwrite.
4248 struct sigaction oldAct;
4249 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4250
4251 void* oldhand = oldAct.sa_sigaction
4252 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4253 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4254 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4255 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4256 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4257 if (AllowUserSignalHandlers || !set_installed) {
4258 // Do not overwrite; user takes responsibility to forward to us.
4259 return;
4260 } else if (UseSignalChaining) {
4261 // save the old handler in jvm
4262 save_preinstalled_handler(sig, oldAct);
4263 // libjsig also interposes the sigaction() call below and saves the
4264 // old sigaction on it own.
4265 } else {
4266 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4267 "%#lx for signal %d.", (long)oldhand, sig));
4268 }
4269 }
4270
4271 struct sigaction sigAct;
4272 sigfillset(&(sigAct.sa_mask));
4273 sigAct.sa_handler = SIG_DFL;
4274 if (!set_installed) {
4275 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4276 } else {
4277 sigAct.sa_sigaction = signalHandler;
4278 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4279 }
4280 // Save flags, which are set by ours
4281 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4282 sigflags[sig] = sigAct.sa_flags;
4283
4284 int ret = sigaction(sig, &sigAct, &oldAct);
4285 assert(ret == 0, "check");
4286
4287 void* oldhand2 = oldAct.sa_sigaction
4288 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4289 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4290 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4291 }
4292
4293 // install signal handlers for signals that HotSpot needs to
4294 // handle in order to support Java-level exception handling.
4295
4296 void os::Linux::install_signal_handlers() {
4297 if (!signal_handlers_are_installed) {
4298 signal_handlers_are_installed = true;
4299
4300 // signal-chaining
4301 typedef void (*signal_setting_t)();
4302 signal_setting_t begin_signal_setting = NULL;
4303 signal_setting_t end_signal_setting = NULL;
4304 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4305 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4306 if (begin_signal_setting != NULL) {
4307 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4308 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4309 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4310 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4311 libjsig_is_loaded = true;
4312 assert(UseSignalChaining, "should enable signal-chaining");
4313 }
4314 if (libjsig_is_loaded) {
4315 // Tell libjsig jvm is setting signal handlers
4316 (*begin_signal_setting)();
4317 }
4318
4319 set_signal_handler(SIGSEGV, true);
4320 set_signal_handler(SIGPIPE, true);
4321 set_signal_handler(SIGBUS, true);
4322 set_signal_handler(SIGILL, true);
4323 set_signal_handler(SIGFPE, true);
4324 #if defined(PPC64)
4325 set_signal_handler(SIGTRAP, true);
4326 #endif
4327 set_signal_handler(SIGXFSZ, true);
4328
4329 if (libjsig_is_loaded) {
4330 // Tell libjsig jvm finishes setting signal handlers
4331 (*end_signal_setting)();
4332 }
4333
4334 // We don't activate signal checker if libjsig is in place, we trust ourselves
4335 // and if UserSignalHandler is installed all bets are off.
4336 // Log that signal checking is off only if -verbose:jni is specified.
4337 if (CheckJNICalls) {
4338 if (libjsig_is_loaded) {
4339 if (PrintJNIResolving) {
4340 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4341 }
4342 check_signals = false;
4343 }
4344 if (AllowUserSignalHandlers) {
4345 if (PrintJNIResolving) {
4346 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4347 }
4348 check_signals = false;
4349 }
4350 }
4351 }
4352 }
4353
4354 // This is the fastest way to get thread cpu time on Linux.
4355 // Returns cpu time (user+sys) for any thread, not only for current.
4356 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4357 // It might work on 2.6.10+ with a special kernel/glibc patch.
4358 // For reference, please, see IEEE Std 1003.1-2004:
4359 // http://www.unix.org/single_unix_specification
4360
4361 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4362 struct timespec tp;
4363 int rc = os::Linux::clock_gettime(clockid, &tp);
4364 assert(rc == 0, "clock_gettime is expected to return 0 code");
4365
4366 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4367 }
4368
4369 /////
4370 // glibc on Linux platform uses non-documented flag
4371 // to indicate, that some special sort of signal
4372 // trampoline is used.
4373 // We will never set this flag, and we should
4374 // ignore this flag in our diagnostic
4375 #ifdef SIGNIFICANT_SIGNAL_MASK
4376 #undef SIGNIFICANT_SIGNAL_MASK
4377 #endif
4378 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4379
4380 static const char* get_signal_handler_name(address handler,
4381 char* buf, int buflen) {
4382 int offset;
4383 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4384 if (found) {
4385 // skip directory names
4386 const char *p1, *p2;
4387 p1 = buf;
4388 size_t len = strlen(os::file_separator());
4389 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4390 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4391 } else {
4392 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4393 }
4394 return buf;
4395 }
4396
4397 static void print_signal_handler(outputStream* st, int sig,
4398 char* buf, size_t buflen) {
4399 struct sigaction sa;
4400
4401 sigaction(sig, NULL, &sa);
4402
4403 // See comment for SIGNIFICANT_SIGNAL_MASK define
4404 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4405
4406 st->print("%s: ", os::exception_name(sig, buf, buflen));
4407
4408 address handler = (sa.sa_flags & SA_SIGINFO)
4409 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4410 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4411
4412 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4413 st->print("SIG_DFL");
4414 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4415 st->print("SIG_IGN");
4416 } else {
4417 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4418 }
4419
4420 st->print(", sa_mask[0]=");
4421 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4422
4423 address rh = VMError::get_resetted_sighandler(sig);
4424 // May be, handler was resetted by VMError?
4425 if (rh != NULL) {
4426 handler = rh;
4427 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4428 }
4429
4430 st->print(", sa_flags=");
4431 os::Posix::print_sa_flags(st, sa.sa_flags);
4432
4433 // Check: is it our handler?
4434 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4435 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4436 // It is our signal handler
4437 // check for flags, reset system-used one!
4438 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4439 st->print(
4440 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4441 os::Linux::get_our_sigflags(sig));
4442 }
4443 }
4444 st->cr();
4445 }
4446
4447
4448 #define DO_SIGNAL_CHECK(sig) \
4449 do { \
4450 if (!sigismember(&check_signal_done, sig)) { \
4451 os::Linux::check_signal_handler(sig); \
4452 } \
4453 } while (0)
4454
4455 // This method is a periodic task to check for misbehaving JNI applications
4456 // under CheckJNI, we can add any periodic checks here
4457
4458 void os::run_periodic_checks() {
4459 if (check_signals == false) return;
4460
4461 // SEGV and BUS if overridden could potentially prevent
4462 // generation of hs*.log in the event of a crash, debugging
4463 // such a case can be very challenging, so we absolutely
4464 // check the following for a good measure:
4465 DO_SIGNAL_CHECK(SIGSEGV);
4466 DO_SIGNAL_CHECK(SIGILL);
4467 DO_SIGNAL_CHECK(SIGFPE);
4468 DO_SIGNAL_CHECK(SIGBUS);
4469 DO_SIGNAL_CHECK(SIGPIPE);
4470 DO_SIGNAL_CHECK(SIGXFSZ);
4471 #if defined(PPC64)
4472 DO_SIGNAL_CHECK(SIGTRAP);
4473 #endif
4474
4475 // ReduceSignalUsage allows the user to override these handlers
4476 // see comments at the very top and jvm_solaris.h
4477 if (!ReduceSignalUsage) {
4478 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4479 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4480 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4481 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4482 }
4483
4484 DO_SIGNAL_CHECK(SR_signum);
4485 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4486 }
4487
4488 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4489
4490 static os_sigaction_t os_sigaction = NULL;
4491
4492 void os::Linux::check_signal_handler(int sig) {
4493 char buf[O_BUFLEN];
4494 address jvmHandler = NULL;
4495
4496
4497 struct sigaction act;
4498 if (os_sigaction == NULL) {
4499 // only trust the default sigaction, in case it has been interposed
4500 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4501 if (os_sigaction == NULL) return;
4502 }
4503
4504 os_sigaction(sig, (struct sigaction*)NULL, &act);
4505
4506
4507 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4508
4509 address thisHandler = (act.sa_flags & SA_SIGINFO)
4510 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4511 : CAST_FROM_FN_PTR(address, act.sa_handler);
4512
4513
4514 switch (sig) {
4515 case SIGSEGV:
4516 case SIGBUS:
4517 case SIGFPE:
4518 case SIGPIPE:
4519 case SIGILL:
4520 case SIGXFSZ:
4521 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4522 break;
4523
4524 case SHUTDOWN1_SIGNAL:
4525 case SHUTDOWN2_SIGNAL:
4526 case SHUTDOWN3_SIGNAL:
4527 case BREAK_SIGNAL:
4528 jvmHandler = (address)user_handler();
4529 break;
4530
4531 case INTERRUPT_SIGNAL:
4532 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4533 break;
4534
4535 default:
4536 if (sig == SR_signum) {
4537 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4538 } else {
4539 return;
4540 }
4541 break;
4542 }
4543
4544 if (thisHandler != jvmHandler) {
4545 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4546 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4547 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4548 // No need to check this sig any longer
4549 sigaddset(&check_signal_done, sig);
4550 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4551 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4552 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4553 exception_name(sig, buf, O_BUFLEN));
4554 }
4555 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4556 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4557 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4558 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4559 // No need to check this sig any longer
4560 sigaddset(&check_signal_done, sig);
4561 }
4562
4563 // Dump all the signal
4564 if (sigismember(&check_signal_done, sig)) {
4565 print_signal_handlers(tty, buf, O_BUFLEN);
4566 }
4567 }
4568
4569 extern void report_error(char* file_name, int line_no, char* title,
4570 char* format, ...);
4571
4572 extern bool signal_name(int signo, char* buf, size_t len);
4573
4574 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4575 if (0 < exception_code && exception_code <= SIGRTMAX) {
4576 // signal
4577 if (!signal_name(exception_code, buf, size)) {
4578 jio_snprintf(buf, size, "SIG%d", exception_code);
4579 }
4580 return buf;
4581 } else {
4582 return NULL;
4583 }
4584 }
4585
4586 // this is called _before_ the most of global arguments have been parsed
4587 void os::init(void) {
4588 char dummy; // used to get a guess on initial stack address
4589 // first_hrtime = gethrtime();
4590
4591 // With LinuxThreads the JavaMain thread pid (primordial thread)
4592 // is different than the pid of the java launcher thread.
4593 // So, on Linux, the launcher thread pid is passed to the VM
4594 // via the sun.java.launcher.pid property.
4595 // Use this property instead of getpid() if it was correctly passed.
4596 // See bug 6351349.
4597 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4598
4599 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4600
4601 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4602
4603 init_random(1234567);
4604
4605 ThreadCritical::initialize();
4606
4607 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4608 if (Linux::page_size() == -1) {
4609 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4610 strerror(errno)));
4611 }
4612 init_page_sizes((size_t) Linux::page_size());
4613
4614 Linux::initialize_system_info();
4615
4616 // main_thread points to the aboriginal thread
4617 Linux::_main_thread = pthread_self();
4618
4619 Linux::clock_init();
4620 initial_time_count = javaTimeNanos();
4621
4622 // pthread_condattr initialization for monotonic clock
4623 int status;
4624 pthread_condattr_t* _condattr = os::Linux::condAttr();
4625 if ((status = pthread_condattr_init(_condattr)) != 0) {
4626 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4627 }
4628 // Only set the clock if CLOCK_MONOTONIC is available
4629 if (os::supports_monotonic_clock()) {
4630 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4631 if (status == EINVAL) {
4632 warning("Unable to use monotonic clock with relative timed-waits" \
4633 " - changes to the time-of-day clock may have adverse affects");
4634 } else {
4635 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4636 }
4637 }
4638 }
4639 // else it defaults to CLOCK_REALTIME
4640
4641 // If the pagesize of the VM is greater than 8K determine the appropriate
4642 // number of initial guard pages. The user can change this with the
4643 // command line arguments, if needed.
4644 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4645 StackYellowPages = 1;
4646 StackRedPages = 1;
4647 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4648 }
4649
4650 // retrieve entry point for pthread_setname_np
4651 Linux::_pthread_setname_np =
4652 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4653
4654 }
4655
4656 // To install functions for atexit system call
4657 extern "C" {
4658 static void perfMemory_exit_helper() {
4659 perfMemory_exit();
4660 }
4661 }
4662
4663 // this is called _after_ the global arguments have been parsed
4664 jint os::init_2(void) {
4665 Linux::fast_thread_clock_init();
4666
4667 // Allocate a single page and mark it as readable for safepoint polling
4668 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4669 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4670
4671 os::set_polling_page(polling_page);
4672
4673 #ifndef PRODUCT
4674 if (Verbose && PrintMiscellaneous) {
4675 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4676 (intptr_t)polling_page);
4677 }
4678 #endif
4679
4680 if (!UseMembar) {
4681 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4682 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4683 os::set_memory_serialize_page(mem_serialize_page);
4684
4685 #ifndef PRODUCT
4686 if (Verbose && PrintMiscellaneous) {
4687 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4688 (intptr_t)mem_serialize_page);
4689 }
4690 #endif
4691 }
4692
4693 // initialize suspend/resume support - must do this before signal_sets_init()
4694 if (SR_initialize() != 0) {
4695 perror("SR_initialize failed");
4696 return JNI_ERR;
4697 }
4698
4699 Linux::signal_sets_init();
4700 Linux::install_signal_handlers();
4701
4702 // Check minimum allowable stack size for thread creation and to initialize
4703 // the java system classes, including StackOverflowError - depends on page
4704 // size. Add a page for compiler2 recursion in main thread.
4705 // Add in 2*BytesPerWord times page size to account for VM stack during
4706 // class initialization depending on 32 or 64 bit VM.
4707 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4708 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4709 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4710
4711 size_t threadStackSizeInBytes = ThreadStackSize * K;
4712 if (threadStackSizeInBytes != 0 &&
4713 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4714 tty->print_cr("\nThe stack size specified is too small, "
4715 "Specify at least %dk",
4716 os::Linux::min_stack_allowed/ K);
4717 return JNI_ERR;
4718 }
4719
4720 // Make the stack size a multiple of the page size so that
4721 // the yellow/red zones can be guarded.
4722 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4723 vm_page_size()));
4724
4725 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4726
4727 #if defined(IA32)
4728 workaround_expand_exec_shield_cs_limit();
4729 #endif
4730
4731 Linux::libpthread_init();
4732 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4733 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4734 Linux::glibc_version(), Linux::libpthread_version(),
4735 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4736 }
4737
4738 if (UseNUMA) {
4739 if (!Linux::libnuma_init()) {
4740 UseNUMA = false;
4741 } else {
4742 if ((Linux::numa_max_node() < 1)) {
4743 // There's only one node(they start from 0), disable NUMA.
4744 UseNUMA = false;
4745 }
4746 }
4747 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4748 // we can make the adaptive lgrp chunk resizing work. If the user specified
4749 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4750 // disable adaptive resizing.
4751 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4752 if (FLAG_IS_DEFAULT(UseNUMA)) {
4753 UseNUMA = false;
4754 } else {
4755 if (FLAG_IS_DEFAULT(UseLargePages) &&
4756 FLAG_IS_DEFAULT(UseSHM) &&
4757 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4758 UseLargePages = false;
4759 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4760 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4761 UseAdaptiveSizePolicy = false;
4762 UseAdaptiveNUMAChunkSizing = false;
4763 }
4764 }
4765 }
4766 if (!UseNUMA && ForceNUMA) {
4767 UseNUMA = true;
4768 }
4769 }
4770
4771 if (MaxFDLimit) {
4772 // set the number of file descriptors to max. print out error
4773 // if getrlimit/setrlimit fails but continue regardless.
4774 struct rlimit nbr_files;
4775 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4776 if (status != 0) {
4777 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4778 perror("os::init_2 getrlimit failed");
4779 }
4780 } else {
4781 nbr_files.rlim_cur = nbr_files.rlim_max;
4782 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4783 if (status != 0) {
4784 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4785 perror("os::init_2 setrlimit failed");
4786 }
4787 }
4788 }
4789 }
4790
4791 // Initialize lock used to serialize thread creation (see os::create_thread)
4792 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4793
4794 // at-exit methods are called in the reverse order of their registration.
4795 // atexit functions are called on return from main or as a result of a
4796 // call to exit(3C). There can be only 32 of these functions registered
4797 // and atexit() does not set errno.
4798
4799 if (PerfAllowAtExitRegistration) {
4800 // only register atexit functions if PerfAllowAtExitRegistration is set.
4801 // atexit functions can be delayed until process exit time, which
4802 // can be problematic for embedded VM situations. Embedded VMs should
4803 // call DestroyJavaVM() to assure that VM resources are released.
4804
4805 // note: perfMemory_exit_helper atexit function may be removed in
4806 // the future if the appropriate cleanup code can be added to the
4807 // VM_Exit VMOperation's doit method.
4808 if (atexit(perfMemory_exit_helper) != 0) {
4809 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4810 }
4811 }
4812
4813 // initialize thread priority policy
4814 prio_init();
4815
4816 return JNI_OK;
4817 }
4818
4819 // Mark the polling page as unreadable
4820 void os::make_polling_page_unreadable(void) {
4821 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4822 fatal("Could not disable polling page");
4823 }
4824 }
4825
4826 // Mark the polling page as readable
4827 void os::make_polling_page_readable(void) {
4828 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4829 fatal("Could not enable polling page");
4830 }
4831 }
4832
4833 int os::active_processor_count() {
4834 // Linux doesn't yet have a (official) notion of processor sets,
4835 // so just return the number of online processors.
4836 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4837 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4838 return online_cpus;
4839 }
4840
4841 void os::set_native_thread_name(const char *name) {
4842 if (Linux::_pthread_setname_np) {
4843 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4844 snprintf(buf, sizeof(buf), "%s", name);
4845 buf[sizeof(buf) - 1] = '\0';
4846 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4847 // ERANGE should not happen; all other errors should just be ignored.
4848 assert(rc != ERANGE, "pthread_setname_np failed");
4849 }
4850 }
4851
4852 bool os::distribute_processes(uint length, uint* distribution) {
4853 // Not yet implemented.
4854 return false;
4855 }
4856
4857 bool os::bind_to_processor(uint processor_id) {
4858 // Not yet implemented.
4859 return false;
4860 }
4861
4862 ///
4863
4864 void os::SuspendedThreadTask::internal_do_task() {
4865 if (do_suspend(_thread->osthread())) {
4866 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4867 do_task(context);
4868 do_resume(_thread->osthread());
4869 }
4870 }
4871
4872 class PcFetcher : public os::SuspendedThreadTask {
4873 public:
4874 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4875 ExtendedPC result();
4876 protected:
4877 void do_task(const os::SuspendedThreadTaskContext& context);
4878 private:
4879 ExtendedPC _epc;
4880 };
4881
4882 ExtendedPC PcFetcher::result() {
4883 guarantee(is_done(), "task is not done yet.");
4884 return _epc;
4885 }
4886
4887 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4888 Thread* thread = context.thread();
4889 OSThread* osthread = thread->osthread();
4890 if (osthread->ucontext() != NULL) {
4891 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4892 } else {
4893 // NULL context is unexpected, double-check this is the VMThread
4894 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4895 }
4896 }
4897
4898 // Suspends the target using the signal mechanism and then grabs the PC before
4899 // resuming the target. Used by the flat-profiler only
4900 ExtendedPC os::get_thread_pc(Thread* thread) {
4901 // Make sure that it is called by the watcher for the VMThread
4902 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4903 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4904
4905 PcFetcher fetcher(thread);
4906 fetcher.run();
4907 return fetcher.result();
4908 }
4909
4910 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4911 pthread_mutex_t *_mutex,
4912 const struct timespec *_abstime) {
4913 if (is_NPTL()) {
4914 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4915 } else {
4916 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4917 // word back to default 64bit precision if condvar is signaled. Java
4918 // wants 53bit precision. Save and restore current value.
4919 int fpu = get_fpu_control_word();
4920 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4921 set_fpu_control_word(fpu);
4922 return status;
4923 }
4924 }
4925
4926 ////////////////////////////////////////////////////////////////////////////////
4927 // debug support
4928
4929 bool os::find(address addr, outputStream* st) {
4930 Dl_info dlinfo;
4931 memset(&dlinfo, 0, sizeof(dlinfo));
4932 if (dladdr(addr, &dlinfo) != 0) {
4933 st->print(PTR_FORMAT ": ", addr);
4934 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4935 st->print("%s+%#x", dlinfo.dli_sname,
4936 addr - (intptr_t)dlinfo.dli_saddr);
4937 } else if (dlinfo.dli_fbase != NULL) {
4938 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4939 } else {
4940 st->print("<absolute address>");
4941 }
4942 if (dlinfo.dli_fname != NULL) {
4943 st->print(" in %s", dlinfo.dli_fname);
4944 }
4945 if (dlinfo.dli_fbase != NULL) {
4946 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4947 }
4948 st->cr();
4949
4950 if (Verbose) {
4951 // decode some bytes around the PC
4952 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4953 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4954 address lowest = (address) dlinfo.dli_sname;
4955 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4956 if (begin < lowest) begin = lowest;
4957 Dl_info dlinfo2;
4958 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4959 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4960 end = (address) dlinfo2.dli_saddr;
4961 }
4962 Disassembler::decode(begin, end, st);
4963 }
4964 return true;
4965 }
4966 return false;
4967 }
4968
4969 ////////////////////////////////////////////////////////////////////////////////
4970 // misc
4971
4972 // This does not do anything on Linux. This is basically a hook for being
4973 // able to use structured exception handling (thread-local exception filters)
4974 // on, e.g., Win32.
4975 void
4976 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4977 JavaCallArguments* args, Thread* thread) {
4978 f(value, method, args, thread);
4979 }
4980
4981 void os::print_statistics() {
4982 }
4983
4984 int os::message_box(const char* title, const char* message) {
4985 int i;
4986 fdStream err(defaultStream::error_fd());
4987 for (i = 0; i < 78; i++) err.print_raw("=");
4988 err.cr();
4989 err.print_raw_cr(title);
4990 for (i = 0; i < 78; i++) err.print_raw("-");
4991 err.cr();
4992 err.print_raw_cr(message);
4993 for (i = 0; i < 78; i++) err.print_raw("=");
4994 err.cr();
4995
4996 char buf[16];
4997 // Prevent process from exiting upon "read error" without consuming all CPU
4998 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4999
5000 return buf[0] == 'y' || buf[0] == 'Y';
5001 }
5002
5003 int os::stat(const char *path, struct stat *sbuf) {
5004 char pathbuf[MAX_PATH];
5005 if (strlen(path) > MAX_PATH - 1) {
5006 errno = ENAMETOOLONG;
5007 return -1;
5008 }
5009 os::native_path(strcpy(pathbuf, path));
5010 return ::stat(pathbuf, sbuf);
5011 }
5012
5013 bool os::check_heap(bool force) {
5014 return true;
5015 }
5016
5017 // Is a (classpath) directory empty?
5018 bool os::dir_is_empty(const char* path) {
5019 DIR *dir = NULL;
5020 struct dirent *ptr;
5021
5022 dir = opendir(path);
5023 if (dir == NULL) return true;
5024
5025 // Scan the directory
5026 bool result = true;
5027 char buf[sizeof(struct dirent) + MAX_PATH];
5028 while (result && (ptr = ::readdir(dir)) != NULL) {
5029 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5030 result = false;
5031 }
5032 }
5033 closedir(dir);
5034 return result;
5035 }
5036
5037 // This code originates from JDK's sysOpen and open64_w
5038 // from src/solaris/hpi/src/system_md.c
5039
5040 int os::open(const char *path, int oflag, int mode) {
5041 if (strlen(path) > MAX_PATH - 1) {
5042 errno = ENAMETOOLONG;
5043 return -1;
5044 }
5045
5046 // All file descriptors that are opened in the Java process and not
5047 // specifically destined for a subprocess should have the close-on-exec
5048 // flag set. If we don't set it, then careless 3rd party native code
5049 // might fork and exec without closing all appropriate file descriptors
5050 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5051 // turn might:
5052 //
5053 // - cause end-of-file to fail to be detected on some file
5054 // descriptors, resulting in mysterious hangs, or
5055 //
5056 // - might cause an fopen in the subprocess to fail on a system
5057 // suffering from bug 1085341.
5058 //
5059 // (Yes, the default setting of the close-on-exec flag is a Unix
5060 // design flaw)
5061 //
5062 // See:
5063 // 1085341: 32-bit stdio routines should support file descriptors >255
5064 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5065 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5066 //
5067 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5068 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5069 // because it saves a system call and removes a small window where the flag
5070 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5071 // and we fall back to using FD_CLOEXEC (see below).
5072 #ifdef O_CLOEXEC
5073 oflag |= O_CLOEXEC;
5074 #endif
5075
5076 int fd = ::open64(path, oflag, mode);
5077 if (fd == -1) return -1;
5078
5079 //If the open succeeded, the file might still be a directory
5080 {
5081 struct stat64 buf64;
5082 int ret = ::fstat64(fd, &buf64);
5083 int st_mode = buf64.st_mode;
5084
5085 if (ret != -1) {
5086 if ((st_mode & S_IFMT) == S_IFDIR) {
5087 errno = EISDIR;
5088 ::close(fd);
5089 return -1;
5090 }
5091 } else {
5092 ::close(fd);
5093 return -1;
5094 }
5095 }
5096
5097 #ifdef FD_CLOEXEC
5098 // Validate that the use of the O_CLOEXEC flag on open above worked.
5099 // With recent kernels, we will perform this check exactly once.
5100 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5101 if (!O_CLOEXEC_is_known_to_work) {
5102 int flags = ::fcntl(fd, F_GETFD);
5103 if (flags != -1) {
5104 if ((flags & FD_CLOEXEC) != 0)
5105 O_CLOEXEC_is_known_to_work = 1;
5106 else
5107 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5108 }
5109 }
5110 #endif
5111
5112 return fd;
5113 }
5114
5115
5116 // create binary file, rewriting existing file if required
5117 int os::create_binary_file(const char* path, bool rewrite_existing) {
5118 int oflags = O_WRONLY | O_CREAT;
5119 if (!rewrite_existing) {
5120 oflags |= O_EXCL;
5121 }
5122 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5123 }
5124
5125 // return current position of file pointer
5126 jlong os::current_file_offset(int fd) {
5127 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5128 }
5129
5130 // move file pointer to the specified offset
5131 jlong os::seek_to_file_offset(int fd, jlong offset) {
5132 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5133 }
5134
5135 // This code originates from JDK's sysAvailable
5136 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5137
5138 int os::available(int fd, jlong *bytes) {
5139 jlong cur, end;
5140 int mode;
5141 struct stat64 buf64;
5142
5143 if (::fstat64(fd, &buf64) >= 0) {
5144 mode = buf64.st_mode;
5145 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5146 // XXX: is the following call interruptible? If so, this might
5147 // need to go through the INTERRUPT_IO() wrapper as for other
5148 // blocking, interruptible calls in this file.
5149 int n;
5150 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5151 *bytes = n;
5152 return 1;
5153 }
5154 }
5155 }
5156 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5157 return 0;
5158 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5159 return 0;
5160 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5161 return 0;
5162 }
5163 *bytes = end - cur;
5164 return 1;
5165 }
5166
5167 // Map a block of memory.
5168 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5169 char *addr, size_t bytes, bool read_only,
5170 bool allow_exec) {
5171 int prot;
5172 int flags = MAP_PRIVATE;
5173
5174 if (read_only) {
5175 prot = PROT_READ;
5176 } else {
5177 prot = PROT_READ | PROT_WRITE;
5178 }
5179
5180 if (allow_exec) {
5181 prot |= PROT_EXEC;
5182 }
5183
5184 if (addr != NULL) {
5185 flags |= MAP_FIXED;
5186 }
5187
5188 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5189 fd, file_offset);
5190 if (mapped_address == MAP_FAILED) {
5191 return NULL;
5192 }
5193 return mapped_address;
5194 }
5195
5196
5197 // Remap a block of memory.
5198 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5199 char *addr, size_t bytes, bool read_only,
5200 bool allow_exec) {
5201 // same as map_memory() on this OS
5202 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5203 allow_exec);
5204 }
5205
5206
5207 // Unmap a block of memory.
5208 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5209 return munmap(addr, bytes) == 0;
5210 }
5211
5212 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5213
5214 static clockid_t thread_cpu_clockid(Thread* thread) {
5215 pthread_t tid = thread->osthread()->pthread_id();
5216 clockid_t clockid;
5217
5218 // Get thread clockid
5219 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5220 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5221 return clockid;
5222 }
5223
5224 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5225 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5226 // of a thread.
5227 //
5228 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5229 // the fast estimate available on the platform.
5230
5231 jlong os::current_thread_cpu_time() {
5232 if (os::Linux::supports_fast_thread_cpu_time()) {
5233 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5234 } else {
5235 // return user + sys since the cost is the same
5236 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5237 }
5238 }
5239
5240 jlong os::thread_cpu_time(Thread* thread) {
5241 // consistent with what current_thread_cpu_time() returns
5242 if (os::Linux::supports_fast_thread_cpu_time()) {
5243 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5244 } else {
5245 return slow_thread_cpu_time(thread, true /* user + sys */);
5246 }
5247 }
5248
5249 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5250 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5251 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5252 } else {
5253 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5254 }
5255 }
5256
5257 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5258 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5259 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5260 } else {
5261 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5262 }
5263 }
5264
5265 // -1 on error.
5266 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5267 pid_t tid = thread->osthread()->thread_id();
5268 char *s;
5269 char stat[2048];
5270 int statlen;
5271 char proc_name[64];
5272 int count;
5273 long sys_time, user_time;
5274 char cdummy;
5275 int idummy;
5276 long ldummy;
5277 FILE *fp;
5278
5279 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5280 fp = fopen(proc_name, "r");
5281 if (fp == NULL) return -1;
5282 statlen = fread(stat, 1, 2047, fp);
5283 stat[statlen] = '\0';
5284 fclose(fp);
5285
5286 // Skip pid and the command string. Note that we could be dealing with
5287 // weird command names, e.g. user could decide to rename java launcher
5288 // to "java 1.4.2 :)", then the stat file would look like
5289 // 1234 (java 1.4.2 :)) R ... ...
5290 // We don't really need to know the command string, just find the last
5291 // occurrence of ")" and then start parsing from there. See bug 4726580.
5292 s = strrchr(stat, ')');
5293 if (s == NULL) return -1;
5294
5295 // Skip blank chars
5296 do { s++; } while (s && isspace(*s));
5297
5298 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5299 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5300 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5301 &user_time, &sys_time);
5302 if (count != 13) return -1;
5303 if (user_sys_cpu_time) {
5304 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5305 } else {
5306 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5307 }
5308 }
5309
5310 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5311 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5312 info_ptr->may_skip_backward = false; // elapsed time not wall time
5313 info_ptr->may_skip_forward = false; // elapsed time not wall time
5314 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5315 }
5316
5317 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5318 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5319 info_ptr->may_skip_backward = false; // elapsed time not wall time
5320 info_ptr->may_skip_forward = false; // elapsed time not wall time
5321 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5322 }
5323
5324 bool os::is_thread_cpu_time_supported() {
5325 return true;
5326 }
5327
5328 // System loadavg support. Returns -1 if load average cannot be obtained.
5329 // Linux doesn't yet have a (official) notion of processor sets,
5330 // so just return the system wide load average.
5331 int os::loadavg(double loadavg[], int nelem) {
5332 return ::getloadavg(loadavg, nelem);
5333 }
5334
5335 void os::pause() {
5336 char filename[MAX_PATH];
5337 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5338 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5339 } else {
5340 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5341 }
5342
5343 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5344 if (fd != -1) {
5345 struct stat buf;
5346 ::close(fd);
5347 while (::stat(filename, &buf) == 0) {
5348 (void)::poll(NULL, 0, 100);
5349 }
5350 } else {
5351 jio_fprintf(stderr,
5352 "Could not open pause file '%s', continuing immediately.\n", filename);
5353 }
5354 }
5355
5356
5357 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5358 // comment paragraphs are worth repeating here:
5359 //
5360 // Assumption:
5361 // Only one parker can exist on an event, which is why we allocate
5362 // them per-thread. Multiple unparkers can coexist.
5363 //
5364 // _Event serves as a restricted-range semaphore.
5365 // -1 : thread is blocked, i.e. there is a waiter
5366 // 0 : neutral: thread is running or ready,
5367 // could have been signaled after a wait started
5368 // 1 : signaled - thread is running or ready
5369 //
5370 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5371 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5372 // For specifics regarding the bug see GLIBC BUGID 261237 :
5373 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5374 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5375 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5376 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5377 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5378 // and monitorenter when we're using 1-0 locking. All those operations may result in
5379 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5380 // of libpthread avoids the problem, but isn't practical.
5381 //
5382 // Possible remedies:
5383 //
5384 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5385 // This is palliative and probabilistic, however. If the thread is preempted
5386 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5387 // than the minimum period may have passed, and the abstime may be stale (in the
5388 // past) resultin in a hang. Using this technique reduces the odds of a hang
5389 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5390 //
5391 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5392 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5393 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5394 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5395 // thread.
5396 //
5397 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5398 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5399 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5400 // This also works well. In fact it avoids kernel-level scalability impediments
5401 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5402 // timers in a graceful fashion.
5403 //
5404 // 4. When the abstime value is in the past it appears that control returns
5405 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5406 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5407 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5408 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5409 // It may be possible to avoid reinitialization by checking the return
5410 // value from pthread_cond_timedwait(). In addition to reinitializing the
5411 // condvar we must establish the invariant that cond_signal() is only called
5412 // within critical sections protected by the adjunct mutex. This prevents
5413 // cond_signal() from "seeing" a condvar that's in the midst of being
5414 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5415 // desirable signal-after-unlock optimization that avoids futile context switching.
5416 //
5417 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5418 // structure when a condvar is used or initialized. cond_destroy() would
5419 // release the helper structure. Our reinitialize-after-timedwait fix
5420 // put excessive stress on malloc/free and locks protecting the c-heap.
5421 //
5422 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5423 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5424 // and only enabling the work-around for vulnerable environments.
5425
5426 // utility to compute the abstime argument to timedwait:
5427 // millis is the relative timeout time
5428 // abstime will be the absolute timeout time
5429 // TODO: replace compute_abstime() with unpackTime()
5430
5431 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5432 if (millis < 0) millis = 0;
5433
5434 jlong seconds = millis / 1000;
5435 millis %= 1000;
5436 if (seconds > 50000000) { // see man cond_timedwait(3T)
5437 seconds = 50000000;
5438 }
5439
5440 if (os::supports_monotonic_clock()) {
5441 struct timespec now;
5442 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5443 assert_status(status == 0, status, "clock_gettime");
5444 abstime->tv_sec = now.tv_sec + seconds;
5445 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5446 if (nanos >= NANOSECS_PER_SEC) {
5447 abstime->tv_sec += 1;
5448 nanos -= NANOSECS_PER_SEC;
5449 }
5450 abstime->tv_nsec = nanos;
5451 } else {
5452 struct timeval now;
5453 int status = gettimeofday(&now, NULL);
5454 assert(status == 0, "gettimeofday");
5455 abstime->tv_sec = now.tv_sec + seconds;
5456 long usec = now.tv_usec + millis * 1000;
5457 if (usec >= 1000000) {
5458 abstime->tv_sec += 1;
5459 usec -= 1000000;
5460 }
5461 abstime->tv_nsec = usec * 1000;
5462 }
5463 return abstime;
5464 }
5465
5466 void os::PlatformEvent::park() { // AKA "down()"
5467 // Transitions for _Event:
5468 // -1 => -1 : illegal
5469 // 1 => 0 : pass - return immediately
5470 // 0 => -1 : block; then set _Event to 0 before returning
5471
5472 // Invariant: Only the thread associated with the Event/PlatformEvent
5473 // may call park().
5474 // TODO: assert that _Assoc != NULL or _Assoc == Self
5475 assert(_nParked == 0, "invariant");
5476
5477 int v;
5478 for (;;) {
5479 v = _Event;
5480 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5481 }
5482 guarantee(v >= 0, "invariant");
5483 if (v == 0) {
5484 // Do this the hard way by blocking ...
5485 int status = pthread_mutex_lock(_mutex);
5486 assert_status(status == 0, status, "mutex_lock");
5487 guarantee(_nParked == 0, "invariant");
5488 ++_nParked;
5489 while (_Event < 0) {
5490 status = pthread_cond_wait(_cond, _mutex);
5491 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5492 // Treat this the same as if the wait was interrupted
5493 if (status == ETIME) { status = EINTR; }
5494 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5495 }
5496 --_nParked;
5497
5498 _Event = 0;
5499 status = pthread_mutex_unlock(_mutex);
5500 assert_status(status == 0, status, "mutex_unlock");
5501 // Paranoia to ensure our locked and lock-free paths interact
5502 // correctly with each other.
5503 OrderAccess::fence();
5504 }
5505 guarantee(_Event >= 0, "invariant");
5506 }
5507
5508 int os::PlatformEvent::park(jlong millis) {
5509 // Transitions for _Event:
5510 // -1 => -1 : illegal
5511 // 1 => 0 : pass - return immediately
5512 // 0 => -1 : block; then set _Event to 0 before returning
5513
5514 guarantee(_nParked == 0, "invariant");
5515
5516 int v;
5517 for (;;) {
5518 v = _Event;
5519 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5520 }
5521 guarantee(v >= 0, "invariant");
5522 if (v != 0) return OS_OK;
5523
5524 // We do this the hard way, by blocking the thread.
5525 // Consider enforcing a minimum timeout value.
5526 struct timespec abst;
5527 compute_abstime(&abst, millis);
5528
5529 int ret = OS_TIMEOUT;
5530 int status = pthread_mutex_lock(_mutex);
5531 assert_status(status == 0, status, "mutex_lock");
5532 guarantee(_nParked == 0, "invariant");
5533 ++_nParked;
5534
5535 // Object.wait(timo) will return because of
5536 // (a) notification
5537 // (b) timeout
5538 // (c) thread.interrupt
5539 //
5540 // Thread.interrupt and object.notify{All} both call Event::set.
5541 // That is, we treat thread.interrupt as a special case of notification.
5542 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5543 // We assume all ETIME returns are valid.
5544 //
5545 // TODO: properly differentiate simultaneous notify+interrupt.
5546 // In that case, we should propagate the notify to another waiter.
5547
5548 while (_Event < 0) {
5549 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5550 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5551 pthread_cond_destroy(_cond);
5552 pthread_cond_init(_cond, os::Linux::condAttr());
5553 }
5554 assert_status(status == 0 || status == EINTR ||
5555 status == ETIME || status == ETIMEDOUT,
5556 status, "cond_timedwait");
5557 if (!FilterSpuriousWakeups) break; // previous semantics
5558 if (status == ETIME || status == ETIMEDOUT) break;
5559 // We consume and ignore EINTR and spurious wakeups.
5560 }
5561 --_nParked;
5562 if (_Event >= 0) {
5563 ret = OS_OK;
5564 }
5565 _Event = 0;
5566 status = pthread_mutex_unlock(_mutex);
5567 assert_status(status == 0, status, "mutex_unlock");
5568 assert(_nParked == 0, "invariant");
5569 // Paranoia to ensure our locked and lock-free paths interact
5570 // correctly with each other.
5571 OrderAccess::fence();
5572 return ret;
5573 }
5574
5575 void os::PlatformEvent::unpark() {
5576 // Transitions for _Event:
5577 // 0 => 1 : just return
5578 // 1 => 1 : just return
5579 // -1 => either 0 or 1; must signal target thread
5580 // That is, we can safely transition _Event from -1 to either
5581 // 0 or 1.
5582 // See also: "Semaphores in Plan 9" by Mullender & Cox
5583 //
5584 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5585 // that it will take two back-to-back park() calls for the owning
5586 // thread to block. This has the benefit of forcing a spurious return
5587 // from the first park() call after an unpark() call which will help
5588 // shake out uses of park() and unpark() without condition variables.
5589
5590 if (Atomic::xchg(1, &_Event) >= 0) return;
5591
5592 // Wait for the thread associated with the event to vacate
5593 int status = pthread_mutex_lock(_mutex);
5594 assert_status(status == 0, status, "mutex_lock");
5595 int AnyWaiters = _nParked;
5596 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5597 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5598 AnyWaiters = 0;
5599 pthread_cond_signal(_cond);
5600 }
5601 status = pthread_mutex_unlock(_mutex);
5602 assert_status(status == 0, status, "mutex_unlock");
5603 if (AnyWaiters != 0) {
5604 // Note that we signal() *after* dropping the lock for "immortal" Events.
5605 // This is safe and avoids a common class of futile wakeups. In rare
5606 // circumstances this can cause a thread to return prematurely from
5607 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5608 // will simply re-test the condition and re-park itself.
5609 // This provides particular benefit if the underlying platform does not
5610 // provide wait morphing.
5611 status = pthread_cond_signal(_cond);
5612 assert_status(status == 0, status, "cond_signal");
5613 }
5614 }
5615
5616
5617 // JSR166
5618 // -------------------------------------------------------
5619
5620 // The solaris and linux implementations of park/unpark are fairly
5621 // conservative for now, but can be improved. They currently use a
5622 // mutex/condvar pair, plus a a count.
5623 // Park decrements count if > 0, else does a condvar wait. Unpark
5624 // sets count to 1 and signals condvar. Only one thread ever waits
5625 // on the condvar. Contention seen when trying to park implies that someone
5626 // is unparking you, so don't wait. And spurious returns are fine, so there
5627 // is no need to track notifications.
5628
5629 // This code is common to linux and solaris and will be moved to a
5630 // common place in dolphin.
5631 //
5632 // The passed in time value is either a relative time in nanoseconds
5633 // or an absolute time in milliseconds. Either way it has to be unpacked
5634 // into suitable seconds and nanoseconds components and stored in the
5635 // given timespec structure.
5636 // Given time is a 64-bit value and the time_t used in the timespec is only
5637 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5638 // overflow if times way in the future are given. Further on Solaris versions
5639 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5640 // number of seconds, in abstime, is less than current_time + 100,000,000.
5641 // As it will be 28 years before "now + 100000000" will overflow we can
5642 // ignore overflow and just impose a hard-limit on seconds using the value
5643 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5644 // years from "now".
5645
5646 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5647 assert(time > 0, "convertTime");
5648 time_t max_secs = 0;
5649
5650 if (!os::supports_monotonic_clock() || isAbsolute) {
5651 struct timeval now;
5652 int status = gettimeofday(&now, NULL);
5653 assert(status == 0, "gettimeofday");
5654
5655 max_secs = now.tv_sec + MAX_SECS;
5656
5657 if (isAbsolute) {
5658 jlong secs = time / 1000;
5659 if (secs > max_secs) {
5660 absTime->tv_sec = max_secs;
5661 } else {
5662 absTime->tv_sec = secs;
5663 }
5664 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5665 } else {
5666 jlong secs = time / NANOSECS_PER_SEC;
5667 if (secs >= MAX_SECS) {
5668 absTime->tv_sec = max_secs;
5669 absTime->tv_nsec = 0;
5670 } else {
5671 absTime->tv_sec = now.tv_sec + secs;
5672 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5673 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5674 absTime->tv_nsec -= NANOSECS_PER_SEC;
5675 ++absTime->tv_sec; // note: this must be <= max_secs
5676 }
5677 }
5678 }
5679 } else {
5680 // must be relative using monotonic clock
5681 struct timespec now;
5682 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5683 assert_status(status == 0, status, "clock_gettime");
5684 max_secs = now.tv_sec + MAX_SECS;
5685 jlong secs = time / NANOSECS_PER_SEC;
5686 if (secs >= MAX_SECS) {
5687 absTime->tv_sec = max_secs;
5688 absTime->tv_nsec = 0;
5689 } else {
5690 absTime->tv_sec = now.tv_sec + secs;
5691 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5692 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5693 absTime->tv_nsec -= NANOSECS_PER_SEC;
5694 ++absTime->tv_sec; // note: this must be <= max_secs
5695 }
5696 }
5697 }
5698 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5699 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5700 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5701 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5702 }
5703
5704 void Parker::park(bool isAbsolute, jlong time) {
5705 // Ideally we'd do something useful while spinning, such
5706 // as calling unpackTime().
5707
5708 // Optional fast-path check:
5709 // Return immediately if a permit is available.
5710 // We depend on Atomic::xchg() having full barrier semantics
5711 // since we are doing a lock-free update to _counter.
5712 if (Atomic::xchg(0, &_counter) > 0) return;
5713
5714 Thread* thread = Thread::current();
5715 assert(thread->is_Java_thread(), "Must be JavaThread");
5716 JavaThread *jt = (JavaThread *)thread;
5717
5718 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5719 // Check interrupt before trying to wait
5720 if (Thread::is_interrupted(thread, false)) {
5721 return;
5722 }
5723
5724 // Next, demultiplex/decode time arguments
5725 timespec absTime;
5726 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5727 return;
5728 }
5729 if (time > 0) {
5730 unpackTime(&absTime, isAbsolute, time);
5731 }
5732
5733
5734 // Enter safepoint region
5735 // Beware of deadlocks such as 6317397.
5736 // The per-thread Parker:: mutex is a classic leaf-lock.
5737 // In particular a thread must never block on the Threads_lock while
5738 // holding the Parker:: mutex. If safepoints are pending both the
5739 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5740 ThreadBlockInVM tbivm(jt);
5741
5742 // Don't wait if cannot get lock since interference arises from
5743 // unblocking. Also. check interrupt before trying wait
5744 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5745 return;
5746 }
5747
5748 int status;
5749 if (_counter > 0) { // no wait needed
5750 _counter = 0;
5751 status = pthread_mutex_unlock(_mutex);
5752 assert(status == 0, "invariant");
5753 // Paranoia to ensure our locked and lock-free paths interact
5754 // correctly with each other and Java-level accesses.
5755 OrderAccess::fence();
5756 return;
5757 }
5758
5759 #ifdef ASSERT
5760 // Don't catch signals while blocked; let the running threads have the signals.
5761 // (This allows a debugger to break into the running thread.)
5762 sigset_t oldsigs;
5763 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5764 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5765 #endif
5766
5767 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5768 jt->set_suspend_equivalent();
5769 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5770
5771 assert(_cur_index == -1, "invariant");
5772 if (time == 0) {
5773 _cur_index = REL_INDEX; // arbitrary choice when not timed
5774 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5775 } else {
5776 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5777 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5778 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5779 pthread_cond_destroy(&_cond[_cur_index]);
5780 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5781 }
5782 }
5783 _cur_index = -1;
5784 assert_status(status == 0 || status == EINTR ||
5785 status == ETIME || status == ETIMEDOUT,
5786 status, "cond_timedwait");
5787
5788 #ifdef ASSERT
5789 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5790 #endif
5791
5792 _counter = 0;
5793 status = pthread_mutex_unlock(_mutex);
5794 assert_status(status == 0, status, "invariant");
5795 // Paranoia to ensure our locked and lock-free paths interact
5796 // correctly with each other and Java-level accesses.
5797 OrderAccess::fence();
5798
5799 // If externally suspended while waiting, re-suspend
5800 if (jt->handle_special_suspend_equivalent_condition()) {
5801 jt->java_suspend_self();
5802 }
5803 }
5804
5805 void Parker::unpark() {
5806 int status = pthread_mutex_lock(_mutex);
5807 assert(status == 0, "invariant");
5808 const int s = _counter;
5809 _counter = 1;
5810 if (s < 1) {
5811 // thread might be parked
5812 if (_cur_index != -1) {
5813 // thread is definitely parked
5814 if (WorkAroundNPTLTimedWaitHang) {
5815 status = pthread_cond_signal(&_cond[_cur_index]);
5816 assert(status == 0, "invariant");
5817 status = pthread_mutex_unlock(_mutex);
5818 assert(status == 0, "invariant");
5819 } else {
5820 status = pthread_mutex_unlock(_mutex);
5821 assert(status == 0, "invariant");
5822 status = pthread_cond_signal(&_cond[_cur_index]);
5823 assert(status == 0, "invariant");
5824 }
5825 } else {
5826 pthread_mutex_unlock(_mutex);
5827 assert(status == 0, "invariant");
5828 }
5829 } else {
5830 pthread_mutex_unlock(_mutex);
5831 assert(status == 0, "invariant");
5832 }
5833 }
5834
5835
5836 extern char** environ;
5837
5838 // Run the specified command in a separate process. Return its exit value,
5839 // or -1 on failure (e.g. can't fork a new process).
5840 // Unlike system(), this function can be called from signal handler. It
5841 // doesn't block SIGINT et al.
5842 int os::fork_and_exec(char* cmd) {
5843 const char * argv[4] = {"sh", "-c", cmd, NULL};
5844
5845 pid_t pid = fork();
5846
5847 if (pid < 0) {
5848 // fork failed
5849 return -1;
5850
5851 } else if (pid == 0) {
5852 // child process
5853
5854 execve("/bin/sh", (char* const*)argv, environ);
5855
5856 // execve failed
5857 _exit(-1);
5858
5859 } else {
5860 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5861 // care about the actual exit code, for now.
5862
5863 int status;
5864
5865 // Wait for the child process to exit. This returns immediately if
5866 // the child has already exited. */
5867 while (waitpid(pid, &status, 0) < 0) {
5868 switch (errno) {
5869 case ECHILD: return 0;
5870 case EINTR: break;
5871 default: return -1;
5872 }
5873 }
5874
5875 if (WIFEXITED(status)) {
5876 // The child exited normally; get its exit code.
5877 return WEXITSTATUS(status);
5878 } else if (WIFSIGNALED(status)) {
5879 // The child exited because of a signal
5880 // The best value to return is 0x80 + signal number,
5881 // because that is what all Unix shells do, and because
5882 // it allows callers to distinguish between process exit and
5883 // process death by signal.
5884 return 0x80 + WTERMSIG(status);
5885 } else {
5886 // Unknown exit code; pass it through
5887 return status;
5888 }
5889 }
5890 }
5891
5892 // is_headless_jre()
5893 //
5894 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5895 // in order to report if we are running in a headless jre
5896 //
5897 // Since JDK8 xawt/libmawt.so was moved into the same directory
5898 // as libawt.so, and renamed libawt_xawt.so
5899 //
5900 bool os::is_headless_jre() {
5901 struct stat statbuf;
5902 char buf[MAXPATHLEN];
5903 char libmawtpath[MAXPATHLEN];
5904 const char *xawtstr = "/xawt/libmawt.so";
5905 const char *new_xawtstr = "/libawt_xawt.so";
5906 char *p;
5907
5908 // Get path to libjvm.so
5909 os::jvm_path(buf, sizeof(buf));
5910
5911 // Get rid of libjvm.so
5912 p = strrchr(buf, '/');
5913 if (p == NULL) {
5914 return false;
5915 } else {
5916 *p = '\0';
5917 }
5918
5919 // Get rid of client or server
5920 p = strrchr(buf, '/');
5921 if (p == NULL) {
5922 return false;
5923 } else {
5924 *p = '\0';
5925 }
5926
5927 // check xawt/libmawt.so
5928 strcpy(libmawtpath, buf);
5929 strcat(libmawtpath, xawtstr);
5930 if (::stat(libmawtpath, &statbuf) == 0) return false;
5931
5932 // check libawt_xawt.so
5933 strcpy(libmawtpath, buf);
5934 strcat(libmawtpath, new_xawtstr);
5935 if (::stat(libmawtpath, &statbuf) == 0) return false;
5936
5937 return true;
5938 }
5939
5940 // Get the default path to the core file
5941 // Returns the length of the string
5942 int os::get_core_path(char* buffer, size_t bufferSize) {
5943 /*
5944 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5945 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5946 */
5947 const int core_pattern_len = 129;
5948 char core_pattern[core_pattern_len] = {0};
5949
5950 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5951 if (core_pattern_file != -1) {
5952 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5953 ::close(core_pattern_file);
5954
5955 if (ret > 0) {
5956 char *last_char = core_pattern + strlen(core_pattern) - 1;
5957
5958 if (*last_char == '\n') {
5959 *last_char = '\0';
5960 }
5961 }
5962 }
5963
5964 if (strlen(core_pattern) == 0) {
5965 return -1;
5966 }
5967
5968 char *pid_pos = strstr(core_pattern, "%p");
5969 int written;
5970
5971 if (core_pattern[0] == '/') {
5972 written = jio_snprintf(buffer, bufferSize, core_pattern);
5973 } else {
5974 char cwd[PATH_MAX];
5975
5976 const char* p = get_current_directory(cwd, PATH_MAX);
5977 if (p == NULL) {
5978 return -1;
5979 }
5980
5981 if (core_pattern[0] == '|') {
5982 written = jio_snprintf(buffer, bufferSize,
5983 "\"%s\" (or dumping to %s/core.%d)",
5984 &core_pattern[1], p, current_process_id());
5985 } else {
5986 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5987 }
5988 }
5989
5990 if (written < 0) {
5991 return -1;
5992 }
5993
5994 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5995 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5996
5997 if (core_uses_pid_file != -1) {
5998 char core_uses_pid = 0;
5999 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6000 ::close(core_uses_pid_file);
6001
6002 if (core_uses_pid == '1') {
6003 jio_snprintf(buffer + written, bufferSize - written,
6004 ".%d", current_process_id());
6005 }
6006 }
6007 }
6008
6009 return strlen(buffer);
6010 }
6011
6012 /////////////// Unit tests ///////////////
6013
6014 #ifndef PRODUCT
6015
6016 #define test_log(...) \
6017 do { \
6018 if (VerboseInternalVMTests) { \
6019 tty->print_cr(__VA_ARGS__); \
6020 tty->flush(); \
6021 } \
6022 } while (false)
6023
6024 class TestReserveMemorySpecial : AllStatic {
6025 public:
6026 static void small_page_write(void* addr, size_t size) {
6027 size_t page_size = os::vm_page_size();
6028
6029 char* end = (char*)addr + size;
6030 for (char* p = (char*)addr; p < end; p += page_size) {
6031 *p = 1;
6032 }
6033 }
6034
6035 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6036 if (!UseHugeTLBFS) {
6037 return;
6038 }
6039
6040 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6041
6042 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6043
6044 if (addr != NULL) {
6045 small_page_write(addr, size);
6046
6047 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6048 }
6049 }
6050
6051 static void test_reserve_memory_special_huge_tlbfs_only() {
6052 if (!UseHugeTLBFS) {
6053 return;
6054 }
6055
6056 size_t lp = os::large_page_size();
6057
6058 for (size_t size = lp; size <= lp * 10; size += lp) {
6059 test_reserve_memory_special_huge_tlbfs_only(size);
6060 }
6061 }
6062
6063 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6064 size_t lp = os::large_page_size();
6065 size_t ag = os::vm_allocation_granularity();
6066
6067 // sizes to test
6068 const size_t sizes[] = {
6069 lp, lp + ag, lp + lp / 2, lp * 2,
6070 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6071 lp * 10, lp * 10 + lp / 2
6072 };
6073 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6074
6075 // For each size/alignment combination, we test three scenarios:
6076 // 1) with req_addr == NULL
6077 // 2) with a non-null req_addr at which we expect to successfully allocate
6078 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6079 // expect the allocation to either fail or to ignore req_addr
6080
6081 // Pre-allocate two areas; they shall be as large as the largest allocation
6082 // and aligned to the largest alignment we will be testing.
6083 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6084 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6085 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6086 -1, 0);
6087 assert(mapping1 != MAP_FAILED, "should work");
6088
6089 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6090 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6091 -1, 0);
6092 assert(mapping2 != MAP_FAILED, "should work");
6093
6094 // Unmap the first mapping, but leave the second mapping intact: the first
6095 // mapping will serve as a value for a "good" req_addr (case 2). The second
6096 // mapping, still intact, as "bad" req_addr (case 3).
6097 ::munmap(mapping1, mapping_size);
6098
6099 // Case 1
6100 test_log("%s, req_addr NULL:", __FUNCTION__);
6101 test_log("size align result");
6102
6103 for (int i = 0; i < num_sizes; i++) {
6104 const size_t size = sizes[i];
6105 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6106 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6107 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6108 size, alignment, p, (p != NULL ? "" : "(failed)"));
6109 if (p != NULL) {
6110 assert(is_ptr_aligned(p, alignment), "must be");
6111 small_page_write(p, size);
6112 os::Linux::release_memory_special_huge_tlbfs(p, size);
6113 }
6114 }
6115 }
6116
6117 // Case 2
6118 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6119 test_log("size align req_addr result");
6120
6121 for (int i = 0; i < num_sizes; i++) {
6122 const size_t size = sizes[i];
6123 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6124 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6125 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6126 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6127 size, alignment, req_addr, p,
6128 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6129 if (p != NULL) {
6130 assert(p == req_addr, "must be");
6131 small_page_write(p, size);
6132 os::Linux::release_memory_special_huge_tlbfs(p, size);
6133 }
6134 }
6135 }
6136
6137 // Case 3
6138 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6139 test_log("size align req_addr result");
6140
6141 for (int i = 0; i < num_sizes; i++) {
6142 const size_t size = sizes[i];
6143 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6144 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6145 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6146 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6147 size, alignment, req_addr, p,
6148 ((p != NULL ? "" : "(failed)")));
6149 // as the area around req_addr contains already existing mappings, the API should always
6150 // return NULL (as per contract, it cannot return another address)
6151 assert(p == NULL, "must be");
6152 }
6153 }
6154
6155 ::munmap(mapping2, mapping_size);
6156
6157 }
6158
6159 static void test_reserve_memory_special_huge_tlbfs() {
6160 if (!UseHugeTLBFS) {
6161 return;
6162 }
6163
6164 test_reserve_memory_special_huge_tlbfs_only();
6165 test_reserve_memory_special_huge_tlbfs_mixed();
6166 }
6167
6168 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6169 if (!UseSHM) {
6170 return;
6171 }
6172
6173 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6174
6175 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6176
6177 if (addr != NULL) {
6178 assert(is_ptr_aligned(addr, alignment), "Check");
6179 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6180
6181 small_page_write(addr, size);
6182
6183 os::Linux::release_memory_special_shm(addr, size);
6184 }
6185 }
6186
6187 static void test_reserve_memory_special_shm() {
6188 size_t lp = os::large_page_size();
6189 size_t ag = os::vm_allocation_granularity();
6190
6191 for (size_t size = ag; size < lp * 3; size += ag) {
6192 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6193 test_reserve_memory_special_shm(size, alignment);
6194 }
6195 }
6196 }
6197
6198 static void test() {
6199 test_reserve_memory_special_huge_tlbfs();
6200 test_reserve_memory_special_shm();
6201 }
6202 };
6203
6204 void TestReserveMemorySpecial_test() {
6205 TestReserveMemorySpecial::test();
6206 }
6207
6208 #endif