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 osthread->set_thread_id(os::current_thread_id());
657
658 if (UseNUMA) {
659 int lgrp_id = os::numa_get_group_id();
660 if (lgrp_id != -1) {
661 thread->set_lgrp_id(lgrp_id);
662 }
663 }
664 // initialize signal mask for this thread
665 os::Linux::hotspot_sigmask(thread);
666
667 // initialize floating point control register
668 os::Linux::init_thread_fpu_state();
669
670 // handshaking with parent thread
671 {
672 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
673
674 // notify parent thread
675 osthread->set_state(INITIALIZED);
676 sync->notify_all();
677
678 // wait until os::start_thread()
679 while (osthread->get_state() == INITIALIZED) {
680 sync->wait(Mutex::_no_safepoint_check_flag);
681 }
682 }
683
684 // call one more level start routine
685 thread->run();
686
687 return 0;
688 }
689
690 bool os::create_thread(Thread* thread, ThreadType thr_type,
691 size_t stack_size) {
692 assert(thread->osthread() == NULL, "caller responsible");
693
694 // Allocate the OSThread object
695 OSThread* osthread = new OSThread(NULL, NULL);
696 if (osthread == NULL) {
697 return false;
698 }
699
700 // set the correct thread state
701 osthread->set_thread_type(thr_type);
702
703 // Initial state is ALLOCATED but not INITIALIZED
704 osthread->set_state(ALLOCATED);
705
706 thread->set_osthread(osthread);
707
708 // init thread attributes
709 pthread_attr_t attr;
710 pthread_attr_init(&attr);
711 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
712
713 // stack size
714 // calculate stack size if it's not specified by caller
715 if (stack_size == 0) {
716 stack_size = os::Linux::default_stack_size(thr_type);
717
718 switch (thr_type) {
719 case os::java_thread:
720 // Java threads use ThreadStackSize which default value can be
721 // changed with the flag -Xss
722 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
723 stack_size = JavaThread::stack_size_at_create();
724 break;
725 case os::compiler_thread:
726 if (CompilerThreadStackSize > 0) {
727 stack_size = (size_t)(CompilerThreadStackSize * K);
728 break;
729 } // else fall through:
730 // use VMThreadStackSize if CompilerThreadStackSize is not defined
731 case os::vm_thread:
732 case os::pgc_thread:
733 case os::cgc_thread:
734 case os::watcher_thread:
735 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
736 break;
737 }
738 }
739
740 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
741 pthread_attr_setstacksize(&attr, stack_size);
742
743 // glibc guard page
744 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
745
746 ThreadState state;
747
748 {
749 pthread_t tid;
750 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
751
752 pthread_attr_destroy(&attr);
753
754 if (ret != 0) {
755 if (PrintMiscellaneous && (Verbose || WizardMode)) {
756 perror("pthread_create()");
757 }
758 // Need to clean up stuff we've allocated so far
759 thread->set_osthread(NULL);
760 delete osthread;
761 return false;
762 }
763
764 // Store pthread info into the OSThread
765 osthread->set_pthread_id(tid);
766
767 // Wait until child thread is either initialized or aborted
768 {
769 Monitor* sync_with_child = osthread->startThread_lock();
770 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
771 while ((state = osthread->get_state()) == ALLOCATED) {
772 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
773 }
774 }
775 }
776
777 // Aborted due to thread limit being reached
778 if (state == ZOMBIE) {
779 thread->set_osthread(NULL);
780 delete osthread;
781 return false;
782 }
783
784 // The thread is returned suspended (in state INITIALIZED),
785 // and is started higher up in the call chain
786 assert(state == INITIALIZED, "race condition");
787 return true;
788 }
789
790 /////////////////////////////////////////////////////////////////////////////
791 // attach existing thread
792
793 // bootstrap the main thread
794 bool os::create_main_thread(JavaThread* thread) {
795 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
796 return create_attached_thread(thread);
797 }
798
799 bool os::create_attached_thread(JavaThread* thread) {
800 #ifdef ASSERT
801 thread->verify_not_published();
802 #endif
803
804 // Allocate the OSThread object
805 OSThread* osthread = new OSThread(NULL, NULL);
806
807 if (osthread == NULL) {
808 return false;
809 }
810
811 // Store pthread info into the OSThread
812 osthread->set_thread_id(os::Linux::gettid());
813 osthread->set_pthread_id(::pthread_self());
814
815 // initialize floating point control register
816 os::Linux::init_thread_fpu_state();
817
818 // Initial thread state is RUNNABLE
819 osthread->set_state(RUNNABLE);
820
821 thread->set_osthread(osthread);
822
823 if (UseNUMA) {
824 int lgrp_id = os::numa_get_group_id();
825 if (lgrp_id != -1) {
826 thread->set_lgrp_id(lgrp_id);
827 }
828 }
829
830 if (os::Linux::is_initial_thread()) {
831 // If current thread is initial thread, its stack is mapped on demand,
832 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
833 // the entire stack region to avoid SEGV in stack banging.
834 // It is also useful to get around the heap-stack-gap problem on SuSE
835 // kernel (see 4821821 for details). We first expand stack to the top
836 // of yellow zone, then enable stack yellow zone (order is significant,
837 // enabling yellow zone first will crash JVM on SuSE Linux), so there
838 // is no gap between the last two virtual memory regions.
839
840 JavaThread *jt = (JavaThread *)thread;
841 address addr = jt->stack_yellow_zone_base();
842 assert(addr != NULL, "initialization problem?");
843 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
844
845 osthread->set_expanding_stack();
846 os::Linux::manually_expand_stack(jt, addr);
847 osthread->clear_expanding_stack();
848 }
849
850 // initialize signal mask for this thread
851 // and save the caller's signal mask
852 os::Linux::hotspot_sigmask(thread);
853
854 return true;
855 }
856
857 void os::pd_start_thread(Thread* thread) {
858 OSThread * osthread = thread->osthread();
859 assert(osthread->get_state() != INITIALIZED, "just checking");
860 Monitor* sync_with_child = osthread->startThread_lock();
861 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
862 sync_with_child->notify();
863 }
864
865 // Free Linux resources related to the OSThread
866 void os::free_thread(OSThread* osthread) {
867 assert(osthread != NULL, "osthread not set");
868
869 if (Thread::current()->osthread() == osthread) {
870 // Restore caller's signal mask
871 sigset_t sigmask = osthread->caller_sigmask();
872 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
873 }
874
875 delete osthread;
876 }
877
878 //////////////////////////////////////////////////////////////////////////////
879 // thread local storage
880
881 // Restore the thread pointer if the destructor is called. This is in case
882 // someone from JNI code sets up a destructor with pthread_key_create to run
883 // detachCurrentThread on thread death. Unless we restore the thread pointer we
884 // will hang or crash. When detachCurrentThread is called the key will be set
885 // to null and we will not be called again. If detachCurrentThread is never
886 // called we could loop forever depending on the pthread implementation.
887 static void restore_thread_pointer(void* p) {
888 Thread* thread = (Thread*) p;
889 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
890 }
891
892 int os::allocate_thread_local_storage() {
893 pthread_key_t key;
894 int rslt = pthread_key_create(&key, restore_thread_pointer);
895 assert(rslt == 0, "cannot allocate thread local storage");
896 return (int)key;
897 }
898
899 // Note: This is currently not used by VM, as we don't destroy TLS key
900 // on VM exit.
901 void os::free_thread_local_storage(int index) {
902 int rslt = pthread_key_delete((pthread_key_t)index);
903 assert(rslt == 0, "invalid index");
904 }
905
906 void os::thread_local_storage_at_put(int index, void* value) {
907 int rslt = pthread_setspecific((pthread_key_t)index, value);
908 assert(rslt == 0, "pthread_setspecific failed");
909 }
910
911 extern "C" Thread* get_thread() {
912 return ThreadLocalStorage::thread();
913 }
914
915 //////////////////////////////////////////////////////////////////////////////
916 // initial thread
917
918 // Check if current thread is the initial thread, similar to Solaris thr_main.
919 bool os::Linux::is_initial_thread(void) {
920 char dummy;
921 // If called before init complete, thread stack bottom will be null.
922 // Can be called if fatal error occurs before initialization.
923 if (initial_thread_stack_bottom() == NULL) return false;
924 assert(initial_thread_stack_bottom() != NULL &&
925 initial_thread_stack_size() != 0,
926 "os::init did not locate initial thread's stack region");
927 if ((address)&dummy >= initial_thread_stack_bottom() &&
928 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
929 return true;
930 } else {
931 return false;
932 }
933 }
934
935 // Find the virtual memory area that contains addr
936 static bool find_vma(address addr, address* vma_low, address* vma_high) {
937 FILE *fp = fopen("/proc/self/maps", "r");
938 if (fp) {
939 address low, high;
940 while (!feof(fp)) {
941 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
942 if (low <= addr && addr < high) {
943 if (vma_low) *vma_low = low;
944 if (vma_high) *vma_high = high;
945 fclose(fp);
946 return true;
947 }
948 }
949 for (;;) {
950 int ch = fgetc(fp);
951 if (ch == EOF || ch == (int)'\n') break;
952 }
953 }
954 fclose(fp);
955 }
956 return false;
957 }
958
959 // Locate initial thread stack. This special handling of initial thread stack
960 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
961 // bogus value for initial thread.
962 void os::Linux::capture_initial_stack(size_t max_size) {
963 // stack size is the easy part, get it from RLIMIT_STACK
964 size_t stack_size;
965 struct rlimit rlim;
966 getrlimit(RLIMIT_STACK, &rlim);
967 stack_size = rlim.rlim_cur;
968
969 // 6308388: a bug in ld.so will relocate its own .data section to the
970 // lower end of primordial stack; reduce ulimit -s value a little bit
971 // so we won't install guard page on ld.so's data section.
972 stack_size -= 2 * page_size();
973
974 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
975 // 7.1, in both cases we will get 2G in return value.
976 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
977 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
978 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
979 // in case other parts in glibc still assumes 2M max stack size.
980 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
981 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
982 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
983 stack_size = 2 * K * K IA64_ONLY(*2);
984 }
985 // Try to figure out where the stack base (top) is. This is harder.
986 //
987 // When an application is started, glibc saves the initial stack pointer in
988 // a global variable "__libc_stack_end", which is then used by system
989 // libraries. __libc_stack_end should be pretty close to stack top. The
990 // variable is available since the very early days. However, because it is
991 // a private interface, it could disappear in the future.
992 //
993 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
994 // to __libc_stack_end, it is very close to stack top, but isn't the real
995 // stack top. Note that /proc may not exist if VM is running as a chroot
996 // program, so reading /proc/<pid>/stat could fail. Also the contents of
997 // /proc/<pid>/stat could change in the future (though unlikely).
998 //
999 // We try __libc_stack_end first. If that doesn't work, look for
1000 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1001 // as a hint, which should work well in most cases.
1002
1003 uintptr_t stack_start;
1004
1005 // try __libc_stack_end first
1006 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1007 if (p && *p) {
1008 stack_start = *p;
1009 } else {
1010 // see if we can get the start_stack field from /proc/self/stat
1011 FILE *fp;
1012 int pid;
1013 char state;
1014 int ppid;
1015 int pgrp;
1016 int session;
1017 int nr;
1018 int tpgrp;
1019 unsigned long flags;
1020 unsigned long minflt;
1021 unsigned long cminflt;
1022 unsigned long majflt;
1023 unsigned long cmajflt;
1024 unsigned long utime;
1025 unsigned long stime;
1026 long cutime;
1027 long cstime;
1028 long prio;
1029 long nice;
1030 long junk;
1031 long it_real;
1032 uintptr_t start;
1033 uintptr_t vsize;
1034 intptr_t rss;
1035 uintptr_t rsslim;
1036 uintptr_t scodes;
1037 uintptr_t ecode;
1038 int i;
1039
1040 // Figure what the primordial thread stack base is. Code is inspired
1041 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1042 // followed by command name surrounded by parentheses, state, etc.
1043 char stat[2048];
1044 int statlen;
1045
1046 fp = fopen("/proc/self/stat", "r");
1047 if (fp) {
1048 statlen = fread(stat, 1, 2047, fp);
1049 stat[statlen] = '\0';
1050 fclose(fp);
1051
1052 // Skip pid and the command string. Note that we could be dealing with
1053 // weird command names, e.g. user could decide to rename java launcher
1054 // to "java 1.4.2 :)", then the stat file would look like
1055 // 1234 (java 1.4.2 :)) R ... ...
1056 // We don't really need to know the command string, just find the last
1057 // occurrence of ")" and then start parsing from there. See bug 4726580.
1058 char * s = strrchr(stat, ')');
1059
1060 i = 0;
1061 if (s) {
1062 // Skip blank chars
1063 do { s++; } while (s && isspace(*s));
1064
1065 #define _UFM UINTX_FORMAT
1066 #define _DFM INTX_FORMAT
1067
1068 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1069 // 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
1070 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,
1071 &state, // 3 %c
1072 &ppid, // 4 %d
1073 &pgrp, // 5 %d
1074 &session, // 6 %d
1075 &nr, // 7 %d
1076 &tpgrp, // 8 %d
1077 &flags, // 9 %lu
1078 &minflt, // 10 %lu
1079 &cminflt, // 11 %lu
1080 &majflt, // 12 %lu
1081 &cmajflt, // 13 %lu
1082 &utime, // 14 %lu
1083 &stime, // 15 %lu
1084 &cutime, // 16 %ld
1085 &cstime, // 17 %ld
1086 &prio, // 18 %ld
1087 &nice, // 19 %ld
1088 &junk, // 20 %ld
1089 &it_real, // 21 %ld
1090 &start, // 22 UINTX_FORMAT
1091 &vsize, // 23 UINTX_FORMAT
1092 &rss, // 24 INTX_FORMAT
1093 &rsslim, // 25 UINTX_FORMAT
1094 &scodes, // 26 UINTX_FORMAT
1095 &ecode, // 27 UINTX_FORMAT
1096 &stack_start); // 28 UINTX_FORMAT
1097 }
1098
1099 #undef _UFM
1100 #undef _DFM
1101
1102 if (i != 28 - 2) {
1103 assert(false, "Bad conversion from /proc/self/stat");
1104 // product mode - assume we are the initial thread, good luck in the
1105 // embedded case.
1106 warning("Can't detect initial thread stack location - bad conversion");
1107 stack_start = (uintptr_t) &rlim;
1108 }
1109 } else {
1110 // For some reason we can't open /proc/self/stat (for example, running on
1111 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1112 // most cases, so don't abort:
1113 warning("Can't detect initial thread stack location - no /proc/self/stat");
1114 stack_start = (uintptr_t) &rlim;
1115 }
1116 }
1117
1118 // Now we have a pointer (stack_start) very close to the stack top, the
1119 // next thing to do is to figure out the exact location of stack top. We
1120 // can find out the virtual memory area that contains stack_start by
1121 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1122 // and its upper limit is the real stack top. (again, this would fail if
1123 // running inside chroot, because /proc may not exist.)
1124
1125 uintptr_t stack_top;
1126 address low, high;
1127 if (find_vma((address)stack_start, &low, &high)) {
1128 // success, "high" is the true stack top. (ignore "low", because initial
1129 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1130 stack_top = (uintptr_t)high;
1131 } else {
1132 // failed, likely because /proc/self/maps does not exist
1133 warning("Can't detect initial thread stack location - find_vma failed");
1134 // best effort: stack_start is normally within a few pages below the real
1135 // stack top, use it as stack top, and reduce stack size so we won't put
1136 // guard page outside stack.
1137 stack_top = stack_start;
1138 stack_size -= 16 * page_size();
1139 }
1140
1141 // stack_top could be partially down the page so align it
1142 stack_top = align_size_up(stack_top, page_size());
1143
1144 if (max_size && stack_size > max_size) {
1145 _initial_thread_stack_size = max_size;
1146 } else {
1147 _initial_thread_stack_size = stack_size;
1148 }
1149
1150 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1151 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1152 }
1153
1154 ////////////////////////////////////////////////////////////////////////////////
1155 // time support
1156
1157 // Time since start-up in seconds to a fine granularity.
1158 // Used by VMSelfDestructTimer and the MemProfiler.
1159 double os::elapsedTime() {
1160
1161 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1162 }
1163
1164 jlong os::elapsed_counter() {
1165 return javaTimeNanos() - initial_time_count;
1166 }
1167
1168 jlong os::elapsed_frequency() {
1169 return NANOSECS_PER_SEC; // nanosecond resolution
1170 }
1171
1172 bool os::supports_vtime() { return true; }
1173 bool os::enable_vtime() { return false; }
1174 bool os::vtime_enabled() { return false; }
1175
1176 double os::elapsedVTime() {
1177 struct rusage usage;
1178 int retval = getrusage(RUSAGE_THREAD, &usage);
1179 if (retval == 0) {
1180 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);
1181 } else {
1182 // better than nothing, but not much
1183 return elapsedTime();
1184 }
1185 }
1186
1187 jlong os::javaTimeMillis() {
1188 timeval time;
1189 int status = gettimeofday(&time, NULL);
1190 assert(status != -1, "linux error");
1191 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1192 }
1193
1194 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1195 timeval time;
1196 int status = gettimeofday(&time, NULL);
1197 assert(status != -1, "linux error");
1198 seconds = jlong(time.tv_sec);
1199 nanos = jlong(time.tv_usec) * 1000;
1200 }
1201
1202
1203 #ifndef CLOCK_MONOTONIC
1204 #define CLOCK_MONOTONIC (1)
1205 #endif
1206
1207 void os::Linux::clock_init() {
1208 // we do dlopen's in this particular order due to bug in linux
1209 // dynamical loader (see 6348968) leading to crash on exit
1210 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1211 if (handle == NULL) {
1212 handle = dlopen("librt.so", RTLD_LAZY);
1213 }
1214
1215 if (handle) {
1216 int (*clock_getres_func)(clockid_t, struct timespec*) =
1217 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1218 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1219 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1220 if (clock_getres_func && clock_gettime_func) {
1221 // See if monotonic clock is supported by the kernel. Note that some
1222 // early implementations simply return kernel jiffies (updated every
1223 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1224 // for nano time (though the monotonic property is still nice to have).
1225 // It's fixed in newer kernels, however clock_getres() still returns
1226 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1227 // resolution for now. Hopefully as people move to new kernels, this
1228 // won't be a problem.
1229 struct timespec res;
1230 struct timespec tp;
1231 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1232 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1233 // yes, monotonic clock is supported
1234 _clock_gettime = clock_gettime_func;
1235 return;
1236 } else {
1237 // close librt if there is no monotonic clock
1238 dlclose(handle);
1239 }
1240 }
1241 }
1242 warning("No monotonic clock was available - timed services may " \
1243 "be adversely affected if the time-of-day clock changes");
1244 }
1245
1246 #ifndef SYS_clock_getres
1247 #if defined(IA32) || defined(AMD64)
1248 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1249 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1250 #else
1251 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1252 #define sys_clock_getres(x,y) -1
1253 #endif
1254 #else
1255 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1256 #endif
1257
1258 void os::Linux::fast_thread_clock_init() {
1259 if (!UseLinuxPosixThreadCPUClocks) {
1260 return;
1261 }
1262 clockid_t clockid;
1263 struct timespec tp;
1264 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1265 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1266
1267 // Switch to using fast clocks for thread cpu time if
1268 // the sys_clock_getres() returns 0 error code.
1269 // Note, that some kernels may support the current thread
1270 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1271 // returned by the pthread_getcpuclockid().
1272 // If the fast Posix clocks are supported then the sys_clock_getres()
1273 // must return at least tp.tv_sec == 0 which means a resolution
1274 // better than 1 sec. This is extra check for reliability.
1275
1276 if (pthread_getcpuclockid_func &&
1277 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1278 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1279 _supports_fast_thread_cpu_time = true;
1280 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1281 }
1282 }
1283
1284 jlong os::javaTimeNanos() {
1285 if (os::supports_monotonic_clock()) {
1286 struct timespec tp;
1287 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1288 assert(status == 0, "gettime error");
1289 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1290 return result;
1291 } else {
1292 timeval time;
1293 int status = gettimeofday(&time, NULL);
1294 assert(status != -1, "linux error");
1295 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1296 return 1000 * usecs;
1297 }
1298 }
1299
1300 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1301 if (os::supports_monotonic_clock()) {
1302 info_ptr->max_value = ALL_64_BITS;
1303
1304 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1305 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1306 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1307 } else {
1308 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1309 info_ptr->max_value = ALL_64_BITS;
1310
1311 // gettimeofday is a real time clock so it skips
1312 info_ptr->may_skip_backward = true;
1313 info_ptr->may_skip_forward = true;
1314 }
1315
1316 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1317 }
1318
1319 // Return the real, user, and system times in seconds from an
1320 // arbitrary fixed point in the past.
1321 bool os::getTimesSecs(double* process_real_time,
1322 double* process_user_time,
1323 double* process_system_time) {
1324 struct tms ticks;
1325 clock_t real_ticks = times(&ticks);
1326
1327 if (real_ticks == (clock_t) (-1)) {
1328 return false;
1329 } else {
1330 double ticks_per_second = (double) clock_tics_per_sec;
1331 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1332 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1333 *process_real_time = ((double) real_ticks) / ticks_per_second;
1334
1335 return true;
1336 }
1337 }
1338
1339
1340 char * os::local_time_string(char *buf, size_t buflen) {
1341 struct tm t;
1342 time_t long_time;
1343 time(&long_time);
1344 localtime_r(&long_time, &t);
1345 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1346 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1347 t.tm_hour, t.tm_min, t.tm_sec);
1348 return buf;
1349 }
1350
1351 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1352 return localtime_r(clock, res);
1353 }
1354
1355 ////////////////////////////////////////////////////////////////////////////////
1356 // runtime exit support
1357
1358 // Note: os::shutdown() might be called very early during initialization, or
1359 // called from signal handler. Before adding something to os::shutdown(), make
1360 // sure it is async-safe and can handle partially initialized VM.
1361 void os::shutdown() {
1362
1363 // allow PerfMemory to attempt cleanup of any persistent resources
1364 perfMemory_exit();
1365
1366 // needs to remove object in file system
1367 AttachListener::abort();
1368
1369 // flush buffered output, finish log files
1370 ostream_abort();
1371
1372 // Check for abort hook
1373 abort_hook_t abort_hook = Arguments::abort_hook();
1374 if (abort_hook != NULL) {
1375 abort_hook();
1376 }
1377
1378 }
1379
1380 // Note: os::abort() might be called very early during initialization, or
1381 // called from signal handler. Before adding something to os::abort(), make
1382 // sure it is async-safe and can handle partially initialized VM.
1383 void os::abort(bool dump_core, void* siginfo, void* context) {
1384 os::shutdown();
1385 if (dump_core) {
1386 #ifndef PRODUCT
1387 fdStream out(defaultStream::output_fd());
1388 out.print_raw("Current thread is ");
1389 char buf[16];
1390 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1391 out.print_raw_cr(buf);
1392 out.print_raw_cr("Dumping core ...");
1393 #endif
1394 ::abort(); // dump core
1395 }
1396
1397 ::exit(1);
1398 }
1399
1400 // Die immediately, no exit hook, no abort hook, no cleanup.
1401 void os::die() {
1402 ::abort();
1403 }
1404
1405
1406 // This method is a copy of JDK's sysGetLastErrorString
1407 // from src/solaris/hpi/src/system_md.c
1408
1409 size_t os::lasterror(char *buf, size_t len) {
1410 if (errno == 0) return 0;
1411
1412 const char *s = ::strerror(errno);
1413 size_t n = ::strlen(s);
1414 if (n >= len) {
1415 n = len - 1;
1416 }
1417 ::strncpy(buf, s, n);
1418 buf[n] = '\0';
1419 return n;
1420 }
1421
1422 // thread_id is kernel thread id (similar to Solaris LWP id)
1423 intx os::current_thread_id() { return os::Linux::gettid(); }
1424 int os::current_process_id() {
1425 return ::getpid();
1426 }
1427
1428 // DLL functions
1429
1430 const char* os::dll_file_extension() { return ".so"; }
1431
1432 // This must be hard coded because it's the system's temporary
1433 // directory not the java application's temp directory, ala java.io.tmpdir.
1434 const char* os::get_temp_directory() { return "/tmp"; }
1435
1436 static bool file_exists(const char* filename) {
1437 struct stat statbuf;
1438 if (filename == NULL || strlen(filename) == 0) {
1439 return false;
1440 }
1441 return os::stat(filename, &statbuf) == 0;
1442 }
1443
1444 bool os::dll_build_name(char* buffer, size_t buflen,
1445 const char* pname, const char* fname) {
1446 bool retval = false;
1447 // Copied from libhpi
1448 const size_t pnamelen = pname ? strlen(pname) : 0;
1449
1450 // Return error on buffer overflow.
1451 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1452 return retval;
1453 }
1454
1455 if (pnamelen == 0) {
1456 snprintf(buffer, buflen, "lib%s.so", fname);
1457 retval = true;
1458 } else if (strchr(pname, *os::path_separator()) != NULL) {
1459 int n;
1460 char** pelements = split_path(pname, &n);
1461 if (pelements == NULL) {
1462 return false;
1463 }
1464 for (int i = 0; i < n; i++) {
1465 // Really shouldn't be NULL, but check can't hurt
1466 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1467 continue; // skip the empty path values
1468 }
1469 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1470 if (file_exists(buffer)) {
1471 retval = true;
1472 break;
1473 }
1474 }
1475 // release the storage
1476 for (int i = 0; i < n; i++) {
1477 if (pelements[i] != NULL) {
1478 FREE_C_HEAP_ARRAY(char, pelements[i]);
1479 }
1480 }
1481 if (pelements != NULL) {
1482 FREE_C_HEAP_ARRAY(char*, pelements);
1483 }
1484 } else {
1485 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1486 retval = true;
1487 }
1488 return retval;
1489 }
1490
1491 // check if addr is inside libjvm.so
1492 bool os::address_is_in_vm(address addr) {
1493 static address libjvm_base_addr;
1494 Dl_info dlinfo;
1495
1496 if (libjvm_base_addr == NULL) {
1497 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1498 libjvm_base_addr = (address)dlinfo.dli_fbase;
1499 }
1500 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1501 }
1502
1503 if (dladdr((void *)addr, &dlinfo) != 0) {
1504 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1505 }
1506
1507 return false;
1508 }
1509
1510 bool os::dll_address_to_function_name(address addr, char *buf,
1511 int buflen, int *offset,
1512 bool demangle) {
1513 // buf is not optional, but offset is optional
1514 assert(buf != NULL, "sanity check");
1515
1516 Dl_info dlinfo;
1517
1518 if (dladdr((void*)addr, &dlinfo) != 0) {
1519 // see if we have a matching symbol
1520 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1521 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1522 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1523 }
1524 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1525 return true;
1526 }
1527 // no matching symbol so try for just file info
1528 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1529 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1530 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1531 return true;
1532 }
1533 }
1534 }
1535
1536 buf[0] = '\0';
1537 if (offset != NULL) *offset = -1;
1538 return false;
1539 }
1540
1541 struct _address_to_library_name {
1542 address addr; // input : memory address
1543 size_t buflen; // size of fname
1544 char* fname; // output: library name
1545 address base; // library base addr
1546 };
1547
1548 static int address_to_library_name_callback(struct dl_phdr_info *info,
1549 size_t size, void *data) {
1550 int i;
1551 bool found = false;
1552 address libbase = NULL;
1553 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1554
1555 // iterate through all loadable segments
1556 for (i = 0; i < info->dlpi_phnum; i++) {
1557 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1558 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1559 // base address of a library is the lowest address of its loaded
1560 // segments.
1561 if (libbase == NULL || libbase > segbase) {
1562 libbase = segbase;
1563 }
1564 // see if 'addr' is within current segment
1565 if (segbase <= d->addr &&
1566 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1567 found = true;
1568 }
1569 }
1570 }
1571
1572 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1573 // so dll_address_to_library_name() can fall through to use dladdr() which
1574 // can figure out executable name from argv[0].
1575 if (found && info->dlpi_name && info->dlpi_name[0]) {
1576 d->base = libbase;
1577 if (d->fname) {
1578 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1579 }
1580 return 1;
1581 }
1582 return 0;
1583 }
1584
1585 bool os::dll_address_to_library_name(address addr, char* buf,
1586 int buflen, int* offset) {
1587 // buf is not optional, but offset is optional
1588 assert(buf != NULL, "sanity check");
1589
1590 Dl_info dlinfo;
1591 struct _address_to_library_name data;
1592
1593 // There is a bug in old glibc dladdr() implementation that it could resolve
1594 // to wrong library name if the .so file has a base address != NULL. Here
1595 // we iterate through the program headers of all loaded libraries to find
1596 // out which library 'addr' really belongs to. This workaround can be
1597 // removed once the minimum requirement for glibc is moved to 2.3.x.
1598 data.addr = addr;
1599 data.fname = buf;
1600 data.buflen = buflen;
1601 data.base = NULL;
1602 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1603
1604 if (rslt) {
1605 // buf already contains library name
1606 if (offset) *offset = addr - data.base;
1607 return true;
1608 }
1609 if (dladdr((void*)addr, &dlinfo) != 0) {
1610 if (dlinfo.dli_fname != NULL) {
1611 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1612 }
1613 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1614 *offset = addr - (address)dlinfo.dli_fbase;
1615 }
1616 return true;
1617 }
1618
1619 buf[0] = '\0';
1620 if (offset) *offset = -1;
1621 return false;
1622 }
1623
1624 // Loads .dll/.so and
1625 // in case of error it checks if .dll/.so was built for the
1626 // same architecture as Hotspot is running on
1627
1628
1629 // Remember the stack's state. The Linux dynamic linker will change
1630 // the stack to 'executable' at most once, so we must safepoint only once.
1631 bool os::Linux::_stack_is_executable = false;
1632
1633 // VM operation that loads a library. This is necessary if stack protection
1634 // of the Java stacks can be lost during loading the library. If we
1635 // do not stop the Java threads, they can stack overflow before the stacks
1636 // are protected again.
1637 class VM_LinuxDllLoad: public VM_Operation {
1638 private:
1639 const char *_filename;
1640 char *_ebuf;
1641 int _ebuflen;
1642 void *_lib;
1643 public:
1644 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1645 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1646 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1647 void doit() {
1648 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1649 os::Linux::_stack_is_executable = true;
1650 }
1651 void* loaded_library() { return _lib; }
1652 };
1653
1654 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1655 void * result = NULL;
1656 bool load_attempted = false;
1657
1658 // Check whether the library to load might change execution rights
1659 // of the stack. If they are changed, the protection of the stack
1660 // guard pages will be lost. We need a safepoint to fix this.
1661 //
1662 // See Linux man page execstack(8) for more info.
1663 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1664 ElfFile ef(filename);
1665 if (!ef.specifies_noexecstack()) {
1666 if (!is_init_completed()) {
1667 os::Linux::_stack_is_executable = true;
1668 // This is OK - No Java threads have been created yet, and hence no
1669 // stack guard pages to fix.
1670 //
1671 // This should happen only when you are building JDK7 using a very
1672 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1673 //
1674 // Dynamic loader will make all stacks executable after
1675 // this function returns, and will not do that again.
1676 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1677 } else {
1678 warning("You have loaded library %s which might have disabled stack guard. "
1679 "The VM will try to fix the stack guard now.\n"
1680 "It's highly recommended that you fix the library with "
1681 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1682 filename);
1683
1684 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1685 JavaThread *jt = JavaThread::current();
1686 if (jt->thread_state() != _thread_in_native) {
1687 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1688 // that requires ExecStack. Cannot enter safe point. Let's give up.
1689 warning("Unable to fix stack guard. Giving up.");
1690 } else {
1691 if (!LoadExecStackDllInVMThread) {
1692 // This is for the case where the DLL has an static
1693 // constructor function that executes JNI code. We cannot
1694 // load such DLLs in the VMThread.
1695 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1696 }
1697
1698 ThreadInVMfromNative tiv(jt);
1699 debug_only(VMNativeEntryWrapper vew;)
1700
1701 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1702 VMThread::execute(&op);
1703 if (LoadExecStackDllInVMThread) {
1704 result = op.loaded_library();
1705 }
1706 load_attempted = true;
1707 }
1708 }
1709 }
1710 }
1711
1712 if (!load_attempted) {
1713 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1714 }
1715
1716 if (result != NULL) {
1717 // Successful loading
1718 return result;
1719 }
1720
1721 Elf32_Ehdr elf_head;
1722 int diag_msg_max_length=ebuflen-strlen(ebuf);
1723 char* diag_msg_buf=ebuf+strlen(ebuf);
1724
1725 if (diag_msg_max_length==0) {
1726 // No more space in ebuf for additional diagnostics message
1727 return NULL;
1728 }
1729
1730
1731 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1732
1733 if (file_descriptor < 0) {
1734 // Can't open library, report dlerror() message
1735 return NULL;
1736 }
1737
1738 bool failed_to_read_elf_head=
1739 (sizeof(elf_head)!=
1740 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1741
1742 ::close(file_descriptor);
1743 if (failed_to_read_elf_head) {
1744 // file i/o error - report dlerror() msg
1745 return NULL;
1746 }
1747
1748 typedef struct {
1749 Elf32_Half code; // Actual value as defined in elf.h
1750 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1751 char elf_class; // 32 or 64 bit
1752 char endianess; // MSB or LSB
1753 char* name; // String representation
1754 } arch_t;
1755
1756 #ifndef EM_486
1757 #define EM_486 6 /* Intel 80486 */
1758 #endif
1759 #ifndef EM_AARCH64
1760 #define EM_AARCH64 183 /* ARM AARCH64 */
1761 #endif
1762
1763 static const arch_t arch_array[]={
1764 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1765 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1766 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1767 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1768 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1769 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1770 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1771 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1772 #if defined(VM_LITTLE_ENDIAN)
1773 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1774 #else
1775 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1776 #endif
1777 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1778 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1779 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1780 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1781 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1782 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1783 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1784 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1785 };
1786
1787 #if (defined IA32)
1788 static Elf32_Half running_arch_code=EM_386;
1789 #elif (defined AMD64)
1790 static Elf32_Half running_arch_code=EM_X86_64;
1791 #elif (defined IA64)
1792 static Elf32_Half running_arch_code=EM_IA_64;
1793 #elif (defined __sparc) && (defined _LP64)
1794 static Elf32_Half running_arch_code=EM_SPARCV9;
1795 #elif (defined __sparc) && (!defined _LP64)
1796 static Elf32_Half running_arch_code=EM_SPARC;
1797 #elif (defined __powerpc64__)
1798 static Elf32_Half running_arch_code=EM_PPC64;
1799 #elif (defined __powerpc__)
1800 static Elf32_Half running_arch_code=EM_PPC;
1801 #elif (defined ARM)
1802 static Elf32_Half running_arch_code=EM_ARM;
1803 #elif (defined S390)
1804 static Elf32_Half running_arch_code=EM_S390;
1805 #elif (defined ALPHA)
1806 static Elf32_Half running_arch_code=EM_ALPHA;
1807 #elif (defined MIPSEL)
1808 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1809 #elif (defined PARISC)
1810 static Elf32_Half running_arch_code=EM_PARISC;
1811 #elif (defined MIPS)
1812 static Elf32_Half running_arch_code=EM_MIPS;
1813 #elif (defined M68K)
1814 static Elf32_Half running_arch_code=EM_68K;
1815 #elif (defined AARCH64)
1816 static Elf32_Half running_arch_code=EM_AARCH64;
1817 #else
1818 #error Method os::dll_load requires that one of following is defined:\
1819 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1820 #endif
1821
1822 // Identify compatability class for VM's architecture and library's architecture
1823 // Obtain string descriptions for architectures
1824
1825 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1826 int running_arch_index=-1;
1827
1828 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1829 if (running_arch_code == arch_array[i].code) {
1830 running_arch_index = i;
1831 }
1832 if (lib_arch.code == arch_array[i].code) {
1833 lib_arch.compat_class = arch_array[i].compat_class;
1834 lib_arch.name = arch_array[i].name;
1835 }
1836 }
1837
1838 assert(running_arch_index != -1,
1839 "Didn't find running architecture code (running_arch_code) in arch_array");
1840 if (running_arch_index == -1) {
1841 // Even though running architecture detection failed
1842 // we may still continue with reporting dlerror() message
1843 return NULL;
1844 }
1845
1846 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1847 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1848 return NULL;
1849 }
1850
1851 #ifndef S390
1852 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1853 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1854 return NULL;
1855 }
1856 #endif // !S390
1857
1858 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1859 if (lib_arch.name!=NULL) {
1860 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1861 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1862 lib_arch.name, arch_array[running_arch_index].name);
1863 } else {
1864 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1865 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1866 lib_arch.code,
1867 arch_array[running_arch_index].name);
1868 }
1869 }
1870
1871 return NULL;
1872 }
1873
1874 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1875 int ebuflen) {
1876 void * result = ::dlopen(filename, RTLD_LAZY);
1877 if (result == NULL) {
1878 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1879 ebuf[ebuflen-1] = '\0';
1880 }
1881 return result;
1882 }
1883
1884 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1885 int ebuflen) {
1886 void * result = NULL;
1887 if (LoadExecStackDllInVMThread) {
1888 result = dlopen_helper(filename, ebuf, ebuflen);
1889 }
1890
1891 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1892 // library that requires an executable stack, or which does not have this
1893 // stack attribute set, dlopen changes the stack attribute to executable. The
1894 // read protection of the guard pages gets lost.
1895 //
1896 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1897 // may have been queued at the same time.
1898
1899 if (!_stack_is_executable) {
1900 JavaThread *jt = Threads::first();
1901
1902 while (jt) {
1903 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1904 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
1905 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1906 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1907 warning("Attempt to reguard stack yellow zone failed.");
1908 }
1909 }
1910 jt = jt->next();
1911 }
1912 }
1913
1914 return result;
1915 }
1916
1917 void* os::dll_lookup(void* handle, const char* name) {
1918 void* res = dlsym(handle, name);
1919 return res;
1920 }
1921
1922 void* os::get_default_process_handle() {
1923 return (void*)::dlopen(NULL, RTLD_LAZY);
1924 }
1925
1926 static bool _print_ascii_file(const char* filename, outputStream* st) {
1927 int fd = ::open(filename, O_RDONLY);
1928 if (fd == -1) {
1929 return false;
1930 }
1931
1932 char buf[32];
1933 int bytes;
1934 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1935 st->print_raw(buf, bytes);
1936 }
1937
1938 ::close(fd);
1939
1940 return true;
1941 }
1942
1943 void os::print_dll_info(outputStream *st) {
1944 st->print_cr("Dynamic libraries:");
1945
1946 char fname[32];
1947 pid_t pid = os::Linux::gettid();
1948
1949 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1950
1951 if (!_print_ascii_file(fname, st)) {
1952 st->print("Can not get library information for pid = %d\n", pid);
1953 }
1954 }
1955
1956 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1957 FILE *procmapsFile = NULL;
1958
1959 // Open the procfs maps file for the current process
1960 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1961 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1962 char line[PATH_MAX + 100];
1963
1964 // Read line by line from 'file'
1965 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1966 u8 base, top, offset, inode;
1967 char permissions[5];
1968 char device[6];
1969 char name[PATH_MAX + 1];
1970
1971 // Parse fields from line
1972 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
1973
1974 // Filter by device id '00:00' so that we only get file system mapped files.
1975 if (strcmp(device, "00:00") != 0) {
1976
1977 // Call callback with the fields of interest
1978 if(callback(name, (address)base, (address)top, param)) {
1979 // Oops abort, callback aborted
1980 fclose(procmapsFile);
1981 return 1;
1982 }
1983 }
1984 }
1985 fclose(procmapsFile);
1986 }
1987 return 0;
1988 }
1989
1990 void os::print_os_info_brief(outputStream* st) {
1991 os::Linux::print_distro_info(st);
1992
1993 os::Posix::print_uname_info(st);
1994
1995 os::Linux::print_libversion_info(st);
1996
1997 }
1998
1999 void os::print_os_info(outputStream* st) {
2000 st->print("OS:");
2001
2002 os::Linux::print_distro_info(st);
2003
2004 os::Posix::print_uname_info(st);
2005
2006 // Print warning if unsafe chroot environment detected
2007 if (unsafe_chroot_detected) {
2008 st->print("WARNING!! ");
2009 st->print_cr("%s", unstable_chroot_error);
2010 }
2011
2012 os::Linux::print_libversion_info(st);
2013
2014 os::Posix::print_rlimit_info(st);
2015
2016 os::Posix::print_load_average(st);
2017
2018 os::Linux::print_full_memory_info(st);
2019 }
2020
2021 // Try to identify popular distros.
2022 // Most Linux distributions have a /etc/XXX-release file, which contains
2023 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2024 // file that also contains the OS version string. Some have more than one
2025 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2026 // /etc/redhat-release.), so the order is important.
2027 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2028 // their own specific XXX-release file as well as a redhat-release file.
2029 // Because of this the XXX-release file needs to be searched for before the
2030 // redhat-release file.
2031 // Since Red Hat has a lsb-release file that is not very descriptive the
2032 // search for redhat-release needs to be before lsb-release.
2033 // Since the lsb-release file is the new standard it needs to be searched
2034 // before the older style release files.
2035 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2036 // next to last resort. The os-release file is a new standard that contains
2037 // distribution information and the system-release file seems to be an old
2038 // standard that has been replaced by the lsb-release and os-release files.
2039 // Searching for the debian_version file is the last resort. It contains
2040 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2041 // "Debian " is printed before the contents of the debian_version file.
2042
2043 const char* distro_files[] = {
2044 "/etc/oracle-release",
2045 "/etc/mandriva-release",
2046 "/etc/mandrake-release",
2047 "/etc/sun-release",
2048 "/etc/redhat-release",
2049 "/etc/lsb-release",
2050 "/etc/SuSE-release",
2051 "/etc/turbolinux-release",
2052 "/etc/gentoo-release",
2053 "/etc/ltib-release",
2054 "/etc/angstrom-version",
2055 "/etc/system-release",
2056 "/etc/os-release",
2057 NULL };
2058
2059 void os::Linux::print_distro_info(outputStream* st) {
2060 for (int i = 0;; i++) {
2061 const char* file = distro_files[i];
2062 if (file == NULL) {
2063 break; // done
2064 }
2065 // If file prints, we found it.
2066 if (_print_ascii_file(file, st)) {
2067 return;
2068 }
2069 }
2070
2071 if (file_exists("/etc/debian_version")) {
2072 st->print("Debian ");
2073 _print_ascii_file("/etc/debian_version", st);
2074 } else {
2075 st->print("Linux");
2076 }
2077 st->cr();
2078 }
2079
2080 static void parse_os_info(char* distro, size_t length, const char* file) {
2081 FILE* fp = fopen(file, "r");
2082 if (fp != NULL) {
2083 char buf[256];
2084 // get last line of the file.
2085 while (fgets(buf, sizeof(buf), fp)) { }
2086 // Edit out extra stuff in expected ubuntu format
2087 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2088 char* ptr = strstr(buf, "\""); // the name is in quotes
2089 if (ptr != NULL) {
2090 ptr++; // go beyond first quote
2091 char* nl = strchr(ptr, '\"');
2092 if (nl != NULL) *nl = '\0';
2093 strncpy(distro, ptr, length);
2094 } else {
2095 ptr = strstr(buf, "=");
2096 ptr++; // go beyond equals then
2097 char* nl = strchr(ptr, '\n');
2098 if (nl != NULL) *nl = '\0';
2099 strncpy(distro, ptr, length);
2100 }
2101 } else {
2102 // if not in expected Ubuntu format, print out whole line minus \n
2103 char* nl = strchr(buf, '\n');
2104 if (nl != NULL) *nl = '\0';
2105 strncpy(distro, buf, length);
2106 }
2107 // close distro file
2108 fclose(fp);
2109 }
2110 }
2111
2112 void os::get_summary_os_info(char* buf, size_t buflen) {
2113 for (int i = 0;; i++) {
2114 const char* file = distro_files[i];
2115 if (file == NULL) {
2116 break; // ran out of distro_files
2117 }
2118 if (file_exists(file)) {
2119 parse_os_info(buf, buflen, file);
2120 return;
2121 }
2122 }
2123 // special case for debian
2124 if (file_exists("/etc/debian_version")) {
2125 strncpy(buf, "Debian ", buflen);
2126 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2127 } else {
2128 strncpy(buf, "Linux", buflen);
2129 }
2130 }
2131
2132 void os::Linux::print_libversion_info(outputStream* st) {
2133 // libc, pthread
2134 st->print("libc:");
2135 st->print("%s ", os::Linux::glibc_version());
2136 st->print("%s ", os::Linux::libpthread_version());
2137 st->cr();
2138 }
2139
2140 void os::Linux::print_full_memory_info(outputStream* st) {
2141 st->print("\n/proc/meminfo:\n");
2142 _print_ascii_file("/proc/meminfo", st);
2143 st->cr();
2144 }
2145
2146 void os::print_memory_info(outputStream* st) {
2147
2148 st->print("Memory:");
2149 st->print(" %dk page", os::vm_page_size()>>10);
2150
2151 // values in struct sysinfo are "unsigned long"
2152 struct sysinfo si;
2153 sysinfo(&si);
2154
2155 st->print(", physical " UINT64_FORMAT "k",
2156 os::physical_memory() >> 10);
2157 st->print("(" UINT64_FORMAT "k free)",
2158 os::available_memory() >> 10);
2159 st->print(", swap " UINT64_FORMAT "k",
2160 ((jlong)si.totalswap * si.mem_unit) >> 10);
2161 st->print("(" UINT64_FORMAT "k free)",
2162 ((jlong)si.freeswap * si.mem_unit) >> 10);
2163 st->cr();
2164 }
2165
2166 // Print the first "model name" line and the first "flags" line
2167 // that we find and nothing more. We assume "model name" comes
2168 // before "flags" so if we find a second "model name", then the
2169 // "flags" field is considered missing.
2170 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2171 #if defined(IA32) || defined(AMD64)
2172 // Other platforms have less repetitive cpuinfo files
2173 FILE *fp = fopen("/proc/cpuinfo", "r");
2174 if (fp) {
2175 while (!feof(fp)) {
2176 if (fgets(buf, buflen, fp)) {
2177 // Assume model name comes before flags
2178 bool model_name_printed = false;
2179 if (strstr(buf, "model name") != NULL) {
2180 if (!model_name_printed) {
2181 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2182 st->print_raw(buf);
2183 model_name_printed = true;
2184 } else {
2185 // model name printed but not flags? Odd, just return
2186 fclose(fp);
2187 return true;
2188 }
2189 }
2190 // print the flags line too
2191 if (strstr(buf, "flags") != NULL) {
2192 st->print_raw(buf);
2193 fclose(fp);
2194 return true;
2195 }
2196 }
2197 }
2198 fclose(fp);
2199 }
2200 #endif // x86 platforms
2201 return false;
2202 }
2203
2204 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2205 // Only print the model name if the platform provides this as a summary
2206 if (!print_model_name_and_flags(st, buf, buflen)) {
2207 st->print("\n/proc/cpuinfo:\n");
2208 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2209 st->print_cr(" <Not Available>");
2210 }
2211 }
2212 }
2213
2214 const char* search_string = IA32_ONLY("model name") AMD64_ONLY("model name")
2215 IA64_ONLY("") SPARC_ONLY("cpu")
2216 ARM32_ONLY("Processor") PPC_ONLY("Processor") AARCH64_ONLY("Processor");
2217
2218 // Parses the cpuinfo file for string representing the model name.
2219 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2220 FILE* fp = fopen("/proc/cpuinfo", "r");
2221 if (fp != NULL) {
2222 while (!feof(fp)) {
2223 char buf[256];
2224 if (fgets(buf, sizeof(buf), fp)) {
2225 char* start = strstr(buf, search_string);
2226 if (start != NULL) {
2227 char *ptr = start + strlen(search_string);
2228 char *end = buf + strlen(buf);
2229 while (ptr != end) {
2230 // skip whitespace and colon for the rest of the name.
2231 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2232 break;
2233 }
2234 ptr++;
2235 }
2236 if (ptr != end) {
2237 // reasonable string, get rid of newline and keep the rest
2238 char* nl = strchr(buf, '\n');
2239 if (nl != NULL) *nl = '\0';
2240 strncpy(cpuinfo, ptr, length);
2241 fclose(fp);
2242 return;
2243 }
2244 }
2245 }
2246 }
2247 fclose(fp);
2248 }
2249 // cpuinfo not found or parsing failed, just print generic string. The entire
2250 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2251 strncpy(cpuinfo, IA32_ONLY("x86_32") AMD64_ONLY("x86_32")
2252 IA64_ONLY("IA64") SPARC_ONLY("sparcv9")
2253 ARM32_ONLY("ARM") PPC_ONLY("PPC64") AARCH64_ONLY("AArch64"), length);
2254 }
2255
2256 void os::print_siginfo(outputStream* st, void* siginfo) {
2257 const siginfo_t* si = (const siginfo_t*)siginfo;
2258
2259 os::Posix::print_siginfo_brief(st, si);
2260 #if INCLUDE_CDS
2261 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2262 UseSharedSpaces) {
2263 FileMapInfo* mapinfo = FileMapInfo::current_info();
2264 if (mapinfo->is_in_shared_space(si->si_addr)) {
2265 st->print("\n\nError accessing class data sharing archive." \
2266 " Mapped file inaccessible during execution, " \
2267 " possible disk/network problem.");
2268 }
2269 }
2270 #endif
2271 st->cr();
2272 }
2273
2274
2275 static void print_signal_handler(outputStream* st, int sig,
2276 char* buf, size_t buflen);
2277
2278 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2279 st->print_cr("Signal Handlers:");
2280 print_signal_handler(st, SIGSEGV, buf, buflen);
2281 print_signal_handler(st, SIGBUS , buf, buflen);
2282 print_signal_handler(st, SIGFPE , buf, buflen);
2283 print_signal_handler(st, SIGPIPE, buf, buflen);
2284 print_signal_handler(st, SIGXFSZ, buf, buflen);
2285 print_signal_handler(st, SIGILL , buf, buflen);
2286 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2287 print_signal_handler(st, SR_signum, buf, buflen);
2288 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2289 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2290 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2291 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2292 #if defined(PPC64)
2293 print_signal_handler(st, SIGTRAP, buf, buflen);
2294 #endif
2295 }
2296
2297 static char saved_jvm_path[MAXPATHLEN] = {0};
2298
2299 // Find the full path to the current module, libjvm.so
2300 void os::jvm_path(char *buf, jint buflen) {
2301 // Error checking.
2302 if (buflen < MAXPATHLEN) {
2303 assert(false, "must use a large-enough buffer");
2304 buf[0] = '\0';
2305 return;
2306 }
2307 // Lazy resolve the path to current module.
2308 if (saved_jvm_path[0] != 0) {
2309 strcpy(buf, saved_jvm_path);
2310 return;
2311 }
2312
2313 char dli_fname[MAXPATHLEN];
2314 bool ret = dll_address_to_library_name(
2315 CAST_FROM_FN_PTR(address, os::jvm_path),
2316 dli_fname, sizeof(dli_fname), NULL);
2317 assert(ret, "cannot locate libjvm");
2318 char *rp = NULL;
2319 if (ret && dli_fname[0] != '\0') {
2320 rp = realpath(dli_fname, buf);
2321 }
2322 if (rp == NULL) {
2323 return;
2324 }
2325
2326 if (Arguments::sun_java_launcher_is_altjvm()) {
2327 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2328 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2329 // If "/jre/lib/" appears at the right place in the string, then
2330 // assume we are installed in a JDK and we're done. Otherwise, check
2331 // for a JAVA_HOME environment variable and fix up the path so it
2332 // looks like libjvm.so is installed there (append a fake suffix
2333 // hotspot/libjvm.so).
2334 const char *p = buf + strlen(buf) - 1;
2335 for (int count = 0; p > buf && count < 5; ++count) {
2336 for (--p; p > buf && *p != '/'; --p)
2337 /* empty */ ;
2338 }
2339
2340 if (strncmp(p, "/jre/lib/", 9) != 0) {
2341 // Look for JAVA_HOME in the environment.
2342 char* java_home_var = ::getenv("JAVA_HOME");
2343 if (java_home_var != NULL && java_home_var[0] != 0) {
2344 char* jrelib_p;
2345 int len;
2346
2347 // Check the current module name "libjvm.so".
2348 p = strrchr(buf, '/');
2349 if (p == NULL) {
2350 return;
2351 }
2352 assert(strstr(p, "/libjvm") == p, "invalid library name");
2353
2354 rp = realpath(java_home_var, buf);
2355 if (rp == NULL) {
2356 return;
2357 }
2358
2359 // determine if this is a legacy image or modules image
2360 // modules image doesn't have "jre" subdirectory
2361 len = strlen(buf);
2362 assert(len < buflen, "Ran out of buffer room");
2363 jrelib_p = buf + len;
2364 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2365 if (0 != access(buf, F_OK)) {
2366 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2367 }
2368
2369 if (0 == access(buf, F_OK)) {
2370 // Use current module name "libjvm.so"
2371 len = strlen(buf);
2372 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2373 } else {
2374 // Go back to path of .so
2375 rp = realpath(dli_fname, buf);
2376 if (rp == NULL) {
2377 return;
2378 }
2379 }
2380 }
2381 }
2382 }
2383
2384 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2385 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2386 }
2387
2388 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2389 // no prefix required, not even "_"
2390 }
2391
2392 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2393 // no suffix required
2394 }
2395
2396 ////////////////////////////////////////////////////////////////////////////////
2397 // sun.misc.Signal support
2398
2399 static volatile jint sigint_count = 0;
2400
2401 static void UserHandler(int sig, void *siginfo, void *context) {
2402 // 4511530 - sem_post is serialized and handled by the manager thread. When
2403 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2404 // don't want to flood the manager thread with sem_post requests.
2405 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2406 return;
2407 }
2408
2409 // Ctrl-C is pressed during error reporting, likely because the error
2410 // handler fails to abort. Let VM die immediately.
2411 if (sig == SIGINT && is_error_reported()) {
2412 os::die();
2413 }
2414
2415 os::signal_notify(sig);
2416 }
2417
2418 void* os::user_handler() {
2419 return CAST_FROM_FN_PTR(void*, UserHandler);
2420 }
2421
2422 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2423 struct timespec ts;
2424 // Semaphore's are always associated with CLOCK_REALTIME
2425 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2426 // see unpackTime for discussion on overflow checking
2427 if (sec >= MAX_SECS) {
2428 ts.tv_sec += MAX_SECS;
2429 ts.tv_nsec = 0;
2430 } else {
2431 ts.tv_sec += sec;
2432 ts.tv_nsec += nsec;
2433 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2434 ts.tv_nsec -= NANOSECS_PER_SEC;
2435 ++ts.tv_sec; // note: this must be <= max_secs
2436 }
2437 }
2438
2439 return ts;
2440 }
2441
2442 extern "C" {
2443 typedef void (*sa_handler_t)(int);
2444 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2445 }
2446
2447 void* os::signal(int signal_number, void* handler) {
2448 struct sigaction sigAct, oldSigAct;
2449
2450 sigfillset(&(sigAct.sa_mask));
2451 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2452 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2453
2454 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2455 // -1 means registration failed
2456 return (void *)-1;
2457 }
2458
2459 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2460 }
2461
2462 void os::signal_raise(int signal_number) {
2463 ::raise(signal_number);
2464 }
2465
2466 // The following code is moved from os.cpp for making this
2467 // code platform specific, which it is by its very nature.
2468
2469 // Will be modified when max signal is changed to be dynamic
2470 int os::sigexitnum_pd() {
2471 return NSIG;
2472 }
2473
2474 // a counter for each possible signal value
2475 static volatile jint pending_signals[NSIG+1] = { 0 };
2476
2477 // Linux(POSIX) specific hand shaking semaphore.
2478 static sem_t sig_sem;
2479 static PosixSemaphore sr_semaphore;
2480
2481 void os::signal_init_pd() {
2482 // Initialize signal structures
2483 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2484
2485 // Initialize signal semaphore
2486 ::sem_init(&sig_sem, 0, 0);
2487 }
2488
2489 void os::signal_notify(int sig) {
2490 Atomic::inc(&pending_signals[sig]);
2491 ::sem_post(&sig_sem);
2492 }
2493
2494 static int check_pending_signals(bool wait) {
2495 Atomic::store(0, &sigint_count);
2496 for (;;) {
2497 for (int i = 0; i < NSIG + 1; i++) {
2498 jint n = pending_signals[i];
2499 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2500 return i;
2501 }
2502 }
2503 if (!wait) {
2504 return -1;
2505 }
2506 JavaThread *thread = JavaThread::current();
2507 ThreadBlockInVM tbivm(thread);
2508
2509 bool threadIsSuspended;
2510 do {
2511 thread->set_suspend_equivalent();
2512 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2513 ::sem_wait(&sig_sem);
2514
2515 // were we externally suspended while we were waiting?
2516 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2517 if (threadIsSuspended) {
2518 // The semaphore has been incremented, but while we were waiting
2519 // another thread suspended us. We don't want to continue running
2520 // while suspended because that would surprise the thread that
2521 // suspended us.
2522 ::sem_post(&sig_sem);
2523
2524 thread->java_suspend_self();
2525 }
2526 } while (threadIsSuspended);
2527 }
2528 }
2529
2530 int os::signal_lookup() {
2531 return check_pending_signals(false);
2532 }
2533
2534 int os::signal_wait() {
2535 return check_pending_signals(true);
2536 }
2537
2538 ////////////////////////////////////////////////////////////////////////////////
2539 // Virtual Memory
2540
2541 int os::vm_page_size() {
2542 // Seems redundant as all get out
2543 assert(os::Linux::page_size() != -1, "must call os::init");
2544 return os::Linux::page_size();
2545 }
2546
2547 // Solaris allocates memory by pages.
2548 int os::vm_allocation_granularity() {
2549 assert(os::Linux::page_size() != -1, "must call os::init");
2550 return os::Linux::page_size();
2551 }
2552
2553 // Rationale behind this function:
2554 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2555 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2556 // samples for JITted code. Here we create private executable mapping over the code cache
2557 // and then we can use standard (well, almost, as mapping can change) way to provide
2558 // info for the reporting script by storing timestamp and location of symbol
2559 void linux_wrap_code(char* base, size_t size) {
2560 static volatile jint cnt = 0;
2561
2562 if (!UseOprofile) {
2563 return;
2564 }
2565
2566 char buf[PATH_MAX+1];
2567 int num = Atomic::add(1, &cnt);
2568
2569 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2570 os::get_temp_directory(), os::current_process_id(), num);
2571 unlink(buf);
2572
2573 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2574
2575 if (fd != -1) {
2576 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2577 if (rv != (off_t)-1) {
2578 if (::write(fd, "", 1) == 1) {
2579 mmap(base, size,
2580 PROT_READ|PROT_WRITE|PROT_EXEC,
2581 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2582 }
2583 }
2584 ::close(fd);
2585 unlink(buf);
2586 }
2587 }
2588
2589 static bool recoverable_mmap_error(int err) {
2590 // See if the error is one we can let the caller handle. This
2591 // list of errno values comes from JBS-6843484. I can't find a
2592 // Linux man page that documents this specific set of errno
2593 // values so while this list currently matches Solaris, it may
2594 // change as we gain experience with this failure mode.
2595 switch (err) {
2596 case EBADF:
2597 case EINVAL:
2598 case ENOTSUP:
2599 // let the caller deal with these errors
2600 return true;
2601
2602 default:
2603 // Any remaining errors on this OS can cause our reserved mapping
2604 // to be lost. That can cause confusion where different data
2605 // structures think they have the same memory mapped. The worst
2606 // scenario is if both the VM and a library think they have the
2607 // same memory mapped.
2608 return false;
2609 }
2610 }
2611
2612 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2613 int err) {
2614 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2616 strerror(err), err);
2617 }
2618
2619 static void warn_fail_commit_memory(char* addr, size_t size,
2620 size_t alignment_hint, bool exec,
2621 int err) {
2622 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2624 alignment_hint, exec, strerror(err), err);
2625 }
2626
2627 // NOTE: Linux kernel does not really reserve the pages for us.
2628 // All it does is to check if there are enough free pages
2629 // left at the time of mmap(). This could be a potential
2630 // problem.
2631 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2632 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2633 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2634 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2635 if (res != (uintptr_t) MAP_FAILED) {
2636 if (UseNUMAInterleaving) {
2637 numa_make_global(addr, size);
2638 }
2639 return 0;
2640 }
2641
2642 int err = errno; // save errno from mmap() call above
2643
2644 if (!recoverable_mmap_error(err)) {
2645 warn_fail_commit_memory(addr, size, exec, err);
2646 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2647 }
2648
2649 return err;
2650 }
2651
2652 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2653 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2654 }
2655
2656 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2657 const char* mesg) {
2658 assert(mesg != NULL, "mesg must be specified");
2659 int err = os::Linux::commit_memory_impl(addr, size, exec);
2660 if (err != 0) {
2661 // the caller wants all commit errors to exit with the specified mesg:
2662 warn_fail_commit_memory(addr, size, exec, err);
2663 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2664 }
2665 }
2666
2667 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2668 #ifndef MAP_HUGETLB
2669 #define MAP_HUGETLB 0x40000
2670 #endif
2671
2672 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2673 #ifndef MADV_HUGEPAGE
2674 #define MADV_HUGEPAGE 14
2675 #endif
2676
2677 int os::Linux::commit_memory_impl(char* addr, size_t size,
2678 size_t alignment_hint, bool exec) {
2679 int err = os::Linux::commit_memory_impl(addr, size, exec);
2680 if (err == 0) {
2681 realign_memory(addr, size, alignment_hint);
2682 }
2683 return err;
2684 }
2685
2686 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2687 bool exec) {
2688 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2689 }
2690
2691 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2692 size_t alignment_hint, bool exec,
2693 const char* mesg) {
2694 assert(mesg != NULL, "mesg must be specified");
2695 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2696 if (err != 0) {
2697 // the caller wants all commit errors to exit with the specified mesg:
2698 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2699 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2700 }
2701 }
2702
2703 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2704 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2705 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2706 // be supported or the memory may already be backed by huge pages.
2707 ::madvise(addr, bytes, MADV_HUGEPAGE);
2708 }
2709 }
2710
2711 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2712 // This method works by doing an mmap over an existing mmaping and effectively discarding
2713 // the existing pages. However it won't work for SHM-based large pages that cannot be
2714 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2715 // small pages on top of the SHM segment. This method always works for small pages, so we
2716 // allow that in any case.
2717 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2718 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2719 }
2720 }
2721
2722 void os::numa_make_global(char *addr, size_t bytes) {
2723 Linux::numa_interleave_memory(addr, bytes);
2724 }
2725
2726 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2727 // bind policy to MPOL_PREFERRED for the current thread.
2728 #define USE_MPOL_PREFERRED 0
2729
2730 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2731 // To make NUMA and large pages more robust when both enabled, we need to ease
2732 // the requirements on where the memory should be allocated. MPOL_BIND is the
2733 // default policy and it will force memory to be allocated on the specified
2734 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2735 // the specified node, but will not force it. Using this policy will prevent
2736 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2737 // free large pages.
2738 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2739 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2740 }
2741
2742 bool os::numa_topology_changed() { return false; }
2743
2744 size_t os::numa_get_groups_num() {
2745 int max_node = Linux::numa_max_node();
2746 return max_node > 0 ? max_node + 1 : 1;
2747 }
2748
2749 int os::numa_get_group_id() {
2750 int cpu_id = Linux::sched_getcpu();
2751 if (cpu_id != -1) {
2752 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2753 if (lgrp_id != -1) {
2754 return lgrp_id;
2755 }
2756 }
2757 return 0;
2758 }
2759
2760 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2761 for (size_t i = 0; i < size; i++) {
2762 ids[i] = i;
2763 }
2764 return size;
2765 }
2766
2767 bool os::get_page_info(char *start, page_info* info) {
2768 return false;
2769 }
2770
2771 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2772 page_info* page_found) {
2773 return end;
2774 }
2775
2776
2777 int os::Linux::sched_getcpu_syscall(void) {
2778 unsigned int cpu;
2779 int retval = -1;
2780
2781 #if defined(IA32)
2782 #ifndef SYS_getcpu
2783 #define SYS_getcpu 318
2784 #endif
2785 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2786 #elif defined(AMD64)
2787 // Unfortunately we have to bring all these macros here from vsyscall.h
2788 // to be able to compile on old linuxes.
2789 #define __NR_vgetcpu 2
2790 #define VSYSCALL_START (-10UL << 20)
2791 #define VSYSCALL_SIZE 1024
2792 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2793 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2794 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2795 retval = vgetcpu(&cpu, NULL, NULL);
2796 #endif
2797
2798 return (retval == -1) ? retval : cpu;
2799 }
2800
2801 // Something to do with the numa-aware allocator needs these symbols
2802 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2803 extern "C" JNIEXPORT void numa_error(char *where) { }
2804 extern "C" JNIEXPORT int fork1() { return fork(); }
2805
2806
2807 // If we are running with libnuma version > 2, then we should
2808 // be trying to use symbols with versions 1.1
2809 // If we are running with earlier version, which did not have symbol versions,
2810 // we should use the base version.
2811 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2812 void *f = dlvsym(handle, name, "libnuma_1.1");
2813 if (f == NULL) {
2814 f = dlsym(handle, name);
2815 }
2816 return f;
2817 }
2818
2819 bool os::Linux::libnuma_init() {
2820 // sched_getcpu() should be in libc.
2821 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2822 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2823
2824 // If it's not, try a direct syscall.
2825 if (sched_getcpu() == -1) {
2826 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2827 (void*)&sched_getcpu_syscall));
2828 }
2829
2830 if (sched_getcpu() != -1) { // Does it work?
2831 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2832 if (handle != NULL) {
2833 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2834 libnuma_dlsym(handle, "numa_node_to_cpus")));
2835 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2836 libnuma_dlsym(handle, "numa_max_node")));
2837 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2838 libnuma_dlsym(handle, "numa_available")));
2839 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2840 libnuma_dlsym(handle, "numa_tonode_memory")));
2841 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2842 libnuma_dlsym(handle, "numa_interleave_memory")));
2843 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2844 libnuma_dlsym(handle, "numa_set_bind_policy")));
2845
2846
2847 if (numa_available() != -1) {
2848 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2849 // Create a cpu -> node mapping
2850 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2851 rebuild_cpu_to_node_map();
2852 return true;
2853 }
2854 }
2855 }
2856 return false;
2857 }
2858
2859 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2860 // The table is later used in get_node_by_cpu().
2861 void os::Linux::rebuild_cpu_to_node_map() {
2862 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2863 // in libnuma (possible values are starting from 16,
2864 // and continuing up with every other power of 2, but less
2865 // than the maximum number of CPUs supported by kernel), and
2866 // is a subject to change (in libnuma version 2 the requirements
2867 // are more reasonable) we'll just hardcode the number they use
2868 // in the library.
2869 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2870
2871 size_t cpu_num = os::active_processor_count();
2872 size_t cpu_map_size = NCPUS / BitsPerCLong;
2873 size_t cpu_map_valid_size =
2874 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2875
2876 cpu_to_node()->clear();
2877 cpu_to_node()->at_grow(cpu_num - 1);
2878 size_t node_num = numa_get_groups_num();
2879
2880 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2881 for (size_t i = 0; i < node_num; i++) {
2882 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2883 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2884 if (cpu_map[j] != 0) {
2885 for (size_t k = 0; k < BitsPerCLong; k++) {
2886 if (cpu_map[j] & (1UL << k)) {
2887 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2888 }
2889 }
2890 }
2891 }
2892 }
2893 }
2894 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2895 }
2896
2897 int os::Linux::get_node_by_cpu(int cpu_id) {
2898 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2899 return cpu_to_node()->at(cpu_id);
2900 }
2901 return -1;
2902 }
2903
2904 GrowableArray<int>* os::Linux::_cpu_to_node;
2905 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2906 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2907 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2908 os::Linux::numa_available_func_t os::Linux::_numa_available;
2909 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2910 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2911 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2912 unsigned long* os::Linux::_numa_all_nodes;
2913
2914 bool os::pd_uncommit_memory(char* addr, size_t size) {
2915 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2916 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2917 return res != (uintptr_t) MAP_FAILED;
2918 }
2919
2920 static address get_stack_commited_bottom(address bottom, size_t size) {
2921 address nbot = bottom;
2922 address ntop = bottom + size;
2923
2924 size_t page_sz = os::vm_page_size();
2925 unsigned pages = size / page_sz;
2926
2927 unsigned char vec[1];
2928 unsigned imin = 1, imax = pages + 1, imid;
2929 int mincore_return_value = 0;
2930
2931 assert(imin <= imax, "Unexpected page size");
2932
2933 while (imin < imax) {
2934 imid = (imax + imin) / 2;
2935 nbot = ntop - (imid * page_sz);
2936
2937 // Use a trick with mincore to check whether the page is mapped or not.
2938 // mincore sets vec to 1 if page resides in memory and to 0 if page
2939 // is swapped output but if page we are asking for is unmapped
2940 // it returns -1,ENOMEM
2941 mincore_return_value = mincore(nbot, page_sz, vec);
2942
2943 if (mincore_return_value == -1) {
2944 // Page is not mapped go up
2945 // to find first mapped page
2946 if (errno != EAGAIN) {
2947 assert(errno == ENOMEM, "Unexpected mincore errno");
2948 imax = imid;
2949 }
2950 } else {
2951 // Page is mapped go down
2952 // to find first not mapped page
2953 imin = imid + 1;
2954 }
2955 }
2956
2957 nbot = nbot + page_sz;
2958
2959 // Adjust stack bottom one page up if last checked page is not mapped
2960 if (mincore_return_value == -1) {
2961 nbot = nbot + page_sz;
2962 }
2963
2964 return nbot;
2965 }
2966
2967
2968 // Linux uses a growable mapping for the stack, and if the mapping for
2969 // the stack guard pages is not removed when we detach a thread the
2970 // stack cannot grow beyond the pages where the stack guard was
2971 // mapped. If at some point later in the process the stack expands to
2972 // that point, the Linux kernel cannot expand the stack any further
2973 // because the guard pages are in the way, and a segfault occurs.
2974 //
2975 // However, it's essential not to split the stack region by unmapping
2976 // a region (leaving a hole) that's already part of the stack mapping,
2977 // so if the stack mapping has already grown beyond the guard pages at
2978 // the time we create them, we have to truncate the stack mapping.
2979 // So, we need to know the extent of the stack mapping when
2980 // create_stack_guard_pages() is called.
2981
2982 // We only need this for stacks that are growable: at the time of
2983 // writing thread stacks don't use growable mappings (i.e. those
2984 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2985 // only applies to the main thread.
2986
2987 // If the (growable) stack mapping already extends beyond the point
2988 // where we're going to put our guard pages, truncate the mapping at
2989 // that point by munmap()ping it. This ensures that when we later
2990 // munmap() the guard pages we don't leave a hole in the stack
2991 // mapping. This only affects the main/initial thread
2992
2993 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2994 if (os::Linux::is_initial_thread()) {
2995 // As we manually grow stack up to bottom inside create_attached_thread(),
2996 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2997 // we don't need to do anything special.
2998 // Check it first, before calling heavy function.
2999 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3000 unsigned char vec[1];
3001
3002 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3003 // Fallback to slow path on all errors, including EAGAIN
3004 stack_extent = (uintptr_t) get_stack_commited_bottom(
3005 os::Linux::initial_thread_stack_bottom(),
3006 (size_t)addr - stack_extent);
3007 }
3008
3009 if (stack_extent < (uintptr_t)addr) {
3010 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3011 }
3012 }
3013
3014 return os::commit_memory(addr, size, !ExecMem);
3015 }
3016
3017 // If this is a growable mapping, remove the guard pages entirely by
3018 // munmap()ping them. If not, just call uncommit_memory(). This only
3019 // affects the main/initial thread, but guard against future OS changes
3020 // It's safe to always unmap guard pages for initial thread because we
3021 // always place it right after end of the mapped region
3022
3023 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3024 uintptr_t stack_extent, stack_base;
3025
3026 if (os::Linux::is_initial_thread()) {
3027 return ::munmap(addr, size) == 0;
3028 }
3029
3030 return os::uncommit_memory(addr, size);
3031 }
3032
3033 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3034 // at 'requested_addr'. If there are existing memory mappings at the same
3035 // location, however, they will be overwritten. If 'fixed' is false,
3036 // 'requested_addr' is only treated as a hint, the return value may or
3037 // may not start from the requested address. Unlike Linux mmap(), this
3038 // function returns NULL to indicate failure.
3039 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3040 char * addr;
3041 int flags;
3042
3043 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3044 if (fixed) {
3045 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3046 flags |= MAP_FIXED;
3047 }
3048
3049 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3050 // touch an uncommitted page. Otherwise, the read/write might
3051 // succeed if we have enough swap space to back the physical page.
3052 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3053 flags, -1, 0);
3054
3055 return addr == MAP_FAILED ? NULL : addr;
3056 }
3057
3058 static int anon_munmap(char * addr, size_t size) {
3059 return ::munmap(addr, size) == 0;
3060 }
3061
3062 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3063 size_t alignment_hint) {
3064 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3065 }
3066
3067 bool os::pd_release_memory(char* addr, size_t size) {
3068 return anon_munmap(addr, size);
3069 }
3070
3071 static bool linux_mprotect(char* addr, size_t size, int prot) {
3072 // Linux wants the mprotect address argument to be page aligned.
3073 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3074
3075 // According to SUSv3, mprotect() should only be used with mappings
3076 // established by mmap(), and mmap() always maps whole pages. Unaligned
3077 // 'addr' likely indicates problem in the VM (e.g. trying to change
3078 // protection of malloc'ed or statically allocated memory). Check the
3079 // caller if you hit this assert.
3080 assert(addr == bottom, "sanity check");
3081
3082 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3083 return ::mprotect(bottom, size, prot) == 0;
3084 }
3085
3086 // Set protections specified
3087 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3088 bool is_committed) {
3089 unsigned int p = 0;
3090 switch (prot) {
3091 case MEM_PROT_NONE: p = PROT_NONE; break;
3092 case MEM_PROT_READ: p = PROT_READ; break;
3093 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3094 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3095 default:
3096 ShouldNotReachHere();
3097 }
3098 // is_committed is unused.
3099 return linux_mprotect(addr, bytes, p);
3100 }
3101
3102 bool os::guard_memory(char* addr, size_t size) {
3103 return linux_mprotect(addr, size, PROT_NONE);
3104 }
3105
3106 bool os::unguard_memory(char* addr, size_t size) {
3107 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3108 }
3109
3110 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3111 size_t page_size) {
3112 bool result = false;
3113 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3114 MAP_ANONYMOUS|MAP_PRIVATE,
3115 -1, 0);
3116 if (p != MAP_FAILED) {
3117 void *aligned_p = align_ptr_up(p, page_size);
3118
3119 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3120
3121 munmap(p, page_size * 2);
3122 }
3123
3124 if (warn && !result) {
3125 warning("TransparentHugePages is not supported by the operating system.");
3126 }
3127
3128 return result;
3129 }
3130
3131 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3132 bool result = false;
3133 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3134 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3135 -1, 0);
3136
3137 if (p != MAP_FAILED) {
3138 // We don't know if this really is a huge page or not.
3139 FILE *fp = fopen("/proc/self/maps", "r");
3140 if (fp) {
3141 while (!feof(fp)) {
3142 char chars[257];
3143 long x = 0;
3144 if (fgets(chars, sizeof(chars), fp)) {
3145 if (sscanf(chars, "%lx-%*x", &x) == 1
3146 && x == (long)p) {
3147 if (strstr (chars, "hugepage")) {
3148 result = true;
3149 break;
3150 }
3151 }
3152 }
3153 }
3154 fclose(fp);
3155 }
3156 munmap(p, page_size);
3157 }
3158
3159 if (warn && !result) {
3160 warning("HugeTLBFS is not supported by the operating system.");
3161 }
3162
3163 return result;
3164 }
3165
3166 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3167 //
3168 // From the coredump_filter documentation:
3169 //
3170 // - (bit 0) anonymous private memory
3171 // - (bit 1) anonymous shared memory
3172 // - (bit 2) file-backed private memory
3173 // - (bit 3) file-backed shared memory
3174 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3175 // effective only if the bit 2 is cleared)
3176 // - (bit 5) hugetlb private memory
3177 // - (bit 6) hugetlb shared memory
3178 //
3179 static void set_coredump_filter(void) {
3180 FILE *f;
3181 long cdm;
3182
3183 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3184 return;
3185 }
3186
3187 if (fscanf(f, "%lx", &cdm) != 1) {
3188 fclose(f);
3189 return;
3190 }
3191
3192 rewind(f);
3193
3194 if ((cdm & LARGEPAGES_BIT) == 0) {
3195 cdm |= LARGEPAGES_BIT;
3196 fprintf(f, "%#lx", cdm);
3197 }
3198
3199 fclose(f);
3200 }
3201
3202 // Large page support
3203
3204 static size_t _large_page_size = 0;
3205
3206 size_t os::Linux::find_large_page_size() {
3207 size_t large_page_size = 0;
3208
3209 // large_page_size on Linux is used to round up heap size. x86 uses either
3210 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3211 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3212 // page as large as 256M.
3213 //
3214 // Here we try to figure out page size by parsing /proc/meminfo and looking
3215 // for a line with the following format:
3216 // Hugepagesize: 2048 kB
3217 //
3218 // If we can't determine the value (e.g. /proc is not mounted, or the text
3219 // format has been changed), we'll use the largest page size supported by
3220 // the processor.
3221
3222 #ifndef ZERO
3223 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3224 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3225 #endif // ZERO
3226
3227 FILE *fp = fopen("/proc/meminfo", "r");
3228 if (fp) {
3229 while (!feof(fp)) {
3230 int x = 0;
3231 char buf[16];
3232 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3233 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3234 large_page_size = x * K;
3235 break;
3236 }
3237 } else {
3238 // skip to next line
3239 for (;;) {
3240 int ch = fgetc(fp);
3241 if (ch == EOF || ch == (int)'\n') break;
3242 }
3243 }
3244 }
3245 fclose(fp);
3246 }
3247
3248 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3249 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3250 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3251 proper_unit_for_byte_size(large_page_size));
3252 }
3253
3254 return large_page_size;
3255 }
3256
3257 size_t os::Linux::setup_large_page_size() {
3258 _large_page_size = Linux::find_large_page_size();
3259 const size_t default_page_size = (size_t)Linux::page_size();
3260 if (_large_page_size > default_page_size) {
3261 _page_sizes[0] = _large_page_size;
3262 _page_sizes[1] = default_page_size;
3263 _page_sizes[2] = 0;
3264 }
3265
3266 return _large_page_size;
3267 }
3268
3269 bool os::Linux::setup_large_page_type(size_t page_size) {
3270 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3271 FLAG_IS_DEFAULT(UseSHM) &&
3272 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3273
3274 // The type of large pages has not been specified by the user.
3275
3276 // Try UseHugeTLBFS and then UseSHM.
3277 UseHugeTLBFS = UseSHM = true;
3278
3279 // Don't try UseTransparentHugePages since there are known
3280 // performance issues with it turned on. This might change in the future.
3281 UseTransparentHugePages = false;
3282 }
3283
3284 if (UseTransparentHugePages) {
3285 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3286 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3287 UseHugeTLBFS = false;
3288 UseSHM = false;
3289 return true;
3290 }
3291 UseTransparentHugePages = false;
3292 }
3293
3294 if (UseHugeTLBFS) {
3295 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3296 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3297 UseSHM = false;
3298 return true;
3299 }
3300 UseHugeTLBFS = false;
3301 }
3302
3303 return UseSHM;
3304 }
3305
3306 void os::large_page_init() {
3307 if (!UseLargePages &&
3308 !UseTransparentHugePages &&
3309 !UseHugeTLBFS &&
3310 !UseSHM) {
3311 // Not using large pages.
3312 return;
3313 }
3314
3315 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3316 // The user explicitly turned off large pages.
3317 // Ignore the rest of the large pages flags.
3318 UseTransparentHugePages = false;
3319 UseHugeTLBFS = false;
3320 UseSHM = false;
3321 return;
3322 }
3323
3324 size_t large_page_size = Linux::setup_large_page_size();
3325 UseLargePages = Linux::setup_large_page_type(large_page_size);
3326
3327 set_coredump_filter();
3328 }
3329
3330 #ifndef SHM_HUGETLB
3331 #define SHM_HUGETLB 04000
3332 #endif
3333
3334 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3335 char* req_addr, bool exec) {
3336 // "exec" is passed in but not used. Creating the shared image for
3337 // the code cache doesn't have an SHM_X executable permission to check.
3338 assert(UseLargePages && UseSHM, "only for SHM large pages");
3339 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3340
3341 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3342 return NULL; // Fallback to small pages.
3343 }
3344
3345 key_t key = IPC_PRIVATE;
3346 char *addr;
3347
3348 bool warn_on_failure = UseLargePages &&
3349 (!FLAG_IS_DEFAULT(UseLargePages) ||
3350 !FLAG_IS_DEFAULT(UseSHM) ||
3351 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3352 char msg[128];
3353
3354 // Create a large shared memory region to attach to based on size.
3355 // Currently, size is the total size of the heap
3356 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3357 if (shmid == -1) {
3358 // Possible reasons for shmget failure:
3359 // 1. shmmax is too small for Java heap.
3360 // > check shmmax value: cat /proc/sys/kernel/shmmax
3361 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3362 // 2. not enough large page memory.
3363 // > check available large pages: cat /proc/meminfo
3364 // > increase amount of large pages:
3365 // echo new_value > /proc/sys/vm/nr_hugepages
3366 // Note 1: different Linux may use different name for this property,
3367 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3368 // Note 2: it's possible there's enough physical memory available but
3369 // they are so fragmented after a long run that they can't
3370 // coalesce into large pages. Try to reserve large pages when
3371 // the system is still "fresh".
3372 if (warn_on_failure) {
3373 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3374 warning("%s", msg);
3375 }
3376 return NULL;
3377 }
3378
3379 // attach to the region
3380 addr = (char*)shmat(shmid, req_addr, 0);
3381 int err = errno;
3382
3383 // Remove shmid. If shmat() is successful, the actual shared memory segment
3384 // will be deleted when it's detached by shmdt() or when the process
3385 // terminates. If shmat() is not successful this will remove the shared
3386 // segment immediately.
3387 shmctl(shmid, IPC_RMID, NULL);
3388
3389 if ((intptr_t)addr == -1) {
3390 if (warn_on_failure) {
3391 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3392 warning("%s", msg);
3393 }
3394 return NULL;
3395 }
3396
3397 return addr;
3398 }
3399
3400 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3401 int error) {
3402 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3403
3404 bool warn_on_failure = UseLargePages &&
3405 (!FLAG_IS_DEFAULT(UseLargePages) ||
3406 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3407 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3408
3409 if (warn_on_failure) {
3410 char msg[128];
3411 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3412 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3413 warning("%s", msg);
3414 }
3415 }
3416
3417 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3418 char* req_addr,
3419 bool exec) {
3420 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3421 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3422 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3423
3424 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3425 char* addr = (char*)::mmap(req_addr, bytes, prot,
3426 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3427 -1, 0);
3428
3429 if (addr == MAP_FAILED) {
3430 warn_on_large_pages_failure(req_addr, bytes, errno);
3431 return NULL;
3432 }
3433
3434 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3435
3436 return addr;
3437 }
3438
3439 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3440 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3441 // (req_addr != NULL) or with a given alignment.
3442 // - bytes shall be a multiple of alignment.
3443 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3444 // - alignment sets the alignment at which memory shall be allocated.
3445 // It must be a multiple of allocation granularity.
3446 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3447 // req_addr or NULL.
3448 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3449
3450 size_t extra_size = bytes;
3451 if (req_addr == NULL && alignment > 0) {
3452 extra_size += alignment;
3453 }
3454
3455 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3456 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3457 -1, 0);
3458 if (start == MAP_FAILED) {
3459 start = NULL;
3460 } else {
3461 if (req_addr != NULL) {
3462 if (start != req_addr) {
3463 ::munmap(start, extra_size);
3464 start = NULL;
3465 }
3466 } else {
3467 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3468 char* const end_aligned = start_aligned + bytes;
3469 char* const end = start + extra_size;
3470 if (start_aligned > start) {
3471 ::munmap(start, start_aligned - start);
3472 }
3473 if (end_aligned < end) {
3474 ::munmap(end_aligned, end - end_aligned);
3475 }
3476 start = start_aligned;
3477 }
3478 }
3479 return start;
3480
3481 }
3482
3483 // Reserve memory using mmap(MAP_HUGETLB).
3484 // - bytes shall be a multiple of alignment.
3485 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3486 // - alignment sets the alignment at which memory shall be allocated.
3487 // It must be a multiple of allocation granularity.
3488 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3489 // req_addr or NULL.
3490 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3491 size_t alignment,
3492 char* req_addr,
3493 bool exec) {
3494 size_t large_page_size = os::large_page_size();
3495 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3496
3497 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3498 assert(is_size_aligned(bytes, alignment), "Must be");
3499
3500 // First reserve - but not commit - the address range in small pages.
3501 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3502
3503 if (start == NULL) {
3504 return NULL;
3505 }
3506
3507 assert(is_ptr_aligned(start, alignment), "Must be");
3508
3509 char* end = start + bytes;
3510
3511 // Find the regions of the allocated chunk that can be promoted to large pages.
3512 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3513 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3514
3515 size_t lp_bytes = lp_end - lp_start;
3516
3517 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3518
3519 if (lp_bytes == 0) {
3520 // The mapped region doesn't even span the start and the end of a large page.
3521 // Fall back to allocate a non-special area.
3522 ::munmap(start, end - start);
3523 return NULL;
3524 }
3525
3526 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3527
3528 void* result;
3529
3530 // Commit small-paged leading area.
3531 if (start != lp_start) {
3532 result = ::mmap(start, lp_start - start, prot,
3533 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3534 -1, 0);
3535 if (result == MAP_FAILED) {
3536 ::munmap(lp_start, end - lp_start);
3537 return NULL;
3538 }
3539 }
3540
3541 // Commit large-paged area.
3542 result = ::mmap(lp_start, lp_bytes, prot,
3543 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3544 -1, 0);
3545 if (result == MAP_FAILED) {
3546 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3547 // If the mmap above fails, the large pages region will be unmapped and we
3548 // have regions before and after with small pages. Release these regions.
3549 //
3550 // | mapped | unmapped | mapped |
3551 // ^ ^ ^ ^
3552 // start lp_start lp_end end
3553 //
3554 ::munmap(start, lp_start - start);
3555 ::munmap(lp_end, end - lp_end);
3556 return NULL;
3557 }
3558
3559 // Commit small-paged trailing area.
3560 if (lp_end != end) {
3561 result = ::mmap(lp_end, end - lp_end, prot,
3562 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3563 -1, 0);
3564 if (result == MAP_FAILED) {
3565 ::munmap(start, lp_end - start);
3566 return NULL;
3567 }
3568 }
3569
3570 return start;
3571 }
3572
3573 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3574 size_t alignment,
3575 char* req_addr,
3576 bool exec) {
3577 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3578 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3579 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3580 assert(is_power_of_2(os::large_page_size()), "Must be");
3581 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3582
3583 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3584 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3585 } else {
3586 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3587 }
3588 }
3589
3590 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3591 char* req_addr, bool exec) {
3592 assert(UseLargePages, "only for large pages");
3593
3594 char* addr;
3595 if (UseSHM) {
3596 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3597 } else {
3598 assert(UseHugeTLBFS, "must be");
3599 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3600 }
3601
3602 if (addr != NULL) {
3603 if (UseNUMAInterleaving) {
3604 numa_make_global(addr, bytes);
3605 }
3606
3607 // The memory is committed
3608 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3609 }
3610
3611 return addr;
3612 }
3613
3614 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3615 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3616 return shmdt(base) == 0;
3617 }
3618
3619 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3620 return pd_release_memory(base, bytes);
3621 }
3622
3623 bool os::release_memory_special(char* base, size_t bytes) {
3624 bool res;
3625 if (MemTracker::tracking_level() > NMT_minimal) {
3626 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3627 res = os::Linux::release_memory_special_impl(base, bytes);
3628 if (res) {
3629 tkr.record((address)base, bytes);
3630 }
3631
3632 } else {
3633 res = os::Linux::release_memory_special_impl(base, bytes);
3634 }
3635 return res;
3636 }
3637
3638 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3639 assert(UseLargePages, "only for large pages");
3640 bool res;
3641
3642 if (UseSHM) {
3643 res = os::Linux::release_memory_special_shm(base, bytes);
3644 } else {
3645 assert(UseHugeTLBFS, "must be");
3646 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3647 }
3648 return res;
3649 }
3650
3651 size_t os::large_page_size() {
3652 return _large_page_size;
3653 }
3654
3655 // With SysV SHM the entire memory region must be allocated as shared
3656 // memory.
3657 // HugeTLBFS allows application to commit large page memory on demand.
3658 // However, when committing memory with HugeTLBFS fails, the region
3659 // that was supposed to be committed will lose the old reservation
3660 // and allow other threads to steal that memory region. Because of this
3661 // behavior we can't commit HugeTLBFS memory.
3662 bool os::can_commit_large_page_memory() {
3663 return UseTransparentHugePages;
3664 }
3665
3666 bool os::can_execute_large_page_memory() {
3667 return UseTransparentHugePages || UseHugeTLBFS;
3668 }
3669
3670 // Reserve memory at an arbitrary address, only if that area is
3671 // available (and not reserved for something else).
3672
3673 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3674 const int max_tries = 10;
3675 char* base[max_tries];
3676 size_t size[max_tries];
3677 const size_t gap = 0x000000;
3678
3679 // Assert only that the size is a multiple of the page size, since
3680 // that's all that mmap requires, and since that's all we really know
3681 // about at this low abstraction level. If we need higher alignment,
3682 // we can either pass an alignment to this method or verify alignment
3683 // in one of the methods further up the call chain. See bug 5044738.
3684 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3685
3686 // Repeatedly allocate blocks until the block is allocated at the
3687 // right spot.
3688
3689 // Linux mmap allows caller to pass an address as hint; give it a try first,
3690 // if kernel honors the hint then we can return immediately.
3691 char * addr = anon_mmap(requested_addr, bytes, false);
3692 if (addr == requested_addr) {
3693 return requested_addr;
3694 }
3695
3696 if (addr != NULL) {
3697 // mmap() is successful but it fails to reserve at the requested address
3698 anon_munmap(addr, bytes);
3699 }
3700
3701 int i;
3702 for (i = 0; i < max_tries; ++i) {
3703 base[i] = reserve_memory(bytes);
3704
3705 if (base[i] != NULL) {
3706 // Is this the block we wanted?
3707 if (base[i] == requested_addr) {
3708 size[i] = bytes;
3709 break;
3710 }
3711
3712 // Does this overlap the block we wanted? Give back the overlapped
3713 // parts and try again.
3714
3715 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3716 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3717 unmap_memory(base[i], top_overlap);
3718 base[i] += top_overlap;
3719 size[i] = bytes - top_overlap;
3720 } else {
3721 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3722 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3723 unmap_memory(requested_addr, bottom_overlap);
3724 size[i] = bytes - bottom_overlap;
3725 } else {
3726 size[i] = bytes;
3727 }
3728 }
3729 }
3730 }
3731
3732 // Give back the unused reserved pieces.
3733
3734 for (int j = 0; j < i; ++j) {
3735 if (base[j] != NULL) {
3736 unmap_memory(base[j], size[j]);
3737 }
3738 }
3739
3740 if (i < max_tries) {
3741 return requested_addr;
3742 } else {
3743 return NULL;
3744 }
3745 }
3746
3747 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3748 return ::read(fd, buf, nBytes);
3749 }
3750
3751 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3752 return ::pread(fd, buf, nBytes, offset);
3753 }
3754
3755 // Short sleep, direct OS call.
3756 //
3757 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3758 // sched_yield(2) will actually give up the CPU:
3759 //
3760 // * Alone on this pariticular CPU, keeps running.
3761 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3762 // (pre 2.6.39).
3763 //
3764 // So calling this with 0 is an alternative.
3765 //
3766 void os::naked_short_sleep(jlong ms) {
3767 struct timespec req;
3768
3769 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3770 req.tv_sec = 0;
3771 if (ms > 0) {
3772 req.tv_nsec = (ms % 1000) * 1000000;
3773 } else {
3774 req.tv_nsec = 1;
3775 }
3776
3777 nanosleep(&req, NULL);
3778
3779 return;
3780 }
3781
3782 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3783 void os::infinite_sleep() {
3784 while (true) { // sleep forever ...
3785 ::sleep(100); // ... 100 seconds at a time
3786 }
3787 }
3788
3789 // Used to convert frequent JVM_Yield() to nops
3790 bool os::dont_yield() {
3791 return DontYieldALot;
3792 }
3793
3794 void os::naked_yield() {
3795 sched_yield();
3796 }
3797
3798 ////////////////////////////////////////////////////////////////////////////////
3799 // thread priority support
3800
3801 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3802 // only supports dynamic priority, static priority must be zero. For real-time
3803 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3804 // However, for large multi-threaded applications, SCHED_RR is not only slower
3805 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3806 // of 5 runs - Sep 2005).
3807 //
3808 // The following code actually changes the niceness of kernel-thread/LWP. It
3809 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3810 // not the entire user process, and user level threads are 1:1 mapped to kernel
3811 // threads. It has always been the case, but could change in the future. For
3812 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3813 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3814
3815 int os::java_to_os_priority[CriticalPriority + 1] = {
3816 19, // 0 Entry should never be used
3817
3818 4, // 1 MinPriority
3819 3, // 2
3820 2, // 3
3821
3822 1, // 4
3823 0, // 5 NormPriority
3824 -1, // 6
3825
3826 -2, // 7
3827 -3, // 8
3828 -4, // 9 NearMaxPriority
3829
3830 -5, // 10 MaxPriority
3831
3832 -5 // 11 CriticalPriority
3833 };
3834
3835 static int prio_init() {
3836 if (ThreadPriorityPolicy == 1) {
3837 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3838 // if effective uid is not root. Perhaps, a more elegant way of doing
3839 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3840 if (geteuid() != 0) {
3841 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3842 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3843 }
3844 ThreadPriorityPolicy = 0;
3845 }
3846 }
3847 if (UseCriticalJavaThreadPriority) {
3848 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3849 }
3850 return 0;
3851 }
3852
3853 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3854 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3855
3856 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3857 return (ret == 0) ? OS_OK : OS_ERR;
3858 }
3859
3860 OSReturn os::get_native_priority(const Thread* const thread,
3861 int *priority_ptr) {
3862 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3863 *priority_ptr = java_to_os_priority[NormPriority];
3864 return OS_OK;
3865 }
3866
3867 errno = 0;
3868 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3869 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3870 }
3871
3872 // Hint to the underlying OS that a task switch would not be good.
3873 // Void return because it's a hint and can fail.
3874 void os::hint_no_preempt() {}
3875
3876 ////////////////////////////////////////////////////////////////////////////////
3877 // suspend/resume support
3878
3879 // the low-level signal-based suspend/resume support is a remnant from the
3880 // old VM-suspension that used to be for java-suspension, safepoints etc,
3881 // within hotspot. Now there is a single use-case for this:
3882 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3883 // that runs in the watcher thread.
3884 // The remaining code is greatly simplified from the more general suspension
3885 // code that used to be used.
3886 //
3887 // The protocol is quite simple:
3888 // - suspend:
3889 // - sends a signal to the target thread
3890 // - polls the suspend state of the osthread using a yield loop
3891 // - target thread signal handler (SR_handler) sets suspend state
3892 // and blocks in sigsuspend until continued
3893 // - resume:
3894 // - sets target osthread state to continue
3895 // - sends signal to end the sigsuspend loop in the SR_handler
3896 //
3897 // Note that the SR_lock plays no role in this suspend/resume protocol.
3898
3899 static void resume_clear_context(OSThread *osthread) {
3900 osthread->set_ucontext(NULL);
3901 osthread->set_siginfo(NULL);
3902 }
3903
3904 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3905 ucontext_t* context) {
3906 osthread->set_ucontext(context);
3907 osthread->set_siginfo(siginfo);
3908 }
3909
3910 // Handler function invoked when a thread's execution is suspended or
3911 // resumed. We have to be careful that only async-safe functions are
3912 // called here (Note: most pthread functions are not async safe and
3913 // should be avoided.)
3914 //
3915 // Note: sigwait() is a more natural fit than sigsuspend() from an
3916 // interface point of view, but sigwait() prevents the signal hander
3917 // from being run. libpthread would get very confused by not having
3918 // its signal handlers run and prevents sigwait()'s use with the
3919 // mutex granting granting signal.
3920 //
3921 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3922 //
3923 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3924 // Save and restore errno to avoid confusing native code with EINTR
3925 // after sigsuspend.
3926 int old_errno = errno;
3927
3928 Thread* thread = Thread::current();
3929 OSThread* osthread = thread->osthread();
3930 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3931
3932 os::SuspendResume::State current = osthread->sr.state();
3933 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3934 suspend_save_context(osthread, siginfo, context);
3935
3936 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3937 os::SuspendResume::State state = osthread->sr.suspended();
3938 if (state == os::SuspendResume::SR_SUSPENDED) {
3939 sigset_t suspend_set; // signals for sigsuspend()
3940
3941 // get current set of blocked signals and unblock resume signal
3942 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3943 sigdelset(&suspend_set, SR_signum);
3944
3945 sr_semaphore.signal();
3946 // wait here until we are resumed
3947 while (1) {
3948 sigsuspend(&suspend_set);
3949
3950 os::SuspendResume::State result = osthread->sr.running();
3951 if (result == os::SuspendResume::SR_RUNNING) {
3952 sr_semaphore.signal();
3953 break;
3954 }
3955 }
3956
3957 } else if (state == os::SuspendResume::SR_RUNNING) {
3958 // request was cancelled, continue
3959 } else {
3960 ShouldNotReachHere();
3961 }
3962
3963 resume_clear_context(osthread);
3964 } else if (current == os::SuspendResume::SR_RUNNING) {
3965 // request was cancelled, continue
3966 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3967 // ignore
3968 } else {
3969 // ignore
3970 }
3971
3972 errno = old_errno;
3973 }
3974
3975
3976 static int SR_initialize() {
3977 struct sigaction act;
3978 char *s;
3979 // Get signal number to use for suspend/resume
3980 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3981 int sig = ::strtol(s, 0, 10);
3982 if (sig > 0 || sig < _NSIG) {
3983 SR_signum = sig;
3984 }
3985 }
3986
3987 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3988 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3989
3990 sigemptyset(&SR_sigset);
3991 sigaddset(&SR_sigset, SR_signum);
3992
3993 // Set up signal handler for suspend/resume
3994 act.sa_flags = SA_RESTART|SA_SIGINFO;
3995 act.sa_handler = (void (*)(int)) SR_handler;
3996
3997 // SR_signum is blocked by default.
3998 // 4528190 - We also need to block pthread restart signal (32 on all
3999 // supported Linux platforms). Note that LinuxThreads need to block
4000 // this signal for all threads to work properly. So we don't have
4001 // to use hard-coded signal number when setting up the mask.
4002 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4003
4004 if (sigaction(SR_signum, &act, 0) == -1) {
4005 return -1;
4006 }
4007
4008 // Save signal flag
4009 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4010 return 0;
4011 }
4012
4013 static int sr_notify(OSThread* osthread) {
4014 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4015 assert_status(status == 0, status, "pthread_kill");
4016 return status;
4017 }
4018
4019 // "Randomly" selected value for how long we want to spin
4020 // before bailing out on suspending a thread, also how often
4021 // we send a signal to a thread we want to resume
4022 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4023 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4024
4025 // returns true on success and false on error - really an error is fatal
4026 // but this seems the normal response to library errors
4027 static bool do_suspend(OSThread* osthread) {
4028 assert(osthread->sr.is_running(), "thread should be running");
4029 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4030
4031 // mark as suspended and send signal
4032 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4033 // failed to switch, state wasn't running?
4034 ShouldNotReachHere();
4035 return false;
4036 }
4037
4038 if (sr_notify(osthread) != 0) {
4039 ShouldNotReachHere();
4040 }
4041
4042 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4043 while (true) {
4044 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4045 break;
4046 } else {
4047 // timeout
4048 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4049 if (cancelled == os::SuspendResume::SR_RUNNING) {
4050 return false;
4051 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4052 // make sure that we consume the signal on the semaphore as well
4053 sr_semaphore.wait();
4054 break;
4055 } else {
4056 ShouldNotReachHere();
4057 return false;
4058 }
4059 }
4060 }
4061
4062 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4063 return true;
4064 }
4065
4066 static void do_resume(OSThread* osthread) {
4067 assert(osthread->sr.is_suspended(), "thread should be suspended");
4068 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4069
4070 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4071 // failed to switch to WAKEUP_REQUEST
4072 ShouldNotReachHere();
4073 return;
4074 }
4075
4076 while (true) {
4077 if (sr_notify(osthread) == 0) {
4078 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4079 if (osthread->sr.is_running()) {
4080 return;
4081 }
4082 }
4083 } else {
4084 ShouldNotReachHere();
4085 }
4086 }
4087
4088 guarantee(osthread->sr.is_running(), "Must be running!");
4089 }
4090
4091 ///////////////////////////////////////////////////////////////////////////////////
4092 // signal handling (except suspend/resume)
4093
4094 // This routine may be used by user applications as a "hook" to catch signals.
4095 // The user-defined signal handler must pass unrecognized signals to this
4096 // routine, and if it returns true (non-zero), then the signal handler must
4097 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4098 // routine will never retun false (zero), but instead will execute a VM panic
4099 // routine kill the process.
4100 //
4101 // If this routine returns false, it is OK to call it again. This allows
4102 // the user-defined signal handler to perform checks either before or after
4103 // the VM performs its own checks. Naturally, the user code would be making
4104 // a serious error if it tried to handle an exception (such as a null check
4105 // or breakpoint) that the VM was generating for its own correct operation.
4106 //
4107 // This routine may recognize any of the following kinds of signals:
4108 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4109 // It should be consulted by handlers for any of those signals.
4110 //
4111 // The caller of this routine must pass in the three arguments supplied
4112 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4113 // field of the structure passed to sigaction(). This routine assumes that
4114 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4115 //
4116 // Note that the VM will print warnings if it detects conflicting signal
4117 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4118 //
4119 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4120 siginfo_t* siginfo,
4121 void* ucontext,
4122 int abort_if_unrecognized);
4123
4124 void signalHandler(int sig, siginfo_t* info, void* uc) {
4125 assert(info != NULL && uc != NULL, "it must be old kernel");
4126 int orig_errno = errno; // Preserve errno value over signal handler.
4127 JVM_handle_linux_signal(sig, info, uc, true);
4128 errno = orig_errno;
4129 }
4130
4131
4132 // This boolean allows users to forward their own non-matching signals
4133 // to JVM_handle_linux_signal, harmlessly.
4134 bool os::Linux::signal_handlers_are_installed = false;
4135
4136 // For signal-chaining
4137 struct sigaction os::Linux::sigact[MAXSIGNUM];
4138 unsigned int os::Linux::sigs = 0;
4139 bool os::Linux::libjsig_is_loaded = false;
4140 typedef struct sigaction *(*get_signal_t)(int);
4141 get_signal_t os::Linux::get_signal_action = NULL;
4142
4143 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4144 struct sigaction *actp = NULL;
4145
4146 if (libjsig_is_loaded) {
4147 // Retrieve the old signal handler from libjsig
4148 actp = (*get_signal_action)(sig);
4149 }
4150 if (actp == NULL) {
4151 // Retrieve the preinstalled signal handler from jvm
4152 actp = get_preinstalled_handler(sig);
4153 }
4154
4155 return actp;
4156 }
4157
4158 static bool call_chained_handler(struct sigaction *actp, int sig,
4159 siginfo_t *siginfo, void *context) {
4160 // Call the old signal handler
4161 if (actp->sa_handler == SIG_DFL) {
4162 // It's more reasonable to let jvm treat it as an unexpected exception
4163 // instead of taking the default action.
4164 return false;
4165 } else if (actp->sa_handler != SIG_IGN) {
4166 if ((actp->sa_flags & SA_NODEFER) == 0) {
4167 // automaticlly block the signal
4168 sigaddset(&(actp->sa_mask), sig);
4169 }
4170
4171 sa_handler_t hand;
4172 sa_sigaction_t sa;
4173 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4174 // retrieve the chained handler
4175 if (siginfo_flag_set) {
4176 sa = actp->sa_sigaction;
4177 } else {
4178 hand = actp->sa_handler;
4179 }
4180
4181 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4182 actp->sa_handler = SIG_DFL;
4183 }
4184
4185 // try to honor the signal mask
4186 sigset_t oset;
4187 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4188
4189 // call into the chained handler
4190 if (siginfo_flag_set) {
4191 (*sa)(sig, siginfo, context);
4192 } else {
4193 (*hand)(sig);
4194 }
4195
4196 // restore the signal mask
4197 pthread_sigmask(SIG_SETMASK, &oset, 0);
4198 }
4199 // Tell jvm's signal handler the signal is taken care of.
4200 return true;
4201 }
4202
4203 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4204 bool chained = false;
4205 // signal-chaining
4206 if (UseSignalChaining) {
4207 struct sigaction *actp = get_chained_signal_action(sig);
4208 if (actp != NULL) {
4209 chained = call_chained_handler(actp, sig, siginfo, context);
4210 }
4211 }
4212 return chained;
4213 }
4214
4215 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4216 if ((((unsigned int)1 << sig) & sigs) != 0) {
4217 return &sigact[sig];
4218 }
4219 return NULL;
4220 }
4221
4222 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4223 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4224 sigact[sig] = oldAct;
4225 sigs |= (unsigned int)1 << sig;
4226 }
4227
4228 // for diagnostic
4229 int os::Linux::sigflags[MAXSIGNUM];
4230
4231 int os::Linux::get_our_sigflags(int sig) {
4232 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4233 return sigflags[sig];
4234 }
4235
4236 void os::Linux::set_our_sigflags(int sig, int flags) {
4237 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4238 sigflags[sig] = flags;
4239 }
4240
4241 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4242 // Check for overwrite.
4243 struct sigaction oldAct;
4244 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4245
4246 void* oldhand = oldAct.sa_sigaction
4247 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4248 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4249 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4250 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4251 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4252 if (AllowUserSignalHandlers || !set_installed) {
4253 // Do not overwrite; user takes responsibility to forward to us.
4254 return;
4255 } else if (UseSignalChaining) {
4256 // save the old handler in jvm
4257 save_preinstalled_handler(sig, oldAct);
4258 // libjsig also interposes the sigaction() call below and saves the
4259 // old sigaction on it own.
4260 } else {
4261 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4262 "%#lx for signal %d.", (long)oldhand, sig));
4263 }
4264 }
4265
4266 struct sigaction sigAct;
4267 sigfillset(&(sigAct.sa_mask));
4268 sigAct.sa_handler = SIG_DFL;
4269 if (!set_installed) {
4270 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4271 } else {
4272 sigAct.sa_sigaction = signalHandler;
4273 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4274 }
4275 // Save flags, which are set by ours
4276 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4277 sigflags[sig] = sigAct.sa_flags;
4278
4279 int ret = sigaction(sig, &sigAct, &oldAct);
4280 assert(ret == 0, "check");
4281
4282 void* oldhand2 = oldAct.sa_sigaction
4283 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4284 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4285 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4286 }
4287
4288 // install signal handlers for signals that HotSpot needs to
4289 // handle in order to support Java-level exception handling.
4290
4291 void os::Linux::install_signal_handlers() {
4292 if (!signal_handlers_are_installed) {
4293 signal_handlers_are_installed = true;
4294
4295 // signal-chaining
4296 typedef void (*signal_setting_t)();
4297 signal_setting_t begin_signal_setting = NULL;
4298 signal_setting_t end_signal_setting = NULL;
4299 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4300 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4301 if (begin_signal_setting != NULL) {
4302 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4303 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4304 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4305 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4306 libjsig_is_loaded = true;
4307 assert(UseSignalChaining, "should enable signal-chaining");
4308 }
4309 if (libjsig_is_loaded) {
4310 // Tell libjsig jvm is setting signal handlers
4311 (*begin_signal_setting)();
4312 }
4313
4314 set_signal_handler(SIGSEGV, true);
4315 set_signal_handler(SIGPIPE, true);
4316 set_signal_handler(SIGBUS, true);
4317 set_signal_handler(SIGILL, true);
4318 set_signal_handler(SIGFPE, true);
4319 #if defined(PPC64)
4320 set_signal_handler(SIGTRAP, true);
4321 #endif
4322 set_signal_handler(SIGXFSZ, true);
4323
4324 if (libjsig_is_loaded) {
4325 // Tell libjsig jvm finishes setting signal handlers
4326 (*end_signal_setting)();
4327 }
4328
4329 // We don't activate signal checker if libjsig is in place, we trust ourselves
4330 // and if UserSignalHandler is installed all bets are off.
4331 // Log that signal checking is off only if -verbose:jni is specified.
4332 if (CheckJNICalls) {
4333 if (libjsig_is_loaded) {
4334 if (PrintJNIResolving) {
4335 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4336 }
4337 check_signals = false;
4338 }
4339 if (AllowUserSignalHandlers) {
4340 if (PrintJNIResolving) {
4341 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4342 }
4343 check_signals = false;
4344 }
4345 }
4346 }
4347 }
4348
4349 // This is the fastest way to get thread cpu time on Linux.
4350 // Returns cpu time (user+sys) for any thread, not only for current.
4351 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4352 // It might work on 2.6.10+ with a special kernel/glibc patch.
4353 // For reference, please, see IEEE Std 1003.1-2004:
4354 // http://www.unix.org/single_unix_specification
4355
4356 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4357 struct timespec tp;
4358 int rc = os::Linux::clock_gettime(clockid, &tp);
4359 assert(rc == 0, "clock_gettime is expected to return 0 code");
4360
4361 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4362 }
4363
4364 /////
4365 // glibc on Linux platform uses non-documented flag
4366 // to indicate, that some special sort of signal
4367 // trampoline is used.
4368 // We will never set this flag, and we should
4369 // ignore this flag in our diagnostic
4370 #ifdef SIGNIFICANT_SIGNAL_MASK
4371 #undef SIGNIFICANT_SIGNAL_MASK
4372 #endif
4373 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4374
4375 static const char* get_signal_handler_name(address handler,
4376 char* buf, int buflen) {
4377 int offset;
4378 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4379 if (found) {
4380 // skip directory names
4381 const char *p1, *p2;
4382 p1 = buf;
4383 size_t len = strlen(os::file_separator());
4384 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4385 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4386 } else {
4387 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4388 }
4389 return buf;
4390 }
4391
4392 static void print_signal_handler(outputStream* st, int sig,
4393 char* buf, size_t buflen) {
4394 struct sigaction sa;
4395
4396 sigaction(sig, NULL, &sa);
4397
4398 // See comment for SIGNIFICANT_SIGNAL_MASK define
4399 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4400
4401 st->print("%s: ", os::exception_name(sig, buf, buflen));
4402
4403 address handler = (sa.sa_flags & SA_SIGINFO)
4404 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4405 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4406
4407 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4408 st->print("SIG_DFL");
4409 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4410 st->print("SIG_IGN");
4411 } else {
4412 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4413 }
4414
4415 st->print(", sa_mask[0]=");
4416 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4417
4418 address rh = VMError::get_resetted_sighandler(sig);
4419 // May be, handler was resetted by VMError?
4420 if (rh != NULL) {
4421 handler = rh;
4422 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4423 }
4424
4425 st->print(", sa_flags=");
4426 os::Posix::print_sa_flags(st, sa.sa_flags);
4427
4428 // Check: is it our handler?
4429 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4430 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4431 // It is our signal handler
4432 // check for flags, reset system-used one!
4433 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4434 st->print(
4435 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4436 os::Linux::get_our_sigflags(sig));
4437 }
4438 }
4439 st->cr();
4440 }
4441
4442
4443 #define DO_SIGNAL_CHECK(sig) \
4444 do { \
4445 if (!sigismember(&check_signal_done, sig)) { \
4446 os::Linux::check_signal_handler(sig); \
4447 } \
4448 } while (0)
4449
4450 // This method is a periodic task to check for misbehaving JNI applications
4451 // under CheckJNI, we can add any periodic checks here
4452
4453 void os::run_periodic_checks() {
4454 if (check_signals == false) return;
4455
4456 // SEGV and BUS if overridden could potentially prevent
4457 // generation of hs*.log in the event of a crash, debugging
4458 // such a case can be very challenging, so we absolutely
4459 // check the following for a good measure:
4460 DO_SIGNAL_CHECK(SIGSEGV);
4461 DO_SIGNAL_CHECK(SIGILL);
4462 DO_SIGNAL_CHECK(SIGFPE);
4463 DO_SIGNAL_CHECK(SIGBUS);
4464 DO_SIGNAL_CHECK(SIGPIPE);
4465 DO_SIGNAL_CHECK(SIGXFSZ);
4466 #if defined(PPC64)
4467 DO_SIGNAL_CHECK(SIGTRAP);
4468 #endif
4469
4470 // ReduceSignalUsage allows the user to override these handlers
4471 // see comments at the very top and jvm_solaris.h
4472 if (!ReduceSignalUsage) {
4473 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4474 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4475 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4476 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4477 }
4478
4479 DO_SIGNAL_CHECK(SR_signum);
4480 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4481 }
4482
4483 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4484
4485 static os_sigaction_t os_sigaction = NULL;
4486
4487 void os::Linux::check_signal_handler(int sig) {
4488 char buf[O_BUFLEN];
4489 address jvmHandler = NULL;
4490
4491
4492 struct sigaction act;
4493 if (os_sigaction == NULL) {
4494 // only trust the default sigaction, in case it has been interposed
4495 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4496 if (os_sigaction == NULL) return;
4497 }
4498
4499 os_sigaction(sig, (struct sigaction*)NULL, &act);
4500
4501
4502 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4503
4504 address thisHandler = (act.sa_flags & SA_SIGINFO)
4505 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4506 : CAST_FROM_FN_PTR(address, act.sa_handler);
4507
4508
4509 switch (sig) {
4510 case SIGSEGV:
4511 case SIGBUS:
4512 case SIGFPE:
4513 case SIGPIPE:
4514 case SIGILL:
4515 case SIGXFSZ:
4516 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4517 break;
4518
4519 case SHUTDOWN1_SIGNAL:
4520 case SHUTDOWN2_SIGNAL:
4521 case SHUTDOWN3_SIGNAL:
4522 case BREAK_SIGNAL:
4523 jvmHandler = (address)user_handler();
4524 break;
4525
4526 case INTERRUPT_SIGNAL:
4527 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4528 break;
4529
4530 default:
4531 if (sig == SR_signum) {
4532 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4533 } else {
4534 return;
4535 }
4536 break;
4537 }
4538
4539 if (thisHandler != jvmHandler) {
4540 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4541 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4542 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4543 // No need to check this sig any longer
4544 sigaddset(&check_signal_done, sig);
4545 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4546 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4547 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4548 exception_name(sig, buf, O_BUFLEN));
4549 }
4550 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4551 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4552 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4553 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4554 // No need to check this sig any longer
4555 sigaddset(&check_signal_done, sig);
4556 }
4557
4558 // Dump all the signal
4559 if (sigismember(&check_signal_done, sig)) {
4560 print_signal_handlers(tty, buf, O_BUFLEN);
4561 }
4562 }
4563
4564 extern void report_error(char* file_name, int line_no, char* title,
4565 char* format, ...);
4566
4567 extern bool signal_name(int signo, char* buf, size_t len);
4568
4569 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4570 if (0 < exception_code && exception_code <= SIGRTMAX) {
4571 // signal
4572 if (!signal_name(exception_code, buf, size)) {
4573 jio_snprintf(buf, size, "SIG%d", exception_code);
4574 }
4575 return buf;
4576 } else {
4577 return NULL;
4578 }
4579 }
4580
4581 // this is called _before_ the most of global arguments have been parsed
4582 void os::init(void) {
4583 char dummy; // used to get a guess on initial stack address
4584 // first_hrtime = gethrtime();
4585
4586 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4587
4588 init_random(1234567);
4589
4590 ThreadCritical::initialize();
4591
4592 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4593 if (Linux::page_size() == -1) {
4594 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4595 strerror(errno)));
4596 }
4597 init_page_sizes((size_t) Linux::page_size());
4598
4599 Linux::initialize_system_info();
4600
4601 // main_thread points to the aboriginal thread
4602 Linux::_main_thread = pthread_self();
4603
4604 Linux::clock_init();
4605 initial_time_count = javaTimeNanos();
4606
4607 // pthread_condattr initialization for monotonic clock
4608 int status;
4609 pthread_condattr_t* _condattr = os::Linux::condAttr();
4610 if ((status = pthread_condattr_init(_condattr)) != 0) {
4611 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4612 }
4613 // Only set the clock if CLOCK_MONOTONIC is available
4614 if (os::supports_monotonic_clock()) {
4615 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4616 if (status == EINVAL) {
4617 warning("Unable to use monotonic clock with relative timed-waits" \
4618 " - changes to the time-of-day clock may have adverse affects");
4619 } else {
4620 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4621 }
4622 }
4623 }
4624 // else it defaults to CLOCK_REALTIME
4625
4626 // If the pagesize of the VM is greater than 8K determine the appropriate
4627 // number of initial guard pages. The user can change this with the
4628 // command line arguments, if needed.
4629 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4630 StackYellowPages = 1;
4631 StackRedPages = 1;
4632 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4633 }
4634
4635 // retrieve entry point for pthread_setname_np
4636 Linux::_pthread_setname_np =
4637 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4638
4639 }
4640
4641 // To install functions for atexit system call
4642 extern "C" {
4643 static void perfMemory_exit_helper() {
4644 perfMemory_exit();
4645 }
4646 }
4647
4648 // this is called _after_ the global arguments have been parsed
4649 jint os::init_2(void) {
4650 Linux::fast_thread_clock_init();
4651
4652 // Allocate a single page and mark it as readable for safepoint polling
4653 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4654 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4655
4656 os::set_polling_page(polling_page);
4657
4658 #ifndef PRODUCT
4659 if (Verbose && PrintMiscellaneous) {
4660 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4661 (intptr_t)polling_page);
4662 }
4663 #endif
4664
4665 if (!UseMembar) {
4666 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4667 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4668 os::set_memory_serialize_page(mem_serialize_page);
4669
4670 #ifndef PRODUCT
4671 if (Verbose && PrintMiscellaneous) {
4672 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4673 (intptr_t)mem_serialize_page);
4674 }
4675 #endif
4676 }
4677
4678 // initialize suspend/resume support - must do this before signal_sets_init()
4679 if (SR_initialize() != 0) {
4680 perror("SR_initialize failed");
4681 return JNI_ERR;
4682 }
4683
4684 Linux::signal_sets_init();
4685 Linux::install_signal_handlers();
4686
4687 // Check minimum allowable stack size for thread creation and to initialize
4688 // the java system classes, including StackOverflowError - depends on page
4689 // size. Add a page for compiler2 recursion in main thread.
4690 // Add in 2*BytesPerWord times page size to account for VM stack during
4691 // class initialization depending on 32 or 64 bit VM.
4692 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4693 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4694 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4695
4696 size_t threadStackSizeInBytes = ThreadStackSize * K;
4697 if (threadStackSizeInBytes != 0 &&
4698 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4699 tty->print_cr("\nThe stack size specified is too small, "
4700 "Specify at least %dk",
4701 os::Linux::min_stack_allowed/ K);
4702 return JNI_ERR;
4703 }
4704
4705 // Make the stack size a multiple of the page size so that
4706 // the yellow/red zones can be guarded.
4707 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4708 vm_page_size()));
4709
4710 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4711
4712 #if defined(IA32)
4713 workaround_expand_exec_shield_cs_limit();
4714 #endif
4715
4716 Linux::libpthread_init();
4717 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4718 tty->print_cr("[HotSpot is running with %s, %s]\n",
4719 Linux::glibc_version(), Linux::libpthread_version());
4720 }
4721
4722 if (UseNUMA) {
4723 if (!Linux::libnuma_init()) {
4724 UseNUMA = false;
4725 } else {
4726 if ((Linux::numa_max_node() < 1)) {
4727 // There's only one node(they start from 0), disable NUMA.
4728 UseNUMA = false;
4729 }
4730 }
4731 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4732 // we can make the adaptive lgrp chunk resizing work. If the user specified
4733 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4734 // disable adaptive resizing.
4735 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4736 if (FLAG_IS_DEFAULT(UseNUMA)) {
4737 UseNUMA = false;
4738 } else {
4739 if (FLAG_IS_DEFAULT(UseLargePages) &&
4740 FLAG_IS_DEFAULT(UseSHM) &&
4741 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4742 UseLargePages = false;
4743 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4744 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4745 UseAdaptiveSizePolicy = false;
4746 UseAdaptiveNUMAChunkSizing = false;
4747 }
4748 }
4749 }
4750 if (!UseNUMA && ForceNUMA) {
4751 UseNUMA = true;
4752 }
4753 }
4754
4755 if (MaxFDLimit) {
4756 // set the number of file descriptors to max. print out error
4757 // if getrlimit/setrlimit fails but continue regardless.
4758 struct rlimit nbr_files;
4759 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4760 if (status != 0) {
4761 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4762 perror("os::init_2 getrlimit failed");
4763 }
4764 } else {
4765 nbr_files.rlim_cur = nbr_files.rlim_max;
4766 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4767 if (status != 0) {
4768 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4769 perror("os::init_2 setrlimit failed");
4770 }
4771 }
4772 }
4773 }
4774
4775 // Initialize lock used to serialize thread creation (see os::create_thread)
4776 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4777
4778 // at-exit methods are called in the reverse order of their registration.
4779 // atexit functions are called on return from main or as a result of a
4780 // call to exit(3C). There can be only 32 of these functions registered
4781 // and atexit() does not set errno.
4782
4783 if (PerfAllowAtExitRegistration) {
4784 // only register atexit functions if PerfAllowAtExitRegistration is set.
4785 // atexit functions can be delayed until process exit time, which
4786 // can be problematic for embedded VM situations. Embedded VMs should
4787 // call DestroyJavaVM() to assure that VM resources are released.
4788
4789 // note: perfMemory_exit_helper atexit function may be removed in
4790 // the future if the appropriate cleanup code can be added to the
4791 // VM_Exit VMOperation's doit method.
4792 if (atexit(perfMemory_exit_helper) != 0) {
4793 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4794 }
4795 }
4796
4797 // initialize thread priority policy
4798 prio_init();
4799
4800 return JNI_OK;
4801 }
4802
4803 // Mark the polling page as unreadable
4804 void os::make_polling_page_unreadable(void) {
4805 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4806 fatal("Could not disable polling page");
4807 }
4808 }
4809
4810 // Mark the polling page as readable
4811 void os::make_polling_page_readable(void) {
4812 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4813 fatal("Could not enable polling page");
4814 }
4815 }
4816
4817 int os::active_processor_count() {
4818 // Linux doesn't yet have a (official) notion of processor sets,
4819 // so just return the number of online processors.
4820 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4821 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4822 return online_cpus;
4823 }
4824
4825 void os::set_native_thread_name(const char *name) {
4826 if (Linux::_pthread_setname_np) {
4827 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4828 snprintf(buf, sizeof(buf), "%s", name);
4829 buf[sizeof(buf) - 1] = '\0';
4830 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4831 // ERANGE should not happen; all other errors should just be ignored.
4832 assert(rc != ERANGE, "pthread_setname_np failed");
4833 }
4834 }
4835
4836 bool os::distribute_processes(uint length, uint* distribution) {
4837 // Not yet implemented.
4838 return false;
4839 }
4840
4841 bool os::bind_to_processor(uint processor_id) {
4842 // Not yet implemented.
4843 return false;
4844 }
4845
4846 ///
4847
4848 void os::SuspendedThreadTask::internal_do_task() {
4849 if (do_suspend(_thread->osthread())) {
4850 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4851 do_task(context);
4852 do_resume(_thread->osthread());
4853 }
4854 }
4855
4856 class PcFetcher : public os::SuspendedThreadTask {
4857 public:
4858 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4859 ExtendedPC result();
4860 protected:
4861 void do_task(const os::SuspendedThreadTaskContext& context);
4862 private:
4863 ExtendedPC _epc;
4864 };
4865
4866 ExtendedPC PcFetcher::result() {
4867 guarantee(is_done(), "task is not done yet.");
4868 return _epc;
4869 }
4870
4871 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4872 Thread* thread = context.thread();
4873 OSThread* osthread = thread->osthread();
4874 if (osthread->ucontext() != NULL) {
4875 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4876 } else {
4877 // NULL context is unexpected, double-check this is the VMThread
4878 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4879 }
4880 }
4881
4882 // Suspends the target using the signal mechanism and then grabs the PC before
4883 // resuming the target. Used by the flat-profiler only
4884 ExtendedPC os::get_thread_pc(Thread* thread) {
4885 // Make sure that it is called by the watcher for the VMThread
4886 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4887 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4888
4889 PcFetcher fetcher(thread);
4890 fetcher.run();
4891 return fetcher.result();
4892 }
4893
4894 ////////////////////////////////////////////////////////////////////////////////
4895 // debug support
4896
4897 bool os::find(address addr, outputStream* st) {
4898 Dl_info dlinfo;
4899 memset(&dlinfo, 0, sizeof(dlinfo));
4900 if (dladdr(addr, &dlinfo) != 0) {
4901 st->print(PTR_FORMAT ": ", addr);
4902 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4903 st->print("%s+%#x", dlinfo.dli_sname,
4904 addr - (intptr_t)dlinfo.dli_saddr);
4905 } else if (dlinfo.dli_fbase != NULL) {
4906 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4907 } else {
4908 st->print("<absolute address>");
4909 }
4910 if (dlinfo.dli_fname != NULL) {
4911 st->print(" in %s", dlinfo.dli_fname);
4912 }
4913 if (dlinfo.dli_fbase != NULL) {
4914 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4915 }
4916 st->cr();
4917
4918 if (Verbose) {
4919 // decode some bytes around the PC
4920 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4921 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4922 address lowest = (address) dlinfo.dli_sname;
4923 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4924 if (begin < lowest) begin = lowest;
4925 Dl_info dlinfo2;
4926 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4927 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4928 end = (address) dlinfo2.dli_saddr;
4929 }
4930 Disassembler::decode(begin, end, st);
4931 }
4932 return true;
4933 }
4934 return false;
4935 }
4936
4937 ////////////////////////////////////////////////////////////////////////////////
4938 // misc
4939
4940 // This does not do anything on Linux. This is basically a hook for being
4941 // able to use structured exception handling (thread-local exception filters)
4942 // on, e.g., Win32.
4943 void
4944 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4945 JavaCallArguments* args, Thread* thread) {
4946 f(value, method, args, thread);
4947 }
4948
4949 void os::print_statistics() {
4950 }
4951
4952 int os::message_box(const char* title, const char* message) {
4953 int i;
4954 fdStream err(defaultStream::error_fd());
4955 for (i = 0; i < 78; i++) err.print_raw("=");
4956 err.cr();
4957 err.print_raw_cr(title);
4958 for (i = 0; i < 78; i++) err.print_raw("-");
4959 err.cr();
4960 err.print_raw_cr(message);
4961 for (i = 0; i < 78; i++) err.print_raw("=");
4962 err.cr();
4963
4964 char buf[16];
4965 // Prevent process from exiting upon "read error" without consuming all CPU
4966 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4967
4968 return buf[0] == 'y' || buf[0] == 'Y';
4969 }
4970
4971 int os::stat(const char *path, struct stat *sbuf) {
4972 char pathbuf[MAX_PATH];
4973 if (strlen(path) > MAX_PATH - 1) {
4974 errno = ENAMETOOLONG;
4975 return -1;
4976 }
4977 os::native_path(strcpy(pathbuf, path));
4978 return ::stat(pathbuf, sbuf);
4979 }
4980
4981 bool os::check_heap(bool force) {
4982 return true;
4983 }
4984
4985 // Is a (classpath) directory empty?
4986 bool os::dir_is_empty(const char* path) {
4987 DIR *dir = NULL;
4988 struct dirent *ptr;
4989
4990 dir = opendir(path);
4991 if (dir == NULL) return true;
4992
4993 // Scan the directory
4994 bool result = true;
4995 char buf[sizeof(struct dirent) + MAX_PATH];
4996 while (result && (ptr = ::readdir(dir)) != NULL) {
4997 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4998 result = false;
4999 }
5000 }
5001 closedir(dir);
5002 return result;
5003 }
5004
5005 // This code originates from JDK's sysOpen and open64_w
5006 // from src/solaris/hpi/src/system_md.c
5007
5008 int os::open(const char *path, int oflag, int mode) {
5009 if (strlen(path) > MAX_PATH - 1) {
5010 errno = ENAMETOOLONG;
5011 return -1;
5012 }
5013
5014 // All file descriptors that are opened in the Java process and not
5015 // specifically destined for a subprocess should have the close-on-exec
5016 // flag set. If we don't set it, then careless 3rd party native code
5017 // might fork and exec without closing all appropriate file descriptors
5018 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5019 // turn might:
5020 //
5021 // - cause end-of-file to fail to be detected on some file
5022 // descriptors, resulting in mysterious hangs, or
5023 //
5024 // - might cause an fopen in the subprocess to fail on a system
5025 // suffering from bug 1085341.
5026 //
5027 // (Yes, the default setting of the close-on-exec flag is a Unix
5028 // design flaw)
5029 //
5030 // See:
5031 // 1085341: 32-bit stdio routines should support file descriptors >255
5032 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5033 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5034 //
5035 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5036 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5037 // because it saves a system call and removes a small window where the flag
5038 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5039 // and we fall back to using FD_CLOEXEC (see below).
5040 #ifdef O_CLOEXEC
5041 oflag |= O_CLOEXEC;
5042 #endif
5043
5044 int fd = ::open64(path, oflag, mode);
5045 if (fd == -1) return -1;
5046
5047 //If the open succeeded, the file might still be a directory
5048 {
5049 struct stat64 buf64;
5050 int ret = ::fstat64(fd, &buf64);
5051 int st_mode = buf64.st_mode;
5052
5053 if (ret != -1) {
5054 if ((st_mode & S_IFMT) == S_IFDIR) {
5055 errno = EISDIR;
5056 ::close(fd);
5057 return -1;
5058 }
5059 } else {
5060 ::close(fd);
5061 return -1;
5062 }
5063 }
5064
5065 #ifdef FD_CLOEXEC
5066 // Validate that the use of the O_CLOEXEC flag on open above worked.
5067 // With recent kernels, we will perform this check exactly once.
5068 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5069 if (!O_CLOEXEC_is_known_to_work) {
5070 int flags = ::fcntl(fd, F_GETFD);
5071 if (flags != -1) {
5072 if ((flags & FD_CLOEXEC) != 0)
5073 O_CLOEXEC_is_known_to_work = 1;
5074 else
5075 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5076 }
5077 }
5078 #endif
5079
5080 return fd;
5081 }
5082
5083
5084 // create binary file, rewriting existing file if required
5085 int os::create_binary_file(const char* path, bool rewrite_existing) {
5086 int oflags = O_WRONLY | O_CREAT;
5087 if (!rewrite_existing) {
5088 oflags |= O_EXCL;
5089 }
5090 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5091 }
5092
5093 // return current position of file pointer
5094 jlong os::current_file_offset(int fd) {
5095 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5096 }
5097
5098 // move file pointer to the specified offset
5099 jlong os::seek_to_file_offset(int fd, jlong offset) {
5100 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5101 }
5102
5103 // This code originates from JDK's sysAvailable
5104 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5105
5106 int os::available(int fd, jlong *bytes) {
5107 jlong cur, end;
5108 int mode;
5109 struct stat64 buf64;
5110
5111 if (::fstat64(fd, &buf64) >= 0) {
5112 mode = buf64.st_mode;
5113 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5114 // XXX: is the following call interruptible? If so, this might
5115 // need to go through the INTERRUPT_IO() wrapper as for other
5116 // blocking, interruptible calls in this file.
5117 int n;
5118 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5119 *bytes = n;
5120 return 1;
5121 }
5122 }
5123 }
5124 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5125 return 0;
5126 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5127 return 0;
5128 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5129 return 0;
5130 }
5131 *bytes = end - cur;
5132 return 1;
5133 }
5134
5135 // Map a block of memory.
5136 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5137 char *addr, size_t bytes, bool read_only,
5138 bool allow_exec) {
5139 int prot;
5140 int flags = MAP_PRIVATE;
5141
5142 if (read_only) {
5143 prot = PROT_READ;
5144 } else {
5145 prot = PROT_READ | PROT_WRITE;
5146 }
5147
5148 if (allow_exec) {
5149 prot |= PROT_EXEC;
5150 }
5151
5152 if (addr != NULL) {
5153 flags |= MAP_FIXED;
5154 }
5155
5156 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5157 fd, file_offset);
5158 if (mapped_address == MAP_FAILED) {
5159 return NULL;
5160 }
5161 return mapped_address;
5162 }
5163
5164
5165 // Remap a block of memory.
5166 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5167 char *addr, size_t bytes, bool read_only,
5168 bool allow_exec) {
5169 // same as map_memory() on this OS
5170 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5171 allow_exec);
5172 }
5173
5174
5175 // Unmap a block of memory.
5176 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5177 return munmap(addr, bytes) == 0;
5178 }
5179
5180 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5181
5182 static clockid_t thread_cpu_clockid(Thread* thread) {
5183 pthread_t tid = thread->osthread()->pthread_id();
5184 clockid_t clockid;
5185
5186 // Get thread clockid
5187 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5188 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5189 return clockid;
5190 }
5191
5192 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5193 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5194 // of a thread.
5195 //
5196 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5197 // the fast estimate available on the platform.
5198
5199 jlong os::current_thread_cpu_time() {
5200 if (os::Linux::supports_fast_thread_cpu_time()) {
5201 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5202 } else {
5203 // return user + sys since the cost is the same
5204 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5205 }
5206 }
5207
5208 jlong os::thread_cpu_time(Thread* thread) {
5209 // consistent with what current_thread_cpu_time() returns
5210 if (os::Linux::supports_fast_thread_cpu_time()) {
5211 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5212 } else {
5213 return slow_thread_cpu_time(thread, true /* user + sys */);
5214 }
5215 }
5216
5217 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5218 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5219 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5220 } else {
5221 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5222 }
5223 }
5224
5225 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5226 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5227 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5228 } else {
5229 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5230 }
5231 }
5232
5233 // -1 on error.
5234 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5235 pid_t tid = thread->osthread()->thread_id();
5236 char *s;
5237 char stat[2048];
5238 int statlen;
5239 char proc_name[64];
5240 int count;
5241 long sys_time, user_time;
5242 char cdummy;
5243 int idummy;
5244 long ldummy;
5245 FILE *fp;
5246
5247 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5248 fp = fopen(proc_name, "r");
5249 if (fp == NULL) return -1;
5250 statlen = fread(stat, 1, 2047, fp);
5251 stat[statlen] = '\0';
5252 fclose(fp);
5253
5254 // Skip pid and the command string. Note that we could be dealing with
5255 // weird command names, e.g. user could decide to rename java launcher
5256 // to "java 1.4.2 :)", then the stat file would look like
5257 // 1234 (java 1.4.2 :)) R ... ...
5258 // We don't really need to know the command string, just find the last
5259 // occurrence of ")" and then start parsing from there. See bug 4726580.
5260 s = strrchr(stat, ')');
5261 if (s == NULL) return -1;
5262
5263 // Skip blank chars
5264 do { s++; } while (s && isspace(*s));
5265
5266 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5267 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5268 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5269 &user_time, &sys_time);
5270 if (count != 13) return -1;
5271 if (user_sys_cpu_time) {
5272 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5273 } else {
5274 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5275 }
5276 }
5277
5278 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5279 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5280 info_ptr->may_skip_backward = false; // elapsed time not wall time
5281 info_ptr->may_skip_forward = false; // elapsed time not wall time
5282 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5283 }
5284
5285 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5286 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5287 info_ptr->may_skip_backward = false; // elapsed time not wall time
5288 info_ptr->may_skip_forward = false; // elapsed time not wall time
5289 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5290 }
5291
5292 bool os::is_thread_cpu_time_supported() {
5293 return true;
5294 }
5295
5296 // System loadavg support. Returns -1 if load average cannot be obtained.
5297 // Linux doesn't yet have a (official) notion of processor sets,
5298 // so just return the system wide load average.
5299 int os::loadavg(double loadavg[], int nelem) {
5300 return ::getloadavg(loadavg, nelem);
5301 }
5302
5303 void os::pause() {
5304 char filename[MAX_PATH];
5305 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5306 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5307 } else {
5308 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5309 }
5310
5311 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5312 if (fd != -1) {
5313 struct stat buf;
5314 ::close(fd);
5315 while (::stat(filename, &buf) == 0) {
5316 (void)::poll(NULL, 0, 100);
5317 }
5318 } else {
5319 jio_fprintf(stderr,
5320 "Could not open pause file '%s', continuing immediately.\n", filename);
5321 }
5322 }
5323
5324
5325 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5326 // comment paragraphs are worth repeating here:
5327 //
5328 // Assumption:
5329 // Only one parker can exist on an event, which is why we allocate
5330 // them per-thread. Multiple unparkers can coexist.
5331 //
5332 // _Event serves as a restricted-range semaphore.
5333 // -1 : thread is blocked, i.e. there is a waiter
5334 // 0 : neutral: thread is running or ready,
5335 // could have been signaled after a wait started
5336 // 1 : signaled - thread is running or ready
5337 //
5338 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5339 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5340 // For specifics regarding the bug see GLIBC BUGID 261237 :
5341 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5342 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5343 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5344 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5345 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5346 // and monitorenter when we're using 1-0 locking. All those operations may result in
5347 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5348 // of libpthread avoids the problem, but isn't practical.
5349 //
5350 // Possible remedies:
5351 //
5352 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5353 // This is palliative and probabilistic, however. If the thread is preempted
5354 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5355 // than the minimum period may have passed, and the abstime may be stale (in the
5356 // past) resultin in a hang. Using this technique reduces the odds of a hang
5357 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5358 //
5359 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5360 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5361 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5362 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5363 // thread.
5364 //
5365 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5366 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5367 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5368 // This also works well. In fact it avoids kernel-level scalability impediments
5369 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5370 // timers in a graceful fashion.
5371 //
5372 // 4. When the abstime value is in the past it appears that control returns
5373 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5374 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5375 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5376 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5377 // It may be possible to avoid reinitialization by checking the return
5378 // value from pthread_cond_timedwait(). In addition to reinitializing the
5379 // condvar we must establish the invariant that cond_signal() is only called
5380 // within critical sections protected by the adjunct mutex. This prevents
5381 // cond_signal() from "seeing" a condvar that's in the midst of being
5382 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5383 // desirable signal-after-unlock optimization that avoids futile context switching.
5384 //
5385 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5386 // structure when a condvar is used or initialized. cond_destroy() would
5387 // release the helper structure. Our reinitialize-after-timedwait fix
5388 // put excessive stress on malloc/free and locks protecting the c-heap.
5389 //
5390 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5391 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5392 // and only enabling the work-around for vulnerable environments.
5393
5394 // utility to compute the abstime argument to timedwait:
5395 // millis is the relative timeout time
5396 // abstime will be the absolute timeout time
5397 // TODO: replace compute_abstime() with unpackTime()
5398
5399 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5400 if (millis < 0) millis = 0;
5401
5402 jlong seconds = millis / 1000;
5403 millis %= 1000;
5404 if (seconds > 50000000) { // see man cond_timedwait(3T)
5405 seconds = 50000000;
5406 }
5407
5408 if (os::supports_monotonic_clock()) {
5409 struct timespec now;
5410 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5411 assert_status(status == 0, status, "clock_gettime");
5412 abstime->tv_sec = now.tv_sec + seconds;
5413 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5414 if (nanos >= NANOSECS_PER_SEC) {
5415 abstime->tv_sec += 1;
5416 nanos -= NANOSECS_PER_SEC;
5417 }
5418 abstime->tv_nsec = nanos;
5419 } else {
5420 struct timeval now;
5421 int status = gettimeofday(&now, NULL);
5422 assert(status == 0, "gettimeofday");
5423 abstime->tv_sec = now.tv_sec + seconds;
5424 long usec = now.tv_usec + millis * 1000;
5425 if (usec >= 1000000) {
5426 abstime->tv_sec += 1;
5427 usec -= 1000000;
5428 }
5429 abstime->tv_nsec = usec * 1000;
5430 }
5431 return abstime;
5432 }
5433
5434 void os::PlatformEvent::park() { // AKA "down()"
5435 // Transitions for _Event:
5436 // -1 => -1 : illegal
5437 // 1 => 0 : pass - return immediately
5438 // 0 => -1 : block; then set _Event to 0 before returning
5439
5440 // Invariant: Only the thread associated with the Event/PlatformEvent
5441 // may call park().
5442 // TODO: assert that _Assoc != NULL or _Assoc == Self
5443 assert(_nParked == 0, "invariant");
5444
5445 int v;
5446 for (;;) {
5447 v = _Event;
5448 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5449 }
5450 guarantee(v >= 0, "invariant");
5451 if (v == 0) {
5452 // Do this the hard way by blocking ...
5453 int status = pthread_mutex_lock(_mutex);
5454 assert_status(status == 0, status, "mutex_lock");
5455 guarantee(_nParked == 0, "invariant");
5456 ++_nParked;
5457 while (_Event < 0) {
5458 status = pthread_cond_wait(_cond, _mutex);
5459 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5460 // Treat this the same as if the wait was interrupted
5461 if (status == ETIME) { status = EINTR; }
5462 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5463 }
5464 --_nParked;
5465
5466 _Event = 0;
5467 status = pthread_mutex_unlock(_mutex);
5468 assert_status(status == 0, status, "mutex_unlock");
5469 // Paranoia to ensure our locked and lock-free paths interact
5470 // correctly with each other.
5471 OrderAccess::fence();
5472 }
5473 guarantee(_Event >= 0, "invariant");
5474 }
5475
5476 int os::PlatformEvent::park(jlong millis) {
5477 // Transitions for _Event:
5478 // -1 => -1 : illegal
5479 // 1 => 0 : pass - return immediately
5480 // 0 => -1 : block; then set _Event to 0 before returning
5481
5482 guarantee(_nParked == 0, "invariant");
5483
5484 int v;
5485 for (;;) {
5486 v = _Event;
5487 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5488 }
5489 guarantee(v >= 0, "invariant");
5490 if (v != 0) return OS_OK;
5491
5492 // We do this the hard way, by blocking the thread.
5493 // Consider enforcing a minimum timeout value.
5494 struct timespec abst;
5495 compute_abstime(&abst, millis);
5496
5497 int ret = OS_TIMEOUT;
5498 int status = pthread_mutex_lock(_mutex);
5499 assert_status(status == 0, status, "mutex_lock");
5500 guarantee(_nParked == 0, "invariant");
5501 ++_nParked;
5502
5503 // Object.wait(timo) will return because of
5504 // (a) notification
5505 // (b) timeout
5506 // (c) thread.interrupt
5507 //
5508 // Thread.interrupt and object.notify{All} both call Event::set.
5509 // That is, we treat thread.interrupt as a special case of notification.
5510 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5511 // We assume all ETIME returns are valid.
5512 //
5513 // TODO: properly differentiate simultaneous notify+interrupt.
5514 // In that case, we should propagate the notify to another waiter.
5515
5516 while (_Event < 0) {
5517 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5518 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5519 pthread_cond_destroy(_cond);
5520 pthread_cond_init(_cond, os::Linux::condAttr());
5521 }
5522 assert_status(status == 0 || status == EINTR ||
5523 status == ETIME || status == ETIMEDOUT,
5524 status, "cond_timedwait");
5525 if (!FilterSpuriousWakeups) break; // previous semantics
5526 if (status == ETIME || status == ETIMEDOUT) break;
5527 // We consume and ignore EINTR and spurious wakeups.
5528 }
5529 --_nParked;
5530 if (_Event >= 0) {
5531 ret = OS_OK;
5532 }
5533 _Event = 0;
5534 status = pthread_mutex_unlock(_mutex);
5535 assert_status(status == 0, status, "mutex_unlock");
5536 assert(_nParked == 0, "invariant");
5537 // Paranoia to ensure our locked and lock-free paths interact
5538 // correctly with each other.
5539 OrderAccess::fence();
5540 return ret;
5541 }
5542
5543 void os::PlatformEvent::unpark() {
5544 // Transitions for _Event:
5545 // 0 => 1 : just return
5546 // 1 => 1 : just return
5547 // -1 => either 0 or 1; must signal target thread
5548 // That is, we can safely transition _Event from -1 to either
5549 // 0 or 1.
5550 // See also: "Semaphores in Plan 9" by Mullender & Cox
5551 //
5552 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5553 // that it will take two back-to-back park() calls for the owning
5554 // thread to block. This has the benefit of forcing a spurious return
5555 // from the first park() call after an unpark() call which will help
5556 // shake out uses of park() and unpark() without condition variables.
5557
5558 if (Atomic::xchg(1, &_Event) >= 0) return;
5559
5560 // Wait for the thread associated with the event to vacate
5561 int status = pthread_mutex_lock(_mutex);
5562 assert_status(status == 0, status, "mutex_lock");
5563 int AnyWaiters = _nParked;
5564 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5565 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5566 AnyWaiters = 0;
5567 pthread_cond_signal(_cond);
5568 }
5569 status = pthread_mutex_unlock(_mutex);
5570 assert_status(status == 0, status, "mutex_unlock");
5571 if (AnyWaiters != 0) {
5572 // Note that we signal() *after* dropping the lock for "immortal" Events.
5573 // This is safe and avoids a common class of futile wakeups. In rare
5574 // circumstances this can cause a thread to return prematurely from
5575 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5576 // will simply re-test the condition and re-park itself.
5577 // This provides particular benefit if the underlying platform does not
5578 // provide wait morphing.
5579 status = pthread_cond_signal(_cond);
5580 assert_status(status == 0, status, "cond_signal");
5581 }
5582 }
5583
5584
5585 // JSR166
5586 // -------------------------------------------------------
5587
5588 // The solaris and linux implementations of park/unpark are fairly
5589 // conservative for now, but can be improved. They currently use a
5590 // mutex/condvar pair, plus a a count.
5591 // Park decrements count if > 0, else does a condvar wait. Unpark
5592 // sets count to 1 and signals condvar. Only one thread ever waits
5593 // on the condvar. Contention seen when trying to park implies that someone
5594 // is unparking you, so don't wait. And spurious returns are fine, so there
5595 // is no need to track notifications.
5596
5597 // This code is common to linux and solaris and will be moved to a
5598 // common place in dolphin.
5599 //
5600 // The passed in time value is either a relative time in nanoseconds
5601 // or an absolute time in milliseconds. Either way it has to be unpacked
5602 // into suitable seconds and nanoseconds components and stored in the
5603 // given timespec structure.
5604 // Given time is a 64-bit value and the time_t used in the timespec is only
5605 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5606 // overflow if times way in the future are given. Further on Solaris versions
5607 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5608 // number of seconds, in abstime, is less than current_time + 100,000,000.
5609 // As it will be 28 years before "now + 100000000" will overflow we can
5610 // ignore overflow and just impose a hard-limit on seconds using the value
5611 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5612 // years from "now".
5613
5614 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5615 assert(time > 0, "convertTime");
5616 time_t max_secs = 0;
5617
5618 if (!os::supports_monotonic_clock() || isAbsolute) {
5619 struct timeval now;
5620 int status = gettimeofday(&now, NULL);
5621 assert(status == 0, "gettimeofday");
5622
5623 max_secs = now.tv_sec + MAX_SECS;
5624
5625 if (isAbsolute) {
5626 jlong secs = time / 1000;
5627 if (secs > max_secs) {
5628 absTime->tv_sec = max_secs;
5629 } else {
5630 absTime->tv_sec = secs;
5631 }
5632 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5633 } else {
5634 jlong secs = time / NANOSECS_PER_SEC;
5635 if (secs >= MAX_SECS) {
5636 absTime->tv_sec = max_secs;
5637 absTime->tv_nsec = 0;
5638 } else {
5639 absTime->tv_sec = now.tv_sec + secs;
5640 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5641 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5642 absTime->tv_nsec -= NANOSECS_PER_SEC;
5643 ++absTime->tv_sec; // note: this must be <= max_secs
5644 }
5645 }
5646 }
5647 } else {
5648 // must be relative using monotonic clock
5649 struct timespec now;
5650 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5651 assert_status(status == 0, status, "clock_gettime");
5652 max_secs = now.tv_sec + MAX_SECS;
5653 jlong secs = time / NANOSECS_PER_SEC;
5654 if (secs >= MAX_SECS) {
5655 absTime->tv_sec = max_secs;
5656 absTime->tv_nsec = 0;
5657 } else {
5658 absTime->tv_sec = now.tv_sec + secs;
5659 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5660 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5661 absTime->tv_nsec -= NANOSECS_PER_SEC;
5662 ++absTime->tv_sec; // note: this must be <= max_secs
5663 }
5664 }
5665 }
5666 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5667 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5668 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5669 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5670 }
5671
5672 void Parker::park(bool isAbsolute, jlong time) {
5673 // Ideally we'd do something useful while spinning, such
5674 // as calling unpackTime().
5675
5676 // Optional fast-path check:
5677 // Return immediately if a permit is available.
5678 // We depend on Atomic::xchg() having full barrier semantics
5679 // since we are doing a lock-free update to _counter.
5680 if (Atomic::xchg(0, &_counter) > 0) return;
5681
5682 Thread* thread = Thread::current();
5683 assert(thread->is_Java_thread(), "Must be JavaThread");
5684 JavaThread *jt = (JavaThread *)thread;
5685
5686 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5687 // Check interrupt before trying to wait
5688 if (Thread::is_interrupted(thread, false)) {
5689 return;
5690 }
5691
5692 // Next, demultiplex/decode time arguments
5693 timespec absTime;
5694 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5695 return;
5696 }
5697 if (time > 0) {
5698 unpackTime(&absTime, isAbsolute, time);
5699 }
5700
5701
5702 // Enter safepoint region
5703 // Beware of deadlocks such as 6317397.
5704 // The per-thread Parker:: mutex is a classic leaf-lock.
5705 // In particular a thread must never block on the Threads_lock while
5706 // holding the Parker:: mutex. If safepoints are pending both the
5707 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5708 ThreadBlockInVM tbivm(jt);
5709
5710 // Don't wait if cannot get lock since interference arises from
5711 // unblocking. Also. check interrupt before trying wait
5712 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5713 return;
5714 }
5715
5716 int status;
5717 if (_counter > 0) { // no wait needed
5718 _counter = 0;
5719 status = pthread_mutex_unlock(_mutex);
5720 assert(status == 0, "invariant");
5721 // Paranoia to ensure our locked and lock-free paths interact
5722 // correctly with each other and Java-level accesses.
5723 OrderAccess::fence();
5724 return;
5725 }
5726
5727 #ifdef ASSERT
5728 // Don't catch signals while blocked; let the running threads have the signals.
5729 // (This allows a debugger to break into the running thread.)
5730 sigset_t oldsigs;
5731 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5732 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5733 #endif
5734
5735 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5736 jt->set_suspend_equivalent();
5737 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5738
5739 assert(_cur_index == -1, "invariant");
5740 if (time == 0) {
5741 _cur_index = REL_INDEX; // arbitrary choice when not timed
5742 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5743 } else {
5744 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5745 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5746 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5747 pthread_cond_destroy(&_cond[_cur_index]);
5748 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5749 }
5750 }
5751 _cur_index = -1;
5752 assert_status(status == 0 || status == EINTR ||
5753 status == ETIME || status == ETIMEDOUT,
5754 status, "cond_timedwait");
5755
5756 #ifdef ASSERT
5757 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5758 #endif
5759
5760 _counter = 0;
5761 status = pthread_mutex_unlock(_mutex);
5762 assert_status(status == 0, status, "invariant");
5763 // Paranoia to ensure our locked and lock-free paths interact
5764 // correctly with each other and Java-level accesses.
5765 OrderAccess::fence();
5766
5767 // If externally suspended while waiting, re-suspend
5768 if (jt->handle_special_suspend_equivalent_condition()) {
5769 jt->java_suspend_self();
5770 }
5771 }
5772
5773 void Parker::unpark() {
5774 int status = pthread_mutex_lock(_mutex);
5775 assert(status == 0, "invariant");
5776 const int s = _counter;
5777 _counter = 1;
5778 if (s < 1) {
5779 // thread might be parked
5780 if (_cur_index != -1) {
5781 // thread is definitely parked
5782 if (WorkAroundNPTLTimedWaitHang) {
5783 status = pthread_cond_signal(&_cond[_cur_index]);
5784 assert(status == 0, "invariant");
5785 status = pthread_mutex_unlock(_mutex);
5786 assert(status == 0, "invariant");
5787 } else {
5788 status = pthread_mutex_unlock(_mutex);
5789 assert(status == 0, "invariant");
5790 status = pthread_cond_signal(&_cond[_cur_index]);
5791 assert(status == 0, "invariant");
5792 }
5793 } else {
5794 pthread_mutex_unlock(_mutex);
5795 assert(status == 0, "invariant");
5796 }
5797 } else {
5798 pthread_mutex_unlock(_mutex);
5799 assert(status == 0, "invariant");
5800 }
5801 }
5802
5803
5804 extern char** environ;
5805
5806 // Run the specified command in a separate process. Return its exit value,
5807 // or -1 on failure (e.g. can't fork a new process).
5808 // Unlike system(), this function can be called from signal handler. It
5809 // doesn't block SIGINT et al.
5810 int os::fork_and_exec(char* cmd) {
5811 const char * argv[4] = {"sh", "-c", cmd, NULL};
5812
5813 pid_t pid = fork();
5814
5815 if (pid < 0) {
5816 // fork failed
5817 return -1;
5818
5819 } else if (pid == 0) {
5820 // child process
5821
5822 execve("/bin/sh", (char* const*)argv, environ);
5823
5824 // execve failed
5825 _exit(-1);
5826
5827 } else {
5828 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5829 // care about the actual exit code, for now.
5830
5831 int status;
5832
5833 // Wait for the child process to exit. This returns immediately if
5834 // the child has already exited. */
5835 while (waitpid(pid, &status, 0) < 0) {
5836 switch (errno) {
5837 case ECHILD: return 0;
5838 case EINTR: break;
5839 default: return -1;
5840 }
5841 }
5842
5843 if (WIFEXITED(status)) {
5844 // The child exited normally; get its exit code.
5845 return WEXITSTATUS(status);
5846 } else if (WIFSIGNALED(status)) {
5847 // The child exited because of a signal
5848 // The best value to return is 0x80 + signal number,
5849 // because that is what all Unix shells do, and because
5850 // it allows callers to distinguish between process exit and
5851 // process death by signal.
5852 return 0x80 + WTERMSIG(status);
5853 } else {
5854 // Unknown exit code; pass it through
5855 return status;
5856 }
5857 }
5858 }
5859
5860 // is_headless_jre()
5861 //
5862 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5863 // in order to report if we are running in a headless jre
5864 //
5865 // Since JDK8 xawt/libmawt.so was moved into the same directory
5866 // as libawt.so, and renamed libawt_xawt.so
5867 //
5868 bool os::is_headless_jre() {
5869 struct stat statbuf;
5870 char buf[MAXPATHLEN];
5871 char libmawtpath[MAXPATHLEN];
5872 const char *xawtstr = "/xawt/libmawt.so";
5873 const char *new_xawtstr = "/libawt_xawt.so";
5874 char *p;
5875
5876 // Get path to libjvm.so
5877 os::jvm_path(buf, sizeof(buf));
5878
5879 // Get rid of libjvm.so
5880 p = strrchr(buf, '/');
5881 if (p == NULL) {
5882 return false;
5883 } else {
5884 *p = '\0';
5885 }
5886
5887 // Get rid of client or server
5888 p = strrchr(buf, '/');
5889 if (p == NULL) {
5890 return false;
5891 } else {
5892 *p = '\0';
5893 }
5894
5895 // check xawt/libmawt.so
5896 strcpy(libmawtpath, buf);
5897 strcat(libmawtpath, xawtstr);
5898 if (::stat(libmawtpath, &statbuf) == 0) return false;
5899
5900 // check libawt_xawt.so
5901 strcpy(libmawtpath, buf);
5902 strcat(libmawtpath, new_xawtstr);
5903 if (::stat(libmawtpath, &statbuf) == 0) return false;
5904
5905 return true;
5906 }
5907
5908 // Get the default path to the core file
5909 // Returns the length of the string
5910 int os::get_core_path(char* buffer, size_t bufferSize) {
5911 /*
5912 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5913 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5914 */
5915 const int core_pattern_len = 129;
5916 char core_pattern[core_pattern_len] = {0};
5917
5918 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5919 if (core_pattern_file != -1) {
5920 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5921 ::close(core_pattern_file);
5922
5923 if (ret > 0) {
5924 char *last_char = core_pattern + strlen(core_pattern) - 1;
5925
5926 if (*last_char == '\n') {
5927 *last_char = '\0';
5928 }
5929 }
5930 }
5931
5932 if (strlen(core_pattern) == 0) {
5933 return -1;
5934 }
5935
5936 char *pid_pos = strstr(core_pattern, "%p");
5937 int written;
5938
5939 if (core_pattern[0] == '/') {
5940 written = jio_snprintf(buffer, bufferSize, core_pattern);
5941 } else {
5942 char cwd[PATH_MAX];
5943
5944 const char* p = get_current_directory(cwd, PATH_MAX);
5945 if (p == NULL) {
5946 return -1;
5947 }
5948
5949 if (core_pattern[0] == '|') {
5950 written = jio_snprintf(buffer, bufferSize,
5951 "\"%s\" (or dumping to %s/core.%d)",
5952 &core_pattern[1], p, current_process_id());
5953 } else {
5954 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5955 }
5956 }
5957
5958 if (written < 0) {
5959 return -1;
5960 }
5961
5962 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5963 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5964
5965 if (core_uses_pid_file != -1) {
5966 char core_uses_pid = 0;
5967 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5968 ::close(core_uses_pid_file);
5969
5970 if (core_uses_pid == '1') {
5971 jio_snprintf(buffer + written, bufferSize - written,
5972 ".%d", current_process_id());
5973 }
5974 }
5975 }
5976
5977 return strlen(buffer);
5978 }
5979
5980 /////////////// Unit tests ///////////////
5981
5982 #ifndef PRODUCT
5983
5984 #define test_log(...) \
5985 do { \
5986 if (VerboseInternalVMTests) { \
5987 tty->print_cr(__VA_ARGS__); \
5988 tty->flush(); \
5989 } \
5990 } while (false)
5991
5992 class TestReserveMemorySpecial : AllStatic {
5993 public:
5994 static void small_page_write(void* addr, size_t size) {
5995 size_t page_size = os::vm_page_size();
5996
5997 char* end = (char*)addr + size;
5998 for (char* p = (char*)addr; p < end; p += page_size) {
5999 *p = 1;
6000 }
6001 }
6002
6003 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6004 if (!UseHugeTLBFS) {
6005 return;
6006 }
6007
6008 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6009
6010 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6011
6012 if (addr != NULL) {
6013 small_page_write(addr, size);
6014
6015 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6016 }
6017 }
6018
6019 static void test_reserve_memory_special_huge_tlbfs_only() {
6020 if (!UseHugeTLBFS) {
6021 return;
6022 }
6023
6024 size_t lp = os::large_page_size();
6025
6026 for (size_t size = lp; size <= lp * 10; size += lp) {
6027 test_reserve_memory_special_huge_tlbfs_only(size);
6028 }
6029 }
6030
6031 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6032 size_t lp = os::large_page_size();
6033 size_t ag = os::vm_allocation_granularity();
6034
6035 // sizes to test
6036 const size_t sizes[] = {
6037 lp, lp + ag, lp + lp / 2, lp * 2,
6038 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6039 lp * 10, lp * 10 + lp / 2
6040 };
6041 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6042
6043 // For each size/alignment combination, we test three scenarios:
6044 // 1) with req_addr == NULL
6045 // 2) with a non-null req_addr at which we expect to successfully allocate
6046 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6047 // expect the allocation to either fail or to ignore req_addr
6048
6049 // Pre-allocate two areas; they shall be as large as the largest allocation
6050 // and aligned to the largest alignment we will be testing.
6051 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6052 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6053 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6054 -1, 0);
6055 assert(mapping1 != MAP_FAILED, "should work");
6056
6057 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6058 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6059 -1, 0);
6060 assert(mapping2 != MAP_FAILED, "should work");
6061
6062 // Unmap the first mapping, but leave the second mapping intact: the first
6063 // mapping will serve as a value for a "good" req_addr (case 2). The second
6064 // mapping, still intact, as "bad" req_addr (case 3).
6065 ::munmap(mapping1, mapping_size);
6066
6067 // Case 1
6068 test_log("%s, req_addr NULL:", __FUNCTION__);
6069 test_log("size align result");
6070
6071 for (int i = 0; i < num_sizes; i++) {
6072 const size_t size = sizes[i];
6073 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6074 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6075 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6076 size, alignment, p, (p != NULL ? "" : "(failed)"));
6077 if (p != NULL) {
6078 assert(is_ptr_aligned(p, alignment), "must be");
6079 small_page_write(p, size);
6080 os::Linux::release_memory_special_huge_tlbfs(p, size);
6081 }
6082 }
6083 }
6084
6085 // Case 2
6086 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6087 test_log("size align req_addr result");
6088
6089 for (int i = 0; i < num_sizes; i++) {
6090 const size_t size = sizes[i];
6091 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6092 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6093 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6094 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6095 size, alignment, req_addr, p,
6096 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6097 if (p != NULL) {
6098 assert(p == req_addr, "must be");
6099 small_page_write(p, size);
6100 os::Linux::release_memory_special_huge_tlbfs(p, size);
6101 }
6102 }
6103 }
6104
6105 // Case 3
6106 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6107 test_log("size align req_addr result");
6108
6109 for (int i = 0; i < num_sizes; i++) {
6110 const size_t size = sizes[i];
6111 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6112 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6113 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6114 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6115 size, alignment, req_addr, p,
6116 ((p != NULL ? "" : "(failed)")));
6117 // as the area around req_addr contains already existing mappings, the API should always
6118 // return NULL (as per contract, it cannot return another address)
6119 assert(p == NULL, "must be");
6120 }
6121 }
6122
6123 ::munmap(mapping2, mapping_size);
6124
6125 }
6126
6127 static void test_reserve_memory_special_huge_tlbfs() {
6128 if (!UseHugeTLBFS) {
6129 return;
6130 }
6131
6132 test_reserve_memory_special_huge_tlbfs_only();
6133 test_reserve_memory_special_huge_tlbfs_mixed();
6134 }
6135
6136 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6137 if (!UseSHM) {
6138 return;
6139 }
6140
6141 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6142
6143 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6144
6145 if (addr != NULL) {
6146 assert(is_ptr_aligned(addr, alignment), "Check");
6147 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6148
6149 small_page_write(addr, size);
6150
6151 os::Linux::release_memory_special_shm(addr, size);
6152 }
6153 }
6154
6155 static void test_reserve_memory_special_shm() {
6156 size_t lp = os::large_page_size();
6157 size_t ag = os::vm_allocation_granularity();
6158
6159 for (size_t size = ag; size < lp * 3; size += ag) {
6160 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6161 test_reserve_memory_special_shm(size, alignment);
6162 }
6163 }
6164 }
6165
6166 static void test() {
6167 test_reserve_memory_special_huge_tlbfs();
6168 test_reserve_memory_special_shm();
6169 }
6170 };
6171
6172 void TestReserveMemorySpecial_test() {
6173 TestReserveMemorySpecial::test();
6174 }
6175
6176 #endif