1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/macros.hpp"
72 #include "utilities/vmError.hpp"
73
74 // put OS-includes here
75 # include <sys/types.h>
76 # include <sys/mman.h>
77 # include <sys/stat.h>
78 # include <sys/select.h>
79 # include <pthread.h>
80 # include <signal.h>
81 # include <errno.h>
82 # include <dlfcn.h>
83 # include <stdio.h>
84 # include <unistd.h>
85 # include <sys/resource.h>
86 # include <pthread.h>
87 # include <sys/stat.h>
88 # include <sys/time.h>
89 # include <sys/times.h>
90 # include <sys/utsname.h>
91 # include <sys/socket.h>
92 # include <sys/wait.h>
93 # include <pwd.h>
94 # include <poll.h>
95 # include <semaphore.h>
96 # include <fcntl.h>
97 # include <string.h>
98 # include <syscall.h>
99 # include <sys/sysinfo.h>
100 # include <gnu/libc-version.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
109
110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
111 // getrusage() is prepared to handle the associated failure.
112 #ifndef RUSAGE_THREAD
113 #define RUSAGE_THREAD (1) /* only the calling thread */
114 #endif
115
116 #define MAX_PATH (2 * K)
117
118 #define MAX_SECS 100000000
119
120 // for timer info max values which include all bits
121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123 #define LARGEPAGES_BIT (1 << 6)
124 ////////////////////////////////////////////////////////////////////////////////
125 // global variables
126 julong os::Linux::_physical_memory = 0;
127
128 address os::Linux::_initial_thread_stack_bottom = NULL;
129 uintptr_t os::Linux::_initial_thread_stack_size = 0;
130
131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 const int os::Linux::_vm_default_page_size = (8 * K);
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
143 pthread_condattr_t os::Linux::_condattr[1];
144
145 static jlong initial_time_count=0;
146
147 static int clock_tics_per_sec = 100;
148
149 // For diagnostics to print a message once. see run_periodic_checks
150 static sigset_t check_signal_done;
151 static bool check_signals = true;
152
153 static pid_t _initial_pid = 0;
154
155 // Signal number used to suspend/resume a thread
156
157 // do not use any signal number less than SIGSEGV, see 4355769
158 static int SR_signum = SIGUSR2;
159 sigset_t SR_sigset;
160
161 // Used to protect dlsym() calls
162 static pthread_mutex_t dl_mutex;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 // Return true if user is running as root.
188
189 bool os::have_special_privileges() {
190 static bool init = false;
191 static bool privileges = false;
192 if (!init) {
193 privileges = (getuid() != geteuid()) || (getgid() != getegid());
194 init = true;
195 }
196 return privileges;
197 }
198
199
200 #ifndef SYS_gettid
201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
202 #ifdef __ia64__
203 #define SYS_gettid 1105
204 #elif __i386__
205 #define SYS_gettid 224
206 #elif __amd64__
207 #define SYS_gettid 186
208 #elif __sparc__
209 #define SYS_gettid 143
210 #else
211 #error define gettid for the arch
212 #endif
213 #endif
214
215 // Cpu architecture string
216 #if defined(ZERO)
217 static char cpu_arch[] = ZERO_LIBARCH;
218 #elif defined(IA64)
219 static char cpu_arch[] = "ia64";
220 #elif defined(IA32)
221 static char cpu_arch[] = "i386";
222 #elif defined(AMD64)
223 static char cpu_arch[] = "amd64";
224 #elif defined(ARM)
225 static char cpu_arch[] = "arm";
226 #elif defined(PPC32)
227 static char cpu_arch[] = "ppc";
228 #elif defined(PPC64)
229 static char cpu_arch[] = "ppc64";
230 #elif defined(SPARC)
231 #ifdef _LP64
232 static char cpu_arch[] = "sparcv9";
233 #else
234 static char cpu_arch[] = "sparc";
235 #endif
236 #else
237 #error Add appropriate cpu_arch setting
238 #endif
239
240
241 // pid_t gettid()
242 //
243 // Returns the kernel thread id of the currently running thread. Kernel
244 // thread id is used to access /proc.
245 //
246 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
247 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
248 //
249 pid_t os::Linux::gettid() {
250 int rslt = syscall(SYS_gettid);
251 if (rslt == -1) {
252 // old kernel, no NPTL support
253 return getpid();
254 } else {
255 return (pid_t)rslt;
256 }
257 }
258
259 // Most versions of linux have a bug where the number of processors are
260 // determined by looking at the /proc file system. In a chroot environment,
261 // the system call returns 1. This causes the VM to act as if it is
262 // a single processor and elide locking (see is_MP() call).
263 static bool unsafe_chroot_detected = false;
264 static const char *unstable_chroot_error = "/proc file system not found.\n"
265 "Java may be unstable running multithreaded in a chroot "
266 "environment on Linux when /proc filesystem is not mounted.";
267
268 void os::Linux::initialize_system_info() {
269 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
270 if (processor_count() == 1) {
271 pid_t pid = os::Linux::gettid();
272 char fname[32];
273 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
274 FILE *fp = fopen(fname, "r");
275 if (fp == NULL) {
276 unsafe_chroot_detected = true;
277 } else {
278 fclose(fp);
279 }
280 }
281 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
282 assert(processor_count() > 0, "linux error");
283 }
284
285 void os::init_system_properties_values() {
286 // The next steps are taken in the product version:
287 //
288 // Obtain the JAVA_HOME value from the location of libjvm.so.
289 // This library should be located at:
290 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
291 //
292 // If "/jre/lib/" appears at the right place in the path, then we
293 // assume libjvm.so is installed in a JDK and we use this path.
294 //
295 // Otherwise exit with message: "Could not create the Java virtual machine."
296 //
297 // The following extra steps are taken in the debugging version:
298 //
299 // If "/jre/lib/" does NOT appear at the right place in the path
300 // instead of exit check for $JAVA_HOME environment variable.
301 //
302 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
303 // then we append a fake suffix "hotspot/libjvm.so" to this path so
304 // it looks like libjvm.so is installed there
305 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
306 //
307 // Otherwise exit.
308 //
309 // Important note: if the location of libjvm.so changes this
310 // code needs to be changed accordingly.
311
312 // See ld(1):
313 // The linker uses the following search paths to locate required
314 // shared libraries:
315 // 1: ...
316 // ...
317 // 7: The default directories, normally /lib and /usr/lib.
318 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
319 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
320 #else
321 #define DEFAULT_LIBPATH "/lib:/usr/lib"
322 #endif
323
324 // Base path of extensions installed on the system.
325 #define SYS_EXT_DIR "/usr/java/packages"
326 #define EXTENSIONS_DIR "/lib/ext"
327
328 // Buffer that fits several sprintfs.
329 // Note that the space for the colon and the trailing null are provided
330 // by the nulls included by the sizeof operator.
331 const size_t bufsize =
332 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
333 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
334 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
335
336 // sysclasspath, java_home, dll_dir
337 {
338 char *pslash;
339 os::jvm_path(buf, bufsize);
340
341 // Found the full path to libjvm.so.
342 // Now cut the path to <java_home>/jre if we can.
343 pslash = strrchr(buf, '/');
344 if (pslash != NULL) {
345 *pslash = '\0'; // Get rid of /libjvm.so.
346 }
347 pslash = strrchr(buf, '/');
348 if (pslash != NULL) {
349 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
350 }
351 Arguments::set_dll_dir(buf);
352
353 if (pslash != NULL) {
354 pslash = strrchr(buf, '/');
355 if (pslash != NULL) {
356 *pslash = '\0'; // Get rid of /<arch>.
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /lib.
360 }
361 }
362 }
363 Arguments::set_java_home(buf);
364 set_boot_path('/', ':');
365 }
366
367 // Where to look for native libraries.
368 //
369 // Note: Due to a legacy implementation, most of the library path
370 // is set in the launcher. This was to accomodate linking restrictions
371 // on legacy Linux implementations (which are no longer supported).
372 // Eventually, all the library path setting will be done here.
373 //
374 // However, to prevent the proliferation of improperly built native
375 // libraries, the new path component /usr/java/packages is added here.
376 // Eventually, all the library path setting will be done here.
377 {
378 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
379 // should always exist (until the legacy problem cited above is
380 // addressed).
381 const char *v = ::getenv("LD_LIBRARY_PATH");
382 const char *v_colon = ":";
383 if (v == NULL) { v = ""; v_colon = ""; }
384 // That's +1 for the colon and +1 for the trailing '\0'.
385 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
386 strlen(v) + 1 +
387 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
388 mtInternal);
389 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
390 Arguments::set_library_path(ld_library_path);
391 FREE_C_HEAP_ARRAY(char, ld_library_path);
392 }
393
394 // Extensions directories.
395 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
396 Arguments::set_ext_dirs(buf);
397
398 FREE_C_HEAP_ARRAY(char, buf);
399
400 #undef DEFAULT_LIBPATH
401 #undef SYS_EXT_DIR
402 #undef EXTENSIONS_DIR
403 }
404
405 ////////////////////////////////////////////////////////////////////////////////
406 // breakpoint support
407
408 void os::breakpoint() {
409 BREAKPOINT;
410 }
411
412 extern "C" void breakpoint() {
413 // use debugger to set breakpoint here
414 }
415
416 ////////////////////////////////////////////////////////////////////////////////
417 // signal support
418
419 debug_only(static bool signal_sets_initialized = false);
420 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
421
422 bool os::Linux::is_sig_ignored(int sig) {
423 struct sigaction oact;
424 sigaction(sig, (struct sigaction*)NULL, &oact);
425 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
426 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
427 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
428 return true;
429 } else {
430 return false;
431 }
432 }
433
434 void os::Linux::signal_sets_init() {
435 // Should also have an assertion stating we are still single-threaded.
436 assert(!signal_sets_initialized, "Already initialized");
437 // Fill in signals that are necessarily unblocked for all threads in
438 // the VM. Currently, we unblock the following signals:
439 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
440 // by -Xrs (=ReduceSignalUsage));
441 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
442 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
443 // the dispositions or masks wrt these signals.
444 // Programs embedding the VM that want to use the above signals for their
445 // own purposes must, at this time, use the "-Xrs" option to prevent
446 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
447 // (See bug 4345157, and other related bugs).
448 // In reality, though, unblocking these signals is really a nop, since
449 // these signals are not blocked by default.
450 sigemptyset(&unblocked_sigs);
451 sigemptyset(&allowdebug_blocked_sigs);
452 sigaddset(&unblocked_sigs, SIGILL);
453 sigaddset(&unblocked_sigs, SIGSEGV);
454 sigaddset(&unblocked_sigs, SIGBUS);
455 sigaddset(&unblocked_sigs, SIGFPE);
456 #if defined(PPC64)
457 sigaddset(&unblocked_sigs, SIGTRAP);
458 #endif
459 sigaddset(&unblocked_sigs, SR_signum);
460
461 if (!ReduceSignalUsage) {
462 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
464 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
468 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
469 }
470 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
471 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
472 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
473 }
474 }
475 // Fill in signals that are blocked by all but the VM thread.
476 sigemptyset(&vm_sigs);
477 if (!ReduceSignalUsage) {
478 sigaddset(&vm_sigs, BREAK_SIGNAL);
479 }
480 debug_only(signal_sets_initialized = true);
481
482 }
483
484 // These are signals that are unblocked while a thread is running Java.
485 // (For some reason, they get blocked by default.)
486 sigset_t* os::Linux::unblocked_signals() {
487 assert(signal_sets_initialized, "Not initialized");
488 return &unblocked_sigs;
489 }
490
491 // These are the signals that are blocked while a (non-VM) thread is
492 // running Java. Only the VM thread handles these signals.
493 sigset_t* os::Linux::vm_signals() {
494 assert(signal_sets_initialized, "Not initialized");
495 return &vm_sigs;
496 }
497
498 // These are signals that are blocked during cond_wait to allow debugger in
499 sigset_t* os::Linux::allowdebug_blocked_signals() {
500 assert(signal_sets_initialized, "Not initialized");
501 return &allowdebug_blocked_sigs;
502 }
503
504 void os::Linux::hotspot_sigmask(Thread* thread) {
505
506 //Save caller's signal mask before setting VM signal mask
507 sigset_t caller_sigmask;
508 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
509
510 OSThread* osthread = thread->osthread();
511 osthread->set_caller_sigmask(caller_sigmask);
512
513 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
514
515 if (!ReduceSignalUsage) {
516 if (thread->is_VM_thread()) {
517 // Only the VM thread handles BREAK_SIGNAL ...
518 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
519 } else {
520 // ... all other threads block BREAK_SIGNAL
521 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
522 }
523 }
524 }
525
526 //////////////////////////////////////////////////////////////////////////////
527 // detecting pthread library
528
529 void os::Linux::libpthread_init() {
530 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
531 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
532 // generic name for earlier versions.
533 // Define macros here so we can build HotSpot on old systems.
534 #ifndef _CS_GNU_LIBC_VERSION
535 #define _CS_GNU_LIBC_VERSION 2
536 #endif
537 #ifndef _CS_GNU_LIBPTHREAD_VERSION
538 #define _CS_GNU_LIBPTHREAD_VERSION 3
539 #endif
540
541 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
542 if (n > 0) {
543 char *str = (char *)malloc(n, mtInternal);
544 confstr(_CS_GNU_LIBC_VERSION, str, n);
545 os::Linux::set_glibc_version(str);
546 } else {
547 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
548 static char _gnu_libc_version[32];
549 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
550 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
551 os::Linux::set_glibc_version(_gnu_libc_version);
552 }
553
554 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
555 if (n > 0) {
556 char *str = (char *)malloc(n, mtInternal);
557 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
558 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
559 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
560 // is the case. LinuxThreads has a hard limit on max number of threads.
561 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
562 // On the other hand, NPTL does not have such a limit, sysconf()
563 // will return -1 and errno is not changed. Check if it is really NPTL.
564 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
565 strstr(str, "NPTL") &&
566 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
567 free(str);
568 os::Linux::set_libpthread_version("linuxthreads");
569 } else {
570 os::Linux::set_libpthread_version(str);
571 }
572 } else {
573 // glibc before 2.3.2 only has LinuxThreads.
574 os::Linux::set_libpthread_version("linuxthreads");
575 }
576
577 if (strstr(libpthread_version(), "NPTL")) {
578 os::Linux::set_is_NPTL();
579 } else {
580 os::Linux::set_is_LinuxThreads();
581 }
582
583 // LinuxThreads have two flavors: floating-stack mode, which allows variable
584 // stack size; and fixed-stack mode. NPTL is always floating-stack.
585 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
586 os::Linux::set_is_floating_stack();
587 }
588 }
589
590 /////////////////////////////////////////////////////////////////////////////
591 // thread stack
592
593 // Force Linux kernel to expand current thread stack. If "bottom" is close
594 // to the stack guard, caller should block all signals.
595 //
596 // MAP_GROWSDOWN:
597 // A special mmap() flag that is used to implement thread stacks. It tells
598 // kernel that the memory region should extend downwards when needed. This
599 // allows early versions of LinuxThreads to only mmap the first few pages
600 // when creating a new thread. Linux kernel will automatically expand thread
601 // stack as needed (on page faults).
602 //
603 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
604 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
605 // region, it's hard to tell if the fault is due to a legitimate stack
606 // access or because of reading/writing non-exist memory (e.g. buffer
607 // overrun). As a rule, if the fault happens below current stack pointer,
608 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
609 // application (see Linux kernel fault.c).
610 //
611 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
612 // stack overflow detection.
613 //
614 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
615 // not use this flag. However, the stack of initial thread is not created
616 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
617 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
618 // and then attach the thread to JVM.
619 //
620 // To get around the problem and allow stack banging on Linux, we need to
621 // manually expand thread stack after receiving the SIGSEGV.
622 //
623 // There are two ways to expand thread stack to address "bottom", we used
624 // both of them in JVM before 1.5:
625 // 1. adjust stack pointer first so that it is below "bottom", and then
626 // touch "bottom"
627 // 2. mmap() the page in question
628 //
629 // Now alternate signal stack is gone, it's harder to use 2. For instance,
630 // if current sp is already near the lower end of page 101, and we need to
631 // call mmap() to map page 100, it is possible that part of the mmap() frame
632 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
633 // That will destroy the mmap() frame and cause VM to crash.
634 //
635 // The following code works by adjusting sp first, then accessing the "bottom"
636 // page to force a page fault. Linux kernel will then automatically expand the
637 // stack mapping.
638 //
639 // _expand_stack_to() assumes its frame size is less than page size, which
640 // should always be true if the function is not inlined.
641
642 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
643 #define NOINLINE
644 #else
645 #define NOINLINE __attribute__ ((noinline))
646 #endif
647
648 static void _expand_stack_to(address bottom) NOINLINE;
649
650 static void _expand_stack_to(address bottom) {
651 address sp;
652 size_t size;
653 volatile char *p;
654
655 // Adjust bottom to point to the largest address within the same page, it
656 // gives us a one-page buffer if alloca() allocates slightly more memory.
657 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
658 bottom += os::Linux::page_size() - 1;
659
660 // sp might be slightly above current stack pointer; if that's the case, we
661 // will alloca() a little more space than necessary, which is OK. Don't use
662 // os::current_stack_pointer(), as its result can be slightly below current
663 // stack pointer, causing us to not alloca enough to reach "bottom".
664 sp = (address)&sp;
665
666 if (sp > bottom) {
667 size = sp - bottom;
668 p = (volatile char *)alloca(size);
669 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
670 p[0] = '\0';
671 }
672 }
673
674 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
675 assert(t!=NULL, "just checking");
676 assert(t->osthread()->expanding_stack(), "expand should be set");
677 assert(t->stack_base() != NULL, "stack_base was not initialized");
678
679 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
680 sigset_t mask_all, old_sigset;
681 sigfillset(&mask_all);
682 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
683 _expand_stack_to(addr);
684 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
685 return true;
686 }
687 return false;
688 }
689
690 //////////////////////////////////////////////////////////////////////////////
691 // create new thread
692
693 static address highest_vm_reserved_address();
694
695 // check if it's safe to start a new thread
696 static bool _thread_safety_check(Thread* thread) {
697 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
698 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
699 // Heap is mmap'ed at lower end of memory space. Thread stacks are
700 // allocated (MAP_FIXED) from high address space. Every thread stack
701 // occupies a fixed size slot (usually 2Mbytes, but user can change
702 // it to other values if they rebuild LinuxThreads).
703 //
704 // Problem with MAP_FIXED is that mmap() can still succeed even part of
705 // the memory region has already been mmap'ed. That means if we have too
706 // many threads and/or very large heap, eventually thread stack will
707 // collide with heap.
708 //
709 // Here we try to prevent heap/stack collision by comparing current
710 // stack bottom with the highest address that has been mmap'ed by JVM
711 // plus a safety margin for memory maps created by native code.
712 //
713 // This feature can be disabled by setting ThreadSafetyMargin to 0
714 //
715 if (ThreadSafetyMargin > 0) {
716 address stack_bottom = os::current_stack_base() - os::current_stack_size();
717
718 // not safe if our stack extends below the safety margin
719 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
720 } else {
721 return true;
722 }
723 } else {
724 // Floating stack LinuxThreads or NPTL:
725 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
726 // there's not enough space left, pthread_create() will fail. If we come
727 // here, that means enough space has been reserved for stack.
728 return true;
729 }
730 }
731
732 // Thread start routine for all newly created threads
733 static void *java_start(Thread *thread) {
734 // Try to randomize the cache line index of hot stack frames.
735 // This helps when threads of the same stack traces evict each other's
736 // cache lines. The threads can be either from the same JVM instance, or
737 // from different JVM instances. The benefit is especially true for
738 // processors with hyperthreading technology.
739 static int counter = 0;
740 int pid = os::current_process_id();
741 alloca(((pid ^ counter++) & 7) * 128);
742
743 ThreadLocalStorage::set_thread(thread);
744
745 OSThread* osthread = thread->osthread();
746 Monitor* sync = osthread->startThread_lock();
747
748 // non floating stack LinuxThreads needs extra check, see above
749 if (!_thread_safety_check(thread)) {
750 // notify parent thread
751 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
752 osthread->set_state(ZOMBIE);
753 sync->notify_all();
754 return NULL;
755 }
756
757 // thread_id is kernel thread id (similar to Solaris LWP id)
758 osthread->set_thread_id(os::Linux::gettid());
759
760 if (UseNUMA) {
761 int lgrp_id = os::numa_get_group_id();
762 if (lgrp_id != -1) {
763 thread->set_lgrp_id(lgrp_id);
764 }
765 }
766 // initialize signal mask for this thread
767 os::Linux::hotspot_sigmask(thread);
768
769 // initialize floating point control register
770 os::Linux::init_thread_fpu_state();
771
772 // handshaking with parent thread
773 {
774 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
775
776 // notify parent thread
777 osthread->set_state(INITIALIZED);
778 sync->notify_all();
779
780 // wait until os::start_thread()
781 while (osthread->get_state() == INITIALIZED) {
782 sync->wait(Mutex::_no_safepoint_check_flag);
783 }
784 }
785
786 // call one more level start routine
787 thread->run();
788
789 return 0;
790 }
791
792 bool os::create_thread(Thread* thread, ThreadType thr_type,
793 size_t stack_size) {
794 assert(thread->osthread() == NULL, "caller responsible");
795
796 // Allocate the OSThread object
797 OSThread* osthread = new OSThread(NULL, NULL);
798 if (osthread == NULL) {
799 return false;
800 }
801
802 // set the correct thread state
803 osthread->set_thread_type(thr_type);
804
805 // Initial state is ALLOCATED but not INITIALIZED
806 osthread->set_state(ALLOCATED);
807
808 thread->set_osthread(osthread);
809
810 // init thread attributes
811 pthread_attr_t attr;
812 pthread_attr_init(&attr);
813 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
814
815 // stack size
816 if (os::Linux::supports_variable_stack_size()) {
817 // calculate stack size if it's not specified by caller
818 if (stack_size == 0) {
819 stack_size = os::Linux::default_stack_size(thr_type);
820
821 switch (thr_type) {
822 case os::java_thread:
823 // Java threads use ThreadStackSize which default value can be
824 // changed with the flag -Xss
825 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
826 stack_size = JavaThread::stack_size_at_create();
827 break;
828 case os::compiler_thread:
829 if (CompilerThreadStackSize > 0) {
830 stack_size = (size_t)(CompilerThreadStackSize * K);
831 break;
832 } // else fall through:
833 // use VMThreadStackSize if CompilerThreadStackSize is not defined
834 case os::vm_thread:
835 case os::pgc_thread:
836 case os::cgc_thread:
837 case os::watcher_thread:
838 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
839 break;
840 }
841 }
842
843 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
844 pthread_attr_setstacksize(&attr, stack_size);
845 } else {
846 // let pthread_create() pick the default value.
847 }
848
849 // glibc guard page
850 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
851
852 ThreadState state;
853
854 {
855 // Serialize thread creation if we are running with fixed stack LinuxThreads
856 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
857 if (lock) {
858 os::Linux::createThread_lock()->lock_without_safepoint_check();
859 }
860
861 pthread_t tid;
862 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
863
864 pthread_attr_destroy(&attr);
865
866 if (ret != 0) {
867 if (PrintMiscellaneous && (Verbose || WizardMode)) {
868 perror("pthread_create()");
869 }
870 // Need to clean up stuff we've allocated so far
871 thread->set_osthread(NULL);
872 delete osthread;
873 if (lock) os::Linux::createThread_lock()->unlock();
874 return false;
875 }
876
877 // Store pthread info into the OSThread
878 osthread->set_pthread_id(tid);
879
880 // Wait until child thread is either initialized or aborted
881 {
882 Monitor* sync_with_child = osthread->startThread_lock();
883 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
884 while ((state = osthread->get_state()) == ALLOCATED) {
885 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
886 }
887 }
888
889 if (lock) {
890 os::Linux::createThread_lock()->unlock();
891 }
892 }
893
894 // Aborted due to thread limit being reached
895 if (state == ZOMBIE) {
896 thread->set_osthread(NULL);
897 delete osthread;
898 return false;
899 }
900
901 // The thread is returned suspended (in state INITIALIZED),
902 // and is started higher up in the call chain
903 assert(state == INITIALIZED, "race condition");
904 return true;
905 }
906
907 /////////////////////////////////////////////////////////////////////////////
908 // attach existing thread
909
910 // bootstrap the main thread
911 bool os::create_main_thread(JavaThread* thread) {
912 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
913 return create_attached_thread(thread);
914 }
915
916 bool os::create_attached_thread(JavaThread* thread) {
917 #ifdef ASSERT
918 thread->verify_not_published();
919 #endif
920
921 // Allocate the OSThread object
922 OSThread* osthread = new OSThread(NULL, NULL);
923
924 if (osthread == NULL) {
925 return false;
926 }
927
928 // Store pthread info into the OSThread
929 osthread->set_thread_id(os::Linux::gettid());
930 osthread->set_pthread_id(::pthread_self());
931
932 // initialize floating point control register
933 os::Linux::init_thread_fpu_state();
934
935 // Initial thread state is RUNNABLE
936 osthread->set_state(RUNNABLE);
937
938 thread->set_osthread(osthread);
939
940 if (UseNUMA) {
941 int lgrp_id = os::numa_get_group_id();
942 if (lgrp_id != -1) {
943 thread->set_lgrp_id(lgrp_id);
944 }
945 }
946
947 if (os::Linux::is_initial_thread()) {
948 // If current thread is initial thread, its stack is mapped on demand,
949 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
950 // the entire stack region to avoid SEGV in stack banging.
951 // It is also useful to get around the heap-stack-gap problem on SuSE
952 // kernel (see 4821821 for details). We first expand stack to the top
953 // of yellow zone, then enable stack yellow zone (order is significant,
954 // enabling yellow zone first will crash JVM on SuSE Linux), so there
955 // is no gap between the last two virtual memory regions.
956
957 JavaThread *jt = (JavaThread *)thread;
958 address addr = jt->stack_yellow_zone_base();
959 assert(addr != NULL, "initialization problem?");
960 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
961
962 osthread->set_expanding_stack();
963 os::Linux::manually_expand_stack(jt, addr);
964 osthread->clear_expanding_stack();
965 }
966
967 // initialize signal mask for this thread
968 // and save the caller's signal mask
969 os::Linux::hotspot_sigmask(thread);
970
971 return true;
972 }
973
974 void os::pd_start_thread(Thread* thread) {
975 OSThread * osthread = thread->osthread();
976 assert(osthread->get_state() != INITIALIZED, "just checking");
977 Monitor* sync_with_child = osthread->startThread_lock();
978 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
979 sync_with_child->notify();
980 }
981
982 // Free Linux resources related to the OSThread
983 void os::free_thread(OSThread* osthread) {
984 assert(osthread != NULL, "osthread not set");
985
986 if (Thread::current()->osthread() == osthread) {
987 // Restore caller's signal mask
988 sigset_t sigmask = osthread->caller_sigmask();
989 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
990 }
991
992 delete osthread;
993 }
994
995 //////////////////////////////////////////////////////////////////////////////
996 // thread local storage
997
998 // Restore the thread pointer if the destructor is called. This is in case
999 // someone from JNI code sets up a destructor with pthread_key_create to run
1000 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1001 // will hang or crash. When detachCurrentThread is called the key will be set
1002 // to null and we will not be called again. If detachCurrentThread is never
1003 // called we could loop forever depending on the pthread implementation.
1004 static void restore_thread_pointer(void* p) {
1005 Thread* thread = (Thread*) p;
1006 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1007 }
1008
1009 int os::allocate_thread_local_storage() {
1010 pthread_key_t key;
1011 int rslt = pthread_key_create(&key, restore_thread_pointer);
1012 assert(rslt == 0, "cannot allocate thread local storage");
1013 return (int)key;
1014 }
1015
1016 // Note: This is currently not used by VM, as we don't destroy TLS key
1017 // on VM exit.
1018 void os::free_thread_local_storage(int index) {
1019 int rslt = pthread_key_delete((pthread_key_t)index);
1020 assert(rslt == 0, "invalid index");
1021 }
1022
1023 void os::thread_local_storage_at_put(int index, void* value) {
1024 int rslt = pthread_setspecific((pthread_key_t)index, value);
1025 assert(rslt == 0, "pthread_setspecific failed");
1026 }
1027
1028 extern "C" Thread* get_thread() {
1029 return ThreadLocalStorage::thread();
1030 }
1031
1032 //////////////////////////////////////////////////////////////////////////////
1033 // initial thread
1034
1035 // Check if current thread is the initial thread, similar to Solaris thr_main.
1036 bool os::Linux::is_initial_thread(void) {
1037 char dummy;
1038 // If called before init complete, thread stack bottom will be null.
1039 // Can be called if fatal error occurs before initialization.
1040 if (initial_thread_stack_bottom() == NULL) return false;
1041 assert(initial_thread_stack_bottom() != NULL &&
1042 initial_thread_stack_size() != 0,
1043 "os::init did not locate initial thread's stack region");
1044 if ((address)&dummy >= initial_thread_stack_bottom() &&
1045 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1046 return true;
1047 } else {
1048 return false;
1049 }
1050 }
1051
1052 // Find the virtual memory area that contains addr
1053 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1054 FILE *fp = fopen("/proc/self/maps", "r");
1055 if (fp) {
1056 address low, high;
1057 while (!feof(fp)) {
1058 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1059 if (low <= addr && addr < high) {
1060 if (vma_low) *vma_low = low;
1061 if (vma_high) *vma_high = high;
1062 fclose(fp);
1063 return true;
1064 }
1065 }
1066 for (;;) {
1067 int ch = fgetc(fp);
1068 if (ch == EOF || ch == (int)'\n') break;
1069 }
1070 }
1071 fclose(fp);
1072 }
1073 return false;
1074 }
1075
1076 // Locate initial thread stack. This special handling of initial thread stack
1077 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1078 // bogus value for initial thread.
1079 void os::Linux::capture_initial_stack(size_t max_size) {
1080 // stack size is the easy part, get it from RLIMIT_STACK
1081 size_t stack_size;
1082 struct rlimit rlim;
1083 getrlimit(RLIMIT_STACK, &rlim);
1084 stack_size = rlim.rlim_cur;
1085
1086 // 6308388: a bug in ld.so will relocate its own .data section to the
1087 // lower end of primordial stack; reduce ulimit -s value a little bit
1088 // so we won't install guard page on ld.so's data section.
1089 stack_size -= 2 * page_size();
1090
1091 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1092 // 7.1, in both cases we will get 2G in return value.
1093 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1094 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1095 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1096 // in case other parts in glibc still assumes 2M max stack size.
1097 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1098 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1099 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1100 stack_size = 2 * K * K IA64_ONLY(*2);
1101 }
1102 // Try to figure out where the stack base (top) is. This is harder.
1103 //
1104 // When an application is started, glibc saves the initial stack pointer in
1105 // a global variable "__libc_stack_end", which is then used by system
1106 // libraries. __libc_stack_end should be pretty close to stack top. The
1107 // variable is available since the very early days. However, because it is
1108 // a private interface, it could disappear in the future.
1109 //
1110 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1111 // to __libc_stack_end, it is very close to stack top, but isn't the real
1112 // stack top. Note that /proc may not exist if VM is running as a chroot
1113 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1114 // /proc/<pid>/stat could change in the future (though unlikely).
1115 //
1116 // We try __libc_stack_end first. If that doesn't work, look for
1117 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1118 // as a hint, which should work well in most cases.
1119
1120 uintptr_t stack_start;
1121
1122 // try __libc_stack_end first
1123 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1124 if (p && *p) {
1125 stack_start = *p;
1126 } else {
1127 // see if we can get the start_stack field from /proc/self/stat
1128 FILE *fp;
1129 int pid;
1130 char state;
1131 int ppid;
1132 int pgrp;
1133 int session;
1134 int nr;
1135 int tpgrp;
1136 unsigned long flags;
1137 unsigned long minflt;
1138 unsigned long cminflt;
1139 unsigned long majflt;
1140 unsigned long cmajflt;
1141 unsigned long utime;
1142 unsigned long stime;
1143 long cutime;
1144 long cstime;
1145 long prio;
1146 long nice;
1147 long junk;
1148 long it_real;
1149 uintptr_t start;
1150 uintptr_t vsize;
1151 intptr_t rss;
1152 uintptr_t rsslim;
1153 uintptr_t scodes;
1154 uintptr_t ecode;
1155 int i;
1156
1157 // Figure what the primordial thread stack base is. Code is inspired
1158 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1159 // followed by command name surrounded by parentheses, state, etc.
1160 char stat[2048];
1161 int statlen;
1162
1163 fp = fopen("/proc/self/stat", "r");
1164 if (fp) {
1165 statlen = fread(stat, 1, 2047, fp);
1166 stat[statlen] = '\0';
1167 fclose(fp);
1168
1169 // Skip pid and the command string. Note that we could be dealing with
1170 // weird command names, e.g. user could decide to rename java launcher
1171 // to "java 1.4.2 :)", then the stat file would look like
1172 // 1234 (java 1.4.2 :)) R ... ...
1173 // We don't really need to know the command string, just find the last
1174 // occurrence of ")" and then start parsing from there. See bug 4726580.
1175 char * s = strrchr(stat, ')');
1176
1177 i = 0;
1178 if (s) {
1179 // Skip blank chars
1180 do { s++; } while (s && isspace(*s));
1181
1182 #define _UFM UINTX_FORMAT
1183 #define _DFM INTX_FORMAT
1184
1185 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1186 // 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
1187 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,
1188 &state, // 3 %c
1189 &ppid, // 4 %d
1190 &pgrp, // 5 %d
1191 &session, // 6 %d
1192 &nr, // 7 %d
1193 &tpgrp, // 8 %d
1194 &flags, // 9 %lu
1195 &minflt, // 10 %lu
1196 &cminflt, // 11 %lu
1197 &majflt, // 12 %lu
1198 &cmajflt, // 13 %lu
1199 &utime, // 14 %lu
1200 &stime, // 15 %lu
1201 &cutime, // 16 %ld
1202 &cstime, // 17 %ld
1203 &prio, // 18 %ld
1204 &nice, // 19 %ld
1205 &junk, // 20 %ld
1206 &it_real, // 21 %ld
1207 &start, // 22 UINTX_FORMAT
1208 &vsize, // 23 UINTX_FORMAT
1209 &rss, // 24 INTX_FORMAT
1210 &rsslim, // 25 UINTX_FORMAT
1211 &scodes, // 26 UINTX_FORMAT
1212 &ecode, // 27 UINTX_FORMAT
1213 &stack_start); // 28 UINTX_FORMAT
1214 }
1215
1216 #undef _UFM
1217 #undef _DFM
1218
1219 if (i != 28 - 2) {
1220 assert(false, "Bad conversion from /proc/self/stat");
1221 // product mode - assume we are the initial thread, good luck in the
1222 // embedded case.
1223 warning("Can't detect initial thread stack location - bad conversion");
1224 stack_start = (uintptr_t) &rlim;
1225 }
1226 } else {
1227 // For some reason we can't open /proc/self/stat (for example, running on
1228 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1229 // most cases, so don't abort:
1230 warning("Can't detect initial thread stack location - no /proc/self/stat");
1231 stack_start = (uintptr_t) &rlim;
1232 }
1233 }
1234
1235 // Now we have a pointer (stack_start) very close to the stack top, the
1236 // next thing to do is to figure out the exact location of stack top. We
1237 // can find out the virtual memory area that contains stack_start by
1238 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1239 // and its upper limit is the real stack top. (again, this would fail if
1240 // running inside chroot, because /proc may not exist.)
1241
1242 uintptr_t stack_top;
1243 address low, high;
1244 if (find_vma((address)stack_start, &low, &high)) {
1245 // success, "high" is the true stack top. (ignore "low", because initial
1246 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1247 stack_top = (uintptr_t)high;
1248 } else {
1249 // failed, likely because /proc/self/maps does not exist
1250 warning("Can't detect initial thread stack location - find_vma failed");
1251 // best effort: stack_start is normally within a few pages below the real
1252 // stack top, use it as stack top, and reduce stack size so we won't put
1253 // guard page outside stack.
1254 stack_top = stack_start;
1255 stack_size -= 16 * page_size();
1256 }
1257
1258 // stack_top could be partially down the page so align it
1259 stack_top = align_size_up(stack_top, page_size());
1260
1261 if (max_size && stack_size > max_size) {
1262 _initial_thread_stack_size = max_size;
1263 } else {
1264 _initial_thread_stack_size = stack_size;
1265 }
1266
1267 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1268 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1269 }
1270
1271 ////////////////////////////////////////////////////////////////////////////////
1272 // time support
1273
1274 // Time since start-up in seconds to a fine granularity.
1275 // Used by VMSelfDestructTimer and the MemProfiler.
1276 double os::elapsedTime() {
1277
1278 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1279 }
1280
1281 jlong os::elapsed_counter() {
1282 return javaTimeNanos() - initial_time_count;
1283 }
1284
1285 jlong os::elapsed_frequency() {
1286 return NANOSECS_PER_SEC; // nanosecond resolution
1287 }
1288
1289 bool os::supports_vtime() { return true; }
1290 bool os::enable_vtime() { return false; }
1291 bool os::vtime_enabled() { return false; }
1292
1293 double os::elapsedVTime() {
1294 struct rusage usage;
1295 int retval = getrusage(RUSAGE_THREAD, &usage);
1296 if (retval == 0) {
1297 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);
1298 } else {
1299 // better than nothing, but not much
1300 return elapsedTime();
1301 }
1302 }
1303
1304 jlong os::javaTimeMillis() {
1305 timeval time;
1306 int status = gettimeofday(&time, NULL);
1307 assert(status != -1, "linux error");
1308 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1309 }
1310
1311 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1312 timeval time;
1313 int status = gettimeofday(&time, NULL);
1314 assert(status != -1, "linux error");
1315 seconds = jlong(time.tv_sec);
1316 nanos = jlong(time.tv_usec) * 1000;
1317 }
1318
1319
1320 #ifndef CLOCK_MONOTONIC
1321 #define CLOCK_MONOTONIC (1)
1322 #endif
1323
1324 void os::Linux::clock_init() {
1325 // we do dlopen's in this particular order due to bug in linux
1326 // dynamical loader (see 6348968) leading to crash on exit
1327 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1328 if (handle == NULL) {
1329 handle = dlopen("librt.so", RTLD_LAZY);
1330 }
1331
1332 if (handle) {
1333 int (*clock_getres_func)(clockid_t, struct timespec*) =
1334 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1335 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1336 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1337 if (clock_getres_func && clock_gettime_func) {
1338 // See if monotonic clock is supported by the kernel. Note that some
1339 // early implementations simply return kernel jiffies (updated every
1340 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1341 // for nano time (though the monotonic property is still nice to have).
1342 // It's fixed in newer kernels, however clock_getres() still returns
1343 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1344 // resolution for now. Hopefully as people move to new kernels, this
1345 // won't be a problem.
1346 struct timespec res;
1347 struct timespec tp;
1348 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1349 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1350 // yes, monotonic clock is supported
1351 _clock_gettime = clock_gettime_func;
1352 return;
1353 } else {
1354 // close librt if there is no monotonic clock
1355 dlclose(handle);
1356 }
1357 }
1358 }
1359 warning("No monotonic clock was available - timed services may " \
1360 "be adversely affected if the time-of-day clock changes");
1361 }
1362
1363 #ifndef SYS_clock_getres
1364 #if defined(IA32) || defined(AMD64)
1365 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1366 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1367 #else
1368 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1369 #define sys_clock_getres(x,y) -1
1370 #endif
1371 #else
1372 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1373 #endif
1374
1375 void os::Linux::fast_thread_clock_init() {
1376 if (!UseLinuxPosixThreadCPUClocks) {
1377 return;
1378 }
1379 clockid_t clockid;
1380 struct timespec tp;
1381 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1382 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1383
1384 // Switch to using fast clocks for thread cpu time if
1385 // the sys_clock_getres() returns 0 error code.
1386 // Note, that some kernels may support the current thread
1387 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1388 // returned by the pthread_getcpuclockid().
1389 // If the fast Posix clocks are supported then the sys_clock_getres()
1390 // must return at least tp.tv_sec == 0 which means a resolution
1391 // better than 1 sec. This is extra check for reliability.
1392
1393 if (pthread_getcpuclockid_func &&
1394 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1395 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1396 _supports_fast_thread_cpu_time = true;
1397 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1398 }
1399 }
1400
1401 jlong os::javaTimeNanos() {
1402 if (os::supports_monotonic_clock()) {
1403 struct timespec tp;
1404 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1405 assert(status == 0, "gettime error");
1406 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1407 return result;
1408 } else {
1409 timeval time;
1410 int status = gettimeofday(&time, NULL);
1411 assert(status != -1, "linux error");
1412 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1413 return 1000 * usecs;
1414 }
1415 }
1416
1417 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1418 if (os::supports_monotonic_clock()) {
1419 info_ptr->max_value = ALL_64_BITS;
1420
1421 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1422 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1423 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1424 } else {
1425 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1426 info_ptr->max_value = ALL_64_BITS;
1427
1428 // gettimeofday is a real time clock so it skips
1429 info_ptr->may_skip_backward = true;
1430 info_ptr->may_skip_forward = true;
1431 }
1432
1433 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1434 }
1435
1436 // Return the real, user, and system times in seconds from an
1437 // arbitrary fixed point in the past.
1438 bool os::getTimesSecs(double* process_real_time,
1439 double* process_user_time,
1440 double* process_system_time) {
1441 struct tms ticks;
1442 clock_t real_ticks = times(&ticks);
1443
1444 if (real_ticks == (clock_t) (-1)) {
1445 return false;
1446 } else {
1447 double ticks_per_second = (double) clock_tics_per_sec;
1448 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1449 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1450 *process_real_time = ((double) real_ticks) / ticks_per_second;
1451
1452 return true;
1453 }
1454 }
1455
1456
1457 char * os::local_time_string(char *buf, size_t buflen) {
1458 struct tm t;
1459 time_t long_time;
1460 time(&long_time);
1461 localtime_r(&long_time, &t);
1462 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1463 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1464 t.tm_hour, t.tm_min, t.tm_sec);
1465 return buf;
1466 }
1467
1468 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1469 return localtime_r(clock, res);
1470 }
1471
1472 ////////////////////////////////////////////////////////////////////////////////
1473 // runtime exit support
1474
1475 // Note: os::shutdown() might be called very early during initialization, or
1476 // called from signal handler. Before adding something to os::shutdown(), make
1477 // sure it is async-safe and can handle partially initialized VM.
1478 void os::shutdown() {
1479
1480 // allow PerfMemory to attempt cleanup of any persistent resources
1481 perfMemory_exit();
1482
1483 // needs to remove object in file system
1484 AttachListener::abort();
1485
1486 // flush buffered output, finish log files
1487 ostream_abort();
1488
1489 // Check for abort hook
1490 abort_hook_t abort_hook = Arguments::abort_hook();
1491 if (abort_hook != NULL) {
1492 abort_hook();
1493 }
1494
1495 }
1496
1497 // Note: os::abort() might be called very early during initialization, or
1498 // called from signal handler. Before adding something to os::abort(), make
1499 // sure it is async-safe and can handle partially initialized VM.
1500 void os::abort(bool dump_core) {
1501 os::shutdown();
1502 if (dump_core) {
1503 #ifndef PRODUCT
1504 fdStream out(defaultStream::output_fd());
1505 out.print_raw("Current thread is ");
1506 char buf[16];
1507 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1508 out.print_raw_cr(buf);
1509 out.print_raw_cr("Dumping core ...");
1510 #endif
1511 ::abort(); // dump core
1512 }
1513
1514 ::exit(1);
1515 }
1516
1517 // Die immediately, no exit hook, no abort hook, no cleanup.
1518 void os::die() {
1519 // _exit() on LinuxThreads only kills current thread
1520 ::abort();
1521 }
1522
1523
1524 // This method is a copy of JDK's sysGetLastErrorString
1525 // from src/solaris/hpi/src/system_md.c
1526
1527 size_t os::lasterror(char *buf, size_t len) {
1528 if (errno == 0) return 0;
1529
1530 const char *s = ::strerror(errno);
1531 size_t n = ::strlen(s);
1532 if (n >= len) {
1533 n = len - 1;
1534 }
1535 ::strncpy(buf, s, n);
1536 buf[n] = '\0';
1537 return n;
1538 }
1539
1540 intx os::current_thread_id() { return (intx)pthread_self(); }
1541 int os::current_process_id() {
1542
1543 // Under the old linux thread library, linux gives each thread
1544 // its own process id. Because of this each thread will return
1545 // a different pid if this method were to return the result
1546 // of getpid(2). Linux provides no api that returns the pid
1547 // of the launcher thread for the vm. This implementation
1548 // returns a unique pid, the pid of the launcher thread
1549 // that starts the vm 'process'.
1550
1551 // Under the NPTL, getpid() returns the same pid as the
1552 // launcher thread rather than a unique pid per thread.
1553 // Use gettid() if you want the old pre NPTL behaviour.
1554
1555 // if you are looking for the result of a call to getpid() that
1556 // returns a unique pid for the calling thread, then look at the
1557 // OSThread::thread_id() method in osThread_linux.hpp file
1558
1559 return (int)(_initial_pid ? _initial_pid : getpid());
1560 }
1561
1562 // DLL functions
1563
1564 const char* os::dll_file_extension() { return ".so"; }
1565
1566 // This must be hard coded because it's the system's temporary
1567 // directory not the java application's temp directory, ala java.io.tmpdir.
1568 const char* os::get_temp_directory() { return "/tmp"; }
1569
1570 static bool file_exists(const char* filename) {
1571 struct stat statbuf;
1572 if (filename == NULL || strlen(filename) == 0) {
1573 return false;
1574 }
1575 return os::stat(filename, &statbuf) == 0;
1576 }
1577
1578 bool os::dll_build_name(char* buffer, size_t buflen,
1579 const char* pname, const char* fname) {
1580 bool retval = false;
1581 // Copied from libhpi
1582 const size_t pnamelen = pname ? strlen(pname) : 0;
1583
1584 // Return error on buffer overflow.
1585 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1586 return retval;
1587 }
1588
1589 if (pnamelen == 0) {
1590 snprintf(buffer, buflen, "lib%s.so", fname);
1591 retval = true;
1592 } else if (strchr(pname, *os::path_separator()) != NULL) {
1593 int n;
1594 char** pelements = split_path(pname, &n);
1595 if (pelements == NULL) {
1596 return false;
1597 }
1598 for (int i = 0; i < n; i++) {
1599 // Really shouldn't be NULL, but check can't hurt
1600 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1601 continue; // skip the empty path values
1602 }
1603 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1604 if (file_exists(buffer)) {
1605 retval = true;
1606 break;
1607 }
1608 }
1609 // release the storage
1610 for (int i = 0; i < n; i++) {
1611 if (pelements[i] != NULL) {
1612 FREE_C_HEAP_ARRAY(char, pelements[i]);
1613 }
1614 }
1615 if (pelements != NULL) {
1616 FREE_C_HEAP_ARRAY(char*, pelements);
1617 }
1618 } else {
1619 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1620 retval = true;
1621 }
1622 return retval;
1623 }
1624
1625 // check if addr is inside libjvm.so
1626 bool os::address_is_in_vm(address addr) {
1627 static address libjvm_base_addr;
1628 Dl_info dlinfo;
1629
1630 if (libjvm_base_addr == NULL) {
1631 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1632 libjvm_base_addr = (address)dlinfo.dli_fbase;
1633 }
1634 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1635 }
1636
1637 if (dladdr((void *)addr, &dlinfo) != 0) {
1638 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1639 }
1640
1641 return false;
1642 }
1643
1644 bool os::dll_address_to_function_name(address addr, char *buf,
1645 int buflen, int *offset) {
1646 // buf is not optional, but offset is optional
1647 assert(buf != NULL, "sanity check");
1648
1649 Dl_info dlinfo;
1650
1651 if (dladdr((void*)addr, &dlinfo) != 0) {
1652 // see if we have a matching symbol
1653 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1654 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1655 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1656 }
1657 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1658 return true;
1659 }
1660 // no matching symbol so try for just file info
1661 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1662 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1663 buf, buflen, offset, dlinfo.dli_fname)) {
1664 return true;
1665 }
1666 }
1667 }
1668
1669 buf[0] = '\0';
1670 if (offset != NULL) *offset = -1;
1671 return false;
1672 }
1673
1674 struct _address_to_library_name {
1675 address addr; // input : memory address
1676 size_t buflen; // size of fname
1677 char* fname; // output: library name
1678 address base; // library base addr
1679 };
1680
1681 static int address_to_library_name_callback(struct dl_phdr_info *info,
1682 size_t size, void *data) {
1683 int i;
1684 bool found = false;
1685 address libbase = NULL;
1686 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1687
1688 // iterate through all loadable segments
1689 for (i = 0; i < info->dlpi_phnum; i++) {
1690 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1691 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1692 // base address of a library is the lowest address of its loaded
1693 // segments.
1694 if (libbase == NULL || libbase > segbase) {
1695 libbase = segbase;
1696 }
1697 // see if 'addr' is within current segment
1698 if (segbase <= d->addr &&
1699 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1700 found = true;
1701 }
1702 }
1703 }
1704
1705 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1706 // so dll_address_to_library_name() can fall through to use dladdr() which
1707 // can figure out executable name from argv[0].
1708 if (found && info->dlpi_name && info->dlpi_name[0]) {
1709 d->base = libbase;
1710 if (d->fname) {
1711 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1712 }
1713 return 1;
1714 }
1715 return 0;
1716 }
1717
1718 bool os::dll_address_to_library_name(address addr, char* buf,
1719 int buflen, int* offset) {
1720 // buf is not optional, but offset is optional
1721 assert(buf != NULL, "sanity check");
1722
1723 Dl_info dlinfo;
1724 struct _address_to_library_name data;
1725
1726 // There is a bug in old glibc dladdr() implementation that it could resolve
1727 // to wrong library name if the .so file has a base address != NULL. Here
1728 // we iterate through the program headers of all loaded libraries to find
1729 // out which library 'addr' really belongs to. This workaround can be
1730 // removed once the minimum requirement for glibc is moved to 2.3.x.
1731 data.addr = addr;
1732 data.fname = buf;
1733 data.buflen = buflen;
1734 data.base = NULL;
1735 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1736
1737 if (rslt) {
1738 // buf already contains library name
1739 if (offset) *offset = addr - data.base;
1740 return true;
1741 }
1742 if (dladdr((void*)addr, &dlinfo) != 0) {
1743 if (dlinfo.dli_fname != NULL) {
1744 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1745 }
1746 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1747 *offset = addr - (address)dlinfo.dli_fbase;
1748 }
1749 return true;
1750 }
1751
1752 buf[0] = '\0';
1753 if (offset) *offset = -1;
1754 return false;
1755 }
1756
1757 // Loads .dll/.so and
1758 // in case of error it checks if .dll/.so was built for the
1759 // same architecture as Hotspot is running on
1760
1761
1762 // Remember the stack's state. The Linux dynamic linker will change
1763 // the stack to 'executable' at most once, so we must safepoint only once.
1764 bool os::Linux::_stack_is_executable = false;
1765
1766 // VM operation that loads a library. This is necessary if stack protection
1767 // of the Java stacks can be lost during loading the library. If we
1768 // do not stop the Java threads, they can stack overflow before the stacks
1769 // are protected again.
1770 class VM_LinuxDllLoad: public VM_Operation {
1771 private:
1772 const char *_filename;
1773 char *_ebuf;
1774 int _ebuflen;
1775 void *_lib;
1776 public:
1777 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1778 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1779 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1780 void doit() {
1781 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1782 os::Linux::_stack_is_executable = true;
1783 }
1784 void* loaded_library() { return _lib; }
1785 };
1786
1787 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1788 void * result = NULL;
1789 bool load_attempted = false;
1790
1791 // Check whether the library to load might change execution rights
1792 // of the stack. If they are changed, the protection of the stack
1793 // guard pages will be lost. We need a safepoint to fix this.
1794 //
1795 // See Linux man page execstack(8) for more info.
1796 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1797 ElfFile ef(filename);
1798 if (!ef.specifies_noexecstack()) {
1799 if (!is_init_completed()) {
1800 os::Linux::_stack_is_executable = true;
1801 // This is OK - No Java threads have been created yet, and hence no
1802 // stack guard pages to fix.
1803 //
1804 // This should happen only when you are building JDK7 using a very
1805 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1806 //
1807 // Dynamic loader will make all stacks executable after
1808 // this function returns, and will not do that again.
1809 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1810 } else {
1811 warning("You have loaded library %s which might have disabled stack guard. "
1812 "The VM will try to fix the stack guard now.\n"
1813 "It's highly recommended that you fix the library with "
1814 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1815 filename);
1816
1817 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1818 JavaThread *jt = JavaThread::current();
1819 if (jt->thread_state() != _thread_in_native) {
1820 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1821 // that requires ExecStack. Cannot enter safe point. Let's give up.
1822 warning("Unable to fix stack guard. Giving up.");
1823 } else {
1824 if (!LoadExecStackDllInVMThread) {
1825 // This is for the case where the DLL has an static
1826 // constructor function that executes JNI code. We cannot
1827 // load such DLLs in the VMThread.
1828 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1829 }
1830
1831 ThreadInVMfromNative tiv(jt);
1832 debug_only(VMNativeEntryWrapper vew;)
1833
1834 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1835 VMThread::execute(&op);
1836 if (LoadExecStackDllInVMThread) {
1837 result = op.loaded_library();
1838 }
1839 load_attempted = true;
1840 }
1841 }
1842 }
1843 }
1844
1845 if (!load_attempted) {
1846 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1847 }
1848
1849 if (result != NULL) {
1850 // Successful loading
1851 return result;
1852 }
1853
1854 Elf32_Ehdr elf_head;
1855 int diag_msg_max_length=ebuflen-strlen(ebuf);
1856 char* diag_msg_buf=ebuf+strlen(ebuf);
1857
1858 if (diag_msg_max_length==0) {
1859 // No more space in ebuf for additional diagnostics message
1860 return NULL;
1861 }
1862
1863
1864 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1865
1866 if (file_descriptor < 0) {
1867 // Can't open library, report dlerror() message
1868 return NULL;
1869 }
1870
1871 bool failed_to_read_elf_head=
1872 (sizeof(elf_head)!=
1873 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1874
1875 ::close(file_descriptor);
1876 if (failed_to_read_elf_head) {
1877 // file i/o error - report dlerror() msg
1878 return NULL;
1879 }
1880
1881 typedef struct {
1882 Elf32_Half code; // Actual value as defined in elf.h
1883 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1884 char elf_class; // 32 or 64 bit
1885 char endianess; // MSB or LSB
1886 char* name; // String representation
1887 } arch_t;
1888
1889 #ifndef EM_486
1890 #define EM_486 6 /* Intel 80486 */
1891 #endif
1892
1893 static const arch_t arch_array[]={
1894 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1895 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1896 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1897 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1898 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1899 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1900 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1901 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1902 #if defined(VM_LITTLE_ENDIAN)
1903 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1904 #else
1905 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1906 #endif
1907 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1908 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1909 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1910 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1911 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1912 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1913 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1914 };
1915
1916 #if (defined IA32)
1917 static Elf32_Half running_arch_code=EM_386;
1918 #elif (defined AMD64)
1919 static Elf32_Half running_arch_code=EM_X86_64;
1920 #elif (defined IA64)
1921 static Elf32_Half running_arch_code=EM_IA_64;
1922 #elif (defined __sparc) && (defined _LP64)
1923 static Elf32_Half running_arch_code=EM_SPARCV9;
1924 #elif (defined __sparc) && (!defined _LP64)
1925 static Elf32_Half running_arch_code=EM_SPARC;
1926 #elif (defined __powerpc64__)
1927 static Elf32_Half running_arch_code=EM_PPC64;
1928 #elif (defined __powerpc__)
1929 static Elf32_Half running_arch_code=EM_PPC;
1930 #elif (defined ARM)
1931 static Elf32_Half running_arch_code=EM_ARM;
1932 #elif (defined S390)
1933 static Elf32_Half running_arch_code=EM_S390;
1934 #elif (defined ALPHA)
1935 static Elf32_Half running_arch_code=EM_ALPHA;
1936 #elif (defined MIPSEL)
1937 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1938 #elif (defined PARISC)
1939 static Elf32_Half running_arch_code=EM_PARISC;
1940 #elif (defined MIPS)
1941 static Elf32_Half running_arch_code=EM_MIPS;
1942 #elif (defined M68K)
1943 static Elf32_Half running_arch_code=EM_68K;
1944 #else
1945 #error Method os::dll_load requires that one of following is defined:\
1946 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1947 #endif
1948
1949 // Identify compatability class for VM's architecture and library's architecture
1950 // Obtain string descriptions for architectures
1951
1952 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1953 int running_arch_index=-1;
1954
1955 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1956 if (running_arch_code == arch_array[i].code) {
1957 running_arch_index = i;
1958 }
1959 if (lib_arch.code == arch_array[i].code) {
1960 lib_arch.compat_class = arch_array[i].compat_class;
1961 lib_arch.name = arch_array[i].name;
1962 }
1963 }
1964
1965 assert(running_arch_index != -1,
1966 "Didn't find running architecture code (running_arch_code) in arch_array");
1967 if (running_arch_index == -1) {
1968 // Even though running architecture detection failed
1969 // we may still continue with reporting dlerror() message
1970 return NULL;
1971 }
1972
1973 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1974 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1975 return NULL;
1976 }
1977
1978 #ifndef S390
1979 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1980 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1981 return NULL;
1982 }
1983 #endif // !S390
1984
1985 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1986 if (lib_arch.name!=NULL) {
1987 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1988 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1989 lib_arch.name, arch_array[running_arch_index].name);
1990 } else {
1991 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1992 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1993 lib_arch.code,
1994 arch_array[running_arch_index].name);
1995 }
1996 }
1997
1998 return NULL;
1999 }
2000
2001 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2002 int ebuflen) {
2003 void * result = ::dlopen(filename, RTLD_LAZY);
2004 if (result == NULL) {
2005 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2006 ebuf[ebuflen-1] = '\0';
2007 }
2008 return result;
2009 }
2010
2011 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2012 int ebuflen) {
2013 void * result = NULL;
2014 if (LoadExecStackDllInVMThread) {
2015 result = dlopen_helper(filename, ebuf, ebuflen);
2016 }
2017
2018 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2019 // library that requires an executable stack, or which does not have this
2020 // stack attribute set, dlopen changes the stack attribute to executable. The
2021 // read protection of the guard pages gets lost.
2022 //
2023 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2024 // may have been queued at the same time.
2025
2026 if (!_stack_is_executable) {
2027 JavaThread *jt = Threads::first();
2028
2029 while (jt) {
2030 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2031 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2032 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2033 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2034 warning("Attempt to reguard stack yellow zone failed.");
2035 }
2036 }
2037 jt = jt->next();
2038 }
2039 }
2040
2041 return result;
2042 }
2043
2044 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2045 // chances are you might want to run the generated bits against glibc-2.0
2046 // libdl.so, so always use locking for any version of glibc.
2047 //
2048 void* os::dll_lookup(void* handle, const char* name) {
2049 pthread_mutex_lock(&dl_mutex);
2050 void* res = dlsym(handle, name);
2051 pthread_mutex_unlock(&dl_mutex);
2052 return res;
2053 }
2054
2055 void* os::get_default_process_handle() {
2056 return (void*)::dlopen(NULL, RTLD_LAZY);
2057 }
2058
2059 static bool _print_ascii_file(const char* filename, outputStream* st) {
2060 int fd = ::open(filename, O_RDONLY);
2061 if (fd == -1) {
2062 return false;
2063 }
2064
2065 char buf[32];
2066 int bytes;
2067 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2068 st->print_raw(buf, bytes);
2069 }
2070
2071 ::close(fd);
2072
2073 return true;
2074 }
2075
2076 void os::print_dll_info(outputStream *st) {
2077 st->print_cr("Dynamic libraries:");
2078
2079 char fname[32];
2080 pid_t pid = os::Linux::gettid();
2081
2082 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2083
2084 if (!_print_ascii_file(fname, st)) {
2085 st->print("Can not get library information for pid = %d\n", pid);
2086 }
2087 }
2088
2089 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2090 FILE *procmapsFile = NULL;
2091
2092 // Open the procfs maps file for the current process
2093 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2094 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2095 char line[PATH_MAX + 100];
2096
2097 // Read line by line from 'file'
2098 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2099 u8 base, top, offset, inode;
2100 char permissions[5];
2101 char device[6];
2102 char name[PATH_MAX + 1];
2103
2104 // Parse fields from line
2105 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2106
2107 // Filter by device id '00:00' so that we only get file system mapped files.
2108 if (strcmp(device, "00:00") != 0) {
2109
2110 // Call callback with the fields of interest
2111 if(callback(name, (address)base, (address)top, param)) {
2112 // Oops abort, callback aborted
2113 fclose(procmapsFile);
2114 return 1;
2115 }
2116 }
2117 }
2118 fclose(procmapsFile);
2119 }
2120 return 0;
2121 }
2122
2123 void os::print_os_info_brief(outputStream* st) {
2124 os::Linux::print_distro_info(st);
2125
2126 os::Posix::print_uname_info(st);
2127
2128 os::Linux::print_libversion_info(st);
2129
2130 }
2131
2132 void os::print_os_info(outputStream* st) {
2133 st->print("OS:");
2134
2135 os::Linux::print_distro_info(st);
2136
2137 os::Posix::print_uname_info(st);
2138
2139 // Print warning if unsafe chroot environment detected
2140 if (unsafe_chroot_detected) {
2141 st->print("WARNING!! ");
2142 st->print_cr("%s", unstable_chroot_error);
2143 }
2144
2145 os::Linux::print_libversion_info(st);
2146
2147 os::Posix::print_rlimit_info(st);
2148
2149 os::Posix::print_load_average(st);
2150
2151 os::Linux::print_full_memory_info(st);
2152 }
2153
2154 // Try to identify popular distros.
2155 // Most Linux distributions have a /etc/XXX-release file, which contains
2156 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2157 // file that also contains the OS version string. Some have more than one
2158 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2159 // /etc/redhat-release.), so the order is important.
2160 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2161 // their own specific XXX-release file as well as a redhat-release file.
2162 // Because of this the XXX-release file needs to be searched for before the
2163 // redhat-release file.
2164 // Since Red Hat has a lsb-release file that is not very descriptive the
2165 // search for redhat-release needs to be before lsb-release.
2166 // Since the lsb-release file is the new standard it needs to be searched
2167 // before the older style release files.
2168 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2169 // next to last resort. The os-release file is a new standard that contains
2170 // distribution information and the system-release file seems to be an old
2171 // standard that has been replaced by the lsb-release and os-release files.
2172 // Searching for the debian_version file is the last resort. It contains
2173 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2174 // "Debian " is printed before the contents of the debian_version file.
2175 void os::Linux::print_distro_info(outputStream* st) {
2176 if (!_print_ascii_file("/etc/oracle-release", st) &&
2177 !_print_ascii_file("/etc/mandriva-release", st) &&
2178 !_print_ascii_file("/etc/mandrake-release", st) &&
2179 !_print_ascii_file("/etc/sun-release", st) &&
2180 !_print_ascii_file("/etc/redhat-release", st) &&
2181 !_print_ascii_file("/etc/lsb-release", st) &&
2182 !_print_ascii_file("/etc/SuSE-release", st) &&
2183 !_print_ascii_file("/etc/turbolinux-release", st) &&
2184 !_print_ascii_file("/etc/gentoo-release", st) &&
2185 !_print_ascii_file("/etc/ltib-release", st) &&
2186 !_print_ascii_file("/etc/angstrom-version", st) &&
2187 !_print_ascii_file("/etc/system-release", st) &&
2188 !_print_ascii_file("/etc/os-release", st)) {
2189
2190 if (file_exists("/etc/debian_version")) {
2191 st->print("Debian ");
2192 _print_ascii_file("/etc/debian_version", st);
2193 } else {
2194 st->print("Linux");
2195 }
2196 }
2197 st->cr();
2198 }
2199
2200 void os::Linux::print_libversion_info(outputStream* st) {
2201 // libc, pthread
2202 st->print("libc:");
2203 st->print("%s ", os::Linux::glibc_version());
2204 st->print("%s ", os::Linux::libpthread_version());
2205 if (os::Linux::is_LinuxThreads()) {
2206 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2207 }
2208 st->cr();
2209 }
2210
2211 void os::Linux::print_full_memory_info(outputStream* st) {
2212 st->print("\n/proc/meminfo:\n");
2213 _print_ascii_file("/proc/meminfo", st);
2214 st->cr();
2215 }
2216
2217 void os::print_memory_info(outputStream* st) {
2218
2219 st->print("Memory:");
2220 st->print(" %dk page", os::vm_page_size()>>10);
2221
2222 // values in struct sysinfo are "unsigned long"
2223 struct sysinfo si;
2224 sysinfo(&si);
2225
2226 st->print(", physical " UINT64_FORMAT "k",
2227 os::physical_memory() >> 10);
2228 st->print("(" UINT64_FORMAT "k free)",
2229 os::available_memory() >> 10);
2230 st->print(", swap " UINT64_FORMAT "k",
2231 ((jlong)si.totalswap * si.mem_unit) >> 10);
2232 st->print("(" UINT64_FORMAT "k free)",
2233 ((jlong)si.freeswap * si.mem_unit) >> 10);
2234 st->cr();
2235 }
2236
2237 void os::pd_print_cpu_info(outputStream* st) {
2238 st->print("\n/proc/cpuinfo:\n");
2239 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2240 st->print(" <Not Available>");
2241 }
2242 st->cr();
2243 }
2244
2245 void os::print_siginfo(outputStream* st, void* siginfo) {
2246 const siginfo_t* si = (const siginfo_t*)siginfo;
2247
2248 os::Posix::print_siginfo_brief(st, si);
2249 #if INCLUDE_CDS
2250 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2251 UseSharedSpaces) {
2252 FileMapInfo* mapinfo = FileMapInfo::current_info();
2253 if (mapinfo->is_in_shared_space(si->si_addr)) {
2254 st->print("\n\nError accessing class data sharing archive." \
2255 " Mapped file inaccessible during execution, " \
2256 " possible disk/network problem.");
2257 }
2258 }
2259 #endif
2260 st->cr();
2261 }
2262
2263
2264 static void print_signal_handler(outputStream* st, int sig,
2265 char* buf, size_t buflen);
2266
2267 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2268 st->print_cr("Signal Handlers:");
2269 print_signal_handler(st, SIGSEGV, buf, buflen);
2270 print_signal_handler(st, SIGBUS , buf, buflen);
2271 print_signal_handler(st, SIGFPE , buf, buflen);
2272 print_signal_handler(st, SIGPIPE, buf, buflen);
2273 print_signal_handler(st, SIGXFSZ, buf, buflen);
2274 print_signal_handler(st, SIGILL , buf, buflen);
2275 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2276 print_signal_handler(st, SR_signum, buf, buflen);
2277 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2278 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2279 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2280 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2281 #if defined(PPC64)
2282 print_signal_handler(st, SIGTRAP, buf, buflen);
2283 #endif
2284 }
2285
2286 static char saved_jvm_path[MAXPATHLEN] = {0};
2287
2288 // Find the full path to the current module, libjvm.so
2289 void os::jvm_path(char *buf, jint buflen) {
2290 // Error checking.
2291 if (buflen < MAXPATHLEN) {
2292 assert(false, "must use a large-enough buffer");
2293 buf[0] = '\0';
2294 return;
2295 }
2296 // Lazy resolve the path to current module.
2297 if (saved_jvm_path[0] != 0) {
2298 strcpy(buf, saved_jvm_path);
2299 return;
2300 }
2301
2302 char dli_fname[MAXPATHLEN];
2303 bool ret = dll_address_to_library_name(
2304 CAST_FROM_FN_PTR(address, os::jvm_path),
2305 dli_fname, sizeof(dli_fname), NULL);
2306 assert(ret, "cannot locate libjvm");
2307 char *rp = NULL;
2308 if (ret && dli_fname[0] != '\0') {
2309 rp = realpath(dli_fname, buf);
2310 }
2311 if (rp == NULL) {
2312 return;
2313 }
2314
2315 if (Arguments::sun_java_launcher_is_altjvm()) {
2316 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2317 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2318 // If "/jre/lib/" appears at the right place in the string, then
2319 // assume we are installed in a JDK and we're done. Otherwise, check
2320 // for a JAVA_HOME environment variable and fix up the path so it
2321 // looks like libjvm.so is installed there (append a fake suffix
2322 // hotspot/libjvm.so).
2323 const char *p = buf + strlen(buf) - 1;
2324 for (int count = 0; p > buf && count < 5; ++count) {
2325 for (--p; p > buf && *p != '/'; --p)
2326 /* empty */ ;
2327 }
2328
2329 if (strncmp(p, "/jre/lib/", 9) != 0) {
2330 // Look for JAVA_HOME in the environment.
2331 char* java_home_var = ::getenv("JAVA_HOME");
2332 if (java_home_var != NULL && java_home_var[0] != 0) {
2333 char* jrelib_p;
2334 int len;
2335
2336 // Check the current module name "libjvm.so".
2337 p = strrchr(buf, '/');
2338 if (p == NULL) {
2339 return;
2340 }
2341 assert(strstr(p, "/libjvm") == p, "invalid library name");
2342
2343 rp = realpath(java_home_var, buf);
2344 if (rp == NULL) {
2345 return;
2346 }
2347
2348 // determine if this is a legacy image or modules image
2349 // modules image doesn't have "jre" subdirectory
2350 len = strlen(buf);
2351 assert(len < buflen, "Ran out of buffer room");
2352 jrelib_p = buf + len;
2353 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2354 if (0 != access(buf, F_OK)) {
2355 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2356 }
2357
2358 if (0 == access(buf, F_OK)) {
2359 // Use current module name "libjvm.so"
2360 len = strlen(buf);
2361 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2362 } else {
2363 // Go back to path of .so
2364 rp = realpath(dli_fname, buf);
2365 if (rp == NULL) {
2366 return;
2367 }
2368 }
2369 }
2370 }
2371 }
2372
2373 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2374 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2375 }
2376
2377 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2378 // no prefix required, not even "_"
2379 }
2380
2381 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2382 // no suffix required
2383 }
2384
2385 ////////////////////////////////////////////////////////////////////////////////
2386 // sun.misc.Signal support
2387
2388 static volatile jint sigint_count = 0;
2389
2390 static void UserHandler(int sig, void *siginfo, void *context) {
2391 // 4511530 - sem_post is serialized and handled by the manager thread. When
2392 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2393 // don't want to flood the manager thread with sem_post requests.
2394 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2395 return;
2396 }
2397
2398 // Ctrl-C is pressed during error reporting, likely because the error
2399 // handler fails to abort. Let VM die immediately.
2400 if (sig == SIGINT && is_error_reported()) {
2401 os::die();
2402 }
2403
2404 os::signal_notify(sig);
2405 }
2406
2407 void* os::user_handler() {
2408 return CAST_FROM_FN_PTR(void*, UserHandler);
2409 }
2410
2411 class Semaphore : public StackObj {
2412 public:
2413 Semaphore();
2414 ~Semaphore();
2415 void signal();
2416 void wait();
2417 bool trywait();
2418 bool timedwait(unsigned int sec, int nsec);
2419 private:
2420 sem_t _semaphore;
2421 };
2422
2423 Semaphore::Semaphore() {
2424 sem_init(&_semaphore, 0, 0);
2425 }
2426
2427 Semaphore::~Semaphore() {
2428 sem_destroy(&_semaphore);
2429 }
2430
2431 void Semaphore::signal() {
2432 sem_post(&_semaphore);
2433 }
2434
2435 void Semaphore::wait() {
2436 sem_wait(&_semaphore);
2437 }
2438
2439 bool Semaphore::trywait() {
2440 return sem_trywait(&_semaphore) == 0;
2441 }
2442
2443 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2444
2445 struct timespec ts;
2446 // Semaphore's are always associated with CLOCK_REALTIME
2447 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2448 // see unpackTime for discussion on overflow checking
2449 if (sec >= MAX_SECS) {
2450 ts.tv_sec += MAX_SECS;
2451 ts.tv_nsec = 0;
2452 } else {
2453 ts.tv_sec += sec;
2454 ts.tv_nsec += nsec;
2455 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2456 ts.tv_nsec -= NANOSECS_PER_SEC;
2457 ++ts.tv_sec; // note: this must be <= max_secs
2458 }
2459 }
2460
2461 while (1) {
2462 int result = sem_timedwait(&_semaphore, &ts);
2463 if (result == 0) {
2464 return true;
2465 } else if (errno == EINTR) {
2466 continue;
2467 } else if (errno == ETIMEDOUT) {
2468 return false;
2469 } else {
2470 return false;
2471 }
2472 }
2473 }
2474
2475 extern "C" {
2476 typedef void (*sa_handler_t)(int);
2477 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2478 }
2479
2480 void* os::signal(int signal_number, void* handler) {
2481 struct sigaction sigAct, oldSigAct;
2482
2483 sigfillset(&(sigAct.sa_mask));
2484 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2485 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2486
2487 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2488 // -1 means registration failed
2489 return (void *)-1;
2490 }
2491
2492 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2493 }
2494
2495 void os::signal_raise(int signal_number) {
2496 ::raise(signal_number);
2497 }
2498
2499 // The following code is moved from os.cpp for making this
2500 // code platform specific, which it is by its very nature.
2501
2502 // Will be modified when max signal is changed to be dynamic
2503 int os::sigexitnum_pd() {
2504 return NSIG;
2505 }
2506
2507 // a counter for each possible signal value
2508 static volatile jint pending_signals[NSIG+1] = { 0 };
2509
2510 // Linux(POSIX) specific hand shaking semaphore.
2511 static sem_t sig_sem;
2512 static Semaphore sr_semaphore;
2513
2514 void os::signal_init_pd() {
2515 // Initialize signal structures
2516 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2517
2518 // Initialize signal semaphore
2519 ::sem_init(&sig_sem, 0, 0);
2520 }
2521
2522 void os::signal_notify(int sig) {
2523 Atomic::inc(&pending_signals[sig]);
2524 ::sem_post(&sig_sem);
2525 }
2526
2527 static int check_pending_signals(bool wait) {
2528 Atomic::store(0, &sigint_count);
2529 for (;;) {
2530 for (int i = 0; i < NSIG + 1; i++) {
2531 jint n = pending_signals[i];
2532 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2533 return i;
2534 }
2535 }
2536 if (!wait) {
2537 return -1;
2538 }
2539 JavaThread *thread = JavaThread::current();
2540 ThreadBlockInVM tbivm(thread);
2541
2542 bool threadIsSuspended;
2543 do {
2544 thread->set_suspend_equivalent();
2545 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2546 ::sem_wait(&sig_sem);
2547
2548 // were we externally suspended while we were waiting?
2549 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2550 if (threadIsSuspended) {
2551 // The semaphore has been incremented, but while we were waiting
2552 // another thread suspended us. We don't want to continue running
2553 // while suspended because that would surprise the thread that
2554 // suspended us.
2555 ::sem_post(&sig_sem);
2556
2557 thread->java_suspend_self();
2558 }
2559 } while (threadIsSuspended);
2560 }
2561 }
2562
2563 int os::signal_lookup() {
2564 return check_pending_signals(false);
2565 }
2566
2567 int os::signal_wait() {
2568 return check_pending_signals(true);
2569 }
2570
2571 ////////////////////////////////////////////////////////////////////////////////
2572 // Virtual Memory
2573
2574 int os::vm_page_size() {
2575 // Seems redundant as all get out
2576 assert(os::Linux::page_size() != -1, "must call os::init");
2577 return os::Linux::page_size();
2578 }
2579
2580 // Solaris allocates memory by pages.
2581 int os::vm_allocation_granularity() {
2582 assert(os::Linux::page_size() != -1, "must call os::init");
2583 return os::Linux::page_size();
2584 }
2585
2586 // Rationale behind this function:
2587 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2588 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2589 // samples for JITted code. Here we create private executable mapping over the code cache
2590 // and then we can use standard (well, almost, as mapping can change) way to provide
2591 // info for the reporting script by storing timestamp and location of symbol
2592 void linux_wrap_code(char* base, size_t size) {
2593 static volatile jint cnt = 0;
2594
2595 if (!UseOprofile) {
2596 return;
2597 }
2598
2599 char buf[PATH_MAX+1];
2600 int num = Atomic::add(1, &cnt);
2601
2602 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2603 os::get_temp_directory(), os::current_process_id(), num);
2604 unlink(buf);
2605
2606 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2607
2608 if (fd != -1) {
2609 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2610 if (rv != (off_t)-1) {
2611 if (::write(fd, "", 1) == 1) {
2612 mmap(base, size,
2613 PROT_READ|PROT_WRITE|PROT_EXEC,
2614 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2615 }
2616 }
2617 ::close(fd);
2618 unlink(buf);
2619 }
2620 }
2621
2622 static bool recoverable_mmap_error(int err) {
2623 // See if the error is one we can let the caller handle. This
2624 // list of errno values comes from JBS-6843484. I can't find a
2625 // Linux man page that documents this specific set of errno
2626 // values so while this list currently matches Solaris, it may
2627 // change as we gain experience with this failure mode.
2628 switch (err) {
2629 case EBADF:
2630 case EINVAL:
2631 case ENOTSUP:
2632 // let the caller deal with these errors
2633 return true;
2634
2635 default:
2636 // Any remaining errors on this OS can cause our reserved mapping
2637 // to be lost. That can cause confusion where different data
2638 // structures think they have the same memory mapped. The worst
2639 // scenario is if both the VM and a library think they have the
2640 // same memory mapped.
2641 return false;
2642 }
2643 }
2644
2645 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2646 int err) {
2647 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2648 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2649 strerror(err), err);
2650 }
2651
2652 static void warn_fail_commit_memory(char* addr, size_t size,
2653 size_t alignment_hint, bool exec,
2654 int err) {
2655 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2656 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2657 alignment_hint, exec, strerror(err), err);
2658 }
2659
2660 // NOTE: Linux kernel does not really reserve the pages for us.
2661 // All it does is to check if there are enough free pages
2662 // left at the time of mmap(). This could be a potential
2663 // problem.
2664 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2665 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2666 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2667 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2668 if (res != (uintptr_t) MAP_FAILED) {
2669 if (UseNUMAInterleaving) {
2670 numa_make_global(addr, size);
2671 }
2672 return 0;
2673 }
2674
2675 int err = errno; // save errno from mmap() call above
2676
2677 if (!recoverable_mmap_error(err)) {
2678 warn_fail_commit_memory(addr, size, exec, err);
2679 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2680 }
2681
2682 return err;
2683 }
2684
2685 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2686 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2687 }
2688
2689 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2690 const char* mesg) {
2691 assert(mesg != NULL, "mesg must be specified");
2692 int err = os::Linux::commit_memory_impl(addr, size, exec);
2693 if (err != 0) {
2694 // the caller wants all commit errors to exit with the specified mesg:
2695 warn_fail_commit_memory(addr, size, exec, err);
2696 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2697 }
2698 }
2699
2700 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2701 #ifndef MAP_HUGETLB
2702 #define MAP_HUGETLB 0x40000
2703 #endif
2704
2705 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2706 #ifndef MADV_HUGEPAGE
2707 #define MADV_HUGEPAGE 14
2708 #endif
2709
2710 int os::Linux::commit_memory_impl(char* addr, size_t size,
2711 size_t alignment_hint, bool exec) {
2712 int err = os::Linux::commit_memory_impl(addr, size, exec);
2713 if (err == 0) {
2714 realign_memory(addr, size, alignment_hint);
2715 }
2716 return err;
2717 }
2718
2719 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2720 bool exec) {
2721 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2722 }
2723
2724 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2725 size_t alignment_hint, bool exec,
2726 const char* mesg) {
2727 assert(mesg != NULL, "mesg must be specified");
2728 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2729 if (err != 0) {
2730 // the caller wants all commit errors to exit with the specified mesg:
2731 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2732 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2733 }
2734 }
2735
2736 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2737 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2738 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2739 // be supported or the memory may already be backed by huge pages.
2740 ::madvise(addr, bytes, MADV_HUGEPAGE);
2741 }
2742 }
2743
2744 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2745 // This method works by doing an mmap over an existing mmaping and effectively discarding
2746 // the existing pages. However it won't work for SHM-based large pages that cannot be
2747 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2748 // small pages on top of the SHM segment. This method always works for small pages, so we
2749 // allow that in any case.
2750 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2751 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2752 }
2753 }
2754
2755 void os::numa_make_global(char *addr, size_t bytes) {
2756 Linux::numa_interleave_memory(addr, bytes);
2757 }
2758
2759 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2760 // bind policy to MPOL_PREFERRED for the current thread.
2761 #define USE_MPOL_PREFERRED 0
2762
2763 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2764 // To make NUMA and large pages more robust when both enabled, we need to ease
2765 // the requirements on where the memory should be allocated. MPOL_BIND is the
2766 // default policy and it will force memory to be allocated on the specified
2767 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2768 // the specified node, but will not force it. Using this policy will prevent
2769 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2770 // free large pages.
2771 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2772 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2773 }
2774
2775 bool os::numa_topology_changed() { return false; }
2776
2777 size_t os::numa_get_groups_num() {
2778 int max_node = Linux::numa_max_node();
2779 return max_node > 0 ? max_node + 1 : 1;
2780 }
2781
2782 int os::numa_get_group_id() {
2783 int cpu_id = Linux::sched_getcpu();
2784 if (cpu_id != -1) {
2785 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2786 if (lgrp_id != -1) {
2787 return lgrp_id;
2788 }
2789 }
2790 return 0;
2791 }
2792
2793 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2794 for (size_t i = 0; i < size; i++) {
2795 ids[i] = i;
2796 }
2797 return size;
2798 }
2799
2800 bool os::get_page_info(char *start, page_info* info) {
2801 return false;
2802 }
2803
2804 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2805 page_info* page_found) {
2806 return end;
2807 }
2808
2809
2810 int os::Linux::sched_getcpu_syscall(void) {
2811 unsigned int cpu;
2812 int retval = -1;
2813
2814 #if defined(IA32)
2815 #ifndef SYS_getcpu
2816 #define SYS_getcpu 318
2817 #endif
2818 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2819 #elif defined(AMD64)
2820 // Unfortunately we have to bring all these macros here from vsyscall.h
2821 // to be able to compile on old linuxes.
2822 #define __NR_vgetcpu 2
2823 #define VSYSCALL_START (-10UL << 20)
2824 #define VSYSCALL_SIZE 1024
2825 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2826 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2827 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2828 retval = vgetcpu(&cpu, NULL, NULL);
2829 #endif
2830
2831 return (retval == -1) ? retval : cpu;
2832 }
2833
2834 // Something to do with the numa-aware allocator needs these symbols
2835 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2836 extern "C" JNIEXPORT void numa_error(char *where) { }
2837 extern "C" JNIEXPORT int fork1() { return fork(); }
2838
2839
2840 // If we are running with libnuma version > 2, then we should
2841 // be trying to use symbols with versions 1.1
2842 // If we are running with earlier version, which did not have symbol versions,
2843 // we should use the base version.
2844 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2845 void *f = dlvsym(handle, name, "libnuma_1.1");
2846 if (f == NULL) {
2847 f = dlsym(handle, name);
2848 }
2849 return f;
2850 }
2851
2852 bool os::Linux::libnuma_init() {
2853 // sched_getcpu() should be in libc.
2854 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2855 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2856
2857 // If it's not, try a direct syscall.
2858 if (sched_getcpu() == -1) {
2859 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2860 (void*)&sched_getcpu_syscall));
2861 }
2862
2863 if (sched_getcpu() != -1) { // Does it work?
2864 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2865 if (handle != NULL) {
2866 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2867 libnuma_dlsym(handle, "numa_node_to_cpus")));
2868 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2869 libnuma_dlsym(handle, "numa_max_node")));
2870 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2871 libnuma_dlsym(handle, "numa_available")));
2872 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2873 libnuma_dlsym(handle, "numa_tonode_memory")));
2874 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2875 libnuma_dlsym(handle, "numa_interleave_memory")));
2876 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2877 libnuma_dlsym(handle, "numa_set_bind_policy")));
2878
2879
2880 if (numa_available() != -1) {
2881 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2882 // Create a cpu -> node mapping
2883 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2884 rebuild_cpu_to_node_map();
2885 return true;
2886 }
2887 }
2888 }
2889 return false;
2890 }
2891
2892 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2893 // The table is later used in get_node_by_cpu().
2894 void os::Linux::rebuild_cpu_to_node_map() {
2895 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2896 // in libnuma (possible values are starting from 16,
2897 // and continuing up with every other power of 2, but less
2898 // than the maximum number of CPUs supported by kernel), and
2899 // is a subject to change (in libnuma version 2 the requirements
2900 // are more reasonable) we'll just hardcode the number they use
2901 // in the library.
2902 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2903
2904 size_t cpu_num = os::active_processor_count();
2905 size_t cpu_map_size = NCPUS / BitsPerCLong;
2906 size_t cpu_map_valid_size =
2907 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2908
2909 cpu_to_node()->clear();
2910 cpu_to_node()->at_grow(cpu_num - 1);
2911 size_t node_num = numa_get_groups_num();
2912
2913 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2914 for (size_t i = 0; i < node_num; i++) {
2915 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2916 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2917 if (cpu_map[j] != 0) {
2918 for (size_t k = 0; k < BitsPerCLong; k++) {
2919 if (cpu_map[j] & (1UL << k)) {
2920 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2921 }
2922 }
2923 }
2924 }
2925 }
2926 }
2927 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2928 }
2929
2930 int os::Linux::get_node_by_cpu(int cpu_id) {
2931 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2932 return cpu_to_node()->at(cpu_id);
2933 }
2934 return -1;
2935 }
2936
2937 GrowableArray<int>* os::Linux::_cpu_to_node;
2938 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2939 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2940 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2941 os::Linux::numa_available_func_t os::Linux::_numa_available;
2942 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2943 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2944 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2945 unsigned long* os::Linux::_numa_all_nodes;
2946
2947 bool os::pd_uncommit_memory(char* addr, size_t size) {
2948 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2949 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2950 return res != (uintptr_t) MAP_FAILED;
2951 }
2952
2953 static address get_stack_commited_bottom(address bottom, size_t size) {
2954 address nbot = bottom;
2955 address ntop = bottom + size;
2956
2957 size_t page_sz = os::vm_page_size();
2958 unsigned pages = size / page_sz;
2959
2960 unsigned char vec[1];
2961 unsigned imin = 1, imax = pages + 1, imid;
2962 int mincore_return_value = 0;
2963
2964 assert(imin <= imax, "Unexpected page size");
2965
2966 while (imin < imax) {
2967 imid = (imax + imin) / 2;
2968 nbot = ntop - (imid * page_sz);
2969
2970 // Use a trick with mincore to check whether the page is mapped or not.
2971 // mincore sets vec to 1 if page resides in memory and to 0 if page
2972 // is swapped output but if page we are asking for is unmapped
2973 // it returns -1,ENOMEM
2974 mincore_return_value = mincore(nbot, page_sz, vec);
2975
2976 if (mincore_return_value == -1) {
2977 // Page is not mapped go up
2978 // to find first mapped page
2979 if (errno != EAGAIN) {
2980 assert(errno == ENOMEM, "Unexpected mincore errno");
2981 imax = imid;
2982 }
2983 } else {
2984 // Page is mapped go down
2985 // to find first not mapped page
2986 imin = imid + 1;
2987 }
2988 }
2989
2990 nbot = nbot + page_sz;
2991
2992 // Adjust stack bottom one page up if last checked page is not mapped
2993 if (mincore_return_value == -1) {
2994 nbot = nbot + page_sz;
2995 }
2996
2997 return nbot;
2998 }
2999
3000
3001 // Linux uses a growable mapping for the stack, and if the mapping for
3002 // the stack guard pages is not removed when we detach a thread the
3003 // stack cannot grow beyond the pages where the stack guard was
3004 // mapped. If at some point later in the process the stack expands to
3005 // that point, the Linux kernel cannot expand the stack any further
3006 // because the guard pages are in the way, and a segfault occurs.
3007 //
3008 // However, it's essential not to split the stack region by unmapping
3009 // a region (leaving a hole) that's already part of the stack mapping,
3010 // so if the stack mapping has already grown beyond the guard pages at
3011 // the time we create them, we have to truncate the stack mapping.
3012 // So, we need to know the extent of the stack mapping when
3013 // create_stack_guard_pages() is called.
3014
3015 // We only need this for stacks that are growable: at the time of
3016 // writing thread stacks don't use growable mappings (i.e. those
3017 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3018 // only applies to the main thread.
3019
3020 // If the (growable) stack mapping already extends beyond the point
3021 // where we're going to put our guard pages, truncate the mapping at
3022 // that point by munmap()ping it. This ensures that when we later
3023 // munmap() the guard pages we don't leave a hole in the stack
3024 // mapping. This only affects the main/initial thread
3025
3026 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3027 if (os::Linux::is_initial_thread()) {
3028 // As we manually grow stack up to bottom inside create_attached_thread(),
3029 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3030 // we don't need to do anything special.
3031 // Check it first, before calling heavy function.
3032 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3033 unsigned char vec[1];
3034
3035 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3036 // Fallback to slow path on all errors, including EAGAIN
3037 stack_extent = (uintptr_t) get_stack_commited_bottom(
3038 os::Linux::initial_thread_stack_bottom(),
3039 (size_t)addr - stack_extent);
3040 }
3041
3042 if (stack_extent < (uintptr_t)addr) {
3043 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3044 }
3045 }
3046
3047 return os::commit_memory(addr, size, !ExecMem);
3048 }
3049
3050 // If this is a growable mapping, remove the guard pages entirely by
3051 // munmap()ping them. If not, just call uncommit_memory(). This only
3052 // affects the main/initial thread, but guard against future OS changes
3053 // It's safe to always unmap guard pages for initial thread because we
3054 // always place it right after end of the mapped region
3055
3056 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3057 uintptr_t stack_extent, stack_base;
3058
3059 if (os::Linux::is_initial_thread()) {
3060 return ::munmap(addr, size) == 0;
3061 }
3062
3063 return os::uncommit_memory(addr, size);
3064 }
3065
3066 static address _highest_vm_reserved_address = NULL;
3067
3068 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3069 // at 'requested_addr'. If there are existing memory mappings at the same
3070 // location, however, they will be overwritten. If 'fixed' is false,
3071 // 'requested_addr' is only treated as a hint, the return value may or
3072 // may not start from the requested address. Unlike Linux mmap(), this
3073 // function returns NULL to indicate failure.
3074 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3075 char * addr;
3076 int flags;
3077
3078 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3079 if (fixed) {
3080 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3081 flags |= MAP_FIXED;
3082 }
3083
3084 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3085 // touch an uncommitted page. Otherwise, the read/write might
3086 // succeed if we have enough swap space to back the physical page.
3087 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3088 flags, -1, 0);
3089
3090 if (addr != MAP_FAILED) {
3091 // anon_mmap() should only get called during VM initialization,
3092 // don't need lock (actually we can skip locking even it can be called
3093 // from multiple threads, because _highest_vm_reserved_address is just a
3094 // hint about the upper limit of non-stack memory regions.)
3095 if ((address)addr + bytes > _highest_vm_reserved_address) {
3096 _highest_vm_reserved_address = (address)addr + bytes;
3097 }
3098 }
3099
3100 return addr == MAP_FAILED ? NULL : addr;
3101 }
3102
3103 // Don't update _highest_vm_reserved_address, because there might be memory
3104 // regions above addr + size. If so, releasing a memory region only creates
3105 // a hole in the address space, it doesn't help prevent heap-stack collision.
3106 //
3107 static int anon_munmap(char * addr, size_t size) {
3108 return ::munmap(addr, size) == 0;
3109 }
3110
3111 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3112 size_t alignment_hint) {
3113 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3114 }
3115
3116 bool os::pd_release_memory(char* addr, size_t size) {
3117 return anon_munmap(addr, size);
3118 }
3119
3120 static address highest_vm_reserved_address() {
3121 return _highest_vm_reserved_address;
3122 }
3123
3124 static bool linux_mprotect(char* addr, size_t size, int prot) {
3125 // Linux wants the mprotect address argument to be page aligned.
3126 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3127
3128 // According to SUSv3, mprotect() should only be used with mappings
3129 // established by mmap(), and mmap() always maps whole pages. Unaligned
3130 // 'addr' likely indicates problem in the VM (e.g. trying to change
3131 // protection of malloc'ed or statically allocated memory). Check the
3132 // caller if you hit this assert.
3133 assert(addr == bottom, "sanity check");
3134
3135 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3136 return ::mprotect(bottom, size, prot) == 0;
3137 }
3138
3139 // Set protections specified
3140 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3141 bool is_committed) {
3142 unsigned int p = 0;
3143 switch (prot) {
3144 case MEM_PROT_NONE: p = PROT_NONE; break;
3145 case MEM_PROT_READ: p = PROT_READ; break;
3146 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3147 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3148 default:
3149 ShouldNotReachHere();
3150 }
3151 // is_committed is unused.
3152 return linux_mprotect(addr, bytes, p);
3153 }
3154
3155 bool os::guard_memory(char* addr, size_t size) {
3156 return linux_mprotect(addr, size, PROT_NONE);
3157 }
3158
3159 bool os::unguard_memory(char* addr, size_t size) {
3160 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3161 }
3162
3163 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3164 size_t page_size) {
3165 bool result = false;
3166 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3167 MAP_ANONYMOUS|MAP_PRIVATE,
3168 -1, 0);
3169 if (p != MAP_FAILED) {
3170 void *aligned_p = align_ptr_up(p, page_size);
3171
3172 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3173
3174 munmap(p, page_size * 2);
3175 }
3176
3177 if (warn && !result) {
3178 warning("TransparentHugePages is not supported by the operating system.");
3179 }
3180
3181 return result;
3182 }
3183
3184 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3185 bool result = false;
3186 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3187 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3188 -1, 0);
3189
3190 if (p != MAP_FAILED) {
3191 // We don't know if this really is a huge page or not.
3192 FILE *fp = fopen("/proc/self/maps", "r");
3193 if (fp) {
3194 while (!feof(fp)) {
3195 char chars[257];
3196 long x = 0;
3197 if (fgets(chars, sizeof(chars), fp)) {
3198 if (sscanf(chars, "%lx-%*x", &x) == 1
3199 && x == (long)p) {
3200 if (strstr (chars, "hugepage")) {
3201 result = true;
3202 break;
3203 }
3204 }
3205 }
3206 }
3207 fclose(fp);
3208 }
3209 munmap(p, page_size);
3210 }
3211
3212 if (warn && !result) {
3213 warning("HugeTLBFS is not supported by the operating system.");
3214 }
3215
3216 return result;
3217 }
3218
3219 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3220 //
3221 // From the coredump_filter documentation:
3222 //
3223 // - (bit 0) anonymous private memory
3224 // - (bit 1) anonymous shared memory
3225 // - (bit 2) file-backed private memory
3226 // - (bit 3) file-backed shared memory
3227 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3228 // effective only if the bit 2 is cleared)
3229 // - (bit 5) hugetlb private memory
3230 // - (bit 6) hugetlb shared memory
3231 //
3232 static void set_coredump_filter(void) {
3233 FILE *f;
3234 long cdm;
3235
3236 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3237 return;
3238 }
3239
3240 if (fscanf(f, "%lx", &cdm) != 1) {
3241 fclose(f);
3242 return;
3243 }
3244
3245 rewind(f);
3246
3247 if ((cdm & LARGEPAGES_BIT) == 0) {
3248 cdm |= LARGEPAGES_BIT;
3249 fprintf(f, "%#lx", cdm);
3250 }
3251
3252 fclose(f);
3253 }
3254
3255 // Large page support
3256
3257 static size_t _large_page_size = 0;
3258
3259 size_t os::Linux::find_large_page_size() {
3260 size_t large_page_size = 0;
3261
3262 // large_page_size on Linux is used to round up heap size. x86 uses either
3263 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3264 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3265 // page as large as 256M.
3266 //
3267 // Here we try to figure out page size by parsing /proc/meminfo and looking
3268 // for a line with the following format:
3269 // Hugepagesize: 2048 kB
3270 //
3271 // If we can't determine the value (e.g. /proc is not mounted, or the text
3272 // format has been changed), we'll use the largest page size supported by
3273 // the processor.
3274
3275 #ifndef ZERO
3276 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3277 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3278 #endif // ZERO
3279
3280 FILE *fp = fopen("/proc/meminfo", "r");
3281 if (fp) {
3282 while (!feof(fp)) {
3283 int x = 0;
3284 char buf[16];
3285 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3286 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3287 large_page_size = x * K;
3288 break;
3289 }
3290 } else {
3291 // skip to next line
3292 for (;;) {
3293 int ch = fgetc(fp);
3294 if (ch == EOF || ch == (int)'\n') break;
3295 }
3296 }
3297 }
3298 fclose(fp);
3299 }
3300
3301 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3302 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3303 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3304 proper_unit_for_byte_size(large_page_size));
3305 }
3306
3307 return large_page_size;
3308 }
3309
3310 size_t os::Linux::setup_large_page_size() {
3311 _large_page_size = Linux::find_large_page_size();
3312 const size_t default_page_size = (size_t)Linux::page_size();
3313 if (_large_page_size > default_page_size) {
3314 _page_sizes[0] = _large_page_size;
3315 _page_sizes[1] = default_page_size;
3316 _page_sizes[2] = 0;
3317 }
3318
3319 return _large_page_size;
3320 }
3321
3322 bool os::Linux::setup_large_page_type(size_t page_size) {
3323 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3324 FLAG_IS_DEFAULT(UseSHM) &&
3325 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3326
3327 // The type of large pages has not been specified by the user.
3328
3329 // Try UseHugeTLBFS and then UseSHM.
3330 UseHugeTLBFS = UseSHM = true;
3331
3332 // Don't try UseTransparentHugePages since there are known
3333 // performance issues with it turned on. This might change in the future.
3334 UseTransparentHugePages = false;
3335 }
3336
3337 if (UseTransparentHugePages) {
3338 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3339 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3340 UseHugeTLBFS = false;
3341 UseSHM = false;
3342 return true;
3343 }
3344 UseTransparentHugePages = false;
3345 }
3346
3347 if (UseHugeTLBFS) {
3348 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3349 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3350 UseSHM = false;
3351 return true;
3352 }
3353 UseHugeTLBFS = false;
3354 }
3355
3356 return UseSHM;
3357 }
3358
3359 void os::large_page_init() {
3360 if (!UseLargePages &&
3361 !UseTransparentHugePages &&
3362 !UseHugeTLBFS &&
3363 !UseSHM) {
3364 // Not using large pages.
3365 return;
3366 }
3367
3368 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3369 // The user explicitly turned off large pages.
3370 // Ignore the rest of the large pages flags.
3371 UseTransparentHugePages = false;
3372 UseHugeTLBFS = false;
3373 UseSHM = false;
3374 return;
3375 }
3376
3377 size_t large_page_size = Linux::setup_large_page_size();
3378 UseLargePages = Linux::setup_large_page_type(large_page_size);
3379
3380 set_coredump_filter();
3381 }
3382
3383 #ifndef SHM_HUGETLB
3384 #define SHM_HUGETLB 04000
3385 #endif
3386
3387 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3388 char* req_addr, bool exec) {
3389 // "exec" is passed in but not used. Creating the shared image for
3390 // the code cache doesn't have an SHM_X executable permission to check.
3391 assert(UseLargePages && UseSHM, "only for SHM large pages");
3392 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3393
3394 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3395 return NULL; // Fallback to small pages.
3396 }
3397
3398 key_t key = IPC_PRIVATE;
3399 char *addr;
3400
3401 bool warn_on_failure = UseLargePages &&
3402 (!FLAG_IS_DEFAULT(UseLargePages) ||
3403 !FLAG_IS_DEFAULT(UseSHM) ||
3404 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3405 char msg[128];
3406
3407 // Create a large shared memory region to attach to based on size.
3408 // Currently, size is the total size of the heap
3409 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3410 if (shmid == -1) {
3411 // Possible reasons for shmget failure:
3412 // 1. shmmax is too small for Java heap.
3413 // > check shmmax value: cat /proc/sys/kernel/shmmax
3414 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3415 // 2. not enough large page memory.
3416 // > check available large pages: cat /proc/meminfo
3417 // > increase amount of large pages:
3418 // echo new_value > /proc/sys/vm/nr_hugepages
3419 // Note 1: different Linux may use different name for this property,
3420 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3421 // Note 2: it's possible there's enough physical memory available but
3422 // they are so fragmented after a long run that they can't
3423 // coalesce into large pages. Try to reserve large pages when
3424 // the system is still "fresh".
3425 if (warn_on_failure) {
3426 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3427 warning("%s", msg);
3428 }
3429 return NULL;
3430 }
3431
3432 // attach to the region
3433 addr = (char*)shmat(shmid, req_addr, 0);
3434 int err = errno;
3435
3436 // Remove shmid. If shmat() is successful, the actual shared memory segment
3437 // will be deleted when it's detached by shmdt() or when the process
3438 // terminates. If shmat() is not successful this will remove the shared
3439 // segment immediately.
3440 shmctl(shmid, IPC_RMID, NULL);
3441
3442 if ((intptr_t)addr == -1) {
3443 if (warn_on_failure) {
3444 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3445 warning("%s", msg);
3446 }
3447 return NULL;
3448 }
3449
3450 return addr;
3451 }
3452
3453 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3454 int error) {
3455 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3456
3457 bool warn_on_failure = UseLargePages &&
3458 (!FLAG_IS_DEFAULT(UseLargePages) ||
3459 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3460 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3461
3462 if (warn_on_failure) {
3463 char msg[128];
3464 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3465 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3466 warning("%s", msg);
3467 }
3468 }
3469
3470 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3471 char* req_addr,
3472 bool exec) {
3473 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3474 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3475 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3476
3477 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3478 char* addr = (char*)::mmap(req_addr, bytes, prot,
3479 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3480 -1, 0);
3481
3482 if (addr == MAP_FAILED) {
3483 warn_on_large_pages_failure(req_addr, bytes, errno);
3484 return NULL;
3485 }
3486
3487 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3488
3489 return addr;
3490 }
3491
3492 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3493 size_t alignment,
3494 char* req_addr,
3495 bool exec) {
3496 size_t large_page_size = os::large_page_size();
3497
3498 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3499
3500 // Allocate small pages.
3501
3502 char* start;
3503 if (req_addr != NULL) {
3504 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3505 assert(is_size_aligned(bytes, alignment), "Must be");
3506 start = os::reserve_memory(bytes, req_addr);
3507 assert(start == NULL || start == req_addr, "Must be");
3508 } else {
3509 start = os::reserve_memory_aligned(bytes, alignment);
3510 }
3511
3512 if (start == NULL) {
3513 return NULL;
3514 }
3515
3516 assert(is_ptr_aligned(start, alignment), "Must be");
3517
3518 if (MemTracker::tracking_level() > NMT_minimal) {
3519 // os::reserve_memory_special will record this memory area.
3520 // Need to release it here to prevent overlapping reservations.
3521 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3522 tkr.record((address)start, bytes);
3523 }
3524
3525 char* end = start + bytes;
3526
3527 // Find the regions of the allocated chunk that can be promoted to large pages.
3528 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3529 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3530
3531 size_t lp_bytes = lp_end - lp_start;
3532
3533 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3534
3535 if (lp_bytes == 0) {
3536 // The mapped region doesn't even span the start and the end of a large page.
3537 // Fall back to allocate a non-special area.
3538 ::munmap(start, end - start);
3539 return NULL;
3540 }
3541
3542 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3543
3544
3545 void* result;
3546
3547 if (start != lp_start) {
3548 result = ::mmap(start, lp_start - start, prot,
3549 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3550 -1, 0);
3551 if (result == MAP_FAILED) {
3552 ::munmap(lp_start, end - lp_start);
3553 return NULL;
3554 }
3555 }
3556
3557 result = ::mmap(lp_start, lp_bytes, prot,
3558 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3559 -1, 0);
3560 if (result == MAP_FAILED) {
3561 warn_on_large_pages_failure(req_addr, bytes, errno);
3562 // If the mmap above fails, the large pages region will be unmapped and we
3563 // have regions before and after with small pages. Release these regions.
3564 //
3565 // | mapped | unmapped | mapped |
3566 // ^ ^ ^ ^
3567 // start lp_start lp_end end
3568 //
3569 ::munmap(start, lp_start - start);
3570 ::munmap(lp_end, end - lp_end);
3571 return NULL;
3572 }
3573
3574 if (lp_end != end) {
3575 result = ::mmap(lp_end, end - lp_end, prot,
3576 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3577 -1, 0);
3578 if (result == MAP_FAILED) {
3579 ::munmap(start, lp_end - start);
3580 return NULL;
3581 }
3582 }
3583
3584 return start;
3585 }
3586
3587 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3588 size_t alignment,
3589 char* req_addr,
3590 bool exec) {
3591 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3592 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3593 assert(is_power_of_2(alignment), "Must be");
3594 assert(is_power_of_2(os::large_page_size()), "Must be");
3595 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3596
3597 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3598 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3599 } else {
3600 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3601 }
3602 }
3603
3604 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3605 char* req_addr, bool exec) {
3606 assert(UseLargePages, "only for large pages");
3607
3608 char* addr;
3609 if (UseSHM) {
3610 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3611 } else {
3612 assert(UseHugeTLBFS, "must be");
3613 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3614 }
3615
3616 if (addr != NULL) {
3617 if (UseNUMAInterleaving) {
3618 numa_make_global(addr, bytes);
3619 }
3620
3621 // The memory is committed
3622 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3623 }
3624
3625 return addr;
3626 }
3627
3628 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3629 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3630 return shmdt(base) == 0;
3631 }
3632
3633 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3634 return pd_release_memory(base, bytes);
3635 }
3636
3637 bool os::release_memory_special(char* base, size_t bytes) {
3638 bool res;
3639 if (MemTracker::tracking_level() > NMT_minimal) {
3640 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3641 res = os::Linux::release_memory_special_impl(base, bytes);
3642 if (res) {
3643 tkr.record((address)base, bytes);
3644 }
3645
3646 } else {
3647 res = os::Linux::release_memory_special_impl(base, bytes);
3648 }
3649 return res;
3650 }
3651
3652 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3653 assert(UseLargePages, "only for large pages");
3654 bool res;
3655
3656 if (UseSHM) {
3657 res = os::Linux::release_memory_special_shm(base, bytes);
3658 } else {
3659 assert(UseHugeTLBFS, "must be");
3660 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3661 }
3662 return res;
3663 }
3664
3665 size_t os::large_page_size() {
3666 return _large_page_size;
3667 }
3668
3669 // With SysV SHM the entire memory region must be allocated as shared
3670 // memory.
3671 // HugeTLBFS allows application to commit large page memory on demand.
3672 // However, when committing memory with HugeTLBFS fails, the region
3673 // that was supposed to be committed will lose the old reservation
3674 // and allow other threads to steal that memory region. Because of this
3675 // behavior we can't commit HugeTLBFS memory.
3676 bool os::can_commit_large_page_memory() {
3677 return UseTransparentHugePages;
3678 }
3679
3680 bool os::can_execute_large_page_memory() {
3681 return UseTransparentHugePages || UseHugeTLBFS;
3682 }
3683
3684 // Reserve memory at an arbitrary address, only if that area is
3685 // available (and not reserved for something else).
3686
3687 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3688 const int max_tries = 10;
3689 char* base[max_tries];
3690 size_t size[max_tries];
3691 const size_t gap = 0x000000;
3692
3693 // Assert only that the size is a multiple of the page size, since
3694 // that's all that mmap requires, and since that's all we really know
3695 // about at this low abstraction level. If we need higher alignment,
3696 // we can either pass an alignment to this method or verify alignment
3697 // in one of the methods further up the call chain. See bug 5044738.
3698 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3699
3700 // Repeatedly allocate blocks until the block is allocated at the
3701 // right spot. Give up after max_tries. Note that reserve_memory() will
3702 // automatically update _highest_vm_reserved_address if the call is
3703 // successful. The variable tracks the highest memory address every reserved
3704 // by JVM. It is used to detect heap-stack collision if running with
3705 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3706 // space than needed, it could confuse the collision detecting code. To
3707 // solve the problem, save current _highest_vm_reserved_address and
3708 // calculate the correct value before return.
3709 address old_highest = _highest_vm_reserved_address;
3710
3711 // Linux mmap allows caller to pass an address as hint; give it a try first,
3712 // if kernel honors the hint then we can return immediately.
3713 char * addr = anon_mmap(requested_addr, bytes, false);
3714 if (addr == requested_addr) {
3715 return requested_addr;
3716 }
3717
3718 if (addr != NULL) {
3719 // mmap() is successful but it fails to reserve at the requested address
3720 anon_munmap(addr, bytes);
3721 }
3722
3723 int i;
3724 for (i = 0; i < max_tries; ++i) {
3725 base[i] = reserve_memory(bytes);
3726
3727 if (base[i] != NULL) {
3728 // Is this the block we wanted?
3729 if (base[i] == requested_addr) {
3730 size[i] = bytes;
3731 break;
3732 }
3733
3734 // Does this overlap the block we wanted? Give back the overlapped
3735 // parts and try again.
3736
3737 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3738 if (top_overlap >= 0 && top_overlap < bytes) {
3739 unmap_memory(base[i], top_overlap);
3740 base[i] += top_overlap;
3741 size[i] = bytes - top_overlap;
3742 } else {
3743 size_t bottom_overlap = base[i] + bytes - requested_addr;
3744 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3745 unmap_memory(requested_addr, bottom_overlap);
3746 size[i] = bytes - bottom_overlap;
3747 } else {
3748 size[i] = bytes;
3749 }
3750 }
3751 }
3752 }
3753
3754 // Give back the unused reserved pieces.
3755
3756 for (int j = 0; j < i; ++j) {
3757 if (base[j] != NULL) {
3758 unmap_memory(base[j], size[j]);
3759 }
3760 }
3761
3762 if (i < max_tries) {
3763 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3764 return requested_addr;
3765 } else {
3766 _highest_vm_reserved_address = old_highest;
3767 return NULL;
3768 }
3769 }
3770
3771 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3772 return ::read(fd, buf, nBytes);
3773 }
3774
3775 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3776 return ::pread(fd, buf, nBytes, offset);
3777 }
3778
3779 // Short sleep, direct OS call.
3780 //
3781 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3782 // sched_yield(2) will actually give up the CPU:
3783 //
3784 // * Alone on this pariticular CPU, keeps running.
3785 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3786 // (pre 2.6.39).
3787 //
3788 // So calling this with 0 is an alternative.
3789 //
3790 void os::naked_short_sleep(jlong ms) {
3791 struct timespec req;
3792
3793 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3794 req.tv_sec = 0;
3795 if (ms > 0) {
3796 req.tv_nsec = (ms % 1000) * 1000000;
3797 } else {
3798 req.tv_nsec = 1;
3799 }
3800
3801 nanosleep(&req, NULL);
3802
3803 return;
3804 }
3805
3806 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3807 void os::infinite_sleep() {
3808 while (true) { // sleep forever ...
3809 ::sleep(100); // ... 100 seconds at a time
3810 }
3811 }
3812
3813 // Used to convert frequent JVM_Yield() to nops
3814 bool os::dont_yield() {
3815 return DontYieldALot;
3816 }
3817
3818 void os::naked_yield() {
3819 sched_yield();
3820 }
3821
3822 ////////////////////////////////////////////////////////////////////////////////
3823 // thread priority support
3824
3825 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3826 // only supports dynamic priority, static priority must be zero. For real-time
3827 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3828 // However, for large multi-threaded applications, SCHED_RR is not only slower
3829 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3830 // of 5 runs - Sep 2005).
3831 //
3832 // The following code actually changes the niceness of kernel-thread/LWP. It
3833 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3834 // not the entire user process, and user level threads are 1:1 mapped to kernel
3835 // threads. It has always been the case, but could change in the future. For
3836 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3837 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3838
3839 int os::java_to_os_priority[CriticalPriority + 1] = {
3840 19, // 0 Entry should never be used
3841
3842 4, // 1 MinPriority
3843 3, // 2
3844 2, // 3
3845
3846 1, // 4
3847 0, // 5 NormPriority
3848 -1, // 6
3849
3850 -2, // 7
3851 -3, // 8
3852 -4, // 9 NearMaxPriority
3853
3854 -5, // 10 MaxPriority
3855
3856 -5 // 11 CriticalPriority
3857 };
3858
3859 static int prio_init() {
3860 if (ThreadPriorityPolicy == 1) {
3861 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3862 // if effective uid is not root. Perhaps, a more elegant way of doing
3863 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3864 if (geteuid() != 0) {
3865 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3866 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3867 }
3868 ThreadPriorityPolicy = 0;
3869 }
3870 }
3871 if (UseCriticalJavaThreadPriority) {
3872 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3873 }
3874 return 0;
3875 }
3876
3877 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3878 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3879
3880 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3881 return (ret == 0) ? OS_OK : OS_ERR;
3882 }
3883
3884 OSReturn os::get_native_priority(const Thread* const thread,
3885 int *priority_ptr) {
3886 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3887 *priority_ptr = java_to_os_priority[NormPriority];
3888 return OS_OK;
3889 }
3890
3891 errno = 0;
3892 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3893 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3894 }
3895
3896 // Hint to the underlying OS that a task switch would not be good.
3897 // Void return because it's a hint and can fail.
3898 void os::hint_no_preempt() {}
3899
3900 ////////////////////////////////////////////////////////////////////////////////
3901 // suspend/resume support
3902
3903 // the low-level signal-based suspend/resume support is a remnant from the
3904 // old VM-suspension that used to be for java-suspension, safepoints etc,
3905 // within hotspot. Now there is a single use-case for this:
3906 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3907 // that runs in the watcher thread.
3908 // The remaining code is greatly simplified from the more general suspension
3909 // code that used to be used.
3910 //
3911 // The protocol is quite simple:
3912 // - suspend:
3913 // - sends a signal to the target thread
3914 // - polls the suspend state of the osthread using a yield loop
3915 // - target thread signal handler (SR_handler) sets suspend state
3916 // and blocks in sigsuspend until continued
3917 // - resume:
3918 // - sets target osthread state to continue
3919 // - sends signal to end the sigsuspend loop in the SR_handler
3920 //
3921 // Note that the SR_lock plays no role in this suspend/resume protocol.
3922
3923 static void resume_clear_context(OSThread *osthread) {
3924 osthread->set_ucontext(NULL);
3925 osthread->set_siginfo(NULL);
3926 }
3927
3928 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3929 ucontext_t* context) {
3930 osthread->set_ucontext(context);
3931 osthread->set_siginfo(siginfo);
3932 }
3933
3934 // Handler function invoked when a thread's execution is suspended or
3935 // resumed. We have to be careful that only async-safe functions are
3936 // called here (Note: most pthread functions are not async safe and
3937 // should be avoided.)
3938 //
3939 // Note: sigwait() is a more natural fit than sigsuspend() from an
3940 // interface point of view, but sigwait() prevents the signal hander
3941 // from being run. libpthread would get very confused by not having
3942 // its signal handlers run and prevents sigwait()'s use with the
3943 // mutex granting granting signal.
3944 //
3945 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3946 //
3947 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3948 // Save and restore errno to avoid confusing native code with EINTR
3949 // after sigsuspend.
3950 int old_errno = errno;
3951
3952 Thread* thread = Thread::current();
3953 OSThread* osthread = thread->osthread();
3954 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3955
3956 os::SuspendResume::State current = osthread->sr.state();
3957 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3958 suspend_save_context(osthread, siginfo, context);
3959
3960 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3961 os::SuspendResume::State state = osthread->sr.suspended();
3962 if (state == os::SuspendResume::SR_SUSPENDED) {
3963 sigset_t suspend_set; // signals for sigsuspend()
3964
3965 // get current set of blocked signals and unblock resume signal
3966 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3967 sigdelset(&suspend_set, SR_signum);
3968
3969 sr_semaphore.signal();
3970 // wait here until we are resumed
3971 while (1) {
3972 sigsuspend(&suspend_set);
3973
3974 os::SuspendResume::State result = osthread->sr.running();
3975 if (result == os::SuspendResume::SR_RUNNING) {
3976 sr_semaphore.signal();
3977 break;
3978 }
3979 }
3980
3981 } else if (state == os::SuspendResume::SR_RUNNING) {
3982 // request was cancelled, continue
3983 } else {
3984 ShouldNotReachHere();
3985 }
3986
3987 resume_clear_context(osthread);
3988 } else if (current == os::SuspendResume::SR_RUNNING) {
3989 // request was cancelled, continue
3990 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3991 // ignore
3992 } else {
3993 // ignore
3994 }
3995
3996 errno = old_errno;
3997 }
3998
3999
4000 static int SR_initialize() {
4001 struct sigaction act;
4002 char *s;
4003 // Get signal number to use for suspend/resume
4004 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4005 int sig = ::strtol(s, 0, 10);
4006 if (sig > 0 || sig < _NSIG) {
4007 SR_signum = sig;
4008 }
4009 }
4010
4011 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4012 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4013
4014 sigemptyset(&SR_sigset);
4015 sigaddset(&SR_sigset, SR_signum);
4016
4017 // Set up signal handler for suspend/resume
4018 act.sa_flags = SA_RESTART|SA_SIGINFO;
4019 act.sa_handler = (void (*)(int)) SR_handler;
4020
4021 // SR_signum is blocked by default.
4022 // 4528190 - We also need to block pthread restart signal (32 on all
4023 // supported Linux platforms). Note that LinuxThreads need to block
4024 // this signal for all threads to work properly. So we don't have
4025 // to use hard-coded signal number when setting up the mask.
4026 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4027
4028 if (sigaction(SR_signum, &act, 0) == -1) {
4029 return -1;
4030 }
4031
4032 // Save signal flag
4033 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4034 return 0;
4035 }
4036
4037 static int sr_notify(OSThread* osthread) {
4038 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4039 assert_status(status == 0, status, "pthread_kill");
4040 return status;
4041 }
4042
4043 // "Randomly" selected value for how long we want to spin
4044 // before bailing out on suspending a thread, also how often
4045 // we send a signal to a thread we want to resume
4046 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4047 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4048
4049 // returns true on success and false on error - really an error is fatal
4050 // but this seems the normal response to library errors
4051 static bool do_suspend(OSThread* osthread) {
4052 assert(osthread->sr.is_running(), "thread should be running");
4053 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4054
4055 // mark as suspended and send signal
4056 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4057 // failed to switch, state wasn't running?
4058 ShouldNotReachHere();
4059 return false;
4060 }
4061
4062 if (sr_notify(osthread) != 0) {
4063 ShouldNotReachHere();
4064 }
4065
4066 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4067 while (true) {
4068 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4069 break;
4070 } else {
4071 // timeout
4072 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4073 if (cancelled == os::SuspendResume::SR_RUNNING) {
4074 return false;
4075 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4076 // make sure that we consume the signal on the semaphore as well
4077 sr_semaphore.wait();
4078 break;
4079 } else {
4080 ShouldNotReachHere();
4081 return false;
4082 }
4083 }
4084 }
4085
4086 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4087 return true;
4088 }
4089
4090 static void do_resume(OSThread* osthread) {
4091 assert(osthread->sr.is_suspended(), "thread should be suspended");
4092 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4093
4094 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4095 // failed to switch to WAKEUP_REQUEST
4096 ShouldNotReachHere();
4097 return;
4098 }
4099
4100 while (true) {
4101 if (sr_notify(osthread) == 0) {
4102 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4103 if (osthread->sr.is_running()) {
4104 return;
4105 }
4106 }
4107 } else {
4108 ShouldNotReachHere();
4109 }
4110 }
4111
4112 guarantee(osthread->sr.is_running(), "Must be running!");
4113 }
4114
4115 ///////////////////////////////////////////////////////////////////////////////////
4116 // signal handling (except suspend/resume)
4117
4118 // This routine may be used by user applications as a "hook" to catch signals.
4119 // The user-defined signal handler must pass unrecognized signals to this
4120 // routine, and if it returns true (non-zero), then the signal handler must
4121 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4122 // routine will never retun false (zero), but instead will execute a VM panic
4123 // routine kill the process.
4124 //
4125 // If this routine returns false, it is OK to call it again. This allows
4126 // the user-defined signal handler to perform checks either before or after
4127 // the VM performs its own checks. Naturally, the user code would be making
4128 // a serious error if it tried to handle an exception (such as a null check
4129 // or breakpoint) that the VM was generating for its own correct operation.
4130 //
4131 // This routine may recognize any of the following kinds of signals:
4132 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4133 // It should be consulted by handlers for any of those signals.
4134 //
4135 // The caller of this routine must pass in the three arguments supplied
4136 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4137 // field of the structure passed to sigaction(). This routine assumes that
4138 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4139 //
4140 // Note that the VM will print warnings if it detects conflicting signal
4141 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4142 //
4143 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4144 siginfo_t* siginfo,
4145 void* ucontext,
4146 int abort_if_unrecognized);
4147
4148 void signalHandler(int sig, siginfo_t* info, void* uc) {
4149 assert(info != NULL && uc != NULL, "it must be old kernel");
4150 int orig_errno = errno; // Preserve errno value over signal handler.
4151 JVM_handle_linux_signal(sig, info, uc, true);
4152 errno = orig_errno;
4153 }
4154
4155
4156 // This boolean allows users to forward their own non-matching signals
4157 // to JVM_handle_linux_signal, harmlessly.
4158 bool os::Linux::signal_handlers_are_installed = false;
4159
4160 // For signal-chaining
4161 struct sigaction os::Linux::sigact[MAXSIGNUM];
4162 unsigned int os::Linux::sigs = 0;
4163 bool os::Linux::libjsig_is_loaded = false;
4164 typedef struct sigaction *(*get_signal_t)(int);
4165 get_signal_t os::Linux::get_signal_action = NULL;
4166
4167 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4168 struct sigaction *actp = NULL;
4169
4170 if (libjsig_is_loaded) {
4171 // Retrieve the old signal handler from libjsig
4172 actp = (*get_signal_action)(sig);
4173 }
4174 if (actp == NULL) {
4175 // Retrieve the preinstalled signal handler from jvm
4176 actp = get_preinstalled_handler(sig);
4177 }
4178
4179 return actp;
4180 }
4181
4182 static bool call_chained_handler(struct sigaction *actp, int sig,
4183 siginfo_t *siginfo, void *context) {
4184 // Call the old signal handler
4185 if (actp->sa_handler == SIG_DFL) {
4186 // It's more reasonable to let jvm treat it as an unexpected exception
4187 // instead of taking the default action.
4188 return false;
4189 } else if (actp->sa_handler != SIG_IGN) {
4190 if ((actp->sa_flags & SA_NODEFER) == 0) {
4191 // automaticlly block the signal
4192 sigaddset(&(actp->sa_mask), sig);
4193 }
4194
4195 sa_handler_t hand;
4196 sa_sigaction_t sa;
4197 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4198 // retrieve the chained handler
4199 if (siginfo_flag_set) {
4200 sa = actp->sa_sigaction;
4201 } else {
4202 hand = actp->sa_handler;
4203 }
4204
4205 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4206 actp->sa_handler = SIG_DFL;
4207 }
4208
4209 // try to honor the signal mask
4210 sigset_t oset;
4211 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4212
4213 // call into the chained handler
4214 if (siginfo_flag_set) {
4215 (*sa)(sig, siginfo, context);
4216 } else {
4217 (*hand)(sig);
4218 }
4219
4220 // restore the signal mask
4221 pthread_sigmask(SIG_SETMASK, &oset, 0);
4222 }
4223 // Tell jvm's signal handler the signal is taken care of.
4224 return true;
4225 }
4226
4227 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4228 bool chained = false;
4229 // signal-chaining
4230 if (UseSignalChaining) {
4231 struct sigaction *actp = get_chained_signal_action(sig);
4232 if (actp != NULL) {
4233 chained = call_chained_handler(actp, sig, siginfo, context);
4234 }
4235 }
4236 return chained;
4237 }
4238
4239 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4240 if ((((unsigned int)1 << sig) & sigs) != 0) {
4241 return &sigact[sig];
4242 }
4243 return NULL;
4244 }
4245
4246 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4247 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4248 sigact[sig] = oldAct;
4249 sigs |= (unsigned int)1 << sig;
4250 }
4251
4252 // for diagnostic
4253 int os::Linux::sigflags[MAXSIGNUM];
4254
4255 int os::Linux::get_our_sigflags(int sig) {
4256 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4257 return sigflags[sig];
4258 }
4259
4260 void os::Linux::set_our_sigflags(int sig, int flags) {
4261 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4262 sigflags[sig] = flags;
4263 }
4264
4265 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4266 // Check for overwrite.
4267 struct sigaction oldAct;
4268 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4269
4270 void* oldhand = oldAct.sa_sigaction
4271 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4272 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4273 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4274 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4275 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4276 if (AllowUserSignalHandlers || !set_installed) {
4277 // Do not overwrite; user takes responsibility to forward to us.
4278 return;
4279 } else if (UseSignalChaining) {
4280 // save the old handler in jvm
4281 save_preinstalled_handler(sig, oldAct);
4282 // libjsig also interposes the sigaction() call below and saves the
4283 // old sigaction on it own.
4284 } else {
4285 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4286 "%#lx for signal %d.", (long)oldhand, sig));
4287 }
4288 }
4289
4290 struct sigaction sigAct;
4291 sigfillset(&(sigAct.sa_mask));
4292 sigAct.sa_handler = SIG_DFL;
4293 if (!set_installed) {
4294 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4295 } else {
4296 sigAct.sa_sigaction = signalHandler;
4297 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4298 }
4299 // Save flags, which are set by ours
4300 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4301 sigflags[sig] = sigAct.sa_flags;
4302
4303 int ret = sigaction(sig, &sigAct, &oldAct);
4304 assert(ret == 0, "check");
4305
4306 void* oldhand2 = oldAct.sa_sigaction
4307 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4308 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4309 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4310 }
4311
4312 // install signal handlers for signals that HotSpot needs to
4313 // handle in order to support Java-level exception handling.
4314
4315 void os::Linux::install_signal_handlers() {
4316 if (!signal_handlers_are_installed) {
4317 signal_handlers_are_installed = true;
4318
4319 // signal-chaining
4320 typedef void (*signal_setting_t)();
4321 signal_setting_t begin_signal_setting = NULL;
4322 signal_setting_t end_signal_setting = NULL;
4323 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4324 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4325 if (begin_signal_setting != NULL) {
4326 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4327 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4328 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4329 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4330 libjsig_is_loaded = true;
4331 assert(UseSignalChaining, "should enable signal-chaining");
4332 }
4333 if (libjsig_is_loaded) {
4334 // Tell libjsig jvm is setting signal handlers
4335 (*begin_signal_setting)();
4336 }
4337
4338 set_signal_handler(SIGSEGV, true);
4339 set_signal_handler(SIGPIPE, true);
4340 set_signal_handler(SIGBUS, true);
4341 set_signal_handler(SIGILL, true);
4342 set_signal_handler(SIGFPE, true);
4343 #if defined(PPC64)
4344 set_signal_handler(SIGTRAP, true);
4345 #endif
4346 set_signal_handler(SIGXFSZ, true);
4347
4348 if (libjsig_is_loaded) {
4349 // Tell libjsig jvm finishes setting signal handlers
4350 (*end_signal_setting)();
4351 }
4352
4353 // We don't activate signal checker if libjsig is in place, we trust ourselves
4354 // and if UserSignalHandler is installed all bets are off.
4355 // Log that signal checking is off only if -verbose:jni is specified.
4356 if (CheckJNICalls) {
4357 if (libjsig_is_loaded) {
4358 if (PrintJNIResolving) {
4359 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4360 }
4361 check_signals = false;
4362 }
4363 if (AllowUserSignalHandlers) {
4364 if (PrintJNIResolving) {
4365 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4366 }
4367 check_signals = false;
4368 }
4369 }
4370 }
4371 }
4372
4373 // This is the fastest way to get thread cpu time on Linux.
4374 // Returns cpu time (user+sys) for any thread, not only for current.
4375 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4376 // It might work on 2.6.10+ with a special kernel/glibc patch.
4377 // For reference, please, see IEEE Std 1003.1-2004:
4378 // http://www.unix.org/single_unix_specification
4379
4380 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4381 struct timespec tp;
4382 int rc = os::Linux::clock_gettime(clockid, &tp);
4383 assert(rc == 0, "clock_gettime is expected to return 0 code");
4384
4385 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4386 }
4387
4388 /////
4389 // glibc on Linux platform uses non-documented flag
4390 // to indicate, that some special sort of signal
4391 // trampoline is used.
4392 // We will never set this flag, and we should
4393 // ignore this flag in our diagnostic
4394 #ifdef SIGNIFICANT_SIGNAL_MASK
4395 #undef SIGNIFICANT_SIGNAL_MASK
4396 #endif
4397 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4398
4399 static const char* get_signal_handler_name(address handler,
4400 char* buf, int buflen) {
4401 int offset;
4402 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4403 if (found) {
4404 // skip directory names
4405 const char *p1, *p2;
4406 p1 = buf;
4407 size_t len = strlen(os::file_separator());
4408 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4409 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4410 } else {
4411 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4412 }
4413 return buf;
4414 }
4415
4416 static void print_signal_handler(outputStream* st, int sig,
4417 char* buf, size_t buflen) {
4418 struct sigaction sa;
4419
4420 sigaction(sig, NULL, &sa);
4421
4422 // See comment for SIGNIFICANT_SIGNAL_MASK define
4423 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4424
4425 st->print("%s: ", os::exception_name(sig, buf, buflen));
4426
4427 address handler = (sa.sa_flags & SA_SIGINFO)
4428 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4429 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4430
4431 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4432 st->print("SIG_DFL");
4433 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4434 st->print("SIG_IGN");
4435 } else {
4436 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4437 }
4438
4439 st->print(", sa_mask[0]=");
4440 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4441
4442 address rh = VMError::get_resetted_sighandler(sig);
4443 // May be, handler was resetted by VMError?
4444 if (rh != NULL) {
4445 handler = rh;
4446 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4447 }
4448
4449 st->print(", sa_flags=");
4450 os::Posix::print_sa_flags(st, sa.sa_flags);
4451
4452 // Check: is it our handler?
4453 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4454 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4455 // It is our signal handler
4456 // check for flags, reset system-used one!
4457 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4458 st->print(
4459 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4460 os::Linux::get_our_sigflags(sig));
4461 }
4462 }
4463 st->cr();
4464 }
4465
4466
4467 #define DO_SIGNAL_CHECK(sig) \
4468 do { \
4469 if (!sigismember(&check_signal_done, sig)) { \
4470 os::Linux::check_signal_handler(sig); \
4471 } \
4472 } while (0)
4473
4474 // This method is a periodic task to check for misbehaving JNI applications
4475 // under CheckJNI, we can add any periodic checks here
4476
4477 void os::run_periodic_checks() {
4478 if (check_signals == false) return;
4479
4480 // SEGV and BUS if overridden could potentially prevent
4481 // generation of hs*.log in the event of a crash, debugging
4482 // such a case can be very challenging, so we absolutely
4483 // check the following for a good measure:
4484 DO_SIGNAL_CHECK(SIGSEGV);
4485 DO_SIGNAL_CHECK(SIGILL);
4486 DO_SIGNAL_CHECK(SIGFPE);
4487 DO_SIGNAL_CHECK(SIGBUS);
4488 DO_SIGNAL_CHECK(SIGPIPE);
4489 DO_SIGNAL_CHECK(SIGXFSZ);
4490 #if defined(PPC64)
4491 DO_SIGNAL_CHECK(SIGTRAP);
4492 #endif
4493
4494 // ReduceSignalUsage allows the user to override these handlers
4495 // see comments at the very top and jvm_solaris.h
4496 if (!ReduceSignalUsage) {
4497 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4498 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4499 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4500 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4501 }
4502
4503 DO_SIGNAL_CHECK(SR_signum);
4504 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4505 }
4506
4507 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4508
4509 static os_sigaction_t os_sigaction = NULL;
4510
4511 void os::Linux::check_signal_handler(int sig) {
4512 char buf[O_BUFLEN];
4513 address jvmHandler = NULL;
4514
4515
4516 struct sigaction act;
4517 if (os_sigaction == NULL) {
4518 // only trust the default sigaction, in case it has been interposed
4519 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4520 if (os_sigaction == NULL) return;
4521 }
4522
4523 os_sigaction(sig, (struct sigaction*)NULL, &act);
4524
4525
4526 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4527
4528 address thisHandler = (act.sa_flags & SA_SIGINFO)
4529 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4530 : CAST_FROM_FN_PTR(address, act.sa_handler);
4531
4532
4533 switch (sig) {
4534 case SIGSEGV:
4535 case SIGBUS:
4536 case SIGFPE:
4537 case SIGPIPE:
4538 case SIGILL:
4539 case SIGXFSZ:
4540 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4541 break;
4542
4543 case SHUTDOWN1_SIGNAL:
4544 case SHUTDOWN2_SIGNAL:
4545 case SHUTDOWN3_SIGNAL:
4546 case BREAK_SIGNAL:
4547 jvmHandler = (address)user_handler();
4548 break;
4549
4550 case INTERRUPT_SIGNAL:
4551 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4552 break;
4553
4554 default:
4555 if (sig == SR_signum) {
4556 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4557 } else {
4558 return;
4559 }
4560 break;
4561 }
4562
4563 if (thisHandler != jvmHandler) {
4564 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4565 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4566 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4567 // No need to check this sig any longer
4568 sigaddset(&check_signal_done, sig);
4569 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4570 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4571 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4572 exception_name(sig, buf, O_BUFLEN));
4573 }
4574 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4575 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4576 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4577 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4578 // No need to check this sig any longer
4579 sigaddset(&check_signal_done, sig);
4580 }
4581
4582 // Dump all the signal
4583 if (sigismember(&check_signal_done, sig)) {
4584 print_signal_handlers(tty, buf, O_BUFLEN);
4585 }
4586 }
4587
4588 extern void report_error(char* file_name, int line_no, char* title,
4589 char* format, ...);
4590
4591 extern bool signal_name(int signo, char* buf, size_t len);
4592
4593 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4594 if (0 < exception_code && exception_code <= SIGRTMAX) {
4595 // signal
4596 if (!signal_name(exception_code, buf, size)) {
4597 jio_snprintf(buf, size, "SIG%d", exception_code);
4598 }
4599 return buf;
4600 } else {
4601 return NULL;
4602 }
4603 }
4604
4605 // this is called _before_ the most of global arguments have been parsed
4606 void os::init(void) {
4607 char dummy; // used to get a guess on initial stack address
4608 // first_hrtime = gethrtime();
4609
4610 // With LinuxThreads the JavaMain thread pid (primordial thread)
4611 // is different than the pid of the java launcher thread.
4612 // So, on Linux, the launcher thread pid is passed to the VM
4613 // via the sun.java.launcher.pid property.
4614 // Use this property instead of getpid() if it was correctly passed.
4615 // See bug 6351349.
4616 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4617
4618 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4619
4620 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4621
4622 init_random(1234567);
4623
4624 ThreadCritical::initialize();
4625
4626 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4627 if (Linux::page_size() == -1) {
4628 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4629 strerror(errno)));
4630 }
4631 init_page_sizes((size_t) Linux::page_size());
4632
4633 Linux::initialize_system_info();
4634
4635 // main_thread points to the aboriginal thread
4636 Linux::_main_thread = pthread_self();
4637
4638 Linux::clock_init();
4639 initial_time_count = javaTimeNanos();
4640
4641 // pthread_condattr initialization for monotonic clock
4642 int status;
4643 pthread_condattr_t* _condattr = os::Linux::condAttr();
4644 if ((status = pthread_condattr_init(_condattr)) != 0) {
4645 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4646 }
4647 // Only set the clock if CLOCK_MONOTONIC is available
4648 if (os::supports_monotonic_clock()) {
4649 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4650 if (status == EINVAL) {
4651 warning("Unable to use monotonic clock with relative timed-waits" \
4652 " - changes to the time-of-day clock may have adverse affects");
4653 } else {
4654 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4655 }
4656 }
4657 }
4658 // else it defaults to CLOCK_REALTIME
4659
4660 pthread_mutex_init(&dl_mutex, NULL);
4661
4662 // If the pagesize of the VM is greater than 8K determine the appropriate
4663 // number of initial guard pages. The user can change this with the
4664 // command line arguments, if needed.
4665 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4666 StackYellowPages = 1;
4667 StackRedPages = 1;
4668 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4669 }
4670
4671 // retrieve entry point for pthread_setname_np
4672 Linux::_pthread_setname_np =
4673 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4674
4675 }
4676
4677 // To install functions for atexit system call
4678 extern "C" {
4679 static void perfMemory_exit_helper() {
4680 perfMemory_exit();
4681 }
4682 }
4683
4684 // this is called _after_ the global arguments have been parsed
4685 jint os::init_2(void) {
4686 Linux::fast_thread_clock_init();
4687
4688 // Allocate a single page and mark it as readable for safepoint polling
4689 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4690 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4691
4692 os::set_polling_page(polling_page);
4693
4694 #ifndef PRODUCT
4695 if (Verbose && PrintMiscellaneous) {
4696 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4697 (intptr_t)polling_page);
4698 }
4699 #endif
4700
4701 if (!UseMembar) {
4702 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4703 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4704 os::set_memory_serialize_page(mem_serialize_page);
4705
4706 #ifndef PRODUCT
4707 if (Verbose && PrintMiscellaneous) {
4708 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4709 (intptr_t)mem_serialize_page);
4710 }
4711 #endif
4712 }
4713
4714 // initialize suspend/resume support - must do this before signal_sets_init()
4715 if (SR_initialize() != 0) {
4716 perror("SR_initialize failed");
4717 return JNI_ERR;
4718 }
4719
4720 Linux::signal_sets_init();
4721 Linux::install_signal_handlers();
4722
4723 // Check minimum allowable stack size for thread creation and to initialize
4724 // the java system classes, including StackOverflowError - depends on page
4725 // size. Add a page for compiler2 recursion in main thread.
4726 // Add in 2*BytesPerWord times page size to account for VM stack during
4727 // class initialization depending on 32 or 64 bit VM.
4728 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4729 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4730 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4731
4732 size_t threadStackSizeInBytes = ThreadStackSize * K;
4733 if (threadStackSizeInBytes != 0 &&
4734 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4735 tty->print_cr("\nThe stack size specified is too small, "
4736 "Specify at least %dk",
4737 os::Linux::min_stack_allowed/ K);
4738 return JNI_ERR;
4739 }
4740
4741 // Make the stack size a multiple of the page size so that
4742 // the yellow/red zones can be guarded.
4743 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4744 vm_page_size()));
4745
4746 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4747
4748 #if defined(IA32)
4749 workaround_expand_exec_shield_cs_limit();
4750 #endif
4751
4752 Linux::libpthread_init();
4753 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4754 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4755 Linux::glibc_version(), Linux::libpthread_version(),
4756 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4757 }
4758
4759 if (UseNUMA) {
4760 if (!Linux::libnuma_init()) {
4761 UseNUMA = false;
4762 } else {
4763 if ((Linux::numa_max_node() < 1)) {
4764 // There's only one node(they start from 0), disable NUMA.
4765 UseNUMA = false;
4766 }
4767 }
4768 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4769 // we can make the adaptive lgrp chunk resizing work. If the user specified
4770 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4771 // disable adaptive resizing.
4772 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4773 if (FLAG_IS_DEFAULT(UseNUMA)) {
4774 UseNUMA = false;
4775 } else {
4776 if (FLAG_IS_DEFAULT(UseLargePages) &&
4777 FLAG_IS_DEFAULT(UseSHM) &&
4778 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4779 UseLargePages = false;
4780 } else {
4781 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4782 UseAdaptiveSizePolicy = false;
4783 UseAdaptiveNUMAChunkSizing = false;
4784 }
4785 }
4786 }
4787 if (!UseNUMA && ForceNUMA) {
4788 UseNUMA = true;
4789 }
4790 }
4791
4792 if (MaxFDLimit) {
4793 // set the number of file descriptors to max. print out error
4794 // if getrlimit/setrlimit fails but continue regardless.
4795 struct rlimit nbr_files;
4796 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4797 if (status != 0) {
4798 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4799 perror("os::init_2 getrlimit failed");
4800 }
4801 } else {
4802 nbr_files.rlim_cur = nbr_files.rlim_max;
4803 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4804 if (status != 0) {
4805 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4806 perror("os::init_2 setrlimit failed");
4807 }
4808 }
4809 }
4810 }
4811
4812 // Initialize lock used to serialize thread creation (see os::create_thread)
4813 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4814
4815 // at-exit methods are called in the reverse order of their registration.
4816 // atexit functions are called on return from main or as a result of a
4817 // call to exit(3C). There can be only 32 of these functions registered
4818 // and atexit() does not set errno.
4819
4820 if (PerfAllowAtExitRegistration) {
4821 // only register atexit functions if PerfAllowAtExitRegistration is set.
4822 // atexit functions can be delayed until process exit time, which
4823 // can be problematic for embedded VM situations. Embedded VMs should
4824 // call DestroyJavaVM() to assure that VM resources are released.
4825
4826 // note: perfMemory_exit_helper atexit function may be removed in
4827 // the future if the appropriate cleanup code can be added to the
4828 // VM_Exit VMOperation's doit method.
4829 if (atexit(perfMemory_exit_helper) != 0) {
4830 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4831 }
4832 }
4833
4834 // initialize thread priority policy
4835 prio_init();
4836
4837 return JNI_OK;
4838 }
4839
4840 // Mark the polling page as unreadable
4841 void os::make_polling_page_unreadable(void) {
4842 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4843 fatal("Could not disable polling page");
4844 }
4845 }
4846
4847 // Mark the polling page as readable
4848 void os::make_polling_page_readable(void) {
4849 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4850 fatal("Could not enable polling page");
4851 }
4852 }
4853
4854 int os::active_processor_count() {
4855 // Linux doesn't yet have a (official) notion of processor sets,
4856 // so just return the number of online processors.
4857 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4858 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4859 return online_cpus;
4860 }
4861
4862 void os::set_native_thread_name(const char *name) {
4863 if (Linux::_pthread_setname_np) {
4864 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4865 snprintf(buf, sizeof(buf), "%s", name);
4866 buf[sizeof(buf) - 1] = '\0';
4867 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4868 // ERANGE should not happen; all other errors should just be ignored.
4869 assert(rc != ERANGE, "pthread_setname_np failed");
4870 }
4871 }
4872
4873 bool os::distribute_processes(uint length, uint* distribution) {
4874 // Not yet implemented.
4875 return false;
4876 }
4877
4878 bool os::bind_to_processor(uint processor_id) {
4879 // Not yet implemented.
4880 return false;
4881 }
4882
4883 ///
4884
4885 void os::SuspendedThreadTask::internal_do_task() {
4886 if (do_suspend(_thread->osthread())) {
4887 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4888 do_task(context);
4889 do_resume(_thread->osthread());
4890 }
4891 }
4892
4893 class PcFetcher : public os::SuspendedThreadTask {
4894 public:
4895 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4896 ExtendedPC result();
4897 protected:
4898 void do_task(const os::SuspendedThreadTaskContext& context);
4899 private:
4900 ExtendedPC _epc;
4901 };
4902
4903 ExtendedPC PcFetcher::result() {
4904 guarantee(is_done(), "task is not done yet.");
4905 return _epc;
4906 }
4907
4908 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4909 Thread* thread = context.thread();
4910 OSThread* osthread = thread->osthread();
4911 if (osthread->ucontext() != NULL) {
4912 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4913 } else {
4914 // NULL context is unexpected, double-check this is the VMThread
4915 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4916 }
4917 }
4918
4919 // Suspends the target using the signal mechanism and then grabs the PC before
4920 // resuming the target. Used by the flat-profiler only
4921 ExtendedPC os::get_thread_pc(Thread* thread) {
4922 // Make sure that it is called by the watcher for the VMThread
4923 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4924 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4925
4926 PcFetcher fetcher(thread);
4927 fetcher.run();
4928 return fetcher.result();
4929 }
4930
4931 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4932 pthread_mutex_t *_mutex,
4933 const struct timespec *_abstime) {
4934 if (is_NPTL()) {
4935 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4936 } else {
4937 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4938 // word back to default 64bit precision if condvar is signaled. Java
4939 // wants 53bit precision. Save and restore current value.
4940 int fpu = get_fpu_control_word();
4941 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4942 set_fpu_control_word(fpu);
4943 return status;
4944 }
4945 }
4946
4947 ////////////////////////////////////////////////////////////////////////////////
4948 // debug support
4949
4950 bool os::find(address addr, outputStream* st) {
4951 Dl_info dlinfo;
4952 memset(&dlinfo, 0, sizeof(dlinfo));
4953 if (dladdr(addr, &dlinfo) != 0) {
4954 st->print(PTR_FORMAT ": ", addr);
4955 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4956 st->print("%s+%#x", dlinfo.dli_sname,
4957 addr - (intptr_t)dlinfo.dli_saddr);
4958 } else if (dlinfo.dli_fbase != NULL) {
4959 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4960 } else {
4961 st->print("<absolute address>");
4962 }
4963 if (dlinfo.dli_fname != NULL) {
4964 st->print(" in %s", dlinfo.dli_fname);
4965 }
4966 if (dlinfo.dli_fbase != NULL) {
4967 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4968 }
4969 st->cr();
4970
4971 if (Verbose) {
4972 // decode some bytes around the PC
4973 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4974 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4975 address lowest = (address) dlinfo.dli_sname;
4976 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4977 if (begin < lowest) begin = lowest;
4978 Dl_info dlinfo2;
4979 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4980 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4981 end = (address) dlinfo2.dli_saddr;
4982 }
4983 Disassembler::decode(begin, end, st);
4984 }
4985 return true;
4986 }
4987 return false;
4988 }
4989
4990 ////////////////////////////////////////////////////////////////////////////////
4991 // misc
4992
4993 // This does not do anything on Linux. This is basically a hook for being
4994 // able to use structured exception handling (thread-local exception filters)
4995 // on, e.g., Win32.
4996 void
4997 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4998 JavaCallArguments* args, Thread* thread) {
4999 f(value, method, args, thread);
5000 }
5001
5002 void os::print_statistics() {
5003 }
5004
5005 int os::message_box(const char* title, const char* message) {
5006 int i;
5007 fdStream err(defaultStream::error_fd());
5008 for (i = 0; i < 78; i++) err.print_raw("=");
5009 err.cr();
5010 err.print_raw_cr(title);
5011 for (i = 0; i < 78; i++) err.print_raw("-");
5012 err.cr();
5013 err.print_raw_cr(message);
5014 for (i = 0; i < 78; i++) err.print_raw("=");
5015 err.cr();
5016
5017 char buf[16];
5018 // Prevent process from exiting upon "read error" without consuming all CPU
5019 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5020
5021 return buf[0] == 'y' || buf[0] == 'Y';
5022 }
5023
5024 int os::stat(const char *path, struct stat *sbuf) {
5025 char pathbuf[MAX_PATH];
5026 if (strlen(path) > MAX_PATH - 1) {
5027 errno = ENAMETOOLONG;
5028 return -1;
5029 }
5030 os::native_path(strcpy(pathbuf, path));
5031 return ::stat(pathbuf, sbuf);
5032 }
5033
5034 bool os::check_heap(bool force) {
5035 return true;
5036 }
5037
5038 // Is a (classpath) directory empty?
5039 bool os::dir_is_empty(const char* path) {
5040 DIR *dir = NULL;
5041 struct dirent *ptr;
5042
5043 dir = opendir(path);
5044 if (dir == NULL) return true;
5045
5046 // Scan the directory
5047 bool result = true;
5048 char buf[sizeof(struct dirent) + MAX_PATH];
5049 while (result && (ptr = ::readdir(dir)) != NULL) {
5050 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5051 result = false;
5052 }
5053 }
5054 closedir(dir);
5055 return result;
5056 }
5057
5058 // This code originates from JDK's sysOpen and open64_w
5059 // from src/solaris/hpi/src/system_md.c
5060
5061 int os::open(const char *path, int oflag, int mode) {
5062 if (strlen(path) > MAX_PATH - 1) {
5063 errno = ENAMETOOLONG;
5064 return -1;
5065 }
5066
5067 // All file descriptors that are opened in the Java process and not
5068 // specifically destined for a subprocess should have the close-on-exec
5069 // flag set. If we don't set it, then careless 3rd party native code
5070 // might fork and exec without closing all appropriate file descriptors
5071 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5072 // turn might:
5073 //
5074 // - cause end-of-file to fail to be detected on some file
5075 // descriptors, resulting in mysterious hangs, or
5076 //
5077 // - might cause an fopen in the subprocess to fail on a system
5078 // suffering from bug 1085341.
5079 //
5080 // (Yes, the default setting of the close-on-exec flag is a Unix
5081 // design flaw)
5082 //
5083 // See:
5084 // 1085341: 32-bit stdio routines should support file descriptors >255
5085 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5086 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5087 //
5088 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5089 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5090 // because it saves a system call and removes a small window where the flag
5091 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5092 // and we fall back to using FD_CLOEXEC (see below).
5093 #ifdef O_CLOEXEC
5094 oflag |= O_CLOEXEC;
5095 #endif
5096
5097 int fd = ::open64(path, oflag, mode);
5098 if (fd == -1) return -1;
5099
5100 //If the open succeeded, the file might still be a directory
5101 {
5102 struct stat64 buf64;
5103 int ret = ::fstat64(fd, &buf64);
5104 int st_mode = buf64.st_mode;
5105
5106 if (ret != -1) {
5107 if ((st_mode & S_IFMT) == S_IFDIR) {
5108 errno = EISDIR;
5109 ::close(fd);
5110 return -1;
5111 }
5112 } else {
5113 ::close(fd);
5114 return -1;
5115 }
5116 }
5117
5118 #ifdef FD_CLOEXEC
5119 // Validate that the use of the O_CLOEXEC flag on open above worked.
5120 // With recent kernels, we will perform this check exactly once.
5121 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5122 if (!O_CLOEXEC_is_known_to_work) {
5123 int flags = ::fcntl(fd, F_GETFD);
5124 if (flags != -1) {
5125 if ((flags & FD_CLOEXEC) != 0)
5126 O_CLOEXEC_is_known_to_work = 1;
5127 else
5128 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5129 }
5130 }
5131 #endif
5132
5133 return fd;
5134 }
5135
5136
5137 // create binary file, rewriting existing file if required
5138 int os::create_binary_file(const char* path, bool rewrite_existing) {
5139 int oflags = O_WRONLY | O_CREAT;
5140 if (!rewrite_existing) {
5141 oflags |= O_EXCL;
5142 }
5143 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5144 }
5145
5146 // return current position of file pointer
5147 jlong os::current_file_offset(int fd) {
5148 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5149 }
5150
5151 // move file pointer to the specified offset
5152 jlong os::seek_to_file_offset(int fd, jlong offset) {
5153 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5154 }
5155
5156 // This code originates from JDK's sysAvailable
5157 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5158
5159 int os::available(int fd, jlong *bytes) {
5160 jlong cur, end;
5161 int mode;
5162 struct stat64 buf64;
5163
5164 if (::fstat64(fd, &buf64) >= 0) {
5165 mode = buf64.st_mode;
5166 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5167 // XXX: is the following call interruptible? If so, this might
5168 // need to go through the INTERRUPT_IO() wrapper as for other
5169 // blocking, interruptible calls in this file.
5170 int n;
5171 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5172 *bytes = n;
5173 return 1;
5174 }
5175 }
5176 }
5177 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5178 return 0;
5179 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5180 return 0;
5181 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5182 return 0;
5183 }
5184 *bytes = end - cur;
5185 return 1;
5186 }
5187
5188 // Map a block of memory.
5189 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5190 char *addr, size_t bytes, bool read_only,
5191 bool allow_exec) {
5192 int prot;
5193 int flags = MAP_PRIVATE;
5194
5195 if (read_only) {
5196 prot = PROT_READ;
5197 } else {
5198 prot = PROT_READ | PROT_WRITE;
5199 }
5200
5201 if (allow_exec) {
5202 prot |= PROT_EXEC;
5203 }
5204
5205 if (addr != NULL) {
5206 flags |= MAP_FIXED;
5207 }
5208
5209 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5210 fd, file_offset);
5211 if (mapped_address == MAP_FAILED) {
5212 return NULL;
5213 }
5214 return mapped_address;
5215 }
5216
5217
5218 // Remap a block of memory.
5219 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5220 char *addr, size_t bytes, bool read_only,
5221 bool allow_exec) {
5222 // same as map_memory() on this OS
5223 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5224 allow_exec);
5225 }
5226
5227
5228 // Unmap a block of memory.
5229 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5230 return munmap(addr, bytes) == 0;
5231 }
5232
5233 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5234
5235 static clockid_t thread_cpu_clockid(Thread* thread) {
5236 pthread_t tid = thread->osthread()->pthread_id();
5237 clockid_t clockid;
5238
5239 // Get thread clockid
5240 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5241 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5242 return clockid;
5243 }
5244
5245 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5246 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5247 // of a thread.
5248 //
5249 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5250 // the fast estimate available on the platform.
5251
5252 jlong os::current_thread_cpu_time() {
5253 if (os::Linux::supports_fast_thread_cpu_time()) {
5254 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5255 } else {
5256 // return user + sys since the cost is the same
5257 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5258 }
5259 }
5260
5261 jlong os::thread_cpu_time(Thread* thread) {
5262 // consistent with what current_thread_cpu_time() returns
5263 if (os::Linux::supports_fast_thread_cpu_time()) {
5264 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5265 } else {
5266 return slow_thread_cpu_time(thread, true /* user + sys */);
5267 }
5268 }
5269
5270 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5271 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5272 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5273 } else {
5274 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5275 }
5276 }
5277
5278 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5279 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5280 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5281 } else {
5282 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5283 }
5284 }
5285
5286 // -1 on error.
5287 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5288 pid_t tid = thread->osthread()->thread_id();
5289 char *s;
5290 char stat[2048];
5291 int statlen;
5292 char proc_name[64];
5293 int count;
5294 long sys_time, user_time;
5295 char cdummy;
5296 int idummy;
5297 long ldummy;
5298 FILE *fp;
5299
5300 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5301 fp = fopen(proc_name, "r");
5302 if (fp == NULL) return -1;
5303 statlen = fread(stat, 1, 2047, fp);
5304 stat[statlen] = '\0';
5305 fclose(fp);
5306
5307 // Skip pid and the command string. Note that we could be dealing with
5308 // weird command names, e.g. user could decide to rename java launcher
5309 // to "java 1.4.2 :)", then the stat file would look like
5310 // 1234 (java 1.4.2 :)) R ... ...
5311 // We don't really need to know the command string, just find the last
5312 // occurrence of ")" and then start parsing from there. See bug 4726580.
5313 s = strrchr(stat, ')');
5314 if (s == NULL) return -1;
5315
5316 // Skip blank chars
5317 do { s++; } while (s && isspace(*s));
5318
5319 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5320 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5321 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5322 &user_time, &sys_time);
5323 if (count != 13) return -1;
5324 if (user_sys_cpu_time) {
5325 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5326 } else {
5327 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5328 }
5329 }
5330
5331 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5332 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5333 info_ptr->may_skip_backward = false; // elapsed time not wall time
5334 info_ptr->may_skip_forward = false; // elapsed time not wall time
5335 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5336 }
5337
5338 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5339 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5340 info_ptr->may_skip_backward = false; // elapsed time not wall time
5341 info_ptr->may_skip_forward = false; // elapsed time not wall time
5342 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5343 }
5344
5345 bool os::is_thread_cpu_time_supported() {
5346 return true;
5347 }
5348
5349 // System loadavg support. Returns -1 if load average cannot be obtained.
5350 // Linux doesn't yet have a (official) notion of processor sets,
5351 // so just return the system wide load average.
5352 int os::loadavg(double loadavg[], int nelem) {
5353 return ::getloadavg(loadavg, nelem);
5354 }
5355
5356 void os::pause() {
5357 char filename[MAX_PATH];
5358 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5359 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5360 } else {
5361 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5362 }
5363
5364 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5365 if (fd != -1) {
5366 struct stat buf;
5367 ::close(fd);
5368 while (::stat(filename, &buf) == 0) {
5369 (void)::poll(NULL, 0, 100);
5370 }
5371 } else {
5372 jio_fprintf(stderr,
5373 "Could not open pause file '%s', continuing immediately.\n", filename);
5374 }
5375 }
5376
5377
5378 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5379 // comment paragraphs are worth repeating here:
5380 //
5381 // Assumption:
5382 // Only one parker can exist on an event, which is why we allocate
5383 // them per-thread. Multiple unparkers can coexist.
5384 //
5385 // _Event serves as a restricted-range semaphore.
5386 // -1 : thread is blocked, i.e. there is a waiter
5387 // 0 : neutral: thread is running or ready,
5388 // could have been signaled after a wait started
5389 // 1 : signaled - thread is running or ready
5390 //
5391 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5392 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5393 // For specifics regarding the bug see GLIBC BUGID 261237 :
5394 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5395 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5396 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5397 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5398 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5399 // and monitorenter when we're using 1-0 locking. All those operations may result in
5400 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5401 // of libpthread avoids the problem, but isn't practical.
5402 //
5403 // Possible remedies:
5404 //
5405 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5406 // This is palliative and probabilistic, however. If the thread is preempted
5407 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5408 // than the minimum period may have passed, and the abstime may be stale (in the
5409 // past) resultin in a hang. Using this technique reduces the odds of a hang
5410 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5411 //
5412 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5413 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5414 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5415 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5416 // thread.
5417 //
5418 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5419 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5420 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5421 // This also works well. In fact it avoids kernel-level scalability impediments
5422 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5423 // timers in a graceful fashion.
5424 //
5425 // 4. When the abstime value is in the past it appears that control returns
5426 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5427 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5428 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5429 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5430 // It may be possible to avoid reinitialization by checking the return
5431 // value from pthread_cond_timedwait(). In addition to reinitializing the
5432 // condvar we must establish the invariant that cond_signal() is only called
5433 // within critical sections protected by the adjunct mutex. This prevents
5434 // cond_signal() from "seeing" a condvar that's in the midst of being
5435 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5436 // desirable signal-after-unlock optimization that avoids futile context switching.
5437 //
5438 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5439 // structure when a condvar is used or initialized. cond_destroy() would
5440 // release the helper structure. Our reinitialize-after-timedwait fix
5441 // put excessive stress on malloc/free and locks protecting the c-heap.
5442 //
5443 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5444 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5445 // and only enabling the work-around for vulnerable environments.
5446
5447 // utility to compute the abstime argument to timedwait:
5448 // millis is the relative timeout time
5449 // abstime will be the absolute timeout time
5450 // TODO: replace compute_abstime() with unpackTime()
5451
5452 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5453 if (millis < 0) millis = 0;
5454
5455 jlong seconds = millis / 1000;
5456 millis %= 1000;
5457 if (seconds > 50000000) { // see man cond_timedwait(3T)
5458 seconds = 50000000;
5459 }
5460
5461 if (os::supports_monotonic_clock()) {
5462 struct timespec now;
5463 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5464 assert_status(status == 0, status, "clock_gettime");
5465 abstime->tv_sec = now.tv_sec + seconds;
5466 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5467 if (nanos >= NANOSECS_PER_SEC) {
5468 abstime->tv_sec += 1;
5469 nanos -= NANOSECS_PER_SEC;
5470 }
5471 abstime->tv_nsec = nanos;
5472 } else {
5473 struct timeval now;
5474 int status = gettimeofday(&now, NULL);
5475 assert(status == 0, "gettimeofday");
5476 abstime->tv_sec = now.tv_sec + seconds;
5477 long usec = now.tv_usec + millis * 1000;
5478 if (usec >= 1000000) {
5479 abstime->tv_sec += 1;
5480 usec -= 1000000;
5481 }
5482 abstime->tv_nsec = usec * 1000;
5483 }
5484 return abstime;
5485 }
5486
5487 void os::PlatformEvent::park() { // AKA "down()"
5488 // Transitions for _Event:
5489 // -1 => -1 : illegal
5490 // 1 => 0 : pass - return immediately
5491 // 0 => -1 : block; then set _Event to 0 before returning
5492
5493 // Invariant: Only the thread associated with the Event/PlatformEvent
5494 // may call park().
5495 // TODO: assert that _Assoc != NULL or _Assoc == Self
5496 assert(_nParked == 0, "invariant");
5497
5498 int v;
5499 for (;;) {
5500 v = _Event;
5501 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5502 }
5503 guarantee(v >= 0, "invariant");
5504 if (v == 0) {
5505 // Do this the hard way by blocking ...
5506 int status = pthread_mutex_lock(_mutex);
5507 assert_status(status == 0, status, "mutex_lock");
5508 guarantee(_nParked == 0, "invariant");
5509 ++_nParked;
5510 while (_Event < 0) {
5511 status = pthread_cond_wait(_cond, _mutex);
5512 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5513 // Treat this the same as if the wait was interrupted
5514 if (status == ETIME) { status = EINTR; }
5515 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5516 }
5517 --_nParked;
5518
5519 _Event = 0;
5520 status = pthread_mutex_unlock(_mutex);
5521 assert_status(status == 0, status, "mutex_unlock");
5522 // Paranoia to ensure our locked and lock-free paths interact
5523 // correctly with each other.
5524 OrderAccess::fence();
5525 }
5526 guarantee(_Event >= 0, "invariant");
5527 }
5528
5529 int os::PlatformEvent::park(jlong millis) {
5530 // Transitions for _Event:
5531 // -1 => -1 : illegal
5532 // 1 => 0 : pass - return immediately
5533 // 0 => -1 : block; then set _Event to 0 before returning
5534
5535 guarantee(_nParked == 0, "invariant");
5536
5537 int v;
5538 for (;;) {
5539 v = _Event;
5540 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5541 }
5542 guarantee(v >= 0, "invariant");
5543 if (v != 0) return OS_OK;
5544
5545 // We do this the hard way, by blocking the thread.
5546 // Consider enforcing a minimum timeout value.
5547 struct timespec abst;
5548 compute_abstime(&abst, millis);
5549
5550 int ret = OS_TIMEOUT;
5551 int status = pthread_mutex_lock(_mutex);
5552 assert_status(status == 0, status, "mutex_lock");
5553 guarantee(_nParked == 0, "invariant");
5554 ++_nParked;
5555
5556 // Object.wait(timo) will return because of
5557 // (a) notification
5558 // (b) timeout
5559 // (c) thread.interrupt
5560 //
5561 // Thread.interrupt and object.notify{All} both call Event::set.
5562 // That is, we treat thread.interrupt as a special case of notification.
5563 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5564 // We assume all ETIME returns are valid.
5565 //
5566 // TODO: properly differentiate simultaneous notify+interrupt.
5567 // In that case, we should propagate the notify to another waiter.
5568
5569 while (_Event < 0) {
5570 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5571 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5572 pthread_cond_destroy(_cond);
5573 pthread_cond_init(_cond, os::Linux::condAttr());
5574 }
5575 assert_status(status == 0 || status == EINTR ||
5576 status == ETIME || status == ETIMEDOUT,
5577 status, "cond_timedwait");
5578 if (!FilterSpuriousWakeups) break; // previous semantics
5579 if (status == ETIME || status == ETIMEDOUT) break;
5580 // We consume and ignore EINTR and spurious wakeups.
5581 }
5582 --_nParked;
5583 if (_Event >= 0) {
5584 ret = OS_OK;
5585 }
5586 _Event = 0;
5587 status = pthread_mutex_unlock(_mutex);
5588 assert_status(status == 0, status, "mutex_unlock");
5589 assert(_nParked == 0, "invariant");
5590 // Paranoia to ensure our locked and lock-free paths interact
5591 // correctly with each other.
5592 OrderAccess::fence();
5593 return ret;
5594 }
5595
5596 void os::PlatformEvent::unpark() {
5597 // Transitions for _Event:
5598 // 0 => 1 : just return
5599 // 1 => 1 : just return
5600 // -1 => either 0 or 1; must signal target thread
5601 // That is, we can safely transition _Event from -1 to either
5602 // 0 or 1.
5603 // See also: "Semaphores in Plan 9" by Mullender & Cox
5604 //
5605 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5606 // that it will take two back-to-back park() calls for the owning
5607 // thread to block. This has the benefit of forcing a spurious return
5608 // from the first park() call after an unpark() call which will help
5609 // shake out uses of park() and unpark() without condition variables.
5610
5611 if (Atomic::xchg(1, &_Event) >= 0) return;
5612
5613 // Wait for the thread associated with the event to vacate
5614 int status = pthread_mutex_lock(_mutex);
5615 assert_status(status == 0, status, "mutex_lock");
5616 int AnyWaiters = _nParked;
5617 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5618 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5619 AnyWaiters = 0;
5620 pthread_cond_signal(_cond);
5621 }
5622 status = pthread_mutex_unlock(_mutex);
5623 assert_status(status == 0, status, "mutex_unlock");
5624 if (AnyWaiters != 0) {
5625 // Note that we signal() *after* dropping the lock for "immortal" Events.
5626 // This is safe and avoids a common class of futile wakeups. In rare
5627 // circumstances this can cause a thread to return prematurely from
5628 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5629 // will simply re-test the condition and re-park itself.
5630 // This provides particular benefit if the underlying platform does not
5631 // provide wait morphing.
5632 status = pthread_cond_signal(_cond);
5633 assert_status(status == 0, status, "cond_signal");
5634 }
5635 }
5636
5637
5638 // JSR166
5639 // -------------------------------------------------------
5640
5641 // The solaris and linux implementations of park/unpark are fairly
5642 // conservative for now, but can be improved. They currently use a
5643 // mutex/condvar pair, plus a a count.
5644 // Park decrements count if > 0, else does a condvar wait. Unpark
5645 // sets count to 1 and signals condvar. Only one thread ever waits
5646 // on the condvar. Contention seen when trying to park implies that someone
5647 // is unparking you, so don't wait. And spurious returns are fine, so there
5648 // is no need to track notifications.
5649
5650 // This code is common to linux and solaris and will be moved to a
5651 // common place in dolphin.
5652 //
5653 // The passed in time value is either a relative time in nanoseconds
5654 // or an absolute time in milliseconds. Either way it has to be unpacked
5655 // into suitable seconds and nanoseconds components and stored in the
5656 // given timespec structure.
5657 // Given time is a 64-bit value and the time_t used in the timespec is only
5658 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5659 // overflow if times way in the future are given. Further on Solaris versions
5660 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5661 // number of seconds, in abstime, is less than current_time + 100,000,000.
5662 // As it will be 28 years before "now + 100000000" will overflow we can
5663 // ignore overflow and just impose a hard-limit on seconds using the value
5664 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5665 // years from "now".
5666
5667 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5668 assert(time > 0, "convertTime");
5669 time_t max_secs = 0;
5670
5671 if (!os::supports_monotonic_clock() || isAbsolute) {
5672 struct timeval now;
5673 int status = gettimeofday(&now, NULL);
5674 assert(status == 0, "gettimeofday");
5675
5676 max_secs = now.tv_sec + MAX_SECS;
5677
5678 if (isAbsolute) {
5679 jlong secs = time / 1000;
5680 if (secs > max_secs) {
5681 absTime->tv_sec = max_secs;
5682 } else {
5683 absTime->tv_sec = secs;
5684 }
5685 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5686 } else {
5687 jlong secs = time / NANOSECS_PER_SEC;
5688 if (secs >= MAX_SECS) {
5689 absTime->tv_sec = max_secs;
5690 absTime->tv_nsec = 0;
5691 } else {
5692 absTime->tv_sec = now.tv_sec + secs;
5693 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5694 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5695 absTime->tv_nsec -= NANOSECS_PER_SEC;
5696 ++absTime->tv_sec; // note: this must be <= max_secs
5697 }
5698 }
5699 }
5700 } else {
5701 // must be relative using monotonic clock
5702 struct timespec now;
5703 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5704 assert_status(status == 0, status, "clock_gettime");
5705 max_secs = now.tv_sec + MAX_SECS;
5706 jlong secs = time / NANOSECS_PER_SEC;
5707 if (secs >= MAX_SECS) {
5708 absTime->tv_sec = max_secs;
5709 absTime->tv_nsec = 0;
5710 } else {
5711 absTime->tv_sec = now.tv_sec + secs;
5712 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5713 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5714 absTime->tv_nsec -= NANOSECS_PER_SEC;
5715 ++absTime->tv_sec; // note: this must be <= max_secs
5716 }
5717 }
5718 }
5719 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5720 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5721 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5722 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5723 }
5724
5725 void Parker::park(bool isAbsolute, jlong time) {
5726 // Ideally we'd do something useful while spinning, such
5727 // as calling unpackTime().
5728
5729 // Optional fast-path check:
5730 // Return immediately if a permit is available.
5731 // We depend on Atomic::xchg() having full barrier semantics
5732 // since we are doing a lock-free update to _counter.
5733 if (Atomic::xchg(0, &_counter) > 0) return;
5734
5735 Thread* thread = Thread::current();
5736 assert(thread->is_Java_thread(), "Must be JavaThread");
5737 JavaThread *jt = (JavaThread *)thread;
5738
5739 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5740 // Check interrupt before trying to wait
5741 if (Thread::is_interrupted(thread, false)) {
5742 return;
5743 }
5744
5745 // Next, demultiplex/decode time arguments
5746 timespec absTime;
5747 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5748 return;
5749 }
5750 if (time > 0) {
5751 unpackTime(&absTime, isAbsolute, time);
5752 }
5753
5754
5755 // Enter safepoint region
5756 // Beware of deadlocks such as 6317397.
5757 // The per-thread Parker:: mutex is a classic leaf-lock.
5758 // In particular a thread must never block on the Threads_lock while
5759 // holding the Parker:: mutex. If safepoints are pending both the
5760 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5761 ThreadBlockInVM tbivm(jt);
5762
5763 // Don't wait if cannot get lock since interference arises from
5764 // unblocking. Also. check interrupt before trying wait
5765 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5766 return;
5767 }
5768
5769 int status;
5770 if (_counter > 0) { // no wait needed
5771 _counter = 0;
5772 status = pthread_mutex_unlock(_mutex);
5773 assert(status == 0, "invariant");
5774 // Paranoia to ensure our locked and lock-free paths interact
5775 // correctly with each other and Java-level accesses.
5776 OrderAccess::fence();
5777 return;
5778 }
5779
5780 #ifdef ASSERT
5781 // Don't catch signals while blocked; let the running threads have the signals.
5782 // (This allows a debugger to break into the running thread.)
5783 sigset_t oldsigs;
5784 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5785 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5786 #endif
5787
5788 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5789 jt->set_suspend_equivalent();
5790 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5791
5792 assert(_cur_index == -1, "invariant");
5793 if (time == 0) {
5794 _cur_index = REL_INDEX; // arbitrary choice when not timed
5795 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5796 } else {
5797 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5798 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5799 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5800 pthread_cond_destroy(&_cond[_cur_index]);
5801 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5802 }
5803 }
5804 _cur_index = -1;
5805 assert_status(status == 0 || status == EINTR ||
5806 status == ETIME || status == ETIMEDOUT,
5807 status, "cond_timedwait");
5808
5809 #ifdef ASSERT
5810 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5811 #endif
5812
5813 _counter = 0;
5814 status = pthread_mutex_unlock(_mutex);
5815 assert_status(status == 0, status, "invariant");
5816 // Paranoia to ensure our locked and lock-free paths interact
5817 // correctly with each other and Java-level accesses.
5818 OrderAccess::fence();
5819
5820 // If externally suspended while waiting, re-suspend
5821 if (jt->handle_special_suspend_equivalent_condition()) {
5822 jt->java_suspend_self();
5823 }
5824 }
5825
5826 void Parker::unpark() {
5827 int status = pthread_mutex_lock(_mutex);
5828 assert(status == 0, "invariant");
5829 const int s = _counter;
5830 _counter = 1;
5831 if (s < 1) {
5832 // thread might be parked
5833 if (_cur_index != -1) {
5834 // thread is definitely parked
5835 if (WorkAroundNPTLTimedWaitHang) {
5836 status = pthread_cond_signal(&_cond[_cur_index]);
5837 assert(status == 0, "invariant");
5838 status = pthread_mutex_unlock(_mutex);
5839 assert(status == 0, "invariant");
5840 } else {
5841 status = pthread_mutex_unlock(_mutex);
5842 assert(status == 0, "invariant");
5843 status = pthread_cond_signal(&_cond[_cur_index]);
5844 assert(status == 0, "invariant");
5845 }
5846 } else {
5847 pthread_mutex_unlock(_mutex);
5848 assert(status == 0, "invariant");
5849 }
5850 } else {
5851 pthread_mutex_unlock(_mutex);
5852 assert(status == 0, "invariant");
5853 }
5854 }
5855
5856
5857 extern char** environ;
5858
5859 #ifndef __NR_fork
5860 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5861 #endif
5862
5863 #ifndef __NR_execve
5864 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5865 #endif
5866
5867 // Run the specified command in a separate process. Return its exit value,
5868 // or -1 on failure (e.g. can't fork a new process).
5869 // Unlike system(), this function can be called from signal handler. It
5870 // doesn't block SIGINT et al.
5871 int os::fork_and_exec(char* cmd) {
5872 const char * argv[4] = {"sh", "-c", cmd, NULL};
5873
5874 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5875 // pthread_atfork handlers and reset pthread library. All we need is a
5876 // separate process to execve. Make a direct syscall to fork process.
5877 // On IA64 there's no fork syscall, we have to use fork() and hope for
5878 // the best...
5879 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5880 IA64_ONLY(fork();)
5881
5882 if (pid < 0) {
5883 // fork failed
5884 return -1;
5885
5886 } else if (pid == 0) {
5887 // child process
5888
5889 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5890 // first to kill every thread on the thread list. Because this list is
5891 // not reset by fork() (see notes above), execve() will instead kill
5892 // every thread in the parent process. We know this is the only thread
5893 // in the new process, so make a system call directly.
5894 // IA64 should use normal execve() from glibc to match the glibc fork()
5895 // above.
5896 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5897 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5898
5899 // execve failed
5900 _exit(-1);
5901
5902 } else {
5903 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5904 // care about the actual exit code, for now.
5905
5906 int status;
5907
5908 // Wait for the child process to exit. This returns immediately if
5909 // the child has already exited. */
5910 while (waitpid(pid, &status, 0) < 0) {
5911 switch (errno) {
5912 case ECHILD: return 0;
5913 case EINTR: break;
5914 default: return -1;
5915 }
5916 }
5917
5918 if (WIFEXITED(status)) {
5919 // The child exited normally; get its exit code.
5920 return WEXITSTATUS(status);
5921 } else if (WIFSIGNALED(status)) {
5922 // The child exited because of a signal
5923 // The best value to return is 0x80 + signal number,
5924 // because that is what all Unix shells do, and because
5925 // it allows callers to distinguish between process exit and
5926 // process death by signal.
5927 return 0x80 + WTERMSIG(status);
5928 } else {
5929 // Unknown exit code; pass it through
5930 return status;
5931 }
5932 }
5933 }
5934
5935 // is_headless_jre()
5936 //
5937 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5938 // in order to report if we are running in a headless jre
5939 //
5940 // Since JDK8 xawt/libmawt.so was moved into the same directory
5941 // as libawt.so, and renamed libawt_xawt.so
5942 //
5943 bool os::is_headless_jre() {
5944 struct stat statbuf;
5945 char buf[MAXPATHLEN];
5946 char libmawtpath[MAXPATHLEN];
5947 const char *xawtstr = "/xawt/libmawt.so";
5948 const char *new_xawtstr = "/libawt_xawt.so";
5949 char *p;
5950
5951 // Get path to libjvm.so
5952 os::jvm_path(buf, sizeof(buf));
5953
5954 // Get rid of libjvm.so
5955 p = strrchr(buf, '/');
5956 if (p == NULL) {
5957 return false;
5958 } else {
5959 *p = '\0';
5960 }
5961
5962 // Get rid of client or server
5963 p = strrchr(buf, '/');
5964 if (p == NULL) {
5965 return false;
5966 } else {
5967 *p = '\0';
5968 }
5969
5970 // check xawt/libmawt.so
5971 strcpy(libmawtpath, buf);
5972 strcat(libmawtpath, xawtstr);
5973 if (::stat(libmawtpath, &statbuf) == 0) return false;
5974
5975 // check libawt_xawt.so
5976 strcpy(libmawtpath, buf);
5977 strcat(libmawtpath, new_xawtstr);
5978 if (::stat(libmawtpath, &statbuf) == 0) return false;
5979
5980 return true;
5981 }
5982
5983 // Get the default path to the core file
5984 // Returns the length of the string
5985 int os::get_core_path(char* buffer, size_t bufferSize) {
5986 /*
5987 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5988 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5989 */
5990 const int core_pattern_len = 129;
5991 char core_pattern[core_pattern_len] = {0};
5992
5993 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5994 if (core_pattern_file != -1) {
5995 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5996 ::close(core_pattern_file);
5997
5998 if (ret > 0) {
5999 char *last_char = core_pattern + strlen(core_pattern) - 1;
6000
6001 if (*last_char == '\n') {
6002 *last_char = '\0';
6003 }
6004 }
6005 }
6006
6007 if (strlen(core_pattern) == 0) {
6008 return 0;
6009 }
6010
6011 char *pid_pos = strstr(core_pattern, "%p");
6012 size_t written;
6013
6014 if (core_pattern[0] == '/') {
6015 written = jio_snprintf(buffer, bufferSize, core_pattern);
6016 } else {
6017 char cwd[PATH_MAX];
6018
6019 const char* p = get_current_directory(cwd, PATH_MAX);
6020 if (p == NULL) {
6021 assert(p != NULL, "failed to get current directory");
6022 return 0;
6023 }
6024
6025 if (core_pattern[0] == '|') {
6026 written = jio_snprintf(buffer, bufferSize,
6027 "\"%s\" (or dumping to %s/core.%d)",
6028 &core_pattern[1], p, current_process_id());
6029 } else {
6030 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6031 }
6032 }
6033
6034 if ((written >= 0) && (written < bufferSize)
6035 && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6036 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6037
6038 if (core_uses_pid_file != -1) {
6039 char core_uses_pid = 0;
6040 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6041 ::close(core_uses_pid_file);
6042
6043 if (core_uses_pid == '1'){
6044 jio_snprintf(buffer + written, bufferSize - written,
6045 ".%d", current_process_id());
6046 }
6047 }
6048 }
6049
6050 return strlen(buffer);
6051 }
6052
6053 /////////////// Unit tests ///////////////
6054
6055 #ifndef PRODUCT
6056
6057 #define test_log(...) \
6058 do { \
6059 if (VerboseInternalVMTests) { \
6060 tty->print_cr(__VA_ARGS__); \
6061 tty->flush(); \
6062 } \
6063 } while (false)
6064
6065 class TestReserveMemorySpecial : AllStatic {
6066 public:
6067 static void small_page_write(void* addr, size_t size) {
6068 size_t page_size = os::vm_page_size();
6069
6070 char* end = (char*)addr + size;
6071 for (char* p = (char*)addr; p < end; p += page_size) {
6072 *p = 1;
6073 }
6074 }
6075
6076 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6077 if (!UseHugeTLBFS) {
6078 return;
6079 }
6080
6081 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6082
6083 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6084
6085 if (addr != NULL) {
6086 small_page_write(addr, size);
6087
6088 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6089 }
6090 }
6091
6092 static void test_reserve_memory_special_huge_tlbfs_only() {
6093 if (!UseHugeTLBFS) {
6094 return;
6095 }
6096
6097 size_t lp = os::large_page_size();
6098
6099 for (size_t size = lp; size <= lp * 10; size += lp) {
6100 test_reserve_memory_special_huge_tlbfs_only(size);
6101 }
6102 }
6103
6104 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6105 if (!UseHugeTLBFS) {
6106 return;
6107 }
6108
6109 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6110 size, alignment);
6111
6112 assert(size >= os::large_page_size(), "Incorrect input to test");
6113
6114 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6115
6116 if (addr != NULL) {
6117 small_page_write(addr, size);
6118
6119 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6120 }
6121 }
6122
6123 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6124 size_t lp = os::large_page_size();
6125 size_t ag = os::vm_allocation_granularity();
6126
6127 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6128 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6129 }
6130 }
6131
6132 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6133 size_t lp = os::large_page_size();
6134 size_t ag = os::vm_allocation_granularity();
6135
6136 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6137 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6138 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6139 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6140 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6141 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6142 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6143 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6144 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6145 }
6146
6147 static void test_reserve_memory_special_huge_tlbfs() {
6148 if (!UseHugeTLBFS) {
6149 return;
6150 }
6151
6152 test_reserve_memory_special_huge_tlbfs_only();
6153 test_reserve_memory_special_huge_tlbfs_mixed();
6154 }
6155
6156 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6157 if (!UseSHM) {
6158 return;
6159 }
6160
6161 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6162
6163 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6164
6165 if (addr != NULL) {
6166 assert(is_ptr_aligned(addr, alignment), "Check");
6167 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6168
6169 small_page_write(addr, size);
6170
6171 os::Linux::release_memory_special_shm(addr, size);
6172 }
6173 }
6174
6175 static void test_reserve_memory_special_shm() {
6176 size_t lp = os::large_page_size();
6177 size_t ag = os::vm_allocation_granularity();
6178
6179 for (size_t size = ag; size < lp * 3; size += ag) {
6180 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6181 test_reserve_memory_special_shm(size, alignment);
6182 }
6183 }
6184 }
6185
6186 static void test() {
6187 test_reserve_memory_special_huge_tlbfs();
6188 test_reserve_memory_special_shm();
6189 }
6190 };
6191
6192 void TestReserveMemorySpecial_test() {
6193 TestReserveMemorySpecial::test();
6194 }
6195
6196 #endif