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 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
140 Mutex* os::Linux::_createThread_lock = NULL;
141 pthread_t os::Linux::_main_thread;
142 int os::Linux::_page_size = -1;
143 const int os::Linux::_vm_default_page_size = (8 * K);
144 bool os::Linux::_is_floating_stack = false;
145 bool os::Linux::_is_NPTL = false;
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149 pthread_condattr_t os::Linux::_condattr[1];
150
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 static pid_t _initial_pid = 0;
160
161 /* Signal number used to suspend/resume a thread */
162
163 /* do not use any signal number less than SIGSEGV, see 4355769 */
164 static int SR_signum = SIGUSR2;
165 sigset_t SR_sigset;
166
167 /* Used to protect dlsym() calls */
168 static pthread_mutex_t dl_mutex;
169
170 // Declarations
171 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
172
173 // utility functions
174
175 static int SR_initialize();
176
177 julong os::available_memory() {
178 return Linux::available_memory();
179 }
180
181 julong os::Linux::available_memory() {
182 // values in struct sysinfo are "unsigned long"
183 struct sysinfo si;
184 julong avail_mem;
185
186 if (OSContainer::is_containerized()) {
187 jlong mem_limit, mem_usage;
188 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
189 if (PrintContainerInfo) {
190 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
191 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
192 }
193 }
194
195 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
196 if (PrintContainerInfo) {
197 tty->print_cr("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
198 }
199 }
200
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 if (PrintContainerInfo) {
204 tty->print_cr("available container memory: " JULONG_FORMAT, avail_mem);
205 }
206 return avail_mem;
207 }
208 }
209
210 sysinfo(&si);
211 avail_mem = (julong)si.freeram * si.mem_unit;
212 if (Verbose) {
213 tty->print_cr("available memory: " JULONG_FORMAT, avail_mem);
214 }
215 return avail_mem;
216 }
217
218 julong os::physical_memory() {
219 jlong phys_mem = 0;
220 if (OSContainer::is_containerized()) {
221 jlong mem_limit;
222 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
223 if (PrintContainerInfo) {
224 tty->print_cr("total container memory: " JLONG_FORMAT, mem_limit);
225 }
226 return mem_limit;
227 }
228
229 if (PrintContainerInfo) {
230 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
231 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
232 }
233 }
234
235 phys_mem = Linux::physical_memory();
236 if (Verbose) {
237 tty->print_cr("total system memory: " JLONG_FORMAT, phys_mem);
238 }
239 return phys_mem;
240 }
241
242 ////////////////////////////////////////////////////////////////////////////////
243 // environment support
244
245 bool os::getenv(const char* name, char* buf, int len) {
246 const char* val = ::getenv(name);
247 if (val != NULL && strlen(val) < (size_t)len) {
248 strcpy(buf, val);
249 return true;
250 }
251 if (len > 0) buf[0] = 0; // return a null string
252 return false;
253 }
254
255
256 // Return true if user is running as root.
257
258 bool os::have_special_privileges() {
259 static bool init = false;
260 static bool privileges = false;
261 if (!init) {
262 privileges = (getuid() != geteuid()) || (getgid() != getegid());
263 init = true;
264 }
265 return privileges;
266 }
267
268
269 #ifndef SYS_gettid
270 // i386: 224, ia64: 1105, amd64: 186, sparc 143
271 #ifdef __ia64__
272 #define SYS_gettid 1105
273 #else
274 #ifdef __i386__
275 #define SYS_gettid 224
276 #else
277 #ifdef __amd64__
278 #define SYS_gettid 186
279 #else
280 #ifdef __sparc__
281 #define SYS_gettid 143
282 #else
283 #error define gettid for the arch
284 #endif
285 #endif
286 #endif
287 #endif
288 #endif
289
290 // Cpu architecture string
291 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
292
293 // pid_t gettid()
294 //
295 // Returns the kernel thread id of the currently running thread. Kernel
296 // thread id is used to access /proc.
297 //
298 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
299 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
300 //
301 pid_t os::Linux::gettid() {
302 int rslt = syscall(SYS_gettid);
303 if (rslt == -1) {
304 // old kernel, no NPTL support
305 return getpid();
306 } else {
307 return (pid_t)rslt;
308 }
309 }
310
311 // Most versions of linux have a bug where the number of processors are
312 // determined by looking at the /proc file system. In a chroot environment,
313 // the system call returns 1. This causes the VM to act as if it is
314 // a single processor and elide locking (see is_MP() call).
315 static bool unsafe_chroot_detected = false;
316 static const char *unstable_chroot_error = "/proc file system not found.\n"
317 "Java may be unstable running multithreaded in a chroot "
318 "environment on Linux when /proc filesystem is not mounted.";
319
320 void os::Linux::initialize_system_info() {
321 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
322 if (processor_count() == 1) {
323 pid_t pid = os::Linux::gettid();
324 char fname[32];
325 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
326 FILE *fp = fopen(fname, "r");
327 if (fp == NULL) {
328 unsafe_chroot_detected = true;
329 } else {
330 fclose(fp);
331 }
332 }
333 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
334 assert(processor_count() > 0, "linux error");
335 }
336
337 void os::init_system_properties_values() {
338 // The next steps are taken in the product version:
339 //
340 // Obtain the JAVA_HOME value from the location of libjvm.so.
341 // This library should be located at:
342 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
343 //
344 // If "/jre/lib/" appears at the right place in the path, then we
345 // assume libjvm.so is installed in a JDK and we use this path.
346 //
347 // Otherwise exit with message: "Could not create the Java virtual machine."
348 //
349 // The following extra steps are taken in the debugging version:
350 //
351 // If "/jre/lib/" does NOT appear at the right place in the path
352 // instead of exit check for $JAVA_HOME environment variable.
353 //
354 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
355 // then we append a fake suffix "hotspot/libjvm.so" to this path so
356 // it looks like libjvm.so is installed there
357 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
358 //
359 // Otherwise exit.
360 //
361 // Important note: if the location of libjvm.so changes this
362 // code needs to be changed accordingly.
363
364 // See ld(1):
365 // The linker uses the following search paths to locate required
366 // shared libraries:
367 // 1: ...
368 // ...
369 // 7: The default directories, normally /lib and /usr/lib.
370 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
371 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
372 #else
373 #define DEFAULT_LIBPATH "/lib:/usr/lib"
374 #endif
375
376 // Base path of extensions installed on the system.
377 #define SYS_EXT_DIR "/usr/java/packages"
378 #define EXTENSIONS_DIR "/lib/ext"
379 #define ENDORSED_DIR "/lib/endorsed"
380
381 // Buffer that fits several sprintfs.
382 // Note that the space for the colon and the trailing null are provided
383 // by the nulls included by the sizeof operator.
384 const size_t bufsize =
385 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
386 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
387 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
388 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
389
390 // sysclasspath, java_home, dll_dir
391 {
392 char *pslash;
393 os::jvm_path(buf, bufsize);
394
395 // Found the full path to libjvm.so.
396 // Now cut the path to <java_home>/jre if we can.
397 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
398 pslash = strrchr(buf, '/');
399 if (pslash != NULL) {
400 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
401 }
402 Arguments::set_dll_dir(buf);
403
404 if (pslash != NULL) {
405 pslash = strrchr(buf, '/');
406 if (pslash != NULL) {
407 *pslash = '\0'; // Get rid of /<arch>.
408 pslash = strrchr(buf, '/');
409 if (pslash != NULL) {
410 *pslash = '\0'; // Get rid of /lib.
411 }
412 }
413 }
414 Arguments::set_java_home(buf);
415 set_boot_path('/', ':');
416 }
417
418 // Where to look for native libraries.
419 //
420 // Note: Due to a legacy implementation, most of the library path
421 // is set in the launcher. This was to accomodate linking restrictions
422 // on legacy Linux implementations (which are no longer supported).
423 // Eventually, all the library path setting will be done here.
424 //
425 // However, to prevent the proliferation of improperly built native
426 // libraries, the new path component /usr/java/packages is added here.
427 // Eventually, all the library path setting will be done here.
428 {
429 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
430 // should always exist (until the legacy problem cited above is
431 // addressed).
432 const char *v = ::getenv("LD_LIBRARY_PATH");
433 const char *v_colon = ":";
434 if (v == NULL) { v = ""; v_colon = ""; }
435 // That's +1 for the colon and +1 for the trailing '\0'.
436 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
437 strlen(v) + 1 +
438 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
439 mtInternal);
440 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
441 Arguments::set_library_path(ld_library_path);
442 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
443 }
444
445 // Extensions directories.
446 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
447 Arguments::set_ext_dirs(buf);
448
449 // Endorsed standards default directory.
450 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
451 Arguments::set_endorsed_dirs(buf);
452
453 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
454
455 #undef DEFAULT_LIBPATH
456 #undef SYS_EXT_DIR
457 #undef EXTENSIONS_DIR
458 #undef ENDORSED_DIR
459 }
460
461 ////////////////////////////////////////////////////////////////////////////////
462 // breakpoint support
463
464 void os::breakpoint() {
465 BREAKPOINT;
466 }
467
468 extern "C" void breakpoint() {
469 // use debugger to set breakpoint here
470 }
471
472 ////////////////////////////////////////////////////////////////////////////////
473 // signal support
474
475 debug_only(static bool signal_sets_initialized = false);
476 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
477
478 bool os::Linux::is_sig_ignored(int sig) {
479 struct sigaction oact;
480 sigaction(sig, (struct sigaction*)NULL, &oact);
481 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
482 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
483 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
484 return true;
485 else
486 return false;
487 }
488
489 void os::Linux::signal_sets_init() {
490 // Should also have an assertion stating we are still single-threaded.
491 assert(!signal_sets_initialized, "Already initialized");
492 // Fill in signals that are necessarily unblocked for all threads in
493 // the VM. Currently, we unblock the following signals:
494 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
495 // by -Xrs (=ReduceSignalUsage));
496 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
497 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
498 // the dispositions or masks wrt these signals.
499 // Programs embedding the VM that want to use the above signals for their
500 // own purposes must, at this time, use the "-Xrs" option to prevent
501 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
502 // (See bug 4345157, and other related bugs).
503 // In reality, though, unblocking these signals is really a nop, since
504 // these signals are not blocked by default.
505 sigemptyset(&unblocked_sigs);
506 sigemptyset(&allowdebug_blocked_sigs);
507 sigaddset(&unblocked_sigs, SIGILL);
508 sigaddset(&unblocked_sigs, SIGSEGV);
509 sigaddset(&unblocked_sigs, SIGBUS);
510 sigaddset(&unblocked_sigs, SIGFPE);
511 #if defined(PPC64)
512 sigaddset(&unblocked_sigs, SIGTRAP);
513 #endif
514 sigaddset(&unblocked_sigs, SR_signum);
515
516 if (!ReduceSignalUsage) {
517 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
518 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
519 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
520 }
521 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
522 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
523 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
524 }
525 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
526 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
527 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
528 }
529 }
530 // Fill in signals that are blocked by all but the VM thread.
531 sigemptyset(&vm_sigs);
532 if (!ReduceSignalUsage)
533 sigaddset(&vm_sigs, BREAK_SIGNAL);
534 debug_only(signal_sets_initialized = true);
535
536 }
537
538 // These are signals that are unblocked while a thread is running Java.
539 // (For some reason, they get blocked by default.)
540 sigset_t* os::Linux::unblocked_signals() {
541 assert(signal_sets_initialized, "Not initialized");
542 return &unblocked_sigs;
543 }
544
545 // These are the signals that are blocked while a (non-VM) thread is
546 // running Java. Only the VM thread handles these signals.
547 sigset_t* os::Linux::vm_signals() {
548 assert(signal_sets_initialized, "Not initialized");
549 return &vm_sigs;
550 }
551
552 // These are signals that are blocked during cond_wait to allow debugger in
553 sigset_t* os::Linux::allowdebug_blocked_signals() {
554 assert(signal_sets_initialized, "Not initialized");
555 return &allowdebug_blocked_sigs;
556 }
557
558 void os::Linux::hotspot_sigmask(Thread* thread) {
559
560 //Save caller's signal mask before setting VM signal mask
561 sigset_t caller_sigmask;
562 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
563
564 OSThread* osthread = thread->osthread();
565 osthread->set_caller_sigmask(caller_sigmask);
566
567 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
568
569 if (!ReduceSignalUsage) {
570 if (thread->is_VM_thread()) {
571 // Only the VM thread handles BREAK_SIGNAL ...
572 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
573 } else {
574 // ... all other threads block BREAK_SIGNAL
575 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
576 }
577 }
578 }
579
580 //////////////////////////////////////////////////////////////////////////////
581 // detecting pthread library
582
583 void os::Linux::libpthread_init() {
584 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
585 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
586 // generic name for earlier versions.
587 // Define macros here so we can build HotSpot on old systems.
588 # ifndef _CS_GNU_LIBC_VERSION
589 # define _CS_GNU_LIBC_VERSION 2
590 # endif
591 # ifndef _CS_GNU_LIBPTHREAD_VERSION
592 # define _CS_GNU_LIBPTHREAD_VERSION 3
593 # endif
594
595 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
596 if (n > 0) {
597 char *str = (char *)malloc(n, mtInternal);
598 confstr(_CS_GNU_LIBC_VERSION, str, n);
599 os::Linux::set_glibc_version(str);
600 } else {
601 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
602 static char _gnu_libc_version[32];
603 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
604 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
605 os::Linux::set_glibc_version(_gnu_libc_version);
606 }
607
608 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
609 if (n > 0) {
610 char *str = (char *)malloc(n, mtInternal);
611 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
612 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
613 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
614 // is the case. LinuxThreads has a hard limit on max number of threads.
615 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
616 // On the other hand, NPTL does not have such a limit, sysconf()
617 // will return -1 and errno is not changed. Check if it is really NPTL.
618 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
619 strstr(str, "NPTL") &&
620 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
621 free(str);
622 os::Linux::set_libpthread_version("linuxthreads");
623 } else {
624 os::Linux::set_libpthread_version(str);
625 }
626 } else {
627 // glibc before 2.3.2 only has LinuxThreads.
628 os::Linux::set_libpthread_version("linuxthreads");
629 }
630
631 if (strstr(libpthread_version(), "NPTL")) {
632 os::Linux::set_is_NPTL();
633 } else {
634 os::Linux::set_is_LinuxThreads();
635 }
636
637 // LinuxThreads have two flavors: floating-stack mode, which allows variable
638 // stack size; and fixed-stack mode. NPTL is always floating-stack.
639 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
640 os::Linux::set_is_floating_stack();
641 }
642 }
643
644 /////////////////////////////////////////////////////////////////////////////
645 // thread stack
646
647 // Force Linux kernel to expand current thread stack. If "bottom" is close
648 // to the stack guard, caller should block all signals.
649 //
650 // MAP_GROWSDOWN:
651 // A special mmap() flag that is used to implement thread stacks. It tells
652 // kernel that the memory region should extend downwards when needed. This
653 // allows early versions of LinuxThreads to only mmap the first few pages
654 // when creating a new thread. Linux kernel will automatically expand thread
655 // stack as needed (on page faults).
656 //
657 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
658 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
659 // region, it's hard to tell if the fault is due to a legitimate stack
660 // access or because of reading/writing non-exist memory (e.g. buffer
661 // overrun). As a rule, if the fault happens below current stack pointer,
662 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
663 // application (see Linux kernel fault.c).
664 //
665 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
666 // stack overflow detection.
667 //
668 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
669 // not use this flag. However, the stack of initial thread is not created
670 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
671 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
672 // and then attach the thread to JVM.
673 //
674 // To get around the problem and allow stack banging on Linux, we need to
675 // manually expand thread stack after receiving the SIGSEGV.
676 //
677 // There are two ways to expand thread stack to address "bottom", we used
678 // both of them in JVM before 1.5:
679 // 1. adjust stack pointer first so that it is below "bottom", and then
680 // touch "bottom"
681 // 2. mmap() the page in question
682 //
683 // Now alternate signal stack is gone, it's harder to use 2. For instance,
684 // if current sp is already near the lower end of page 101, and we need to
685 // call mmap() to map page 100, it is possible that part of the mmap() frame
686 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
687 // That will destroy the mmap() frame and cause VM to crash.
688 //
689 // The following code works by adjusting sp first, then accessing the "bottom"
690 // page to force a page fault. Linux kernel will then automatically expand the
691 // stack mapping.
692 //
693 // _expand_stack_to() assumes its frame size is less than page size, which
694 // should always be true if the function is not inlined.
695
696 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
697 #define NOINLINE
698 #else
699 #define NOINLINE __attribute__ ((noinline))
700 #endif
701
702 static void _expand_stack_to(address bottom) NOINLINE;
703
704 static void _expand_stack_to(address bottom) {
705 address sp;
706 size_t size;
707 volatile char *p;
708
709 // Adjust bottom to point to the largest address within the same page, it
710 // gives us a one-page buffer if alloca() allocates slightly more memory.
711 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
712 bottom += os::Linux::page_size() - 1;
713
714 // sp might be slightly above current stack pointer; if that's the case, we
715 // will alloca() a little more space than necessary, which is OK. Don't use
716 // os::current_stack_pointer(), as its result can be slightly below current
717 // stack pointer, causing us to not alloca enough to reach "bottom".
718 sp = (address)&sp;
719
720 if (sp > bottom) {
721 size = sp - bottom;
722 p = (volatile char *)alloca(size);
723 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
724 p[0] = '\0';
725 }
726 }
727
728 void os::Linux::expand_stack_to(address bottom) {
729 _expand_stack_to(bottom);
730 }
731
732 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
733 assert(t!=NULL, "just checking");
734 assert(t->osthread()->expanding_stack(), "expand should be set");
735 assert(t->stack_base() != NULL, "stack_base was not initialized");
736
737 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
738 sigset_t mask_all, old_sigset;
739 sigfillset(&mask_all);
740 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
741 _expand_stack_to(addr);
742 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
743 return true;
744 }
745 return false;
746 }
747
748 //////////////////////////////////////////////////////////////////////////////
749 // create new thread
750
751 static address highest_vm_reserved_address();
752
753 // check if it's safe to start a new thread
754 static bool _thread_safety_check(Thread* thread) {
755 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
756 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
757 // Heap is mmap'ed at lower end of memory space. Thread stacks are
758 // allocated (MAP_FIXED) from high address space. Every thread stack
759 // occupies a fixed size slot (usually 2Mbytes, but user can change
760 // it to other values if they rebuild LinuxThreads).
761 //
762 // Problem with MAP_FIXED is that mmap() can still succeed even part of
763 // the memory region has already been mmap'ed. That means if we have too
764 // many threads and/or very large heap, eventually thread stack will
765 // collide with heap.
766 //
767 // Here we try to prevent heap/stack collision by comparing current
768 // stack bottom with the highest address that has been mmap'ed by JVM
769 // plus a safety margin for memory maps created by native code.
770 //
771 // This feature can be disabled by setting ThreadSafetyMargin to 0
772 //
773 if (ThreadSafetyMargin > 0) {
774 address stack_bottom = os::current_stack_base() - os::current_stack_size();
775
776 // not safe if our stack extends below the safety margin
777 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
778 } else {
779 return true;
780 }
781 } else {
782 // Floating stack LinuxThreads or NPTL:
783 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
784 // there's not enough space left, pthread_create() will fail. If we come
785 // here, that means enough space has been reserved for stack.
786 return true;
787 }
788 }
789
790 // Thread start routine for all newly created threads
791 static void *java_start(Thread *thread) {
792 // Try to randomize the cache line index of hot stack frames.
793 // This helps when threads of the same stack traces evict each other's
794 // cache lines. The threads can be either from the same JVM instance, or
795 // from different JVM instances. The benefit is especially true for
796 // processors with hyperthreading technology.
797 static int counter = 0;
798 int pid = os::current_process_id();
799 alloca(((pid ^ counter++) & 7) * 128);
800
801 ThreadLocalStorage::set_thread(thread);
802
803 OSThread* osthread = thread->osthread();
804 Monitor* sync = osthread->startThread_lock();
805
806 // non floating stack LinuxThreads needs extra check, see above
807 if (!_thread_safety_check(thread)) {
808 // notify parent thread
809 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
810 osthread->set_state(ZOMBIE);
811 sync->notify_all();
812 return NULL;
813 }
814
815 // thread_id is kernel thread id (similar to Solaris LWP id)
816 osthread->set_thread_id(os::Linux::gettid());
817
818 if (UseNUMA) {
819 int lgrp_id = os::numa_get_group_id();
820 if (lgrp_id != -1) {
821 thread->set_lgrp_id(lgrp_id);
822 }
823 }
824 // initialize signal mask for this thread
825 os::Linux::hotspot_sigmask(thread);
826
827 // initialize floating point control register
828 os::Linux::init_thread_fpu_state();
829
830 // handshaking with parent thread
831 {
832 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
833
834 // notify parent thread
835 osthread->set_state(INITIALIZED);
836 sync->notify_all();
837
838 // wait until os::start_thread()
839 while (osthread->get_state() == INITIALIZED) {
840 sync->wait(Mutex::_no_safepoint_check_flag);
841 }
842 }
843
844 // call one more level start routine
845 thread->run();
846
847 return 0;
848 }
849
850 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
851 assert(thread->osthread() == NULL, "caller responsible");
852
853 // Allocate the OSThread object
854 OSThread* osthread = new OSThread(NULL, NULL);
855 if (osthread == NULL) {
856 return false;
857 }
858
859 // set the correct thread state
860 osthread->set_thread_type(thr_type);
861
862 // Initial state is ALLOCATED but not INITIALIZED
863 osthread->set_state(ALLOCATED);
864
865 thread->set_osthread(osthread);
866
867 // init thread attributes
868 pthread_attr_t attr;
869 pthread_attr_init(&attr);
870 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
871
872 // stack size
873 if (os::Linux::supports_variable_stack_size()) {
874 // calculate stack size if it's not specified by caller
875 if (stack_size == 0) {
876 stack_size = os::Linux::default_stack_size(thr_type);
877
878 switch (thr_type) {
879 case os::java_thread:
880 // Java threads use ThreadStackSize which default value can be
881 // changed with the flag -Xss
882 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
883 stack_size = JavaThread::stack_size_at_create();
884 break;
885 case os::compiler_thread:
886 if (CompilerThreadStackSize > 0) {
887 stack_size = (size_t)(CompilerThreadStackSize * K);
888 break;
889 } // else fall through:
890 // use VMThreadStackSize if CompilerThreadStackSize is not defined
891 case os::vm_thread:
892 case os::pgc_thread:
893 case os::cgc_thread:
894 case os::watcher_thread:
895 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
896 break;
897 }
898 }
899
900 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
901 pthread_attr_setstacksize(&attr, stack_size);
902 } else {
903 // let pthread_create() pick the default value.
904 }
905
906 // glibc guard page
907 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
908
909 ThreadState state;
910
911 {
912 // Serialize thread creation if we are running with fixed stack LinuxThreads
913 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
914 if (lock) {
915 os::Linux::createThread_lock()->lock_without_safepoint_check();
916 }
917
918 pthread_t tid;
919 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
920
921 pthread_attr_destroy(&attr);
922
923 if (ret != 0) {
924 if (PrintMiscellaneous && (Verbose || WizardMode)) {
925 perror("pthread_create()");
926 }
927 // Need to clean up stuff we've allocated so far
928 thread->set_osthread(NULL);
929 delete osthread;
930 if (lock) os::Linux::createThread_lock()->unlock();
931 return false;
932 }
933
934 // Store pthread info into the OSThread
935 osthread->set_pthread_id(tid);
936
937 // Wait until child thread is either initialized or aborted
938 {
939 Monitor* sync_with_child = osthread->startThread_lock();
940 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
941 while ((state = osthread->get_state()) == ALLOCATED) {
942 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
943 }
944 }
945
946 if (lock) {
947 os::Linux::createThread_lock()->unlock();
948 }
949 }
950
951 // Aborted due to thread limit being reached
952 if (state == ZOMBIE) {
953 thread->set_osthread(NULL);
954 delete osthread;
955 return false;
956 }
957
958 // The thread is returned suspended (in state INITIALIZED),
959 // and is started higher up in the call chain
960 assert(state == INITIALIZED, "race condition");
961 return true;
962 }
963
964 /////////////////////////////////////////////////////////////////////////////
965 // attach existing thread
966
967 // bootstrap the main thread
968 bool os::create_main_thread(JavaThread* thread) {
969 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
970 return create_attached_thread(thread);
971 }
972
973 bool os::create_attached_thread(JavaThread* thread) {
974 #ifdef ASSERT
975 thread->verify_not_published();
976 #endif
977
978 // Allocate the OSThread object
979 OSThread* osthread = new OSThread(NULL, NULL);
980
981 if (osthread == NULL) {
982 return false;
983 }
984
985 // Store pthread info into the OSThread
986 osthread->set_thread_id(os::Linux::gettid());
987 osthread->set_pthread_id(::pthread_self());
988
989 // initialize floating point control register
990 os::Linux::init_thread_fpu_state();
991
992 // Initial thread state is RUNNABLE
993 osthread->set_state(RUNNABLE);
994
995 thread->set_osthread(osthread);
996
997 if (UseNUMA) {
998 int lgrp_id = os::numa_get_group_id();
999 if (lgrp_id != -1) {
1000 thread->set_lgrp_id(lgrp_id);
1001 }
1002 }
1003
1004 if (os::is_primordial_thread()) {
1005 // If current thread is primordial thread, its stack is mapped on demand,
1006 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1007 // the entire stack region to avoid SEGV in stack banging.
1008 // It is also useful to get around the heap-stack-gap problem on SuSE
1009 // kernel (see 4821821 for details). We first expand stack to the top
1010 // of yellow zone, then enable stack yellow zone (order is significant,
1011 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1012 // is no gap between the last two virtual memory regions.
1013
1014 JavaThread *jt = (JavaThread *)thread;
1015 address addr = jt->stack_yellow_zone_base();
1016 assert(addr != NULL, "initialization problem?");
1017 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1018
1019 osthread->set_expanding_stack();
1020 os::Linux::manually_expand_stack(jt, addr);
1021 osthread->clear_expanding_stack();
1022 }
1023
1024 // initialize signal mask for this thread
1025 // and save the caller's signal mask
1026 os::Linux::hotspot_sigmask(thread);
1027
1028 return true;
1029 }
1030
1031 void os::pd_start_thread(Thread* thread) {
1032 OSThread * osthread = thread->osthread();
1033 assert(osthread->get_state() != INITIALIZED, "just checking");
1034 Monitor* sync_with_child = osthread->startThread_lock();
1035 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1036 sync_with_child->notify();
1037 }
1038
1039 // Free Linux resources related to the OSThread
1040 void os::free_thread(OSThread* osthread) {
1041 assert(osthread != NULL, "osthread not set");
1042
1043 if (Thread::current()->osthread() == osthread) {
1044 // Restore caller's signal mask
1045 sigset_t sigmask = osthread->caller_sigmask();
1046 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1047 }
1048
1049 delete osthread;
1050 }
1051
1052 //////////////////////////////////////////////////////////////////////////////
1053 // thread local storage
1054
1055 // Restore the thread pointer if the destructor is called. This is in case
1056 // someone from JNI code sets up a destructor with pthread_key_create to run
1057 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1058 // will hang or crash. When detachCurrentThread is called the key will be set
1059 // to null and we will not be called again. If detachCurrentThread is never
1060 // called we could loop forever depending on the pthread implementation.
1061 static void restore_thread_pointer(void* p) {
1062 Thread* thread = (Thread*) p;
1063 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1064 }
1065
1066 int os::allocate_thread_local_storage() {
1067 pthread_key_t key;
1068 int rslt = pthread_key_create(&key, restore_thread_pointer);
1069 assert(rslt == 0, "cannot allocate thread local storage");
1070 return (int)key;
1071 }
1072
1073 // Note: This is currently not used by VM, as we don't destroy TLS key
1074 // on VM exit.
1075 void os::free_thread_local_storage(int index) {
1076 int rslt = pthread_key_delete((pthread_key_t)index);
1077 assert(rslt == 0, "invalid index");
1078 }
1079
1080 void os::thread_local_storage_at_put(int index, void* value) {
1081 int rslt = pthread_setspecific((pthread_key_t)index, value);
1082 assert(rslt == 0, "pthread_setspecific failed");
1083 }
1084
1085 extern "C" Thread* get_thread() {
1086 return ThreadLocalStorage::thread();
1087 }
1088
1089 //////////////////////////////////////////////////////////////////////////////
1090 // primordial thread
1091
1092 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1093 bool os::is_primordial_thread(void) {
1094 char dummy;
1095 // If called before init complete, thread stack bottom will be null.
1096 // Can be called if fatal error occurs before initialization.
1097 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1098 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1099 os::Linux::initial_thread_stack_size() != 0,
1100 "os::init did not locate primordial thread's stack region");
1101 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1102 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1103 os::Linux::initial_thread_stack_size()) {
1104 return true;
1105 } else {
1106 return false;
1107 }
1108 }
1109
1110 // Find the virtual memory area that contains addr
1111 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1112 FILE *fp = fopen("/proc/self/maps", "r");
1113 if (fp) {
1114 address low, high;
1115 while (!feof(fp)) {
1116 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1117 if (low <= addr && addr < high) {
1118 if (vma_low) *vma_low = low;
1119 if (vma_high) *vma_high = high;
1120 fclose (fp);
1121 return true;
1122 }
1123 }
1124 for (;;) {
1125 int ch = fgetc(fp);
1126 if (ch == EOF || ch == (int)'\n') break;
1127 }
1128 }
1129 fclose(fp);
1130 }
1131 return false;
1132 }
1133
1134 // Locate primordial thread stack. This special handling of primordial thread stack
1135 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1136 // bogus value for the primordial process thread. While the launcher has created
1137 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1138 // JNI invocation API from a primordial thread.
1139 void os::Linux::capture_initial_stack(size_t max_size) {
1140
1141 // max_size is either 0 (which means accept OS default for thread stacks) or
1142 // a user-specified value known to be at least the minimum needed. If we
1143 // are actually on the primordial thread we can make it appear that we have a
1144 // smaller max_size stack by inserting the guard pages at that location. But we
1145 // cannot do anything to emulate a larger stack than what has been provided by
1146 // the OS or threading library. In fact if we try to use a stack greater than
1147 // what is set by rlimit then we will crash the hosting process.
1148
1149 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1150 // If this is "unlimited" then it will be a huge value.
1151 struct rlimit rlim;
1152 getrlimit(RLIMIT_STACK, &rlim);
1153 size_t stack_size = rlim.rlim_cur;
1154
1155 // 6308388: a bug in ld.so will relocate its own .data section to the
1156 // lower end of primordial stack; reduce ulimit -s value a little bit
1157 // so we won't install guard page on ld.so's data section.
1158 // But ensure we don't underflow the stack size - allow 1 page spare
1159 if (stack_size >= (size_t)(3 * page_size())) {
1160 stack_size -= 2 * page_size();
1161 }
1162
1163 // Try to figure out where the stack base (top) is. This is harder.
1164 //
1165 // When an application is started, glibc saves the initial stack pointer in
1166 // a global variable "__libc_stack_end", which is then used by system
1167 // libraries. __libc_stack_end should be pretty close to stack top. The
1168 // variable is available since the very early days. However, because it is
1169 // a private interface, it could disappear in the future.
1170 //
1171 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1172 // to __libc_stack_end, it is very close to stack top, but isn't the real
1173 // stack top. Note that /proc may not exist if VM is running as a chroot
1174 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1175 // /proc/<pid>/stat could change in the future (though unlikely).
1176 //
1177 // We try __libc_stack_end first. If that doesn't work, look for
1178 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1179 // as a hint, which should work well in most cases.
1180
1181 uintptr_t stack_start;
1182
1183 // try __libc_stack_end first
1184 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1185 if (p && *p) {
1186 stack_start = *p;
1187 } else {
1188 // see if we can get the start_stack field from /proc/self/stat
1189 FILE *fp;
1190 int pid;
1191 char state;
1192 int ppid;
1193 int pgrp;
1194 int session;
1195 int nr;
1196 int tpgrp;
1197 unsigned long flags;
1198 unsigned long minflt;
1199 unsigned long cminflt;
1200 unsigned long majflt;
1201 unsigned long cmajflt;
1202 unsigned long utime;
1203 unsigned long stime;
1204 long cutime;
1205 long cstime;
1206 long prio;
1207 long nice;
1208 long junk;
1209 long it_real;
1210 uintptr_t start;
1211 uintptr_t vsize;
1212 intptr_t rss;
1213 uintptr_t rsslim;
1214 uintptr_t scodes;
1215 uintptr_t ecode;
1216 int i;
1217
1218 // Figure what the primordial thread stack base is. Code is inspired
1219 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1220 // followed by command name surrounded by parentheses, state, etc.
1221 char stat[2048];
1222 int statlen;
1223
1224 fp = fopen("/proc/self/stat", "r");
1225 if (fp) {
1226 statlen = fread(stat, 1, 2047, fp);
1227 stat[statlen] = '\0';
1228 fclose(fp);
1229
1230 // Skip pid and the command string. Note that we could be dealing with
1231 // weird command names, e.g. user could decide to rename java launcher
1232 // to "java 1.4.2 :)", then the stat file would look like
1233 // 1234 (java 1.4.2 :)) R ... ...
1234 // We don't really need to know the command string, just find the last
1235 // occurrence of ")" and then start parsing from there. See bug 4726580.
1236 char * s = strrchr(stat, ')');
1237
1238 i = 0;
1239 if (s) {
1240 // Skip blank chars
1241 do s++; while (isspace(*s));
1242
1243 #define _UFM UINTX_FORMAT
1244 #define _DFM INTX_FORMAT
1245
1246 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1247 /* 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 */
1248 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,
1249 &state, /* 3 %c */
1250 &ppid, /* 4 %d */
1251 &pgrp, /* 5 %d */
1252 &session, /* 6 %d */
1253 &nr, /* 7 %d */
1254 &tpgrp, /* 8 %d */
1255 &flags, /* 9 %lu */
1256 &minflt, /* 10 %lu */
1257 &cminflt, /* 11 %lu */
1258 &majflt, /* 12 %lu */
1259 &cmajflt, /* 13 %lu */
1260 &utime, /* 14 %lu */
1261 &stime, /* 15 %lu */
1262 &cutime, /* 16 %ld */
1263 &cstime, /* 17 %ld */
1264 &prio, /* 18 %ld */
1265 &nice, /* 19 %ld */
1266 &junk, /* 20 %ld */
1267 &it_real, /* 21 %ld */
1268 &start, /* 22 UINTX_FORMAT */
1269 &vsize, /* 23 UINTX_FORMAT */
1270 &rss, /* 24 INTX_FORMAT */
1271 &rsslim, /* 25 UINTX_FORMAT */
1272 &scodes, /* 26 UINTX_FORMAT */
1273 &ecode, /* 27 UINTX_FORMAT */
1274 &stack_start); /* 28 UINTX_FORMAT */
1275 }
1276
1277 #undef _UFM
1278 #undef _DFM
1279
1280 if (i != 28 - 2) {
1281 assert(false, "Bad conversion from /proc/self/stat");
1282 // product mode - assume we are the primordial thread, good luck in the
1283 // embedded case.
1284 warning("Can't detect primordial thread stack location - bad conversion");
1285 stack_start = (uintptr_t) &rlim;
1286 }
1287 } else {
1288 // For some reason we can't open /proc/self/stat (for example, running on
1289 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1290 // most cases, so don't abort:
1291 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1292 stack_start = (uintptr_t) &rlim;
1293 }
1294 }
1295
1296 // Now we have a pointer (stack_start) very close to the stack top, the
1297 // next thing to do is to figure out the exact location of stack top. We
1298 // can find out the virtual memory area that contains stack_start by
1299 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1300 // and its upper limit is the real stack top. (again, this would fail if
1301 // running inside chroot, because /proc may not exist.)
1302
1303 uintptr_t stack_top;
1304 address low, high;
1305 if (find_vma((address)stack_start, &low, &high)) {
1306 // success, "high" is the true stack top. (ignore "low", because initial
1307 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1308 stack_top = (uintptr_t)high;
1309 } else {
1310 // failed, likely because /proc/self/maps does not exist
1311 warning("Can't detect primordial thread stack location - find_vma failed");
1312 // best effort: stack_start is normally within a few pages below the real
1313 // stack top, use it as stack top, and reduce stack size so we won't put
1314 // guard page outside stack.
1315 stack_top = stack_start;
1316 stack_size -= 16 * page_size();
1317 }
1318
1319 // stack_top could be partially down the page so align it
1320 stack_top = align_size_up(stack_top, page_size());
1321
1322 // Allowed stack value is minimum of max_size and what we derived from rlimit
1323 if (max_size > 0) {
1324 _initial_thread_stack_size = MIN2(max_size, stack_size);
1325 } else {
1326 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1327 // clamp it at 8MB as we do on Solaris
1328 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1329 }
1330
1331 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1332 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1333 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1334 }
1335
1336 ////////////////////////////////////////////////////////////////////////////////
1337 // time support
1338
1339 // Time since start-up in seconds to a fine granularity.
1340 // Used by VMSelfDestructTimer and the MemProfiler.
1341 double os::elapsedTime() {
1342
1343 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1344 }
1345
1346 jlong os::elapsed_counter() {
1347 return javaTimeNanos() - initial_time_count;
1348 }
1349
1350 jlong os::elapsed_frequency() {
1351 return NANOSECS_PER_SEC; // nanosecond resolution
1352 }
1353
1354 bool os::supports_vtime() { return true; }
1355 bool os::enable_vtime() { return false; }
1356 bool os::vtime_enabled() { return false; }
1357
1358 double os::elapsedVTime() {
1359 struct rusage usage;
1360 int retval = getrusage(RUSAGE_THREAD, &usage);
1361 if (retval == 0) {
1362 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);
1363 } else {
1364 // better than nothing, but not much
1365 return elapsedTime();
1366 }
1367 }
1368
1369 jlong os::javaTimeMillis() {
1370 timeval time;
1371 int status = gettimeofday(&time, NULL);
1372 assert(status != -1, "linux error");
1373 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1374 }
1375
1376 #ifndef CLOCK_MONOTONIC
1377 #define CLOCK_MONOTONIC (1)
1378 #endif
1379
1380 void os::Linux::clock_init() {
1381 // we do dlopen's in this particular order due to bug in linux
1382 // dynamical loader (see 6348968) leading to crash on exit
1383 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1384 if (handle == NULL) {
1385 handle = dlopen("librt.so", RTLD_LAZY);
1386 }
1387
1388 if (handle) {
1389 int (*clock_getres_func)(clockid_t, struct timespec*) =
1390 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1391 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1392 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1393 if (clock_getres_func && clock_gettime_func) {
1394 // See if monotonic clock is supported by the kernel. Note that some
1395 // early implementations simply return kernel jiffies (updated every
1396 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1397 // for nano time (though the monotonic property is still nice to have).
1398 // It's fixed in newer kernels, however clock_getres() still returns
1399 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1400 // resolution for now. Hopefully as people move to new kernels, this
1401 // won't be a problem.
1402 struct timespec res;
1403 struct timespec tp;
1404 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1405 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1406 // yes, monotonic clock is supported
1407 _clock_gettime = clock_gettime_func;
1408 return;
1409 } else {
1410 // close librt if there is no monotonic clock
1411 dlclose(handle);
1412 }
1413 }
1414 }
1415 warning("No monotonic clock was available - timed services may " \
1416 "be adversely affected if the time-of-day clock changes");
1417 }
1418
1419 #ifndef SYS_clock_getres
1420
1421 #if defined(IA32) || defined(AMD64)
1422 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1423 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1424 #else
1425 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1426 #define sys_clock_getres(x,y) -1
1427 #endif
1428
1429 #else
1430 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1431 #endif
1432
1433 void os::Linux::fast_thread_clock_init() {
1434 if (!UseLinuxPosixThreadCPUClocks) {
1435 return;
1436 }
1437 clockid_t clockid;
1438 struct timespec tp;
1439 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1440 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1441
1442 // Switch to using fast clocks for thread cpu time if
1443 // the sys_clock_getres() returns 0 error code.
1444 // Note, that some kernels may support the current thread
1445 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1446 // returned by the pthread_getcpuclockid().
1447 // If the fast Posix clocks are supported then the sys_clock_getres()
1448 // must return at least tp.tv_sec == 0 which means a resolution
1449 // better than 1 sec. This is extra check for reliability.
1450
1451 if(pthread_getcpuclockid_func &&
1452 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1453 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1454
1455 _supports_fast_thread_cpu_time = true;
1456 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1457 }
1458 }
1459
1460 jlong os::javaTimeNanos() {
1461 if (Linux::supports_monotonic_clock()) {
1462 struct timespec tp;
1463 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1464 assert(status == 0, "gettime error");
1465 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1466 return result;
1467 } else {
1468 timeval time;
1469 int status = gettimeofday(&time, NULL);
1470 assert(status != -1, "linux error");
1471 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1472 return 1000 * usecs;
1473 }
1474 }
1475
1476 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1477 if (Linux::supports_monotonic_clock()) {
1478 info_ptr->max_value = ALL_64_BITS;
1479
1480 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1481 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1482 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1483 } else {
1484 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1485 info_ptr->max_value = ALL_64_BITS;
1486
1487 // gettimeofday is a real time clock so it skips
1488 info_ptr->may_skip_backward = true;
1489 info_ptr->may_skip_forward = true;
1490 }
1491
1492 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1493 }
1494
1495 // Return the real, user, and system times in seconds from an
1496 // arbitrary fixed point in the past.
1497 bool os::getTimesSecs(double* process_real_time,
1498 double* process_user_time,
1499 double* process_system_time) {
1500 struct tms ticks;
1501 clock_t real_ticks = times(&ticks);
1502
1503 if (real_ticks == (clock_t) (-1)) {
1504 return false;
1505 } else {
1506 double ticks_per_second = (double) clock_tics_per_sec;
1507 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1508 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1509 *process_real_time = ((double) real_ticks) / ticks_per_second;
1510
1511 return true;
1512 }
1513 }
1514
1515
1516 char * os::local_time_string(char *buf, size_t buflen) {
1517 struct tm t;
1518 time_t long_time;
1519 time(&long_time);
1520 localtime_r(&long_time, &t);
1521 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1522 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1523 t.tm_hour, t.tm_min, t.tm_sec);
1524 return buf;
1525 }
1526
1527 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1528 return localtime_r(clock, res);
1529 }
1530
1531 ////////////////////////////////////////////////////////////////////////////////
1532 // runtime exit support
1533
1534 // Note: os::shutdown() might be called very early during initialization, or
1535 // called from signal handler. Before adding something to os::shutdown(), make
1536 // sure it is async-safe and can handle partially initialized VM.
1537 void os::shutdown() {
1538
1539 // allow PerfMemory to attempt cleanup of any persistent resources
1540 perfMemory_exit();
1541
1542 // needs to remove object in file system
1543 AttachListener::abort();
1544
1545 // flush buffered output, finish log files
1546 ostream_abort();
1547
1548 // Check for abort hook
1549 abort_hook_t abort_hook = Arguments::abort_hook();
1550 if (abort_hook != NULL) {
1551 abort_hook();
1552 }
1553
1554 }
1555
1556 // Note: os::abort() might be called very early during initialization, or
1557 // called from signal handler. Before adding something to os::abort(), make
1558 // sure it is async-safe and can handle partially initialized VM.
1559 void os::abort(bool dump_core) {
1560 os::shutdown();
1561 if (dump_core) {
1562 #ifndef PRODUCT
1563 fdStream out(defaultStream::output_fd());
1564 out.print_raw("Current thread is ");
1565 char buf[16];
1566 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1567 out.print_raw_cr(buf);
1568 out.print_raw_cr("Dumping core ...");
1569 #endif
1570 ::abort(); // dump core
1571 }
1572
1573 ::exit(1);
1574 }
1575
1576 // Die immediately, no exit hook, no abort hook, no cleanup.
1577 void os::die() {
1578 // _exit() on LinuxThreads only kills current thread
1579 ::abort();
1580 }
1581
1582
1583 // This method is a copy of JDK's sysGetLastErrorString
1584 // from src/solaris/hpi/src/system_md.c
1585
1586 size_t os::lasterror(char *buf, size_t len) {
1587
1588 if (errno == 0) return 0;
1589
1590 const char *s = ::strerror(errno);
1591 size_t n = ::strlen(s);
1592 if (n >= len) {
1593 n = len - 1;
1594 }
1595 ::strncpy(buf, s, n);
1596 buf[n] = '\0';
1597 return n;
1598 }
1599
1600 intx os::current_thread_id() { return (intx)pthread_self(); }
1601 int os::current_process_id() {
1602
1603 // Under the old linux thread library, linux gives each thread
1604 // its own process id. Because of this each thread will return
1605 // a different pid if this method were to return the result
1606 // of getpid(2). Linux provides no api that returns the pid
1607 // of the launcher thread for the vm. This implementation
1608 // returns a unique pid, the pid of the launcher thread
1609 // that starts the vm 'process'.
1610
1611 // Under the NPTL, getpid() returns the same pid as the
1612 // launcher thread rather than a unique pid per thread.
1613 // Use gettid() if you want the old pre NPTL behaviour.
1614
1615 // if you are looking for the result of a call to getpid() that
1616 // returns a unique pid for the calling thread, then look at the
1617 // OSThread::thread_id() method in osThread_linux.hpp file
1618
1619 return (int)(_initial_pid ? _initial_pid : getpid());
1620 }
1621
1622 // DLL functions
1623
1624 const char* os::dll_file_extension() { return ".so"; }
1625
1626 // This must be hard coded because it's the system's temporary
1627 // directory not the java application's temp directory, ala java.io.tmpdir.
1628 const char* os::get_temp_directory() { return "/tmp"; }
1629
1630 static bool file_exists(const char* filename) {
1631 struct stat statbuf;
1632 if (filename == NULL || strlen(filename) == 0) {
1633 return false;
1634 }
1635 return os::stat(filename, &statbuf) == 0;
1636 }
1637
1638 bool os::dll_build_name(char* buffer, size_t buflen,
1639 const char* pname, const char* fname) {
1640 bool retval = false;
1641 // Copied from libhpi
1642 const size_t pnamelen = pname ? strlen(pname) : 0;
1643
1644 // Return error on buffer overflow.
1645 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1646 return retval;
1647 }
1648
1649 if (pnamelen == 0) {
1650 snprintf(buffer, buflen, "lib%s.so", fname);
1651 retval = true;
1652 } else if (strchr(pname, *os::path_separator()) != NULL) {
1653 int n;
1654 char** pelements = split_path(pname, &n);
1655 if (pelements == NULL) {
1656 return false;
1657 }
1658 for (int i = 0 ; i < n ; i++) {
1659 // Really shouldn't be NULL, but check can't hurt
1660 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1661 continue; // skip the empty path values
1662 }
1663 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1664 if (file_exists(buffer)) {
1665 retval = true;
1666 break;
1667 }
1668 }
1669 // release the storage
1670 for (int i = 0 ; i < n ; i++) {
1671 if (pelements[i] != NULL) {
1672 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1673 }
1674 }
1675 if (pelements != NULL) {
1676 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1677 }
1678 } else {
1679 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1680 retval = true;
1681 }
1682 return retval;
1683 }
1684
1685 // check if addr is inside libjvm.so
1686 bool os::address_is_in_vm(address addr) {
1687 static address libjvm_base_addr;
1688 Dl_info dlinfo;
1689
1690 if (libjvm_base_addr == NULL) {
1691 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1692 libjvm_base_addr = (address)dlinfo.dli_fbase;
1693 }
1694 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1695 }
1696
1697 if (dladdr((void *)addr, &dlinfo) != 0) {
1698 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1699 }
1700
1701 return false;
1702 }
1703
1704 bool os::dll_address_to_function_name(address addr, char *buf,
1705 int buflen, int *offset) {
1706 // buf is not optional, but offset is optional
1707 assert(buf != NULL, "sanity check");
1708
1709 Dl_info dlinfo;
1710
1711 if (dladdr((void*)addr, &dlinfo) != 0) {
1712 // see if we have a matching symbol
1713 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1714 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1715 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1716 }
1717 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1718 return true;
1719 }
1720 // no matching symbol so try for just file info
1721 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1722 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1723 buf, buflen, offset, dlinfo.dli_fname)) {
1724 return true;
1725 }
1726 }
1727 }
1728
1729 buf[0] = '\0';
1730 if (offset != NULL) *offset = -1;
1731 return false;
1732 }
1733
1734 struct _address_to_library_name {
1735 address addr; // input : memory address
1736 size_t buflen; // size of fname
1737 char* fname; // output: library name
1738 address base; // library base addr
1739 };
1740
1741 static int address_to_library_name_callback(struct dl_phdr_info *info,
1742 size_t size, void *data) {
1743 int i;
1744 bool found = false;
1745 address libbase = NULL;
1746 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1747
1748 // iterate through all loadable segments
1749 for (i = 0; i < info->dlpi_phnum; i++) {
1750 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1751 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1752 // base address of a library is the lowest address of its loaded
1753 // segments.
1754 if (libbase == NULL || libbase > segbase) {
1755 libbase = segbase;
1756 }
1757 // see if 'addr' is within current segment
1758 if (segbase <= d->addr &&
1759 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1760 found = true;
1761 }
1762 }
1763 }
1764
1765 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1766 // so dll_address_to_library_name() can fall through to use dladdr() which
1767 // can figure out executable name from argv[0].
1768 if (found && info->dlpi_name && info->dlpi_name[0]) {
1769 d->base = libbase;
1770 if (d->fname) {
1771 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1772 }
1773 return 1;
1774 }
1775 return 0;
1776 }
1777
1778 bool os::dll_address_to_library_name(address addr, char* buf,
1779 int buflen, int* offset) {
1780 // buf is not optional, but offset is optional
1781 assert(buf != NULL, "sanity check");
1782
1783 Dl_info dlinfo;
1784 struct _address_to_library_name data;
1785
1786 // There is a bug in old glibc dladdr() implementation that it could resolve
1787 // to wrong library name if the .so file has a base address != NULL. Here
1788 // we iterate through the program headers of all loaded libraries to find
1789 // out which library 'addr' really belongs to. This workaround can be
1790 // removed once the minimum requirement for glibc is moved to 2.3.x.
1791 data.addr = addr;
1792 data.fname = buf;
1793 data.buflen = buflen;
1794 data.base = NULL;
1795 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1796
1797 if (rslt) {
1798 // buf already contains library name
1799 if (offset) *offset = addr - data.base;
1800 return true;
1801 }
1802 if (dladdr((void*)addr, &dlinfo) != 0) {
1803 if (dlinfo.dli_fname != NULL) {
1804 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1805 }
1806 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1807 *offset = addr - (address)dlinfo.dli_fbase;
1808 }
1809 return true;
1810 }
1811
1812 buf[0] = '\0';
1813 if (offset) *offset = -1;
1814 return false;
1815 }
1816
1817 // Loads .dll/.so and
1818 // in case of error it checks if .dll/.so was built for the
1819 // same architecture as Hotspot is running on
1820
1821
1822 // Remember the stack's state. The Linux dynamic linker will change
1823 // the stack to 'executable' at most once, so we must safepoint only once.
1824 bool os::Linux::_stack_is_executable = false;
1825
1826 // VM operation that loads a library. This is necessary if stack protection
1827 // of the Java stacks can be lost during loading the library. If we
1828 // do not stop the Java threads, they can stack overflow before the stacks
1829 // are protected again.
1830 class VM_LinuxDllLoad: public VM_Operation {
1831 private:
1832 const char *_filename;
1833 char *_ebuf;
1834 int _ebuflen;
1835 void *_lib;
1836 public:
1837 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1838 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1839 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1840 void doit() {
1841 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1842 os::Linux::_stack_is_executable = true;
1843 }
1844 void* loaded_library() { return _lib; }
1845 };
1846
1847 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1848 {
1849 void * result = NULL;
1850 bool load_attempted = false;
1851
1852 // Check whether the library to load might change execution rights
1853 // of the stack. If they are changed, the protection of the stack
1854 // guard pages will be lost. We need a safepoint to fix this.
1855 //
1856 // See Linux man page execstack(8) for more info.
1857 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1858 ElfFile ef(filename);
1859 if (!ef.specifies_noexecstack()) {
1860 if (!is_init_completed()) {
1861 os::Linux::_stack_is_executable = true;
1862 // This is OK - No Java threads have been created yet, and hence no
1863 // stack guard pages to fix.
1864 //
1865 // This should happen only when you are building JDK7 using a very
1866 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1867 //
1868 // Dynamic loader will make all stacks executable after
1869 // this function returns, and will not do that again.
1870 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1871 } else {
1872 warning("You have loaded library %s which might have disabled stack guard. "
1873 "The VM will try to fix the stack guard now.\n"
1874 "It's highly recommended that you fix the library with "
1875 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1876 filename);
1877
1878 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1879 JavaThread *jt = JavaThread::current();
1880 if (jt->thread_state() != _thread_in_native) {
1881 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1882 // that requires ExecStack. Cannot enter safe point. Let's give up.
1883 warning("Unable to fix stack guard. Giving up.");
1884 } else {
1885 if (!LoadExecStackDllInVMThread) {
1886 // This is for the case where the DLL has an static
1887 // constructor function that executes JNI code. We cannot
1888 // load such DLLs in the VMThread.
1889 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1890 }
1891
1892 ThreadInVMfromNative tiv(jt);
1893 debug_only(VMNativeEntryWrapper vew;)
1894
1895 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1896 VMThread::execute(&op);
1897 if (LoadExecStackDllInVMThread) {
1898 result = op.loaded_library();
1899 }
1900 load_attempted = true;
1901 }
1902 }
1903 }
1904 }
1905
1906 if (!load_attempted) {
1907 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1908 }
1909
1910 if (result != NULL) {
1911 // Successful loading
1912 return result;
1913 }
1914
1915 Elf32_Ehdr elf_head;
1916 int diag_msg_max_length=ebuflen-strlen(ebuf);
1917 char* diag_msg_buf=ebuf+strlen(ebuf);
1918
1919 if (diag_msg_max_length==0) {
1920 // No more space in ebuf for additional diagnostics message
1921 return NULL;
1922 }
1923
1924
1925 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1926
1927 if (file_descriptor < 0) {
1928 // Can't open library, report dlerror() message
1929 return NULL;
1930 }
1931
1932 bool failed_to_read_elf_head=
1933 (sizeof(elf_head)!=
1934 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1935
1936 ::close(file_descriptor);
1937 if (failed_to_read_elf_head) {
1938 // file i/o error - report dlerror() msg
1939 return NULL;
1940 }
1941
1942 typedef struct {
1943 Elf32_Half code; // Actual value as defined in elf.h
1944 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1945 char elf_class; // 32 or 64 bit
1946 char endianess; // MSB or LSB
1947 char* name; // String representation
1948 } arch_t;
1949
1950 #ifndef EM_486
1951 #define EM_486 6 /* Intel 80486 */
1952 #endif
1953 #ifndef EM_AARCH64
1954 #define EM_AARCH64 183 /* ARM AARCH64 */
1955 #endif
1956
1957 static const arch_t arch_array[]={
1958 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1959 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1960 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1961 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1962 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1963 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1964 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1965 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1966 #if defined(VM_LITTLE_ENDIAN)
1967 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1968 #else
1969 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1970 #endif
1971 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1972 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1973 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1974 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1975 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1976 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1977 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1978 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1979 };
1980
1981 #if (defined IA32)
1982 static Elf32_Half running_arch_code=EM_386;
1983 #elif (defined AMD64)
1984 static Elf32_Half running_arch_code=EM_X86_64;
1985 #elif (defined IA64)
1986 static Elf32_Half running_arch_code=EM_IA_64;
1987 #elif (defined __sparc) && (defined _LP64)
1988 static Elf32_Half running_arch_code=EM_SPARCV9;
1989 #elif (defined __sparc) && (!defined _LP64)
1990 static Elf32_Half running_arch_code=EM_SPARC;
1991 #elif (defined __powerpc64__)
1992 static Elf32_Half running_arch_code=EM_PPC64;
1993 #elif (defined __powerpc__)
1994 static Elf32_Half running_arch_code=EM_PPC;
1995 #elif (defined ARM)
1996 static Elf32_Half running_arch_code=EM_ARM;
1997 #elif (defined S390)
1998 static Elf32_Half running_arch_code=EM_S390;
1999 #elif (defined ALPHA)
2000 static Elf32_Half running_arch_code=EM_ALPHA;
2001 #elif (defined MIPSEL)
2002 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2003 #elif (defined PARISC)
2004 static Elf32_Half running_arch_code=EM_PARISC;
2005 #elif (defined MIPS)
2006 static Elf32_Half running_arch_code=EM_MIPS;
2007 #elif (defined M68K)
2008 static Elf32_Half running_arch_code=EM_68K;
2009 #elif (defined AARCH64)
2010 static Elf32_Half running_arch_code=EM_AARCH64;
2011 #else
2012 #error Method os::dll_load requires that one of following is defined:\
2013 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2014 #endif
2015
2016 // Identify compatability class for VM's architecture and library's architecture
2017 // Obtain string descriptions for architectures
2018
2019 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2020 int running_arch_index=-1;
2021
2022 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2023 if (running_arch_code == arch_array[i].code) {
2024 running_arch_index = i;
2025 }
2026 if (lib_arch.code == arch_array[i].code) {
2027 lib_arch.compat_class = arch_array[i].compat_class;
2028 lib_arch.name = arch_array[i].name;
2029 }
2030 }
2031
2032 assert(running_arch_index != -1,
2033 "Didn't find running architecture code (running_arch_code) in arch_array");
2034 if (running_arch_index == -1) {
2035 // Even though running architecture detection failed
2036 // we may still continue with reporting dlerror() message
2037 return NULL;
2038 }
2039
2040 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2041 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2042 return NULL;
2043 }
2044
2045 #ifndef S390
2046 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2047 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2048 return NULL;
2049 }
2050 #endif // !S390
2051
2052 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2053 if ( lib_arch.name!=NULL ) {
2054 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2055 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2056 lib_arch.name, arch_array[running_arch_index].name);
2057 } else {
2058 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2059 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2060 lib_arch.code,
2061 arch_array[running_arch_index].name);
2062 }
2063 }
2064
2065 return NULL;
2066 }
2067
2068 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2069 void * result = ::dlopen(filename, RTLD_LAZY);
2070 if (result == NULL) {
2071 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2072 ebuf[ebuflen-1] = '\0';
2073 }
2074 return result;
2075 }
2076
2077 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2078 void * result = NULL;
2079 if (LoadExecStackDllInVMThread) {
2080 result = dlopen_helper(filename, ebuf, ebuflen);
2081 }
2082
2083 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2084 // library that requires an executable stack, or which does not have this
2085 // stack attribute set, dlopen changes the stack attribute to executable. The
2086 // read protection of the guard pages gets lost.
2087 //
2088 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2089 // may have been queued at the same time.
2090
2091 if (!_stack_is_executable) {
2092 JavaThread *jt = Threads::first();
2093
2094 while (jt) {
2095 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2096 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2097 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2098 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2099 warning("Attempt to reguard stack yellow zone failed.");
2100 }
2101 }
2102 jt = jt->next();
2103 }
2104 }
2105
2106 return result;
2107 }
2108
2109 /*
2110 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2111 * chances are you might want to run the generated bits against glibc-2.0
2112 * libdl.so, so always use locking for any version of glibc.
2113 */
2114 void* os::dll_lookup(void* handle, const char* name) {
2115 pthread_mutex_lock(&dl_mutex);
2116 void* res = dlsym(handle, name);
2117 pthread_mutex_unlock(&dl_mutex);
2118 return res;
2119 }
2120
2121 void* os::get_default_process_handle() {
2122 return (void*)::dlopen(NULL, RTLD_LAZY);
2123 }
2124
2125 static bool _print_ascii_file(const char* filename, outputStream* st) {
2126 int fd = ::open(filename, O_RDONLY);
2127 if (fd == -1) {
2128 return false;
2129 }
2130
2131 char buf[32];
2132 int bytes;
2133 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2134 st->print_raw(buf, bytes);
2135 }
2136
2137 ::close(fd);
2138
2139 return true;
2140 }
2141
2142 void os::print_dll_info(outputStream *st) {
2143 st->print_cr("Dynamic libraries:");
2144
2145 char fname[32];
2146 pid_t pid = os::Linux::gettid();
2147
2148 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2149
2150 if (!_print_ascii_file(fname, st)) {
2151 st->print("Can not get library information for pid = %d\n", pid);
2152 }
2153 }
2154
2155 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2156 FILE *procmapsFile = NULL;
2157
2158 // Open the procfs maps file for the current process
2159 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2160 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2161 char line[PATH_MAX + 100];
2162
2163 // Read line by line from 'file'
2164 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2165 u8 base, top, offset, inode;
2166 char permissions[5];
2167 char device[6];
2168 char name[PATH_MAX + 1];
2169
2170 // Parse fields from line
2171 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2172 &base, &top, permissions, &offset, device, &inode, name);
2173
2174 // Filter by device id '00:00' so that we only get file system mapped files.
2175 if (strcmp(device, "00:00") != 0) {
2176
2177 // Call callback with the fields of interest
2178 if(callback(name, (address)base, (address)top, param)) {
2179 // Oops abort, callback aborted
2180 fclose(procmapsFile);
2181 return 1;
2182 }
2183 }
2184 }
2185 fclose(procmapsFile);
2186 }
2187 return 0;
2188 }
2189
2190 void os::print_os_info_brief(outputStream* st) {
2191 os::Linux::print_distro_info(st);
2192
2193 os::Posix::print_uname_info(st);
2194
2195 os::Linux::print_libversion_info(st);
2196
2197 }
2198
2199 void os::print_os_info(outputStream* st) {
2200 st->print("OS:");
2201
2202 os::Linux::print_distro_info(st);
2203
2204 os::Posix::print_uname_info(st);
2205
2206 // Print warning if unsafe chroot environment detected
2207 if (unsafe_chroot_detected) {
2208 st->print("WARNING!! ");
2209 st->print_cr("%s", unstable_chroot_error);
2210 }
2211
2212 os::Linux::print_libversion_info(st);
2213
2214 os::Posix::print_rlimit_info(st);
2215
2216 os::Posix::print_load_average(st);
2217
2218 os::Linux::print_full_memory_info(st);
2219
2220 os::Linux::print_container_info(st);
2221 }
2222
2223 // Try to identify popular distros.
2224 // Most Linux distributions have a /etc/XXX-release file, which contains
2225 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2226 // file that also contains the OS version string. Some have more than one
2227 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2228 // /etc/redhat-release.), so the order is important.
2229 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2230 // their own specific XXX-release file as well as a redhat-release file.
2231 // Because of this the XXX-release file needs to be searched for before the
2232 // redhat-release file.
2233 // Since Red Hat has a lsb-release file that is not very descriptive the
2234 // search for redhat-release needs to be before lsb-release.
2235 // Since the lsb-release file is the new standard it needs to be searched
2236 // before the older style release files.
2237 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2238 // next to last resort. The os-release file is a new standard that contains
2239 // distribution information and the system-release file seems to be an old
2240 // standard that has been replaced by the lsb-release and os-release files.
2241 // Searching for the debian_version file is the last resort. It contains
2242 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2243 // "Debian " is printed before the contents of the debian_version file.
2244 void os::Linux::print_distro_info(outputStream* st) {
2245 if (!_print_ascii_file("/etc/oracle-release", st) &&
2246 !_print_ascii_file("/etc/mandriva-release", st) &&
2247 !_print_ascii_file("/etc/mandrake-release", st) &&
2248 !_print_ascii_file("/etc/sun-release", st) &&
2249 !_print_ascii_file("/etc/redhat-release", st) &&
2250 !_print_ascii_file("/etc/lsb-release", st) &&
2251 !_print_ascii_file("/etc/SuSE-release", st) &&
2252 !_print_ascii_file("/etc/turbolinux-release", st) &&
2253 !_print_ascii_file("/etc/gentoo-release", st) &&
2254 !_print_ascii_file("/etc/ltib-release", st) &&
2255 !_print_ascii_file("/etc/angstrom-version", st) &&
2256 !_print_ascii_file("/etc/system-release", st) &&
2257 !_print_ascii_file("/etc/os-release", st)) {
2258
2259 if (file_exists("/etc/debian_version")) {
2260 st->print("Debian ");
2261 _print_ascii_file("/etc/debian_version", st);
2262 } else {
2263 st->print("Linux");
2264 }
2265 }
2266 st->cr();
2267 }
2268
2269 void os::Linux::print_libversion_info(outputStream* st) {
2270 // libc, pthread
2271 st->print("libc:");
2272 st->print("%s ", os::Linux::glibc_version());
2273 st->print("%s ", os::Linux::libpthread_version());
2274 if (os::Linux::is_LinuxThreads()) {
2275 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2276 }
2277 st->cr();
2278 }
2279
2280 void os::Linux::print_full_memory_info(outputStream* st) {
2281 st->print("\n/proc/meminfo:\n");
2282 _print_ascii_file("/proc/meminfo", st);
2283 st->cr();
2284 }
2285
2286 void os::Linux::print_container_info(outputStream* st) {
2287 if (!OSContainer::is_containerized()) {
2288 return;
2289 }
2290
2291 st->print("container (cgroup) information:\n");
2292
2293 const char *p_ct = OSContainer::container_type();
2294 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2295
2296 char *p = OSContainer::cpu_cpuset_cpus();
2297 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2298 free(p);
2299
2300 p = OSContainer::cpu_cpuset_memory_nodes();
2301 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2302 free(p);
2303
2304 int i = OSContainer::active_processor_count();
2305 if (i > 0) {
2306 st->print("active_processor_count: %d\n", i);
2307 } else {
2308 st->print("active_processor_count: failed\n");
2309 }
2310
2311 i = OSContainer::cpu_quota();
2312 st->print("cpu_quota: %d\n", i);
2313
2314 i = OSContainer::cpu_period();
2315 st->print("cpu_period: %d\n", i);
2316
2317 i = OSContainer::cpu_shares();
2318 st->print("cpu_shares: %d\n", i);
2319
2320 jlong j = OSContainer::memory_limit_in_bytes();
2321 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2322
2323 j = OSContainer::memory_and_swap_limit_in_bytes();
2324 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2325
2326 j = OSContainer::memory_soft_limit_in_bytes();
2327 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2328
2329 j = OSContainer::OSContainer::memory_usage_in_bytes();
2330 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2331
2332 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2333 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2334 st->cr();
2335 }
2336
2337 void os::print_memory_info(outputStream* st) {
2338
2339 st->print("Memory:");
2340 st->print(" %dk page", os::vm_page_size()>>10);
2341
2342 // values in struct sysinfo are "unsigned long"
2343 struct sysinfo si;
2344 sysinfo(&si);
2345
2346 st->print(", physical " UINT64_FORMAT "k",
2347 os::physical_memory() >> 10);
2348 st->print("(" UINT64_FORMAT "k free)",
2349 os::available_memory() >> 10);
2350 st->print(", swap " UINT64_FORMAT "k",
2351 ((jlong)si.totalswap * si.mem_unit) >> 10);
2352 st->print("(" UINT64_FORMAT "k free)",
2353 ((jlong)si.freeswap * si.mem_unit) >> 10);
2354 st->cr();
2355 }
2356
2357 void os::pd_print_cpu_info(outputStream* st) {
2358 st->print("\n/proc/cpuinfo:\n");
2359 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2360 st->print(" <Not Available>");
2361 }
2362 st->cr();
2363 }
2364
2365 void os::print_siginfo(outputStream* st, void* siginfo) {
2366 const siginfo_t* si = (const siginfo_t*)siginfo;
2367
2368 os::Posix::print_siginfo_brief(st, si);
2369 #if INCLUDE_CDS
2370 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2371 UseSharedSpaces) {
2372 FileMapInfo* mapinfo = FileMapInfo::current_info();
2373 if (mapinfo->is_in_shared_space(si->si_addr)) {
2374 st->print("\n\nError accessing class data sharing archive." \
2375 " Mapped file inaccessible during execution, " \
2376 " possible disk/network problem.");
2377 }
2378 }
2379 #endif
2380 st->cr();
2381 }
2382
2383
2384 static void print_signal_handler(outputStream* st, int sig,
2385 char* buf, size_t buflen);
2386
2387 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2388 st->print_cr("Signal Handlers:");
2389 print_signal_handler(st, SIGSEGV, buf, buflen);
2390 print_signal_handler(st, SIGBUS , buf, buflen);
2391 print_signal_handler(st, SIGFPE , buf, buflen);
2392 print_signal_handler(st, SIGPIPE, buf, buflen);
2393 print_signal_handler(st, SIGXFSZ, buf, buflen);
2394 print_signal_handler(st, SIGILL , buf, buflen);
2395 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2396 print_signal_handler(st, SR_signum, buf, buflen);
2397 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2398 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2399 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2400 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2401 #if defined(PPC64)
2402 print_signal_handler(st, SIGTRAP, buf, buflen);
2403 #endif
2404 }
2405
2406 static char saved_jvm_path[MAXPATHLEN] = {0};
2407
2408 // Find the full path to the current module, libjvm.so
2409 void os::jvm_path(char *buf, jint buflen) {
2410 // Error checking.
2411 if (buflen < MAXPATHLEN) {
2412 assert(false, "must use a large-enough buffer");
2413 buf[0] = '\0';
2414 return;
2415 }
2416 // Lazy resolve the path to current module.
2417 if (saved_jvm_path[0] != 0) {
2418 strcpy(buf, saved_jvm_path);
2419 return;
2420 }
2421
2422 char dli_fname[MAXPATHLEN];
2423 bool ret = dll_address_to_library_name(
2424 CAST_FROM_FN_PTR(address, os::jvm_path),
2425 dli_fname, sizeof(dli_fname), NULL);
2426 assert(ret, "cannot locate libjvm");
2427 char *rp = NULL;
2428 if (ret && dli_fname[0] != '\0') {
2429 rp = realpath(dli_fname, buf);
2430 }
2431 if (rp == NULL)
2432 return;
2433
2434 if (Arguments::created_by_gamma_launcher()) {
2435 // Support for the gamma launcher. Typical value for buf is
2436 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2437 // the right place in the string, then assume we are installed in a JDK and
2438 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2439 // up the path so it looks like libjvm.so is installed there (append a
2440 // fake suffix hotspot/libjvm.so).
2441 const char *p = buf + strlen(buf) - 1;
2442 for (int count = 0; p > buf && count < 5; ++count) {
2443 for (--p; p > buf && *p != '/'; --p)
2444 /* empty */ ;
2445 }
2446
2447 if (strncmp(p, "/jre/lib/", 9) != 0) {
2448 // Look for JAVA_HOME in the environment.
2449 char* java_home_var = ::getenv("JAVA_HOME");
2450 if (java_home_var != NULL && java_home_var[0] != 0) {
2451 char* jrelib_p;
2452 int len;
2453
2454 // Check the current module name "libjvm.so".
2455 p = strrchr(buf, '/');
2456 assert(strstr(p, "/libjvm") == p, "invalid library name");
2457
2458 rp = realpath(java_home_var, buf);
2459 if (rp == NULL)
2460 return;
2461
2462 // determine if this is a legacy image or modules image
2463 // modules image doesn't have "jre" subdirectory
2464 len = strlen(buf);
2465 assert(len < buflen, "Ran out of buffer room");
2466 jrelib_p = buf + len;
2467 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2468 if (0 != access(buf, F_OK)) {
2469 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2470 }
2471
2472 if (0 == access(buf, F_OK)) {
2473 // Use current module name "libjvm.so"
2474 len = strlen(buf);
2475 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2476 } else {
2477 // Go back to path of .so
2478 rp = realpath(dli_fname, buf);
2479 if (rp == NULL)
2480 return;
2481 }
2482 }
2483 }
2484 }
2485
2486 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2487 }
2488
2489 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2490 // no prefix required, not even "_"
2491 }
2492
2493 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2494 // no suffix required
2495 }
2496
2497 ////////////////////////////////////////////////////////////////////////////////
2498 // sun.misc.Signal support
2499
2500 static volatile jint sigint_count = 0;
2501
2502 static void
2503 UserHandler(int sig, void *siginfo, void *context) {
2504 // 4511530 - sem_post is serialized and handled by the manager thread. When
2505 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2506 // don't want to flood the manager thread with sem_post requests.
2507 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2508 return;
2509
2510 // Ctrl-C is pressed during error reporting, likely because the error
2511 // handler fails to abort. Let VM die immediately.
2512 if (sig == SIGINT && is_error_reported()) {
2513 os::die();
2514 }
2515
2516 os::signal_notify(sig);
2517 }
2518
2519 void* os::user_handler() {
2520 return CAST_FROM_FN_PTR(void*, UserHandler);
2521 }
2522
2523 class Semaphore : public StackObj {
2524 public:
2525 Semaphore();
2526 ~Semaphore();
2527 void signal();
2528 void wait();
2529 bool trywait();
2530 bool timedwait(unsigned int sec, int nsec);
2531 private:
2532 sem_t _semaphore;
2533 };
2534
2535 Semaphore::Semaphore() {
2536 sem_init(&_semaphore, 0, 0);
2537 }
2538
2539 Semaphore::~Semaphore() {
2540 sem_destroy(&_semaphore);
2541 }
2542
2543 void Semaphore::signal() {
2544 sem_post(&_semaphore);
2545 }
2546
2547 void Semaphore::wait() {
2548 sem_wait(&_semaphore);
2549 }
2550
2551 bool Semaphore::trywait() {
2552 return sem_trywait(&_semaphore) == 0;
2553 }
2554
2555 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2556
2557 struct timespec ts;
2558 // Semaphore's are always associated with CLOCK_REALTIME
2559 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2560 // see unpackTime for discussion on overflow checking
2561 if (sec >= MAX_SECS) {
2562 ts.tv_sec += MAX_SECS;
2563 ts.tv_nsec = 0;
2564 } else {
2565 ts.tv_sec += sec;
2566 ts.tv_nsec += nsec;
2567 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2568 ts.tv_nsec -= NANOSECS_PER_SEC;
2569 ++ts.tv_sec; // note: this must be <= max_secs
2570 }
2571 }
2572
2573 while (1) {
2574 int result = sem_timedwait(&_semaphore, &ts);
2575 if (result == 0) {
2576 return true;
2577 } else if (errno == EINTR) {
2578 continue;
2579 } else if (errno == ETIMEDOUT) {
2580 return false;
2581 } else {
2582 return false;
2583 }
2584 }
2585 }
2586
2587 extern "C" {
2588 typedef void (*sa_handler_t)(int);
2589 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2590 }
2591
2592 void* os::signal(int signal_number, void* handler) {
2593 struct sigaction sigAct, oldSigAct;
2594
2595 sigfillset(&(sigAct.sa_mask));
2596 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2597 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2598
2599 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2600 // -1 means registration failed
2601 return (void *)-1;
2602 }
2603
2604 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2605 }
2606
2607 void os::signal_raise(int signal_number) {
2608 ::raise(signal_number);
2609 }
2610
2611 /*
2612 * The following code is moved from os.cpp for making this
2613 * code platform specific, which it is by its very nature.
2614 */
2615
2616 // Will be modified when max signal is changed to be dynamic
2617 int os::sigexitnum_pd() {
2618 return NSIG;
2619 }
2620
2621 // a counter for each possible signal value
2622 static volatile jint pending_signals[NSIG+1] = { 0 };
2623
2624 // Linux(POSIX) specific hand shaking semaphore.
2625 static sem_t sig_sem;
2626 static Semaphore sr_semaphore;
2627
2628 void os::signal_init_pd() {
2629 // Initialize signal structures
2630 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2631
2632 // Initialize signal semaphore
2633 ::sem_init(&sig_sem, 0, 0);
2634 }
2635
2636 void os::signal_notify(int sig) {
2637 Atomic::inc(&pending_signals[sig]);
2638 ::sem_post(&sig_sem);
2639 }
2640
2641 static int check_pending_signals(bool wait) {
2642 Atomic::store(0, &sigint_count);
2643 for (;;) {
2644 for (int i = 0; i < NSIG + 1; i++) {
2645 jint n = pending_signals[i];
2646 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2647 return i;
2648 }
2649 }
2650 if (!wait) {
2651 return -1;
2652 }
2653 JavaThread *thread = JavaThread::current();
2654 ThreadBlockInVM tbivm(thread);
2655
2656 bool threadIsSuspended;
2657 do {
2658 thread->set_suspend_equivalent();
2659 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2660 ::sem_wait(&sig_sem);
2661
2662 // were we externally suspended while we were waiting?
2663 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2664 if (threadIsSuspended) {
2665 //
2666 // The semaphore has been incremented, but while we were waiting
2667 // another thread suspended us. We don't want to continue running
2668 // while suspended because that would surprise the thread that
2669 // suspended us.
2670 //
2671 ::sem_post(&sig_sem);
2672
2673 thread->java_suspend_self();
2674 }
2675 } while (threadIsSuspended);
2676 }
2677 }
2678
2679 int os::signal_lookup() {
2680 return check_pending_signals(false);
2681 }
2682
2683 int os::signal_wait() {
2684 return check_pending_signals(true);
2685 }
2686
2687 ////////////////////////////////////////////////////////////////////////////////
2688 // Virtual Memory
2689
2690 int os::vm_page_size() {
2691 // Seems redundant as all get out
2692 assert(os::Linux::page_size() != -1, "must call os::init");
2693 return os::Linux::page_size();
2694 }
2695
2696 // Solaris allocates memory by pages.
2697 int os::vm_allocation_granularity() {
2698 assert(os::Linux::page_size() != -1, "must call os::init");
2699 return os::Linux::page_size();
2700 }
2701
2702 // Rationale behind this function:
2703 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2704 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2705 // samples for JITted code. Here we create private executable mapping over the code cache
2706 // and then we can use standard (well, almost, as mapping can change) way to provide
2707 // info for the reporting script by storing timestamp and location of symbol
2708 void linux_wrap_code(char* base, size_t size) {
2709 static volatile jint cnt = 0;
2710
2711 if (!UseOprofile) {
2712 return;
2713 }
2714
2715 char buf[PATH_MAX+1];
2716 int num = Atomic::add(1, &cnt);
2717
2718 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2719 os::get_temp_directory(), os::current_process_id(), num);
2720 unlink(buf);
2721
2722 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2723
2724 if (fd != -1) {
2725 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2726 if (rv != (off_t)-1) {
2727 if (::write(fd, "", 1) == 1) {
2728 mmap(base, size,
2729 PROT_READ|PROT_WRITE|PROT_EXEC,
2730 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2731 }
2732 }
2733 ::close(fd);
2734 unlink(buf);
2735 }
2736 }
2737
2738 static bool recoverable_mmap_error(int err) {
2739 // See if the error is one we can let the caller handle. This
2740 // list of errno values comes from JBS-6843484. I can't find a
2741 // Linux man page that documents this specific set of errno
2742 // values so while this list currently matches Solaris, it may
2743 // change as we gain experience with this failure mode.
2744 switch (err) {
2745 case EBADF:
2746 case EINVAL:
2747 case ENOTSUP:
2748 // let the caller deal with these errors
2749 return true;
2750
2751 default:
2752 // Any remaining errors on this OS can cause our reserved mapping
2753 // to be lost. That can cause confusion where different data
2754 // structures think they have the same memory mapped. The worst
2755 // scenario is if both the VM and a library think they have the
2756 // same memory mapped.
2757 return false;
2758 }
2759 }
2760
2761 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2762 int err) {
2763 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2764 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2765 strerror(err), err);
2766 }
2767
2768 static void warn_fail_commit_memory(char* addr, size_t size,
2769 size_t alignment_hint, bool exec,
2770 int err) {
2771 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2772 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2773 alignment_hint, exec, strerror(err), err);
2774 }
2775
2776 // NOTE: Linux kernel does not really reserve the pages for us.
2777 // All it does is to check if there are enough free pages
2778 // left at the time of mmap(). This could be a potential
2779 // problem.
2780 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2781 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2782 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2783 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2784 if (res != (uintptr_t) MAP_FAILED) {
2785 if (UseNUMAInterleaving) {
2786 numa_make_global(addr, size);
2787 }
2788 return 0;
2789 }
2790
2791 int err = errno; // save errno from mmap() call above
2792
2793 if (!recoverable_mmap_error(err)) {
2794 warn_fail_commit_memory(addr, size, exec, err);
2795 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2796 }
2797
2798 return err;
2799 }
2800
2801 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2802 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2803 }
2804
2805 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2806 const char* mesg) {
2807 assert(mesg != NULL, "mesg must be specified");
2808 int err = os::Linux::commit_memory_impl(addr, size, exec);
2809 if (err != 0) {
2810 // the caller wants all commit errors to exit with the specified mesg:
2811 warn_fail_commit_memory(addr, size, exec, err);
2812 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2813 }
2814 }
2815
2816 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2817 #ifndef MAP_HUGETLB
2818 #define MAP_HUGETLB 0x40000
2819 #endif
2820
2821 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2822 #ifndef MADV_HUGEPAGE
2823 #define MADV_HUGEPAGE 14
2824 #endif
2825
2826 int os::Linux::commit_memory_impl(char* addr, size_t size,
2827 size_t alignment_hint, bool exec) {
2828 int err = os::Linux::commit_memory_impl(addr, size, exec);
2829 if (err == 0) {
2830 realign_memory(addr, size, alignment_hint);
2831 }
2832 return err;
2833 }
2834
2835 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2836 bool exec) {
2837 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2838 }
2839
2840 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2841 size_t alignment_hint, bool exec,
2842 const char* mesg) {
2843 assert(mesg != NULL, "mesg must be specified");
2844 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2845 if (err != 0) {
2846 // the caller wants all commit errors to exit with the specified mesg:
2847 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2848 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2849 }
2850 }
2851
2852 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2853 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2854 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2855 // be supported or the memory may already be backed by huge pages.
2856 ::madvise(addr, bytes, MADV_HUGEPAGE);
2857 }
2858 }
2859
2860 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2861 // This method works by doing an mmap over an existing mmaping and effectively discarding
2862 // the existing pages. However it won't work for SHM-based large pages that cannot be
2863 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2864 // small pages on top of the SHM segment. This method always works for small pages, so we
2865 // allow that in any case.
2866 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2867 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2868 }
2869 }
2870
2871 void os::numa_make_global(char *addr, size_t bytes) {
2872 Linux::numa_interleave_memory(addr, bytes);
2873 }
2874
2875 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2876 // bind policy to MPOL_PREFERRED for the current thread.
2877 #define USE_MPOL_PREFERRED 0
2878
2879 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2880 // To make NUMA and large pages more robust when both enabled, we need to ease
2881 // the requirements on where the memory should be allocated. MPOL_BIND is the
2882 // default policy and it will force memory to be allocated on the specified
2883 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2884 // the specified node, but will not force it. Using this policy will prevent
2885 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2886 // free large pages.
2887 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2888 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2889 }
2890
2891 bool os::numa_topology_changed() { return false; }
2892
2893 size_t os::numa_get_groups_num() {
2894 // Return just the number of nodes in which it's possible to allocate memory
2895 // (in numa terminology, configured nodes).
2896 return Linux::numa_num_configured_nodes();
2897 }
2898
2899 int os::numa_get_group_id() {
2900 int cpu_id = Linux::sched_getcpu();
2901 if (cpu_id != -1) {
2902 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2903 if (lgrp_id != -1) {
2904 return lgrp_id;
2905 }
2906 }
2907 return 0;
2908 }
2909
2910 int os::Linux::get_existing_num_nodes() {
2911 size_t node;
2912 size_t highest_node_number = Linux::numa_max_node();
2913 int num_nodes = 0;
2914
2915 // Get the total number of nodes in the system including nodes without memory.
2916 for (node = 0; node <= highest_node_number; node++) {
2917 if (isnode_in_existing_nodes(node)) {
2918 num_nodes++;
2919 }
2920 }
2921 return num_nodes;
2922 }
2923
2924 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2925 size_t highest_node_number = Linux::numa_max_node();
2926 size_t i = 0;
2927
2928 // Map all node ids in which is possible to allocate memory. Also nodes are
2929 // not always consecutively available, i.e. available from 0 to the highest
2930 // node number.
2931 for (size_t node = 0; node <= highest_node_number; node++) {
2932 if (Linux::isnode_in_configured_nodes(node)) {
2933 ids[i++] = node;
2934 }
2935 }
2936 return i;
2937 }
2938
2939 bool os::get_page_info(char *start, page_info* info) {
2940 return false;
2941 }
2942
2943 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2944 return end;
2945 }
2946
2947
2948 int os::Linux::sched_getcpu_syscall(void) {
2949 unsigned int cpu = 0;
2950 int retval = -1;
2951
2952 #if defined(IA32)
2953 # ifndef SYS_getcpu
2954 # define SYS_getcpu 318
2955 # endif
2956 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2957 #elif defined(AMD64)
2958 // Unfortunately we have to bring all these macros here from vsyscall.h
2959 // to be able to compile on old linuxes.
2960 # define __NR_vgetcpu 2
2961 # define VSYSCALL_START (-10UL << 20)
2962 # define VSYSCALL_SIZE 1024
2963 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2964 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2965 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2966 retval = vgetcpu(&cpu, NULL, NULL);
2967 #endif
2968
2969 return (retval == -1) ? retval : cpu;
2970 }
2971
2972 // Something to do with the numa-aware allocator needs these symbols
2973 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2974 extern "C" JNIEXPORT void numa_error(char *where) { }
2975 extern "C" JNIEXPORT int fork1() { return fork(); }
2976
2977 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2978 // load symbol from base version instead.
2979 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2980 void *f = dlvsym(handle, name, "libnuma_1.1");
2981 if (f == NULL) {
2982 f = dlsym(handle, name);
2983 }
2984 return f;
2985 }
2986
2987 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2988 // Return NULL if the symbol is not defined in this particular version.
2989 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2990 return dlvsym(handle, name, "libnuma_1.2");
2991 }
2992
2993 bool os::Linux::libnuma_init() {
2994 // sched_getcpu() should be in libc.
2995 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2996 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2997
2998 // If it's not, try a direct syscall.
2999 if (sched_getcpu() == -1)
3000 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
3001
3002 if (sched_getcpu() != -1) { // Does it work?
3003 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3004 if (handle != NULL) {
3005 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3006 libnuma_dlsym(handle, "numa_node_to_cpus")));
3007 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3008 libnuma_dlsym(handle, "numa_max_node")));
3009 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3010 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3011 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3012 libnuma_dlsym(handle, "numa_available")));
3013 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3014 libnuma_dlsym(handle, "numa_tonode_memory")));
3015 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3016 libnuma_dlsym(handle, "numa_interleave_memory")));
3017 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3018 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3019 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3020 libnuma_dlsym(handle, "numa_set_bind_policy")));
3021 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3022 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3023 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3024 libnuma_dlsym(handle, "numa_distance")));
3025
3026 if (numa_available() != -1) {
3027 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3028 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3029 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3030 // Create an index -> node mapping, since nodes are not always consecutive
3031 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3032 rebuild_nindex_to_node_map();
3033 // Create a cpu -> node mapping
3034 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3035 rebuild_cpu_to_node_map();
3036 return true;
3037 }
3038 }
3039 }
3040 return false;
3041 }
3042
3043 void os::Linux::rebuild_nindex_to_node_map() {
3044 int highest_node_number = Linux::numa_max_node();
3045
3046 nindex_to_node()->clear();
3047 for (int node = 0; node <= highest_node_number; node++) {
3048 if (Linux::isnode_in_existing_nodes(node)) {
3049 nindex_to_node()->append(node);
3050 }
3051 }
3052 }
3053
3054 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3055 // The table is later used in get_node_by_cpu().
3056 void os::Linux::rebuild_cpu_to_node_map() {
3057 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3058 // in libnuma (possible values are starting from 16,
3059 // and continuing up with every other power of 2, but less
3060 // than the maximum number of CPUs supported by kernel), and
3061 // is a subject to change (in libnuma version 2 the requirements
3062 // are more reasonable) we'll just hardcode the number they use
3063 // in the library.
3064 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3065
3066 size_t cpu_num = processor_count();
3067 size_t cpu_map_size = NCPUS / BitsPerCLong;
3068 size_t cpu_map_valid_size =
3069 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3070
3071 cpu_to_node()->clear();
3072 cpu_to_node()->at_grow(cpu_num - 1);
3073
3074 size_t node_num = get_existing_num_nodes();
3075
3076 int distance = 0;
3077 int closest_distance = INT_MAX;
3078 int closest_node = 0;
3079 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3080 for (size_t i = 0; i < node_num; i++) {
3081 // Check if node is configured (not a memory-less node). If it is not, find
3082 // the closest configured node.
3083 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3084 closest_distance = INT_MAX;
3085 // Check distance from all remaining nodes in the system. Ignore distance
3086 // from itself and from another non-configured node.
3087 for (size_t m = 0; m < node_num; m++) {
3088 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3089 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3090 // If a closest node is found, update. There is always at least one
3091 // configured node in the system so there is always at least one node
3092 // close.
3093 if (distance != 0 && distance < closest_distance) {
3094 closest_distance = distance;
3095 closest_node = nindex_to_node()->at(m);
3096 }
3097 }
3098 }
3099 } else {
3100 // Current node is already a configured node.
3101 closest_node = nindex_to_node()->at(i);
3102 }
3103
3104 // Get cpus from the original node and map them to the closest node. If node
3105 // is a configured node (not a memory-less node), then original node and
3106 // closest node are the same.
3107 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3108 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3109 if (cpu_map[j] != 0) {
3110 for (size_t k = 0; k < BitsPerCLong; k++) {
3111 if (cpu_map[j] & (1UL << k)) {
3112 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3113 }
3114 }
3115 }
3116 }
3117 }
3118 }
3119 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3120 }
3121
3122 int os::Linux::get_node_by_cpu(int cpu_id) {
3123 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3124 return cpu_to_node()->at(cpu_id);
3125 }
3126 return -1;
3127 }
3128
3129 GrowableArray<int>* os::Linux::_cpu_to_node;
3130 GrowableArray<int>* os::Linux::_nindex_to_node;
3131 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3132 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3133 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3134 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3135 os::Linux::numa_available_func_t os::Linux::_numa_available;
3136 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3137 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3138 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3139 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3140 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3141 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3142 unsigned long* os::Linux::_numa_all_nodes;
3143 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3144 struct bitmask* os::Linux::_numa_nodes_ptr;
3145
3146 bool os::pd_uncommit_memory(char* addr, size_t size) {
3147 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3148 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3149 return res != (uintptr_t) MAP_FAILED;
3150 }
3151
3152 static
3153 address get_stack_commited_bottom(address bottom, size_t size) {
3154 address nbot = bottom;
3155 address ntop = bottom + size;
3156
3157 size_t page_sz = os::vm_page_size();
3158 unsigned pages = size / page_sz;
3159
3160 unsigned char vec[1];
3161 unsigned imin = 1, imax = pages + 1, imid;
3162 int mincore_return_value = 0;
3163
3164 assert(imin <= imax, "Unexpected page size");
3165
3166 while (imin < imax) {
3167 imid = (imax + imin) / 2;
3168 nbot = ntop - (imid * page_sz);
3169
3170 // Use a trick with mincore to check whether the page is mapped or not.
3171 // mincore sets vec to 1 if page resides in memory and to 0 if page
3172 // is swapped output but if page we are asking for is unmapped
3173 // it returns -1,ENOMEM
3174 mincore_return_value = mincore(nbot, page_sz, vec);
3175
3176 if (mincore_return_value == -1) {
3177 // Page is not mapped go up
3178 // to find first mapped page
3179 if (errno != EAGAIN) {
3180 assert(errno == ENOMEM, "Unexpected mincore errno");
3181 imax = imid;
3182 }
3183 } else {
3184 // Page is mapped go down
3185 // to find first not mapped page
3186 imin = imid + 1;
3187 }
3188 }
3189
3190 nbot = nbot + page_sz;
3191
3192 // Adjust stack bottom one page up if last checked page is not mapped
3193 if (mincore_return_value == -1) {
3194 nbot = nbot + page_sz;
3195 }
3196
3197 return nbot;
3198 }
3199
3200
3201 // Linux uses a growable mapping for the stack, and if the mapping for
3202 // the stack guard pages is not removed when we detach a thread the
3203 // stack cannot grow beyond the pages where the stack guard was
3204 // mapped. If at some point later in the process the stack expands to
3205 // that point, the Linux kernel cannot expand the stack any further
3206 // because the guard pages are in the way, and a segfault occurs.
3207 //
3208 // However, it's essential not to split the stack region by unmapping
3209 // a region (leaving a hole) that's already part of the stack mapping,
3210 // so if the stack mapping has already grown beyond the guard pages at
3211 // the time we create them, we have to truncate the stack mapping.
3212 // So, we need to know the extent of the stack mapping when
3213 // create_stack_guard_pages() is called.
3214
3215 // We only need this for stacks that are growable: at the time of
3216 // writing thread stacks don't use growable mappings (i.e. those
3217 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3218 // only applies to the main thread.
3219
3220 // If the (growable) stack mapping already extends beyond the point
3221 // where we're going to put our guard pages, truncate the mapping at
3222 // that point by munmap()ping it. This ensures that when we later
3223 // munmap() the guard pages we don't leave a hole in the stack
3224 // mapping. This only affects the main/primordial thread
3225
3226 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3227
3228 if (os::is_primordial_thread()) {
3229 // As we manually grow stack up to bottom inside create_attached_thread(),
3230 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3231 // we don't need to do anything special.
3232 // Check it first, before calling heavy function.
3233 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3234 unsigned char vec[1];
3235
3236 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3237 // Fallback to slow path on all errors, including EAGAIN
3238 stack_extent = (uintptr_t) get_stack_commited_bottom(
3239 os::Linux::initial_thread_stack_bottom(),
3240 (size_t)addr - stack_extent);
3241 }
3242
3243 if (stack_extent < (uintptr_t)addr) {
3244 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3245 }
3246 }
3247
3248 return os::commit_memory(addr, size, !ExecMem);
3249 }
3250
3251 // If this is a growable mapping, remove the guard pages entirely by
3252 // munmap()ping them. If not, just call uncommit_memory(). This only
3253 // affects the main/primordial thread, but guard against future OS changes.
3254 // It's safe to always unmap guard pages for primordial thread because we
3255 // always place it right after end of the mapped region.
3256
3257 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3258 uintptr_t stack_extent, stack_base;
3259
3260 if (os::is_primordial_thread()) {
3261 return ::munmap(addr, size) == 0;
3262 }
3263
3264 return os::uncommit_memory(addr, size);
3265 }
3266
3267 static address _highest_vm_reserved_address = NULL;
3268
3269 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3270 // at 'requested_addr'. If there are existing memory mappings at the same
3271 // location, however, they will be overwritten. If 'fixed' is false,
3272 // 'requested_addr' is only treated as a hint, the return value may or
3273 // may not start from the requested address. Unlike Linux mmap(), this
3274 // function returns NULL to indicate failure.
3275 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3276 char * addr;
3277 int flags;
3278
3279 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3280 if (fixed) {
3281 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3282 flags |= MAP_FIXED;
3283 }
3284
3285 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3286 // touch an uncommitted page. Otherwise, the read/write might
3287 // succeed if we have enough swap space to back the physical page.
3288 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3289 flags, -1, 0);
3290
3291 if (addr != MAP_FAILED) {
3292 // anon_mmap() should only get called during VM initialization,
3293 // don't need lock (actually we can skip locking even it can be called
3294 // from multiple threads, because _highest_vm_reserved_address is just a
3295 // hint about the upper limit of non-stack memory regions.)
3296 if ((address)addr + bytes > _highest_vm_reserved_address) {
3297 _highest_vm_reserved_address = (address)addr + bytes;
3298 }
3299 }
3300
3301 return addr == MAP_FAILED ? NULL : addr;
3302 }
3303
3304 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3305 // (req_addr != NULL) or with a given alignment.
3306 // - bytes shall be a multiple of alignment.
3307 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3308 // - alignment sets the alignment at which memory shall be allocated.
3309 // It must be a multiple of allocation granularity.
3310 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3311 // req_addr or NULL.
3312 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3313
3314 size_t extra_size = bytes;
3315 if (req_addr == NULL && alignment > 0) {
3316 extra_size += alignment;
3317 }
3318
3319 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3320 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3321 -1, 0);
3322 if (start == MAP_FAILED) {
3323 start = NULL;
3324 } else {
3325 if (req_addr != NULL) {
3326 if (start != req_addr) {
3327 ::munmap(start, extra_size);
3328 start = NULL;
3329 }
3330 } else {
3331 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3332 char* const end_aligned = start_aligned + bytes;
3333 char* const end = start + extra_size;
3334 if (start_aligned > start) {
3335 ::munmap(start, start_aligned - start);
3336 }
3337 if (end_aligned < end) {
3338 ::munmap(end_aligned, end - end_aligned);
3339 }
3340 start = start_aligned;
3341 }
3342 }
3343 return start;
3344 }
3345
3346 // Don't update _highest_vm_reserved_address, because there might be memory
3347 // regions above addr + size. If so, releasing a memory region only creates
3348 // a hole in the address space, it doesn't help prevent heap-stack collision.
3349 //
3350 static int anon_munmap(char * addr, size_t size) {
3351 return ::munmap(addr, size) == 0;
3352 }
3353
3354 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3355 size_t alignment_hint) {
3356 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3357 }
3358
3359 bool os::pd_release_memory(char* addr, size_t size) {
3360 return anon_munmap(addr, size);
3361 }
3362
3363 static address highest_vm_reserved_address() {
3364 return _highest_vm_reserved_address;
3365 }
3366
3367 static bool linux_mprotect(char* addr, size_t size, int prot) {
3368 // Linux wants the mprotect address argument to be page aligned.
3369 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3370
3371 // According to SUSv3, mprotect() should only be used with mappings
3372 // established by mmap(), and mmap() always maps whole pages. Unaligned
3373 // 'addr' likely indicates problem in the VM (e.g. trying to change
3374 // protection of malloc'ed or statically allocated memory). Check the
3375 // caller if you hit this assert.
3376 assert(addr == bottom, "sanity check");
3377
3378 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3379 return ::mprotect(bottom, size, prot) == 0;
3380 }
3381
3382 // Set protections specified
3383 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3384 bool is_committed) {
3385 unsigned int p = 0;
3386 switch (prot) {
3387 case MEM_PROT_NONE: p = PROT_NONE; break;
3388 case MEM_PROT_READ: p = PROT_READ; break;
3389 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3390 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3391 default:
3392 ShouldNotReachHere();
3393 }
3394 // is_committed is unused.
3395 return linux_mprotect(addr, bytes, p);
3396 }
3397
3398 bool os::guard_memory(char* addr, size_t size) {
3399 return linux_mprotect(addr, size, PROT_NONE);
3400 }
3401
3402 bool os::unguard_memory(char* addr, size_t size) {
3403 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3404 }
3405
3406 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3407 bool result = false;
3408 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3409 MAP_ANONYMOUS|MAP_PRIVATE,
3410 -1, 0);
3411 if (p != MAP_FAILED) {
3412 void *aligned_p = align_ptr_up(p, page_size);
3413
3414 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3415
3416 munmap(p, page_size * 2);
3417 }
3418
3419 if (warn && !result) {
3420 warning("TransparentHugePages is not supported by the operating system.");
3421 }
3422
3423 return result;
3424 }
3425
3426 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3427 bool result = false;
3428 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3429 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3430 -1, 0);
3431
3432 if (p != MAP_FAILED) {
3433 // We don't know if this really is a huge page or not.
3434 FILE *fp = fopen("/proc/self/maps", "r");
3435 if (fp) {
3436 while (!feof(fp)) {
3437 char chars[257];
3438 long x = 0;
3439 if (fgets(chars, sizeof(chars), fp)) {
3440 if (sscanf(chars, "%lx-%*x", &x) == 1
3441 && x == (long)p) {
3442 if (strstr (chars, "hugepage")) {
3443 result = true;
3444 break;
3445 }
3446 }
3447 }
3448 }
3449 fclose(fp);
3450 }
3451 munmap(p, page_size);
3452 }
3453
3454 if (warn && !result) {
3455 warning("HugeTLBFS is not supported by the operating system.");
3456 }
3457
3458 return result;
3459 }
3460
3461 /*
3462 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3463 *
3464 * From the coredump_filter documentation:
3465 *
3466 * - (bit 0) anonymous private memory
3467 * - (bit 1) anonymous shared memory
3468 * - (bit 2) file-backed private memory
3469 * - (bit 3) file-backed shared memory
3470 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3471 * effective only if the bit 2 is cleared)
3472 * - (bit 5) hugetlb private memory
3473 * - (bit 6) hugetlb shared memory
3474 */
3475 static void set_coredump_filter(void) {
3476 FILE *f;
3477 long cdm;
3478
3479 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3480 return;
3481 }
3482
3483 if (fscanf(f, "%lx", &cdm) != 1) {
3484 fclose(f);
3485 return;
3486 }
3487
3488 rewind(f);
3489
3490 if ((cdm & LARGEPAGES_BIT) == 0) {
3491 cdm |= LARGEPAGES_BIT;
3492 fprintf(f, "%#lx", cdm);
3493 }
3494
3495 fclose(f);
3496 }
3497
3498 // Large page support
3499
3500 static size_t _large_page_size = 0;
3501
3502 size_t os::Linux::find_large_page_size() {
3503 size_t large_page_size = 0;
3504
3505 // large_page_size on Linux is used to round up heap size. x86 uses either
3506 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3507 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3508 // page as large as 256M.
3509 //
3510 // Here we try to figure out page size by parsing /proc/meminfo and looking
3511 // for a line with the following format:
3512 // Hugepagesize: 2048 kB
3513 //
3514 // If we can't determine the value (e.g. /proc is not mounted, or the text
3515 // format has been changed), we'll use the largest page size supported by
3516 // the processor.
3517
3518 #ifndef ZERO
3519 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3520 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3521 #endif // ZERO
3522
3523 FILE *fp = fopen("/proc/meminfo", "r");
3524 if (fp) {
3525 while (!feof(fp)) {
3526 int x = 0;
3527 char buf[16];
3528 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3529 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3530 large_page_size = x * K;
3531 break;
3532 }
3533 } else {
3534 // skip to next line
3535 for (;;) {
3536 int ch = fgetc(fp);
3537 if (ch == EOF || ch == (int)'\n') break;
3538 }
3539 }
3540 }
3541 fclose(fp);
3542 }
3543
3544 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3545 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3546 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3547 proper_unit_for_byte_size(large_page_size));
3548 }
3549
3550 return large_page_size;
3551 }
3552
3553 size_t os::Linux::setup_large_page_size() {
3554 _large_page_size = Linux::find_large_page_size();
3555 const size_t default_page_size = (size_t)Linux::page_size();
3556 if (_large_page_size > default_page_size) {
3557 _page_sizes[0] = _large_page_size;
3558 _page_sizes[1] = default_page_size;
3559 _page_sizes[2] = 0;
3560 }
3561
3562 return _large_page_size;
3563 }
3564
3565 bool os::Linux::setup_large_page_type(size_t page_size) {
3566 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3567 FLAG_IS_DEFAULT(UseSHM) &&
3568 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3569
3570 // The type of large pages has not been specified by the user.
3571
3572 // Try UseHugeTLBFS and then UseSHM.
3573 UseHugeTLBFS = UseSHM = true;
3574
3575 // Don't try UseTransparentHugePages since there are known
3576 // performance issues with it turned on. This might change in the future.
3577 UseTransparentHugePages = false;
3578 }
3579
3580 if (UseTransparentHugePages) {
3581 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3582 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3583 UseHugeTLBFS = false;
3584 UseSHM = false;
3585 return true;
3586 }
3587 UseTransparentHugePages = false;
3588 }
3589
3590 if (UseHugeTLBFS) {
3591 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3592 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3593 UseSHM = false;
3594 return true;
3595 }
3596 UseHugeTLBFS = false;
3597 }
3598
3599 return UseSHM;
3600 }
3601
3602 void os::large_page_init() {
3603 if (!UseLargePages &&
3604 !UseTransparentHugePages &&
3605 !UseHugeTLBFS &&
3606 !UseSHM) {
3607 // Not using large pages.
3608 return;
3609 }
3610
3611 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3612 // The user explicitly turned off large pages.
3613 // Ignore the rest of the large pages flags.
3614 UseTransparentHugePages = false;
3615 UseHugeTLBFS = false;
3616 UseSHM = false;
3617 return;
3618 }
3619
3620 size_t large_page_size = Linux::setup_large_page_size();
3621 UseLargePages = Linux::setup_large_page_type(large_page_size);
3622
3623 set_coredump_filter();
3624 }
3625
3626 #ifndef SHM_HUGETLB
3627 #define SHM_HUGETLB 04000
3628 #endif
3629
3630 #define shm_warning_format(format, ...) \
3631 do { \
3632 if (UseLargePages && \
3633 (!FLAG_IS_DEFAULT(UseLargePages) || \
3634 !FLAG_IS_DEFAULT(UseSHM) || \
3635 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3636 warning(format, __VA_ARGS__); \
3637 } \
3638 } while (0)
3639
3640 #define shm_warning(str) shm_warning_format("%s", str)
3641
3642 #define shm_warning_with_errno(str) \
3643 do { \
3644 int err = errno; \
3645 shm_warning_format(str " (error = %d)", err); \
3646 } while (0)
3647
3648 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3649 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3650
3651 if (!is_size_aligned(alignment, SHMLBA)) {
3652 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3653 return NULL;
3654 }
3655
3656 // To ensure that we get 'alignment' aligned memory from shmat,
3657 // we pre-reserve aligned virtual memory and then attach to that.
3658
3659 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3660 if (pre_reserved_addr == NULL) {
3661 // Couldn't pre-reserve aligned memory.
3662 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3663 return NULL;
3664 }
3665
3666 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3667 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3668
3669 if ((intptr_t)addr == -1) {
3670 int err = errno;
3671 shm_warning_with_errno("Failed to attach shared memory.");
3672
3673 assert(err != EACCES, "Unexpected error");
3674 assert(err != EIDRM, "Unexpected error");
3675 assert(err != EINVAL, "Unexpected error");
3676
3677 // Since we don't know if the kernel unmapped the pre-reserved memory area
3678 // we can't unmap it, since that would potentially unmap memory that was
3679 // mapped from other threads.
3680 return NULL;
3681 }
3682
3683 return addr;
3684 }
3685
3686 static char* shmat_at_address(int shmid, char* req_addr) {
3687 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3688 assert(false, "Requested address needs to be SHMLBA aligned");
3689 return NULL;
3690 }
3691
3692 char* addr = (char*)shmat(shmid, req_addr, 0);
3693
3694 if ((intptr_t)addr == -1) {
3695 shm_warning_with_errno("Failed to attach shared memory.");
3696 return NULL;
3697 }
3698
3699 return addr;
3700 }
3701
3702 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3703 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3704 if (req_addr != NULL) {
3705 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3706 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3707 return shmat_at_address(shmid, req_addr);
3708 }
3709
3710 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3711 // return large page size aligned memory addresses when req_addr == NULL.
3712 // However, if the alignment is larger than the large page size, we have
3713 // to manually ensure that the memory returned is 'alignment' aligned.
3714 if (alignment > os::large_page_size()) {
3715 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3716 return shmat_with_alignment(shmid, bytes, alignment);
3717 } else {
3718 return shmat_at_address(shmid, NULL);
3719 }
3720 }
3721
3722 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3723 // "exec" is passed in but not used. Creating the shared image for
3724 // the code cache doesn't have an SHM_X executable permission to check.
3725 assert(UseLargePages && UseSHM, "only for SHM large pages");
3726 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3727 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3728
3729 if (!is_size_aligned(bytes, os::large_page_size())) {
3730 return NULL; // Fallback to small pages.
3731 }
3732
3733 // Create a large shared memory region to attach to based on size.
3734 // Currently, size is the total size of the heap.
3735 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3736 if (shmid == -1) {
3737 // Possible reasons for shmget failure:
3738 // 1. shmmax is too small for Java heap.
3739 // > check shmmax value: cat /proc/sys/kernel/shmmax
3740 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3741 // 2. not enough large page memory.
3742 // > check available large pages: cat /proc/meminfo
3743 // > increase amount of large pages:
3744 // echo new_value > /proc/sys/vm/nr_hugepages
3745 // Note 1: different Linux may use different name for this property,
3746 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3747 // Note 2: it's possible there's enough physical memory available but
3748 // they are so fragmented after a long run that they can't
3749 // coalesce into large pages. Try to reserve large pages when
3750 // the system is still "fresh".
3751 shm_warning_with_errno("Failed to reserve shared memory.");
3752 return NULL;
3753 }
3754
3755 // Attach to the region.
3756 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3757
3758 // Remove shmid. If shmat() is successful, the actual shared memory segment
3759 // will be deleted when it's detached by shmdt() or when the process
3760 // terminates. If shmat() is not successful this will remove the shared
3761 // segment immediately.
3762 shmctl(shmid, IPC_RMID, NULL);
3763
3764 return addr;
3765 }
3766
3767 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3768 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3769
3770 bool warn_on_failure = UseLargePages &&
3771 (!FLAG_IS_DEFAULT(UseLargePages) ||
3772 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3773 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3774
3775 if (warn_on_failure) {
3776 char msg[128];
3777 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3778 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3779 warning("%s", msg);
3780 }
3781 }
3782
3783 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3784 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3785 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3786 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3787
3788 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3789 char* addr = (char*)::mmap(req_addr, bytes, prot,
3790 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3791 -1, 0);
3792
3793 if (addr == MAP_FAILED) {
3794 warn_on_large_pages_failure(req_addr, bytes, errno);
3795 return NULL;
3796 }
3797
3798 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3799
3800 return addr;
3801 }
3802
3803 // Reserve memory using mmap(MAP_HUGETLB).
3804 // - bytes shall be a multiple of alignment.
3805 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3806 // - alignment sets the alignment at which memory shall be allocated.
3807 // It must be a multiple of allocation granularity.
3808 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3809 // req_addr or NULL.
3810 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3811 size_t large_page_size = os::large_page_size();
3812 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3813
3814 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3815 assert(is_size_aligned(bytes, alignment), "Must be");
3816
3817 // First reserve - but not commit - the address range in small pages.
3818 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3819
3820 if (start == NULL) {
3821 return NULL;
3822 }
3823
3824 assert(is_ptr_aligned(start, alignment), "Must be");
3825
3826 char* end = start + bytes;
3827
3828 // Find the regions of the allocated chunk that can be promoted to large pages.
3829 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3830 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3831
3832 size_t lp_bytes = lp_end - lp_start;
3833
3834 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3835
3836 if (lp_bytes == 0) {
3837 // The mapped region doesn't even span the start and the end of a large page.
3838 // Fall back to allocate a non-special area.
3839 ::munmap(start, end - start);
3840 return NULL;
3841 }
3842
3843 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3844
3845 void* result;
3846
3847 // Commit small-paged leading area.
3848 if (start != lp_start) {
3849 result = ::mmap(start, lp_start - start, prot,
3850 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3851 -1, 0);
3852 if (result == MAP_FAILED) {
3853 ::munmap(lp_start, end - lp_start);
3854 return NULL;
3855 }
3856 }
3857
3858 // Commit large-paged area.
3859 result = ::mmap(lp_start, lp_bytes, prot,
3860 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3861 -1, 0);
3862 if (result == MAP_FAILED) {
3863 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3864 // If the mmap above fails, the large pages region will be unmapped and we
3865 // have regions before and after with small pages. Release these regions.
3866 //
3867 // | mapped | unmapped | mapped |
3868 // ^ ^ ^ ^
3869 // start lp_start lp_end end
3870 //
3871 ::munmap(start, lp_start - start);
3872 ::munmap(lp_end, end - lp_end);
3873 return NULL;
3874 }
3875
3876 // Commit small-paged trailing area.
3877 if (lp_end != end) {
3878 result = ::mmap(lp_end, end - lp_end, prot,
3879 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3880 -1, 0);
3881 if (result == MAP_FAILED) {
3882 ::munmap(start, lp_end - start);
3883 return NULL;
3884 }
3885 }
3886
3887 return start;
3888 }
3889
3890 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3891 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3892 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3893 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3894 assert(is_power_of_2(os::large_page_size()), "Must be");
3895 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3896
3897 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3898 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3899 } else {
3900 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3901 }
3902 }
3903
3904 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3905 assert(UseLargePages, "only for large pages");
3906
3907 char* addr;
3908 if (UseSHM) {
3909 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3910 } else {
3911 assert(UseHugeTLBFS, "must be");
3912 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3913 }
3914
3915 if (addr != NULL) {
3916 if (UseNUMAInterleaving) {
3917 numa_make_global(addr, bytes);
3918 }
3919
3920 // The memory is committed
3921 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3922 }
3923
3924 return addr;
3925 }
3926
3927 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3928 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3929 return shmdt(base) == 0;
3930 }
3931
3932 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3933 return pd_release_memory(base, bytes);
3934 }
3935
3936 bool os::release_memory_special(char* base, size_t bytes) {
3937 bool res;
3938 if (MemTracker::tracking_level() > NMT_minimal) {
3939 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3940 res = os::Linux::release_memory_special_impl(base, bytes);
3941 if (res) {
3942 tkr.record((address)base, bytes);
3943 }
3944
3945 } else {
3946 res = os::Linux::release_memory_special_impl(base, bytes);
3947 }
3948 return res;
3949 }
3950
3951 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3952 assert(UseLargePages, "only for large pages");
3953 bool res;
3954
3955 if (UseSHM) {
3956 res = os::Linux::release_memory_special_shm(base, bytes);
3957 } else {
3958 assert(UseHugeTLBFS, "must be");
3959 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3960 }
3961 return res;
3962 }
3963
3964 size_t os::large_page_size() {
3965 return _large_page_size;
3966 }
3967
3968 // With SysV SHM the entire memory region must be allocated as shared
3969 // memory.
3970 // HugeTLBFS allows application to commit large page memory on demand.
3971 // However, when committing memory with HugeTLBFS fails, the region
3972 // that was supposed to be committed will lose the old reservation
3973 // and allow other threads to steal that memory region. Because of this
3974 // behavior we can't commit HugeTLBFS memory.
3975 bool os::can_commit_large_page_memory() {
3976 return UseTransparentHugePages;
3977 }
3978
3979 bool os::can_execute_large_page_memory() {
3980 return UseTransparentHugePages || UseHugeTLBFS;
3981 }
3982
3983 // Reserve memory at an arbitrary address, only if that area is
3984 // available (and not reserved for something else).
3985
3986 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3987 const int max_tries = 10;
3988 char* base[max_tries];
3989 size_t size[max_tries];
3990 const size_t gap = 0x000000;
3991
3992 // Assert only that the size is a multiple of the page size, since
3993 // that's all that mmap requires, and since that's all we really know
3994 // about at this low abstraction level. If we need higher alignment,
3995 // we can either pass an alignment to this method or verify alignment
3996 // in one of the methods further up the call chain. See bug 5044738.
3997 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3998
3999 // Repeatedly allocate blocks until the block is allocated at the
4000 // right spot. Give up after max_tries. Note that reserve_memory() will
4001 // automatically update _highest_vm_reserved_address if the call is
4002 // successful. The variable tracks the highest memory address every reserved
4003 // by JVM. It is used to detect heap-stack collision if running with
4004 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
4005 // space than needed, it could confuse the collision detecting code. To
4006 // solve the problem, save current _highest_vm_reserved_address and
4007 // calculate the correct value before return.
4008 address old_highest = _highest_vm_reserved_address;
4009
4010 // Linux mmap allows caller to pass an address as hint; give it a try first,
4011 // if kernel honors the hint then we can return immediately.
4012 char * addr = anon_mmap(requested_addr, bytes, false);
4013 if (addr == requested_addr) {
4014 return requested_addr;
4015 }
4016
4017 if (addr != NULL) {
4018 // mmap() is successful but it fails to reserve at the requested address
4019 anon_munmap(addr, bytes);
4020 }
4021
4022 int i;
4023 for (i = 0; i < max_tries; ++i) {
4024 base[i] = reserve_memory(bytes);
4025
4026 if (base[i] != NULL) {
4027 // Is this the block we wanted?
4028 if (base[i] == requested_addr) {
4029 size[i] = bytes;
4030 break;
4031 }
4032
4033 // Does this overlap the block we wanted? Give back the overlapped
4034 // parts and try again.
4035
4036 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4037 if (top_overlap >= 0 && top_overlap < bytes) {
4038 unmap_memory(base[i], top_overlap);
4039 base[i] += top_overlap;
4040 size[i] = bytes - top_overlap;
4041 } else {
4042 size_t bottom_overlap = base[i] + bytes - requested_addr;
4043 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4044 unmap_memory(requested_addr, bottom_overlap);
4045 size[i] = bytes - bottom_overlap;
4046 } else {
4047 size[i] = bytes;
4048 }
4049 }
4050 }
4051 }
4052
4053 // Give back the unused reserved pieces.
4054
4055 for (int j = 0; j < i; ++j) {
4056 if (base[j] != NULL) {
4057 unmap_memory(base[j], size[j]);
4058 }
4059 }
4060
4061 if (i < max_tries) {
4062 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4063 return requested_addr;
4064 } else {
4065 _highest_vm_reserved_address = old_highest;
4066 return NULL;
4067 }
4068 }
4069
4070 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4071 return ::read(fd, buf, nBytes);
4072 }
4073
4074 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4075 return ::pread(fd, buf, nBytes, offset);
4076 }
4077
4078 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4079 // Solaris uses poll(), linux uses park().
4080 // Poll() is likely a better choice, assuming that Thread.interrupt()
4081 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4082 // SIGSEGV, see 4355769.
4083
4084 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4085 assert(thread == Thread::current(), "thread consistency check");
4086
4087 ParkEvent * const slp = thread->_SleepEvent ;
4088 slp->reset() ;
4089 OrderAccess::fence() ;
4090
4091 if (interruptible) {
4092 jlong prevtime = javaTimeNanos();
4093
4094 for (;;) {
4095 if (os::is_interrupted(thread, true)) {
4096 return OS_INTRPT;
4097 }
4098
4099 jlong newtime = javaTimeNanos();
4100
4101 if (newtime - prevtime < 0) {
4102 // time moving backwards, should only happen if no monotonic clock
4103 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4104 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4105 } else {
4106 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4107 }
4108
4109 if(millis <= 0) {
4110 return OS_OK;
4111 }
4112
4113 prevtime = newtime;
4114
4115 {
4116 assert(thread->is_Java_thread(), "sanity check");
4117 JavaThread *jt = (JavaThread *) thread;
4118 ThreadBlockInVM tbivm(jt);
4119 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4120
4121 jt->set_suspend_equivalent();
4122 // cleared by handle_special_suspend_equivalent_condition() or
4123 // java_suspend_self() via check_and_wait_while_suspended()
4124
4125 slp->park(millis);
4126
4127 // were we externally suspended while we were waiting?
4128 jt->check_and_wait_while_suspended();
4129 }
4130 }
4131 } else {
4132 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4133 jlong prevtime = javaTimeNanos();
4134
4135 for (;;) {
4136 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4137 // the 1st iteration ...
4138 jlong newtime = javaTimeNanos();
4139
4140 if (newtime - prevtime < 0) {
4141 // time moving backwards, should only happen if no monotonic clock
4142 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4143 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4144 } else {
4145 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4146 }
4147
4148 if(millis <= 0) break ;
4149
4150 prevtime = newtime;
4151 slp->park(millis);
4152 }
4153 return OS_OK ;
4154 }
4155 }
4156
4157 //
4158 // Short sleep, direct OS call.
4159 //
4160 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4161 // sched_yield(2) will actually give up the CPU:
4162 //
4163 // * Alone on this pariticular CPU, keeps running.
4164 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4165 // (pre 2.6.39).
4166 //
4167 // So calling this with 0 is an alternative.
4168 //
4169 void os::naked_short_sleep(jlong ms) {
4170 struct timespec req;
4171
4172 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4173 req.tv_sec = 0;
4174 if (ms > 0) {
4175 req.tv_nsec = (ms % 1000) * 1000000;
4176 }
4177 else {
4178 req.tv_nsec = 1;
4179 }
4180
4181 nanosleep(&req, NULL);
4182
4183 return;
4184 }
4185
4186 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4187 void os::infinite_sleep() {
4188 while (true) { // sleep forever ...
4189 ::sleep(100); // ... 100 seconds at a time
4190 }
4191 }
4192
4193 // Used to convert frequent JVM_Yield() to nops
4194 bool os::dont_yield() {
4195 return DontYieldALot;
4196 }
4197
4198 void os::yield() {
4199 sched_yield();
4200 }
4201
4202 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4203
4204 void os::yield_all(int attempts) {
4205 // Yields to all threads, including threads with lower priorities
4206 // Threads on Linux are all with same priority. The Solaris style
4207 // os::yield_all() with nanosleep(1ms) is not necessary.
4208 sched_yield();
4209 }
4210
4211 // Called from the tight loops to possibly influence time-sharing heuristics
4212 void os::loop_breaker(int attempts) {
4213 os::yield_all(attempts);
4214 }
4215
4216 ////////////////////////////////////////////////////////////////////////////////
4217 // thread priority support
4218
4219 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4220 // only supports dynamic priority, static priority must be zero. For real-time
4221 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4222 // However, for large multi-threaded applications, SCHED_RR is not only slower
4223 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4224 // of 5 runs - Sep 2005).
4225 //
4226 // The following code actually changes the niceness of kernel-thread/LWP. It
4227 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4228 // not the entire user process, and user level threads are 1:1 mapped to kernel
4229 // threads. It has always been the case, but could change in the future. For
4230 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4231 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4232
4233 int os::java_to_os_priority[CriticalPriority + 1] = {
4234 19, // 0 Entry should never be used
4235
4236 4, // 1 MinPriority
4237 3, // 2
4238 2, // 3
4239
4240 1, // 4
4241 0, // 5 NormPriority
4242 -1, // 6
4243
4244 -2, // 7
4245 -3, // 8
4246 -4, // 9 NearMaxPriority
4247
4248 -5, // 10 MaxPriority
4249
4250 -5 // 11 CriticalPriority
4251 };
4252
4253 static int prio_init() {
4254 if (ThreadPriorityPolicy == 1) {
4255 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4256 // if effective uid is not root. Perhaps, a more elegant way of doing
4257 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4258 if (geteuid() != 0) {
4259 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4260 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4261 }
4262 ThreadPriorityPolicy = 0;
4263 }
4264 }
4265 if (UseCriticalJavaThreadPriority) {
4266 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4267 }
4268 return 0;
4269 }
4270
4271 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4272 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4273
4274 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4275 return (ret == 0) ? OS_OK : OS_ERR;
4276 }
4277
4278 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4279 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4280 *priority_ptr = java_to_os_priority[NormPriority];
4281 return OS_OK;
4282 }
4283
4284 errno = 0;
4285 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4286 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4287 }
4288
4289 // Hint to the underlying OS that a task switch would not be good.
4290 // Void return because it's a hint and can fail.
4291 void os::hint_no_preempt() {}
4292
4293 ////////////////////////////////////////////////////////////////////////////////
4294 // suspend/resume support
4295
4296 // the low-level signal-based suspend/resume support is a remnant from the
4297 // old VM-suspension that used to be for java-suspension, safepoints etc,
4298 // within hotspot. Now there is a single use-case for this:
4299 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4300 // that runs in the watcher thread.
4301 // The remaining code is greatly simplified from the more general suspension
4302 // code that used to be used.
4303 //
4304 // The protocol is quite simple:
4305 // - suspend:
4306 // - sends a signal to the target thread
4307 // - polls the suspend state of the osthread using a yield loop
4308 // - target thread signal handler (SR_handler) sets suspend state
4309 // and blocks in sigsuspend until continued
4310 // - resume:
4311 // - sets target osthread state to continue
4312 // - sends signal to end the sigsuspend loop in the SR_handler
4313 //
4314 // Note that the SR_lock plays no role in this suspend/resume protocol.
4315 //
4316
4317 static void resume_clear_context(OSThread *osthread) {
4318 osthread->set_ucontext(NULL);
4319 osthread->set_siginfo(NULL);
4320 }
4321
4322 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4323 osthread->set_ucontext(context);
4324 osthread->set_siginfo(siginfo);
4325 }
4326
4327 //
4328 // Handler function invoked when a thread's execution is suspended or
4329 // resumed. We have to be careful that only async-safe functions are
4330 // called here (Note: most pthread functions are not async safe and
4331 // should be avoided.)
4332 //
4333 // Note: sigwait() is a more natural fit than sigsuspend() from an
4334 // interface point of view, but sigwait() prevents the signal hander
4335 // from being run. libpthread would get very confused by not having
4336 // its signal handlers run and prevents sigwait()'s use with the
4337 // mutex granting granting signal.
4338 //
4339 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4340 //
4341 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4342 // Save and restore errno to avoid confusing native code with EINTR
4343 // after sigsuspend.
4344 int old_errno = errno;
4345
4346 Thread* thread = Thread::current();
4347 OSThread* osthread = thread->osthread();
4348 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4349
4350 os::SuspendResume::State current = osthread->sr.state();
4351 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4352 suspend_save_context(osthread, siginfo, context);
4353
4354 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4355 os::SuspendResume::State state = osthread->sr.suspended();
4356 if (state == os::SuspendResume::SR_SUSPENDED) {
4357 sigset_t suspend_set; // signals for sigsuspend()
4358
4359 // get current set of blocked signals and unblock resume signal
4360 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4361 sigdelset(&suspend_set, SR_signum);
4362
4363 sr_semaphore.signal();
4364 // wait here until we are resumed
4365 while (1) {
4366 sigsuspend(&suspend_set);
4367
4368 os::SuspendResume::State result = osthread->sr.running();
4369 if (result == os::SuspendResume::SR_RUNNING) {
4370 sr_semaphore.signal();
4371 break;
4372 }
4373 }
4374
4375 } else if (state == os::SuspendResume::SR_RUNNING) {
4376 // request was cancelled, continue
4377 } else {
4378 ShouldNotReachHere();
4379 }
4380
4381 resume_clear_context(osthread);
4382 } else if (current == os::SuspendResume::SR_RUNNING) {
4383 // request was cancelled, continue
4384 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4385 // ignore
4386 } else {
4387 // ignore
4388 }
4389
4390 errno = old_errno;
4391 }
4392
4393
4394 static int SR_initialize() {
4395 struct sigaction act;
4396 char *s;
4397 /* Get signal number to use for suspend/resume */
4398 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4399 int sig = ::strtol(s, 0, 10);
4400 if (sig > 0 || sig < _NSIG) {
4401 SR_signum = sig;
4402 }
4403 }
4404
4405 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4406 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4407
4408 sigemptyset(&SR_sigset);
4409 sigaddset(&SR_sigset, SR_signum);
4410
4411 /* Set up signal handler for suspend/resume */
4412 act.sa_flags = SA_RESTART|SA_SIGINFO;
4413 act.sa_handler = (void (*)(int)) SR_handler;
4414
4415 // SR_signum is blocked by default.
4416 // 4528190 - We also need to block pthread restart signal (32 on all
4417 // supported Linux platforms). Note that LinuxThreads need to block
4418 // this signal for all threads to work properly. So we don't have
4419 // to use hard-coded signal number when setting up the mask.
4420 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4421
4422 if (sigaction(SR_signum, &act, 0) == -1) {
4423 return -1;
4424 }
4425
4426 // Save signal flag
4427 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4428 return 0;
4429 }
4430
4431 static int sr_notify(OSThread* osthread) {
4432 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4433 assert_status(status == 0, status, "pthread_kill");
4434 return status;
4435 }
4436
4437 // "Randomly" selected value for how long we want to spin
4438 // before bailing out on suspending a thread, also how often
4439 // we send a signal to a thread we want to resume
4440 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4441 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4442
4443 // returns true on success and false on error - really an error is fatal
4444 // but this seems the normal response to library errors
4445 static bool do_suspend(OSThread* osthread) {
4446 assert(osthread->sr.is_running(), "thread should be running");
4447 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4448
4449 // mark as suspended and send signal
4450 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4451 // failed to switch, state wasn't running?
4452 ShouldNotReachHere();
4453 return false;
4454 }
4455
4456 if (sr_notify(osthread) != 0) {
4457 ShouldNotReachHere();
4458 }
4459
4460 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4461 while (true) {
4462 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4463 break;
4464 } else {
4465 // timeout
4466 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4467 if (cancelled == os::SuspendResume::SR_RUNNING) {
4468 return false;
4469 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4470 // make sure that we consume the signal on the semaphore as well
4471 sr_semaphore.wait();
4472 break;
4473 } else {
4474 ShouldNotReachHere();
4475 return false;
4476 }
4477 }
4478 }
4479
4480 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4481 return true;
4482 }
4483
4484 static void do_resume(OSThread* osthread) {
4485 assert(osthread->sr.is_suspended(), "thread should be suspended");
4486 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4487
4488 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4489 // failed to switch to WAKEUP_REQUEST
4490 ShouldNotReachHere();
4491 return;
4492 }
4493
4494 while (true) {
4495 if (sr_notify(osthread) == 0) {
4496 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4497 if (osthread->sr.is_running()) {
4498 return;
4499 }
4500 }
4501 } else {
4502 ShouldNotReachHere();
4503 }
4504 }
4505
4506 guarantee(osthread->sr.is_running(), "Must be running!");
4507 }
4508
4509 ////////////////////////////////////////////////////////////////////////////////
4510 // interrupt support
4511
4512 void os::interrupt(Thread* thread) {
4513 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4514 "possibility of dangling Thread pointer");
4515
4516 OSThread* osthread = thread->osthread();
4517
4518 if (!osthread->interrupted()) {
4519 osthread->set_interrupted(true);
4520 // More than one thread can get here with the same value of osthread,
4521 // resulting in multiple notifications. We do, however, want the store
4522 // to interrupted() to be visible to other threads before we execute unpark().
4523 OrderAccess::fence();
4524 ParkEvent * const slp = thread->_SleepEvent ;
4525 if (slp != NULL) slp->unpark() ;
4526 }
4527
4528 // For JSR166. Unpark even if interrupt status already was set
4529 if (thread->is_Java_thread())
4530 ((JavaThread*)thread)->parker()->unpark();
4531
4532 ParkEvent * ev = thread->_ParkEvent ;
4533 if (ev != NULL) ev->unpark() ;
4534
4535 }
4536
4537 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4538 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4539 "possibility of dangling Thread pointer");
4540
4541 OSThread* osthread = thread->osthread();
4542
4543 bool interrupted = osthread->interrupted();
4544
4545 if (interrupted && clear_interrupted) {
4546 osthread->set_interrupted(false);
4547 // consider thread->_SleepEvent->reset() ... optional optimization
4548 }
4549
4550 return interrupted;
4551 }
4552
4553 ///////////////////////////////////////////////////////////////////////////////////
4554 // signal handling (except suspend/resume)
4555
4556 // This routine may be used by user applications as a "hook" to catch signals.
4557 // The user-defined signal handler must pass unrecognized signals to this
4558 // routine, and if it returns true (non-zero), then the signal handler must
4559 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4560 // routine will never retun false (zero), but instead will execute a VM panic
4561 // routine kill the process.
4562 //
4563 // If this routine returns false, it is OK to call it again. This allows
4564 // the user-defined signal handler to perform checks either before or after
4565 // the VM performs its own checks. Naturally, the user code would be making
4566 // a serious error if it tried to handle an exception (such as a null check
4567 // or breakpoint) that the VM was generating for its own correct operation.
4568 //
4569 // This routine may recognize any of the following kinds of signals:
4570 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4571 // It should be consulted by handlers for any of those signals.
4572 //
4573 // The caller of this routine must pass in the three arguments supplied
4574 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4575 // field of the structure passed to sigaction(). This routine assumes that
4576 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4577 //
4578 // Note that the VM will print warnings if it detects conflicting signal
4579 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4580 //
4581 extern "C" JNIEXPORT int
4582 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4583 void* ucontext, int abort_if_unrecognized);
4584
4585 void signalHandler(int sig, siginfo_t* info, void* uc) {
4586 assert(info != NULL && uc != NULL, "it must be old kernel");
4587 int orig_errno = errno; // Preserve errno value over signal handler.
4588 JVM_handle_linux_signal(sig, info, uc, true);
4589 errno = orig_errno;
4590 }
4591
4592
4593 // This boolean allows users to forward their own non-matching signals
4594 // to JVM_handle_linux_signal, harmlessly.
4595 bool os::Linux::signal_handlers_are_installed = false;
4596
4597 // For signal-chaining
4598 struct sigaction os::Linux::sigact[MAXSIGNUM];
4599 unsigned int os::Linux::sigs = 0;
4600 bool os::Linux::libjsig_is_loaded = false;
4601 typedef struct sigaction *(*get_signal_t)(int);
4602 get_signal_t os::Linux::get_signal_action = NULL;
4603
4604 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4605 struct sigaction *actp = NULL;
4606
4607 if (libjsig_is_loaded) {
4608 // Retrieve the old signal handler from libjsig
4609 actp = (*get_signal_action)(sig);
4610 }
4611 if (actp == NULL) {
4612 // Retrieve the preinstalled signal handler from jvm
4613 actp = get_preinstalled_handler(sig);
4614 }
4615
4616 return actp;
4617 }
4618
4619 static bool call_chained_handler(struct sigaction *actp, int sig,
4620 siginfo_t *siginfo, void *context) {
4621 // Call the old signal handler
4622 if (actp->sa_handler == SIG_DFL) {
4623 // It's more reasonable to let jvm treat it as an unexpected exception
4624 // instead of taking the default action.
4625 return false;
4626 } else if (actp->sa_handler != SIG_IGN) {
4627 if ((actp->sa_flags & SA_NODEFER) == 0) {
4628 // automaticlly block the signal
4629 sigaddset(&(actp->sa_mask), sig);
4630 }
4631
4632 sa_handler_t hand = NULL;
4633 sa_sigaction_t sa = NULL;
4634 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4635 // retrieve the chained handler
4636 if (siginfo_flag_set) {
4637 sa = actp->sa_sigaction;
4638 } else {
4639 hand = actp->sa_handler;
4640 }
4641
4642 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4643 actp->sa_handler = SIG_DFL;
4644 }
4645
4646 // try to honor the signal mask
4647 sigset_t oset;
4648 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4649
4650 // call into the chained handler
4651 if (siginfo_flag_set) {
4652 (*sa)(sig, siginfo, context);
4653 } else {
4654 (*hand)(sig);
4655 }
4656
4657 // restore the signal mask
4658 pthread_sigmask(SIG_SETMASK, &oset, 0);
4659 }
4660 // Tell jvm's signal handler the signal is taken care of.
4661 return true;
4662 }
4663
4664 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4665 bool chained = false;
4666 // signal-chaining
4667 if (UseSignalChaining) {
4668 struct sigaction *actp = get_chained_signal_action(sig);
4669 if (actp != NULL) {
4670 chained = call_chained_handler(actp, sig, siginfo, context);
4671 }
4672 }
4673 return chained;
4674 }
4675
4676 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4677 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4678 return &sigact[sig];
4679 }
4680 return NULL;
4681 }
4682
4683 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4684 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4685 sigact[sig] = oldAct;
4686 sigs |= (unsigned int)1 << sig;
4687 }
4688
4689 // for diagnostic
4690 int os::Linux::sigflags[MAXSIGNUM];
4691
4692 int os::Linux::get_our_sigflags(int sig) {
4693 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4694 return sigflags[sig];
4695 }
4696
4697 void os::Linux::set_our_sigflags(int sig, int flags) {
4698 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4699 sigflags[sig] = flags;
4700 }
4701
4702 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4703 // Check for overwrite.
4704 struct sigaction oldAct;
4705 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4706
4707 void* oldhand = oldAct.sa_sigaction
4708 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4709 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4710 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4711 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4712 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4713 if (AllowUserSignalHandlers || !set_installed) {
4714 // Do not overwrite; user takes responsibility to forward to us.
4715 return;
4716 } else if (UseSignalChaining) {
4717 // save the old handler in jvm
4718 save_preinstalled_handler(sig, oldAct);
4719 // libjsig also interposes the sigaction() call below and saves the
4720 // old sigaction on it own.
4721 } else {
4722 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4723 "%#lx for signal %d.", (long)oldhand, sig));
4724 }
4725 }
4726
4727 struct sigaction sigAct;
4728 sigfillset(&(sigAct.sa_mask));
4729 sigAct.sa_handler = SIG_DFL;
4730 if (!set_installed) {
4731 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4732 } else {
4733 sigAct.sa_sigaction = signalHandler;
4734 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4735 }
4736 // Save flags, which are set by ours
4737 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4738 sigflags[sig] = sigAct.sa_flags;
4739
4740 int ret = sigaction(sig, &sigAct, &oldAct);
4741 assert(ret == 0, "check");
4742
4743 void* oldhand2 = oldAct.sa_sigaction
4744 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4745 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4746 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4747 }
4748
4749 // install signal handlers for signals that HotSpot needs to
4750 // handle in order to support Java-level exception handling.
4751
4752 void os::Linux::install_signal_handlers() {
4753 if (!signal_handlers_are_installed) {
4754 signal_handlers_are_installed = true;
4755
4756 // signal-chaining
4757 typedef void (*signal_setting_t)();
4758 signal_setting_t begin_signal_setting = NULL;
4759 signal_setting_t end_signal_setting = NULL;
4760 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4761 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4762 if (begin_signal_setting != NULL) {
4763 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4764 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4765 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4766 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4767 libjsig_is_loaded = true;
4768 assert(UseSignalChaining, "should enable signal-chaining");
4769 }
4770 if (libjsig_is_loaded) {
4771 // Tell libjsig jvm is setting signal handlers
4772 (*begin_signal_setting)();
4773 }
4774
4775 set_signal_handler(SIGSEGV, true);
4776 set_signal_handler(SIGPIPE, true);
4777 set_signal_handler(SIGBUS, true);
4778 set_signal_handler(SIGILL, true);
4779 set_signal_handler(SIGFPE, true);
4780 #if defined(PPC64)
4781 set_signal_handler(SIGTRAP, true);
4782 #endif
4783 set_signal_handler(SIGXFSZ, true);
4784
4785 if (libjsig_is_loaded) {
4786 // Tell libjsig jvm finishes setting signal handlers
4787 (*end_signal_setting)();
4788 }
4789
4790 // We don't activate signal checker if libjsig is in place, we trust ourselves
4791 // and if UserSignalHandler is installed all bets are off.
4792 // Log that signal checking is off only if -verbose:jni is specified.
4793 if (CheckJNICalls) {
4794 if (libjsig_is_loaded) {
4795 if (PrintJNIResolving) {
4796 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4797 }
4798 check_signals = false;
4799 }
4800 if (AllowUserSignalHandlers) {
4801 if (PrintJNIResolving) {
4802 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4803 }
4804 check_signals = false;
4805 }
4806 }
4807 }
4808 }
4809
4810 // This is the fastest way to get thread cpu time on Linux.
4811 // Returns cpu time (user+sys) for any thread, not only for current.
4812 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4813 // It might work on 2.6.10+ with a special kernel/glibc patch.
4814 // For reference, please, see IEEE Std 1003.1-2004:
4815 // http://www.unix.org/single_unix_specification
4816
4817 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4818 struct timespec tp;
4819 int rc = os::Linux::clock_gettime(clockid, &tp);
4820 assert(rc == 0, "clock_gettime is expected to return 0 code");
4821
4822 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4823 }
4824
4825 /////
4826 // glibc on Linux platform uses non-documented flag
4827 // to indicate, that some special sort of signal
4828 // trampoline is used.
4829 // We will never set this flag, and we should
4830 // ignore this flag in our diagnostic
4831 #ifdef SIGNIFICANT_SIGNAL_MASK
4832 #undef SIGNIFICANT_SIGNAL_MASK
4833 #endif
4834 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4835
4836 static const char* get_signal_handler_name(address handler,
4837 char* buf, int buflen) {
4838 int offset = 0;
4839 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4840 if (found) {
4841 // skip directory names
4842 const char *p1, *p2;
4843 p1 = buf;
4844 size_t len = strlen(os::file_separator());
4845 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4846 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4847 } else {
4848 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4849 }
4850 return buf;
4851 }
4852
4853 static void print_signal_handler(outputStream* st, int sig,
4854 char* buf, size_t buflen) {
4855 struct sigaction sa;
4856
4857 sigaction(sig, NULL, &sa);
4858
4859 // See comment for SIGNIFICANT_SIGNAL_MASK define
4860 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4861
4862 st->print("%s: ", os::exception_name(sig, buf, buflen));
4863
4864 address handler = (sa.sa_flags & SA_SIGINFO)
4865 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4866 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4867
4868 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4869 st->print("SIG_DFL");
4870 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4871 st->print("SIG_IGN");
4872 } else {
4873 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4874 }
4875
4876 st->print(", sa_mask[0]=");
4877 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4878
4879 address rh = VMError::get_resetted_sighandler(sig);
4880 // May be, handler was resetted by VMError?
4881 if(rh != NULL) {
4882 handler = rh;
4883 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4884 }
4885
4886 st->print(", sa_flags=");
4887 os::Posix::print_sa_flags(st, sa.sa_flags);
4888
4889 // Check: is it our handler?
4890 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4891 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4892 // It is our signal handler
4893 // check for flags, reset system-used one!
4894 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4895 st->print(
4896 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4897 os::Linux::get_our_sigflags(sig));
4898 }
4899 }
4900 st->cr();
4901 }
4902
4903
4904 #define DO_SIGNAL_CHECK(sig) \
4905 if (!sigismember(&check_signal_done, sig)) \
4906 os::Linux::check_signal_handler(sig)
4907
4908 // This method is a periodic task to check for misbehaving JNI applications
4909 // under CheckJNI, we can add any periodic checks here
4910
4911 void os::run_periodic_checks() {
4912
4913 if (check_signals == false) return;
4914
4915 // SEGV and BUS if overridden could potentially prevent
4916 // generation of hs*.log in the event of a crash, debugging
4917 // such a case can be very challenging, so we absolutely
4918 // check the following for a good measure:
4919 DO_SIGNAL_CHECK(SIGSEGV);
4920 DO_SIGNAL_CHECK(SIGILL);
4921 DO_SIGNAL_CHECK(SIGFPE);
4922 DO_SIGNAL_CHECK(SIGBUS);
4923 DO_SIGNAL_CHECK(SIGPIPE);
4924 DO_SIGNAL_CHECK(SIGXFSZ);
4925 #if defined(PPC64)
4926 DO_SIGNAL_CHECK(SIGTRAP);
4927 #endif
4928
4929 // ReduceSignalUsage allows the user to override these handlers
4930 // see comments at the very top and jvm_solaris.h
4931 if (!ReduceSignalUsage) {
4932 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4933 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4934 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4935 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4936 }
4937
4938 DO_SIGNAL_CHECK(SR_signum);
4939 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4940 }
4941
4942 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4943
4944 static os_sigaction_t os_sigaction = NULL;
4945
4946 void os::Linux::check_signal_handler(int sig) {
4947 char buf[O_BUFLEN];
4948 address jvmHandler = NULL;
4949
4950
4951 struct sigaction act;
4952 if (os_sigaction == NULL) {
4953 // only trust the default sigaction, in case it has been interposed
4954 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4955 if (os_sigaction == NULL) return;
4956 }
4957
4958 os_sigaction(sig, (struct sigaction*)NULL, &act);
4959
4960
4961 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4962
4963 address thisHandler = (act.sa_flags & SA_SIGINFO)
4964 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4965 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4966
4967
4968 switch(sig) {
4969 case SIGSEGV:
4970 case SIGBUS:
4971 case SIGFPE:
4972 case SIGPIPE:
4973 case SIGILL:
4974 case SIGXFSZ:
4975 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4976 break;
4977
4978 case SHUTDOWN1_SIGNAL:
4979 case SHUTDOWN2_SIGNAL:
4980 case SHUTDOWN3_SIGNAL:
4981 case BREAK_SIGNAL:
4982 jvmHandler = (address)user_handler();
4983 break;
4984
4985 case INTERRUPT_SIGNAL:
4986 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4987 break;
4988
4989 default:
4990 if (sig == SR_signum) {
4991 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4992 } else {
4993 return;
4994 }
4995 break;
4996 }
4997
4998 if (thisHandler != jvmHandler) {
4999 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5000 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5001 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5002 // No need to check this sig any longer
5003 sigaddset(&check_signal_done, sig);
5004 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5005 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5006 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5007 exception_name(sig, buf, O_BUFLEN));
5008 }
5009 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5010 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5011 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
5012 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
5013 // No need to check this sig any longer
5014 sigaddset(&check_signal_done, sig);
5015 }
5016
5017 // Dump all the signal
5018 if (sigismember(&check_signal_done, sig)) {
5019 print_signal_handlers(tty, buf, O_BUFLEN);
5020 }
5021 }
5022
5023 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
5024
5025 extern bool signal_name(int signo, char* buf, size_t len);
5026
5027 const char* os::exception_name(int exception_code, char* buf, size_t size) {
5028 if (0 < exception_code && exception_code <= SIGRTMAX) {
5029 // signal
5030 if (!signal_name(exception_code, buf, size)) {
5031 jio_snprintf(buf, size, "SIG%d", exception_code);
5032 }
5033 return buf;
5034 } else {
5035 return NULL;
5036 }
5037 }
5038
5039 // this is called _before_ most of the global arguments have been parsed
5040 void os::init(void) {
5041 char dummy; /* used to get a guess on initial stack address */
5042
5043 // With LinuxThreads the JavaMain thread pid (primordial thread)
5044 // is different than the pid of the java launcher thread.
5045 // So, on Linux, the launcher thread pid is passed to the VM
5046 // via the sun.java.launcher.pid property.
5047 // Use this property instead of getpid() if it was correctly passed.
5048 // See bug 6351349.
5049 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5050
5051 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5052
5053 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5054
5055 init_random(1234567);
5056
5057 ThreadCritical::initialize();
5058
5059 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5060 if (Linux::page_size() == -1) {
5061 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5062 strerror(errno)));
5063 }
5064 init_page_sizes((size_t) Linux::page_size());
5065
5066 Linux::initialize_system_info();
5067
5068 // _main_thread points to the thread that created/loaded the JVM.
5069 Linux::_main_thread = pthread_self();
5070
5071 Linux::clock_init();
5072 initial_time_count = javaTimeNanos();
5073
5074 // pthread_condattr initialization for monotonic clock
5075 int status;
5076 pthread_condattr_t* _condattr = os::Linux::condAttr();
5077 if ((status = pthread_condattr_init(_condattr)) != 0) {
5078 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5079 }
5080 // Only set the clock if CLOCK_MONOTONIC is available
5081 if (Linux::supports_monotonic_clock()) {
5082 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5083 if (status == EINVAL) {
5084 warning("Unable to use monotonic clock with relative timed-waits" \
5085 " - changes to the time-of-day clock may have adverse affects");
5086 } else {
5087 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5088 }
5089 }
5090 }
5091 // else it defaults to CLOCK_REALTIME
5092
5093 pthread_mutex_init(&dl_mutex, NULL);
5094
5095 // If the pagesize of the VM is greater than 8K determine the appropriate
5096 // number of initial guard pages. The user can change this with the
5097 // command line arguments, if needed.
5098 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5099 StackYellowPages = 1;
5100 StackRedPages = 1;
5101 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5102 }
5103
5104 // retrieve entry point for pthread_setname_np
5105 Linux::_pthread_setname_np =
5106 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5107
5108 }
5109
5110 // To install functions for atexit system call
5111 extern "C" {
5112 static void perfMemory_exit_helper() {
5113 perfMemory_exit();
5114 }
5115 }
5116
5117 void os::pd_init_container_support() {
5118 OSContainer::init();
5119 }
5120
5121 // this is called _after_ the global arguments have been parsed
5122 jint os::init_2(void)
5123 {
5124 Linux::fast_thread_clock_init();
5125
5126 // Allocate a single page and mark it as readable for safepoint polling
5127 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5128 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5129
5130 os::set_polling_page( polling_page );
5131
5132 #ifndef PRODUCT
5133 if(Verbose && PrintMiscellaneous)
5134 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5135 #endif
5136
5137 if (!UseMembar) {
5138 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5139 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5140 os::set_memory_serialize_page( mem_serialize_page );
5141
5142 #ifndef PRODUCT
5143 if(Verbose && PrintMiscellaneous)
5144 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5145 #endif
5146 }
5147
5148 // initialize suspend/resume support - must do this before signal_sets_init()
5149 if (SR_initialize() != 0) {
5150 perror("SR_initialize failed");
5151 return JNI_ERR;
5152 }
5153
5154 Linux::signal_sets_init();
5155 Linux::install_signal_handlers();
5156
5157 // Check minimum allowable stack size for thread creation and to initialize
5158 // the java system classes, including StackOverflowError - depends on page
5159 // size. Add a page for compiler2 recursion in main thread.
5160 // Add in 2*BytesPerWord times page size to account for VM stack during
5161 // class initialization depending on 32 or 64 bit VM.
5162 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5163 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5164 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5165
5166 size_t threadStackSizeInBytes = ThreadStackSize * K;
5167 if (threadStackSizeInBytes != 0 &&
5168 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
5169 tty->print_cr("\nThe stack size specified is too small, "
5170 "Specify at least %dk",
5171 os::Linux::min_stack_allowed/ K);
5172 return JNI_ERR;
5173 }
5174
5175 // Make the stack size a multiple of the page size so that
5176 // the yellow/red zones can be guarded.
5177 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5178 vm_page_size()));
5179
5180 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5181
5182 #if defined(IA32)
5183 workaround_expand_exec_shield_cs_limit();
5184 #endif
5185
5186 Linux::libpthread_init();
5187 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5188 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5189 Linux::glibc_version(), Linux::libpthread_version(),
5190 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5191 }
5192
5193 if (UseNUMA) {
5194 if (!Linux::libnuma_init()) {
5195 UseNUMA = false;
5196 } else {
5197 if ((Linux::numa_max_node() < 1)) {
5198 // There's only one node(they start from 0), disable NUMA.
5199 UseNUMA = false;
5200 }
5201 }
5202 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5203 // we can make the adaptive lgrp chunk resizing work. If the user specified
5204 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5205 // disable adaptive resizing.
5206 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5207 if (FLAG_IS_DEFAULT(UseNUMA)) {
5208 UseNUMA = false;
5209 } else {
5210 if (FLAG_IS_DEFAULT(UseLargePages) &&
5211 FLAG_IS_DEFAULT(UseSHM) &&
5212 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5213 UseLargePages = false;
5214 } else {
5215 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5216 UseAdaptiveSizePolicy = false;
5217 UseAdaptiveNUMAChunkSizing = false;
5218 }
5219 }
5220 }
5221 if (!UseNUMA && ForceNUMA) {
5222 UseNUMA = true;
5223 }
5224 }
5225
5226 if (MaxFDLimit) {
5227 // set the number of file descriptors to max. print out error
5228 // if getrlimit/setrlimit fails but continue regardless.
5229 struct rlimit nbr_files;
5230 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5231 if (status != 0) {
5232 if (PrintMiscellaneous && (Verbose || WizardMode))
5233 perror("os::init_2 getrlimit failed");
5234 } else {
5235 nbr_files.rlim_cur = nbr_files.rlim_max;
5236 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5237 if (status != 0) {
5238 if (PrintMiscellaneous && (Verbose || WizardMode))
5239 perror("os::init_2 setrlimit failed");
5240 }
5241 }
5242 }
5243
5244 // Initialize lock used to serialize thread creation (see os::create_thread)
5245 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5246
5247 // at-exit methods are called in the reverse order of their registration.
5248 // atexit functions are called on return from main or as a result of a
5249 // call to exit(3C). There can be only 32 of these functions registered
5250 // and atexit() does not set errno.
5251
5252 if (PerfAllowAtExitRegistration) {
5253 // only register atexit functions if PerfAllowAtExitRegistration is set.
5254 // atexit functions can be delayed until process exit time, which
5255 // can be problematic for embedded VM situations. Embedded VMs should
5256 // call DestroyJavaVM() to assure that VM resources are released.
5257
5258 // note: perfMemory_exit_helper atexit function may be removed in
5259 // the future if the appropriate cleanup code can be added to the
5260 // VM_Exit VMOperation's doit method.
5261 if (atexit(perfMemory_exit_helper) != 0) {
5262 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5263 }
5264 }
5265
5266 // initialize thread priority policy
5267 prio_init();
5268
5269 return JNI_OK;
5270 }
5271
5272 // Mark the polling page as unreadable
5273 void os::make_polling_page_unreadable(void) {
5274 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5275 fatal("Could not disable polling page");
5276 };
5277
5278 // Mark the polling page as readable
5279 void os::make_polling_page_readable(void) {
5280 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5281 fatal("Could not enable polling page");
5282 }
5283 };
5284
5285 static int os_cpu_count(const cpu_set_t* cpus) {
5286 int count = 0;
5287 // only look up to the number of configured processors
5288 for (int i = 0; i < os::processor_count(); i++) {
5289 if (CPU_ISSET(i, cpus)) {
5290 count++;
5291 }
5292 }
5293 return count;
5294 }
5295
5296 // Get the current number of available processors for this process.
5297 // This value can change at any time during a process's lifetime.
5298 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5299 // If anything goes wrong we fallback to returning the number of online
5300 // processors - which can be greater than the number available to the process.
5301 int os::Linux::active_processor_count() {
5302 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5303 int cpus_size = sizeof(cpu_set_t);
5304 int cpu_count = 0;
5305
5306 // pid 0 means the current thread - which we have to assume represents the process
5307 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5308 cpu_count = os_cpu_count(&cpus);
5309 if (PrintActiveCpus) {
5310 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5311 }
5312 }
5313 else {
5314 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5315 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5316 "which may exceed available processors", strerror(errno), cpu_count);
5317 }
5318
5319 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5320 return cpu_count;
5321 }
5322
5323 // Determine the active processor count from one of
5324 // three different sources:
5325 //
5326 // 1. User option -XX:ActiveProcessorCount
5327 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5328 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5329 //
5330 // Option 1, if specified, will always override.
5331 // If the cgroup subsystem is active and configured, we
5332 // will return the min of the cgroup and option 2 results.
5333 // This is required since tools, such as numactl, that
5334 // alter cpu affinity do not update cgroup subsystem
5335 // cpuset configuration files.
5336 int os::active_processor_count() {
5337 // User has overridden the number of active processors
5338 if (ActiveProcessorCount > 0) {
5339 if (PrintActiveCpus) {
5340 tty->print_cr("active_processor_count: "
5341 "active processor count set by user : %d",
5342 ActiveProcessorCount);
5343 }
5344 return ActiveProcessorCount;
5345 }
5346
5347 int active_cpus;
5348 if (OSContainer::is_containerized()) {
5349 active_cpus = OSContainer::active_processor_count();
5350 if (PrintActiveCpus) {
5351 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5352 active_cpus);
5353 }
5354 } else {
5355 active_cpus = os::Linux::active_processor_count();
5356 }
5357
5358 return active_cpus;
5359 }
5360
5361 void os::set_native_thread_name(const char *name) {
5362 if (Linux::_pthread_setname_np) {
5363 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5364 snprintf(buf, sizeof(buf), "%s", name);
5365 buf[sizeof(buf) - 1] = '\0';
5366 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5367 // ERANGE should not happen; all other errors should just be ignored.
5368 assert(rc != ERANGE, "pthread_setname_np failed");
5369 }
5370 }
5371
5372 bool os::distribute_processes(uint length, uint* distribution) {
5373 // Not yet implemented.
5374 return false;
5375 }
5376
5377 bool os::bind_to_processor(uint processor_id) {
5378 // Not yet implemented.
5379 return false;
5380 }
5381
5382 ///
5383
5384 void os::SuspendedThreadTask::internal_do_task() {
5385 if (do_suspend(_thread->osthread())) {
5386 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5387 do_task(context);
5388 do_resume(_thread->osthread());
5389 }
5390 }
5391
5392 class PcFetcher : public os::SuspendedThreadTask {
5393 public:
5394 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5395 ExtendedPC result();
5396 protected:
5397 void do_task(const os::SuspendedThreadTaskContext& context);
5398 private:
5399 ExtendedPC _epc;
5400 };
5401
5402 ExtendedPC PcFetcher::result() {
5403 guarantee(is_done(), "task is not done yet.");
5404 return _epc;
5405 }
5406
5407 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5408 Thread* thread = context.thread();
5409 OSThread* osthread = thread->osthread();
5410 if (osthread->ucontext() != NULL) {
5411 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5412 } else {
5413 // NULL context is unexpected, double-check this is the VMThread
5414 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5415 }
5416 }
5417
5418 // Suspends the target using the signal mechanism and then grabs the PC before
5419 // resuming the target. Used by the flat-profiler only
5420 ExtendedPC os::get_thread_pc(Thread* thread) {
5421 // Make sure that it is called by the watcher for the VMThread
5422 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5423 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5424
5425 PcFetcher fetcher(thread);
5426 fetcher.run();
5427 return fetcher.result();
5428 }
5429
5430 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5431 {
5432 if (is_NPTL()) {
5433 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5434 } else {
5435 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5436 // word back to default 64bit precision if condvar is signaled. Java
5437 // wants 53bit precision. Save and restore current value.
5438 int fpu = get_fpu_control_word();
5439 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5440 set_fpu_control_word(fpu);
5441 return status;
5442 }
5443 }
5444
5445 ////////////////////////////////////////////////////////////////////////////////
5446 // debug support
5447
5448 bool os::find(address addr, outputStream* st) {
5449 Dl_info dlinfo;
5450 memset(&dlinfo, 0, sizeof(dlinfo));
5451 if (dladdr(addr, &dlinfo) != 0) {
5452 st->print(PTR_FORMAT ": ", addr);
5453 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5454 st->print("%s+%#x", dlinfo.dli_sname,
5455 addr - (intptr_t)dlinfo.dli_saddr);
5456 } else if (dlinfo.dli_fbase != NULL) {
5457 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5458 } else {
5459 st->print("<absolute address>");
5460 }
5461 if (dlinfo.dli_fname != NULL) {
5462 st->print(" in %s", dlinfo.dli_fname);
5463 }
5464 if (dlinfo.dli_fbase != NULL) {
5465 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5466 }
5467 st->cr();
5468
5469 if (Verbose) {
5470 // decode some bytes around the PC
5471 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5472 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5473 address lowest = (address) dlinfo.dli_sname;
5474 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5475 if (begin < lowest) begin = lowest;
5476 Dl_info dlinfo2;
5477 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5478 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5479 end = (address) dlinfo2.dli_saddr;
5480 Disassembler::decode(begin, end, st);
5481 }
5482 return true;
5483 }
5484 return false;
5485 }
5486
5487 ////////////////////////////////////////////////////////////////////////////////
5488 // misc
5489
5490 // This does not do anything on Linux. This is basically a hook for being
5491 // able to use structured exception handling (thread-local exception filters)
5492 // on, e.g., Win32.
5493 void
5494 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5495 JavaCallArguments* args, Thread* thread) {
5496 f(value, method, args, thread);
5497 }
5498
5499 void os::print_statistics() {
5500 }
5501
5502 int os::message_box(const char* title, const char* message) {
5503 int i;
5504 fdStream err(defaultStream::error_fd());
5505 for (i = 0; i < 78; i++) err.print_raw("=");
5506 err.cr();
5507 err.print_raw_cr(title);
5508 for (i = 0; i < 78; i++) err.print_raw("-");
5509 err.cr();
5510 err.print_raw_cr(message);
5511 for (i = 0; i < 78; i++) err.print_raw("=");
5512 err.cr();
5513
5514 char buf[16];
5515 // Prevent process from exiting upon "read error" without consuming all CPU
5516 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5517
5518 return buf[0] == 'y' || buf[0] == 'Y';
5519 }
5520
5521 int os::stat(const char *path, struct stat *sbuf) {
5522 char pathbuf[MAX_PATH];
5523 if (strlen(path) > MAX_PATH - 1) {
5524 errno = ENAMETOOLONG;
5525 return -1;
5526 }
5527 os::native_path(strcpy(pathbuf, path));
5528 return ::stat(pathbuf, sbuf);
5529 }
5530
5531 bool os::check_heap(bool force) {
5532 return true;
5533 }
5534
5535 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5536 return ::vsnprintf(buf, count, format, args);
5537 }
5538
5539 // Is a (classpath) directory empty?
5540 bool os::dir_is_empty(const char* path) {
5541 DIR *dir = NULL;
5542 struct dirent *ptr;
5543
5544 dir = opendir(path);
5545 if (dir == NULL) return true;
5546
5547 /* Scan the directory */
5548 bool result = true;
5549 while (result && (ptr = readdir(dir)) != NULL) {
5550 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5551 result = false;
5552 }
5553 }
5554 closedir(dir);
5555 return result;
5556 }
5557
5558 // This code originates from JDK's sysOpen and open64_w
5559 // from src/solaris/hpi/src/system_md.c
5560
5561 #ifndef O_DELETE
5562 #define O_DELETE 0x10000
5563 #endif
5564
5565 // Open a file. Unlink the file immediately after open returns
5566 // if the specified oflag has the O_DELETE flag set.
5567 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5568
5569 int os::open(const char *path, int oflag, int mode) {
5570
5571 if (strlen(path) > MAX_PATH - 1) {
5572 errno = ENAMETOOLONG;
5573 return -1;
5574 }
5575 int fd;
5576 int o_delete = (oflag & O_DELETE);
5577 oflag = oflag & ~O_DELETE;
5578
5579 fd = ::open64(path, oflag, mode);
5580 if (fd == -1) return -1;
5581
5582 //If the open succeeded, the file might still be a directory
5583 {
5584 struct stat64 buf64;
5585 int ret = ::fstat64(fd, &buf64);
5586 int st_mode = buf64.st_mode;
5587
5588 if (ret != -1) {
5589 if ((st_mode & S_IFMT) == S_IFDIR) {
5590 errno = EISDIR;
5591 ::close(fd);
5592 return -1;
5593 }
5594 } else {
5595 ::close(fd);
5596 return -1;
5597 }
5598 }
5599
5600 /*
5601 * All file descriptors that are opened in the JVM and not
5602 * specifically destined for a subprocess should have the
5603 * close-on-exec flag set. If we don't set it, then careless 3rd
5604 * party native code might fork and exec without closing all
5605 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5606 * UNIXProcess.c), and this in turn might:
5607 *
5608 * - cause end-of-file to fail to be detected on some file
5609 * descriptors, resulting in mysterious hangs, or
5610 *
5611 * - might cause an fopen in the subprocess to fail on a system
5612 * suffering from bug 1085341.
5613 *
5614 * (Yes, the default setting of the close-on-exec flag is a Unix
5615 * design flaw)
5616 *
5617 * See:
5618 * 1085341: 32-bit stdio routines should support file descriptors >255
5619 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5620 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5621 */
5622 #ifdef FD_CLOEXEC
5623 {
5624 int flags = ::fcntl(fd, F_GETFD);
5625 if (flags != -1)
5626 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5627 }
5628 #endif
5629
5630 if (o_delete != 0) {
5631 ::unlink(path);
5632 }
5633 return fd;
5634 }
5635
5636
5637 // create binary file, rewriting existing file if required
5638 int os::create_binary_file(const char* path, bool rewrite_existing) {
5639 int oflags = O_WRONLY | O_CREAT;
5640 if (!rewrite_existing) {
5641 oflags |= O_EXCL;
5642 }
5643 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5644 }
5645
5646 // return current position of file pointer
5647 jlong os::current_file_offset(int fd) {
5648 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5649 }
5650
5651 // move file pointer to the specified offset
5652 jlong os::seek_to_file_offset(int fd, jlong offset) {
5653 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5654 }
5655
5656 // This code originates from JDK's sysAvailable
5657 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5658
5659 int os::available(int fd, jlong *bytes) {
5660 jlong cur, end;
5661 int mode;
5662 struct stat64 buf64;
5663
5664 if (::fstat64(fd, &buf64) >= 0) {
5665 mode = buf64.st_mode;
5666 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5667 /*
5668 * XXX: is the following call interruptible? If so, this might
5669 * need to go through the INTERRUPT_IO() wrapper as for other
5670 * blocking, interruptible calls in this file.
5671 */
5672 int n;
5673 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5674 *bytes = n;
5675 return 1;
5676 }
5677 }
5678 }
5679 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5680 return 0;
5681 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5682 return 0;
5683 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5684 return 0;
5685 }
5686 *bytes = end - cur;
5687 return 1;
5688 }
5689
5690 int os::socket_available(int fd, jint *pbytes) {
5691 // Linux doc says EINTR not returned, unlike Solaris
5692 int ret = ::ioctl(fd, FIONREAD, pbytes);
5693
5694 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5695 // is expected to return 0 on failure and 1 on success to the jdk.
5696 return (ret < 0) ? 0 : 1;
5697 }
5698
5699 // Map a block of memory.
5700 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5701 char *addr, size_t bytes, bool read_only,
5702 bool allow_exec) {
5703 int prot;
5704 int flags = MAP_PRIVATE;
5705
5706 if (read_only) {
5707 prot = PROT_READ;
5708 } else {
5709 prot = PROT_READ | PROT_WRITE;
5710 }
5711
5712 if (allow_exec) {
5713 prot |= PROT_EXEC;
5714 }
5715
5716 if (addr != NULL) {
5717 flags |= MAP_FIXED;
5718 }
5719
5720 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5721 fd, file_offset);
5722 if (mapped_address == MAP_FAILED) {
5723 return NULL;
5724 }
5725 return mapped_address;
5726 }
5727
5728
5729 // Remap a block of memory.
5730 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5731 char *addr, size_t bytes, bool read_only,
5732 bool allow_exec) {
5733 // same as map_memory() on this OS
5734 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5735 allow_exec);
5736 }
5737
5738
5739 // Unmap a block of memory.
5740 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5741 return munmap(addr, bytes) == 0;
5742 }
5743
5744 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5745
5746 static clockid_t thread_cpu_clockid(Thread* thread) {
5747 pthread_t tid = thread->osthread()->pthread_id();
5748 clockid_t clockid;
5749
5750 // Get thread clockid
5751 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5752 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5753 return clockid;
5754 }
5755
5756 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5757 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5758 // of a thread.
5759 //
5760 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5761 // the fast estimate available on the platform.
5762
5763 jlong os::current_thread_cpu_time() {
5764 if (os::Linux::supports_fast_thread_cpu_time()) {
5765 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5766 } else {
5767 // return user + sys since the cost is the same
5768 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5769 }
5770 }
5771
5772 jlong os::thread_cpu_time(Thread* thread) {
5773 // consistent with what current_thread_cpu_time() returns
5774 if (os::Linux::supports_fast_thread_cpu_time()) {
5775 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5776 } else {
5777 return slow_thread_cpu_time(thread, true /* user + sys */);
5778 }
5779 }
5780
5781 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5782 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5783 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5784 } else {
5785 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5786 }
5787 }
5788
5789 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5790 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5791 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5792 } else {
5793 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5794 }
5795 }
5796
5797 //
5798 // -1 on error.
5799 //
5800
5801 PRAGMA_DIAG_PUSH
5802 PRAGMA_FORMAT_NONLITERAL_IGNORED
5803 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5804 static bool proc_task_unchecked = true;
5805 pid_t tid = thread->osthread()->thread_id();
5806 char *s;
5807 char stat[2048];
5808 int statlen;
5809 char proc_name[64];
5810 int count;
5811 long sys_time, user_time;
5812 char cdummy;
5813 int idummy;
5814 long ldummy;
5815 FILE *fp;
5816
5817 snprintf(proc_name, 64, "/proc/%d/stat", tid);
5818
5819 // The /proc/<tid>/stat aggregates per-process usage on
5820 // new Linux kernels 2.6+ where NPTL is supported.
5821 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5822 // See bug 6328462.
5823 // There possibly can be cases where there is no directory
5824 // /proc/self/task, so we check its availability.
5825 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5826 // This is executed only once
5827 proc_task_unchecked = false;
5828 fp = fopen("/proc/self/task", "r");
5829 if (fp != NULL) {
5830 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5831 fclose(fp);
5832 }
5833 }
5834
5835 fp = fopen(proc_name, "r");
5836 if ( fp == NULL ) return -1;
5837 statlen = fread(stat, 1, 2047, fp);
5838 stat[statlen] = '\0';
5839 fclose(fp);
5840
5841 // Skip pid and the command string. Note that we could be dealing with
5842 // weird command names, e.g. user could decide to rename java launcher
5843 // to "java 1.4.2 :)", then the stat file would look like
5844 // 1234 (java 1.4.2 :)) R ... ...
5845 // We don't really need to know the command string, just find the last
5846 // occurrence of ")" and then start parsing from there. See bug 4726580.
5847 s = strrchr(stat, ')');
5848 if (s == NULL ) return -1;
5849
5850 // Skip blank chars
5851 do s++; while (isspace(*s));
5852
5853 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5854 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5855 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5856 &user_time, &sys_time);
5857 if ( count != 13 ) return -1;
5858 if (user_sys_cpu_time) {
5859 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5860 } else {
5861 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5862 }
5863 }
5864 PRAGMA_DIAG_POP
5865
5866 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5867 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5868 info_ptr->may_skip_backward = false; // elapsed time not wall time
5869 info_ptr->may_skip_forward = false; // elapsed time not wall time
5870 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5871 }
5872
5873 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5874 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5875 info_ptr->may_skip_backward = false; // elapsed time not wall time
5876 info_ptr->may_skip_forward = false; // elapsed time not wall time
5877 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5878 }
5879
5880 bool os::is_thread_cpu_time_supported() {
5881 return true;
5882 }
5883
5884 // System loadavg support. Returns -1 if load average cannot be obtained.
5885 // Linux doesn't yet have a (official) notion of processor sets,
5886 // so just return the system wide load average.
5887 int os::loadavg(double loadavg[], int nelem) {
5888 return ::getloadavg(loadavg, nelem);
5889 }
5890
5891 void os::pause() {
5892 char filename[MAX_PATH];
5893 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5894 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5895 } else {
5896 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5897 }
5898
5899 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5900 if (fd != -1) {
5901 struct stat buf;
5902 ::close(fd);
5903 while (::stat(filename, &buf) == 0) {
5904 (void)::poll(NULL, 0, 100);
5905 }
5906 } else {
5907 jio_fprintf(stderr,
5908 "Could not open pause file '%s', continuing immediately.\n", filename);
5909 }
5910 }
5911
5912
5913 // Refer to the comments in os_solaris.cpp park-unpark.
5914 //
5915 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5916 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5917 // For specifics regarding the bug see GLIBC BUGID 261237 :
5918 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5919 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5920 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5921 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5922 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5923 // and monitorenter when we're using 1-0 locking. All those operations may result in
5924 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5925 // of libpthread avoids the problem, but isn't practical.
5926 //
5927 // Possible remedies:
5928 //
5929 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5930 // This is palliative and probabilistic, however. If the thread is preempted
5931 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5932 // than the minimum period may have passed, and the abstime may be stale (in the
5933 // past) resultin in a hang. Using this technique reduces the odds of a hang
5934 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5935 //
5936 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5937 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5938 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5939 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5940 // thread.
5941 //
5942 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5943 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5944 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5945 // This also works well. In fact it avoids kernel-level scalability impediments
5946 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5947 // timers in a graceful fashion.
5948 //
5949 // 4. When the abstime value is in the past it appears that control returns
5950 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5951 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5952 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5953 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5954 // It may be possible to avoid reinitialization by checking the return
5955 // value from pthread_cond_timedwait(). In addition to reinitializing the
5956 // condvar we must establish the invariant that cond_signal() is only called
5957 // within critical sections protected by the adjunct mutex. This prevents
5958 // cond_signal() from "seeing" a condvar that's in the midst of being
5959 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5960 // desirable signal-after-unlock optimization that avoids futile context switching.
5961 //
5962 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5963 // structure when a condvar is used or initialized. cond_destroy() would
5964 // release the helper structure. Our reinitialize-after-timedwait fix
5965 // put excessive stress on malloc/free and locks protecting the c-heap.
5966 //
5967 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5968 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5969 // and only enabling the work-around for vulnerable environments.
5970
5971 // utility to compute the abstime argument to timedwait:
5972 // millis is the relative timeout time
5973 // abstime will be the absolute timeout time
5974 // TODO: replace compute_abstime() with unpackTime()
5975
5976 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5977 if (millis < 0) millis = 0;
5978
5979 jlong seconds = millis / 1000;
5980 millis %= 1000;
5981 if (seconds > 50000000) { // see man cond_timedwait(3T)
5982 seconds = 50000000;
5983 }
5984
5985 if (os::Linux::supports_monotonic_clock()) {
5986 struct timespec now;
5987 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5988 assert_status(status == 0, status, "clock_gettime");
5989 abstime->tv_sec = now.tv_sec + seconds;
5990 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5991 if (nanos >= NANOSECS_PER_SEC) {
5992 abstime->tv_sec += 1;
5993 nanos -= NANOSECS_PER_SEC;
5994 }
5995 abstime->tv_nsec = nanos;
5996 } else {
5997 struct timeval now;
5998 int status = gettimeofday(&now, NULL);
5999 assert(status == 0, "gettimeofday");
6000 abstime->tv_sec = now.tv_sec + seconds;
6001 long usec = now.tv_usec + millis * 1000;
6002 if (usec >= 1000000) {
6003 abstime->tv_sec += 1;
6004 usec -= 1000000;
6005 }
6006 abstime->tv_nsec = usec * 1000;
6007 }
6008 return abstime;
6009 }
6010
6011
6012 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6013 // Conceptually TryPark() should be equivalent to park(0).
6014
6015 int os::PlatformEvent::TryPark() {
6016 for (;;) {
6017 const int v = _Event ;
6018 guarantee ((v == 0) || (v == 1), "invariant") ;
6019 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
6020 }
6021 }
6022
6023 void os::PlatformEvent::park() { // AKA "down()"
6024 // Invariant: Only the thread associated with the Event/PlatformEvent
6025 // may call park().
6026 // TODO: assert that _Assoc != NULL or _Assoc == Self
6027 int v ;
6028 for (;;) {
6029 v = _Event ;
6030 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6031 }
6032 guarantee (v >= 0, "invariant") ;
6033 if (v == 0) {
6034 // Do this the hard way by blocking ...
6035 int status = pthread_mutex_lock(_mutex);
6036 assert_status(status == 0, status, "mutex_lock");
6037 guarantee (_nParked == 0, "invariant") ;
6038 ++ _nParked ;
6039 while (_Event < 0) {
6040 status = pthread_cond_wait(_cond, _mutex);
6041 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6042 // Treat this the same as if the wait was interrupted
6043 if (status == ETIME) { status = EINTR; }
6044 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6045 }
6046 -- _nParked ;
6047
6048 _Event = 0 ;
6049 status = pthread_mutex_unlock(_mutex);
6050 assert_status(status == 0, status, "mutex_unlock");
6051 // Paranoia to ensure our locked and lock-free paths interact
6052 // correctly with each other.
6053 OrderAccess::fence();
6054 }
6055 guarantee (_Event >= 0, "invariant") ;
6056 }
6057
6058 int os::PlatformEvent::park(jlong millis) {
6059 guarantee (_nParked == 0, "invariant") ;
6060
6061 int v ;
6062 for (;;) {
6063 v = _Event ;
6064 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6065 }
6066 guarantee (v >= 0, "invariant") ;
6067 if (v != 0) return OS_OK ;
6068
6069 // We do this the hard way, by blocking the thread.
6070 // Consider enforcing a minimum timeout value.
6071 struct timespec abst;
6072 compute_abstime(&abst, millis);
6073
6074 int ret = OS_TIMEOUT;
6075 int status = pthread_mutex_lock(_mutex);
6076 assert_status(status == 0, status, "mutex_lock");
6077 guarantee (_nParked == 0, "invariant") ;
6078 ++_nParked ;
6079
6080 // Object.wait(timo) will return because of
6081 // (a) notification
6082 // (b) timeout
6083 // (c) thread.interrupt
6084 //
6085 // Thread.interrupt and object.notify{All} both call Event::set.
6086 // That is, we treat thread.interrupt as a special case of notification.
6087 // The underlying Solaris implementation, cond_timedwait, admits
6088 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6089 // JVM from making those visible to Java code. As such, we must
6090 // filter out spurious wakeups. We assume all ETIME returns are valid.
6091 //
6092 // TODO: properly differentiate simultaneous notify+interrupt.
6093 // In that case, we should propagate the notify to another waiter.
6094
6095 while (_Event < 0) {
6096 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6097 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6098 pthread_cond_destroy (_cond);
6099 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6100 }
6101 assert_status(status == 0 || status == EINTR ||
6102 status == ETIME || status == ETIMEDOUT,
6103 status, "cond_timedwait");
6104 if (!FilterSpuriousWakeups) break ; // previous semantics
6105 if (status == ETIME || status == ETIMEDOUT) break ;
6106 // We consume and ignore EINTR and spurious wakeups.
6107 }
6108 --_nParked ;
6109 if (_Event >= 0) {
6110 ret = OS_OK;
6111 }
6112 _Event = 0 ;
6113 status = pthread_mutex_unlock(_mutex);
6114 assert_status(status == 0, status, "mutex_unlock");
6115 assert (_nParked == 0, "invariant") ;
6116 // Paranoia to ensure our locked and lock-free paths interact
6117 // correctly with each other.
6118 OrderAccess::fence();
6119 return ret;
6120 }
6121
6122 void os::PlatformEvent::unpark() {
6123 // Transitions for _Event:
6124 // 0 :=> 1
6125 // 1 :=> 1
6126 // -1 :=> either 0 or 1; must signal target thread
6127 // That is, we can safely transition _Event from -1 to either
6128 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6129 // unpark() calls.
6130 // See also: "Semaphores in Plan 9" by Mullender & Cox
6131 //
6132 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6133 // that it will take two back-to-back park() calls for the owning
6134 // thread to block. This has the benefit of forcing a spurious return
6135 // from the first park() call after an unpark() call which will help
6136 // shake out uses of park() and unpark() without condition variables.
6137
6138 if (Atomic::xchg(1, &_Event) >= 0) return;
6139
6140 // Wait for the thread associated with the event to vacate
6141 int status = pthread_mutex_lock(_mutex);
6142 assert_status(status == 0, status, "mutex_lock");
6143 int AnyWaiters = _nParked;
6144 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6145 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6146 AnyWaiters = 0;
6147 pthread_cond_signal(_cond);
6148 }
6149 status = pthread_mutex_unlock(_mutex);
6150 assert_status(status == 0, status, "mutex_unlock");
6151 if (AnyWaiters != 0) {
6152 status = pthread_cond_signal(_cond);
6153 assert_status(status == 0, status, "cond_signal");
6154 }
6155
6156 // Note that we signal() _after dropping the lock for "immortal" Events.
6157 // This is safe and avoids a common class of futile wakeups. In rare
6158 // circumstances this can cause a thread to return prematurely from
6159 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6160 // simply re-test the condition and re-park itself.
6161 }
6162
6163
6164 // JSR166
6165 // -------------------------------------------------------
6166
6167 /*
6168 * The solaris and linux implementations of park/unpark are fairly
6169 * conservative for now, but can be improved. They currently use a
6170 * mutex/condvar pair, plus a a count.
6171 * Park decrements count if > 0, else does a condvar wait. Unpark
6172 * sets count to 1 and signals condvar. Only one thread ever waits
6173 * on the condvar. Contention seen when trying to park implies that someone
6174 * is unparking you, so don't wait. And spurious returns are fine, so there
6175 * is no need to track notifications.
6176 */
6177
6178 /*
6179 * This code is common to linux and solaris and will be moved to a
6180 * common place in dolphin.
6181 *
6182 * The passed in time value is either a relative time in nanoseconds
6183 * or an absolute time in milliseconds. Either way it has to be unpacked
6184 * into suitable seconds and nanoseconds components and stored in the
6185 * given timespec structure.
6186 * Given time is a 64-bit value and the time_t used in the timespec is only
6187 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6188 * overflow if times way in the future are given. Further on Solaris versions
6189 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6190 * number of seconds, in abstime, is less than current_time + 100,000,000.
6191 * As it will be 28 years before "now + 100000000" will overflow we can
6192 * ignore overflow and just impose a hard-limit on seconds using the value
6193 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6194 * years from "now".
6195 */
6196
6197 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6198 assert (time > 0, "convertTime");
6199 time_t max_secs = 0;
6200
6201 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6202 struct timeval now;
6203 int status = gettimeofday(&now, NULL);
6204 assert(status == 0, "gettimeofday");
6205
6206 max_secs = now.tv_sec + MAX_SECS;
6207
6208 if (isAbsolute) {
6209 jlong secs = time / 1000;
6210 if (secs > max_secs) {
6211 absTime->tv_sec = max_secs;
6212 } else {
6213 absTime->tv_sec = secs;
6214 }
6215 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6216 } else {
6217 jlong secs = time / NANOSECS_PER_SEC;
6218 if (secs >= MAX_SECS) {
6219 absTime->tv_sec = max_secs;
6220 absTime->tv_nsec = 0;
6221 } else {
6222 absTime->tv_sec = now.tv_sec + secs;
6223 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6224 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6225 absTime->tv_nsec -= NANOSECS_PER_SEC;
6226 ++absTime->tv_sec; // note: this must be <= max_secs
6227 }
6228 }
6229 }
6230 } else {
6231 // must be relative using monotonic clock
6232 struct timespec now;
6233 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6234 assert_status(status == 0, status, "clock_gettime");
6235 max_secs = now.tv_sec + MAX_SECS;
6236 jlong secs = time / NANOSECS_PER_SEC;
6237 if (secs >= MAX_SECS) {
6238 absTime->tv_sec = max_secs;
6239 absTime->tv_nsec = 0;
6240 } else {
6241 absTime->tv_sec = now.tv_sec + secs;
6242 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6243 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6244 absTime->tv_nsec -= NANOSECS_PER_SEC;
6245 ++absTime->tv_sec; // note: this must be <= max_secs
6246 }
6247 }
6248 }
6249 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6250 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6251 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6252 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6253 }
6254
6255 void Parker::park(bool isAbsolute, jlong time) {
6256 // Ideally we'd do something useful while spinning, such
6257 // as calling unpackTime().
6258
6259 // Optional fast-path check:
6260 // Return immediately if a permit is available.
6261 // We depend on Atomic::xchg() having full barrier semantics
6262 // since we are doing a lock-free update to _counter.
6263 if (Atomic::xchg(0, &_counter) > 0) return;
6264
6265 Thread* thread = Thread::current();
6266 assert(thread->is_Java_thread(), "Must be JavaThread");
6267 JavaThread *jt = (JavaThread *)thread;
6268
6269 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6270 // Check interrupt before trying to wait
6271 if (Thread::is_interrupted(thread, false)) {
6272 return;
6273 }
6274
6275 // Next, demultiplex/decode time arguments
6276 timespec absTime;
6277 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6278 return;
6279 }
6280 if (time > 0) {
6281 unpackTime(&absTime, isAbsolute, time);
6282 }
6283
6284
6285 // Enter safepoint region
6286 // Beware of deadlocks such as 6317397.
6287 // The per-thread Parker:: mutex is a classic leaf-lock.
6288 // In particular a thread must never block on the Threads_lock while
6289 // holding the Parker:: mutex. If safepoints are pending both the
6290 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6291 ThreadBlockInVM tbivm(jt);
6292
6293 // Don't wait if cannot get lock since interference arises from
6294 // unblocking. Also. check interrupt before trying wait
6295 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6296 return;
6297 }
6298
6299 int status ;
6300 if (_counter > 0) { // no wait needed
6301 _counter = 0;
6302 status = pthread_mutex_unlock(_mutex);
6303 assert (status == 0, "invariant") ;
6304 // Paranoia to ensure our locked and lock-free paths interact
6305 // correctly with each other and Java-level accesses.
6306 OrderAccess::fence();
6307 return;
6308 }
6309
6310 #ifdef ASSERT
6311 // Don't catch signals while blocked; let the running threads have the signals.
6312 // (This allows a debugger to break into the running thread.)
6313 sigset_t oldsigs;
6314 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6315 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6316 #endif
6317
6318 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6319 jt->set_suspend_equivalent();
6320 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6321
6322 assert(_cur_index == -1, "invariant");
6323 if (time == 0) {
6324 _cur_index = REL_INDEX; // arbitrary choice when not timed
6325 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6326 } else {
6327 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6328 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6329 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6330 pthread_cond_destroy (&_cond[_cur_index]) ;
6331 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6332 }
6333 }
6334 _cur_index = -1;
6335 assert_status(status == 0 || status == EINTR ||
6336 status == ETIME || status == ETIMEDOUT,
6337 status, "cond_timedwait");
6338
6339 #ifdef ASSERT
6340 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6341 #endif
6342
6343 _counter = 0 ;
6344 status = pthread_mutex_unlock(_mutex) ;
6345 assert_status(status == 0, status, "invariant") ;
6346 // Paranoia to ensure our locked and lock-free paths interact
6347 // correctly with each other and Java-level accesses.
6348 OrderAccess::fence();
6349
6350 // If externally suspended while waiting, re-suspend
6351 if (jt->handle_special_suspend_equivalent_condition()) {
6352 jt->java_suspend_self();
6353 }
6354 }
6355
6356 void Parker::unpark() {
6357 int s, status ;
6358 status = pthread_mutex_lock(_mutex);
6359 assert (status == 0, "invariant") ;
6360 s = _counter;
6361 _counter = 1;
6362 if (s < 1) {
6363 // thread might be parked
6364 if (_cur_index != -1) {
6365 // thread is definitely parked
6366 if (WorkAroundNPTLTimedWaitHang) {
6367 status = pthread_cond_signal (&_cond[_cur_index]);
6368 assert (status == 0, "invariant");
6369 status = pthread_mutex_unlock(_mutex);
6370 assert (status == 0, "invariant");
6371 } else {
6372 // must capture correct index before unlocking
6373 int index = _cur_index;
6374 status = pthread_mutex_unlock(_mutex);
6375 assert (status == 0, "invariant");
6376 status = pthread_cond_signal (&_cond[index]);
6377 assert (status == 0, "invariant");
6378 }
6379 } else {
6380 pthread_mutex_unlock(_mutex);
6381 assert (status == 0, "invariant") ;
6382 }
6383 } else {
6384 pthread_mutex_unlock(_mutex);
6385 assert (status == 0, "invariant") ;
6386 }
6387 }
6388
6389
6390 extern char** environ;
6391
6392 // Run the specified command in a separate process. Return its exit value,
6393 // or -1 on failure (e.g. can't fork a new process).
6394 // Unlike system(), this function can be called from signal handler. It
6395 // doesn't block SIGINT et al.
6396 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
6397 const char * argv[4] = {"sh", "-c", cmd, NULL};
6398
6399 pid_t pid ;
6400
6401 if (use_vfork_if_available) {
6402 pid = vfork();
6403 } else {
6404 pid = fork();
6405 }
6406
6407 if (pid < 0) {
6408 // fork failed
6409 return -1;
6410
6411 } else if (pid == 0) {
6412 // child process
6413
6414 execve("/bin/sh", (char* const*)argv, environ);
6415
6416 // execve failed
6417 _exit(-1);
6418
6419 } else {
6420 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6421 // care about the actual exit code, for now.
6422
6423 int status;
6424
6425 // Wait for the child process to exit. This returns immediately if
6426 // the child has already exited. */
6427 while (waitpid(pid, &status, 0) < 0) {
6428 switch (errno) {
6429 case ECHILD: return 0;
6430 case EINTR: break;
6431 default: return -1;
6432 }
6433 }
6434
6435 if (WIFEXITED(status)) {
6436 // The child exited normally; get its exit code.
6437 return WEXITSTATUS(status);
6438 } else if (WIFSIGNALED(status)) {
6439 // The child exited because of a signal
6440 // The best value to return is 0x80 + signal number,
6441 // because that is what all Unix shells do, and because
6442 // it allows callers to distinguish between process exit and
6443 // process death by signal.
6444 return 0x80 + WTERMSIG(status);
6445 } else {
6446 // Unknown exit code; pass it through
6447 return status;
6448 }
6449 }
6450 }
6451
6452 // is_headless_jre()
6453 //
6454 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6455 // in order to report if we are running in a headless jre
6456 //
6457 // Since JDK8 xawt/libmawt.so was moved into the same directory
6458 // as libawt.so, and renamed libawt_xawt.so
6459 //
6460 bool os::is_headless_jre() {
6461 struct stat statbuf;
6462 char buf[MAXPATHLEN];
6463 char libmawtpath[MAXPATHLEN];
6464 const char *xawtstr = "/xawt/libmawt.so";
6465 const char *new_xawtstr = "/libawt_xawt.so";
6466 char *p;
6467
6468 // Get path to libjvm.so
6469 os::jvm_path(buf, sizeof(buf));
6470
6471 // Get rid of libjvm.so
6472 p = strrchr(buf, '/');
6473 if (p == NULL) return false;
6474 else *p = '\0';
6475
6476 // Get rid of client or server
6477 p = strrchr(buf, '/');
6478 if (p == NULL) return false;
6479 else *p = '\0';
6480
6481 // check xawt/libmawt.so
6482 strcpy(libmawtpath, buf);
6483 strcat(libmawtpath, xawtstr);
6484 if (::stat(libmawtpath, &statbuf) == 0) return false;
6485
6486 // check libawt_xawt.so
6487 strcpy(libmawtpath, buf);
6488 strcat(libmawtpath, new_xawtstr);
6489 if (::stat(libmawtpath, &statbuf) == 0) return false;
6490
6491 return true;
6492 }
6493
6494 // Get the default path to the core file
6495 // Returns the length of the string
6496 int os::get_core_path(char* buffer, size_t bufferSize) {
6497 const char* p = get_current_directory(buffer, bufferSize);
6498
6499 if (p == NULL) {
6500 assert(p != NULL, "failed to get current directory");
6501 return 0;
6502 }
6503
6504 return strlen(buffer);
6505 }
6506
6507 /////////////// Unit tests ///////////////
6508
6509 #ifndef PRODUCT
6510
6511 #define test_log(...) \
6512 do {\
6513 if (VerboseInternalVMTests) { \
6514 tty->print_cr(__VA_ARGS__); \
6515 tty->flush(); \
6516 }\
6517 } while (false)
6518
6519 class TestReserveMemorySpecial : AllStatic {
6520 public:
6521 static void small_page_write(void* addr, size_t size) {
6522 size_t page_size = os::vm_page_size();
6523
6524 char* end = (char*)addr + size;
6525 for (char* p = (char*)addr; p < end; p += page_size) {
6526 *p = 1;
6527 }
6528 }
6529
6530 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6531 if (!UseHugeTLBFS) {
6532 return;
6533 }
6534
6535 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6536
6537 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6538
6539 if (addr != NULL) {
6540 small_page_write(addr, size);
6541
6542 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6543 }
6544 }
6545
6546 static void test_reserve_memory_special_huge_tlbfs_only() {
6547 if (!UseHugeTLBFS) {
6548 return;
6549 }
6550
6551 size_t lp = os::large_page_size();
6552
6553 for (size_t size = lp; size <= lp * 10; size += lp) {
6554 test_reserve_memory_special_huge_tlbfs_only(size);
6555 }
6556 }
6557
6558 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6559 size_t lp = os::large_page_size();
6560 size_t ag = os::vm_allocation_granularity();
6561
6562 // sizes to test
6563 const size_t sizes[] = {
6564 lp, lp + ag, lp + lp / 2, lp * 2,
6565 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6566 lp * 10, lp * 10 + lp / 2
6567 };
6568 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6569
6570 // For each size/alignment combination, we test three scenarios:
6571 // 1) with req_addr == NULL
6572 // 2) with a non-null req_addr at which we expect to successfully allocate
6573 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6574 // expect the allocation to either fail or to ignore req_addr
6575
6576 // Pre-allocate two areas; they shall be as large as the largest allocation
6577 // and aligned to the largest alignment we will be testing.
6578 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6579 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6580 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6581 -1, 0);
6582 assert(mapping1 != MAP_FAILED, "should work");
6583
6584 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6585 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6586 -1, 0);
6587 assert(mapping2 != MAP_FAILED, "should work");
6588
6589 // Unmap the first mapping, but leave the second mapping intact: the first
6590 // mapping will serve as a value for a "good" req_addr (case 2). The second
6591 // mapping, still intact, as "bad" req_addr (case 3).
6592 ::munmap(mapping1, mapping_size);
6593
6594 // Case 1
6595 test_log("%s, req_addr NULL:", __FUNCTION__);
6596 test_log("size align result");
6597
6598 for (int i = 0; i < num_sizes; i++) {
6599 const size_t size = sizes[i];
6600 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6601 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6602 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6603 size, alignment, p, (p != NULL ? "" : "(failed)"));
6604 if (p != NULL) {
6605 assert(is_ptr_aligned(p, alignment), "must be");
6606 small_page_write(p, size);
6607 os::Linux::release_memory_special_huge_tlbfs(p, size);
6608 }
6609 }
6610 }
6611
6612 // Case 2
6613 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6614 test_log("size align req_addr result");
6615
6616 for (int i = 0; i < num_sizes; i++) {
6617 const size_t size = sizes[i];
6618 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6619 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6620 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6621 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6622 size, alignment, req_addr, p,
6623 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6624 if (p != NULL) {
6625 assert(p == req_addr, "must be");
6626 small_page_write(p, size);
6627 os::Linux::release_memory_special_huge_tlbfs(p, size);
6628 }
6629 }
6630 }
6631
6632 // Case 3
6633 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6634 test_log("size align req_addr result");
6635
6636 for (int i = 0; i < num_sizes; i++) {
6637 const size_t size = sizes[i];
6638 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6639 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6640 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6641 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6642 size, alignment, req_addr, p,
6643 ((p != NULL ? "" : "(failed)")));
6644 // as the area around req_addr contains already existing mappings, the API should always
6645 // return NULL (as per contract, it cannot return another address)
6646 assert(p == NULL, "must be");
6647 }
6648 }
6649
6650 ::munmap(mapping2, mapping_size);
6651
6652 }
6653
6654 static void test_reserve_memory_special_huge_tlbfs() {
6655 if (!UseHugeTLBFS) {
6656 return;
6657 }
6658
6659 test_reserve_memory_special_huge_tlbfs_only();
6660 test_reserve_memory_special_huge_tlbfs_mixed();
6661 }
6662
6663 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6664 if (!UseSHM) {
6665 return;
6666 }
6667
6668 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6669
6670 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6671
6672 if (addr != NULL) {
6673 assert(is_ptr_aligned(addr, alignment), "Check");
6674 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6675
6676 small_page_write(addr, size);
6677
6678 os::Linux::release_memory_special_shm(addr, size);
6679 }
6680 }
6681
6682 static void test_reserve_memory_special_shm() {
6683 size_t lp = os::large_page_size();
6684 size_t ag = os::vm_allocation_granularity();
6685
6686 for (size_t size = ag; size < lp * 3; size += ag) {
6687 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6688 test_reserve_memory_special_shm(size, alignment);
6689 }
6690 }
6691 }
6692
6693 static void test() {
6694 test_reserve_memory_special_huge_tlbfs();
6695 test_reserve_memory_special_shm();
6696 }
6697 };
6698
6699 void TestReserveMemorySpecial_test() {
6700 TestReserveMemorySpecial::test();
6701 }
6702
6703 #endif