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 static char cpu_arch[] = "ppc64";
244 #elif defined(SPARC)
245 #ifdef _LP64
246 static char cpu_arch[] = "sparcv9";
247 #else
248 static char cpu_arch[] = "sparc";
249 #endif
250 #elif defined(AARCH64)
251 static char cpu_arch[] = "aarch64";
252 #else
253 #error Add appropriate cpu_arch setting
254 #endif
255
256
257 // pid_t gettid()
258 //
259 // Returns the kernel thread id of the currently running thread. Kernel
260 // thread id is used to access /proc.
261 //
262 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
263 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
264 //
265 pid_t os::Linux::gettid() {
266 int rslt = syscall(SYS_gettid);
267 if (rslt == -1) {
268 // old kernel, no NPTL support
269 return getpid();
270 } else {
271 return (pid_t)rslt;
272 }
273 }
274
275 // Most versions of linux have a bug where the number of processors are
276 // determined by looking at the /proc file system. In a chroot environment,
277 // the system call returns 1. This causes the VM to act as if it is
278 // a single processor and elide locking (see is_MP() call).
279 static bool unsafe_chroot_detected = false;
280 static const char *unstable_chroot_error = "/proc file system not found.\n"
281 "Java may be unstable running multithreaded in a chroot "
282 "environment on Linux when /proc filesystem is not mounted.";
283
284 void os::Linux::initialize_system_info() {
285 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
286 if (processor_count() == 1) {
287 pid_t pid = os::Linux::gettid();
288 char fname[32];
289 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
290 FILE *fp = fopen(fname, "r");
291 if (fp == NULL) {
292 unsafe_chroot_detected = true;
293 } else {
294 fclose(fp);
295 }
296 }
297 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
298 assert(processor_count() > 0, "linux error");
299 }
300
301 void os::init_system_properties_values() {
302 // The next steps are taken in the product version:
303 //
304 // Obtain the JAVA_HOME value from the location of libjvm.so.
305 // This library should be located at:
306 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
307 //
308 // If "/jre/lib/" appears at the right place in the path, then we
309 // assume libjvm.so is installed in a JDK and we use this path.
310 //
311 // Otherwise exit with message: "Could not create the Java virtual machine."
312 //
313 // The following extra steps are taken in the debugging version:
314 //
315 // If "/jre/lib/" does NOT appear at the right place in the path
316 // instead of exit check for $JAVA_HOME environment variable.
317 //
318 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
319 // then we append a fake suffix "hotspot/libjvm.so" to this path so
320 // it looks like libjvm.so is installed there
321 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
322 //
323 // Otherwise exit.
324 //
325 // Important note: if the location of libjvm.so changes this
326 // code needs to be changed accordingly.
327
328 // See ld(1):
329 // The linker uses the following search paths to locate required
330 // shared libraries:
331 // 1: ...
332 // ...
333 // 7: The default directories, normally /lib and /usr/lib.
334 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
335 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
336 #else
337 #define DEFAULT_LIBPATH "/lib:/usr/lib"
338 #endif
339
340 // Base path of extensions installed on the system.
341 #define SYS_EXT_DIR "/usr/java/packages"
342 #define EXTENSIONS_DIR "/lib/ext"
343
344 // Buffer that fits several sprintfs.
345 // Note that the space for the colon and the trailing null are provided
346 // by the nulls included by the sizeof operator.
347 const size_t bufsize =
348 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
349 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
350 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
351
352 // sysclasspath, java_home, dll_dir
353 {
354 char *pslash;
355 os::jvm_path(buf, bufsize);
356
357 // Found the full path to libjvm.so.
358 // Now cut the path to <java_home>/jre if we can.
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /libjvm.so.
362 }
363 pslash = strrchr(buf, '/');
364 if (pslash != NULL) {
365 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
366 }
367 Arguments::set_dll_dir(buf);
368
369 if (pslash != NULL) {
370 pslash = strrchr(buf, '/');
371 if (pslash != NULL) {
372 *pslash = '\0'; // Get rid of /<arch>.
373 pslash = strrchr(buf, '/');
374 if (pslash != NULL) {
375 *pslash = '\0'; // Get rid of /lib.
376 }
377 }
378 }
379 Arguments::set_java_home(buf);
380 set_boot_path('/', ':');
381 }
382
383 // Where to look for native libraries.
384 //
385 // Note: Due to a legacy implementation, most of the library path
386 // is set in the launcher. This was to accomodate linking restrictions
387 // on legacy Linux implementations (which are no longer supported).
388 // Eventually, all the library path setting will be done here.
389 //
390 // However, to prevent the proliferation of improperly built native
391 // libraries, the new path component /usr/java/packages is added here.
392 // Eventually, all the library path setting will be done here.
393 {
394 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
395 // should always exist (until the legacy problem cited above is
396 // addressed).
397 const char *v = ::getenv("LD_LIBRARY_PATH");
398 const char *v_colon = ":";
399 if (v == NULL) { v = ""; v_colon = ""; }
400 // That's +1 for the colon and +1 for the trailing '\0'.
401 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
402 strlen(v) + 1 +
403 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
404 mtInternal);
405 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
406 Arguments::set_library_path(ld_library_path);
407 FREE_C_HEAP_ARRAY(char, ld_library_path);
408 }
409
410 // Extensions directories.
411 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
412 Arguments::set_ext_dirs(buf);
413
414 FREE_C_HEAP_ARRAY(char, buf);
415
416 #undef DEFAULT_LIBPATH
417 #undef SYS_EXT_DIR
418 #undef EXTENSIONS_DIR
419 }
420
421 ////////////////////////////////////////////////////////////////////////////////
422 // breakpoint support
423
424 void os::breakpoint() {
425 BREAKPOINT;
426 }
427
428 extern "C" void breakpoint() {
429 // use debugger to set breakpoint here
430 }
431
432 ////////////////////////////////////////////////////////////////////////////////
433 // signal support
434
435 debug_only(static bool signal_sets_initialized = false);
436 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
437
438 bool os::Linux::is_sig_ignored(int sig) {
439 struct sigaction oact;
440 sigaction(sig, (struct sigaction*)NULL, &oact);
441 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
442 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
443 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
444 return true;
445 } else {
446 return false;
447 }
448 }
449
450 void os::Linux::signal_sets_init() {
451 // Should also have an assertion stating we are still single-threaded.
452 assert(!signal_sets_initialized, "Already initialized");
453 // Fill in signals that are necessarily unblocked for all threads in
454 // the VM. Currently, we unblock the following signals:
455 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
456 // by -Xrs (=ReduceSignalUsage));
457 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
458 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
459 // the dispositions or masks wrt these signals.
460 // Programs embedding the VM that want to use the above signals for their
461 // own purposes must, at this time, use the "-Xrs" option to prevent
462 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
463 // (See bug 4345157, and other related bugs).
464 // In reality, though, unblocking these signals is really a nop, since
465 // these signals are not blocked by default.
466 sigemptyset(&unblocked_sigs);
467 sigemptyset(&allowdebug_blocked_sigs);
468 sigaddset(&unblocked_sigs, SIGILL);
469 sigaddset(&unblocked_sigs, SIGSEGV);
470 sigaddset(&unblocked_sigs, SIGBUS);
471 sigaddset(&unblocked_sigs, SIGFPE);
472 #if defined(PPC64)
473 sigaddset(&unblocked_sigs, SIGTRAP);
474 #endif
475 sigaddset(&unblocked_sigs, SR_signum);
476
477 if (!ReduceSignalUsage) {
478 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
479 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
480 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
481 }
482 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
483 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
484 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
485 }
486 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
487 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
488 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
489 }
490 }
491 // Fill in signals that are blocked by all but the VM thread.
492 sigemptyset(&vm_sigs);
493 if (!ReduceSignalUsage) {
494 sigaddset(&vm_sigs, BREAK_SIGNAL);
495 }
496 debug_only(signal_sets_initialized = true);
497
498 }
499
500 // These are signals that are unblocked while a thread is running Java.
501 // (For some reason, they get blocked by default.)
502 sigset_t* os::Linux::unblocked_signals() {
503 assert(signal_sets_initialized, "Not initialized");
504 return &unblocked_sigs;
505 }
506
507 // These are the signals that are blocked while a (non-VM) thread is
508 // running Java. Only the VM thread handles these signals.
509 sigset_t* os::Linux::vm_signals() {
510 assert(signal_sets_initialized, "Not initialized");
511 return &vm_sigs;
512 }
513
514 // These are signals that are blocked during cond_wait to allow debugger in
515 sigset_t* os::Linux::allowdebug_blocked_signals() {
516 assert(signal_sets_initialized, "Not initialized");
517 return &allowdebug_blocked_sigs;
518 }
519
520 void os::Linux::hotspot_sigmask(Thread* thread) {
521
522 //Save caller's signal mask before setting VM signal mask
523 sigset_t caller_sigmask;
524 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
525
526 OSThread* osthread = thread->osthread();
527 osthread->set_caller_sigmask(caller_sigmask);
528
529 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
530
531 if (!ReduceSignalUsage) {
532 if (thread->is_VM_thread()) {
533 // Only the VM thread handles BREAK_SIGNAL ...
534 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
535 } else {
536 // ... all other threads block BREAK_SIGNAL
537 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
538 }
539 }
540 }
541
542 //////////////////////////////////////////////////////////////////////////////
543 // detecting pthread library
544
545 void os::Linux::libpthread_init() {
546 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
547 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
548 // generic name for earlier versions.
549 // Define macros here so we can build HotSpot on old systems.
550 #ifndef _CS_GNU_LIBC_VERSION
551 #define _CS_GNU_LIBC_VERSION 2
552 #endif
553 #ifndef _CS_GNU_LIBPTHREAD_VERSION
554 #define _CS_GNU_LIBPTHREAD_VERSION 3
555 #endif
556
557 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
558 if (n > 0) {
559 char *str = (char *)malloc(n, mtInternal);
560 confstr(_CS_GNU_LIBC_VERSION, str, n);
561 os::Linux::set_glibc_version(str);
562 } else {
563 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
564 static char _gnu_libc_version[32];
565 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
566 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
567 os::Linux::set_glibc_version(_gnu_libc_version);
568 }
569
570 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
571 if (n > 0) {
572 char *str = (char *)malloc(n, mtInternal);
573 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
574 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
575 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
576 // is the case. LinuxThreads has a hard limit on max number of threads.
577 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
578 // On the other hand, NPTL does not have such a limit, sysconf()
579 // will return -1 and errno is not changed. Check if it is really NPTL.
580 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
581 strstr(str, "NPTL") &&
582 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
583 free(str);
584 os::Linux::set_libpthread_version("linuxthreads");
585 } else {
586 os::Linux::set_libpthread_version(str);
587 }
588 } else {
589 // glibc before 2.3.2 only has LinuxThreads.
590 os::Linux::set_libpthread_version("linuxthreads");
591 }
592
593 if (strstr(libpthread_version(), "NPTL")) {
594 os::Linux::set_is_NPTL();
595 } else {
596 os::Linux::set_is_LinuxThreads();
597 }
598
599 // LinuxThreads have two flavors: floating-stack mode, which allows variable
600 // stack size; and fixed-stack mode. NPTL is always floating-stack.
601 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
602 os::Linux::set_is_floating_stack();
603 }
604 }
605
606 /////////////////////////////////////////////////////////////////////////////
607 // thread stack
608
609 // Force Linux kernel to expand current thread stack. If "bottom" is close
610 // to the stack guard, caller should block all signals.
611 //
612 // MAP_GROWSDOWN:
613 // A special mmap() flag that is used to implement thread stacks. It tells
614 // kernel that the memory region should extend downwards when needed. This
615 // allows early versions of LinuxThreads to only mmap the first few pages
616 // when creating a new thread. Linux kernel will automatically expand thread
617 // stack as needed (on page faults).
618 //
619 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
620 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
621 // region, it's hard to tell if the fault is due to a legitimate stack
622 // access or because of reading/writing non-exist memory (e.g. buffer
623 // overrun). As a rule, if the fault happens below current stack pointer,
624 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
625 // application (see Linux kernel fault.c).
626 //
627 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
628 // stack overflow detection.
629 //
630 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
631 // not use this flag. However, the stack of initial thread is not created
632 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
633 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
634 // and then attach the thread to JVM.
635 //
636 // To get around the problem and allow stack banging on Linux, we need to
637 // manually expand thread stack after receiving the SIGSEGV.
638 //
639 // There are two ways to expand thread stack to address "bottom", we used
640 // both of them in JVM before 1.5:
641 // 1. adjust stack pointer first so that it is below "bottom", and then
642 // touch "bottom"
643 // 2. mmap() the page in question
644 //
645 // Now alternate signal stack is gone, it's harder to use 2. For instance,
646 // if current sp is already near the lower end of page 101, and we need to
647 // call mmap() to map page 100, it is possible that part of the mmap() frame
648 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
649 // That will destroy the mmap() frame and cause VM to crash.
650 //
651 // The following code works by adjusting sp first, then accessing the "bottom"
652 // page to force a page fault. Linux kernel will then automatically expand the
653 // stack mapping.
654 //
655 // _expand_stack_to() assumes its frame size is less than page size, which
656 // should always be true if the function is not inlined.
657
658 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
659 #define NOINLINE
660 #else
661 #define NOINLINE __attribute__ ((noinline))
662 #endif
663
664 static void _expand_stack_to(address bottom) NOINLINE;
665
666 static void _expand_stack_to(address bottom) {
667 address sp;
668 size_t size;
669 volatile char *p;
670
671 // Adjust bottom to point to the largest address within the same page, it
672 // gives us a one-page buffer if alloca() allocates slightly more memory.
673 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
674 bottom += os::Linux::page_size() - 1;
675
676 // sp might be slightly above current stack pointer; if that's the case, we
677 // will alloca() a little more space than necessary, which is OK. Don't use
678 // os::current_stack_pointer(), as its result can be slightly below current
679 // stack pointer, causing us to not alloca enough to reach "bottom".
680 sp = (address)&sp;
681
682 if (sp > bottom) {
683 size = sp - bottom;
684 p = (volatile char *)alloca(size);
685 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
686 p[0] = '\0';
687 }
688 }
689
690 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
691 assert(t!=NULL, "just checking");
692 assert(t->osthread()->expanding_stack(), "expand should be set");
693 assert(t->stack_base() != NULL, "stack_base was not initialized");
694
695 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
696 sigset_t mask_all, old_sigset;
697 sigfillset(&mask_all);
698 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
699 _expand_stack_to(addr);
700 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
701 return true;
702 }
703 return false;
704 }
705
706 //////////////////////////////////////////////////////////////////////////////
707 // create new thread
708
709 static address highest_vm_reserved_address();
710
711 // check if it's safe to start a new thread
712 static bool _thread_safety_check(Thread* thread) {
713 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
714 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
715 // Heap is mmap'ed at lower end of memory space. Thread stacks are
716 // allocated (MAP_FIXED) from high address space. Every thread stack
717 // occupies a fixed size slot (usually 2Mbytes, but user can change
718 // it to other values if they rebuild LinuxThreads).
719 //
720 // Problem with MAP_FIXED is that mmap() can still succeed even part of
721 // the memory region has already been mmap'ed. That means if we have too
722 // many threads and/or very large heap, eventually thread stack will
723 // collide with heap.
724 //
725 // Here we try to prevent heap/stack collision by comparing current
726 // stack bottom with the highest address that has been mmap'ed by JVM
727 // plus a safety margin for memory maps created by native code.
728 //
729 // This feature can be disabled by setting ThreadSafetyMargin to 0
730 //
731 if (ThreadSafetyMargin > 0) {
732 address stack_bottom = os::current_stack_base() - os::current_stack_size();
733
734 // not safe if our stack extends below the safety margin
735 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
736 } else {
737 return true;
738 }
739 } else {
740 // Floating stack LinuxThreads or NPTL:
741 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
742 // there's not enough space left, pthread_create() will fail. If we come
743 // here, that means enough space has been reserved for stack.
744 return true;
745 }
746 }
747
748 // Thread start routine for all newly created threads
749 static void *java_start(Thread *thread) {
750 // Try to randomize the cache line index of hot stack frames.
751 // This helps when threads of the same stack traces evict each other's
752 // cache lines. The threads can be either from the same JVM instance, or
753 // from different JVM instances. The benefit is especially true for
754 // processors with hyperthreading technology.
755 static int counter = 0;
756 int pid = os::current_process_id();
757 alloca(((pid ^ counter++) & 7) * 128);
758
759 ThreadLocalStorage::set_thread(thread);
760
761 OSThread* osthread = thread->osthread();
762 Monitor* sync = osthread->startThread_lock();
763
764 // non floating stack LinuxThreads needs extra check, see above
765 if (!_thread_safety_check(thread)) {
766 // notify parent thread
767 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
768 osthread->set_state(ZOMBIE);
769 sync->notify_all();
770 return NULL;
771 }
772
773 // thread_id is kernel thread id (similar to Solaris LWP id)
774 osthread->set_thread_id(os::Linux::gettid());
775
776 if (UseNUMA) {
777 int lgrp_id = os::numa_get_group_id();
778 if (lgrp_id != -1) {
779 thread->set_lgrp_id(lgrp_id);
780 }
781 }
782 // initialize signal mask for this thread
783 os::Linux::hotspot_sigmask(thread);
784
785 // initialize floating point control register
786 os::Linux::init_thread_fpu_state();
787
788 // handshaking with parent thread
789 {
790 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
791
792 // notify parent thread
793 osthread->set_state(INITIALIZED);
794 sync->notify_all();
795
796 // wait until os::start_thread()
797 while (osthread->get_state() == INITIALIZED) {
798 sync->wait(Mutex::_no_safepoint_check_flag);
799 }
800 }
801
802 // call one more level start routine
803 thread->run();
804
805 return 0;
806 }
807
808 bool os::create_thread(Thread* thread, ThreadType thr_type,
809 size_t stack_size) {
810 assert(thread->osthread() == NULL, "caller responsible");
811
812 // Allocate the OSThread object
813 OSThread* osthread = new OSThread(NULL, NULL);
814 if (osthread == NULL) {
815 return false;
816 }
817
818 // set the correct thread state
819 osthread->set_thread_type(thr_type);
820
821 // Initial state is ALLOCATED but not INITIALIZED
822 osthread->set_state(ALLOCATED);
823
824 thread->set_osthread(osthread);
825
826 // init thread attributes
827 pthread_attr_t attr;
828 pthread_attr_init(&attr);
829 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
830
831 // stack size
832 if (os::Linux::supports_variable_stack_size()) {
833 // calculate stack size if it's not specified by caller
834 if (stack_size == 0) {
835 stack_size = os::Linux::default_stack_size(thr_type);
836
837 switch (thr_type) {
838 case os::java_thread:
839 // Java threads use ThreadStackSize which default value can be
840 // changed with the flag -Xss
841 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
842 stack_size = JavaThread::stack_size_at_create();
843 break;
844 case os::compiler_thread:
845 if (CompilerThreadStackSize > 0) {
846 stack_size = (size_t)(CompilerThreadStackSize * K);
847 break;
848 } // else fall through:
849 // use VMThreadStackSize if CompilerThreadStackSize is not defined
850 case os::vm_thread:
851 case os::pgc_thread:
852 case os::cgc_thread:
853 case os::watcher_thread:
854 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
855 break;
856 }
857 }
858
859 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
860 pthread_attr_setstacksize(&attr, stack_size);
861 } else {
862 // let pthread_create() pick the default value.
863 }
864
865 // glibc guard page
866 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
867
868 ThreadState state;
869
870 {
871 // Serialize thread creation if we are running with fixed stack LinuxThreads
872 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
873 if (lock) {
874 os::Linux::createThread_lock()->lock_without_safepoint_check();
875 }
876
877 pthread_t tid;
878 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
879
880 pthread_attr_destroy(&attr);
881
882 if (ret != 0) {
883 if (PrintMiscellaneous && (Verbose || WizardMode)) {
884 perror("pthread_create()");
885 }
886 // Need to clean up stuff we've allocated so far
887 thread->set_osthread(NULL);
888 delete osthread;
889 if (lock) os::Linux::createThread_lock()->unlock();
890 return false;
891 }
892
893 // Store pthread info into the OSThread
894 osthread->set_pthread_id(tid);
895
896 // Wait until child thread is either initialized or aborted
897 {
898 Monitor* sync_with_child = osthread->startThread_lock();
899 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
900 while ((state = osthread->get_state()) == ALLOCATED) {
901 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
902 }
903 }
904
905 if (lock) {
906 os::Linux::createThread_lock()->unlock();
907 }
908 }
909
910 // Aborted due to thread limit being reached
911 if (state == ZOMBIE) {
912 thread->set_osthread(NULL);
913 delete osthread;
914 return false;
915 }
916
917 // The thread is returned suspended (in state INITIALIZED),
918 // and is started higher up in the call chain
919 assert(state == INITIALIZED, "race condition");
920 return true;
921 }
922
923 /////////////////////////////////////////////////////////////////////////////
924 // attach existing thread
925
926 // bootstrap the main thread
927 bool os::create_main_thread(JavaThread* thread) {
928 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
929 return create_attached_thread(thread);
930 }
931
932 bool os::create_attached_thread(JavaThread* thread) {
933 #ifdef ASSERT
934 thread->verify_not_published();
935 #endif
936
937 // Allocate the OSThread object
938 OSThread* osthread = new OSThread(NULL, NULL);
939
940 if (osthread == NULL) {
941 return false;
942 }
943
944 // Store pthread info into the OSThread
945 osthread->set_thread_id(os::Linux::gettid());
946 osthread->set_pthread_id(::pthread_self());
947
948 // initialize floating point control register
949 os::Linux::init_thread_fpu_state();
950
951 // Initial thread state is RUNNABLE
952 osthread->set_state(RUNNABLE);
953
954 thread->set_osthread(osthread);
955
956 if (UseNUMA) {
957 int lgrp_id = os::numa_get_group_id();
958 if (lgrp_id != -1) {
959 thread->set_lgrp_id(lgrp_id);
960 }
961 }
962
963 if (os::Linux::is_initial_thread()) {
964 // If current thread is initial thread, its stack is mapped on demand,
965 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
966 // the entire stack region to avoid SEGV in stack banging.
967 // It is also useful to get around the heap-stack-gap problem on SuSE
968 // kernel (see 4821821 for details). We first expand stack to the top
969 // of yellow zone, then enable stack yellow zone (order is significant,
970 // enabling yellow zone first will crash JVM on SuSE Linux), so there
971 // is no gap between the last two virtual memory regions.
972
973 JavaThread *jt = (JavaThread *)thread;
974 address addr = jt->stack_yellow_zone_base();
975 assert(addr != NULL, "initialization problem?");
976 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
977
978 osthread->set_expanding_stack();
979 os::Linux::manually_expand_stack(jt, addr);
980 osthread->clear_expanding_stack();
981 }
982
983 // initialize signal mask for this thread
984 // and save the caller's signal mask
985 os::Linux::hotspot_sigmask(thread);
986
987 return true;
988 }
989
990 void os::pd_start_thread(Thread* thread) {
991 OSThread * osthread = thread->osthread();
992 assert(osthread->get_state() != INITIALIZED, "just checking");
993 Monitor* sync_with_child = osthread->startThread_lock();
994 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
995 sync_with_child->notify();
996 }
997
998 // Free Linux resources related to the OSThread
999 void os::free_thread(OSThread* osthread) {
1000 assert(osthread != NULL, "osthread not set");
1001
1002 if (Thread::current()->osthread() == osthread) {
1003 // Restore caller's signal mask
1004 sigset_t sigmask = osthread->caller_sigmask();
1005 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1006 }
1007
1008 delete osthread;
1009 }
1010
1011 //////////////////////////////////////////////////////////////////////////////
1012 // thread local storage
1013
1014 // Restore the thread pointer if the destructor is called. This is in case
1015 // someone from JNI code sets up a destructor with pthread_key_create to run
1016 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1017 // will hang or crash. When detachCurrentThread is called the key will be set
1018 // to null and we will not be called again. If detachCurrentThread is never
1019 // called we could loop forever depending on the pthread implementation.
1020 static void restore_thread_pointer(void* p) {
1021 Thread* thread = (Thread*) p;
1022 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1023 }
1024
1025 int os::allocate_thread_local_storage() {
1026 pthread_key_t key;
1027 int rslt = pthread_key_create(&key, restore_thread_pointer);
1028 assert(rslt == 0, "cannot allocate thread local storage");
1029 return (int)key;
1030 }
1031
1032 // Note: This is currently not used by VM, as we don't destroy TLS key
1033 // on VM exit.
1034 void os::free_thread_local_storage(int index) {
1035 int rslt = pthread_key_delete((pthread_key_t)index);
1036 assert(rslt == 0, "invalid index");
1037 }
1038
1039 void os::thread_local_storage_at_put(int index, void* value) {
1040 int rslt = pthread_setspecific((pthread_key_t)index, value);
1041 assert(rslt == 0, "pthread_setspecific failed");
1042 }
1043
1044 extern "C" Thread* get_thread() {
1045 return ThreadLocalStorage::thread();
1046 }
1047
1048 //////////////////////////////////////////////////////////////////////////////
1049 // initial thread
1050
1051 // Check if current thread is the initial thread, similar to Solaris thr_main.
1052 bool os::Linux::is_initial_thread(void) {
1053 char dummy;
1054 // If called before init complete, thread stack bottom will be null.
1055 // Can be called if fatal error occurs before initialization.
1056 if (initial_thread_stack_bottom() == NULL) return false;
1057 assert(initial_thread_stack_bottom() != NULL &&
1058 initial_thread_stack_size() != 0,
1059 "os::init did not locate initial thread's stack region");
1060 if ((address)&dummy >= initial_thread_stack_bottom() &&
1061 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1062 return true;
1063 } else {
1064 return false;
1065 }
1066 }
1067
1068 // Find the virtual memory area that contains addr
1069 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1070 FILE *fp = fopen("/proc/self/maps", "r");
1071 if (fp) {
1072 address low, high;
1073 while (!feof(fp)) {
1074 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1075 if (low <= addr && addr < high) {
1076 if (vma_low) *vma_low = low;
1077 if (vma_high) *vma_high = high;
1078 fclose(fp);
1079 return true;
1080 }
1081 }
1082 for (;;) {
1083 int ch = fgetc(fp);
1084 if (ch == EOF || ch == (int)'\n') break;
1085 }
1086 }
1087 fclose(fp);
1088 }
1089 return false;
1090 }
1091
1092 // Locate initial thread stack. This special handling of initial thread stack
1093 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1094 // bogus value for initial thread.
1095 void os::Linux::capture_initial_stack(size_t max_size) {
1096 // stack size is the easy part, get it from RLIMIT_STACK
1097 size_t stack_size;
1098 struct rlimit rlim;
1099 getrlimit(RLIMIT_STACK, &rlim);
1100 stack_size = rlim.rlim_cur;
1101
1102 // 6308388: a bug in ld.so will relocate its own .data section to the
1103 // lower end of primordial stack; reduce ulimit -s value a little bit
1104 // so we won't install guard page on ld.so's data section.
1105 stack_size -= 2 * page_size();
1106
1107 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1108 // 7.1, in both cases we will get 2G in return value.
1109 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1110 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1111 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1112 // in case other parts in glibc still assumes 2M max stack size.
1113 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1114 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1115 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1116 stack_size = 2 * K * K IA64_ONLY(*2);
1117 }
1118 // Try to figure out where the stack base (top) is. This is harder.
1119 //
1120 // When an application is started, glibc saves the initial stack pointer in
1121 // a global variable "__libc_stack_end", which is then used by system
1122 // libraries. __libc_stack_end should be pretty close to stack top. The
1123 // variable is available since the very early days. However, because it is
1124 // a private interface, it could disappear in the future.
1125 //
1126 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1127 // to __libc_stack_end, it is very close to stack top, but isn't the real
1128 // stack top. Note that /proc may not exist if VM is running as a chroot
1129 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1130 // /proc/<pid>/stat could change in the future (though unlikely).
1131 //
1132 // We try __libc_stack_end first. If that doesn't work, look for
1133 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1134 // as a hint, which should work well in most cases.
1135
1136 uintptr_t stack_start;
1137
1138 // try __libc_stack_end first
1139 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1140 if (p && *p) {
1141 stack_start = *p;
1142 } else {
1143 // see if we can get the start_stack field from /proc/self/stat
1144 FILE *fp;
1145 int pid;
1146 char state;
1147 int ppid;
1148 int pgrp;
1149 int session;
1150 int nr;
1151 int tpgrp;
1152 unsigned long flags;
1153 unsigned long minflt;
1154 unsigned long cminflt;
1155 unsigned long majflt;
1156 unsigned long cmajflt;
1157 unsigned long utime;
1158 unsigned long stime;
1159 long cutime;
1160 long cstime;
1161 long prio;
1162 long nice;
1163 long junk;
1164 long it_real;
1165 uintptr_t start;
1166 uintptr_t vsize;
1167 intptr_t rss;
1168 uintptr_t rsslim;
1169 uintptr_t scodes;
1170 uintptr_t ecode;
1171 int i;
1172
1173 // Figure what the primordial thread stack base is. Code is inspired
1174 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1175 // followed by command name surrounded by parentheses, state, etc.
1176 char stat[2048];
1177 int statlen;
1178
1179 fp = fopen("/proc/self/stat", "r");
1180 if (fp) {
1181 statlen = fread(stat, 1, 2047, fp);
1182 stat[statlen] = '\0';
1183 fclose(fp);
1184
1185 // Skip pid and the command string. Note that we could be dealing with
1186 // weird command names, e.g. user could decide to rename java launcher
1187 // to "java 1.4.2 :)", then the stat file would look like
1188 // 1234 (java 1.4.2 :)) R ... ...
1189 // We don't really need to know the command string, just find the last
1190 // occurrence of ")" and then start parsing from there. See bug 4726580.
1191 char * s = strrchr(stat, ')');
1192
1193 i = 0;
1194 if (s) {
1195 // Skip blank chars
1196 do { s++; } while (s && isspace(*s));
1197
1198 #define _UFM UINTX_FORMAT
1199 #define _DFM INTX_FORMAT
1200
1201 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1202 // 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
1203 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,
1204 &state, // 3 %c
1205 &ppid, // 4 %d
1206 &pgrp, // 5 %d
1207 &session, // 6 %d
1208 &nr, // 7 %d
1209 &tpgrp, // 8 %d
1210 &flags, // 9 %lu
1211 &minflt, // 10 %lu
1212 &cminflt, // 11 %lu
1213 &majflt, // 12 %lu
1214 &cmajflt, // 13 %lu
1215 &utime, // 14 %lu
1216 &stime, // 15 %lu
1217 &cutime, // 16 %ld
1218 &cstime, // 17 %ld
1219 &prio, // 18 %ld
1220 &nice, // 19 %ld
1221 &junk, // 20 %ld
1222 &it_real, // 21 %ld
1223 &start, // 22 UINTX_FORMAT
1224 &vsize, // 23 UINTX_FORMAT
1225 &rss, // 24 INTX_FORMAT
1226 &rsslim, // 25 UINTX_FORMAT
1227 &scodes, // 26 UINTX_FORMAT
1228 &ecode, // 27 UINTX_FORMAT
1229 &stack_start); // 28 UINTX_FORMAT
1230 }
1231
1232 #undef _UFM
1233 #undef _DFM
1234
1235 if (i != 28 - 2) {
1236 assert(false, "Bad conversion from /proc/self/stat");
1237 // product mode - assume we are the initial thread, good luck in the
1238 // embedded case.
1239 warning("Can't detect initial thread stack location - bad conversion");
1240 stack_start = (uintptr_t) &rlim;
1241 }
1242 } else {
1243 // For some reason we can't open /proc/self/stat (for example, running on
1244 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1245 // most cases, so don't abort:
1246 warning("Can't detect initial thread stack location - no /proc/self/stat");
1247 stack_start = (uintptr_t) &rlim;
1248 }
1249 }
1250
1251 // Now we have a pointer (stack_start) very close to the stack top, the
1252 // next thing to do is to figure out the exact location of stack top. We
1253 // can find out the virtual memory area that contains stack_start by
1254 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1255 // and its upper limit is the real stack top. (again, this would fail if
1256 // running inside chroot, because /proc may not exist.)
1257
1258 uintptr_t stack_top;
1259 address low, high;
1260 if (find_vma((address)stack_start, &low, &high)) {
1261 // success, "high" is the true stack top. (ignore "low", because initial
1262 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1263 stack_top = (uintptr_t)high;
1264 } else {
1265 // failed, likely because /proc/self/maps does not exist
1266 warning("Can't detect initial thread stack location - find_vma failed");
1267 // best effort: stack_start is normally within a few pages below the real
1268 // stack top, use it as stack top, and reduce stack size so we won't put
1269 // guard page outside stack.
1270 stack_top = stack_start;
1271 stack_size -= 16 * page_size();
1272 }
1273
1274 // stack_top could be partially down the page so align it
1275 stack_top = align_size_up(stack_top, page_size());
1276
1277 if (max_size && stack_size > max_size) {
1278 _initial_thread_stack_size = max_size;
1279 } else {
1280 _initial_thread_stack_size = stack_size;
1281 }
1282
1283 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1284 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1285 }
1286
1287 ////////////////////////////////////////////////////////////////////////////////
1288 // time support
1289
1290 // Time since start-up in seconds to a fine granularity.
1291 // Used by VMSelfDestructTimer and the MemProfiler.
1292 double os::elapsedTime() {
1293
1294 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1295 }
1296
1297 jlong os::elapsed_counter() {
1298 return javaTimeNanos() - initial_time_count;
1299 }
1300
1301 jlong os::elapsed_frequency() {
1302 return NANOSECS_PER_SEC; // nanosecond resolution
1303 }
1304
1305 bool os::supports_vtime() { return true; }
1306 bool os::enable_vtime() { return false; }
1307 bool os::vtime_enabled() { return false; }
1308
1309 double os::elapsedVTime() {
1310 struct rusage usage;
1311 int retval = getrusage(RUSAGE_THREAD, &usage);
1312 if (retval == 0) {
1313 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);
1314 } else {
1315 // better than nothing, but not much
1316 return elapsedTime();
1317 }
1318 }
1319
1320 jlong os::javaTimeMillis() {
1321 timeval time;
1322 int status = gettimeofday(&time, NULL);
1323 assert(status != -1, "linux error");
1324 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1325 }
1326
1327 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1328 timeval time;
1329 int status = gettimeofday(&time, NULL);
1330 assert(status != -1, "linux error");
1331 seconds = jlong(time.tv_sec);
1332 nanos = jlong(time.tv_usec) * 1000;
1333 }
1334
1335
1336 #ifndef CLOCK_MONOTONIC
1337 #define CLOCK_MONOTONIC (1)
1338 #endif
1339
1340 void os::Linux::clock_init() {
1341 // we do dlopen's in this particular order due to bug in linux
1342 // dynamical loader (see 6348968) leading to crash on exit
1343 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1344 if (handle == NULL) {
1345 handle = dlopen("librt.so", RTLD_LAZY);
1346 }
1347
1348 if (handle) {
1349 int (*clock_getres_func)(clockid_t, struct timespec*) =
1350 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1351 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1352 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1353 if (clock_getres_func && clock_gettime_func) {
1354 // See if monotonic clock is supported by the kernel. Note that some
1355 // early implementations simply return kernel jiffies (updated every
1356 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1357 // for nano time (though the monotonic property is still nice to have).
1358 // It's fixed in newer kernels, however clock_getres() still returns
1359 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1360 // resolution for now. Hopefully as people move to new kernels, this
1361 // won't be a problem.
1362 struct timespec res;
1363 struct timespec tp;
1364 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1365 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1366 // yes, monotonic clock is supported
1367 _clock_gettime = clock_gettime_func;
1368 return;
1369 } else {
1370 // close librt if there is no monotonic clock
1371 dlclose(handle);
1372 }
1373 }
1374 }
1375 warning("No monotonic clock was available - timed services may " \
1376 "be adversely affected if the time-of-day clock changes");
1377 }
1378
1379 #ifndef SYS_clock_getres
1380 #if defined(IA32) || defined(AMD64)
1381 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1382 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1383 #else
1384 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1385 #define sys_clock_getres(x,y) -1
1386 #endif
1387 #else
1388 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1389 #endif
1390
1391 void os::Linux::fast_thread_clock_init() {
1392 if (!UseLinuxPosixThreadCPUClocks) {
1393 return;
1394 }
1395 clockid_t clockid;
1396 struct timespec tp;
1397 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1398 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1399
1400 // Switch to using fast clocks for thread cpu time if
1401 // the sys_clock_getres() returns 0 error code.
1402 // Note, that some kernels may support the current thread
1403 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1404 // returned by the pthread_getcpuclockid().
1405 // If the fast Posix clocks are supported then the sys_clock_getres()
1406 // must return at least tp.tv_sec == 0 which means a resolution
1407 // better than 1 sec. This is extra check for reliability.
1408
1409 if (pthread_getcpuclockid_func &&
1410 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1411 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1412 _supports_fast_thread_cpu_time = true;
1413 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1414 }
1415 }
1416
1417 jlong os::javaTimeNanos() {
1418 if (os::supports_monotonic_clock()) {
1419 struct timespec tp;
1420 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1421 assert(status == 0, "gettime error");
1422 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1423 return result;
1424 } else {
1425 timeval time;
1426 int status = gettimeofday(&time, NULL);
1427 assert(status != -1, "linux error");
1428 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1429 return 1000 * usecs;
1430 }
1431 }
1432
1433 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1434 if (os::supports_monotonic_clock()) {
1435 info_ptr->max_value = ALL_64_BITS;
1436
1437 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1438 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1439 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1440 } else {
1441 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1442 info_ptr->max_value = ALL_64_BITS;
1443
1444 // gettimeofday is a real time clock so it skips
1445 info_ptr->may_skip_backward = true;
1446 info_ptr->may_skip_forward = true;
1447 }
1448
1449 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1450 }
1451
1452 // Return the real, user, and system times in seconds from an
1453 // arbitrary fixed point in the past.
1454 bool os::getTimesSecs(double* process_real_time,
1455 double* process_user_time,
1456 double* process_system_time) {
1457 struct tms ticks;
1458 clock_t real_ticks = times(&ticks);
1459
1460 if (real_ticks == (clock_t) (-1)) {
1461 return false;
1462 } else {
1463 double ticks_per_second = (double) clock_tics_per_sec;
1464 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1465 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1466 *process_real_time = ((double) real_ticks) / ticks_per_second;
1467
1468 return true;
1469 }
1470 }
1471
1472
1473 char * os::local_time_string(char *buf, size_t buflen) {
1474 struct tm t;
1475 time_t long_time;
1476 time(&long_time);
1477 localtime_r(&long_time, &t);
1478 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1479 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1480 t.tm_hour, t.tm_min, t.tm_sec);
1481 return buf;
1482 }
1483
1484 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1485 return localtime_r(clock, res);
1486 }
1487
1488 ////////////////////////////////////////////////////////////////////////////////
1489 // runtime exit support
1490
1491 // Note: os::shutdown() might be called very early during initialization, or
1492 // called from signal handler. Before adding something to os::shutdown(), make
1493 // sure it is async-safe and can handle partially initialized VM.
1494 void os::shutdown() {
1495
1496 // allow PerfMemory to attempt cleanup of any persistent resources
1497 perfMemory_exit();
1498
1499 // needs to remove object in file system
1500 AttachListener::abort();
1501
1502 // flush buffered output, finish log files
1503 ostream_abort();
1504
1505 // Check for abort hook
1506 abort_hook_t abort_hook = Arguments::abort_hook();
1507 if (abort_hook != NULL) {
1508 abort_hook();
1509 }
1510
1511 }
1512
1513 // Note: os::abort() might be called very early during initialization, or
1514 // called from signal handler. Before adding something to os::abort(), make
1515 // sure it is async-safe and can handle partially initialized VM.
1516 void os::abort(bool dump_core) {
1517 os::shutdown();
1518 if (dump_core) {
1519 #ifndef PRODUCT
1520 fdStream out(defaultStream::output_fd());
1521 out.print_raw("Current thread is ");
1522 char buf[16];
1523 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1524 out.print_raw_cr(buf);
1525 out.print_raw_cr("Dumping core ...");
1526 #endif
1527 ::abort(); // dump core
1528 }
1529
1530 ::exit(1);
1531 }
1532
1533 // Die immediately, no exit hook, no abort hook, no cleanup.
1534 void os::die() {
1535 // _exit() on LinuxThreads only kills current thread
1536 ::abort();
1537 }
1538
1539
1540 // This method is a copy of JDK's sysGetLastErrorString
1541 // from src/solaris/hpi/src/system_md.c
1542
1543 size_t os::lasterror(char *buf, size_t len) {
1544 if (errno == 0) return 0;
1545
1546 const char *s = ::strerror(errno);
1547 size_t n = ::strlen(s);
1548 if (n >= len) {
1549 n = len - 1;
1550 }
1551 ::strncpy(buf, s, n);
1552 buf[n] = '\0';
1553 return n;
1554 }
1555
1556 intx os::current_thread_id() { return (intx)pthread_self(); }
1557 int os::current_process_id() {
1558
1559 // Under the old linux thread library, linux gives each thread
1560 // its own process id. Because of this each thread will return
1561 // a different pid if this method were to return the result
1562 // of getpid(2). Linux provides no api that returns the pid
1563 // of the launcher thread for the vm. This implementation
1564 // returns a unique pid, the pid of the launcher thread
1565 // that starts the vm 'process'.
1566
1567 // Under the NPTL, getpid() returns the same pid as the
1568 // launcher thread rather than a unique pid per thread.
1569 // Use gettid() if you want the old pre NPTL behaviour.
1570
1571 // if you are looking for the result of a call to getpid() that
1572 // returns a unique pid for the calling thread, then look at the
1573 // OSThread::thread_id() method in osThread_linux.hpp file
1574
1575 return (int)(_initial_pid ? _initial_pid : getpid());
1576 }
1577
1578 // DLL functions
1579
1580 const char* os::dll_file_extension() { return ".so"; }
1581
1582 // This must be hard coded because it's the system's temporary
1583 // directory not the java application's temp directory, ala java.io.tmpdir.
1584 const char* os::get_temp_directory() { return "/tmp"; }
1585
1586 static bool file_exists(const char* filename) {
1587 struct stat statbuf;
1588 if (filename == NULL || strlen(filename) == 0) {
1589 return false;
1590 }
1591 return os::stat(filename, &statbuf) == 0;
1592 }
1593
1594 bool os::dll_build_name(char* buffer, size_t buflen,
1595 const char* pname, const char* fname) {
1596 bool retval = false;
1597 // Copied from libhpi
1598 const size_t pnamelen = pname ? strlen(pname) : 0;
1599
1600 // Return error on buffer overflow.
1601 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1602 return retval;
1603 }
1604
1605 if (pnamelen == 0) {
1606 snprintf(buffer, buflen, "lib%s.so", fname);
1607 retval = true;
1608 } else if (strchr(pname, *os::path_separator()) != NULL) {
1609 int n;
1610 char** pelements = split_path(pname, &n);
1611 if (pelements == NULL) {
1612 return false;
1613 }
1614 for (int i = 0; i < n; i++) {
1615 // Really shouldn't be NULL, but check can't hurt
1616 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1617 continue; // skip the empty path values
1618 }
1619 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1620 if (file_exists(buffer)) {
1621 retval = true;
1622 break;
1623 }
1624 }
1625 // release the storage
1626 for (int i = 0; i < n; i++) {
1627 if (pelements[i] != NULL) {
1628 FREE_C_HEAP_ARRAY(char, pelements[i]);
1629 }
1630 }
1631 if (pelements != NULL) {
1632 FREE_C_HEAP_ARRAY(char*, pelements);
1633 }
1634 } else {
1635 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1636 retval = true;
1637 }
1638 return retval;
1639 }
1640
1641 // check if addr is inside libjvm.so
1642 bool os::address_is_in_vm(address addr) {
1643 static address libjvm_base_addr;
1644 Dl_info dlinfo;
1645
1646 if (libjvm_base_addr == NULL) {
1647 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1648 libjvm_base_addr = (address)dlinfo.dli_fbase;
1649 }
1650 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1651 }
1652
1653 if (dladdr((void *)addr, &dlinfo) != 0) {
1654 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1655 }
1656
1657 return false;
1658 }
1659
1660 bool os::dll_address_to_function_name(address addr, char *buf,
1661 int buflen, int *offset) {
1662 // buf is not optional, but offset is optional
1663 assert(buf != NULL, "sanity check");
1664
1665 Dl_info dlinfo;
1666
1667 if (dladdr((void*)addr, &dlinfo) != 0) {
1668 // see if we have a matching symbol
1669 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1670 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1671 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1672 }
1673 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1674 return true;
1675 }
1676 // no matching symbol so try for just file info
1677 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1678 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1679 buf, buflen, offset, dlinfo.dli_fname)) {
1680 return true;
1681 }
1682 }
1683 }
1684
1685 buf[0] = '\0';
1686 if (offset != NULL) *offset = -1;
1687 return false;
1688 }
1689
1690 struct _address_to_library_name {
1691 address addr; // input : memory address
1692 size_t buflen; // size of fname
1693 char* fname; // output: library name
1694 address base; // library base addr
1695 };
1696
1697 static int address_to_library_name_callback(struct dl_phdr_info *info,
1698 size_t size, void *data) {
1699 int i;
1700 bool found = false;
1701 address libbase = NULL;
1702 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1703
1704 // iterate through all loadable segments
1705 for (i = 0; i < info->dlpi_phnum; i++) {
1706 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1707 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1708 // base address of a library is the lowest address of its loaded
1709 // segments.
1710 if (libbase == NULL || libbase > segbase) {
1711 libbase = segbase;
1712 }
1713 // see if 'addr' is within current segment
1714 if (segbase <= d->addr &&
1715 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1716 found = true;
1717 }
1718 }
1719 }
1720
1721 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1722 // so dll_address_to_library_name() can fall through to use dladdr() which
1723 // can figure out executable name from argv[0].
1724 if (found && info->dlpi_name && info->dlpi_name[0]) {
1725 d->base = libbase;
1726 if (d->fname) {
1727 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1728 }
1729 return 1;
1730 }
1731 return 0;
1732 }
1733
1734 bool os::dll_address_to_library_name(address addr, char* buf,
1735 int buflen, int* offset) {
1736 // buf is not optional, but offset is optional
1737 assert(buf != NULL, "sanity check");
1738
1739 Dl_info dlinfo;
1740 struct _address_to_library_name data;
1741
1742 // There is a bug in old glibc dladdr() implementation that it could resolve
1743 // to wrong library name if the .so file has a base address != NULL. Here
1744 // we iterate through the program headers of all loaded libraries to find
1745 // out which library 'addr' really belongs to. This workaround can be
1746 // removed once the minimum requirement for glibc is moved to 2.3.x.
1747 data.addr = addr;
1748 data.fname = buf;
1749 data.buflen = buflen;
1750 data.base = NULL;
1751 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1752
1753 if (rslt) {
1754 // buf already contains library name
1755 if (offset) *offset = addr - data.base;
1756 return true;
1757 }
1758 if (dladdr((void*)addr, &dlinfo) != 0) {
1759 if (dlinfo.dli_fname != NULL) {
1760 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1761 }
1762 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1763 *offset = addr - (address)dlinfo.dli_fbase;
1764 }
1765 return true;
1766 }
1767
1768 buf[0] = '\0';
1769 if (offset) *offset = -1;
1770 return false;
1771 }
1772
1773 // Loads .dll/.so and
1774 // in case of error it checks if .dll/.so was built for the
1775 // same architecture as Hotspot is running on
1776
1777
1778 // Remember the stack's state. The Linux dynamic linker will change
1779 // the stack to 'executable' at most once, so we must safepoint only once.
1780 bool os::Linux::_stack_is_executable = false;
1781
1782 // VM operation that loads a library. This is necessary if stack protection
1783 // of the Java stacks can be lost during loading the library. If we
1784 // do not stop the Java threads, they can stack overflow before the stacks
1785 // are protected again.
1786 class VM_LinuxDllLoad: public VM_Operation {
1787 private:
1788 const char *_filename;
1789 char *_ebuf;
1790 int _ebuflen;
1791 void *_lib;
1792 public:
1793 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1794 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1795 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1796 void doit() {
1797 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1798 os::Linux::_stack_is_executable = true;
1799 }
1800 void* loaded_library() { return _lib; }
1801 };
1802
1803 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1804 void * result = NULL;
1805 bool load_attempted = false;
1806
1807 // Check whether the library to load might change execution rights
1808 // of the stack. If they are changed, the protection of the stack
1809 // guard pages will be lost. We need a safepoint to fix this.
1810 //
1811 // See Linux man page execstack(8) for more info.
1812 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1813 ElfFile ef(filename);
1814 if (!ef.specifies_noexecstack()) {
1815 if (!is_init_completed()) {
1816 os::Linux::_stack_is_executable = true;
1817 // This is OK - No Java threads have been created yet, and hence no
1818 // stack guard pages to fix.
1819 //
1820 // This should happen only when you are building JDK7 using a very
1821 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1822 //
1823 // Dynamic loader will make all stacks executable after
1824 // this function returns, and will not do that again.
1825 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1826 } else {
1827 warning("You have loaded library %s which might have disabled stack guard. "
1828 "The VM will try to fix the stack guard now.\n"
1829 "It's highly recommended that you fix the library with "
1830 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1831 filename);
1832
1833 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1834 JavaThread *jt = JavaThread::current();
1835 if (jt->thread_state() != _thread_in_native) {
1836 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1837 // that requires ExecStack. Cannot enter safe point. Let's give up.
1838 warning("Unable to fix stack guard. Giving up.");
1839 } else {
1840 if (!LoadExecStackDllInVMThread) {
1841 // This is for the case where the DLL has an static
1842 // constructor function that executes JNI code. We cannot
1843 // load such DLLs in the VMThread.
1844 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1845 }
1846
1847 ThreadInVMfromNative tiv(jt);
1848 debug_only(VMNativeEntryWrapper vew;)
1849
1850 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1851 VMThread::execute(&op);
1852 if (LoadExecStackDllInVMThread) {
1853 result = op.loaded_library();
1854 }
1855 load_attempted = true;
1856 }
1857 }
1858 }
1859 }
1860
1861 if (!load_attempted) {
1862 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1863 }
1864
1865 if (result != NULL) {
1866 // Successful loading
1867 return result;
1868 }
1869
1870 Elf32_Ehdr elf_head;
1871 int diag_msg_max_length=ebuflen-strlen(ebuf);
1872 char* diag_msg_buf=ebuf+strlen(ebuf);
1873
1874 if (diag_msg_max_length==0) {
1875 // No more space in ebuf for additional diagnostics message
1876 return NULL;
1877 }
1878
1879
1880 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1881
1882 if (file_descriptor < 0) {
1883 // Can't open library, report dlerror() message
1884 return NULL;
1885 }
1886
1887 bool failed_to_read_elf_head=
1888 (sizeof(elf_head)!=
1889 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1890
1891 ::close(file_descriptor);
1892 if (failed_to_read_elf_head) {
1893 // file i/o error - report dlerror() msg
1894 return NULL;
1895 }
1896
1897 typedef struct {
1898 Elf32_Half code; // Actual value as defined in elf.h
1899 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1900 char elf_class; // 32 or 64 bit
1901 char endianess; // MSB or LSB
1902 char* name; // String representation
1903 } arch_t;
1904
1905 #ifndef EM_486
1906 #define EM_486 6 /* Intel 80486 */
1907 #endif
1908 #ifndef EM_AARCH64
1909 #define EM_AARCH64 183 /* ARM AARCH64 */
1910 #endif
1911
1912 static const arch_t arch_array[]={
1913 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1914 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1915 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1916 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1917 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1918 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1919 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1920 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1921 #if defined(VM_LITTLE_ENDIAN)
1922 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1923 #else
1924 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1925 #endif
1926 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1927 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1928 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1929 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1930 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1931 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1932 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1933 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1934 };
1935
1936 #if (defined IA32)
1937 static Elf32_Half running_arch_code=EM_386;
1938 #elif (defined AMD64)
1939 static Elf32_Half running_arch_code=EM_X86_64;
1940 #elif (defined IA64)
1941 static Elf32_Half running_arch_code=EM_IA_64;
1942 #elif (defined __sparc) && (defined _LP64)
1943 static Elf32_Half running_arch_code=EM_SPARCV9;
1944 #elif (defined __sparc) && (!defined _LP64)
1945 static Elf32_Half running_arch_code=EM_SPARC;
1946 #elif (defined __powerpc64__)
1947 static Elf32_Half running_arch_code=EM_PPC64;
1948 #elif (defined __powerpc__)
1949 static Elf32_Half running_arch_code=EM_PPC;
1950 #elif (defined ARM)
1951 static Elf32_Half running_arch_code=EM_ARM;
1952 #elif (defined S390)
1953 static Elf32_Half running_arch_code=EM_S390;
1954 #elif (defined ALPHA)
1955 static Elf32_Half running_arch_code=EM_ALPHA;
1956 #elif (defined MIPSEL)
1957 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1958 #elif (defined PARISC)
1959 static Elf32_Half running_arch_code=EM_PARISC;
1960 #elif (defined MIPS)
1961 static Elf32_Half running_arch_code=EM_MIPS;
1962 #elif (defined M68K)
1963 static Elf32_Half running_arch_code=EM_68K;
1964 #elif (defined AARCH64)
1965 static Elf32_Half running_arch_code=EM_AARCH64;
1966 #else
1967 #error Method os::dll_load requires that one of following is defined:\
1968 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1969 #endif
1970
1971 // Identify compatability class for VM's architecture and library's architecture
1972 // Obtain string descriptions for architectures
1973
1974 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1975 int running_arch_index=-1;
1976
1977 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1978 if (running_arch_code == arch_array[i].code) {
1979 running_arch_index = i;
1980 }
1981 if (lib_arch.code == arch_array[i].code) {
1982 lib_arch.compat_class = arch_array[i].compat_class;
1983 lib_arch.name = arch_array[i].name;
1984 }
1985 }
1986
1987 assert(running_arch_index != -1,
1988 "Didn't find running architecture code (running_arch_code) in arch_array");
1989 if (running_arch_index == -1) {
1990 // Even though running architecture detection failed
1991 // we may still continue with reporting dlerror() message
1992 return NULL;
1993 }
1994
1995 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1996 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1997 return NULL;
1998 }
1999
2000 #ifndef S390
2001 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2002 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2003 return NULL;
2004 }
2005 #endif // !S390
2006
2007 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2008 if (lib_arch.name!=NULL) {
2009 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2010 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2011 lib_arch.name, arch_array[running_arch_index].name);
2012 } else {
2013 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2014 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2015 lib_arch.code,
2016 arch_array[running_arch_index].name);
2017 }
2018 }
2019
2020 return NULL;
2021 }
2022
2023 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2024 int ebuflen) {
2025 void * result = ::dlopen(filename, RTLD_LAZY);
2026 if (result == NULL) {
2027 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2028 ebuf[ebuflen-1] = '\0';
2029 }
2030 return result;
2031 }
2032
2033 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2034 int ebuflen) {
2035 void * result = NULL;
2036 if (LoadExecStackDllInVMThread) {
2037 result = dlopen_helper(filename, ebuf, ebuflen);
2038 }
2039
2040 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2041 // library that requires an executable stack, or which does not have this
2042 // stack attribute set, dlopen changes the stack attribute to executable. The
2043 // read protection of the guard pages gets lost.
2044 //
2045 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2046 // may have been queued at the same time.
2047
2048 if (!_stack_is_executable) {
2049 JavaThread *jt = Threads::first();
2050
2051 while (jt) {
2052 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2053 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2054 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2055 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2056 warning("Attempt to reguard stack yellow zone failed.");
2057 }
2058 }
2059 jt = jt->next();
2060 }
2061 }
2062
2063 return result;
2064 }
2065
2066 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2067 // chances are you might want to run the generated bits against glibc-2.0
2068 // libdl.so, so always use locking for any version of glibc.
2069 //
2070 void* os::dll_lookup(void* handle, const char* name) {
2071 pthread_mutex_lock(&dl_mutex);
2072 void* res = dlsym(handle, name);
2073 pthread_mutex_unlock(&dl_mutex);
2074 return res;
2075 }
2076
2077 void* os::get_default_process_handle() {
2078 return (void*)::dlopen(NULL, RTLD_LAZY);
2079 }
2080
2081 static bool _print_ascii_file(const char* filename, outputStream* st) {
2082 int fd = ::open(filename, O_RDONLY);
2083 if (fd == -1) {
2084 return false;
2085 }
2086
2087 char buf[32];
2088 int bytes;
2089 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2090 st->print_raw(buf, bytes);
2091 }
2092
2093 ::close(fd);
2094
2095 return true;
2096 }
2097
2098 void os::print_dll_info(outputStream *st) {
2099 st->print_cr("Dynamic libraries:");
2100
2101 char fname[32];
2102 pid_t pid = os::Linux::gettid();
2103
2104 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2105
2106 if (!_print_ascii_file(fname, st)) {
2107 st->print("Can not get library information for pid = %d\n", pid);
2108 }
2109 }
2110
2111 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2112 FILE *procmapsFile = NULL;
2113
2114 // Open the procfs maps file for the current process
2115 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2116 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2117 char line[PATH_MAX + 100];
2118
2119 // Read line by line from 'file'
2120 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2121 u8 base, top, offset, inode;
2122 char permissions[5];
2123 char device[6];
2124 char name[PATH_MAX + 1];
2125
2126 // Parse fields from line
2127 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2128
2129 // Filter by device id '00:00' so that we only get file system mapped files.
2130 if (strcmp(device, "00:00") != 0) {
2131
2132 // Call callback with the fields of interest
2133 if(callback(name, (address)base, (address)top, param)) {
2134 // Oops abort, callback aborted
2135 fclose(procmapsFile);
2136 return 1;
2137 }
2138 }
2139 }
2140 fclose(procmapsFile);
2141 }
2142 return 0;
2143 }
2144
2145 void os::print_os_info_brief(outputStream* st) {
2146 os::Linux::print_distro_info(st);
2147
2148 os::Posix::print_uname_info(st);
2149
2150 os::Linux::print_libversion_info(st);
2151
2152 }
2153
2154 void os::print_os_info(outputStream* st) {
2155 st->print("OS:");
2156
2157 os::Linux::print_distro_info(st);
2158
2159 os::Posix::print_uname_info(st);
2160
2161 // Print warning if unsafe chroot environment detected
2162 if (unsafe_chroot_detected) {
2163 st->print("WARNING!! ");
2164 st->print_cr("%s", unstable_chroot_error);
2165 }
2166
2167 os::Linux::print_libversion_info(st);
2168
2169 os::Posix::print_rlimit_info(st);
2170
2171 os::Posix::print_load_average(st);
2172
2173 os::Linux::print_full_memory_info(st);
2174 }
2175
2176 // Try to identify popular distros.
2177 // Most Linux distributions have a /etc/XXX-release file, which contains
2178 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2179 // file that also contains the OS version string. Some have more than one
2180 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2181 // /etc/redhat-release.), so the order is important.
2182 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2183 // their own specific XXX-release file as well as a redhat-release file.
2184 // Because of this the XXX-release file needs to be searched for before the
2185 // redhat-release file.
2186 // Since Red Hat has a lsb-release file that is not very descriptive the
2187 // search for redhat-release needs to be before lsb-release.
2188 // Since the lsb-release file is the new standard it needs to be searched
2189 // before the older style release files.
2190 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2191 // next to last resort. The os-release file is a new standard that contains
2192 // distribution information and the system-release file seems to be an old
2193 // standard that has been replaced by the lsb-release and os-release files.
2194 // Searching for the debian_version file is the last resort. It contains
2195 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2196 // "Debian " is printed before the contents of the debian_version file.
2197 void os::Linux::print_distro_info(outputStream* st) {
2198 if (!_print_ascii_file("/etc/oracle-release", st) &&
2199 !_print_ascii_file("/etc/mandriva-release", st) &&
2200 !_print_ascii_file("/etc/mandrake-release", st) &&
2201 !_print_ascii_file("/etc/sun-release", st) &&
2202 !_print_ascii_file("/etc/redhat-release", st) &&
2203 !_print_ascii_file("/etc/lsb-release", st) &&
2204 !_print_ascii_file("/etc/SuSE-release", st) &&
2205 !_print_ascii_file("/etc/turbolinux-release", st) &&
2206 !_print_ascii_file("/etc/gentoo-release", st) &&
2207 !_print_ascii_file("/etc/ltib-release", st) &&
2208 !_print_ascii_file("/etc/angstrom-version", st) &&
2209 !_print_ascii_file("/etc/system-release", st) &&
2210 !_print_ascii_file("/etc/os-release", st)) {
2211
2212 if (file_exists("/etc/debian_version")) {
2213 st->print("Debian ");
2214 _print_ascii_file("/etc/debian_version", st);
2215 } else {
2216 st->print("Linux");
2217 }
2218 }
2219 st->cr();
2220 }
2221
2222 void os::Linux::print_libversion_info(outputStream* st) {
2223 // libc, pthread
2224 st->print("libc:");
2225 st->print("%s ", os::Linux::glibc_version());
2226 st->print("%s ", os::Linux::libpthread_version());
2227 if (os::Linux::is_LinuxThreads()) {
2228 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2229 }
2230 st->cr();
2231 }
2232
2233 void os::Linux::print_full_memory_info(outputStream* st) {
2234 st->print("\n/proc/meminfo:\n");
2235 _print_ascii_file("/proc/meminfo", st);
2236 st->cr();
2237 }
2238
2239 void os::print_memory_info(outputStream* st) {
2240
2241 st->print("Memory:");
2242 st->print(" %dk page", os::vm_page_size()>>10);
2243
2244 // values in struct sysinfo are "unsigned long"
2245 struct sysinfo si;
2246 sysinfo(&si);
2247
2248 st->print(", physical " UINT64_FORMAT "k",
2249 os::physical_memory() >> 10);
2250 st->print("(" UINT64_FORMAT "k free)",
2251 os::available_memory() >> 10);
2252 st->print(", swap " UINT64_FORMAT "k",
2253 ((jlong)si.totalswap * si.mem_unit) >> 10);
2254 st->print("(" UINT64_FORMAT "k free)",
2255 ((jlong)si.freeswap * si.mem_unit) >> 10);
2256 st->cr();
2257 }
2258
2259 void os::pd_print_cpu_info(outputStream* st) {
2260 st->print("\n/proc/cpuinfo:\n");
2261 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2262 st->print(" <Not Available>");
2263 }
2264 st->cr();
2265 }
2266
2267 void os::print_siginfo(outputStream* st, void* siginfo) {
2268 const siginfo_t* si = (const siginfo_t*)siginfo;
2269
2270 os::Posix::print_siginfo_brief(st, si);
2271 #if INCLUDE_CDS
2272 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2273 UseSharedSpaces) {
2274 FileMapInfo* mapinfo = FileMapInfo::current_info();
2275 if (mapinfo->is_in_shared_space(si->si_addr)) {
2276 st->print("\n\nError accessing class data sharing archive." \
2277 " Mapped file inaccessible during execution, " \
2278 " possible disk/network problem.");
2279 }
2280 }
2281 #endif
2282 st->cr();
2283 }
2284
2285
2286 static void print_signal_handler(outputStream* st, int sig,
2287 char* buf, size_t buflen);
2288
2289 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2290 st->print_cr("Signal Handlers:");
2291 print_signal_handler(st, SIGSEGV, buf, buflen);
2292 print_signal_handler(st, SIGBUS , buf, buflen);
2293 print_signal_handler(st, SIGFPE , buf, buflen);
2294 print_signal_handler(st, SIGPIPE, buf, buflen);
2295 print_signal_handler(st, SIGXFSZ, buf, buflen);
2296 print_signal_handler(st, SIGILL , buf, buflen);
2297 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2298 print_signal_handler(st, SR_signum, buf, buflen);
2299 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2300 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2301 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2302 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2303 #if defined(PPC64)
2304 print_signal_handler(st, SIGTRAP, buf, buflen);
2305 #endif
2306 }
2307
2308 static char saved_jvm_path[MAXPATHLEN] = {0};
2309
2310 // Find the full path to the current module, libjvm.so
2311 void os::jvm_path(char *buf, jint buflen) {
2312 // Error checking.
2313 if (buflen < MAXPATHLEN) {
2314 assert(false, "must use a large-enough buffer");
2315 buf[0] = '\0';
2316 return;
2317 }
2318 // Lazy resolve the path to current module.
2319 if (saved_jvm_path[0] != 0) {
2320 strcpy(buf, saved_jvm_path);
2321 return;
2322 }
2323
2324 char dli_fname[MAXPATHLEN];
2325 bool ret = dll_address_to_library_name(
2326 CAST_FROM_FN_PTR(address, os::jvm_path),
2327 dli_fname, sizeof(dli_fname), NULL);
2328 assert(ret, "cannot locate libjvm");
2329 char *rp = NULL;
2330 if (ret && dli_fname[0] != '\0') {
2331 rp = realpath(dli_fname, buf);
2332 }
2333 if (rp == NULL) {
2334 return;
2335 }
2336
2337 if (Arguments::sun_java_launcher_is_altjvm()) {
2338 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2339 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2340 // If "/jre/lib/" appears at the right place in the string, then
2341 // assume we are installed in a JDK and we're done. Otherwise, check
2342 // for a JAVA_HOME environment variable and fix up the path so it
2343 // looks like libjvm.so is installed there (append a fake suffix
2344 // hotspot/libjvm.so).
2345 const char *p = buf + strlen(buf) - 1;
2346 for (int count = 0; p > buf && count < 5; ++count) {
2347 for (--p; p > buf && *p != '/'; --p)
2348 /* empty */ ;
2349 }
2350
2351 if (strncmp(p, "/jre/lib/", 9) != 0) {
2352 // Look for JAVA_HOME in the environment.
2353 char* java_home_var = ::getenv("JAVA_HOME");
2354 if (java_home_var != NULL && java_home_var[0] != 0) {
2355 char* jrelib_p;
2356 int len;
2357
2358 // Check the current module name "libjvm.so".
2359 p = strrchr(buf, '/');
2360 if (p == NULL) {
2361 return;
2362 }
2363 assert(strstr(p, "/libjvm") == p, "invalid library name");
2364
2365 rp = realpath(java_home_var, buf);
2366 if (rp == NULL) {
2367 return;
2368 }
2369
2370 // determine if this is a legacy image or modules image
2371 // modules image doesn't have "jre" subdirectory
2372 len = strlen(buf);
2373 assert(len < buflen, "Ran out of buffer room");
2374 jrelib_p = buf + len;
2375 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2376 if (0 != access(buf, F_OK)) {
2377 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2378 }
2379
2380 if (0 == access(buf, F_OK)) {
2381 // Use current module name "libjvm.so"
2382 len = strlen(buf);
2383 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2384 } else {
2385 // Go back to path of .so
2386 rp = realpath(dli_fname, buf);
2387 if (rp == NULL) {
2388 return;
2389 }
2390 }
2391 }
2392 }
2393 }
2394
2395 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2396 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2397 }
2398
2399 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2400 // no prefix required, not even "_"
2401 }
2402
2403 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2404 // no suffix required
2405 }
2406
2407 ////////////////////////////////////////////////////////////////////////////////
2408 // sun.misc.Signal support
2409
2410 static volatile jint sigint_count = 0;
2411
2412 static void UserHandler(int sig, void *siginfo, void *context) {
2413 // 4511530 - sem_post is serialized and handled by the manager thread. When
2414 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2415 // don't want to flood the manager thread with sem_post requests.
2416 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2417 return;
2418 }
2419
2420 // Ctrl-C is pressed during error reporting, likely because the error
2421 // handler fails to abort. Let VM die immediately.
2422 if (sig == SIGINT && is_error_reported()) {
2423 os::die();
2424 }
2425
2426 os::signal_notify(sig);
2427 }
2428
2429 void* os::user_handler() {
2430 return CAST_FROM_FN_PTR(void*, UserHandler);
2431 }
2432
2433 class Semaphore : public StackObj {
2434 public:
2435 Semaphore();
2436 ~Semaphore();
2437 void signal();
2438 void wait();
2439 bool trywait();
2440 bool timedwait(unsigned int sec, int nsec);
2441 private:
2442 sem_t _semaphore;
2443 };
2444
2445 Semaphore::Semaphore() {
2446 sem_init(&_semaphore, 0, 0);
2447 }
2448
2449 Semaphore::~Semaphore() {
2450 sem_destroy(&_semaphore);
2451 }
2452
2453 void Semaphore::signal() {
2454 sem_post(&_semaphore);
2455 }
2456
2457 void Semaphore::wait() {
2458 sem_wait(&_semaphore);
2459 }
2460
2461 bool Semaphore::trywait() {
2462 return sem_trywait(&_semaphore) == 0;
2463 }
2464
2465 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2466
2467 struct timespec ts;
2468 // Semaphore's are always associated with CLOCK_REALTIME
2469 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2470 // see unpackTime for discussion on overflow checking
2471 if (sec >= MAX_SECS) {
2472 ts.tv_sec += MAX_SECS;
2473 ts.tv_nsec = 0;
2474 } else {
2475 ts.tv_sec += sec;
2476 ts.tv_nsec += nsec;
2477 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2478 ts.tv_nsec -= NANOSECS_PER_SEC;
2479 ++ts.tv_sec; // note: this must be <= max_secs
2480 }
2481 }
2482
2483 while (1) {
2484 int result = sem_timedwait(&_semaphore, &ts);
2485 if (result == 0) {
2486 return true;
2487 } else if (errno == EINTR) {
2488 continue;
2489 } else if (errno == ETIMEDOUT) {
2490 return false;
2491 } else {
2492 return false;
2493 }
2494 }
2495 }
2496
2497 extern "C" {
2498 typedef void (*sa_handler_t)(int);
2499 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2500 }
2501
2502 void* os::signal(int signal_number, void* handler) {
2503 struct sigaction sigAct, oldSigAct;
2504
2505 sigfillset(&(sigAct.sa_mask));
2506 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2507 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2508
2509 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2510 // -1 means registration failed
2511 return (void *)-1;
2512 }
2513
2514 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2515 }
2516
2517 void os::signal_raise(int signal_number) {
2518 ::raise(signal_number);
2519 }
2520
2521 // The following code is moved from os.cpp for making this
2522 // code platform specific, which it is by its very nature.
2523
2524 // Will be modified when max signal is changed to be dynamic
2525 int os::sigexitnum_pd() {
2526 return NSIG;
2527 }
2528
2529 // a counter for each possible signal value
2530 static volatile jint pending_signals[NSIG+1] = { 0 };
2531
2532 // Linux(POSIX) specific hand shaking semaphore.
2533 static sem_t sig_sem;
2534 static Semaphore sr_semaphore;
2535
2536 void os::signal_init_pd() {
2537 // Initialize signal structures
2538 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2539
2540 // Initialize signal semaphore
2541 ::sem_init(&sig_sem, 0, 0);
2542 }
2543
2544 void os::signal_notify(int sig) {
2545 Atomic::inc(&pending_signals[sig]);
2546 ::sem_post(&sig_sem);
2547 }
2548
2549 static int check_pending_signals(bool wait) {
2550 Atomic::store(0, &sigint_count);
2551 for (;;) {
2552 for (int i = 0; i < NSIG + 1; i++) {
2553 jint n = pending_signals[i];
2554 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2555 return i;
2556 }
2557 }
2558 if (!wait) {
2559 return -1;
2560 }
2561 JavaThread *thread = JavaThread::current();
2562 ThreadBlockInVM tbivm(thread);
2563
2564 bool threadIsSuspended;
2565 do {
2566 thread->set_suspend_equivalent();
2567 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2568 ::sem_wait(&sig_sem);
2569
2570 // were we externally suspended while we were waiting?
2571 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2572 if (threadIsSuspended) {
2573 // The semaphore has been incremented, but while we were waiting
2574 // another thread suspended us. We don't want to continue running
2575 // while suspended because that would surprise the thread that
2576 // suspended us.
2577 ::sem_post(&sig_sem);
2578
2579 thread->java_suspend_self();
2580 }
2581 } while (threadIsSuspended);
2582 }
2583 }
2584
2585 int os::signal_lookup() {
2586 return check_pending_signals(false);
2587 }
2588
2589 int os::signal_wait() {
2590 return check_pending_signals(true);
2591 }
2592
2593 ////////////////////////////////////////////////////////////////////////////////
2594 // Virtual Memory
2595
2596 int os::vm_page_size() {
2597 // Seems redundant as all get out
2598 assert(os::Linux::page_size() != -1, "must call os::init");
2599 return os::Linux::page_size();
2600 }
2601
2602 // Solaris allocates memory by pages.
2603 int os::vm_allocation_granularity() {
2604 assert(os::Linux::page_size() != -1, "must call os::init");
2605 return os::Linux::page_size();
2606 }
2607
2608 // Rationale behind this function:
2609 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2610 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2611 // samples for JITted code. Here we create private executable mapping over the code cache
2612 // and then we can use standard (well, almost, as mapping can change) way to provide
2613 // info for the reporting script by storing timestamp and location of symbol
2614 void linux_wrap_code(char* base, size_t size) {
2615 static volatile jint cnt = 0;
2616
2617 if (!UseOprofile) {
2618 return;
2619 }
2620
2621 char buf[PATH_MAX+1];
2622 int num = Atomic::add(1, &cnt);
2623
2624 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2625 os::get_temp_directory(), os::current_process_id(), num);
2626 unlink(buf);
2627
2628 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2629
2630 if (fd != -1) {
2631 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2632 if (rv != (off_t)-1) {
2633 if (::write(fd, "", 1) == 1) {
2634 mmap(base, size,
2635 PROT_READ|PROT_WRITE|PROT_EXEC,
2636 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2637 }
2638 }
2639 ::close(fd);
2640 unlink(buf);
2641 }
2642 }
2643
2644 static bool recoverable_mmap_error(int err) {
2645 // See if the error is one we can let the caller handle. This
2646 // list of errno values comes from JBS-6843484. I can't find a
2647 // Linux man page that documents this specific set of errno
2648 // values so while this list currently matches Solaris, it may
2649 // change as we gain experience with this failure mode.
2650 switch (err) {
2651 case EBADF:
2652 case EINVAL:
2653 case ENOTSUP:
2654 // let the caller deal with these errors
2655 return true;
2656
2657 default:
2658 // Any remaining errors on this OS can cause our reserved mapping
2659 // to be lost. That can cause confusion where different data
2660 // structures think they have the same memory mapped. The worst
2661 // scenario is if both the VM and a library think they have the
2662 // same memory mapped.
2663 return false;
2664 }
2665 }
2666
2667 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2668 int err) {
2669 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2670 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2671 strerror(err), err);
2672 }
2673
2674 static void warn_fail_commit_memory(char* addr, size_t size,
2675 size_t alignment_hint, bool exec,
2676 int err) {
2677 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2678 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2679 alignment_hint, exec, strerror(err), err);
2680 }
2681
2682 // NOTE: Linux kernel does not really reserve the pages for us.
2683 // All it does is to check if there are enough free pages
2684 // left at the time of mmap(). This could be a potential
2685 // problem.
2686 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2687 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2688 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2689 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2690 if (res != (uintptr_t) MAP_FAILED) {
2691 if (UseNUMAInterleaving) {
2692 numa_make_global(addr, size);
2693 }
2694 return 0;
2695 }
2696
2697 int err = errno; // save errno from mmap() call above
2698
2699 if (!recoverable_mmap_error(err)) {
2700 warn_fail_commit_memory(addr, size, exec, err);
2701 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2702 }
2703
2704 return err;
2705 }
2706
2707 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2708 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2709 }
2710
2711 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2712 const char* mesg) {
2713 assert(mesg != NULL, "mesg must be specified");
2714 int err = os::Linux::commit_memory_impl(addr, size, exec);
2715 if (err != 0) {
2716 // the caller wants all commit errors to exit with the specified mesg:
2717 warn_fail_commit_memory(addr, size, exec, err);
2718 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2719 }
2720 }
2721
2722 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2723 #ifndef MAP_HUGETLB
2724 #define MAP_HUGETLB 0x40000
2725 #endif
2726
2727 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2728 #ifndef MADV_HUGEPAGE
2729 #define MADV_HUGEPAGE 14
2730 #endif
2731
2732 int os::Linux::commit_memory_impl(char* addr, size_t size,
2733 size_t alignment_hint, bool exec) {
2734 int err = os::Linux::commit_memory_impl(addr, size, exec);
2735 if (err == 0) {
2736 realign_memory(addr, size, alignment_hint);
2737 }
2738 return err;
2739 }
2740
2741 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2742 bool exec) {
2743 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2744 }
2745
2746 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2747 size_t alignment_hint, bool exec,
2748 const char* mesg) {
2749 assert(mesg != NULL, "mesg must be specified");
2750 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2751 if (err != 0) {
2752 // the caller wants all commit errors to exit with the specified mesg:
2753 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2754 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2755 }
2756 }
2757
2758 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2759 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2760 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2761 // be supported or the memory may already be backed by huge pages.
2762 ::madvise(addr, bytes, MADV_HUGEPAGE);
2763 }
2764 }
2765
2766 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2767 // This method works by doing an mmap over an existing mmaping and effectively discarding
2768 // the existing pages. However it won't work for SHM-based large pages that cannot be
2769 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2770 // small pages on top of the SHM segment. This method always works for small pages, so we
2771 // allow that in any case.
2772 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2773 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2774 }
2775 }
2776
2777 void os::numa_make_global(char *addr, size_t bytes) {
2778 Linux::numa_interleave_memory(addr, bytes);
2779 }
2780
2781 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2782 // bind policy to MPOL_PREFERRED for the current thread.
2783 #define USE_MPOL_PREFERRED 0
2784
2785 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2786 // To make NUMA and large pages more robust when both enabled, we need to ease
2787 // the requirements on where the memory should be allocated. MPOL_BIND is the
2788 // default policy and it will force memory to be allocated on the specified
2789 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2790 // the specified node, but will not force it. Using this policy will prevent
2791 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2792 // free large pages.
2793 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2794 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2795 }
2796
2797 bool os::numa_topology_changed() { return false; }
2798
2799 size_t os::numa_get_groups_num() {
2800 int max_node = Linux::numa_max_node();
2801 return max_node > 0 ? max_node + 1 : 1;
2802 }
2803
2804 int os::numa_get_group_id() {
2805 int cpu_id = Linux::sched_getcpu();
2806 if (cpu_id != -1) {
2807 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2808 if (lgrp_id != -1) {
2809 return lgrp_id;
2810 }
2811 }
2812 return 0;
2813 }
2814
2815 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2816 for (size_t i = 0; i < size; i++) {
2817 ids[i] = i;
2818 }
2819 return size;
2820 }
2821
2822 bool os::get_page_info(char *start, page_info* info) {
2823 return false;
2824 }
2825
2826 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2827 page_info* page_found) {
2828 return end;
2829 }
2830
2831
2832 int os::Linux::sched_getcpu_syscall(void) {
2833 unsigned int cpu;
2834 int retval = -1;
2835
2836 #if defined(IA32)
2837 #ifndef SYS_getcpu
2838 #define SYS_getcpu 318
2839 #endif
2840 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2841 #elif defined(AMD64)
2842 // Unfortunately we have to bring all these macros here from vsyscall.h
2843 // to be able to compile on old linuxes.
2844 #define __NR_vgetcpu 2
2845 #define VSYSCALL_START (-10UL << 20)
2846 #define VSYSCALL_SIZE 1024
2847 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2848 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2849 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2850 retval = vgetcpu(&cpu, NULL, NULL);
2851 #endif
2852
2853 return (retval == -1) ? retval : cpu;
2854 }
2855
2856 // Something to do with the numa-aware allocator needs these symbols
2857 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2858 extern "C" JNIEXPORT void numa_error(char *where) { }
2859 extern "C" JNIEXPORT int fork1() { return fork(); }
2860
2861
2862 // If we are running with libnuma version > 2, then we should
2863 // be trying to use symbols with versions 1.1
2864 // If we are running with earlier version, which did not have symbol versions,
2865 // we should use the base version.
2866 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2867 void *f = dlvsym(handle, name, "libnuma_1.1");
2868 if (f == NULL) {
2869 f = dlsym(handle, name);
2870 }
2871 return f;
2872 }
2873
2874 bool os::Linux::libnuma_init() {
2875 // sched_getcpu() should be in libc.
2876 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2877 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2878
2879 // If it's not, try a direct syscall.
2880 if (sched_getcpu() == -1) {
2881 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2882 (void*)&sched_getcpu_syscall));
2883 }
2884
2885 if (sched_getcpu() != -1) { // Does it work?
2886 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2887 if (handle != NULL) {
2888 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2889 libnuma_dlsym(handle, "numa_node_to_cpus")));
2890 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2891 libnuma_dlsym(handle, "numa_max_node")));
2892 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2893 libnuma_dlsym(handle, "numa_available")));
2894 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2895 libnuma_dlsym(handle, "numa_tonode_memory")));
2896 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2897 libnuma_dlsym(handle, "numa_interleave_memory")));
2898 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2899 libnuma_dlsym(handle, "numa_set_bind_policy")));
2900
2901
2902 if (numa_available() != -1) {
2903 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2904 // Create a cpu -> node mapping
2905 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2906 rebuild_cpu_to_node_map();
2907 return true;
2908 }
2909 }
2910 }
2911 return false;
2912 }
2913
2914 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2915 // The table is later used in get_node_by_cpu().
2916 void os::Linux::rebuild_cpu_to_node_map() {
2917 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2918 // in libnuma (possible values are starting from 16,
2919 // and continuing up with every other power of 2, but less
2920 // than the maximum number of CPUs supported by kernel), and
2921 // is a subject to change (in libnuma version 2 the requirements
2922 // are more reasonable) we'll just hardcode the number they use
2923 // in the library.
2924 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2925
2926 size_t cpu_num = os::active_processor_count();
2927 size_t cpu_map_size = NCPUS / BitsPerCLong;
2928 size_t cpu_map_valid_size =
2929 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2930
2931 cpu_to_node()->clear();
2932 cpu_to_node()->at_grow(cpu_num - 1);
2933 size_t node_num = numa_get_groups_num();
2934
2935 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2936 for (size_t i = 0; i < node_num; i++) {
2937 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2938 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2939 if (cpu_map[j] != 0) {
2940 for (size_t k = 0; k < BitsPerCLong; k++) {
2941 if (cpu_map[j] & (1UL << k)) {
2942 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2943 }
2944 }
2945 }
2946 }
2947 }
2948 }
2949 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2950 }
2951
2952 int os::Linux::get_node_by_cpu(int cpu_id) {
2953 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2954 return cpu_to_node()->at(cpu_id);
2955 }
2956 return -1;
2957 }
2958
2959 GrowableArray<int>* os::Linux::_cpu_to_node;
2960 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2961 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2962 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2963 os::Linux::numa_available_func_t os::Linux::_numa_available;
2964 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2965 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2966 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2967 unsigned long* os::Linux::_numa_all_nodes;
2968
2969 bool os::pd_uncommit_memory(char* addr, size_t size) {
2970 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2971 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2972 return res != (uintptr_t) MAP_FAILED;
2973 }
2974
2975 static address get_stack_commited_bottom(address bottom, size_t size) {
2976 address nbot = bottom;
2977 address ntop = bottom + size;
2978
2979 size_t page_sz = os::vm_page_size();
2980 unsigned pages = size / page_sz;
2981
2982 unsigned char vec[1];
2983 unsigned imin = 1, imax = pages + 1, imid;
2984 int mincore_return_value = 0;
2985
2986 assert(imin <= imax, "Unexpected page size");
2987
2988 while (imin < imax) {
2989 imid = (imax + imin) / 2;
2990 nbot = ntop - (imid * page_sz);
2991
2992 // Use a trick with mincore to check whether the page is mapped or not.
2993 // mincore sets vec to 1 if page resides in memory and to 0 if page
2994 // is swapped output but if page we are asking for is unmapped
2995 // it returns -1,ENOMEM
2996 mincore_return_value = mincore(nbot, page_sz, vec);
2997
2998 if (mincore_return_value == -1) {
2999 // Page is not mapped go up
3000 // to find first mapped page
3001 if (errno != EAGAIN) {
3002 assert(errno == ENOMEM, "Unexpected mincore errno");
3003 imax = imid;
3004 }
3005 } else {
3006 // Page is mapped go down
3007 // to find first not mapped page
3008 imin = imid + 1;
3009 }
3010 }
3011
3012 nbot = nbot + page_sz;
3013
3014 // Adjust stack bottom one page up if last checked page is not mapped
3015 if (mincore_return_value == -1) {
3016 nbot = nbot + page_sz;
3017 }
3018
3019 return nbot;
3020 }
3021
3022
3023 // Linux uses a growable mapping for the stack, and if the mapping for
3024 // the stack guard pages is not removed when we detach a thread the
3025 // stack cannot grow beyond the pages where the stack guard was
3026 // mapped. If at some point later in the process the stack expands to
3027 // that point, the Linux kernel cannot expand the stack any further
3028 // because the guard pages are in the way, and a segfault occurs.
3029 //
3030 // However, it's essential not to split the stack region by unmapping
3031 // a region (leaving a hole) that's already part of the stack mapping,
3032 // so if the stack mapping has already grown beyond the guard pages at
3033 // the time we create them, we have to truncate the stack mapping.
3034 // So, we need to know the extent of the stack mapping when
3035 // create_stack_guard_pages() is called.
3036
3037 // We only need this for stacks that are growable: at the time of
3038 // writing thread stacks don't use growable mappings (i.e. those
3039 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3040 // only applies to the main thread.
3041
3042 // If the (growable) stack mapping already extends beyond the point
3043 // where we're going to put our guard pages, truncate the mapping at
3044 // that point by munmap()ping it. This ensures that when we later
3045 // munmap() the guard pages we don't leave a hole in the stack
3046 // mapping. This only affects the main/initial thread
3047
3048 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3049 if (os::Linux::is_initial_thread()) {
3050 // As we manually grow stack up to bottom inside create_attached_thread(),
3051 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3052 // we don't need to do anything special.
3053 // Check it first, before calling heavy function.
3054 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3055 unsigned char vec[1];
3056
3057 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3058 // Fallback to slow path on all errors, including EAGAIN
3059 stack_extent = (uintptr_t) get_stack_commited_bottom(
3060 os::Linux::initial_thread_stack_bottom(),
3061 (size_t)addr - stack_extent);
3062 }
3063
3064 if (stack_extent < (uintptr_t)addr) {
3065 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3066 }
3067 }
3068
3069 return os::commit_memory(addr, size, !ExecMem);
3070 }
3071
3072 // If this is a growable mapping, remove the guard pages entirely by
3073 // munmap()ping them. If not, just call uncommit_memory(). This only
3074 // affects the main/initial thread, but guard against future OS changes
3075 // It's safe to always unmap guard pages for initial thread because we
3076 // always place it right after end of the mapped region
3077
3078 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3079 uintptr_t stack_extent, stack_base;
3080
3081 if (os::Linux::is_initial_thread()) {
3082 return ::munmap(addr, size) == 0;
3083 }
3084
3085 return os::uncommit_memory(addr, size);
3086 }
3087
3088 static address _highest_vm_reserved_address = NULL;
3089
3090 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3091 // at 'requested_addr'. If there are existing memory mappings at the same
3092 // location, however, they will be overwritten. If 'fixed' is false,
3093 // 'requested_addr' is only treated as a hint, the return value may or
3094 // may not start from the requested address. Unlike Linux mmap(), this
3095 // function returns NULL to indicate failure.
3096 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3097 char * addr;
3098 int flags;
3099
3100 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3101 if (fixed) {
3102 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3103 flags |= MAP_FIXED;
3104 }
3105
3106 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3107 // touch an uncommitted page. Otherwise, the read/write might
3108 // succeed if we have enough swap space to back the physical page.
3109 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3110 flags, -1, 0);
3111
3112 if (addr != MAP_FAILED) {
3113 // anon_mmap() should only get called during VM initialization,
3114 // don't need lock (actually we can skip locking even it can be called
3115 // from multiple threads, because _highest_vm_reserved_address is just a
3116 // hint about the upper limit of non-stack memory regions.)
3117 if ((address)addr + bytes > _highest_vm_reserved_address) {
3118 _highest_vm_reserved_address = (address)addr + bytes;
3119 }
3120 }
3121
3122 return addr == MAP_FAILED ? NULL : addr;
3123 }
3124
3125 // Don't update _highest_vm_reserved_address, because there might be memory
3126 // regions above addr + size. If so, releasing a memory region only creates
3127 // a hole in the address space, it doesn't help prevent heap-stack collision.
3128 //
3129 static int anon_munmap(char * addr, size_t size) {
3130 return ::munmap(addr, size) == 0;
3131 }
3132
3133 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3134 size_t alignment_hint) {
3135 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3136 }
3137
3138 bool os::pd_release_memory(char* addr, size_t size) {
3139 return anon_munmap(addr, size);
3140 }
3141
3142 static address highest_vm_reserved_address() {
3143 return _highest_vm_reserved_address;
3144 }
3145
3146 static bool linux_mprotect(char* addr, size_t size, int prot) {
3147 // Linux wants the mprotect address argument to be page aligned.
3148 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3149
3150 // According to SUSv3, mprotect() should only be used with mappings
3151 // established by mmap(), and mmap() always maps whole pages. Unaligned
3152 // 'addr' likely indicates problem in the VM (e.g. trying to change
3153 // protection of malloc'ed or statically allocated memory). Check the
3154 // caller if you hit this assert.
3155 assert(addr == bottom, "sanity check");
3156
3157 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3158 return ::mprotect(bottom, size, prot) == 0;
3159 }
3160
3161 // Set protections specified
3162 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3163 bool is_committed) {
3164 unsigned int p = 0;
3165 switch (prot) {
3166 case MEM_PROT_NONE: p = PROT_NONE; break;
3167 case MEM_PROT_READ: p = PROT_READ; break;
3168 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3169 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3170 default:
3171 ShouldNotReachHere();
3172 }
3173 // is_committed is unused.
3174 return linux_mprotect(addr, bytes, p);
3175 }
3176
3177 bool os::guard_memory(char* addr, size_t size) {
3178 return linux_mprotect(addr, size, PROT_NONE);
3179 }
3180
3181 bool os::unguard_memory(char* addr, size_t size) {
3182 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3183 }
3184
3185 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3186 size_t page_size) {
3187 bool result = false;
3188 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3189 MAP_ANONYMOUS|MAP_PRIVATE,
3190 -1, 0);
3191 if (p != MAP_FAILED) {
3192 void *aligned_p = align_ptr_up(p, page_size);
3193
3194 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3195
3196 munmap(p, page_size * 2);
3197 }
3198
3199 if (warn && !result) {
3200 warning("TransparentHugePages is not supported by the operating system.");
3201 }
3202
3203 return result;
3204 }
3205
3206 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3207 bool result = false;
3208 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3209 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3210 -1, 0);
3211
3212 if (p != MAP_FAILED) {
3213 // We don't know if this really is a huge page or not.
3214 FILE *fp = fopen("/proc/self/maps", "r");
3215 if (fp) {
3216 while (!feof(fp)) {
3217 char chars[257];
3218 long x = 0;
3219 if (fgets(chars, sizeof(chars), fp)) {
3220 if (sscanf(chars, "%lx-%*x", &x) == 1
3221 && x == (long)p) {
3222 if (strstr (chars, "hugepage")) {
3223 result = true;
3224 break;
3225 }
3226 }
3227 }
3228 }
3229 fclose(fp);
3230 }
3231 munmap(p, page_size);
3232 }
3233
3234 if (warn && !result) {
3235 warning("HugeTLBFS is not supported by the operating system.");
3236 }
3237
3238 return result;
3239 }
3240
3241 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3242 //
3243 // From the coredump_filter documentation:
3244 //
3245 // - (bit 0) anonymous private memory
3246 // - (bit 1) anonymous shared memory
3247 // - (bit 2) file-backed private memory
3248 // - (bit 3) file-backed shared memory
3249 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3250 // effective only if the bit 2 is cleared)
3251 // - (bit 5) hugetlb private memory
3252 // - (bit 6) hugetlb shared memory
3253 //
3254 static void set_coredump_filter(void) {
3255 FILE *f;
3256 long cdm;
3257
3258 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3259 return;
3260 }
3261
3262 if (fscanf(f, "%lx", &cdm) != 1) {
3263 fclose(f);
3264 return;
3265 }
3266
3267 rewind(f);
3268
3269 if ((cdm & LARGEPAGES_BIT) == 0) {
3270 cdm |= LARGEPAGES_BIT;
3271 fprintf(f, "%#lx", cdm);
3272 }
3273
3274 fclose(f);
3275 }
3276
3277 // Large page support
3278
3279 static size_t _large_page_size = 0;
3280
3281 size_t os::Linux::find_large_page_size() {
3282 size_t large_page_size = 0;
3283
3284 // large_page_size on Linux is used to round up heap size. x86 uses either
3285 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3286 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3287 // page as large as 256M.
3288 //
3289 // Here we try to figure out page size by parsing /proc/meminfo and looking
3290 // for a line with the following format:
3291 // Hugepagesize: 2048 kB
3292 //
3293 // If we can't determine the value (e.g. /proc is not mounted, or the text
3294 // format has been changed), we'll use the largest page size supported by
3295 // the processor.
3296
3297 #ifndef ZERO
3298 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3299 ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3300 #endif // ZERO
3301
3302 FILE *fp = fopen("/proc/meminfo", "r");
3303 if (fp) {
3304 while (!feof(fp)) {
3305 int x = 0;
3306 char buf[16];
3307 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3308 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3309 large_page_size = x * K;
3310 break;
3311 }
3312 } else {
3313 // skip to next line
3314 for (;;) {
3315 int ch = fgetc(fp);
3316 if (ch == EOF || ch == (int)'\n') break;
3317 }
3318 }
3319 }
3320 fclose(fp);
3321 }
3322
3323 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3324 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3325 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3326 proper_unit_for_byte_size(large_page_size));
3327 }
3328
3329 return large_page_size;
3330 }
3331
3332 size_t os::Linux::setup_large_page_size() {
3333 _large_page_size = Linux::find_large_page_size();
3334 const size_t default_page_size = (size_t)Linux::page_size();
3335 if (_large_page_size > default_page_size) {
3336 _page_sizes[0] = _large_page_size;
3337 _page_sizes[1] = default_page_size;
3338 _page_sizes[2] = 0;
3339 }
3340
3341 return _large_page_size;
3342 }
3343
3344 bool os::Linux::setup_large_page_type(size_t page_size) {
3345 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3346 FLAG_IS_DEFAULT(UseSHM) &&
3347 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3348
3349 // The type of large pages has not been specified by the user.
3350
3351 // Try UseHugeTLBFS and then UseSHM.
3352 UseHugeTLBFS = UseSHM = true;
3353
3354 // Don't try UseTransparentHugePages since there are known
3355 // performance issues with it turned on. This might change in the future.
3356 UseTransparentHugePages = false;
3357 }
3358
3359 if (UseTransparentHugePages) {
3360 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3361 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3362 UseHugeTLBFS = false;
3363 UseSHM = false;
3364 return true;
3365 }
3366 UseTransparentHugePages = false;
3367 }
3368
3369 if (UseHugeTLBFS) {
3370 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3371 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3372 UseSHM = false;
3373 return true;
3374 }
3375 UseHugeTLBFS = false;
3376 }
3377
3378 return UseSHM;
3379 }
3380
3381 void os::large_page_init() {
3382 if (!UseLargePages &&
3383 !UseTransparentHugePages &&
3384 !UseHugeTLBFS &&
3385 !UseSHM) {
3386 // Not using large pages.
3387 return;
3388 }
3389
3390 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3391 // The user explicitly turned off large pages.
3392 // Ignore the rest of the large pages flags.
3393 UseTransparentHugePages = false;
3394 UseHugeTLBFS = false;
3395 UseSHM = false;
3396 return;
3397 }
3398
3399 size_t large_page_size = Linux::setup_large_page_size();
3400 UseLargePages = Linux::setup_large_page_type(large_page_size);
3401
3402 set_coredump_filter();
3403 }
3404
3405 #ifndef SHM_HUGETLB
3406 #define SHM_HUGETLB 04000
3407 #endif
3408
3409 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3410 char* req_addr, bool exec) {
3411 // "exec" is passed in but not used. Creating the shared image for
3412 // the code cache doesn't have an SHM_X executable permission to check.
3413 assert(UseLargePages && UseSHM, "only for SHM large pages");
3414 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3415
3416 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3417 return NULL; // Fallback to small pages.
3418 }
3419
3420 key_t key = IPC_PRIVATE;
3421 char *addr;
3422
3423 bool warn_on_failure = UseLargePages &&
3424 (!FLAG_IS_DEFAULT(UseLargePages) ||
3425 !FLAG_IS_DEFAULT(UseSHM) ||
3426 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3427 char msg[128];
3428
3429 // Create a large shared memory region to attach to based on size.
3430 // Currently, size is the total size of the heap
3431 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3432 if (shmid == -1) {
3433 // Possible reasons for shmget failure:
3434 // 1. shmmax is too small for Java heap.
3435 // > check shmmax value: cat /proc/sys/kernel/shmmax
3436 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3437 // 2. not enough large page memory.
3438 // > check available large pages: cat /proc/meminfo
3439 // > increase amount of large pages:
3440 // echo new_value > /proc/sys/vm/nr_hugepages
3441 // Note 1: different Linux may use different name for this property,
3442 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3443 // Note 2: it's possible there's enough physical memory available but
3444 // they are so fragmented after a long run that they can't
3445 // coalesce into large pages. Try to reserve large pages when
3446 // the system is still "fresh".
3447 if (warn_on_failure) {
3448 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3449 warning("%s", msg);
3450 }
3451 return NULL;
3452 }
3453
3454 // attach to the region
3455 addr = (char*)shmat(shmid, req_addr, 0);
3456 int err = errno;
3457
3458 // Remove shmid. If shmat() is successful, the actual shared memory segment
3459 // will be deleted when it's detached by shmdt() or when the process
3460 // terminates. If shmat() is not successful this will remove the shared
3461 // segment immediately.
3462 shmctl(shmid, IPC_RMID, NULL);
3463
3464 if ((intptr_t)addr == -1) {
3465 if (warn_on_failure) {
3466 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3467 warning("%s", msg);
3468 }
3469 return NULL;
3470 }
3471
3472 return addr;
3473 }
3474
3475 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3476 int error) {
3477 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3478
3479 bool warn_on_failure = UseLargePages &&
3480 (!FLAG_IS_DEFAULT(UseLargePages) ||
3481 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3482 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3483
3484 if (warn_on_failure) {
3485 char msg[128];
3486 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3487 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3488 warning("%s", msg);
3489 }
3490 }
3491
3492 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3493 char* req_addr,
3494 bool exec) {
3495 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3496 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3497 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3498
3499 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3500 char* addr = (char*)::mmap(req_addr, bytes, prot,
3501 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3502 -1, 0);
3503
3504 if (addr == MAP_FAILED) {
3505 warn_on_large_pages_failure(req_addr, bytes, errno);
3506 return NULL;
3507 }
3508
3509 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3510
3511 return addr;
3512 }
3513
3514 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3515 size_t alignment,
3516 char* req_addr,
3517 bool exec) {
3518 size_t large_page_size = os::large_page_size();
3519
3520 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3521
3522 // Allocate small pages.
3523
3524 char* start;
3525 if (req_addr != NULL) {
3526 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3527 assert(is_size_aligned(bytes, alignment), "Must be");
3528 start = os::reserve_memory(bytes, req_addr);
3529 assert(start == NULL || start == req_addr, "Must be");
3530 } else {
3531 start = os::reserve_memory_aligned(bytes, alignment);
3532 }
3533
3534 if (start == NULL) {
3535 return NULL;
3536 }
3537
3538 assert(is_ptr_aligned(start, alignment), "Must be");
3539
3540 if (MemTracker::tracking_level() > NMT_minimal) {
3541 // os::reserve_memory_special will record this memory area.
3542 // Need to release it here to prevent overlapping reservations.
3543 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3544 tkr.record((address)start, bytes);
3545 }
3546
3547 char* end = start + bytes;
3548
3549 // Find the regions of the allocated chunk that can be promoted to large pages.
3550 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3551 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3552
3553 size_t lp_bytes = lp_end - lp_start;
3554
3555 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3556
3557 if (lp_bytes == 0) {
3558 // The mapped region doesn't even span the start and the end of a large page.
3559 // Fall back to allocate a non-special area.
3560 ::munmap(start, end - start);
3561 return NULL;
3562 }
3563
3564 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3565
3566
3567 void* result;
3568
3569 if (start != lp_start) {
3570 result = ::mmap(start, lp_start - start, prot,
3571 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3572 -1, 0);
3573 if (result == MAP_FAILED) {
3574 ::munmap(lp_start, end - lp_start);
3575 return NULL;
3576 }
3577 }
3578
3579 result = ::mmap(lp_start, lp_bytes, prot,
3580 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3581 -1, 0);
3582 if (result == MAP_FAILED) {
3583 warn_on_large_pages_failure(req_addr, bytes, errno);
3584 // If the mmap above fails, the large pages region will be unmapped and we
3585 // have regions before and after with small pages. Release these regions.
3586 //
3587 // | mapped | unmapped | mapped |
3588 // ^ ^ ^ ^
3589 // start lp_start lp_end end
3590 //
3591 ::munmap(start, lp_start - start);
3592 ::munmap(lp_end, end - lp_end);
3593 return NULL;
3594 }
3595
3596 if (lp_end != end) {
3597 result = ::mmap(lp_end, end - lp_end, prot,
3598 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3599 -1, 0);
3600 if (result == MAP_FAILED) {
3601 ::munmap(start, lp_end - start);
3602 return NULL;
3603 }
3604 }
3605
3606 return start;
3607 }
3608
3609 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3610 size_t alignment,
3611 char* req_addr,
3612 bool exec) {
3613 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3614 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3615 assert(is_power_of_2(alignment), "Must be");
3616 assert(is_power_of_2(os::large_page_size()), "Must be");
3617 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3618
3619 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3620 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3621 } else {
3622 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3623 }
3624 }
3625
3626 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3627 char* req_addr, bool exec) {
3628 assert(UseLargePages, "only for large pages");
3629
3630 char* addr;
3631 if (UseSHM) {
3632 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3633 } else {
3634 assert(UseHugeTLBFS, "must be");
3635 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3636 }
3637
3638 if (addr != NULL) {
3639 if (UseNUMAInterleaving) {
3640 numa_make_global(addr, bytes);
3641 }
3642
3643 // The memory is committed
3644 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3645 }
3646
3647 return addr;
3648 }
3649
3650 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3651 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3652 return shmdt(base) == 0;
3653 }
3654
3655 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3656 return pd_release_memory(base, bytes);
3657 }
3658
3659 bool os::release_memory_special(char* base, size_t bytes) {
3660 bool res;
3661 if (MemTracker::tracking_level() > NMT_minimal) {
3662 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3663 res = os::Linux::release_memory_special_impl(base, bytes);
3664 if (res) {
3665 tkr.record((address)base, bytes);
3666 }
3667
3668 } else {
3669 res = os::Linux::release_memory_special_impl(base, bytes);
3670 }
3671 return res;
3672 }
3673
3674 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3675 assert(UseLargePages, "only for large pages");
3676 bool res;
3677
3678 if (UseSHM) {
3679 res = os::Linux::release_memory_special_shm(base, bytes);
3680 } else {
3681 assert(UseHugeTLBFS, "must be");
3682 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3683 }
3684 return res;
3685 }
3686
3687 size_t os::large_page_size() {
3688 return _large_page_size;
3689 }
3690
3691 // With SysV SHM the entire memory region must be allocated as shared
3692 // memory.
3693 // HugeTLBFS allows application to commit large page memory on demand.
3694 // However, when committing memory with HugeTLBFS fails, the region
3695 // that was supposed to be committed will lose the old reservation
3696 // and allow other threads to steal that memory region. Because of this
3697 // behavior we can't commit HugeTLBFS memory.
3698 bool os::can_commit_large_page_memory() {
3699 return UseTransparentHugePages;
3700 }
3701
3702 bool os::can_execute_large_page_memory() {
3703 return UseTransparentHugePages || UseHugeTLBFS;
3704 }
3705
3706 // Reserve memory at an arbitrary address, only if that area is
3707 // available (and not reserved for something else).
3708
3709 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3710 const int max_tries = 10;
3711 char* base[max_tries];
3712 size_t size[max_tries];
3713 const size_t gap = 0x000000;
3714
3715 // Assert only that the size is a multiple of the page size, since
3716 // that's all that mmap requires, and since that's all we really know
3717 // about at this low abstraction level. If we need higher alignment,
3718 // we can either pass an alignment to this method or verify alignment
3719 // in one of the methods further up the call chain. See bug 5044738.
3720 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3721
3722 // Repeatedly allocate blocks until the block is allocated at the
3723 // right spot. Give up after max_tries. Note that reserve_memory() will
3724 // automatically update _highest_vm_reserved_address if the call is
3725 // successful. The variable tracks the highest memory address every reserved
3726 // by JVM. It is used to detect heap-stack collision if running with
3727 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3728 // space than needed, it could confuse the collision detecting code. To
3729 // solve the problem, save current _highest_vm_reserved_address and
3730 // calculate the correct value before return.
3731 address old_highest = _highest_vm_reserved_address;
3732
3733 // Linux mmap allows caller to pass an address as hint; give it a try first,
3734 // if kernel honors the hint then we can return immediately.
3735 char * addr = anon_mmap(requested_addr, bytes, false);
3736 if (addr == requested_addr) {
3737 return requested_addr;
3738 }
3739
3740 if (addr != NULL) {
3741 // mmap() is successful but it fails to reserve at the requested address
3742 anon_munmap(addr, bytes);
3743 }
3744
3745 int i;
3746 for (i = 0; i < max_tries; ++i) {
3747 base[i] = reserve_memory(bytes);
3748
3749 if (base[i] != NULL) {
3750 // Is this the block we wanted?
3751 if (base[i] == requested_addr) {
3752 size[i] = bytes;
3753 break;
3754 }
3755
3756 // Does this overlap the block we wanted? Give back the overlapped
3757 // parts and try again.
3758
3759 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3760 if (top_overlap >= 0 && top_overlap < bytes) {
3761 unmap_memory(base[i], top_overlap);
3762 base[i] += top_overlap;
3763 size[i] = bytes - top_overlap;
3764 } else {
3765 size_t bottom_overlap = base[i] + bytes - requested_addr;
3766 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3767 unmap_memory(requested_addr, bottom_overlap);
3768 size[i] = bytes - bottom_overlap;
3769 } else {
3770 size[i] = bytes;
3771 }
3772 }
3773 }
3774 }
3775
3776 // Give back the unused reserved pieces.
3777
3778 for (int j = 0; j < i; ++j) {
3779 if (base[j] != NULL) {
3780 unmap_memory(base[j], size[j]);
3781 }
3782 }
3783
3784 if (i < max_tries) {
3785 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3786 return requested_addr;
3787 } else {
3788 _highest_vm_reserved_address = old_highest;
3789 return NULL;
3790 }
3791 }
3792
3793 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3794 return ::read(fd, buf, nBytes);
3795 }
3796
3797 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3798 return ::pread(fd, buf, nBytes, offset);
3799 }
3800
3801 // Short sleep, direct OS call.
3802 //
3803 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3804 // sched_yield(2) will actually give up the CPU:
3805 //
3806 // * Alone on this pariticular CPU, keeps running.
3807 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3808 // (pre 2.6.39).
3809 //
3810 // So calling this with 0 is an alternative.
3811 //
3812 void os::naked_short_sleep(jlong ms) {
3813 struct timespec req;
3814
3815 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3816 req.tv_sec = 0;
3817 if (ms > 0) {
3818 req.tv_nsec = (ms % 1000) * 1000000;
3819 } else {
3820 req.tv_nsec = 1;
3821 }
3822
3823 nanosleep(&req, NULL);
3824
3825 return;
3826 }
3827
3828 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3829 void os::infinite_sleep() {
3830 while (true) { // sleep forever ...
3831 ::sleep(100); // ... 100 seconds at a time
3832 }
3833 }
3834
3835 // Used to convert frequent JVM_Yield() to nops
3836 bool os::dont_yield() {
3837 return DontYieldALot;
3838 }
3839
3840 void os::naked_yield() {
3841 sched_yield();
3842 }
3843
3844 ////////////////////////////////////////////////////////////////////////////////
3845 // thread priority support
3846
3847 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3848 // only supports dynamic priority, static priority must be zero. For real-time
3849 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3850 // However, for large multi-threaded applications, SCHED_RR is not only slower
3851 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3852 // of 5 runs - Sep 2005).
3853 //
3854 // The following code actually changes the niceness of kernel-thread/LWP. It
3855 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3856 // not the entire user process, and user level threads are 1:1 mapped to kernel
3857 // threads. It has always been the case, but could change in the future. For
3858 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3859 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3860
3861 int os::java_to_os_priority[CriticalPriority + 1] = {
3862 19, // 0 Entry should never be used
3863
3864 4, // 1 MinPriority
3865 3, // 2
3866 2, // 3
3867
3868 1, // 4
3869 0, // 5 NormPriority
3870 -1, // 6
3871
3872 -2, // 7
3873 -3, // 8
3874 -4, // 9 NearMaxPriority
3875
3876 -5, // 10 MaxPriority
3877
3878 -5 // 11 CriticalPriority
3879 };
3880
3881 static int prio_init() {
3882 if (ThreadPriorityPolicy == 1) {
3883 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3884 // if effective uid is not root. Perhaps, a more elegant way of doing
3885 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3886 if (geteuid() != 0) {
3887 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3888 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3889 }
3890 ThreadPriorityPolicy = 0;
3891 }
3892 }
3893 if (UseCriticalJavaThreadPriority) {
3894 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3895 }
3896 return 0;
3897 }
3898
3899 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3900 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3901
3902 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3903 return (ret == 0) ? OS_OK : OS_ERR;
3904 }
3905
3906 OSReturn os::get_native_priority(const Thread* const thread,
3907 int *priority_ptr) {
3908 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3909 *priority_ptr = java_to_os_priority[NormPriority];
3910 return OS_OK;
3911 }
3912
3913 errno = 0;
3914 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3915 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3916 }
3917
3918 // Hint to the underlying OS that a task switch would not be good.
3919 // Void return because it's a hint and can fail.
3920 void os::hint_no_preempt() {}
3921
3922 ////////////////////////////////////////////////////////////////////////////////
3923 // suspend/resume support
3924
3925 // the low-level signal-based suspend/resume support is a remnant from the
3926 // old VM-suspension that used to be for java-suspension, safepoints etc,
3927 // within hotspot. Now there is a single use-case for this:
3928 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3929 // that runs in the watcher thread.
3930 // The remaining code is greatly simplified from the more general suspension
3931 // code that used to be used.
3932 //
3933 // The protocol is quite simple:
3934 // - suspend:
3935 // - sends a signal to the target thread
3936 // - polls the suspend state of the osthread using a yield loop
3937 // - target thread signal handler (SR_handler) sets suspend state
3938 // and blocks in sigsuspend until continued
3939 // - resume:
3940 // - sets target osthread state to continue
3941 // - sends signal to end the sigsuspend loop in the SR_handler
3942 //
3943 // Note that the SR_lock plays no role in this suspend/resume protocol.
3944
3945 static void resume_clear_context(OSThread *osthread) {
3946 osthread->set_ucontext(NULL);
3947 osthread->set_siginfo(NULL);
3948 }
3949
3950 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3951 ucontext_t* context) {
3952 osthread->set_ucontext(context);
3953 osthread->set_siginfo(siginfo);
3954 }
3955
3956 // Handler function invoked when a thread's execution is suspended or
3957 // resumed. We have to be careful that only async-safe functions are
3958 // called here (Note: most pthread functions are not async safe and
3959 // should be avoided.)
3960 //
3961 // Note: sigwait() is a more natural fit than sigsuspend() from an
3962 // interface point of view, but sigwait() prevents the signal hander
3963 // from being run. libpthread would get very confused by not having
3964 // its signal handlers run and prevents sigwait()'s use with the
3965 // mutex granting granting signal.
3966 //
3967 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3968 //
3969 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3970 // Save and restore errno to avoid confusing native code with EINTR
3971 // after sigsuspend.
3972 int old_errno = errno;
3973
3974 Thread* thread = Thread::current();
3975 OSThread* osthread = thread->osthread();
3976 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3977
3978 os::SuspendResume::State current = osthread->sr.state();
3979 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3980 suspend_save_context(osthread, siginfo, context);
3981
3982 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3983 os::SuspendResume::State state = osthread->sr.suspended();
3984 if (state == os::SuspendResume::SR_SUSPENDED) {
3985 sigset_t suspend_set; // signals for sigsuspend()
3986
3987 // get current set of blocked signals and unblock resume signal
3988 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3989 sigdelset(&suspend_set, SR_signum);
3990
3991 sr_semaphore.signal();
3992 // wait here until we are resumed
3993 while (1) {
3994 sigsuspend(&suspend_set);
3995
3996 os::SuspendResume::State result = osthread->sr.running();
3997 if (result == os::SuspendResume::SR_RUNNING) {
3998 sr_semaphore.signal();
3999 break;
4000 }
4001 }
4002
4003 } else if (state == os::SuspendResume::SR_RUNNING) {
4004 // request was cancelled, continue
4005 } else {
4006 ShouldNotReachHere();
4007 }
4008
4009 resume_clear_context(osthread);
4010 } else if (current == os::SuspendResume::SR_RUNNING) {
4011 // request was cancelled, continue
4012 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4013 // ignore
4014 } else {
4015 // ignore
4016 }
4017
4018 errno = old_errno;
4019 }
4020
4021
4022 static int SR_initialize() {
4023 struct sigaction act;
4024 char *s;
4025 // Get signal number to use for suspend/resume
4026 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4027 int sig = ::strtol(s, 0, 10);
4028 if (sig > 0 || sig < _NSIG) {
4029 SR_signum = sig;
4030 }
4031 }
4032
4033 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4034 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4035
4036 sigemptyset(&SR_sigset);
4037 sigaddset(&SR_sigset, SR_signum);
4038
4039 // Set up signal handler for suspend/resume
4040 act.sa_flags = SA_RESTART|SA_SIGINFO;
4041 act.sa_handler = (void (*)(int)) SR_handler;
4042
4043 // SR_signum is blocked by default.
4044 // 4528190 - We also need to block pthread restart signal (32 on all
4045 // supported Linux platforms). Note that LinuxThreads need to block
4046 // this signal for all threads to work properly. So we don't have
4047 // to use hard-coded signal number when setting up the mask.
4048 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4049
4050 if (sigaction(SR_signum, &act, 0) == -1) {
4051 return -1;
4052 }
4053
4054 // Save signal flag
4055 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4056 return 0;
4057 }
4058
4059 static int sr_notify(OSThread* osthread) {
4060 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4061 assert_status(status == 0, status, "pthread_kill");
4062 return status;
4063 }
4064
4065 // "Randomly" selected value for how long we want to spin
4066 // before bailing out on suspending a thread, also how often
4067 // we send a signal to a thread we want to resume
4068 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4069 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4070
4071 // returns true on success and false on error - really an error is fatal
4072 // but this seems the normal response to library errors
4073 static bool do_suspend(OSThread* osthread) {
4074 assert(osthread->sr.is_running(), "thread should be running");
4075 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4076
4077 // mark as suspended and send signal
4078 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4079 // failed to switch, state wasn't running?
4080 ShouldNotReachHere();
4081 return false;
4082 }
4083
4084 if (sr_notify(osthread) != 0) {
4085 ShouldNotReachHere();
4086 }
4087
4088 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4089 while (true) {
4090 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4091 break;
4092 } else {
4093 // timeout
4094 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4095 if (cancelled == os::SuspendResume::SR_RUNNING) {
4096 return false;
4097 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4098 // make sure that we consume the signal on the semaphore as well
4099 sr_semaphore.wait();
4100 break;
4101 } else {
4102 ShouldNotReachHere();
4103 return false;
4104 }
4105 }
4106 }
4107
4108 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4109 return true;
4110 }
4111
4112 static void do_resume(OSThread* osthread) {
4113 assert(osthread->sr.is_suspended(), "thread should be suspended");
4114 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4115
4116 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4117 // failed to switch to WAKEUP_REQUEST
4118 ShouldNotReachHere();
4119 return;
4120 }
4121
4122 while (true) {
4123 if (sr_notify(osthread) == 0) {
4124 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4125 if (osthread->sr.is_running()) {
4126 return;
4127 }
4128 }
4129 } else {
4130 ShouldNotReachHere();
4131 }
4132 }
4133
4134 guarantee(osthread->sr.is_running(), "Must be running!");
4135 }
4136
4137 ///////////////////////////////////////////////////////////////////////////////////
4138 // signal handling (except suspend/resume)
4139
4140 // This routine may be used by user applications as a "hook" to catch signals.
4141 // The user-defined signal handler must pass unrecognized signals to this
4142 // routine, and if it returns true (non-zero), then the signal handler must
4143 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4144 // routine will never retun false (zero), but instead will execute a VM panic
4145 // routine kill the process.
4146 //
4147 // If this routine returns false, it is OK to call it again. This allows
4148 // the user-defined signal handler to perform checks either before or after
4149 // the VM performs its own checks. Naturally, the user code would be making
4150 // a serious error if it tried to handle an exception (such as a null check
4151 // or breakpoint) that the VM was generating for its own correct operation.
4152 //
4153 // This routine may recognize any of the following kinds of signals:
4154 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4155 // It should be consulted by handlers for any of those signals.
4156 //
4157 // The caller of this routine must pass in the three arguments supplied
4158 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4159 // field of the structure passed to sigaction(). This routine assumes that
4160 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4161 //
4162 // Note that the VM will print warnings if it detects conflicting signal
4163 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4164 //
4165 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4166 siginfo_t* siginfo,
4167 void* ucontext,
4168 int abort_if_unrecognized);
4169
4170 void signalHandler(int sig, siginfo_t* info, void* uc) {
4171 assert(info != NULL && uc != NULL, "it must be old kernel");
4172 int orig_errno = errno; // Preserve errno value over signal handler.
4173 JVM_handle_linux_signal(sig, info, uc, true);
4174 errno = orig_errno;
4175 }
4176
4177
4178 // This boolean allows users to forward their own non-matching signals
4179 // to JVM_handle_linux_signal, harmlessly.
4180 bool os::Linux::signal_handlers_are_installed = false;
4181
4182 // For signal-chaining
4183 struct sigaction os::Linux::sigact[MAXSIGNUM];
4184 unsigned int os::Linux::sigs = 0;
4185 bool os::Linux::libjsig_is_loaded = false;
4186 typedef struct sigaction *(*get_signal_t)(int);
4187 get_signal_t os::Linux::get_signal_action = NULL;
4188
4189 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4190 struct sigaction *actp = NULL;
4191
4192 if (libjsig_is_loaded) {
4193 // Retrieve the old signal handler from libjsig
4194 actp = (*get_signal_action)(sig);
4195 }
4196 if (actp == NULL) {
4197 // Retrieve the preinstalled signal handler from jvm
4198 actp = get_preinstalled_handler(sig);
4199 }
4200
4201 return actp;
4202 }
4203
4204 static bool call_chained_handler(struct sigaction *actp, int sig,
4205 siginfo_t *siginfo, void *context) {
4206 // Call the old signal handler
4207 if (actp->sa_handler == SIG_DFL) {
4208 // It's more reasonable to let jvm treat it as an unexpected exception
4209 // instead of taking the default action.
4210 return false;
4211 } else if (actp->sa_handler != SIG_IGN) {
4212 if ((actp->sa_flags & SA_NODEFER) == 0) {
4213 // automaticlly block the signal
4214 sigaddset(&(actp->sa_mask), sig);
4215 }
4216
4217 sa_handler_t hand;
4218 sa_sigaction_t sa;
4219 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4220 // retrieve the chained handler
4221 if (siginfo_flag_set) {
4222 sa = actp->sa_sigaction;
4223 } else {
4224 hand = actp->sa_handler;
4225 }
4226
4227 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4228 actp->sa_handler = SIG_DFL;
4229 }
4230
4231 // try to honor the signal mask
4232 sigset_t oset;
4233 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4234
4235 // call into the chained handler
4236 if (siginfo_flag_set) {
4237 (*sa)(sig, siginfo, context);
4238 } else {
4239 (*hand)(sig);
4240 }
4241
4242 // restore the signal mask
4243 pthread_sigmask(SIG_SETMASK, &oset, 0);
4244 }
4245 // Tell jvm's signal handler the signal is taken care of.
4246 return true;
4247 }
4248
4249 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4250 bool chained = false;
4251 // signal-chaining
4252 if (UseSignalChaining) {
4253 struct sigaction *actp = get_chained_signal_action(sig);
4254 if (actp != NULL) {
4255 chained = call_chained_handler(actp, sig, siginfo, context);
4256 }
4257 }
4258 return chained;
4259 }
4260
4261 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4262 if ((((unsigned int)1 << sig) & sigs) != 0) {
4263 return &sigact[sig];
4264 }
4265 return NULL;
4266 }
4267
4268 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4269 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4270 sigact[sig] = oldAct;
4271 sigs |= (unsigned int)1 << sig;
4272 }
4273
4274 // for diagnostic
4275 int os::Linux::sigflags[MAXSIGNUM];
4276
4277 int os::Linux::get_our_sigflags(int sig) {
4278 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4279 return sigflags[sig];
4280 }
4281
4282 void os::Linux::set_our_sigflags(int sig, int flags) {
4283 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4284 sigflags[sig] = flags;
4285 }
4286
4287 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4288 // Check for overwrite.
4289 struct sigaction oldAct;
4290 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4291
4292 void* oldhand = oldAct.sa_sigaction
4293 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4294 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4295 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4296 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4297 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4298 if (AllowUserSignalHandlers || !set_installed) {
4299 // Do not overwrite; user takes responsibility to forward to us.
4300 return;
4301 } else if (UseSignalChaining) {
4302 // save the old handler in jvm
4303 save_preinstalled_handler(sig, oldAct);
4304 // libjsig also interposes the sigaction() call below and saves the
4305 // old sigaction on it own.
4306 } else {
4307 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4308 "%#lx for signal %d.", (long)oldhand, sig));
4309 }
4310 }
4311
4312 struct sigaction sigAct;
4313 sigfillset(&(sigAct.sa_mask));
4314 sigAct.sa_handler = SIG_DFL;
4315 if (!set_installed) {
4316 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4317 } else {
4318 sigAct.sa_sigaction = signalHandler;
4319 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4320 }
4321 // Save flags, which are set by ours
4322 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4323 sigflags[sig] = sigAct.sa_flags;
4324
4325 int ret = sigaction(sig, &sigAct, &oldAct);
4326 assert(ret == 0, "check");
4327
4328 void* oldhand2 = oldAct.sa_sigaction
4329 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4330 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4331 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4332 }
4333
4334 // install signal handlers for signals that HotSpot needs to
4335 // handle in order to support Java-level exception handling.
4336
4337 void os::Linux::install_signal_handlers() {
4338 if (!signal_handlers_are_installed) {
4339 signal_handlers_are_installed = true;
4340
4341 // signal-chaining
4342 typedef void (*signal_setting_t)();
4343 signal_setting_t begin_signal_setting = NULL;
4344 signal_setting_t end_signal_setting = NULL;
4345 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4346 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4347 if (begin_signal_setting != NULL) {
4348 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4349 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4350 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4351 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4352 libjsig_is_loaded = true;
4353 assert(UseSignalChaining, "should enable signal-chaining");
4354 }
4355 if (libjsig_is_loaded) {
4356 // Tell libjsig jvm is setting signal handlers
4357 (*begin_signal_setting)();
4358 }
4359
4360 set_signal_handler(SIGSEGV, true);
4361 set_signal_handler(SIGPIPE, true);
4362 set_signal_handler(SIGBUS, true);
4363 set_signal_handler(SIGILL, true);
4364 set_signal_handler(SIGFPE, true);
4365 #if defined(PPC64)
4366 set_signal_handler(SIGTRAP, true);
4367 #endif
4368 set_signal_handler(SIGXFSZ, true);
4369
4370 if (libjsig_is_loaded) {
4371 // Tell libjsig jvm finishes setting signal handlers
4372 (*end_signal_setting)();
4373 }
4374
4375 // We don't activate signal checker if libjsig is in place, we trust ourselves
4376 // and if UserSignalHandler is installed all bets are off.
4377 // Log that signal checking is off only if -verbose:jni is specified.
4378 if (CheckJNICalls) {
4379 if (libjsig_is_loaded) {
4380 if (PrintJNIResolving) {
4381 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4382 }
4383 check_signals = false;
4384 }
4385 if (AllowUserSignalHandlers) {
4386 if (PrintJNIResolving) {
4387 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4388 }
4389 check_signals = false;
4390 }
4391 }
4392 }
4393 }
4394
4395 // This is the fastest way to get thread cpu time on Linux.
4396 // Returns cpu time (user+sys) for any thread, not only for current.
4397 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4398 // It might work on 2.6.10+ with a special kernel/glibc patch.
4399 // For reference, please, see IEEE Std 1003.1-2004:
4400 // http://www.unix.org/single_unix_specification
4401
4402 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4403 struct timespec tp;
4404 int rc = os::Linux::clock_gettime(clockid, &tp);
4405 assert(rc == 0, "clock_gettime is expected to return 0 code");
4406
4407 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4408 }
4409
4410 /////
4411 // glibc on Linux platform uses non-documented flag
4412 // to indicate, that some special sort of signal
4413 // trampoline is used.
4414 // We will never set this flag, and we should
4415 // ignore this flag in our diagnostic
4416 #ifdef SIGNIFICANT_SIGNAL_MASK
4417 #undef SIGNIFICANT_SIGNAL_MASK
4418 #endif
4419 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4420
4421 static const char* get_signal_handler_name(address handler,
4422 char* buf, int buflen) {
4423 int offset;
4424 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4425 if (found) {
4426 // skip directory names
4427 const char *p1, *p2;
4428 p1 = buf;
4429 size_t len = strlen(os::file_separator());
4430 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4431 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4432 } else {
4433 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4434 }
4435 return buf;
4436 }
4437
4438 static void print_signal_handler(outputStream* st, int sig,
4439 char* buf, size_t buflen) {
4440 struct sigaction sa;
4441
4442 sigaction(sig, NULL, &sa);
4443
4444 // See comment for SIGNIFICANT_SIGNAL_MASK define
4445 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4446
4447 st->print("%s: ", os::exception_name(sig, buf, buflen));
4448
4449 address handler = (sa.sa_flags & SA_SIGINFO)
4450 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4451 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4452
4453 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4454 st->print("SIG_DFL");
4455 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4456 st->print("SIG_IGN");
4457 } else {
4458 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4459 }
4460
4461 st->print(", sa_mask[0]=");
4462 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4463
4464 address rh = VMError::get_resetted_sighandler(sig);
4465 // May be, handler was resetted by VMError?
4466 if (rh != NULL) {
4467 handler = rh;
4468 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4469 }
4470
4471 st->print(", sa_flags=");
4472 os::Posix::print_sa_flags(st, sa.sa_flags);
4473
4474 // Check: is it our handler?
4475 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4476 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4477 // It is our signal handler
4478 // check for flags, reset system-used one!
4479 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4480 st->print(
4481 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4482 os::Linux::get_our_sigflags(sig));
4483 }
4484 }
4485 st->cr();
4486 }
4487
4488
4489 #define DO_SIGNAL_CHECK(sig) \
4490 do { \
4491 if (!sigismember(&check_signal_done, sig)) { \
4492 os::Linux::check_signal_handler(sig); \
4493 } \
4494 } while (0)
4495
4496 // This method is a periodic task to check for misbehaving JNI applications
4497 // under CheckJNI, we can add any periodic checks here
4498
4499 void os::run_periodic_checks() {
4500 if (check_signals == false) return;
4501
4502 // SEGV and BUS if overridden could potentially prevent
4503 // generation of hs*.log in the event of a crash, debugging
4504 // such a case can be very challenging, so we absolutely
4505 // check the following for a good measure:
4506 DO_SIGNAL_CHECK(SIGSEGV);
4507 DO_SIGNAL_CHECK(SIGILL);
4508 DO_SIGNAL_CHECK(SIGFPE);
4509 DO_SIGNAL_CHECK(SIGBUS);
4510 DO_SIGNAL_CHECK(SIGPIPE);
4511 DO_SIGNAL_CHECK(SIGXFSZ);
4512 #if defined(PPC64)
4513 DO_SIGNAL_CHECK(SIGTRAP);
4514 #endif
4515
4516 // ReduceSignalUsage allows the user to override these handlers
4517 // see comments at the very top and jvm_solaris.h
4518 if (!ReduceSignalUsage) {
4519 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4520 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4521 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4522 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4523 }
4524
4525 DO_SIGNAL_CHECK(SR_signum);
4526 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4527 }
4528
4529 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4530
4531 static os_sigaction_t os_sigaction = NULL;
4532
4533 void os::Linux::check_signal_handler(int sig) {
4534 char buf[O_BUFLEN];
4535 address jvmHandler = NULL;
4536
4537
4538 struct sigaction act;
4539 if (os_sigaction == NULL) {
4540 // only trust the default sigaction, in case it has been interposed
4541 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4542 if (os_sigaction == NULL) return;
4543 }
4544
4545 os_sigaction(sig, (struct sigaction*)NULL, &act);
4546
4547
4548 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4549
4550 address thisHandler = (act.sa_flags & SA_SIGINFO)
4551 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4552 : CAST_FROM_FN_PTR(address, act.sa_handler);
4553
4554
4555 switch (sig) {
4556 case SIGSEGV:
4557 case SIGBUS:
4558 case SIGFPE:
4559 case SIGPIPE:
4560 case SIGILL:
4561 case SIGXFSZ:
4562 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4563 break;
4564
4565 case SHUTDOWN1_SIGNAL:
4566 case SHUTDOWN2_SIGNAL:
4567 case SHUTDOWN3_SIGNAL:
4568 case BREAK_SIGNAL:
4569 jvmHandler = (address)user_handler();
4570 break;
4571
4572 case INTERRUPT_SIGNAL:
4573 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4574 break;
4575
4576 default:
4577 if (sig == SR_signum) {
4578 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4579 } else {
4580 return;
4581 }
4582 break;
4583 }
4584
4585 if (thisHandler != jvmHandler) {
4586 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4587 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4588 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4589 // No need to check this sig any longer
4590 sigaddset(&check_signal_done, sig);
4591 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4592 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4593 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4594 exception_name(sig, buf, O_BUFLEN));
4595 }
4596 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4597 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4598 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4599 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4600 // No need to check this sig any longer
4601 sigaddset(&check_signal_done, sig);
4602 }
4603
4604 // Dump all the signal
4605 if (sigismember(&check_signal_done, sig)) {
4606 print_signal_handlers(tty, buf, O_BUFLEN);
4607 }
4608 }
4609
4610 extern void report_error(char* file_name, int line_no, char* title,
4611 char* format, ...);
4612
4613 extern bool signal_name(int signo, char* buf, size_t len);
4614
4615 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4616 if (0 < exception_code && exception_code <= SIGRTMAX) {
4617 // signal
4618 if (!signal_name(exception_code, buf, size)) {
4619 jio_snprintf(buf, size, "SIG%d", exception_code);
4620 }
4621 return buf;
4622 } else {
4623 return NULL;
4624 }
4625 }
4626
4627 // this is called _before_ the most of global arguments have been parsed
4628 void os::init(void) {
4629 char dummy; // used to get a guess on initial stack address
4630 // first_hrtime = gethrtime();
4631
4632 // With LinuxThreads the JavaMain thread pid (primordial thread)
4633 // is different than the pid of the java launcher thread.
4634 // So, on Linux, the launcher thread pid is passed to the VM
4635 // via the sun.java.launcher.pid property.
4636 // Use this property instead of getpid() if it was correctly passed.
4637 // See bug 6351349.
4638 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4639
4640 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4641
4642 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4643
4644 init_random(1234567);
4645
4646 ThreadCritical::initialize();
4647
4648 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4649 if (Linux::page_size() == -1) {
4650 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4651 strerror(errno)));
4652 }
4653 init_page_sizes((size_t) Linux::page_size());
4654
4655 Linux::initialize_system_info();
4656
4657 // main_thread points to the aboriginal thread
4658 Linux::_main_thread = pthread_self();
4659
4660 Linux::clock_init();
4661 initial_time_count = javaTimeNanos();
4662
4663 // pthread_condattr initialization for monotonic clock
4664 int status;
4665 pthread_condattr_t* _condattr = os::Linux::condAttr();
4666 if ((status = pthread_condattr_init(_condattr)) != 0) {
4667 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4668 }
4669 // Only set the clock if CLOCK_MONOTONIC is available
4670 if (os::supports_monotonic_clock()) {
4671 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4672 if (status == EINVAL) {
4673 warning("Unable to use monotonic clock with relative timed-waits" \
4674 " - changes to the time-of-day clock may have adverse affects");
4675 } else {
4676 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4677 }
4678 }
4679 }
4680 // else it defaults to CLOCK_REALTIME
4681
4682 pthread_mutex_init(&dl_mutex, NULL);
4683
4684 // If the pagesize of the VM is greater than 8K determine the appropriate
4685 // number of initial guard pages. The user can change this with the
4686 // command line arguments, if needed.
4687 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4688 StackYellowPages = 1;
4689 StackRedPages = 1;
4690 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4691 }
4692
4693 // retrieve entry point for pthread_setname_np
4694 Linux::_pthread_setname_np =
4695 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4696
4697 }
4698
4699 // To install functions for atexit system call
4700 extern "C" {
4701 static void perfMemory_exit_helper() {
4702 perfMemory_exit();
4703 }
4704 }
4705
4706 // this is called _after_ the global arguments have been parsed
4707 jint os::init_2(void) {
4708 Linux::fast_thread_clock_init();
4709
4710 // Allocate a single page and mark it as readable for safepoint polling
4711 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4712 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4713
4714 os::set_polling_page(polling_page);
4715
4716 #ifndef PRODUCT
4717 if (Verbose && PrintMiscellaneous) {
4718 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4719 (intptr_t)polling_page);
4720 }
4721 #endif
4722
4723 if (!UseMembar) {
4724 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4725 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4726 os::set_memory_serialize_page(mem_serialize_page);
4727
4728 #ifndef PRODUCT
4729 if (Verbose && PrintMiscellaneous) {
4730 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4731 (intptr_t)mem_serialize_page);
4732 }
4733 #endif
4734 }
4735
4736 // initialize suspend/resume support - must do this before signal_sets_init()
4737 if (SR_initialize() != 0) {
4738 perror("SR_initialize failed");
4739 return JNI_ERR;
4740 }
4741
4742 Linux::signal_sets_init();
4743 Linux::install_signal_handlers();
4744
4745 // Check minimum allowable stack size for thread creation and to initialize
4746 // the java system classes, including StackOverflowError - depends on page
4747 // size. Add a page for compiler2 recursion in main thread.
4748 // Add in 2*BytesPerWord times page size to account for VM stack during
4749 // class initialization depending on 32 or 64 bit VM.
4750 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4751 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4752 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4753
4754 size_t threadStackSizeInBytes = ThreadStackSize * K;
4755 if (threadStackSizeInBytes != 0 &&
4756 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4757 tty->print_cr("\nThe stack size specified is too small, "
4758 "Specify at least %dk",
4759 os::Linux::min_stack_allowed/ K);
4760 return JNI_ERR;
4761 }
4762
4763 // Make the stack size a multiple of the page size so that
4764 // the yellow/red zones can be guarded.
4765 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4766 vm_page_size()));
4767
4768 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4769
4770 #if defined(IA32)
4771 workaround_expand_exec_shield_cs_limit();
4772 #endif
4773
4774 Linux::libpthread_init();
4775 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4776 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4777 Linux::glibc_version(), Linux::libpthread_version(),
4778 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4779 }
4780
4781 if (UseNUMA) {
4782 if (!Linux::libnuma_init()) {
4783 UseNUMA = false;
4784 } else {
4785 if ((Linux::numa_max_node() < 1)) {
4786 // There's only one node(they start from 0), disable NUMA.
4787 UseNUMA = false;
4788 }
4789 }
4790 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4791 // we can make the adaptive lgrp chunk resizing work. If the user specified
4792 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4793 // disable adaptive resizing.
4794 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4795 if (FLAG_IS_DEFAULT(UseNUMA)) {
4796 UseNUMA = false;
4797 } else {
4798 if (FLAG_IS_DEFAULT(UseLargePages) &&
4799 FLAG_IS_DEFAULT(UseSHM) &&
4800 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4801 UseLargePages = false;
4802 } else {
4803 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4804 UseAdaptiveSizePolicy = false;
4805 UseAdaptiveNUMAChunkSizing = false;
4806 }
4807 }
4808 }
4809 if (!UseNUMA && ForceNUMA) {
4810 UseNUMA = true;
4811 }
4812 }
4813
4814 if (MaxFDLimit) {
4815 // set the number of file descriptors to max. print out error
4816 // if getrlimit/setrlimit fails but continue regardless.
4817 struct rlimit nbr_files;
4818 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4819 if (status != 0) {
4820 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4821 perror("os::init_2 getrlimit failed");
4822 }
4823 } else {
4824 nbr_files.rlim_cur = nbr_files.rlim_max;
4825 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4826 if (status != 0) {
4827 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4828 perror("os::init_2 setrlimit failed");
4829 }
4830 }
4831 }
4832 }
4833
4834 // Initialize lock used to serialize thread creation (see os::create_thread)
4835 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4836
4837 // at-exit methods are called in the reverse order of their registration.
4838 // atexit functions are called on return from main or as a result of a
4839 // call to exit(3C). There can be only 32 of these functions registered
4840 // and atexit() does not set errno.
4841
4842 if (PerfAllowAtExitRegistration) {
4843 // only register atexit functions if PerfAllowAtExitRegistration is set.
4844 // atexit functions can be delayed until process exit time, which
4845 // can be problematic for embedded VM situations. Embedded VMs should
4846 // call DestroyJavaVM() to assure that VM resources are released.
4847
4848 // note: perfMemory_exit_helper atexit function may be removed in
4849 // the future if the appropriate cleanup code can be added to the
4850 // VM_Exit VMOperation's doit method.
4851 if (atexit(perfMemory_exit_helper) != 0) {
4852 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4853 }
4854 }
4855
4856 // initialize thread priority policy
4857 prio_init();
4858
4859 return JNI_OK;
4860 }
4861
4862 // Mark the polling page as unreadable
4863 void os::make_polling_page_unreadable(void) {
4864 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4865 fatal("Could not disable polling page");
4866 }
4867 }
4868
4869 // Mark the polling page as readable
4870 void os::make_polling_page_readable(void) {
4871 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4872 fatal("Could not enable polling page");
4873 }
4874 }
4875
4876 int os::active_processor_count() {
4877 // Linux doesn't yet have a (official) notion of processor sets,
4878 // so just return the number of online processors.
4879 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4880 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4881 return online_cpus;
4882 }
4883
4884 void os::set_native_thread_name(const char *name) {
4885 if (Linux::_pthread_setname_np) {
4886 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4887 snprintf(buf, sizeof(buf), "%s", name);
4888 buf[sizeof(buf) - 1] = '\0';
4889 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4890 // ERANGE should not happen; all other errors should just be ignored.
4891 assert(rc != ERANGE, "pthread_setname_np failed");
4892 }
4893 }
4894
4895 bool os::distribute_processes(uint length, uint* distribution) {
4896 // Not yet implemented.
4897 return false;
4898 }
4899
4900 bool os::bind_to_processor(uint processor_id) {
4901 // Not yet implemented.
4902 return false;
4903 }
4904
4905 ///
4906
4907 void os::SuspendedThreadTask::internal_do_task() {
4908 if (do_suspend(_thread->osthread())) {
4909 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4910 do_task(context);
4911 do_resume(_thread->osthread());
4912 }
4913 }
4914
4915 class PcFetcher : public os::SuspendedThreadTask {
4916 public:
4917 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4918 ExtendedPC result();
4919 protected:
4920 void do_task(const os::SuspendedThreadTaskContext& context);
4921 private:
4922 ExtendedPC _epc;
4923 };
4924
4925 ExtendedPC PcFetcher::result() {
4926 guarantee(is_done(), "task is not done yet.");
4927 return _epc;
4928 }
4929
4930 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4931 Thread* thread = context.thread();
4932 OSThread* osthread = thread->osthread();
4933 if (osthread->ucontext() != NULL) {
4934 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4935 } else {
4936 // NULL context is unexpected, double-check this is the VMThread
4937 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4938 }
4939 }
4940
4941 // Suspends the target using the signal mechanism and then grabs the PC before
4942 // resuming the target. Used by the flat-profiler only
4943 ExtendedPC os::get_thread_pc(Thread* thread) {
4944 // Make sure that it is called by the watcher for the VMThread
4945 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4946 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4947
4948 PcFetcher fetcher(thread);
4949 fetcher.run();
4950 return fetcher.result();
4951 }
4952
4953 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4954 pthread_mutex_t *_mutex,
4955 const struct timespec *_abstime) {
4956 if (is_NPTL()) {
4957 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4958 } else {
4959 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4960 // word back to default 64bit precision if condvar is signaled. Java
4961 // wants 53bit precision. Save and restore current value.
4962 int fpu = get_fpu_control_word();
4963 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4964 set_fpu_control_word(fpu);
4965 return status;
4966 }
4967 }
4968
4969 ////////////////////////////////////////////////////////////////////////////////
4970 // debug support
4971
4972 bool os::find(address addr, outputStream* st) {
4973 Dl_info dlinfo;
4974 memset(&dlinfo, 0, sizeof(dlinfo));
4975 if (dladdr(addr, &dlinfo) != 0) {
4976 st->print(PTR_FORMAT ": ", addr);
4977 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4978 st->print("%s+%#x", dlinfo.dli_sname,
4979 addr - (intptr_t)dlinfo.dli_saddr);
4980 } else if (dlinfo.dli_fbase != NULL) {
4981 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4982 } else {
4983 st->print("<absolute address>");
4984 }
4985 if (dlinfo.dli_fname != NULL) {
4986 st->print(" in %s", dlinfo.dli_fname);
4987 }
4988 if (dlinfo.dli_fbase != NULL) {
4989 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4990 }
4991 st->cr();
4992
4993 if (Verbose) {
4994 // decode some bytes around the PC
4995 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4996 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4997 address lowest = (address) dlinfo.dli_sname;
4998 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4999 if (begin < lowest) begin = lowest;
5000 Dl_info dlinfo2;
5001 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5002 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5003 end = (address) dlinfo2.dli_saddr;
5004 }
5005 Disassembler::decode(begin, end, st);
5006 }
5007 return true;
5008 }
5009 return false;
5010 }
5011
5012 ////////////////////////////////////////////////////////////////////////////////
5013 // misc
5014
5015 // This does not do anything on Linux. This is basically a hook for being
5016 // able to use structured exception handling (thread-local exception filters)
5017 // on, e.g., Win32.
5018 void
5019 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5020 JavaCallArguments* args, Thread* thread) {
5021 f(value, method, args, thread);
5022 }
5023
5024 void os::print_statistics() {
5025 }
5026
5027 int os::message_box(const char* title, const char* message) {
5028 int i;
5029 fdStream err(defaultStream::error_fd());
5030 for (i = 0; i < 78; i++) err.print_raw("=");
5031 err.cr();
5032 err.print_raw_cr(title);
5033 for (i = 0; i < 78; i++) err.print_raw("-");
5034 err.cr();
5035 err.print_raw_cr(message);
5036 for (i = 0; i < 78; i++) err.print_raw("=");
5037 err.cr();
5038
5039 char buf[16];
5040 // Prevent process from exiting upon "read error" without consuming all CPU
5041 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5042
5043 return buf[0] == 'y' || buf[0] == 'Y';
5044 }
5045
5046 int os::stat(const char *path, struct stat *sbuf) {
5047 char pathbuf[MAX_PATH];
5048 if (strlen(path) > MAX_PATH - 1) {
5049 errno = ENAMETOOLONG;
5050 return -1;
5051 }
5052 os::native_path(strcpy(pathbuf, path));
5053 return ::stat(pathbuf, sbuf);
5054 }
5055
5056 bool os::check_heap(bool force) {
5057 return true;
5058 }
5059
5060 // Is a (classpath) directory empty?
5061 bool os::dir_is_empty(const char* path) {
5062 DIR *dir = NULL;
5063 struct dirent *ptr;
5064
5065 dir = opendir(path);
5066 if (dir == NULL) return true;
5067
5068 // Scan the directory
5069 bool result = true;
5070 char buf[sizeof(struct dirent) + MAX_PATH];
5071 while (result && (ptr = ::readdir(dir)) != NULL) {
5072 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5073 result = false;
5074 }
5075 }
5076 closedir(dir);
5077 return result;
5078 }
5079
5080 // This code originates from JDK's sysOpen and open64_w
5081 // from src/solaris/hpi/src/system_md.c
5082
5083 int os::open(const char *path, int oflag, int mode) {
5084 if (strlen(path) > MAX_PATH - 1) {
5085 errno = ENAMETOOLONG;
5086 return -1;
5087 }
5088
5089 // All file descriptors that are opened in the Java process and not
5090 // specifically destined for a subprocess should have the close-on-exec
5091 // flag set. If we don't set it, then careless 3rd party native code
5092 // might fork and exec without closing all appropriate file descriptors
5093 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5094 // turn might:
5095 //
5096 // - cause end-of-file to fail to be detected on some file
5097 // descriptors, resulting in mysterious hangs, or
5098 //
5099 // - might cause an fopen in the subprocess to fail on a system
5100 // suffering from bug 1085341.
5101 //
5102 // (Yes, the default setting of the close-on-exec flag is a Unix
5103 // design flaw)
5104 //
5105 // See:
5106 // 1085341: 32-bit stdio routines should support file descriptors >255
5107 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5108 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5109 //
5110 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5111 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5112 // because it saves a system call and removes a small window where the flag
5113 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5114 // and we fall back to using FD_CLOEXEC (see below).
5115 #ifdef O_CLOEXEC
5116 oflag |= O_CLOEXEC;
5117 #endif
5118
5119 int fd = ::open64(path, oflag, mode);
5120 if (fd == -1) return -1;
5121
5122 //If the open succeeded, the file might still be a directory
5123 {
5124 struct stat64 buf64;
5125 int ret = ::fstat64(fd, &buf64);
5126 int st_mode = buf64.st_mode;
5127
5128 if (ret != -1) {
5129 if ((st_mode & S_IFMT) == S_IFDIR) {
5130 errno = EISDIR;
5131 ::close(fd);
5132 return -1;
5133 }
5134 } else {
5135 ::close(fd);
5136 return -1;
5137 }
5138 }
5139
5140 #ifdef FD_CLOEXEC
5141 // Validate that the use of the O_CLOEXEC flag on open above worked.
5142 // With recent kernels, we will perform this check exactly once.
5143 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5144 if (!O_CLOEXEC_is_known_to_work) {
5145 int flags = ::fcntl(fd, F_GETFD);
5146 if (flags != -1) {
5147 if ((flags & FD_CLOEXEC) != 0)
5148 O_CLOEXEC_is_known_to_work = 1;
5149 else
5150 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5151 }
5152 }
5153 #endif
5154
5155 return fd;
5156 }
5157
5158
5159 // create binary file, rewriting existing file if required
5160 int os::create_binary_file(const char* path, bool rewrite_existing) {
5161 int oflags = O_WRONLY | O_CREAT;
5162 if (!rewrite_existing) {
5163 oflags |= O_EXCL;
5164 }
5165 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5166 }
5167
5168 // return current position of file pointer
5169 jlong os::current_file_offset(int fd) {
5170 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5171 }
5172
5173 // move file pointer to the specified offset
5174 jlong os::seek_to_file_offset(int fd, jlong offset) {
5175 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5176 }
5177
5178 // This code originates from JDK's sysAvailable
5179 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5180
5181 int os::available(int fd, jlong *bytes) {
5182 jlong cur, end;
5183 int mode;
5184 struct stat64 buf64;
5185
5186 if (::fstat64(fd, &buf64) >= 0) {
5187 mode = buf64.st_mode;
5188 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5189 // XXX: is the following call interruptible? If so, this might
5190 // need to go through the INTERRUPT_IO() wrapper as for other
5191 // blocking, interruptible calls in this file.
5192 int n;
5193 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5194 *bytes = n;
5195 return 1;
5196 }
5197 }
5198 }
5199 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5200 return 0;
5201 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5202 return 0;
5203 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5204 return 0;
5205 }
5206 *bytes = end - cur;
5207 return 1;
5208 }
5209
5210 // Map a block of memory.
5211 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5212 char *addr, size_t bytes, bool read_only,
5213 bool allow_exec) {
5214 int prot;
5215 int flags = MAP_PRIVATE;
5216
5217 if (read_only) {
5218 prot = PROT_READ;
5219 } else {
5220 prot = PROT_READ | PROT_WRITE;
5221 }
5222
5223 if (allow_exec) {
5224 prot |= PROT_EXEC;
5225 }
5226
5227 if (addr != NULL) {
5228 flags |= MAP_FIXED;
5229 }
5230
5231 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5232 fd, file_offset);
5233 if (mapped_address == MAP_FAILED) {
5234 return NULL;
5235 }
5236 return mapped_address;
5237 }
5238
5239
5240 // Remap a block of memory.
5241 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5242 char *addr, size_t bytes, bool read_only,
5243 bool allow_exec) {
5244 // same as map_memory() on this OS
5245 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5246 allow_exec);
5247 }
5248
5249
5250 // Unmap a block of memory.
5251 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5252 return munmap(addr, bytes) == 0;
5253 }
5254
5255 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5256
5257 static clockid_t thread_cpu_clockid(Thread* thread) {
5258 pthread_t tid = thread->osthread()->pthread_id();
5259 clockid_t clockid;
5260
5261 // Get thread clockid
5262 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5263 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5264 return clockid;
5265 }
5266
5267 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5268 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5269 // of a thread.
5270 //
5271 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5272 // the fast estimate available on the platform.
5273
5274 jlong os::current_thread_cpu_time() {
5275 if (os::Linux::supports_fast_thread_cpu_time()) {
5276 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5277 } else {
5278 // return user + sys since the cost is the same
5279 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5280 }
5281 }
5282
5283 jlong os::thread_cpu_time(Thread* thread) {
5284 // consistent with what current_thread_cpu_time() returns
5285 if (os::Linux::supports_fast_thread_cpu_time()) {
5286 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5287 } else {
5288 return slow_thread_cpu_time(thread, true /* user + sys */);
5289 }
5290 }
5291
5292 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5293 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5294 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5295 } else {
5296 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5297 }
5298 }
5299
5300 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5301 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5302 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5303 } else {
5304 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5305 }
5306 }
5307
5308 // -1 on error.
5309 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5310 pid_t tid = thread->osthread()->thread_id();
5311 char *s;
5312 char stat[2048];
5313 int statlen;
5314 char proc_name[64];
5315 int count;
5316 long sys_time, user_time;
5317 char cdummy;
5318 int idummy;
5319 long ldummy;
5320 FILE *fp;
5321
5322 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5323 fp = fopen(proc_name, "r");
5324 if (fp == NULL) return -1;
5325 statlen = fread(stat, 1, 2047, fp);
5326 stat[statlen] = '\0';
5327 fclose(fp);
5328
5329 // Skip pid and the command string. Note that we could be dealing with
5330 // weird command names, e.g. user could decide to rename java launcher
5331 // to "java 1.4.2 :)", then the stat file would look like
5332 // 1234 (java 1.4.2 :)) R ... ...
5333 // We don't really need to know the command string, just find the last
5334 // occurrence of ")" and then start parsing from there. See bug 4726580.
5335 s = strrchr(stat, ')');
5336 if (s == NULL) return -1;
5337
5338 // Skip blank chars
5339 do { s++; } while (s && isspace(*s));
5340
5341 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5342 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5343 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5344 &user_time, &sys_time);
5345 if (count != 13) return -1;
5346 if (user_sys_cpu_time) {
5347 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5348 } else {
5349 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5350 }
5351 }
5352
5353 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5354 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5355 info_ptr->may_skip_backward = false; // elapsed time not wall time
5356 info_ptr->may_skip_forward = false; // elapsed time not wall time
5357 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5358 }
5359
5360 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5361 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5362 info_ptr->may_skip_backward = false; // elapsed time not wall time
5363 info_ptr->may_skip_forward = false; // elapsed time not wall time
5364 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5365 }
5366
5367 bool os::is_thread_cpu_time_supported() {
5368 return true;
5369 }
5370
5371 // System loadavg support. Returns -1 if load average cannot be obtained.
5372 // Linux doesn't yet have a (official) notion of processor sets,
5373 // so just return the system wide load average.
5374 int os::loadavg(double loadavg[], int nelem) {
5375 return ::getloadavg(loadavg, nelem);
5376 }
5377
5378 void os::pause() {
5379 char filename[MAX_PATH];
5380 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5381 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5382 } else {
5383 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5384 }
5385
5386 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5387 if (fd != -1) {
5388 struct stat buf;
5389 ::close(fd);
5390 while (::stat(filename, &buf) == 0) {
5391 (void)::poll(NULL, 0, 100);
5392 }
5393 } else {
5394 jio_fprintf(stderr,
5395 "Could not open pause file '%s', continuing immediately.\n", filename);
5396 }
5397 }
5398
5399
5400 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5401 // comment paragraphs are worth repeating here:
5402 //
5403 // Assumption:
5404 // Only one parker can exist on an event, which is why we allocate
5405 // them per-thread. Multiple unparkers can coexist.
5406 //
5407 // _Event serves as a restricted-range semaphore.
5408 // -1 : thread is blocked, i.e. there is a waiter
5409 // 0 : neutral: thread is running or ready,
5410 // could have been signaled after a wait started
5411 // 1 : signaled - thread is running or ready
5412 //
5413 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5414 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5415 // For specifics regarding the bug see GLIBC BUGID 261237 :
5416 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5417 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5418 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5419 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5420 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5421 // and monitorenter when we're using 1-0 locking. All those operations may result in
5422 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5423 // of libpthread avoids the problem, but isn't practical.
5424 //
5425 // Possible remedies:
5426 //
5427 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5428 // This is palliative and probabilistic, however. If the thread is preempted
5429 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5430 // than the minimum period may have passed, and the abstime may be stale (in the
5431 // past) resultin in a hang. Using this technique reduces the odds of a hang
5432 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5433 //
5434 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5435 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5436 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5437 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5438 // thread.
5439 //
5440 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5441 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5442 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5443 // This also works well. In fact it avoids kernel-level scalability impediments
5444 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5445 // timers in a graceful fashion.
5446 //
5447 // 4. When the abstime value is in the past it appears that control returns
5448 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5449 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5450 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5451 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5452 // It may be possible to avoid reinitialization by checking the return
5453 // value from pthread_cond_timedwait(). In addition to reinitializing the
5454 // condvar we must establish the invariant that cond_signal() is only called
5455 // within critical sections protected by the adjunct mutex. This prevents
5456 // cond_signal() from "seeing" a condvar that's in the midst of being
5457 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5458 // desirable signal-after-unlock optimization that avoids futile context switching.
5459 //
5460 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5461 // structure when a condvar is used or initialized. cond_destroy() would
5462 // release the helper structure. Our reinitialize-after-timedwait fix
5463 // put excessive stress on malloc/free and locks protecting the c-heap.
5464 //
5465 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5466 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5467 // and only enabling the work-around for vulnerable environments.
5468
5469 // utility to compute the abstime argument to timedwait:
5470 // millis is the relative timeout time
5471 // abstime will be the absolute timeout time
5472 // TODO: replace compute_abstime() with unpackTime()
5473
5474 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5475 if (millis < 0) millis = 0;
5476
5477 jlong seconds = millis / 1000;
5478 millis %= 1000;
5479 if (seconds > 50000000) { // see man cond_timedwait(3T)
5480 seconds = 50000000;
5481 }
5482
5483 if (os::supports_monotonic_clock()) {
5484 struct timespec now;
5485 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5486 assert_status(status == 0, status, "clock_gettime");
5487 abstime->tv_sec = now.tv_sec + seconds;
5488 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5489 if (nanos >= NANOSECS_PER_SEC) {
5490 abstime->tv_sec += 1;
5491 nanos -= NANOSECS_PER_SEC;
5492 }
5493 abstime->tv_nsec = nanos;
5494 } else {
5495 struct timeval now;
5496 int status = gettimeofday(&now, NULL);
5497 assert(status == 0, "gettimeofday");
5498 abstime->tv_sec = now.tv_sec + seconds;
5499 long usec = now.tv_usec + millis * 1000;
5500 if (usec >= 1000000) {
5501 abstime->tv_sec += 1;
5502 usec -= 1000000;
5503 }
5504 abstime->tv_nsec = usec * 1000;
5505 }
5506 return abstime;
5507 }
5508
5509 void os::PlatformEvent::park() { // AKA "down()"
5510 // Transitions for _Event:
5511 // -1 => -1 : illegal
5512 // 1 => 0 : pass - return immediately
5513 // 0 => -1 : block; then set _Event to 0 before returning
5514
5515 // Invariant: Only the thread associated with the Event/PlatformEvent
5516 // may call park().
5517 // TODO: assert that _Assoc != NULL or _Assoc == Self
5518 assert(_nParked == 0, "invariant");
5519
5520 int v;
5521 for (;;) {
5522 v = _Event;
5523 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5524 }
5525 guarantee(v >= 0, "invariant");
5526 if (v == 0) {
5527 // Do this the hard way by blocking ...
5528 int status = pthread_mutex_lock(_mutex);
5529 assert_status(status == 0, status, "mutex_lock");
5530 guarantee(_nParked == 0, "invariant");
5531 ++_nParked;
5532 while (_Event < 0) {
5533 status = pthread_cond_wait(_cond, _mutex);
5534 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5535 // Treat this the same as if the wait was interrupted
5536 if (status == ETIME) { status = EINTR; }
5537 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5538 }
5539 --_nParked;
5540
5541 _Event = 0;
5542 status = pthread_mutex_unlock(_mutex);
5543 assert_status(status == 0, status, "mutex_unlock");
5544 // Paranoia to ensure our locked and lock-free paths interact
5545 // correctly with each other.
5546 OrderAccess::fence();
5547 }
5548 guarantee(_Event >= 0, "invariant");
5549 }
5550
5551 int os::PlatformEvent::park(jlong millis) {
5552 // Transitions for _Event:
5553 // -1 => -1 : illegal
5554 // 1 => 0 : pass - return immediately
5555 // 0 => -1 : block; then set _Event to 0 before returning
5556
5557 guarantee(_nParked == 0, "invariant");
5558
5559 int v;
5560 for (;;) {
5561 v = _Event;
5562 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5563 }
5564 guarantee(v >= 0, "invariant");
5565 if (v != 0) return OS_OK;
5566
5567 // We do this the hard way, by blocking the thread.
5568 // Consider enforcing a minimum timeout value.
5569 struct timespec abst;
5570 compute_abstime(&abst, millis);
5571
5572 int ret = OS_TIMEOUT;
5573 int status = pthread_mutex_lock(_mutex);
5574 assert_status(status == 0, status, "mutex_lock");
5575 guarantee(_nParked == 0, "invariant");
5576 ++_nParked;
5577
5578 // Object.wait(timo) will return because of
5579 // (a) notification
5580 // (b) timeout
5581 // (c) thread.interrupt
5582 //
5583 // Thread.interrupt and object.notify{All} both call Event::set.
5584 // That is, we treat thread.interrupt as a special case of notification.
5585 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5586 // We assume all ETIME returns are valid.
5587 //
5588 // TODO: properly differentiate simultaneous notify+interrupt.
5589 // In that case, we should propagate the notify to another waiter.
5590
5591 while (_Event < 0) {
5592 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5593 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5594 pthread_cond_destroy(_cond);
5595 pthread_cond_init(_cond, os::Linux::condAttr());
5596 }
5597 assert_status(status == 0 || status == EINTR ||
5598 status == ETIME || status == ETIMEDOUT,
5599 status, "cond_timedwait");
5600 if (!FilterSpuriousWakeups) break; // previous semantics
5601 if (status == ETIME || status == ETIMEDOUT) break;
5602 // We consume and ignore EINTR and spurious wakeups.
5603 }
5604 --_nParked;
5605 if (_Event >= 0) {
5606 ret = OS_OK;
5607 }
5608 _Event = 0;
5609 status = pthread_mutex_unlock(_mutex);
5610 assert_status(status == 0, status, "mutex_unlock");
5611 assert(_nParked == 0, "invariant");
5612 // Paranoia to ensure our locked and lock-free paths interact
5613 // correctly with each other.
5614 OrderAccess::fence();
5615 return ret;
5616 }
5617
5618 void os::PlatformEvent::unpark() {
5619 // Transitions for _Event:
5620 // 0 => 1 : just return
5621 // 1 => 1 : just return
5622 // -1 => either 0 or 1; must signal target thread
5623 // That is, we can safely transition _Event from -1 to either
5624 // 0 or 1.
5625 // See also: "Semaphores in Plan 9" by Mullender & Cox
5626 //
5627 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5628 // that it will take two back-to-back park() calls for the owning
5629 // thread to block. This has the benefit of forcing a spurious return
5630 // from the first park() call after an unpark() call which will help
5631 // shake out uses of park() and unpark() without condition variables.
5632
5633 if (Atomic::xchg(1, &_Event) >= 0) return;
5634
5635 // Wait for the thread associated with the event to vacate
5636 int status = pthread_mutex_lock(_mutex);
5637 assert_status(status == 0, status, "mutex_lock");
5638 int AnyWaiters = _nParked;
5639 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5640 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5641 AnyWaiters = 0;
5642 pthread_cond_signal(_cond);
5643 }
5644 status = pthread_mutex_unlock(_mutex);
5645 assert_status(status == 0, status, "mutex_unlock");
5646 if (AnyWaiters != 0) {
5647 // Note that we signal() *after* dropping the lock for "immortal" Events.
5648 // This is safe and avoids a common class of futile wakeups. In rare
5649 // circumstances this can cause a thread to return prematurely from
5650 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5651 // will simply re-test the condition and re-park itself.
5652 // This provides particular benefit if the underlying platform does not
5653 // provide wait morphing.
5654 status = pthread_cond_signal(_cond);
5655 assert_status(status == 0, status, "cond_signal");
5656 }
5657 }
5658
5659
5660 // JSR166
5661 // -------------------------------------------------------
5662
5663 // The solaris and linux implementations of park/unpark are fairly
5664 // conservative for now, but can be improved. They currently use a
5665 // mutex/condvar pair, plus a a count.
5666 // Park decrements count if > 0, else does a condvar wait. Unpark
5667 // sets count to 1 and signals condvar. Only one thread ever waits
5668 // on the condvar. Contention seen when trying to park implies that someone
5669 // is unparking you, so don't wait. And spurious returns are fine, so there
5670 // is no need to track notifications.
5671
5672 // This code is common to linux and solaris and will be moved to a
5673 // common place in dolphin.
5674 //
5675 // The passed in time value is either a relative time in nanoseconds
5676 // or an absolute time in milliseconds. Either way it has to be unpacked
5677 // into suitable seconds and nanoseconds components and stored in the
5678 // given timespec structure.
5679 // Given time is a 64-bit value and the time_t used in the timespec is only
5680 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5681 // overflow if times way in the future are given. Further on Solaris versions
5682 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5683 // number of seconds, in abstime, is less than current_time + 100,000,000.
5684 // As it will be 28 years before "now + 100000000" will overflow we can
5685 // ignore overflow and just impose a hard-limit on seconds using the value
5686 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5687 // years from "now".
5688
5689 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5690 assert(time > 0, "convertTime");
5691 time_t max_secs = 0;
5692
5693 if (!os::supports_monotonic_clock() || isAbsolute) {
5694 struct timeval now;
5695 int status = gettimeofday(&now, NULL);
5696 assert(status == 0, "gettimeofday");
5697
5698 max_secs = now.tv_sec + MAX_SECS;
5699
5700 if (isAbsolute) {
5701 jlong secs = time / 1000;
5702 if (secs > max_secs) {
5703 absTime->tv_sec = max_secs;
5704 } else {
5705 absTime->tv_sec = secs;
5706 }
5707 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5708 } else {
5709 jlong secs = time / NANOSECS_PER_SEC;
5710 if (secs >= MAX_SECS) {
5711 absTime->tv_sec = max_secs;
5712 absTime->tv_nsec = 0;
5713 } else {
5714 absTime->tv_sec = now.tv_sec + secs;
5715 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5716 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5717 absTime->tv_nsec -= NANOSECS_PER_SEC;
5718 ++absTime->tv_sec; // note: this must be <= max_secs
5719 }
5720 }
5721 }
5722 } else {
5723 // must be relative using monotonic clock
5724 struct timespec now;
5725 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5726 assert_status(status == 0, status, "clock_gettime");
5727 max_secs = now.tv_sec + MAX_SECS;
5728 jlong secs = time / NANOSECS_PER_SEC;
5729 if (secs >= MAX_SECS) {
5730 absTime->tv_sec = max_secs;
5731 absTime->tv_nsec = 0;
5732 } else {
5733 absTime->tv_sec = now.tv_sec + secs;
5734 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5735 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5736 absTime->tv_nsec -= NANOSECS_PER_SEC;
5737 ++absTime->tv_sec; // note: this must be <= max_secs
5738 }
5739 }
5740 }
5741 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5742 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5743 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5744 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5745 }
5746
5747 void Parker::park(bool isAbsolute, jlong time) {
5748 // Ideally we'd do something useful while spinning, such
5749 // as calling unpackTime().
5750
5751 // Optional fast-path check:
5752 // Return immediately if a permit is available.
5753 // We depend on Atomic::xchg() having full barrier semantics
5754 // since we are doing a lock-free update to _counter.
5755 if (Atomic::xchg(0, &_counter) > 0) return;
5756
5757 Thread* thread = Thread::current();
5758 assert(thread->is_Java_thread(), "Must be JavaThread");
5759 JavaThread *jt = (JavaThread *)thread;
5760
5761 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5762 // Check interrupt before trying to wait
5763 if (Thread::is_interrupted(thread, false)) {
5764 return;
5765 }
5766
5767 // Next, demultiplex/decode time arguments
5768 timespec absTime;
5769 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5770 return;
5771 }
5772 if (time > 0) {
5773 unpackTime(&absTime, isAbsolute, time);
5774 }
5775
5776
5777 // Enter safepoint region
5778 // Beware of deadlocks such as 6317397.
5779 // The per-thread Parker:: mutex is a classic leaf-lock.
5780 // In particular a thread must never block on the Threads_lock while
5781 // holding the Parker:: mutex. If safepoints are pending both the
5782 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5783 ThreadBlockInVM tbivm(jt);
5784
5785 // Don't wait if cannot get lock since interference arises from
5786 // unblocking. Also. check interrupt before trying wait
5787 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5788 return;
5789 }
5790
5791 int status;
5792 if (_counter > 0) { // no wait needed
5793 _counter = 0;
5794 status = pthread_mutex_unlock(_mutex);
5795 assert(status == 0, "invariant");
5796 // Paranoia to ensure our locked and lock-free paths interact
5797 // correctly with each other and Java-level accesses.
5798 OrderAccess::fence();
5799 return;
5800 }
5801
5802 #ifdef ASSERT
5803 // Don't catch signals while blocked; let the running threads have the signals.
5804 // (This allows a debugger to break into the running thread.)
5805 sigset_t oldsigs;
5806 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5807 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5808 #endif
5809
5810 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5811 jt->set_suspend_equivalent();
5812 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5813
5814 assert(_cur_index == -1, "invariant");
5815 if (time == 0) {
5816 _cur_index = REL_INDEX; // arbitrary choice when not timed
5817 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5818 } else {
5819 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5820 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5821 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5822 pthread_cond_destroy(&_cond[_cur_index]);
5823 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5824 }
5825 }
5826 _cur_index = -1;
5827 assert_status(status == 0 || status == EINTR ||
5828 status == ETIME || status == ETIMEDOUT,
5829 status, "cond_timedwait");
5830
5831 #ifdef ASSERT
5832 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5833 #endif
5834
5835 _counter = 0;
5836 status = pthread_mutex_unlock(_mutex);
5837 assert_status(status == 0, status, "invariant");
5838 // Paranoia to ensure our locked and lock-free paths interact
5839 // correctly with each other and Java-level accesses.
5840 OrderAccess::fence();
5841
5842 // If externally suspended while waiting, re-suspend
5843 if (jt->handle_special_suspend_equivalent_condition()) {
5844 jt->java_suspend_self();
5845 }
5846 }
5847
5848 void Parker::unpark() {
5849 int status = pthread_mutex_lock(_mutex);
5850 assert(status == 0, "invariant");
5851 const int s = _counter;
5852 _counter = 1;
5853 if (s < 1) {
5854 // thread might be parked
5855 if (_cur_index != -1) {
5856 // thread is definitely parked
5857 if (WorkAroundNPTLTimedWaitHang) {
5858 status = pthread_cond_signal(&_cond[_cur_index]);
5859 assert(status == 0, "invariant");
5860 status = pthread_mutex_unlock(_mutex);
5861 assert(status == 0, "invariant");
5862 } else {
5863 status = pthread_mutex_unlock(_mutex);
5864 assert(status == 0, "invariant");
5865 status = pthread_cond_signal(&_cond[_cur_index]);
5866 assert(status == 0, "invariant");
5867 }
5868 } else {
5869 pthread_mutex_unlock(_mutex);
5870 assert(status == 0, "invariant");
5871 }
5872 } else {
5873 pthread_mutex_unlock(_mutex);
5874 assert(status == 0, "invariant");
5875 }
5876 }
5877
5878
5879 extern char** environ;
5880
5881 #ifndef __NR_fork
5882 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5883 #endif
5884
5885 #ifndef __NR_execve
5886 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5887 #endif
5888
5889 // Run the specified command in a separate process. Return its exit value,
5890 // or -1 on failure (e.g. can't fork a new process).
5891 // Unlike system(), this function can be called from signal handler. It
5892 // doesn't block SIGINT et al.
5893 int os::fork_and_exec(char* cmd) {
5894 const char * argv[4] = {"sh", "-c", cmd, NULL};
5895
5896 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5897 // pthread_atfork handlers and reset pthread library. All we need is a
5898 // separate process to execve. Make a direct syscall to fork process.
5899 // On IA64 there's no fork syscall, we have to use fork() and hope for
5900 // the best...
5901 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5902 IA64_ONLY(fork();)
5903
5904 if (pid < 0) {
5905 // fork failed
5906 return -1;
5907
5908 } else if (pid == 0) {
5909 // child process
5910
5911 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5912 // first to kill every thread on the thread list. Because this list is
5913 // not reset by fork() (see notes above), execve() will instead kill
5914 // every thread in the parent process. We know this is the only thread
5915 // in the new process, so make a system call directly.
5916 // IA64 should use normal execve() from glibc to match the glibc fork()
5917 // above.
5918 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5919 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5920
5921 // execve failed
5922 _exit(-1);
5923
5924 } else {
5925 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5926 // care about the actual exit code, for now.
5927
5928 int status;
5929
5930 // Wait for the child process to exit. This returns immediately if
5931 // the child has already exited. */
5932 while (waitpid(pid, &status, 0) < 0) {
5933 switch (errno) {
5934 case ECHILD: return 0;
5935 case EINTR: break;
5936 default: return -1;
5937 }
5938 }
5939
5940 if (WIFEXITED(status)) {
5941 // The child exited normally; get its exit code.
5942 return WEXITSTATUS(status);
5943 } else if (WIFSIGNALED(status)) {
5944 // The child exited because of a signal
5945 // The best value to return is 0x80 + signal number,
5946 // because that is what all Unix shells do, and because
5947 // it allows callers to distinguish between process exit and
5948 // process death by signal.
5949 return 0x80 + WTERMSIG(status);
5950 } else {
5951 // Unknown exit code; pass it through
5952 return status;
5953 }
5954 }
5955 }
5956
5957 // is_headless_jre()
5958 //
5959 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5960 // in order to report if we are running in a headless jre
5961 //
5962 // Since JDK8 xawt/libmawt.so was moved into the same directory
5963 // as libawt.so, and renamed libawt_xawt.so
5964 //
5965 bool os::is_headless_jre() {
5966 struct stat statbuf;
5967 char buf[MAXPATHLEN];
5968 char libmawtpath[MAXPATHLEN];
5969 const char *xawtstr = "/xawt/libmawt.so";
5970 const char *new_xawtstr = "/libawt_xawt.so";
5971 char *p;
5972
5973 // Get path to libjvm.so
5974 os::jvm_path(buf, sizeof(buf));
5975
5976 // Get rid of libjvm.so
5977 p = strrchr(buf, '/');
5978 if (p == NULL) {
5979 return false;
5980 } else {
5981 *p = '\0';
5982 }
5983
5984 // Get rid of client or server
5985 p = strrchr(buf, '/');
5986 if (p == NULL) {
5987 return false;
5988 } else {
5989 *p = '\0';
5990 }
5991
5992 // check xawt/libmawt.so
5993 strcpy(libmawtpath, buf);
5994 strcat(libmawtpath, xawtstr);
5995 if (::stat(libmawtpath, &statbuf) == 0) return false;
5996
5997 // check libawt_xawt.so
5998 strcpy(libmawtpath, buf);
5999 strcat(libmawtpath, new_xawtstr);
6000 if (::stat(libmawtpath, &statbuf) == 0) return false;
6001
6002 return true;
6003 }
6004
6005 // Get the default path to the core file
6006 // Returns the length of the string
6007 int os::get_core_path(char* buffer, size_t bufferSize) {
6008 /*
6009 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6010 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6011 */
6012 const int core_pattern_len = 129;
6013 char core_pattern[core_pattern_len] = {0};
6014
6015 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6016 if (core_pattern_file != -1) {
6017 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6018 ::close(core_pattern_file);
6019
6020 if (ret > 0) {
6021 char *last_char = core_pattern + strlen(core_pattern) - 1;
6022
6023 if (*last_char == '\n') {
6024 *last_char = '\0';
6025 }
6026 }
6027 }
6028
6029 if (strlen(core_pattern) == 0) {
6030 return 0;
6031 }
6032
6033 char *pid_pos = strstr(core_pattern, "%p");
6034 size_t written;
6035
6036 if (core_pattern[0] == '/') {
6037 written = jio_snprintf(buffer, bufferSize, core_pattern);
6038 } else {
6039 char cwd[PATH_MAX];
6040
6041 const char* p = get_current_directory(cwd, PATH_MAX);
6042 if (p == NULL) {
6043 assert(p != NULL, "failed to get current directory");
6044 return 0;
6045 }
6046
6047 if (core_pattern[0] == '|') {
6048 written = jio_snprintf(buffer, bufferSize,
6049 "\"%s\" (or dumping to %s/core.%d)",
6050 &core_pattern[1], p, current_process_id());
6051 } else {
6052 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6053 }
6054 }
6055
6056 if ((written >= 0) && (written < bufferSize)
6057 && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6058 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6059
6060 if (core_uses_pid_file != -1) {
6061 char core_uses_pid = 0;
6062 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6063 ::close(core_uses_pid_file);
6064
6065 if (core_uses_pid == '1'){
6066 jio_snprintf(buffer + written, bufferSize - written,
6067 ".%d", current_process_id());
6068 }
6069 }
6070 }
6071
6072 return strlen(buffer);
6073 }
6074
6075 /////////////// Unit tests ///////////////
6076
6077 #ifndef PRODUCT
6078
6079 #define test_log(...) \
6080 do { \
6081 if (VerboseInternalVMTests) { \
6082 tty->print_cr(__VA_ARGS__); \
6083 tty->flush(); \
6084 } \
6085 } while (false)
6086
6087 class TestReserveMemorySpecial : AllStatic {
6088 public:
6089 static void small_page_write(void* addr, size_t size) {
6090 size_t page_size = os::vm_page_size();
6091
6092 char* end = (char*)addr + size;
6093 for (char* p = (char*)addr; p < end; p += page_size) {
6094 *p = 1;
6095 }
6096 }
6097
6098 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6099 if (!UseHugeTLBFS) {
6100 return;
6101 }
6102
6103 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6104
6105 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6106
6107 if (addr != NULL) {
6108 small_page_write(addr, size);
6109
6110 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6111 }
6112 }
6113
6114 static void test_reserve_memory_special_huge_tlbfs_only() {
6115 if (!UseHugeTLBFS) {
6116 return;
6117 }
6118
6119 size_t lp = os::large_page_size();
6120
6121 for (size_t size = lp; size <= lp * 10; size += lp) {
6122 test_reserve_memory_special_huge_tlbfs_only(size);
6123 }
6124 }
6125
6126 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6127 if (!UseHugeTLBFS) {
6128 return;
6129 }
6130
6131 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6132 size, alignment);
6133
6134 assert(size >= os::large_page_size(), "Incorrect input to test");
6135
6136 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6137
6138 if (addr != NULL) {
6139 small_page_write(addr, size);
6140
6141 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6142 }
6143 }
6144
6145 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6146 size_t lp = os::large_page_size();
6147 size_t ag = os::vm_allocation_granularity();
6148
6149 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6150 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6151 }
6152 }
6153
6154 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6155 size_t lp = os::large_page_size();
6156 size_t ag = os::vm_allocation_granularity();
6157
6158 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6159 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6160 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6161 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6162 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6163 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6164 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6165 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6166 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6167 }
6168
6169 static void test_reserve_memory_special_huge_tlbfs() {
6170 if (!UseHugeTLBFS) {
6171 return;
6172 }
6173
6174 test_reserve_memory_special_huge_tlbfs_only();
6175 test_reserve_memory_special_huge_tlbfs_mixed();
6176 }
6177
6178 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6179 if (!UseSHM) {
6180 return;
6181 }
6182
6183 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6184
6185 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6186
6187 if (addr != NULL) {
6188 assert(is_ptr_aligned(addr, alignment), "Check");
6189 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6190
6191 small_page_write(addr, size);
6192
6193 os::Linux::release_memory_special_shm(addr, size);
6194 }
6195 }
6196
6197 static void test_reserve_memory_special_shm() {
6198 size_t lp = os::large_page_size();
6199 size_t ag = os::vm_allocation_granularity();
6200
6201 for (size_t size = ag; size < lp * 3; size += ag) {
6202 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6203 test_reserve_memory_special_shm(size, alignment);
6204 }
6205 }
6206 }
6207
6208 static void test() {
6209 test_reserve_memory_special_huge_tlbfs();
6210 test_reserve_memory_special_shm();
6211 }
6212 };
6213
6214 void TestReserveMemorySpecial_test() {
6215 TestReserveMemorySpecial::test();
6216 }
6217
6218 #endif