1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "osContainer_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/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "services/attachListener.hpp"
63 #include "services/memTracker.hpp"
64 #include "services/runtimeService.hpp"
65 #include "utilities/decoder.hpp"
66 #include "utilities/defaultStream.hpp"
67 #include "utilities/events.hpp"
68 #include "utilities/elfFile.hpp"
69 #include "utilities/growableArray.hpp"
70 #include "utilities/vmError.hpp"
71
72 // put OS-includes here
73 # include <sys/types.h>
74 # include <sys/mman.h>
75 # include <sys/stat.h>
76 # include <sys/select.h>
77 # include <pthread.h>
78 # include <signal.h>
79 # include <errno.h>
80 # include <dlfcn.h>
81 # include <stdio.h>
82 # include <unistd.h>
83 # include <sys/resource.h>
84 # include <pthread.h>
85 # include <sys/stat.h>
86 # include <sys/time.h>
87 # include <sys/times.h>
88 # include <sys/utsname.h>
89 # include <sys/socket.h>
90 # include <sys/wait.h>
91 # include <pwd.h>
92 # include <poll.h>
93 # include <semaphore.h>
94 # include <fcntl.h>
95 # include <string.h>
96 # include <syscall.h>
97 # include <sys/sysinfo.h>
98 # include <gnu/libc-version.h>
99 # include <sys/ipc.h>
100 # include <sys/shm.h>
101 # include <link.h>
102 # include <stdint.h>
103 # include <inttypes.h>
104 # include <sys/ioctl.h>
105
106 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
107
108 #ifndef _GNU_SOURCE
109 #define _GNU_SOURCE
110 #include <sched.h>
111 #undef _GNU_SOURCE
112 #else
113 #include <sched.h>
114 #endif
115
116 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
117 // getrusage() is prepared to handle the associated failure.
118 #ifndef RUSAGE_THREAD
119 #define RUSAGE_THREAD (1) /* only the calling thread */
120 #endif
121
122 #define MAX_PATH (2 * K)
123
124 #define MAX_SECS 100000000
125
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
128
129 #define LARGEPAGES_BIT (1 << 6)
130 ////////////////////////////////////////////////////////////////////////////////
131 // global variables
132 julong os::Linux::_physical_memory = 0;
133
134 address os::Linux::_initial_thread_stack_bottom = NULL;
135 uintptr_t os::Linux::_initial_thread_stack_size = 0;
136
137 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
138 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
139 Mutex* os::Linux::_createThread_lock = NULL;
140 pthread_t os::Linux::_main_thread;
141 int os::Linux::_page_size = -1;
142 const int os::Linux::_vm_default_page_size = (8 * K);
143 bool os::Linux::_is_floating_stack = false;
144 bool os::Linux::_is_NPTL = false;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148 pthread_condattr_t os::Linux::_condattr[1];
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 static pid_t _initial_pid = 0;
159
160 /* Signal number used to suspend/resume a thread */
161
162 /* do not use any signal number less than SIGSEGV, see 4355769 */
163 static int SR_signum = SIGUSR2;
164 sigset_t SR_sigset;
165
166 /* Used to protect dlsym() calls */
167 static pthread_mutex_t dl_mutex;
168
169 // Declarations
170 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
171
172 // utility functions
173
174 static int SR_initialize();
175
176 julong os::available_memory() {
177 return Linux::available_memory();
178 }
179
180 julong os::Linux::available_memory() {
181 // values in struct sysinfo are "unsigned long"
182 struct sysinfo si;
183 julong avail_mem;
184
185 if (OSContainer::is_containerized()) {
186 jlong mem_limit, mem_usage;
187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
188 if (PrintContainerInfo) {
189 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
190 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
191 }
192 }
193
194 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
195 if (PrintContainerInfo) {
196 tty->print_cr("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
197 }
198 }
199
200 if (mem_limit > 0 && mem_usage > 0 ) {
201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
202 if (PrintContainerInfo) {
203 tty->print_cr("available container memory: " JULONG_FORMAT, avail_mem);
204 }
205 return avail_mem;
206 }
207 }
208
209 sysinfo(&si);
210 avail_mem = (julong)si.freeram * si.mem_unit;
211 if (Verbose) {
212 tty->print_cr("available memory: " JULONG_FORMAT, avail_mem);
213 }
214 return avail_mem;
215 }
216
217 julong os::physical_memory() {
218 jlong phys_mem = 0;
219 if (OSContainer::is_containerized()) {
220 jlong mem_limit;
221 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
222 if (PrintContainerInfo) {
223 tty->print_cr("total container memory: " JLONG_FORMAT, mem_limit);
224 }
225 return mem_limit;
226 }
227
228 if (PrintContainerInfo) {
229 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
230 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
231 }
232 }
233
234 phys_mem = Linux::physical_memory();
235 if (Verbose) {
236 tty->print_cr("total system memory: " JLONG_FORMAT, phys_mem);
237 }
238 return phys_mem;
239 }
240
241 ////////////////////////////////////////////////////////////////////////////////
242 // environment support
243
244 bool os::getenv(const char* name, char* buf, int len) {
245 const char* val = ::getenv(name);
246 if (val != NULL && strlen(val) < (size_t)len) {
247 strcpy(buf, val);
248 return true;
249 }
250 if (len > 0) buf[0] = 0; // return a null string
251 return false;
252 }
253
254
255 // Return true if user is running as root.
256
257 bool os::have_special_privileges() {
258 static bool init = false;
259 static bool privileges = false;
260 if (!init) {
261 privileges = (getuid() != geteuid()) || (getgid() != getegid());
262 init = true;
263 }
264 return privileges;
265 }
266
267
268 #ifndef SYS_gettid
269 // i386: 224, ia64: 1105, amd64: 186, sparc 143
270 #ifdef __ia64__
271 #define SYS_gettid 1105
272 #else
273 #ifdef __i386__
274 #define SYS_gettid 224
275 #else
276 #ifdef __amd64__
277 #define SYS_gettid 186
278 #else
279 #ifdef __sparc__
280 #define SYS_gettid 143
281 #else
282 #error define gettid for the arch
283 #endif
284 #endif
285 #endif
286 #endif
287 #endif
288
289 // Cpu architecture string
290 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
291
292 // pid_t gettid()
293 //
294 // Returns the kernel thread id of the currently running thread. Kernel
295 // thread id is used to access /proc.
296 //
297 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
298 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
299 //
300 pid_t os::Linux::gettid() {
301 int rslt = syscall(SYS_gettid);
302 if (rslt == -1) {
303 // old kernel, no NPTL support
304 return getpid();
305 } else {
306 return (pid_t)rslt;
307 }
308 }
309
310 // Most versions of linux have a bug where the number of processors are
311 // determined by looking at the /proc file system. In a chroot environment,
312 // the system call returns 1. This causes the VM to act as if it is
313 // a single processor and elide locking (see is_MP() call).
314 static bool unsafe_chroot_detected = false;
315 static const char *unstable_chroot_error = "/proc file system not found.\n"
316 "Java may be unstable running multithreaded in a chroot "
317 "environment on Linux when /proc filesystem is not mounted.";
318
319 void os::Linux::initialize_system_info() {
320 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
321 if (processor_count() == 1) {
322 pid_t pid = os::Linux::gettid();
323 char fname[32];
324 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
325 FILE *fp = fopen(fname, "r");
326 if (fp == NULL) {
327 unsafe_chroot_detected = true;
328 } else {
329 fclose(fp);
330 }
331 }
332 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
333 assert(processor_count() > 0, "linux error");
334 }
335
336 void os::init_system_properties_values() {
337 // The next steps are taken in the product version:
338 //
339 // Obtain the JAVA_HOME value from the location of libjvm.so.
340 // This library should be located at:
341 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
342 //
343 // If "/jre/lib/" appears at the right place in the path, then we
344 // assume libjvm.so is installed in a JDK and we use this path.
345 //
346 // Otherwise exit with message: "Could not create the Java virtual machine."
347 //
348 // The following extra steps are taken in the debugging version:
349 //
350 // If "/jre/lib/" does NOT appear at the right place in the path
351 // instead of exit check for $JAVA_HOME environment variable.
352 //
353 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
354 // then we append a fake suffix "hotspot/libjvm.so" to this path so
355 // it looks like libjvm.so is installed there
356 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
357 //
358 // Otherwise exit.
359 //
360 // Important note: if the location of libjvm.so changes this
361 // code needs to be changed accordingly.
362
363 // See ld(1):
364 // The linker uses the following search paths to locate required
365 // shared libraries:
366 // 1: ...
367 // ...
368 // 7: The default directories, normally /lib and /usr/lib.
369 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
370 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
371 #else
372 #define DEFAULT_LIBPATH "/lib:/usr/lib"
373 #endif
374
375 // Base path of extensions installed on the system.
376 #define SYS_EXT_DIR "/usr/java/packages"
377 #define EXTENSIONS_DIR "/lib/ext"
378 #define ENDORSED_DIR "/lib/endorsed"
379
380 // Buffer that fits several sprintfs.
381 // Note that the space for the colon and the trailing null are provided
382 // by the nulls included by the sizeof operator.
383 const size_t bufsize =
384 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
385 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
386 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
387 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
388
389 // sysclasspath, java_home, dll_dir
390 {
391 char *pslash;
392 os::jvm_path(buf, bufsize);
393
394 // Found the full path to libjvm.so.
395 // Now cut the path to <java_home>/jre if we can.
396 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
397 pslash = strrchr(buf, '/');
398 if (pslash != NULL) {
399 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
400 }
401 Arguments::set_dll_dir(buf);
402
403 if (pslash != NULL) {
404 pslash = strrchr(buf, '/');
405 if (pslash != NULL) {
406 *pslash = '\0'; // Get rid of /<arch>.
407 pslash = strrchr(buf, '/');
408 if (pslash != NULL) {
409 *pslash = '\0'; // Get rid of /lib.
410 }
411 }
412 }
413 Arguments::set_java_home(buf);
414 set_boot_path('/', ':');
415 }
416
417 // Where to look for native libraries.
418 //
419 // Note: Due to a legacy implementation, most of the library path
420 // is set in the launcher. This was to accomodate linking restrictions
421 // on legacy Linux implementations (which are no longer supported).
422 // Eventually, all the library path setting will be done here.
423 //
424 // However, to prevent the proliferation of improperly built native
425 // libraries, the new path component /usr/java/packages is added here.
426 // Eventually, all the library path setting will be done here.
427 {
428 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
429 // should always exist (until the legacy problem cited above is
430 // addressed).
431 const char *v = ::getenv("LD_LIBRARY_PATH");
432 const char *v_colon = ":";
433 if (v == NULL) { v = ""; v_colon = ""; }
434 // That's +1 for the colon and +1 for the trailing '\0'.
435 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
436 strlen(v) + 1 +
437 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
438 mtInternal);
439 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
440 Arguments::set_library_path(ld_library_path);
441 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
442 }
443
444 // Extensions directories.
445 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
446 Arguments::set_ext_dirs(buf);
447
448 // Endorsed standards default directory.
449 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
450 Arguments::set_endorsed_dirs(buf);
451
452 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
453
454 #undef DEFAULT_LIBPATH
455 #undef SYS_EXT_DIR
456 #undef EXTENSIONS_DIR
457 #undef ENDORSED_DIR
458 }
459
460 ////////////////////////////////////////////////////////////////////////////////
461 // breakpoint support
462
463 void os::breakpoint() {
464 BREAKPOINT;
465 }
466
467 extern "C" void breakpoint() {
468 // use debugger to set breakpoint here
469 }
470
471 ////////////////////////////////////////////////////////////////////////////////
472 // signal support
473
474 debug_only(static bool signal_sets_initialized = false);
475 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
476
477 bool os::Linux::is_sig_ignored(int sig) {
478 struct sigaction oact;
479 sigaction(sig, (struct sigaction*)NULL, &oact);
480 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
481 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
482 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
483 return true;
484 else
485 return false;
486 }
487
488 void os::Linux::signal_sets_init() {
489 // Should also have an assertion stating we are still single-threaded.
490 assert(!signal_sets_initialized, "Already initialized");
491 // Fill in signals that are necessarily unblocked for all threads in
492 // the VM. Currently, we unblock the following signals:
493 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
494 // by -Xrs (=ReduceSignalUsage));
495 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
496 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
497 // the dispositions or masks wrt these signals.
498 // Programs embedding the VM that want to use the above signals for their
499 // own purposes must, at this time, use the "-Xrs" option to prevent
500 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
501 // (See bug 4345157, and other related bugs).
502 // In reality, though, unblocking these signals is really a nop, since
503 // these signals are not blocked by default.
504 sigemptyset(&unblocked_sigs);
505 sigemptyset(&allowdebug_blocked_sigs);
506 sigaddset(&unblocked_sigs, SIGILL);
507 sigaddset(&unblocked_sigs, SIGSEGV);
508 sigaddset(&unblocked_sigs, SIGBUS);
509 sigaddset(&unblocked_sigs, SIGFPE);
510 #if defined(PPC64)
511 sigaddset(&unblocked_sigs, SIGTRAP);
512 #endif
513 sigaddset(&unblocked_sigs, SR_signum);
514
515 if (!ReduceSignalUsage) {
516 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
517 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
518 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
519 }
520 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
521 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
522 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
523 }
524 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
525 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
526 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
527 }
528 }
529 // Fill in signals that are blocked by all but the VM thread.
530 sigemptyset(&vm_sigs);
531 if (!ReduceSignalUsage)
532 sigaddset(&vm_sigs, BREAK_SIGNAL);
533 debug_only(signal_sets_initialized = true);
534
535 }
536
537 // These are signals that are unblocked while a thread is running Java.
538 // (For some reason, they get blocked by default.)
539 sigset_t* os::Linux::unblocked_signals() {
540 assert(signal_sets_initialized, "Not initialized");
541 return &unblocked_sigs;
542 }
543
544 // These are the signals that are blocked while a (non-VM) thread is
545 // running Java. Only the VM thread handles these signals.
546 sigset_t* os::Linux::vm_signals() {
547 assert(signal_sets_initialized, "Not initialized");
548 return &vm_sigs;
549 }
550
551 // These are signals that are blocked during cond_wait to allow debugger in
552 sigset_t* os::Linux::allowdebug_blocked_signals() {
553 assert(signal_sets_initialized, "Not initialized");
554 return &allowdebug_blocked_sigs;
555 }
556
557 void os::Linux::hotspot_sigmask(Thread* thread) {
558
559 //Save caller's signal mask before setting VM signal mask
560 sigset_t caller_sigmask;
561 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
562
563 OSThread* osthread = thread->osthread();
564 osthread->set_caller_sigmask(caller_sigmask);
565
566 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
567
568 if (!ReduceSignalUsage) {
569 if (thread->is_VM_thread()) {
570 // Only the VM thread handles BREAK_SIGNAL ...
571 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
572 } else {
573 // ... all other threads block BREAK_SIGNAL
574 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
575 }
576 }
577 }
578
579 //////////////////////////////////////////////////////////////////////////////
580 // detecting pthread library
581
582 void os::Linux::libpthread_init() {
583 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
584 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
585 // generic name for earlier versions.
586 // Define macros here so we can build HotSpot on old systems.
587 # ifndef _CS_GNU_LIBC_VERSION
588 # define _CS_GNU_LIBC_VERSION 2
589 # endif
590 # ifndef _CS_GNU_LIBPTHREAD_VERSION
591 # define _CS_GNU_LIBPTHREAD_VERSION 3
592 # endif
593
594 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
595 if (n > 0) {
596 char *str = (char *)malloc(n, mtInternal);
597 confstr(_CS_GNU_LIBC_VERSION, str, n);
598 os::Linux::set_glibc_version(str);
599 } else {
600 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
601 static char _gnu_libc_version[32];
602 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
603 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
604 os::Linux::set_glibc_version(_gnu_libc_version);
605 }
606
607 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
608 if (n > 0) {
609 char *str = (char *)malloc(n, mtInternal);
610 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
611 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
612 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
613 // is the case. LinuxThreads has a hard limit on max number of threads.
614 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
615 // On the other hand, NPTL does not have such a limit, sysconf()
616 // will return -1 and errno is not changed. Check if it is really NPTL.
617 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
618 strstr(str, "NPTL") &&
619 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
620 free(str);
621 os::Linux::set_libpthread_version("linuxthreads");
622 } else {
623 os::Linux::set_libpthread_version(str);
624 }
625 } else {
626 // glibc before 2.3.2 only has LinuxThreads.
627 os::Linux::set_libpthread_version("linuxthreads");
628 }
629
630 if (strstr(libpthread_version(), "NPTL")) {
631 os::Linux::set_is_NPTL();
632 } else {
633 os::Linux::set_is_LinuxThreads();
634 }
635
636 // LinuxThreads have two flavors: floating-stack mode, which allows variable
637 // stack size; and fixed-stack mode. NPTL is always floating-stack.
638 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
639 os::Linux::set_is_floating_stack();
640 }
641 }
642
643 /////////////////////////////////////////////////////////////////////////////
644 // thread stack
645
646 // Force Linux kernel to expand current thread stack. If "bottom" is close
647 // to the stack guard, caller should block all signals.
648 //
649 // MAP_GROWSDOWN:
650 // A special mmap() flag that is used to implement thread stacks. It tells
651 // kernel that the memory region should extend downwards when needed. This
652 // allows early versions of LinuxThreads to only mmap the first few pages
653 // when creating a new thread. Linux kernel will automatically expand thread
654 // stack as needed (on page faults).
655 //
656 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
657 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
658 // region, it's hard to tell if the fault is due to a legitimate stack
659 // access or because of reading/writing non-exist memory (e.g. buffer
660 // overrun). As a rule, if the fault happens below current stack pointer,
661 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
662 // application (see Linux kernel fault.c).
663 //
664 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
665 // stack overflow detection.
666 //
667 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
668 // not use this flag. However, the stack of initial thread is not created
669 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
670 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
671 // and then attach the thread to JVM.
672 //
673 // To get around the problem and allow stack banging on Linux, we need to
674 // manually expand thread stack after receiving the SIGSEGV.
675 //
676 // There are two ways to expand thread stack to address "bottom", we used
677 // both of them in JVM before 1.5:
678 // 1. adjust stack pointer first so that it is below "bottom", and then
679 // touch "bottom"
680 // 2. mmap() the page in question
681 //
682 // Now alternate signal stack is gone, it's harder to use 2. For instance,
683 // if current sp is already near the lower end of page 101, and we need to
684 // call mmap() to map page 100, it is possible that part of the mmap() frame
685 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
686 // That will destroy the mmap() frame and cause VM to crash.
687 //
688 // The following code works by adjusting sp first, then accessing the "bottom"
689 // page to force a page fault. Linux kernel will then automatically expand the
690 // stack mapping.
691 //
692 // _expand_stack_to() assumes its frame size is less than page size, which
693 // should always be true if the function is not inlined.
694
695 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
696 #define NOINLINE
697 #else
698 #define NOINLINE __attribute__ ((noinline))
699 #endif
700
701 static void _expand_stack_to(address bottom) NOINLINE;
702
703 static void _expand_stack_to(address bottom) {
704 address sp;
705 size_t size;
706 volatile char *p;
707
708 // Adjust bottom to point to the largest address within the same page, it
709 // gives us a one-page buffer if alloca() allocates slightly more memory.
710 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
711 bottom += os::Linux::page_size() - 1;
712
713 // sp might be slightly above current stack pointer; if that's the case, we
714 // will alloca() a little more space than necessary, which is OK. Don't use
715 // os::current_stack_pointer(), as its result can be slightly below current
716 // stack pointer, causing us to not alloca enough to reach "bottom".
717 sp = (address)&sp;
718
719 if (sp > bottom) {
720 size = sp - bottom;
721 p = (volatile char *)alloca(size);
722 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
723 p[0] = '\0';
724 }
725 }
726
727 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
728 assert(t!=NULL, "just checking");
729 assert(t->osthread()->expanding_stack(), "expand should be set");
730 assert(t->stack_base() != NULL, "stack_base was not initialized");
731
732 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
733 sigset_t mask_all, old_sigset;
734 sigfillset(&mask_all);
735 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
736 _expand_stack_to(addr);
737 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
738 return true;
739 }
740 return false;
741 }
742
743 //////////////////////////////////////////////////////////////////////////////
744 // create new thread
745
746 static address highest_vm_reserved_address();
747
748 // check if it's safe to start a new thread
749 static bool _thread_safety_check(Thread* thread) {
750 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
751 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
752 // Heap is mmap'ed at lower end of memory space. Thread stacks are
753 // allocated (MAP_FIXED) from high address space. Every thread stack
754 // occupies a fixed size slot (usually 2Mbytes, but user can change
755 // it to other values if they rebuild LinuxThreads).
756 //
757 // Problem with MAP_FIXED is that mmap() can still succeed even part of
758 // the memory region has already been mmap'ed. That means if we have too
759 // many threads and/or very large heap, eventually thread stack will
760 // collide with heap.
761 //
762 // Here we try to prevent heap/stack collision by comparing current
763 // stack bottom with the highest address that has been mmap'ed by JVM
764 // plus a safety margin for memory maps created by native code.
765 //
766 // This feature can be disabled by setting ThreadSafetyMargin to 0
767 //
768 if (ThreadSafetyMargin > 0) {
769 address stack_bottom = os::current_stack_base() - os::current_stack_size();
770
771 // not safe if our stack extends below the safety margin
772 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
773 } else {
774 return true;
775 }
776 } else {
777 // Floating stack LinuxThreads or NPTL:
778 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
779 // there's not enough space left, pthread_create() will fail. If we come
780 // here, that means enough space has been reserved for stack.
781 return true;
782 }
783 }
784
785 // Thread start routine for all newly created threads
786 static void *java_start(Thread *thread) {
787 // Try to randomize the cache line index of hot stack frames.
788 // This helps when threads of the same stack traces evict each other's
789 // cache lines. The threads can be either from the same JVM instance, or
790 // from different JVM instances. The benefit is especially true for
791 // processors with hyperthreading technology.
792 static int counter = 0;
793 int pid = os::current_process_id();
794 alloca(((pid ^ counter++) & 7) * 128);
795
796 ThreadLocalStorage::set_thread(thread);
797
798 OSThread* osthread = thread->osthread();
799 Monitor* sync = osthread->startThread_lock();
800
801 // non floating stack LinuxThreads needs extra check, see above
802 if (!_thread_safety_check(thread)) {
803 // notify parent thread
804 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
805 osthread->set_state(ZOMBIE);
806 sync->notify_all();
807 return NULL;
808 }
809
810 // thread_id is kernel thread id (similar to Solaris LWP id)
811 osthread->set_thread_id(os::Linux::gettid());
812
813 if (UseNUMA) {
814 int lgrp_id = os::numa_get_group_id();
815 if (lgrp_id != -1) {
816 thread->set_lgrp_id(lgrp_id);
817 }
818 }
819 // initialize signal mask for this thread
820 os::Linux::hotspot_sigmask(thread);
821
822 // initialize floating point control register
823 os::Linux::init_thread_fpu_state();
824
825 // handshaking with parent thread
826 {
827 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
828
829 // notify parent thread
830 osthread->set_state(INITIALIZED);
831 sync->notify_all();
832
833 // wait until os::start_thread()
834 while (osthread->get_state() == INITIALIZED) {
835 sync->wait(Mutex::_no_safepoint_check_flag);
836 }
837 }
838
839 // call one more level start routine
840 thread->run();
841
842 return 0;
843 }
844
845 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
846 assert(thread->osthread() == NULL, "caller responsible");
847
848 // Allocate the OSThread object
849 OSThread* osthread = new OSThread(NULL, NULL);
850 if (osthread == NULL) {
851 return false;
852 }
853
854 // set the correct thread state
855 osthread->set_thread_type(thr_type);
856
857 // Initial state is ALLOCATED but not INITIALIZED
858 osthread->set_state(ALLOCATED);
859
860 thread->set_osthread(osthread);
861
862 // init thread attributes
863 pthread_attr_t attr;
864 pthread_attr_init(&attr);
865 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
866
867 // stack size
868 if (os::Linux::supports_variable_stack_size()) {
869 // calculate stack size if it's not specified by caller
870 if (stack_size == 0) {
871 stack_size = os::Linux::default_stack_size(thr_type);
872
873 switch (thr_type) {
874 case os::java_thread:
875 // Java threads use ThreadStackSize which default value can be
876 // changed with the flag -Xss
877 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
878 stack_size = JavaThread::stack_size_at_create();
879 break;
880 case os::compiler_thread:
881 if (CompilerThreadStackSize > 0) {
882 stack_size = (size_t)(CompilerThreadStackSize * K);
883 break;
884 } // else fall through:
885 // use VMThreadStackSize if CompilerThreadStackSize is not defined
886 case os::vm_thread:
887 case os::pgc_thread:
888 case os::cgc_thread:
889 case os::watcher_thread:
890 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
891 break;
892 }
893 }
894
895 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
896 pthread_attr_setstacksize(&attr, stack_size);
897 } else {
898 // let pthread_create() pick the default value.
899 }
900
901 // glibc guard page
902 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
903
904 ThreadState state;
905
906 {
907 // Serialize thread creation if we are running with fixed stack LinuxThreads
908 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
909 if (lock) {
910 os::Linux::createThread_lock()->lock_without_safepoint_check();
911 }
912
913 pthread_t tid;
914 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
915
916 pthread_attr_destroy(&attr);
917
918 if (ret != 0) {
919 if (PrintMiscellaneous && (Verbose || WizardMode)) {
920 perror("pthread_create()");
921 }
922 // Need to clean up stuff we've allocated so far
923 thread->set_osthread(NULL);
924 delete osthread;
925 if (lock) os::Linux::createThread_lock()->unlock();
926 return false;
927 }
928
929 // Store pthread info into the OSThread
930 osthread->set_pthread_id(tid);
931
932 // Wait until child thread is either initialized or aborted
933 {
934 Monitor* sync_with_child = osthread->startThread_lock();
935 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
936 while ((state = osthread->get_state()) == ALLOCATED) {
937 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
938 }
939 }
940
941 if (lock) {
942 os::Linux::createThread_lock()->unlock();
943 }
944 }
945
946 // Aborted due to thread limit being reached
947 if (state == ZOMBIE) {
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
952
953 // The thread is returned suspended (in state INITIALIZED),
954 // and is started higher up in the call chain
955 assert(state == INITIALIZED, "race condition");
956 return true;
957 }
958
959 /////////////////////////////////////////////////////////////////////////////
960 // attach existing thread
961
962 // bootstrap the main thread
963 bool os::create_main_thread(JavaThread* thread) {
964 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
965 return create_attached_thread(thread);
966 }
967
968 bool os::create_attached_thread(JavaThread* thread) {
969 #ifdef ASSERT
970 thread->verify_not_published();
971 #endif
972
973 // Allocate the OSThread object
974 OSThread* osthread = new OSThread(NULL, NULL);
975
976 if (osthread == NULL) {
977 return false;
978 }
979
980 // Store pthread info into the OSThread
981 osthread->set_thread_id(os::Linux::gettid());
982 osthread->set_pthread_id(::pthread_self());
983
984 // initialize floating point control register
985 os::Linux::init_thread_fpu_state();
986
987 // Initial thread state is RUNNABLE
988 osthread->set_state(RUNNABLE);
989
990 thread->set_osthread(osthread);
991
992 if (UseNUMA) {
993 int lgrp_id = os::numa_get_group_id();
994 if (lgrp_id != -1) {
995 thread->set_lgrp_id(lgrp_id);
996 }
997 }
998
999 if (os::is_primordial_thread()) {
1000 // If current thread is primordial thread, its stack is mapped on demand,
1001 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1002 // the entire stack region to avoid SEGV in stack banging.
1003 // It is also useful to get around the heap-stack-gap problem on SuSE
1004 // kernel (see 4821821 for details). We first expand stack to the top
1005 // of yellow zone, then enable stack yellow zone (order is significant,
1006 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1007 // is no gap between the last two virtual memory regions.
1008
1009 JavaThread *jt = (JavaThread *)thread;
1010 address addr = jt->stack_yellow_zone_base();
1011 assert(addr != NULL, "initialization problem?");
1012 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1013
1014 osthread->set_expanding_stack();
1015 os::Linux::manually_expand_stack(jt, addr);
1016 osthread->clear_expanding_stack();
1017 }
1018
1019 // initialize signal mask for this thread
1020 // and save the caller's signal mask
1021 os::Linux::hotspot_sigmask(thread);
1022
1023 return true;
1024 }
1025
1026 void os::pd_start_thread(Thread* thread) {
1027 OSThread * osthread = thread->osthread();
1028 assert(osthread->get_state() != INITIALIZED, "just checking");
1029 Monitor* sync_with_child = osthread->startThread_lock();
1030 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1031 sync_with_child->notify();
1032 }
1033
1034 // Free Linux resources related to the OSThread
1035 void os::free_thread(OSThread* osthread) {
1036 assert(osthread != NULL, "osthread not set");
1037
1038 if (Thread::current()->osthread() == osthread) {
1039 // Restore caller's signal mask
1040 sigset_t sigmask = osthread->caller_sigmask();
1041 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1042 }
1043
1044 delete osthread;
1045 }
1046
1047 //////////////////////////////////////////////////////////////////////////////
1048 // thread local storage
1049
1050 // Restore the thread pointer if the destructor is called. This is in case
1051 // someone from JNI code sets up a destructor with pthread_key_create to run
1052 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1053 // will hang or crash. When detachCurrentThread is called the key will be set
1054 // to null and we will not be called again. If detachCurrentThread is never
1055 // called we could loop forever depending on the pthread implementation.
1056 static void restore_thread_pointer(void* p) {
1057 Thread* thread = (Thread*) p;
1058 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1059 }
1060
1061 int os::allocate_thread_local_storage() {
1062 pthread_key_t key;
1063 int rslt = pthread_key_create(&key, restore_thread_pointer);
1064 assert(rslt == 0, "cannot allocate thread local storage");
1065 return (int)key;
1066 }
1067
1068 // Note: This is currently not used by VM, as we don't destroy TLS key
1069 // on VM exit.
1070 void os::free_thread_local_storage(int index) {
1071 int rslt = pthread_key_delete((pthread_key_t)index);
1072 assert(rslt == 0, "invalid index");
1073 }
1074
1075 void os::thread_local_storage_at_put(int index, void* value) {
1076 int rslt = pthread_setspecific((pthread_key_t)index, value);
1077 assert(rslt == 0, "pthread_setspecific failed");
1078 }
1079
1080 extern "C" Thread* get_thread() {
1081 return ThreadLocalStorage::thread();
1082 }
1083
1084 //////////////////////////////////////////////////////////////////////////////
1085 // primordial thread
1086
1087 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1088 bool os::is_primordial_thread(void) {
1089 char dummy;
1090 // If called before init complete, thread stack bottom will be null.
1091 // Can be called if fatal error occurs before initialization.
1092 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1093 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1094 os::Linux::initial_thread_stack_size() != 0,
1095 "os::init did not locate primordial thread's stack region");
1096 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1097 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1098 os::Linux::initial_thread_stack_size()) {
1099 return true;
1100 } else {
1101 return false;
1102 }
1103 }
1104
1105 // Find the virtual memory area that contains addr
1106 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1107 FILE *fp = fopen("/proc/self/maps", "r");
1108 if (fp) {
1109 address low, high;
1110 while (!feof(fp)) {
1111 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1112 if (low <= addr && addr < high) {
1113 if (vma_low) *vma_low = low;
1114 if (vma_high) *vma_high = high;
1115 fclose (fp);
1116 return true;
1117 }
1118 }
1119 for (;;) {
1120 int ch = fgetc(fp);
1121 if (ch == EOF || ch == (int)'\n') break;
1122 }
1123 }
1124 fclose(fp);
1125 }
1126 return false;
1127 }
1128
1129 // Locate primordial thread stack. This special handling of primordial thread stack
1130 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1131 // bogus value for the primordial process thread. While the launcher has created
1132 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1133 // JNI invocation API from a primordial thread.
1134 void os::Linux::capture_initial_stack(size_t max_size) {
1135
1136 // max_size is either 0 (which means accept OS default for thread stacks) or
1137 // a user-specified value known to be at least the minimum needed. If we
1138 // are actually on the primordial thread we can make it appear that we have a
1139 // smaller max_size stack by inserting the guard pages at that location. But we
1140 // cannot do anything to emulate a larger stack than what has been provided by
1141 // the OS or threading library. In fact if we try to use a stack greater than
1142 // what is set by rlimit then we will crash the hosting process.
1143
1144 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1145 // If this is "unlimited" then it will be a huge value.
1146 struct rlimit rlim;
1147 getrlimit(RLIMIT_STACK, &rlim);
1148 size_t stack_size = rlim.rlim_cur;
1149
1150 // 6308388: a bug in ld.so will relocate its own .data section to the
1151 // lower end of primordial stack; reduce ulimit -s value a little bit
1152 // so we won't install guard page on ld.so's data section.
1153 // But ensure we don't underflow the stack size - allow 1 page spare
1154 if (stack_size >= (size_t)(3 * page_size())) {
1155 stack_size -= 2 * page_size();
1156 }
1157
1158 // Try to figure out where the stack base (top) is. This is harder.
1159 //
1160 // When an application is started, glibc saves the initial stack pointer in
1161 // a global variable "__libc_stack_end", which is then used by system
1162 // libraries. __libc_stack_end should be pretty close to stack top. The
1163 // variable is available since the very early days. However, because it is
1164 // a private interface, it could disappear in the future.
1165 //
1166 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1167 // to __libc_stack_end, it is very close to stack top, but isn't the real
1168 // stack top. Note that /proc may not exist if VM is running as a chroot
1169 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1170 // /proc/<pid>/stat could change in the future (though unlikely).
1171 //
1172 // We try __libc_stack_end first. If that doesn't work, look for
1173 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1174 // as a hint, which should work well in most cases.
1175
1176 uintptr_t stack_start;
1177
1178 // try __libc_stack_end first
1179 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1180 if (p && *p) {
1181 stack_start = *p;
1182 } else {
1183 // see if we can get the start_stack field from /proc/self/stat
1184 FILE *fp;
1185 int pid;
1186 char state;
1187 int ppid;
1188 int pgrp;
1189 int session;
1190 int nr;
1191 int tpgrp;
1192 unsigned long flags;
1193 unsigned long minflt;
1194 unsigned long cminflt;
1195 unsigned long majflt;
1196 unsigned long cmajflt;
1197 unsigned long utime;
1198 unsigned long stime;
1199 long cutime;
1200 long cstime;
1201 long prio;
1202 long nice;
1203 long junk;
1204 long it_real;
1205 uintptr_t start;
1206 uintptr_t vsize;
1207 intptr_t rss;
1208 uintptr_t rsslim;
1209 uintptr_t scodes;
1210 uintptr_t ecode;
1211 int i;
1212
1213 // Figure what the primordial thread stack base is. Code is inspired
1214 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1215 // followed by command name surrounded by parentheses, state, etc.
1216 char stat[2048];
1217 int statlen;
1218
1219 fp = fopen("/proc/self/stat", "r");
1220 if (fp) {
1221 statlen = fread(stat, 1, 2047, fp);
1222 stat[statlen] = '\0';
1223 fclose(fp);
1224
1225 // Skip pid and the command string. Note that we could be dealing with
1226 // weird command names, e.g. user could decide to rename java launcher
1227 // to "java 1.4.2 :)", then the stat file would look like
1228 // 1234 (java 1.4.2 :)) R ... ...
1229 // We don't really need to know the command string, just find the last
1230 // occurrence of ")" and then start parsing from there. See bug 4726580.
1231 char * s = strrchr(stat, ')');
1232
1233 i = 0;
1234 if (s) {
1235 // Skip blank chars
1236 do s++; while (isspace(*s));
1237
1238 #define _UFM UINTX_FORMAT
1239 #define _DFM INTX_FORMAT
1240
1241 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1242 /* 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 */
1243 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,
1244 &state, /* 3 %c */
1245 &ppid, /* 4 %d */
1246 &pgrp, /* 5 %d */
1247 &session, /* 6 %d */
1248 &nr, /* 7 %d */
1249 &tpgrp, /* 8 %d */
1250 &flags, /* 9 %lu */
1251 &minflt, /* 10 %lu */
1252 &cminflt, /* 11 %lu */
1253 &majflt, /* 12 %lu */
1254 &cmajflt, /* 13 %lu */
1255 &utime, /* 14 %lu */
1256 &stime, /* 15 %lu */
1257 &cutime, /* 16 %ld */
1258 &cstime, /* 17 %ld */
1259 &prio, /* 18 %ld */
1260 &nice, /* 19 %ld */
1261 &junk, /* 20 %ld */
1262 &it_real, /* 21 %ld */
1263 &start, /* 22 UINTX_FORMAT */
1264 &vsize, /* 23 UINTX_FORMAT */
1265 &rss, /* 24 INTX_FORMAT */
1266 &rsslim, /* 25 UINTX_FORMAT */
1267 &scodes, /* 26 UINTX_FORMAT */
1268 &ecode, /* 27 UINTX_FORMAT */
1269 &stack_start); /* 28 UINTX_FORMAT */
1270 }
1271
1272 #undef _UFM
1273 #undef _DFM
1274
1275 if (i != 28 - 2) {
1276 assert(false, "Bad conversion from /proc/self/stat");
1277 // product mode - assume we are the primordial thread, good luck in the
1278 // embedded case.
1279 warning("Can't detect primordial thread stack location - bad conversion");
1280 stack_start = (uintptr_t) &rlim;
1281 }
1282 } else {
1283 // For some reason we can't open /proc/self/stat (for example, running on
1284 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1285 // most cases, so don't abort:
1286 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 }
1290
1291 // Now we have a pointer (stack_start) very close to the stack top, the
1292 // next thing to do is to figure out the exact location of stack top. We
1293 // can find out the virtual memory area that contains stack_start by
1294 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1295 // and its upper limit is the real stack top. (again, this would fail if
1296 // running inside chroot, because /proc may not exist.)
1297
1298 uintptr_t stack_top;
1299 address low, high;
1300 if (find_vma((address)stack_start, &low, &high)) {
1301 // success, "high" is the true stack top. (ignore "low", because initial
1302 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1303 stack_top = (uintptr_t)high;
1304 } else {
1305 // failed, likely because /proc/self/maps does not exist
1306 warning("Can't detect primordial thread stack location - find_vma failed");
1307 // best effort: stack_start is normally within a few pages below the real
1308 // stack top, use it as stack top, and reduce stack size so we won't put
1309 // guard page outside stack.
1310 stack_top = stack_start;
1311 stack_size -= 16 * page_size();
1312 }
1313
1314 // stack_top could be partially down the page so align it
1315 stack_top = align_size_up(stack_top, page_size());
1316
1317 // Allowed stack value is minimum of max_size and what we derived from rlimit
1318 if (max_size > 0) {
1319 _initial_thread_stack_size = MIN2(max_size, stack_size);
1320 } else {
1321 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1322 // clamp it at 8MB as we do on Solaris
1323 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1324 }
1325
1326 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1327 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1328 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1329 }
1330
1331 ////////////////////////////////////////////////////////////////////////////////
1332 // time support
1333
1334 // Time since start-up in seconds to a fine granularity.
1335 // Used by VMSelfDestructTimer and the MemProfiler.
1336 double os::elapsedTime() {
1337
1338 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1339 }
1340
1341 jlong os::elapsed_counter() {
1342 return javaTimeNanos() - initial_time_count;
1343 }
1344
1345 jlong os::elapsed_frequency() {
1346 return NANOSECS_PER_SEC; // nanosecond resolution
1347 }
1348
1349 bool os::supports_vtime() { return true; }
1350 bool os::enable_vtime() { return false; }
1351 bool os::vtime_enabled() { return false; }
1352
1353 double os::elapsedVTime() {
1354 struct rusage usage;
1355 int retval = getrusage(RUSAGE_THREAD, &usage);
1356 if (retval == 0) {
1357 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);
1358 } else {
1359 // better than nothing, but not much
1360 return elapsedTime();
1361 }
1362 }
1363
1364 jlong os::javaTimeMillis() {
1365 timeval time;
1366 int status = gettimeofday(&time, NULL);
1367 assert(status != -1, "linux error");
1368 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1369 }
1370
1371 #ifndef CLOCK_MONOTONIC
1372 #define CLOCK_MONOTONIC (1)
1373 #endif
1374
1375 void os::Linux::clock_init() {
1376 // we do dlopen's in this particular order due to bug in linux
1377 // dynamical loader (see 6348968) leading to crash on exit
1378 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1379 if (handle == NULL) {
1380 handle = dlopen("librt.so", RTLD_LAZY);
1381 }
1382
1383 if (handle) {
1384 int (*clock_getres_func)(clockid_t, struct timespec*) =
1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1386 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1387 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1388 if (clock_getres_func && clock_gettime_func) {
1389 // See if monotonic clock is supported by the kernel. Note that some
1390 // early implementations simply return kernel jiffies (updated every
1391 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1392 // for nano time (though the monotonic property is still nice to have).
1393 // It's fixed in newer kernels, however clock_getres() still returns
1394 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1395 // resolution for now. Hopefully as people move to new kernels, this
1396 // won't be a problem.
1397 struct timespec res;
1398 struct timespec tp;
1399 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1400 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1401 // yes, monotonic clock is supported
1402 _clock_gettime = clock_gettime_func;
1403 return;
1404 } else {
1405 // close librt if there is no monotonic clock
1406 dlclose(handle);
1407 }
1408 }
1409 }
1410 warning("No monotonic clock was available - timed services may " \
1411 "be adversely affected if the time-of-day clock changes");
1412 }
1413
1414 #ifndef SYS_clock_getres
1415
1416 #if defined(IA32) || defined(AMD64)
1417 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1418 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1419 #else
1420 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1421 #define sys_clock_getres(x,y) -1
1422 #endif
1423
1424 #else
1425 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1426 #endif
1427
1428 void os::Linux::fast_thread_clock_init() {
1429 if (!UseLinuxPosixThreadCPUClocks) {
1430 return;
1431 }
1432 clockid_t clockid;
1433 struct timespec tp;
1434 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1435 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1436
1437 // Switch to using fast clocks for thread cpu time if
1438 // the sys_clock_getres() returns 0 error code.
1439 // Note, that some kernels may support the current thread
1440 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1441 // returned by the pthread_getcpuclockid().
1442 // If the fast Posix clocks are supported then the sys_clock_getres()
1443 // must return at least tp.tv_sec == 0 which means a resolution
1444 // better than 1 sec. This is extra check for reliability.
1445
1446 if(pthread_getcpuclockid_func &&
1447 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1448 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1449
1450 _supports_fast_thread_cpu_time = true;
1451 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1452 }
1453 }
1454
1455 jlong os::javaTimeNanos() {
1456 if (Linux::supports_monotonic_clock()) {
1457 struct timespec tp;
1458 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1459 assert(status == 0, "gettime error");
1460 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1461 return result;
1462 } else {
1463 timeval time;
1464 int status = gettimeofday(&time, NULL);
1465 assert(status != -1, "linux error");
1466 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1467 return 1000 * usecs;
1468 }
1469 }
1470
1471 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1472 if (Linux::supports_monotonic_clock()) {
1473 info_ptr->max_value = ALL_64_BITS;
1474
1475 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1476 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1477 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1478 } else {
1479 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1480 info_ptr->max_value = ALL_64_BITS;
1481
1482 // gettimeofday is a real time clock so it skips
1483 info_ptr->may_skip_backward = true;
1484 info_ptr->may_skip_forward = true;
1485 }
1486
1487 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1488 }
1489
1490 // Return the real, user, and system times in seconds from an
1491 // arbitrary fixed point in the past.
1492 bool os::getTimesSecs(double* process_real_time,
1493 double* process_user_time,
1494 double* process_system_time) {
1495 struct tms ticks;
1496 clock_t real_ticks = times(&ticks);
1497
1498 if (real_ticks == (clock_t) (-1)) {
1499 return false;
1500 } else {
1501 double ticks_per_second = (double) clock_tics_per_sec;
1502 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1503 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1504 *process_real_time = ((double) real_ticks) / ticks_per_second;
1505
1506 return true;
1507 }
1508 }
1509
1510
1511 char * os::local_time_string(char *buf, size_t buflen) {
1512 struct tm t;
1513 time_t long_time;
1514 time(&long_time);
1515 localtime_r(&long_time, &t);
1516 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1517 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1518 t.tm_hour, t.tm_min, t.tm_sec);
1519 return buf;
1520 }
1521
1522 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1523 return localtime_r(clock, res);
1524 }
1525
1526 ////////////////////////////////////////////////////////////////////////////////
1527 // runtime exit support
1528
1529 // Note: os::shutdown() might be called very early during initialization, or
1530 // called from signal handler. Before adding something to os::shutdown(), make
1531 // sure it is async-safe and can handle partially initialized VM.
1532 void os::shutdown() {
1533
1534 // allow PerfMemory to attempt cleanup of any persistent resources
1535 perfMemory_exit();
1536
1537 // needs to remove object in file system
1538 AttachListener::abort();
1539
1540 // flush buffered output, finish log files
1541 ostream_abort();
1542
1543 // Check for abort hook
1544 abort_hook_t abort_hook = Arguments::abort_hook();
1545 if (abort_hook != NULL) {
1546 abort_hook();
1547 }
1548
1549 }
1550
1551 // Note: os::abort() might be called very early during initialization, or
1552 // called from signal handler. Before adding something to os::abort(), make
1553 // sure it is async-safe and can handle partially initialized VM.
1554 void os::abort(bool dump_core) {
1555 os::shutdown();
1556 if (dump_core) {
1557 #ifndef PRODUCT
1558 fdStream out(defaultStream::output_fd());
1559 out.print_raw("Current thread is ");
1560 char buf[16];
1561 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1562 out.print_raw_cr(buf);
1563 out.print_raw_cr("Dumping core ...");
1564 #endif
1565 ::abort(); // dump core
1566 }
1567
1568 ::exit(1);
1569 }
1570
1571 // Die immediately, no exit hook, no abort hook, no cleanup.
1572 void os::die() {
1573 // _exit() on LinuxThreads only kills current thread
1574 ::abort();
1575 }
1576
1577
1578 // This method is a copy of JDK's sysGetLastErrorString
1579 // from src/solaris/hpi/src/system_md.c
1580
1581 size_t os::lasterror(char *buf, size_t len) {
1582
1583 if (errno == 0) return 0;
1584
1585 const char *s = ::strerror(errno);
1586 size_t n = ::strlen(s);
1587 if (n >= len) {
1588 n = len - 1;
1589 }
1590 ::strncpy(buf, s, n);
1591 buf[n] = '\0';
1592 return n;
1593 }
1594
1595 intx os::current_thread_id() { return (intx)pthread_self(); }
1596 int os::current_process_id() {
1597
1598 // Under the old linux thread library, linux gives each thread
1599 // its own process id. Because of this each thread will return
1600 // a different pid if this method were to return the result
1601 // of getpid(2). Linux provides no api that returns the pid
1602 // of the launcher thread for the vm. This implementation
1603 // returns a unique pid, the pid of the launcher thread
1604 // that starts the vm 'process'.
1605
1606 // Under the NPTL, getpid() returns the same pid as the
1607 // launcher thread rather than a unique pid per thread.
1608 // Use gettid() if you want the old pre NPTL behaviour.
1609
1610 // if you are looking for the result of a call to getpid() that
1611 // returns a unique pid for the calling thread, then look at the
1612 // OSThread::thread_id() method in osThread_linux.hpp file
1613
1614 return (int)(_initial_pid ? _initial_pid : getpid());
1615 }
1616
1617 // DLL functions
1618
1619 const char* os::dll_file_extension() { return ".so"; }
1620
1621 // This must be hard coded because it's the system's temporary
1622 // directory not the java application's temp directory, ala java.io.tmpdir.
1623 const char* os::get_temp_directory() { return "/tmp"; }
1624
1625 static bool file_exists(const char* filename) {
1626 struct stat statbuf;
1627 if (filename == NULL || strlen(filename) == 0) {
1628 return false;
1629 }
1630 return os::stat(filename, &statbuf) == 0;
1631 }
1632
1633 bool os::dll_build_name(char* buffer, size_t buflen,
1634 const char* pname, const char* fname) {
1635 bool retval = false;
1636 // Copied from libhpi
1637 const size_t pnamelen = pname ? strlen(pname) : 0;
1638
1639 // Return error on buffer overflow.
1640 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1641 return retval;
1642 }
1643
1644 if (pnamelen == 0) {
1645 snprintf(buffer, buflen, "lib%s.so", fname);
1646 retval = true;
1647 } else if (strchr(pname, *os::path_separator()) != NULL) {
1648 int n;
1649 char** pelements = split_path(pname, &n);
1650 if (pelements == NULL) {
1651 return false;
1652 }
1653 for (int i = 0 ; i < n ; i++) {
1654 // Really shouldn't be NULL, but check can't hurt
1655 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1656 continue; // skip the empty path values
1657 }
1658 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1659 if (file_exists(buffer)) {
1660 retval = true;
1661 break;
1662 }
1663 }
1664 // release the storage
1665 for (int i = 0 ; i < n ; i++) {
1666 if (pelements[i] != NULL) {
1667 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1668 }
1669 }
1670 if (pelements != NULL) {
1671 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1672 }
1673 } else {
1674 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1675 retval = true;
1676 }
1677 return retval;
1678 }
1679
1680 // check if addr is inside libjvm.so
1681 bool os::address_is_in_vm(address addr) {
1682 static address libjvm_base_addr;
1683 Dl_info dlinfo;
1684
1685 if (libjvm_base_addr == NULL) {
1686 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1687 libjvm_base_addr = (address)dlinfo.dli_fbase;
1688 }
1689 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1690 }
1691
1692 if (dladdr((void *)addr, &dlinfo) != 0) {
1693 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1694 }
1695
1696 return false;
1697 }
1698
1699 bool os::dll_address_to_function_name(address addr, char *buf,
1700 int buflen, int *offset) {
1701 // buf is not optional, but offset is optional
1702 assert(buf != NULL, "sanity check");
1703
1704 Dl_info dlinfo;
1705
1706 if (dladdr((void*)addr, &dlinfo) != 0) {
1707 // see if we have a matching symbol
1708 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1709 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1710 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1711 }
1712 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1713 return true;
1714 }
1715 // no matching symbol so try for just file info
1716 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1717 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1718 buf, buflen, offset, dlinfo.dli_fname)) {
1719 return true;
1720 }
1721 }
1722 }
1723
1724 buf[0] = '\0';
1725 if (offset != NULL) *offset = -1;
1726 return false;
1727 }
1728
1729 struct _address_to_library_name {
1730 address addr; // input : memory address
1731 size_t buflen; // size of fname
1732 char* fname; // output: library name
1733 address base; // library base addr
1734 };
1735
1736 static int address_to_library_name_callback(struct dl_phdr_info *info,
1737 size_t size, void *data) {
1738 int i;
1739 bool found = false;
1740 address libbase = NULL;
1741 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1742
1743 // iterate through all loadable segments
1744 for (i = 0; i < info->dlpi_phnum; i++) {
1745 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1746 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1747 // base address of a library is the lowest address of its loaded
1748 // segments.
1749 if (libbase == NULL || libbase > segbase) {
1750 libbase = segbase;
1751 }
1752 // see if 'addr' is within current segment
1753 if (segbase <= d->addr &&
1754 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1755 found = true;
1756 }
1757 }
1758 }
1759
1760 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1761 // so dll_address_to_library_name() can fall through to use dladdr() which
1762 // can figure out executable name from argv[0].
1763 if (found && info->dlpi_name && info->dlpi_name[0]) {
1764 d->base = libbase;
1765 if (d->fname) {
1766 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1767 }
1768 return 1;
1769 }
1770 return 0;
1771 }
1772
1773 bool os::dll_address_to_library_name(address addr, char* buf,
1774 int buflen, int* offset) {
1775 // buf is not optional, but offset is optional
1776 assert(buf != NULL, "sanity check");
1777
1778 Dl_info dlinfo;
1779 struct _address_to_library_name data;
1780
1781 // There is a bug in old glibc dladdr() implementation that it could resolve
1782 // to wrong library name if the .so file has a base address != NULL. Here
1783 // we iterate through the program headers of all loaded libraries to find
1784 // out which library 'addr' really belongs to. This workaround can be
1785 // removed once the minimum requirement for glibc is moved to 2.3.x.
1786 data.addr = addr;
1787 data.fname = buf;
1788 data.buflen = buflen;
1789 data.base = NULL;
1790 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1791
1792 if (rslt) {
1793 // buf already contains library name
1794 if (offset) *offset = addr - data.base;
1795 return true;
1796 }
1797 if (dladdr((void*)addr, &dlinfo) != 0) {
1798 if (dlinfo.dli_fname != NULL) {
1799 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1800 }
1801 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1802 *offset = addr - (address)dlinfo.dli_fbase;
1803 }
1804 return true;
1805 }
1806
1807 buf[0] = '\0';
1808 if (offset) *offset = -1;
1809 return false;
1810 }
1811
1812 // Loads .dll/.so and
1813 // in case of error it checks if .dll/.so was built for the
1814 // same architecture as Hotspot is running on
1815
1816
1817 // Remember the stack's state. The Linux dynamic linker will change
1818 // the stack to 'executable' at most once, so we must safepoint only once.
1819 bool os::Linux::_stack_is_executable = false;
1820
1821 // VM operation that loads a library. This is necessary if stack protection
1822 // of the Java stacks can be lost during loading the library. If we
1823 // do not stop the Java threads, they can stack overflow before the stacks
1824 // are protected again.
1825 class VM_LinuxDllLoad: public VM_Operation {
1826 private:
1827 const char *_filename;
1828 char *_ebuf;
1829 int _ebuflen;
1830 void *_lib;
1831 public:
1832 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1833 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1834 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1835 void doit() {
1836 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1837 os::Linux::_stack_is_executable = true;
1838 }
1839 void* loaded_library() { return _lib; }
1840 };
1841
1842 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1843 {
1844 void * result = NULL;
1845 bool load_attempted = false;
1846
1847 // Check whether the library to load might change execution rights
1848 // of the stack. If they are changed, the protection of the stack
1849 // guard pages will be lost. We need a safepoint to fix this.
1850 //
1851 // See Linux man page execstack(8) for more info.
1852 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1853 ElfFile ef(filename);
1854 if (!ef.specifies_noexecstack()) {
1855 if (!is_init_completed()) {
1856 os::Linux::_stack_is_executable = true;
1857 // This is OK - No Java threads have been created yet, and hence no
1858 // stack guard pages to fix.
1859 //
1860 // This should happen only when you are building JDK7 using a very
1861 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1862 //
1863 // Dynamic loader will make all stacks executable after
1864 // this function returns, and will not do that again.
1865 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1866 } else {
1867 warning("You have loaded library %s which might have disabled stack guard. "
1868 "The VM will try to fix the stack guard now.\n"
1869 "It's highly recommended that you fix the library with "
1870 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1871 filename);
1872
1873 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1874 JavaThread *jt = JavaThread::current();
1875 if (jt->thread_state() != _thread_in_native) {
1876 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1877 // that requires ExecStack. Cannot enter safe point. Let's give up.
1878 warning("Unable to fix stack guard. Giving up.");
1879 } else {
1880 if (!LoadExecStackDllInVMThread) {
1881 // This is for the case where the DLL has an static
1882 // constructor function that executes JNI code. We cannot
1883 // load such DLLs in the VMThread.
1884 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1885 }
1886
1887 ThreadInVMfromNative tiv(jt);
1888 debug_only(VMNativeEntryWrapper vew;)
1889
1890 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1891 VMThread::execute(&op);
1892 if (LoadExecStackDllInVMThread) {
1893 result = op.loaded_library();
1894 }
1895 load_attempted = true;
1896 }
1897 }
1898 }
1899 }
1900
1901 if (!load_attempted) {
1902 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1903 }
1904
1905 if (result != NULL) {
1906 // Successful loading
1907 return result;
1908 }
1909
1910 Elf32_Ehdr elf_head;
1911 int diag_msg_max_length=ebuflen-strlen(ebuf);
1912 char* diag_msg_buf=ebuf+strlen(ebuf);
1913
1914 if (diag_msg_max_length==0) {
1915 // No more space in ebuf for additional diagnostics message
1916 return NULL;
1917 }
1918
1919
1920 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1921
1922 if (file_descriptor < 0) {
1923 // Can't open library, report dlerror() message
1924 return NULL;
1925 }
1926
1927 bool failed_to_read_elf_head=
1928 (sizeof(elf_head)!=
1929 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1930
1931 ::close(file_descriptor);
1932 if (failed_to_read_elf_head) {
1933 // file i/o error - report dlerror() msg
1934 return NULL;
1935 }
1936
1937 typedef struct {
1938 Elf32_Half code; // Actual value as defined in elf.h
1939 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1940 char elf_class; // 32 or 64 bit
1941 char endianess; // MSB or LSB
1942 char* name; // String representation
1943 } arch_t;
1944
1945 #ifndef EM_486
1946 #define EM_486 6 /* Intel 80486 */
1947 #endif
1948
1949 static const arch_t arch_array[]={
1950 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1951 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1953 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1954 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1955 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1957 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1958 #if defined(VM_LITTLE_ENDIAN)
1959 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1960 #else
1961 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1962 #endif
1963 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1964 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1965 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1966 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1967 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1968 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1969 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1970 };
1971
1972 #if (defined IA32)
1973 static Elf32_Half running_arch_code=EM_386;
1974 #elif (defined AMD64)
1975 static Elf32_Half running_arch_code=EM_X86_64;
1976 #elif (defined IA64)
1977 static Elf32_Half running_arch_code=EM_IA_64;
1978 #elif (defined __sparc) && (defined _LP64)
1979 static Elf32_Half running_arch_code=EM_SPARCV9;
1980 #elif (defined __sparc) && (!defined _LP64)
1981 static Elf32_Half running_arch_code=EM_SPARC;
1982 #elif (defined __powerpc64__)
1983 static Elf32_Half running_arch_code=EM_PPC64;
1984 #elif (defined __powerpc__)
1985 static Elf32_Half running_arch_code=EM_PPC;
1986 #elif (defined ARM)
1987 static Elf32_Half running_arch_code=EM_ARM;
1988 #elif (defined S390)
1989 static Elf32_Half running_arch_code=EM_S390;
1990 #elif (defined ALPHA)
1991 static Elf32_Half running_arch_code=EM_ALPHA;
1992 #elif (defined MIPSEL)
1993 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1994 #elif (defined PARISC)
1995 static Elf32_Half running_arch_code=EM_PARISC;
1996 #elif (defined MIPS)
1997 static Elf32_Half running_arch_code=EM_MIPS;
1998 #elif (defined M68K)
1999 static Elf32_Half running_arch_code=EM_68K;
2000 #else
2001 #error Method os::dll_load requires that one of following is defined:\
2002 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2003 #endif
2004
2005 // Identify compatability class for VM's architecture and library's architecture
2006 // Obtain string descriptions for architectures
2007
2008 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2009 int running_arch_index=-1;
2010
2011 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2012 if (running_arch_code == arch_array[i].code) {
2013 running_arch_index = i;
2014 }
2015 if (lib_arch.code == arch_array[i].code) {
2016 lib_arch.compat_class = arch_array[i].compat_class;
2017 lib_arch.name = arch_array[i].name;
2018 }
2019 }
2020
2021 assert(running_arch_index != -1,
2022 "Didn't find running architecture code (running_arch_code) in arch_array");
2023 if (running_arch_index == -1) {
2024 // Even though running architecture detection failed
2025 // we may still continue with reporting dlerror() message
2026 return NULL;
2027 }
2028
2029 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2030 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2031 return NULL;
2032 }
2033
2034 #ifndef S390
2035 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2036 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2037 return NULL;
2038 }
2039 #endif // !S390
2040
2041 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2042 if ( lib_arch.name!=NULL ) {
2043 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2044 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2045 lib_arch.name, arch_array[running_arch_index].name);
2046 } else {
2047 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2048 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2049 lib_arch.code,
2050 arch_array[running_arch_index].name);
2051 }
2052 }
2053
2054 return NULL;
2055 }
2056
2057 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2058 void * result = ::dlopen(filename, RTLD_LAZY);
2059 if (result == NULL) {
2060 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2061 ebuf[ebuflen-1] = '\0';
2062 }
2063 return result;
2064 }
2065
2066 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2067 void * result = NULL;
2068 if (LoadExecStackDllInVMThread) {
2069 result = dlopen_helper(filename, ebuf, ebuflen);
2070 }
2071
2072 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2073 // library that requires an executable stack, or which does not have this
2074 // stack attribute set, dlopen changes the stack attribute to executable. The
2075 // read protection of the guard pages gets lost.
2076 //
2077 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2078 // may have been queued at the same time.
2079
2080 if (!_stack_is_executable) {
2081 JavaThread *jt = Threads::first();
2082
2083 while (jt) {
2084 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2085 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2086 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2087 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2088 warning("Attempt to reguard stack yellow zone failed.");
2089 }
2090 }
2091 jt = jt->next();
2092 }
2093 }
2094
2095 return result;
2096 }
2097
2098 /*
2099 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2100 * chances are you might want to run the generated bits against glibc-2.0
2101 * libdl.so, so always use locking for any version of glibc.
2102 */
2103 void* os::dll_lookup(void* handle, const char* name) {
2104 pthread_mutex_lock(&dl_mutex);
2105 void* res = dlsym(handle, name);
2106 pthread_mutex_unlock(&dl_mutex);
2107 return res;
2108 }
2109
2110 void* os::get_default_process_handle() {
2111 return (void*)::dlopen(NULL, RTLD_LAZY);
2112 }
2113
2114 static bool _print_ascii_file(const char* filename, outputStream* st) {
2115 int fd = ::open(filename, O_RDONLY);
2116 if (fd == -1) {
2117 return false;
2118 }
2119
2120 char buf[32];
2121 int bytes;
2122 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2123 st->print_raw(buf, bytes);
2124 }
2125
2126 ::close(fd);
2127
2128 return true;
2129 }
2130
2131 void os::print_dll_info(outputStream *st) {
2132 st->print_cr("Dynamic libraries:");
2133
2134 char fname[32];
2135 pid_t pid = os::Linux::gettid();
2136
2137 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2138
2139 if (!_print_ascii_file(fname, st)) {
2140 st->print("Can not get library information for pid = %d\n", pid);
2141 }
2142 }
2143
2144 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2145 FILE *procmapsFile = NULL;
2146
2147 // Open the procfs maps file for the current process
2148 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2149 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2150 char line[PATH_MAX + 100];
2151
2152 // Read line by line from 'file'
2153 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2154 u8 base, top, offset, inode;
2155 char permissions[5];
2156 char device[6];
2157 char name[PATH_MAX + 1];
2158
2159 // Parse fields from line
2160 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
2161 &base, &top, permissions, &offset, device, &inode, name);
2162
2163 // Filter by device id '00:00' so that we only get file system mapped files.
2164 if (strcmp(device, "00:00") != 0) {
2165
2166 // Call callback with the fields of interest
2167 if(callback(name, (address)base, (address)top, param)) {
2168 // Oops abort, callback aborted
2169 fclose(procmapsFile);
2170 return 1;
2171 }
2172 }
2173 }
2174 fclose(procmapsFile);
2175 }
2176 return 0;
2177 }
2178
2179 void os::print_os_info_brief(outputStream* st) {
2180 os::Linux::print_distro_info(st);
2181
2182 os::Posix::print_uname_info(st);
2183
2184 os::Linux::print_libversion_info(st);
2185
2186 }
2187
2188 void os::print_os_info(outputStream* st) {
2189 st->print("OS:");
2190
2191 os::Linux::print_distro_info(st);
2192
2193 os::Posix::print_uname_info(st);
2194
2195 // Print warning if unsafe chroot environment detected
2196 if (unsafe_chroot_detected) {
2197 st->print("WARNING!! ");
2198 st->print_cr("%s", unstable_chroot_error);
2199 }
2200
2201 os::Linux::print_libversion_info(st);
2202
2203 os::Posix::print_rlimit_info(st);
2204
2205 os::Posix::print_load_average(st);
2206
2207 os::Linux::print_full_memory_info(st);
2208
2209 os::Linux::print_container_info(st);
2210 }
2211
2212 // Try to identify popular distros.
2213 // Most Linux distributions have a /etc/XXX-release file, which contains
2214 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2215 // file that also contains the OS version string. Some have more than one
2216 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2217 // /etc/redhat-release.), so the order is important.
2218 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2219 // their own specific XXX-release file as well as a redhat-release file.
2220 // Because of this the XXX-release file needs to be searched for before the
2221 // redhat-release file.
2222 // Since Red Hat has a lsb-release file that is not very descriptive the
2223 // search for redhat-release needs to be before lsb-release.
2224 // Since the lsb-release file is the new standard it needs to be searched
2225 // before the older style release files.
2226 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2227 // next to last resort. The os-release file is a new standard that contains
2228 // distribution information and the system-release file seems to be an old
2229 // standard that has been replaced by the lsb-release and os-release files.
2230 // Searching for the debian_version file is the last resort. It contains
2231 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2232 // "Debian " is printed before the contents of the debian_version file.
2233 void os::Linux::print_distro_info(outputStream* st) {
2234 if (!_print_ascii_file("/etc/oracle-release", st) &&
2235 !_print_ascii_file("/etc/mandriva-release", st) &&
2236 !_print_ascii_file("/etc/mandrake-release", st) &&
2237 !_print_ascii_file("/etc/sun-release", st) &&
2238 !_print_ascii_file("/etc/redhat-release", st) &&
2239 !_print_ascii_file("/etc/lsb-release", st) &&
2240 !_print_ascii_file("/etc/SuSE-release", st) &&
2241 !_print_ascii_file("/etc/turbolinux-release", st) &&
2242 !_print_ascii_file("/etc/gentoo-release", st) &&
2243 !_print_ascii_file("/etc/ltib-release", st) &&
2244 !_print_ascii_file("/etc/angstrom-version", st) &&
2245 !_print_ascii_file("/etc/system-release", st) &&
2246 !_print_ascii_file("/etc/os-release", st)) {
2247
2248 if (file_exists("/etc/debian_version")) {
2249 st->print("Debian ");
2250 _print_ascii_file("/etc/debian_version", st);
2251 } else {
2252 st->print("Linux");
2253 }
2254 }
2255 st->cr();
2256 }
2257
2258 void os::Linux::print_libversion_info(outputStream* st) {
2259 // libc, pthread
2260 st->print("libc:");
2261 st->print("%s ", os::Linux::glibc_version());
2262 st->print("%s ", os::Linux::libpthread_version());
2263 if (os::Linux::is_LinuxThreads()) {
2264 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2265 }
2266 st->cr();
2267 }
2268
2269 void os::Linux::print_full_memory_info(outputStream* st) {
2270 st->print("\n/proc/meminfo:\n");
2271 _print_ascii_file("/proc/meminfo", st);
2272 st->cr();
2273 }
2274
2275 void os::Linux::print_container_info(outputStream* st) {
2276 if (!OSContainer::is_containerized()) {
2277 return;
2278 }
2279
2280 st->print("container (cgroup) information:\n");
2281
2282 const char *p_ct = OSContainer::container_type();
2283 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2284
2285 char *p = OSContainer::cpu_cpuset_cpus();
2286 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2287 free(p);
2288
2289 p = OSContainer::cpu_cpuset_memory_nodes();
2290 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2291 free(p);
2292
2293 int i = OSContainer::active_processor_count();
2294 if (i > 0) {
2295 st->print("active_processor_count: %d\n", i);
2296 } else {
2297 st->print("active_processor_count: failed\n");
2298 }
2299
2300 i = OSContainer::cpu_quota();
2301 st->print("cpu_quota: %d\n", i);
2302
2303 i = OSContainer::cpu_period();
2304 st->print("cpu_period: %d\n", i);
2305
2306 i = OSContainer::cpu_shares();
2307 st->print("cpu_shares: %d\n", i);
2308
2309 jlong j = OSContainer::memory_limit_in_bytes();
2310 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2311
2312 j = OSContainer::memory_and_swap_limit_in_bytes();
2313 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2314
2315 j = OSContainer::memory_soft_limit_in_bytes();
2316 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2317
2318 j = OSContainer::OSContainer::memory_usage_in_bytes();
2319 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2320
2321 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2322 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2323 st->cr();
2324 }
2325
2326 void os::print_memory_info(outputStream* st) {
2327
2328 st->print("Memory:");
2329 st->print(" %dk page", os::vm_page_size()>>10);
2330
2331 // values in struct sysinfo are "unsigned long"
2332 struct sysinfo si;
2333 sysinfo(&si);
2334
2335 st->print(", physical " UINT64_FORMAT "k",
2336 os::physical_memory() >> 10);
2337 st->print("(" UINT64_FORMAT "k free)",
2338 os::available_memory() >> 10);
2339 st->print(", swap " UINT64_FORMAT "k",
2340 ((jlong)si.totalswap * si.mem_unit) >> 10);
2341 st->print("(" UINT64_FORMAT "k free)",
2342 ((jlong)si.freeswap * si.mem_unit) >> 10);
2343 st->cr();
2344 }
2345
2346 void os::pd_print_cpu_info(outputStream* st) {
2347 st->print("\n/proc/cpuinfo:\n");
2348 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2349 st->print(" <Not Available>");
2350 }
2351 st->cr();
2352 }
2353
2354 void os::print_siginfo(outputStream* st, void* siginfo) {
2355 const siginfo_t* si = (const siginfo_t*)siginfo;
2356
2357 os::Posix::print_siginfo_brief(st, si);
2358 #if INCLUDE_CDS
2359 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2360 UseSharedSpaces) {
2361 FileMapInfo* mapinfo = FileMapInfo::current_info();
2362 if (mapinfo->is_in_shared_space(si->si_addr)) {
2363 st->print("\n\nError accessing class data sharing archive." \
2364 " Mapped file inaccessible during execution, " \
2365 " possible disk/network problem.");
2366 }
2367 }
2368 #endif
2369 st->cr();
2370 }
2371
2372
2373 static void print_signal_handler(outputStream* st, int sig,
2374 char* buf, size_t buflen);
2375
2376 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2377 st->print_cr("Signal Handlers:");
2378 print_signal_handler(st, SIGSEGV, buf, buflen);
2379 print_signal_handler(st, SIGBUS , buf, buflen);
2380 print_signal_handler(st, SIGFPE , buf, buflen);
2381 print_signal_handler(st, SIGPIPE, buf, buflen);
2382 print_signal_handler(st, SIGXFSZ, buf, buflen);
2383 print_signal_handler(st, SIGILL , buf, buflen);
2384 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2385 print_signal_handler(st, SR_signum, buf, buflen);
2386 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2387 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2388 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2389 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2390 #if defined(PPC64)
2391 print_signal_handler(st, SIGTRAP, buf, buflen);
2392 #endif
2393 }
2394
2395 static char saved_jvm_path[MAXPATHLEN] = {0};
2396
2397 // Find the full path to the current module, libjvm.so
2398 void os::jvm_path(char *buf, jint buflen) {
2399 // Error checking.
2400 if (buflen < MAXPATHLEN) {
2401 assert(false, "must use a large-enough buffer");
2402 buf[0] = '\0';
2403 return;
2404 }
2405 // Lazy resolve the path to current module.
2406 if (saved_jvm_path[0] != 0) {
2407 strcpy(buf, saved_jvm_path);
2408 return;
2409 }
2410
2411 char dli_fname[MAXPATHLEN];
2412 bool ret = dll_address_to_library_name(
2413 CAST_FROM_FN_PTR(address, os::jvm_path),
2414 dli_fname, sizeof(dli_fname), NULL);
2415 assert(ret, "cannot locate libjvm");
2416 char *rp = NULL;
2417 if (ret && dli_fname[0] != '\0') {
2418 rp = realpath(dli_fname, buf);
2419 }
2420 if (rp == NULL)
2421 return;
2422
2423 if (Arguments::created_by_gamma_launcher()) {
2424 // Support for the gamma launcher. Typical value for buf is
2425 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2426 // the right place in the string, then assume we are installed in a JDK and
2427 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2428 // up the path so it looks like libjvm.so is installed there (append a
2429 // fake suffix hotspot/libjvm.so).
2430 const char *p = buf + strlen(buf) - 1;
2431 for (int count = 0; p > buf && count < 5; ++count) {
2432 for (--p; p > buf && *p != '/'; --p)
2433 /* empty */ ;
2434 }
2435
2436 if (strncmp(p, "/jre/lib/", 9) != 0) {
2437 // Look for JAVA_HOME in the environment.
2438 char* java_home_var = ::getenv("JAVA_HOME");
2439 if (java_home_var != NULL && java_home_var[0] != 0) {
2440 char* jrelib_p;
2441 int len;
2442
2443 // Check the current module name "libjvm.so".
2444 p = strrchr(buf, '/');
2445 assert(strstr(p, "/libjvm") == p, "invalid library name");
2446
2447 rp = realpath(java_home_var, buf);
2448 if (rp == NULL)
2449 return;
2450
2451 // determine if this is a legacy image or modules image
2452 // modules image doesn't have "jre" subdirectory
2453 len = strlen(buf);
2454 assert(len < buflen, "Ran out of buffer room");
2455 jrelib_p = buf + len;
2456 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2457 if (0 != access(buf, F_OK)) {
2458 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2459 }
2460
2461 if (0 == access(buf, F_OK)) {
2462 // Use current module name "libjvm.so"
2463 len = strlen(buf);
2464 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2465 } else {
2466 // Go back to path of .so
2467 rp = realpath(dli_fname, buf);
2468 if (rp == NULL)
2469 return;
2470 }
2471 }
2472 }
2473 }
2474
2475 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2476 }
2477
2478 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2479 // no prefix required, not even "_"
2480 }
2481
2482 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2483 // no suffix required
2484 }
2485
2486 ////////////////////////////////////////////////////////////////////////////////
2487 // sun.misc.Signal support
2488
2489 static volatile jint sigint_count = 0;
2490
2491 static void
2492 UserHandler(int sig, void *siginfo, void *context) {
2493 // 4511530 - sem_post is serialized and handled by the manager thread. When
2494 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2495 // don't want to flood the manager thread with sem_post requests.
2496 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2497 return;
2498
2499 // Ctrl-C is pressed during error reporting, likely because the error
2500 // handler fails to abort. Let VM die immediately.
2501 if (sig == SIGINT && is_error_reported()) {
2502 os::die();
2503 }
2504
2505 os::signal_notify(sig);
2506 }
2507
2508 void* os::user_handler() {
2509 return CAST_FROM_FN_PTR(void*, UserHandler);
2510 }
2511
2512 class Semaphore : public StackObj {
2513 public:
2514 Semaphore();
2515 ~Semaphore();
2516 void signal();
2517 void wait();
2518 bool trywait();
2519 bool timedwait(unsigned int sec, int nsec);
2520 private:
2521 sem_t _semaphore;
2522 };
2523
2524 Semaphore::Semaphore() {
2525 sem_init(&_semaphore, 0, 0);
2526 }
2527
2528 Semaphore::~Semaphore() {
2529 sem_destroy(&_semaphore);
2530 }
2531
2532 void Semaphore::signal() {
2533 sem_post(&_semaphore);
2534 }
2535
2536 void Semaphore::wait() {
2537 sem_wait(&_semaphore);
2538 }
2539
2540 bool Semaphore::trywait() {
2541 return sem_trywait(&_semaphore) == 0;
2542 }
2543
2544 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2545
2546 struct timespec ts;
2547 // Semaphore's are always associated with CLOCK_REALTIME
2548 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2549 // see unpackTime for discussion on overflow checking
2550 if (sec >= MAX_SECS) {
2551 ts.tv_sec += MAX_SECS;
2552 ts.tv_nsec = 0;
2553 } else {
2554 ts.tv_sec += sec;
2555 ts.tv_nsec += nsec;
2556 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2557 ts.tv_nsec -= NANOSECS_PER_SEC;
2558 ++ts.tv_sec; // note: this must be <= max_secs
2559 }
2560 }
2561
2562 while (1) {
2563 int result = sem_timedwait(&_semaphore, &ts);
2564 if (result == 0) {
2565 return true;
2566 } else if (errno == EINTR) {
2567 continue;
2568 } else if (errno == ETIMEDOUT) {
2569 return false;
2570 } else {
2571 return false;
2572 }
2573 }
2574 }
2575
2576 extern "C" {
2577 typedef void (*sa_handler_t)(int);
2578 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2579 }
2580
2581 void* os::signal(int signal_number, void* handler) {
2582 struct sigaction sigAct, oldSigAct;
2583
2584 sigfillset(&(sigAct.sa_mask));
2585 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2586 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2587
2588 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2589 // -1 means registration failed
2590 return (void *)-1;
2591 }
2592
2593 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2594 }
2595
2596 void os::signal_raise(int signal_number) {
2597 ::raise(signal_number);
2598 }
2599
2600 /*
2601 * The following code is moved from os.cpp for making this
2602 * code platform specific, which it is by its very nature.
2603 */
2604
2605 // Will be modified when max signal is changed to be dynamic
2606 int os::sigexitnum_pd() {
2607 return NSIG;
2608 }
2609
2610 // a counter for each possible signal value
2611 static volatile jint pending_signals[NSIG+1] = { 0 };
2612
2613 // Linux(POSIX) specific hand shaking semaphore.
2614 static sem_t sig_sem;
2615 static Semaphore sr_semaphore;
2616
2617 void os::signal_init_pd() {
2618 // Initialize signal structures
2619 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2620
2621 // Initialize signal semaphore
2622 ::sem_init(&sig_sem, 0, 0);
2623 }
2624
2625 void os::signal_notify(int sig) {
2626 Atomic::inc(&pending_signals[sig]);
2627 ::sem_post(&sig_sem);
2628 }
2629
2630 static int check_pending_signals(bool wait) {
2631 Atomic::store(0, &sigint_count);
2632 for (;;) {
2633 for (int i = 0; i < NSIG + 1; i++) {
2634 jint n = pending_signals[i];
2635 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2636 return i;
2637 }
2638 }
2639 if (!wait) {
2640 return -1;
2641 }
2642 JavaThread *thread = JavaThread::current();
2643 ThreadBlockInVM tbivm(thread);
2644
2645 bool threadIsSuspended;
2646 do {
2647 thread->set_suspend_equivalent();
2648 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2649 ::sem_wait(&sig_sem);
2650
2651 // were we externally suspended while we were waiting?
2652 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2653 if (threadIsSuspended) {
2654 //
2655 // The semaphore has been incremented, but while we were waiting
2656 // another thread suspended us. We don't want to continue running
2657 // while suspended because that would surprise the thread that
2658 // suspended us.
2659 //
2660 ::sem_post(&sig_sem);
2661
2662 thread->java_suspend_self();
2663 }
2664 } while (threadIsSuspended);
2665 }
2666 }
2667
2668 int os::signal_lookup() {
2669 return check_pending_signals(false);
2670 }
2671
2672 int os::signal_wait() {
2673 return check_pending_signals(true);
2674 }
2675
2676 ////////////////////////////////////////////////////////////////////////////////
2677 // Virtual Memory
2678
2679 int os::vm_page_size() {
2680 // Seems redundant as all get out
2681 assert(os::Linux::page_size() != -1, "must call os::init");
2682 return os::Linux::page_size();
2683 }
2684
2685 // Solaris allocates memory by pages.
2686 int os::vm_allocation_granularity() {
2687 assert(os::Linux::page_size() != -1, "must call os::init");
2688 return os::Linux::page_size();
2689 }
2690
2691 // Rationale behind this function:
2692 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2693 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2694 // samples for JITted code. Here we create private executable mapping over the code cache
2695 // and then we can use standard (well, almost, as mapping can change) way to provide
2696 // info for the reporting script by storing timestamp and location of symbol
2697 void linux_wrap_code(char* base, size_t size) {
2698 static volatile jint cnt = 0;
2699
2700 if (!UseOprofile) {
2701 return;
2702 }
2703
2704 char buf[PATH_MAX+1];
2705 int num = Atomic::add(1, &cnt);
2706
2707 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2708 os::get_temp_directory(), os::current_process_id(), num);
2709 unlink(buf);
2710
2711 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2712
2713 if (fd != -1) {
2714 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2715 if (rv != (off_t)-1) {
2716 if (::write(fd, "", 1) == 1) {
2717 mmap(base, size,
2718 PROT_READ|PROT_WRITE|PROT_EXEC,
2719 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2720 }
2721 }
2722 ::close(fd);
2723 unlink(buf);
2724 }
2725 }
2726
2727 static bool recoverable_mmap_error(int err) {
2728 // See if the error is one we can let the caller handle. This
2729 // list of errno values comes from JBS-6843484. I can't find a
2730 // Linux man page that documents this specific set of errno
2731 // values so while this list currently matches Solaris, it may
2732 // change as we gain experience with this failure mode.
2733 switch (err) {
2734 case EBADF:
2735 case EINVAL:
2736 case ENOTSUP:
2737 // let the caller deal with these errors
2738 return true;
2739
2740 default:
2741 // Any remaining errors on this OS can cause our reserved mapping
2742 // to be lost. That can cause confusion where different data
2743 // structures think they have the same memory mapped. The worst
2744 // scenario is if both the VM and a library think they have the
2745 // same memory mapped.
2746 return false;
2747 }
2748 }
2749
2750 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2751 int err) {
2752 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2753 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2754 strerror(err), err);
2755 }
2756
2757 static void warn_fail_commit_memory(char* addr, size_t size,
2758 size_t alignment_hint, bool exec,
2759 int err) {
2760 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2761 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2762 alignment_hint, exec, strerror(err), err);
2763 }
2764
2765 // NOTE: Linux kernel does not really reserve the pages for us.
2766 // All it does is to check if there are enough free pages
2767 // left at the time of mmap(). This could be a potential
2768 // problem.
2769 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2770 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2771 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2772 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2773 if (res != (uintptr_t) MAP_FAILED) {
2774 if (UseNUMAInterleaving) {
2775 numa_make_global(addr, size);
2776 }
2777 return 0;
2778 }
2779
2780 int err = errno; // save errno from mmap() call above
2781
2782 if (!recoverable_mmap_error(err)) {
2783 warn_fail_commit_memory(addr, size, exec, err);
2784 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2785 }
2786
2787 return err;
2788 }
2789
2790 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2791 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2792 }
2793
2794 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2795 const char* mesg) {
2796 assert(mesg != NULL, "mesg must be specified");
2797 int err = os::Linux::commit_memory_impl(addr, size, exec);
2798 if (err != 0) {
2799 // the caller wants all commit errors to exit with the specified mesg:
2800 warn_fail_commit_memory(addr, size, exec, err);
2801 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2802 }
2803 }
2804
2805 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2806 #ifndef MAP_HUGETLB
2807 #define MAP_HUGETLB 0x40000
2808 #endif
2809
2810 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2811 #ifndef MADV_HUGEPAGE
2812 #define MADV_HUGEPAGE 14
2813 #endif
2814
2815 int os::Linux::commit_memory_impl(char* addr, size_t size,
2816 size_t alignment_hint, bool exec) {
2817 int err = os::Linux::commit_memory_impl(addr, size, exec);
2818 if (err == 0) {
2819 realign_memory(addr, size, alignment_hint);
2820 }
2821 return err;
2822 }
2823
2824 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2825 bool exec) {
2826 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2827 }
2828
2829 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2830 size_t alignment_hint, bool exec,
2831 const char* mesg) {
2832 assert(mesg != NULL, "mesg must be specified");
2833 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2834 if (err != 0) {
2835 // the caller wants all commit errors to exit with the specified mesg:
2836 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2837 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2838 }
2839 }
2840
2841 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2842 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2843 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2844 // be supported or the memory may already be backed by huge pages.
2845 ::madvise(addr, bytes, MADV_HUGEPAGE);
2846 }
2847 }
2848
2849 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2850 // This method works by doing an mmap over an existing mmaping and effectively discarding
2851 // the existing pages. However it won't work for SHM-based large pages that cannot be
2852 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2853 // small pages on top of the SHM segment. This method always works for small pages, so we
2854 // allow that in any case.
2855 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2856 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2857 }
2858 }
2859
2860 void os::numa_make_global(char *addr, size_t bytes) {
2861 Linux::numa_interleave_memory(addr, bytes);
2862 }
2863
2864 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2865 // bind policy to MPOL_PREFERRED for the current thread.
2866 #define USE_MPOL_PREFERRED 0
2867
2868 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2869 // To make NUMA and large pages more robust when both enabled, we need to ease
2870 // the requirements on where the memory should be allocated. MPOL_BIND is the
2871 // default policy and it will force memory to be allocated on the specified
2872 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2873 // the specified node, but will not force it. Using this policy will prevent
2874 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2875 // free large pages.
2876 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2877 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2878 }
2879
2880 bool os::numa_topology_changed() { return false; }
2881
2882 size_t os::numa_get_groups_num() {
2883 // Return just the number of nodes in which it's possible to allocate memory
2884 // (in numa terminology, configured nodes).
2885 return Linux::numa_num_configured_nodes();
2886 }
2887
2888 int os::numa_get_group_id() {
2889 int cpu_id = Linux::sched_getcpu();
2890 if (cpu_id != -1) {
2891 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2892 if (lgrp_id != -1) {
2893 return lgrp_id;
2894 }
2895 }
2896 return 0;
2897 }
2898
2899 int os::Linux::get_existing_num_nodes() {
2900 size_t node;
2901 size_t highest_node_number = Linux::numa_max_node();
2902 int num_nodes = 0;
2903
2904 // Get the total number of nodes in the system including nodes without memory.
2905 for (node = 0; node <= highest_node_number; node++) {
2906 if (isnode_in_existing_nodes(node)) {
2907 num_nodes++;
2908 }
2909 }
2910 return num_nodes;
2911 }
2912
2913 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2914 size_t highest_node_number = Linux::numa_max_node();
2915 size_t i = 0;
2916
2917 // Map all node ids in which is possible to allocate memory. Also nodes are
2918 // not always consecutively available, i.e. available from 0 to the highest
2919 // node number.
2920 for (size_t node = 0; node <= highest_node_number; node++) {
2921 if (Linux::isnode_in_configured_nodes(node)) {
2922 ids[i++] = node;
2923 }
2924 }
2925 return i;
2926 }
2927
2928 bool os::get_page_info(char *start, page_info* info) {
2929 return false;
2930 }
2931
2932 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2933 return end;
2934 }
2935
2936
2937 int os::Linux::sched_getcpu_syscall(void) {
2938 unsigned int cpu = 0;
2939 int retval = -1;
2940
2941 #if defined(IA32)
2942 # ifndef SYS_getcpu
2943 # define SYS_getcpu 318
2944 # endif
2945 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2946 #elif defined(AMD64)
2947 // Unfortunately we have to bring all these macros here from vsyscall.h
2948 // to be able to compile on old linuxes.
2949 # define __NR_vgetcpu 2
2950 # define VSYSCALL_START (-10UL << 20)
2951 # define VSYSCALL_SIZE 1024
2952 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2953 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2954 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2955 retval = vgetcpu(&cpu, NULL, NULL);
2956 #endif
2957
2958 return (retval == -1) ? retval : cpu;
2959 }
2960
2961 // Something to do with the numa-aware allocator needs these symbols
2962 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2963 extern "C" JNIEXPORT void numa_error(char *where) { }
2964 extern "C" JNIEXPORT int fork1() { return fork(); }
2965
2966 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2967 // load symbol from base version instead.
2968 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2969 void *f = dlvsym(handle, name, "libnuma_1.1");
2970 if (f == NULL) {
2971 f = dlsym(handle, name);
2972 }
2973 return f;
2974 }
2975
2976 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2977 // Return NULL if the symbol is not defined in this particular version.
2978 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2979 return dlvsym(handle, name, "libnuma_1.2");
2980 }
2981
2982 bool os::Linux::libnuma_init() {
2983 // sched_getcpu() should be in libc.
2984 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2985 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2986
2987 // If it's not, try a direct syscall.
2988 if (sched_getcpu() == -1)
2989 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2990
2991 if (sched_getcpu() != -1) { // Does it work?
2992 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2993 if (handle != NULL) {
2994 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2995 libnuma_dlsym(handle, "numa_node_to_cpus")));
2996 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2997 libnuma_dlsym(handle, "numa_max_node")));
2998 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2999 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3000 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3001 libnuma_dlsym(handle, "numa_available")));
3002 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3003 libnuma_dlsym(handle, "numa_tonode_memory")));
3004 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3005 libnuma_dlsym(handle, "numa_interleave_memory")));
3006 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3007 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3008 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3009 libnuma_dlsym(handle, "numa_set_bind_policy")));
3010 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3011 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3012 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3013 libnuma_dlsym(handle, "numa_distance")));
3014
3015 if (numa_available() != -1) {
3016 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3017 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3018 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3019 // Create an index -> node mapping, since nodes are not always consecutive
3020 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3021 rebuild_nindex_to_node_map();
3022 // Create a cpu -> node mapping
3023 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3024 rebuild_cpu_to_node_map();
3025 return true;
3026 }
3027 }
3028 }
3029 return false;
3030 }
3031
3032 void os::Linux::rebuild_nindex_to_node_map() {
3033 int highest_node_number = Linux::numa_max_node();
3034
3035 nindex_to_node()->clear();
3036 for (int node = 0; node <= highest_node_number; node++) {
3037 if (Linux::isnode_in_existing_nodes(node)) {
3038 nindex_to_node()->append(node);
3039 }
3040 }
3041 }
3042
3043 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3044 // The table is later used in get_node_by_cpu().
3045 void os::Linux::rebuild_cpu_to_node_map() {
3046 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3047 // in libnuma (possible values are starting from 16,
3048 // and continuing up with every other power of 2, but less
3049 // than the maximum number of CPUs supported by kernel), and
3050 // is a subject to change (in libnuma version 2 the requirements
3051 // are more reasonable) we'll just hardcode the number they use
3052 // in the library.
3053 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3054
3055 size_t cpu_num = processor_count();
3056 size_t cpu_map_size = NCPUS / BitsPerCLong;
3057 size_t cpu_map_valid_size =
3058 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3059
3060 cpu_to_node()->clear();
3061 cpu_to_node()->at_grow(cpu_num - 1);
3062
3063 size_t node_num = get_existing_num_nodes();
3064
3065 int distance = 0;
3066 int closest_distance = INT_MAX;
3067 int closest_node = 0;
3068 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3069 for (size_t i = 0; i < node_num; i++) {
3070 // Check if node is configured (not a memory-less node). If it is not, find
3071 // the closest configured node.
3072 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3073 closest_distance = INT_MAX;
3074 // Check distance from all remaining nodes in the system. Ignore distance
3075 // from itself and from another non-configured node.
3076 for (size_t m = 0; m < node_num; m++) {
3077 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3078 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3079 // If a closest node is found, update. There is always at least one
3080 // configured node in the system so there is always at least one node
3081 // close.
3082 if (distance != 0 && distance < closest_distance) {
3083 closest_distance = distance;
3084 closest_node = nindex_to_node()->at(m);
3085 }
3086 }
3087 }
3088 } else {
3089 // Current node is already a configured node.
3090 closest_node = nindex_to_node()->at(i);
3091 }
3092
3093 // Get cpus from the original node and map them to the closest node. If node
3094 // is a configured node (not a memory-less node), then original node and
3095 // closest node are the same.
3096 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3097 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3098 if (cpu_map[j] != 0) {
3099 for (size_t k = 0; k < BitsPerCLong; k++) {
3100 if (cpu_map[j] & (1UL << k)) {
3101 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3102 }
3103 }
3104 }
3105 }
3106 }
3107 }
3108 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3109 }
3110
3111 int os::Linux::get_node_by_cpu(int cpu_id) {
3112 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3113 return cpu_to_node()->at(cpu_id);
3114 }
3115 return -1;
3116 }
3117
3118 GrowableArray<int>* os::Linux::_cpu_to_node;
3119 GrowableArray<int>* os::Linux::_nindex_to_node;
3120 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3121 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3122 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3123 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3124 os::Linux::numa_available_func_t os::Linux::_numa_available;
3125 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3126 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3127 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3128 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3129 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3130 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3131 unsigned long* os::Linux::_numa_all_nodes;
3132 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3133 struct bitmask* os::Linux::_numa_nodes_ptr;
3134
3135 bool os::pd_uncommit_memory(char* addr, size_t size) {
3136 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3137 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3138 return res != (uintptr_t) MAP_FAILED;
3139 }
3140
3141 static
3142 address get_stack_commited_bottom(address bottom, size_t size) {
3143 address nbot = bottom;
3144 address ntop = bottom + size;
3145
3146 size_t page_sz = os::vm_page_size();
3147 unsigned pages = size / page_sz;
3148
3149 unsigned char vec[1];
3150 unsigned imin = 1, imax = pages + 1, imid;
3151 int mincore_return_value = 0;
3152
3153 assert(imin <= imax, "Unexpected page size");
3154
3155 while (imin < imax) {
3156 imid = (imax + imin) / 2;
3157 nbot = ntop - (imid * page_sz);
3158
3159 // Use a trick with mincore to check whether the page is mapped or not.
3160 // mincore sets vec to 1 if page resides in memory and to 0 if page
3161 // is swapped output but if page we are asking for is unmapped
3162 // it returns -1,ENOMEM
3163 mincore_return_value = mincore(nbot, page_sz, vec);
3164
3165 if (mincore_return_value == -1) {
3166 // Page is not mapped go up
3167 // to find first mapped page
3168 if (errno != EAGAIN) {
3169 assert(errno == ENOMEM, "Unexpected mincore errno");
3170 imax = imid;
3171 }
3172 } else {
3173 // Page is mapped go down
3174 // to find first not mapped page
3175 imin = imid + 1;
3176 }
3177 }
3178
3179 nbot = nbot + page_sz;
3180
3181 // Adjust stack bottom one page up if last checked page is not mapped
3182 if (mincore_return_value == -1) {
3183 nbot = nbot + page_sz;
3184 }
3185
3186 return nbot;
3187 }
3188
3189
3190 // Linux uses a growable mapping for the stack, and if the mapping for
3191 // the stack guard pages is not removed when we detach a thread the
3192 // stack cannot grow beyond the pages where the stack guard was
3193 // mapped. If at some point later in the process the stack expands to
3194 // that point, the Linux kernel cannot expand the stack any further
3195 // because the guard pages are in the way, and a segfault occurs.
3196 //
3197 // However, it's essential not to split the stack region by unmapping
3198 // a region (leaving a hole) that's already part of the stack mapping,
3199 // so if the stack mapping has already grown beyond the guard pages at
3200 // the time we create them, we have to truncate the stack mapping.
3201 // So, we need to know the extent of the stack mapping when
3202 // create_stack_guard_pages() is called.
3203
3204 // We only need this for stacks that are growable: at the time of
3205 // writing thread stacks don't use growable mappings (i.e. those
3206 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3207 // only applies to the main thread.
3208
3209 // If the (growable) stack mapping already extends beyond the point
3210 // where we're going to put our guard pages, truncate the mapping at
3211 // that point by munmap()ping it. This ensures that when we later
3212 // munmap() the guard pages we don't leave a hole in the stack
3213 // mapping. This only affects the main/primordial thread
3214
3215 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3216
3217 if (os::is_primordial_thread()) {
3218 // As we manually grow stack up to bottom inside create_attached_thread(),
3219 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3220 // we don't need to do anything special.
3221 // Check it first, before calling heavy function.
3222 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3223 unsigned char vec[1];
3224
3225 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3226 // Fallback to slow path on all errors, including EAGAIN
3227 stack_extent = (uintptr_t) get_stack_commited_bottom(
3228 os::Linux::initial_thread_stack_bottom(),
3229 (size_t)addr - stack_extent);
3230 }
3231
3232 if (stack_extent < (uintptr_t)addr) {
3233 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3234 }
3235 }
3236
3237 return os::commit_memory(addr, size, !ExecMem);
3238 }
3239
3240 // If this is a growable mapping, remove the guard pages entirely by
3241 // munmap()ping them. If not, just call uncommit_memory(). This only
3242 // affects the main/primordial thread, but guard against future OS changes.
3243 // It's safe to always unmap guard pages for primordial thread because we
3244 // always place it right after end of the mapped region.
3245
3246 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3247 uintptr_t stack_extent, stack_base;
3248
3249 if (os::is_primordial_thread()) {
3250 return ::munmap(addr, size) == 0;
3251 }
3252
3253 return os::uncommit_memory(addr, size);
3254 }
3255
3256 static address _highest_vm_reserved_address = NULL;
3257
3258 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3259 // at 'requested_addr'. If there are existing memory mappings at the same
3260 // location, however, they will be overwritten. If 'fixed' is false,
3261 // 'requested_addr' is only treated as a hint, the return value may or
3262 // may not start from the requested address. Unlike Linux mmap(), this
3263 // function returns NULL to indicate failure.
3264 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3265 char * addr;
3266 int flags;
3267
3268 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3269 if (fixed) {
3270 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3271 flags |= MAP_FIXED;
3272 }
3273
3274 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3275 // touch an uncommitted page. Otherwise, the read/write might
3276 // succeed if we have enough swap space to back the physical page.
3277 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3278 flags, -1, 0);
3279
3280 if (addr != MAP_FAILED) {
3281 // anon_mmap() should only get called during VM initialization,
3282 // don't need lock (actually we can skip locking even it can be called
3283 // from multiple threads, because _highest_vm_reserved_address is just a
3284 // hint about the upper limit of non-stack memory regions.)
3285 if ((address)addr + bytes > _highest_vm_reserved_address) {
3286 _highest_vm_reserved_address = (address)addr + bytes;
3287 }
3288 }
3289
3290 return addr == MAP_FAILED ? NULL : addr;
3291 }
3292
3293 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3294 // (req_addr != NULL) or with a given alignment.
3295 // - bytes shall be a multiple of alignment.
3296 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3297 // - alignment sets the alignment at which memory shall be allocated.
3298 // It must be a multiple of allocation granularity.
3299 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3300 // req_addr or NULL.
3301 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3302
3303 size_t extra_size = bytes;
3304 if (req_addr == NULL && alignment > 0) {
3305 extra_size += alignment;
3306 }
3307
3308 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3309 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3310 -1, 0);
3311 if (start == MAP_FAILED) {
3312 start = NULL;
3313 } else {
3314 if (req_addr != NULL) {
3315 if (start != req_addr) {
3316 ::munmap(start, extra_size);
3317 start = NULL;
3318 }
3319 } else {
3320 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3321 char* const end_aligned = start_aligned + bytes;
3322 char* const end = start + extra_size;
3323 if (start_aligned > start) {
3324 ::munmap(start, start_aligned - start);
3325 }
3326 if (end_aligned < end) {
3327 ::munmap(end_aligned, end - end_aligned);
3328 }
3329 start = start_aligned;
3330 }
3331 }
3332 return start;
3333 }
3334
3335 // Don't update _highest_vm_reserved_address, because there might be memory
3336 // regions above addr + size. If so, releasing a memory region only creates
3337 // a hole in the address space, it doesn't help prevent heap-stack collision.
3338 //
3339 static int anon_munmap(char * addr, size_t size) {
3340 return ::munmap(addr, size) == 0;
3341 }
3342
3343 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3344 size_t alignment_hint) {
3345 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3346 }
3347
3348 bool os::pd_release_memory(char* addr, size_t size) {
3349 return anon_munmap(addr, size);
3350 }
3351
3352 static address highest_vm_reserved_address() {
3353 return _highest_vm_reserved_address;
3354 }
3355
3356 static bool linux_mprotect(char* addr, size_t size, int prot) {
3357 // Linux wants the mprotect address argument to be page aligned.
3358 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3359
3360 // According to SUSv3, mprotect() should only be used with mappings
3361 // established by mmap(), and mmap() always maps whole pages. Unaligned
3362 // 'addr' likely indicates problem in the VM (e.g. trying to change
3363 // protection of malloc'ed or statically allocated memory). Check the
3364 // caller if you hit this assert.
3365 assert(addr == bottom, "sanity check");
3366
3367 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3368 return ::mprotect(bottom, size, prot) == 0;
3369 }
3370
3371 // Set protections specified
3372 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3373 bool is_committed) {
3374 unsigned int p = 0;
3375 switch (prot) {
3376 case MEM_PROT_NONE: p = PROT_NONE; break;
3377 case MEM_PROT_READ: p = PROT_READ; break;
3378 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3379 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3380 default:
3381 ShouldNotReachHere();
3382 }
3383 // is_committed is unused.
3384 return linux_mprotect(addr, bytes, p);
3385 }
3386
3387 bool os::guard_memory(char* addr, size_t size) {
3388 return linux_mprotect(addr, size, PROT_NONE);
3389 }
3390
3391 bool os::unguard_memory(char* addr, size_t size) {
3392 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3393 }
3394
3395 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3396 bool result = false;
3397 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3398 MAP_ANONYMOUS|MAP_PRIVATE,
3399 -1, 0);
3400 if (p != MAP_FAILED) {
3401 void *aligned_p = align_ptr_up(p, page_size);
3402
3403 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3404
3405 munmap(p, page_size * 2);
3406 }
3407
3408 if (warn && !result) {
3409 warning("TransparentHugePages is not supported by the operating system.");
3410 }
3411
3412 return result;
3413 }
3414
3415 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3416 bool result = false;
3417 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3418 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3419 -1, 0);
3420
3421 if (p != MAP_FAILED) {
3422 // We don't know if this really is a huge page or not.
3423 FILE *fp = fopen("/proc/self/maps", "r");
3424 if (fp) {
3425 while (!feof(fp)) {
3426 char chars[257];
3427 long x = 0;
3428 if (fgets(chars, sizeof(chars), fp)) {
3429 if (sscanf(chars, "%lx-%*x", &x) == 1
3430 && x == (long)p) {
3431 if (strstr (chars, "hugepage")) {
3432 result = true;
3433 break;
3434 }
3435 }
3436 }
3437 }
3438 fclose(fp);
3439 }
3440 munmap(p, page_size);
3441 }
3442
3443 if (warn && !result) {
3444 warning("HugeTLBFS is not supported by the operating system.");
3445 }
3446
3447 return result;
3448 }
3449
3450 /*
3451 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3452 *
3453 * From the coredump_filter documentation:
3454 *
3455 * - (bit 0) anonymous private memory
3456 * - (bit 1) anonymous shared memory
3457 * - (bit 2) file-backed private memory
3458 * - (bit 3) file-backed shared memory
3459 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3460 * effective only if the bit 2 is cleared)
3461 * - (bit 5) hugetlb private memory
3462 * - (bit 6) hugetlb shared memory
3463 */
3464 static void set_coredump_filter(void) {
3465 FILE *f;
3466 long cdm;
3467
3468 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3469 return;
3470 }
3471
3472 if (fscanf(f, "%lx", &cdm) != 1) {
3473 fclose(f);
3474 return;
3475 }
3476
3477 rewind(f);
3478
3479 if ((cdm & LARGEPAGES_BIT) == 0) {
3480 cdm |= LARGEPAGES_BIT;
3481 fprintf(f, "%#lx", cdm);
3482 }
3483
3484 fclose(f);
3485 }
3486
3487 // Large page support
3488
3489 static size_t _large_page_size = 0;
3490
3491 size_t os::Linux::find_large_page_size() {
3492 size_t large_page_size = 0;
3493
3494 // large_page_size on Linux is used to round up heap size. x86 uses either
3495 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3496 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3497 // page as large as 256M.
3498 //
3499 // Here we try to figure out page size by parsing /proc/meminfo and looking
3500 // for a line with the following format:
3501 // Hugepagesize: 2048 kB
3502 //
3503 // If we can't determine the value (e.g. /proc is not mounted, or the text
3504 // format has been changed), we'll use the largest page size supported by
3505 // the processor.
3506
3507 #ifndef ZERO
3508 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3509 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3510 #endif // ZERO
3511
3512 FILE *fp = fopen("/proc/meminfo", "r");
3513 if (fp) {
3514 while (!feof(fp)) {
3515 int x = 0;
3516 char buf[16];
3517 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3518 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3519 large_page_size = x * K;
3520 break;
3521 }
3522 } else {
3523 // skip to next line
3524 for (;;) {
3525 int ch = fgetc(fp);
3526 if (ch == EOF || ch == (int)'\n') break;
3527 }
3528 }
3529 }
3530 fclose(fp);
3531 }
3532
3533 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3534 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3535 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3536 proper_unit_for_byte_size(large_page_size));
3537 }
3538
3539 return large_page_size;
3540 }
3541
3542 size_t os::Linux::setup_large_page_size() {
3543 _large_page_size = Linux::find_large_page_size();
3544 const size_t default_page_size = (size_t)Linux::page_size();
3545 if (_large_page_size > default_page_size) {
3546 _page_sizes[0] = _large_page_size;
3547 _page_sizes[1] = default_page_size;
3548 _page_sizes[2] = 0;
3549 }
3550
3551 return _large_page_size;
3552 }
3553
3554 bool os::Linux::setup_large_page_type(size_t page_size) {
3555 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3556 FLAG_IS_DEFAULT(UseSHM) &&
3557 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3558
3559 // The type of large pages has not been specified by the user.
3560
3561 // Try UseHugeTLBFS and then UseSHM.
3562 UseHugeTLBFS = UseSHM = true;
3563
3564 // Don't try UseTransparentHugePages since there are known
3565 // performance issues with it turned on. This might change in the future.
3566 UseTransparentHugePages = false;
3567 }
3568
3569 if (UseTransparentHugePages) {
3570 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3571 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3572 UseHugeTLBFS = false;
3573 UseSHM = false;
3574 return true;
3575 }
3576 UseTransparentHugePages = false;
3577 }
3578
3579 if (UseHugeTLBFS) {
3580 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3581 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3582 UseSHM = false;
3583 return true;
3584 }
3585 UseHugeTLBFS = false;
3586 }
3587
3588 return UseSHM;
3589 }
3590
3591 void os::large_page_init() {
3592 if (!UseLargePages &&
3593 !UseTransparentHugePages &&
3594 !UseHugeTLBFS &&
3595 !UseSHM) {
3596 // Not using large pages.
3597 return;
3598 }
3599
3600 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3601 // The user explicitly turned off large pages.
3602 // Ignore the rest of the large pages flags.
3603 UseTransparentHugePages = false;
3604 UseHugeTLBFS = false;
3605 UseSHM = false;
3606 return;
3607 }
3608
3609 size_t large_page_size = Linux::setup_large_page_size();
3610 UseLargePages = Linux::setup_large_page_type(large_page_size);
3611
3612 set_coredump_filter();
3613 }
3614
3615 #ifndef SHM_HUGETLB
3616 #define SHM_HUGETLB 04000
3617 #endif
3618
3619 #define shm_warning_format(format, ...) \
3620 do { \
3621 if (UseLargePages && \
3622 (!FLAG_IS_DEFAULT(UseLargePages) || \
3623 !FLAG_IS_DEFAULT(UseSHM) || \
3624 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3625 warning(format, __VA_ARGS__); \
3626 } \
3627 } while (0)
3628
3629 #define shm_warning(str) shm_warning_format("%s", str)
3630
3631 #define shm_warning_with_errno(str) \
3632 do { \
3633 int err = errno; \
3634 shm_warning_format(str " (error = %d)", err); \
3635 } while (0)
3636
3637 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3638 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3639
3640 if (!is_size_aligned(alignment, SHMLBA)) {
3641 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3642 return NULL;
3643 }
3644
3645 // To ensure that we get 'alignment' aligned memory from shmat,
3646 // we pre-reserve aligned virtual memory and then attach to that.
3647
3648 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3649 if (pre_reserved_addr == NULL) {
3650 // Couldn't pre-reserve aligned memory.
3651 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3652 return NULL;
3653 }
3654
3655 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3656 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3657
3658 if ((intptr_t)addr == -1) {
3659 int err = errno;
3660 shm_warning_with_errno("Failed to attach shared memory.");
3661
3662 assert(err != EACCES, "Unexpected error");
3663 assert(err != EIDRM, "Unexpected error");
3664 assert(err != EINVAL, "Unexpected error");
3665
3666 // Since we don't know if the kernel unmapped the pre-reserved memory area
3667 // we can't unmap it, since that would potentially unmap memory that was
3668 // mapped from other threads.
3669 return NULL;
3670 }
3671
3672 return addr;
3673 }
3674
3675 static char* shmat_at_address(int shmid, char* req_addr) {
3676 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3677 assert(false, "Requested address needs to be SHMLBA aligned");
3678 return NULL;
3679 }
3680
3681 char* addr = (char*)shmat(shmid, req_addr, 0);
3682
3683 if ((intptr_t)addr == -1) {
3684 shm_warning_with_errno("Failed to attach shared memory.");
3685 return NULL;
3686 }
3687
3688 return addr;
3689 }
3690
3691 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3692 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3693 if (req_addr != NULL) {
3694 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3695 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3696 return shmat_at_address(shmid, req_addr);
3697 }
3698
3699 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3700 // return large page size aligned memory addresses when req_addr == NULL.
3701 // However, if the alignment is larger than the large page size, we have
3702 // to manually ensure that the memory returned is 'alignment' aligned.
3703 if (alignment > os::large_page_size()) {
3704 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3705 return shmat_with_alignment(shmid, bytes, alignment);
3706 } else {
3707 return shmat_at_address(shmid, NULL);
3708 }
3709 }
3710
3711 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3712 // "exec" is passed in but not used. Creating the shared image for
3713 // the code cache doesn't have an SHM_X executable permission to check.
3714 assert(UseLargePages && UseSHM, "only for SHM large pages");
3715 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3716 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3717
3718 if (!is_size_aligned(bytes, os::large_page_size())) {
3719 return NULL; // Fallback to small pages.
3720 }
3721
3722 // Create a large shared memory region to attach to based on size.
3723 // Currently, size is the total size of the heap.
3724 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3725 if (shmid == -1) {
3726 // Possible reasons for shmget failure:
3727 // 1. shmmax is too small for Java heap.
3728 // > check shmmax value: cat /proc/sys/kernel/shmmax
3729 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3730 // 2. not enough large page memory.
3731 // > check available large pages: cat /proc/meminfo
3732 // > increase amount of large pages:
3733 // echo new_value > /proc/sys/vm/nr_hugepages
3734 // Note 1: different Linux may use different name for this property,
3735 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3736 // Note 2: it's possible there's enough physical memory available but
3737 // they are so fragmented after a long run that they can't
3738 // coalesce into large pages. Try to reserve large pages when
3739 // the system is still "fresh".
3740 shm_warning_with_errno("Failed to reserve shared memory.");
3741 return NULL;
3742 }
3743
3744 // Attach to the region.
3745 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3746
3747 // Remove shmid. If shmat() is successful, the actual shared memory segment
3748 // will be deleted when it's detached by shmdt() or when the process
3749 // terminates. If shmat() is not successful this will remove the shared
3750 // segment immediately.
3751 shmctl(shmid, IPC_RMID, NULL);
3752
3753 return addr;
3754 }
3755
3756 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3757 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3758
3759 bool warn_on_failure = UseLargePages &&
3760 (!FLAG_IS_DEFAULT(UseLargePages) ||
3761 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3762 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3763
3764 if (warn_on_failure) {
3765 char msg[128];
3766 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3767 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3768 warning("%s", msg);
3769 }
3770 }
3771
3772 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3773 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3774 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3775 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3776
3777 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3778 char* addr = (char*)::mmap(req_addr, bytes, prot,
3779 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3780 -1, 0);
3781
3782 if (addr == MAP_FAILED) {
3783 warn_on_large_pages_failure(req_addr, bytes, errno);
3784 return NULL;
3785 }
3786
3787 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3788
3789 return addr;
3790 }
3791
3792 // Reserve memory using mmap(MAP_HUGETLB).
3793 // - bytes shall be a multiple of alignment.
3794 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3795 // - alignment sets the alignment at which memory shall be allocated.
3796 // It must be a multiple of allocation granularity.
3797 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3798 // req_addr or NULL.
3799 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3800 size_t large_page_size = os::large_page_size();
3801 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3802
3803 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3804 assert(is_size_aligned(bytes, alignment), "Must be");
3805
3806 // First reserve - but not commit - the address range in small pages.
3807 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3808
3809 if (start == NULL) {
3810 return NULL;
3811 }
3812
3813 assert(is_ptr_aligned(start, alignment), "Must be");
3814
3815 char* end = start + bytes;
3816
3817 // Find the regions of the allocated chunk that can be promoted to large pages.
3818 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3819 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3820
3821 size_t lp_bytes = lp_end - lp_start;
3822
3823 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3824
3825 if (lp_bytes == 0) {
3826 // The mapped region doesn't even span the start and the end of a large page.
3827 // Fall back to allocate a non-special area.
3828 ::munmap(start, end - start);
3829 return NULL;
3830 }
3831
3832 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3833
3834 void* result;
3835
3836 // Commit small-paged leading area.
3837 if (start != lp_start) {
3838 result = ::mmap(start, lp_start - start, prot,
3839 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3840 -1, 0);
3841 if (result == MAP_FAILED) {
3842 ::munmap(lp_start, end - lp_start);
3843 return NULL;
3844 }
3845 }
3846
3847 // Commit large-paged area.
3848 result = ::mmap(lp_start, lp_bytes, prot,
3849 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3850 -1, 0);
3851 if (result == MAP_FAILED) {
3852 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3853 // If the mmap above fails, the large pages region will be unmapped and we
3854 // have regions before and after with small pages. Release these regions.
3855 //
3856 // | mapped | unmapped | mapped |
3857 // ^ ^ ^ ^
3858 // start lp_start lp_end end
3859 //
3860 ::munmap(start, lp_start - start);
3861 ::munmap(lp_end, end - lp_end);
3862 return NULL;
3863 }
3864
3865 // Commit small-paged trailing area.
3866 if (lp_end != end) {
3867 result = ::mmap(lp_end, end - lp_end, prot,
3868 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3869 -1, 0);
3870 if (result == MAP_FAILED) {
3871 ::munmap(start, lp_end - start);
3872 return NULL;
3873 }
3874 }
3875
3876 return start;
3877 }
3878
3879 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3880 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3881 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3882 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3883 assert(is_power_of_2(os::large_page_size()), "Must be");
3884 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3885
3886 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3887 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3888 } else {
3889 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3890 }
3891 }
3892
3893 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3894 assert(UseLargePages, "only for large pages");
3895
3896 char* addr;
3897 if (UseSHM) {
3898 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3899 } else {
3900 assert(UseHugeTLBFS, "must be");
3901 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3902 }
3903
3904 if (addr != NULL) {
3905 if (UseNUMAInterleaving) {
3906 numa_make_global(addr, bytes);
3907 }
3908
3909 // The memory is committed
3910 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3911 }
3912
3913 return addr;
3914 }
3915
3916 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3917 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3918 return shmdt(base) == 0;
3919 }
3920
3921 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3922 return pd_release_memory(base, bytes);
3923 }
3924
3925 bool os::release_memory_special(char* base, size_t bytes) {
3926 bool res;
3927 if (MemTracker::tracking_level() > NMT_minimal) {
3928 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3929 res = os::Linux::release_memory_special_impl(base, bytes);
3930 if (res) {
3931 tkr.record((address)base, bytes);
3932 }
3933
3934 } else {
3935 res = os::Linux::release_memory_special_impl(base, bytes);
3936 }
3937 return res;
3938 }
3939
3940 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3941 assert(UseLargePages, "only for large pages");
3942 bool res;
3943
3944 if (UseSHM) {
3945 res = os::Linux::release_memory_special_shm(base, bytes);
3946 } else {
3947 assert(UseHugeTLBFS, "must be");
3948 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3949 }
3950 return res;
3951 }
3952
3953 size_t os::large_page_size() {
3954 return _large_page_size;
3955 }
3956
3957 // With SysV SHM the entire memory region must be allocated as shared
3958 // memory.
3959 // HugeTLBFS allows application to commit large page memory on demand.
3960 // However, when committing memory with HugeTLBFS fails, the region
3961 // that was supposed to be committed will lose the old reservation
3962 // and allow other threads to steal that memory region. Because of this
3963 // behavior we can't commit HugeTLBFS memory.
3964 bool os::can_commit_large_page_memory() {
3965 return UseTransparentHugePages;
3966 }
3967
3968 bool os::can_execute_large_page_memory() {
3969 return UseTransparentHugePages || UseHugeTLBFS;
3970 }
3971
3972 // Reserve memory at an arbitrary address, only if that area is
3973 // available (and not reserved for something else).
3974
3975 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3976 const int max_tries = 10;
3977 char* base[max_tries];
3978 size_t size[max_tries];
3979 const size_t gap = 0x000000;
3980
3981 // Assert only that the size is a multiple of the page size, since
3982 // that's all that mmap requires, and since that's all we really know
3983 // about at this low abstraction level. If we need higher alignment,
3984 // we can either pass an alignment to this method or verify alignment
3985 // in one of the methods further up the call chain. See bug 5044738.
3986 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3987
3988 // Repeatedly allocate blocks until the block is allocated at the
3989 // right spot. Give up after max_tries. Note that reserve_memory() will
3990 // automatically update _highest_vm_reserved_address if the call is
3991 // successful. The variable tracks the highest memory address every reserved
3992 // by JVM. It is used to detect heap-stack collision if running with
3993 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3994 // space than needed, it could confuse the collision detecting code. To
3995 // solve the problem, save current _highest_vm_reserved_address and
3996 // calculate the correct value before return.
3997 address old_highest = _highest_vm_reserved_address;
3998
3999 // Linux mmap allows caller to pass an address as hint; give it a try first,
4000 // if kernel honors the hint then we can return immediately.
4001 char * addr = anon_mmap(requested_addr, bytes, false);
4002 if (addr == requested_addr) {
4003 return requested_addr;
4004 }
4005
4006 if (addr != NULL) {
4007 // mmap() is successful but it fails to reserve at the requested address
4008 anon_munmap(addr, bytes);
4009 }
4010
4011 int i;
4012 for (i = 0; i < max_tries; ++i) {
4013 base[i] = reserve_memory(bytes);
4014
4015 if (base[i] != NULL) {
4016 // Is this the block we wanted?
4017 if (base[i] == requested_addr) {
4018 size[i] = bytes;
4019 break;
4020 }
4021
4022 // Does this overlap the block we wanted? Give back the overlapped
4023 // parts and try again.
4024
4025 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4026 if (top_overlap >= 0 && top_overlap < bytes) {
4027 unmap_memory(base[i], top_overlap);
4028 base[i] += top_overlap;
4029 size[i] = bytes - top_overlap;
4030 } else {
4031 size_t bottom_overlap = base[i] + bytes - requested_addr;
4032 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4033 unmap_memory(requested_addr, bottom_overlap);
4034 size[i] = bytes - bottom_overlap;
4035 } else {
4036 size[i] = bytes;
4037 }
4038 }
4039 }
4040 }
4041
4042 // Give back the unused reserved pieces.
4043
4044 for (int j = 0; j < i; ++j) {
4045 if (base[j] != NULL) {
4046 unmap_memory(base[j], size[j]);
4047 }
4048 }
4049
4050 if (i < max_tries) {
4051 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4052 return requested_addr;
4053 } else {
4054 _highest_vm_reserved_address = old_highest;
4055 return NULL;
4056 }
4057 }
4058
4059 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4060 return ::read(fd, buf, nBytes);
4061 }
4062
4063 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4064 return ::pread(fd, buf, nBytes, offset);
4065 }
4066
4067 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4068 // Solaris uses poll(), linux uses park().
4069 // Poll() is likely a better choice, assuming that Thread.interrupt()
4070 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4071 // SIGSEGV, see 4355769.
4072
4073 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4074 assert(thread == Thread::current(), "thread consistency check");
4075
4076 ParkEvent * const slp = thread->_SleepEvent ;
4077 slp->reset() ;
4078 OrderAccess::fence() ;
4079
4080 if (interruptible) {
4081 jlong prevtime = javaTimeNanos();
4082
4083 for (;;) {
4084 if (os::is_interrupted(thread, true)) {
4085 return OS_INTRPT;
4086 }
4087
4088 jlong newtime = javaTimeNanos();
4089
4090 if (newtime - prevtime < 0) {
4091 // time moving backwards, should only happen if no monotonic clock
4092 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4093 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4094 } else {
4095 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4096 }
4097
4098 if(millis <= 0) {
4099 return OS_OK;
4100 }
4101
4102 prevtime = newtime;
4103
4104 {
4105 assert(thread->is_Java_thread(), "sanity check");
4106 JavaThread *jt = (JavaThread *) thread;
4107 ThreadBlockInVM tbivm(jt);
4108 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4109
4110 jt->set_suspend_equivalent();
4111 // cleared by handle_special_suspend_equivalent_condition() or
4112 // java_suspend_self() via check_and_wait_while_suspended()
4113
4114 slp->park(millis);
4115
4116 // were we externally suspended while we were waiting?
4117 jt->check_and_wait_while_suspended();
4118 }
4119 }
4120 } else {
4121 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4122 jlong prevtime = javaTimeNanos();
4123
4124 for (;;) {
4125 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4126 // the 1st iteration ...
4127 jlong newtime = javaTimeNanos();
4128
4129 if (newtime - prevtime < 0) {
4130 // time moving backwards, should only happen if no monotonic clock
4131 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4132 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4133 } else {
4134 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4135 }
4136
4137 if(millis <= 0) break ;
4138
4139 prevtime = newtime;
4140 slp->park(millis);
4141 }
4142 return OS_OK ;
4143 }
4144 }
4145
4146 //
4147 // Short sleep, direct OS call.
4148 //
4149 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4150 // sched_yield(2) will actually give up the CPU:
4151 //
4152 // * Alone on this pariticular CPU, keeps running.
4153 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4154 // (pre 2.6.39).
4155 //
4156 // So calling this with 0 is an alternative.
4157 //
4158 void os::naked_short_sleep(jlong ms) {
4159 struct timespec req;
4160
4161 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4162 req.tv_sec = 0;
4163 if (ms > 0) {
4164 req.tv_nsec = (ms % 1000) * 1000000;
4165 }
4166 else {
4167 req.tv_nsec = 1;
4168 }
4169
4170 nanosleep(&req, NULL);
4171
4172 return;
4173 }
4174
4175 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4176 void os::infinite_sleep() {
4177 while (true) { // sleep forever ...
4178 ::sleep(100); // ... 100 seconds at a time
4179 }
4180 }
4181
4182 // Used to convert frequent JVM_Yield() to nops
4183 bool os::dont_yield() {
4184 return DontYieldALot;
4185 }
4186
4187 void os::yield() {
4188 sched_yield();
4189 }
4190
4191 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4192
4193 void os::yield_all(int attempts) {
4194 // Yields to all threads, including threads with lower priorities
4195 // Threads on Linux are all with same priority. The Solaris style
4196 // os::yield_all() with nanosleep(1ms) is not necessary.
4197 sched_yield();
4198 }
4199
4200 // Called from the tight loops to possibly influence time-sharing heuristics
4201 void os::loop_breaker(int attempts) {
4202 os::yield_all(attempts);
4203 }
4204
4205 ////////////////////////////////////////////////////////////////////////////////
4206 // thread priority support
4207
4208 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4209 // only supports dynamic priority, static priority must be zero. For real-time
4210 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4211 // However, for large multi-threaded applications, SCHED_RR is not only slower
4212 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4213 // of 5 runs - Sep 2005).
4214 //
4215 // The following code actually changes the niceness of kernel-thread/LWP. It
4216 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4217 // not the entire user process, and user level threads are 1:1 mapped to kernel
4218 // threads. It has always been the case, but could change in the future. For
4219 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4220 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4221
4222 int os::java_to_os_priority[CriticalPriority + 1] = {
4223 19, // 0 Entry should never be used
4224
4225 4, // 1 MinPriority
4226 3, // 2
4227 2, // 3
4228
4229 1, // 4
4230 0, // 5 NormPriority
4231 -1, // 6
4232
4233 -2, // 7
4234 -3, // 8
4235 -4, // 9 NearMaxPriority
4236
4237 -5, // 10 MaxPriority
4238
4239 -5 // 11 CriticalPriority
4240 };
4241
4242 static int prio_init() {
4243 if (ThreadPriorityPolicy == 1) {
4244 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4245 // if effective uid is not root. Perhaps, a more elegant way of doing
4246 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4247 if (geteuid() != 0) {
4248 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4249 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4250 }
4251 ThreadPriorityPolicy = 0;
4252 }
4253 }
4254 if (UseCriticalJavaThreadPriority) {
4255 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4256 }
4257 return 0;
4258 }
4259
4260 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4261 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4262
4263 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4264 return (ret == 0) ? OS_OK : OS_ERR;
4265 }
4266
4267 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4268 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4269 *priority_ptr = java_to_os_priority[NormPriority];
4270 return OS_OK;
4271 }
4272
4273 errno = 0;
4274 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4275 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4276 }
4277
4278 // Hint to the underlying OS that a task switch would not be good.
4279 // Void return because it's a hint and can fail.
4280 void os::hint_no_preempt() {}
4281
4282 ////////////////////////////////////////////////////////////////////////////////
4283 // suspend/resume support
4284
4285 // the low-level signal-based suspend/resume support is a remnant from the
4286 // old VM-suspension that used to be for java-suspension, safepoints etc,
4287 // within hotspot. Now there is a single use-case for this:
4288 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4289 // that runs in the watcher thread.
4290 // The remaining code is greatly simplified from the more general suspension
4291 // code that used to be used.
4292 //
4293 // The protocol is quite simple:
4294 // - suspend:
4295 // - sends a signal to the target thread
4296 // - polls the suspend state of the osthread using a yield loop
4297 // - target thread signal handler (SR_handler) sets suspend state
4298 // and blocks in sigsuspend until continued
4299 // - resume:
4300 // - sets target osthread state to continue
4301 // - sends signal to end the sigsuspend loop in the SR_handler
4302 //
4303 // Note that the SR_lock plays no role in this suspend/resume protocol.
4304 //
4305
4306 static void resume_clear_context(OSThread *osthread) {
4307 osthread->set_ucontext(NULL);
4308 osthread->set_siginfo(NULL);
4309 }
4310
4311 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4312 osthread->set_ucontext(context);
4313 osthread->set_siginfo(siginfo);
4314 }
4315
4316 //
4317 // Handler function invoked when a thread's execution is suspended or
4318 // resumed. We have to be careful that only async-safe functions are
4319 // called here (Note: most pthread functions are not async safe and
4320 // should be avoided.)
4321 //
4322 // Note: sigwait() is a more natural fit than sigsuspend() from an
4323 // interface point of view, but sigwait() prevents the signal hander
4324 // from being run. libpthread would get very confused by not having
4325 // its signal handlers run and prevents sigwait()'s use with the
4326 // mutex granting granting signal.
4327 //
4328 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4329 //
4330 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4331 // Save and restore errno to avoid confusing native code with EINTR
4332 // after sigsuspend.
4333 int old_errno = errno;
4334
4335 Thread* thread = Thread::current();
4336 OSThread* osthread = thread->osthread();
4337 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4338
4339 os::SuspendResume::State current = osthread->sr.state();
4340 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4341 suspend_save_context(osthread, siginfo, context);
4342
4343 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4344 os::SuspendResume::State state = osthread->sr.suspended();
4345 if (state == os::SuspendResume::SR_SUSPENDED) {
4346 sigset_t suspend_set; // signals for sigsuspend()
4347
4348 // get current set of blocked signals and unblock resume signal
4349 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4350 sigdelset(&suspend_set, SR_signum);
4351
4352 sr_semaphore.signal();
4353 // wait here until we are resumed
4354 while (1) {
4355 sigsuspend(&suspend_set);
4356
4357 os::SuspendResume::State result = osthread->sr.running();
4358 if (result == os::SuspendResume::SR_RUNNING) {
4359 sr_semaphore.signal();
4360 break;
4361 }
4362 }
4363
4364 } else if (state == os::SuspendResume::SR_RUNNING) {
4365 // request was cancelled, continue
4366 } else {
4367 ShouldNotReachHere();
4368 }
4369
4370 resume_clear_context(osthread);
4371 } else if (current == os::SuspendResume::SR_RUNNING) {
4372 // request was cancelled, continue
4373 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4374 // ignore
4375 } else {
4376 // ignore
4377 }
4378
4379 errno = old_errno;
4380 }
4381
4382
4383 static int SR_initialize() {
4384 struct sigaction act;
4385 char *s;
4386 /* Get signal number to use for suspend/resume */
4387 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4388 int sig = ::strtol(s, 0, 10);
4389 if (sig > 0 || sig < _NSIG) {
4390 SR_signum = sig;
4391 }
4392 }
4393
4394 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4395 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4396
4397 sigemptyset(&SR_sigset);
4398 sigaddset(&SR_sigset, SR_signum);
4399
4400 /* Set up signal handler for suspend/resume */
4401 act.sa_flags = SA_RESTART|SA_SIGINFO;
4402 act.sa_handler = (void (*)(int)) SR_handler;
4403
4404 // SR_signum is blocked by default.
4405 // 4528190 - We also need to block pthread restart signal (32 on all
4406 // supported Linux platforms). Note that LinuxThreads need to block
4407 // this signal for all threads to work properly. So we don't have
4408 // to use hard-coded signal number when setting up the mask.
4409 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4410
4411 if (sigaction(SR_signum, &act, 0) == -1) {
4412 return -1;
4413 }
4414
4415 // Save signal flag
4416 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4417 return 0;
4418 }
4419
4420 static int sr_notify(OSThread* osthread) {
4421 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4422 assert_status(status == 0, status, "pthread_kill");
4423 return status;
4424 }
4425
4426 // "Randomly" selected value for how long we want to spin
4427 // before bailing out on suspending a thread, also how often
4428 // we send a signal to a thread we want to resume
4429 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4430 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4431
4432 // returns true on success and false on error - really an error is fatal
4433 // but this seems the normal response to library errors
4434 static bool do_suspend(OSThread* osthread) {
4435 assert(osthread->sr.is_running(), "thread should be running");
4436 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4437
4438 // mark as suspended and send signal
4439 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4440 // failed to switch, state wasn't running?
4441 ShouldNotReachHere();
4442 return false;
4443 }
4444
4445 if (sr_notify(osthread) != 0) {
4446 ShouldNotReachHere();
4447 }
4448
4449 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4450 while (true) {
4451 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4452 break;
4453 } else {
4454 // timeout
4455 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4456 if (cancelled == os::SuspendResume::SR_RUNNING) {
4457 return false;
4458 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4459 // make sure that we consume the signal on the semaphore as well
4460 sr_semaphore.wait();
4461 break;
4462 } else {
4463 ShouldNotReachHere();
4464 return false;
4465 }
4466 }
4467 }
4468
4469 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4470 return true;
4471 }
4472
4473 static void do_resume(OSThread* osthread) {
4474 assert(osthread->sr.is_suspended(), "thread should be suspended");
4475 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4476
4477 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4478 // failed to switch to WAKEUP_REQUEST
4479 ShouldNotReachHere();
4480 return;
4481 }
4482
4483 while (true) {
4484 if (sr_notify(osthread) == 0) {
4485 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4486 if (osthread->sr.is_running()) {
4487 return;
4488 }
4489 }
4490 } else {
4491 ShouldNotReachHere();
4492 }
4493 }
4494
4495 guarantee(osthread->sr.is_running(), "Must be running!");
4496 }
4497
4498 ////////////////////////////////////////////////////////////////////////////////
4499 // interrupt support
4500
4501 void os::interrupt(Thread* thread) {
4502 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4503 "possibility of dangling Thread pointer");
4504
4505 OSThread* osthread = thread->osthread();
4506
4507 if (!osthread->interrupted()) {
4508 osthread->set_interrupted(true);
4509 // More than one thread can get here with the same value of osthread,
4510 // resulting in multiple notifications. We do, however, want the store
4511 // to interrupted() to be visible to other threads before we execute unpark().
4512 OrderAccess::fence();
4513 ParkEvent * const slp = thread->_SleepEvent ;
4514 if (slp != NULL) slp->unpark() ;
4515 }
4516
4517 // For JSR166. Unpark even if interrupt status already was set
4518 if (thread->is_Java_thread())
4519 ((JavaThread*)thread)->parker()->unpark();
4520
4521 ParkEvent * ev = thread->_ParkEvent ;
4522 if (ev != NULL) ev->unpark() ;
4523
4524 }
4525
4526 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4527 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4528 "possibility of dangling Thread pointer");
4529
4530 OSThread* osthread = thread->osthread();
4531
4532 bool interrupted = osthread->interrupted();
4533
4534 if (interrupted && clear_interrupted) {
4535 osthread->set_interrupted(false);
4536 // consider thread->_SleepEvent->reset() ... optional optimization
4537 }
4538
4539 return interrupted;
4540 }
4541
4542 ///////////////////////////////////////////////////////////////////////////////////
4543 // signal handling (except suspend/resume)
4544
4545 // This routine may be used by user applications as a "hook" to catch signals.
4546 // The user-defined signal handler must pass unrecognized signals to this
4547 // routine, and if it returns true (non-zero), then the signal handler must
4548 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4549 // routine will never retun false (zero), but instead will execute a VM panic
4550 // routine kill the process.
4551 //
4552 // If this routine returns false, it is OK to call it again. This allows
4553 // the user-defined signal handler to perform checks either before or after
4554 // the VM performs its own checks. Naturally, the user code would be making
4555 // a serious error if it tried to handle an exception (such as a null check
4556 // or breakpoint) that the VM was generating for its own correct operation.
4557 //
4558 // This routine may recognize any of the following kinds of signals:
4559 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4560 // It should be consulted by handlers for any of those signals.
4561 //
4562 // The caller of this routine must pass in the three arguments supplied
4563 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4564 // field of the structure passed to sigaction(). This routine assumes that
4565 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4566 //
4567 // Note that the VM will print warnings if it detects conflicting signal
4568 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4569 //
4570 extern "C" JNIEXPORT int
4571 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4572 void* ucontext, int abort_if_unrecognized);
4573
4574 void signalHandler(int sig, siginfo_t* info, void* uc) {
4575 assert(info != NULL && uc != NULL, "it must be old kernel");
4576 int orig_errno = errno; // Preserve errno value over signal handler.
4577 JVM_handle_linux_signal(sig, info, uc, true);
4578 errno = orig_errno;
4579 }
4580
4581
4582 // This boolean allows users to forward their own non-matching signals
4583 // to JVM_handle_linux_signal, harmlessly.
4584 bool os::Linux::signal_handlers_are_installed = false;
4585
4586 // For signal-chaining
4587 struct sigaction os::Linux::sigact[MAXSIGNUM];
4588 unsigned int os::Linux::sigs = 0;
4589 bool os::Linux::libjsig_is_loaded = false;
4590 typedef struct sigaction *(*get_signal_t)(int);
4591 get_signal_t os::Linux::get_signal_action = NULL;
4592
4593 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4594 struct sigaction *actp = NULL;
4595
4596 if (libjsig_is_loaded) {
4597 // Retrieve the old signal handler from libjsig
4598 actp = (*get_signal_action)(sig);
4599 }
4600 if (actp == NULL) {
4601 // Retrieve the preinstalled signal handler from jvm
4602 actp = get_preinstalled_handler(sig);
4603 }
4604
4605 return actp;
4606 }
4607
4608 static bool call_chained_handler(struct sigaction *actp, int sig,
4609 siginfo_t *siginfo, void *context) {
4610 // Call the old signal handler
4611 if (actp->sa_handler == SIG_DFL) {
4612 // It's more reasonable to let jvm treat it as an unexpected exception
4613 // instead of taking the default action.
4614 return false;
4615 } else if (actp->sa_handler != SIG_IGN) {
4616 if ((actp->sa_flags & SA_NODEFER) == 0) {
4617 // automaticlly block the signal
4618 sigaddset(&(actp->sa_mask), sig);
4619 }
4620
4621 sa_handler_t hand = NULL;
4622 sa_sigaction_t sa = NULL;
4623 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4624 // retrieve the chained handler
4625 if (siginfo_flag_set) {
4626 sa = actp->sa_sigaction;
4627 } else {
4628 hand = actp->sa_handler;
4629 }
4630
4631 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4632 actp->sa_handler = SIG_DFL;
4633 }
4634
4635 // try to honor the signal mask
4636 sigset_t oset;
4637 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4638
4639 // call into the chained handler
4640 if (siginfo_flag_set) {
4641 (*sa)(sig, siginfo, context);
4642 } else {
4643 (*hand)(sig);
4644 }
4645
4646 // restore the signal mask
4647 pthread_sigmask(SIG_SETMASK, &oset, 0);
4648 }
4649 // Tell jvm's signal handler the signal is taken care of.
4650 return true;
4651 }
4652
4653 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4654 bool chained = false;
4655 // signal-chaining
4656 if (UseSignalChaining) {
4657 struct sigaction *actp = get_chained_signal_action(sig);
4658 if (actp != NULL) {
4659 chained = call_chained_handler(actp, sig, siginfo, context);
4660 }
4661 }
4662 return chained;
4663 }
4664
4665 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4666 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4667 return &sigact[sig];
4668 }
4669 return NULL;
4670 }
4671
4672 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4673 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4674 sigact[sig] = oldAct;
4675 sigs |= (unsigned int)1 << sig;
4676 }
4677
4678 // for diagnostic
4679 int os::Linux::sigflags[MAXSIGNUM];
4680
4681 int os::Linux::get_our_sigflags(int sig) {
4682 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4683 return sigflags[sig];
4684 }
4685
4686 void os::Linux::set_our_sigflags(int sig, int flags) {
4687 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4688 sigflags[sig] = flags;
4689 }
4690
4691 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4692 // Check for overwrite.
4693 struct sigaction oldAct;
4694 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4695
4696 void* oldhand = oldAct.sa_sigaction
4697 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4698 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4699 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4700 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4701 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4702 if (AllowUserSignalHandlers || !set_installed) {
4703 // Do not overwrite; user takes responsibility to forward to us.
4704 return;
4705 } else if (UseSignalChaining) {
4706 // save the old handler in jvm
4707 save_preinstalled_handler(sig, oldAct);
4708 // libjsig also interposes the sigaction() call below and saves the
4709 // old sigaction on it own.
4710 } else {
4711 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4712 "%#lx for signal %d.", (long)oldhand, sig));
4713 }
4714 }
4715
4716 struct sigaction sigAct;
4717 sigfillset(&(sigAct.sa_mask));
4718 sigAct.sa_handler = SIG_DFL;
4719 if (!set_installed) {
4720 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4721 } else {
4722 sigAct.sa_sigaction = signalHandler;
4723 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4724 }
4725 // Save flags, which are set by ours
4726 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4727 sigflags[sig] = sigAct.sa_flags;
4728
4729 int ret = sigaction(sig, &sigAct, &oldAct);
4730 assert(ret == 0, "check");
4731
4732 void* oldhand2 = oldAct.sa_sigaction
4733 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4734 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4735 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4736 }
4737
4738 // install signal handlers for signals that HotSpot needs to
4739 // handle in order to support Java-level exception handling.
4740
4741 void os::Linux::install_signal_handlers() {
4742 if (!signal_handlers_are_installed) {
4743 signal_handlers_are_installed = true;
4744
4745 // signal-chaining
4746 typedef void (*signal_setting_t)();
4747 signal_setting_t begin_signal_setting = NULL;
4748 signal_setting_t end_signal_setting = NULL;
4749 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4750 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4751 if (begin_signal_setting != NULL) {
4752 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4753 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4754 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4755 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4756 libjsig_is_loaded = true;
4757 assert(UseSignalChaining, "should enable signal-chaining");
4758 }
4759 if (libjsig_is_loaded) {
4760 // Tell libjsig jvm is setting signal handlers
4761 (*begin_signal_setting)();
4762 }
4763
4764 set_signal_handler(SIGSEGV, true);
4765 set_signal_handler(SIGPIPE, true);
4766 set_signal_handler(SIGBUS, true);
4767 set_signal_handler(SIGILL, true);
4768 set_signal_handler(SIGFPE, true);
4769 #if defined(PPC64)
4770 set_signal_handler(SIGTRAP, true);
4771 #endif
4772 set_signal_handler(SIGXFSZ, true);
4773
4774 if (libjsig_is_loaded) {
4775 // Tell libjsig jvm finishes setting signal handlers
4776 (*end_signal_setting)();
4777 }
4778
4779 // We don't activate signal checker if libjsig is in place, we trust ourselves
4780 // and if UserSignalHandler is installed all bets are off.
4781 // Log that signal checking is off only if -verbose:jni is specified.
4782 if (CheckJNICalls) {
4783 if (libjsig_is_loaded) {
4784 if (PrintJNIResolving) {
4785 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4786 }
4787 check_signals = false;
4788 }
4789 if (AllowUserSignalHandlers) {
4790 if (PrintJNIResolving) {
4791 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4792 }
4793 check_signals = false;
4794 }
4795 }
4796 }
4797 }
4798
4799 // This is the fastest way to get thread cpu time on Linux.
4800 // Returns cpu time (user+sys) for any thread, not only for current.
4801 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4802 // It might work on 2.6.10+ with a special kernel/glibc patch.
4803 // For reference, please, see IEEE Std 1003.1-2004:
4804 // http://www.unix.org/single_unix_specification
4805
4806 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4807 struct timespec tp;
4808 int rc = os::Linux::clock_gettime(clockid, &tp);
4809 assert(rc == 0, "clock_gettime is expected to return 0 code");
4810
4811 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4812 }
4813
4814 /////
4815 // glibc on Linux platform uses non-documented flag
4816 // to indicate, that some special sort of signal
4817 // trampoline is used.
4818 // We will never set this flag, and we should
4819 // ignore this flag in our diagnostic
4820 #ifdef SIGNIFICANT_SIGNAL_MASK
4821 #undef SIGNIFICANT_SIGNAL_MASK
4822 #endif
4823 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4824
4825 static const char* get_signal_handler_name(address handler,
4826 char* buf, int buflen) {
4827 int offset = 0;
4828 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4829 if (found) {
4830 // skip directory names
4831 const char *p1, *p2;
4832 p1 = buf;
4833 size_t len = strlen(os::file_separator());
4834 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4835 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4836 } else {
4837 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4838 }
4839 return buf;
4840 }
4841
4842 static void print_signal_handler(outputStream* st, int sig,
4843 char* buf, size_t buflen) {
4844 struct sigaction sa;
4845
4846 sigaction(sig, NULL, &sa);
4847
4848 // See comment for SIGNIFICANT_SIGNAL_MASK define
4849 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4850
4851 st->print("%s: ", os::exception_name(sig, buf, buflen));
4852
4853 address handler = (sa.sa_flags & SA_SIGINFO)
4854 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4855 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4856
4857 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4858 st->print("SIG_DFL");
4859 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4860 st->print("SIG_IGN");
4861 } else {
4862 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4863 }
4864
4865 st->print(", sa_mask[0]=");
4866 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4867
4868 address rh = VMError::get_resetted_sighandler(sig);
4869 // May be, handler was resetted by VMError?
4870 if(rh != NULL) {
4871 handler = rh;
4872 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4873 }
4874
4875 st->print(", sa_flags=");
4876 os::Posix::print_sa_flags(st, sa.sa_flags);
4877
4878 // Check: is it our handler?
4879 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4880 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4881 // It is our signal handler
4882 // check for flags, reset system-used one!
4883 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4884 st->print(
4885 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4886 os::Linux::get_our_sigflags(sig));
4887 }
4888 }
4889 st->cr();
4890 }
4891
4892
4893 #define DO_SIGNAL_CHECK(sig) \
4894 if (!sigismember(&check_signal_done, sig)) \
4895 os::Linux::check_signal_handler(sig)
4896
4897 // This method is a periodic task to check for misbehaving JNI applications
4898 // under CheckJNI, we can add any periodic checks here
4899
4900 void os::run_periodic_checks() {
4901
4902 if (check_signals == false) return;
4903
4904 // SEGV and BUS if overridden could potentially prevent
4905 // generation of hs*.log in the event of a crash, debugging
4906 // such a case can be very challenging, so we absolutely
4907 // check the following for a good measure:
4908 DO_SIGNAL_CHECK(SIGSEGV);
4909 DO_SIGNAL_CHECK(SIGILL);
4910 DO_SIGNAL_CHECK(SIGFPE);
4911 DO_SIGNAL_CHECK(SIGBUS);
4912 DO_SIGNAL_CHECK(SIGPIPE);
4913 DO_SIGNAL_CHECK(SIGXFSZ);
4914 #if defined(PPC64)
4915 DO_SIGNAL_CHECK(SIGTRAP);
4916 #endif
4917
4918 // ReduceSignalUsage allows the user to override these handlers
4919 // see comments at the very top and jvm_solaris.h
4920 if (!ReduceSignalUsage) {
4921 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4922 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4923 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4924 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4925 }
4926
4927 DO_SIGNAL_CHECK(SR_signum);
4928 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4929 }
4930
4931 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4932
4933 static os_sigaction_t os_sigaction = NULL;
4934
4935 void os::Linux::check_signal_handler(int sig) {
4936 char buf[O_BUFLEN];
4937 address jvmHandler = NULL;
4938
4939
4940 struct sigaction act;
4941 if (os_sigaction == NULL) {
4942 // only trust the default sigaction, in case it has been interposed
4943 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4944 if (os_sigaction == NULL) return;
4945 }
4946
4947 os_sigaction(sig, (struct sigaction*)NULL, &act);
4948
4949
4950 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4951
4952 address thisHandler = (act.sa_flags & SA_SIGINFO)
4953 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4954 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4955
4956
4957 switch(sig) {
4958 case SIGSEGV:
4959 case SIGBUS:
4960 case SIGFPE:
4961 case SIGPIPE:
4962 case SIGILL:
4963 case SIGXFSZ:
4964 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4965 break;
4966
4967 case SHUTDOWN1_SIGNAL:
4968 case SHUTDOWN2_SIGNAL:
4969 case SHUTDOWN3_SIGNAL:
4970 case BREAK_SIGNAL:
4971 jvmHandler = (address)user_handler();
4972 break;
4973
4974 case INTERRUPT_SIGNAL:
4975 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4976 break;
4977
4978 default:
4979 if (sig == SR_signum) {
4980 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4981 } else {
4982 return;
4983 }
4984 break;
4985 }
4986
4987 if (thisHandler != jvmHandler) {
4988 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4989 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4990 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4991 // No need to check this sig any longer
4992 sigaddset(&check_signal_done, sig);
4993 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4994 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4995 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4996 exception_name(sig, buf, O_BUFLEN));
4997 }
4998 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4999 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5000 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
5001 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
5002 // No need to check this sig any longer
5003 sigaddset(&check_signal_done, sig);
5004 }
5005
5006 // Dump all the signal
5007 if (sigismember(&check_signal_done, sig)) {
5008 print_signal_handlers(tty, buf, O_BUFLEN);
5009 }
5010 }
5011
5012 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
5013
5014 extern bool signal_name(int signo, char* buf, size_t len);
5015
5016 const char* os::exception_name(int exception_code, char* buf, size_t size) {
5017 if (0 < exception_code && exception_code <= SIGRTMAX) {
5018 // signal
5019 if (!signal_name(exception_code, buf, size)) {
5020 jio_snprintf(buf, size, "SIG%d", exception_code);
5021 }
5022 return buf;
5023 } else {
5024 return NULL;
5025 }
5026 }
5027
5028 // this is called _before_ most of the global arguments have been parsed
5029 void os::init(void) {
5030 char dummy; /* used to get a guess on initial stack address */
5031
5032 // With LinuxThreads the JavaMain thread pid (primordial thread)
5033 // is different than the pid of the java launcher thread.
5034 // So, on Linux, the launcher thread pid is passed to the VM
5035 // via the sun.java.launcher.pid property.
5036 // Use this property instead of getpid() if it was correctly passed.
5037 // See bug 6351349.
5038 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5039
5040 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5041
5042 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5043
5044 init_random(1234567);
5045
5046 ThreadCritical::initialize();
5047
5048 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5049 if (Linux::page_size() == -1) {
5050 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5051 strerror(errno)));
5052 }
5053 init_page_sizes((size_t) Linux::page_size());
5054
5055 Linux::initialize_system_info();
5056
5057 // _main_thread points to the thread that created/loaded the JVM.
5058 Linux::_main_thread = pthread_self();
5059
5060 Linux::clock_init();
5061 initial_time_count = javaTimeNanos();
5062
5063 // pthread_condattr initialization for monotonic clock
5064 int status;
5065 pthread_condattr_t* _condattr = os::Linux::condAttr();
5066 if ((status = pthread_condattr_init(_condattr)) != 0) {
5067 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5068 }
5069 // Only set the clock if CLOCK_MONOTONIC is available
5070 if (Linux::supports_monotonic_clock()) {
5071 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5072 if (status == EINVAL) {
5073 warning("Unable to use monotonic clock with relative timed-waits" \
5074 " - changes to the time-of-day clock may have adverse affects");
5075 } else {
5076 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5077 }
5078 }
5079 }
5080 // else it defaults to CLOCK_REALTIME
5081
5082 pthread_mutex_init(&dl_mutex, NULL);
5083
5084 // If the pagesize of the VM is greater than 8K determine the appropriate
5085 // number of initial guard pages. The user can change this with the
5086 // command line arguments, if needed.
5087 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5088 StackYellowPages = 1;
5089 StackRedPages = 1;
5090 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5091 }
5092 }
5093
5094 // To install functions for atexit system call
5095 extern "C" {
5096 static void perfMemory_exit_helper() {
5097 perfMemory_exit();
5098 }
5099 }
5100
5101 void os::pd_init_container_support() {
5102 OSContainer::init();
5103 }
5104
5105 // this is called _after_ the global arguments have been parsed
5106 jint os::init_2(void)
5107 {
5108 Linux::fast_thread_clock_init();
5109
5110 // Allocate a single page and mark it as readable for safepoint polling
5111 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5112 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5113
5114 os::set_polling_page( polling_page );
5115
5116 #ifndef PRODUCT
5117 if(Verbose && PrintMiscellaneous)
5118 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5119 #endif
5120
5121 if (!UseMembar) {
5122 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5123 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5124 os::set_memory_serialize_page( mem_serialize_page );
5125
5126 #ifndef PRODUCT
5127 if(Verbose && PrintMiscellaneous)
5128 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5129 #endif
5130 }
5131
5132 // initialize suspend/resume support - must do this before signal_sets_init()
5133 if (SR_initialize() != 0) {
5134 perror("SR_initialize failed");
5135 return JNI_ERR;
5136 }
5137
5138 Linux::signal_sets_init();
5139 Linux::install_signal_handlers();
5140
5141 // Check minimum allowable stack size for thread creation and to initialize
5142 // the java system classes, including StackOverflowError - depends on page
5143 // size. Add a page for compiler2 recursion in main thread.
5144 // Add in 2*BytesPerWord times page size to account for VM stack during
5145 // class initialization depending on 32 or 64 bit VM.
5146 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5147 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5148 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5149
5150 size_t threadStackSizeInBytes = ThreadStackSize * K;
5151 if (threadStackSizeInBytes != 0 &&
5152 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
5153 tty->print_cr("\nThe stack size specified is too small, "
5154 "Specify at least %dk",
5155 os::Linux::min_stack_allowed/ K);
5156 return JNI_ERR;
5157 }
5158
5159 // Make the stack size a multiple of the page size so that
5160 // the yellow/red zones can be guarded.
5161 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5162 vm_page_size()));
5163
5164 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5165
5166 #if defined(IA32)
5167 workaround_expand_exec_shield_cs_limit();
5168 #endif
5169
5170 Linux::libpthread_init();
5171 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5172 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5173 Linux::glibc_version(), Linux::libpthread_version(),
5174 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5175 }
5176
5177 if (UseNUMA) {
5178 if (!Linux::libnuma_init()) {
5179 UseNUMA = false;
5180 } else {
5181 if ((Linux::numa_max_node() < 1)) {
5182 // There's only one node(they start from 0), disable NUMA.
5183 UseNUMA = false;
5184 }
5185 }
5186 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5187 // we can make the adaptive lgrp chunk resizing work. If the user specified
5188 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5189 // disable adaptive resizing.
5190 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5191 if (FLAG_IS_DEFAULT(UseNUMA)) {
5192 UseNUMA = false;
5193 } else {
5194 if (FLAG_IS_DEFAULT(UseLargePages) &&
5195 FLAG_IS_DEFAULT(UseSHM) &&
5196 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5197 UseLargePages = false;
5198 } else {
5199 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5200 UseAdaptiveSizePolicy = false;
5201 UseAdaptiveNUMAChunkSizing = false;
5202 }
5203 }
5204 }
5205 if (!UseNUMA && ForceNUMA) {
5206 UseNUMA = true;
5207 }
5208 }
5209
5210 if (MaxFDLimit) {
5211 // set the number of file descriptors to max. print out error
5212 // if getrlimit/setrlimit fails but continue regardless.
5213 struct rlimit nbr_files;
5214 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5215 if (status != 0) {
5216 if (PrintMiscellaneous && (Verbose || WizardMode))
5217 perror("os::init_2 getrlimit failed");
5218 } else {
5219 nbr_files.rlim_cur = nbr_files.rlim_max;
5220 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5221 if (status != 0) {
5222 if (PrintMiscellaneous && (Verbose || WizardMode))
5223 perror("os::init_2 setrlimit failed");
5224 }
5225 }
5226 }
5227
5228 // Initialize lock used to serialize thread creation (see os::create_thread)
5229 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5230
5231 // at-exit methods are called in the reverse order of their registration.
5232 // atexit functions are called on return from main or as a result of a
5233 // call to exit(3C). There can be only 32 of these functions registered
5234 // and atexit() does not set errno.
5235
5236 if (PerfAllowAtExitRegistration) {
5237 // only register atexit functions if PerfAllowAtExitRegistration is set.
5238 // atexit functions can be delayed until process exit time, which
5239 // can be problematic for embedded VM situations. Embedded VMs should
5240 // call DestroyJavaVM() to assure that VM resources are released.
5241
5242 // note: perfMemory_exit_helper atexit function may be removed in
5243 // the future if the appropriate cleanup code can be added to the
5244 // VM_Exit VMOperation's doit method.
5245 if (atexit(perfMemory_exit_helper) != 0) {
5246 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5247 }
5248 }
5249
5250 // initialize thread priority policy
5251 prio_init();
5252
5253 return JNI_OK;
5254 }
5255
5256 // Mark the polling page as unreadable
5257 void os::make_polling_page_unreadable(void) {
5258 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5259 fatal("Could not disable polling page");
5260 };
5261
5262 // Mark the polling page as readable
5263 void os::make_polling_page_readable(void) {
5264 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5265 fatal("Could not enable polling page");
5266 }
5267 };
5268
5269 static int os_cpu_count(const cpu_set_t* cpus) {
5270 int count = 0;
5271 // only look up to the number of configured processors
5272 for (int i = 0; i < os::processor_count(); i++) {
5273 if (CPU_ISSET(i, cpus)) {
5274 count++;
5275 }
5276 }
5277 return count;
5278 }
5279
5280 // Get the current number of available processors for this process.
5281 // This value can change at any time during a process's lifetime.
5282 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5283 // If anything goes wrong we fallback to returning the number of online
5284 // processors - which can be greater than the number available to the process.
5285 int os::Linux::active_processor_count() {
5286 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5287 int cpus_size = sizeof(cpu_set_t);
5288 int cpu_count = 0;
5289
5290 // pid 0 means the current thread - which we have to assume represents the process
5291 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5292 cpu_count = os_cpu_count(&cpus);
5293 if (PrintActiveCpus) {
5294 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5295 }
5296 }
5297 else {
5298 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5299 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5300 "which may exceed available processors", strerror(errno), cpu_count);
5301 }
5302
5303 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5304 return cpu_count;
5305 }
5306
5307 // Determine the active processor count from one of
5308 // three different sources:
5309 //
5310 // 1. User option -XX:ActiveProcessorCount
5311 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5312 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5313 //
5314 // Option 1, if specified, will always override.
5315 // If the cgroup subsystem is active and configured, we
5316 // will return the min of the cgroup and option 2 results.
5317 // This is required since tools, such as numactl, that
5318 // alter cpu affinity do not update cgroup subsystem
5319 // cpuset configuration files.
5320 int os::active_processor_count() {
5321 // User has overridden the number of active processors
5322 if (ActiveProcessorCount > 0) {
5323 if (PrintActiveCpus) {
5324 tty->print_cr("active_processor_count: "
5325 "active processor count set by user : %d",
5326 ActiveProcessorCount);
5327 }
5328 return ActiveProcessorCount;
5329 }
5330
5331 int active_cpus;
5332 if (OSContainer::is_containerized()) {
5333 active_cpus = OSContainer::active_processor_count();
5334 if (PrintActiveCpus) {
5335 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5336 active_cpus);
5337 }
5338 } else {
5339 active_cpus = os::Linux::active_processor_count();
5340 }
5341
5342 return active_cpus;
5343 }
5344
5345 void os::set_native_thread_name(const char *name) {
5346 // Not yet implemented.
5347 return;
5348 }
5349
5350 bool os::distribute_processes(uint length, uint* distribution) {
5351 // Not yet implemented.
5352 return false;
5353 }
5354
5355 bool os::bind_to_processor(uint processor_id) {
5356 // Not yet implemented.
5357 return false;
5358 }
5359
5360 ///
5361
5362 void os::SuspendedThreadTask::internal_do_task() {
5363 if (do_suspend(_thread->osthread())) {
5364 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5365 do_task(context);
5366 do_resume(_thread->osthread());
5367 }
5368 }
5369
5370 class PcFetcher : public os::SuspendedThreadTask {
5371 public:
5372 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5373 ExtendedPC result();
5374 protected:
5375 void do_task(const os::SuspendedThreadTaskContext& context);
5376 private:
5377 ExtendedPC _epc;
5378 };
5379
5380 ExtendedPC PcFetcher::result() {
5381 guarantee(is_done(), "task is not done yet.");
5382 return _epc;
5383 }
5384
5385 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5386 Thread* thread = context.thread();
5387 OSThread* osthread = thread->osthread();
5388 if (osthread->ucontext() != NULL) {
5389 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5390 } else {
5391 // NULL context is unexpected, double-check this is the VMThread
5392 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5393 }
5394 }
5395
5396 // Suspends the target using the signal mechanism and then grabs the PC before
5397 // resuming the target. Used by the flat-profiler only
5398 ExtendedPC os::get_thread_pc(Thread* thread) {
5399 // Make sure that it is called by the watcher for the VMThread
5400 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5401 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5402
5403 PcFetcher fetcher(thread);
5404 fetcher.run();
5405 return fetcher.result();
5406 }
5407
5408 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5409 {
5410 if (is_NPTL()) {
5411 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5412 } else {
5413 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5414 // word back to default 64bit precision if condvar is signaled. Java
5415 // wants 53bit precision. Save and restore current value.
5416 int fpu = get_fpu_control_word();
5417 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5418 set_fpu_control_word(fpu);
5419 return status;
5420 }
5421 }
5422
5423 ////////////////////////////////////////////////////////////////////////////////
5424 // debug support
5425
5426 bool os::find(address addr, outputStream* st) {
5427 Dl_info dlinfo;
5428 memset(&dlinfo, 0, sizeof(dlinfo));
5429 if (dladdr(addr, &dlinfo) != 0) {
5430 st->print(PTR_FORMAT ": ", addr);
5431 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5432 st->print("%s+%#x", dlinfo.dli_sname,
5433 addr - (intptr_t)dlinfo.dli_saddr);
5434 } else if (dlinfo.dli_fbase != NULL) {
5435 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5436 } else {
5437 st->print("<absolute address>");
5438 }
5439 if (dlinfo.dli_fname != NULL) {
5440 st->print(" in %s", dlinfo.dli_fname);
5441 }
5442 if (dlinfo.dli_fbase != NULL) {
5443 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5444 }
5445 st->cr();
5446
5447 if (Verbose) {
5448 // decode some bytes around the PC
5449 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5450 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5451 address lowest = (address) dlinfo.dli_sname;
5452 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5453 if (begin < lowest) begin = lowest;
5454 Dl_info dlinfo2;
5455 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5456 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5457 end = (address) dlinfo2.dli_saddr;
5458 Disassembler::decode(begin, end, st);
5459 }
5460 return true;
5461 }
5462 return false;
5463 }
5464
5465 ////////////////////////////////////////////////////////////////////////////////
5466 // misc
5467
5468 // This does not do anything on Linux. This is basically a hook for being
5469 // able to use structured exception handling (thread-local exception filters)
5470 // on, e.g., Win32.
5471 void
5472 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5473 JavaCallArguments* args, Thread* thread) {
5474 f(value, method, args, thread);
5475 }
5476
5477 void os::print_statistics() {
5478 }
5479
5480 int os::message_box(const char* title, const char* message) {
5481 int i;
5482 fdStream err(defaultStream::error_fd());
5483 for (i = 0; i < 78; i++) err.print_raw("=");
5484 err.cr();
5485 err.print_raw_cr(title);
5486 for (i = 0; i < 78; i++) err.print_raw("-");
5487 err.cr();
5488 err.print_raw_cr(message);
5489 for (i = 0; i < 78; i++) err.print_raw("=");
5490 err.cr();
5491
5492 char buf[16];
5493 // Prevent process from exiting upon "read error" without consuming all CPU
5494 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5495
5496 return buf[0] == 'y' || buf[0] == 'Y';
5497 }
5498
5499 int os::stat(const char *path, struct stat *sbuf) {
5500 char pathbuf[MAX_PATH];
5501 if (strlen(path) > MAX_PATH - 1) {
5502 errno = ENAMETOOLONG;
5503 return -1;
5504 }
5505 os::native_path(strcpy(pathbuf, path));
5506 return ::stat(pathbuf, sbuf);
5507 }
5508
5509 bool os::check_heap(bool force) {
5510 return true;
5511 }
5512
5513 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5514 return ::vsnprintf(buf, count, format, args);
5515 }
5516
5517 // Is a (classpath) directory empty?
5518 bool os::dir_is_empty(const char* path) {
5519 DIR *dir = NULL;
5520 struct dirent *ptr;
5521
5522 dir = opendir(path);
5523 if (dir == NULL) return true;
5524
5525 /* Scan the directory */
5526 bool result = true;
5527 char buf[sizeof(struct dirent) + MAX_PATH];
5528 while (result && (ptr = ::readdir(dir)) != NULL) {
5529 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5530 result = false;
5531 }
5532 }
5533 closedir(dir);
5534 return result;
5535 }
5536
5537 // This code originates from JDK's sysOpen and open64_w
5538 // from src/solaris/hpi/src/system_md.c
5539
5540 #ifndef O_DELETE
5541 #define O_DELETE 0x10000
5542 #endif
5543
5544 // Open a file. Unlink the file immediately after open returns
5545 // if the specified oflag has the O_DELETE flag set.
5546 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5547
5548 int os::open(const char *path, int oflag, int mode) {
5549
5550 if (strlen(path) > MAX_PATH - 1) {
5551 errno = ENAMETOOLONG;
5552 return -1;
5553 }
5554 int fd;
5555 int o_delete = (oflag & O_DELETE);
5556 oflag = oflag & ~O_DELETE;
5557
5558 fd = ::open64(path, oflag, mode);
5559 if (fd == -1) return -1;
5560
5561 //If the open succeeded, the file might still be a directory
5562 {
5563 struct stat64 buf64;
5564 int ret = ::fstat64(fd, &buf64);
5565 int st_mode = buf64.st_mode;
5566
5567 if (ret != -1) {
5568 if ((st_mode & S_IFMT) == S_IFDIR) {
5569 errno = EISDIR;
5570 ::close(fd);
5571 return -1;
5572 }
5573 } else {
5574 ::close(fd);
5575 return -1;
5576 }
5577 }
5578
5579 /*
5580 * All file descriptors that are opened in the JVM and not
5581 * specifically destined for a subprocess should have the
5582 * close-on-exec flag set. If we don't set it, then careless 3rd
5583 * party native code might fork and exec without closing all
5584 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5585 * UNIXProcess.c), and this in turn might:
5586 *
5587 * - cause end-of-file to fail to be detected on some file
5588 * descriptors, resulting in mysterious hangs, or
5589 *
5590 * - might cause an fopen in the subprocess to fail on a system
5591 * suffering from bug 1085341.
5592 *
5593 * (Yes, the default setting of the close-on-exec flag is a Unix
5594 * design flaw)
5595 *
5596 * See:
5597 * 1085341: 32-bit stdio routines should support file descriptors >255
5598 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5599 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5600 */
5601 #ifdef FD_CLOEXEC
5602 {
5603 int flags = ::fcntl(fd, F_GETFD);
5604 if (flags != -1)
5605 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5606 }
5607 #endif
5608
5609 if (o_delete != 0) {
5610 ::unlink(path);
5611 }
5612 return fd;
5613 }
5614
5615
5616 // create binary file, rewriting existing file if required
5617 int os::create_binary_file(const char* path, bool rewrite_existing) {
5618 int oflags = O_WRONLY | O_CREAT;
5619 if (!rewrite_existing) {
5620 oflags |= O_EXCL;
5621 }
5622 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5623 }
5624
5625 // return current position of file pointer
5626 jlong os::current_file_offset(int fd) {
5627 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5628 }
5629
5630 // move file pointer to the specified offset
5631 jlong os::seek_to_file_offset(int fd, jlong offset) {
5632 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5633 }
5634
5635 // This code originates from JDK's sysAvailable
5636 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5637
5638 int os::available(int fd, jlong *bytes) {
5639 jlong cur, end;
5640 int mode;
5641 struct stat64 buf64;
5642
5643 if (::fstat64(fd, &buf64) >= 0) {
5644 mode = buf64.st_mode;
5645 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5646 /*
5647 * XXX: is the following call interruptible? If so, this might
5648 * need to go through the INTERRUPT_IO() wrapper as for other
5649 * blocking, interruptible calls in this file.
5650 */
5651 int n;
5652 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5653 *bytes = n;
5654 return 1;
5655 }
5656 }
5657 }
5658 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5659 return 0;
5660 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5661 return 0;
5662 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5663 return 0;
5664 }
5665 *bytes = end - cur;
5666 return 1;
5667 }
5668
5669 int os::socket_available(int fd, jint *pbytes) {
5670 // Linux doc says EINTR not returned, unlike Solaris
5671 int ret = ::ioctl(fd, FIONREAD, pbytes);
5672
5673 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5674 // is expected to return 0 on failure and 1 on success to the jdk.
5675 return (ret < 0) ? 0 : 1;
5676 }
5677
5678 // Map a block of memory.
5679 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5680 char *addr, size_t bytes, bool read_only,
5681 bool allow_exec) {
5682 int prot;
5683 int flags = MAP_PRIVATE;
5684
5685 if (read_only) {
5686 prot = PROT_READ;
5687 } else {
5688 prot = PROT_READ | PROT_WRITE;
5689 }
5690
5691 if (allow_exec) {
5692 prot |= PROT_EXEC;
5693 }
5694
5695 if (addr != NULL) {
5696 flags |= MAP_FIXED;
5697 }
5698
5699 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5700 fd, file_offset);
5701 if (mapped_address == MAP_FAILED) {
5702 return NULL;
5703 }
5704 return mapped_address;
5705 }
5706
5707
5708 // Remap a block of memory.
5709 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5710 char *addr, size_t bytes, bool read_only,
5711 bool allow_exec) {
5712 // same as map_memory() on this OS
5713 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5714 allow_exec);
5715 }
5716
5717
5718 // Unmap a block of memory.
5719 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5720 return munmap(addr, bytes) == 0;
5721 }
5722
5723 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5724
5725 static clockid_t thread_cpu_clockid(Thread* thread) {
5726 pthread_t tid = thread->osthread()->pthread_id();
5727 clockid_t clockid;
5728
5729 // Get thread clockid
5730 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5731 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5732 return clockid;
5733 }
5734
5735 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5736 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5737 // of a thread.
5738 //
5739 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5740 // the fast estimate available on the platform.
5741
5742 jlong os::current_thread_cpu_time() {
5743 if (os::Linux::supports_fast_thread_cpu_time()) {
5744 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5745 } else {
5746 // return user + sys since the cost is the same
5747 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5748 }
5749 }
5750
5751 jlong os::thread_cpu_time(Thread* thread) {
5752 // consistent with what current_thread_cpu_time() returns
5753 if (os::Linux::supports_fast_thread_cpu_time()) {
5754 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5755 } else {
5756 return slow_thread_cpu_time(thread, true /* user + sys */);
5757 }
5758 }
5759
5760 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5761 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5762 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5763 } else {
5764 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5765 }
5766 }
5767
5768 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5769 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5770 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5771 } else {
5772 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5773 }
5774 }
5775
5776 //
5777 // -1 on error.
5778 //
5779
5780 PRAGMA_DIAG_PUSH
5781 PRAGMA_FORMAT_NONLITERAL_IGNORED
5782 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5783 static bool proc_task_unchecked = true;
5784 static const char *proc_stat_path = "/proc/%d/stat";
5785 pid_t tid = thread->osthread()->thread_id();
5786 char *s;
5787 char stat[2048];
5788 int statlen;
5789 char proc_name[64];
5790 int count;
5791 long sys_time, user_time;
5792 char cdummy;
5793 int idummy;
5794 long ldummy;
5795 FILE *fp;
5796
5797 // The /proc/<tid>/stat aggregates per-process usage on
5798 // new Linux kernels 2.6+ where NPTL is supported.
5799 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5800 // See bug 6328462.
5801 // There possibly can be cases where there is no directory
5802 // /proc/self/task, so we check its availability.
5803 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5804 // This is executed only once
5805 proc_task_unchecked = false;
5806 fp = fopen("/proc/self/task", "r");
5807 if (fp != NULL) {
5808 proc_stat_path = "/proc/self/task/%d/stat";
5809 fclose(fp);
5810 }
5811 }
5812
5813 sprintf(proc_name, proc_stat_path, tid);
5814 fp = fopen(proc_name, "r");
5815 if ( fp == NULL ) return -1;
5816 statlen = fread(stat, 1, 2047, fp);
5817 stat[statlen] = '\0';
5818 fclose(fp);
5819
5820 // Skip pid and the command string. Note that we could be dealing with
5821 // weird command names, e.g. user could decide to rename java launcher
5822 // to "java 1.4.2 :)", then the stat file would look like
5823 // 1234 (java 1.4.2 :)) R ... ...
5824 // We don't really need to know the command string, just find the last
5825 // occurrence of ")" and then start parsing from there. See bug 4726580.
5826 s = strrchr(stat, ')');
5827 if (s == NULL ) return -1;
5828
5829 // Skip blank chars
5830 do s++; while (isspace(*s));
5831
5832 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5833 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5834 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5835 &user_time, &sys_time);
5836 if ( count != 13 ) return -1;
5837 if (user_sys_cpu_time) {
5838 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5839 } else {
5840 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5841 }
5842 }
5843 PRAGMA_DIAG_POP
5844
5845 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5846 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5847 info_ptr->may_skip_backward = false; // elapsed time not wall time
5848 info_ptr->may_skip_forward = false; // elapsed time not wall time
5849 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5850 }
5851
5852 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5853 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5854 info_ptr->may_skip_backward = false; // elapsed time not wall time
5855 info_ptr->may_skip_forward = false; // elapsed time not wall time
5856 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5857 }
5858
5859 bool os::is_thread_cpu_time_supported() {
5860 return true;
5861 }
5862
5863 // System loadavg support. Returns -1 if load average cannot be obtained.
5864 // Linux doesn't yet have a (official) notion of processor sets,
5865 // so just return the system wide load average.
5866 int os::loadavg(double loadavg[], int nelem) {
5867 return ::getloadavg(loadavg, nelem);
5868 }
5869
5870 void os::pause() {
5871 char filename[MAX_PATH];
5872 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5873 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5874 } else {
5875 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5876 }
5877
5878 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5879 if (fd != -1) {
5880 struct stat buf;
5881 ::close(fd);
5882 while (::stat(filename, &buf) == 0) {
5883 (void)::poll(NULL, 0, 100);
5884 }
5885 } else {
5886 jio_fprintf(stderr,
5887 "Could not open pause file '%s', continuing immediately.\n", filename);
5888 }
5889 }
5890
5891
5892 // Refer to the comments in os_solaris.cpp park-unpark.
5893 //
5894 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5895 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5896 // For specifics regarding the bug see GLIBC BUGID 261237 :
5897 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5898 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5899 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5900 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5901 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5902 // and monitorenter when we're using 1-0 locking. All those operations may result in
5903 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5904 // of libpthread avoids the problem, but isn't practical.
5905 //
5906 // Possible remedies:
5907 //
5908 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5909 // This is palliative and probabilistic, however. If the thread is preempted
5910 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5911 // than the minimum period may have passed, and the abstime may be stale (in the
5912 // past) resultin in a hang. Using this technique reduces the odds of a hang
5913 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5914 //
5915 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5916 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5917 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5918 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5919 // thread.
5920 //
5921 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5922 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5923 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5924 // This also works well. In fact it avoids kernel-level scalability impediments
5925 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5926 // timers in a graceful fashion.
5927 //
5928 // 4. When the abstime value is in the past it appears that control returns
5929 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5930 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5931 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5932 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5933 // It may be possible to avoid reinitialization by checking the return
5934 // value from pthread_cond_timedwait(). In addition to reinitializing the
5935 // condvar we must establish the invariant that cond_signal() is only called
5936 // within critical sections protected by the adjunct mutex. This prevents
5937 // cond_signal() from "seeing" a condvar that's in the midst of being
5938 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5939 // desirable signal-after-unlock optimization that avoids futile context switching.
5940 //
5941 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5942 // structure when a condvar is used or initialized. cond_destroy() would
5943 // release the helper structure. Our reinitialize-after-timedwait fix
5944 // put excessive stress on malloc/free and locks protecting the c-heap.
5945 //
5946 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5947 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5948 // and only enabling the work-around for vulnerable environments.
5949
5950 // utility to compute the abstime argument to timedwait:
5951 // millis is the relative timeout time
5952 // abstime will be the absolute timeout time
5953 // TODO: replace compute_abstime() with unpackTime()
5954
5955 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5956 if (millis < 0) millis = 0;
5957
5958 jlong seconds = millis / 1000;
5959 millis %= 1000;
5960 if (seconds > 50000000) { // see man cond_timedwait(3T)
5961 seconds = 50000000;
5962 }
5963
5964 if (os::Linux::supports_monotonic_clock()) {
5965 struct timespec now;
5966 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5967 assert_status(status == 0, status, "clock_gettime");
5968 abstime->tv_sec = now.tv_sec + seconds;
5969 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5970 if (nanos >= NANOSECS_PER_SEC) {
5971 abstime->tv_sec += 1;
5972 nanos -= NANOSECS_PER_SEC;
5973 }
5974 abstime->tv_nsec = nanos;
5975 } else {
5976 struct timeval now;
5977 int status = gettimeofday(&now, NULL);
5978 assert(status == 0, "gettimeofday");
5979 abstime->tv_sec = now.tv_sec + seconds;
5980 long usec = now.tv_usec + millis * 1000;
5981 if (usec >= 1000000) {
5982 abstime->tv_sec += 1;
5983 usec -= 1000000;
5984 }
5985 abstime->tv_nsec = usec * 1000;
5986 }
5987 return abstime;
5988 }
5989
5990
5991 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5992 // Conceptually TryPark() should be equivalent to park(0).
5993
5994 int os::PlatformEvent::TryPark() {
5995 for (;;) {
5996 const int v = _Event ;
5997 guarantee ((v == 0) || (v == 1), "invariant") ;
5998 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5999 }
6000 }
6001
6002 void os::PlatformEvent::park() { // AKA "down()"
6003 // Invariant: Only the thread associated with the Event/PlatformEvent
6004 // may call park().
6005 // TODO: assert that _Assoc != NULL or _Assoc == Self
6006 int v ;
6007 for (;;) {
6008 v = _Event ;
6009 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6010 }
6011 guarantee (v >= 0, "invariant") ;
6012 if (v == 0) {
6013 // Do this the hard way by blocking ...
6014 int status = pthread_mutex_lock(_mutex);
6015 assert_status(status == 0, status, "mutex_lock");
6016 guarantee (_nParked == 0, "invariant") ;
6017 ++ _nParked ;
6018 while (_Event < 0) {
6019 status = pthread_cond_wait(_cond, _mutex);
6020 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6021 // Treat this the same as if the wait was interrupted
6022 if (status == ETIME) { status = EINTR; }
6023 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6024 }
6025 -- _nParked ;
6026
6027 _Event = 0 ;
6028 status = pthread_mutex_unlock(_mutex);
6029 assert_status(status == 0, status, "mutex_unlock");
6030 // Paranoia to ensure our locked and lock-free paths interact
6031 // correctly with each other.
6032 OrderAccess::fence();
6033 }
6034 guarantee (_Event >= 0, "invariant") ;
6035 }
6036
6037 int os::PlatformEvent::park(jlong millis) {
6038 guarantee (_nParked == 0, "invariant") ;
6039
6040 int v ;
6041 for (;;) {
6042 v = _Event ;
6043 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6044 }
6045 guarantee (v >= 0, "invariant") ;
6046 if (v != 0) return OS_OK ;
6047
6048 // We do this the hard way, by blocking the thread.
6049 // Consider enforcing a minimum timeout value.
6050 struct timespec abst;
6051 compute_abstime(&abst, millis);
6052
6053 int ret = OS_TIMEOUT;
6054 int status = pthread_mutex_lock(_mutex);
6055 assert_status(status == 0, status, "mutex_lock");
6056 guarantee (_nParked == 0, "invariant") ;
6057 ++_nParked ;
6058
6059 // Object.wait(timo) will return because of
6060 // (a) notification
6061 // (b) timeout
6062 // (c) thread.interrupt
6063 //
6064 // Thread.interrupt and object.notify{All} both call Event::set.
6065 // That is, we treat thread.interrupt as a special case of notification.
6066 // The underlying Solaris implementation, cond_timedwait, admits
6067 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6068 // JVM from making those visible to Java code. As such, we must
6069 // filter out spurious wakeups. We assume all ETIME returns are valid.
6070 //
6071 // TODO: properly differentiate simultaneous notify+interrupt.
6072 // In that case, we should propagate the notify to another waiter.
6073
6074 while (_Event < 0) {
6075 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6076 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6077 pthread_cond_destroy (_cond);
6078 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6079 }
6080 assert_status(status == 0 || status == EINTR ||
6081 status == ETIME || status == ETIMEDOUT,
6082 status, "cond_timedwait");
6083 if (!FilterSpuriousWakeups) break ; // previous semantics
6084 if (status == ETIME || status == ETIMEDOUT) break ;
6085 // We consume and ignore EINTR and spurious wakeups.
6086 }
6087 --_nParked ;
6088 if (_Event >= 0) {
6089 ret = OS_OK;
6090 }
6091 _Event = 0 ;
6092 status = pthread_mutex_unlock(_mutex);
6093 assert_status(status == 0, status, "mutex_unlock");
6094 assert (_nParked == 0, "invariant") ;
6095 // Paranoia to ensure our locked and lock-free paths interact
6096 // correctly with each other.
6097 OrderAccess::fence();
6098 return ret;
6099 }
6100
6101 void os::PlatformEvent::unpark() {
6102 // Transitions for _Event:
6103 // 0 :=> 1
6104 // 1 :=> 1
6105 // -1 :=> either 0 or 1; must signal target thread
6106 // That is, we can safely transition _Event from -1 to either
6107 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6108 // unpark() calls.
6109 // See also: "Semaphores in Plan 9" by Mullender & Cox
6110 //
6111 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6112 // that it will take two back-to-back park() calls for the owning
6113 // thread to block. This has the benefit of forcing a spurious return
6114 // from the first park() call after an unpark() call which will help
6115 // shake out uses of park() and unpark() without condition variables.
6116
6117 if (Atomic::xchg(1, &_Event) >= 0) return;
6118
6119 // Wait for the thread associated with the event to vacate
6120 int status = pthread_mutex_lock(_mutex);
6121 assert_status(status == 0, status, "mutex_lock");
6122 int AnyWaiters = _nParked;
6123 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6124 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6125 AnyWaiters = 0;
6126 pthread_cond_signal(_cond);
6127 }
6128 status = pthread_mutex_unlock(_mutex);
6129 assert_status(status == 0, status, "mutex_unlock");
6130 if (AnyWaiters != 0) {
6131 status = pthread_cond_signal(_cond);
6132 assert_status(status == 0, status, "cond_signal");
6133 }
6134
6135 // Note that we signal() _after dropping the lock for "immortal" Events.
6136 // This is safe and avoids a common class of futile wakeups. In rare
6137 // circumstances this can cause a thread to return prematurely from
6138 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6139 // simply re-test the condition and re-park itself.
6140 }
6141
6142
6143 // JSR166
6144 // -------------------------------------------------------
6145
6146 /*
6147 * The solaris and linux implementations of park/unpark are fairly
6148 * conservative for now, but can be improved. They currently use a
6149 * mutex/condvar pair, plus a a count.
6150 * Park decrements count if > 0, else does a condvar wait. Unpark
6151 * sets count to 1 and signals condvar. Only one thread ever waits
6152 * on the condvar. Contention seen when trying to park implies that someone
6153 * is unparking you, so don't wait. And spurious returns are fine, so there
6154 * is no need to track notifications.
6155 */
6156
6157 /*
6158 * This code is common to linux and solaris and will be moved to a
6159 * common place in dolphin.
6160 *
6161 * The passed in time value is either a relative time in nanoseconds
6162 * or an absolute time in milliseconds. Either way it has to be unpacked
6163 * into suitable seconds and nanoseconds components and stored in the
6164 * given timespec structure.
6165 * Given time is a 64-bit value and the time_t used in the timespec is only
6166 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6167 * overflow if times way in the future are given. Further on Solaris versions
6168 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6169 * number of seconds, in abstime, is less than current_time + 100,000,000.
6170 * As it will be 28 years before "now + 100000000" will overflow we can
6171 * ignore overflow and just impose a hard-limit on seconds using the value
6172 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6173 * years from "now".
6174 */
6175
6176 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6177 assert (time > 0, "convertTime");
6178 time_t max_secs = 0;
6179
6180 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6181 struct timeval now;
6182 int status = gettimeofday(&now, NULL);
6183 assert(status == 0, "gettimeofday");
6184
6185 max_secs = now.tv_sec + MAX_SECS;
6186
6187 if (isAbsolute) {
6188 jlong secs = time / 1000;
6189 if (secs > max_secs) {
6190 absTime->tv_sec = max_secs;
6191 } else {
6192 absTime->tv_sec = secs;
6193 }
6194 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6195 } else {
6196 jlong secs = time / NANOSECS_PER_SEC;
6197 if (secs >= MAX_SECS) {
6198 absTime->tv_sec = max_secs;
6199 absTime->tv_nsec = 0;
6200 } else {
6201 absTime->tv_sec = now.tv_sec + secs;
6202 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6203 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6204 absTime->tv_nsec -= NANOSECS_PER_SEC;
6205 ++absTime->tv_sec; // note: this must be <= max_secs
6206 }
6207 }
6208 }
6209 } else {
6210 // must be relative using monotonic clock
6211 struct timespec now;
6212 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6213 assert_status(status == 0, status, "clock_gettime");
6214 max_secs = now.tv_sec + MAX_SECS;
6215 jlong secs = time / NANOSECS_PER_SEC;
6216 if (secs >= MAX_SECS) {
6217 absTime->tv_sec = max_secs;
6218 absTime->tv_nsec = 0;
6219 } else {
6220 absTime->tv_sec = now.tv_sec + secs;
6221 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6222 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6223 absTime->tv_nsec -= NANOSECS_PER_SEC;
6224 ++absTime->tv_sec; // note: this must be <= max_secs
6225 }
6226 }
6227 }
6228 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6229 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6230 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6231 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6232 }
6233
6234 void Parker::park(bool isAbsolute, jlong time) {
6235 // Ideally we'd do something useful while spinning, such
6236 // as calling unpackTime().
6237
6238 // Optional fast-path check:
6239 // Return immediately if a permit is available.
6240 // We depend on Atomic::xchg() having full barrier semantics
6241 // since we are doing a lock-free update to _counter.
6242 if (Atomic::xchg(0, &_counter) > 0) return;
6243
6244 Thread* thread = Thread::current();
6245 assert(thread->is_Java_thread(), "Must be JavaThread");
6246 JavaThread *jt = (JavaThread *)thread;
6247
6248 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6249 // Check interrupt before trying to wait
6250 if (Thread::is_interrupted(thread, false)) {
6251 return;
6252 }
6253
6254 // Next, demultiplex/decode time arguments
6255 timespec absTime;
6256 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6257 return;
6258 }
6259 if (time > 0) {
6260 unpackTime(&absTime, isAbsolute, time);
6261 }
6262
6263
6264 // Enter safepoint region
6265 // Beware of deadlocks such as 6317397.
6266 // The per-thread Parker:: mutex is a classic leaf-lock.
6267 // In particular a thread must never block on the Threads_lock while
6268 // holding the Parker:: mutex. If safepoints are pending both the
6269 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6270 ThreadBlockInVM tbivm(jt);
6271
6272 // Don't wait if cannot get lock since interference arises from
6273 // unblocking. Also. check interrupt before trying wait
6274 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6275 return;
6276 }
6277
6278 int status ;
6279 if (_counter > 0) { // no wait needed
6280 _counter = 0;
6281 status = pthread_mutex_unlock(_mutex);
6282 assert (status == 0, "invariant") ;
6283 // Paranoia to ensure our locked and lock-free paths interact
6284 // correctly with each other and Java-level accesses.
6285 OrderAccess::fence();
6286 return;
6287 }
6288
6289 #ifdef ASSERT
6290 // Don't catch signals while blocked; let the running threads have the signals.
6291 // (This allows a debugger to break into the running thread.)
6292 sigset_t oldsigs;
6293 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6294 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6295 #endif
6296
6297 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6298 jt->set_suspend_equivalent();
6299 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6300
6301 assert(_cur_index == -1, "invariant");
6302 if (time == 0) {
6303 _cur_index = REL_INDEX; // arbitrary choice when not timed
6304 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6305 } else {
6306 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6307 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6308 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6309 pthread_cond_destroy (&_cond[_cur_index]) ;
6310 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6311 }
6312 }
6313 _cur_index = -1;
6314 assert_status(status == 0 || status == EINTR ||
6315 status == ETIME || status == ETIMEDOUT,
6316 status, "cond_timedwait");
6317
6318 #ifdef ASSERT
6319 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6320 #endif
6321
6322 _counter = 0 ;
6323 status = pthread_mutex_unlock(_mutex) ;
6324 assert_status(status == 0, status, "invariant") ;
6325 // Paranoia to ensure our locked and lock-free paths interact
6326 // correctly with each other and Java-level accesses.
6327 OrderAccess::fence();
6328
6329 // If externally suspended while waiting, re-suspend
6330 if (jt->handle_special_suspend_equivalent_condition()) {
6331 jt->java_suspend_self();
6332 }
6333 }
6334
6335 void Parker::unpark() {
6336 int s, status ;
6337 status = pthread_mutex_lock(_mutex);
6338 assert (status == 0, "invariant") ;
6339 s = _counter;
6340 _counter = 1;
6341 if (s < 1) {
6342 // thread might be parked
6343 if (_cur_index != -1) {
6344 // thread is definitely parked
6345 if (WorkAroundNPTLTimedWaitHang) {
6346 status = pthread_cond_signal (&_cond[_cur_index]);
6347 assert (status == 0, "invariant");
6348 status = pthread_mutex_unlock(_mutex);
6349 assert (status == 0, "invariant");
6350 } else {
6351 // must capture correct index before unlocking
6352 int index = _cur_index;
6353 status = pthread_mutex_unlock(_mutex);
6354 assert (status == 0, "invariant");
6355 status = pthread_cond_signal (&_cond[index]);
6356 assert (status == 0, "invariant");
6357 }
6358 } else {
6359 pthread_mutex_unlock(_mutex);
6360 assert (status == 0, "invariant") ;
6361 }
6362 } else {
6363 pthread_mutex_unlock(_mutex);
6364 assert (status == 0, "invariant") ;
6365 }
6366 }
6367
6368
6369 extern char** environ;
6370
6371 // Run the specified command in a separate process. Return its exit value,
6372 // or -1 on failure (e.g. can't fork a new process).
6373 // Unlike system(), this function can be called from signal handler. It
6374 // doesn't block SIGINT et al.
6375 int os::fork_and_exec(char* cmd) {
6376 const char * argv[4] = {"sh", "-c", cmd, NULL};
6377
6378 pid_t pid = fork();
6379
6380 if (pid < 0) {
6381 // fork failed
6382 return -1;
6383
6384 } else if (pid == 0) {
6385 // child process
6386
6387 execve("/bin/sh", (char* const*)argv, environ);
6388
6389 // execve failed
6390 _exit(-1);
6391
6392 } else {
6393 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6394 // care about the actual exit code, for now.
6395
6396 int status;
6397
6398 // Wait for the child process to exit. This returns immediately if
6399 // the child has already exited. */
6400 while (waitpid(pid, &status, 0) < 0) {
6401 switch (errno) {
6402 case ECHILD: return 0;
6403 case EINTR: break;
6404 default: return -1;
6405 }
6406 }
6407
6408 if (WIFEXITED(status)) {
6409 // The child exited normally; get its exit code.
6410 return WEXITSTATUS(status);
6411 } else if (WIFSIGNALED(status)) {
6412 // The child exited because of a signal
6413 // The best value to return is 0x80 + signal number,
6414 // because that is what all Unix shells do, and because
6415 // it allows callers to distinguish between process exit and
6416 // process death by signal.
6417 return 0x80 + WTERMSIG(status);
6418 } else {
6419 // Unknown exit code; pass it through
6420 return status;
6421 }
6422 }
6423 }
6424
6425 // is_headless_jre()
6426 //
6427 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6428 // in order to report if we are running in a headless jre
6429 //
6430 // Since JDK8 xawt/libmawt.so was moved into the same directory
6431 // as libawt.so, and renamed libawt_xawt.so
6432 //
6433 bool os::is_headless_jre() {
6434 struct stat statbuf;
6435 char buf[MAXPATHLEN];
6436 char libmawtpath[MAXPATHLEN];
6437 const char *xawtstr = "/xawt/libmawt.so";
6438 const char *new_xawtstr = "/libawt_xawt.so";
6439 char *p;
6440
6441 // Get path to libjvm.so
6442 os::jvm_path(buf, sizeof(buf));
6443
6444 // Get rid of libjvm.so
6445 p = strrchr(buf, '/');
6446 if (p == NULL) return false;
6447 else *p = '\0';
6448
6449 // Get rid of client or server
6450 p = strrchr(buf, '/');
6451 if (p == NULL) return false;
6452 else *p = '\0';
6453
6454 // check xawt/libmawt.so
6455 strcpy(libmawtpath, buf);
6456 strcat(libmawtpath, xawtstr);
6457 if (::stat(libmawtpath, &statbuf) == 0) return false;
6458
6459 // check libawt_xawt.so
6460 strcpy(libmawtpath, buf);
6461 strcat(libmawtpath, new_xawtstr);
6462 if (::stat(libmawtpath, &statbuf) == 0) return false;
6463
6464 return true;
6465 }
6466
6467 // Get the default path to the core file
6468 // Returns the length of the string
6469 int os::get_core_path(char* buffer, size_t bufferSize) {
6470 const char* p = get_current_directory(buffer, bufferSize);
6471
6472 if (p == NULL) {
6473 assert(p != NULL, "failed to get current directory");
6474 return 0;
6475 }
6476
6477 return strlen(buffer);
6478 }
6479
6480 /////////////// Unit tests ///////////////
6481
6482 #ifndef PRODUCT
6483
6484 #define test_log(...) \
6485 do {\
6486 if (VerboseInternalVMTests) { \
6487 tty->print_cr(__VA_ARGS__); \
6488 tty->flush(); \
6489 }\
6490 } while (false)
6491
6492 class TestReserveMemorySpecial : AllStatic {
6493 public:
6494 static void small_page_write(void* addr, size_t size) {
6495 size_t page_size = os::vm_page_size();
6496
6497 char* end = (char*)addr + size;
6498 for (char* p = (char*)addr; p < end; p += page_size) {
6499 *p = 1;
6500 }
6501 }
6502
6503 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6504 if (!UseHugeTLBFS) {
6505 return;
6506 }
6507
6508 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6509
6510 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6511
6512 if (addr != NULL) {
6513 small_page_write(addr, size);
6514
6515 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6516 }
6517 }
6518
6519 static void test_reserve_memory_special_huge_tlbfs_only() {
6520 if (!UseHugeTLBFS) {
6521 return;
6522 }
6523
6524 size_t lp = os::large_page_size();
6525
6526 for (size_t size = lp; size <= lp * 10; size += lp) {
6527 test_reserve_memory_special_huge_tlbfs_only(size);
6528 }
6529 }
6530
6531 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6532 size_t lp = os::large_page_size();
6533 size_t ag = os::vm_allocation_granularity();
6534
6535 // sizes to test
6536 const size_t sizes[] = {
6537 lp, lp + ag, lp + lp / 2, lp * 2,
6538 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6539 lp * 10, lp * 10 + lp / 2
6540 };
6541 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6542
6543 // For each size/alignment combination, we test three scenarios:
6544 // 1) with req_addr == NULL
6545 // 2) with a non-null req_addr at which we expect to successfully allocate
6546 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6547 // expect the allocation to either fail or to ignore req_addr
6548
6549 // Pre-allocate two areas; they shall be as large as the largest allocation
6550 // and aligned to the largest alignment we will be testing.
6551 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6552 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6553 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6554 -1, 0);
6555 assert(mapping1 != MAP_FAILED, "should work");
6556
6557 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6558 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6559 -1, 0);
6560 assert(mapping2 != MAP_FAILED, "should work");
6561
6562 // Unmap the first mapping, but leave the second mapping intact: the first
6563 // mapping will serve as a value for a "good" req_addr (case 2). The second
6564 // mapping, still intact, as "bad" req_addr (case 3).
6565 ::munmap(mapping1, mapping_size);
6566
6567 // Case 1
6568 test_log("%s, req_addr NULL:", __FUNCTION__);
6569 test_log("size align result");
6570
6571 for (int i = 0; i < num_sizes; i++) {
6572 const size_t size = sizes[i];
6573 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6574 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6575 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6576 size, alignment, p, (p != NULL ? "" : "(failed)"));
6577 if (p != NULL) {
6578 assert(is_ptr_aligned(p, alignment), "must be");
6579 small_page_write(p, size);
6580 os::Linux::release_memory_special_huge_tlbfs(p, size);
6581 }
6582 }
6583 }
6584
6585 // Case 2
6586 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6587 test_log("size align req_addr result");
6588
6589 for (int i = 0; i < num_sizes; i++) {
6590 const size_t size = sizes[i];
6591 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6592 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6593 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6594 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6595 size, alignment, req_addr, p,
6596 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6597 if (p != NULL) {
6598 assert(p == req_addr, "must be");
6599 small_page_write(p, size);
6600 os::Linux::release_memory_special_huge_tlbfs(p, size);
6601 }
6602 }
6603 }
6604
6605 // Case 3
6606 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6607 test_log("size align req_addr result");
6608
6609 for (int i = 0; i < num_sizes; i++) {
6610 const size_t size = sizes[i];
6611 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6612 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6613 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6614 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6615 size, alignment, req_addr, p,
6616 ((p != NULL ? "" : "(failed)")));
6617 // as the area around req_addr contains already existing mappings, the API should always
6618 // return NULL (as per contract, it cannot return another address)
6619 assert(p == NULL, "must be");
6620 }
6621 }
6622
6623 ::munmap(mapping2, mapping_size);
6624
6625 }
6626
6627 static void test_reserve_memory_special_huge_tlbfs() {
6628 if (!UseHugeTLBFS) {
6629 return;
6630 }
6631
6632 test_reserve_memory_special_huge_tlbfs_only();
6633 test_reserve_memory_special_huge_tlbfs_mixed();
6634 }
6635
6636 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6637 if (!UseSHM) {
6638 return;
6639 }
6640
6641 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6642
6643 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6644
6645 if (addr != NULL) {
6646 assert(is_ptr_aligned(addr, alignment), "Check");
6647 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6648
6649 small_page_write(addr, size);
6650
6651 os::Linux::release_memory_special_shm(addr, size);
6652 }
6653 }
6654
6655 static void test_reserve_memory_special_shm() {
6656 size_t lp = os::large_page_size();
6657 size_t ag = os::vm_allocation_granularity();
6658
6659 for (size_t size = ag; size < lp * 3; size += ag) {
6660 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6661 test_reserve_memory_special_shm(size, alignment);
6662 }
6663 }
6664 }
6665
6666 static void test() {
6667 test_reserve_memory_special_huge_tlbfs();
6668 test_reserve_memory_special_shm();
6669 }
6670 };
6671
6672 void TestReserveMemorySpecial_test() {
6673 TestReserveMemorySpecial::test();
6674 }
6675
6676 #endif