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