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_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/orderAccess.inline.hpp"
53 #include "runtime/osThread.hpp"
54 #include "runtime/perfMemory.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/statSampler.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "runtime/thread.inline.hpp"
59 #include "runtime/threadCritical.hpp"
60 #include "runtime/timer.hpp"
61 #include "services/attachListener.hpp"
62 #include "services/memTracker.hpp"
63 #include "services/runtimeService.hpp"
64 #include "utilities/decoder.hpp"
65 #include "utilities/defaultStream.hpp"
66 #include "utilities/events.hpp"
67 #include "utilities/elfFile.hpp"
68 #include "utilities/growableArray.hpp"
69 #include "utilities/vmError.hpp"
70
71 // put OS-includes here
72 # include <sys/types.h>
73 # include <sys/mman.h>
74 # include <sys/stat.h>
75 # include <sys/select.h>
76 # include <pthread.h>
77 # include <signal.h>
78 # include <errno.h>
79 # include <dlfcn.h>
80 # include <stdio.h>
81 # include <unistd.h>
82 # include <sys/resource.h>
83 # include <pthread.h>
84 # include <sys/stat.h>
85 # include <sys/time.h>
86 # include <sys/times.h>
87 # include <sys/utsname.h>
88 # include <sys/socket.h>
89 # include <sys/wait.h>
90 # include <pwd.h>
91 # include <poll.h>
92 # include <semaphore.h>
93 # include <fcntl.h>
94 # include <string.h>
95 # include <syscall.h>
96 # include <sys/sysinfo.h>
97 # include <gnu/libc-version.h>
98 # include <sys/ipc.h>
99 # include <sys/shm.h>
100 # include <link.h>
101 # include <stdint.h>
102 # include <inttypes.h>
103 # include <sys/ioctl.h>
104
105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
106
107 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
108 // getrusage() is prepared to handle the associated failure.
109 #ifndef RUSAGE_THREAD
110 #define RUSAGE_THREAD (1) /* only the calling thread */
111 #endif
112
113 #define MAX_PATH (2 * K)
114
115 #define MAX_SECS 100000000
116
117 // for timer info max values which include all bits
118 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
119
120 #define LARGEPAGES_BIT (1 << 6)
121 ////////////////////////////////////////////////////////////////////////////////
122 // global variables
123 julong os::Linux::_physical_memory = 0;
124
125 address os::Linux::_initial_thread_stack_bottom = NULL;
126 uintptr_t os::Linux::_initial_thread_stack_size = 0;
127
128 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
129 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
130 Mutex* os::Linux::_createThread_lock = NULL;
131 pthread_t os::Linux::_main_thread;
132 int os::Linux::_page_size = -1;
133 const int os::Linux::_vm_default_page_size = (8 * K);
134 bool os::Linux::_is_floating_stack = false;
135 bool os::Linux::_is_NPTL = false;
136 bool os::Linux::_supports_fast_thread_cpu_time = false;
137 const char * os::Linux::_glibc_version = NULL;
138 const char * os::Linux::_libpthread_version = NULL;
139 pthread_condattr_t os::Linux::_condattr[1];
140
141 static jlong initial_time_count=0;
142
143 static int clock_tics_per_sec = 100;
144
145 // For diagnostics to print a message once. see run_periodic_checks
146 static sigset_t check_signal_done;
147 static bool check_signals = true;
148
149 static pid_t _initial_pid = 0;
150
151 /* Signal number used to suspend/resume a thread */
152
153 /* do not use any signal number less than SIGSEGV, see 4355769 */
154 static int SR_signum = SIGUSR2;
155 sigset_t SR_sigset;
156
157 /* Used to protect dlsym() calls */
158 static pthread_mutex_t dl_mutex;
159
160 // Declarations
161 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
162
163 // utility functions
164
165 static int SR_initialize();
166
167 julong os::available_memory() {
168 return Linux::available_memory();
169 }
170
171 julong os::Linux::available_memory() {
172 // values in struct sysinfo are "unsigned long"
173 struct sysinfo si;
174 sysinfo(&si);
175
176 return (julong)si.freeram * si.mem_unit;
177 }
178
179 julong os::physical_memory() {
180 return Linux::physical_memory();
181 }
182
183 ////////////////////////////////////////////////////////////////////////////////
184 // environment support
185
186 bool os::getenv(const char* name, char* buf, int len) {
187 const char* val = ::getenv(name);
188 if (val != NULL && strlen(val) < (size_t)len) {
189 strcpy(buf, val);
190 return true;
191 }
192 if (len > 0) buf[0] = 0; // return a null string
193 return false;
194 }
195
196
197 // Return true if user is running as root.
198
199 bool os::have_special_privileges() {
200 static bool init = false;
201 static bool privileges = false;
202 if (!init) {
203 privileges = (getuid() != geteuid()) || (getgid() != getegid());
204 init = true;
205 }
206 return privileges;
207 }
208
209
210 #ifndef SYS_gettid
211 // i386: 224, ia64: 1105, amd64: 186, sparc 143
212 #ifdef __ia64__
213 #define SYS_gettid 1105
214 #else
215 #ifdef __i386__
216 #define SYS_gettid 224
217 #else
218 #ifdef __amd64__
219 #define SYS_gettid 186
220 #else
221 #ifdef __sparc__
222 #define SYS_gettid 143
223 #else
224 #error define gettid for the arch
225 #endif
226 #endif
227 #endif
228 #endif
229 #endif
230
231 // Cpu architecture string
232 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
233
234 // pid_t gettid()
235 //
236 // Returns the kernel thread id of the currently running thread. Kernel
237 // thread id is used to access /proc.
238 //
239 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
240 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
241 //
242 pid_t os::Linux::gettid() {
243 int rslt = syscall(SYS_gettid);
244 if (rslt == -1) {
245 // old kernel, no NPTL support
246 return getpid();
247 } else {
248 return (pid_t)rslt;
249 }
250 }
251
252 // Most versions of linux have a bug where the number of processors are
253 // determined by looking at the /proc file system. In a chroot environment,
254 // the system call returns 1. This causes the VM to act as if it is
255 // a single processor and elide locking (see is_MP() call).
256 static bool unsafe_chroot_detected = false;
257 static const char *unstable_chroot_error = "/proc file system not found.\n"
258 "Java may be unstable running multithreaded in a chroot "
259 "environment on Linux when /proc filesystem is not mounted.";
260
261 void os::Linux::initialize_system_info() {
262 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
263 if (processor_count() == 1) {
264 pid_t pid = os::Linux::gettid();
265 char fname[32];
266 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
267 FILE *fp = fopen(fname, "r");
268 if (fp == NULL) {
269 unsafe_chroot_detected = true;
270 } else {
271 fclose(fp);
272 }
273 }
274 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
275 assert(processor_count() > 0, "linux error");
276 }
277
278 void os::init_system_properties_values() {
279 // The next steps are taken in the product version:
280 //
281 // Obtain the JAVA_HOME value from the location of libjvm.so.
282 // This library should be located at:
283 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
284 //
285 // If "/jre/lib/" appears at the right place in the path, then we
286 // assume libjvm.so is installed in a JDK and we use this path.
287 //
288 // Otherwise exit with message: "Could not create the Java virtual machine."
289 //
290 // The following extra steps are taken in the debugging version:
291 //
292 // If "/jre/lib/" does NOT appear at the right place in the path
293 // instead of exit check for $JAVA_HOME environment variable.
294 //
295 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
296 // then we append a fake suffix "hotspot/libjvm.so" to this path so
297 // it looks like libjvm.so is installed there
298 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
299 //
300 // Otherwise exit.
301 //
302 // Important note: if the location of libjvm.so changes this
303 // code needs to be changed accordingly.
304
305 // See ld(1):
306 // The linker uses the following search paths to locate required
307 // shared libraries:
308 // 1: ...
309 // ...
310 // 7: The default directories, normally /lib and /usr/lib.
311 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
312 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
313 #else
314 #define DEFAULT_LIBPATH "/lib:/usr/lib"
315 #endif
316
317 // Base path of extensions installed on the system.
318 #define SYS_EXT_DIR "/usr/java/packages"
319 #define EXTENSIONS_DIR "/lib/ext"
320 #define ENDORSED_DIR "/lib/endorsed"
321
322 // Buffer that fits several sprintfs.
323 // Note that the space for the colon and the trailing null are provided
324 // by the nulls included by the sizeof operator.
325 const size_t bufsize =
326 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
327 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
328 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
329 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
330
331 // sysclasspath, java_home, dll_dir
332 {
333 char *pslash;
334 os::jvm_path(buf, bufsize);
335
336 // Found the full path to libjvm.so.
337 // Now cut the path to <java_home>/jre if we can.
338 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
339 pslash = strrchr(buf, '/');
340 if (pslash != NULL) {
341 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
342 }
343 Arguments::set_dll_dir(buf);
344
345 if (pslash != NULL) {
346 pslash = strrchr(buf, '/');
347 if (pslash != NULL) {
348 *pslash = '\0'; // Get rid of /<arch>.
349 pslash = strrchr(buf, '/');
350 if (pslash != NULL) {
351 *pslash = '\0'; // Get rid of /lib.
352 }
353 }
354 }
355 Arguments::set_java_home(buf);
356 set_boot_path('/', ':');
357 }
358
359 // Where to look for native libraries.
360 //
361 // Note: Due to a legacy implementation, most of the library path
362 // is set in the launcher. This was to accomodate linking restrictions
363 // on legacy Linux implementations (which are no longer supported).
364 // Eventually, all the library path setting will be done here.
365 //
366 // However, to prevent the proliferation of improperly built native
367 // libraries, the new path component /usr/java/packages is added here.
368 // Eventually, all the library path setting will be done here.
369 {
370 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
371 // should always exist (until the legacy problem cited above is
372 // addressed).
373 const char *v = ::getenv("LD_LIBRARY_PATH");
374 const char *v_colon = ":";
375 if (v == NULL) { v = ""; v_colon = ""; }
376 // That's +1 for the colon and +1 for the trailing '\0'.
377 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
378 strlen(v) + 1 +
379 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
380 mtInternal);
381 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
382 Arguments::set_library_path(ld_library_path);
383 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
384 }
385
386 // Extensions directories.
387 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
388 Arguments::set_ext_dirs(buf);
389
390 // Endorsed standards default directory.
391 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
392 Arguments::set_endorsed_dirs(buf);
393
394 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
395
396 #undef DEFAULT_LIBPATH
397 #undef SYS_EXT_DIR
398 #undef EXTENSIONS_DIR
399 #undef ENDORSED_DIR
400 }
401
402 ////////////////////////////////////////////////////////////////////////////////
403 // breakpoint support
404
405 void os::breakpoint() {
406 BREAKPOINT;
407 }
408
409 extern "C" void breakpoint() {
410 // use debugger to set breakpoint here
411 }
412
413 ////////////////////////////////////////////////////////////////////////////////
414 // signal support
415
416 debug_only(static bool signal_sets_initialized = false);
417 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
418
419 bool os::Linux::is_sig_ignored(int sig) {
420 struct sigaction oact;
421 sigaction(sig, (struct sigaction*)NULL, &oact);
422 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
423 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
424 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
425 return true;
426 else
427 return false;
428 }
429
430 void os::Linux::signal_sets_init() {
431 // Should also have an assertion stating we are still single-threaded.
432 assert(!signal_sets_initialized, "Already initialized");
433 // Fill in signals that are necessarily unblocked for all threads in
434 // the VM. Currently, we unblock the following signals:
435 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
436 // by -Xrs (=ReduceSignalUsage));
437 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
438 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
439 // the dispositions or masks wrt these signals.
440 // Programs embedding the VM that want to use the above signals for their
441 // own purposes must, at this time, use the "-Xrs" option to prevent
442 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
443 // (See bug 4345157, and other related bugs).
444 // In reality, though, unblocking these signals is really a nop, since
445 // these signals are not blocked by default.
446 sigemptyset(&unblocked_sigs);
447 sigemptyset(&allowdebug_blocked_sigs);
448 sigaddset(&unblocked_sigs, SIGILL);
449 sigaddset(&unblocked_sigs, SIGSEGV);
450 sigaddset(&unblocked_sigs, SIGBUS);
451 sigaddset(&unblocked_sigs, SIGFPE);
452 #if defined(PPC64)
453 sigaddset(&unblocked_sigs, SIGTRAP);
454 #endif
455 sigaddset(&unblocked_sigs, SR_signum);
456
457 if (!ReduceSignalUsage) {
458 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
459 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
460 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
461 }
462 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
464 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
469 }
470 }
471 // Fill in signals that are blocked by all but the VM thread.
472 sigemptyset(&vm_sigs);
473 if (!ReduceSignalUsage)
474 sigaddset(&vm_sigs, BREAK_SIGNAL);
475 debug_only(signal_sets_initialized = true);
476
477 }
478
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
485
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
492
493 // These are signals that are blocked during cond_wait to allow debugger in
494 sigset_t* os::Linux::allowdebug_blocked_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &allowdebug_blocked_sigs;
497 }
498
499 void os::Linux::hotspot_sigmask(Thread* thread) {
500
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
504
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
507
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
509
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
520
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
523
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
526 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
527 // generic name for earlier versions.
528 // Define macros here so we can build HotSpot on old systems.
529 # ifndef _CS_GNU_LIBC_VERSION
530 # define _CS_GNU_LIBC_VERSION 2
531 # endif
532 # ifndef _CS_GNU_LIBPTHREAD_VERSION
533 # define _CS_GNU_LIBPTHREAD_VERSION 3
534 # endif
535
536 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
537 if (n > 0) {
538 char *str = (char *)malloc(n, mtInternal);
539 confstr(_CS_GNU_LIBC_VERSION, str, n);
540 os::Linux::set_glibc_version(str);
541 } else {
542 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
543 static char _gnu_libc_version[32];
544 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
545 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
546 os::Linux::set_glibc_version(_gnu_libc_version);
547 }
548
549 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
550 if (n > 0) {
551 char *str = (char *)malloc(n, mtInternal);
552 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
553 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
554 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
555 // is the case. LinuxThreads has a hard limit on max number of threads.
556 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
557 // On the other hand, NPTL does not have such a limit, sysconf()
558 // will return -1 and errno is not changed. Check if it is really NPTL.
559 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
560 strstr(str, "NPTL") &&
561 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
562 free(str);
563 os::Linux::set_libpthread_version("linuxthreads");
564 } else {
565 os::Linux::set_libpthread_version(str);
566 }
567 } else {
568 // glibc before 2.3.2 only has LinuxThreads.
569 os::Linux::set_libpthread_version("linuxthreads");
570 }
571
572 if (strstr(libpthread_version(), "NPTL")) {
573 os::Linux::set_is_NPTL();
574 } else {
575 os::Linux::set_is_LinuxThreads();
576 }
577
578 // LinuxThreads have two flavors: floating-stack mode, which allows variable
579 // stack size; and fixed-stack mode. NPTL is always floating-stack.
580 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
581 os::Linux::set_is_floating_stack();
582 }
583 }
584
585 /////////////////////////////////////////////////////////////////////////////
586 // thread stack
587
588 // Force Linux kernel to expand current thread stack. If "bottom" is close
589 // to the stack guard, caller should block all signals.
590 //
591 // MAP_GROWSDOWN:
592 // A special mmap() flag that is used to implement thread stacks. It tells
593 // kernel that the memory region should extend downwards when needed. This
594 // allows early versions of LinuxThreads to only mmap the first few pages
595 // when creating a new thread. Linux kernel will automatically expand thread
596 // stack as needed (on page faults).
597 //
598 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
599 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
600 // region, it's hard to tell if the fault is due to a legitimate stack
601 // access or because of reading/writing non-exist memory (e.g. buffer
602 // overrun). As a rule, if the fault happens below current stack pointer,
603 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
604 // application (see Linux kernel fault.c).
605 //
606 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
607 // stack overflow detection.
608 //
609 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
610 // not use this flag. However, the stack of initial thread is not created
611 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
612 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
613 // and then attach the thread to JVM.
614 //
615 // To get around the problem and allow stack banging on Linux, we need to
616 // manually expand thread stack after receiving the SIGSEGV.
617 //
618 // There are two ways to expand thread stack to address "bottom", we used
619 // both of them in JVM before 1.5:
620 // 1. adjust stack pointer first so that it is below "bottom", and then
621 // touch "bottom"
622 // 2. mmap() the page in question
623 //
624 // Now alternate signal stack is gone, it's harder to use 2. For instance,
625 // if current sp is already near the lower end of page 101, and we need to
626 // call mmap() to map page 100, it is possible that part of the mmap() frame
627 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
628 // That will destroy the mmap() frame and cause VM to crash.
629 //
630 // The following code works by adjusting sp first, then accessing the "bottom"
631 // page to force a page fault. Linux kernel will then automatically expand the
632 // stack mapping.
633 //
634 // _expand_stack_to() assumes its frame size is less than page size, which
635 // should always be true if the function is not inlined.
636
637 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
638 #define NOINLINE
639 #else
640 #define NOINLINE __attribute__ ((noinline))
641 #endif
642
643 static void _expand_stack_to(address bottom) NOINLINE;
644
645 static void _expand_stack_to(address bottom) {
646 address sp;
647 size_t size;
648 volatile char *p;
649
650 // Adjust bottom to point to the largest address within the same page, it
651 // gives us a one-page buffer if alloca() allocates slightly more memory.
652 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
653 bottom += os::Linux::page_size() - 1;
654
655 // sp might be slightly above current stack pointer; if that's the case, we
656 // will alloca() a little more space than necessary, which is OK. Don't use
657 // os::current_stack_pointer(), as its result can be slightly below current
658 // stack pointer, causing us to not alloca enough to reach "bottom".
659 sp = (address)&sp;
660
661 if (sp > bottom) {
662 size = sp - bottom;
663 p = (volatile char *)alloca(size);
664 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
665 p[0] = '\0';
666 }
667 }
668
669 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
670 assert(t!=NULL, "just checking");
671 assert(t->osthread()->expanding_stack(), "expand should be set");
672 assert(t->stack_base() != NULL, "stack_base was not initialized");
673
674 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
675 sigset_t mask_all, old_sigset;
676 sigfillset(&mask_all);
677 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
678 _expand_stack_to(addr);
679 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
680 return true;
681 }
682 return false;
683 }
684
685 //////////////////////////////////////////////////////////////////////////////
686 // create new thread
687
688 static address highest_vm_reserved_address();
689
690 // check if it's safe to start a new thread
691 static bool _thread_safety_check(Thread* thread) {
692 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
693 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
694 // Heap is mmap'ed at lower end of memory space. Thread stacks are
695 // allocated (MAP_FIXED) from high address space. Every thread stack
696 // occupies a fixed size slot (usually 2Mbytes, but user can change
697 // it to other values if they rebuild LinuxThreads).
698 //
699 // Problem with MAP_FIXED is that mmap() can still succeed even part of
700 // the memory region has already been mmap'ed. That means if we have too
701 // many threads and/or very large heap, eventually thread stack will
702 // collide with heap.
703 //
704 // Here we try to prevent heap/stack collision by comparing current
705 // stack bottom with the highest address that has been mmap'ed by JVM
706 // plus a safety margin for memory maps created by native code.
707 //
708 // This feature can be disabled by setting ThreadSafetyMargin to 0
709 //
710 if (ThreadSafetyMargin > 0) {
711 address stack_bottom = os::current_stack_base() - os::current_stack_size();
712
713 // not safe if our stack extends below the safety margin
714 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
715 } else {
716 return true;
717 }
718 } else {
719 // Floating stack LinuxThreads or NPTL:
720 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
721 // there's not enough space left, pthread_create() will fail. If we come
722 // here, that means enough space has been reserved for stack.
723 return true;
724 }
725 }
726
727 // Thread start routine for all newly created threads
728 static void *java_start(Thread *thread) {
729 // Try to randomize the cache line index of hot stack frames.
730 // This helps when threads of the same stack traces evict each other's
731 // cache lines. The threads can be either from the same JVM instance, or
732 // from different JVM instances. The benefit is especially true for
733 // processors with hyperthreading technology.
734 static int counter = 0;
735 int pid = os::current_process_id();
736 alloca(((pid ^ counter++) & 7) * 128);
737
738 ThreadLocalStorage::set_thread(thread);
739
740 OSThread* osthread = thread->osthread();
741 Monitor* sync = osthread->startThread_lock();
742
743 // non floating stack LinuxThreads needs extra check, see above
744 if (!_thread_safety_check(thread)) {
745 // notify parent thread
746 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
747 osthread->set_state(ZOMBIE);
748 sync->notify_all();
749 return NULL;
750 }
751
752 // thread_id is kernel thread id (similar to Solaris LWP id)
753 osthread->set_thread_id(os::Linux::gettid());
754
755 if (UseNUMA) {
756 int lgrp_id = os::numa_get_group_id();
757 if (lgrp_id != -1) {
758 thread->set_lgrp_id(lgrp_id);
759 }
760 }
761 // initialize signal mask for this thread
762 os::Linux::hotspot_sigmask(thread);
763
764 // initialize floating point control register
765 os::Linux::init_thread_fpu_state();
766
767 // handshaking with parent thread
768 {
769 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
770
771 // notify parent thread
772 osthread->set_state(INITIALIZED);
773 sync->notify_all();
774
775 // wait until os::start_thread()
776 while (osthread->get_state() == INITIALIZED) {
777 sync->wait(Mutex::_no_safepoint_check_flag);
778 }
779 }
780
781 // call one more level start routine
782 thread->run();
783
784 return 0;
785 }
786
787 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
788 assert(thread->osthread() == NULL, "caller responsible");
789
790 // Allocate the OSThread object
791 OSThread* osthread = new OSThread(NULL, NULL);
792 if (osthread == NULL) {
793 return false;
794 }
795
796 // set the correct thread state
797 osthread->set_thread_type(thr_type);
798
799 // Initial state is ALLOCATED but not INITIALIZED
800 osthread->set_state(ALLOCATED);
801
802 thread->set_osthread(osthread);
803
804 // init thread attributes
805 pthread_attr_t attr;
806 pthread_attr_init(&attr);
807 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
808
809 // stack size
810 if (os::Linux::supports_variable_stack_size()) {
811 // calculate stack size if it's not specified by caller
812 if (stack_size == 0) {
813 stack_size = os::Linux::default_stack_size(thr_type);
814
815 switch (thr_type) {
816 case os::java_thread:
817 // Java threads use ThreadStackSize which default value can be
818 // changed with the flag -Xss
819 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
820 stack_size = JavaThread::stack_size_at_create();
821 break;
822 case os::compiler_thread:
823 if (CompilerThreadStackSize > 0) {
824 stack_size = (size_t)(CompilerThreadStackSize * K);
825 break;
826 } // else fall through:
827 // use VMThreadStackSize if CompilerThreadStackSize is not defined
828 case os::vm_thread:
829 case os::pgc_thread:
830 case os::cgc_thread:
831 case os::watcher_thread:
832 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
833 break;
834 }
835 }
836
837 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
838 pthread_attr_setstacksize(&attr, stack_size);
839 } else {
840 // let pthread_create() pick the default value.
841 }
842
843 // glibc guard page
844 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
845
846 ThreadState state;
847
848 {
849 // Serialize thread creation if we are running with fixed stack LinuxThreads
850 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
851 if (lock) {
852 os::Linux::createThread_lock()->lock_without_safepoint_check();
853 }
854
855 pthread_t tid;
856 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
857
858 pthread_attr_destroy(&attr);
859
860 if (ret != 0) {
861 if (PrintMiscellaneous && (Verbose || WizardMode)) {
862 perror("pthread_create()");
863 }
864 // Need to clean up stuff we've allocated so far
865 thread->set_osthread(NULL);
866 delete osthread;
867 if (lock) os::Linux::createThread_lock()->unlock();
868 return false;
869 }
870
871 // Store pthread info into the OSThread
872 osthread->set_pthread_id(tid);
873
874 // Wait until child thread is either initialized or aborted
875 {
876 Monitor* sync_with_child = osthread->startThread_lock();
877 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
878 while ((state = osthread->get_state()) == ALLOCATED) {
879 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
880 }
881 }
882
883 if (lock) {
884 os::Linux::createThread_lock()->unlock();
885 }
886 }
887
888 // Aborted due to thread limit being reached
889 if (state == ZOMBIE) {
890 thread->set_osthread(NULL);
891 delete osthread;
892 return false;
893 }
894
895 // The thread is returned suspended (in state INITIALIZED),
896 // and is started higher up in the call chain
897 assert(state == INITIALIZED, "race condition");
898 return true;
899 }
900
901 /////////////////////////////////////////////////////////////////////////////
902 // attach existing thread
903
904 // bootstrap the main thread
905 bool os::create_main_thread(JavaThread* thread) {
906 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
907 return create_attached_thread(thread);
908 }
909
910 bool os::create_attached_thread(JavaThread* thread) {
911 #ifdef ASSERT
912 thread->verify_not_published();
913 #endif
914
915 // Allocate the OSThread object
916 OSThread* osthread = new OSThread(NULL, NULL);
917
918 if (osthread == NULL) {
919 return false;
920 }
921
922 // Store pthread info into the OSThread
923 osthread->set_thread_id(os::Linux::gettid());
924 osthread->set_pthread_id(::pthread_self());
925
926 // initialize floating point control register
927 os::Linux::init_thread_fpu_state();
928
929 // Initial thread state is RUNNABLE
930 osthread->set_state(RUNNABLE);
931
932 thread->set_osthread(osthread);
933
934 if (UseNUMA) {
935 int lgrp_id = os::numa_get_group_id();
936 if (lgrp_id != -1) {
937 thread->set_lgrp_id(lgrp_id);
938 }
939 }
940
941 if (os::Linux::is_initial_thread()) {
942 // If current thread is initial thread, its stack is mapped on demand,
943 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
944 // the entire stack region to avoid SEGV in stack banging.
945 // It is also useful to get around the heap-stack-gap problem on SuSE
946 // kernel (see 4821821 for details). We first expand stack to the top
947 // of yellow zone, then enable stack yellow zone (order is significant,
948 // enabling yellow zone first will crash JVM on SuSE Linux), so there
949 // is no gap between the last two virtual memory regions.
950
951 JavaThread *jt = (JavaThread *)thread;
952 address addr = jt->stack_yellow_zone_base();
953 assert(addr != NULL, "initialization problem?");
954 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
955
956 osthread->set_expanding_stack();
957 os::Linux::manually_expand_stack(jt, addr);
958 osthread->clear_expanding_stack();
959 }
960
961 // initialize signal mask for this thread
962 // and save the caller's signal mask
963 os::Linux::hotspot_sigmask(thread);
964
965 return true;
966 }
967
968 void os::pd_start_thread(Thread* thread) {
969 OSThread * osthread = thread->osthread();
970 assert(osthread->get_state() != INITIALIZED, "just checking");
971 Monitor* sync_with_child = osthread->startThread_lock();
972 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
973 sync_with_child->notify();
974 }
975
976 // Free Linux resources related to the OSThread
977 void os::free_thread(OSThread* osthread) {
978 assert(osthread != NULL, "osthread not set");
979
980 if (Thread::current()->osthread() == osthread) {
981 // Restore caller's signal mask
982 sigset_t sigmask = osthread->caller_sigmask();
983 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
984 }
985
986 delete osthread;
987 }
988
989 //////////////////////////////////////////////////////////////////////////////
990 // thread local storage
991
992 // Restore the thread pointer if the destructor is called. This is in case
993 // someone from JNI code sets up a destructor with pthread_key_create to run
994 // detachCurrentThread on thread death. Unless we restore the thread pointer we
995 // will hang or crash. When detachCurrentThread is called the key will be set
996 // to null and we will not be called again. If detachCurrentThread is never
997 // called we could loop forever depending on the pthread implementation.
998 static void restore_thread_pointer(void* p) {
999 Thread* thread = (Thread*) p;
1000 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1001 }
1002
1003 int os::allocate_thread_local_storage() {
1004 pthread_key_t key;
1005 int rslt = pthread_key_create(&key, restore_thread_pointer);
1006 assert(rslt == 0, "cannot allocate thread local storage");
1007 return (int)key;
1008 }
1009
1010 // Note: This is currently not used by VM, as we don't destroy TLS key
1011 // on VM exit.
1012 void os::free_thread_local_storage(int index) {
1013 int rslt = pthread_key_delete((pthread_key_t)index);
1014 assert(rslt == 0, "invalid index");
1015 }
1016
1017 void os::thread_local_storage_at_put(int index, void* value) {
1018 int rslt = pthread_setspecific((pthread_key_t)index, value);
1019 assert(rslt == 0, "pthread_setspecific failed");
1020 }
1021
1022 extern "C" Thread* get_thread() {
1023 return ThreadLocalStorage::thread();
1024 }
1025
1026 //////////////////////////////////////////////////////////////////////////////
1027 // initial thread
1028
1029 // Check if current thread is the initial thread, similar to Solaris thr_main.
1030 bool os::Linux::is_initial_thread(void) {
1031 char dummy;
1032 // If called before init complete, thread stack bottom will be null.
1033 // Can be called if fatal error occurs before initialization.
1034 if (initial_thread_stack_bottom() == NULL) return false;
1035 assert(initial_thread_stack_bottom() != NULL &&
1036 initial_thread_stack_size() != 0,
1037 "os::init did not locate initial thread's stack region");
1038 if ((address)&dummy >= initial_thread_stack_bottom() &&
1039 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1040 return true;
1041 else return false;
1042 }
1043
1044 // Find the virtual memory area that contains addr
1045 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1046 FILE *fp = fopen("/proc/self/maps", "r");
1047 if (fp) {
1048 address low, high;
1049 while (!feof(fp)) {
1050 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1051 if (low <= addr && addr < high) {
1052 if (vma_low) *vma_low = low;
1053 if (vma_high) *vma_high = high;
1054 fclose (fp);
1055 return true;
1056 }
1057 }
1058 for (;;) {
1059 int ch = fgetc(fp);
1060 if (ch == EOF || ch == (int)'\n') break;
1061 }
1062 }
1063 fclose(fp);
1064 }
1065 return false;
1066 }
1067
1068 // Locate initial thread stack. This special handling of initial thread stack
1069 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1070 // bogus value for initial thread.
1071 void os::Linux::capture_initial_stack(size_t max_size) {
1072 // stack size is the easy part, get it from RLIMIT_STACK
1073 size_t stack_size;
1074 struct rlimit rlim;
1075 getrlimit(RLIMIT_STACK, &rlim);
1076 stack_size = rlim.rlim_cur;
1077
1078 // 6308388: a bug in ld.so will relocate its own .data section to the
1079 // lower end of primordial stack; reduce ulimit -s value a little bit
1080 // so we won't install guard page on ld.so's data section.
1081 stack_size -= 2 * page_size();
1082
1083 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1084 // 7.1, in both cases we will get 2G in return value.
1085 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1086 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1087 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1088 // in case other parts in glibc still assumes 2M max stack size.
1089 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1090 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1091 if (stack_size > 2 * K * K IA64_ONLY(*2))
1092 stack_size = 2 * K * K IA64_ONLY(*2);
1093 // Try to figure out where the stack base (top) is. This is harder.
1094 //
1095 // When an application is started, glibc saves the initial stack pointer in
1096 // a global variable "__libc_stack_end", which is then used by system
1097 // libraries. __libc_stack_end should be pretty close to stack top. The
1098 // variable is available since the very early days. However, because it is
1099 // a private interface, it could disappear in the future.
1100 //
1101 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1102 // to __libc_stack_end, it is very close to stack top, but isn't the real
1103 // stack top. Note that /proc may not exist if VM is running as a chroot
1104 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1105 // /proc/<pid>/stat could change in the future (though unlikely).
1106 //
1107 // We try __libc_stack_end first. If that doesn't work, look for
1108 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1109 // as a hint, which should work well in most cases.
1110
1111 uintptr_t stack_start;
1112
1113 // try __libc_stack_end first
1114 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1115 if (p && *p) {
1116 stack_start = *p;
1117 } else {
1118 // see if we can get the start_stack field from /proc/self/stat
1119 FILE *fp;
1120 int pid;
1121 char state;
1122 int ppid;
1123 int pgrp;
1124 int session;
1125 int nr;
1126 int tpgrp;
1127 unsigned long flags;
1128 unsigned long minflt;
1129 unsigned long cminflt;
1130 unsigned long majflt;
1131 unsigned long cmajflt;
1132 unsigned long utime;
1133 unsigned long stime;
1134 long cutime;
1135 long cstime;
1136 long prio;
1137 long nice;
1138 long junk;
1139 long it_real;
1140 uintptr_t start;
1141 uintptr_t vsize;
1142 intptr_t rss;
1143 uintptr_t rsslim;
1144 uintptr_t scodes;
1145 uintptr_t ecode;
1146 int i;
1147
1148 // Figure what the primordial thread stack base is. Code is inspired
1149 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1150 // followed by command name surrounded by parentheses, state, etc.
1151 char stat[2048];
1152 int statlen;
1153
1154 fp = fopen("/proc/self/stat", "r");
1155 if (fp) {
1156 statlen = fread(stat, 1, 2047, fp);
1157 stat[statlen] = '\0';
1158 fclose(fp);
1159
1160 // Skip pid and the command string. Note that we could be dealing with
1161 // weird command names, e.g. user could decide to rename java launcher
1162 // to "java 1.4.2 :)", then the stat file would look like
1163 // 1234 (java 1.4.2 :)) R ... ...
1164 // We don't really need to know the command string, just find the last
1165 // occurrence of ")" and then start parsing from there. See bug 4726580.
1166 char * s = strrchr(stat, ')');
1167
1168 i = 0;
1169 if (s) {
1170 // Skip blank chars
1171 do s++; while (isspace(*s));
1172
1173 #define _UFM UINTX_FORMAT
1174 #define _DFM INTX_FORMAT
1175
1176 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1177 /* 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 */
1178 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,
1179 &state, /* 3 %c */
1180 &ppid, /* 4 %d */
1181 &pgrp, /* 5 %d */
1182 &session, /* 6 %d */
1183 &nr, /* 7 %d */
1184 &tpgrp, /* 8 %d */
1185 &flags, /* 9 %lu */
1186 &minflt, /* 10 %lu */
1187 &cminflt, /* 11 %lu */
1188 &majflt, /* 12 %lu */
1189 &cmajflt, /* 13 %lu */
1190 &utime, /* 14 %lu */
1191 &stime, /* 15 %lu */
1192 &cutime, /* 16 %ld */
1193 &cstime, /* 17 %ld */
1194 &prio, /* 18 %ld */
1195 &nice, /* 19 %ld */
1196 &junk, /* 20 %ld */
1197 &it_real, /* 21 %ld */
1198 &start, /* 22 UINTX_FORMAT */
1199 &vsize, /* 23 UINTX_FORMAT */
1200 &rss, /* 24 INTX_FORMAT */
1201 &rsslim, /* 25 UINTX_FORMAT */
1202 &scodes, /* 26 UINTX_FORMAT */
1203 &ecode, /* 27 UINTX_FORMAT */
1204 &stack_start); /* 28 UINTX_FORMAT */
1205 }
1206
1207 #undef _UFM
1208 #undef _DFM
1209
1210 if (i != 28 - 2) {
1211 assert(false, "Bad conversion from /proc/self/stat");
1212 // product mode - assume we are the initial thread, good luck in the
1213 // embedded case.
1214 warning("Can't detect initial thread stack location - bad conversion");
1215 stack_start = (uintptr_t) &rlim;
1216 }
1217 } else {
1218 // For some reason we can't open /proc/self/stat (for example, running on
1219 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1220 // most cases, so don't abort:
1221 warning("Can't detect initial thread stack location - no /proc/self/stat");
1222 stack_start = (uintptr_t) &rlim;
1223 }
1224 }
1225
1226 // Now we have a pointer (stack_start) very close to the stack top, the
1227 // next thing to do is to figure out the exact location of stack top. We
1228 // can find out the virtual memory area that contains stack_start by
1229 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1230 // and its upper limit is the real stack top. (again, this would fail if
1231 // running inside chroot, because /proc may not exist.)
1232
1233 uintptr_t stack_top;
1234 address low, high;
1235 if (find_vma((address)stack_start, &low, &high)) {
1236 // success, "high" is the true stack top. (ignore "low", because initial
1237 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1238 stack_top = (uintptr_t)high;
1239 } else {
1240 // failed, likely because /proc/self/maps does not exist
1241 warning("Can't detect initial thread stack location - find_vma failed");
1242 // best effort: stack_start is normally within a few pages below the real
1243 // stack top, use it as stack top, and reduce stack size so we won't put
1244 // guard page outside stack.
1245 stack_top = stack_start;
1246 stack_size -= 16 * page_size();
1247 }
1248
1249 // stack_top could be partially down the page so align it
1250 stack_top = align_size_up(stack_top, page_size());
1251
1252 if (max_size && stack_size > max_size) {
1253 _initial_thread_stack_size = max_size;
1254 } else {
1255 _initial_thread_stack_size = stack_size;
1256 }
1257
1258 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1259 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1260 }
1261
1262 ////////////////////////////////////////////////////////////////////////////////
1263 // time support
1264
1265 // Time since start-up in seconds to a fine granularity.
1266 // Used by VMSelfDestructTimer and the MemProfiler.
1267 double os::elapsedTime() {
1268
1269 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1270 }
1271
1272 jlong os::elapsed_counter() {
1273 return javaTimeNanos() - initial_time_count;
1274 }
1275
1276 jlong os::elapsed_frequency() {
1277 return NANOSECS_PER_SEC; // nanosecond resolution
1278 }
1279
1280 bool os::supports_vtime() { return true; }
1281 bool os::enable_vtime() { return false; }
1282 bool os::vtime_enabled() { return false; }
1283
1284 double os::elapsedVTime() {
1285 struct rusage usage;
1286 int retval = getrusage(RUSAGE_THREAD, &usage);
1287 if (retval == 0) {
1288 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);
1289 } else {
1290 // better than nothing, but not much
1291 return elapsedTime();
1292 }
1293 }
1294
1295 jlong os::javaTimeMillis() {
1296 timeval time;
1297 int status = gettimeofday(&time, NULL);
1298 assert(status != -1, "linux error");
1299 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1300 }
1301
1302 #ifndef CLOCK_MONOTONIC
1303 #define CLOCK_MONOTONIC (1)
1304 #endif
1305
1306 void os::Linux::clock_init() {
1307 // we do dlopen's in this particular order due to bug in linux
1308 // dynamical loader (see 6348968) leading to crash on exit
1309 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1310 if (handle == NULL) {
1311 handle = dlopen("librt.so", RTLD_LAZY);
1312 }
1313
1314 if (handle) {
1315 int (*clock_getres_func)(clockid_t, struct timespec*) =
1316 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1317 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1318 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1319 if (clock_getres_func && clock_gettime_func) {
1320 // See if monotonic clock is supported by the kernel. Note that some
1321 // early implementations simply return kernel jiffies (updated every
1322 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1323 // for nano time (though the monotonic property is still nice to have).
1324 // It's fixed in newer kernels, however clock_getres() still returns
1325 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1326 // resolution for now. Hopefully as people move to new kernels, this
1327 // won't be a problem.
1328 struct timespec res;
1329 struct timespec tp;
1330 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1331 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1332 // yes, monotonic clock is supported
1333 _clock_gettime = clock_gettime_func;
1334 return;
1335 } else {
1336 // close librt if there is no monotonic clock
1337 dlclose(handle);
1338 }
1339 }
1340 }
1341 warning("No monotonic clock was available - timed services may " \
1342 "be adversely affected if the time-of-day clock changes");
1343 }
1344
1345 #ifndef SYS_clock_getres
1346
1347 #if defined(IA32) || defined(AMD64)
1348 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1349 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1350 #else
1351 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1352 #define sys_clock_getres(x,y) -1
1353 #endif
1354
1355 #else
1356 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1357 #endif
1358
1359 void os::Linux::fast_thread_clock_init() {
1360 if (!UseLinuxPosixThreadCPUClocks) {
1361 return;
1362 }
1363 clockid_t clockid;
1364 struct timespec tp;
1365 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1366 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1367
1368 // Switch to using fast clocks for thread cpu time if
1369 // the sys_clock_getres() returns 0 error code.
1370 // Note, that some kernels may support the current thread
1371 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1372 // returned by the pthread_getcpuclockid().
1373 // If the fast Posix clocks are supported then the sys_clock_getres()
1374 // must return at least tp.tv_sec == 0 which means a resolution
1375 // better than 1 sec. This is extra check for reliability.
1376
1377 if(pthread_getcpuclockid_func &&
1378 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1379 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1380
1381 _supports_fast_thread_cpu_time = true;
1382 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1383 }
1384 }
1385
1386 jlong os::javaTimeNanos() {
1387 if (Linux::supports_monotonic_clock()) {
1388 struct timespec tp;
1389 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1390 assert(status == 0, "gettime error");
1391 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1392 return result;
1393 } else {
1394 timeval time;
1395 int status = gettimeofday(&time, NULL);
1396 assert(status != -1, "linux error");
1397 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1398 return 1000 * usecs;
1399 }
1400 }
1401
1402 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1403 if (Linux::supports_monotonic_clock()) {
1404 info_ptr->max_value = ALL_64_BITS;
1405
1406 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1407 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1408 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1409 } else {
1410 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1411 info_ptr->max_value = ALL_64_BITS;
1412
1413 // gettimeofday is a real time clock so it skips
1414 info_ptr->may_skip_backward = true;
1415 info_ptr->may_skip_forward = true;
1416 }
1417
1418 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1419 }
1420
1421 // Return the real, user, and system times in seconds from an
1422 // arbitrary fixed point in the past.
1423 bool os::getTimesSecs(double* process_real_time,
1424 double* process_user_time,
1425 double* process_system_time) {
1426 struct tms ticks;
1427 clock_t real_ticks = times(&ticks);
1428
1429 if (real_ticks == (clock_t) (-1)) {
1430 return false;
1431 } else {
1432 double ticks_per_second = (double) clock_tics_per_sec;
1433 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1434 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1435 *process_real_time = ((double) real_ticks) / ticks_per_second;
1436
1437 return true;
1438 }
1439 }
1440
1441
1442 char * os::local_time_string(char *buf, size_t buflen) {
1443 struct tm t;
1444 time_t long_time;
1445 time(&long_time);
1446 localtime_r(&long_time, &t);
1447 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1448 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1449 t.tm_hour, t.tm_min, t.tm_sec);
1450 return buf;
1451 }
1452
1453 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1454 return localtime_r(clock, res);
1455 }
1456
1457 ////////////////////////////////////////////////////////////////////////////////
1458 // runtime exit support
1459
1460 // Note: os::shutdown() might be called very early during initialization, or
1461 // called from signal handler. Before adding something to os::shutdown(), make
1462 // sure it is async-safe and can handle partially initialized VM.
1463 void os::shutdown() {
1464
1465 // allow PerfMemory to attempt cleanup of any persistent resources
1466 perfMemory_exit();
1467
1468 // needs to remove object in file system
1469 AttachListener::abort();
1470
1471 // flush buffered output, finish log files
1472 ostream_abort();
1473
1474 // Check for abort hook
1475 abort_hook_t abort_hook = Arguments::abort_hook();
1476 if (abort_hook != NULL) {
1477 abort_hook();
1478 }
1479
1480 }
1481
1482 // Note: os::abort() might be called very early during initialization, or
1483 // called from signal handler. Before adding something to os::abort(), make
1484 // sure it is async-safe and can handle partially initialized VM.
1485 void os::abort(bool dump_core) {
1486 os::shutdown();
1487 if (dump_core) {
1488 #ifndef PRODUCT
1489 fdStream out(defaultStream::output_fd());
1490 out.print_raw("Current thread is ");
1491 char buf[16];
1492 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1493 out.print_raw_cr(buf);
1494 out.print_raw_cr("Dumping core ...");
1495 #endif
1496 ::abort(); // dump core
1497 }
1498
1499 ::exit(1);
1500 }
1501
1502 // Die immediately, no exit hook, no abort hook, no cleanup.
1503 void os::die() {
1504 // _exit() on LinuxThreads only kills current thread
1505 ::abort();
1506 }
1507
1508
1509 // This method is a copy of JDK's sysGetLastErrorString
1510 // from src/solaris/hpi/src/system_md.c
1511
1512 size_t os::lasterror(char *buf, size_t len) {
1513
1514 if (errno == 0) return 0;
1515
1516 const char *s = ::strerror(errno);
1517 size_t n = ::strlen(s);
1518 if (n >= len) {
1519 n = len - 1;
1520 }
1521 ::strncpy(buf, s, n);
1522 buf[n] = '\0';
1523 return n;
1524 }
1525
1526 intx os::current_thread_id() { return (intx)pthread_self(); }
1527 int os::current_process_id() {
1528
1529 // Under the old linux thread library, linux gives each thread
1530 // its own process id. Because of this each thread will return
1531 // a different pid if this method were to return the result
1532 // of getpid(2). Linux provides no api that returns the pid
1533 // of the launcher thread for the vm. This implementation
1534 // returns a unique pid, the pid of the launcher thread
1535 // that starts the vm 'process'.
1536
1537 // Under the NPTL, getpid() returns the same pid as the
1538 // launcher thread rather than a unique pid per thread.
1539 // Use gettid() if you want the old pre NPTL behaviour.
1540
1541 // if you are looking for the result of a call to getpid() that
1542 // returns a unique pid for the calling thread, then look at the
1543 // OSThread::thread_id() method in osThread_linux.hpp file
1544
1545 return (int)(_initial_pid ? _initial_pid : getpid());
1546 }
1547
1548 // DLL functions
1549
1550 const char* os::dll_file_extension() { return ".so"; }
1551
1552 // This must be hard coded because it's the system's temporary
1553 // directory not the java application's temp directory, ala java.io.tmpdir.
1554 const char* os::get_temp_directory() { return "/tmp"; }
1555
1556 static bool file_exists(const char* filename) {
1557 struct stat statbuf;
1558 if (filename == NULL || strlen(filename) == 0) {
1559 return false;
1560 }
1561 return os::stat(filename, &statbuf) == 0;
1562 }
1563
1564 bool os::dll_build_name(char* buffer, size_t buflen,
1565 const char* pname, const char* fname) {
1566 bool retval = false;
1567 // Copied from libhpi
1568 const size_t pnamelen = pname ? strlen(pname) : 0;
1569
1570 // Return error on buffer overflow.
1571 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1572 return retval;
1573 }
1574
1575 if (pnamelen == 0) {
1576 snprintf(buffer, buflen, "lib%s.so", fname);
1577 retval = true;
1578 } else if (strchr(pname, *os::path_separator()) != NULL) {
1579 int n;
1580 char** pelements = split_path(pname, &n);
1581 if (pelements == NULL) {
1582 return false;
1583 }
1584 for (int i = 0 ; i < n ; i++) {
1585 // Really shouldn't be NULL, but check can't hurt
1586 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1587 continue; // skip the empty path values
1588 }
1589 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1590 if (file_exists(buffer)) {
1591 retval = true;
1592 break;
1593 }
1594 }
1595 // release the storage
1596 for (int i = 0 ; i < n ; i++) {
1597 if (pelements[i] != NULL) {
1598 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1599 }
1600 }
1601 if (pelements != NULL) {
1602 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1603 }
1604 } else {
1605 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1606 retval = true;
1607 }
1608 return retval;
1609 }
1610
1611 // check if addr is inside libjvm.so
1612 bool os::address_is_in_vm(address addr) {
1613 static address libjvm_base_addr;
1614 Dl_info dlinfo;
1615
1616 if (libjvm_base_addr == NULL) {
1617 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1618 libjvm_base_addr = (address)dlinfo.dli_fbase;
1619 }
1620 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1621 }
1622
1623 if (dladdr((void *)addr, &dlinfo) != 0) {
1624 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1625 }
1626
1627 return false;
1628 }
1629
1630 bool os::dll_address_to_function_name(address addr, char *buf,
1631 int buflen, int *offset) {
1632 // buf is not optional, but offset is optional
1633 assert(buf != NULL, "sanity check");
1634
1635 Dl_info dlinfo;
1636
1637 if (dladdr((void*)addr, &dlinfo) != 0) {
1638 // see if we have a matching symbol
1639 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1640 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1641 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1642 }
1643 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1644 return true;
1645 }
1646 // no matching symbol so try for just file info
1647 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1648 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1649 buf, buflen, offset, dlinfo.dli_fname)) {
1650 return true;
1651 }
1652 }
1653 }
1654
1655 buf[0] = '\0';
1656 if (offset != NULL) *offset = -1;
1657 return false;
1658 }
1659
1660 struct _address_to_library_name {
1661 address addr; // input : memory address
1662 size_t buflen; // size of fname
1663 char* fname; // output: library name
1664 address base; // library base addr
1665 };
1666
1667 static int address_to_library_name_callback(struct dl_phdr_info *info,
1668 size_t size, void *data) {
1669 int i;
1670 bool found = false;
1671 address libbase = NULL;
1672 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1673
1674 // iterate through all loadable segments
1675 for (i = 0; i < info->dlpi_phnum; i++) {
1676 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1677 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1678 // base address of a library is the lowest address of its loaded
1679 // segments.
1680 if (libbase == NULL || libbase > segbase) {
1681 libbase = segbase;
1682 }
1683 // see if 'addr' is within current segment
1684 if (segbase <= d->addr &&
1685 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1686 found = true;
1687 }
1688 }
1689 }
1690
1691 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1692 // so dll_address_to_library_name() can fall through to use dladdr() which
1693 // can figure out executable name from argv[0].
1694 if (found && info->dlpi_name && info->dlpi_name[0]) {
1695 d->base = libbase;
1696 if (d->fname) {
1697 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1698 }
1699 return 1;
1700 }
1701 return 0;
1702 }
1703
1704 bool os::dll_address_to_library_name(address addr, char* buf,
1705 int buflen, int* offset) {
1706 // buf is not optional, but offset is optional
1707 assert(buf != NULL, "sanity check");
1708
1709 Dl_info dlinfo;
1710 struct _address_to_library_name data;
1711
1712 // There is a bug in old glibc dladdr() implementation that it could resolve
1713 // to wrong library name if the .so file has a base address != NULL. Here
1714 // we iterate through the program headers of all loaded libraries to find
1715 // out which library 'addr' really belongs to. This workaround can be
1716 // removed once the minimum requirement for glibc is moved to 2.3.x.
1717 data.addr = addr;
1718 data.fname = buf;
1719 data.buflen = buflen;
1720 data.base = NULL;
1721 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1722
1723 if (rslt) {
1724 // buf already contains library name
1725 if (offset) *offset = addr - data.base;
1726 return true;
1727 }
1728 if (dladdr((void*)addr, &dlinfo) != 0) {
1729 if (dlinfo.dli_fname != NULL) {
1730 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1731 }
1732 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1733 *offset = addr - (address)dlinfo.dli_fbase;
1734 }
1735 return true;
1736 }
1737
1738 buf[0] = '\0';
1739 if (offset) *offset = -1;
1740 return false;
1741 }
1742
1743 // Loads .dll/.so and
1744 // in case of error it checks if .dll/.so was built for the
1745 // same architecture as Hotspot is running on
1746
1747
1748 // Remember the stack's state. The Linux dynamic linker will change
1749 // the stack to 'executable' at most once, so we must safepoint only once.
1750 bool os::Linux::_stack_is_executable = false;
1751
1752 // VM operation that loads a library. This is necessary if stack protection
1753 // of the Java stacks can be lost during loading the library. If we
1754 // do not stop the Java threads, they can stack overflow before the stacks
1755 // are protected again.
1756 class VM_LinuxDllLoad: public VM_Operation {
1757 private:
1758 const char *_filename;
1759 char *_ebuf;
1760 int _ebuflen;
1761 void *_lib;
1762 public:
1763 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1764 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1765 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1766 void doit() {
1767 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1768 os::Linux::_stack_is_executable = true;
1769 }
1770 void* loaded_library() { return _lib; }
1771 };
1772
1773 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1774 {
1775 void * result = NULL;
1776 bool load_attempted = false;
1777
1778 // Check whether the library to load might change execution rights
1779 // of the stack. If they are changed, the protection of the stack
1780 // guard pages will be lost. We need a safepoint to fix this.
1781 //
1782 // See Linux man page execstack(8) for more info.
1783 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1784 ElfFile ef(filename);
1785 if (!ef.specifies_noexecstack()) {
1786 if (!is_init_completed()) {
1787 os::Linux::_stack_is_executable = true;
1788 // This is OK - No Java threads have been created yet, and hence no
1789 // stack guard pages to fix.
1790 //
1791 // This should happen only when you are building JDK7 using a very
1792 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1793 //
1794 // Dynamic loader will make all stacks executable after
1795 // this function returns, and will not do that again.
1796 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1797 } else {
1798 warning("You have loaded library %s which might have disabled stack guard. "
1799 "The VM will try to fix the stack guard now.\n"
1800 "It's highly recommended that you fix the library with "
1801 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1802 filename);
1803
1804 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1805 JavaThread *jt = JavaThread::current();
1806 if (jt->thread_state() != _thread_in_native) {
1807 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1808 // that requires ExecStack. Cannot enter safe point. Let's give up.
1809 warning("Unable to fix stack guard. Giving up.");
1810 } else {
1811 if (!LoadExecStackDllInVMThread) {
1812 // This is for the case where the DLL has an static
1813 // constructor function that executes JNI code. We cannot
1814 // load such DLLs in the VMThread.
1815 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1816 }
1817
1818 ThreadInVMfromNative tiv(jt);
1819 debug_only(VMNativeEntryWrapper vew;)
1820
1821 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1822 VMThread::execute(&op);
1823 if (LoadExecStackDllInVMThread) {
1824 result = op.loaded_library();
1825 }
1826 load_attempted = true;
1827 }
1828 }
1829 }
1830 }
1831
1832 if (!load_attempted) {
1833 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1834 }
1835
1836 if (result != NULL) {
1837 // Successful loading
1838 return result;
1839 }
1840
1841 Elf32_Ehdr elf_head;
1842 int diag_msg_max_length=ebuflen-strlen(ebuf);
1843 char* diag_msg_buf=ebuf+strlen(ebuf);
1844
1845 if (diag_msg_max_length==0) {
1846 // No more space in ebuf for additional diagnostics message
1847 return NULL;
1848 }
1849
1850
1851 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1852
1853 if (file_descriptor < 0) {
1854 // Can't open library, report dlerror() message
1855 return NULL;
1856 }
1857
1858 bool failed_to_read_elf_head=
1859 (sizeof(elf_head)!=
1860 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1861
1862 ::close(file_descriptor);
1863 if (failed_to_read_elf_head) {
1864 // file i/o error - report dlerror() msg
1865 return NULL;
1866 }
1867
1868 typedef struct {
1869 Elf32_Half code; // Actual value as defined in elf.h
1870 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1871 char elf_class; // 32 or 64 bit
1872 char endianess; // MSB or LSB
1873 char* name; // String representation
1874 } arch_t;
1875
1876 #ifndef EM_486
1877 #define EM_486 6 /* Intel 80486 */
1878 #endif
1879
1880 static const arch_t arch_array[]={
1881 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1882 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1883 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1884 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1885 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1886 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1887 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1888 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1889 #if defined(VM_LITTLE_ENDIAN)
1890 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1891 #else
1892 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1893 #endif
1894 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1895 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1896 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1897 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1898 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1899 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1900 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1901 };
1902
1903 #if (defined IA32)
1904 static Elf32_Half running_arch_code=EM_386;
1905 #elif (defined AMD64)
1906 static Elf32_Half running_arch_code=EM_X86_64;
1907 #elif (defined IA64)
1908 static Elf32_Half running_arch_code=EM_IA_64;
1909 #elif (defined __sparc) && (defined _LP64)
1910 static Elf32_Half running_arch_code=EM_SPARCV9;
1911 #elif (defined __sparc) && (!defined _LP64)
1912 static Elf32_Half running_arch_code=EM_SPARC;
1913 #elif (defined __powerpc64__)
1914 static Elf32_Half running_arch_code=EM_PPC64;
1915 #elif (defined __powerpc__)
1916 static Elf32_Half running_arch_code=EM_PPC;
1917 #elif (defined ARM)
1918 static Elf32_Half running_arch_code=EM_ARM;
1919 #elif (defined S390)
1920 static Elf32_Half running_arch_code=EM_S390;
1921 #elif (defined ALPHA)
1922 static Elf32_Half running_arch_code=EM_ALPHA;
1923 #elif (defined MIPSEL)
1924 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1925 #elif (defined PARISC)
1926 static Elf32_Half running_arch_code=EM_PARISC;
1927 #elif (defined MIPS)
1928 static Elf32_Half running_arch_code=EM_MIPS;
1929 #elif (defined M68K)
1930 static Elf32_Half running_arch_code=EM_68K;
1931 #else
1932 #error Method os::dll_load requires that one of following is defined:\
1933 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1934 #endif
1935
1936 // Identify compatability class for VM's architecture and library's architecture
1937 // Obtain string descriptions for architectures
1938
1939 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1940 int running_arch_index=-1;
1941
1942 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1943 if (running_arch_code == arch_array[i].code) {
1944 running_arch_index = i;
1945 }
1946 if (lib_arch.code == arch_array[i].code) {
1947 lib_arch.compat_class = arch_array[i].compat_class;
1948 lib_arch.name = arch_array[i].name;
1949 }
1950 }
1951
1952 assert(running_arch_index != -1,
1953 "Didn't find running architecture code (running_arch_code) in arch_array");
1954 if (running_arch_index == -1) {
1955 // Even though running architecture detection failed
1956 // we may still continue with reporting dlerror() message
1957 return NULL;
1958 }
1959
1960 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1961 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1962 return NULL;
1963 }
1964
1965 #ifndef S390
1966 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1967 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1968 return NULL;
1969 }
1970 #endif // !S390
1971
1972 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1973 if ( lib_arch.name!=NULL ) {
1974 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1975 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1976 lib_arch.name, arch_array[running_arch_index].name);
1977 } else {
1978 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1979 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1980 lib_arch.code,
1981 arch_array[running_arch_index].name);
1982 }
1983 }
1984
1985 return NULL;
1986 }
1987
1988 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
1989 void * result = ::dlopen(filename, RTLD_LAZY);
1990 if (result == NULL) {
1991 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1992 ebuf[ebuflen-1] = '\0';
1993 }
1994 return result;
1995 }
1996
1997 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
1998 void * result = NULL;
1999 if (LoadExecStackDllInVMThread) {
2000 result = dlopen_helper(filename, ebuf, ebuflen);
2001 }
2002
2003 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2004 // library that requires an executable stack, or which does not have this
2005 // stack attribute set, dlopen changes the stack attribute to executable. The
2006 // read protection of the guard pages gets lost.
2007 //
2008 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2009 // may have been queued at the same time.
2010
2011 if (!_stack_is_executable) {
2012 JavaThread *jt = Threads::first();
2013
2014 while (jt) {
2015 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2016 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2017 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2018 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2019 warning("Attempt to reguard stack yellow zone failed.");
2020 }
2021 }
2022 jt = jt->next();
2023 }
2024 }
2025
2026 return result;
2027 }
2028
2029 /*
2030 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2031 * chances are you might want to run the generated bits against glibc-2.0
2032 * libdl.so, so always use locking for any version of glibc.
2033 */
2034 void* os::dll_lookup(void* handle, const char* name) {
2035 pthread_mutex_lock(&dl_mutex);
2036 void* res = dlsym(handle, name);
2037 pthread_mutex_unlock(&dl_mutex);
2038 return res;
2039 }
2040
2041 void* os::get_default_process_handle() {
2042 return (void*)::dlopen(NULL, RTLD_LAZY);
2043 }
2044
2045 static bool _print_ascii_file(const char* filename, outputStream* st) {
2046 int fd = ::open(filename, O_RDONLY);
2047 if (fd == -1) {
2048 return false;
2049 }
2050
2051 char buf[32];
2052 int bytes;
2053 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2054 st->print_raw(buf, bytes);
2055 }
2056
2057 ::close(fd);
2058
2059 return true;
2060 }
2061
2062 void os::print_dll_info(outputStream *st) {
2063 st->print_cr("Dynamic libraries:");
2064
2065 char fname[32];
2066 pid_t pid = os::Linux::gettid();
2067
2068 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2069
2070 if (!_print_ascii_file(fname, st)) {
2071 st->print("Can not get library information for pid = %d\n", pid);
2072 }
2073 }
2074
2075 void os::print_os_info_brief(outputStream* st) {
2076 os::Linux::print_distro_info(st);
2077
2078 os::Posix::print_uname_info(st);
2079
2080 os::Linux::print_libversion_info(st);
2081
2082 }
2083
2084 void os::print_os_info(outputStream* st) {
2085 st->print("OS:");
2086
2087 os::Linux::print_distro_info(st);
2088
2089 os::Posix::print_uname_info(st);
2090
2091 // Print warning if unsafe chroot environment detected
2092 if (unsafe_chroot_detected) {
2093 st->print("WARNING!! ");
2094 st->print_cr("%s", unstable_chroot_error);
2095 }
2096
2097 os::Linux::print_libversion_info(st);
2098
2099 os::Posix::print_rlimit_info(st);
2100
2101 os::Posix::print_load_average(st);
2102
2103 os::Linux::print_full_memory_info(st);
2104 }
2105
2106 // Try to identify popular distros.
2107 // Most Linux distributions have a /etc/XXX-release file, which contains
2108 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2109 // file that also contains the OS version string. Some have more than one
2110 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2111 // /etc/redhat-release.), so the order is important.
2112 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2113 // their own specific XXX-release file as well as a redhat-release file.
2114 // Because of this the XXX-release file needs to be searched for before the
2115 // redhat-release file.
2116 // Since Red Hat has a lsb-release file that is not very descriptive the
2117 // search for redhat-release needs to be before lsb-release.
2118 // Since the lsb-release file is the new standard it needs to be searched
2119 // before the older style release files.
2120 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2121 // next to last resort. The os-release file is a new standard that contains
2122 // distribution information and the system-release file seems to be an old
2123 // standard that has been replaced by the lsb-release and os-release files.
2124 // Searching for the debian_version file is the last resort. It contains
2125 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2126 // "Debian " is printed before the contents of the debian_version file.
2127 void os::Linux::print_distro_info(outputStream* st) {
2128 if (!_print_ascii_file("/etc/oracle-release", st) &&
2129 !_print_ascii_file("/etc/mandriva-release", st) &&
2130 !_print_ascii_file("/etc/mandrake-release", st) &&
2131 !_print_ascii_file("/etc/sun-release", st) &&
2132 !_print_ascii_file("/etc/redhat-release", st) &&
2133 !_print_ascii_file("/etc/lsb-release", st) &&
2134 !_print_ascii_file("/etc/SuSE-release", st) &&
2135 !_print_ascii_file("/etc/turbolinux-release", st) &&
2136 !_print_ascii_file("/etc/gentoo-release", st) &&
2137 !_print_ascii_file("/etc/ltib-release", st) &&
2138 !_print_ascii_file("/etc/angstrom-version", st) &&
2139 !_print_ascii_file("/etc/system-release", st) &&
2140 !_print_ascii_file("/etc/os-release", st)) {
2141
2142 if (file_exists("/etc/debian_version")) {
2143 st->print("Debian ");
2144 _print_ascii_file("/etc/debian_version", st);
2145 } else {
2146 st->print("Linux");
2147 }
2148 }
2149 st->cr();
2150 }
2151
2152 void os::Linux::print_libversion_info(outputStream* st) {
2153 // libc, pthread
2154 st->print("libc:");
2155 st->print("%s ", os::Linux::glibc_version());
2156 st->print("%s ", os::Linux::libpthread_version());
2157 if (os::Linux::is_LinuxThreads()) {
2158 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2159 }
2160 st->cr();
2161 }
2162
2163 void os::Linux::print_full_memory_info(outputStream* st) {
2164 st->print("\n/proc/meminfo:\n");
2165 _print_ascii_file("/proc/meminfo", st);
2166 st->cr();
2167 }
2168
2169 void os::print_memory_info(outputStream* st) {
2170
2171 st->print("Memory:");
2172 st->print(" %dk page", os::vm_page_size()>>10);
2173
2174 // values in struct sysinfo are "unsigned long"
2175 struct sysinfo si;
2176 sysinfo(&si);
2177
2178 st->print(", physical " UINT64_FORMAT "k",
2179 os::physical_memory() >> 10);
2180 st->print("(" UINT64_FORMAT "k free)",
2181 os::available_memory() >> 10);
2182 st->print(", swap " UINT64_FORMAT "k",
2183 ((jlong)si.totalswap * si.mem_unit) >> 10);
2184 st->print("(" UINT64_FORMAT "k free)",
2185 ((jlong)si.freeswap * si.mem_unit) >> 10);
2186 st->cr();
2187 }
2188
2189 void os::pd_print_cpu_info(outputStream* st) {
2190 st->print("\n/proc/cpuinfo:\n");
2191 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2192 st->print(" <Not Available>");
2193 }
2194 st->cr();
2195 }
2196
2197 void os::print_siginfo(outputStream* st, void* siginfo) {
2198 const siginfo_t* si = (const siginfo_t*)siginfo;
2199
2200 os::Posix::print_siginfo_brief(st, si);
2201 #if INCLUDE_CDS
2202 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2203 UseSharedSpaces) {
2204 FileMapInfo* mapinfo = FileMapInfo::current_info();
2205 if (mapinfo->is_in_shared_space(si->si_addr)) {
2206 st->print("\n\nError accessing class data sharing archive." \
2207 " Mapped file inaccessible during execution, " \
2208 " possible disk/network problem.");
2209 }
2210 }
2211 #endif
2212 st->cr();
2213 }
2214
2215
2216 static void print_signal_handler(outputStream* st, int sig,
2217 char* buf, size_t buflen);
2218
2219 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2220 st->print_cr("Signal Handlers:");
2221 print_signal_handler(st, SIGSEGV, buf, buflen);
2222 print_signal_handler(st, SIGBUS , buf, buflen);
2223 print_signal_handler(st, SIGFPE , buf, buflen);
2224 print_signal_handler(st, SIGPIPE, buf, buflen);
2225 print_signal_handler(st, SIGXFSZ, buf, buflen);
2226 print_signal_handler(st, SIGILL , buf, buflen);
2227 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2228 print_signal_handler(st, SR_signum, buf, buflen);
2229 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2230 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2231 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2232 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2233 #if defined(PPC64)
2234 print_signal_handler(st, SIGTRAP, buf, buflen);
2235 #endif
2236 }
2237
2238 static char saved_jvm_path[MAXPATHLEN] = {0};
2239
2240 // Find the full path to the current module, libjvm.so
2241 void os::jvm_path(char *buf, jint buflen) {
2242 // Error checking.
2243 if (buflen < MAXPATHLEN) {
2244 assert(false, "must use a large-enough buffer");
2245 buf[0] = '\0';
2246 return;
2247 }
2248 // Lazy resolve the path to current module.
2249 if (saved_jvm_path[0] != 0) {
2250 strcpy(buf, saved_jvm_path);
2251 return;
2252 }
2253
2254 char dli_fname[MAXPATHLEN];
2255 bool ret = dll_address_to_library_name(
2256 CAST_FROM_FN_PTR(address, os::jvm_path),
2257 dli_fname, sizeof(dli_fname), NULL);
2258 assert(ret, "cannot locate libjvm");
2259 char *rp = NULL;
2260 if (ret && dli_fname[0] != '\0') {
2261 rp = realpath(dli_fname, buf);
2262 }
2263 if (rp == NULL)
2264 return;
2265
2266 if (Arguments::created_by_gamma_launcher()) {
2267 // Support for the gamma launcher. Typical value for buf is
2268 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2269 // the right place in the string, then assume we are installed in a JDK and
2270 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2271 // up the path so it looks like libjvm.so is installed there (append a
2272 // fake suffix hotspot/libjvm.so).
2273 const char *p = buf + strlen(buf) - 1;
2274 for (int count = 0; p > buf && count < 5; ++count) {
2275 for (--p; p > buf && *p != '/'; --p)
2276 /* empty */ ;
2277 }
2278
2279 if (strncmp(p, "/jre/lib/", 9) != 0) {
2280 // Look for JAVA_HOME in the environment.
2281 char* java_home_var = ::getenv("JAVA_HOME");
2282 if (java_home_var != NULL && java_home_var[0] != 0) {
2283 char* jrelib_p;
2284 int len;
2285
2286 // Check the current module name "libjvm.so".
2287 p = strrchr(buf, '/');
2288 assert(strstr(p, "/libjvm") == p, "invalid library name");
2289
2290 rp = realpath(java_home_var, buf);
2291 if (rp == NULL)
2292 return;
2293
2294 // determine if this is a legacy image or modules image
2295 // modules image doesn't have "jre" subdirectory
2296 len = strlen(buf);
2297 assert(len < buflen, "Ran out of buffer room");
2298 jrelib_p = buf + len;
2299 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2300 if (0 != access(buf, F_OK)) {
2301 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2302 }
2303
2304 if (0 == access(buf, F_OK)) {
2305 // Use current module name "libjvm.so"
2306 len = strlen(buf);
2307 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2308 } else {
2309 // Go back to path of .so
2310 rp = realpath(dli_fname, buf);
2311 if (rp == NULL)
2312 return;
2313 }
2314 }
2315 }
2316 }
2317
2318 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2319 }
2320
2321 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2322 // no prefix required, not even "_"
2323 }
2324
2325 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2326 // no suffix required
2327 }
2328
2329 ////////////////////////////////////////////////////////////////////////////////
2330 // sun.misc.Signal support
2331
2332 static volatile jint sigint_count = 0;
2333
2334 static void
2335 UserHandler(int sig, void *siginfo, void *context) {
2336 // 4511530 - sem_post is serialized and handled by the manager thread. When
2337 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2338 // don't want to flood the manager thread with sem_post requests.
2339 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2340 return;
2341
2342 // Ctrl-C is pressed during error reporting, likely because the error
2343 // handler fails to abort. Let VM die immediately.
2344 if (sig == SIGINT && is_error_reported()) {
2345 os::die();
2346 }
2347
2348 os::signal_notify(sig);
2349 }
2350
2351 void* os::user_handler() {
2352 return CAST_FROM_FN_PTR(void*, UserHandler);
2353 }
2354
2355 class Semaphore : public StackObj {
2356 public:
2357 Semaphore();
2358 ~Semaphore();
2359 void signal();
2360 void wait();
2361 bool trywait();
2362 bool timedwait(unsigned int sec, int nsec);
2363 private:
2364 sem_t _semaphore;
2365 };
2366
2367 Semaphore::Semaphore() {
2368 sem_init(&_semaphore, 0, 0);
2369 }
2370
2371 Semaphore::~Semaphore() {
2372 sem_destroy(&_semaphore);
2373 }
2374
2375 void Semaphore::signal() {
2376 sem_post(&_semaphore);
2377 }
2378
2379 void Semaphore::wait() {
2380 sem_wait(&_semaphore);
2381 }
2382
2383 bool Semaphore::trywait() {
2384 return sem_trywait(&_semaphore) == 0;
2385 }
2386
2387 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2388
2389 struct timespec ts;
2390 // Semaphore's are always associated with CLOCK_REALTIME
2391 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2392 // see unpackTime for discussion on overflow checking
2393 if (sec >= MAX_SECS) {
2394 ts.tv_sec += MAX_SECS;
2395 ts.tv_nsec = 0;
2396 } else {
2397 ts.tv_sec += sec;
2398 ts.tv_nsec += nsec;
2399 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2400 ts.tv_nsec -= NANOSECS_PER_SEC;
2401 ++ts.tv_sec; // note: this must be <= max_secs
2402 }
2403 }
2404
2405 while (1) {
2406 int result = sem_timedwait(&_semaphore, &ts);
2407 if (result == 0) {
2408 return true;
2409 } else if (errno == EINTR) {
2410 continue;
2411 } else if (errno == ETIMEDOUT) {
2412 return false;
2413 } else {
2414 return false;
2415 }
2416 }
2417 }
2418
2419 extern "C" {
2420 typedef void (*sa_handler_t)(int);
2421 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2422 }
2423
2424 void* os::signal(int signal_number, void* handler) {
2425 struct sigaction sigAct, oldSigAct;
2426
2427 sigfillset(&(sigAct.sa_mask));
2428 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2429 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2430
2431 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2432 // -1 means registration failed
2433 return (void *)-1;
2434 }
2435
2436 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2437 }
2438
2439 void os::signal_raise(int signal_number) {
2440 ::raise(signal_number);
2441 }
2442
2443 /*
2444 * The following code is moved from os.cpp for making this
2445 * code platform specific, which it is by its very nature.
2446 */
2447
2448 // Will be modified when max signal is changed to be dynamic
2449 int os::sigexitnum_pd() {
2450 return NSIG;
2451 }
2452
2453 // a counter for each possible signal value
2454 static volatile jint pending_signals[NSIG+1] = { 0 };
2455
2456 // Linux(POSIX) specific hand shaking semaphore.
2457 static sem_t sig_sem;
2458 static Semaphore sr_semaphore;
2459
2460 void os::signal_init_pd() {
2461 // Initialize signal structures
2462 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2463
2464 // Initialize signal semaphore
2465 ::sem_init(&sig_sem, 0, 0);
2466 }
2467
2468 void os::signal_notify(int sig) {
2469 Atomic::inc(&pending_signals[sig]);
2470 ::sem_post(&sig_sem);
2471 }
2472
2473 static int check_pending_signals(bool wait) {
2474 Atomic::store(0, &sigint_count);
2475 for (;;) {
2476 for (int i = 0; i < NSIG + 1; i++) {
2477 jint n = pending_signals[i];
2478 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2479 return i;
2480 }
2481 }
2482 if (!wait) {
2483 return -1;
2484 }
2485 JavaThread *thread = JavaThread::current();
2486 ThreadBlockInVM tbivm(thread);
2487
2488 bool threadIsSuspended;
2489 do {
2490 thread->set_suspend_equivalent();
2491 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2492 ::sem_wait(&sig_sem);
2493
2494 // were we externally suspended while we were waiting?
2495 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2496 if (threadIsSuspended) {
2497 //
2498 // The semaphore has been incremented, but while we were waiting
2499 // another thread suspended us. We don't want to continue running
2500 // while suspended because that would surprise the thread that
2501 // suspended us.
2502 //
2503 ::sem_post(&sig_sem);
2504
2505 thread->java_suspend_self();
2506 }
2507 } while (threadIsSuspended);
2508 }
2509 }
2510
2511 int os::signal_lookup() {
2512 return check_pending_signals(false);
2513 }
2514
2515 int os::signal_wait() {
2516 return check_pending_signals(true);
2517 }
2518
2519 ////////////////////////////////////////////////////////////////////////////////
2520 // Virtual Memory
2521
2522 int os::vm_page_size() {
2523 // Seems redundant as all get out
2524 assert(os::Linux::page_size() != -1, "must call os::init");
2525 return os::Linux::page_size();
2526 }
2527
2528 // Solaris allocates memory by pages.
2529 int os::vm_allocation_granularity() {
2530 assert(os::Linux::page_size() != -1, "must call os::init");
2531 return os::Linux::page_size();
2532 }
2533
2534 // Rationale behind this function:
2535 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2536 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2537 // samples for JITted code. Here we create private executable mapping over the code cache
2538 // and then we can use standard (well, almost, as mapping can change) way to provide
2539 // info for the reporting script by storing timestamp and location of symbol
2540 void linux_wrap_code(char* base, size_t size) {
2541 static volatile jint cnt = 0;
2542
2543 if (!UseOprofile) {
2544 return;
2545 }
2546
2547 char buf[PATH_MAX+1];
2548 int num = Atomic::add(1, &cnt);
2549
2550 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2551 os::get_temp_directory(), os::current_process_id(), num);
2552 unlink(buf);
2553
2554 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2555
2556 if (fd != -1) {
2557 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2558 if (rv != (off_t)-1) {
2559 if (::write(fd, "", 1) == 1) {
2560 mmap(base, size,
2561 PROT_READ|PROT_WRITE|PROT_EXEC,
2562 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2563 }
2564 }
2565 ::close(fd);
2566 unlink(buf);
2567 }
2568 }
2569
2570 static bool recoverable_mmap_error(int err) {
2571 // See if the error is one we can let the caller handle. This
2572 // list of errno values comes from JBS-6843484. I can't find a
2573 // Linux man page that documents this specific set of errno
2574 // values so while this list currently matches Solaris, it may
2575 // change as we gain experience with this failure mode.
2576 switch (err) {
2577 case EBADF:
2578 case EINVAL:
2579 case ENOTSUP:
2580 // let the caller deal with these errors
2581 return true;
2582
2583 default:
2584 // Any remaining errors on this OS can cause our reserved mapping
2585 // to be lost. That can cause confusion where different data
2586 // structures think they have the same memory mapped. The worst
2587 // scenario is if both the VM and a library think they have the
2588 // same memory mapped.
2589 return false;
2590 }
2591 }
2592
2593 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2594 int err) {
2595 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2596 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2597 strerror(err), err);
2598 }
2599
2600 static void warn_fail_commit_memory(char* addr, size_t size,
2601 size_t alignment_hint, bool exec,
2602 int err) {
2603 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2604 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2605 alignment_hint, exec, strerror(err), err);
2606 }
2607
2608 // NOTE: Linux kernel does not really reserve the pages for us.
2609 // All it does is to check if there are enough free pages
2610 // left at the time of mmap(). This could be a potential
2611 // problem.
2612 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2613 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2614 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2615 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2616 if (res != (uintptr_t) MAP_FAILED) {
2617 if (UseNUMAInterleaving) {
2618 numa_make_global(addr, size);
2619 }
2620 return 0;
2621 }
2622
2623 int err = errno; // save errno from mmap() call above
2624
2625 if (!recoverable_mmap_error(err)) {
2626 warn_fail_commit_memory(addr, size, exec, err);
2627 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2628 }
2629
2630 return err;
2631 }
2632
2633 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2634 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2635 }
2636
2637 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2638 const char* mesg) {
2639 assert(mesg != NULL, "mesg must be specified");
2640 int err = os::Linux::commit_memory_impl(addr, size, exec);
2641 if (err != 0) {
2642 // the caller wants all commit errors to exit with the specified mesg:
2643 warn_fail_commit_memory(addr, size, exec, err);
2644 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2645 }
2646 }
2647
2648 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2649 #ifndef MAP_HUGETLB
2650 #define MAP_HUGETLB 0x40000
2651 #endif
2652
2653 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2654 #ifndef MADV_HUGEPAGE
2655 #define MADV_HUGEPAGE 14
2656 #endif
2657
2658 int os::Linux::commit_memory_impl(char* addr, size_t size,
2659 size_t alignment_hint, bool exec) {
2660 int err = os::Linux::commit_memory_impl(addr, size, exec);
2661 if (err == 0) {
2662 realign_memory(addr, size, alignment_hint);
2663 }
2664 return err;
2665 }
2666
2667 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2668 bool exec) {
2669 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2670 }
2671
2672 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2673 size_t alignment_hint, bool exec,
2674 const char* mesg) {
2675 assert(mesg != NULL, "mesg must be specified");
2676 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2677 if (err != 0) {
2678 // the caller wants all commit errors to exit with the specified mesg:
2679 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2680 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2681 }
2682 }
2683
2684 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2685 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2686 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2687 // be supported or the memory may already be backed by huge pages.
2688 ::madvise(addr, bytes, MADV_HUGEPAGE);
2689 }
2690 }
2691
2692 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2693 // This method works by doing an mmap over an existing mmaping and effectively discarding
2694 // the existing pages. However it won't work for SHM-based large pages that cannot be
2695 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2696 // small pages on top of the SHM segment. This method always works for small pages, so we
2697 // allow that in any case.
2698 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2699 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2700 }
2701 }
2702
2703 void os::numa_make_global(char *addr, size_t bytes) {
2704 Linux::numa_interleave_memory(addr, bytes);
2705 }
2706
2707 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2708 // bind policy to MPOL_PREFERRED for the current thread.
2709 #define USE_MPOL_PREFERRED 0
2710
2711 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2712 // To make NUMA and large pages more robust when both enabled, we need to ease
2713 // the requirements on where the memory should be allocated. MPOL_BIND is the
2714 // default policy and it will force memory to be allocated on the specified
2715 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2716 // the specified node, but will not force it. Using this policy will prevent
2717 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2718 // free large pages.
2719 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2720 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2721 }
2722
2723 bool os::numa_topology_changed() { return false; }
2724
2725 size_t os::numa_get_groups_num() {
2726 int max_node = Linux::numa_max_node();
2727 return max_node > 0 ? max_node + 1 : 1;
2728 }
2729
2730 int os::numa_get_group_id() {
2731 int cpu_id = Linux::sched_getcpu();
2732 if (cpu_id != -1) {
2733 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2734 if (lgrp_id != -1) {
2735 return lgrp_id;
2736 }
2737 }
2738 return 0;
2739 }
2740
2741 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2742 for (size_t i = 0; i < size; i++) {
2743 ids[i] = i;
2744 }
2745 return size;
2746 }
2747
2748 bool os::get_page_info(char *start, page_info* info) {
2749 return false;
2750 }
2751
2752 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2753 return end;
2754 }
2755
2756
2757 int os::Linux::sched_getcpu_syscall(void) {
2758 unsigned int cpu = 0;
2759 int retval = -1;
2760
2761 #if defined(IA32)
2762 # ifndef SYS_getcpu
2763 # define SYS_getcpu 318
2764 # endif
2765 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2766 #elif defined(AMD64)
2767 // Unfortunately we have to bring all these macros here from vsyscall.h
2768 // to be able to compile on old linuxes.
2769 # define __NR_vgetcpu 2
2770 # define VSYSCALL_START (-10UL << 20)
2771 # define VSYSCALL_SIZE 1024
2772 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2773 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2774 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2775 retval = vgetcpu(&cpu, NULL, NULL);
2776 #endif
2777
2778 return (retval == -1) ? retval : cpu;
2779 }
2780
2781 // Something to do with the numa-aware allocator needs these symbols
2782 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2783 extern "C" JNIEXPORT void numa_error(char *where) { }
2784 extern "C" JNIEXPORT int fork1() { return fork(); }
2785
2786
2787 // If we are running with libnuma version > 2, then we should
2788 // be trying to use symbols with versions 1.1
2789 // If we are running with earlier version, which did not have symbol versions,
2790 // we should use the base version.
2791 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2792 void *f = dlvsym(handle, name, "libnuma_1.1");
2793 if (f == NULL) {
2794 f = dlsym(handle, name);
2795 }
2796 return f;
2797 }
2798
2799 bool os::Linux::libnuma_init() {
2800 // sched_getcpu() should be in libc.
2801 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2802 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2803
2804 // If it's not, try a direct syscall.
2805 if (sched_getcpu() == -1)
2806 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2807
2808 if (sched_getcpu() != -1) { // Does it work?
2809 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2810 if (handle != NULL) {
2811 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2812 libnuma_dlsym(handle, "numa_node_to_cpus")));
2813 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2814 libnuma_dlsym(handle, "numa_max_node")));
2815 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2816 libnuma_dlsym(handle, "numa_available")));
2817 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2818 libnuma_dlsym(handle, "numa_tonode_memory")));
2819 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2820 libnuma_dlsym(handle, "numa_interleave_memory")));
2821 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2822 libnuma_dlsym(handle, "numa_set_bind_policy")));
2823
2824
2825 if (numa_available() != -1) {
2826 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2827 // Create a cpu -> node mapping
2828 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2829 rebuild_cpu_to_node_map();
2830 return true;
2831 }
2832 }
2833 }
2834 return false;
2835 }
2836
2837 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2838 // The table is later used in get_node_by_cpu().
2839 void os::Linux::rebuild_cpu_to_node_map() {
2840 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2841 // in libnuma (possible values are starting from 16,
2842 // and continuing up with every other power of 2, but less
2843 // than the maximum number of CPUs supported by kernel), and
2844 // is a subject to change (in libnuma version 2 the requirements
2845 // are more reasonable) we'll just hardcode the number they use
2846 // in the library.
2847 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2848
2849 size_t cpu_num = os::active_processor_count();
2850 size_t cpu_map_size = NCPUS / BitsPerCLong;
2851 size_t cpu_map_valid_size =
2852 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2853
2854 cpu_to_node()->clear();
2855 cpu_to_node()->at_grow(cpu_num - 1);
2856 size_t node_num = numa_get_groups_num();
2857
2858 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2859 for (size_t i = 0; i < node_num; i++) {
2860 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2861 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2862 if (cpu_map[j] != 0) {
2863 for (size_t k = 0; k < BitsPerCLong; k++) {
2864 if (cpu_map[j] & (1UL << k)) {
2865 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2866 }
2867 }
2868 }
2869 }
2870 }
2871 }
2872 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2873 }
2874
2875 int os::Linux::get_node_by_cpu(int cpu_id) {
2876 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2877 return cpu_to_node()->at(cpu_id);
2878 }
2879 return -1;
2880 }
2881
2882 GrowableArray<int>* os::Linux::_cpu_to_node;
2883 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2884 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2885 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2886 os::Linux::numa_available_func_t os::Linux::_numa_available;
2887 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2888 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2889 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2890 unsigned long* os::Linux::_numa_all_nodes;
2891
2892 bool os::pd_uncommit_memory(char* addr, size_t size) {
2893 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2894 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2895 return res != (uintptr_t) MAP_FAILED;
2896 }
2897
2898 static
2899 address get_stack_commited_bottom(address bottom, size_t size) {
2900 address nbot = bottom;
2901 address ntop = bottom + size;
2902
2903 size_t page_sz = os::vm_page_size();
2904 unsigned pages = size / page_sz;
2905
2906 unsigned char vec[1];
2907 unsigned imin = 1, imax = pages + 1, imid;
2908 int mincore_return_value = 0;
2909
2910 assert(imin <= imax, "Unexpected page size");
2911
2912 while (imin < imax) {
2913 imid = (imax + imin) / 2;
2914 nbot = ntop - (imid * page_sz);
2915
2916 // Use a trick with mincore to check whether the page is mapped or not.
2917 // mincore sets vec to 1 if page resides in memory and to 0 if page
2918 // is swapped output but if page we are asking for is unmapped
2919 // it returns -1,ENOMEM
2920 mincore_return_value = mincore(nbot, page_sz, vec);
2921
2922 if (mincore_return_value == -1) {
2923 // Page is not mapped go up
2924 // to find first mapped page
2925 if (errno != EAGAIN) {
2926 assert(errno == ENOMEM, "Unexpected mincore errno");
2927 imax = imid;
2928 }
2929 } else {
2930 // Page is mapped go down
2931 // to find first not mapped page
2932 imin = imid + 1;
2933 }
2934 }
2935
2936 nbot = nbot + page_sz;
2937
2938 // Adjust stack bottom one page up if last checked page is not mapped
2939 if (mincore_return_value == -1) {
2940 nbot = nbot + page_sz;
2941 }
2942
2943 return nbot;
2944 }
2945
2946
2947 // Linux uses a growable mapping for the stack, and if the mapping for
2948 // the stack guard pages is not removed when we detach a thread the
2949 // stack cannot grow beyond the pages where the stack guard was
2950 // mapped. If at some point later in the process the stack expands to
2951 // that point, the Linux kernel cannot expand the stack any further
2952 // because the guard pages are in the way, and a segfault occurs.
2953 //
2954 // However, it's essential not to split the stack region by unmapping
2955 // a region (leaving a hole) that's already part of the stack mapping,
2956 // so if the stack mapping has already grown beyond the guard pages at
2957 // the time we create them, we have to truncate the stack mapping.
2958 // So, we need to know the extent of the stack mapping when
2959 // create_stack_guard_pages() is called.
2960
2961 // We only need this for stacks that are growable: at the time of
2962 // writing thread stacks don't use growable mappings (i.e. those
2963 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2964 // only applies to the main thread.
2965
2966 // If the (growable) stack mapping already extends beyond the point
2967 // where we're going to put our guard pages, truncate the mapping at
2968 // that point by munmap()ping it. This ensures that when we later
2969 // munmap() the guard pages we don't leave a hole in the stack
2970 // mapping. This only affects the main/initial thread
2971
2972 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2973
2974 if (os::Linux::is_initial_thread()) {
2975 // As we manually grow stack up to bottom inside create_attached_thread(),
2976 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2977 // we don't need to do anything special.
2978 // Check it first, before calling heavy function.
2979 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2980 unsigned char vec[1];
2981
2982 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2983 // Fallback to slow path on all errors, including EAGAIN
2984 stack_extent = (uintptr_t) get_stack_commited_bottom(
2985 os::Linux::initial_thread_stack_bottom(),
2986 (size_t)addr - stack_extent);
2987 }
2988
2989 if (stack_extent < (uintptr_t)addr) {
2990 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2991 }
2992 }
2993
2994 return os::commit_memory(addr, size, !ExecMem);
2995 }
2996
2997 // If this is a growable mapping, remove the guard pages entirely by
2998 // munmap()ping them. If not, just call uncommit_memory(). This only
2999 // affects the main/initial thread, but guard against future OS changes
3000 // It's safe to always unmap guard pages for initial thread because we
3001 // always place it right after end of the mapped region
3002
3003 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3004 uintptr_t stack_extent, stack_base;
3005
3006 if (os::Linux::is_initial_thread()) {
3007 return ::munmap(addr, size) == 0;
3008 }
3009
3010 return os::uncommit_memory(addr, size);
3011 }
3012
3013 static address _highest_vm_reserved_address = NULL;
3014
3015 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3016 // at 'requested_addr'. If there are existing memory mappings at the same
3017 // location, however, they will be overwritten. If 'fixed' is false,
3018 // 'requested_addr' is only treated as a hint, the return value may or
3019 // may not start from the requested address. Unlike Linux mmap(), this
3020 // function returns NULL to indicate failure.
3021 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3022 char * addr;
3023 int flags;
3024
3025 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3026 if (fixed) {
3027 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3028 flags |= MAP_FIXED;
3029 }
3030
3031 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3032 // touch an uncommitted page. Otherwise, the read/write might
3033 // succeed if we have enough swap space to back the physical page.
3034 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3035 flags, -1, 0);
3036
3037 if (addr != MAP_FAILED) {
3038 // anon_mmap() should only get called during VM initialization,
3039 // don't need lock (actually we can skip locking even it can be called
3040 // from multiple threads, because _highest_vm_reserved_address is just a
3041 // hint about the upper limit of non-stack memory regions.)
3042 if ((address)addr + bytes > _highest_vm_reserved_address) {
3043 _highest_vm_reserved_address = (address)addr + bytes;
3044 }
3045 }
3046
3047 return addr == MAP_FAILED ? NULL : addr;
3048 }
3049
3050 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3051 // (req_addr != NULL) or with a given alignment.
3052 // - bytes shall be a multiple of alignment.
3053 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3054 // - alignment sets the alignment at which memory shall be allocated.
3055 // It must be a multiple of allocation granularity.
3056 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3057 // req_addr or NULL.
3058 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3059
3060 size_t extra_size = bytes;
3061 if (req_addr == NULL && alignment > 0) {
3062 extra_size += alignment;
3063 }
3064
3065 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3066 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3067 -1, 0);
3068 if (start == MAP_FAILED) {
3069 start = NULL;
3070 } else {
3071 if (req_addr != NULL) {
3072 if (start != req_addr) {
3073 ::munmap(start, extra_size);
3074 start = NULL;
3075 }
3076 } else {
3077 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3078 char* const end_aligned = start_aligned + bytes;
3079 char* const end = start + extra_size;
3080 if (start_aligned > start) {
3081 ::munmap(start, start_aligned - start);
3082 }
3083 if (end_aligned < end) {
3084 ::munmap(end_aligned, end - end_aligned);
3085 }
3086 start = start_aligned;
3087 }
3088 }
3089 return start;
3090 }
3091
3092 // Don't update _highest_vm_reserved_address, because there might be memory
3093 // regions above addr + size. If so, releasing a memory region only creates
3094 // a hole in the address space, it doesn't help prevent heap-stack collision.
3095 //
3096 static int anon_munmap(char * addr, size_t size) {
3097 return ::munmap(addr, size) == 0;
3098 }
3099
3100 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3101 size_t alignment_hint) {
3102 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3103 }
3104
3105 bool os::pd_release_memory(char* addr, size_t size) {
3106 return anon_munmap(addr, size);
3107 }
3108
3109 static address highest_vm_reserved_address() {
3110 return _highest_vm_reserved_address;
3111 }
3112
3113 static bool linux_mprotect(char* addr, size_t size, int prot) {
3114 // Linux wants the mprotect address argument to be page aligned.
3115 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3116
3117 // According to SUSv3, mprotect() should only be used with mappings
3118 // established by mmap(), and mmap() always maps whole pages. Unaligned
3119 // 'addr' likely indicates problem in the VM (e.g. trying to change
3120 // protection of malloc'ed or statically allocated memory). Check the
3121 // caller if you hit this assert.
3122 assert(addr == bottom, "sanity check");
3123
3124 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3125 return ::mprotect(bottom, size, prot) == 0;
3126 }
3127
3128 // Set protections specified
3129 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3130 bool is_committed) {
3131 unsigned int p = 0;
3132 switch (prot) {
3133 case MEM_PROT_NONE: p = PROT_NONE; break;
3134 case MEM_PROT_READ: p = PROT_READ; break;
3135 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3136 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3137 default:
3138 ShouldNotReachHere();
3139 }
3140 // is_committed is unused.
3141 return linux_mprotect(addr, bytes, p);
3142 }
3143
3144 bool os::guard_memory(char* addr, size_t size) {
3145 return linux_mprotect(addr, size, PROT_NONE);
3146 }
3147
3148 bool os::unguard_memory(char* addr, size_t size) {
3149 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3150 }
3151
3152 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3153 bool result = false;
3154 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3155 MAP_ANONYMOUS|MAP_PRIVATE,
3156 -1, 0);
3157 if (p != MAP_FAILED) {
3158 void *aligned_p = align_ptr_up(p, page_size);
3159
3160 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3161
3162 munmap(p, page_size * 2);
3163 }
3164
3165 if (warn && !result) {
3166 warning("TransparentHugePages is not supported by the operating system.");
3167 }
3168
3169 return result;
3170 }
3171
3172 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3173 bool result = false;
3174 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3175 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3176 -1, 0);
3177
3178 if (p != MAP_FAILED) {
3179 // We don't know if this really is a huge page or not.
3180 FILE *fp = fopen("/proc/self/maps", "r");
3181 if (fp) {
3182 while (!feof(fp)) {
3183 char chars[257];
3184 long x = 0;
3185 if (fgets(chars, sizeof(chars), fp)) {
3186 if (sscanf(chars, "%lx-%*x", &x) == 1
3187 && x == (long)p) {
3188 if (strstr (chars, "hugepage")) {
3189 result = true;
3190 break;
3191 }
3192 }
3193 }
3194 }
3195 fclose(fp);
3196 }
3197 munmap(p, page_size);
3198 }
3199
3200 if (warn && !result) {
3201 warning("HugeTLBFS is not supported by the operating system.");
3202 }
3203
3204 return result;
3205 }
3206
3207 /*
3208 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3209 *
3210 * From the coredump_filter documentation:
3211 *
3212 * - (bit 0) anonymous private memory
3213 * - (bit 1) anonymous shared memory
3214 * - (bit 2) file-backed private memory
3215 * - (bit 3) file-backed shared memory
3216 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3217 * effective only if the bit 2 is cleared)
3218 * - (bit 5) hugetlb private memory
3219 * - (bit 6) hugetlb shared memory
3220 */
3221 static void set_coredump_filter(void) {
3222 FILE *f;
3223 long cdm;
3224
3225 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3226 return;
3227 }
3228
3229 if (fscanf(f, "%lx", &cdm) != 1) {
3230 fclose(f);
3231 return;
3232 }
3233
3234 rewind(f);
3235
3236 if ((cdm & LARGEPAGES_BIT) == 0) {
3237 cdm |= LARGEPAGES_BIT;
3238 fprintf(f, "%#lx", cdm);
3239 }
3240
3241 fclose(f);
3242 }
3243
3244 // Large page support
3245
3246 static size_t _large_page_size = 0;
3247
3248 size_t os::Linux::find_large_page_size() {
3249 size_t large_page_size = 0;
3250
3251 // large_page_size on Linux is used to round up heap size. x86 uses either
3252 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3253 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3254 // page as large as 256M.
3255 //
3256 // Here we try to figure out page size by parsing /proc/meminfo and looking
3257 // for a line with the following format:
3258 // Hugepagesize: 2048 kB
3259 //
3260 // If we can't determine the value (e.g. /proc is not mounted, or the text
3261 // format has been changed), we'll use the largest page size supported by
3262 // the processor.
3263
3264 #ifndef ZERO
3265 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3266 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3267 #endif // ZERO
3268
3269 FILE *fp = fopen("/proc/meminfo", "r");
3270 if (fp) {
3271 while (!feof(fp)) {
3272 int x = 0;
3273 char buf[16];
3274 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3275 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3276 large_page_size = x * K;
3277 break;
3278 }
3279 } else {
3280 // skip to next line
3281 for (;;) {
3282 int ch = fgetc(fp);
3283 if (ch == EOF || ch == (int)'\n') break;
3284 }
3285 }
3286 }
3287 fclose(fp);
3288 }
3289
3290 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3291 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3292 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3293 proper_unit_for_byte_size(large_page_size));
3294 }
3295
3296 return large_page_size;
3297 }
3298
3299 size_t os::Linux::setup_large_page_size() {
3300 _large_page_size = Linux::find_large_page_size();
3301 const size_t default_page_size = (size_t)Linux::page_size();
3302 if (_large_page_size > default_page_size) {
3303 _page_sizes[0] = _large_page_size;
3304 _page_sizes[1] = default_page_size;
3305 _page_sizes[2] = 0;
3306 }
3307
3308 return _large_page_size;
3309 }
3310
3311 bool os::Linux::setup_large_page_type(size_t page_size) {
3312 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3313 FLAG_IS_DEFAULT(UseSHM) &&
3314 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3315
3316 // The type of large pages has not been specified by the user.
3317
3318 // Try UseHugeTLBFS and then UseSHM.
3319 UseHugeTLBFS = UseSHM = true;
3320
3321 // Don't try UseTransparentHugePages since there are known
3322 // performance issues with it turned on. This might change in the future.
3323 UseTransparentHugePages = false;
3324 }
3325
3326 if (UseTransparentHugePages) {
3327 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3328 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3329 UseHugeTLBFS = false;
3330 UseSHM = false;
3331 return true;
3332 }
3333 UseTransparentHugePages = false;
3334 }
3335
3336 if (UseHugeTLBFS) {
3337 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3338 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3339 UseSHM = false;
3340 return true;
3341 }
3342 UseHugeTLBFS = false;
3343 }
3344
3345 return UseSHM;
3346 }
3347
3348 void os::large_page_init() {
3349 if (!UseLargePages &&
3350 !UseTransparentHugePages &&
3351 !UseHugeTLBFS &&
3352 !UseSHM) {
3353 // Not using large pages.
3354 return;
3355 }
3356
3357 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3358 // The user explicitly turned off large pages.
3359 // Ignore the rest of the large pages flags.
3360 UseTransparentHugePages = false;
3361 UseHugeTLBFS = false;
3362 UseSHM = false;
3363 return;
3364 }
3365
3366 size_t large_page_size = Linux::setup_large_page_size();
3367 UseLargePages = Linux::setup_large_page_type(large_page_size);
3368
3369 set_coredump_filter();
3370 }
3371
3372 #ifndef SHM_HUGETLB
3373 #define SHM_HUGETLB 04000
3374 #endif
3375
3376 #define shm_warning_format(format, ...) \
3377 do { \
3378 if (UseLargePages && \
3379 (!FLAG_IS_DEFAULT(UseLargePages) || \
3380 !FLAG_IS_DEFAULT(UseSHM) || \
3381 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3382 warning(format, __VA_ARGS__); \
3383 } \
3384 } while (0)
3385
3386 #define shm_warning(str) shm_warning_format("%s", str)
3387
3388 #define shm_warning_with_errno(str) \
3389 do { \
3390 int err = errno; \
3391 shm_warning_format(str " (error = %d)", err); \
3392 } while (0)
3393
3394 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3395 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3396
3397 if (!is_size_aligned(alignment, SHMLBA)) {
3398 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3399 return NULL;
3400 }
3401
3402 // To ensure that we get 'alignment' aligned memory from shmat,
3403 // we pre-reserve aligned virtual memory and then attach to that.
3404
3405 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3406 if (pre_reserved_addr == NULL) {
3407 // Couldn't pre-reserve aligned memory.
3408 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3409 return NULL;
3410 }
3411
3412 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3413 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3414
3415 if ((intptr_t)addr == -1) {
3416 int err = errno;
3417 shm_warning_with_errno("Failed to attach shared memory.");
3418
3419 assert(err != EACCES, "Unexpected error");
3420 assert(err != EIDRM, "Unexpected error");
3421 assert(err != EINVAL, "Unexpected error");
3422
3423 // Since we don't know if the kernel unmapped the pre-reserved memory area
3424 // we can't unmap it, since that would potentially unmap memory that was
3425 // mapped from other threads.
3426 return NULL;
3427 }
3428
3429 return addr;
3430 }
3431
3432 static char* shmat_at_address(int shmid, char* req_addr) {
3433 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3434 assert(false, "Requested address needs to be SHMLBA aligned");
3435 return NULL;
3436 }
3437
3438 char* addr = (char*)shmat(shmid, req_addr, 0);
3439
3440 if ((intptr_t)addr == -1) {
3441 shm_warning_with_errno("Failed to attach shared memory.");
3442 return NULL;
3443 }
3444
3445 return addr;
3446 }
3447
3448 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3449 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3450 if (req_addr != NULL) {
3451 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3452 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3453 return shmat_at_address(shmid, req_addr);
3454 }
3455
3456 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3457 // return large page size aligned memory addresses when req_addr == NULL.
3458 // However, if the alignment is larger than the large page size, we have
3459 // to manually ensure that the memory returned is 'alignment' aligned.
3460 if (alignment > os::large_page_size()) {
3461 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3462 return shmat_with_alignment(shmid, bytes, alignment);
3463 } else {
3464 return shmat_at_address(shmid, NULL);
3465 }
3466 }
3467
3468 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3469 // "exec" is passed in but not used. Creating the shared image for
3470 // the code cache doesn't have an SHM_X executable permission to check.
3471 assert(UseLargePages && UseSHM, "only for SHM large pages");
3472 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3473 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3474
3475 if (!is_size_aligned(bytes, os::large_page_size())) {
3476 return NULL; // Fallback to small pages.
3477 }
3478
3479 // Create a large shared memory region to attach to based on size.
3480 // Currently, size is the total size of the heap.
3481 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3482 if (shmid == -1) {
3483 // Possible reasons for shmget failure:
3484 // 1. shmmax is too small for Java heap.
3485 // > check shmmax value: cat /proc/sys/kernel/shmmax
3486 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3487 // 2. not enough large page memory.
3488 // > check available large pages: cat /proc/meminfo
3489 // > increase amount of large pages:
3490 // echo new_value > /proc/sys/vm/nr_hugepages
3491 // Note 1: different Linux may use different name for this property,
3492 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3493 // Note 2: it's possible there's enough physical memory available but
3494 // they are so fragmented after a long run that they can't
3495 // coalesce into large pages. Try to reserve large pages when
3496 // the system is still "fresh".
3497 shm_warning_with_errno("Failed to reserve shared memory.");
3498 return NULL;
3499 }
3500
3501 // Attach to the region.
3502 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3503
3504 // Remove shmid. If shmat() is successful, the actual shared memory segment
3505 // will be deleted when it's detached by shmdt() or when the process
3506 // terminates. If shmat() is not successful this will remove the shared
3507 // segment immediately.
3508 shmctl(shmid, IPC_RMID, NULL);
3509
3510 return addr;
3511 }
3512
3513 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3514 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3515
3516 bool warn_on_failure = UseLargePages &&
3517 (!FLAG_IS_DEFAULT(UseLargePages) ||
3518 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3519 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3520
3521 if (warn_on_failure) {
3522 char msg[128];
3523 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3524 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3525 warning("%s", msg);
3526 }
3527 }
3528
3529 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3530 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3531 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3532 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3533
3534 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3535 char* addr = (char*)::mmap(req_addr, bytes, prot,
3536 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3537 -1, 0);
3538
3539 if (addr == MAP_FAILED) {
3540 warn_on_large_pages_failure(req_addr, bytes, errno);
3541 return NULL;
3542 }
3543
3544 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3545
3546 return addr;
3547 }
3548
3549 // Reserve memory using mmap(MAP_HUGETLB).
3550 // - bytes shall be a multiple of alignment.
3551 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3552 // - alignment sets the alignment at which memory shall be allocated.
3553 // It must be a multiple of allocation granularity.
3554 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3555 // req_addr or NULL.
3556 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3557 size_t large_page_size = os::large_page_size();
3558 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3559
3560 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3561 assert(is_size_aligned(bytes, alignment), "Must be");
3562
3563 // First reserve - but not commit - the address range in small pages.
3564 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3565
3566 if (start == NULL) {
3567 return NULL;
3568 }
3569
3570 assert(is_ptr_aligned(start, alignment), "Must be");
3571
3572 char* end = start + bytes;
3573
3574 // Find the regions of the allocated chunk that can be promoted to large pages.
3575 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3576 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3577
3578 size_t lp_bytes = lp_end - lp_start;
3579
3580 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3581
3582 if (lp_bytes == 0) {
3583 // The mapped region doesn't even span the start and the end of a large page.
3584 // Fall back to allocate a non-special area.
3585 ::munmap(start, end - start);
3586 return NULL;
3587 }
3588
3589 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3590
3591 void* result;
3592
3593 // Commit small-paged leading area.
3594 if (start != lp_start) {
3595 result = ::mmap(start, lp_start - start, prot,
3596 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3597 -1, 0);
3598 if (result == MAP_FAILED) {
3599 ::munmap(lp_start, end - lp_start);
3600 return NULL;
3601 }
3602 }
3603
3604 // Commit large-paged area.
3605 result = ::mmap(lp_start, lp_bytes, prot,
3606 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3607 -1, 0);
3608 if (result == MAP_FAILED) {
3609 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3610 // If the mmap above fails, the large pages region will be unmapped and we
3611 // have regions before and after with small pages. Release these regions.
3612 //
3613 // | mapped | unmapped | mapped |
3614 // ^ ^ ^ ^
3615 // start lp_start lp_end end
3616 //
3617 ::munmap(start, lp_start - start);
3618 ::munmap(lp_end, end - lp_end);
3619 return NULL;
3620 }
3621
3622 // Commit small-paged trailing area.
3623 if (lp_end != end) {
3624 result = ::mmap(lp_end, end - lp_end, prot,
3625 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3626 -1, 0);
3627 if (result == MAP_FAILED) {
3628 ::munmap(start, lp_end - start);
3629 return NULL;
3630 }
3631 }
3632
3633 return start;
3634 }
3635
3636 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3637 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3638 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3639 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3640 assert(is_power_of_2(os::large_page_size()), "Must be");
3641 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3642
3643 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3644 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3645 } else {
3646 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3647 }
3648 }
3649
3650 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3651 assert(UseLargePages, "only for large pages");
3652
3653 char* addr;
3654 if (UseSHM) {
3655 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3656 } else {
3657 assert(UseHugeTLBFS, "must be");
3658 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3659 }
3660
3661 if (addr != NULL) {
3662 if (UseNUMAInterleaving) {
3663 numa_make_global(addr, bytes);
3664 }
3665
3666 // The memory is committed
3667 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3668 }
3669
3670 return addr;
3671 }
3672
3673 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3674 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3675 return shmdt(base) == 0;
3676 }
3677
3678 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3679 return pd_release_memory(base, bytes);
3680 }
3681
3682 bool os::release_memory_special(char* base, size_t bytes) {
3683 bool res;
3684 if (MemTracker::tracking_level() > NMT_minimal) {
3685 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3686 res = os::Linux::release_memory_special_impl(base, bytes);
3687 if (res) {
3688 tkr.record((address)base, bytes);
3689 }
3690
3691 } else {
3692 res = os::Linux::release_memory_special_impl(base, bytes);
3693 }
3694 return res;
3695 }
3696
3697 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3698 assert(UseLargePages, "only for large pages");
3699 bool res;
3700
3701 if (UseSHM) {
3702 res = os::Linux::release_memory_special_shm(base, bytes);
3703 } else {
3704 assert(UseHugeTLBFS, "must be");
3705 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3706 }
3707 return res;
3708 }
3709
3710 size_t os::large_page_size() {
3711 return _large_page_size;
3712 }
3713
3714 // With SysV SHM the entire memory region must be allocated as shared
3715 // memory.
3716 // HugeTLBFS allows application to commit large page memory on demand.
3717 // However, when committing memory with HugeTLBFS fails, the region
3718 // that was supposed to be committed will lose the old reservation
3719 // and allow other threads to steal that memory region. Because of this
3720 // behavior we can't commit HugeTLBFS memory.
3721 bool os::can_commit_large_page_memory() {
3722 return UseTransparentHugePages;
3723 }
3724
3725 bool os::can_execute_large_page_memory() {
3726 return UseTransparentHugePages || UseHugeTLBFS;
3727 }
3728
3729 // Reserve memory at an arbitrary address, only if that area is
3730 // available (and not reserved for something else).
3731
3732 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3733 const int max_tries = 10;
3734 char* base[max_tries];
3735 size_t size[max_tries];
3736 const size_t gap = 0x000000;
3737
3738 // Assert only that the size is a multiple of the page size, since
3739 // that's all that mmap requires, and since that's all we really know
3740 // about at this low abstraction level. If we need higher alignment,
3741 // we can either pass an alignment to this method or verify alignment
3742 // in one of the methods further up the call chain. See bug 5044738.
3743 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3744
3745 // Repeatedly allocate blocks until the block is allocated at the
3746 // right spot. Give up after max_tries. Note that reserve_memory() will
3747 // automatically update _highest_vm_reserved_address if the call is
3748 // successful. The variable tracks the highest memory address every reserved
3749 // by JVM. It is used to detect heap-stack collision if running with
3750 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3751 // space than needed, it could confuse the collision detecting code. To
3752 // solve the problem, save current _highest_vm_reserved_address and
3753 // calculate the correct value before return.
3754 address old_highest = _highest_vm_reserved_address;
3755
3756 // Linux mmap allows caller to pass an address as hint; give it a try first,
3757 // if kernel honors the hint then we can return immediately.
3758 char * addr = anon_mmap(requested_addr, bytes, false);
3759 if (addr == requested_addr) {
3760 return requested_addr;
3761 }
3762
3763 if (addr != NULL) {
3764 // mmap() is successful but it fails to reserve at the requested address
3765 anon_munmap(addr, bytes);
3766 }
3767
3768 int i;
3769 for (i = 0; i < max_tries; ++i) {
3770 base[i] = reserve_memory(bytes);
3771
3772 if (base[i] != NULL) {
3773 // Is this the block we wanted?
3774 if (base[i] == requested_addr) {
3775 size[i] = bytes;
3776 break;
3777 }
3778
3779 // Does this overlap the block we wanted? Give back the overlapped
3780 // parts and try again.
3781
3782 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3783 if (top_overlap >= 0 && top_overlap < bytes) {
3784 unmap_memory(base[i], top_overlap);
3785 base[i] += top_overlap;
3786 size[i] = bytes - top_overlap;
3787 } else {
3788 size_t bottom_overlap = base[i] + bytes - requested_addr;
3789 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3790 unmap_memory(requested_addr, bottom_overlap);
3791 size[i] = bytes - bottom_overlap;
3792 } else {
3793 size[i] = bytes;
3794 }
3795 }
3796 }
3797 }
3798
3799 // Give back the unused reserved pieces.
3800
3801 for (int j = 0; j < i; ++j) {
3802 if (base[j] != NULL) {
3803 unmap_memory(base[j], size[j]);
3804 }
3805 }
3806
3807 if (i < max_tries) {
3808 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3809 return requested_addr;
3810 } else {
3811 _highest_vm_reserved_address = old_highest;
3812 return NULL;
3813 }
3814 }
3815
3816 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3817 return ::read(fd, buf, nBytes);
3818 }
3819
3820 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3821 // Solaris uses poll(), linux uses park().
3822 // Poll() is likely a better choice, assuming that Thread.interrupt()
3823 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3824 // SIGSEGV, see 4355769.
3825
3826 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3827 assert(thread == Thread::current(), "thread consistency check");
3828
3829 ParkEvent * const slp = thread->_SleepEvent ;
3830 slp->reset() ;
3831 OrderAccess::fence() ;
3832
3833 if (interruptible) {
3834 jlong prevtime = javaTimeNanos();
3835
3836 for (;;) {
3837 if (os::is_interrupted(thread, true)) {
3838 return OS_INTRPT;
3839 }
3840
3841 jlong newtime = javaTimeNanos();
3842
3843 if (newtime - prevtime < 0) {
3844 // time moving backwards, should only happen if no monotonic clock
3845 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3846 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3847 } else {
3848 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3849 }
3850
3851 if(millis <= 0) {
3852 return OS_OK;
3853 }
3854
3855 prevtime = newtime;
3856
3857 {
3858 assert(thread->is_Java_thread(), "sanity check");
3859 JavaThread *jt = (JavaThread *) thread;
3860 ThreadBlockInVM tbivm(jt);
3861 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3862
3863 jt->set_suspend_equivalent();
3864 // cleared by handle_special_suspend_equivalent_condition() or
3865 // java_suspend_self() via check_and_wait_while_suspended()
3866
3867 slp->park(millis);
3868
3869 // were we externally suspended while we were waiting?
3870 jt->check_and_wait_while_suspended();
3871 }
3872 }
3873 } else {
3874 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3875 jlong prevtime = javaTimeNanos();
3876
3877 for (;;) {
3878 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3879 // the 1st iteration ...
3880 jlong newtime = javaTimeNanos();
3881
3882 if (newtime - prevtime < 0) {
3883 // time moving backwards, should only happen if no monotonic clock
3884 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3885 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3886 } else {
3887 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3888 }
3889
3890 if(millis <= 0) break ;
3891
3892 prevtime = newtime;
3893 slp->park(millis);
3894 }
3895 return OS_OK ;
3896 }
3897 }
3898
3899 //
3900 // Short sleep, direct OS call.
3901 //
3902 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3903 // sched_yield(2) will actually give up the CPU:
3904 //
3905 // * Alone on this pariticular CPU, keeps running.
3906 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3907 // (pre 2.6.39).
3908 //
3909 // So calling this with 0 is an alternative.
3910 //
3911 void os::naked_short_sleep(jlong ms) {
3912 struct timespec req;
3913
3914 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3915 req.tv_sec = 0;
3916 if (ms > 0) {
3917 req.tv_nsec = (ms % 1000) * 1000000;
3918 }
3919 else {
3920 req.tv_nsec = 1;
3921 }
3922
3923 nanosleep(&req, NULL);
3924
3925 return;
3926 }
3927
3928 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3929 void os::infinite_sleep() {
3930 while (true) { // sleep forever ...
3931 ::sleep(100); // ... 100 seconds at a time
3932 }
3933 }
3934
3935 // Used to convert frequent JVM_Yield() to nops
3936 bool os::dont_yield() {
3937 return DontYieldALot;
3938 }
3939
3940 void os::yield() {
3941 sched_yield();
3942 }
3943
3944 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3945
3946 void os::yield_all(int attempts) {
3947 // Yields to all threads, including threads with lower priorities
3948 // Threads on Linux are all with same priority. The Solaris style
3949 // os::yield_all() with nanosleep(1ms) is not necessary.
3950 sched_yield();
3951 }
3952
3953 // Called from the tight loops to possibly influence time-sharing heuristics
3954 void os::loop_breaker(int attempts) {
3955 os::yield_all(attempts);
3956 }
3957
3958 ////////////////////////////////////////////////////////////////////////////////
3959 // thread priority support
3960
3961 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3962 // only supports dynamic priority, static priority must be zero. For real-time
3963 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3964 // However, for large multi-threaded applications, SCHED_RR is not only slower
3965 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3966 // of 5 runs - Sep 2005).
3967 //
3968 // The following code actually changes the niceness of kernel-thread/LWP. It
3969 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3970 // not the entire user process, and user level threads are 1:1 mapped to kernel
3971 // threads. It has always been the case, but could change in the future. For
3972 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3973 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3974
3975 int os::java_to_os_priority[CriticalPriority + 1] = {
3976 19, // 0 Entry should never be used
3977
3978 4, // 1 MinPriority
3979 3, // 2
3980 2, // 3
3981
3982 1, // 4
3983 0, // 5 NormPriority
3984 -1, // 6
3985
3986 -2, // 7
3987 -3, // 8
3988 -4, // 9 NearMaxPriority
3989
3990 -5, // 10 MaxPriority
3991
3992 -5 // 11 CriticalPriority
3993 };
3994
3995 static int prio_init() {
3996 if (ThreadPriorityPolicy == 1) {
3997 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3998 // if effective uid is not root. Perhaps, a more elegant way of doing
3999 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4000 if (geteuid() != 0) {
4001 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4002 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4003 }
4004 ThreadPriorityPolicy = 0;
4005 }
4006 }
4007 if (UseCriticalJavaThreadPriority) {
4008 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4009 }
4010 return 0;
4011 }
4012
4013 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4014 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4015
4016 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4017 return (ret == 0) ? OS_OK : OS_ERR;
4018 }
4019
4020 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4021 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4022 *priority_ptr = java_to_os_priority[NormPriority];
4023 return OS_OK;
4024 }
4025
4026 errno = 0;
4027 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4028 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4029 }
4030
4031 // Hint to the underlying OS that a task switch would not be good.
4032 // Void return because it's a hint and can fail.
4033 void os::hint_no_preempt() {}
4034
4035 ////////////////////////////////////////////////////////////////////////////////
4036 // suspend/resume support
4037
4038 // the low-level signal-based suspend/resume support is a remnant from the
4039 // old VM-suspension that used to be for java-suspension, safepoints etc,
4040 // within hotspot. Now there is a single use-case for this:
4041 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4042 // that runs in the watcher thread.
4043 // The remaining code is greatly simplified from the more general suspension
4044 // code that used to be used.
4045 //
4046 // The protocol is quite simple:
4047 // - suspend:
4048 // - sends a signal to the target thread
4049 // - polls the suspend state of the osthread using a yield loop
4050 // - target thread signal handler (SR_handler) sets suspend state
4051 // and blocks in sigsuspend until continued
4052 // - resume:
4053 // - sets target osthread state to continue
4054 // - sends signal to end the sigsuspend loop in the SR_handler
4055 //
4056 // Note that the SR_lock plays no role in this suspend/resume protocol.
4057 //
4058
4059 static void resume_clear_context(OSThread *osthread) {
4060 osthread->set_ucontext(NULL);
4061 osthread->set_siginfo(NULL);
4062 }
4063
4064 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4065 osthread->set_ucontext(context);
4066 osthread->set_siginfo(siginfo);
4067 }
4068
4069 //
4070 // Handler function invoked when a thread's execution is suspended or
4071 // resumed. We have to be careful that only async-safe functions are
4072 // called here (Note: most pthread functions are not async safe and
4073 // should be avoided.)
4074 //
4075 // Note: sigwait() is a more natural fit than sigsuspend() from an
4076 // interface point of view, but sigwait() prevents the signal hander
4077 // from being run. libpthread would get very confused by not having
4078 // its signal handlers run and prevents sigwait()'s use with the
4079 // mutex granting granting signal.
4080 //
4081 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4082 //
4083 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4084 // Save and restore errno to avoid confusing native code with EINTR
4085 // after sigsuspend.
4086 int old_errno = errno;
4087
4088 Thread* thread = Thread::current();
4089 OSThread* osthread = thread->osthread();
4090 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4091
4092 os::SuspendResume::State current = osthread->sr.state();
4093 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4094 suspend_save_context(osthread, siginfo, context);
4095
4096 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4097 os::SuspendResume::State state = osthread->sr.suspended();
4098 if (state == os::SuspendResume::SR_SUSPENDED) {
4099 sigset_t suspend_set; // signals for sigsuspend()
4100
4101 // get current set of blocked signals and unblock resume signal
4102 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4103 sigdelset(&suspend_set, SR_signum);
4104
4105 sr_semaphore.signal();
4106 // wait here until we are resumed
4107 while (1) {
4108 sigsuspend(&suspend_set);
4109
4110 os::SuspendResume::State result = osthread->sr.running();
4111 if (result == os::SuspendResume::SR_RUNNING) {
4112 sr_semaphore.signal();
4113 break;
4114 }
4115 }
4116
4117 } else if (state == os::SuspendResume::SR_RUNNING) {
4118 // request was cancelled, continue
4119 } else {
4120 ShouldNotReachHere();
4121 }
4122
4123 resume_clear_context(osthread);
4124 } else if (current == os::SuspendResume::SR_RUNNING) {
4125 // request was cancelled, continue
4126 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4127 // ignore
4128 } else {
4129 // ignore
4130 }
4131
4132 errno = old_errno;
4133 }
4134
4135
4136 static int SR_initialize() {
4137 struct sigaction act;
4138 char *s;
4139 /* Get signal number to use for suspend/resume */
4140 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4141 int sig = ::strtol(s, 0, 10);
4142 if (sig > 0 || sig < _NSIG) {
4143 SR_signum = sig;
4144 }
4145 }
4146
4147 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4148 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4149
4150 sigemptyset(&SR_sigset);
4151 sigaddset(&SR_sigset, SR_signum);
4152
4153 /* Set up signal handler for suspend/resume */
4154 act.sa_flags = SA_RESTART|SA_SIGINFO;
4155 act.sa_handler = (void (*)(int)) SR_handler;
4156
4157 // SR_signum is blocked by default.
4158 // 4528190 - We also need to block pthread restart signal (32 on all
4159 // supported Linux platforms). Note that LinuxThreads need to block
4160 // this signal for all threads to work properly. So we don't have
4161 // to use hard-coded signal number when setting up the mask.
4162 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4163
4164 if (sigaction(SR_signum, &act, 0) == -1) {
4165 return -1;
4166 }
4167
4168 // Save signal flag
4169 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4170 return 0;
4171 }
4172
4173 static int sr_notify(OSThread* osthread) {
4174 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4175 assert_status(status == 0, status, "pthread_kill");
4176 return status;
4177 }
4178
4179 // "Randomly" selected value for how long we want to spin
4180 // before bailing out on suspending a thread, also how often
4181 // we send a signal to a thread we want to resume
4182 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4183 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4184
4185 // returns true on success and false on error - really an error is fatal
4186 // but this seems the normal response to library errors
4187 static bool do_suspend(OSThread* osthread) {
4188 assert(osthread->sr.is_running(), "thread should be running");
4189 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4190
4191 // mark as suspended and send signal
4192 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4193 // failed to switch, state wasn't running?
4194 ShouldNotReachHere();
4195 return false;
4196 }
4197
4198 if (sr_notify(osthread) != 0) {
4199 ShouldNotReachHere();
4200 }
4201
4202 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4203 while (true) {
4204 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4205 break;
4206 } else {
4207 // timeout
4208 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4209 if (cancelled == os::SuspendResume::SR_RUNNING) {
4210 return false;
4211 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4212 // make sure that we consume the signal on the semaphore as well
4213 sr_semaphore.wait();
4214 break;
4215 } else {
4216 ShouldNotReachHere();
4217 return false;
4218 }
4219 }
4220 }
4221
4222 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4223 return true;
4224 }
4225
4226 static void do_resume(OSThread* osthread) {
4227 assert(osthread->sr.is_suspended(), "thread should be suspended");
4228 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4229
4230 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4231 // failed to switch to WAKEUP_REQUEST
4232 ShouldNotReachHere();
4233 return;
4234 }
4235
4236 while (true) {
4237 if (sr_notify(osthread) == 0) {
4238 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4239 if (osthread->sr.is_running()) {
4240 return;
4241 }
4242 }
4243 } else {
4244 ShouldNotReachHere();
4245 }
4246 }
4247
4248 guarantee(osthread->sr.is_running(), "Must be running!");
4249 }
4250
4251 ////////////////////////////////////////////////////////////////////////////////
4252 // interrupt support
4253
4254 void os::interrupt(Thread* thread) {
4255 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4256 "possibility of dangling Thread pointer");
4257
4258 OSThread* osthread = thread->osthread();
4259
4260 if (!osthread->interrupted()) {
4261 osthread->set_interrupted(true);
4262 // More than one thread can get here with the same value of osthread,
4263 // resulting in multiple notifications. We do, however, want the store
4264 // to interrupted() to be visible to other threads before we execute unpark().
4265 OrderAccess::fence();
4266 ParkEvent * const slp = thread->_SleepEvent ;
4267 if (slp != NULL) slp->unpark() ;
4268 }
4269
4270 // For JSR166. Unpark even if interrupt status already was set
4271 if (thread->is_Java_thread())
4272 ((JavaThread*)thread)->parker()->unpark();
4273
4274 ParkEvent * ev = thread->_ParkEvent ;
4275 if (ev != NULL) ev->unpark() ;
4276
4277 }
4278
4279 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4280 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4281 "possibility of dangling Thread pointer");
4282
4283 OSThread* osthread = thread->osthread();
4284
4285 bool interrupted = osthread->interrupted();
4286
4287 if (interrupted && clear_interrupted) {
4288 osthread->set_interrupted(false);
4289 // consider thread->_SleepEvent->reset() ... optional optimization
4290 }
4291
4292 return interrupted;
4293 }
4294
4295 ///////////////////////////////////////////////////////////////////////////////////
4296 // signal handling (except suspend/resume)
4297
4298 // This routine may be used by user applications as a "hook" to catch signals.
4299 // The user-defined signal handler must pass unrecognized signals to this
4300 // routine, and if it returns true (non-zero), then the signal handler must
4301 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4302 // routine will never retun false (zero), but instead will execute a VM panic
4303 // routine kill the process.
4304 //
4305 // If this routine returns false, it is OK to call it again. This allows
4306 // the user-defined signal handler to perform checks either before or after
4307 // the VM performs its own checks. Naturally, the user code would be making
4308 // a serious error if it tried to handle an exception (such as a null check
4309 // or breakpoint) that the VM was generating for its own correct operation.
4310 //
4311 // This routine may recognize any of the following kinds of signals:
4312 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4313 // It should be consulted by handlers for any of those signals.
4314 //
4315 // The caller of this routine must pass in the three arguments supplied
4316 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4317 // field of the structure passed to sigaction(). This routine assumes that
4318 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4319 //
4320 // Note that the VM will print warnings if it detects conflicting signal
4321 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4322 //
4323 extern "C" JNIEXPORT int
4324 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4325 void* ucontext, int abort_if_unrecognized);
4326
4327 void signalHandler(int sig, siginfo_t* info, void* uc) {
4328 assert(info != NULL && uc != NULL, "it must be old kernel");
4329 int orig_errno = errno; // Preserve errno value over signal handler.
4330 JVM_handle_linux_signal(sig, info, uc, true);
4331 errno = orig_errno;
4332 }
4333
4334
4335 // This boolean allows users to forward their own non-matching signals
4336 // to JVM_handle_linux_signal, harmlessly.
4337 bool os::Linux::signal_handlers_are_installed = false;
4338
4339 // For signal-chaining
4340 struct sigaction os::Linux::sigact[MAXSIGNUM];
4341 unsigned int os::Linux::sigs = 0;
4342 bool os::Linux::libjsig_is_loaded = false;
4343 typedef struct sigaction *(*get_signal_t)(int);
4344 get_signal_t os::Linux::get_signal_action = NULL;
4345
4346 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4347 struct sigaction *actp = NULL;
4348
4349 if (libjsig_is_loaded) {
4350 // Retrieve the old signal handler from libjsig
4351 actp = (*get_signal_action)(sig);
4352 }
4353 if (actp == NULL) {
4354 // Retrieve the preinstalled signal handler from jvm
4355 actp = get_preinstalled_handler(sig);
4356 }
4357
4358 return actp;
4359 }
4360
4361 static bool call_chained_handler(struct sigaction *actp, int sig,
4362 siginfo_t *siginfo, void *context) {
4363 // Call the old signal handler
4364 if (actp->sa_handler == SIG_DFL) {
4365 // It's more reasonable to let jvm treat it as an unexpected exception
4366 // instead of taking the default action.
4367 return false;
4368 } else if (actp->sa_handler != SIG_IGN) {
4369 if ((actp->sa_flags & SA_NODEFER) == 0) {
4370 // automaticlly block the signal
4371 sigaddset(&(actp->sa_mask), sig);
4372 }
4373
4374 sa_handler_t hand = NULL;
4375 sa_sigaction_t sa = NULL;
4376 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4377 // retrieve the chained handler
4378 if (siginfo_flag_set) {
4379 sa = actp->sa_sigaction;
4380 } else {
4381 hand = actp->sa_handler;
4382 }
4383
4384 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4385 actp->sa_handler = SIG_DFL;
4386 }
4387
4388 // try to honor the signal mask
4389 sigset_t oset;
4390 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4391
4392 // call into the chained handler
4393 if (siginfo_flag_set) {
4394 (*sa)(sig, siginfo, context);
4395 } else {
4396 (*hand)(sig);
4397 }
4398
4399 // restore the signal mask
4400 pthread_sigmask(SIG_SETMASK, &oset, 0);
4401 }
4402 // Tell jvm's signal handler the signal is taken care of.
4403 return true;
4404 }
4405
4406 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4407 bool chained = false;
4408 // signal-chaining
4409 if (UseSignalChaining) {
4410 struct sigaction *actp = get_chained_signal_action(sig);
4411 if (actp != NULL) {
4412 chained = call_chained_handler(actp, sig, siginfo, context);
4413 }
4414 }
4415 return chained;
4416 }
4417
4418 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4419 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4420 return &sigact[sig];
4421 }
4422 return NULL;
4423 }
4424
4425 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4426 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4427 sigact[sig] = oldAct;
4428 sigs |= (unsigned int)1 << sig;
4429 }
4430
4431 // for diagnostic
4432 int os::Linux::sigflags[MAXSIGNUM];
4433
4434 int os::Linux::get_our_sigflags(int sig) {
4435 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4436 return sigflags[sig];
4437 }
4438
4439 void os::Linux::set_our_sigflags(int sig, int flags) {
4440 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4441 sigflags[sig] = flags;
4442 }
4443
4444 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4445 // Check for overwrite.
4446 struct sigaction oldAct;
4447 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4448
4449 void* oldhand = oldAct.sa_sigaction
4450 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4451 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4452 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4453 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4454 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4455 if (AllowUserSignalHandlers || !set_installed) {
4456 // Do not overwrite; user takes responsibility to forward to us.
4457 return;
4458 } else if (UseSignalChaining) {
4459 // save the old handler in jvm
4460 save_preinstalled_handler(sig, oldAct);
4461 // libjsig also interposes the sigaction() call below and saves the
4462 // old sigaction on it own.
4463 } else {
4464 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4465 "%#lx for signal %d.", (long)oldhand, sig));
4466 }
4467 }
4468
4469 struct sigaction sigAct;
4470 sigfillset(&(sigAct.sa_mask));
4471 sigAct.sa_handler = SIG_DFL;
4472 if (!set_installed) {
4473 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4474 } else {
4475 sigAct.sa_sigaction = signalHandler;
4476 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4477 }
4478 // Save flags, which are set by ours
4479 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4480 sigflags[sig] = sigAct.sa_flags;
4481
4482 int ret = sigaction(sig, &sigAct, &oldAct);
4483 assert(ret == 0, "check");
4484
4485 void* oldhand2 = oldAct.sa_sigaction
4486 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4487 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4488 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4489 }
4490
4491 // install signal handlers for signals that HotSpot needs to
4492 // handle in order to support Java-level exception handling.
4493
4494 void os::Linux::install_signal_handlers() {
4495 if (!signal_handlers_are_installed) {
4496 signal_handlers_are_installed = true;
4497
4498 // signal-chaining
4499 typedef void (*signal_setting_t)();
4500 signal_setting_t begin_signal_setting = NULL;
4501 signal_setting_t end_signal_setting = NULL;
4502 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4503 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4504 if (begin_signal_setting != NULL) {
4505 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4506 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4507 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4508 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4509 libjsig_is_loaded = true;
4510 assert(UseSignalChaining, "should enable signal-chaining");
4511 }
4512 if (libjsig_is_loaded) {
4513 // Tell libjsig jvm is setting signal handlers
4514 (*begin_signal_setting)();
4515 }
4516
4517 set_signal_handler(SIGSEGV, true);
4518 set_signal_handler(SIGPIPE, true);
4519 set_signal_handler(SIGBUS, true);
4520 set_signal_handler(SIGILL, true);
4521 set_signal_handler(SIGFPE, true);
4522 #if defined(PPC64)
4523 set_signal_handler(SIGTRAP, true);
4524 #endif
4525 set_signal_handler(SIGXFSZ, true);
4526
4527 if (libjsig_is_loaded) {
4528 // Tell libjsig jvm finishes setting signal handlers
4529 (*end_signal_setting)();
4530 }
4531
4532 // We don't activate signal checker if libjsig is in place, we trust ourselves
4533 // and if UserSignalHandler is installed all bets are off.
4534 // Log that signal checking is off only if -verbose:jni is specified.
4535 if (CheckJNICalls) {
4536 if (libjsig_is_loaded) {
4537 if (PrintJNIResolving) {
4538 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4539 }
4540 check_signals = false;
4541 }
4542 if (AllowUserSignalHandlers) {
4543 if (PrintJNIResolving) {
4544 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4545 }
4546 check_signals = false;
4547 }
4548 }
4549 }
4550 }
4551
4552 // This is the fastest way to get thread cpu time on Linux.
4553 // Returns cpu time (user+sys) for any thread, not only for current.
4554 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4555 // It might work on 2.6.10+ with a special kernel/glibc patch.
4556 // For reference, please, see IEEE Std 1003.1-2004:
4557 // http://www.unix.org/single_unix_specification
4558
4559 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4560 struct timespec tp;
4561 int rc = os::Linux::clock_gettime(clockid, &tp);
4562 assert(rc == 0, "clock_gettime is expected to return 0 code");
4563
4564 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4565 }
4566
4567 /////
4568 // glibc on Linux platform uses non-documented flag
4569 // to indicate, that some special sort of signal
4570 // trampoline is used.
4571 // We will never set this flag, and we should
4572 // ignore this flag in our diagnostic
4573 #ifdef SIGNIFICANT_SIGNAL_MASK
4574 #undef SIGNIFICANT_SIGNAL_MASK
4575 #endif
4576 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4577
4578 static const char* get_signal_handler_name(address handler,
4579 char* buf, int buflen) {
4580 int offset = 0;
4581 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4582 if (found) {
4583 // skip directory names
4584 const char *p1, *p2;
4585 p1 = buf;
4586 size_t len = strlen(os::file_separator());
4587 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4588 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4589 } else {
4590 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4591 }
4592 return buf;
4593 }
4594
4595 static void print_signal_handler(outputStream* st, int sig,
4596 char* buf, size_t buflen) {
4597 struct sigaction sa;
4598
4599 sigaction(sig, NULL, &sa);
4600
4601 // See comment for SIGNIFICANT_SIGNAL_MASK define
4602 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4603
4604 st->print("%s: ", os::exception_name(sig, buf, buflen));
4605
4606 address handler = (sa.sa_flags & SA_SIGINFO)
4607 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4608 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4609
4610 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4611 st->print("SIG_DFL");
4612 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4613 st->print("SIG_IGN");
4614 } else {
4615 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4616 }
4617
4618 st->print(", sa_mask[0]=");
4619 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4620
4621 address rh = VMError::get_resetted_sighandler(sig);
4622 // May be, handler was resetted by VMError?
4623 if(rh != NULL) {
4624 handler = rh;
4625 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4626 }
4627
4628 st->print(", sa_flags=");
4629 os::Posix::print_sa_flags(st, sa.sa_flags);
4630
4631 // Check: is it our handler?
4632 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4633 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4634 // It is our signal handler
4635 // check for flags, reset system-used one!
4636 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4637 st->print(
4638 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4639 os::Linux::get_our_sigflags(sig));
4640 }
4641 }
4642 st->cr();
4643 }
4644
4645
4646 #define DO_SIGNAL_CHECK(sig) \
4647 if (!sigismember(&check_signal_done, sig)) \
4648 os::Linux::check_signal_handler(sig)
4649
4650 // This method is a periodic task to check for misbehaving JNI applications
4651 // under CheckJNI, we can add any periodic checks here
4652
4653 void os::run_periodic_checks() {
4654
4655 if (check_signals == false) return;
4656
4657 // SEGV and BUS if overridden could potentially prevent
4658 // generation of hs*.log in the event of a crash, debugging
4659 // such a case can be very challenging, so we absolutely
4660 // check the following for a good measure:
4661 DO_SIGNAL_CHECK(SIGSEGV);
4662 DO_SIGNAL_CHECK(SIGILL);
4663 DO_SIGNAL_CHECK(SIGFPE);
4664 DO_SIGNAL_CHECK(SIGBUS);
4665 DO_SIGNAL_CHECK(SIGPIPE);
4666 DO_SIGNAL_CHECK(SIGXFSZ);
4667 #if defined(PPC64)
4668 DO_SIGNAL_CHECK(SIGTRAP);
4669 #endif
4670
4671 // ReduceSignalUsage allows the user to override these handlers
4672 // see comments at the very top and jvm_solaris.h
4673 if (!ReduceSignalUsage) {
4674 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4675 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4676 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4677 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4678 }
4679
4680 DO_SIGNAL_CHECK(SR_signum);
4681 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4682 }
4683
4684 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4685
4686 static os_sigaction_t os_sigaction = NULL;
4687
4688 void os::Linux::check_signal_handler(int sig) {
4689 char buf[O_BUFLEN];
4690 address jvmHandler = NULL;
4691
4692
4693 struct sigaction act;
4694 if (os_sigaction == NULL) {
4695 // only trust the default sigaction, in case it has been interposed
4696 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4697 if (os_sigaction == NULL) return;
4698 }
4699
4700 os_sigaction(sig, (struct sigaction*)NULL, &act);
4701
4702
4703 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4704
4705 address thisHandler = (act.sa_flags & SA_SIGINFO)
4706 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4707 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4708
4709
4710 switch(sig) {
4711 case SIGSEGV:
4712 case SIGBUS:
4713 case SIGFPE:
4714 case SIGPIPE:
4715 case SIGILL:
4716 case SIGXFSZ:
4717 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4718 break;
4719
4720 case SHUTDOWN1_SIGNAL:
4721 case SHUTDOWN2_SIGNAL:
4722 case SHUTDOWN3_SIGNAL:
4723 case BREAK_SIGNAL:
4724 jvmHandler = (address)user_handler();
4725 break;
4726
4727 case INTERRUPT_SIGNAL:
4728 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4729 break;
4730
4731 default:
4732 if (sig == SR_signum) {
4733 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4734 } else {
4735 return;
4736 }
4737 break;
4738 }
4739
4740 if (thisHandler != jvmHandler) {
4741 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4742 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4743 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4744 // No need to check this sig any longer
4745 sigaddset(&check_signal_done, sig);
4746 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4747 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4748 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4749 exception_name(sig, buf, O_BUFLEN));
4750 }
4751 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4752 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4753 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4754 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4755 // No need to check this sig any longer
4756 sigaddset(&check_signal_done, sig);
4757 }
4758
4759 // Dump all the signal
4760 if (sigismember(&check_signal_done, sig)) {
4761 print_signal_handlers(tty, buf, O_BUFLEN);
4762 }
4763 }
4764
4765 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4766
4767 extern bool signal_name(int signo, char* buf, size_t len);
4768
4769 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4770 if (0 < exception_code && exception_code <= SIGRTMAX) {
4771 // signal
4772 if (!signal_name(exception_code, buf, size)) {
4773 jio_snprintf(buf, size, "SIG%d", exception_code);
4774 }
4775 return buf;
4776 } else {
4777 return NULL;
4778 }
4779 }
4780
4781 // this is called _before_ the most of global arguments have been parsed
4782 void os::init(void) {
4783 char dummy; /* used to get a guess on initial stack address */
4784 // first_hrtime = gethrtime();
4785
4786 // With LinuxThreads the JavaMain thread pid (primordial thread)
4787 // is different than the pid of the java launcher thread.
4788 // So, on Linux, the launcher thread pid is passed to the VM
4789 // via the sun.java.launcher.pid property.
4790 // Use this property instead of getpid() if it was correctly passed.
4791 // See bug 6351349.
4792 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4793
4794 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4795
4796 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4797
4798 init_random(1234567);
4799
4800 ThreadCritical::initialize();
4801
4802 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4803 if (Linux::page_size() == -1) {
4804 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4805 strerror(errno)));
4806 }
4807 init_page_sizes((size_t) Linux::page_size());
4808
4809 Linux::initialize_system_info();
4810
4811 // main_thread points to the aboriginal thread
4812 Linux::_main_thread = pthread_self();
4813
4814 Linux::clock_init();
4815 initial_time_count = javaTimeNanos();
4816
4817 // pthread_condattr initialization for monotonic clock
4818 int status;
4819 pthread_condattr_t* _condattr = os::Linux::condAttr();
4820 if ((status = pthread_condattr_init(_condattr)) != 0) {
4821 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4822 }
4823 // Only set the clock if CLOCK_MONOTONIC is available
4824 if (Linux::supports_monotonic_clock()) {
4825 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4826 if (status == EINVAL) {
4827 warning("Unable to use monotonic clock with relative timed-waits" \
4828 " - changes to the time-of-day clock may have adverse affects");
4829 } else {
4830 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4831 }
4832 }
4833 }
4834 // else it defaults to CLOCK_REALTIME
4835
4836 pthread_mutex_init(&dl_mutex, NULL);
4837
4838 // If the pagesize of the VM is greater than 8K determine the appropriate
4839 // number of initial guard pages. The user can change this with the
4840 // command line arguments, if needed.
4841 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4842 StackYellowPages = 1;
4843 StackRedPages = 1;
4844 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4845 }
4846 }
4847
4848 // To install functions for atexit system call
4849 extern "C" {
4850 static void perfMemory_exit_helper() {
4851 perfMemory_exit();
4852 }
4853 }
4854
4855 // this is called _after_ the global arguments have been parsed
4856 jint os::init_2(void)
4857 {
4858 Linux::fast_thread_clock_init();
4859
4860 // Allocate a single page and mark it as readable for safepoint polling
4861 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4862 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4863
4864 os::set_polling_page( polling_page );
4865
4866 #ifndef PRODUCT
4867 if(Verbose && PrintMiscellaneous)
4868 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4869 #endif
4870
4871 if (!UseMembar) {
4872 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4873 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4874 os::set_memory_serialize_page( mem_serialize_page );
4875
4876 #ifndef PRODUCT
4877 if(Verbose && PrintMiscellaneous)
4878 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4879 #endif
4880 }
4881
4882 // initialize suspend/resume support - must do this before signal_sets_init()
4883 if (SR_initialize() != 0) {
4884 perror("SR_initialize failed");
4885 return JNI_ERR;
4886 }
4887
4888 Linux::signal_sets_init();
4889 Linux::install_signal_handlers();
4890
4891 // Check minimum allowable stack size for thread creation and to initialize
4892 // the java system classes, including StackOverflowError - depends on page
4893 // size. Add a page for compiler2 recursion in main thread.
4894 // Add in 2*BytesPerWord times page size to account for VM stack during
4895 // class initialization depending on 32 or 64 bit VM.
4896 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4897 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4898 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4899
4900 size_t threadStackSizeInBytes = ThreadStackSize * K;
4901 if (threadStackSizeInBytes != 0 &&
4902 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4903 tty->print_cr("\nThe stack size specified is too small, "
4904 "Specify at least %dk",
4905 os::Linux::min_stack_allowed/ K);
4906 return JNI_ERR;
4907 }
4908
4909 // Make the stack size a multiple of the page size so that
4910 // the yellow/red zones can be guarded.
4911 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4912 vm_page_size()));
4913
4914 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4915
4916 #if defined(IA32)
4917 workaround_expand_exec_shield_cs_limit();
4918 #endif
4919
4920 Linux::libpthread_init();
4921 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4922 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4923 Linux::glibc_version(), Linux::libpthread_version(),
4924 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4925 }
4926
4927 if (UseNUMA) {
4928 if (!Linux::libnuma_init()) {
4929 UseNUMA = false;
4930 } else {
4931 if ((Linux::numa_max_node() < 1)) {
4932 // There's only one node(they start from 0), disable NUMA.
4933 UseNUMA = false;
4934 }
4935 }
4936 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4937 // we can make the adaptive lgrp chunk resizing work. If the user specified
4938 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4939 // disable adaptive resizing.
4940 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4941 if (FLAG_IS_DEFAULT(UseNUMA)) {
4942 UseNUMA = false;
4943 } else {
4944 if (FLAG_IS_DEFAULT(UseLargePages) &&
4945 FLAG_IS_DEFAULT(UseSHM) &&
4946 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4947 UseLargePages = false;
4948 } else {
4949 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4950 UseAdaptiveSizePolicy = false;
4951 UseAdaptiveNUMAChunkSizing = false;
4952 }
4953 }
4954 }
4955 if (!UseNUMA && ForceNUMA) {
4956 UseNUMA = true;
4957 }
4958 }
4959
4960 if (MaxFDLimit) {
4961 // set the number of file descriptors to max. print out error
4962 // if getrlimit/setrlimit fails but continue regardless.
4963 struct rlimit nbr_files;
4964 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4965 if (status != 0) {
4966 if (PrintMiscellaneous && (Verbose || WizardMode))
4967 perror("os::init_2 getrlimit failed");
4968 } else {
4969 nbr_files.rlim_cur = nbr_files.rlim_max;
4970 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4971 if (status != 0) {
4972 if (PrintMiscellaneous && (Verbose || WizardMode))
4973 perror("os::init_2 setrlimit failed");
4974 }
4975 }
4976 }
4977
4978 // Initialize lock used to serialize thread creation (see os::create_thread)
4979 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4980
4981 // at-exit methods are called in the reverse order of their registration.
4982 // atexit functions are called on return from main or as a result of a
4983 // call to exit(3C). There can be only 32 of these functions registered
4984 // and atexit() does not set errno.
4985
4986 if (PerfAllowAtExitRegistration) {
4987 // only register atexit functions if PerfAllowAtExitRegistration is set.
4988 // atexit functions can be delayed until process exit time, which
4989 // can be problematic for embedded VM situations. Embedded VMs should
4990 // call DestroyJavaVM() to assure that VM resources are released.
4991
4992 // note: perfMemory_exit_helper atexit function may be removed in
4993 // the future if the appropriate cleanup code can be added to the
4994 // VM_Exit VMOperation's doit method.
4995 if (atexit(perfMemory_exit_helper) != 0) {
4996 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4997 }
4998 }
4999
5000 // initialize thread priority policy
5001 prio_init();
5002
5003 return JNI_OK;
5004 }
5005
5006 // Mark the polling page as unreadable
5007 void os::make_polling_page_unreadable(void) {
5008 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5009 fatal("Could not disable polling page");
5010 };
5011
5012 // Mark the polling page as readable
5013 void os::make_polling_page_readable(void) {
5014 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5015 fatal("Could not enable polling page");
5016 }
5017 };
5018
5019 int os::active_processor_count() {
5020 // Linux doesn't yet have a (official) notion of processor sets,
5021 // so just return the number of online processors.
5022 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5023 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5024 return online_cpus;
5025 }
5026
5027 void os::set_native_thread_name(const char *name) {
5028 // Not yet implemented.
5029 return;
5030 }
5031
5032 bool os::distribute_processes(uint length, uint* distribution) {
5033 // Not yet implemented.
5034 return false;
5035 }
5036
5037 bool os::bind_to_processor(uint processor_id) {
5038 // Not yet implemented.
5039 return false;
5040 }
5041
5042 ///
5043
5044 void os::SuspendedThreadTask::internal_do_task() {
5045 if (do_suspend(_thread->osthread())) {
5046 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5047 do_task(context);
5048 do_resume(_thread->osthread());
5049 }
5050 }
5051
5052 class PcFetcher : public os::SuspendedThreadTask {
5053 public:
5054 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5055 ExtendedPC result();
5056 protected:
5057 void do_task(const os::SuspendedThreadTaskContext& context);
5058 private:
5059 ExtendedPC _epc;
5060 };
5061
5062 ExtendedPC PcFetcher::result() {
5063 guarantee(is_done(), "task is not done yet.");
5064 return _epc;
5065 }
5066
5067 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5068 Thread* thread = context.thread();
5069 OSThread* osthread = thread->osthread();
5070 if (osthread->ucontext() != NULL) {
5071 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5072 } else {
5073 // NULL context is unexpected, double-check this is the VMThread
5074 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5075 }
5076 }
5077
5078 // Suspends the target using the signal mechanism and then grabs the PC before
5079 // resuming the target. Used by the flat-profiler only
5080 ExtendedPC os::get_thread_pc(Thread* thread) {
5081 // Make sure that it is called by the watcher for the VMThread
5082 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5083 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5084
5085 PcFetcher fetcher(thread);
5086 fetcher.run();
5087 return fetcher.result();
5088 }
5089
5090 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5091 {
5092 if (is_NPTL()) {
5093 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5094 } else {
5095 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5096 // word back to default 64bit precision if condvar is signaled. Java
5097 // wants 53bit precision. Save and restore current value.
5098 int fpu = get_fpu_control_word();
5099 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5100 set_fpu_control_word(fpu);
5101 return status;
5102 }
5103 }
5104
5105 ////////////////////////////////////////////////////////////////////////////////
5106 // debug support
5107
5108 bool os::find(address addr, outputStream* st) {
5109 Dl_info dlinfo;
5110 memset(&dlinfo, 0, sizeof(dlinfo));
5111 if (dladdr(addr, &dlinfo) != 0) {
5112 st->print(PTR_FORMAT ": ", addr);
5113 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5114 st->print("%s+%#x", dlinfo.dli_sname,
5115 addr - (intptr_t)dlinfo.dli_saddr);
5116 } else if (dlinfo.dli_fbase != NULL) {
5117 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5118 } else {
5119 st->print("<absolute address>");
5120 }
5121 if (dlinfo.dli_fname != NULL) {
5122 st->print(" in %s", dlinfo.dli_fname);
5123 }
5124 if (dlinfo.dli_fbase != NULL) {
5125 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5126 }
5127 st->cr();
5128
5129 if (Verbose) {
5130 // decode some bytes around the PC
5131 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5132 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5133 address lowest = (address) dlinfo.dli_sname;
5134 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5135 if (begin < lowest) begin = lowest;
5136 Dl_info dlinfo2;
5137 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5138 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5139 end = (address) dlinfo2.dli_saddr;
5140 Disassembler::decode(begin, end, st);
5141 }
5142 return true;
5143 }
5144 return false;
5145 }
5146
5147 ////////////////////////////////////////////////////////////////////////////////
5148 // misc
5149
5150 // This does not do anything on Linux. This is basically a hook for being
5151 // able to use structured exception handling (thread-local exception filters)
5152 // on, e.g., Win32.
5153 void
5154 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5155 JavaCallArguments* args, Thread* thread) {
5156 f(value, method, args, thread);
5157 }
5158
5159 void os::print_statistics() {
5160 }
5161
5162 int os::message_box(const char* title, const char* message) {
5163 int i;
5164 fdStream err(defaultStream::error_fd());
5165 for (i = 0; i < 78; i++) err.print_raw("=");
5166 err.cr();
5167 err.print_raw_cr(title);
5168 for (i = 0; i < 78; i++) err.print_raw("-");
5169 err.cr();
5170 err.print_raw_cr(message);
5171 for (i = 0; i < 78; i++) err.print_raw("=");
5172 err.cr();
5173
5174 char buf[16];
5175 // Prevent process from exiting upon "read error" without consuming all CPU
5176 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5177
5178 return buf[0] == 'y' || buf[0] == 'Y';
5179 }
5180
5181 int os::stat(const char *path, struct stat *sbuf) {
5182 char pathbuf[MAX_PATH];
5183 if (strlen(path) > MAX_PATH - 1) {
5184 errno = ENAMETOOLONG;
5185 return -1;
5186 }
5187 os::native_path(strcpy(pathbuf, path));
5188 return ::stat(pathbuf, sbuf);
5189 }
5190
5191 bool os::check_heap(bool force) {
5192 return true;
5193 }
5194
5195 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5196 return ::vsnprintf(buf, count, format, args);
5197 }
5198
5199 // Is a (classpath) directory empty?
5200 bool os::dir_is_empty(const char* path) {
5201 DIR *dir = NULL;
5202 struct dirent *ptr;
5203
5204 dir = opendir(path);
5205 if (dir == NULL) return true;
5206
5207 /* Scan the directory */
5208 bool result = true;
5209 char buf[sizeof(struct dirent) + MAX_PATH];
5210 while (result && (ptr = ::readdir(dir)) != NULL) {
5211 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5212 result = false;
5213 }
5214 }
5215 closedir(dir);
5216 return result;
5217 }
5218
5219 // This code originates from JDK's sysOpen and open64_w
5220 // from src/solaris/hpi/src/system_md.c
5221
5222 #ifndef O_DELETE
5223 #define O_DELETE 0x10000
5224 #endif
5225
5226 // Open a file. Unlink the file immediately after open returns
5227 // if the specified oflag has the O_DELETE flag set.
5228 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5229
5230 int os::open(const char *path, int oflag, int mode) {
5231
5232 if (strlen(path) > MAX_PATH - 1) {
5233 errno = ENAMETOOLONG;
5234 return -1;
5235 }
5236 int fd;
5237 int o_delete = (oflag & O_DELETE);
5238 oflag = oflag & ~O_DELETE;
5239
5240 fd = ::open64(path, oflag, mode);
5241 if (fd == -1) return -1;
5242
5243 //If the open succeeded, the file might still be a directory
5244 {
5245 struct stat64 buf64;
5246 int ret = ::fstat64(fd, &buf64);
5247 int st_mode = buf64.st_mode;
5248
5249 if (ret != -1) {
5250 if ((st_mode & S_IFMT) == S_IFDIR) {
5251 errno = EISDIR;
5252 ::close(fd);
5253 return -1;
5254 }
5255 } else {
5256 ::close(fd);
5257 return -1;
5258 }
5259 }
5260
5261 /*
5262 * All file descriptors that are opened in the JVM and not
5263 * specifically destined for a subprocess should have the
5264 * close-on-exec flag set. If we don't set it, then careless 3rd
5265 * party native code might fork and exec without closing all
5266 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5267 * UNIXProcess.c), and this in turn might:
5268 *
5269 * - cause end-of-file to fail to be detected on some file
5270 * descriptors, resulting in mysterious hangs, or
5271 *
5272 * - might cause an fopen in the subprocess to fail on a system
5273 * suffering from bug 1085341.
5274 *
5275 * (Yes, the default setting of the close-on-exec flag is a Unix
5276 * design flaw)
5277 *
5278 * See:
5279 * 1085341: 32-bit stdio routines should support file descriptors >255
5280 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5281 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5282 */
5283 #ifdef FD_CLOEXEC
5284 {
5285 int flags = ::fcntl(fd, F_GETFD);
5286 if (flags != -1)
5287 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5288 }
5289 #endif
5290
5291 if (o_delete != 0) {
5292 ::unlink(path);
5293 }
5294 return fd;
5295 }
5296
5297
5298 // create binary file, rewriting existing file if required
5299 int os::create_binary_file(const char* path, bool rewrite_existing) {
5300 int oflags = O_WRONLY | O_CREAT;
5301 if (!rewrite_existing) {
5302 oflags |= O_EXCL;
5303 }
5304 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5305 }
5306
5307 // return current position of file pointer
5308 jlong os::current_file_offset(int fd) {
5309 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5310 }
5311
5312 // move file pointer to the specified offset
5313 jlong os::seek_to_file_offset(int fd, jlong offset) {
5314 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5315 }
5316
5317 // This code originates from JDK's sysAvailable
5318 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5319
5320 int os::available(int fd, jlong *bytes) {
5321 jlong cur, end;
5322 int mode;
5323 struct stat64 buf64;
5324
5325 if (::fstat64(fd, &buf64) >= 0) {
5326 mode = buf64.st_mode;
5327 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5328 /*
5329 * XXX: is the following call interruptible? If so, this might
5330 * need to go through the INTERRUPT_IO() wrapper as for other
5331 * blocking, interruptible calls in this file.
5332 */
5333 int n;
5334 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5335 *bytes = n;
5336 return 1;
5337 }
5338 }
5339 }
5340 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5341 return 0;
5342 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5343 return 0;
5344 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5345 return 0;
5346 }
5347 *bytes = end - cur;
5348 return 1;
5349 }
5350
5351 int os::socket_available(int fd, jint *pbytes) {
5352 // Linux doc says EINTR not returned, unlike Solaris
5353 int ret = ::ioctl(fd, FIONREAD, pbytes);
5354
5355 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5356 // is expected to return 0 on failure and 1 on success to the jdk.
5357 return (ret < 0) ? 0 : 1;
5358 }
5359
5360 // Map a block of memory.
5361 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5362 char *addr, size_t bytes, bool read_only,
5363 bool allow_exec) {
5364 int prot;
5365 int flags = MAP_PRIVATE;
5366
5367 if (read_only) {
5368 prot = PROT_READ;
5369 } else {
5370 prot = PROT_READ | PROT_WRITE;
5371 }
5372
5373 if (allow_exec) {
5374 prot |= PROT_EXEC;
5375 }
5376
5377 if (addr != NULL) {
5378 flags |= MAP_FIXED;
5379 }
5380
5381 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5382 fd, file_offset);
5383 if (mapped_address == MAP_FAILED) {
5384 return NULL;
5385 }
5386 return mapped_address;
5387 }
5388
5389
5390 // Remap a block of memory.
5391 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5392 char *addr, size_t bytes, bool read_only,
5393 bool allow_exec) {
5394 // same as map_memory() on this OS
5395 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5396 allow_exec);
5397 }
5398
5399
5400 // Unmap a block of memory.
5401 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5402 return munmap(addr, bytes) == 0;
5403 }
5404
5405 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5406
5407 static clockid_t thread_cpu_clockid(Thread* thread) {
5408 pthread_t tid = thread->osthread()->pthread_id();
5409 clockid_t clockid;
5410
5411 // Get thread clockid
5412 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5413 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5414 return clockid;
5415 }
5416
5417 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5418 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5419 // of a thread.
5420 //
5421 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5422 // the fast estimate available on the platform.
5423
5424 jlong os::current_thread_cpu_time() {
5425 if (os::Linux::supports_fast_thread_cpu_time()) {
5426 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5427 } else {
5428 // return user + sys since the cost is the same
5429 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5430 }
5431 }
5432
5433 jlong os::thread_cpu_time(Thread* thread) {
5434 // consistent with what current_thread_cpu_time() returns
5435 if (os::Linux::supports_fast_thread_cpu_time()) {
5436 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5437 } else {
5438 return slow_thread_cpu_time(thread, true /* user + sys */);
5439 }
5440 }
5441
5442 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5443 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5444 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5445 } else {
5446 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5447 }
5448 }
5449
5450 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5451 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5452 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5453 } else {
5454 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5455 }
5456 }
5457
5458 //
5459 // -1 on error.
5460 //
5461
5462 PRAGMA_DIAG_PUSH
5463 PRAGMA_FORMAT_NONLITERAL_IGNORED
5464 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5465 static bool proc_task_unchecked = true;
5466 static const char *proc_stat_path = "/proc/%d/stat";
5467 pid_t tid = thread->osthread()->thread_id();
5468 char *s;
5469 char stat[2048];
5470 int statlen;
5471 char proc_name[64];
5472 int count;
5473 long sys_time, user_time;
5474 char cdummy;
5475 int idummy;
5476 long ldummy;
5477 FILE *fp;
5478
5479 // The /proc/<tid>/stat aggregates per-process usage on
5480 // new Linux kernels 2.6+ where NPTL is supported.
5481 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5482 // See bug 6328462.
5483 // There possibly can be cases where there is no directory
5484 // /proc/self/task, so we check its availability.
5485 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5486 // This is executed only once
5487 proc_task_unchecked = false;
5488 fp = fopen("/proc/self/task", "r");
5489 if (fp != NULL) {
5490 proc_stat_path = "/proc/self/task/%d/stat";
5491 fclose(fp);
5492 }
5493 }
5494
5495 sprintf(proc_name, proc_stat_path, tid);
5496 fp = fopen(proc_name, "r");
5497 if ( fp == NULL ) return -1;
5498 statlen = fread(stat, 1, 2047, fp);
5499 stat[statlen] = '\0';
5500 fclose(fp);
5501
5502 // Skip pid and the command string. Note that we could be dealing with
5503 // weird command names, e.g. user could decide to rename java launcher
5504 // to "java 1.4.2 :)", then the stat file would look like
5505 // 1234 (java 1.4.2 :)) R ... ...
5506 // We don't really need to know the command string, just find the last
5507 // occurrence of ")" and then start parsing from there. See bug 4726580.
5508 s = strrchr(stat, ')');
5509 if (s == NULL ) return -1;
5510
5511 // Skip blank chars
5512 do s++; while (isspace(*s));
5513
5514 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5515 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5516 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5517 &user_time, &sys_time);
5518 if ( count != 13 ) return -1;
5519 if (user_sys_cpu_time) {
5520 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5521 } else {
5522 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5523 }
5524 }
5525 PRAGMA_DIAG_POP
5526
5527 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5528 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5529 info_ptr->may_skip_backward = false; // elapsed time not wall time
5530 info_ptr->may_skip_forward = false; // elapsed time not wall time
5531 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5532 }
5533
5534 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5535 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5536 info_ptr->may_skip_backward = false; // elapsed time not wall time
5537 info_ptr->may_skip_forward = false; // elapsed time not wall time
5538 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5539 }
5540
5541 bool os::is_thread_cpu_time_supported() {
5542 return true;
5543 }
5544
5545 // System loadavg support. Returns -1 if load average cannot be obtained.
5546 // Linux doesn't yet have a (official) notion of processor sets,
5547 // so just return the system wide load average.
5548 int os::loadavg(double loadavg[], int nelem) {
5549 return ::getloadavg(loadavg, nelem);
5550 }
5551
5552 void os::pause() {
5553 char filename[MAX_PATH];
5554 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5555 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5556 } else {
5557 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5558 }
5559
5560 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5561 if (fd != -1) {
5562 struct stat buf;
5563 ::close(fd);
5564 while (::stat(filename, &buf) == 0) {
5565 (void)::poll(NULL, 0, 100);
5566 }
5567 } else {
5568 jio_fprintf(stderr,
5569 "Could not open pause file '%s', continuing immediately.\n", filename);
5570 }
5571 }
5572
5573
5574 // Refer to the comments in os_solaris.cpp park-unpark.
5575 //
5576 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5577 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5578 // For specifics regarding the bug see GLIBC BUGID 261237 :
5579 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5580 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5581 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5582 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5583 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5584 // and monitorenter when we're using 1-0 locking. All those operations may result in
5585 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5586 // of libpthread avoids the problem, but isn't practical.
5587 //
5588 // Possible remedies:
5589 //
5590 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5591 // This is palliative and probabilistic, however. If the thread is preempted
5592 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5593 // than the minimum period may have passed, and the abstime may be stale (in the
5594 // past) resultin in a hang. Using this technique reduces the odds of a hang
5595 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5596 //
5597 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5598 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5599 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5600 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5601 // thread.
5602 //
5603 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5604 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5605 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5606 // This also works well. In fact it avoids kernel-level scalability impediments
5607 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5608 // timers in a graceful fashion.
5609 //
5610 // 4. When the abstime value is in the past it appears that control returns
5611 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5612 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5613 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5614 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5615 // It may be possible to avoid reinitialization by checking the return
5616 // value from pthread_cond_timedwait(). In addition to reinitializing the
5617 // condvar we must establish the invariant that cond_signal() is only called
5618 // within critical sections protected by the adjunct mutex. This prevents
5619 // cond_signal() from "seeing" a condvar that's in the midst of being
5620 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5621 // desirable signal-after-unlock optimization that avoids futile context switching.
5622 //
5623 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5624 // structure when a condvar is used or initialized. cond_destroy() would
5625 // release the helper structure. Our reinitialize-after-timedwait fix
5626 // put excessive stress on malloc/free and locks protecting the c-heap.
5627 //
5628 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5629 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5630 // and only enabling the work-around for vulnerable environments.
5631
5632 // utility to compute the abstime argument to timedwait:
5633 // millis is the relative timeout time
5634 // abstime will be the absolute timeout time
5635 // TODO: replace compute_abstime() with unpackTime()
5636
5637 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5638 if (millis < 0) millis = 0;
5639
5640 jlong seconds = millis / 1000;
5641 millis %= 1000;
5642 if (seconds > 50000000) { // see man cond_timedwait(3T)
5643 seconds = 50000000;
5644 }
5645
5646 if (os::Linux::supports_monotonic_clock()) {
5647 struct timespec now;
5648 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5649 assert_status(status == 0, status, "clock_gettime");
5650 abstime->tv_sec = now.tv_sec + seconds;
5651 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5652 if (nanos >= NANOSECS_PER_SEC) {
5653 abstime->tv_sec += 1;
5654 nanos -= NANOSECS_PER_SEC;
5655 }
5656 abstime->tv_nsec = nanos;
5657 } else {
5658 struct timeval now;
5659 int status = gettimeofday(&now, NULL);
5660 assert(status == 0, "gettimeofday");
5661 abstime->tv_sec = now.tv_sec + seconds;
5662 long usec = now.tv_usec + millis * 1000;
5663 if (usec >= 1000000) {
5664 abstime->tv_sec += 1;
5665 usec -= 1000000;
5666 }
5667 abstime->tv_nsec = usec * 1000;
5668 }
5669 return abstime;
5670 }
5671
5672
5673 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5674 // Conceptually TryPark() should be equivalent to park(0).
5675
5676 int os::PlatformEvent::TryPark() {
5677 for (;;) {
5678 const int v = _Event ;
5679 guarantee ((v == 0) || (v == 1), "invariant") ;
5680 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5681 }
5682 }
5683
5684 void os::PlatformEvent::park() { // AKA "down()"
5685 // Invariant: Only the thread associated with the Event/PlatformEvent
5686 // may call park().
5687 // TODO: assert that _Assoc != NULL or _Assoc == Self
5688 int v ;
5689 for (;;) {
5690 v = _Event ;
5691 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5692 }
5693 guarantee (v >= 0, "invariant") ;
5694 if (v == 0) {
5695 // Do this the hard way by blocking ...
5696 int status = pthread_mutex_lock(_mutex);
5697 assert_status(status == 0, status, "mutex_lock");
5698 guarantee (_nParked == 0, "invariant") ;
5699 ++ _nParked ;
5700 while (_Event < 0) {
5701 status = pthread_cond_wait(_cond, _mutex);
5702 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5703 // Treat this the same as if the wait was interrupted
5704 if (status == ETIME) { status = EINTR; }
5705 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5706 }
5707 -- _nParked ;
5708
5709 _Event = 0 ;
5710 status = pthread_mutex_unlock(_mutex);
5711 assert_status(status == 0, status, "mutex_unlock");
5712 // Paranoia to ensure our locked and lock-free paths interact
5713 // correctly with each other.
5714 OrderAccess::fence();
5715 }
5716 guarantee (_Event >= 0, "invariant") ;
5717 }
5718
5719 int os::PlatformEvent::park(jlong millis) {
5720 guarantee (_nParked == 0, "invariant") ;
5721
5722 int v ;
5723 for (;;) {
5724 v = _Event ;
5725 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5726 }
5727 guarantee (v >= 0, "invariant") ;
5728 if (v != 0) return OS_OK ;
5729
5730 // We do this the hard way, by blocking the thread.
5731 // Consider enforcing a minimum timeout value.
5732 struct timespec abst;
5733 compute_abstime(&abst, millis);
5734
5735 int ret = OS_TIMEOUT;
5736 int status = pthread_mutex_lock(_mutex);
5737 assert_status(status == 0, status, "mutex_lock");
5738 guarantee (_nParked == 0, "invariant") ;
5739 ++_nParked ;
5740
5741 // Object.wait(timo) will return because of
5742 // (a) notification
5743 // (b) timeout
5744 // (c) thread.interrupt
5745 //
5746 // Thread.interrupt and object.notify{All} both call Event::set.
5747 // That is, we treat thread.interrupt as a special case of notification.
5748 // The underlying Solaris implementation, cond_timedwait, admits
5749 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5750 // JVM from making those visible to Java code. As such, we must
5751 // filter out spurious wakeups. We assume all ETIME returns are valid.
5752 //
5753 // TODO: properly differentiate simultaneous notify+interrupt.
5754 // In that case, we should propagate the notify to another waiter.
5755
5756 while (_Event < 0) {
5757 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5758 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5759 pthread_cond_destroy (_cond);
5760 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5761 }
5762 assert_status(status == 0 || status == EINTR ||
5763 status == ETIME || status == ETIMEDOUT,
5764 status, "cond_timedwait");
5765 if (!FilterSpuriousWakeups) break ; // previous semantics
5766 if (status == ETIME || status == ETIMEDOUT) break ;
5767 // We consume and ignore EINTR and spurious wakeups.
5768 }
5769 --_nParked ;
5770 if (_Event >= 0) {
5771 ret = OS_OK;
5772 }
5773 _Event = 0 ;
5774 status = pthread_mutex_unlock(_mutex);
5775 assert_status(status == 0, status, "mutex_unlock");
5776 assert (_nParked == 0, "invariant") ;
5777 // Paranoia to ensure our locked and lock-free paths interact
5778 // correctly with each other.
5779 OrderAccess::fence();
5780 return ret;
5781 }
5782
5783 void os::PlatformEvent::unpark() {
5784 // Transitions for _Event:
5785 // 0 :=> 1
5786 // 1 :=> 1
5787 // -1 :=> either 0 or 1; must signal target thread
5788 // That is, we can safely transition _Event from -1 to either
5789 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5790 // unpark() calls.
5791 // See also: "Semaphores in Plan 9" by Mullender & Cox
5792 //
5793 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5794 // that it will take two back-to-back park() calls for the owning
5795 // thread to block. This has the benefit of forcing a spurious return
5796 // from the first park() call after an unpark() call which will help
5797 // shake out uses of park() and unpark() without condition variables.
5798
5799 if (Atomic::xchg(1, &_Event) >= 0) return;
5800
5801 // Wait for the thread associated with the event to vacate
5802 int status = pthread_mutex_lock(_mutex);
5803 assert_status(status == 0, status, "mutex_lock");
5804 int AnyWaiters = _nParked;
5805 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5806 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5807 AnyWaiters = 0;
5808 pthread_cond_signal(_cond);
5809 }
5810 status = pthread_mutex_unlock(_mutex);
5811 assert_status(status == 0, status, "mutex_unlock");
5812 if (AnyWaiters != 0) {
5813 status = pthread_cond_signal(_cond);
5814 assert_status(status == 0, status, "cond_signal");
5815 }
5816
5817 // Note that we signal() _after dropping the lock for "immortal" Events.
5818 // This is safe and avoids a common class of futile wakeups. In rare
5819 // circumstances this can cause a thread to return prematurely from
5820 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5821 // simply re-test the condition and re-park itself.
5822 }
5823
5824
5825 // JSR166
5826 // -------------------------------------------------------
5827
5828 /*
5829 * The solaris and linux implementations of park/unpark are fairly
5830 * conservative for now, but can be improved. They currently use a
5831 * mutex/condvar pair, plus a a count.
5832 * Park decrements count if > 0, else does a condvar wait. Unpark
5833 * sets count to 1 and signals condvar. Only one thread ever waits
5834 * on the condvar. Contention seen when trying to park implies that someone
5835 * is unparking you, so don't wait. And spurious returns are fine, so there
5836 * is no need to track notifications.
5837 */
5838
5839 /*
5840 * This code is common to linux and solaris and will be moved to a
5841 * common place in dolphin.
5842 *
5843 * The passed in time value is either a relative time in nanoseconds
5844 * or an absolute time in milliseconds. Either way it has to be unpacked
5845 * into suitable seconds and nanoseconds components and stored in the
5846 * given timespec structure.
5847 * Given time is a 64-bit value and the time_t used in the timespec is only
5848 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5849 * overflow if times way in the future are given. Further on Solaris versions
5850 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5851 * number of seconds, in abstime, is less than current_time + 100,000,000.
5852 * As it will be 28 years before "now + 100000000" will overflow we can
5853 * ignore overflow and just impose a hard-limit on seconds using the value
5854 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5855 * years from "now".
5856 */
5857
5858 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5859 assert (time > 0, "convertTime");
5860 time_t max_secs = 0;
5861
5862 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5863 struct timeval now;
5864 int status = gettimeofday(&now, NULL);
5865 assert(status == 0, "gettimeofday");
5866
5867 max_secs = now.tv_sec + MAX_SECS;
5868
5869 if (isAbsolute) {
5870 jlong secs = time / 1000;
5871 if (secs > max_secs) {
5872 absTime->tv_sec = max_secs;
5873 } else {
5874 absTime->tv_sec = secs;
5875 }
5876 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5877 } else {
5878 jlong secs = time / NANOSECS_PER_SEC;
5879 if (secs >= MAX_SECS) {
5880 absTime->tv_sec = max_secs;
5881 absTime->tv_nsec = 0;
5882 } else {
5883 absTime->tv_sec = now.tv_sec + secs;
5884 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5885 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5886 absTime->tv_nsec -= NANOSECS_PER_SEC;
5887 ++absTime->tv_sec; // note: this must be <= max_secs
5888 }
5889 }
5890 }
5891 } else {
5892 // must be relative using monotonic clock
5893 struct timespec now;
5894 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5895 assert_status(status == 0, status, "clock_gettime");
5896 max_secs = now.tv_sec + MAX_SECS;
5897 jlong secs = time / NANOSECS_PER_SEC;
5898 if (secs >= MAX_SECS) {
5899 absTime->tv_sec = max_secs;
5900 absTime->tv_nsec = 0;
5901 } else {
5902 absTime->tv_sec = now.tv_sec + secs;
5903 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5904 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5905 absTime->tv_nsec -= NANOSECS_PER_SEC;
5906 ++absTime->tv_sec; // note: this must be <= max_secs
5907 }
5908 }
5909 }
5910 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5911 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5912 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5913 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5914 }
5915
5916 void Parker::park(bool isAbsolute, jlong time) {
5917 // Ideally we'd do something useful while spinning, such
5918 // as calling unpackTime().
5919
5920 // Optional fast-path check:
5921 // Return immediately if a permit is available.
5922 // We depend on Atomic::xchg() having full barrier semantics
5923 // since we are doing a lock-free update to _counter.
5924 if (Atomic::xchg(0, &_counter) > 0) return;
5925
5926 Thread* thread = Thread::current();
5927 assert(thread->is_Java_thread(), "Must be JavaThread");
5928 JavaThread *jt = (JavaThread *)thread;
5929
5930 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5931 // Check interrupt before trying to wait
5932 if (Thread::is_interrupted(thread, false)) {
5933 return;
5934 }
5935
5936 // Next, demultiplex/decode time arguments
5937 timespec absTime;
5938 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5939 return;
5940 }
5941 if (time > 0) {
5942 unpackTime(&absTime, isAbsolute, time);
5943 }
5944
5945
5946 // Enter safepoint region
5947 // Beware of deadlocks such as 6317397.
5948 // The per-thread Parker:: mutex is a classic leaf-lock.
5949 // In particular a thread must never block on the Threads_lock while
5950 // holding the Parker:: mutex. If safepoints are pending both the
5951 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5952 ThreadBlockInVM tbivm(jt);
5953
5954 // Don't wait if cannot get lock since interference arises from
5955 // unblocking. Also. check interrupt before trying wait
5956 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5957 return;
5958 }
5959
5960 int status ;
5961 if (_counter > 0) { // no wait needed
5962 _counter = 0;
5963 status = pthread_mutex_unlock(_mutex);
5964 assert (status == 0, "invariant") ;
5965 // Paranoia to ensure our locked and lock-free paths interact
5966 // correctly with each other and Java-level accesses.
5967 OrderAccess::fence();
5968 return;
5969 }
5970
5971 #ifdef ASSERT
5972 // Don't catch signals while blocked; let the running threads have the signals.
5973 // (This allows a debugger to break into the running thread.)
5974 sigset_t oldsigs;
5975 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5976 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5977 #endif
5978
5979 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5980 jt->set_suspend_equivalent();
5981 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5982
5983 assert(_cur_index == -1, "invariant");
5984 if (time == 0) {
5985 _cur_index = REL_INDEX; // arbitrary choice when not timed
5986 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5987 } else {
5988 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5989 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5990 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5991 pthread_cond_destroy (&_cond[_cur_index]) ;
5992 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5993 }
5994 }
5995 _cur_index = -1;
5996 assert_status(status == 0 || status == EINTR ||
5997 status == ETIME || status == ETIMEDOUT,
5998 status, "cond_timedwait");
5999
6000 #ifdef ASSERT
6001 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6002 #endif
6003
6004 _counter = 0 ;
6005 status = pthread_mutex_unlock(_mutex) ;
6006 assert_status(status == 0, status, "invariant") ;
6007 // Paranoia to ensure our locked and lock-free paths interact
6008 // correctly with each other and Java-level accesses.
6009 OrderAccess::fence();
6010
6011 // If externally suspended while waiting, re-suspend
6012 if (jt->handle_special_suspend_equivalent_condition()) {
6013 jt->java_suspend_self();
6014 }
6015 }
6016
6017 void Parker::unpark() {
6018 int s, status ;
6019 status = pthread_mutex_lock(_mutex);
6020 assert (status == 0, "invariant") ;
6021 s = _counter;
6022 _counter = 1;
6023 if (s < 1) {
6024 // thread might be parked
6025 if (_cur_index != -1) {
6026 // thread is definitely parked
6027 if (WorkAroundNPTLTimedWaitHang) {
6028 status = pthread_cond_signal (&_cond[_cur_index]);
6029 assert (status == 0, "invariant");
6030 status = pthread_mutex_unlock(_mutex);
6031 assert (status == 0, "invariant");
6032 } else {
6033 // must capture correct index before unlocking
6034 int index = _cur_index;
6035 status = pthread_mutex_unlock(_mutex);
6036 assert (status == 0, "invariant");
6037 status = pthread_cond_signal (&_cond[index]);
6038 assert (status == 0, "invariant");
6039 }
6040 } else {
6041 pthread_mutex_unlock(_mutex);
6042 assert (status == 0, "invariant") ;
6043 }
6044 } else {
6045 pthread_mutex_unlock(_mutex);
6046 assert (status == 0, "invariant") ;
6047 }
6048 }
6049
6050
6051 extern char** environ;
6052
6053 // Run the specified command in a separate process. Return its exit value,
6054 // or -1 on failure (e.g. can't fork a new process).
6055 // Unlike system(), this function can be called from signal handler. It
6056 // doesn't block SIGINT et al.
6057 int os::fork_and_exec(char* cmd) {
6058 const char * argv[4] = {"sh", "-c", cmd, NULL};
6059
6060 pid_t pid = fork();
6061
6062 if (pid < 0) {
6063 // fork failed
6064 return -1;
6065
6066 } else if (pid == 0) {
6067 // child process
6068
6069 execve("/bin/sh", (char* const*)argv, environ);
6070
6071 // execve failed
6072 _exit(-1);
6073
6074 } else {
6075 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6076 // care about the actual exit code, for now.
6077
6078 int status;
6079
6080 // Wait for the child process to exit. This returns immediately if
6081 // the child has already exited. */
6082 while (waitpid(pid, &status, 0) < 0) {
6083 switch (errno) {
6084 case ECHILD: return 0;
6085 case EINTR: break;
6086 default: return -1;
6087 }
6088 }
6089
6090 if (WIFEXITED(status)) {
6091 // The child exited normally; get its exit code.
6092 return WEXITSTATUS(status);
6093 } else if (WIFSIGNALED(status)) {
6094 // The child exited because of a signal
6095 // The best value to return is 0x80 + signal number,
6096 // because that is what all Unix shells do, and because
6097 // it allows callers to distinguish between process exit and
6098 // process death by signal.
6099 return 0x80 + WTERMSIG(status);
6100 } else {
6101 // Unknown exit code; pass it through
6102 return status;
6103 }
6104 }
6105 }
6106
6107 // is_headless_jre()
6108 //
6109 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6110 // in order to report if we are running in a headless jre
6111 //
6112 // Since JDK8 xawt/libmawt.so was moved into the same directory
6113 // as libawt.so, and renamed libawt_xawt.so
6114 //
6115 bool os::is_headless_jre() {
6116 struct stat statbuf;
6117 char buf[MAXPATHLEN];
6118 char libmawtpath[MAXPATHLEN];
6119 const char *xawtstr = "/xawt/libmawt.so";
6120 const char *new_xawtstr = "/libawt_xawt.so";
6121 char *p;
6122
6123 // Get path to libjvm.so
6124 os::jvm_path(buf, sizeof(buf));
6125
6126 // Get rid of libjvm.so
6127 p = strrchr(buf, '/');
6128 if (p == NULL) return false;
6129 else *p = '\0';
6130
6131 // Get rid of client or server
6132 p = strrchr(buf, '/');
6133 if (p == NULL) return false;
6134 else *p = '\0';
6135
6136 // check xawt/libmawt.so
6137 strcpy(libmawtpath, buf);
6138 strcat(libmawtpath, xawtstr);
6139 if (::stat(libmawtpath, &statbuf) == 0) return false;
6140
6141 // check libawt_xawt.so
6142 strcpy(libmawtpath, buf);
6143 strcat(libmawtpath, new_xawtstr);
6144 if (::stat(libmawtpath, &statbuf) == 0) return false;
6145
6146 return true;
6147 }
6148
6149 // Get the default path to the core file
6150 // Returns the length of the string
6151 int os::get_core_path(char* buffer, size_t bufferSize) {
6152 const char* p = get_current_directory(buffer, bufferSize);
6153
6154 if (p == NULL) {
6155 assert(p != NULL, "failed to get current directory");
6156 return 0;
6157 }
6158
6159 return strlen(buffer);
6160 }
6161
6162 /////////////// Unit tests ///////////////
6163
6164 #ifndef PRODUCT
6165
6166 #define test_log(...) \
6167 do {\
6168 if (VerboseInternalVMTests) { \
6169 tty->print_cr(__VA_ARGS__); \
6170 tty->flush(); \
6171 }\
6172 } while (false)
6173
6174 class TestReserveMemorySpecial : AllStatic {
6175 public:
6176 static void small_page_write(void* addr, size_t size) {
6177 size_t page_size = os::vm_page_size();
6178
6179 char* end = (char*)addr + size;
6180 for (char* p = (char*)addr; p < end; p += page_size) {
6181 *p = 1;
6182 }
6183 }
6184
6185 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6186 if (!UseHugeTLBFS) {
6187 return;
6188 }
6189
6190 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6191
6192 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6193
6194 if (addr != NULL) {
6195 small_page_write(addr, size);
6196
6197 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6198 }
6199 }
6200
6201 static void test_reserve_memory_special_huge_tlbfs_only() {
6202 if (!UseHugeTLBFS) {
6203 return;
6204 }
6205
6206 size_t lp = os::large_page_size();
6207
6208 for (size_t size = lp; size <= lp * 10; size += lp) {
6209 test_reserve_memory_special_huge_tlbfs_only(size);
6210 }
6211 }
6212
6213 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6214 size_t lp = os::large_page_size();
6215 size_t ag = os::vm_allocation_granularity();
6216
6217 // sizes to test
6218 const size_t sizes[] = {
6219 lp, lp + ag, lp + lp / 2, lp * 2,
6220 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6221 lp * 10, lp * 10 + lp / 2
6222 };
6223 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6224
6225 // For each size/alignment combination, we test three scenarios:
6226 // 1) with req_addr == NULL
6227 // 2) with a non-null req_addr at which we expect to successfully allocate
6228 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6229 // expect the allocation to either fail or to ignore req_addr
6230
6231 // Pre-allocate two areas; they shall be as large as the largest allocation
6232 // and aligned to the largest alignment we will be testing.
6233 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6234 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6235 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6236 -1, 0);
6237 assert(mapping1 != MAP_FAILED, "should work");
6238
6239 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6240 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6241 -1, 0);
6242 assert(mapping2 != MAP_FAILED, "should work");
6243
6244 // Unmap the first mapping, but leave the second mapping intact: the first
6245 // mapping will serve as a value for a "good" req_addr (case 2). The second
6246 // mapping, still intact, as "bad" req_addr (case 3).
6247 ::munmap(mapping1, mapping_size);
6248
6249 // Case 1
6250 test_log("%s, req_addr NULL:", __FUNCTION__);
6251 test_log("size align result");
6252
6253 for (int i = 0; i < num_sizes; i++) {
6254 const size_t size = sizes[i];
6255 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6256 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6257 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6258 size, alignment, p, (p != NULL ? "" : "(failed)"));
6259 if (p != NULL) {
6260 assert(is_ptr_aligned(p, alignment), "must be");
6261 small_page_write(p, size);
6262 os::Linux::release_memory_special_huge_tlbfs(p, size);
6263 }
6264 }
6265 }
6266
6267 // Case 2
6268 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6269 test_log("size align req_addr result");
6270
6271 for (int i = 0; i < num_sizes; i++) {
6272 const size_t size = sizes[i];
6273 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6274 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6275 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6276 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6277 size, alignment, req_addr, p,
6278 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6279 if (p != NULL) {
6280 assert(p == req_addr, "must be");
6281 small_page_write(p, size);
6282 os::Linux::release_memory_special_huge_tlbfs(p, size);
6283 }
6284 }
6285 }
6286
6287 // Case 3
6288 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6289 test_log("size align req_addr result");
6290
6291 for (int i = 0; i < num_sizes; i++) {
6292 const size_t size = sizes[i];
6293 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6294 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6295 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6296 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6297 size, alignment, req_addr, p,
6298 ((p != NULL ? "" : "(failed)")));
6299 // as the area around req_addr contains already existing mappings, the API should always
6300 // return NULL (as per contract, it cannot return another address)
6301 assert(p == NULL, "must be");
6302 }
6303 }
6304
6305 ::munmap(mapping2, mapping_size);
6306
6307 }
6308
6309 static void test_reserve_memory_special_huge_tlbfs() {
6310 if (!UseHugeTLBFS) {
6311 return;
6312 }
6313
6314 test_reserve_memory_special_huge_tlbfs_only();
6315 test_reserve_memory_special_huge_tlbfs_mixed();
6316 }
6317
6318 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6319 if (!UseSHM) {
6320 return;
6321 }
6322
6323 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6324
6325 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6326
6327 if (addr != NULL) {
6328 assert(is_ptr_aligned(addr, alignment), "Check");
6329 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6330
6331 small_page_write(addr, size);
6332
6333 os::Linux::release_memory_special_shm(addr, size);
6334 }
6335 }
6336
6337 static void test_reserve_memory_special_shm() {
6338 size_t lp = os::large_page_size();
6339 size_t ag = os::vm_allocation_granularity();
6340
6341 for (size_t size = ag; size < lp * 3; size += ag) {
6342 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6343 test_reserve_memory_special_shm(size, alignment);
6344 }
6345 }
6346 }
6347
6348 static void test() {
6349 test_reserve_memory_special_huge_tlbfs();
6350 test_reserve_memory_special_shm();
6351 }
6352 };
6353
6354 void TestReserveMemorySpecial_test() {
6355 TestReserveMemorySpecial::test();
6356 }
6357
6358 #endif