1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/macros.hpp"
72 #include "utilities/vmError.hpp"
73
74 // put OS-includes here
75 # include <sys/types.h>
76 # include <sys/mman.h>
77 # include <sys/stat.h>
78 # include <sys/select.h>
79 # include <pthread.h>
80 # include <signal.h>
81 # include <errno.h>
82 # include <dlfcn.h>
83 # include <stdio.h>
84 # include <unistd.h>
85 # include <sys/resource.h>
86 # include <pthread.h>
87 # include <sys/stat.h>
88 # include <sys/time.h>
89 # include <sys/times.h>
90 # include <sys/utsname.h>
91 # include <sys/socket.h>
92 # include <sys/wait.h>
93 # include <pwd.h>
94 # include <poll.h>
95 # include <semaphore.h>
96 # include <fcntl.h>
97 # include <string.h>
98 # include <syscall.h>
99 # include <sys/sysinfo.h>
100 # include <gnu/libc-version.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
109
110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
111 // getrusage() is prepared to handle the associated failure.
112 #ifndef RUSAGE_THREAD
113 #define RUSAGE_THREAD (1) /* only the calling thread */
114 #endif
115
116 #define MAX_PATH (2 * K)
117
118 #define MAX_SECS 100000000
119
120 // for timer info max values which include all bits
121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123 #define LARGEPAGES_BIT (1 << 6)
124 ////////////////////////////////////////////////////////////////////////////////
125 // global variables
126 julong os::Linux::_physical_memory = 0;
127
128 address os::Linux::_initial_thread_stack_bottom = NULL;
129 uintptr_t os::Linux::_initial_thread_stack_size = 0;
130
131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 const int os::Linux::_vm_default_page_size = (8 * K);
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
143 pthread_condattr_t os::Linux::_condattr[1];
144
145 static jlong initial_time_count=0;
146
147 static int clock_tics_per_sec = 100;
148
149 // For diagnostics to print a message once. see run_periodic_checks
150 static sigset_t check_signal_done;
151 static bool check_signals = true;
152
153 static pid_t _initial_pid = 0;
154
155 // Signal number used to suspend/resume a thread
156
157 // do not use any signal number less than SIGSEGV, see 4355769
158 static int SR_signum = SIGUSR2;
159 sigset_t SR_sigset;
160
161 // Used to protect dlsym() calls
162 static pthread_mutex_t dl_mutex;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 ////////////////////////////////////////////////////////////////////////////////
188 // environment support
189
190 bool os::getenv(const char* name, char* buf, int len) {
191 const char* val = ::getenv(name);
192 if (val != NULL && strlen(val) < (size_t)len) {
193 strcpy(buf, val);
194 return true;
195 }
196 if (len > 0) buf[0] = 0; // return a null string
197 return false;
198 }
199
200
201 // Return true if user is running as root.
202
203 bool os::have_special_privileges() {
204 static bool init = false;
205 static bool privileges = false;
206 if (!init) {
207 privileges = (getuid() != geteuid()) || (getgid() != getegid());
208 init = true;
209 }
210 return privileges;
211 }
212
213
214 #ifndef SYS_gettid
215 // i386: 224, ia64: 1105, amd64: 186, sparc 143
216 #ifdef __ia64__
217 #define SYS_gettid 1105
218 #elif __i386__
219 #define SYS_gettid 224
220 #elif __amd64__
221 #define SYS_gettid 186
222 #elif __sparc__
223 #define SYS_gettid 143
224 #else
225 #error define gettid for the arch
226 #endif
227 #endif
228
229 // Cpu architecture string
230 #if defined(ZERO)
231 static char cpu_arch[] = ZERO_LIBARCH;
232 #elif defined(IA64)
233 static char cpu_arch[] = "ia64";
234 #elif defined(IA32)
235 static char cpu_arch[] = "i386";
236 #elif defined(AMD64)
237 static char cpu_arch[] = "amd64";
238 #elif defined(ARM)
239 static char cpu_arch[] = "arm";
240 #elif defined(PPC32)
241 static char cpu_arch[] = "ppc";
242 #elif defined(PPC64)
243 #if defined(VM_LITTLE_ENDIAN)
244 static char cpu_arch[] = "ppc64le";
245 #else
246 static char cpu_arch[] = "ppc64";
247 #endif
248 #elif defined(SPARC)
249 #ifdef _LP64
250 static char cpu_arch[] = "sparcv9";
251 #else
252 static char cpu_arch[] = "sparc";
253 #endif
254 #else
255 #error Add appropriate cpu_arch setting
256 #endif
257
258
259 // pid_t gettid()
260 //
261 // Returns the kernel thread id of the currently running thread. Kernel
262 // thread id is used to access /proc.
263 //
264 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
265 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
266 //
267 pid_t os::Linux::gettid() {
268 int rslt = syscall(SYS_gettid);
269 if (rslt == -1) {
270 // old kernel, no NPTL support
271 return getpid();
272 } else {
273 return (pid_t)rslt;
274 }
275 }
276
277 // Most versions of linux have a bug where the number of processors are
278 // determined by looking at the /proc file system. In a chroot environment,
279 // the system call returns 1. This causes the VM to act as if it is
280 // a single processor and elide locking (see is_MP() call).
281 static bool unsafe_chroot_detected = false;
282 static const char *unstable_chroot_error = "/proc file system not found.\n"
283 "Java may be unstable running multithreaded in a chroot "
284 "environment on Linux when /proc filesystem is not mounted.";
285
286 void os::Linux::initialize_system_info() {
287 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
288 if (processor_count() == 1) {
289 pid_t pid = os::Linux::gettid();
290 char fname[32];
291 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
292 FILE *fp = fopen(fname, "r");
293 if (fp == NULL) {
294 unsafe_chroot_detected = true;
295 } else {
296 fclose(fp);
297 }
298 }
299 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
300 assert(processor_count() > 0, "linux error");
301 }
302
303 void os::init_system_properties_values() {
304 // The next steps are taken in the product version:
305 //
306 // Obtain the JAVA_HOME value from the location of libjvm.so.
307 // This library should be located at:
308 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
309 //
310 // If "/jre/lib/" appears at the right place in the path, then we
311 // assume libjvm.so is installed in a JDK and we use this path.
312 //
313 // Otherwise exit with message: "Could not create the Java virtual machine."
314 //
315 // The following extra steps are taken in the debugging version:
316 //
317 // If "/jre/lib/" does NOT appear at the right place in the path
318 // instead of exit check for $JAVA_HOME environment variable.
319 //
320 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
321 // then we append a fake suffix "hotspot/libjvm.so" to this path so
322 // it looks like libjvm.so is installed there
323 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
324 //
325 // Otherwise exit.
326 //
327 // Important note: if the location of libjvm.so changes this
328 // code needs to be changed accordingly.
329
330 // See ld(1):
331 // The linker uses the following search paths to locate required
332 // shared libraries:
333 // 1: ...
334 // ...
335 // 7: The default directories, normally /lib and /usr/lib.
336 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
337 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
338 #else
339 #define DEFAULT_LIBPATH "/lib:/usr/lib"
340 #endif
341
342 // Base path of extensions installed on the system.
343 #define SYS_EXT_DIR "/usr/java/packages"
344 #define EXTENSIONS_DIR "/lib/ext"
345
346 // Buffer that fits several sprintfs.
347 // Note that the space for the colon and the trailing null are provided
348 // by the nulls included by the sizeof operator.
349 const size_t bufsize =
350 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
351 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
352 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
353
354 // sysclasspath, java_home, dll_dir
355 {
356 char *pslash;
357 os::jvm_path(buf, bufsize);
358
359 // Found the full path to libjvm.so.
360 // Now cut the path to <java_home>/jre if we can.
361 pslash = strrchr(buf, '/');
362 if (pslash != NULL) {
363 *pslash = '\0'; // Get rid of /libjvm.so.
364 }
365 pslash = strrchr(buf, '/');
366 if (pslash != NULL) {
367 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
368 }
369 Arguments::set_dll_dir(buf);
370
371 if (pslash != NULL) {
372 pslash = strrchr(buf, '/');
373 if (pslash != NULL) {
374 *pslash = '\0'; // Get rid of /<arch>.
375 pslash = strrchr(buf, '/');
376 if (pslash != NULL) {
377 *pslash = '\0'; // Get rid of /lib.
378 }
379 }
380 }
381 Arguments::set_java_home(buf);
382 set_boot_path('/', ':');
383 }
384
385 // Where to look for native libraries.
386 //
387 // Note: Due to a legacy implementation, most of the library path
388 // is set in the launcher. This was to accomodate linking restrictions
389 // on legacy Linux implementations (which are no longer supported).
390 // Eventually, all the library path setting will be done here.
391 //
392 // However, to prevent the proliferation of improperly built native
393 // libraries, the new path component /usr/java/packages is added here.
394 // Eventually, all the library path setting will be done here.
395 {
396 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
397 // should always exist (until the legacy problem cited above is
398 // addressed).
399 const char *v = ::getenv("LD_LIBRARY_PATH");
400 const char *v_colon = ":";
401 if (v == NULL) { v = ""; v_colon = ""; }
402 // That's +1 for the colon and +1 for the trailing '\0'.
403 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
404 strlen(v) + 1 +
405 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
406 mtInternal);
407 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
408 Arguments::set_library_path(ld_library_path);
409 FREE_C_HEAP_ARRAY(char, ld_library_path);
410 }
411
412 // Extensions directories.
413 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
414 Arguments::set_ext_dirs(buf);
415
416 FREE_C_HEAP_ARRAY(char, buf);
417
418 #undef DEFAULT_LIBPATH
419 #undef SYS_EXT_DIR
420 #undef EXTENSIONS_DIR
421 }
422
423 ////////////////////////////////////////////////////////////////////////////////
424 // breakpoint support
425
426 void os::breakpoint() {
427 BREAKPOINT;
428 }
429
430 extern "C" void breakpoint() {
431 // use debugger to set breakpoint here
432 }
433
434 ////////////////////////////////////////////////////////////////////////////////
435 // signal support
436
437 debug_only(static bool signal_sets_initialized = false);
438 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
439
440 bool os::Linux::is_sig_ignored(int sig) {
441 struct sigaction oact;
442 sigaction(sig, (struct sigaction*)NULL, &oact);
443 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
444 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
445 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
446 return true;
447 } else {
448 return false;
449 }
450 }
451
452 void os::Linux::signal_sets_init() {
453 // Should also have an assertion stating we are still single-threaded.
454 assert(!signal_sets_initialized, "Already initialized");
455 // Fill in signals that are necessarily unblocked for all threads in
456 // the VM. Currently, we unblock the following signals:
457 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
458 // by -Xrs (=ReduceSignalUsage));
459 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
460 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
461 // the dispositions or masks wrt these signals.
462 // Programs embedding the VM that want to use the above signals for their
463 // own purposes must, at this time, use the "-Xrs" option to prevent
464 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
465 // (See bug 4345157, and other related bugs).
466 // In reality, though, unblocking these signals is really a nop, since
467 // these signals are not blocked by default.
468 sigemptyset(&unblocked_sigs);
469 sigemptyset(&allowdebug_blocked_sigs);
470 sigaddset(&unblocked_sigs, SIGILL);
471 sigaddset(&unblocked_sigs, SIGSEGV);
472 sigaddset(&unblocked_sigs, SIGBUS);
473 sigaddset(&unblocked_sigs, SIGFPE);
474 #if defined(PPC64)
475 sigaddset(&unblocked_sigs, SIGTRAP);
476 #endif
477 sigaddset(&unblocked_sigs, SR_signum);
478
479 if (!ReduceSignalUsage) {
480 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
481 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
482 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
483 }
484 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
485 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
486 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
487 }
488 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
489 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
490 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
491 }
492 }
493 // Fill in signals that are blocked by all but the VM thread.
494 sigemptyset(&vm_sigs);
495 if (!ReduceSignalUsage) {
496 sigaddset(&vm_sigs, BREAK_SIGNAL);
497 }
498 debug_only(signal_sets_initialized = true);
499
500 }
501
502 // These are signals that are unblocked while a thread is running Java.
503 // (For some reason, they get blocked by default.)
504 sigset_t* os::Linux::unblocked_signals() {
505 assert(signal_sets_initialized, "Not initialized");
506 return &unblocked_sigs;
507 }
508
509 // These are the signals that are blocked while a (non-VM) thread is
510 // running Java. Only the VM thread handles these signals.
511 sigset_t* os::Linux::vm_signals() {
512 assert(signal_sets_initialized, "Not initialized");
513 return &vm_sigs;
514 }
515
516 // These are signals that are blocked during cond_wait to allow debugger in
517 sigset_t* os::Linux::allowdebug_blocked_signals() {
518 assert(signal_sets_initialized, "Not initialized");
519 return &allowdebug_blocked_sigs;
520 }
521
522 void os::Linux::hotspot_sigmask(Thread* thread) {
523
524 //Save caller's signal mask before setting VM signal mask
525 sigset_t caller_sigmask;
526 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
527
528 OSThread* osthread = thread->osthread();
529 osthread->set_caller_sigmask(caller_sigmask);
530
531 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
532
533 if (!ReduceSignalUsage) {
534 if (thread->is_VM_thread()) {
535 // Only the VM thread handles BREAK_SIGNAL ...
536 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
537 } else {
538 // ... all other threads block BREAK_SIGNAL
539 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
540 }
541 }
542 }
543
544 //////////////////////////////////////////////////////////////////////////////
545 // detecting pthread library
546
547 void os::Linux::libpthread_init() {
548 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
549 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
550 // generic name for earlier versions.
551 // Define macros here so we can build HotSpot on old systems.
552 #ifndef _CS_GNU_LIBC_VERSION
553 #define _CS_GNU_LIBC_VERSION 2
554 #endif
555 #ifndef _CS_GNU_LIBPTHREAD_VERSION
556 #define _CS_GNU_LIBPTHREAD_VERSION 3
557 #endif
558
559 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
560 if (n > 0) {
561 char *str = (char *)malloc(n, mtInternal);
562 confstr(_CS_GNU_LIBC_VERSION, str, n);
563 os::Linux::set_glibc_version(str);
564 } else {
565 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
566 static char _gnu_libc_version[32];
567 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
568 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
569 os::Linux::set_glibc_version(_gnu_libc_version);
570 }
571
572 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
573 if (n > 0) {
574 char *str = (char *)malloc(n, mtInternal);
575 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
576 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
577 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
578 // is the case. LinuxThreads has a hard limit on max number of threads.
579 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
580 // On the other hand, NPTL does not have such a limit, sysconf()
581 // will return -1 and errno is not changed. Check if it is really NPTL.
582 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
583 strstr(str, "NPTL") &&
584 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
585 free(str);
586 os::Linux::set_libpthread_version("linuxthreads");
587 } else {
588 os::Linux::set_libpthread_version(str);
589 }
590 } else {
591 // glibc before 2.3.2 only has LinuxThreads.
592 os::Linux::set_libpthread_version("linuxthreads");
593 }
594
595 if (strstr(libpthread_version(), "NPTL")) {
596 os::Linux::set_is_NPTL();
597 } else {
598 os::Linux::set_is_LinuxThreads();
599 }
600
601 // LinuxThreads have two flavors: floating-stack mode, which allows variable
602 // stack size; and fixed-stack mode. NPTL is always floating-stack.
603 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
604 os::Linux::set_is_floating_stack();
605 }
606 }
607
608 /////////////////////////////////////////////////////////////////////////////
609 // thread stack
610
611 // Force Linux kernel to expand current thread stack. If "bottom" is close
612 // to the stack guard, caller should block all signals.
613 //
614 // MAP_GROWSDOWN:
615 // A special mmap() flag that is used to implement thread stacks. It tells
616 // kernel that the memory region should extend downwards when needed. This
617 // allows early versions of LinuxThreads to only mmap the first few pages
618 // when creating a new thread. Linux kernel will automatically expand thread
619 // stack as needed (on page faults).
620 //
621 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
622 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
623 // region, it's hard to tell if the fault is due to a legitimate stack
624 // access or because of reading/writing non-exist memory (e.g. buffer
625 // overrun). As a rule, if the fault happens below current stack pointer,
626 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
627 // application (see Linux kernel fault.c).
628 //
629 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
630 // stack overflow detection.
631 //
632 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
633 // not use this flag. However, the stack of initial thread is not created
634 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
635 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
636 // and then attach the thread to JVM.
637 //
638 // To get around the problem and allow stack banging on Linux, we need to
639 // manually expand thread stack after receiving the SIGSEGV.
640 //
641 // There are two ways to expand thread stack to address "bottom", we used
642 // both of them in JVM before 1.5:
643 // 1. adjust stack pointer first so that it is below "bottom", and then
644 // touch "bottom"
645 // 2. mmap() the page in question
646 //
647 // Now alternate signal stack is gone, it's harder to use 2. For instance,
648 // if current sp is already near the lower end of page 101, and we need to
649 // call mmap() to map page 100, it is possible that part of the mmap() frame
650 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
651 // That will destroy the mmap() frame and cause VM to crash.
652 //
653 // The following code works by adjusting sp first, then accessing the "bottom"
654 // page to force a page fault. Linux kernel will then automatically expand the
655 // stack mapping.
656 //
657 // _expand_stack_to() assumes its frame size is less than page size, which
658 // should always be true if the function is not inlined.
659
660 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
661 #define NOINLINE
662 #else
663 #define NOINLINE __attribute__ ((noinline))
664 #endif
665
666 static void _expand_stack_to(address bottom) NOINLINE;
667
668 static void _expand_stack_to(address bottom) {
669 address sp;
670 size_t size;
671 volatile char *p;
672
673 // Adjust bottom to point to the largest address within the same page, it
674 // gives us a one-page buffer if alloca() allocates slightly more memory.
675 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
676 bottom += os::Linux::page_size() - 1;
677
678 // sp might be slightly above current stack pointer; if that's the case, we
679 // will alloca() a little more space than necessary, which is OK. Don't use
680 // os::current_stack_pointer(), as its result can be slightly below current
681 // stack pointer, causing us to not alloca enough to reach "bottom".
682 sp = (address)&sp;
683
684 if (sp > bottom) {
685 size = sp - bottom;
686 p = (volatile char *)alloca(size);
687 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
688 p[0] = '\0';
689 }
690 }
691
692 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
693 assert(t!=NULL, "just checking");
694 assert(t->osthread()->expanding_stack(), "expand should be set");
695 assert(t->stack_base() != NULL, "stack_base was not initialized");
696
697 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
698 sigset_t mask_all, old_sigset;
699 sigfillset(&mask_all);
700 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
701 _expand_stack_to(addr);
702 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
703 return true;
704 }
705 return false;
706 }
707
708 //////////////////////////////////////////////////////////////////////////////
709 // create new thread
710
711 static address highest_vm_reserved_address();
712
713 // check if it's safe to start a new thread
714 static bool _thread_safety_check(Thread* thread) {
715 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
716 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
717 // Heap is mmap'ed at lower end of memory space. Thread stacks are
718 // allocated (MAP_FIXED) from high address space. Every thread stack
719 // occupies a fixed size slot (usually 2Mbytes, but user can change
720 // it to other values if they rebuild LinuxThreads).
721 //
722 // Problem with MAP_FIXED is that mmap() can still succeed even part of
723 // the memory region has already been mmap'ed. That means if we have too
724 // many threads and/or very large heap, eventually thread stack will
725 // collide with heap.
726 //
727 // Here we try to prevent heap/stack collision by comparing current
728 // stack bottom with the highest address that has been mmap'ed by JVM
729 // plus a safety margin for memory maps created by native code.
730 //
731 // This feature can be disabled by setting ThreadSafetyMargin to 0
732 //
733 if (ThreadSafetyMargin > 0) {
734 address stack_bottom = os::current_stack_base() - os::current_stack_size();
735
736 // not safe if our stack extends below the safety margin
737 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
738 } else {
739 return true;
740 }
741 } else {
742 // Floating stack LinuxThreads or NPTL:
743 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
744 // there's not enough space left, pthread_create() will fail. If we come
745 // here, that means enough space has been reserved for stack.
746 return true;
747 }
748 }
749
750 // Thread start routine for all newly created threads
751 static void *java_start(Thread *thread) {
752 // Try to randomize the cache line index of hot stack frames.
753 // This helps when threads of the same stack traces evict each other's
754 // cache lines. The threads can be either from the same JVM instance, or
755 // from different JVM instances. The benefit is especially true for
756 // processors with hyperthreading technology.
757 static int counter = 0;
758 int pid = os::current_process_id();
759 alloca(((pid ^ counter++) & 7) * 128);
760
761 ThreadLocalStorage::set_thread(thread);
762
763 OSThread* osthread = thread->osthread();
764 Monitor* sync = osthread->startThread_lock();
765
766 // non floating stack LinuxThreads needs extra check, see above
767 if (!_thread_safety_check(thread)) {
768 // notify parent thread
769 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
770 osthread->set_state(ZOMBIE);
771 sync->notify_all();
772 return NULL;
773 }
774
775 // thread_id is kernel thread id (similar to Solaris LWP id)
776 osthread->set_thread_id(os::Linux::gettid());
777
778 if (UseNUMA) {
779 int lgrp_id = os::numa_get_group_id();
780 if (lgrp_id != -1) {
781 thread->set_lgrp_id(lgrp_id);
782 }
783 }
784 // initialize signal mask for this thread
785 os::Linux::hotspot_sigmask(thread);
786
787 // initialize floating point control register
788 os::Linux::init_thread_fpu_state();
789
790 // handshaking with parent thread
791 {
792 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
793
794 // notify parent thread
795 osthread->set_state(INITIALIZED);
796 sync->notify_all();
797
798 // wait until os::start_thread()
799 while (osthread->get_state() == INITIALIZED) {
800 sync->wait(Mutex::_no_safepoint_check_flag);
801 }
802 }
803
804 // call one more level start routine
805 thread->run();
806
807 return 0;
808 }
809
810 bool os::create_thread(Thread* thread, ThreadType thr_type,
811 size_t stack_size) {
812 assert(thread->osthread() == NULL, "caller responsible");
813
814 // Allocate the OSThread object
815 OSThread* osthread = new OSThread(NULL, NULL);
816 if (osthread == NULL) {
817 return false;
818 }
819
820 // set the correct thread state
821 osthread->set_thread_type(thr_type);
822
823 // Initial state is ALLOCATED but not INITIALIZED
824 osthread->set_state(ALLOCATED);
825
826 thread->set_osthread(osthread);
827
828 // init thread attributes
829 pthread_attr_t attr;
830 pthread_attr_init(&attr);
831 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
832
833 // stack size
834 if (os::Linux::supports_variable_stack_size()) {
835 // calculate stack size if it's not specified by caller
836 if (stack_size == 0) {
837 stack_size = os::Linux::default_stack_size(thr_type);
838
839 switch (thr_type) {
840 case os::java_thread:
841 // Java threads use ThreadStackSize which default value can be
842 // changed with the flag -Xss
843 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
844 stack_size = JavaThread::stack_size_at_create();
845 break;
846 case os::compiler_thread:
847 if (CompilerThreadStackSize > 0) {
848 stack_size = (size_t)(CompilerThreadStackSize * K);
849 break;
850 } // else fall through:
851 // use VMThreadStackSize if CompilerThreadStackSize is not defined
852 case os::vm_thread:
853 case os::pgc_thread:
854 case os::cgc_thread:
855 case os::watcher_thread:
856 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
857 break;
858 }
859 }
860
861 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
862 pthread_attr_setstacksize(&attr, stack_size);
863 } else {
864 // let pthread_create() pick the default value.
865 }
866
867 // glibc guard page
868 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
869
870 ThreadState state;
871
872 {
873 // Serialize thread creation if we are running with fixed stack LinuxThreads
874 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
875 if (lock) {
876 os::Linux::createThread_lock()->lock_without_safepoint_check();
877 }
878
879 pthread_t tid;
880 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
881
882 pthread_attr_destroy(&attr);
883
884 if (ret != 0) {
885 if (PrintMiscellaneous && (Verbose || WizardMode)) {
886 perror("pthread_create()");
887 }
888 // Need to clean up stuff we've allocated so far
889 thread->set_osthread(NULL);
890 delete osthread;
891 if (lock) os::Linux::createThread_lock()->unlock();
892 return false;
893 }
894
895 // Store pthread info into the OSThread
896 osthread->set_pthread_id(tid);
897
898 // Wait until child thread is either initialized or aborted
899 {
900 Monitor* sync_with_child = osthread->startThread_lock();
901 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
902 while ((state = osthread->get_state()) == ALLOCATED) {
903 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
904 }
905 }
906
907 if (lock) {
908 os::Linux::createThread_lock()->unlock();
909 }
910 }
911
912 // Aborted due to thread limit being reached
913 if (state == ZOMBIE) {
914 thread->set_osthread(NULL);
915 delete osthread;
916 return false;
917 }
918
919 // The thread is returned suspended (in state INITIALIZED),
920 // and is started higher up in the call chain
921 assert(state == INITIALIZED, "race condition");
922 return true;
923 }
924
925 /////////////////////////////////////////////////////////////////////////////
926 // attach existing thread
927
928 // bootstrap the main thread
929 bool os::create_main_thread(JavaThread* thread) {
930 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
931 return create_attached_thread(thread);
932 }
933
934 bool os::create_attached_thread(JavaThread* thread) {
935 #ifdef ASSERT
936 thread->verify_not_published();
937 #endif
938
939 // Allocate the OSThread object
940 OSThread* osthread = new OSThread(NULL, NULL);
941
942 if (osthread == NULL) {
943 return false;
944 }
945
946 // Store pthread info into the OSThread
947 osthread->set_thread_id(os::Linux::gettid());
948 osthread->set_pthread_id(::pthread_self());
949
950 // initialize floating point control register
951 os::Linux::init_thread_fpu_state();
952
953 // Initial thread state is RUNNABLE
954 osthread->set_state(RUNNABLE);
955
956 thread->set_osthread(osthread);
957
958 if (UseNUMA) {
959 int lgrp_id = os::numa_get_group_id();
960 if (lgrp_id != -1) {
961 thread->set_lgrp_id(lgrp_id);
962 }
963 }
964
965 if (os::Linux::is_initial_thread()) {
966 // If current thread is initial thread, its stack is mapped on demand,
967 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
968 // the entire stack region to avoid SEGV in stack banging.
969 // It is also useful to get around the heap-stack-gap problem on SuSE
970 // kernel (see 4821821 for details). We first expand stack to the top
971 // of yellow zone, then enable stack yellow zone (order is significant,
972 // enabling yellow zone first will crash JVM on SuSE Linux), so there
973 // is no gap between the last two virtual memory regions.
974
975 JavaThread *jt = (JavaThread *)thread;
976 address addr = jt->stack_yellow_zone_base();
977 assert(addr != NULL, "initialization problem?");
978 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
979
980 osthread->set_expanding_stack();
981 os::Linux::manually_expand_stack(jt, addr);
982 osthread->clear_expanding_stack();
983 }
984
985 // initialize signal mask for this thread
986 // and save the caller's signal mask
987 os::Linux::hotspot_sigmask(thread);
988
989 return true;
990 }
991
992 void os::pd_start_thread(Thread* thread) {
993 OSThread * osthread = thread->osthread();
994 assert(osthread->get_state() != INITIALIZED, "just checking");
995 Monitor* sync_with_child = osthread->startThread_lock();
996 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
997 sync_with_child->notify();
998 }
999
1000 // Free Linux resources related to the OSThread
1001 void os::free_thread(OSThread* osthread) {
1002 assert(osthread != NULL, "osthread not set");
1003
1004 if (Thread::current()->osthread() == osthread) {
1005 // Restore caller's signal mask
1006 sigset_t sigmask = osthread->caller_sigmask();
1007 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1008 }
1009
1010 delete osthread;
1011 }
1012
1013 //////////////////////////////////////////////////////////////////////////////
1014 // thread local storage
1015
1016 // Restore the thread pointer if the destructor is called. This is in case
1017 // someone from JNI code sets up a destructor with pthread_key_create to run
1018 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1019 // will hang or crash. When detachCurrentThread is called the key will be set
1020 // to null and we will not be called again. If detachCurrentThread is never
1021 // called we could loop forever depending on the pthread implementation.
1022 static void restore_thread_pointer(void* p) {
1023 Thread* thread = (Thread*) p;
1024 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1025 }
1026
1027 int os::allocate_thread_local_storage() {
1028 pthread_key_t key;
1029 int rslt = pthread_key_create(&key, restore_thread_pointer);
1030 assert(rslt == 0, "cannot allocate thread local storage");
1031 return (int)key;
1032 }
1033
1034 // Note: This is currently not used by VM, as we don't destroy TLS key
1035 // on VM exit.
1036 void os::free_thread_local_storage(int index) {
1037 int rslt = pthread_key_delete((pthread_key_t)index);
1038 assert(rslt == 0, "invalid index");
1039 }
1040
1041 void os::thread_local_storage_at_put(int index, void* value) {
1042 int rslt = pthread_setspecific((pthread_key_t)index, value);
1043 assert(rslt == 0, "pthread_setspecific failed");
1044 }
1045
1046 extern "C" Thread* get_thread() {
1047 return ThreadLocalStorage::thread();
1048 }
1049
1050 //////////////////////////////////////////////////////////////////////////////
1051 // initial thread
1052
1053 // Check if current thread is the initial thread, similar to Solaris thr_main.
1054 bool os::Linux::is_initial_thread(void) {
1055 char dummy;
1056 // If called before init complete, thread stack bottom will be null.
1057 // Can be called if fatal error occurs before initialization.
1058 if (initial_thread_stack_bottom() == NULL) return false;
1059 assert(initial_thread_stack_bottom() != NULL &&
1060 initial_thread_stack_size() != 0,
1061 "os::init did not locate initial thread's stack region");
1062 if ((address)&dummy >= initial_thread_stack_bottom() &&
1063 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1064 return true;
1065 } else {
1066 return false;
1067 }
1068 }
1069
1070 // Find the virtual memory area that contains addr
1071 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1072 FILE *fp = fopen("/proc/self/maps", "r");
1073 if (fp) {
1074 address low, high;
1075 while (!feof(fp)) {
1076 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1077 if (low <= addr && addr < high) {
1078 if (vma_low) *vma_low = low;
1079 if (vma_high) *vma_high = high;
1080 fclose(fp);
1081 return true;
1082 }
1083 }
1084 for (;;) {
1085 int ch = fgetc(fp);
1086 if (ch == EOF || ch == (int)'\n') break;
1087 }
1088 }
1089 fclose(fp);
1090 }
1091 return false;
1092 }
1093
1094 // Locate initial thread stack. This special handling of initial thread stack
1095 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1096 // bogus value for initial thread.
1097 void os::Linux::capture_initial_stack(size_t max_size) {
1098 // stack size is the easy part, get it from RLIMIT_STACK
1099 size_t stack_size;
1100 struct rlimit rlim;
1101 getrlimit(RLIMIT_STACK, &rlim);
1102 stack_size = rlim.rlim_cur;
1103
1104 // 6308388: a bug in ld.so will relocate its own .data section to the
1105 // lower end of primordial stack; reduce ulimit -s value a little bit
1106 // so we won't install guard page on ld.so's data section.
1107 stack_size -= 2 * page_size();
1108
1109 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1110 // 7.1, in both cases we will get 2G in return value.
1111 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1112 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1113 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1114 // in case other parts in glibc still assumes 2M max stack size.
1115 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1116 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1117 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1118 stack_size = 2 * K * K IA64_ONLY(*2);
1119 }
1120 // Try to figure out where the stack base (top) is. This is harder.
1121 //
1122 // When an application is started, glibc saves the initial stack pointer in
1123 // a global variable "__libc_stack_end", which is then used by system
1124 // libraries. __libc_stack_end should be pretty close to stack top. The
1125 // variable is available since the very early days. However, because it is
1126 // a private interface, it could disappear in the future.
1127 //
1128 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1129 // to __libc_stack_end, it is very close to stack top, but isn't the real
1130 // stack top. Note that /proc may not exist if VM is running as a chroot
1131 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1132 // /proc/<pid>/stat could change in the future (though unlikely).
1133 //
1134 // We try __libc_stack_end first. If that doesn't work, look for
1135 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1136 // as a hint, which should work well in most cases.
1137
1138 uintptr_t stack_start;
1139
1140 // try __libc_stack_end first
1141 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1142 if (p && *p) {
1143 stack_start = *p;
1144 } else {
1145 // see if we can get the start_stack field from /proc/self/stat
1146 FILE *fp;
1147 int pid;
1148 char state;
1149 int ppid;
1150 int pgrp;
1151 int session;
1152 int nr;
1153 int tpgrp;
1154 unsigned long flags;
1155 unsigned long minflt;
1156 unsigned long cminflt;
1157 unsigned long majflt;
1158 unsigned long cmajflt;
1159 unsigned long utime;
1160 unsigned long stime;
1161 long cutime;
1162 long cstime;
1163 long prio;
1164 long nice;
1165 long junk;
1166 long it_real;
1167 uintptr_t start;
1168 uintptr_t vsize;
1169 intptr_t rss;
1170 uintptr_t rsslim;
1171 uintptr_t scodes;
1172 uintptr_t ecode;
1173 int i;
1174
1175 // Figure what the primordial thread stack base is. Code is inspired
1176 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1177 // followed by command name surrounded by parentheses, state, etc.
1178 char stat[2048];
1179 int statlen;
1180
1181 fp = fopen("/proc/self/stat", "r");
1182 if (fp) {
1183 statlen = fread(stat, 1, 2047, fp);
1184 stat[statlen] = '\0';
1185 fclose(fp);
1186
1187 // Skip pid and the command string. Note that we could be dealing with
1188 // weird command names, e.g. user could decide to rename java launcher
1189 // to "java 1.4.2 :)", then the stat file would look like
1190 // 1234 (java 1.4.2 :)) R ... ...
1191 // We don't really need to know the command string, just find the last
1192 // occurrence of ")" and then start parsing from there. See bug 4726580.
1193 char * s = strrchr(stat, ')');
1194
1195 i = 0;
1196 if (s) {
1197 // Skip blank chars
1198 do { s++; } while (s && isspace(*s));
1199
1200 #define _UFM UINTX_FORMAT
1201 #define _DFM INTX_FORMAT
1202
1203 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1204 // 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
1205 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,
1206 &state, // 3 %c
1207 &ppid, // 4 %d
1208 &pgrp, // 5 %d
1209 &session, // 6 %d
1210 &nr, // 7 %d
1211 &tpgrp, // 8 %d
1212 &flags, // 9 %lu
1213 &minflt, // 10 %lu
1214 &cminflt, // 11 %lu
1215 &majflt, // 12 %lu
1216 &cmajflt, // 13 %lu
1217 &utime, // 14 %lu
1218 &stime, // 15 %lu
1219 &cutime, // 16 %ld
1220 &cstime, // 17 %ld
1221 &prio, // 18 %ld
1222 &nice, // 19 %ld
1223 &junk, // 20 %ld
1224 &it_real, // 21 %ld
1225 &start, // 22 UINTX_FORMAT
1226 &vsize, // 23 UINTX_FORMAT
1227 &rss, // 24 INTX_FORMAT
1228 &rsslim, // 25 UINTX_FORMAT
1229 &scodes, // 26 UINTX_FORMAT
1230 &ecode, // 27 UINTX_FORMAT
1231 &stack_start); // 28 UINTX_FORMAT
1232 }
1233
1234 #undef _UFM
1235 #undef _DFM
1236
1237 if (i != 28 - 2) {
1238 assert(false, "Bad conversion from /proc/self/stat");
1239 // product mode - assume we are the initial thread, good luck in the
1240 // embedded case.
1241 warning("Can't detect initial thread stack location - bad conversion");
1242 stack_start = (uintptr_t) &rlim;
1243 }
1244 } else {
1245 // For some reason we can't open /proc/self/stat (for example, running on
1246 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1247 // most cases, so don't abort:
1248 warning("Can't detect initial thread stack location - no /proc/self/stat");
1249 stack_start = (uintptr_t) &rlim;
1250 }
1251 }
1252
1253 // Now we have a pointer (stack_start) very close to the stack top, the
1254 // next thing to do is to figure out the exact location of stack top. We
1255 // can find out the virtual memory area that contains stack_start by
1256 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1257 // and its upper limit is the real stack top. (again, this would fail if
1258 // running inside chroot, because /proc may not exist.)
1259
1260 uintptr_t stack_top;
1261 address low, high;
1262 if (find_vma((address)stack_start, &low, &high)) {
1263 // success, "high" is the true stack top. (ignore "low", because initial
1264 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1265 stack_top = (uintptr_t)high;
1266 } else {
1267 // failed, likely because /proc/self/maps does not exist
1268 warning("Can't detect initial thread stack location - find_vma failed");
1269 // best effort: stack_start is normally within a few pages below the real
1270 // stack top, use it as stack top, and reduce stack size so we won't put
1271 // guard page outside stack.
1272 stack_top = stack_start;
1273 stack_size -= 16 * page_size();
1274 }
1275
1276 // stack_top could be partially down the page so align it
1277 stack_top = align_size_up(stack_top, page_size());
1278
1279 if (max_size && stack_size > max_size) {
1280 _initial_thread_stack_size = max_size;
1281 } else {
1282 _initial_thread_stack_size = stack_size;
1283 }
1284
1285 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1286 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1287 }
1288
1289 ////////////////////////////////////////////////////////////////////////////////
1290 // time support
1291
1292 // Time since start-up in seconds to a fine granularity.
1293 // Used by VMSelfDestructTimer and the MemProfiler.
1294 double os::elapsedTime() {
1295
1296 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1297 }
1298
1299 jlong os::elapsed_counter() {
1300 return javaTimeNanos() - initial_time_count;
1301 }
1302
1303 jlong os::elapsed_frequency() {
1304 return NANOSECS_PER_SEC; // nanosecond resolution
1305 }
1306
1307 bool os::supports_vtime() { return true; }
1308 bool os::enable_vtime() { return false; }
1309 bool os::vtime_enabled() { return false; }
1310
1311 double os::elapsedVTime() {
1312 struct rusage usage;
1313 int retval = getrusage(RUSAGE_THREAD, &usage);
1314 if (retval == 0) {
1315 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);
1316 } else {
1317 // better than nothing, but not much
1318 return elapsedTime();
1319 }
1320 }
1321
1322 jlong os::javaTimeMillis() {
1323 timeval time;
1324 int status = gettimeofday(&time, NULL);
1325 assert(status != -1, "linux error");
1326 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1327 }
1328
1329 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1330 timeval time;
1331 int status = gettimeofday(&time, NULL);
1332 assert(status != -1, "linux error");
1333 seconds = jlong(time.tv_sec);
1334 nanos = jlong(time.tv_usec) * 1000;
1335 }
1336
1337
1338 #ifndef CLOCK_MONOTONIC
1339 #define CLOCK_MONOTONIC (1)
1340 #endif
1341
1342 void os::Linux::clock_init() {
1343 // we do dlopen's in this particular order due to bug in linux
1344 // dynamical loader (see 6348968) leading to crash on exit
1345 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1346 if (handle == NULL) {
1347 handle = dlopen("librt.so", RTLD_LAZY);
1348 }
1349
1350 if (handle) {
1351 int (*clock_getres_func)(clockid_t, struct timespec*) =
1352 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1353 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1354 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1355 if (clock_getres_func && clock_gettime_func) {
1356 // See if monotonic clock is supported by the kernel. Note that some
1357 // early implementations simply return kernel jiffies (updated every
1358 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1359 // for nano time (though the monotonic property is still nice to have).
1360 // It's fixed in newer kernels, however clock_getres() still returns
1361 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1362 // resolution for now. Hopefully as people move to new kernels, this
1363 // won't be a problem.
1364 struct timespec res;
1365 struct timespec tp;
1366 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1367 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1368 // yes, monotonic clock is supported
1369 _clock_gettime = clock_gettime_func;
1370 return;
1371 } else {
1372 // close librt if there is no monotonic clock
1373 dlclose(handle);
1374 }
1375 }
1376 }
1377 warning("No monotonic clock was available - timed services may " \
1378 "be adversely affected if the time-of-day clock changes");
1379 }
1380
1381 #ifndef SYS_clock_getres
1382 #if defined(IA32) || defined(AMD64)
1383 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1384 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1385 #else
1386 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1387 #define sys_clock_getres(x,y) -1
1388 #endif
1389 #else
1390 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1391 #endif
1392
1393 void os::Linux::fast_thread_clock_init() {
1394 if (!UseLinuxPosixThreadCPUClocks) {
1395 return;
1396 }
1397 clockid_t clockid;
1398 struct timespec tp;
1399 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1400 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1401
1402 // Switch to using fast clocks for thread cpu time if
1403 // the sys_clock_getres() returns 0 error code.
1404 // Note, that some kernels may support the current thread
1405 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1406 // returned by the pthread_getcpuclockid().
1407 // If the fast Posix clocks are supported then the sys_clock_getres()
1408 // must return at least tp.tv_sec == 0 which means a resolution
1409 // better than 1 sec. This is extra check for reliability.
1410
1411 if (pthread_getcpuclockid_func &&
1412 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1413 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1414 _supports_fast_thread_cpu_time = true;
1415 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1416 }
1417 }
1418
1419 jlong os::javaTimeNanos() {
1420 if (os::supports_monotonic_clock()) {
1421 struct timespec tp;
1422 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1423 assert(status == 0, "gettime error");
1424 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1425 return result;
1426 } else {
1427 timeval time;
1428 int status = gettimeofday(&time, NULL);
1429 assert(status != -1, "linux error");
1430 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1431 return 1000 * usecs;
1432 }
1433 }
1434
1435 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1436 if (os::supports_monotonic_clock()) {
1437 info_ptr->max_value = ALL_64_BITS;
1438
1439 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1440 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1441 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1442 } else {
1443 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1444 info_ptr->max_value = ALL_64_BITS;
1445
1446 // gettimeofday is a real time clock so it skips
1447 info_ptr->may_skip_backward = true;
1448 info_ptr->may_skip_forward = true;
1449 }
1450
1451 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1452 }
1453
1454 // Return the real, user, and system times in seconds from an
1455 // arbitrary fixed point in the past.
1456 bool os::getTimesSecs(double* process_real_time,
1457 double* process_user_time,
1458 double* process_system_time) {
1459 struct tms ticks;
1460 clock_t real_ticks = times(&ticks);
1461
1462 if (real_ticks == (clock_t) (-1)) {
1463 return false;
1464 } else {
1465 double ticks_per_second = (double) clock_tics_per_sec;
1466 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1467 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1468 *process_real_time = ((double) real_ticks) / ticks_per_second;
1469
1470 return true;
1471 }
1472 }
1473
1474
1475 char * os::local_time_string(char *buf, size_t buflen) {
1476 struct tm t;
1477 time_t long_time;
1478 time(&long_time);
1479 localtime_r(&long_time, &t);
1480 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1481 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1482 t.tm_hour, t.tm_min, t.tm_sec);
1483 return buf;
1484 }
1485
1486 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1487 return localtime_r(clock, res);
1488 }
1489
1490 ////////////////////////////////////////////////////////////////////////////////
1491 // runtime exit support
1492
1493 // Note: os::shutdown() might be called very early during initialization, or
1494 // called from signal handler. Before adding something to os::shutdown(), make
1495 // sure it is async-safe and can handle partially initialized VM.
1496 void os::shutdown() {
1497
1498 // allow PerfMemory to attempt cleanup of any persistent resources
1499 perfMemory_exit();
1500
1501 // needs to remove object in file system
1502 AttachListener::abort();
1503
1504 // flush buffered output, finish log files
1505 ostream_abort();
1506
1507 // Check for abort hook
1508 abort_hook_t abort_hook = Arguments::abort_hook();
1509 if (abort_hook != NULL) {
1510 abort_hook();
1511 }
1512
1513 }
1514
1515 // Note: os::abort() might be called very early during initialization, or
1516 // called from signal handler. Before adding something to os::abort(), make
1517 // sure it is async-safe and can handle partially initialized VM.
1518 void os::abort(bool dump_core) {
1519 os::shutdown();
1520 if (dump_core) {
1521 #ifndef PRODUCT
1522 fdStream out(defaultStream::output_fd());
1523 out.print_raw("Current thread is ");
1524 char buf[16];
1525 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1526 out.print_raw_cr(buf);
1527 out.print_raw_cr("Dumping core ...");
1528 #endif
1529 ::abort(); // dump core
1530 }
1531
1532 ::exit(1);
1533 }
1534
1535 // Die immediately, no exit hook, no abort hook, no cleanup.
1536 void os::die() {
1537 // _exit() on LinuxThreads only kills current thread
1538 ::abort();
1539 }
1540
1541
1542 // This method is a copy of JDK's sysGetLastErrorString
1543 // from src/solaris/hpi/src/system_md.c
1544
1545 size_t os::lasterror(char *buf, size_t len) {
1546 if (errno == 0) return 0;
1547
1548 const char *s = ::strerror(errno);
1549 size_t n = ::strlen(s);
1550 if (n >= len) {
1551 n = len - 1;
1552 }
1553 ::strncpy(buf, s, n);
1554 buf[n] = '\0';
1555 return n;
1556 }
1557
1558 intx os::current_thread_id() { return (intx)pthread_self(); }
1559 int os::current_process_id() {
1560
1561 // Under the old linux thread library, linux gives each thread
1562 // its own process id. Because of this each thread will return
1563 // a different pid if this method were to return the result
1564 // of getpid(2). Linux provides no api that returns the pid
1565 // of the launcher thread for the vm. This implementation
1566 // returns a unique pid, the pid of the launcher thread
1567 // that starts the vm 'process'.
1568
1569 // Under the NPTL, getpid() returns the same pid as the
1570 // launcher thread rather than a unique pid per thread.
1571 // Use gettid() if you want the old pre NPTL behaviour.
1572
1573 // if you are looking for the result of a call to getpid() that
1574 // returns a unique pid for the calling thread, then look at the
1575 // OSThread::thread_id() method in osThread_linux.hpp file
1576
1577 return (int)(_initial_pid ? _initial_pid : getpid());
1578 }
1579
1580 // DLL functions
1581
1582 const char* os::dll_file_extension() { return ".so"; }
1583
1584 // This must be hard coded because it's the system's temporary
1585 // directory not the java application's temp directory, ala java.io.tmpdir.
1586 const char* os::get_temp_directory() { return "/tmp"; }
1587
1588 static bool file_exists(const char* filename) {
1589 struct stat statbuf;
1590 if (filename == NULL || strlen(filename) == 0) {
1591 return false;
1592 }
1593 return os::stat(filename, &statbuf) == 0;
1594 }
1595
1596 bool os::dll_build_name(char* buffer, size_t buflen,
1597 const char* pname, const char* fname) {
1598 bool retval = false;
1599 // Copied from libhpi
1600 const size_t pnamelen = pname ? strlen(pname) : 0;
1601
1602 // Return error on buffer overflow.
1603 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1604 return retval;
1605 }
1606
1607 if (pnamelen == 0) {
1608 snprintf(buffer, buflen, "lib%s.so", fname);
1609 retval = true;
1610 } else if (strchr(pname, *os::path_separator()) != NULL) {
1611 int n;
1612 char** pelements = split_path(pname, &n);
1613 if (pelements == NULL) {
1614 return false;
1615 }
1616 for (int i = 0; i < n; i++) {
1617 // Really shouldn't be NULL, but check can't hurt
1618 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1619 continue; // skip the empty path values
1620 }
1621 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1622 if (file_exists(buffer)) {
1623 retval = true;
1624 break;
1625 }
1626 }
1627 // release the storage
1628 for (int i = 0; i < n; i++) {
1629 if (pelements[i] != NULL) {
1630 FREE_C_HEAP_ARRAY(char, pelements[i]);
1631 }
1632 }
1633 if (pelements != NULL) {
1634 FREE_C_HEAP_ARRAY(char*, pelements);
1635 }
1636 } else {
1637 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1638 retval = true;
1639 }
1640 return retval;
1641 }
1642
1643 // check if addr is inside libjvm.so
1644 bool os::address_is_in_vm(address addr) {
1645 static address libjvm_base_addr;
1646 Dl_info dlinfo;
1647
1648 if (libjvm_base_addr == NULL) {
1649 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1650 libjvm_base_addr = (address)dlinfo.dli_fbase;
1651 }
1652 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1653 }
1654
1655 if (dladdr((void *)addr, &dlinfo) != 0) {
1656 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1657 }
1658
1659 return false;
1660 }
1661
1662 bool os::dll_address_to_function_name(address addr, char *buf,
1663 int buflen, int *offset) {
1664 // buf is not optional, but offset is optional
1665 assert(buf != NULL, "sanity check");
1666
1667 Dl_info dlinfo;
1668
1669 if (dladdr((void*)addr, &dlinfo) != 0) {
1670 // see if we have a matching symbol
1671 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1672 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1673 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1674 }
1675 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1676 return true;
1677 }
1678 // no matching symbol so try for just file info
1679 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1680 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1681 buf, buflen, offset, dlinfo.dli_fname)) {
1682 return true;
1683 }
1684 }
1685 }
1686
1687 buf[0] = '\0';
1688 if (offset != NULL) *offset = -1;
1689 return false;
1690 }
1691
1692 struct _address_to_library_name {
1693 address addr; // input : memory address
1694 size_t buflen; // size of fname
1695 char* fname; // output: library name
1696 address base; // library base addr
1697 };
1698
1699 static int address_to_library_name_callback(struct dl_phdr_info *info,
1700 size_t size, void *data) {
1701 int i;
1702 bool found = false;
1703 address libbase = NULL;
1704 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1705
1706 // iterate through all loadable segments
1707 for (i = 0; i < info->dlpi_phnum; i++) {
1708 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1709 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1710 // base address of a library is the lowest address of its loaded
1711 // segments.
1712 if (libbase == NULL || libbase > segbase) {
1713 libbase = segbase;
1714 }
1715 // see if 'addr' is within current segment
1716 if (segbase <= d->addr &&
1717 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1718 found = true;
1719 }
1720 }
1721 }
1722
1723 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1724 // so dll_address_to_library_name() can fall through to use dladdr() which
1725 // can figure out executable name from argv[0].
1726 if (found && info->dlpi_name && info->dlpi_name[0]) {
1727 d->base = libbase;
1728 if (d->fname) {
1729 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1730 }
1731 return 1;
1732 }
1733 return 0;
1734 }
1735
1736 bool os::dll_address_to_library_name(address addr, char* buf,
1737 int buflen, int* offset) {
1738 // buf is not optional, but offset is optional
1739 assert(buf != NULL, "sanity check");
1740
1741 Dl_info dlinfo;
1742 struct _address_to_library_name data;
1743
1744 // There is a bug in old glibc dladdr() implementation that it could resolve
1745 // to wrong library name if the .so file has a base address != NULL. Here
1746 // we iterate through the program headers of all loaded libraries to find
1747 // out which library 'addr' really belongs to. This workaround can be
1748 // removed once the minimum requirement for glibc is moved to 2.3.x.
1749 data.addr = addr;
1750 data.fname = buf;
1751 data.buflen = buflen;
1752 data.base = NULL;
1753 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1754
1755 if (rslt) {
1756 // buf already contains library name
1757 if (offset) *offset = addr - data.base;
1758 return true;
1759 }
1760 if (dladdr((void*)addr, &dlinfo) != 0) {
1761 if (dlinfo.dli_fname != NULL) {
1762 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1763 }
1764 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1765 *offset = addr - (address)dlinfo.dli_fbase;
1766 }
1767 return true;
1768 }
1769
1770 buf[0] = '\0';
1771 if (offset) *offset = -1;
1772 return false;
1773 }
1774
1775 // Loads .dll/.so and
1776 // in case of error it checks if .dll/.so was built for the
1777 // same architecture as Hotspot is running on
1778
1779
1780 // Remember the stack's state. The Linux dynamic linker will change
1781 // the stack to 'executable' at most once, so we must safepoint only once.
1782 bool os::Linux::_stack_is_executable = false;
1783
1784 // VM operation that loads a library. This is necessary if stack protection
1785 // of the Java stacks can be lost during loading the library. If we
1786 // do not stop the Java threads, they can stack overflow before the stacks
1787 // are protected again.
1788 class VM_LinuxDllLoad: public VM_Operation {
1789 private:
1790 const char *_filename;
1791 char *_ebuf;
1792 int _ebuflen;
1793 void *_lib;
1794 public:
1795 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1796 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1797 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1798 void doit() {
1799 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1800 os::Linux::_stack_is_executable = true;
1801 }
1802 void* loaded_library() { return _lib; }
1803 };
1804
1805 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1806 void * result = NULL;
1807 bool load_attempted = false;
1808
1809 // Check whether the library to load might change execution rights
1810 // of the stack. If they are changed, the protection of the stack
1811 // guard pages will be lost. We need a safepoint to fix this.
1812 //
1813 // See Linux man page execstack(8) for more info.
1814 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1815 ElfFile ef(filename);
1816 if (!ef.specifies_noexecstack()) {
1817 if (!is_init_completed()) {
1818 os::Linux::_stack_is_executable = true;
1819 // This is OK - No Java threads have been created yet, and hence no
1820 // stack guard pages to fix.
1821 //
1822 // This should happen only when you are building JDK7 using a very
1823 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1824 //
1825 // Dynamic loader will make all stacks executable after
1826 // this function returns, and will not do that again.
1827 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1828 } else {
1829 warning("You have loaded library %s which might have disabled stack guard. "
1830 "The VM will try to fix the stack guard now.\n"
1831 "It's highly recommended that you fix the library with "
1832 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1833 filename);
1834
1835 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1836 JavaThread *jt = JavaThread::current();
1837 if (jt->thread_state() != _thread_in_native) {
1838 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1839 // that requires ExecStack. Cannot enter safe point. Let's give up.
1840 warning("Unable to fix stack guard. Giving up.");
1841 } else {
1842 if (!LoadExecStackDllInVMThread) {
1843 // This is for the case where the DLL has an static
1844 // constructor function that executes JNI code. We cannot
1845 // load such DLLs in the VMThread.
1846 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1847 }
1848
1849 ThreadInVMfromNative tiv(jt);
1850 debug_only(VMNativeEntryWrapper vew;)
1851
1852 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1853 VMThread::execute(&op);
1854 if (LoadExecStackDllInVMThread) {
1855 result = op.loaded_library();
1856 }
1857 load_attempted = true;
1858 }
1859 }
1860 }
1861 }
1862
1863 if (!load_attempted) {
1864 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1865 }
1866
1867 if (result != NULL) {
1868 // Successful loading
1869 return result;
1870 }
1871
1872 Elf32_Ehdr elf_head;
1873 int diag_msg_max_length=ebuflen-strlen(ebuf);
1874 char* diag_msg_buf=ebuf+strlen(ebuf);
1875
1876 if (diag_msg_max_length==0) {
1877 // No more space in ebuf for additional diagnostics message
1878 return NULL;
1879 }
1880
1881
1882 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1883
1884 if (file_descriptor < 0) {
1885 // Can't open library, report dlerror() message
1886 return NULL;
1887 }
1888
1889 bool failed_to_read_elf_head=
1890 (sizeof(elf_head)!=
1891 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1892
1893 ::close(file_descriptor);
1894 if (failed_to_read_elf_head) {
1895 // file i/o error - report dlerror() msg
1896 return NULL;
1897 }
1898
1899 typedef struct {
1900 Elf32_Half code; // Actual value as defined in elf.h
1901 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1902 char elf_class; // 32 or 64 bit
1903 char endianess; // MSB or LSB
1904 char* name; // String representation
1905 } arch_t;
1906
1907 #ifndef EM_486
1908 #define EM_486 6 /* Intel 80486 */
1909 #endif
1910
1911 static const arch_t arch_array[]={
1912 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1913 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1914 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1915 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1916 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1917 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1918 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1919 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1920 #if defined(VM_LITTLE_ENDIAN)
1921 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1922 #else
1923 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1924 #endif
1925 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1926 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1927 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1928 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1929 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1930 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1931 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1932 };
1933
1934 #if (defined IA32)
1935 static Elf32_Half running_arch_code=EM_386;
1936 #elif (defined AMD64)
1937 static Elf32_Half running_arch_code=EM_X86_64;
1938 #elif (defined IA64)
1939 static Elf32_Half running_arch_code=EM_IA_64;
1940 #elif (defined __sparc) && (defined _LP64)
1941 static Elf32_Half running_arch_code=EM_SPARCV9;
1942 #elif (defined __sparc) && (!defined _LP64)
1943 static Elf32_Half running_arch_code=EM_SPARC;
1944 #elif (defined __powerpc64__)
1945 static Elf32_Half running_arch_code=EM_PPC64;
1946 #elif (defined __powerpc__)
1947 static Elf32_Half running_arch_code=EM_PPC;
1948 #elif (defined ARM)
1949 static Elf32_Half running_arch_code=EM_ARM;
1950 #elif (defined S390)
1951 static Elf32_Half running_arch_code=EM_S390;
1952 #elif (defined ALPHA)
1953 static Elf32_Half running_arch_code=EM_ALPHA;
1954 #elif (defined MIPSEL)
1955 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1956 #elif (defined PARISC)
1957 static Elf32_Half running_arch_code=EM_PARISC;
1958 #elif (defined MIPS)
1959 static Elf32_Half running_arch_code=EM_MIPS;
1960 #elif (defined M68K)
1961 static Elf32_Half running_arch_code=EM_68K;
1962 #else
1963 #error Method os::dll_load requires that one of following is defined:\
1964 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1965 #endif
1966
1967 // Identify compatability class for VM's architecture and library's architecture
1968 // Obtain string descriptions for architectures
1969
1970 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1971 int running_arch_index=-1;
1972
1973 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1974 if (running_arch_code == arch_array[i].code) {
1975 running_arch_index = i;
1976 }
1977 if (lib_arch.code == arch_array[i].code) {
1978 lib_arch.compat_class = arch_array[i].compat_class;
1979 lib_arch.name = arch_array[i].name;
1980 }
1981 }
1982
1983 assert(running_arch_index != -1,
1984 "Didn't find running architecture code (running_arch_code) in arch_array");
1985 if (running_arch_index == -1) {
1986 // Even though running architecture detection failed
1987 // we may still continue with reporting dlerror() message
1988 return NULL;
1989 }
1990
1991 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1992 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1993 return NULL;
1994 }
1995
1996 #ifndef S390
1997 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1998 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1999 return NULL;
2000 }
2001 #endif // !S390
2002
2003 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2004 if (lib_arch.name!=NULL) {
2005 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2006 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2007 lib_arch.name, arch_array[running_arch_index].name);
2008 } else {
2009 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2010 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2011 lib_arch.code,
2012 arch_array[running_arch_index].name);
2013 }
2014 }
2015
2016 return NULL;
2017 }
2018
2019 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2020 int ebuflen) {
2021 void * result = ::dlopen(filename, RTLD_LAZY);
2022 if (result == NULL) {
2023 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2024 ebuf[ebuflen-1] = '\0';
2025 }
2026 return result;
2027 }
2028
2029 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2030 int ebuflen) {
2031 void * result = NULL;
2032 if (LoadExecStackDllInVMThread) {
2033 result = dlopen_helper(filename, ebuf, ebuflen);
2034 }
2035
2036 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2037 // library that requires an executable stack, or which does not have this
2038 // stack attribute set, dlopen changes the stack attribute to executable. The
2039 // read protection of the guard pages gets lost.
2040 //
2041 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2042 // may have been queued at the same time.
2043
2044 if (!_stack_is_executable) {
2045 JavaThread *jt = Threads::first();
2046
2047 while (jt) {
2048 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2049 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2050 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2051 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2052 warning("Attempt to reguard stack yellow zone failed.");
2053 }
2054 }
2055 jt = jt->next();
2056 }
2057 }
2058
2059 return result;
2060 }
2061
2062 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2063 // chances are you might want to run the generated bits against glibc-2.0
2064 // libdl.so, so always use locking for any version of glibc.
2065 //
2066 void* os::dll_lookup(void* handle, const char* name) {
2067 pthread_mutex_lock(&dl_mutex);
2068 void* res = dlsym(handle, name);
2069 pthread_mutex_unlock(&dl_mutex);
2070 return res;
2071 }
2072
2073 void* os::get_default_process_handle() {
2074 return (void*)::dlopen(NULL, RTLD_LAZY);
2075 }
2076
2077 static bool _print_ascii_file(const char* filename, outputStream* st) {
2078 int fd = ::open(filename, O_RDONLY);
2079 if (fd == -1) {
2080 return false;
2081 }
2082
2083 char buf[32];
2084 int bytes;
2085 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2086 st->print_raw(buf, bytes);
2087 }
2088
2089 ::close(fd);
2090
2091 return true;
2092 }
2093
2094 void os::print_dll_info(outputStream *st) {
2095 st->print_cr("Dynamic libraries:");
2096
2097 char fname[32];
2098 pid_t pid = os::Linux::gettid();
2099
2100 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2101
2102 if (!_print_ascii_file(fname, st)) {
2103 st->print("Can not get library information for pid = %d\n", pid);
2104 }
2105 }
2106
2107 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2108 FILE *procmapsFile = NULL;
2109
2110 // Open the procfs maps file for the current process
2111 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2112 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2113 char line[PATH_MAX + 100];
2114
2115 // Read line by line from 'file'
2116 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2117 u8 base, top, offset, inode;
2118 char permissions[5];
2119 char device[6];
2120 char name[PATH_MAX + 1];
2121
2122 // Parse fields from line
2123 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2124
2125 // Filter by device id '00:00' so that we only get file system mapped files.
2126 if (strcmp(device, "00:00") != 0) {
2127
2128 // Call callback with the fields of interest
2129 if(callback(name, (address)base, (address)top, param)) {
2130 // Oops abort, callback aborted
2131 fclose(procmapsFile);
2132 return 1;
2133 }
2134 }
2135 }
2136 fclose(procmapsFile);
2137 }
2138 return 0;
2139 }
2140
2141 void os::print_os_info_brief(outputStream* st) {
2142 os::Linux::print_distro_info(st);
2143
2144 os::Posix::print_uname_info(st);
2145
2146 os::Linux::print_libversion_info(st);
2147
2148 }
2149
2150 void os::print_os_info(outputStream* st) {
2151 st->print("OS:");
2152
2153 os::Linux::print_distro_info(st);
2154
2155 os::Posix::print_uname_info(st);
2156
2157 // Print warning if unsafe chroot environment detected
2158 if (unsafe_chroot_detected) {
2159 st->print("WARNING!! ");
2160 st->print_cr("%s", unstable_chroot_error);
2161 }
2162
2163 os::Linux::print_libversion_info(st);
2164
2165 os::Posix::print_rlimit_info(st);
2166
2167 os::Posix::print_load_average(st);
2168
2169 os::Linux::print_full_memory_info(st);
2170 }
2171
2172 // Try to identify popular distros.
2173 // Most Linux distributions have a /etc/XXX-release file, which contains
2174 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2175 // file that also contains the OS version string. Some have more than one
2176 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2177 // /etc/redhat-release.), so the order is important.
2178 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2179 // their own specific XXX-release file as well as a redhat-release file.
2180 // Because of this the XXX-release file needs to be searched for before the
2181 // redhat-release file.
2182 // Since Red Hat has a lsb-release file that is not very descriptive the
2183 // search for redhat-release needs to be before lsb-release.
2184 // Since the lsb-release file is the new standard it needs to be searched
2185 // before the older style release files.
2186 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2187 // next to last resort. The os-release file is a new standard that contains
2188 // distribution information and the system-release file seems to be an old
2189 // standard that has been replaced by the lsb-release and os-release files.
2190 // Searching for the debian_version file is the last resort. It contains
2191 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2192 // "Debian " is printed before the contents of the debian_version file.
2193 void os::Linux::print_distro_info(outputStream* st) {
2194 if (!_print_ascii_file("/etc/oracle-release", st) &&
2195 !_print_ascii_file("/etc/mandriva-release", st) &&
2196 !_print_ascii_file("/etc/mandrake-release", st) &&
2197 !_print_ascii_file("/etc/sun-release", st) &&
2198 !_print_ascii_file("/etc/redhat-release", st) &&
2199 !_print_ascii_file("/etc/lsb-release", st) &&
2200 !_print_ascii_file("/etc/SuSE-release", st) &&
2201 !_print_ascii_file("/etc/turbolinux-release", st) &&
2202 !_print_ascii_file("/etc/gentoo-release", st) &&
2203 !_print_ascii_file("/etc/ltib-release", st) &&
2204 !_print_ascii_file("/etc/angstrom-version", st) &&
2205 !_print_ascii_file("/etc/system-release", st) &&
2206 !_print_ascii_file("/etc/os-release", st)) {
2207
2208 if (file_exists("/etc/debian_version")) {
2209 st->print("Debian ");
2210 _print_ascii_file("/etc/debian_version", st);
2211 } else {
2212 st->print("Linux");
2213 }
2214 }
2215 st->cr();
2216 }
2217
2218 void os::Linux::print_libversion_info(outputStream* st) {
2219 // libc, pthread
2220 st->print("libc:");
2221 st->print("%s ", os::Linux::glibc_version());
2222 st->print("%s ", os::Linux::libpthread_version());
2223 if (os::Linux::is_LinuxThreads()) {
2224 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2225 }
2226 st->cr();
2227 }
2228
2229 void os::Linux::print_full_memory_info(outputStream* st) {
2230 st->print("\n/proc/meminfo:\n");
2231 _print_ascii_file("/proc/meminfo", st);
2232 st->cr();
2233 }
2234
2235 void os::print_memory_info(outputStream* st) {
2236
2237 st->print("Memory:");
2238 st->print(" %dk page", os::vm_page_size()>>10);
2239
2240 // values in struct sysinfo are "unsigned long"
2241 struct sysinfo si;
2242 sysinfo(&si);
2243
2244 st->print(", physical " UINT64_FORMAT "k",
2245 os::physical_memory() >> 10);
2246 st->print("(" UINT64_FORMAT "k free)",
2247 os::available_memory() >> 10);
2248 st->print(", swap " UINT64_FORMAT "k",
2249 ((jlong)si.totalswap * si.mem_unit) >> 10);
2250 st->print("(" UINT64_FORMAT "k free)",
2251 ((jlong)si.freeswap * si.mem_unit) >> 10);
2252 st->cr();
2253 }
2254
2255 void os::pd_print_cpu_info(outputStream* st) {
2256 st->print("\n/proc/cpuinfo:\n");
2257 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2258 st->print(" <Not Available>");
2259 }
2260 st->cr();
2261 }
2262
2263 void os::print_siginfo(outputStream* st, void* siginfo) {
2264 const siginfo_t* si = (const siginfo_t*)siginfo;
2265
2266 os::Posix::print_siginfo_brief(st, si);
2267 #if INCLUDE_CDS
2268 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2269 UseSharedSpaces) {
2270 FileMapInfo* mapinfo = FileMapInfo::current_info();
2271 if (mapinfo->is_in_shared_space(si->si_addr)) {
2272 st->print("\n\nError accessing class data sharing archive." \
2273 " Mapped file inaccessible during execution, " \
2274 " possible disk/network problem.");
2275 }
2276 }
2277 #endif
2278 st->cr();
2279 }
2280
2281
2282 static void print_signal_handler(outputStream* st, int sig,
2283 char* buf, size_t buflen);
2284
2285 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2286 st->print_cr("Signal Handlers:");
2287 print_signal_handler(st, SIGSEGV, buf, buflen);
2288 print_signal_handler(st, SIGBUS , buf, buflen);
2289 print_signal_handler(st, SIGFPE , buf, buflen);
2290 print_signal_handler(st, SIGPIPE, buf, buflen);
2291 print_signal_handler(st, SIGXFSZ, buf, buflen);
2292 print_signal_handler(st, SIGILL , buf, buflen);
2293 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2294 print_signal_handler(st, SR_signum, buf, buflen);
2295 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2296 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2297 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2298 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2299 #if defined(PPC64)
2300 print_signal_handler(st, SIGTRAP, buf, buflen);
2301 #endif
2302 }
2303
2304 static char saved_jvm_path[MAXPATHLEN] = {0};
2305
2306 // Find the full path to the current module, libjvm.so
2307 void os::jvm_path(char *buf, jint buflen) {
2308 // Error checking.
2309 if (buflen < MAXPATHLEN) {
2310 assert(false, "must use a large-enough buffer");
2311 buf[0] = '\0';
2312 return;
2313 }
2314 // Lazy resolve the path to current module.
2315 if (saved_jvm_path[0] != 0) {
2316 strcpy(buf, saved_jvm_path);
2317 return;
2318 }
2319
2320 char dli_fname[MAXPATHLEN];
2321 bool ret = dll_address_to_library_name(
2322 CAST_FROM_FN_PTR(address, os::jvm_path),
2323 dli_fname, sizeof(dli_fname), NULL);
2324 assert(ret, "cannot locate libjvm");
2325 char *rp = NULL;
2326 if (ret && dli_fname[0] != '\0') {
2327 rp = realpath(dli_fname, buf);
2328 }
2329 if (rp == NULL) {
2330 return;
2331 }
2332
2333 if (Arguments::sun_java_launcher_is_altjvm()) {
2334 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2335 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2336 // If "/jre/lib/" appears at the right place in the string, then
2337 // assume we are installed in a JDK and we're done. Otherwise, check
2338 // for a JAVA_HOME environment variable and fix up the path so it
2339 // looks like libjvm.so is installed there (append a fake suffix
2340 // hotspot/libjvm.so).
2341 const char *p = buf + strlen(buf) - 1;
2342 for (int count = 0; p > buf && count < 5; ++count) {
2343 for (--p; p > buf && *p != '/'; --p)
2344 /* empty */ ;
2345 }
2346
2347 if (strncmp(p, "/jre/lib/", 9) != 0) {
2348 // Look for JAVA_HOME in the environment.
2349 char* java_home_var = ::getenv("JAVA_HOME");
2350 if (java_home_var != NULL && java_home_var[0] != 0) {
2351 char* jrelib_p;
2352 int len;
2353
2354 // Check the current module name "libjvm.so".
2355 p = strrchr(buf, '/');
2356 if (p == NULL) {
2357 return;
2358 }
2359 assert(strstr(p, "/libjvm") == p, "invalid library name");
2360
2361 rp = realpath(java_home_var, buf);
2362 if (rp == NULL) {
2363 return;
2364 }
2365
2366 // determine if this is a legacy image or modules image
2367 // modules image doesn't have "jre" subdirectory
2368 len = strlen(buf);
2369 assert(len < buflen, "Ran out of buffer room");
2370 jrelib_p = buf + len;
2371 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2372 if (0 != access(buf, F_OK)) {
2373 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2374 }
2375
2376 if (0 == access(buf, F_OK)) {
2377 // Use current module name "libjvm.so"
2378 len = strlen(buf);
2379 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2380 } else {
2381 // Go back to path of .so
2382 rp = realpath(dli_fname, buf);
2383 if (rp == NULL) {
2384 return;
2385 }
2386 }
2387 }
2388 }
2389 }
2390
2391 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2392 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2393 }
2394
2395 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2396 // no prefix required, not even "_"
2397 }
2398
2399 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2400 // no suffix required
2401 }
2402
2403 ////////////////////////////////////////////////////////////////////////////////
2404 // sun.misc.Signal support
2405
2406 static volatile jint sigint_count = 0;
2407
2408 static void UserHandler(int sig, void *siginfo, void *context) {
2409 // 4511530 - sem_post is serialized and handled by the manager thread. When
2410 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2411 // don't want to flood the manager thread with sem_post requests.
2412 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2413 return;
2414 }
2415
2416 // Ctrl-C is pressed during error reporting, likely because the error
2417 // handler fails to abort. Let VM die immediately.
2418 if (sig == SIGINT && is_error_reported()) {
2419 os::die();
2420 }
2421
2422 os::signal_notify(sig);
2423 }
2424
2425 void* os::user_handler() {
2426 return CAST_FROM_FN_PTR(void*, UserHandler);
2427 }
2428
2429 class Semaphore : public StackObj {
2430 public:
2431 Semaphore();
2432 ~Semaphore();
2433 void signal();
2434 void wait();
2435 bool trywait();
2436 bool timedwait(unsigned int sec, int nsec);
2437 private:
2438 sem_t _semaphore;
2439 };
2440
2441 Semaphore::Semaphore() {
2442 sem_init(&_semaphore, 0, 0);
2443 }
2444
2445 Semaphore::~Semaphore() {
2446 sem_destroy(&_semaphore);
2447 }
2448
2449 void Semaphore::signal() {
2450 sem_post(&_semaphore);
2451 }
2452
2453 void Semaphore::wait() {
2454 sem_wait(&_semaphore);
2455 }
2456
2457 bool Semaphore::trywait() {
2458 return sem_trywait(&_semaphore) == 0;
2459 }
2460
2461 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2462
2463 struct timespec ts;
2464 // Semaphore's are always associated with CLOCK_REALTIME
2465 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2466 // see unpackTime for discussion on overflow checking
2467 if (sec >= MAX_SECS) {
2468 ts.tv_sec += MAX_SECS;
2469 ts.tv_nsec = 0;
2470 } else {
2471 ts.tv_sec += sec;
2472 ts.tv_nsec += nsec;
2473 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2474 ts.tv_nsec -= NANOSECS_PER_SEC;
2475 ++ts.tv_sec; // note: this must be <= max_secs
2476 }
2477 }
2478
2479 while (1) {
2480 int result = sem_timedwait(&_semaphore, &ts);
2481 if (result == 0) {
2482 return true;
2483 } else if (errno == EINTR) {
2484 continue;
2485 } else if (errno == ETIMEDOUT) {
2486 return false;
2487 } else {
2488 return false;
2489 }
2490 }
2491 }
2492
2493 extern "C" {
2494 typedef void (*sa_handler_t)(int);
2495 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2496 }
2497
2498 void* os::signal(int signal_number, void* handler) {
2499 struct sigaction sigAct, oldSigAct;
2500
2501 sigfillset(&(sigAct.sa_mask));
2502 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2503 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2504
2505 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2506 // -1 means registration failed
2507 return (void *)-1;
2508 }
2509
2510 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2511 }
2512
2513 void os::signal_raise(int signal_number) {
2514 ::raise(signal_number);
2515 }
2516
2517 // The following code is moved from os.cpp for making this
2518 // code platform specific, which it is by its very nature.
2519
2520 // Will be modified when max signal is changed to be dynamic
2521 int os::sigexitnum_pd() {
2522 return NSIG;
2523 }
2524
2525 // a counter for each possible signal value
2526 static volatile jint pending_signals[NSIG+1] = { 0 };
2527
2528 // Linux(POSIX) specific hand shaking semaphore.
2529 static sem_t sig_sem;
2530 static Semaphore sr_semaphore;
2531
2532 void os::signal_init_pd() {
2533 // Initialize signal structures
2534 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2535
2536 // Initialize signal semaphore
2537 ::sem_init(&sig_sem, 0, 0);
2538 }
2539
2540 void os::signal_notify(int sig) {
2541 Atomic::inc(&pending_signals[sig]);
2542 ::sem_post(&sig_sem);
2543 }
2544
2545 static int check_pending_signals(bool wait) {
2546 Atomic::store(0, &sigint_count);
2547 for (;;) {
2548 for (int i = 0; i < NSIG + 1; i++) {
2549 jint n = pending_signals[i];
2550 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2551 return i;
2552 }
2553 }
2554 if (!wait) {
2555 return -1;
2556 }
2557 JavaThread *thread = JavaThread::current();
2558 ThreadBlockInVM tbivm(thread);
2559
2560 bool threadIsSuspended;
2561 do {
2562 thread->set_suspend_equivalent();
2563 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2564 ::sem_wait(&sig_sem);
2565
2566 // were we externally suspended while we were waiting?
2567 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2568 if (threadIsSuspended) {
2569 // The semaphore has been incremented, but while we were waiting
2570 // another thread suspended us. We don't want to continue running
2571 // while suspended because that would surprise the thread that
2572 // suspended us.
2573 ::sem_post(&sig_sem);
2574
2575 thread->java_suspend_self();
2576 }
2577 } while (threadIsSuspended);
2578 }
2579 }
2580
2581 int os::signal_lookup() {
2582 return check_pending_signals(false);
2583 }
2584
2585 int os::signal_wait() {
2586 return check_pending_signals(true);
2587 }
2588
2589 ////////////////////////////////////////////////////////////////////////////////
2590 // Virtual Memory
2591
2592 int os::vm_page_size() {
2593 // Seems redundant as all get out
2594 assert(os::Linux::page_size() != -1, "must call os::init");
2595 return os::Linux::page_size();
2596 }
2597
2598 // Solaris allocates memory by pages.
2599 int os::vm_allocation_granularity() {
2600 assert(os::Linux::page_size() != -1, "must call os::init");
2601 return os::Linux::page_size();
2602 }
2603
2604 // Rationale behind this function:
2605 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2606 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2607 // samples for JITted code. Here we create private executable mapping over the code cache
2608 // and then we can use standard (well, almost, as mapping can change) way to provide
2609 // info for the reporting script by storing timestamp and location of symbol
2610 void linux_wrap_code(char* base, size_t size) {
2611 static volatile jint cnt = 0;
2612
2613 if (!UseOprofile) {
2614 return;
2615 }
2616
2617 char buf[PATH_MAX+1];
2618 int num = Atomic::add(1, &cnt);
2619
2620 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2621 os::get_temp_directory(), os::current_process_id(), num);
2622 unlink(buf);
2623
2624 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2625
2626 if (fd != -1) {
2627 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2628 if (rv != (off_t)-1) {
2629 if (::write(fd, "", 1) == 1) {
2630 mmap(base, size,
2631 PROT_READ|PROT_WRITE|PROT_EXEC,
2632 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2633 }
2634 }
2635 ::close(fd);
2636 unlink(buf);
2637 }
2638 }
2639
2640 static bool recoverable_mmap_error(int err) {
2641 // See if the error is one we can let the caller handle. This
2642 // list of errno values comes from JBS-6843484. I can't find a
2643 // Linux man page that documents this specific set of errno
2644 // values so while this list currently matches Solaris, it may
2645 // change as we gain experience with this failure mode.
2646 switch (err) {
2647 case EBADF:
2648 case EINVAL:
2649 case ENOTSUP:
2650 // let the caller deal with these errors
2651 return true;
2652
2653 default:
2654 // Any remaining errors on this OS can cause our reserved mapping
2655 // to be lost. That can cause confusion where different data
2656 // structures think they have the same memory mapped. The worst
2657 // scenario is if both the VM and a library think they have the
2658 // same memory mapped.
2659 return false;
2660 }
2661 }
2662
2663 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2664 int err) {
2665 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2666 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2667 strerror(err), err);
2668 }
2669
2670 static void warn_fail_commit_memory(char* addr, size_t size,
2671 size_t alignment_hint, bool exec,
2672 int err) {
2673 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2674 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2675 alignment_hint, exec, strerror(err), err);
2676 }
2677
2678 // NOTE: Linux kernel does not really reserve the pages for us.
2679 // All it does is to check if there are enough free pages
2680 // left at the time of mmap(). This could be a potential
2681 // problem.
2682 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2683 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2684 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2685 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2686 if (res != (uintptr_t) MAP_FAILED) {
2687 if (UseNUMAInterleaving) {
2688 numa_make_global(addr, size);
2689 }
2690 return 0;
2691 }
2692
2693 int err = errno; // save errno from mmap() call above
2694
2695 if (!recoverable_mmap_error(err)) {
2696 warn_fail_commit_memory(addr, size, exec, err);
2697 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2698 }
2699
2700 return err;
2701 }
2702
2703 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2704 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2705 }
2706
2707 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2708 const char* mesg) {
2709 assert(mesg != NULL, "mesg must be specified");
2710 int err = os::Linux::commit_memory_impl(addr, size, exec);
2711 if (err != 0) {
2712 // the caller wants all commit errors to exit with the specified mesg:
2713 warn_fail_commit_memory(addr, size, exec, err);
2714 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2715 }
2716 }
2717
2718 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2719 #ifndef MAP_HUGETLB
2720 #define MAP_HUGETLB 0x40000
2721 #endif
2722
2723 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2724 #ifndef MADV_HUGEPAGE
2725 #define MADV_HUGEPAGE 14
2726 #endif
2727
2728 int os::Linux::commit_memory_impl(char* addr, size_t size,
2729 size_t alignment_hint, bool exec) {
2730 int err = os::Linux::commit_memory_impl(addr, size, exec);
2731 if (err == 0) {
2732 realign_memory(addr, size, alignment_hint);
2733 }
2734 return err;
2735 }
2736
2737 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2738 bool exec) {
2739 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2740 }
2741
2742 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2743 size_t alignment_hint, bool exec,
2744 const char* mesg) {
2745 assert(mesg != NULL, "mesg must be specified");
2746 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2747 if (err != 0) {
2748 // the caller wants all commit errors to exit with the specified mesg:
2749 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2750 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2751 }
2752 }
2753
2754 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2755 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2756 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2757 // be supported or the memory may already be backed by huge pages.
2758 ::madvise(addr, bytes, MADV_HUGEPAGE);
2759 }
2760 }
2761
2762 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2763 // This method works by doing an mmap over an existing mmaping and effectively discarding
2764 // the existing pages. However it won't work for SHM-based large pages that cannot be
2765 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2766 // small pages on top of the SHM segment. This method always works for small pages, so we
2767 // allow that in any case.
2768 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2769 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2770 }
2771 }
2772
2773 void os::numa_make_global(char *addr, size_t bytes) {
2774 Linux::numa_interleave_memory(addr, bytes);
2775 }
2776
2777 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2778 // bind policy to MPOL_PREFERRED for the current thread.
2779 #define USE_MPOL_PREFERRED 0
2780
2781 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2782 // To make NUMA and large pages more robust when both enabled, we need to ease
2783 // the requirements on where the memory should be allocated. MPOL_BIND is the
2784 // default policy and it will force memory to be allocated on the specified
2785 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2786 // the specified node, but will not force it. Using this policy will prevent
2787 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2788 // free large pages.
2789 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2790 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2791 }
2792
2793 bool os::numa_topology_changed() { return false; }
2794
2795 size_t os::numa_get_groups_num() {
2796 int max_node = Linux::numa_max_node();
2797 return max_node > 0 ? max_node + 1 : 1;
2798 }
2799
2800 int os::numa_get_group_id() {
2801 int cpu_id = Linux::sched_getcpu();
2802 if (cpu_id != -1) {
2803 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2804 if (lgrp_id != -1) {
2805 return lgrp_id;
2806 }
2807 }
2808 return 0;
2809 }
2810
2811 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2812 for (size_t i = 0; i < size; i++) {
2813 ids[i] = i;
2814 }
2815 return size;
2816 }
2817
2818 bool os::get_page_info(char *start, page_info* info) {
2819 return false;
2820 }
2821
2822 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2823 page_info* page_found) {
2824 return end;
2825 }
2826
2827
2828 int os::Linux::sched_getcpu_syscall(void) {
2829 unsigned int cpu;
2830 int retval = -1;
2831
2832 #if defined(IA32)
2833 #ifndef SYS_getcpu
2834 #define SYS_getcpu 318
2835 #endif
2836 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2837 #elif defined(AMD64)
2838 // Unfortunately we have to bring all these macros here from vsyscall.h
2839 // to be able to compile on old linuxes.
2840 #define __NR_vgetcpu 2
2841 #define VSYSCALL_START (-10UL << 20)
2842 #define VSYSCALL_SIZE 1024
2843 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2844 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2845 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2846 retval = vgetcpu(&cpu, NULL, NULL);
2847 #endif
2848
2849 return (retval == -1) ? retval : cpu;
2850 }
2851
2852 // Something to do with the numa-aware allocator needs these symbols
2853 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2854 extern "C" JNIEXPORT void numa_error(char *where) { }
2855 extern "C" JNIEXPORT int fork1() { return fork(); }
2856
2857
2858 // If we are running with libnuma version > 2, then we should
2859 // be trying to use symbols with versions 1.1
2860 // If we are running with earlier version, which did not have symbol versions,
2861 // we should use the base version.
2862 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2863 void *f = dlvsym(handle, name, "libnuma_1.1");
2864 if (f == NULL) {
2865 f = dlsym(handle, name);
2866 }
2867 return f;
2868 }
2869
2870 bool os::Linux::libnuma_init() {
2871 // sched_getcpu() should be in libc.
2872 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2873 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2874
2875 // If it's not, try a direct syscall.
2876 if (sched_getcpu() == -1) {
2877 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2878 (void*)&sched_getcpu_syscall));
2879 }
2880
2881 if (sched_getcpu() != -1) { // Does it work?
2882 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2883 if (handle != NULL) {
2884 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2885 libnuma_dlsym(handle, "numa_node_to_cpus")));
2886 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2887 libnuma_dlsym(handle, "numa_max_node")));
2888 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2889 libnuma_dlsym(handle, "numa_available")));
2890 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2891 libnuma_dlsym(handle, "numa_tonode_memory")));
2892 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2893 libnuma_dlsym(handle, "numa_interleave_memory")));
2894 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2895 libnuma_dlsym(handle, "numa_set_bind_policy")));
2896
2897
2898 if (numa_available() != -1) {
2899 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2900 // Create a cpu -> node mapping
2901 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2902 rebuild_cpu_to_node_map();
2903 return true;
2904 }
2905 }
2906 }
2907 return false;
2908 }
2909
2910 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2911 // The table is later used in get_node_by_cpu().
2912 void os::Linux::rebuild_cpu_to_node_map() {
2913 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2914 // in libnuma (possible values are starting from 16,
2915 // and continuing up with every other power of 2, but less
2916 // than the maximum number of CPUs supported by kernel), and
2917 // is a subject to change (in libnuma version 2 the requirements
2918 // are more reasonable) we'll just hardcode the number they use
2919 // in the library.
2920 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2921
2922 size_t cpu_num = os::active_processor_count();
2923 size_t cpu_map_size = NCPUS / BitsPerCLong;
2924 size_t cpu_map_valid_size =
2925 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2926
2927 cpu_to_node()->clear();
2928 cpu_to_node()->at_grow(cpu_num - 1);
2929 size_t node_num = numa_get_groups_num();
2930
2931 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2932 for (size_t i = 0; i < node_num; i++) {
2933 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2934 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2935 if (cpu_map[j] != 0) {
2936 for (size_t k = 0; k < BitsPerCLong; k++) {
2937 if (cpu_map[j] & (1UL << k)) {
2938 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2939 }
2940 }
2941 }
2942 }
2943 }
2944 }
2945 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2946 }
2947
2948 int os::Linux::get_node_by_cpu(int cpu_id) {
2949 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2950 return cpu_to_node()->at(cpu_id);
2951 }
2952 return -1;
2953 }
2954
2955 GrowableArray<int>* os::Linux::_cpu_to_node;
2956 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2957 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2958 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2959 os::Linux::numa_available_func_t os::Linux::_numa_available;
2960 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2961 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2962 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2963 unsigned long* os::Linux::_numa_all_nodes;
2964
2965 bool os::pd_uncommit_memory(char* addr, size_t size) {
2966 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2967 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2968 return res != (uintptr_t) MAP_FAILED;
2969 }
2970
2971 static address get_stack_commited_bottom(address bottom, size_t size) {
2972 address nbot = bottom;
2973 address ntop = bottom + size;
2974
2975 size_t page_sz = os::vm_page_size();
2976 unsigned pages = size / page_sz;
2977
2978 unsigned char vec[1];
2979 unsigned imin = 1, imax = pages + 1, imid;
2980 int mincore_return_value = 0;
2981
2982 assert(imin <= imax, "Unexpected page size");
2983
2984 while (imin < imax) {
2985 imid = (imax + imin) / 2;
2986 nbot = ntop - (imid * page_sz);
2987
2988 // Use a trick with mincore to check whether the page is mapped or not.
2989 // mincore sets vec to 1 if page resides in memory and to 0 if page
2990 // is swapped output but if page we are asking for is unmapped
2991 // it returns -1,ENOMEM
2992 mincore_return_value = mincore(nbot, page_sz, vec);
2993
2994 if (mincore_return_value == -1) {
2995 // Page is not mapped go up
2996 // to find first mapped page
2997 if (errno != EAGAIN) {
2998 assert(errno == ENOMEM, "Unexpected mincore errno");
2999 imax = imid;
3000 }
3001 } else {
3002 // Page is mapped go down
3003 // to find first not mapped page
3004 imin = imid + 1;
3005 }
3006 }
3007
3008 nbot = nbot + page_sz;
3009
3010 // Adjust stack bottom one page up if last checked page is not mapped
3011 if (mincore_return_value == -1) {
3012 nbot = nbot + page_sz;
3013 }
3014
3015 return nbot;
3016 }
3017
3018
3019 // Linux uses a growable mapping for the stack, and if the mapping for
3020 // the stack guard pages is not removed when we detach a thread the
3021 // stack cannot grow beyond the pages where the stack guard was
3022 // mapped. If at some point later in the process the stack expands to
3023 // that point, the Linux kernel cannot expand the stack any further
3024 // because the guard pages are in the way, and a segfault occurs.
3025 //
3026 // However, it's essential not to split the stack region by unmapping
3027 // a region (leaving a hole) that's already part of the stack mapping,
3028 // so if the stack mapping has already grown beyond the guard pages at
3029 // the time we create them, we have to truncate the stack mapping.
3030 // So, we need to know the extent of the stack mapping when
3031 // create_stack_guard_pages() is called.
3032
3033 // We only need this for stacks that are growable: at the time of
3034 // writing thread stacks don't use growable mappings (i.e. those
3035 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3036 // only applies to the main thread.
3037
3038 // If the (growable) stack mapping already extends beyond the point
3039 // where we're going to put our guard pages, truncate the mapping at
3040 // that point by munmap()ping it. This ensures that when we later
3041 // munmap() the guard pages we don't leave a hole in the stack
3042 // mapping. This only affects the main/initial thread
3043
3044 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3045 if (os::Linux::is_initial_thread()) {
3046 // As we manually grow stack up to bottom inside create_attached_thread(),
3047 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3048 // we don't need to do anything special.
3049 // Check it first, before calling heavy function.
3050 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3051 unsigned char vec[1];
3052
3053 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3054 // Fallback to slow path on all errors, including EAGAIN
3055 stack_extent = (uintptr_t) get_stack_commited_bottom(
3056 os::Linux::initial_thread_stack_bottom(),
3057 (size_t)addr - stack_extent);
3058 }
3059
3060 if (stack_extent < (uintptr_t)addr) {
3061 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3062 }
3063 }
3064
3065 return os::commit_memory(addr, size, !ExecMem);
3066 }
3067
3068 // If this is a growable mapping, remove the guard pages entirely by
3069 // munmap()ping them. If not, just call uncommit_memory(). This only
3070 // affects the main/initial thread, but guard against future OS changes
3071 // It's safe to always unmap guard pages for initial thread because we
3072 // always place it right after end of the mapped region
3073
3074 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3075 uintptr_t stack_extent, stack_base;
3076
3077 if (os::Linux::is_initial_thread()) {
3078 return ::munmap(addr, size) == 0;
3079 }
3080
3081 return os::uncommit_memory(addr, size);
3082 }
3083
3084 static address _highest_vm_reserved_address = NULL;
3085
3086 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3087 // at 'requested_addr'. If there are existing memory mappings at the same
3088 // location, however, they will be overwritten. If 'fixed' is false,
3089 // 'requested_addr' is only treated as a hint, the return value may or
3090 // may not start from the requested address. Unlike Linux mmap(), this
3091 // function returns NULL to indicate failure.
3092 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3093 char * addr;
3094 int flags;
3095
3096 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3097 if (fixed) {
3098 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3099 flags |= MAP_FIXED;
3100 }
3101
3102 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3103 // touch an uncommitted page. Otherwise, the read/write might
3104 // succeed if we have enough swap space to back the physical page.
3105 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3106 flags, -1, 0);
3107
3108 if (addr != MAP_FAILED) {
3109 // anon_mmap() should only get called during VM initialization,
3110 // don't need lock (actually we can skip locking even it can be called
3111 // from multiple threads, because _highest_vm_reserved_address is just a
3112 // hint about the upper limit of non-stack memory regions.)
3113 if ((address)addr + bytes > _highest_vm_reserved_address) {
3114 _highest_vm_reserved_address = (address)addr + bytes;
3115 }
3116 }
3117
3118 return addr == MAP_FAILED ? NULL : addr;
3119 }
3120
3121 // Don't update _highest_vm_reserved_address, because there might be memory
3122 // regions above addr + size. If so, releasing a memory region only creates
3123 // a hole in the address space, it doesn't help prevent heap-stack collision.
3124 //
3125 static int anon_munmap(char * addr, size_t size) {
3126 return ::munmap(addr, size) == 0;
3127 }
3128
3129 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3130 size_t alignment_hint) {
3131 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3132 }
3133
3134 bool os::pd_release_memory(char* addr, size_t size) {
3135 return anon_munmap(addr, size);
3136 }
3137
3138 static address highest_vm_reserved_address() {
3139 return _highest_vm_reserved_address;
3140 }
3141
3142 static bool linux_mprotect(char* addr, size_t size, int prot) {
3143 // Linux wants the mprotect address argument to be page aligned.
3144 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3145
3146 // According to SUSv3, mprotect() should only be used with mappings
3147 // established by mmap(), and mmap() always maps whole pages. Unaligned
3148 // 'addr' likely indicates problem in the VM (e.g. trying to change
3149 // protection of malloc'ed or statically allocated memory). Check the
3150 // caller if you hit this assert.
3151 assert(addr == bottom, "sanity check");
3152
3153 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3154 return ::mprotect(bottom, size, prot) == 0;
3155 }
3156
3157 // Set protections specified
3158 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3159 bool is_committed) {
3160 unsigned int p = 0;
3161 switch (prot) {
3162 case MEM_PROT_NONE: p = PROT_NONE; break;
3163 case MEM_PROT_READ: p = PROT_READ; break;
3164 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3165 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3166 default:
3167 ShouldNotReachHere();
3168 }
3169 // is_committed is unused.
3170 return linux_mprotect(addr, bytes, p);
3171 }
3172
3173 bool os::guard_memory(char* addr, size_t size) {
3174 return linux_mprotect(addr, size, PROT_NONE);
3175 }
3176
3177 bool os::unguard_memory(char* addr, size_t size) {
3178 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3179 }
3180
3181 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3182 size_t page_size) {
3183 bool result = false;
3184 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3185 MAP_ANONYMOUS|MAP_PRIVATE,
3186 -1, 0);
3187 if (p != MAP_FAILED) {
3188 void *aligned_p = align_ptr_up(p, page_size);
3189
3190 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3191
3192 munmap(p, page_size * 2);
3193 }
3194
3195 if (warn && !result) {
3196 warning("TransparentHugePages is not supported by the operating system.");
3197 }
3198
3199 return result;
3200 }
3201
3202 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3203 bool result = false;
3204 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3205 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3206 -1, 0);
3207
3208 if (p != MAP_FAILED) {
3209 // We don't know if this really is a huge page or not.
3210 FILE *fp = fopen("/proc/self/maps", "r");
3211 if (fp) {
3212 while (!feof(fp)) {
3213 char chars[257];
3214 long x = 0;
3215 if (fgets(chars, sizeof(chars), fp)) {
3216 if (sscanf(chars, "%lx-%*x", &x) == 1
3217 && x == (long)p) {
3218 if (strstr (chars, "hugepage")) {
3219 result = true;
3220 break;
3221 }
3222 }
3223 }
3224 }
3225 fclose(fp);
3226 }
3227 munmap(p, page_size);
3228 }
3229
3230 if (warn && !result) {
3231 warning("HugeTLBFS is not supported by the operating system.");
3232 }
3233
3234 return result;
3235 }
3236
3237 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3238 //
3239 // From the coredump_filter documentation:
3240 //
3241 // - (bit 0) anonymous private memory
3242 // - (bit 1) anonymous shared memory
3243 // - (bit 2) file-backed private memory
3244 // - (bit 3) file-backed shared memory
3245 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3246 // effective only if the bit 2 is cleared)
3247 // - (bit 5) hugetlb private memory
3248 // - (bit 6) hugetlb shared memory
3249 //
3250 static void set_coredump_filter(void) {
3251 FILE *f;
3252 long cdm;
3253
3254 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3255 return;
3256 }
3257
3258 if (fscanf(f, "%lx", &cdm) != 1) {
3259 fclose(f);
3260 return;
3261 }
3262
3263 rewind(f);
3264
3265 if ((cdm & LARGEPAGES_BIT) == 0) {
3266 cdm |= LARGEPAGES_BIT;
3267 fprintf(f, "%#lx", cdm);
3268 }
3269
3270 fclose(f);
3271 }
3272
3273 // Large page support
3274
3275 static size_t _large_page_size = 0;
3276
3277 size_t os::Linux::find_large_page_size() {
3278 size_t large_page_size = 0;
3279
3280 // large_page_size on Linux is used to round up heap size. x86 uses either
3281 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3282 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3283 // page as large as 256M.
3284 //
3285 // Here we try to figure out page size by parsing /proc/meminfo and looking
3286 // for a line with the following format:
3287 // Hugepagesize: 2048 kB
3288 //
3289 // If we can't determine the value (e.g. /proc is not mounted, or the text
3290 // format has been changed), we'll use the largest page size supported by
3291 // the processor.
3292
3293 #ifndef ZERO
3294 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3295 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3296 #endif // ZERO
3297
3298 FILE *fp = fopen("/proc/meminfo", "r");
3299 if (fp) {
3300 while (!feof(fp)) {
3301 int x = 0;
3302 char buf[16];
3303 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3304 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3305 large_page_size = x * K;
3306 break;
3307 }
3308 } else {
3309 // skip to next line
3310 for (;;) {
3311 int ch = fgetc(fp);
3312 if (ch == EOF || ch == (int)'\n') break;
3313 }
3314 }
3315 }
3316 fclose(fp);
3317 }
3318
3319 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3320 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3321 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3322 proper_unit_for_byte_size(large_page_size));
3323 }
3324
3325 return large_page_size;
3326 }
3327
3328 size_t os::Linux::setup_large_page_size() {
3329 _large_page_size = Linux::find_large_page_size();
3330 const size_t default_page_size = (size_t)Linux::page_size();
3331 if (_large_page_size > default_page_size) {
3332 _page_sizes[0] = _large_page_size;
3333 _page_sizes[1] = default_page_size;
3334 _page_sizes[2] = 0;
3335 }
3336
3337 return _large_page_size;
3338 }
3339
3340 bool os::Linux::setup_large_page_type(size_t page_size) {
3341 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3342 FLAG_IS_DEFAULT(UseSHM) &&
3343 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3344
3345 // The type of large pages has not been specified by the user.
3346
3347 // Try UseHugeTLBFS and then UseSHM.
3348 UseHugeTLBFS = UseSHM = true;
3349
3350 // Don't try UseTransparentHugePages since there are known
3351 // performance issues with it turned on. This might change in the future.
3352 UseTransparentHugePages = false;
3353 }
3354
3355 if (UseTransparentHugePages) {
3356 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3357 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3358 UseHugeTLBFS = false;
3359 UseSHM = false;
3360 return true;
3361 }
3362 UseTransparentHugePages = false;
3363 }
3364
3365 if (UseHugeTLBFS) {
3366 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3367 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3368 UseSHM = false;
3369 return true;
3370 }
3371 UseHugeTLBFS = false;
3372 }
3373
3374 return UseSHM;
3375 }
3376
3377 void os::large_page_init() {
3378 if (!UseLargePages &&
3379 !UseTransparentHugePages &&
3380 !UseHugeTLBFS &&
3381 !UseSHM) {
3382 // Not using large pages.
3383 return;
3384 }
3385
3386 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3387 // The user explicitly turned off large pages.
3388 // Ignore the rest of the large pages flags.
3389 UseTransparentHugePages = false;
3390 UseHugeTLBFS = false;
3391 UseSHM = false;
3392 return;
3393 }
3394
3395 size_t large_page_size = Linux::setup_large_page_size();
3396 UseLargePages = Linux::setup_large_page_type(large_page_size);
3397
3398 set_coredump_filter();
3399 }
3400
3401 #ifndef SHM_HUGETLB
3402 #define SHM_HUGETLB 04000
3403 #endif
3404
3405 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3406 char* req_addr, bool exec) {
3407 // "exec" is passed in but not used. Creating the shared image for
3408 // the code cache doesn't have an SHM_X executable permission to check.
3409 assert(UseLargePages && UseSHM, "only for SHM large pages");
3410 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3411
3412 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3413 return NULL; // Fallback to small pages.
3414 }
3415
3416 key_t key = IPC_PRIVATE;
3417 char *addr;
3418
3419 bool warn_on_failure = UseLargePages &&
3420 (!FLAG_IS_DEFAULT(UseLargePages) ||
3421 !FLAG_IS_DEFAULT(UseSHM) ||
3422 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3423 char msg[128];
3424
3425 // Create a large shared memory region to attach to based on size.
3426 // Currently, size is the total size of the heap
3427 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3428 if (shmid == -1) {
3429 // Possible reasons for shmget failure:
3430 // 1. shmmax is too small for Java heap.
3431 // > check shmmax value: cat /proc/sys/kernel/shmmax
3432 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3433 // 2. not enough large page memory.
3434 // > check available large pages: cat /proc/meminfo
3435 // > increase amount of large pages:
3436 // echo new_value > /proc/sys/vm/nr_hugepages
3437 // Note 1: different Linux may use different name for this property,
3438 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3439 // Note 2: it's possible there's enough physical memory available but
3440 // they are so fragmented after a long run that they can't
3441 // coalesce into large pages. Try to reserve large pages when
3442 // the system is still "fresh".
3443 if (warn_on_failure) {
3444 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3445 warning("%s", msg);
3446 }
3447 return NULL;
3448 }
3449
3450 // attach to the region
3451 addr = (char*)shmat(shmid, req_addr, 0);
3452 int err = errno;
3453
3454 // Remove shmid. If shmat() is successful, the actual shared memory segment
3455 // will be deleted when it's detached by shmdt() or when the process
3456 // terminates. If shmat() is not successful this will remove the shared
3457 // segment immediately.
3458 shmctl(shmid, IPC_RMID, NULL);
3459
3460 if ((intptr_t)addr == -1) {
3461 if (warn_on_failure) {
3462 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3463 warning("%s", msg);
3464 }
3465 return NULL;
3466 }
3467
3468 return addr;
3469 }
3470
3471 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3472 int error) {
3473 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3474
3475 bool warn_on_failure = UseLargePages &&
3476 (!FLAG_IS_DEFAULT(UseLargePages) ||
3477 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3478 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3479
3480 if (warn_on_failure) {
3481 char msg[128];
3482 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3483 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3484 warning("%s", msg);
3485 }
3486 }
3487
3488 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3489 char* req_addr,
3490 bool exec) {
3491 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3492 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3493 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3494
3495 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3496 char* addr = (char*)::mmap(req_addr, bytes, prot,
3497 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3498 -1, 0);
3499
3500 if (addr == MAP_FAILED) {
3501 warn_on_large_pages_failure(req_addr, bytes, errno);
3502 return NULL;
3503 }
3504
3505 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3506
3507 return addr;
3508 }
3509
3510 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3511 size_t alignment,
3512 char* req_addr,
3513 bool exec) {
3514 size_t large_page_size = os::large_page_size();
3515
3516 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3517
3518 // Allocate small pages.
3519
3520 char* start;
3521 if (req_addr != NULL) {
3522 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3523 assert(is_size_aligned(bytes, alignment), "Must be");
3524 start = os::reserve_memory(bytes, req_addr);
3525 assert(start == NULL || start == req_addr, "Must be");
3526 } else {
3527 start = os::reserve_memory_aligned(bytes, alignment);
3528 }
3529
3530 if (start == NULL) {
3531 return NULL;
3532 }
3533
3534 assert(is_ptr_aligned(start, alignment), "Must be");
3535
3536 if (MemTracker::tracking_level() > NMT_minimal) {
3537 // os::reserve_memory_special will record this memory area.
3538 // Need to release it here to prevent overlapping reservations.
3539 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3540 tkr.record((address)start, bytes);
3541 }
3542
3543 char* end = start + bytes;
3544
3545 // Find the regions of the allocated chunk that can be promoted to large pages.
3546 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3547 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3548
3549 size_t lp_bytes = lp_end - lp_start;
3550
3551 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3552
3553 if (lp_bytes == 0) {
3554 // The mapped region doesn't even span the start and the end of a large page.
3555 // Fall back to allocate a non-special area.
3556 ::munmap(start, end - start);
3557 return NULL;
3558 }
3559
3560 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3561
3562
3563 void* result;
3564
3565 if (start != lp_start) {
3566 result = ::mmap(start, lp_start - start, prot,
3567 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3568 -1, 0);
3569 if (result == MAP_FAILED) {
3570 ::munmap(lp_start, end - lp_start);
3571 return NULL;
3572 }
3573 }
3574
3575 result = ::mmap(lp_start, lp_bytes, prot,
3576 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3577 -1, 0);
3578 if (result == MAP_FAILED) {
3579 warn_on_large_pages_failure(req_addr, bytes, errno);
3580 // If the mmap above fails, the large pages region will be unmapped and we
3581 // have regions before and after with small pages. Release these regions.
3582 //
3583 // | mapped | unmapped | mapped |
3584 // ^ ^ ^ ^
3585 // start lp_start lp_end end
3586 //
3587 ::munmap(start, lp_start - start);
3588 ::munmap(lp_end, end - lp_end);
3589 return NULL;
3590 }
3591
3592 if (lp_end != end) {
3593 result = ::mmap(lp_end, end - lp_end, prot,
3594 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3595 -1, 0);
3596 if (result == MAP_FAILED) {
3597 ::munmap(start, lp_end - start);
3598 return NULL;
3599 }
3600 }
3601
3602 return start;
3603 }
3604
3605 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3606 size_t alignment,
3607 char* req_addr,
3608 bool exec) {
3609 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3610 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3611 assert(is_power_of_2(alignment), "Must be");
3612 assert(is_power_of_2(os::large_page_size()), "Must be");
3613 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3614
3615 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3616 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3617 } else {
3618 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3619 }
3620 }
3621
3622 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3623 char* req_addr, bool exec) {
3624 assert(UseLargePages, "only for large pages");
3625
3626 char* addr;
3627 if (UseSHM) {
3628 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3629 } else {
3630 assert(UseHugeTLBFS, "must be");
3631 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3632 }
3633
3634 if (addr != NULL) {
3635 if (UseNUMAInterleaving) {
3636 numa_make_global(addr, bytes);
3637 }
3638
3639 // The memory is committed
3640 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3641 }
3642
3643 return addr;
3644 }
3645
3646 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3647 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3648 return shmdt(base) == 0;
3649 }
3650
3651 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3652 return pd_release_memory(base, bytes);
3653 }
3654
3655 bool os::release_memory_special(char* base, size_t bytes) {
3656 bool res;
3657 if (MemTracker::tracking_level() > NMT_minimal) {
3658 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3659 res = os::Linux::release_memory_special_impl(base, bytes);
3660 if (res) {
3661 tkr.record((address)base, bytes);
3662 }
3663
3664 } else {
3665 res = os::Linux::release_memory_special_impl(base, bytes);
3666 }
3667 return res;
3668 }
3669
3670 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3671 assert(UseLargePages, "only for large pages");
3672 bool res;
3673
3674 if (UseSHM) {
3675 res = os::Linux::release_memory_special_shm(base, bytes);
3676 } else {
3677 assert(UseHugeTLBFS, "must be");
3678 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3679 }
3680 return res;
3681 }
3682
3683 size_t os::large_page_size() {
3684 return _large_page_size;
3685 }
3686
3687 // With SysV SHM the entire memory region must be allocated as shared
3688 // memory.
3689 // HugeTLBFS allows application to commit large page memory on demand.
3690 // However, when committing memory with HugeTLBFS fails, the region
3691 // that was supposed to be committed will lose the old reservation
3692 // and allow other threads to steal that memory region. Because of this
3693 // behavior we can't commit HugeTLBFS memory.
3694 bool os::can_commit_large_page_memory() {
3695 return UseTransparentHugePages;
3696 }
3697
3698 bool os::can_execute_large_page_memory() {
3699 return UseTransparentHugePages || UseHugeTLBFS;
3700 }
3701
3702 // Reserve memory at an arbitrary address, only if that area is
3703 // available (and not reserved for something else).
3704
3705 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3706 const int max_tries = 10;
3707 char* base[max_tries];
3708 size_t size[max_tries];
3709 const size_t gap = 0x000000;
3710
3711 // Assert only that the size is a multiple of the page size, since
3712 // that's all that mmap requires, and since that's all we really know
3713 // about at this low abstraction level. If we need higher alignment,
3714 // we can either pass an alignment to this method or verify alignment
3715 // in one of the methods further up the call chain. See bug 5044738.
3716 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3717
3718 // Repeatedly allocate blocks until the block is allocated at the
3719 // right spot. Give up after max_tries. Note that reserve_memory() will
3720 // automatically update _highest_vm_reserved_address if the call is
3721 // successful. The variable tracks the highest memory address every reserved
3722 // by JVM. It is used to detect heap-stack collision if running with
3723 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3724 // space than needed, it could confuse the collision detecting code. To
3725 // solve the problem, save current _highest_vm_reserved_address and
3726 // calculate the correct value before return.
3727 address old_highest = _highest_vm_reserved_address;
3728
3729 // Linux mmap allows caller to pass an address as hint; give it a try first,
3730 // if kernel honors the hint then we can return immediately.
3731 char * addr = anon_mmap(requested_addr, bytes, false);
3732 if (addr == requested_addr) {
3733 return requested_addr;
3734 }
3735
3736 if (addr != NULL) {
3737 // mmap() is successful but it fails to reserve at the requested address
3738 anon_munmap(addr, bytes);
3739 }
3740
3741 int i;
3742 for (i = 0; i < max_tries; ++i) {
3743 base[i] = reserve_memory(bytes);
3744
3745 if (base[i] != NULL) {
3746 // Is this the block we wanted?
3747 if (base[i] == requested_addr) {
3748 size[i] = bytes;
3749 break;
3750 }
3751
3752 // Does this overlap the block we wanted? Give back the overlapped
3753 // parts and try again.
3754
3755 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3756 if (top_overlap >= 0 && top_overlap < bytes) {
3757 unmap_memory(base[i], top_overlap);
3758 base[i] += top_overlap;
3759 size[i] = bytes - top_overlap;
3760 } else {
3761 size_t bottom_overlap = base[i] + bytes - requested_addr;
3762 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3763 unmap_memory(requested_addr, bottom_overlap);
3764 size[i] = bytes - bottom_overlap;
3765 } else {
3766 size[i] = bytes;
3767 }
3768 }
3769 }
3770 }
3771
3772 // Give back the unused reserved pieces.
3773
3774 for (int j = 0; j < i; ++j) {
3775 if (base[j] != NULL) {
3776 unmap_memory(base[j], size[j]);
3777 }
3778 }
3779
3780 if (i < max_tries) {
3781 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3782 return requested_addr;
3783 } else {
3784 _highest_vm_reserved_address = old_highest;
3785 return NULL;
3786 }
3787 }
3788
3789 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3790 return ::read(fd, buf, nBytes);
3791 }
3792
3793 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3794 return ::pread(fd, buf, nBytes, offset);
3795 }
3796
3797 // Short sleep, direct OS call.
3798 //
3799 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3800 // sched_yield(2) will actually give up the CPU:
3801 //
3802 // * Alone on this pariticular CPU, keeps running.
3803 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3804 // (pre 2.6.39).
3805 //
3806 // So calling this with 0 is an alternative.
3807 //
3808 void os::naked_short_sleep(jlong ms) {
3809 struct timespec req;
3810
3811 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3812 req.tv_sec = 0;
3813 if (ms > 0) {
3814 req.tv_nsec = (ms % 1000) * 1000000;
3815 } else {
3816 req.tv_nsec = 1;
3817 }
3818
3819 nanosleep(&req, NULL);
3820
3821 return;
3822 }
3823
3824 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3825 void os::infinite_sleep() {
3826 while (true) { // sleep forever ...
3827 ::sleep(100); // ... 100 seconds at a time
3828 }
3829 }
3830
3831 // Used to convert frequent JVM_Yield() to nops
3832 bool os::dont_yield() {
3833 return DontYieldALot;
3834 }
3835
3836 void os::naked_yield() {
3837 sched_yield();
3838 }
3839
3840 ////////////////////////////////////////////////////////////////////////////////
3841 // thread priority support
3842
3843 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3844 // only supports dynamic priority, static priority must be zero. For real-time
3845 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3846 // However, for large multi-threaded applications, SCHED_RR is not only slower
3847 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3848 // of 5 runs - Sep 2005).
3849 //
3850 // The following code actually changes the niceness of kernel-thread/LWP. It
3851 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3852 // not the entire user process, and user level threads are 1:1 mapped to kernel
3853 // threads. It has always been the case, but could change in the future. For
3854 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3855 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3856
3857 int os::java_to_os_priority[CriticalPriority + 1] = {
3858 19, // 0 Entry should never be used
3859
3860 4, // 1 MinPriority
3861 3, // 2
3862 2, // 3
3863
3864 1, // 4
3865 0, // 5 NormPriority
3866 -1, // 6
3867
3868 -2, // 7
3869 -3, // 8
3870 -4, // 9 NearMaxPriority
3871
3872 -5, // 10 MaxPriority
3873
3874 -5 // 11 CriticalPriority
3875 };
3876
3877 static int prio_init() {
3878 if (ThreadPriorityPolicy == 1) {
3879 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3880 // if effective uid is not root. Perhaps, a more elegant way of doing
3881 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3882 if (geteuid() != 0) {
3883 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3884 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3885 }
3886 ThreadPriorityPolicy = 0;
3887 }
3888 }
3889 if (UseCriticalJavaThreadPriority) {
3890 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3891 }
3892 return 0;
3893 }
3894
3895 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3896 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3897
3898 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3899 return (ret == 0) ? OS_OK : OS_ERR;
3900 }
3901
3902 OSReturn os::get_native_priority(const Thread* const thread,
3903 int *priority_ptr) {
3904 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3905 *priority_ptr = java_to_os_priority[NormPriority];
3906 return OS_OK;
3907 }
3908
3909 errno = 0;
3910 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3911 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3912 }
3913
3914 // Hint to the underlying OS that a task switch would not be good.
3915 // Void return because it's a hint and can fail.
3916 void os::hint_no_preempt() {}
3917
3918 ////////////////////////////////////////////////////////////////////////////////
3919 // suspend/resume support
3920
3921 // the low-level signal-based suspend/resume support is a remnant from the
3922 // old VM-suspension that used to be for java-suspension, safepoints etc,
3923 // within hotspot. Now there is a single use-case for this:
3924 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3925 // that runs in the watcher thread.
3926 // The remaining code is greatly simplified from the more general suspension
3927 // code that used to be used.
3928 //
3929 // The protocol is quite simple:
3930 // - suspend:
3931 // - sends a signal to the target thread
3932 // - polls the suspend state of the osthread using a yield loop
3933 // - target thread signal handler (SR_handler) sets suspend state
3934 // and blocks in sigsuspend until continued
3935 // - resume:
3936 // - sets target osthread state to continue
3937 // - sends signal to end the sigsuspend loop in the SR_handler
3938 //
3939 // Note that the SR_lock plays no role in this suspend/resume protocol.
3940
3941 static void resume_clear_context(OSThread *osthread) {
3942 osthread->set_ucontext(NULL);
3943 osthread->set_siginfo(NULL);
3944 }
3945
3946 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3947 ucontext_t* context) {
3948 osthread->set_ucontext(context);
3949 osthread->set_siginfo(siginfo);
3950 }
3951
3952 // Handler function invoked when a thread's execution is suspended or
3953 // resumed. We have to be careful that only async-safe functions are
3954 // called here (Note: most pthread functions are not async safe and
3955 // should be avoided.)
3956 //
3957 // Note: sigwait() is a more natural fit than sigsuspend() from an
3958 // interface point of view, but sigwait() prevents the signal hander
3959 // from being run. libpthread would get very confused by not having
3960 // its signal handlers run and prevents sigwait()'s use with the
3961 // mutex granting granting signal.
3962 //
3963 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3964 //
3965 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3966 // Save and restore errno to avoid confusing native code with EINTR
3967 // after sigsuspend.
3968 int old_errno = errno;
3969
3970 Thread* thread = Thread::current();
3971 OSThread* osthread = thread->osthread();
3972 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3973
3974 os::SuspendResume::State current = osthread->sr.state();
3975 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3976 suspend_save_context(osthread, siginfo, context);
3977
3978 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3979 os::SuspendResume::State state = osthread->sr.suspended();
3980 if (state == os::SuspendResume::SR_SUSPENDED) {
3981 sigset_t suspend_set; // signals for sigsuspend()
3982
3983 // get current set of blocked signals and unblock resume signal
3984 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3985 sigdelset(&suspend_set, SR_signum);
3986
3987 sr_semaphore.signal();
3988 // wait here until we are resumed
3989 while (1) {
3990 sigsuspend(&suspend_set);
3991
3992 os::SuspendResume::State result = osthread->sr.running();
3993 if (result == os::SuspendResume::SR_RUNNING) {
3994 sr_semaphore.signal();
3995 break;
3996 }
3997 }
3998
3999 } else if (state == os::SuspendResume::SR_RUNNING) {
4000 // request was cancelled, continue
4001 } else {
4002 ShouldNotReachHere();
4003 }
4004
4005 resume_clear_context(osthread);
4006 } else if (current == os::SuspendResume::SR_RUNNING) {
4007 // request was cancelled, continue
4008 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4009 // ignore
4010 } else {
4011 // ignore
4012 }
4013
4014 errno = old_errno;
4015 }
4016
4017
4018 static int SR_initialize() {
4019 struct sigaction act;
4020 char *s;
4021 // Get signal number to use for suspend/resume
4022 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4023 int sig = ::strtol(s, 0, 10);
4024 if (sig > 0 || sig < _NSIG) {
4025 SR_signum = sig;
4026 }
4027 }
4028
4029 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4030 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4031
4032 sigemptyset(&SR_sigset);
4033 sigaddset(&SR_sigset, SR_signum);
4034
4035 // Set up signal handler for suspend/resume
4036 act.sa_flags = SA_RESTART|SA_SIGINFO;
4037 act.sa_handler = (void (*)(int)) SR_handler;
4038
4039 // SR_signum is blocked by default.
4040 // 4528190 - We also need to block pthread restart signal (32 on all
4041 // supported Linux platforms). Note that LinuxThreads need to block
4042 // this signal for all threads to work properly. So we don't have
4043 // to use hard-coded signal number when setting up the mask.
4044 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4045
4046 if (sigaction(SR_signum, &act, 0) == -1) {
4047 return -1;
4048 }
4049
4050 // Save signal flag
4051 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4052 return 0;
4053 }
4054
4055 static int sr_notify(OSThread* osthread) {
4056 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4057 assert_status(status == 0, status, "pthread_kill");
4058 return status;
4059 }
4060
4061 // "Randomly" selected value for how long we want to spin
4062 // before bailing out on suspending a thread, also how often
4063 // we send a signal to a thread we want to resume
4064 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4065 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4066
4067 // returns true on success and false on error - really an error is fatal
4068 // but this seems the normal response to library errors
4069 static bool do_suspend(OSThread* osthread) {
4070 assert(osthread->sr.is_running(), "thread should be running");
4071 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4072
4073 // mark as suspended and send signal
4074 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4075 // failed to switch, state wasn't running?
4076 ShouldNotReachHere();
4077 return false;
4078 }
4079
4080 if (sr_notify(osthread) != 0) {
4081 ShouldNotReachHere();
4082 }
4083
4084 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4085 while (true) {
4086 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4087 break;
4088 } else {
4089 // timeout
4090 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4091 if (cancelled == os::SuspendResume::SR_RUNNING) {
4092 return false;
4093 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4094 // make sure that we consume the signal on the semaphore as well
4095 sr_semaphore.wait();
4096 break;
4097 } else {
4098 ShouldNotReachHere();
4099 return false;
4100 }
4101 }
4102 }
4103
4104 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4105 return true;
4106 }
4107
4108 static void do_resume(OSThread* osthread) {
4109 assert(osthread->sr.is_suspended(), "thread should be suspended");
4110 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4111
4112 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4113 // failed to switch to WAKEUP_REQUEST
4114 ShouldNotReachHere();
4115 return;
4116 }
4117
4118 while (true) {
4119 if (sr_notify(osthread) == 0) {
4120 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4121 if (osthread->sr.is_running()) {
4122 return;
4123 }
4124 }
4125 } else {
4126 ShouldNotReachHere();
4127 }
4128 }
4129
4130 guarantee(osthread->sr.is_running(), "Must be running!");
4131 }
4132
4133 ///////////////////////////////////////////////////////////////////////////////////
4134 // signal handling (except suspend/resume)
4135
4136 // This routine may be used by user applications as a "hook" to catch signals.
4137 // The user-defined signal handler must pass unrecognized signals to this
4138 // routine, and if it returns true (non-zero), then the signal handler must
4139 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4140 // routine will never retun false (zero), but instead will execute a VM panic
4141 // routine kill the process.
4142 //
4143 // If this routine returns false, it is OK to call it again. This allows
4144 // the user-defined signal handler to perform checks either before or after
4145 // the VM performs its own checks. Naturally, the user code would be making
4146 // a serious error if it tried to handle an exception (such as a null check
4147 // or breakpoint) that the VM was generating for its own correct operation.
4148 //
4149 // This routine may recognize any of the following kinds of signals:
4150 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4151 // It should be consulted by handlers for any of those signals.
4152 //
4153 // The caller of this routine must pass in the three arguments supplied
4154 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4155 // field of the structure passed to sigaction(). This routine assumes that
4156 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4157 //
4158 // Note that the VM will print warnings if it detects conflicting signal
4159 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4160 //
4161 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4162 siginfo_t* siginfo,
4163 void* ucontext,
4164 int abort_if_unrecognized);
4165
4166 void signalHandler(int sig, siginfo_t* info, void* uc) {
4167 assert(info != NULL && uc != NULL, "it must be old kernel");
4168 int orig_errno = errno; // Preserve errno value over signal handler.
4169 JVM_handle_linux_signal(sig, info, uc, true);
4170 errno = orig_errno;
4171 }
4172
4173
4174 // This boolean allows users to forward their own non-matching signals
4175 // to JVM_handle_linux_signal, harmlessly.
4176 bool os::Linux::signal_handlers_are_installed = false;
4177
4178 // For signal-chaining
4179 struct sigaction os::Linux::sigact[MAXSIGNUM];
4180 unsigned int os::Linux::sigs = 0;
4181 bool os::Linux::libjsig_is_loaded = false;
4182 typedef struct sigaction *(*get_signal_t)(int);
4183 get_signal_t os::Linux::get_signal_action = NULL;
4184
4185 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4186 struct sigaction *actp = NULL;
4187
4188 if (libjsig_is_loaded) {
4189 // Retrieve the old signal handler from libjsig
4190 actp = (*get_signal_action)(sig);
4191 }
4192 if (actp == NULL) {
4193 // Retrieve the preinstalled signal handler from jvm
4194 actp = get_preinstalled_handler(sig);
4195 }
4196
4197 return actp;
4198 }
4199
4200 static bool call_chained_handler(struct sigaction *actp, int sig,
4201 siginfo_t *siginfo, void *context) {
4202 // Call the old signal handler
4203 if (actp->sa_handler == SIG_DFL) {
4204 // It's more reasonable to let jvm treat it as an unexpected exception
4205 // instead of taking the default action.
4206 return false;
4207 } else if (actp->sa_handler != SIG_IGN) {
4208 if ((actp->sa_flags & SA_NODEFER) == 0) {
4209 // automaticlly block the signal
4210 sigaddset(&(actp->sa_mask), sig);
4211 }
4212
4213 sa_handler_t hand;
4214 sa_sigaction_t sa;
4215 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4216 // retrieve the chained handler
4217 if (siginfo_flag_set) {
4218 sa = actp->sa_sigaction;
4219 } else {
4220 hand = actp->sa_handler;
4221 }
4222
4223 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4224 actp->sa_handler = SIG_DFL;
4225 }
4226
4227 // try to honor the signal mask
4228 sigset_t oset;
4229 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4230
4231 // call into the chained handler
4232 if (siginfo_flag_set) {
4233 (*sa)(sig, siginfo, context);
4234 } else {
4235 (*hand)(sig);
4236 }
4237
4238 // restore the signal mask
4239 pthread_sigmask(SIG_SETMASK, &oset, 0);
4240 }
4241 // Tell jvm's signal handler the signal is taken care of.
4242 return true;
4243 }
4244
4245 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4246 bool chained = false;
4247 // signal-chaining
4248 if (UseSignalChaining) {
4249 struct sigaction *actp = get_chained_signal_action(sig);
4250 if (actp != NULL) {
4251 chained = call_chained_handler(actp, sig, siginfo, context);
4252 }
4253 }
4254 return chained;
4255 }
4256
4257 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4258 if ((((unsigned int)1 << sig) & sigs) != 0) {
4259 return &sigact[sig];
4260 }
4261 return NULL;
4262 }
4263
4264 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4265 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4266 sigact[sig] = oldAct;
4267 sigs |= (unsigned int)1 << sig;
4268 }
4269
4270 // for diagnostic
4271 int os::Linux::sigflags[MAXSIGNUM];
4272
4273 int os::Linux::get_our_sigflags(int sig) {
4274 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4275 return sigflags[sig];
4276 }
4277
4278 void os::Linux::set_our_sigflags(int sig, int flags) {
4279 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4280 sigflags[sig] = flags;
4281 }
4282
4283 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4284 // Check for overwrite.
4285 struct sigaction oldAct;
4286 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4287
4288 void* oldhand = oldAct.sa_sigaction
4289 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4290 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4291 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4292 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4293 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4294 if (AllowUserSignalHandlers || !set_installed) {
4295 // Do not overwrite; user takes responsibility to forward to us.
4296 return;
4297 } else if (UseSignalChaining) {
4298 // save the old handler in jvm
4299 save_preinstalled_handler(sig, oldAct);
4300 // libjsig also interposes the sigaction() call below and saves the
4301 // old sigaction on it own.
4302 } else {
4303 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4304 "%#lx for signal %d.", (long)oldhand, sig));
4305 }
4306 }
4307
4308 struct sigaction sigAct;
4309 sigfillset(&(sigAct.sa_mask));
4310 sigAct.sa_handler = SIG_DFL;
4311 if (!set_installed) {
4312 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4313 } else {
4314 sigAct.sa_sigaction = signalHandler;
4315 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4316 }
4317 // Save flags, which are set by ours
4318 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4319 sigflags[sig] = sigAct.sa_flags;
4320
4321 int ret = sigaction(sig, &sigAct, &oldAct);
4322 assert(ret == 0, "check");
4323
4324 void* oldhand2 = oldAct.sa_sigaction
4325 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4326 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4327 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4328 }
4329
4330 // install signal handlers for signals that HotSpot needs to
4331 // handle in order to support Java-level exception handling.
4332
4333 void os::Linux::install_signal_handlers() {
4334 if (!signal_handlers_are_installed) {
4335 signal_handlers_are_installed = true;
4336
4337 // signal-chaining
4338 typedef void (*signal_setting_t)();
4339 signal_setting_t begin_signal_setting = NULL;
4340 signal_setting_t end_signal_setting = NULL;
4341 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4342 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4343 if (begin_signal_setting != NULL) {
4344 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4345 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4346 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4347 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4348 libjsig_is_loaded = true;
4349 assert(UseSignalChaining, "should enable signal-chaining");
4350 }
4351 if (libjsig_is_loaded) {
4352 // Tell libjsig jvm is setting signal handlers
4353 (*begin_signal_setting)();
4354 }
4355
4356 set_signal_handler(SIGSEGV, true);
4357 set_signal_handler(SIGPIPE, true);
4358 set_signal_handler(SIGBUS, true);
4359 set_signal_handler(SIGILL, true);
4360 set_signal_handler(SIGFPE, true);
4361 #if defined(PPC64)
4362 set_signal_handler(SIGTRAP, true);
4363 #endif
4364 set_signal_handler(SIGXFSZ, true);
4365
4366 if (libjsig_is_loaded) {
4367 // Tell libjsig jvm finishes setting signal handlers
4368 (*end_signal_setting)();
4369 }
4370
4371 // We don't activate signal checker if libjsig is in place, we trust ourselves
4372 // and if UserSignalHandler is installed all bets are off.
4373 // Log that signal checking is off only if -verbose:jni is specified.
4374 if (CheckJNICalls) {
4375 if (libjsig_is_loaded) {
4376 if (PrintJNIResolving) {
4377 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4378 }
4379 check_signals = false;
4380 }
4381 if (AllowUserSignalHandlers) {
4382 if (PrintJNIResolving) {
4383 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4384 }
4385 check_signals = false;
4386 }
4387 }
4388 }
4389 }
4390
4391 // This is the fastest way to get thread cpu time on Linux.
4392 // Returns cpu time (user+sys) for any thread, not only for current.
4393 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4394 // It might work on 2.6.10+ with a special kernel/glibc patch.
4395 // For reference, please, see IEEE Std 1003.1-2004:
4396 // http://www.unix.org/single_unix_specification
4397
4398 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4399 struct timespec tp;
4400 int rc = os::Linux::clock_gettime(clockid, &tp);
4401 assert(rc == 0, "clock_gettime is expected to return 0 code");
4402
4403 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4404 }
4405
4406 /////
4407 // glibc on Linux platform uses non-documented flag
4408 // to indicate, that some special sort of signal
4409 // trampoline is used.
4410 // We will never set this flag, and we should
4411 // ignore this flag in our diagnostic
4412 #ifdef SIGNIFICANT_SIGNAL_MASK
4413 #undef SIGNIFICANT_SIGNAL_MASK
4414 #endif
4415 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4416
4417 static const char* get_signal_handler_name(address handler,
4418 char* buf, int buflen) {
4419 int offset;
4420 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4421 if (found) {
4422 // skip directory names
4423 const char *p1, *p2;
4424 p1 = buf;
4425 size_t len = strlen(os::file_separator());
4426 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4427 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4428 } else {
4429 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4430 }
4431 return buf;
4432 }
4433
4434 static void print_signal_handler(outputStream* st, int sig,
4435 char* buf, size_t buflen) {
4436 struct sigaction sa;
4437
4438 sigaction(sig, NULL, &sa);
4439
4440 // See comment for SIGNIFICANT_SIGNAL_MASK define
4441 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4442
4443 st->print("%s: ", os::exception_name(sig, buf, buflen));
4444
4445 address handler = (sa.sa_flags & SA_SIGINFO)
4446 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4447 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4448
4449 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4450 st->print("SIG_DFL");
4451 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4452 st->print("SIG_IGN");
4453 } else {
4454 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4455 }
4456
4457 st->print(", sa_mask[0]=");
4458 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4459
4460 address rh = VMError::get_resetted_sighandler(sig);
4461 // May be, handler was resetted by VMError?
4462 if (rh != NULL) {
4463 handler = rh;
4464 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4465 }
4466
4467 st->print(", sa_flags=");
4468 os::Posix::print_sa_flags(st, sa.sa_flags);
4469
4470 // Check: is it our handler?
4471 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4472 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4473 // It is our signal handler
4474 // check for flags, reset system-used one!
4475 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4476 st->print(
4477 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4478 os::Linux::get_our_sigflags(sig));
4479 }
4480 }
4481 st->cr();
4482 }
4483
4484
4485 #define DO_SIGNAL_CHECK(sig) \
4486 do { \
4487 if (!sigismember(&check_signal_done, sig)) { \
4488 os::Linux::check_signal_handler(sig); \
4489 } \
4490 } while (0)
4491
4492 // This method is a periodic task to check for misbehaving JNI applications
4493 // under CheckJNI, we can add any periodic checks here
4494
4495 void os::run_periodic_checks() {
4496 if (check_signals == false) return;
4497
4498 // SEGV and BUS if overridden could potentially prevent
4499 // generation of hs*.log in the event of a crash, debugging
4500 // such a case can be very challenging, so we absolutely
4501 // check the following for a good measure:
4502 DO_SIGNAL_CHECK(SIGSEGV);
4503 DO_SIGNAL_CHECK(SIGILL);
4504 DO_SIGNAL_CHECK(SIGFPE);
4505 DO_SIGNAL_CHECK(SIGBUS);
4506 DO_SIGNAL_CHECK(SIGPIPE);
4507 DO_SIGNAL_CHECK(SIGXFSZ);
4508 #if defined(PPC64)
4509 DO_SIGNAL_CHECK(SIGTRAP);
4510 #endif
4511
4512 // ReduceSignalUsage allows the user to override these handlers
4513 // see comments at the very top and jvm_solaris.h
4514 if (!ReduceSignalUsage) {
4515 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4516 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4517 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4518 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4519 }
4520
4521 DO_SIGNAL_CHECK(SR_signum);
4522 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4523 }
4524
4525 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4526
4527 static os_sigaction_t os_sigaction = NULL;
4528
4529 void os::Linux::check_signal_handler(int sig) {
4530 char buf[O_BUFLEN];
4531 address jvmHandler = NULL;
4532
4533
4534 struct sigaction act;
4535 if (os_sigaction == NULL) {
4536 // only trust the default sigaction, in case it has been interposed
4537 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4538 if (os_sigaction == NULL) return;
4539 }
4540
4541 os_sigaction(sig, (struct sigaction*)NULL, &act);
4542
4543
4544 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4545
4546 address thisHandler = (act.sa_flags & SA_SIGINFO)
4547 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4548 : CAST_FROM_FN_PTR(address, act.sa_handler);
4549
4550
4551 switch (sig) {
4552 case SIGSEGV:
4553 case SIGBUS:
4554 case SIGFPE:
4555 case SIGPIPE:
4556 case SIGILL:
4557 case SIGXFSZ:
4558 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4559 break;
4560
4561 case SHUTDOWN1_SIGNAL:
4562 case SHUTDOWN2_SIGNAL:
4563 case SHUTDOWN3_SIGNAL:
4564 case BREAK_SIGNAL:
4565 jvmHandler = (address)user_handler();
4566 break;
4567
4568 case INTERRUPT_SIGNAL:
4569 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4570 break;
4571
4572 default:
4573 if (sig == SR_signum) {
4574 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4575 } else {
4576 return;
4577 }
4578 break;
4579 }
4580
4581 if (thisHandler != jvmHandler) {
4582 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4583 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4584 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4585 // No need to check this sig any longer
4586 sigaddset(&check_signal_done, sig);
4587 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4588 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4589 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4590 exception_name(sig, buf, O_BUFLEN));
4591 }
4592 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4593 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4594 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4595 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4596 // No need to check this sig any longer
4597 sigaddset(&check_signal_done, sig);
4598 }
4599
4600 // Dump all the signal
4601 if (sigismember(&check_signal_done, sig)) {
4602 print_signal_handlers(tty, buf, O_BUFLEN);
4603 }
4604 }
4605
4606 extern void report_error(char* file_name, int line_no, char* title,
4607 char* format, ...);
4608
4609 extern bool signal_name(int signo, char* buf, size_t len);
4610
4611 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4612 if (0 < exception_code && exception_code <= SIGRTMAX) {
4613 // signal
4614 if (!signal_name(exception_code, buf, size)) {
4615 jio_snprintf(buf, size, "SIG%d", exception_code);
4616 }
4617 return buf;
4618 } else {
4619 return NULL;
4620 }
4621 }
4622
4623 // this is called _before_ the most of global arguments have been parsed
4624 void os::init(void) {
4625 char dummy; // used to get a guess on initial stack address
4626 // first_hrtime = gethrtime();
4627
4628 // With LinuxThreads the JavaMain thread pid (primordial thread)
4629 // is different than the pid of the java launcher thread.
4630 // So, on Linux, the launcher thread pid is passed to the VM
4631 // via the sun.java.launcher.pid property.
4632 // Use this property instead of getpid() if it was correctly passed.
4633 // See bug 6351349.
4634 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4635
4636 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4637
4638 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4639
4640 init_random(1234567);
4641
4642 ThreadCritical::initialize();
4643
4644 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4645 if (Linux::page_size() == -1) {
4646 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4647 strerror(errno)));
4648 }
4649 init_page_sizes((size_t) Linux::page_size());
4650
4651 Linux::initialize_system_info();
4652
4653 // main_thread points to the aboriginal thread
4654 Linux::_main_thread = pthread_self();
4655
4656 Linux::clock_init();
4657 initial_time_count = javaTimeNanos();
4658
4659 // pthread_condattr initialization for monotonic clock
4660 int status;
4661 pthread_condattr_t* _condattr = os::Linux::condAttr();
4662 if ((status = pthread_condattr_init(_condattr)) != 0) {
4663 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4664 }
4665 // Only set the clock if CLOCK_MONOTONIC is available
4666 if (os::supports_monotonic_clock()) {
4667 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4668 if (status == EINVAL) {
4669 warning("Unable to use monotonic clock with relative timed-waits" \
4670 " - changes to the time-of-day clock may have adverse affects");
4671 } else {
4672 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4673 }
4674 }
4675 }
4676 // else it defaults to CLOCK_REALTIME
4677
4678 pthread_mutex_init(&dl_mutex, NULL);
4679
4680 // If the pagesize of the VM is greater than 8K determine the appropriate
4681 // number of initial guard pages. The user can change this with the
4682 // command line arguments, if needed.
4683 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4684 StackYellowPages = 1;
4685 StackRedPages = 1;
4686 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4687 }
4688
4689 // retrieve entry point for pthread_setname_np
4690 Linux::_pthread_setname_np =
4691 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4692
4693 }
4694
4695 // To install functions for atexit system call
4696 extern "C" {
4697 static void perfMemory_exit_helper() {
4698 perfMemory_exit();
4699 }
4700 }
4701
4702 // this is called _after_ the global arguments have been parsed
4703 jint os::init_2(void) {
4704 Linux::fast_thread_clock_init();
4705
4706 // Allocate a single page and mark it as readable for safepoint polling
4707 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4708 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4709
4710 os::set_polling_page(polling_page);
4711
4712 #ifndef PRODUCT
4713 if (Verbose && PrintMiscellaneous) {
4714 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4715 (intptr_t)polling_page);
4716 }
4717 #endif
4718
4719 if (!UseMembar) {
4720 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4721 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4722 os::set_memory_serialize_page(mem_serialize_page);
4723
4724 #ifndef PRODUCT
4725 if (Verbose && PrintMiscellaneous) {
4726 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4727 (intptr_t)mem_serialize_page);
4728 }
4729 #endif
4730 }
4731
4732 // initialize suspend/resume support - must do this before signal_sets_init()
4733 if (SR_initialize() != 0) {
4734 perror("SR_initialize failed");
4735 return JNI_ERR;
4736 }
4737
4738 Linux::signal_sets_init();
4739 Linux::install_signal_handlers();
4740
4741 // Check minimum allowable stack size for thread creation and to initialize
4742 // the java system classes, including StackOverflowError - depends on page
4743 // size. Add a page for compiler2 recursion in main thread.
4744 // Add in 2*BytesPerWord times page size to account for VM stack during
4745 // class initialization depending on 32 or 64 bit VM.
4746 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4747 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4748 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4749
4750 size_t threadStackSizeInBytes = ThreadStackSize * K;
4751 if (threadStackSizeInBytes != 0 &&
4752 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4753 tty->print_cr("\nThe stack size specified is too small, "
4754 "Specify at least %dk",
4755 os::Linux::min_stack_allowed/ K);
4756 return JNI_ERR;
4757 }
4758
4759 // Make the stack size a multiple of the page size so that
4760 // the yellow/red zones can be guarded.
4761 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4762 vm_page_size()));
4763
4764 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4765
4766 #if defined(IA32)
4767 workaround_expand_exec_shield_cs_limit();
4768 #endif
4769
4770 Linux::libpthread_init();
4771 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4772 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4773 Linux::glibc_version(), Linux::libpthread_version(),
4774 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4775 }
4776
4777 if (UseNUMA) {
4778 if (!Linux::libnuma_init()) {
4779 UseNUMA = false;
4780 } else {
4781 if ((Linux::numa_max_node() < 1)) {
4782 // There's only one node(they start from 0), disable NUMA.
4783 UseNUMA = false;
4784 }
4785 }
4786 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4787 // we can make the adaptive lgrp chunk resizing work. If the user specified
4788 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4789 // disable adaptive resizing.
4790 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4791 if (FLAG_IS_DEFAULT(UseNUMA)) {
4792 UseNUMA = false;
4793 } else {
4794 if (FLAG_IS_DEFAULT(UseLargePages) &&
4795 FLAG_IS_DEFAULT(UseSHM) &&
4796 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4797 UseLargePages = false;
4798 } else {
4799 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4800 UseAdaptiveSizePolicy = false;
4801 UseAdaptiveNUMAChunkSizing = false;
4802 }
4803 }
4804 }
4805 if (!UseNUMA && ForceNUMA) {
4806 UseNUMA = true;
4807 }
4808 }
4809
4810 if (MaxFDLimit) {
4811 // set the number of file descriptors to max. print out error
4812 // if getrlimit/setrlimit fails but continue regardless.
4813 struct rlimit nbr_files;
4814 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4815 if (status != 0) {
4816 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4817 perror("os::init_2 getrlimit failed");
4818 }
4819 } else {
4820 nbr_files.rlim_cur = nbr_files.rlim_max;
4821 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4822 if (status != 0) {
4823 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4824 perror("os::init_2 setrlimit failed");
4825 }
4826 }
4827 }
4828 }
4829
4830 // Initialize lock used to serialize thread creation (see os::create_thread)
4831 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4832
4833 // at-exit methods are called in the reverse order of their registration.
4834 // atexit functions are called on return from main or as a result of a
4835 // call to exit(3C). There can be only 32 of these functions registered
4836 // and atexit() does not set errno.
4837
4838 if (PerfAllowAtExitRegistration) {
4839 // only register atexit functions if PerfAllowAtExitRegistration is set.
4840 // atexit functions can be delayed until process exit time, which
4841 // can be problematic for embedded VM situations. Embedded VMs should
4842 // call DestroyJavaVM() to assure that VM resources are released.
4843
4844 // note: perfMemory_exit_helper atexit function may be removed in
4845 // the future if the appropriate cleanup code can be added to the
4846 // VM_Exit VMOperation's doit method.
4847 if (atexit(perfMemory_exit_helper) != 0) {
4848 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4849 }
4850 }
4851
4852 // initialize thread priority policy
4853 prio_init();
4854
4855 return JNI_OK;
4856 }
4857
4858 // Mark the polling page as unreadable
4859 void os::make_polling_page_unreadable(void) {
4860 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4861 fatal("Could not disable polling page");
4862 }
4863 }
4864
4865 // Mark the polling page as readable
4866 void os::make_polling_page_readable(void) {
4867 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4868 fatal("Could not enable polling page");
4869 }
4870 }
4871
4872 int os::active_processor_count() {
4873 // Linux doesn't yet have a (official) notion of processor sets,
4874 // so just return the number of online processors.
4875 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4876 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4877 return online_cpus;
4878 }
4879
4880 void os::set_native_thread_name(const char *name) {
4881 if (Linux::_pthread_setname_np) {
4882 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4883 snprintf(buf, sizeof(buf), "%s", name);
4884 buf[sizeof(buf) - 1] = '\0';
4885 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4886 // ERANGE should not happen; all other errors should just be ignored.
4887 assert(rc != ERANGE, "pthread_setname_np failed");
4888 }
4889 }
4890
4891 bool os::distribute_processes(uint length, uint* distribution) {
4892 // Not yet implemented.
4893 return false;
4894 }
4895
4896 bool os::bind_to_processor(uint processor_id) {
4897 // Not yet implemented.
4898 return false;
4899 }
4900
4901 ///
4902
4903 void os::SuspendedThreadTask::internal_do_task() {
4904 if (do_suspend(_thread->osthread())) {
4905 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4906 do_task(context);
4907 do_resume(_thread->osthread());
4908 }
4909 }
4910
4911 class PcFetcher : public os::SuspendedThreadTask {
4912 public:
4913 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4914 ExtendedPC result();
4915 protected:
4916 void do_task(const os::SuspendedThreadTaskContext& context);
4917 private:
4918 ExtendedPC _epc;
4919 };
4920
4921 ExtendedPC PcFetcher::result() {
4922 guarantee(is_done(), "task is not done yet.");
4923 return _epc;
4924 }
4925
4926 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4927 Thread* thread = context.thread();
4928 OSThread* osthread = thread->osthread();
4929 if (osthread->ucontext() != NULL) {
4930 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4931 } else {
4932 // NULL context is unexpected, double-check this is the VMThread
4933 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4934 }
4935 }
4936
4937 // Suspends the target using the signal mechanism and then grabs the PC before
4938 // resuming the target. Used by the flat-profiler only
4939 ExtendedPC os::get_thread_pc(Thread* thread) {
4940 // Make sure that it is called by the watcher for the VMThread
4941 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4942 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4943
4944 PcFetcher fetcher(thread);
4945 fetcher.run();
4946 return fetcher.result();
4947 }
4948
4949 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4950 pthread_mutex_t *_mutex,
4951 const struct timespec *_abstime) {
4952 if (is_NPTL()) {
4953 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4954 } else {
4955 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4956 // word back to default 64bit precision if condvar is signaled. Java
4957 // wants 53bit precision. Save and restore current value.
4958 int fpu = get_fpu_control_word();
4959 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4960 set_fpu_control_word(fpu);
4961 return status;
4962 }
4963 }
4964
4965 ////////////////////////////////////////////////////////////////////////////////
4966 // debug support
4967
4968 bool os::find(address addr, outputStream* st) {
4969 Dl_info dlinfo;
4970 memset(&dlinfo, 0, sizeof(dlinfo));
4971 if (dladdr(addr, &dlinfo) != 0) {
4972 st->print(PTR_FORMAT ": ", addr);
4973 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4974 st->print("%s+%#x", dlinfo.dli_sname,
4975 addr - (intptr_t)dlinfo.dli_saddr);
4976 } else if (dlinfo.dli_fbase != NULL) {
4977 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4978 } else {
4979 st->print("<absolute address>");
4980 }
4981 if (dlinfo.dli_fname != NULL) {
4982 st->print(" in %s", dlinfo.dli_fname);
4983 }
4984 if (dlinfo.dli_fbase != NULL) {
4985 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4986 }
4987 st->cr();
4988
4989 if (Verbose) {
4990 // decode some bytes around the PC
4991 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4992 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4993 address lowest = (address) dlinfo.dli_sname;
4994 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4995 if (begin < lowest) begin = lowest;
4996 Dl_info dlinfo2;
4997 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4998 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4999 end = (address) dlinfo2.dli_saddr;
5000 }
5001 Disassembler::decode(begin, end, st);
5002 }
5003 return true;
5004 }
5005 return false;
5006 }
5007
5008 ////////////////////////////////////////////////////////////////////////////////
5009 // misc
5010
5011 // This does not do anything on Linux. This is basically a hook for being
5012 // able to use structured exception handling (thread-local exception filters)
5013 // on, e.g., Win32.
5014 void
5015 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5016 JavaCallArguments* args, Thread* thread) {
5017 f(value, method, args, thread);
5018 }
5019
5020 void os::print_statistics() {
5021 }
5022
5023 int os::message_box(const char* title, const char* message) {
5024 int i;
5025 fdStream err(defaultStream::error_fd());
5026 for (i = 0; i < 78; i++) err.print_raw("=");
5027 err.cr();
5028 err.print_raw_cr(title);
5029 for (i = 0; i < 78; i++) err.print_raw("-");
5030 err.cr();
5031 err.print_raw_cr(message);
5032 for (i = 0; i < 78; i++) err.print_raw("=");
5033 err.cr();
5034
5035 char buf[16];
5036 // Prevent process from exiting upon "read error" without consuming all CPU
5037 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5038
5039 return buf[0] == 'y' || buf[0] == 'Y';
5040 }
5041
5042 int os::stat(const char *path, struct stat *sbuf) {
5043 char pathbuf[MAX_PATH];
5044 if (strlen(path) > MAX_PATH - 1) {
5045 errno = ENAMETOOLONG;
5046 return -1;
5047 }
5048 os::native_path(strcpy(pathbuf, path));
5049 return ::stat(pathbuf, sbuf);
5050 }
5051
5052 bool os::check_heap(bool force) {
5053 return true;
5054 }
5055
5056 // Is a (classpath) directory empty?
5057 bool os::dir_is_empty(const char* path) {
5058 DIR *dir = NULL;
5059 struct dirent *ptr;
5060
5061 dir = opendir(path);
5062 if (dir == NULL) return true;
5063
5064 // Scan the directory
5065 bool result = true;
5066 char buf[sizeof(struct dirent) + MAX_PATH];
5067 while (result && (ptr = ::readdir(dir)) != NULL) {
5068 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5069 result = false;
5070 }
5071 }
5072 closedir(dir);
5073 return result;
5074 }
5075
5076 // This code originates from JDK's sysOpen and open64_w
5077 // from src/solaris/hpi/src/system_md.c
5078
5079 int os::open(const char *path, int oflag, int mode) {
5080 if (strlen(path) > MAX_PATH - 1) {
5081 errno = ENAMETOOLONG;
5082 return -1;
5083 }
5084
5085 // All file descriptors that are opened in the Java process and not
5086 // specifically destined for a subprocess should have the close-on-exec
5087 // flag set. If we don't set it, then careless 3rd party native code
5088 // might fork and exec without closing all appropriate file descriptors
5089 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5090 // turn might:
5091 //
5092 // - cause end-of-file to fail to be detected on some file
5093 // descriptors, resulting in mysterious hangs, or
5094 //
5095 // - might cause an fopen in the subprocess to fail on a system
5096 // suffering from bug 1085341.
5097 //
5098 // (Yes, the default setting of the close-on-exec flag is a Unix
5099 // design flaw)
5100 //
5101 // See:
5102 // 1085341: 32-bit stdio routines should support file descriptors >255
5103 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5104 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5105 //
5106 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5107 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5108 // because it saves a system call and removes a small window where the flag
5109 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5110 // and we fall back to using FD_CLOEXEC (see below).
5111 #ifdef O_CLOEXEC
5112 oflag |= O_CLOEXEC;
5113 #endif
5114
5115 int fd = ::open64(path, oflag, mode);
5116 if (fd == -1) return -1;
5117
5118 //If the open succeeded, the file might still be a directory
5119 {
5120 struct stat64 buf64;
5121 int ret = ::fstat64(fd, &buf64);
5122 int st_mode = buf64.st_mode;
5123
5124 if (ret != -1) {
5125 if ((st_mode & S_IFMT) == S_IFDIR) {
5126 errno = EISDIR;
5127 ::close(fd);
5128 return -1;
5129 }
5130 } else {
5131 ::close(fd);
5132 return -1;
5133 }
5134 }
5135
5136 #ifdef FD_CLOEXEC
5137 // Validate that the use of the O_CLOEXEC flag on open above worked.
5138 // With recent kernels, we will perform this check exactly once.
5139 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5140 if (!O_CLOEXEC_is_known_to_work) {
5141 int flags = ::fcntl(fd, F_GETFD);
5142 if (flags != -1) {
5143 if ((flags & FD_CLOEXEC) != 0)
5144 O_CLOEXEC_is_known_to_work = 1;
5145 else
5146 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5147 }
5148 }
5149 #endif
5150
5151 return fd;
5152 }
5153
5154
5155 // create binary file, rewriting existing file if required
5156 int os::create_binary_file(const char* path, bool rewrite_existing) {
5157 int oflags = O_WRONLY | O_CREAT;
5158 if (!rewrite_existing) {
5159 oflags |= O_EXCL;
5160 }
5161 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5162 }
5163
5164 // return current position of file pointer
5165 jlong os::current_file_offset(int fd) {
5166 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5167 }
5168
5169 // move file pointer to the specified offset
5170 jlong os::seek_to_file_offset(int fd, jlong offset) {
5171 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5172 }
5173
5174 // This code originates from JDK's sysAvailable
5175 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5176
5177 int os::available(int fd, jlong *bytes) {
5178 jlong cur, end;
5179 int mode;
5180 struct stat64 buf64;
5181
5182 if (::fstat64(fd, &buf64) >= 0) {
5183 mode = buf64.st_mode;
5184 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5185 // XXX: is the following call interruptible? If so, this might
5186 // need to go through the INTERRUPT_IO() wrapper as for other
5187 // blocking, interruptible calls in this file.
5188 int n;
5189 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5190 *bytes = n;
5191 return 1;
5192 }
5193 }
5194 }
5195 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5196 return 0;
5197 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5198 return 0;
5199 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5200 return 0;
5201 }
5202 *bytes = end - cur;
5203 return 1;
5204 }
5205
5206 // Map a block of memory.
5207 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5208 char *addr, size_t bytes, bool read_only,
5209 bool allow_exec) {
5210 int prot;
5211 int flags = MAP_PRIVATE;
5212
5213 if (read_only) {
5214 prot = PROT_READ;
5215 } else {
5216 prot = PROT_READ | PROT_WRITE;
5217 }
5218
5219 if (allow_exec) {
5220 prot |= PROT_EXEC;
5221 }
5222
5223 if (addr != NULL) {
5224 flags |= MAP_FIXED;
5225 }
5226
5227 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5228 fd, file_offset);
5229 if (mapped_address == MAP_FAILED) {
5230 return NULL;
5231 }
5232 return mapped_address;
5233 }
5234
5235
5236 // Remap a block of memory.
5237 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5238 char *addr, size_t bytes, bool read_only,
5239 bool allow_exec) {
5240 // same as map_memory() on this OS
5241 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5242 allow_exec);
5243 }
5244
5245
5246 // Unmap a block of memory.
5247 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5248 return munmap(addr, bytes) == 0;
5249 }
5250
5251 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5252
5253 static clockid_t thread_cpu_clockid(Thread* thread) {
5254 pthread_t tid = thread->osthread()->pthread_id();
5255 clockid_t clockid;
5256
5257 // Get thread clockid
5258 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5259 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5260 return clockid;
5261 }
5262
5263 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5264 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5265 // of a thread.
5266 //
5267 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5268 // the fast estimate available on the platform.
5269
5270 jlong os::current_thread_cpu_time() {
5271 if (os::Linux::supports_fast_thread_cpu_time()) {
5272 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5273 } else {
5274 // return user + sys since the cost is the same
5275 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5276 }
5277 }
5278
5279 jlong os::thread_cpu_time(Thread* thread) {
5280 // consistent with what current_thread_cpu_time() returns
5281 if (os::Linux::supports_fast_thread_cpu_time()) {
5282 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5283 } else {
5284 return slow_thread_cpu_time(thread, true /* user + sys */);
5285 }
5286 }
5287
5288 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5289 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5290 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5291 } else {
5292 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5293 }
5294 }
5295
5296 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5297 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5298 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5299 } else {
5300 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5301 }
5302 }
5303
5304 // -1 on error.
5305 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5306 pid_t tid = thread->osthread()->thread_id();
5307 char *s;
5308 char stat[2048];
5309 int statlen;
5310 char proc_name[64];
5311 int count;
5312 long sys_time, user_time;
5313 char cdummy;
5314 int idummy;
5315 long ldummy;
5316 FILE *fp;
5317
5318 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5319 fp = fopen(proc_name, "r");
5320 if (fp == NULL) return -1;
5321 statlen = fread(stat, 1, 2047, fp);
5322 stat[statlen] = '\0';
5323 fclose(fp);
5324
5325 // Skip pid and the command string. Note that we could be dealing with
5326 // weird command names, e.g. user could decide to rename java launcher
5327 // to "java 1.4.2 :)", then the stat file would look like
5328 // 1234 (java 1.4.2 :)) R ... ...
5329 // We don't really need to know the command string, just find the last
5330 // occurrence of ")" and then start parsing from there. See bug 4726580.
5331 s = strrchr(stat, ')');
5332 if (s == NULL) return -1;
5333
5334 // Skip blank chars
5335 do { s++; } while (s && isspace(*s));
5336
5337 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5338 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5339 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5340 &user_time, &sys_time);
5341 if (count != 13) return -1;
5342 if (user_sys_cpu_time) {
5343 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5344 } else {
5345 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5346 }
5347 }
5348
5349 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5350 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5351 info_ptr->may_skip_backward = false; // elapsed time not wall time
5352 info_ptr->may_skip_forward = false; // elapsed time not wall time
5353 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5354 }
5355
5356 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5357 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5358 info_ptr->may_skip_backward = false; // elapsed time not wall time
5359 info_ptr->may_skip_forward = false; // elapsed time not wall time
5360 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5361 }
5362
5363 bool os::is_thread_cpu_time_supported() {
5364 return true;
5365 }
5366
5367 // System loadavg support. Returns -1 if load average cannot be obtained.
5368 // Linux doesn't yet have a (official) notion of processor sets,
5369 // so just return the system wide load average.
5370 int os::loadavg(double loadavg[], int nelem) {
5371 return ::getloadavg(loadavg, nelem);
5372 }
5373
5374 void os::pause() {
5375 char filename[MAX_PATH];
5376 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5377 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5378 } else {
5379 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5380 }
5381
5382 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5383 if (fd != -1) {
5384 struct stat buf;
5385 ::close(fd);
5386 while (::stat(filename, &buf) == 0) {
5387 (void)::poll(NULL, 0, 100);
5388 }
5389 } else {
5390 jio_fprintf(stderr,
5391 "Could not open pause file '%s', continuing immediately.\n", filename);
5392 }
5393 }
5394
5395
5396 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5397 // comment paragraphs are worth repeating here:
5398 //
5399 // Assumption:
5400 // Only one parker can exist on an event, which is why we allocate
5401 // them per-thread. Multiple unparkers can coexist.
5402 //
5403 // _Event serves as a restricted-range semaphore.
5404 // -1 : thread is blocked, i.e. there is a waiter
5405 // 0 : neutral: thread is running or ready,
5406 // could have been signaled after a wait started
5407 // 1 : signaled - thread is running or ready
5408 //
5409 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5410 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5411 // For specifics regarding the bug see GLIBC BUGID 261237 :
5412 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5413 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5414 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5415 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5416 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5417 // and monitorenter when we're using 1-0 locking. All those operations may result in
5418 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5419 // of libpthread avoids the problem, but isn't practical.
5420 //
5421 // Possible remedies:
5422 //
5423 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5424 // This is palliative and probabilistic, however. If the thread is preempted
5425 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5426 // than the minimum period may have passed, and the abstime may be stale (in the
5427 // past) resultin in a hang. Using this technique reduces the odds of a hang
5428 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5429 //
5430 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5431 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5432 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5433 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5434 // thread.
5435 //
5436 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5437 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5438 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5439 // This also works well. In fact it avoids kernel-level scalability impediments
5440 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5441 // timers in a graceful fashion.
5442 //
5443 // 4. When the abstime value is in the past it appears that control returns
5444 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5445 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5446 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5447 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5448 // It may be possible to avoid reinitialization by checking the return
5449 // value from pthread_cond_timedwait(). In addition to reinitializing the
5450 // condvar we must establish the invariant that cond_signal() is only called
5451 // within critical sections protected by the adjunct mutex. This prevents
5452 // cond_signal() from "seeing" a condvar that's in the midst of being
5453 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5454 // desirable signal-after-unlock optimization that avoids futile context switching.
5455 //
5456 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5457 // structure when a condvar is used or initialized. cond_destroy() would
5458 // release the helper structure. Our reinitialize-after-timedwait fix
5459 // put excessive stress on malloc/free and locks protecting the c-heap.
5460 //
5461 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5462 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5463 // and only enabling the work-around for vulnerable environments.
5464
5465 // utility to compute the abstime argument to timedwait:
5466 // millis is the relative timeout time
5467 // abstime will be the absolute timeout time
5468 // TODO: replace compute_abstime() with unpackTime()
5469
5470 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5471 if (millis < 0) millis = 0;
5472
5473 jlong seconds = millis / 1000;
5474 millis %= 1000;
5475 if (seconds > 50000000) { // see man cond_timedwait(3T)
5476 seconds = 50000000;
5477 }
5478
5479 if (os::supports_monotonic_clock()) {
5480 struct timespec now;
5481 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5482 assert_status(status == 0, status, "clock_gettime");
5483 abstime->tv_sec = now.tv_sec + seconds;
5484 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5485 if (nanos >= NANOSECS_PER_SEC) {
5486 abstime->tv_sec += 1;
5487 nanos -= NANOSECS_PER_SEC;
5488 }
5489 abstime->tv_nsec = nanos;
5490 } else {
5491 struct timeval now;
5492 int status = gettimeofday(&now, NULL);
5493 assert(status == 0, "gettimeofday");
5494 abstime->tv_sec = now.tv_sec + seconds;
5495 long usec = now.tv_usec + millis * 1000;
5496 if (usec >= 1000000) {
5497 abstime->tv_sec += 1;
5498 usec -= 1000000;
5499 }
5500 abstime->tv_nsec = usec * 1000;
5501 }
5502 return abstime;
5503 }
5504
5505 void os::PlatformEvent::park() { // AKA "down()"
5506 // Transitions for _Event:
5507 // -1 => -1 : illegal
5508 // 1 => 0 : pass - return immediately
5509 // 0 => -1 : block; then set _Event to 0 before returning
5510
5511 // Invariant: Only the thread associated with the Event/PlatformEvent
5512 // may call park().
5513 // TODO: assert that _Assoc != NULL or _Assoc == Self
5514 assert(_nParked == 0, "invariant");
5515
5516 int v;
5517 for (;;) {
5518 v = _Event;
5519 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5520 }
5521 guarantee(v >= 0, "invariant");
5522 if (v == 0) {
5523 // Do this the hard way by blocking ...
5524 int status = pthread_mutex_lock(_mutex);
5525 assert_status(status == 0, status, "mutex_lock");
5526 guarantee(_nParked == 0, "invariant");
5527 ++_nParked;
5528 while (_Event < 0) {
5529 status = pthread_cond_wait(_cond, _mutex);
5530 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5531 // Treat this the same as if the wait was interrupted
5532 if (status == ETIME) { status = EINTR; }
5533 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5534 }
5535 --_nParked;
5536
5537 _Event = 0;
5538 status = pthread_mutex_unlock(_mutex);
5539 assert_status(status == 0, status, "mutex_unlock");
5540 // Paranoia to ensure our locked and lock-free paths interact
5541 // correctly with each other.
5542 OrderAccess::fence();
5543 }
5544 guarantee(_Event >= 0, "invariant");
5545 }
5546
5547 int os::PlatformEvent::park(jlong millis) {
5548 // Transitions for _Event:
5549 // -1 => -1 : illegal
5550 // 1 => 0 : pass - return immediately
5551 // 0 => -1 : block; then set _Event to 0 before returning
5552
5553 guarantee(_nParked == 0, "invariant");
5554
5555 int v;
5556 for (;;) {
5557 v = _Event;
5558 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5559 }
5560 guarantee(v >= 0, "invariant");
5561 if (v != 0) return OS_OK;
5562
5563 // We do this the hard way, by blocking the thread.
5564 // Consider enforcing a minimum timeout value.
5565 struct timespec abst;
5566 compute_abstime(&abst, millis);
5567
5568 int ret = OS_TIMEOUT;
5569 int status = pthread_mutex_lock(_mutex);
5570 assert_status(status == 0, status, "mutex_lock");
5571 guarantee(_nParked == 0, "invariant");
5572 ++_nParked;
5573
5574 // Object.wait(timo) will return because of
5575 // (a) notification
5576 // (b) timeout
5577 // (c) thread.interrupt
5578 //
5579 // Thread.interrupt and object.notify{All} both call Event::set.
5580 // That is, we treat thread.interrupt as a special case of notification.
5581 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5582 // We assume all ETIME returns are valid.
5583 //
5584 // TODO: properly differentiate simultaneous notify+interrupt.
5585 // In that case, we should propagate the notify to another waiter.
5586
5587 while (_Event < 0) {
5588 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5589 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5590 pthread_cond_destroy(_cond);
5591 pthread_cond_init(_cond, os::Linux::condAttr());
5592 }
5593 assert_status(status == 0 || status == EINTR ||
5594 status == ETIME || status == ETIMEDOUT,
5595 status, "cond_timedwait");
5596 if (!FilterSpuriousWakeups) break; // previous semantics
5597 if (status == ETIME || status == ETIMEDOUT) break;
5598 // We consume and ignore EINTR and spurious wakeups.
5599 }
5600 --_nParked;
5601 if (_Event >= 0) {
5602 ret = OS_OK;
5603 }
5604 _Event = 0;
5605 status = pthread_mutex_unlock(_mutex);
5606 assert_status(status == 0, status, "mutex_unlock");
5607 assert(_nParked == 0, "invariant");
5608 // Paranoia to ensure our locked and lock-free paths interact
5609 // correctly with each other.
5610 OrderAccess::fence();
5611 return ret;
5612 }
5613
5614 void os::PlatformEvent::unpark() {
5615 // Transitions for _Event:
5616 // 0 => 1 : just return
5617 // 1 => 1 : just return
5618 // -1 => either 0 or 1; must signal target thread
5619 // That is, we can safely transition _Event from -1 to either
5620 // 0 or 1.
5621 // See also: "Semaphores in Plan 9" by Mullender & Cox
5622 //
5623 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5624 // that it will take two back-to-back park() calls for the owning
5625 // thread to block. This has the benefit of forcing a spurious return
5626 // from the first park() call after an unpark() call which will help
5627 // shake out uses of park() and unpark() without condition variables.
5628
5629 if (Atomic::xchg(1, &_Event) >= 0) return;
5630
5631 // Wait for the thread associated with the event to vacate
5632 int status = pthread_mutex_lock(_mutex);
5633 assert_status(status == 0, status, "mutex_lock");
5634 int AnyWaiters = _nParked;
5635 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5636 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5637 AnyWaiters = 0;
5638 pthread_cond_signal(_cond);
5639 }
5640 status = pthread_mutex_unlock(_mutex);
5641 assert_status(status == 0, status, "mutex_unlock");
5642 if (AnyWaiters != 0) {
5643 // Note that we signal() *after* dropping the lock for "immortal" Events.
5644 // This is safe and avoids a common class of futile wakeups. In rare
5645 // circumstances this can cause a thread to return prematurely from
5646 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5647 // will simply re-test the condition and re-park itself.
5648 // This provides particular benefit if the underlying platform does not
5649 // provide wait morphing.
5650 status = pthread_cond_signal(_cond);
5651 assert_status(status == 0, status, "cond_signal");
5652 }
5653 }
5654
5655
5656 // JSR166
5657 // -------------------------------------------------------
5658
5659 // The solaris and linux implementations of park/unpark are fairly
5660 // conservative for now, but can be improved. They currently use a
5661 // mutex/condvar pair, plus a a count.
5662 // Park decrements count if > 0, else does a condvar wait. Unpark
5663 // sets count to 1 and signals condvar. Only one thread ever waits
5664 // on the condvar. Contention seen when trying to park implies that someone
5665 // is unparking you, so don't wait. And spurious returns are fine, so there
5666 // is no need to track notifications.
5667
5668 // This code is common to linux and solaris and will be moved to a
5669 // common place in dolphin.
5670 //
5671 // The passed in time value is either a relative time in nanoseconds
5672 // or an absolute time in milliseconds. Either way it has to be unpacked
5673 // into suitable seconds and nanoseconds components and stored in the
5674 // given timespec structure.
5675 // Given time is a 64-bit value and the time_t used in the timespec is only
5676 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5677 // overflow if times way in the future are given. Further on Solaris versions
5678 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5679 // number of seconds, in abstime, is less than current_time + 100,000,000.
5680 // As it will be 28 years before "now + 100000000" will overflow we can
5681 // ignore overflow and just impose a hard-limit on seconds using the value
5682 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5683 // years from "now".
5684
5685 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5686 assert(time > 0, "convertTime");
5687 time_t max_secs = 0;
5688
5689 if (!os::supports_monotonic_clock() || isAbsolute) {
5690 struct timeval now;
5691 int status = gettimeofday(&now, NULL);
5692 assert(status == 0, "gettimeofday");
5693
5694 max_secs = now.tv_sec + MAX_SECS;
5695
5696 if (isAbsolute) {
5697 jlong secs = time / 1000;
5698 if (secs > max_secs) {
5699 absTime->tv_sec = max_secs;
5700 } else {
5701 absTime->tv_sec = secs;
5702 }
5703 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5704 } else {
5705 jlong secs = time / NANOSECS_PER_SEC;
5706 if (secs >= MAX_SECS) {
5707 absTime->tv_sec = max_secs;
5708 absTime->tv_nsec = 0;
5709 } else {
5710 absTime->tv_sec = now.tv_sec + secs;
5711 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5712 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5713 absTime->tv_nsec -= NANOSECS_PER_SEC;
5714 ++absTime->tv_sec; // note: this must be <= max_secs
5715 }
5716 }
5717 }
5718 } else {
5719 // must be relative using monotonic clock
5720 struct timespec now;
5721 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5722 assert_status(status == 0, status, "clock_gettime");
5723 max_secs = now.tv_sec + MAX_SECS;
5724 jlong secs = time / NANOSECS_PER_SEC;
5725 if (secs >= MAX_SECS) {
5726 absTime->tv_sec = max_secs;
5727 absTime->tv_nsec = 0;
5728 } else {
5729 absTime->tv_sec = now.tv_sec + secs;
5730 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5731 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5732 absTime->tv_nsec -= NANOSECS_PER_SEC;
5733 ++absTime->tv_sec; // note: this must be <= max_secs
5734 }
5735 }
5736 }
5737 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5738 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5739 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5740 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5741 }
5742
5743 void Parker::park(bool isAbsolute, jlong time) {
5744 // Ideally we'd do something useful while spinning, such
5745 // as calling unpackTime().
5746
5747 // Optional fast-path check:
5748 // Return immediately if a permit is available.
5749 // We depend on Atomic::xchg() having full barrier semantics
5750 // since we are doing a lock-free update to _counter.
5751 if (Atomic::xchg(0, &_counter) > 0) return;
5752
5753 Thread* thread = Thread::current();
5754 assert(thread->is_Java_thread(), "Must be JavaThread");
5755 JavaThread *jt = (JavaThread *)thread;
5756
5757 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5758 // Check interrupt before trying to wait
5759 if (Thread::is_interrupted(thread, false)) {
5760 return;
5761 }
5762
5763 // Next, demultiplex/decode time arguments
5764 timespec absTime;
5765 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5766 return;
5767 }
5768 if (time > 0) {
5769 unpackTime(&absTime, isAbsolute, time);
5770 }
5771
5772
5773 // Enter safepoint region
5774 // Beware of deadlocks such as 6317397.
5775 // The per-thread Parker:: mutex is a classic leaf-lock.
5776 // In particular a thread must never block on the Threads_lock while
5777 // holding the Parker:: mutex. If safepoints are pending both the
5778 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5779 ThreadBlockInVM tbivm(jt);
5780
5781 // Don't wait if cannot get lock since interference arises from
5782 // unblocking. Also. check interrupt before trying wait
5783 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5784 return;
5785 }
5786
5787 int status;
5788 if (_counter > 0) { // no wait needed
5789 _counter = 0;
5790 status = pthread_mutex_unlock(_mutex);
5791 assert(status == 0, "invariant");
5792 // Paranoia to ensure our locked and lock-free paths interact
5793 // correctly with each other and Java-level accesses.
5794 OrderAccess::fence();
5795 return;
5796 }
5797
5798 #ifdef ASSERT
5799 // Don't catch signals while blocked; let the running threads have the signals.
5800 // (This allows a debugger to break into the running thread.)
5801 sigset_t oldsigs;
5802 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5803 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5804 #endif
5805
5806 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5807 jt->set_suspend_equivalent();
5808 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5809
5810 assert(_cur_index == -1, "invariant");
5811 if (time == 0) {
5812 _cur_index = REL_INDEX; // arbitrary choice when not timed
5813 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5814 } else {
5815 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5816 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5817 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5818 pthread_cond_destroy(&_cond[_cur_index]);
5819 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5820 }
5821 }
5822 _cur_index = -1;
5823 assert_status(status == 0 || status == EINTR ||
5824 status == ETIME || status == ETIMEDOUT,
5825 status, "cond_timedwait");
5826
5827 #ifdef ASSERT
5828 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5829 #endif
5830
5831 _counter = 0;
5832 status = pthread_mutex_unlock(_mutex);
5833 assert_status(status == 0, status, "invariant");
5834 // Paranoia to ensure our locked and lock-free paths interact
5835 // correctly with each other and Java-level accesses.
5836 OrderAccess::fence();
5837
5838 // If externally suspended while waiting, re-suspend
5839 if (jt->handle_special_suspend_equivalent_condition()) {
5840 jt->java_suspend_self();
5841 }
5842 }
5843
5844 void Parker::unpark() {
5845 int status = pthread_mutex_lock(_mutex);
5846 assert(status == 0, "invariant");
5847 const int s = _counter;
5848 _counter = 1;
5849 if (s < 1) {
5850 // thread might be parked
5851 if (_cur_index != -1) {
5852 // thread is definitely parked
5853 if (WorkAroundNPTLTimedWaitHang) {
5854 status = pthread_cond_signal(&_cond[_cur_index]);
5855 assert(status == 0, "invariant");
5856 status = pthread_mutex_unlock(_mutex);
5857 assert(status == 0, "invariant");
5858 } else {
5859 status = pthread_mutex_unlock(_mutex);
5860 assert(status == 0, "invariant");
5861 status = pthread_cond_signal(&_cond[_cur_index]);
5862 assert(status == 0, "invariant");
5863 }
5864 } else {
5865 pthread_mutex_unlock(_mutex);
5866 assert(status == 0, "invariant");
5867 }
5868 } else {
5869 pthread_mutex_unlock(_mutex);
5870 assert(status == 0, "invariant");
5871 }
5872 }
5873
5874
5875 extern char** environ;
5876
5877 #ifndef __NR_fork
5878 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5879 #endif
5880
5881 #ifndef __NR_execve
5882 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5883 #endif
5884
5885 // Run the specified command in a separate process. Return its exit value,
5886 // or -1 on failure (e.g. can't fork a new process).
5887 // Unlike system(), this function can be called from signal handler. It
5888 // doesn't block SIGINT et al.
5889 int os::fork_and_exec(char* cmd) {
5890 const char * argv[4] = {"sh", "-c", cmd, NULL};
5891
5892 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5893 // pthread_atfork handlers and reset pthread library. All we need is a
5894 // separate process to execve. Make a direct syscall to fork process.
5895 // On IA64 there's no fork syscall, we have to use fork() and hope for
5896 // the best...
5897 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5898 IA64_ONLY(fork();)
5899
5900 if (pid < 0) {
5901 // fork failed
5902 return -1;
5903
5904 } else if (pid == 0) {
5905 // child process
5906
5907 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5908 // first to kill every thread on the thread list. Because this list is
5909 // not reset by fork() (see notes above), execve() will instead kill
5910 // every thread in the parent process. We know this is the only thread
5911 // in the new process, so make a system call directly.
5912 // IA64 should use normal execve() from glibc to match the glibc fork()
5913 // above.
5914 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5915 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5916
5917 // execve failed
5918 _exit(-1);
5919
5920 } else {
5921 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5922 // care about the actual exit code, for now.
5923
5924 int status;
5925
5926 // Wait for the child process to exit. This returns immediately if
5927 // the child has already exited. */
5928 while (waitpid(pid, &status, 0) < 0) {
5929 switch (errno) {
5930 case ECHILD: return 0;
5931 case EINTR: break;
5932 default: return -1;
5933 }
5934 }
5935
5936 if (WIFEXITED(status)) {
5937 // The child exited normally; get its exit code.
5938 return WEXITSTATUS(status);
5939 } else if (WIFSIGNALED(status)) {
5940 // The child exited because of a signal
5941 // The best value to return is 0x80 + signal number,
5942 // because that is what all Unix shells do, and because
5943 // it allows callers to distinguish between process exit and
5944 // process death by signal.
5945 return 0x80 + WTERMSIG(status);
5946 } else {
5947 // Unknown exit code; pass it through
5948 return status;
5949 }
5950 }
5951 }
5952
5953 // is_headless_jre()
5954 //
5955 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5956 // in order to report if we are running in a headless jre
5957 //
5958 // Since JDK8 xawt/libmawt.so was moved into the same directory
5959 // as libawt.so, and renamed libawt_xawt.so
5960 //
5961 bool os::is_headless_jre() {
5962 struct stat statbuf;
5963 char buf[MAXPATHLEN];
5964 char libmawtpath[MAXPATHLEN];
5965 const char *xawtstr = "/xawt/libmawt.so";
5966 const char *new_xawtstr = "/libawt_xawt.so";
5967 char *p;
5968
5969 // Get path to libjvm.so
5970 os::jvm_path(buf, sizeof(buf));
5971
5972 // Get rid of libjvm.so
5973 p = strrchr(buf, '/');
5974 if (p == NULL) {
5975 return false;
5976 } else {
5977 *p = '\0';
5978 }
5979
5980 // Get rid of client or server
5981 p = strrchr(buf, '/');
5982 if (p == NULL) {
5983 return false;
5984 } else {
5985 *p = '\0';
5986 }
5987
5988 // check xawt/libmawt.so
5989 strcpy(libmawtpath, buf);
5990 strcat(libmawtpath, xawtstr);
5991 if (::stat(libmawtpath, &statbuf) == 0) return false;
5992
5993 // check libawt_xawt.so
5994 strcpy(libmawtpath, buf);
5995 strcat(libmawtpath, new_xawtstr);
5996 if (::stat(libmawtpath, &statbuf) == 0) return false;
5997
5998 return true;
5999 }
6000
6001 // Get the default path to the core file
6002 // Returns the length of the string
6003 int os::get_core_path(char* buffer, size_t bufferSize) {
6004 /*
6005 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6006 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6007 */
6008 const int core_pattern_len = 129;
6009 char core_pattern[core_pattern_len] = {0};
6010
6011 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6012 if (core_pattern_file != -1) {
6013 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6014 ::close(core_pattern_file);
6015
6016 if (ret > 0) {
6017 char *last_char = core_pattern + strlen(core_pattern) - 1;
6018
6019 if (*last_char == '\n') {
6020 *last_char = '\0';
6021 }
6022 }
6023 }
6024
6025 if (strlen(core_pattern) == 0) {
6026 return 0;
6027 }
6028
6029 char *pid_pos = strstr(core_pattern, "%p");
6030 size_t written;
6031
6032 if (core_pattern[0] == '/') {
6033 written = jio_snprintf(buffer, bufferSize, core_pattern);
6034 } else {
6035 char cwd[PATH_MAX];
6036
6037 const char* p = get_current_directory(cwd, PATH_MAX);
6038 if (p == NULL) {
6039 assert(p != NULL, "failed to get current directory");
6040 return 0;
6041 }
6042
6043 if (core_pattern[0] == '|') {
6044 written = jio_snprintf(buffer, bufferSize,
6045 "\"%s\" (or dumping to %s/core.%d)",
6046 &core_pattern[1], p, current_process_id());
6047 } else {
6048 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6049 }
6050 }
6051
6052 if ((written >= 0) && (written < bufferSize)
6053 && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6054 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6055
6056 if (core_uses_pid_file != -1) {
6057 char core_uses_pid = 0;
6058 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6059 ::close(core_uses_pid_file);
6060
6061 if (core_uses_pid == '1'){
6062 jio_snprintf(buffer + written, bufferSize - written,
6063 ".%d", current_process_id());
6064 }
6065 }
6066 }
6067
6068 return strlen(buffer);
6069 }
6070
6071 /////////////// Unit tests ///////////////
6072
6073 #ifndef PRODUCT
6074
6075 #define test_log(...) \
6076 do { \
6077 if (VerboseInternalVMTests) { \
6078 tty->print_cr(__VA_ARGS__); \
6079 tty->flush(); \
6080 } \
6081 } while (false)
6082
6083 class TestReserveMemorySpecial : AllStatic {
6084 public:
6085 static void small_page_write(void* addr, size_t size) {
6086 size_t page_size = os::vm_page_size();
6087
6088 char* end = (char*)addr + size;
6089 for (char* p = (char*)addr; p < end; p += page_size) {
6090 *p = 1;
6091 }
6092 }
6093
6094 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6095 if (!UseHugeTLBFS) {
6096 return;
6097 }
6098
6099 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6100
6101 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6102
6103 if (addr != NULL) {
6104 small_page_write(addr, size);
6105
6106 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6107 }
6108 }
6109
6110 static void test_reserve_memory_special_huge_tlbfs_only() {
6111 if (!UseHugeTLBFS) {
6112 return;
6113 }
6114
6115 size_t lp = os::large_page_size();
6116
6117 for (size_t size = lp; size <= lp * 10; size += lp) {
6118 test_reserve_memory_special_huge_tlbfs_only(size);
6119 }
6120 }
6121
6122 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6123 if (!UseHugeTLBFS) {
6124 return;
6125 }
6126
6127 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6128 size, alignment);
6129
6130 assert(size >= os::large_page_size(), "Incorrect input to test");
6131
6132 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6133
6134 if (addr != NULL) {
6135 small_page_write(addr, size);
6136
6137 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6138 }
6139 }
6140
6141 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6142 size_t lp = os::large_page_size();
6143 size_t ag = os::vm_allocation_granularity();
6144
6145 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6146 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6147 }
6148 }
6149
6150 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6151 size_t lp = os::large_page_size();
6152 size_t ag = os::vm_allocation_granularity();
6153
6154 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6155 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6156 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6157 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6158 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6159 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6160 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6161 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6162 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6163 }
6164
6165 static void test_reserve_memory_special_huge_tlbfs() {
6166 if (!UseHugeTLBFS) {
6167 return;
6168 }
6169
6170 test_reserve_memory_special_huge_tlbfs_only();
6171 test_reserve_memory_special_huge_tlbfs_mixed();
6172 }
6173
6174 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6175 if (!UseSHM) {
6176 return;
6177 }
6178
6179 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6180
6181 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6182
6183 if (addr != NULL) {
6184 assert(is_ptr_aligned(addr, alignment), "Check");
6185 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6186
6187 small_page_write(addr, size);
6188
6189 os::Linux::release_memory_special_shm(addr, size);
6190 }
6191 }
6192
6193 static void test_reserve_memory_special_shm() {
6194 size_t lp = os::large_page_size();
6195 size_t ag = os::vm_allocation_granularity();
6196
6197 for (size_t size = ag; size < lp * 3; size += ag) {
6198 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6199 test_reserve_memory_special_shm(size, alignment);
6200 }
6201 }
6202 }
6203
6204 static void test() {
6205 test_reserve_memory_special_huge_tlbfs();
6206 test_reserve_memory_special_shm();
6207 }
6208 };
6209
6210 void TestReserveMemorySpecial_test() {
6211 TestReserveMemorySpecial::test();
6212 }
6213
6214 #endif