1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/macros.hpp"
72 #include "utilities/vmError.hpp"
73
74 // put OS-includes here
75 # include <sys/types.h>
76 # include <sys/mman.h>
77 # include <sys/stat.h>
78 # include <sys/select.h>
79 # include <pthread.h>
80 # include <signal.h>
81 # include <errno.h>
82 # include <dlfcn.h>
83 # include <stdio.h>
84 # include <unistd.h>
85 # include <sys/resource.h>
86 # include <pthread.h>
87 # include <sys/stat.h>
88 # include <sys/time.h>
89 # include <sys/times.h>
90 # include <sys/utsname.h>
91 # include <sys/socket.h>
92 # include <sys/wait.h>
93 # include <pwd.h>
94 # include <poll.h>
95 # include <semaphore.h>
96 # include <fcntl.h>
97 # include <string.h>
98 # include <syscall.h>
99 # include <sys/sysinfo.h>
100 # include <gnu/libc-version.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
109
110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
111 // getrusage() is prepared to handle the associated failure.
112 #ifndef RUSAGE_THREAD
113 #define RUSAGE_THREAD (1) /* only the calling thread */
114 #endif
115
116 #define MAX_PATH (2 * K)
117
118 #define MAX_SECS 100000000
119
120 // for timer info max values which include all bits
121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123 #define LARGEPAGES_BIT (1 << 6)
124 ////////////////////////////////////////////////////////////////////////////////
125 // global variables
126 julong os::Linux::_physical_memory = 0;
127
128 address os::Linux::_initial_thread_stack_bottom = NULL;
129 uintptr_t os::Linux::_initial_thread_stack_size = 0;
130
131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 const int os::Linux::_vm_default_page_size = (8 * K);
138 bool os::Linux::_supports_fast_thread_cpu_time = false;
139 const char * os::Linux::_glibc_version = NULL;
140 const char * os::Linux::_libpthread_version = NULL;
141 pthread_condattr_t os::Linux::_condattr[1];
142
143 static jlong initial_time_count=0;
144
145 static int clock_tics_per_sec = 100;
146
147 // For diagnostics to print a message once. see run_periodic_checks
148 static sigset_t check_signal_done;
149 static bool check_signals = true;
150
151 // Signal number used to suspend/resume a thread
152
153 // do not use any signal number less than SIGSEGV, see 4355769
154 static int SR_signum = SIGUSR2;
155 sigset_t SR_sigset;
156
157 // Declarations
158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
159
160 // utility functions
161
162 static int SR_initialize();
163
164 julong os::available_memory() {
165 return Linux::available_memory();
166 }
167
168 julong os::Linux::available_memory() {
169 // values in struct sysinfo are "unsigned long"
170 struct sysinfo si;
171 sysinfo(&si);
172
173 return (julong)si.freeram * si.mem_unit;
174 }
175
176 julong os::physical_memory() {
177 return Linux::physical_memory();
178 }
179
180 // Return true if user is running as root.
181
182 bool os::have_special_privileges() {
183 static bool init = false;
184 static bool privileges = false;
185 if (!init) {
186 privileges = (getuid() != geteuid()) || (getgid() != getegid());
187 init = true;
188 }
189 return privileges;
190 }
191
192
193 #ifndef SYS_gettid
194 // i386: 224, ia64: 1105, amd64: 186, sparc 143
195 #ifdef __ia64__
196 #define SYS_gettid 1105
197 #else
198 #ifdef __i386__
199 #define SYS_gettid 224
200 #else
201 #ifdef __amd64__
202 #define SYS_gettid 186
203 #else
204 #ifdef __sparc__
205 #define SYS_gettid 143
206 #else
207 #error define gettid for the arch
208 #endif
209 #endif
210 #endif
211 #endif
212 #endif
213
214 // Cpu architecture string
215 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
216
217
218 // pid_t gettid()
219 //
220 // Returns the kernel thread id of the currently running thread. Kernel
221 // thread id is used to access /proc.
222 pid_t os::Linux::gettid() {
223 int rslt = syscall(SYS_gettid);
224 assert(rslt != -1, "must be."); // old linuxthreads implementation?
225 return (pid_t)rslt;
226 }
227
228 // Most versions of linux have a bug where the number of processors are
229 // determined by looking at the /proc file system. In a chroot environment,
230 // the system call returns 1. This causes the VM to act as if it is
231 // a single processor and elide locking (see is_MP() call).
232 static bool unsafe_chroot_detected = false;
233 static const char *unstable_chroot_error = "/proc file system not found.\n"
234 "Java may be unstable running multithreaded in a chroot "
235 "environment on Linux when /proc filesystem is not mounted.";
236
237 void os::Linux::initialize_system_info() {
238 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
239 if (processor_count() == 1) {
240 pid_t pid = os::Linux::gettid();
241 char fname[32];
242 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
243 FILE *fp = fopen(fname, "r");
244 if (fp == NULL) {
245 unsafe_chroot_detected = true;
246 } else {
247 fclose(fp);
248 }
249 }
250 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
251 assert(processor_count() > 0, "linux error");
252 }
253
254 void os::init_system_properties_values() {
255 // The next steps are taken in the product version:
256 //
257 // Obtain the JAVA_HOME value from the location of libjvm.so.
258 // This library should be located at:
259 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
260 //
261 // If "/jre/lib/" appears at the right place in the path, then we
262 // assume libjvm.so is installed in a JDK and we use this path.
263 //
264 // Otherwise exit with message: "Could not create the Java virtual machine."
265 //
266 // The following extra steps are taken in the debugging version:
267 //
268 // If "/jre/lib/" does NOT appear at the right place in the path
269 // instead of exit check for $JAVA_HOME environment variable.
270 //
271 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
272 // then we append a fake suffix "hotspot/libjvm.so" to this path so
273 // it looks like libjvm.so is installed there
274 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
275 //
276 // Otherwise exit.
277 //
278 // Important note: if the location of libjvm.so changes this
279 // code needs to be changed accordingly.
280
281 // See ld(1):
282 // The linker uses the following search paths to locate required
283 // shared libraries:
284 // 1: ...
285 // ...
286 // 7: The default directories, normally /lib and /usr/lib.
287 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
288 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
289 #else
290 #define DEFAULT_LIBPATH "/lib:/usr/lib"
291 #endif
292
293 // Base path of extensions installed on the system.
294 #define SYS_EXT_DIR "/usr/java/packages"
295 #define EXTENSIONS_DIR "/lib/ext"
296
297 // Buffer that fits several sprintfs.
298 // Note that the space for the colon and the trailing null are provided
299 // by the nulls included by the sizeof operator.
300 const size_t bufsize =
301 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
302 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
303 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
304
305 // sysclasspath, java_home, dll_dir
306 {
307 char *pslash;
308 os::jvm_path(buf, bufsize);
309
310 // Found the full path to libjvm.so.
311 // Now cut the path to <java_home>/jre if we can.
312 pslash = strrchr(buf, '/');
313 if (pslash != NULL) {
314 *pslash = '\0'; // Get rid of /libjvm.so.
315 }
316 pslash = strrchr(buf, '/');
317 if (pslash != NULL) {
318 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
319 }
320 Arguments::set_dll_dir(buf);
321
322 if (pslash != NULL) {
323 pslash = strrchr(buf, '/');
324 if (pslash != NULL) {
325 *pslash = '\0'; // Get rid of /<arch>.
326 pslash = strrchr(buf, '/');
327 if (pslash != NULL) {
328 *pslash = '\0'; // Get rid of /lib.
329 }
330 }
331 }
332 Arguments::set_java_home(buf);
333 set_boot_path('/', ':');
334 }
335
336 // Where to look for native libraries.
337 //
338 // Note: Due to a legacy implementation, most of the library path
339 // is set in the launcher. This was to accomodate linking restrictions
340 // on legacy Linux implementations (which are no longer supported).
341 // Eventually, all the library path setting will be done here.
342 //
343 // However, to prevent the proliferation of improperly built native
344 // libraries, the new path component /usr/java/packages is added here.
345 // Eventually, all the library path setting will be done here.
346 {
347 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
348 // should always exist (until the legacy problem cited above is
349 // addressed).
350 const char *v = ::getenv("LD_LIBRARY_PATH");
351 const char *v_colon = ":";
352 if (v == NULL) { v = ""; v_colon = ""; }
353 // That's +1 for the colon and +1 for the trailing '\0'.
354 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
355 strlen(v) + 1 +
356 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
357 mtInternal);
358 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
359 Arguments::set_library_path(ld_library_path);
360 FREE_C_HEAP_ARRAY(char, ld_library_path);
361 }
362
363 // Extensions directories.
364 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
365 Arguments::set_ext_dirs(buf);
366
367 FREE_C_HEAP_ARRAY(char, buf);
368
369 #undef DEFAULT_LIBPATH
370 #undef SYS_EXT_DIR
371 #undef EXTENSIONS_DIR
372 }
373
374 ////////////////////////////////////////////////////////////////////////////////
375 // breakpoint support
376
377 void os::breakpoint() {
378 BREAKPOINT;
379 }
380
381 extern "C" void breakpoint() {
382 // use debugger to set breakpoint here
383 }
384
385 ////////////////////////////////////////////////////////////////////////////////
386 // signal support
387
388 debug_only(static bool signal_sets_initialized = false);
389 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
390
391 bool os::Linux::is_sig_ignored(int sig) {
392 struct sigaction oact;
393 sigaction(sig, (struct sigaction*)NULL, &oact);
394 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
395 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
396 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
397 return true;
398 } else {
399 return false;
400 }
401 }
402
403 void os::Linux::signal_sets_init() {
404 // Should also have an assertion stating we are still single-threaded.
405 assert(!signal_sets_initialized, "Already initialized");
406 // Fill in signals that are necessarily unblocked for all threads in
407 // the VM. Currently, we unblock the following signals:
408 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
409 // by -Xrs (=ReduceSignalUsage));
410 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
411 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
412 // the dispositions or masks wrt these signals.
413 // Programs embedding the VM that want to use the above signals for their
414 // own purposes must, at this time, use the "-Xrs" option to prevent
415 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
416 // (See bug 4345157, and other related bugs).
417 // In reality, though, unblocking these signals is really a nop, since
418 // these signals are not blocked by default.
419 sigemptyset(&unblocked_sigs);
420 sigemptyset(&allowdebug_blocked_sigs);
421 sigaddset(&unblocked_sigs, SIGILL);
422 sigaddset(&unblocked_sigs, SIGSEGV);
423 sigaddset(&unblocked_sigs, SIGBUS);
424 sigaddset(&unblocked_sigs, SIGFPE);
425 #if defined(PPC64)
426 sigaddset(&unblocked_sigs, SIGTRAP);
427 #endif
428 sigaddset(&unblocked_sigs, SR_signum);
429
430 if (!ReduceSignalUsage) {
431 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
432 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
433 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
434 }
435 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
436 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
437 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
438 }
439 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
440 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
441 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
442 }
443 }
444 // Fill in signals that are blocked by all but the VM thread.
445 sigemptyset(&vm_sigs);
446 if (!ReduceSignalUsage) {
447 sigaddset(&vm_sigs, BREAK_SIGNAL);
448 }
449 debug_only(signal_sets_initialized = true);
450
451 }
452
453 // These are signals that are unblocked while a thread is running Java.
454 // (For some reason, they get blocked by default.)
455 sigset_t* os::Linux::unblocked_signals() {
456 assert(signal_sets_initialized, "Not initialized");
457 return &unblocked_sigs;
458 }
459
460 // These are the signals that are blocked while a (non-VM) thread is
461 // running Java. Only the VM thread handles these signals.
462 sigset_t* os::Linux::vm_signals() {
463 assert(signal_sets_initialized, "Not initialized");
464 return &vm_sigs;
465 }
466
467 // These are signals that are blocked during cond_wait to allow debugger in
468 sigset_t* os::Linux::allowdebug_blocked_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &allowdebug_blocked_sigs;
471 }
472
473 void os::Linux::hotspot_sigmask(Thread* thread) {
474
475 //Save caller's signal mask before setting VM signal mask
476 sigset_t caller_sigmask;
477 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
478
479 OSThread* osthread = thread->osthread();
480 osthread->set_caller_sigmask(caller_sigmask);
481
482 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
483
484 if (!ReduceSignalUsage) {
485 if (thread->is_VM_thread()) {
486 // Only the VM thread handles BREAK_SIGNAL ...
487 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
488 } else {
489 // ... all other threads block BREAK_SIGNAL
490 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
491 }
492 }
493 }
494
495 //////////////////////////////////////////////////////////////////////////////
496 // detecting pthread library
497
498 void os::Linux::libpthread_init() {
499 // Save glibc and pthread version strings.
500 #if !defined(_CS_GNU_LIBC_VERSION) || \
501 !defined(_CS_GNU_LIBPTHREAD_VERSION)
502 #error "glibc too old (< 2.3.2)"
503 #endif
504
505 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
506 assert(n > 0, "cannot retrieve glibc version");
507 char *str = (char *)malloc(n, mtInternal);
508 confstr(_CS_GNU_LIBC_VERSION, str, n);
509 os::Linux::set_glibc_version(str);
510
511 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
512 assert(n > 0, "cannot retrieve pthread version");
513 str = (char *)malloc(n, mtInternal);
514 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
515 os::Linux::set_libpthread_version(str);
516 }
517
518 /////////////////////////////////////////////////////////////////////////////
519 // thread stack expansion
520
521 // os::Linux::manually_expand_stack() takes care of expanding the thread
522 // stack. Note that this is normally not needed: pthread stacks allocate
523 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
524 // committed. Therefore it is not necessary to expand the stack manually.
525 //
526 // Manually expanding the stack was historically needed on LinuxThreads
527 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
528 // it is kept to deal with very rare corner cases:
529 //
530 // For one, user may run the VM on an own implementation of threads
531 // whose stacks are - like the old LinuxThreads - implemented using
532 // mmap(MAP_GROWSDOWN).
533 //
534 // Also, this coding may be needed if the VM is running on the primordial
535 // thread. Normally we avoid running on the primordial thread; however,
536 // user may still invoke the VM on the primordial thread.
537 //
538 // The following historical comment describes the details about running
539 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
540
541
542 // Force Linux kernel to expand current thread stack. If "bottom" is close
543 // to the stack guard, caller should block all signals.
544 //
545 // MAP_GROWSDOWN:
546 // A special mmap() flag that is used to implement thread stacks. It tells
547 // kernel that the memory region should extend downwards when needed. This
548 // allows early versions of LinuxThreads to only mmap the first few pages
549 // when creating a new thread. Linux kernel will automatically expand thread
550 // stack as needed (on page faults).
551 //
552 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
553 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
554 // region, it's hard to tell if the fault is due to a legitimate stack
555 // access or because of reading/writing non-exist memory (e.g. buffer
556 // overrun). As a rule, if the fault happens below current stack pointer,
557 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
558 // application (see Linux kernel fault.c).
559 //
560 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
561 // stack overflow detection.
562 //
563 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
564 // not use MAP_GROWSDOWN.
565 //
566 // To get around the problem and allow stack banging on Linux, we need to
567 // manually expand thread stack after receiving the SIGSEGV.
568 //
569 // There are two ways to expand thread stack to address "bottom", we used
570 // both of them in JVM before 1.5:
571 // 1. adjust stack pointer first so that it is below "bottom", and then
572 // touch "bottom"
573 // 2. mmap() the page in question
574 //
575 // Now alternate signal stack is gone, it's harder to use 2. For instance,
576 // if current sp is already near the lower end of page 101, and we need to
577 // call mmap() to map page 100, it is possible that part of the mmap() frame
578 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
579 // That will destroy the mmap() frame and cause VM to crash.
580 //
581 // The following code works by adjusting sp first, then accessing the "bottom"
582 // page to force a page fault. Linux kernel will then automatically expand the
583 // stack mapping.
584 //
585 // _expand_stack_to() assumes its frame size is less than page size, which
586 // should always be true if the function is not inlined.
587
588 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
589 #define NOINLINE
590 #else
591 #define NOINLINE __attribute__ ((noinline))
592 #endif
593
594 static void _expand_stack_to(address bottom) NOINLINE;
595
596 static void _expand_stack_to(address bottom) {
597 address sp;
598 size_t size;
599 volatile char *p;
600
601 // Adjust bottom to point to the largest address within the same page, it
602 // gives us a one-page buffer if alloca() allocates slightly more memory.
603 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
604 bottom += os::Linux::page_size() - 1;
605
606 // sp might be slightly above current stack pointer; if that's the case, we
607 // will alloca() a little more space than necessary, which is OK. Don't use
608 // os::current_stack_pointer(), as its result can be slightly below current
609 // stack pointer, causing us to not alloca enough to reach "bottom".
610 sp = (address)&sp;
611
612 if (sp > bottom) {
613 size = sp - bottom;
614 p = (volatile char *)alloca(size);
615 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
616 p[0] = '\0';
617 }
618 }
619
620 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
621 assert(t!=NULL, "just checking");
622 assert(t->osthread()->expanding_stack(), "expand should be set");
623 assert(t->stack_base() != NULL, "stack_base was not initialized");
624
625 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
626 sigset_t mask_all, old_sigset;
627 sigfillset(&mask_all);
628 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
629 _expand_stack_to(addr);
630 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
631 return true;
632 }
633 return false;
634 }
635
636 //////////////////////////////////////////////////////////////////////////////
637 // create new thread
638
639 // Thread start routine for all newly created threads
640 static void *java_start(Thread *thread) {
641 // Try to randomize the cache line index of hot stack frames.
642 // This helps when threads of the same stack traces evict each other's
643 // cache lines. The threads can be either from the same JVM instance, or
644 // from different JVM instances. The benefit is especially true for
645 // processors with hyperthreading technology.
646 static int counter = 0;
647 int pid = os::current_process_id();
648 alloca(((pid ^ counter++) & 7) * 128);
649
650 ThreadLocalStorage::set_thread(thread);
651
652 OSThread* osthread = thread->osthread();
653 Monitor* sync = osthread->startThread_lock();
654
655 // thread_id is kernel thread id (similar to Solaris LWP id)
656 osthread->set_thread_id(os::Linux::gettid());
657
658 if (UseNUMA) {
659 int lgrp_id = os::numa_get_group_id();
660 if (lgrp_id != -1) {
661 thread->set_lgrp_id(lgrp_id);
662 }
663 }
664 // initialize signal mask for this thread
665 os::Linux::hotspot_sigmask(thread);
666
667 // initialize floating point control register
668 os::Linux::init_thread_fpu_state();
669
670 // handshaking with parent thread
671 {
672 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
673
674 // notify parent thread
675 osthread->set_state(INITIALIZED);
676 sync->notify_all();
677
678 // wait until os::start_thread()
679 while (osthread->get_state() == INITIALIZED) {
680 sync->wait(Mutex::_no_safepoint_check_flag);
681 }
682 }
683
684 // call one more level start routine
685 thread->run();
686
687 return 0;
688 }
689
690 bool os::create_thread(Thread* thread, ThreadType thr_type,
691 size_t stack_size) {
692 assert(thread->osthread() == NULL, "caller responsible");
693
694 // Allocate the OSThread object
695 OSThread* osthread = new OSThread(NULL, NULL);
696 if (osthread == NULL) {
697 return false;
698 }
699
700 // set the correct thread state
701 osthread->set_thread_type(thr_type);
702
703 // Initial state is ALLOCATED but not INITIALIZED
704 osthread->set_state(ALLOCATED);
705
706 thread->set_osthread(osthread);
707
708 // init thread attributes
709 pthread_attr_t attr;
710 pthread_attr_init(&attr);
711 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
712
713 // stack size
714 if (os::Linux::supports_variable_stack_size()) {
715 // calculate stack size if it's not specified by caller
716 if (stack_size == 0) {
717 stack_size = os::Linux::default_stack_size(thr_type);
718
719 switch (thr_type) {
720 case os::java_thread:
721 // Java threads use ThreadStackSize which default value can be
722 // changed with the flag -Xss
723 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
724 stack_size = JavaThread::stack_size_at_create();
725 break;
726 case os::compiler_thread:
727 if (CompilerThreadStackSize > 0) {
728 stack_size = (size_t)(CompilerThreadStackSize * K);
729 break;
730 } // else fall through:
731 // use VMThreadStackSize if CompilerThreadStackSize is not defined
732 case os::vm_thread:
733 case os::pgc_thread:
734 case os::cgc_thread:
735 case os::watcher_thread:
736 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
737 break;
738 }
739 }
740
741 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
742 pthread_attr_setstacksize(&attr, stack_size);
743 } else {
744 // let pthread_create() pick the default value.
745 }
746
747 // glibc guard page
748 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
749
750 ThreadState state;
751
752 {
753 pthread_t tid;
754 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
755
756 pthread_attr_destroy(&attr);
757
758 if (ret != 0) {
759 if (PrintMiscellaneous && (Verbose || WizardMode)) {
760 perror("pthread_create()");
761 }
762 // Need to clean up stuff we've allocated so far
763 thread->set_osthread(NULL);
764 delete osthread;
765 return false;
766 }
767
768 // Store pthread info into the OSThread
769 osthread->set_pthread_id(tid);
770
771 // Wait until child thread is either initialized or aborted
772 {
773 Monitor* sync_with_child = osthread->startThread_lock();
774 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
775 while ((state = osthread->get_state()) == ALLOCATED) {
776 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
777 }
778 }
779 }
780
781 // Aborted due to thread limit being reached
782 if (state == ZOMBIE) {
783 thread->set_osthread(NULL);
784 delete osthread;
785 return false;
786 }
787
788 // The thread is returned suspended (in state INITIALIZED),
789 // and is started higher up in the call chain
790 assert(state == INITIALIZED, "race condition");
791 return true;
792 }
793
794 /////////////////////////////////////////////////////////////////////////////
795 // attach existing thread
796
797 // bootstrap the main thread
798 bool os::create_main_thread(JavaThread* thread) {
799 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
800 return create_attached_thread(thread);
801 }
802
803 bool os::create_attached_thread(JavaThread* thread) {
804 #ifdef ASSERT
805 thread->verify_not_published();
806 #endif
807
808 // Allocate the OSThread object
809 OSThread* osthread = new OSThread(NULL, NULL);
810
811 if (osthread == NULL) {
812 return false;
813 }
814
815 // Store pthread info into the OSThread
816 osthread->set_thread_id(os::Linux::gettid());
817 osthread->set_pthread_id(::pthread_self());
818
819 // initialize floating point control register
820 os::Linux::init_thread_fpu_state();
821
822 // Initial thread state is RUNNABLE
823 osthread->set_state(RUNNABLE);
824
825 thread->set_osthread(osthread);
826
827 if (UseNUMA) {
828 int lgrp_id = os::numa_get_group_id();
829 if (lgrp_id != -1) {
830 thread->set_lgrp_id(lgrp_id);
831 }
832 }
833
834 if (os::Linux::is_initial_thread()) {
835 // If current thread is initial thread, its stack is mapped on demand,
836 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
837 // the entire stack region to avoid SEGV in stack banging.
838 // It is also useful to get around the heap-stack-gap problem on SuSE
839 // kernel (see 4821821 for details). We first expand stack to the top
840 // of yellow zone, then enable stack yellow zone (order is significant,
841 // enabling yellow zone first will crash JVM on SuSE Linux), so there
842 // is no gap between the last two virtual memory regions.
843
844 JavaThread *jt = (JavaThread *)thread;
845 address addr = jt->stack_yellow_zone_base();
846 assert(addr != NULL, "initialization problem?");
847 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
848
849 osthread->set_expanding_stack();
850 os::Linux::manually_expand_stack(jt, addr);
851 osthread->clear_expanding_stack();
852 }
853
854 // initialize signal mask for this thread
855 // and save the caller's signal mask
856 os::Linux::hotspot_sigmask(thread);
857
858 return true;
859 }
860
861 void os::pd_start_thread(Thread* thread) {
862 OSThread * osthread = thread->osthread();
863 assert(osthread->get_state() != INITIALIZED, "just checking");
864 Monitor* sync_with_child = osthread->startThread_lock();
865 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
866 sync_with_child->notify();
867 }
868
869 // Free Linux resources related to the OSThread
870 void os::free_thread(OSThread* osthread) {
871 assert(osthread != NULL, "osthread not set");
872
873 if (Thread::current()->osthread() == osthread) {
874 // Restore caller's signal mask
875 sigset_t sigmask = osthread->caller_sigmask();
876 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
877 }
878
879 delete osthread;
880 }
881
882 //////////////////////////////////////////////////////////////////////////////
883 // thread local storage
884
885 // Restore the thread pointer if the destructor is called. This is in case
886 // someone from JNI code sets up a destructor with pthread_key_create to run
887 // detachCurrentThread on thread death. Unless we restore the thread pointer we
888 // will hang or crash. When detachCurrentThread is called the key will be set
889 // to null and we will not be called again. If detachCurrentThread is never
890 // called we could loop forever depending on the pthread implementation.
891 static void restore_thread_pointer(void* p) {
892 Thread* thread = (Thread*) p;
893 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
894 }
895
896 int os::allocate_thread_local_storage() {
897 pthread_key_t key;
898 int rslt = pthread_key_create(&key, restore_thread_pointer);
899 assert(rslt == 0, "cannot allocate thread local storage");
900 return (int)key;
901 }
902
903 // Note: This is currently not used by VM, as we don't destroy TLS key
904 // on VM exit.
905 void os::free_thread_local_storage(int index) {
906 int rslt = pthread_key_delete((pthread_key_t)index);
907 assert(rslt == 0, "invalid index");
908 }
909
910 void os::thread_local_storage_at_put(int index, void* value) {
911 int rslt = pthread_setspecific((pthread_key_t)index, value);
912 assert(rslt == 0, "pthread_setspecific failed");
913 }
914
915 extern "C" Thread* get_thread() {
916 return ThreadLocalStorage::thread();
917 }
918
919 //////////////////////////////////////////////////////////////////////////////
920 // initial thread
921
922 // Check if current thread is the initial thread, similar to Solaris thr_main.
923 bool os::Linux::is_initial_thread(void) {
924 char dummy;
925 // If called before init complete, thread stack bottom will be null.
926 // Can be called if fatal error occurs before initialization.
927 if (initial_thread_stack_bottom() == NULL) return false;
928 assert(initial_thread_stack_bottom() != NULL &&
929 initial_thread_stack_size() != 0,
930 "os::init did not locate initial thread's stack region");
931 if ((address)&dummy >= initial_thread_stack_bottom() &&
932 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
933 return true;
934 } else {
935 return false;
936 }
937 }
938
939 // Find the virtual memory area that contains addr
940 static bool find_vma(address addr, address* vma_low, address* vma_high) {
941 FILE *fp = fopen("/proc/self/maps", "r");
942 if (fp) {
943 address low, high;
944 while (!feof(fp)) {
945 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
946 if (low <= addr && addr < high) {
947 if (vma_low) *vma_low = low;
948 if (vma_high) *vma_high = high;
949 fclose(fp);
950 return true;
951 }
952 }
953 for (;;) {
954 int ch = fgetc(fp);
955 if (ch == EOF || ch == (int)'\n') break;
956 }
957 }
958 fclose(fp);
959 }
960 return false;
961 }
962
963 // Locate initial thread stack. This special handling of initial thread stack
964 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
965 // bogus value for initial thread.
966 void os::Linux::capture_initial_stack(size_t max_size) {
967 // stack size is the easy part, get it from RLIMIT_STACK
968 size_t stack_size;
969 struct rlimit rlim;
970 getrlimit(RLIMIT_STACK, &rlim);
971 stack_size = rlim.rlim_cur;
972
973 // 6308388: a bug in ld.so will relocate its own .data section to the
974 // lower end of primordial stack; reduce ulimit -s value a little bit
975 // so we won't install guard page on ld.so's data section.
976 stack_size -= 2 * page_size();
977
978 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
979 // 7.1, in both cases we will get 2G in return value.
980 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
981 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
982 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
983 // in case other parts in glibc still assumes 2M max stack size.
984 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
985 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
986 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
987 stack_size = 2 * K * K IA64_ONLY(*2);
988 }
989 // Try to figure out where the stack base (top) is. This is harder.
990 //
991 // When an application is started, glibc saves the initial stack pointer in
992 // a global variable "__libc_stack_end", which is then used by system
993 // libraries. __libc_stack_end should be pretty close to stack top. The
994 // variable is available since the very early days. However, because it is
995 // a private interface, it could disappear in the future.
996 //
997 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
998 // to __libc_stack_end, it is very close to stack top, but isn't the real
999 // stack top. Note that /proc may not exist if VM is running as a chroot
1000 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1001 // /proc/<pid>/stat could change in the future (though unlikely).
1002 //
1003 // We try __libc_stack_end first. If that doesn't work, look for
1004 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1005 // as a hint, which should work well in most cases.
1006
1007 uintptr_t stack_start;
1008
1009 // try __libc_stack_end first
1010 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1011 if (p && *p) {
1012 stack_start = *p;
1013 } else {
1014 // see if we can get the start_stack field from /proc/self/stat
1015 FILE *fp;
1016 int pid;
1017 char state;
1018 int ppid;
1019 int pgrp;
1020 int session;
1021 int nr;
1022 int tpgrp;
1023 unsigned long flags;
1024 unsigned long minflt;
1025 unsigned long cminflt;
1026 unsigned long majflt;
1027 unsigned long cmajflt;
1028 unsigned long utime;
1029 unsigned long stime;
1030 long cutime;
1031 long cstime;
1032 long prio;
1033 long nice;
1034 long junk;
1035 long it_real;
1036 uintptr_t start;
1037 uintptr_t vsize;
1038 intptr_t rss;
1039 uintptr_t rsslim;
1040 uintptr_t scodes;
1041 uintptr_t ecode;
1042 int i;
1043
1044 // Figure what the primordial thread stack base is. Code is inspired
1045 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1046 // followed by command name surrounded by parentheses, state, etc.
1047 char stat[2048];
1048 int statlen;
1049
1050 fp = fopen("/proc/self/stat", "r");
1051 if (fp) {
1052 statlen = fread(stat, 1, 2047, fp);
1053 stat[statlen] = '\0';
1054 fclose(fp);
1055
1056 // Skip pid and the command string. Note that we could be dealing with
1057 // weird command names, e.g. user could decide to rename java launcher
1058 // to "java 1.4.2 :)", then the stat file would look like
1059 // 1234 (java 1.4.2 :)) R ... ...
1060 // We don't really need to know the command string, just find the last
1061 // occurrence of ")" and then start parsing from there. See bug 4726580.
1062 char * s = strrchr(stat, ')');
1063
1064 i = 0;
1065 if (s) {
1066 // Skip blank chars
1067 do { s++; } while (s && isspace(*s));
1068
1069 #define _UFM UINTX_FORMAT
1070 #define _DFM INTX_FORMAT
1071
1072 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1073 // 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
1074 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,
1075 &state, // 3 %c
1076 &ppid, // 4 %d
1077 &pgrp, // 5 %d
1078 &session, // 6 %d
1079 &nr, // 7 %d
1080 &tpgrp, // 8 %d
1081 &flags, // 9 %lu
1082 &minflt, // 10 %lu
1083 &cminflt, // 11 %lu
1084 &majflt, // 12 %lu
1085 &cmajflt, // 13 %lu
1086 &utime, // 14 %lu
1087 &stime, // 15 %lu
1088 &cutime, // 16 %ld
1089 &cstime, // 17 %ld
1090 &prio, // 18 %ld
1091 &nice, // 19 %ld
1092 &junk, // 20 %ld
1093 &it_real, // 21 %ld
1094 &start, // 22 UINTX_FORMAT
1095 &vsize, // 23 UINTX_FORMAT
1096 &rss, // 24 INTX_FORMAT
1097 &rsslim, // 25 UINTX_FORMAT
1098 &scodes, // 26 UINTX_FORMAT
1099 &ecode, // 27 UINTX_FORMAT
1100 &stack_start); // 28 UINTX_FORMAT
1101 }
1102
1103 #undef _UFM
1104 #undef _DFM
1105
1106 if (i != 28 - 2) {
1107 assert(false, "Bad conversion from /proc/self/stat");
1108 // product mode - assume we are the initial thread, good luck in the
1109 // embedded case.
1110 warning("Can't detect initial thread stack location - bad conversion");
1111 stack_start = (uintptr_t) &rlim;
1112 }
1113 } else {
1114 // For some reason we can't open /proc/self/stat (for example, running on
1115 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1116 // most cases, so don't abort:
1117 warning("Can't detect initial thread stack location - no /proc/self/stat");
1118 stack_start = (uintptr_t) &rlim;
1119 }
1120 }
1121
1122 // Now we have a pointer (stack_start) very close to the stack top, the
1123 // next thing to do is to figure out the exact location of stack top. We
1124 // can find out the virtual memory area that contains stack_start by
1125 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1126 // and its upper limit is the real stack top. (again, this would fail if
1127 // running inside chroot, because /proc may not exist.)
1128
1129 uintptr_t stack_top;
1130 address low, high;
1131 if (find_vma((address)stack_start, &low, &high)) {
1132 // success, "high" is the true stack top. (ignore "low", because initial
1133 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1134 stack_top = (uintptr_t)high;
1135 } else {
1136 // failed, likely because /proc/self/maps does not exist
1137 warning("Can't detect initial thread stack location - find_vma failed");
1138 // best effort: stack_start is normally within a few pages below the real
1139 // stack top, use it as stack top, and reduce stack size so we won't put
1140 // guard page outside stack.
1141 stack_top = stack_start;
1142 stack_size -= 16 * page_size();
1143 }
1144
1145 // stack_top could be partially down the page so align it
1146 stack_top = align_size_up(stack_top, page_size());
1147
1148 if (max_size && stack_size > max_size) {
1149 _initial_thread_stack_size = max_size;
1150 } else {
1151 _initial_thread_stack_size = stack_size;
1152 }
1153
1154 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1155 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156 }
1157
1158 ////////////////////////////////////////////////////////////////////////////////
1159 // time support
1160
1161 // Time since start-up in seconds to a fine granularity.
1162 // Used by VMSelfDestructTimer and the MemProfiler.
1163 double os::elapsedTime() {
1164
1165 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1166 }
1167
1168 jlong os::elapsed_counter() {
1169 return javaTimeNanos() - initial_time_count;
1170 }
1171
1172 jlong os::elapsed_frequency() {
1173 return NANOSECS_PER_SEC; // nanosecond resolution
1174 }
1175
1176 bool os::supports_vtime() { return true; }
1177 bool os::enable_vtime() { return false; }
1178 bool os::vtime_enabled() { return false; }
1179
1180 double os::elapsedVTime() {
1181 struct rusage usage;
1182 int retval = getrusage(RUSAGE_THREAD, &usage);
1183 if (retval == 0) {
1184 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);
1185 } else {
1186 // better than nothing, but not much
1187 return elapsedTime();
1188 }
1189 }
1190
1191 jlong os::javaTimeMillis() {
1192 timeval time;
1193 int status = gettimeofday(&time, NULL);
1194 assert(status != -1, "linux error");
1195 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1196 }
1197
1198 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1199 timeval time;
1200 int status = gettimeofday(&time, NULL);
1201 assert(status != -1, "linux error");
1202 seconds = jlong(time.tv_sec);
1203 nanos = jlong(time.tv_usec) * 1000;
1204 }
1205
1206
1207 #ifndef CLOCK_MONOTONIC
1208 #define CLOCK_MONOTONIC (1)
1209 #endif
1210
1211 void os::Linux::clock_init() {
1212 // we do dlopen's in this particular order due to bug in linux
1213 // dynamical loader (see 6348968) leading to crash on exit
1214 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1215 if (handle == NULL) {
1216 handle = dlopen("librt.so", RTLD_LAZY);
1217 }
1218
1219 if (handle) {
1220 int (*clock_getres_func)(clockid_t, struct timespec*) =
1221 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1222 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1223 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1224 if (clock_getres_func && clock_gettime_func) {
1225 // See if monotonic clock is supported by the kernel. Note that some
1226 // early implementations simply return kernel jiffies (updated every
1227 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1228 // for nano time (though the monotonic property is still nice to have).
1229 // It's fixed in newer kernels, however clock_getres() still returns
1230 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1231 // resolution for now. Hopefully as people move to new kernels, this
1232 // won't be a problem.
1233 struct timespec res;
1234 struct timespec tp;
1235 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1236 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1237 // yes, monotonic clock is supported
1238 _clock_gettime = clock_gettime_func;
1239 return;
1240 } else {
1241 // close librt if there is no monotonic clock
1242 dlclose(handle);
1243 }
1244 }
1245 }
1246 warning("No monotonic clock was available - timed services may " \
1247 "be adversely affected if the time-of-day clock changes");
1248 }
1249
1250 #ifndef SYS_clock_getres
1251 #if defined(IA32) || defined(AMD64)
1252 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1253 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1254 #else
1255 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1256 #define sys_clock_getres(x,y) -1
1257 #endif
1258 #else
1259 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1260 #endif
1261
1262 void os::Linux::fast_thread_clock_init() {
1263 if (!UseLinuxPosixThreadCPUClocks) {
1264 return;
1265 }
1266 clockid_t clockid;
1267 struct timespec tp;
1268 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1269 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1270
1271 // Switch to using fast clocks for thread cpu time if
1272 // the sys_clock_getres() returns 0 error code.
1273 // Note, that some kernels may support the current thread
1274 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1275 // returned by the pthread_getcpuclockid().
1276 // If the fast Posix clocks are supported then the sys_clock_getres()
1277 // must return at least tp.tv_sec == 0 which means a resolution
1278 // better than 1 sec. This is extra check for reliability.
1279
1280 if (pthread_getcpuclockid_func &&
1281 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1282 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1283 _supports_fast_thread_cpu_time = true;
1284 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1285 }
1286 }
1287
1288 jlong os::javaTimeNanos() {
1289 if (os::supports_monotonic_clock()) {
1290 struct timespec tp;
1291 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1292 assert(status == 0, "gettime error");
1293 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1294 return result;
1295 } else {
1296 timeval time;
1297 int status = gettimeofday(&time, NULL);
1298 assert(status != -1, "linux error");
1299 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1300 return 1000 * usecs;
1301 }
1302 }
1303
1304 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1305 if (os::supports_monotonic_clock()) {
1306 info_ptr->max_value = ALL_64_BITS;
1307
1308 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1309 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1310 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1311 } else {
1312 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1313 info_ptr->max_value = ALL_64_BITS;
1314
1315 // gettimeofday is a real time clock so it skips
1316 info_ptr->may_skip_backward = true;
1317 info_ptr->may_skip_forward = true;
1318 }
1319
1320 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1321 }
1322
1323 // Return the real, user, and system times in seconds from an
1324 // arbitrary fixed point in the past.
1325 bool os::getTimesSecs(double* process_real_time,
1326 double* process_user_time,
1327 double* process_system_time) {
1328 struct tms ticks;
1329 clock_t real_ticks = times(&ticks);
1330
1331 if (real_ticks == (clock_t) (-1)) {
1332 return false;
1333 } else {
1334 double ticks_per_second = (double) clock_tics_per_sec;
1335 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1336 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1337 *process_real_time = ((double) real_ticks) / ticks_per_second;
1338
1339 return true;
1340 }
1341 }
1342
1343
1344 char * os::local_time_string(char *buf, size_t buflen) {
1345 struct tm t;
1346 time_t long_time;
1347 time(&long_time);
1348 localtime_r(&long_time, &t);
1349 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1350 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1351 t.tm_hour, t.tm_min, t.tm_sec);
1352 return buf;
1353 }
1354
1355 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1356 return localtime_r(clock, res);
1357 }
1358
1359 ////////////////////////////////////////////////////////////////////////////////
1360 // runtime exit support
1361
1362 // Note: os::shutdown() might be called very early during initialization, or
1363 // called from signal handler. Before adding something to os::shutdown(), make
1364 // sure it is async-safe and can handle partially initialized VM.
1365 void os::shutdown() {
1366
1367 // allow PerfMemory to attempt cleanup of any persistent resources
1368 perfMemory_exit();
1369
1370 // needs to remove object in file system
1371 AttachListener::abort();
1372
1373 // flush buffered output, finish log files
1374 ostream_abort();
1375
1376 // Check for abort hook
1377 abort_hook_t abort_hook = Arguments::abort_hook();
1378 if (abort_hook != NULL) {
1379 abort_hook();
1380 }
1381
1382 }
1383
1384 // Note: os::abort() might be called very early during initialization, or
1385 // called from signal handler. Before adding something to os::abort(), make
1386 // sure it is async-safe and can handle partially initialized VM.
1387 void os::abort(bool dump_core, void* siginfo, void* context) {
1388 os::shutdown();
1389 if (dump_core) {
1390 #ifndef PRODUCT
1391 fdStream out(defaultStream::output_fd());
1392 out.print_raw("Current thread is ");
1393 char buf[16];
1394 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1395 out.print_raw_cr(buf);
1396 out.print_raw_cr("Dumping core ...");
1397 #endif
1398 ::abort(); // dump core
1399 }
1400
1401 ::exit(1);
1402 }
1403
1404 // Die immediately, no exit hook, no abort hook, no cleanup.
1405 void os::die() {
1406 ::abort();
1407 }
1408
1409
1410 // This method is a copy of JDK's sysGetLastErrorString
1411 // from src/solaris/hpi/src/system_md.c
1412
1413 size_t os::lasterror(char *buf, size_t len) {
1414 if (errno == 0) return 0;
1415
1416 const char *s = ::strerror(errno);
1417 size_t n = ::strlen(s);
1418 if (n >= len) {
1419 n = len - 1;
1420 }
1421 ::strncpy(buf, s, n);
1422 buf[n] = '\0';
1423 return n;
1424 }
1425
1426 intx os::current_thread_id() { return (intx)pthread_self(); }
1427 int os::current_process_id() {
1428 return ::getpid();
1429 }
1430
1431 // DLL functions
1432
1433 const char* os::dll_file_extension() { return ".so"; }
1434
1435 // This must be hard coded because it's the system's temporary
1436 // directory not the java application's temp directory, ala java.io.tmpdir.
1437 const char* os::get_temp_directory() { return "/tmp"; }
1438
1439 static bool file_exists(const char* filename) {
1440 struct stat statbuf;
1441 if (filename == NULL || strlen(filename) == 0) {
1442 return false;
1443 }
1444 return os::stat(filename, &statbuf) == 0;
1445 }
1446
1447 bool os::dll_build_name(char* buffer, size_t buflen,
1448 const char* pname, const char* fname) {
1449 bool retval = false;
1450 // Copied from libhpi
1451 const size_t pnamelen = pname ? strlen(pname) : 0;
1452
1453 // Return error on buffer overflow.
1454 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1455 return retval;
1456 }
1457
1458 if (pnamelen == 0) {
1459 snprintf(buffer, buflen, "lib%s.so", fname);
1460 retval = true;
1461 } else if (strchr(pname, *os::path_separator()) != NULL) {
1462 int n;
1463 char** pelements = split_path(pname, &n);
1464 if (pelements == NULL) {
1465 return false;
1466 }
1467 for (int i = 0; i < n; i++) {
1468 // Really shouldn't be NULL, but check can't hurt
1469 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1470 continue; // skip the empty path values
1471 }
1472 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1473 if (file_exists(buffer)) {
1474 retval = true;
1475 break;
1476 }
1477 }
1478 // release the storage
1479 for (int i = 0; i < n; i++) {
1480 if (pelements[i] != NULL) {
1481 FREE_C_HEAP_ARRAY(char, pelements[i]);
1482 }
1483 }
1484 if (pelements != NULL) {
1485 FREE_C_HEAP_ARRAY(char*, pelements);
1486 }
1487 } else {
1488 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1489 retval = true;
1490 }
1491 return retval;
1492 }
1493
1494 // check if addr is inside libjvm.so
1495 bool os::address_is_in_vm(address addr) {
1496 static address libjvm_base_addr;
1497 Dl_info dlinfo;
1498
1499 if (libjvm_base_addr == NULL) {
1500 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1501 libjvm_base_addr = (address)dlinfo.dli_fbase;
1502 }
1503 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1504 }
1505
1506 if (dladdr((void *)addr, &dlinfo) != 0) {
1507 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1508 }
1509
1510 return false;
1511 }
1512
1513 bool os::dll_address_to_function_name(address addr, char *buf,
1514 int buflen, int *offset) {
1515 // buf is not optional, but offset is optional
1516 assert(buf != NULL, "sanity check");
1517
1518 Dl_info dlinfo;
1519
1520 if (dladdr((void*)addr, &dlinfo) != 0) {
1521 // see if we have a matching symbol
1522 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1523 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1524 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1525 }
1526 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1527 return true;
1528 }
1529 // no matching symbol so try for just file info
1530 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1531 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1532 buf, buflen, offset, dlinfo.dli_fname)) {
1533 return true;
1534 }
1535 }
1536 }
1537
1538 buf[0] = '\0';
1539 if (offset != NULL) *offset = -1;
1540 return false;
1541 }
1542
1543 struct _address_to_library_name {
1544 address addr; // input : memory address
1545 size_t buflen; // size of fname
1546 char* fname; // output: library name
1547 address base; // library base addr
1548 };
1549
1550 static int address_to_library_name_callback(struct dl_phdr_info *info,
1551 size_t size, void *data) {
1552 int i;
1553 bool found = false;
1554 address libbase = NULL;
1555 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1556
1557 // iterate through all loadable segments
1558 for (i = 0; i < info->dlpi_phnum; i++) {
1559 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1560 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1561 // base address of a library is the lowest address of its loaded
1562 // segments.
1563 if (libbase == NULL || libbase > segbase) {
1564 libbase = segbase;
1565 }
1566 // see if 'addr' is within current segment
1567 if (segbase <= d->addr &&
1568 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1569 found = true;
1570 }
1571 }
1572 }
1573
1574 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1575 // so dll_address_to_library_name() can fall through to use dladdr() which
1576 // can figure out executable name from argv[0].
1577 if (found && info->dlpi_name && info->dlpi_name[0]) {
1578 d->base = libbase;
1579 if (d->fname) {
1580 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1581 }
1582 return 1;
1583 }
1584 return 0;
1585 }
1586
1587 bool os::dll_address_to_library_name(address addr, char* buf,
1588 int buflen, int* offset) {
1589 // buf is not optional, but offset is optional
1590 assert(buf != NULL, "sanity check");
1591
1592 Dl_info dlinfo;
1593 struct _address_to_library_name data;
1594
1595 // There is a bug in old glibc dladdr() implementation that it could resolve
1596 // to wrong library name if the .so file has a base address != NULL. Here
1597 // we iterate through the program headers of all loaded libraries to find
1598 // out which library 'addr' really belongs to. This workaround can be
1599 // removed once the minimum requirement for glibc is moved to 2.3.x.
1600 data.addr = addr;
1601 data.fname = buf;
1602 data.buflen = buflen;
1603 data.base = NULL;
1604 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1605
1606 if (rslt) {
1607 // buf already contains library name
1608 if (offset) *offset = addr - data.base;
1609 return true;
1610 }
1611 if (dladdr((void*)addr, &dlinfo) != 0) {
1612 if (dlinfo.dli_fname != NULL) {
1613 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1614 }
1615 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1616 *offset = addr - (address)dlinfo.dli_fbase;
1617 }
1618 return true;
1619 }
1620
1621 buf[0] = '\0';
1622 if (offset) *offset = -1;
1623 return false;
1624 }
1625
1626 // Loads .dll/.so and
1627 // in case of error it checks if .dll/.so was built for the
1628 // same architecture as Hotspot is running on
1629
1630
1631 // Remember the stack's state. The Linux dynamic linker will change
1632 // the stack to 'executable' at most once, so we must safepoint only once.
1633 bool os::Linux::_stack_is_executable = false;
1634
1635 // VM operation that loads a library. This is necessary if stack protection
1636 // of the Java stacks can be lost during loading the library. If we
1637 // do not stop the Java threads, they can stack overflow before the stacks
1638 // are protected again.
1639 class VM_LinuxDllLoad: public VM_Operation {
1640 private:
1641 const char *_filename;
1642 char *_ebuf;
1643 int _ebuflen;
1644 void *_lib;
1645 public:
1646 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1647 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1648 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1649 void doit() {
1650 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1651 os::Linux::_stack_is_executable = true;
1652 }
1653 void* loaded_library() { return _lib; }
1654 };
1655
1656 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1657 void * result = NULL;
1658 bool load_attempted = false;
1659
1660 // Check whether the library to load might change execution rights
1661 // of the stack. If they are changed, the protection of the stack
1662 // guard pages will be lost. We need a safepoint to fix this.
1663 //
1664 // See Linux man page execstack(8) for more info.
1665 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1666 ElfFile ef(filename);
1667 if (!ef.specifies_noexecstack()) {
1668 if (!is_init_completed()) {
1669 os::Linux::_stack_is_executable = true;
1670 // This is OK - No Java threads have been created yet, and hence no
1671 // stack guard pages to fix.
1672 //
1673 // This should happen only when you are building JDK7 using a very
1674 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1675 //
1676 // Dynamic loader will make all stacks executable after
1677 // this function returns, and will not do that again.
1678 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1679 } else {
1680 warning("You have loaded library %s which might have disabled stack guard. "
1681 "The VM will try to fix the stack guard now.\n"
1682 "It's highly recommended that you fix the library with "
1683 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1684 filename);
1685
1686 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1687 JavaThread *jt = JavaThread::current();
1688 if (jt->thread_state() != _thread_in_native) {
1689 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1690 // that requires ExecStack. Cannot enter safe point. Let's give up.
1691 warning("Unable to fix stack guard. Giving up.");
1692 } else {
1693 if (!LoadExecStackDllInVMThread) {
1694 // This is for the case where the DLL has an static
1695 // constructor function that executes JNI code. We cannot
1696 // load such DLLs in the VMThread.
1697 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1698 }
1699
1700 ThreadInVMfromNative tiv(jt);
1701 debug_only(VMNativeEntryWrapper vew;)
1702
1703 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1704 VMThread::execute(&op);
1705 if (LoadExecStackDllInVMThread) {
1706 result = op.loaded_library();
1707 }
1708 load_attempted = true;
1709 }
1710 }
1711 }
1712 }
1713
1714 if (!load_attempted) {
1715 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1716 }
1717
1718 if (result != NULL) {
1719 // Successful loading
1720 return result;
1721 }
1722
1723 Elf32_Ehdr elf_head;
1724 int diag_msg_max_length=ebuflen-strlen(ebuf);
1725 char* diag_msg_buf=ebuf+strlen(ebuf);
1726
1727 if (diag_msg_max_length==0) {
1728 // No more space in ebuf for additional diagnostics message
1729 return NULL;
1730 }
1731
1732
1733 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1734
1735 if (file_descriptor < 0) {
1736 // Can't open library, report dlerror() message
1737 return NULL;
1738 }
1739
1740 bool failed_to_read_elf_head=
1741 (sizeof(elf_head)!=
1742 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1743
1744 ::close(file_descriptor);
1745 if (failed_to_read_elf_head) {
1746 // file i/o error - report dlerror() msg
1747 return NULL;
1748 }
1749
1750 typedef struct {
1751 Elf32_Half code; // Actual value as defined in elf.h
1752 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1753 char elf_class; // 32 or 64 bit
1754 char endianess; // MSB or LSB
1755 char* name; // String representation
1756 } arch_t;
1757
1758 #ifndef EM_486
1759 #define EM_486 6 /* Intel 80486 */
1760 #endif
1761 #ifndef EM_AARCH64
1762 #define EM_AARCH64 183 /* ARM AARCH64 */
1763 #endif
1764
1765 static const arch_t arch_array[]={
1766 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1767 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1768 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1769 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1770 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1771 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1772 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1773 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1774 #if defined(VM_LITTLE_ENDIAN)
1775 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1776 #else
1777 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1778 #endif
1779 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1780 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1781 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1782 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1783 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1784 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1785 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1786 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1787 };
1788
1789 #if (defined IA32)
1790 static Elf32_Half running_arch_code=EM_386;
1791 #elif (defined AMD64)
1792 static Elf32_Half running_arch_code=EM_X86_64;
1793 #elif (defined IA64)
1794 static Elf32_Half running_arch_code=EM_IA_64;
1795 #elif (defined __sparc) && (defined _LP64)
1796 static Elf32_Half running_arch_code=EM_SPARCV9;
1797 #elif (defined __sparc) && (!defined _LP64)
1798 static Elf32_Half running_arch_code=EM_SPARC;
1799 #elif (defined __powerpc64__)
1800 static Elf32_Half running_arch_code=EM_PPC64;
1801 #elif (defined __powerpc__)
1802 static Elf32_Half running_arch_code=EM_PPC;
1803 #elif (defined ARM)
1804 static Elf32_Half running_arch_code=EM_ARM;
1805 #elif (defined S390)
1806 static Elf32_Half running_arch_code=EM_S390;
1807 #elif (defined ALPHA)
1808 static Elf32_Half running_arch_code=EM_ALPHA;
1809 #elif (defined MIPSEL)
1810 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1811 #elif (defined PARISC)
1812 static Elf32_Half running_arch_code=EM_PARISC;
1813 #elif (defined MIPS)
1814 static Elf32_Half running_arch_code=EM_MIPS;
1815 #elif (defined M68K)
1816 static Elf32_Half running_arch_code=EM_68K;
1817 #elif (defined AARCH64)
1818 static Elf32_Half running_arch_code=EM_AARCH64;
1819 #else
1820 #error Method os::dll_load requires that one of following is defined:\
1821 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1822 #endif
1823
1824 // Identify compatability class for VM's architecture and library's architecture
1825 // Obtain string descriptions for architectures
1826
1827 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1828 int running_arch_index=-1;
1829
1830 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1831 if (running_arch_code == arch_array[i].code) {
1832 running_arch_index = i;
1833 }
1834 if (lib_arch.code == arch_array[i].code) {
1835 lib_arch.compat_class = arch_array[i].compat_class;
1836 lib_arch.name = arch_array[i].name;
1837 }
1838 }
1839
1840 assert(running_arch_index != -1,
1841 "Didn't find running architecture code (running_arch_code) in arch_array");
1842 if (running_arch_index == -1) {
1843 // Even though running architecture detection failed
1844 // we may still continue with reporting dlerror() message
1845 return NULL;
1846 }
1847
1848 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1849 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1850 return NULL;
1851 }
1852
1853 #ifndef S390
1854 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1855 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1856 return NULL;
1857 }
1858 #endif // !S390
1859
1860 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1861 if (lib_arch.name!=NULL) {
1862 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1863 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1864 lib_arch.name, arch_array[running_arch_index].name);
1865 } else {
1866 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1867 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1868 lib_arch.code,
1869 arch_array[running_arch_index].name);
1870 }
1871 }
1872
1873 return NULL;
1874 }
1875
1876 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1877 int ebuflen) {
1878 void * result = ::dlopen(filename, RTLD_LAZY);
1879 if (result == NULL) {
1880 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1881 ebuf[ebuflen-1] = '\0';
1882 }
1883 return result;
1884 }
1885
1886 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1887 int ebuflen) {
1888 void * result = NULL;
1889 if (LoadExecStackDllInVMThread) {
1890 result = dlopen_helper(filename, ebuf, ebuflen);
1891 }
1892
1893 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1894 // library that requires an executable stack, or which does not have this
1895 // stack attribute set, dlopen changes the stack attribute to executable. The
1896 // read protection of the guard pages gets lost.
1897 //
1898 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1899 // may have been queued at the same time.
1900
1901 if (!_stack_is_executable) {
1902 JavaThread *jt = Threads::first();
1903
1904 while (jt) {
1905 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1906 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
1907 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1908 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1909 warning("Attempt to reguard stack yellow zone failed.");
1910 }
1911 }
1912 jt = jt->next();
1913 }
1914 }
1915
1916 return result;
1917 }
1918
1919 void* os::dll_lookup(void* handle, const char* name) {
1920 void* res = dlsym(handle, name);
1921 return res;
1922 }
1923
1924 void* os::get_default_process_handle() {
1925 return (void*)::dlopen(NULL, RTLD_LAZY);
1926 }
1927
1928 static bool _print_ascii_file(const char* filename, outputStream* st) {
1929 int fd = ::open(filename, O_RDONLY);
1930 if (fd == -1) {
1931 return false;
1932 }
1933
1934 char buf[32];
1935 int bytes;
1936 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1937 st->print_raw(buf, bytes);
1938 }
1939
1940 ::close(fd);
1941
1942 return true;
1943 }
1944
1945 void os::print_dll_info(outputStream *st) {
1946 st->print_cr("Dynamic libraries:");
1947
1948 char fname[32];
1949 pid_t pid = os::Linux::gettid();
1950
1951 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1952
1953 if (!_print_ascii_file(fname, st)) {
1954 st->print("Can not get library information for pid = %d\n", pid);
1955 }
1956 }
1957
1958 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1959 FILE *procmapsFile = NULL;
1960
1961 // Open the procfs maps file for the current process
1962 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1963 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1964 char line[PATH_MAX + 100];
1965
1966 // Read line by line from 'file'
1967 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1968 u8 base, top, offset, inode;
1969 char permissions[5];
1970 char device[6];
1971 char name[PATH_MAX + 1];
1972
1973 // Parse fields from line
1974 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
1975
1976 // Filter by device id '00:00' so that we only get file system mapped files.
1977 if (strcmp(device, "00:00") != 0) {
1978
1979 // Call callback with the fields of interest
1980 if(callback(name, (address)base, (address)top, param)) {
1981 // Oops abort, callback aborted
1982 fclose(procmapsFile);
1983 return 1;
1984 }
1985 }
1986 }
1987 fclose(procmapsFile);
1988 }
1989 return 0;
1990 }
1991
1992 void os::print_os_info_brief(outputStream* st) {
1993 os::Linux::print_distro_info(st);
1994
1995 os::Posix::print_uname_info(st);
1996
1997 os::Linux::print_libversion_info(st);
1998
1999 }
2000
2001 void os::print_os_info(outputStream* st) {
2002 st->print("OS:");
2003
2004 os::Linux::print_distro_info(st);
2005
2006 os::Posix::print_uname_info(st);
2007
2008 // Print warning if unsafe chroot environment detected
2009 if (unsafe_chroot_detected) {
2010 st->print("WARNING!! ");
2011 st->print_cr("%s", unstable_chroot_error);
2012 }
2013
2014 os::Linux::print_libversion_info(st);
2015
2016 os::Posix::print_rlimit_info(st);
2017
2018 os::Posix::print_load_average(st);
2019
2020 os::Linux::print_full_memory_info(st);
2021 }
2022
2023 // Try to identify popular distros.
2024 // Most Linux distributions have a /etc/XXX-release file, which contains
2025 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2026 // file that also contains the OS version string. Some have more than one
2027 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2028 // /etc/redhat-release.), so the order is important.
2029 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2030 // their own specific XXX-release file as well as a redhat-release file.
2031 // Because of this the XXX-release file needs to be searched for before the
2032 // redhat-release file.
2033 // Since Red Hat has a lsb-release file that is not very descriptive the
2034 // search for redhat-release needs to be before lsb-release.
2035 // Since the lsb-release file is the new standard it needs to be searched
2036 // before the older style release files.
2037 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2038 // next to last resort. The os-release file is a new standard that contains
2039 // distribution information and the system-release file seems to be an old
2040 // standard that has been replaced by the lsb-release and os-release files.
2041 // Searching for the debian_version file is the last resort. It contains
2042 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2043 // "Debian " is printed before the contents of the debian_version file.
2044 void os::Linux::print_distro_info(outputStream* st) {
2045 if (!_print_ascii_file("/etc/oracle-release", st) &&
2046 !_print_ascii_file("/etc/mandriva-release", st) &&
2047 !_print_ascii_file("/etc/mandrake-release", st) &&
2048 !_print_ascii_file("/etc/sun-release", st) &&
2049 !_print_ascii_file("/etc/redhat-release", st) &&
2050 !_print_ascii_file("/etc/lsb-release", st) &&
2051 !_print_ascii_file("/etc/SuSE-release", st) &&
2052 !_print_ascii_file("/etc/turbolinux-release", st) &&
2053 !_print_ascii_file("/etc/gentoo-release", st) &&
2054 !_print_ascii_file("/etc/ltib-release", st) &&
2055 !_print_ascii_file("/etc/angstrom-version", st) &&
2056 !_print_ascii_file("/etc/system-release", st) &&
2057 !_print_ascii_file("/etc/os-release", st)) {
2058
2059 if (file_exists("/etc/debian_version")) {
2060 st->print("Debian ");
2061 _print_ascii_file("/etc/debian_version", st);
2062 } else {
2063 st->print("Linux");
2064 }
2065 }
2066 st->cr();
2067 }
2068
2069 void os::Linux::print_libversion_info(outputStream* st) {
2070 // libc, pthread
2071 st->print("libc:");
2072 st->print("%s ", os::Linux::glibc_version());
2073 st->print("%s ", os::Linux::libpthread_version());
2074 st->cr();
2075 }
2076
2077 void os::Linux::print_full_memory_info(outputStream* st) {
2078 st->print("\n/proc/meminfo:\n");
2079 _print_ascii_file("/proc/meminfo", st);
2080 st->cr();
2081 }
2082
2083 void os::print_memory_info(outputStream* st) {
2084
2085 st->print("Memory:");
2086 st->print(" %dk page", os::vm_page_size()>>10);
2087
2088 // values in struct sysinfo are "unsigned long"
2089 struct sysinfo si;
2090 sysinfo(&si);
2091
2092 st->print(", physical " UINT64_FORMAT "k",
2093 os::physical_memory() >> 10);
2094 st->print("(" UINT64_FORMAT "k free)",
2095 os::available_memory() >> 10);
2096 st->print(", swap " UINT64_FORMAT "k",
2097 ((jlong)si.totalswap * si.mem_unit) >> 10);
2098 st->print("(" UINT64_FORMAT "k free)",
2099 ((jlong)si.freeswap * si.mem_unit) >> 10);
2100 st->cr();
2101 }
2102
2103 void os::pd_print_cpu_info(outputStream* st) {
2104 st->print("\n/proc/cpuinfo:\n");
2105 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2106 st->print(" <Not Available>");
2107 }
2108 st->cr();
2109 }
2110
2111 void os::print_siginfo(outputStream* st, void* siginfo) {
2112 const siginfo_t* si = (const siginfo_t*)siginfo;
2113
2114 os::Posix::print_siginfo_brief(st, si);
2115 #if INCLUDE_CDS
2116 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2117 UseSharedSpaces) {
2118 FileMapInfo* mapinfo = FileMapInfo::current_info();
2119 if (mapinfo->is_in_shared_space(si->si_addr)) {
2120 st->print("\n\nError accessing class data sharing archive." \
2121 " Mapped file inaccessible during execution, " \
2122 " possible disk/network problem.");
2123 }
2124 }
2125 #endif
2126 st->cr();
2127 }
2128
2129
2130 static void print_signal_handler(outputStream* st, int sig,
2131 char* buf, size_t buflen);
2132
2133 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2134 st->print_cr("Signal Handlers:");
2135 print_signal_handler(st, SIGSEGV, buf, buflen);
2136 print_signal_handler(st, SIGBUS , buf, buflen);
2137 print_signal_handler(st, SIGFPE , buf, buflen);
2138 print_signal_handler(st, SIGPIPE, buf, buflen);
2139 print_signal_handler(st, SIGXFSZ, buf, buflen);
2140 print_signal_handler(st, SIGILL , buf, buflen);
2141 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2142 print_signal_handler(st, SR_signum, buf, buflen);
2143 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2144 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2145 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2146 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2147 #if defined(PPC64)
2148 print_signal_handler(st, SIGTRAP, buf, buflen);
2149 #endif
2150 }
2151
2152 static char saved_jvm_path[MAXPATHLEN] = {0};
2153
2154 // Find the full path to the current module, libjvm.so
2155 void os::jvm_path(char *buf, jint buflen) {
2156 // Error checking.
2157 if (buflen < MAXPATHLEN) {
2158 assert(false, "must use a large-enough buffer");
2159 buf[0] = '\0';
2160 return;
2161 }
2162 // Lazy resolve the path to current module.
2163 if (saved_jvm_path[0] != 0) {
2164 strcpy(buf, saved_jvm_path);
2165 return;
2166 }
2167
2168 char dli_fname[MAXPATHLEN];
2169 bool ret = dll_address_to_library_name(
2170 CAST_FROM_FN_PTR(address, os::jvm_path),
2171 dli_fname, sizeof(dli_fname), NULL);
2172 assert(ret, "cannot locate libjvm");
2173 char *rp = NULL;
2174 if (ret && dli_fname[0] != '\0') {
2175 rp = realpath(dli_fname, buf);
2176 }
2177 if (rp == NULL) {
2178 return;
2179 }
2180
2181 if (Arguments::sun_java_launcher_is_altjvm()) {
2182 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2183 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2184 // If "/jre/lib/" appears at the right place in the string, then
2185 // assume we are installed in a JDK and we're done. Otherwise, check
2186 // for a JAVA_HOME environment variable and fix up the path so it
2187 // looks like libjvm.so is installed there (append a fake suffix
2188 // hotspot/libjvm.so).
2189 const char *p = buf + strlen(buf) - 1;
2190 for (int count = 0; p > buf && count < 5; ++count) {
2191 for (--p; p > buf && *p != '/'; --p)
2192 /* empty */ ;
2193 }
2194
2195 if (strncmp(p, "/jre/lib/", 9) != 0) {
2196 // Look for JAVA_HOME in the environment.
2197 char* java_home_var = ::getenv("JAVA_HOME");
2198 if (java_home_var != NULL && java_home_var[0] != 0) {
2199 char* jrelib_p;
2200 int len;
2201
2202 // Check the current module name "libjvm.so".
2203 p = strrchr(buf, '/');
2204 if (p == NULL) {
2205 return;
2206 }
2207 assert(strstr(p, "/libjvm") == p, "invalid library name");
2208
2209 rp = realpath(java_home_var, buf);
2210 if (rp == NULL) {
2211 return;
2212 }
2213
2214 // determine if this is a legacy image or modules image
2215 // modules image doesn't have "jre" subdirectory
2216 len = strlen(buf);
2217 assert(len < buflen, "Ran out of buffer room");
2218 jrelib_p = buf + len;
2219 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2220 if (0 != access(buf, F_OK)) {
2221 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2222 }
2223
2224 if (0 == access(buf, F_OK)) {
2225 // Use current module name "libjvm.so"
2226 len = strlen(buf);
2227 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2228 } else {
2229 // Go back to path of .so
2230 rp = realpath(dli_fname, buf);
2231 if (rp == NULL) {
2232 return;
2233 }
2234 }
2235 }
2236 }
2237 }
2238
2239 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2240 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2241 }
2242
2243 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2244 // no prefix required, not even "_"
2245 }
2246
2247 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2248 // no suffix required
2249 }
2250
2251 ////////////////////////////////////////////////////////////////////////////////
2252 // sun.misc.Signal support
2253
2254 static volatile jint sigint_count = 0;
2255
2256 static void UserHandler(int sig, void *siginfo, void *context) {
2257 // 4511530 - sem_post is serialized and handled by the manager thread. When
2258 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2259 // don't want to flood the manager thread with sem_post requests.
2260 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2261 return;
2262 }
2263
2264 // Ctrl-C is pressed during error reporting, likely because the error
2265 // handler fails to abort. Let VM die immediately.
2266 if (sig == SIGINT && is_error_reported()) {
2267 os::die();
2268 }
2269
2270 os::signal_notify(sig);
2271 }
2272
2273 void* os::user_handler() {
2274 return CAST_FROM_FN_PTR(void*, UserHandler);
2275 }
2276
2277 class Semaphore : public StackObj {
2278 public:
2279 Semaphore();
2280 ~Semaphore();
2281 void signal();
2282 void wait();
2283 bool trywait();
2284 bool timedwait(unsigned int sec, int nsec);
2285 private:
2286 sem_t _semaphore;
2287 };
2288
2289 Semaphore::Semaphore() {
2290 sem_init(&_semaphore, 0, 0);
2291 }
2292
2293 Semaphore::~Semaphore() {
2294 sem_destroy(&_semaphore);
2295 }
2296
2297 void Semaphore::signal() {
2298 sem_post(&_semaphore);
2299 }
2300
2301 void Semaphore::wait() {
2302 sem_wait(&_semaphore);
2303 }
2304
2305 bool Semaphore::trywait() {
2306 return sem_trywait(&_semaphore) == 0;
2307 }
2308
2309 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2310
2311 struct timespec ts;
2312 // Semaphore's are always associated with CLOCK_REALTIME
2313 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2314 // see unpackTime for discussion on overflow checking
2315 if (sec >= MAX_SECS) {
2316 ts.tv_sec += MAX_SECS;
2317 ts.tv_nsec = 0;
2318 } else {
2319 ts.tv_sec += sec;
2320 ts.tv_nsec += nsec;
2321 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2322 ts.tv_nsec -= NANOSECS_PER_SEC;
2323 ++ts.tv_sec; // note: this must be <= max_secs
2324 }
2325 }
2326
2327 while (1) {
2328 int result = sem_timedwait(&_semaphore, &ts);
2329 if (result == 0) {
2330 return true;
2331 } else if (errno == EINTR) {
2332 continue;
2333 } else if (errno == ETIMEDOUT) {
2334 return false;
2335 } else {
2336 return false;
2337 }
2338 }
2339 }
2340
2341 extern "C" {
2342 typedef void (*sa_handler_t)(int);
2343 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2344 }
2345
2346 void* os::signal(int signal_number, void* handler) {
2347 struct sigaction sigAct, oldSigAct;
2348
2349 sigfillset(&(sigAct.sa_mask));
2350 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2351 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2352
2353 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2354 // -1 means registration failed
2355 return (void *)-1;
2356 }
2357
2358 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2359 }
2360
2361 void os::signal_raise(int signal_number) {
2362 ::raise(signal_number);
2363 }
2364
2365 // The following code is moved from os.cpp for making this
2366 // code platform specific, which it is by its very nature.
2367
2368 // Will be modified when max signal is changed to be dynamic
2369 int os::sigexitnum_pd() {
2370 return NSIG;
2371 }
2372
2373 // a counter for each possible signal value
2374 static volatile jint pending_signals[NSIG+1] = { 0 };
2375
2376 // Linux(POSIX) specific hand shaking semaphore.
2377 static sem_t sig_sem;
2378 static Semaphore sr_semaphore;
2379
2380 void os::signal_init_pd() {
2381 // Initialize signal structures
2382 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2383
2384 // Initialize signal semaphore
2385 ::sem_init(&sig_sem, 0, 0);
2386 }
2387
2388 void os::signal_notify(int sig) {
2389 Atomic::inc(&pending_signals[sig]);
2390 ::sem_post(&sig_sem);
2391 }
2392
2393 static int check_pending_signals(bool wait) {
2394 Atomic::store(0, &sigint_count);
2395 for (;;) {
2396 for (int i = 0; i < NSIG + 1; i++) {
2397 jint n = pending_signals[i];
2398 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2399 return i;
2400 }
2401 }
2402 if (!wait) {
2403 return -1;
2404 }
2405 JavaThread *thread = JavaThread::current();
2406 ThreadBlockInVM tbivm(thread);
2407
2408 bool threadIsSuspended;
2409 do {
2410 thread->set_suspend_equivalent();
2411 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2412 ::sem_wait(&sig_sem);
2413
2414 // were we externally suspended while we were waiting?
2415 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2416 if (threadIsSuspended) {
2417 // The semaphore has been incremented, but while we were waiting
2418 // another thread suspended us. We don't want to continue running
2419 // while suspended because that would surprise the thread that
2420 // suspended us.
2421 ::sem_post(&sig_sem);
2422
2423 thread->java_suspend_self();
2424 }
2425 } while (threadIsSuspended);
2426 }
2427 }
2428
2429 int os::signal_lookup() {
2430 return check_pending_signals(false);
2431 }
2432
2433 int os::signal_wait() {
2434 return check_pending_signals(true);
2435 }
2436
2437 ////////////////////////////////////////////////////////////////////////////////
2438 // Virtual Memory
2439
2440 int os::vm_page_size() {
2441 // Seems redundant as all get out
2442 assert(os::Linux::page_size() != -1, "must call os::init");
2443 return os::Linux::page_size();
2444 }
2445
2446 // Solaris allocates memory by pages.
2447 int os::vm_allocation_granularity() {
2448 assert(os::Linux::page_size() != -1, "must call os::init");
2449 return os::Linux::page_size();
2450 }
2451
2452 // Rationale behind this function:
2453 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2454 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2455 // samples for JITted code. Here we create private executable mapping over the code cache
2456 // and then we can use standard (well, almost, as mapping can change) way to provide
2457 // info for the reporting script by storing timestamp and location of symbol
2458 void linux_wrap_code(char* base, size_t size) {
2459 static volatile jint cnt = 0;
2460
2461 if (!UseOprofile) {
2462 return;
2463 }
2464
2465 char buf[PATH_MAX+1];
2466 int num = Atomic::add(1, &cnt);
2467
2468 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2469 os::get_temp_directory(), os::current_process_id(), num);
2470 unlink(buf);
2471
2472 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2473
2474 if (fd != -1) {
2475 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2476 if (rv != (off_t)-1) {
2477 if (::write(fd, "", 1) == 1) {
2478 mmap(base, size,
2479 PROT_READ|PROT_WRITE|PROT_EXEC,
2480 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2481 }
2482 }
2483 ::close(fd);
2484 unlink(buf);
2485 }
2486 }
2487
2488 static bool recoverable_mmap_error(int err) {
2489 // See if the error is one we can let the caller handle. This
2490 // list of errno values comes from JBS-6843484. I can't find a
2491 // Linux man page that documents this specific set of errno
2492 // values so while this list currently matches Solaris, it may
2493 // change as we gain experience with this failure mode.
2494 switch (err) {
2495 case EBADF:
2496 case EINVAL:
2497 case ENOTSUP:
2498 // let the caller deal with these errors
2499 return true;
2500
2501 default:
2502 // Any remaining errors on this OS can cause our reserved mapping
2503 // to be lost. That can cause confusion where different data
2504 // structures think they have the same memory mapped. The worst
2505 // scenario is if both the VM and a library think they have the
2506 // same memory mapped.
2507 return false;
2508 }
2509 }
2510
2511 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2512 int err) {
2513 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2514 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2515 strerror(err), err);
2516 }
2517
2518 static void warn_fail_commit_memory(char* addr, size_t size,
2519 size_t alignment_hint, bool exec,
2520 int err) {
2521 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2522 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2523 alignment_hint, exec, strerror(err), err);
2524 }
2525
2526 // NOTE: Linux kernel does not really reserve the pages for us.
2527 // All it does is to check if there are enough free pages
2528 // left at the time of mmap(). This could be a potential
2529 // problem.
2530 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2531 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2532 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2533 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2534 if (res != (uintptr_t) MAP_FAILED) {
2535 if (UseNUMAInterleaving) {
2536 numa_make_global(addr, size);
2537 }
2538 return 0;
2539 }
2540
2541 int err = errno; // save errno from mmap() call above
2542
2543 if (!recoverable_mmap_error(err)) {
2544 warn_fail_commit_memory(addr, size, exec, err);
2545 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2546 }
2547
2548 return err;
2549 }
2550
2551 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2552 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2553 }
2554
2555 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2556 const char* mesg) {
2557 assert(mesg != NULL, "mesg must be specified");
2558 int err = os::Linux::commit_memory_impl(addr, size, exec);
2559 if (err != 0) {
2560 // the caller wants all commit errors to exit with the specified mesg:
2561 warn_fail_commit_memory(addr, size, exec, err);
2562 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2563 }
2564 }
2565
2566 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2567 #ifndef MAP_HUGETLB
2568 #define MAP_HUGETLB 0x40000
2569 #endif
2570
2571 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2572 #ifndef MADV_HUGEPAGE
2573 #define MADV_HUGEPAGE 14
2574 #endif
2575
2576 int os::Linux::commit_memory_impl(char* addr, size_t size,
2577 size_t alignment_hint, bool exec) {
2578 int err = os::Linux::commit_memory_impl(addr, size, exec);
2579 if (err == 0) {
2580 realign_memory(addr, size, alignment_hint);
2581 }
2582 return err;
2583 }
2584
2585 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2586 bool exec) {
2587 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2588 }
2589
2590 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2591 size_t alignment_hint, bool exec,
2592 const char* mesg) {
2593 assert(mesg != NULL, "mesg must be specified");
2594 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2595 if (err != 0) {
2596 // the caller wants all commit errors to exit with the specified mesg:
2597 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2598 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2599 }
2600 }
2601
2602 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2603 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2604 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2605 // be supported or the memory may already be backed by huge pages.
2606 ::madvise(addr, bytes, MADV_HUGEPAGE);
2607 }
2608 }
2609
2610 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2611 // This method works by doing an mmap over an existing mmaping and effectively discarding
2612 // the existing pages. However it won't work for SHM-based large pages that cannot be
2613 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2614 // small pages on top of the SHM segment. This method always works for small pages, so we
2615 // allow that in any case.
2616 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2617 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2618 }
2619 }
2620
2621 void os::numa_make_global(char *addr, size_t bytes) {
2622 Linux::numa_interleave_memory(addr, bytes);
2623 }
2624
2625 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2626 // bind policy to MPOL_PREFERRED for the current thread.
2627 #define USE_MPOL_PREFERRED 0
2628
2629 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2630 // To make NUMA and large pages more robust when both enabled, we need to ease
2631 // the requirements on where the memory should be allocated. MPOL_BIND is the
2632 // default policy and it will force memory to be allocated on the specified
2633 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2634 // the specified node, but will not force it. Using this policy will prevent
2635 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2636 // free large pages.
2637 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2638 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2639 }
2640
2641 bool os::numa_topology_changed() { return false; }
2642
2643 size_t os::numa_get_groups_num() {
2644 int max_node = Linux::numa_max_node();
2645 return max_node > 0 ? max_node + 1 : 1;
2646 }
2647
2648 int os::numa_get_group_id() {
2649 int cpu_id = Linux::sched_getcpu();
2650 if (cpu_id != -1) {
2651 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2652 if (lgrp_id != -1) {
2653 return lgrp_id;
2654 }
2655 }
2656 return 0;
2657 }
2658
2659 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2660 for (size_t i = 0; i < size; i++) {
2661 ids[i] = i;
2662 }
2663 return size;
2664 }
2665
2666 bool os::get_page_info(char *start, page_info* info) {
2667 return false;
2668 }
2669
2670 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2671 page_info* page_found) {
2672 return end;
2673 }
2674
2675
2676 int os::Linux::sched_getcpu_syscall(void) {
2677 unsigned int cpu;
2678 int retval = -1;
2679
2680 #if defined(IA32)
2681 #ifndef SYS_getcpu
2682 #define SYS_getcpu 318
2683 #endif
2684 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2685 #elif defined(AMD64)
2686 // Unfortunately we have to bring all these macros here from vsyscall.h
2687 // to be able to compile on old linuxes.
2688 #define __NR_vgetcpu 2
2689 #define VSYSCALL_START (-10UL << 20)
2690 #define VSYSCALL_SIZE 1024
2691 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2692 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2693 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2694 retval = vgetcpu(&cpu, NULL, NULL);
2695 #endif
2696
2697 return (retval == -1) ? retval : cpu;
2698 }
2699
2700 // Something to do with the numa-aware allocator needs these symbols
2701 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2702 extern "C" JNIEXPORT void numa_error(char *where) { }
2703 extern "C" JNIEXPORT int fork1() { return fork(); }
2704
2705
2706 // If we are running with libnuma version > 2, then we should
2707 // be trying to use symbols with versions 1.1
2708 // If we are running with earlier version, which did not have symbol versions,
2709 // we should use the base version.
2710 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2711 void *f = dlvsym(handle, name, "libnuma_1.1");
2712 if (f == NULL) {
2713 f = dlsym(handle, name);
2714 }
2715 return f;
2716 }
2717
2718 bool os::Linux::libnuma_init() {
2719 // sched_getcpu() should be in libc.
2720 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2721 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2722
2723 // If it's not, try a direct syscall.
2724 if (sched_getcpu() == -1) {
2725 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2726 (void*)&sched_getcpu_syscall));
2727 }
2728
2729 if (sched_getcpu() != -1) { // Does it work?
2730 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2731 if (handle != NULL) {
2732 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2733 libnuma_dlsym(handle, "numa_node_to_cpus")));
2734 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2735 libnuma_dlsym(handle, "numa_max_node")));
2736 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2737 libnuma_dlsym(handle, "numa_available")));
2738 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2739 libnuma_dlsym(handle, "numa_tonode_memory")));
2740 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2741 libnuma_dlsym(handle, "numa_interleave_memory")));
2742 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2743 libnuma_dlsym(handle, "numa_set_bind_policy")));
2744
2745
2746 if (numa_available() != -1) {
2747 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2748 // Create a cpu -> node mapping
2749 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2750 rebuild_cpu_to_node_map();
2751 return true;
2752 }
2753 }
2754 }
2755 return false;
2756 }
2757
2758 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2759 // The table is later used in get_node_by_cpu().
2760 void os::Linux::rebuild_cpu_to_node_map() {
2761 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2762 // in libnuma (possible values are starting from 16,
2763 // and continuing up with every other power of 2, but less
2764 // than the maximum number of CPUs supported by kernel), and
2765 // is a subject to change (in libnuma version 2 the requirements
2766 // are more reasonable) we'll just hardcode the number they use
2767 // in the library.
2768 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2769
2770 size_t cpu_num = os::active_processor_count();
2771 size_t cpu_map_size = NCPUS / BitsPerCLong;
2772 size_t cpu_map_valid_size =
2773 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2774
2775 cpu_to_node()->clear();
2776 cpu_to_node()->at_grow(cpu_num - 1);
2777 size_t node_num = numa_get_groups_num();
2778
2779 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2780 for (size_t i = 0; i < node_num; i++) {
2781 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2782 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2783 if (cpu_map[j] != 0) {
2784 for (size_t k = 0; k < BitsPerCLong; k++) {
2785 if (cpu_map[j] & (1UL << k)) {
2786 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2787 }
2788 }
2789 }
2790 }
2791 }
2792 }
2793 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2794 }
2795
2796 int os::Linux::get_node_by_cpu(int cpu_id) {
2797 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2798 return cpu_to_node()->at(cpu_id);
2799 }
2800 return -1;
2801 }
2802
2803 GrowableArray<int>* os::Linux::_cpu_to_node;
2804 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2805 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2806 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2807 os::Linux::numa_available_func_t os::Linux::_numa_available;
2808 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2809 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2810 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2811 unsigned long* os::Linux::_numa_all_nodes;
2812
2813 bool os::pd_uncommit_memory(char* addr, size_t size) {
2814 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2815 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2816 return res != (uintptr_t) MAP_FAILED;
2817 }
2818
2819 static address get_stack_commited_bottom(address bottom, size_t size) {
2820 address nbot = bottom;
2821 address ntop = bottom + size;
2822
2823 size_t page_sz = os::vm_page_size();
2824 unsigned pages = size / page_sz;
2825
2826 unsigned char vec[1];
2827 unsigned imin = 1, imax = pages + 1, imid;
2828 int mincore_return_value = 0;
2829
2830 assert(imin <= imax, "Unexpected page size");
2831
2832 while (imin < imax) {
2833 imid = (imax + imin) / 2;
2834 nbot = ntop - (imid * page_sz);
2835
2836 // Use a trick with mincore to check whether the page is mapped or not.
2837 // mincore sets vec to 1 if page resides in memory and to 0 if page
2838 // is swapped output but if page we are asking for is unmapped
2839 // it returns -1,ENOMEM
2840 mincore_return_value = mincore(nbot, page_sz, vec);
2841
2842 if (mincore_return_value == -1) {
2843 // Page is not mapped go up
2844 // to find first mapped page
2845 if (errno != EAGAIN) {
2846 assert(errno == ENOMEM, "Unexpected mincore errno");
2847 imax = imid;
2848 }
2849 } else {
2850 // Page is mapped go down
2851 // to find first not mapped page
2852 imin = imid + 1;
2853 }
2854 }
2855
2856 nbot = nbot + page_sz;
2857
2858 // Adjust stack bottom one page up if last checked page is not mapped
2859 if (mincore_return_value == -1) {
2860 nbot = nbot + page_sz;
2861 }
2862
2863 return nbot;
2864 }
2865
2866
2867 // Linux uses a growable mapping for the stack, and if the mapping for
2868 // the stack guard pages is not removed when we detach a thread the
2869 // stack cannot grow beyond the pages where the stack guard was
2870 // mapped. If at some point later in the process the stack expands to
2871 // that point, the Linux kernel cannot expand the stack any further
2872 // because the guard pages are in the way, and a segfault occurs.
2873 //
2874 // However, it's essential not to split the stack region by unmapping
2875 // a region (leaving a hole) that's already part of the stack mapping,
2876 // so if the stack mapping has already grown beyond the guard pages at
2877 // the time we create them, we have to truncate the stack mapping.
2878 // So, we need to know the extent of the stack mapping when
2879 // create_stack_guard_pages() is called.
2880
2881 // We only need this for stacks that are growable: at the time of
2882 // writing thread stacks don't use growable mappings (i.e. those
2883 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2884 // only applies to the main thread.
2885
2886 // If the (growable) stack mapping already extends beyond the point
2887 // where we're going to put our guard pages, truncate the mapping at
2888 // that point by munmap()ping it. This ensures that when we later
2889 // munmap() the guard pages we don't leave a hole in the stack
2890 // mapping. This only affects the main/initial thread
2891
2892 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2893 if (os::Linux::is_initial_thread()) {
2894 // As we manually grow stack up to bottom inside create_attached_thread(),
2895 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2896 // we don't need to do anything special.
2897 // Check it first, before calling heavy function.
2898 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2899 unsigned char vec[1];
2900
2901 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2902 // Fallback to slow path on all errors, including EAGAIN
2903 stack_extent = (uintptr_t) get_stack_commited_bottom(
2904 os::Linux::initial_thread_stack_bottom(),
2905 (size_t)addr - stack_extent);
2906 }
2907
2908 if (stack_extent < (uintptr_t)addr) {
2909 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2910 }
2911 }
2912
2913 return os::commit_memory(addr, size, !ExecMem);
2914 }
2915
2916 // If this is a growable mapping, remove the guard pages entirely by
2917 // munmap()ping them. If not, just call uncommit_memory(). This only
2918 // affects the main/initial thread, but guard against future OS changes
2919 // It's safe to always unmap guard pages for initial thread because we
2920 // always place it right after end of the mapped region
2921
2922 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2923 uintptr_t stack_extent, stack_base;
2924
2925 if (os::Linux::is_initial_thread()) {
2926 return ::munmap(addr, size) == 0;
2927 }
2928
2929 return os::uncommit_memory(addr, size);
2930 }
2931
2932 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2933 // at 'requested_addr'. If there are existing memory mappings at the same
2934 // location, however, they will be overwritten. If 'fixed' is false,
2935 // 'requested_addr' is only treated as a hint, the return value may or
2936 // may not start from the requested address. Unlike Linux mmap(), this
2937 // function returns NULL to indicate failure.
2938 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2939 char * addr;
2940 int flags;
2941
2942 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2943 if (fixed) {
2944 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2945 flags |= MAP_FIXED;
2946 }
2947
2948 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
2949 // touch an uncommitted page. Otherwise, the read/write might
2950 // succeed if we have enough swap space to back the physical page.
2951 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
2952 flags, -1, 0);
2953
2954 return addr == MAP_FAILED ? NULL : addr;
2955 }
2956
2957 static int anon_munmap(char * addr, size_t size) {
2958 return ::munmap(addr, size) == 0;
2959 }
2960
2961 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2962 size_t alignment_hint) {
2963 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2964 }
2965
2966 bool os::pd_release_memory(char* addr, size_t size) {
2967 return anon_munmap(addr, size);
2968 }
2969
2970 static bool linux_mprotect(char* addr, size_t size, int prot) {
2971 // Linux wants the mprotect address argument to be page aligned.
2972 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2973
2974 // According to SUSv3, mprotect() should only be used with mappings
2975 // established by mmap(), and mmap() always maps whole pages. Unaligned
2976 // 'addr' likely indicates problem in the VM (e.g. trying to change
2977 // protection of malloc'ed or statically allocated memory). Check the
2978 // caller if you hit this assert.
2979 assert(addr == bottom, "sanity check");
2980
2981 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2982 return ::mprotect(bottom, size, prot) == 0;
2983 }
2984
2985 // Set protections specified
2986 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2987 bool is_committed) {
2988 unsigned int p = 0;
2989 switch (prot) {
2990 case MEM_PROT_NONE: p = PROT_NONE; break;
2991 case MEM_PROT_READ: p = PROT_READ; break;
2992 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2993 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2994 default:
2995 ShouldNotReachHere();
2996 }
2997 // is_committed is unused.
2998 return linux_mprotect(addr, bytes, p);
2999 }
3000
3001 bool os::guard_memory(char* addr, size_t size) {
3002 return linux_mprotect(addr, size, PROT_NONE);
3003 }
3004
3005 bool os::unguard_memory(char* addr, size_t size) {
3006 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3007 }
3008
3009 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3010 size_t page_size) {
3011 bool result = false;
3012 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3013 MAP_ANONYMOUS|MAP_PRIVATE,
3014 -1, 0);
3015 if (p != MAP_FAILED) {
3016 void *aligned_p = align_ptr_up(p, page_size);
3017
3018 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3019
3020 munmap(p, page_size * 2);
3021 }
3022
3023 if (warn && !result) {
3024 warning("TransparentHugePages is not supported by the operating system.");
3025 }
3026
3027 return result;
3028 }
3029
3030 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3031 bool result = false;
3032 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3033 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3034 -1, 0);
3035
3036 if (p != MAP_FAILED) {
3037 // We don't know if this really is a huge page or not.
3038 FILE *fp = fopen("/proc/self/maps", "r");
3039 if (fp) {
3040 while (!feof(fp)) {
3041 char chars[257];
3042 long x = 0;
3043 if (fgets(chars, sizeof(chars), fp)) {
3044 if (sscanf(chars, "%lx-%*x", &x) == 1
3045 && x == (long)p) {
3046 if (strstr (chars, "hugepage")) {
3047 result = true;
3048 break;
3049 }
3050 }
3051 }
3052 }
3053 fclose(fp);
3054 }
3055 munmap(p, page_size);
3056 }
3057
3058 if (warn && !result) {
3059 warning("HugeTLBFS is not supported by the operating system.");
3060 }
3061
3062 return result;
3063 }
3064
3065 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3066 //
3067 // From the coredump_filter documentation:
3068 //
3069 // - (bit 0) anonymous private memory
3070 // - (bit 1) anonymous shared memory
3071 // - (bit 2) file-backed private memory
3072 // - (bit 3) file-backed shared memory
3073 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3074 // effective only if the bit 2 is cleared)
3075 // - (bit 5) hugetlb private memory
3076 // - (bit 6) hugetlb shared memory
3077 //
3078 static void set_coredump_filter(void) {
3079 FILE *f;
3080 long cdm;
3081
3082 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3083 return;
3084 }
3085
3086 if (fscanf(f, "%lx", &cdm) != 1) {
3087 fclose(f);
3088 return;
3089 }
3090
3091 rewind(f);
3092
3093 if ((cdm & LARGEPAGES_BIT) == 0) {
3094 cdm |= LARGEPAGES_BIT;
3095 fprintf(f, "%#lx", cdm);
3096 }
3097
3098 fclose(f);
3099 }
3100
3101 // Large page support
3102
3103 static size_t _large_page_size = 0;
3104
3105 size_t os::Linux::find_large_page_size() {
3106 size_t large_page_size = 0;
3107
3108 // large_page_size on Linux is used to round up heap size. x86 uses either
3109 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3110 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3111 // page as large as 256M.
3112 //
3113 // Here we try to figure out page size by parsing /proc/meminfo and looking
3114 // for a line with the following format:
3115 // Hugepagesize: 2048 kB
3116 //
3117 // If we can't determine the value (e.g. /proc is not mounted, or the text
3118 // format has been changed), we'll use the largest page size supported by
3119 // the processor.
3120
3121 #ifndef ZERO
3122 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3123 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3124 #endif // ZERO
3125
3126 FILE *fp = fopen("/proc/meminfo", "r");
3127 if (fp) {
3128 while (!feof(fp)) {
3129 int x = 0;
3130 char buf[16];
3131 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3132 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3133 large_page_size = x * K;
3134 break;
3135 }
3136 } else {
3137 // skip to next line
3138 for (;;) {
3139 int ch = fgetc(fp);
3140 if (ch == EOF || ch == (int)'\n') break;
3141 }
3142 }
3143 }
3144 fclose(fp);
3145 }
3146
3147 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3148 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3149 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3150 proper_unit_for_byte_size(large_page_size));
3151 }
3152
3153 return large_page_size;
3154 }
3155
3156 size_t os::Linux::setup_large_page_size() {
3157 _large_page_size = Linux::find_large_page_size();
3158 const size_t default_page_size = (size_t)Linux::page_size();
3159 if (_large_page_size > default_page_size) {
3160 _page_sizes[0] = _large_page_size;
3161 _page_sizes[1] = default_page_size;
3162 _page_sizes[2] = 0;
3163 }
3164
3165 return _large_page_size;
3166 }
3167
3168 bool os::Linux::setup_large_page_type(size_t page_size) {
3169 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3170 FLAG_IS_DEFAULT(UseSHM) &&
3171 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3172
3173 // The type of large pages has not been specified by the user.
3174
3175 // Try UseHugeTLBFS and then UseSHM.
3176 UseHugeTLBFS = UseSHM = true;
3177
3178 // Don't try UseTransparentHugePages since there are known
3179 // performance issues with it turned on. This might change in the future.
3180 UseTransparentHugePages = false;
3181 }
3182
3183 if (UseTransparentHugePages) {
3184 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3185 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3186 UseHugeTLBFS = false;
3187 UseSHM = false;
3188 return true;
3189 }
3190 UseTransparentHugePages = false;
3191 }
3192
3193 if (UseHugeTLBFS) {
3194 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3195 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3196 UseSHM = false;
3197 return true;
3198 }
3199 UseHugeTLBFS = false;
3200 }
3201
3202 return UseSHM;
3203 }
3204
3205 void os::large_page_init() {
3206 if (!UseLargePages &&
3207 !UseTransparentHugePages &&
3208 !UseHugeTLBFS &&
3209 !UseSHM) {
3210 // Not using large pages.
3211 return;
3212 }
3213
3214 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3215 // The user explicitly turned off large pages.
3216 // Ignore the rest of the large pages flags.
3217 UseTransparentHugePages = false;
3218 UseHugeTLBFS = false;
3219 UseSHM = false;
3220 return;
3221 }
3222
3223 size_t large_page_size = Linux::setup_large_page_size();
3224 UseLargePages = Linux::setup_large_page_type(large_page_size);
3225
3226 set_coredump_filter();
3227 }
3228
3229 #ifndef SHM_HUGETLB
3230 #define SHM_HUGETLB 04000
3231 #endif
3232
3233 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3234 char* req_addr, bool exec) {
3235 // "exec" is passed in but not used. Creating the shared image for
3236 // the code cache doesn't have an SHM_X executable permission to check.
3237 assert(UseLargePages && UseSHM, "only for SHM large pages");
3238 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3239
3240 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3241 return NULL; // Fallback to small pages.
3242 }
3243
3244 key_t key = IPC_PRIVATE;
3245 char *addr;
3246
3247 bool warn_on_failure = UseLargePages &&
3248 (!FLAG_IS_DEFAULT(UseLargePages) ||
3249 !FLAG_IS_DEFAULT(UseSHM) ||
3250 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3251 char msg[128];
3252
3253 // Create a large shared memory region to attach to based on size.
3254 // Currently, size is the total size of the heap
3255 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3256 if (shmid == -1) {
3257 // Possible reasons for shmget failure:
3258 // 1. shmmax is too small for Java heap.
3259 // > check shmmax value: cat /proc/sys/kernel/shmmax
3260 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3261 // 2. not enough large page memory.
3262 // > check available large pages: cat /proc/meminfo
3263 // > increase amount of large pages:
3264 // echo new_value > /proc/sys/vm/nr_hugepages
3265 // Note 1: different Linux may use different name for this property,
3266 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3267 // Note 2: it's possible there's enough physical memory available but
3268 // they are so fragmented after a long run that they can't
3269 // coalesce into large pages. Try to reserve large pages when
3270 // the system is still "fresh".
3271 if (warn_on_failure) {
3272 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3273 warning("%s", msg);
3274 }
3275 return NULL;
3276 }
3277
3278 // attach to the region
3279 addr = (char*)shmat(shmid, req_addr, 0);
3280 int err = errno;
3281
3282 // Remove shmid. If shmat() is successful, the actual shared memory segment
3283 // will be deleted when it's detached by shmdt() or when the process
3284 // terminates. If shmat() is not successful this will remove the shared
3285 // segment immediately.
3286 shmctl(shmid, IPC_RMID, NULL);
3287
3288 if ((intptr_t)addr == -1) {
3289 if (warn_on_failure) {
3290 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3291 warning("%s", msg);
3292 }
3293 return NULL;
3294 }
3295
3296 return addr;
3297 }
3298
3299 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3300 int error) {
3301 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3302
3303 bool warn_on_failure = UseLargePages &&
3304 (!FLAG_IS_DEFAULT(UseLargePages) ||
3305 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3306 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3307
3308 if (warn_on_failure) {
3309 char msg[128];
3310 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3311 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3312 warning("%s", msg);
3313 }
3314 }
3315
3316 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3317 char* req_addr,
3318 bool exec) {
3319 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3320 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3321 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3322
3323 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3324 char* addr = (char*)::mmap(req_addr, bytes, prot,
3325 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3326 -1, 0);
3327
3328 if (addr == MAP_FAILED) {
3329 warn_on_large_pages_failure(req_addr, bytes, errno);
3330 return NULL;
3331 }
3332
3333 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3334
3335 return addr;
3336 }
3337
3338 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3339 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3340 // (req_addr != NULL) or with a given alignment.
3341 // - bytes shall be a multiple of alignment.
3342 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3343 // - alignment sets the alignment at which memory shall be allocated.
3344 // It must be a multiple of allocation granularity.
3345 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3346 // req_addr or NULL.
3347 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3348
3349 size_t extra_size = bytes;
3350 if (req_addr == NULL && alignment > 0) {
3351 extra_size += alignment;
3352 }
3353
3354 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3355 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3356 -1, 0);
3357 if (start == MAP_FAILED) {
3358 start = NULL;
3359 } else {
3360 if (req_addr != NULL) {
3361 if (start != req_addr) {
3362 ::munmap(start, extra_size);
3363 start = NULL;
3364 }
3365 } else {
3366 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3367 char* const end_aligned = start_aligned + bytes;
3368 char* const end = start + extra_size;
3369 if (start_aligned > start) {
3370 ::munmap(start, start_aligned - start);
3371 }
3372 if (end_aligned < end) {
3373 ::munmap(end_aligned, end - end_aligned);
3374 }
3375 start = start_aligned;
3376 }
3377 }
3378 return start;
3379
3380 }
3381
3382 // Reserve memory using mmap(MAP_HUGETLB).
3383 // - bytes shall be a multiple of alignment.
3384 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3385 // - alignment sets the alignment at which memory shall be allocated.
3386 // It must be a multiple of allocation granularity.
3387 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3388 // req_addr or NULL.
3389 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3390 size_t alignment,
3391 char* req_addr,
3392 bool exec) {
3393 size_t large_page_size = os::large_page_size();
3394 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3395
3396 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3397 assert(is_size_aligned(bytes, alignment), "Must be");
3398
3399 // First reserve - but not commit - the address range in small pages.
3400 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3401
3402 if (start == NULL) {
3403 return NULL;
3404 }
3405
3406 assert(is_ptr_aligned(start, alignment), "Must be");
3407
3408 char* end = start + bytes;
3409
3410 // Find the regions of the allocated chunk that can be promoted to large pages.
3411 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3412 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3413
3414 size_t lp_bytes = lp_end - lp_start;
3415
3416 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3417
3418 if (lp_bytes == 0) {
3419 // The mapped region doesn't even span the start and the end of a large page.
3420 // Fall back to allocate a non-special area.
3421 ::munmap(start, end - start);
3422 return NULL;
3423 }
3424
3425 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3426
3427 void* result;
3428
3429 // Commit small-paged leading area.
3430 if (start != lp_start) {
3431 result = ::mmap(start, lp_start - start, prot,
3432 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3433 -1, 0);
3434 if (result == MAP_FAILED) {
3435 ::munmap(lp_start, end - lp_start);
3436 return NULL;
3437 }
3438 }
3439
3440 // Commit large-paged area.
3441 result = ::mmap(lp_start, lp_bytes, prot,
3442 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3443 -1, 0);
3444 if (result == MAP_FAILED) {
3445 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3446 // If the mmap above fails, the large pages region will be unmapped and we
3447 // have regions before and after with small pages. Release these regions.
3448 //
3449 // | mapped | unmapped | mapped |
3450 // ^ ^ ^ ^
3451 // start lp_start lp_end end
3452 //
3453 ::munmap(start, lp_start - start);
3454 ::munmap(lp_end, end - lp_end);
3455 return NULL;
3456 }
3457
3458 // Commit small-paged trailing area.
3459 if (lp_end != end) {
3460 result = ::mmap(lp_end, end - lp_end, prot,
3461 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3462 -1, 0);
3463 if (result == MAP_FAILED) {
3464 ::munmap(start, lp_end - start);
3465 return NULL;
3466 }
3467 }
3468
3469 return start;
3470 }
3471
3472 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3473 size_t alignment,
3474 char* req_addr,
3475 bool exec) {
3476 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3477 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3478 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3479 assert(is_power_of_2(os::large_page_size()), "Must be");
3480 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3481
3482 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3483 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3484 } else {
3485 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3486 }
3487 }
3488
3489 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3490 char* req_addr, bool exec) {
3491 assert(UseLargePages, "only for large pages");
3492
3493 char* addr;
3494 if (UseSHM) {
3495 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3496 } else {
3497 assert(UseHugeTLBFS, "must be");
3498 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3499 }
3500
3501 if (addr != NULL) {
3502 if (UseNUMAInterleaving) {
3503 numa_make_global(addr, bytes);
3504 }
3505
3506 // The memory is committed
3507 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3508 }
3509
3510 return addr;
3511 }
3512
3513 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3514 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3515 return shmdt(base) == 0;
3516 }
3517
3518 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3519 return pd_release_memory(base, bytes);
3520 }
3521
3522 bool os::release_memory_special(char* base, size_t bytes) {
3523 bool res;
3524 if (MemTracker::tracking_level() > NMT_minimal) {
3525 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3526 res = os::Linux::release_memory_special_impl(base, bytes);
3527 if (res) {
3528 tkr.record((address)base, bytes);
3529 }
3530
3531 } else {
3532 res = os::Linux::release_memory_special_impl(base, bytes);
3533 }
3534 return res;
3535 }
3536
3537 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3538 assert(UseLargePages, "only for large pages");
3539 bool res;
3540
3541 if (UseSHM) {
3542 res = os::Linux::release_memory_special_shm(base, bytes);
3543 } else {
3544 assert(UseHugeTLBFS, "must be");
3545 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3546 }
3547 return res;
3548 }
3549
3550 size_t os::large_page_size() {
3551 return _large_page_size;
3552 }
3553
3554 // With SysV SHM the entire memory region must be allocated as shared
3555 // memory.
3556 // HugeTLBFS allows application to commit large page memory on demand.
3557 // However, when committing memory with HugeTLBFS fails, the region
3558 // that was supposed to be committed will lose the old reservation
3559 // and allow other threads to steal that memory region. Because of this
3560 // behavior we can't commit HugeTLBFS memory.
3561 bool os::can_commit_large_page_memory() {
3562 return UseTransparentHugePages;
3563 }
3564
3565 bool os::can_execute_large_page_memory() {
3566 return UseTransparentHugePages || UseHugeTLBFS;
3567 }
3568
3569 // Reserve memory at an arbitrary address, only if that area is
3570 // available (and not reserved for something else).
3571
3572 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3573 const int max_tries = 10;
3574 char* base[max_tries];
3575 size_t size[max_tries];
3576 const size_t gap = 0x000000;
3577
3578 // Assert only that the size is a multiple of the page size, since
3579 // that's all that mmap requires, and since that's all we really know
3580 // about at this low abstraction level. If we need higher alignment,
3581 // we can either pass an alignment to this method or verify alignment
3582 // in one of the methods further up the call chain. See bug 5044738.
3583 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3584
3585 // Repeatedly allocate blocks until the block is allocated at the
3586 // right spot.
3587
3588 // Linux mmap allows caller to pass an address as hint; give it a try first,
3589 // if kernel honors the hint then we can return immediately.
3590 char * addr = anon_mmap(requested_addr, bytes, false);
3591 if (addr == requested_addr) {
3592 return requested_addr;
3593 }
3594
3595 if (addr != NULL) {
3596 // mmap() is successful but it fails to reserve at the requested address
3597 anon_munmap(addr, bytes);
3598 }
3599
3600 int i;
3601 for (i = 0; i < max_tries; ++i) {
3602 base[i] = reserve_memory(bytes);
3603
3604 if (base[i] != NULL) {
3605 // Is this the block we wanted?
3606 if (base[i] == requested_addr) {
3607 size[i] = bytes;
3608 break;
3609 }
3610
3611 // Does this overlap the block we wanted? Give back the overlapped
3612 // parts and try again.
3613
3614 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3615 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3616 unmap_memory(base[i], top_overlap);
3617 base[i] += top_overlap;
3618 size[i] = bytes - top_overlap;
3619 } else {
3620 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3621 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3622 unmap_memory(requested_addr, bottom_overlap);
3623 size[i] = bytes - bottom_overlap;
3624 } else {
3625 size[i] = bytes;
3626 }
3627 }
3628 }
3629 }
3630
3631 // Give back the unused reserved pieces.
3632
3633 for (int j = 0; j < i; ++j) {
3634 if (base[j] != NULL) {
3635 unmap_memory(base[j], size[j]);
3636 }
3637 }
3638
3639 if (i < max_tries) {
3640 return requested_addr;
3641 } else {
3642 return NULL;
3643 }
3644 }
3645
3646 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3647 return ::read(fd, buf, nBytes);
3648 }
3649
3650 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3651 return ::pread(fd, buf, nBytes, offset);
3652 }
3653
3654 // Short sleep, direct OS call.
3655 //
3656 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3657 // sched_yield(2) will actually give up the CPU:
3658 //
3659 // * Alone on this pariticular CPU, keeps running.
3660 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3661 // (pre 2.6.39).
3662 //
3663 // So calling this with 0 is an alternative.
3664 //
3665 void os::naked_short_sleep(jlong ms) {
3666 struct timespec req;
3667
3668 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3669 req.tv_sec = 0;
3670 if (ms > 0) {
3671 req.tv_nsec = (ms % 1000) * 1000000;
3672 } else {
3673 req.tv_nsec = 1;
3674 }
3675
3676 nanosleep(&req, NULL);
3677
3678 return;
3679 }
3680
3681 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3682 void os::infinite_sleep() {
3683 while (true) { // sleep forever ...
3684 ::sleep(100); // ... 100 seconds at a time
3685 }
3686 }
3687
3688 // Used to convert frequent JVM_Yield() to nops
3689 bool os::dont_yield() {
3690 return DontYieldALot;
3691 }
3692
3693 void os::naked_yield() {
3694 sched_yield();
3695 }
3696
3697 ////////////////////////////////////////////////////////////////////////////////
3698 // thread priority support
3699
3700 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3701 // only supports dynamic priority, static priority must be zero. For real-time
3702 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3703 // However, for large multi-threaded applications, SCHED_RR is not only slower
3704 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3705 // of 5 runs - Sep 2005).
3706 //
3707 // The following code actually changes the niceness of kernel-thread/LWP. It
3708 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3709 // not the entire user process, and user level threads are 1:1 mapped to kernel
3710 // threads. It has always been the case, but could change in the future. For
3711 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3712 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3713
3714 int os::java_to_os_priority[CriticalPriority + 1] = {
3715 19, // 0 Entry should never be used
3716
3717 4, // 1 MinPriority
3718 3, // 2
3719 2, // 3
3720
3721 1, // 4
3722 0, // 5 NormPriority
3723 -1, // 6
3724
3725 -2, // 7
3726 -3, // 8
3727 -4, // 9 NearMaxPriority
3728
3729 -5, // 10 MaxPriority
3730
3731 -5 // 11 CriticalPriority
3732 };
3733
3734 static int prio_init() {
3735 if (ThreadPriorityPolicy == 1) {
3736 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3737 // if effective uid is not root. Perhaps, a more elegant way of doing
3738 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3739 if (geteuid() != 0) {
3740 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3741 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3742 }
3743 ThreadPriorityPolicy = 0;
3744 }
3745 }
3746 if (UseCriticalJavaThreadPriority) {
3747 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3748 }
3749 return 0;
3750 }
3751
3752 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3753 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3754
3755 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3756 return (ret == 0) ? OS_OK : OS_ERR;
3757 }
3758
3759 OSReturn os::get_native_priority(const Thread* const thread,
3760 int *priority_ptr) {
3761 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3762 *priority_ptr = java_to_os_priority[NormPriority];
3763 return OS_OK;
3764 }
3765
3766 errno = 0;
3767 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3768 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3769 }
3770
3771 // Hint to the underlying OS that a task switch would not be good.
3772 // Void return because it's a hint and can fail.
3773 void os::hint_no_preempt() {}
3774
3775 ////////////////////////////////////////////////////////////////////////////////
3776 // suspend/resume support
3777
3778 // the low-level signal-based suspend/resume support is a remnant from the
3779 // old VM-suspension that used to be for java-suspension, safepoints etc,
3780 // within hotspot. Now there is a single use-case for this:
3781 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3782 // that runs in the watcher thread.
3783 // The remaining code is greatly simplified from the more general suspension
3784 // code that used to be used.
3785 //
3786 // The protocol is quite simple:
3787 // - suspend:
3788 // - sends a signal to the target thread
3789 // - polls the suspend state of the osthread using a yield loop
3790 // - target thread signal handler (SR_handler) sets suspend state
3791 // and blocks in sigsuspend until continued
3792 // - resume:
3793 // - sets target osthread state to continue
3794 // - sends signal to end the sigsuspend loop in the SR_handler
3795 //
3796 // Note that the SR_lock plays no role in this suspend/resume protocol.
3797
3798 static void resume_clear_context(OSThread *osthread) {
3799 osthread->set_ucontext(NULL);
3800 osthread->set_siginfo(NULL);
3801 }
3802
3803 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3804 ucontext_t* context) {
3805 osthread->set_ucontext(context);
3806 osthread->set_siginfo(siginfo);
3807 }
3808
3809 // Handler function invoked when a thread's execution is suspended or
3810 // resumed. We have to be careful that only async-safe functions are
3811 // called here (Note: most pthread functions are not async safe and
3812 // should be avoided.)
3813 //
3814 // Note: sigwait() is a more natural fit than sigsuspend() from an
3815 // interface point of view, but sigwait() prevents the signal hander
3816 // from being run. libpthread would get very confused by not having
3817 // its signal handlers run and prevents sigwait()'s use with the
3818 // mutex granting granting signal.
3819 //
3820 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3821 //
3822 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3823 // Save and restore errno to avoid confusing native code with EINTR
3824 // after sigsuspend.
3825 int old_errno = errno;
3826
3827 Thread* thread = Thread::current();
3828 OSThread* osthread = thread->osthread();
3829 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3830
3831 os::SuspendResume::State current = osthread->sr.state();
3832 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3833 suspend_save_context(osthread, siginfo, context);
3834
3835 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3836 os::SuspendResume::State state = osthread->sr.suspended();
3837 if (state == os::SuspendResume::SR_SUSPENDED) {
3838 sigset_t suspend_set; // signals for sigsuspend()
3839
3840 // get current set of blocked signals and unblock resume signal
3841 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3842 sigdelset(&suspend_set, SR_signum);
3843
3844 sr_semaphore.signal();
3845 // wait here until we are resumed
3846 while (1) {
3847 sigsuspend(&suspend_set);
3848
3849 os::SuspendResume::State result = osthread->sr.running();
3850 if (result == os::SuspendResume::SR_RUNNING) {
3851 sr_semaphore.signal();
3852 break;
3853 }
3854 }
3855
3856 } else if (state == os::SuspendResume::SR_RUNNING) {
3857 // request was cancelled, continue
3858 } else {
3859 ShouldNotReachHere();
3860 }
3861
3862 resume_clear_context(osthread);
3863 } else if (current == os::SuspendResume::SR_RUNNING) {
3864 // request was cancelled, continue
3865 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3866 // ignore
3867 } else {
3868 // ignore
3869 }
3870
3871 errno = old_errno;
3872 }
3873
3874
3875 static int SR_initialize() {
3876 struct sigaction act;
3877 char *s;
3878 // Get signal number to use for suspend/resume
3879 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3880 int sig = ::strtol(s, 0, 10);
3881 if (sig > 0 || sig < _NSIG) {
3882 SR_signum = sig;
3883 }
3884 }
3885
3886 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3887 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3888
3889 sigemptyset(&SR_sigset);
3890 sigaddset(&SR_sigset, SR_signum);
3891
3892 // Set up signal handler for suspend/resume
3893 act.sa_flags = SA_RESTART|SA_SIGINFO;
3894 act.sa_handler = (void (*)(int)) SR_handler;
3895
3896 // SR_signum is blocked by default.
3897 // 4528190 - We also need to block pthread restart signal (32 on all
3898 // supported Linux platforms). Note that LinuxThreads need to block
3899 // this signal for all threads to work properly. So we don't have
3900 // to use hard-coded signal number when setting up the mask.
3901 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3902
3903 if (sigaction(SR_signum, &act, 0) == -1) {
3904 return -1;
3905 }
3906
3907 // Save signal flag
3908 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3909 return 0;
3910 }
3911
3912 static int sr_notify(OSThread* osthread) {
3913 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3914 assert_status(status == 0, status, "pthread_kill");
3915 return status;
3916 }
3917
3918 // "Randomly" selected value for how long we want to spin
3919 // before bailing out on suspending a thread, also how often
3920 // we send a signal to a thread we want to resume
3921 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3922 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3923
3924 // returns true on success and false on error - really an error is fatal
3925 // but this seems the normal response to library errors
3926 static bool do_suspend(OSThread* osthread) {
3927 assert(osthread->sr.is_running(), "thread should be running");
3928 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3929
3930 // mark as suspended and send signal
3931 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3932 // failed to switch, state wasn't running?
3933 ShouldNotReachHere();
3934 return false;
3935 }
3936
3937 if (sr_notify(osthread) != 0) {
3938 ShouldNotReachHere();
3939 }
3940
3941 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3942 while (true) {
3943 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3944 break;
3945 } else {
3946 // timeout
3947 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3948 if (cancelled == os::SuspendResume::SR_RUNNING) {
3949 return false;
3950 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3951 // make sure that we consume the signal on the semaphore as well
3952 sr_semaphore.wait();
3953 break;
3954 } else {
3955 ShouldNotReachHere();
3956 return false;
3957 }
3958 }
3959 }
3960
3961 guarantee(osthread->sr.is_suspended(), "Must be suspended");
3962 return true;
3963 }
3964
3965 static void do_resume(OSThread* osthread) {
3966 assert(osthread->sr.is_suspended(), "thread should be suspended");
3967 assert(!sr_semaphore.trywait(), "invalid semaphore state");
3968
3969 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3970 // failed to switch to WAKEUP_REQUEST
3971 ShouldNotReachHere();
3972 return;
3973 }
3974
3975 while (true) {
3976 if (sr_notify(osthread) == 0) {
3977 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3978 if (osthread->sr.is_running()) {
3979 return;
3980 }
3981 }
3982 } else {
3983 ShouldNotReachHere();
3984 }
3985 }
3986
3987 guarantee(osthread->sr.is_running(), "Must be running!");
3988 }
3989
3990 ///////////////////////////////////////////////////////////////////////////////////
3991 // signal handling (except suspend/resume)
3992
3993 // This routine may be used by user applications as a "hook" to catch signals.
3994 // The user-defined signal handler must pass unrecognized signals to this
3995 // routine, and if it returns true (non-zero), then the signal handler must
3996 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3997 // routine will never retun false (zero), but instead will execute a VM panic
3998 // routine kill the process.
3999 //
4000 // If this routine returns false, it is OK to call it again. This allows
4001 // the user-defined signal handler to perform checks either before or after
4002 // the VM performs its own checks. Naturally, the user code would be making
4003 // a serious error if it tried to handle an exception (such as a null check
4004 // or breakpoint) that the VM was generating for its own correct operation.
4005 //
4006 // This routine may recognize any of the following kinds of signals:
4007 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4008 // It should be consulted by handlers for any of those signals.
4009 //
4010 // The caller of this routine must pass in the three arguments supplied
4011 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4012 // field of the structure passed to sigaction(). This routine assumes that
4013 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4014 //
4015 // Note that the VM will print warnings if it detects conflicting signal
4016 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4017 //
4018 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4019 siginfo_t* siginfo,
4020 void* ucontext,
4021 int abort_if_unrecognized);
4022
4023 void signalHandler(int sig, siginfo_t* info, void* uc) {
4024 assert(info != NULL && uc != NULL, "it must be old kernel");
4025 int orig_errno = errno; // Preserve errno value over signal handler.
4026 JVM_handle_linux_signal(sig, info, uc, true);
4027 errno = orig_errno;
4028 }
4029
4030
4031 // This boolean allows users to forward their own non-matching signals
4032 // to JVM_handle_linux_signal, harmlessly.
4033 bool os::Linux::signal_handlers_are_installed = false;
4034
4035 // For signal-chaining
4036 struct sigaction os::Linux::sigact[MAXSIGNUM];
4037 unsigned int os::Linux::sigs = 0;
4038 bool os::Linux::libjsig_is_loaded = false;
4039 typedef struct sigaction *(*get_signal_t)(int);
4040 get_signal_t os::Linux::get_signal_action = NULL;
4041
4042 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4043 struct sigaction *actp = NULL;
4044
4045 if (libjsig_is_loaded) {
4046 // Retrieve the old signal handler from libjsig
4047 actp = (*get_signal_action)(sig);
4048 }
4049 if (actp == NULL) {
4050 // Retrieve the preinstalled signal handler from jvm
4051 actp = get_preinstalled_handler(sig);
4052 }
4053
4054 return actp;
4055 }
4056
4057 static bool call_chained_handler(struct sigaction *actp, int sig,
4058 siginfo_t *siginfo, void *context) {
4059 // Call the old signal handler
4060 if (actp->sa_handler == SIG_DFL) {
4061 // It's more reasonable to let jvm treat it as an unexpected exception
4062 // instead of taking the default action.
4063 return false;
4064 } else if (actp->sa_handler != SIG_IGN) {
4065 if ((actp->sa_flags & SA_NODEFER) == 0) {
4066 // automaticlly block the signal
4067 sigaddset(&(actp->sa_mask), sig);
4068 }
4069
4070 sa_handler_t hand;
4071 sa_sigaction_t sa;
4072 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4073 // retrieve the chained handler
4074 if (siginfo_flag_set) {
4075 sa = actp->sa_sigaction;
4076 } else {
4077 hand = actp->sa_handler;
4078 }
4079
4080 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4081 actp->sa_handler = SIG_DFL;
4082 }
4083
4084 // try to honor the signal mask
4085 sigset_t oset;
4086 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4087
4088 // call into the chained handler
4089 if (siginfo_flag_set) {
4090 (*sa)(sig, siginfo, context);
4091 } else {
4092 (*hand)(sig);
4093 }
4094
4095 // restore the signal mask
4096 pthread_sigmask(SIG_SETMASK, &oset, 0);
4097 }
4098 // Tell jvm's signal handler the signal is taken care of.
4099 return true;
4100 }
4101
4102 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4103 bool chained = false;
4104 // signal-chaining
4105 if (UseSignalChaining) {
4106 struct sigaction *actp = get_chained_signal_action(sig);
4107 if (actp != NULL) {
4108 chained = call_chained_handler(actp, sig, siginfo, context);
4109 }
4110 }
4111 return chained;
4112 }
4113
4114 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4115 if ((((unsigned int)1 << sig) & sigs) != 0) {
4116 return &sigact[sig];
4117 }
4118 return NULL;
4119 }
4120
4121 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4122 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4123 sigact[sig] = oldAct;
4124 sigs |= (unsigned int)1 << sig;
4125 }
4126
4127 // for diagnostic
4128 int os::Linux::sigflags[MAXSIGNUM];
4129
4130 int os::Linux::get_our_sigflags(int sig) {
4131 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4132 return sigflags[sig];
4133 }
4134
4135 void os::Linux::set_our_sigflags(int sig, int flags) {
4136 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4137 sigflags[sig] = flags;
4138 }
4139
4140 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4141 // Check for overwrite.
4142 struct sigaction oldAct;
4143 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4144
4145 void* oldhand = oldAct.sa_sigaction
4146 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4147 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4148 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4149 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4150 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4151 if (AllowUserSignalHandlers || !set_installed) {
4152 // Do not overwrite; user takes responsibility to forward to us.
4153 return;
4154 } else if (UseSignalChaining) {
4155 // save the old handler in jvm
4156 save_preinstalled_handler(sig, oldAct);
4157 // libjsig also interposes the sigaction() call below and saves the
4158 // old sigaction on it own.
4159 } else {
4160 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4161 "%#lx for signal %d.", (long)oldhand, sig));
4162 }
4163 }
4164
4165 struct sigaction sigAct;
4166 sigfillset(&(sigAct.sa_mask));
4167 sigAct.sa_handler = SIG_DFL;
4168 if (!set_installed) {
4169 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4170 } else {
4171 sigAct.sa_sigaction = signalHandler;
4172 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4173 }
4174 // Save flags, which are set by ours
4175 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4176 sigflags[sig] = sigAct.sa_flags;
4177
4178 int ret = sigaction(sig, &sigAct, &oldAct);
4179 assert(ret == 0, "check");
4180
4181 void* oldhand2 = oldAct.sa_sigaction
4182 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4183 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4184 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4185 }
4186
4187 // install signal handlers for signals that HotSpot needs to
4188 // handle in order to support Java-level exception handling.
4189
4190 void os::Linux::install_signal_handlers() {
4191 if (!signal_handlers_are_installed) {
4192 signal_handlers_are_installed = true;
4193
4194 // signal-chaining
4195 typedef void (*signal_setting_t)();
4196 signal_setting_t begin_signal_setting = NULL;
4197 signal_setting_t end_signal_setting = NULL;
4198 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4199 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4200 if (begin_signal_setting != NULL) {
4201 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4202 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4203 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4204 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4205 libjsig_is_loaded = true;
4206 assert(UseSignalChaining, "should enable signal-chaining");
4207 }
4208 if (libjsig_is_loaded) {
4209 // Tell libjsig jvm is setting signal handlers
4210 (*begin_signal_setting)();
4211 }
4212
4213 set_signal_handler(SIGSEGV, true);
4214 set_signal_handler(SIGPIPE, true);
4215 set_signal_handler(SIGBUS, true);
4216 set_signal_handler(SIGILL, true);
4217 set_signal_handler(SIGFPE, true);
4218 #if defined(PPC64)
4219 set_signal_handler(SIGTRAP, true);
4220 #endif
4221 set_signal_handler(SIGXFSZ, true);
4222
4223 if (libjsig_is_loaded) {
4224 // Tell libjsig jvm finishes setting signal handlers
4225 (*end_signal_setting)();
4226 }
4227
4228 // We don't activate signal checker if libjsig is in place, we trust ourselves
4229 // and if UserSignalHandler is installed all bets are off.
4230 // Log that signal checking is off only if -verbose:jni is specified.
4231 if (CheckJNICalls) {
4232 if (libjsig_is_loaded) {
4233 if (PrintJNIResolving) {
4234 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4235 }
4236 check_signals = false;
4237 }
4238 if (AllowUserSignalHandlers) {
4239 if (PrintJNIResolving) {
4240 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4241 }
4242 check_signals = false;
4243 }
4244 }
4245 }
4246 }
4247
4248 // This is the fastest way to get thread cpu time on Linux.
4249 // Returns cpu time (user+sys) for any thread, not only for current.
4250 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4251 // It might work on 2.6.10+ with a special kernel/glibc patch.
4252 // For reference, please, see IEEE Std 1003.1-2004:
4253 // http://www.unix.org/single_unix_specification
4254
4255 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4256 struct timespec tp;
4257 int rc = os::Linux::clock_gettime(clockid, &tp);
4258 assert(rc == 0, "clock_gettime is expected to return 0 code");
4259
4260 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4261 }
4262
4263 /////
4264 // glibc on Linux platform uses non-documented flag
4265 // to indicate, that some special sort of signal
4266 // trampoline is used.
4267 // We will never set this flag, and we should
4268 // ignore this flag in our diagnostic
4269 #ifdef SIGNIFICANT_SIGNAL_MASK
4270 #undef SIGNIFICANT_SIGNAL_MASK
4271 #endif
4272 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4273
4274 static const char* get_signal_handler_name(address handler,
4275 char* buf, int buflen) {
4276 int offset;
4277 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4278 if (found) {
4279 // skip directory names
4280 const char *p1, *p2;
4281 p1 = buf;
4282 size_t len = strlen(os::file_separator());
4283 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4284 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4285 } else {
4286 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4287 }
4288 return buf;
4289 }
4290
4291 static void print_signal_handler(outputStream* st, int sig,
4292 char* buf, size_t buflen) {
4293 struct sigaction sa;
4294
4295 sigaction(sig, NULL, &sa);
4296
4297 // See comment for SIGNIFICANT_SIGNAL_MASK define
4298 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4299
4300 st->print("%s: ", os::exception_name(sig, buf, buflen));
4301
4302 address handler = (sa.sa_flags & SA_SIGINFO)
4303 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4304 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4305
4306 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4307 st->print("SIG_DFL");
4308 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4309 st->print("SIG_IGN");
4310 } else {
4311 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4312 }
4313
4314 st->print(", sa_mask[0]=");
4315 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4316
4317 address rh = VMError::get_resetted_sighandler(sig);
4318 // May be, handler was resetted by VMError?
4319 if (rh != NULL) {
4320 handler = rh;
4321 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4322 }
4323
4324 st->print(", sa_flags=");
4325 os::Posix::print_sa_flags(st, sa.sa_flags);
4326
4327 // Check: is it our handler?
4328 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4329 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4330 // It is our signal handler
4331 // check for flags, reset system-used one!
4332 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4333 st->print(
4334 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4335 os::Linux::get_our_sigflags(sig));
4336 }
4337 }
4338 st->cr();
4339 }
4340
4341
4342 #define DO_SIGNAL_CHECK(sig) \
4343 do { \
4344 if (!sigismember(&check_signal_done, sig)) { \
4345 os::Linux::check_signal_handler(sig); \
4346 } \
4347 } while (0)
4348
4349 // This method is a periodic task to check for misbehaving JNI applications
4350 // under CheckJNI, we can add any periodic checks here
4351
4352 void os::run_periodic_checks() {
4353 if (check_signals == false) return;
4354
4355 // SEGV and BUS if overridden could potentially prevent
4356 // generation of hs*.log in the event of a crash, debugging
4357 // such a case can be very challenging, so we absolutely
4358 // check the following for a good measure:
4359 DO_SIGNAL_CHECK(SIGSEGV);
4360 DO_SIGNAL_CHECK(SIGILL);
4361 DO_SIGNAL_CHECK(SIGFPE);
4362 DO_SIGNAL_CHECK(SIGBUS);
4363 DO_SIGNAL_CHECK(SIGPIPE);
4364 DO_SIGNAL_CHECK(SIGXFSZ);
4365 #if defined(PPC64)
4366 DO_SIGNAL_CHECK(SIGTRAP);
4367 #endif
4368
4369 // ReduceSignalUsage allows the user to override these handlers
4370 // see comments at the very top and jvm_solaris.h
4371 if (!ReduceSignalUsage) {
4372 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4373 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4374 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4375 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4376 }
4377
4378 DO_SIGNAL_CHECK(SR_signum);
4379 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4380 }
4381
4382 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4383
4384 static os_sigaction_t os_sigaction = NULL;
4385
4386 void os::Linux::check_signal_handler(int sig) {
4387 char buf[O_BUFLEN];
4388 address jvmHandler = NULL;
4389
4390
4391 struct sigaction act;
4392 if (os_sigaction == NULL) {
4393 // only trust the default sigaction, in case it has been interposed
4394 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4395 if (os_sigaction == NULL) return;
4396 }
4397
4398 os_sigaction(sig, (struct sigaction*)NULL, &act);
4399
4400
4401 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4402
4403 address thisHandler = (act.sa_flags & SA_SIGINFO)
4404 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4405 : CAST_FROM_FN_PTR(address, act.sa_handler);
4406
4407
4408 switch (sig) {
4409 case SIGSEGV:
4410 case SIGBUS:
4411 case SIGFPE:
4412 case SIGPIPE:
4413 case SIGILL:
4414 case SIGXFSZ:
4415 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4416 break;
4417
4418 case SHUTDOWN1_SIGNAL:
4419 case SHUTDOWN2_SIGNAL:
4420 case SHUTDOWN3_SIGNAL:
4421 case BREAK_SIGNAL:
4422 jvmHandler = (address)user_handler();
4423 break;
4424
4425 case INTERRUPT_SIGNAL:
4426 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4427 break;
4428
4429 default:
4430 if (sig == SR_signum) {
4431 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4432 } else {
4433 return;
4434 }
4435 break;
4436 }
4437
4438 if (thisHandler != jvmHandler) {
4439 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4440 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4441 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4442 // No need to check this sig any longer
4443 sigaddset(&check_signal_done, sig);
4444 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4445 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4446 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4447 exception_name(sig, buf, O_BUFLEN));
4448 }
4449 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4450 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4451 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4452 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4453 // No need to check this sig any longer
4454 sigaddset(&check_signal_done, sig);
4455 }
4456
4457 // Dump all the signal
4458 if (sigismember(&check_signal_done, sig)) {
4459 print_signal_handlers(tty, buf, O_BUFLEN);
4460 }
4461 }
4462
4463 extern void report_error(char* file_name, int line_no, char* title,
4464 char* format, ...);
4465
4466 extern bool signal_name(int signo, char* buf, size_t len);
4467
4468 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4469 if (0 < exception_code && exception_code <= SIGRTMAX) {
4470 // signal
4471 if (!signal_name(exception_code, buf, size)) {
4472 jio_snprintf(buf, size, "SIG%d", exception_code);
4473 }
4474 return buf;
4475 } else {
4476 return NULL;
4477 }
4478 }
4479
4480 // this is called _before_ the most of global arguments have been parsed
4481 void os::init(void) {
4482 char dummy; // used to get a guess on initial stack address
4483 // first_hrtime = gethrtime();
4484
4485 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4486
4487 init_random(1234567);
4488
4489 ThreadCritical::initialize();
4490
4491 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4492 if (Linux::page_size() == -1) {
4493 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4494 strerror(errno)));
4495 }
4496 init_page_sizes((size_t) Linux::page_size());
4497
4498 Linux::initialize_system_info();
4499
4500 // main_thread points to the aboriginal thread
4501 Linux::_main_thread = pthread_self();
4502
4503 Linux::clock_init();
4504 initial_time_count = javaTimeNanos();
4505
4506 // pthread_condattr initialization for monotonic clock
4507 int status;
4508 pthread_condattr_t* _condattr = os::Linux::condAttr();
4509 if ((status = pthread_condattr_init(_condattr)) != 0) {
4510 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4511 }
4512 // Only set the clock if CLOCK_MONOTONIC is available
4513 if (os::supports_monotonic_clock()) {
4514 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4515 if (status == EINVAL) {
4516 warning("Unable to use monotonic clock with relative timed-waits" \
4517 " - changes to the time-of-day clock may have adverse affects");
4518 } else {
4519 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4520 }
4521 }
4522 }
4523 // else it defaults to CLOCK_REALTIME
4524
4525 // If the pagesize of the VM is greater than 8K determine the appropriate
4526 // number of initial guard pages. The user can change this with the
4527 // command line arguments, if needed.
4528 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4529 StackYellowPages = 1;
4530 StackRedPages = 1;
4531 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4532 }
4533
4534 // retrieve entry point for pthread_setname_np
4535 Linux::_pthread_setname_np =
4536 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4537
4538 }
4539
4540 // To install functions for atexit system call
4541 extern "C" {
4542 static void perfMemory_exit_helper() {
4543 perfMemory_exit();
4544 }
4545 }
4546
4547 // this is called _after_ the global arguments have been parsed
4548 jint os::init_2(void) {
4549 Linux::fast_thread_clock_init();
4550
4551 // Allocate a single page and mark it as readable for safepoint polling
4552 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4553 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4554
4555 os::set_polling_page(polling_page);
4556
4557 #ifndef PRODUCT
4558 if (Verbose && PrintMiscellaneous) {
4559 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4560 (intptr_t)polling_page);
4561 }
4562 #endif
4563
4564 if (!UseMembar) {
4565 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4566 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4567 os::set_memory_serialize_page(mem_serialize_page);
4568
4569 #ifndef PRODUCT
4570 if (Verbose && PrintMiscellaneous) {
4571 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4572 (intptr_t)mem_serialize_page);
4573 }
4574 #endif
4575 }
4576
4577 // initialize suspend/resume support - must do this before signal_sets_init()
4578 if (SR_initialize() != 0) {
4579 perror("SR_initialize failed");
4580 return JNI_ERR;
4581 }
4582
4583 Linux::signal_sets_init();
4584 Linux::install_signal_handlers();
4585
4586 // Check minimum allowable stack size for thread creation and to initialize
4587 // the java system classes, including StackOverflowError - depends on page
4588 // size. Add a page for compiler2 recursion in main thread.
4589 // Add in 2*BytesPerWord times page size to account for VM stack during
4590 // class initialization depending on 32 or 64 bit VM.
4591 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4592 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4593 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4594
4595 size_t threadStackSizeInBytes = ThreadStackSize * K;
4596 if (threadStackSizeInBytes != 0 &&
4597 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4598 tty->print_cr("\nThe stack size specified is too small, "
4599 "Specify at least %dk",
4600 os::Linux::min_stack_allowed/ K);
4601 return JNI_ERR;
4602 }
4603
4604 // Make the stack size a multiple of the page size so that
4605 // the yellow/red zones can be guarded.
4606 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4607 vm_page_size()));
4608
4609 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4610
4611 #if defined(IA32)
4612 workaround_expand_exec_shield_cs_limit();
4613 #endif
4614
4615 Linux::libpthread_init();
4616 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4617 tty->print_cr("[HotSpot is running with %s, %s]\n",
4618 Linux::glibc_version(), Linux::libpthread_version());
4619 }
4620
4621 if (UseNUMA) {
4622 if (!Linux::libnuma_init()) {
4623 UseNUMA = false;
4624 } else {
4625 if ((Linux::numa_max_node() < 1)) {
4626 // There's only one node(they start from 0), disable NUMA.
4627 UseNUMA = false;
4628 }
4629 }
4630 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4631 // we can make the adaptive lgrp chunk resizing work. If the user specified
4632 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4633 // disable adaptive resizing.
4634 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4635 if (FLAG_IS_DEFAULT(UseNUMA)) {
4636 UseNUMA = false;
4637 } else {
4638 if (FLAG_IS_DEFAULT(UseLargePages) &&
4639 FLAG_IS_DEFAULT(UseSHM) &&
4640 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4641 UseLargePages = false;
4642 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4643 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4644 UseAdaptiveSizePolicy = false;
4645 UseAdaptiveNUMAChunkSizing = false;
4646 }
4647 }
4648 }
4649 if (!UseNUMA && ForceNUMA) {
4650 UseNUMA = true;
4651 }
4652 }
4653
4654 if (MaxFDLimit) {
4655 // set the number of file descriptors to max. print out error
4656 // if getrlimit/setrlimit fails but continue regardless.
4657 struct rlimit nbr_files;
4658 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4659 if (status != 0) {
4660 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4661 perror("os::init_2 getrlimit failed");
4662 }
4663 } else {
4664 nbr_files.rlim_cur = nbr_files.rlim_max;
4665 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4666 if (status != 0) {
4667 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4668 perror("os::init_2 setrlimit failed");
4669 }
4670 }
4671 }
4672 }
4673
4674 // Initialize lock used to serialize thread creation (see os::create_thread)
4675 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4676
4677 // at-exit methods are called in the reverse order of their registration.
4678 // atexit functions are called on return from main or as a result of a
4679 // call to exit(3C). There can be only 32 of these functions registered
4680 // and atexit() does not set errno.
4681
4682 if (PerfAllowAtExitRegistration) {
4683 // only register atexit functions if PerfAllowAtExitRegistration is set.
4684 // atexit functions can be delayed until process exit time, which
4685 // can be problematic for embedded VM situations. Embedded VMs should
4686 // call DestroyJavaVM() to assure that VM resources are released.
4687
4688 // note: perfMemory_exit_helper atexit function may be removed in
4689 // the future if the appropriate cleanup code can be added to the
4690 // VM_Exit VMOperation's doit method.
4691 if (atexit(perfMemory_exit_helper) != 0) {
4692 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4693 }
4694 }
4695
4696 // initialize thread priority policy
4697 prio_init();
4698
4699 return JNI_OK;
4700 }
4701
4702 // Mark the polling page as unreadable
4703 void os::make_polling_page_unreadable(void) {
4704 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4705 fatal("Could not disable polling page");
4706 }
4707 }
4708
4709 // Mark the polling page as readable
4710 void os::make_polling_page_readable(void) {
4711 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4712 fatal("Could not enable polling page");
4713 }
4714 }
4715
4716 int os::active_processor_count() {
4717 // Linux doesn't yet have a (official) notion of processor sets,
4718 // so just return the number of online processors.
4719 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4720 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4721 return online_cpus;
4722 }
4723
4724 void os::set_native_thread_name(const char *name) {
4725 if (Linux::_pthread_setname_np) {
4726 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4727 snprintf(buf, sizeof(buf), "%s", name);
4728 buf[sizeof(buf) - 1] = '\0';
4729 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4730 // ERANGE should not happen; all other errors should just be ignored.
4731 assert(rc != ERANGE, "pthread_setname_np failed");
4732 }
4733 }
4734
4735 bool os::distribute_processes(uint length, uint* distribution) {
4736 // Not yet implemented.
4737 return false;
4738 }
4739
4740 bool os::bind_to_processor(uint processor_id) {
4741 // Not yet implemented.
4742 return false;
4743 }
4744
4745 ///
4746
4747 void os::SuspendedThreadTask::internal_do_task() {
4748 if (do_suspend(_thread->osthread())) {
4749 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4750 do_task(context);
4751 do_resume(_thread->osthread());
4752 }
4753 }
4754
4755 class PcFetcher : public os::SuspendedThreadTask {
4756 public:
4757 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4758 ExtendedPC result();
4759 protected:
4760 void do_task(const os::SuspendedThreadTaskContext& context);
4761 private:
4762 ExtendedPC _epc;
4763 };
4764
4765 ExtendedPC PcFetcher::result() {
4766 guarantee(is_done(), "task is not done yet.");
4767 return _epc;
4768 }
4769
4770 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4771 Thread* thread = context.thread();
4772 OSThread* osthread = thread->osthread();
4773 if (osthread->ucontext() != NULL) {
4774 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4775 } else {
4776 // NULL context is unexpected, double-check this is the VMThread
4777 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4778 }
4779 }
4780
4781 // Suspends the target using the signal mechanism and then grabs the PC before
4782 // resuming the target. Used by the flat-profiler only
4783 ExtendedPC os::get_thread_pc(Thread* thread) {
4784 // Make sure that it is called by the watcher for the VMThread
4785 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4786 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4787
4788 PcFetcher fetcher(thread);
4789 fetcher.run();
4790 return fetcher.result();
4791 }
4792
4793 ////////////////////////////////////////////////////////////////////////////////
4794 // debug support
4795
4796 bool os::find(address addr, outputStream* st) {
4797 Dl_info dlinfo;
4798 memset(&dlinfo, 0, sizeof(dlinfo));
4799 if (dladdr(addr, &dlinfo) != 0) {
4800 st->print(PTR_FORMAT ": ", addr);
4801 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4802 st->print("%s+%#x", dlinfo.dli_sname,
4803 addr - (intptr_t)dlinfo.dli_saddr);
4804 } else if (dlinfo.dli_fbase != NULL) {
4805 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4806 } else {
4807 st->print("<absolute address>");
4808 }
4809 if (dlinfo.dli_fname != NULL) {
4810 st->print(" in %s", dlinfo.dli_fname);
4811 }
4812 if (dlinfo.dli_fbase != NULL) {
4813 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4814 }
4815 st->cr();
4816
4817 if (Verbose) {
4818 // decode some bytes around the PC
4819 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4820 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4821 address lowest = (address) dlinfo.dli_sname;
4822 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4823 if (begin < lowest) begin = lowest;
4824 Dl_info dlinfo2;
4825 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4826 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4827 end = (address) dlinfo2.dli_saddr;
4828 }
4829 Disassembler::decode(begin, end, st);
4830 }
4831 return true;
4832 }
4833 return false;
4834 }
4835
4836 ////////////////////////////////////////////////////////////////////////////////
4837 // misc
4838
4839 // This does not do anything on Linux. This is basically a hook for being
4840 // able to use structured exception handling (thread-local exception filters)
4841 // on, e.g., Win32.
4842 void
4843 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4844 JavaCallArguments* args, Thread* thread) {
4845 f(value, method, args, thread);
4846 }
4847
4848 void os::print_statistics() {
4849 }
4850
4851 int os::message_box(const char* title, const char* message) {
4852 int i;
4853 fdStream err(defaultStream::error_fd());
4854 for (i = 0; i < 78; i++) err.print_raw("=");
4855 err.cr();
4856 err.print_raw_cr(title);
4857 for (i = 0; i < 78; i++) err.print_raw("-");
4858 err.cr();
4859 err.print_raw_cr(message);
4860 for (i = 0; i < 78; i++) err.print_raw("=");
4861 err.cr();
4862
4863 char buf[16];
4864 // Prevent process from exiting upon "read error" without consuming all CPU
4865 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4866
4867 return buf[0] == 'y' || buf[0] == 'Y';
4868 }
4869
4870 int os::stat(const char *path, struct stat *sbuf) {
4871 char pathbuf[MAX_PATH];
4872 if (strlen(path) > MAX_PATH - 1) {
4873 errno = ENAMETOOLONG;
4874 return -1;
4875 }
4876 os::native_path(strcpy(pathbuf, path));
4877 return ::stat(pathbuf, sbuf);
4878 }
4879
4880 bool os::check_heap(bool force) {
4881 return true;
4882 }
4883
4884 // Is a (classpath) directory empty?
4885 bool os::dir_is_empty(const char* path) {
4886 DIR *dir = NULL;
4887 struct dirent *ptr;
4888
4889 dir = opendir(path);
4890 if (dir == NULL) return true;
4891
4892 // Scan the directory
4893 bool result = true;
4894 char buf[sizeof(struct dirent) + MAX_PATH];
4895 while (result && (ptr = ::readdir(dir)) != NULL) {
4896 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4897 result = false;
4898 }
4899 }
4900 closedir(dir);
4901 return result;
4902 }
4903
4904 // This code originates from JDK's sysOpen and open64_w
4905 // from src/solaris/hpi/src/system_md.c
4906
4907 int os::open(const char *path, int oflag, int mode) {
4908 if (strlen(path) > MAX_PATH - 1) {
4909 errno = ENAMETOOLONG;
4910 return -1;
4911 }
4912
4913 // All file descriptors that are opened in the Java process and not
4914 // specifically destined for a subprocess should have the close-on-exec
4915 // flag set. If we don't set it, then careless 3rd party native code
4916 // might fork and exec without closing all appropriate file descriptors
4917 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
4918 // turn might:
4919 //
4920 // - cause end-of-file to fail to be detected on some file
4921 // descriptors, resulting in mysterious hangs, or
4922 //
4923 // - might cause an fopen in the subprocess to fail on a system
4924 // suffering from bug 1085341.
4925 //
4926 // (Yes, the default setting of the close-on-exec flag is a Unix
4927 // design flaw)
4928 //
4929 // See:
4930 // 1085341: 32-bit stdio routines should support file descriptors >255
4931 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4932 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4933 //
4934 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
4935 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
4936 // because it saves a system call and removes a small window where the flag
4937 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
4938 // and we fall back to using FD_CLOEXEC (see below).
4939 #ifdef O_CLOEXEC
4940 oflag |= O_CLOEXEC;
4941 #endif
4942
4943 int fd = ::open64(path, oflag, mode);
4944 if (fd == -1) return -1;
4945
4946 //If the open succeeded, the file might still be a directory
4947 {
4948 struct stat64 buf64;
4949 int ret = ::fstat64(fd, &buf64);
4950 int st_mode = buf64.st_mode;
4951
4952 if (ret != -1) {
4953 if ((st_mode & S_IFMT) == S_IFDIR) {
4954 errno = EISDIR;
4955 ::close(fd);
4956 return -1;
4957 }
4958 } else {
4959 ::close(fd);
4960 return -1;
4961 }
4962 }
4963
4964 #ifdef FD_CLOEXEC
4965 // Validate that the use of the O_CLOEXEC flag on open above worked.
4966 // With recent kernels, we will perform this check exactly once.
4967 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
4968 if (!O_CLOEXEC_is_known_to_work) {
4969 int flags = ::fcntl(fd, F_GETFD);
4970 if (flags != -1) {
4971 if ((flags & FD_CLOEXEC) != 0)
4972 O_CLOEXEC_is_known_to_work = 1;
4973 else
4974 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4975 }
4976 }
4977 #endif
4978
4979 return fd;
4980 }
4981
4982
4983 // create binary file, rewriting existing file if required
4984 int os::create_binary_file(const char* path, bool rewrite_existing) {
4985 int oflags = O_WRONLY | O_CREAT;
4986 if (!rewrite_existing) {
4987 oflags |= O_EXCL;
4988 }
4989 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4990 }
4991
4992 // return current position of file pointer
4993 jlong os::current_file_offset(int fd) {
4994 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4995 }
4996
4997 // move file pointer to the specified offset
4998 jlong os::seek_to_file_offset(int fd, jlong offset) {
4999 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5000 }
5001
5002 // This code originates from JDK's sysAvailable
5003 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5004
5005 int os::available(int fd, jlong *bytes) {
5006 jlong cur, end;
5007 int mode;
5008 struct stat64 buf64;
5009
5010 if (::fstat64(fd, &buf64) >= 0) {
5011 mode = buf64.st_mode;
5012 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5013 // XXX: is the following call interruptible? If so, this might
5014 // need to go through the INTERRUPT_IO() wrapper as for other
5015 // blocking, interruptible calls in this file.
5016 int n;
5017 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5018 *bytes = n;
5019 return 1;
5020 }
5021 }
5022 }
5023 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5024 return 0;
5025 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5026 return 0;
5027 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5028 return 0;
5029 }
5030 *bytes = end - cur;
5031 return 1;
5032 }
5033
5034 // Map a block of memory.
5035 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5036 char *addr, size_t bytes, bool read_only,
5037 bool allow_exec) {
5038 int prot;
5039 int flags = MAP_PRIVATE;
5040
5041 if (read_only) {
5042 prot = PROT_READ;
5043 } else {
5044 prot = PROT_READ | PROT_WRITE;
5045 }
5046
5047 if (allow_exec) {
5048 prot |= PROT_EXEC;
5049 }
5050
5051 if (addr != NULL) {
5052 flags |= MAP_FIXED;
5053 }
5054
5055 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5056 fd, file_offset);
5057 if (mapped_address == MAP_FAILED) {
5058 return NULL;
5059 }
5060 return mapped_address;
5061 }
5062
5063
5064 // Remap a block of memory.
5065 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5066 char *addr, size_t bytes, bool read_only,
5067 bool allow_exec) {
5068 // same as map_memory() on this OS
5069 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5070 allow_exec);
5071 }
5072
5073
5074 // Unmap a block of memory.
5075 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5076 return munmap(addr, bytes) == 0;
5077 }
5078
5079 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5080
5081 static clockid_t thread_cpu_clockid(Thread* thread) {
5082 pthread_t tid = thread->osthread()->pthread_id();
5083 clockid_t clockid;
5084
5085 // Get thread clockid
5086 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5087 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5088 return clockid;
5089 }
5090
5091 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5092 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5093 // of a thread.
5094 //
5095 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5096 // the fast estimate available on the platform.
5097
5098 jlong os::current_thread_cpu_time() {
5099 if (os::Linux::supports_fast_thread_cpu_time()) {
5100 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5101 } else {
5102 // return user + sys since the cost is the same
5103 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5104 }
5105 }
5106
5107 jlong os::thread_cpu_time(Thread* thread) {
5108 // consistent with what current_thread_cpu_time() returns
5109 if (os::Linux::supports_fast_thread_cpu_time()) {
5110 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5111 } else {
5112 return slow_thread_cpu_time(thread, true /* user + sys */);
5113 }
5114 }
5115
5116 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5117 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5118 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5119 } else {
5120 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5121 }
5122 }
5123
5124 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5125 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5126 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5127 } else {
5128 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5129 }
5130 }
5131
5132 // -1 on error.
5133 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5134 pid_t tid = thread->osthread()->thread_id();
5135 char *s;
5136 char stat[2048];
5137 int statlen;
5138 char proc_name[64];
5139 int count;
5140 long sys_time, user_time;
5141 char cdummy;
5142 int idummy;
5143 long ldummy;
5144 FILE *fp;
5145
5146 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5147 fp = fopen(proc_name, "r");
5148 if (fp == NULL) return -1;
5149 statlen = fread(stat, 1, 2047, fp);
5150 stat[statlen] = '\0';
5151 fclose(fp);
5152
5153 // Skip pid and the command string. Note that we could be dealing with
5154 // weird command names, e.g. user could decide to rename java launcher
5155 // to "java 1.4.2 :)", then the stat file would look like
5156 // 1234 (java 1.4.2 :)) R ... ...
5157 // We don't really need to know the command string, just find the last
5158 // occurrence of ")" and then start parsing from there. See bug 4726580.
5159 s = strrchr(stat, ')');
5160 if (s == NULL) return -1;
5161
5162 // Skip blank chars
5163 do { s++; } while (s && isspace(*s));
5164
5165 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5166 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5167 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5168 &user_time, &sys_time);
5169 if (count != 13) return -1;
5170 if (user_sys_cpu_time) {
5171 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5172 } else {
5173 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5174 }
5175 }
5176
5177 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5178 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5179 info_ptr->may_skip_backward = false; // elapsed time not wall time
5180 info_ptr->may_skip_forward = false; // elapsed time not wall time
5181 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5182 }
5183
5184 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5185 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5186 info_ptr->may_skip_backward = false; // elapsed time not wall time
5187 info_ptr->may_skip_forward = false; // elapsed time not wall time
5188 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5189 }
5190
5191 bool os::is_thread_cpu_time_supported() {
5192 return true;
5193 }
5194
5195 // System loadavg support. Returns -1 if load average cannot be obtained.
5196 // Linux doesn't yet have a (official) notion of processor sets,
5197 // so just return the system wide load average.
5198 int os::loadavg(double loadavg[], int nelem) {
5199 return ::getloadavg(loadavg, nelem);
5200 }
5201
5202 void os::pause() {
5203 char filename[MAX_PATH];
5204 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5205 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5206 } else {
5207 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5208 }
5209
5210 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5211 if (fd != -1) {
5212 struct stat buf;
5213 ::close(fd);
5214 while (::stat(filename, &buf) == 0) {
5215 (void)::poll(NULL, 0, 100);
5216 }
5217 } else {
5218 jio_fprintf(stderr,
5219 "Could not open pause file '%s', continuing immediately.\n", filename);
5220 }
5221 }
5222
5223
5224 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5225 // comment paragraphs are worth repeating here:
5226 //
5227 // Assumption:
5228 // Only one parker can exist on an event, which is why we allocate
5229 // them per-thread. Multiple unparkers can coexist.
5230 //
5231 // _Event serves as a restricted-range semaphore.
5232 // -1 : thread is blocked, i.e. there is a waiter
5233 // 0 : neutral: thread is running or ready,
5234 // could have been signaled after a wait started
5235 // 1 : signaled - thread is running or ready
5236 //
5237 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5238 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5239 // For specifics regarding the bug see GLIBC BUGID 261237 :
5240 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5241 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5242 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5243 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5244 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5245 // and monitorenter when we're using 1-0 locking. All those operations may result in
5246 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5247 // of libpthread avoids the problem, but isn't practical.
5248 //
5249 // Possible remedies:
5250 //
5251 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5252 // This is palliative and probabilistic, however. If the thread is preempted
5253 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5254 // than the minimum period may have passed, and the abstime may be stale (in the
5255 // past) resultin in a hang. Using this technique reduces the odds of a hang
5256 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5257 //
5258 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5259 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5260 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5261 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5262 // thread.
5263 //
5264 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5265 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5266 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5267 // This also works well. In fact it avoids kernel-level scalability impediments
5268 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5269 // timers in a graceful fashion.
5270 //
5271 // 4. When the abstime value is in the past it appears that control returns
5272 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5273 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5274 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5275 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5276 // It may be possible to avoid reinitialization by checking the return
5277 // value from pthread_cond_timedwait(). In addition to reinitializing the
5278 // condvar we must establish the invariant that cond_signal() is only called
5279 // within critical sections protected by the adjunct mutex. This prevents
5280 // cond_signal() from "seeing" a condvar that's in the midst of being
5281 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5282 // desirable signal-after-unlock optimization that avoids futile context switching.
5283 //
5284 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5285 // structure when a condvar is used or initialized. cond_destroy() would
5286 // release the helper structure. Our reinitialize-after-timedwait fix
5287 // put excessive stress on malloc/free and locks protecting the c-heap.
5288 //
5289 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5290 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5291 // and only enabling the work-around for vulnerable environments.
5292
5293 // utility to compute the abstime argument to timedwait:
5294 // millis is the relative timeout time
5295 // abstime will be the absolute timeout time
5296 // TODO: replace compute_abstime() with unpackTime()
5297
5298 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5299 if (millis < 0) millis = 0;
5300
5301 jlong seconds = millis / 1000;
5302 millis %= 1000;
5303 if (seconds > 50000000) { // see man cond_timedwait(3T)
5304 seconds = 50000000;
5305 }
5306
5307 if (os::supports_monotonic_clock()) {
5308 struct timespec now;
5309 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5310 assert_status(status == 0, status, "clock_gettime");
5311 abstime->tv_sec = now.tv_sec + seconds;
5312 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5313 if (nanos >= NANOSECS_PER_SEC) {
5314 abstime->tv_sec += 1;
5315 nanos -= NANOSECS_PER_SEC;
5316 }
5317 abstime->tv_nsec = nanos;
5318 } else {
5319 struct timeval now;
5320 int status = gettimeofday(&now, NULL);
5321 assert(status == 0, "gettimeofday");
5322 abstime->tv_sec = now.tv_sec + seconds;
5323 long usec = now.tv_usec + millis * 1000;
5324 if (usec >= 1000000) {
5325 abstime->tv_sec += 1;
5326 usec -= 1000000;
5327 }
5328 abstime->tv_nsec = usec * 1000;
5329 }
5330 return abstime;
5331 }
5332
5333 void os::PlatformEvent::park() { // AKA "down()"
5334 // Transitions for _Event:
5335 // -1 => -1 : illegal
5336 // 1 => 0 : pass - return immediately
5337 // 0 => -1 : block; then set _Event to 0 before returning
5338
5339 // Invariant: Only the thread associated with the Event/PlatformEvent
5340 // may call park().
5341 // TODO: assert that _Assoc != NULL or _Assoc == Self
5342 assert(_nParked == 0, "invariant");
5343
5344 int v;
5345 for (;;) {
5346 v = _Event;
5347 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5348 }
5349 guarantee(v >= 0, "invariant");
5350 if (v == 0) {
5351 // Do this the hard way by blocking ...
5352 int status = pthread_mutex_lock(_mutex);
5353 assert_status(status == 0, status, "mutex_lock");
5354 guarantee(_nParked == 0, "invariant");
5355 ++_nParked;
5356 while (_Event < 0) {
5357 status = pthread_cond_wait(_cond, _mutex);
5358 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5359 // Treat this the same as if the wait was interrupted
5360 if (status == ETIME) { status = EINTR; }
5361 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5362 }
5363 --_nParked;
5364
5365 _Event = 0;
5366 status = pthread_mutex_unlock(_mutex);
5367 assert_status(status == 0, status, "mutex_unlock");
5368 // Paranoia to ensure our locked and lock-free paths interact
5369 // correctly with each other.
5370 OrderAccess::fence();
5371 }
5372 guarantee(_Event >= 0, "invariant");
5373 }
5374
5375 int os::PlatformEvent::park(jlong millis) {
5376 // Transitions for _Event:
5377 // -1 => -1 : illegal
5378 // 1 => 0 : pass - return immediately
5379 // 0 => -1 : block; then set _Event to 0 before returning
5380
5381 guarantee(_nParked == 0, "invariant");
5382
5383 int v;
5384 for (;;) {
5385 v = _Event;
5386 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5387 }
5388 guarantee(v >= 0, "invariant");
5389 if (v != 0) return OS_OK;
5390
5391 // We do this the hard way, by blocking the thread.
5392 // Consider enforcing a minimum timeout value.
5393 struct timespec abst;
5394 compute_abstime(&abst, millis);
5395
5396 int ret = OS_TIMEOUT;
5397 int status = pthread_mutex_lock(_mutex);
5398 assert_status(status == 0, status, "mutex_lock");
5399 guarantee(_nParked == 0, "invariant");
5400 ++_nParked;
5401
5402 // Object.wait(timo) will return because of
5403 // (a) notification
5404 // (b) timeout
5405 // (c) thread.interrupt
5406 //
5407 // Thread.interrupt and object.notify{All} both call Event::set.
5408 // That is, we treat thread.interrupt as a special case of notification.
5409 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5410 // We assume all ETIME returns are valid.
5411 //
5412 // TODO: properly differentiate simultaneous notify+interrupt.
5413 // In that case, we should propagate the notify to another waiter.
5414
5415 while (_Event < 0) {
5416 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5417 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5418 pthread_cond_destroy(_cond);
5419 pthread_cond_init(_cond, os::Linux::condAttr());
5420 }
5421 assert_status(status == 0 || status == EINTR ||
5422 status == ETIME || status == ETIMEDOUT,
5423 status, "cond_timedwait");
5424 if (!FilterSpuriousWakeups) break; // previous semantics
5425 if (status == ETIME || status == ETIMEDOUT) break;
5426 // We consume and ignore EINTR and spurious wakeups.
5427 }
5428 --_nParked;
5429 if (_Event >= 0) {
5430 ret = OS_OK;
5431 }
5432 _Event = 0;
5433 status = pthread_mutex_unlock(_mutex);
5434 assert_status(status == 0, status, "mutex_unlock");
5435 assert(_nParked == 0, "invariant");
5436 // Paranoia to ensure our locked and lock-free paths interact
5437 // correctly with each other.
5438 OrderAccess::fence();
5439 return ret;
5440 }
5441
5442 void os::PlatformEvent::unpark() {
5443 // Transitions for _Event:
5444 // 0 => 1 : just return
5445 // 1 => 1 : just return
5446 // -1 => either 0 or 1; must signal target thread
5447 // That is, we can safely transition _Event from -1 to either
5448 // 0 or 1.
5449 // See also: "Semaphores in Plan 9" by Mullender & Cox
5450 //
5451 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5452 // that it will take two back-to-back park() calls for the owning
5453 // thread to block. This has the benefit of forcing a spurious return
5454 // from the first park() call after an unpark() call which will help
5455 // shake out uses of park() and unpark() without condition variables.
5456
5457 if (Atomic::xchg(1, &_Event) >= 0) return;
5458
5459 // Wait for the thread associated with the event to vacate
5460 int status = pthread_mutex_lock(_mutex);
5461 assert_status(status == 0, status, "mutex_lock");
5462 int AnyWaiters = _nParked;
5463 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5464 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5465 AnyWaiters = 0;
5466 pthread_cond_signal(_cond);
5467 }
5468 status = pthread_mutex_unlock(_mutex);
5469 assert_status(status == 0, status, "mutex_unlock");
5470 if (AnyWaiters != 0) {
5471 // Note that we signal() *after* dropping the lock for "immortal" Events.
5472 // This is safe and avoids a common class of futile wakeups. In rare
5473 // circumstances this can cause a thread to return prematurely from
5474 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5475 // will simply re-test the condition and re-park itself.
5476 // This provides particular benefit if the underlying platform does not
5477 // provide wait morphing.
5478 status = pthread_cond_signal(_cond);
5479 assert_status(status == 0, status, "cond_signal");
5480 }
5481 }
5482
5483
5484 // JSR166
5485 // -------------------------------------------------------
5486
5487 // The solaris and linux implementations of park/unpark are fairly
5488 // conservative for now, but can be improved. They currently use a
5489 // mutex/condvar pair, plus a a count.
5490 // Park decrements count if > 0, else does a condvar wait. Unpark
5491 // sets count to 1 and signals condvar. Only one thread ever waits
5492 // on the condvar. Contention seen when trying to park implies that someone
5493 // is unparking you, so don't wait. And spurious returns are fine, so there
5494 // is no need to track notifications.
5495
5496 // This code is common to linux and solaris and will be moved to a
5497 // common place in dolphin.
5498 //
5499 // The passed in time value is either a relative time in nanoseconds
5500 // or an absolute time in milliseconds. Either way it has to be unpacked
5501 // into suitable seconds and nanoseconds components and stored in the
5502 // given timespec structure.
5503 // Given time is a 64-bit value and the time_t used in the timespec is only
5504 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5505 // overflow if times way in the future are given. Further on Solaris versions
5506 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5507 // number of seconds, in abstime, is less than current_time + 100,000,000.
5508 // As it will be 28 years before "now + 100000000" will overflow we can
5509 // ignore overflow and just impose a hard-limit on seconds using the value
5510 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5511 // years from "now".
5512
5513 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5514 assert(time > 0, "convertTime");
5515 time_t max_secs = 0;
5516
5517 if (!os::supports_monotonic_clock() || isAbsolute) {
5518 struct timeval now;
5519 int status = gettimeofday(&now, NULL);
5520 assert(status == 0, "gettimeofday");
5521
5522 max_secs = now.tv_sec + MAX_SECS;
5523
5524 if (isAbsolute) {
5525 jlong secs = time / 1000;
5526 if (secs > max_secs) {
5527 absTime->tv_sec = max_secs;
5528 } else {
5529 absTime->tv_sec = secs;
5530 }
5531 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5532 } else {
5533 jlong secs = time / NANOSECS_PER_SEC;
5534 if (secs >= MAX_SECS) {
5535 absTime->tv_sec = max_secs;
5536 absTime->tv_nsec = 0;
5537 } else {
5538 absTime->tv_sec = now.tv_sec + secs;
5539 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5540 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5541 absTime->tv_nsec -= NANOSECS_PER_SEC;
5542 ++absTime->tv_sec; // note: this must be <= max_secs
5543 }
5544 }
5545 }
5546 } else {
5547 // must be relative using monotonic clock
5548 struct timespec now;
5549 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5550 assert_status(status == 0, status, "clock_gettime");
5551 max_secs = now.tv_sec + MAX_SECS;
5552 jlong secs = time / NANOSECS_PER_SEC;
5553 if (secs >= MAX_SECS) {
5554 absTime->tv_sec = max_secs;
5555 absTime->tv_nsec = 0;
5556 } else {
5557 absTime->tv_sec = now.tv_sec + secs;
5558 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5559 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5560 absTime->tv_nsec -= NANOSECS_PER_SEC;
5561 ++absTime->tv_sec; // note: this must be <= max_secs
5562 }
5563 }
5564 }
5565 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5566 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5567 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5568 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5569 }
5570
5571 void Parker::park(bool isAbsolute, jlong time) {
5572 // Ideally we'd do something useful while spinning, such
5573 // as calling unpackTime().
5574
5575 // Optional fast-path check:
5576 // Return immediately if a permit is available.
5577 // We depend on Atomic::xchg() having full barrier semantics
5578 // since we are doing a lock-free update to _counter.
5579 if (Atomic::xchg(0, &_counter) > 0) return;
5580
5581 Thread* thread = Thread::current();
5582 assert(thread->is_Java_thread(), "Must be JavaThread");
5583 JavaThread *jt = (JavaThread *)thread;
5584
5585 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5586 // Check interrupt before trying to wait
5587 if (Thread::is_interrupted(thread, false)) {
5588 return;
5589 }
5590
5591 // Next, demultiplex/decode time arguments
5592 timespec absTime;
5593 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5594 return;
5595 }
5596 if (time > 0) {
5597 unpackTime(&absTime, isAbsolute, time);
5598 }
5599
5600
5601 // Enter safepoint region
5602 // Beware of deadlocks such as 6317397.
5603 // The per-thread Parker:: mutex is a classic leaf-lock.
5604 // In particular a thread must never block on the Threads_lock while
5605 // holding the Parker:: mutex. If safepoints are pending both the
5606 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5607 ThreadBlockInVM tbivm(jt);
5608
5609 // Don't wait if cannot get lock since interference arises from
5610 // unblocking. Also. check interrupt before trying wait
5611 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5612 return;
5613 }
5614
5615 int status;
5616 if (_counter > 0) { // no wait needed
5617 _counter = 0;
5618 status = pthread_mutex_unlock(_mutex);
5619 assert(status == 0, "invariant");
5620 // Paranoia to ensure our locked and lock-free paths interact
5621 // correctly with each other and Java-level accesses.
5622 OrderAccess::fence();
5623 return;
5624 }
5625
5626 #ifdef ASSERT
5627 // Don't catch signals while blocked; let the running threads have the signals.
5628 // (This allows a debugger to break into the running thread.)
5629 sigset_t oldsigs;
5630 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5631 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5632 #endif
5633
5634 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5635 jt->set_suspend_equivalent();
5636 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5637
5638 assert(_cur_index == -1, "invariant");
5639 if (time == 0) {
5640 _cur_index = REL_INDEX; // arbitrary choice when not timed
5641 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5642 } else {
5643 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5644 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5645 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5646 pthread_cond_destroy(&_cond[_cur_index]);
5647 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5648 }
5649 }
5650 _cur_index = -1;
5651 assert_status(status == 0 || status == EINTR ||
5652 status == ETIME || status == ETIMEDOUT,
5653 status, "cond_timedwait");
5654
5655 #ifdef ASSERT
5656 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5657 #endif
5658
5659 _counter = 0;
5660 status = pthread_mutex_unlock(_mutex);
5661 assert_status(status == 0, status, "invariant");
5662 // Paranoia to ensure our locked and lock-free paths interact
5663 // correctly with each other and Java-level accesses.
5664 OrderAccess::fence();
5665
5666 // If externally suspended while waiting, re-suspend
5667 if (jt->handle_special_suspend_equivalent_condition()) {
5668 jt->java_suspend_self();
5669 }
5670 }
5671
5672 void Parker::unpark() {
5673 int status = pthread_mutex_lock(_mutex);
5674 assert(status == 0, "invariant");
5675 const int s = _counter;
5676 _counter = 1;
5677 if (s < 1) {
5678 // thread might be parked
5679 if (_cur_index != -1) {
5680 // thread is definitely parked
5681 if (WorkAroundNPTLTimedWaitHang) {
5682 status = pthread_cond_signal(&_cond[_cur_index]);
5683 assert(status == 0, "invariant");
5684 status = pthread_mutex_unlock(_mutex);
5685 assert(status == 0, "invariant");
5686 } else {
5687 status = pthread_mutex_unlock(_mutex);
5688 assert(status == 0, "invariant");
5689 status = pthread_cond_signal(&_cond[_cur_index]);
5690 assert(status == 0, "invariant");
5691 }
5692 } else {
5693 pthread_mutex_unlock(_mutex);
5694 assert(status == 0, "invariant");
5695 }
5696 } else {
5697 pthread_mutex_unlock(_mutex);
5698 assert(status == 0, "invariant");
5699 }
5700 }
5701
5702
5703 extern char** environ;
5704
5705 // Run the specified command in a separate process. Return its exit value,
5706 // or -1 on failure (e.g. can't fork a new process).
5707 // Unlike system(), this function can be called from signal handler. It
5708 // doesn't block SIGINT et al.
5709 int os::fork_and_exec(char* cmd) {
5710 const char * argv[4] = {"sh", "-c", cmd, NULL};
5711
5712 pid_t pid = fork();
5713
5714 if (pid < 0) {
5715 // fork failed
5716 return -1;
5717
5718 } else if (pid == 0) {
5719 // child process
5720
5721 execve("/bin/sh", (char* const*)argv, environ);
5722
5723 // execve failed
5724 _exit(-1);
5725
5726 } else {
5727 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5728 // care about the actual exit code, for now.
5729
5730 int status;
5731
5732 // Wait for the child process to exit. This returns immediately if
5733 // the child has already exited. */
5734 while (waitpid(pid, &status, 0) < 0) {
5735 switch (errno) {
5736 case ECHILD: return 0;
5737 case EINTR: break;
5738 default: return -1;
5739 }
5740 }
5741
5742 if (WIFEXITED(status)) {
5743 // The child exited normally; get its exit code.
5744 return WEXITSTATUS(status);
5745 } else if (WIFSIGNALED(status)) {
5746 // The child exited because of a signal
5747 // The best value to return is 0x80 + signal number,
5748 // because that is what all Unix shells do, and because
5749 // it allows callers to distinguish between process exit and
5750 // process death by signal.
5751 return 0x80 + WTERMSIG(status);
5752 } else {
5753 // Unknown exit code; pass it through
5754 return status;
5755 }
5756 }
5757 }
5758
5759 // is_headless_jre()
5760 //
5761 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5762 // in order to report if we are running in a headless jre
5763 //
5764 // Since JDK8 xawt/libmawt.so was moved into the same directory
5765 // as libawt.so, and renamed libawt_xawt.so
5766 //
5767 bool os::is_headless_jre() {
5768 struct stat statbuf;
5769 char buf[MAXPATHLEN];
5770 char libmawtpath[MAXPATHLEN];
5771 const char *xawtstr = "/xawt/libmawt.so";
5772 const char *new_xawtstr = "/libawt_xawt.so";
5773 char *p;
5774
5775 // Get path to libjvm.so
5776 os::jvm_path(buf, sizeof(buf));
5777
5778 // Get rid of libjvm.so
5779 p = strrchr(buf, '/');
5780 if (p == NULL) {
5781 return false;
5782 } else {
5783 *p = '\0';
5784 }
5785
5786 // Get rid of client or server
5787 p = strrchr(buf, '/');
5788 if (p == NULL) {
5789 return false;
5790 } else {
5791 *p = '\0';
5792 }
5793
5794 // check xawt/libmawt.so
5795 strcpy(libmawtpath, buf);
5796 strcat(libmawtpath, xawtstr);
5797 if (::stat(libmawtpath, &statbuf) == 0) return false;
5798
5799 // check libawt_xawt.so
5800 strcpy(libmawtpath, buf);
5801 strcat(libmawtpath, new_xawtstr);
5802 if (::stat(libmawtpath, &statbuf) == 0) return false;
5803
5804 return true;
5805 }
5806
5807 // Get the default path to the core file
5808 // Returns the length of the string
5809 int os::get_core_path(char* buffer, size_t bufferSize) {
5810 /*
5811 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5812 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5813 */
5814 const int core_pattern_len = 129;
5815 char core_pattern[core_pattern_len] = {0};
5816
5817 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5818 if (core_pattern_file != -1) {
5819 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5820 ::close(core_pattern_file);
5821
5822 if (ret > 0) {
5823 char *last_char = core_pattern + strlen(core_pattern) - 1;
5824
5825 if (*last_char == '\n') {
5826 *last_char = '\0';
5827 }
5828 }
5829 }
5830
5831 if (strlen(core_pattern) == 0) {
5832 return -1;
5833 }
5834
5835 char *pid_pos = strstr(core_pattern, "%p");
5836 int written;
5837
5838 if (core_pattern[0] == '/') {
5839 written = jio_snprintf(buffer, bufferSize, core_pattern);
5840 } else {
5841 char cwd[PATH_MAX];
5842
5843 const char* p = get_current_directory(cwd, PATH_MAX);
5844 if (p == NULL) {
5845 return -1;
5846 }
5847
5848 if (core_pattern[0] == '|') {
5849 written = jio_snprintf(buffer, bufferSize,
5850 "\"%s\" (or dumping to %s/core.%d)",
5851 &core_pattern[1], p, current_process_id());
5852 } else {
5853 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5854 }
5855 }
5856
5857 if (written < 0) {
5858 return -1;
5859 }
5860
5861 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5862 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5863
5864 if (core_uses_pid_file != -1) {
5865 char core_uses_pid = 0;
5866 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5867 ::close(core_uses_pid_file);
5868
5869 if (core_uses_pid == '1') {
5870 jio_snprintf(buffer + written, bufferSize - written,
5871 ".%d", current_process_id());
5872 }
5873 }
5874 }
5875
5876 return strlen(buffer);
5877 }
5878
5879 /////////////// Unit tests ///////////////
5880
5881 #ifndef PRODUCT
5882
5883 #define test_log(...) \
5884 do { \
5885 if (VerboseInternalVMTests) { \
5886 tty->print_cr(__VA_ARGS__); \
5887 tty->flush(); \
5888 } \
5889 } while (false)
5890
5891 class TestReserveMemorySpecial : AllStatic {
5892 public:
5893 static void small_page_write(void* addr, size_t size) {
5894 size_t page_size = os::vm_page_size();
5895
5896 char* end = (char*)addr + size;
5897 for (char* p = (char*)addr; p < end; p += page_size) {
5898 *p = 1;
5899 }
5900 }
5901
5902 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5903 if (!UseHugeTLBFS) {
5904 return;
5905 }
5906
5907 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5908
5909 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5910
5911 if (addr != NULL) {
5912 small_page_write(addr, size);
5913
5914 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5915 }
5916 }
5917
5918 static void test_reserve_memory_special_huge_tlbfs_only() {
5919 if (!UseHugeTLBFS) {
5920 return;
5921 }
5922
5923 size_t lp = os::large_page_size();
5924
5925 for (size_t size = lp; size <= lp * 10; size += lp) {
5926 test_reserve_memory_special_huge_tlbfs_only(size);
5927 }
5928 }
5929
5930 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5931 size_t lp = os::large_page_size();
5932 size_t ag = os::vm_allocation_granularity();
5933
5934 // sizes to test
5935 const size_t sizes[] = {
5936 lp, lp + ag, lp + lp / 2, lp * 2,
5937 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5938 lp * 10, lp * 10 + lp / 2
5939 };
5940 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5941
5942 // For each size/alignment combination, we test three scenarios:
5943 // 1) with req_addr == NULL
5944 // 2) with a non-null req_addr at which we expect to successfully allocate
5945 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5946 // expect the allocation to either fail or to ignore req_addr
5947
5948 // Pre-allocate two areas; they shall be as large as the largest allocation
5949 // and aligned to the largest alignment we will be testing.
5950 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5951 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5952 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5953 -1, 0);
5954 assert(mapping1 != MAP_FAILED, "should work");
5955
5956 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5957 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5958 -1, 0);
5959 assert(mapping2 != MAP_FAILED, "should work");
5960
5961 // Unmap the first mapping, but leave the second mapping intact: the first
5962 // mapping will serve as a value for a "good" req_addr (case 2). The second
5963 // mapping, still intact, as "bad" req_addr (case 3).
5964 ::munmap(mapping1, mapping_size);
5965
5966 // Case 1
5967 test_log("%s, req_addr NULL:", __FUNCTION__);
5968 test_log("size align result");
5969
5970 for (int i = 0; i < num_sizes; i++) {
5971 const size_t size = sizes[i];
5972 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
5973 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5974 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5975 size, alignment, p, (p != NULL ? "" : "(failed)"));
5976 if (p != NULL) {
5977 assert(is_ptr_aligned(p, alignment), "must be");
5978 small_page_write(p, size);
5979 os::Linux::release_memory_special_huge_tlbfs(p, size);
5980 }
5981 }
5982 }
5983
5984 // Case 2
5985 test_log("%s, req_addr non-NULL:", __FUNCTION__);
5986 test_log("size align req_addr result");
5987
5988 for (int i = 0; i < num_sizes; i++) {
5989 const size_t size = sizes[i];
5990 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
5991 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
5992 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5993 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
5994 size, alignment, req_addr, p,
5995 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
5996 if (p != NULL) {
5997 assert(p == req_addr, "must be");
5998 small_page_write(p, size);
5999 os::Linux::release_memory_special_huge_tlbfs(p, size);
6000 }
6001 }
6002 }
6003
6004 // Case 3
6005 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6006 test_log("size align req_addr result");
6007
6008 for (int i = 0; i < num_sizes; i++) {
6009 const size_t size = sizes[i];
6010 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6011 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6012 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6013 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6014 size, alignment, req_addr, p,
6015 ((p != NULL ? "" : "(failed)")));
6016 // as the area around req_addr contains already existing mappings, the API should always
6017 // return NULL (as per contract, it cannot return another address)
6018 assert(p == NULL, "must be");
6019 }
6020 }
6021
6022 ::munmap(mapping2, mapping_size);
6023
6024 }
6025
6026 static void test_reserve_memory_special_huge_tlbfs() {
6027 if (!UseHugeTLBFS) {
6028 return;
6029 }
6030
6031 test_reserve_memory_special_huge_tlbfs_only();
6032 test_reserve_memory_special_huge_tlbfs_mixed();
6033 }
6034
6035 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6036 if (!UseSHM) {
6037 return;
6038 }
6039
6040 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6041
6042 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6043
6044 if (addr != NULL) {
6045 assert(is_ptr_aligned(addr, alignment), "Check");
6046 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6047
6048 small_page_write(addr, size);
6049
6050 os::Linux::release_memory_special_shm(addr, size);
6051 }
6052 }
6053
6054 static void test_reserve_memory_special_shm() {
6055 size_t lp = os::large_page_size();
6056 size_t ag = os::vm_allocation_granularity();
6057
6058 for (size_t size = ag; size < lp * 3; size += ag) {
6059 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6060 test_reserve_memory_special_shm(size, alignment);
6061 }
6062 }
6063 }
6064
6065 static void test() {
6066 test_reserve_memory_special_huge_tlbfs();
6067 test_reserve_memory_special_shm();
6068 }
6069 };
6070
6071 void TestReserveMemorySpecial_test() {
6072 TestReserveMemorySpecial::test();
6073 }
6074
6075 #endif