1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
110
111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
112 // getrusage() is prepared to handle the associated failure.
113 #ifndef RUSAGE_THREAD
114 #define RUSAGE_THREAD (1) /* only the calling thread */
115 #endif
116
117 #define MAX_PATH (2 * K)
118
119 #define MAX_SECS 100000000
120
121 // for timer info max values which include all bits
122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123
124 #define LARGEPAGES_BIT (1 << 6)
125 ////////////////////////////////////////////////////////////////////////////////
126 // global variables
127 julong os::Linux::_physical_memory = 0;
128
129 address os::Linux::_initial_thread_stack_bottom = NULL;
130 uintptr_t os::Linux::_initial_thread_stack_size = 0;
131
132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 const int os::Linux::_vm_default_page_size = (8 * K);
139 bool os::Linux::_supports_fast_thread_cpu_time = false;
140 const char * os::Linux::_glibc_version = NULL;
141 const char * os::Linux::_libpthread_version = NULL;
142 pthread_condattr_t os::Linux::_condattr[1];
143
144 static jlong initial_time_count=0;
145
146 static int clock_tics_per_sec = 100;
147
148 // For diagnostics to print a message once. see run_periodic_checks
149 static sigset_t check_signal_done;
150 static bool check_signals = true;
151
152 // Signal number used to suspend/resume a thread
153
154 // do not use any signal number less than SIGSEGV, see 4355769
155 static int SR_signum = SIGUSR2;
156 sigset_t SR_sigset;
157
158 // Declarations
159 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
160
161 // utility functions
162
163 static int SR_initialize();
164
165 julong os::available_memory() {
166 return Linux::available_memory();
167 }
168
169 julong os::Linux::available_memory() {
170 // values in struct sysinfo are "unsigned long"
171 struct sysinfo si;
172 sysinfo(&si);
173
174 return (julong)si.freeram * si.mem_unit;
175 }
176
177 julong os::physical_memory() {
178 return Linux::physical_memory();
179 }
180
181 // Return true if user is running as root.
182
183 bool os::have_special_privileges() {
184 static bool init = false;
185 static bool privileges = false;
186 if (!init) {
187 privileges = (getuid() != geteuid()) || (getgid() != getegid());
188 init = true;
189 }
190 return privileges;
191 }
192
193
194 #ifndef SYS_gettid
195 // i386: 224, ia64: 1105, amd64: 186, sparc 143
196 #ifdef __ia64__
197 #define SYS_gettid 1105
198 #else
199 #ifdef __i386__
200 #define SYS_gettid 224
201 #else
202 #ifdef __amd64__
203 #define SYS_gettid 186
204 #else
205 #ifdef __sparc__
206 #define SYS_gettid 143
207 #else
208 #error define gettid for the arch
209 #endif
210 #endif
211 #endif
212 #endif
213 #endif
214
215 // Cpu architecture string
216 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
217
218
219 // pid_t gettid()
220 //
221 // Returns the kernel thread id of the currently running thread. Kernel
222 // thread id is used to access /proc.
223 pid_t os::Linux::gettid() {
224 int rslt = syscall(SYS_gettid);
225 assert(rslt != -1, "must be."); // old linuxthreads implementation?
226 return (pid_t)rslt;
227 }
228
229 // Most versions of linux have a bug where the number of processors are
230 // determined by looking at the /proc file system. In a chroot environment,
231 // the system call returns 1. This causes the VM to act as if it is
232 // a single processor and elide locking (see is_MP() call).
233 static bool unsafe_chroot_detected = false;
234 static const char *unstable_chroot_error = "/proc file system not found.\n"
235 "Java may be unstable running multithreaded in a chroot "
236 "environment on Linux when /proc filesystem is not mounted.";
237
238 void os::Linux::initialize_system_info() {
239 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
240 if (processor_count() == 1) {
241 pid_t pid = os::Linux::gettid();
242 char fname[32];
243 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
244 FILE *fp = fopen(fname, "r");
245 if (fp == NULL) {
246 unsafe_chroot_detected = true;
247 } else {
248 fclose(fp);
249 }
250 }
251 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
252 assert(processor_count() > 0, "linux error");
253 }
254
255 void os::init_system_properties_values() {
256 // The next steps are taken in the product version:
257 //
258 // Obtain the JAVA_HOME value from the location of libjvm.so.
259 // This library should be located at:
260 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
261 //
262 // If "/jre/lib/" appears at the right place in the path, then we
263 // assume libjvm.so is installed in a JDK and we use this path.
264 //
265 // Otherwise exit with message: "Could not create the Java virtual machine."
266 //
267 // The following extra steps are taken in the debugging version:
268 //
269 // If "/jre/lib/" does NOT appear at the right place in the path
270 // instead of exit check for $JAVA_HOME environment variable.
271 //
272 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
273 // then we append a fake suffix "hotspot/libjvm.so" to this path so
274 // it looks like libjvm.so is installed there
275 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
276 //
277 // Otherwise exit.
278 //
279 // Important note: if the location of libjvm.so changes this
280 // code needs to be changed accordingly.
281
282 // See ld(1):
283 // The linker uses the following search paths to locate required
284 // shared libraries:
285 // 1: ...
286 // ...
287 // 7: The default directories, normally /lib and /usr/lib.
288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
289 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
290 #else
291 #define DEFAULT_LIBPATH "/lib:/usr/lib"
292 #endif
293
294 // Base path of extensions installed on the system.
295 #define SYS_EXT_DIR "/usr/java/packages"
296 #define EXTENSIONS_DIR "/lib/ext"
297
298 // Buffer that fits several sprintfs.
299 // Note that the space for the colon and the trailing null are provided
300 // by the nulls included by the sizeof operator.
301 const size_t bufsize =
302 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
303 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
304 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
305
306 // sysclasspath, java_home, dll_dir
307 {
308 char *pslash;
309 os::jvm_path(buf, bufsize);
310
311 // Found the full path to libjvm.so.
312 // Now cut the path to <java_home>/jre if we can.
313 pslash = strrchr(buf, '/');
314 if (pslash != NULL) {
315 *pslash = '\0'; // Get rid of /libjvm.so.
316 }
317 pslash = strrchr(buf, '/');
318 if (pslash != NULL) {
319 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
320 }
321 Arguments::set_dll_dir(buf);
322
323 if (pslash != NULL) {
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /<arch>.
327 pslash = strrchr(buf, '/');
328 if (pslash != NULL) {
329 *pslash = '\0'; // Get rid of /lib.
330 }
331 }
332 }
333 Arguments::set_java_home(buf);
334 set_boot_path('/', ':');
335 }
336
337 // Where to look for native libraries.
338 //
339 // Note: Due to a legacy implementation, most of the library path
340 // is set in the launcher. This was to accomodate linking restrictions
341 // on legacy Linux implementations (which are no longer supported).
342 // Eventually, all the library path setting will be done here.
343 //
344 // However, to prevent the proliferation of improperly built native
345 // libraries, the new path component /usr/java/packages is added here.
346 // Eventually, all the library path setting will be done here.
347 {
348 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
349 // should always exist (until the legacy problem cited above is
350 // addressed).
351 const char *v = ::getenv("LD_LIBRARY_PATH");
352 const char *v_colon = ":";
353 if (v == NULL) { v = ""; v_colon = ""; }
354 // That's +1 for the colon and +1 for the trailing '\0'.
355 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
356 strlen(v) + 1 +
357 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
358 mtInternal);
359 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
360 Arguments::set_library_path(ld_library_path);
361 FREE_C_HEAP_ARRAY(char, ld_library_path);
362 }
363
364 // Extensions directories.
365 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
366 Arguments::set_ext_dirs(buf);
367
368 FREE_C_HEAP_ARRAY(char, buf);
369
370 #undef DEFAULT_LIBPATH
371 #undef SYS_EXT_DIR
372 #undef EXTENSIONS_DIR
373 }
374
375 ////////////////////////////////////////////////////////////////////////////////
376 // breakpoint support
377
378 void os::breakpoint() {
379 BREAKPOINT;
380 }
381
382 extern "C" void breakpoint() {
383 // use debugger to set breakpoint here
384 }
385
386 ////////////////////////////////////////////////////////////////////////////////
387 // signal support
388
389 debug_only(static bool signal_sets_initialized = false);
390 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
391
392 bool os::Linux::is_sig_ignored(int sig) {
393 struct sigaction oact;
394 sigaction(sig, (struct sigaction*)NULL, &oact);
395 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
396 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
397 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
398 return true;
399 } else {
400 return false;
401 }
402 }
403
404 void os::Linux::signal_sets_init() {
405 // Should also have an assertion stating we are still single-threaded.
406 assert(!signal_sets_initialized, "Already initialized");
407 // Fill in signals that are necessarily unblocked for all threads in
408 // the VM. Currently, we unblock the following signals:
409 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
410 // by -Xrs (=ReduceSignalUsage));
411 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
412 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
413 // the dispositions or masks wrt these signals.
414 // Programs embedding the VM that want to use the above signals for their
415 // own purposes must, at this time, use the "-Xrs" option to prevent
416 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
417 // (See bug 4345157, and other related bugs).
418 // In reality, though, unblocking these signals is really a nop, since
419 // these signals are not blocked by default.
420 sigemptyset(&unblocked_sigs);
421 sigemptyset(&allowdebug_blocked_sigs);
422 sigaddset(&unblocked_sigs, SIGILL);
423 sigaddset(&unblocked_sigs, SIGSEGV);
424 sigaddset(&unblocked_sigs, SIGBUS);
425 sigaddset(&unblocked_sigs, SIGFPE);
426 #if defined(PPC64)
427 sigaddset(&unblocked_sigs, SIGTRAP);
428 #endif
429 sigaddset(&unblocked_sigs, SR_signum);
430
431 if (!ReduceSignalUsage) {
432 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
433 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
434 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
435 }
436 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
437 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
438 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
439 }
440 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
441 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
443 }
444 }
445 // Fill in signals that are blocked by all but the VM thread.
446 sigemptyset(&vm_sigs);
447 if (!ReduceSignalUsage) {
448 sigaddset(&vm_sigs, BREAK_SIGNAL);
449 }
450 debug_only(signal_sets_initialized = true);
451
452 }
453
454 // These are signals that are unblocked while a thread is running Java.
455 // (For some reason, they get blocked by default.)
456 sigset_t* os::Linux::unblocked_signals() {
457 assert(signal_sets_initialized, "Not initialized");
458 return &unblocked_sigs;
459 }
460
461 // These are the signals that are blocked while a (non-VM) thread is
462 // running Java. Only the VM thread handles these signals.
463 sigset_t* os::Linux::vm_signals() {
464 assert(signal_sets_initialized, "Not initialized");
465 return &vm_sigs;
466 }
467
468 // These are signals that are blocked during cond_wait to allow debugger in
469 sigset_t* os::Linux::allowdebug_blocked_signals() {
470 assert(signal_sets_initialized, "Not initialized");
471 return &allowdebug_blocked_sigs;
472 }
473
474 void os::Linux::hotspot_sigmask(Thread* thread) {
475
476 //Save caller's signal mask before setting VM signal mask
477 sigset_t caller_sigmask;
478 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
479
480 OSThread* osthread = thread->osthread();
481 osthread->set_caller_sigmask(caller_sigmask);
482
483 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
484
485 if (!ReduceSignalUsage) {
486 if (thread->is_VM_thread()) {
487 // Only the VM thread handles BREAK_SIGNAL ...
488 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
489 } else {
490 // ... all other threads block BREAK_SIGNAL
491 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
492 }
493 }
494 }
495
496 //////////////////////////////////////////////////////////////////////////////
497 // detecting pthread library
498
499 void os::Linux::libpthread_init() {
500 // Save glibc and pthread version strings.
501 #if !defined(_CS_GNU_LIBC_VERSION) || \
502 !defined(_CS_GNU_LIBPTHREAD_VERSION)
503 #error "glibc too old (< 2.3.2)"
504 #endif
505
506 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
507 assert(n > 0, "cannot retrieve glibc version");
508 char *str = (char *)malloc(n, mtInternal);
509 confstr(_CS_GNU_LIBC_VERSION, str, n);
510 os::Linux::set_glibc_version(str);
511
512 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
513 assert(n > 0, "cannot retrieve pthread version");
514 str = (char *)malloc(n, mtInternal);
515 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
516 os::Linux::set_libpthread_version(str);
517 }
518
519 /////////////////////////////////////////////////////////////////////////////
520 // thread stack expansion
521
522 // os::Linux::manually_expand_stack() takes care of expanding the thread
523 // stack. Note that this is normally not needed: pthread stacks allocate
524 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
525 // committed. Therefore it is not necessary to expand the stack manually.
526 //
527 // Manually expanding the stack was historically needed on LinuxThreads
528 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
529 // it is kept to deal with very rare corner cases:
530 //
531 // For one, user may run the VM on an own implementation of threads
532 // whose stacks are - like the old LinuxThreads - implemented using
533 // mmap(MAP_GROWSDOWN).
534 //
535 // Also, this coding may be needed if the VM is running on the primordial
536 // thread. Normally we avoid running on the primordial thread; however,
537 // user may still invoke the VM on the primordial thread.
538 //
539 // The following historical comment describes the details about running
540 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
541
542
543 // Force Linux kernel to expand current thread stack. If "bottom" is close
544 // to the stack guard, caller should block all signals.
545 //
546 // MAP_GROWSDOWN:
547 // A special mmap() flag that is used to implement thread stacks. It tells
548 // kernel that the memory region should extend downwards when needed. This
549 // allows early versions of LinuxThreads to only mmap the first few pages
550 // when creating a new thread. Linux kernel will automatically expand thread
551 // stack as needed (on page faults).
552 //
553 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
554 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
555 // region, it's hard to tell if the fault is due to a legitimate stack
556 // access or because of reading/writing non-exist memory (e.g. buffer
557 // overrun). As a rule, if the fault happens below current stack pointer,
558 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
559 // application (see Linux kernel fault.c).
560 //
561 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
562 // stack overflow detection.
563 //
564 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
565 // not use MAP_GROWSDOWN.
566 //
567 // To get around the problem and allow stack banging on Linux, we need to
568 // manually expand thread stack after receiving the SIGSEGV.
569 //
570 // There are two ways to expand thread stack to address "bottom", we used
571 // both of them in JVM before 1.5:
572 // 1. adjust stack pointer first so that it is below "bottom", and then
573 // touch "bottom"
574 // 2. mmap() the page in question
575 //
576 // Now alternate signal stack is gone, it's harder to use 2. For instance,
577 // if current sp is already near the lower end of page 101, and we need to
578 // call mmap() to map page 100, it is possible that part of the mmap() frame
579 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
580 // That will destroy the mmap() frame and cause VM to crash.
581 //
582 // The following code works by adjusting sp first, then accessing the "bottom"
583 // page to force a page fault. Linux kernel will then automatically expand the
584 // stack mapping.
585 //
586 // _expand_stack_to() assumes its frame size is less than page size, which
587 // should always be true if the function is not inlined.
588
589 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
590 #define NOINLINE
591 #else
592 #define NOINLINE __attribute__ ((noinline))
593 #endif
594
595 static void _expand_stack_to(address bottom) NOINLINE;
596
597 static void _expand_stack_to(address bottom) {
598 address sp;
599 size_t size;
600 volatile char *p;
601
602 // Adjust bottom to point to the largest address within the same page, it
603 // gives us a one-page buffer if alloca() allocates slightly more memory.
604 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
605 bottom += os::Linux::page_size() - 1;
606
607 // sp might be slightly above current stack pointer; if that's the case, we
608 // will alloca() a little more space than necessary, which is OK. Don't use
609 // os::current_stack_pointer(), as its result can be slightly below current
610 // stack pointer, causing us to not alloca enough to reach "bottom".
611 sp = (address)&sp;
612
613 if (sp > bottom) {
614 size = sp - bottom;
615 p = (volatile char *)alloca(size);
616 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
617 p[0] = '\0';
618 }
619 }
620
621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
622 assert(t!=NULL, "just checking");
623 assert(t->osthread()->expanding_stack(), "expand should be set");
624 assert(t->stack_base() != NULL, "stack_base was not initialized");
625
626 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
627 sigset_t mask_all, old_sigset;
628 sigfillset(&mask_all);
629 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
630 _expand_stack_to(addr);
631 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
632 return true;
633 }
634 return false;
635 }
636
637 //////////////////////////////////////////////////////////////////////////////
638 // create new thread
639
640 // Thread start routine for all newly created threads
641 static void *java_start(Thread *thread) {
642 // Try to randomize the cache line index of hot stack frames.
643 // This helps when threads of the same stack traces evict each other's
644 // cache lines. The threads can be either from the same JVM instance, or
645 // from different JVM instances. The benefit is especially true for
646 // processors with hyperthreading technology.
647 static int counter = 0;
648 int pid = os::current_process_id();
649 alloca(((pid ^ counter++) & 7) * 128);
650
651 ThreadLocalStorage::set_thread(thread);
652
653 OSThread* osthread = thread->osthread();
654 Monitor* sync = osthread->startThread_lock();
655
656 // thread_id is kernel thread id (similar to Solaris LWP id)
657 osthread->set_thread_id(os::Linux::gettid());
658
659 if (UseNUMA) {
660 int lgrp_id = os::numa_get_group_id();
661 if (lgrp_id != -1) {
662 thread->set_lgrp_id(lgrp_id);
663 }
664 }
665 // initialize signal mask for this thread
666 os::Linux::hotspot_sigmask(thread);
667
668 // initialize floating point control register
669 os::Linux::init_thread_fpu_state();
670
671 // handshaking with parent thread
672 {
673 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
674
675 // notify parent thread
676 osthread->set_state(INITIALIZED);
677 sync->notify_all();
678
679 // wait until os::start_thread()
680 while (osthread->get_state() == INITIALIZED) {
681 sync->wait(Mutex::_no_safepoint_check_flag);
682 }
683 }
684
685 // call one more level start routine
686 thread->run();
687
688 return 0;
689 }
690
691 bool os::create_thread(Thread* thread, ThreadType thr_type,
692 size_t stack_size) {
693 assert(thread->osthread() == NULL, "caller responsible");
694
695 // Allocate the OSThread object
696 OSThread* osthread = new OSThread(NULL, NULL);
697 if (osthread == NULL) {
698 return false;
699 }
700
701 // set the correct thread state
702 osthread->set_thread_type(thr_type);
703
704 // Initial state is ALLOCATED but not INITIALIZED
705 osthread->set_state(ALLOCATED);
706
707 thread->set_osthread(osthread);
708
709 // init thread attributes
710 pthread_attr_t attr;
711 pthread_attr_init(&attr);
712 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
713
714 // stack size
715 if (os::Linux::supports_variable_stack_size()) {
716 // calculate stack size if it's not specified by caller
717 if (stack_size == 0) {
718 stack_size = os::Linux::default_stack_size(thr_type);
719
720 switch (thr_type) {
721 case os::java_thread:
722 // Java threads use ThreadStackSize which default value can be
723 // changed with the flag -Xss
724 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
725 stack_size = JavaThread::stack_size_at_create();
726 break;
727 case os::compiler_thread:
728 if (CompilerThreadStackSize > 0) {
729 stack_size = (size_t)(CompilerThreadStackSize * K);
730 break;
731 } // else fall through:
732 // use VMThreadStackSize if CompilerThreadStackSize is not defined
733 case os::vm_thread:
734 case os::pgc_thread:
735 case os::cgc_thread:
736 case os::watcher_thread:
737 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
738 break;
739 }
740 }
741
742 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
743 pthread_attr_setstacksize(&attr, stack_size);
744 } else {
745 // let pthread_create() pick the default value.
746 }
747
748 // glibc guard page
749 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
750
751 ThreadState state;
752
753 {
754 pthread_t tid;
755 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
756
757 pthread_attr_destroy(&attr);
758
759 if (ret != 0) {
760 if (PrintMiscellaneous && (Verbose || WizardMode)) {
761 perror("pthread_create()");
762 }
763 // Need to clean up stuff we've allocated so far
764 thread->set_osthread(NULL);
765 delete osthread;
766 return false;
767 }
768
769 // Store pthread info into the OSThread
770 osthread->set_pthread_id(tid);
771
772 // Wait until child thread is either initialized or aborted
773 {
774 Monitor* sync_with_child = osthread->startThread_lock();
775 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
776 while ((state = osthread->get_state()) == ALLOCATED) {
777 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
778 }
779 }
780 }
781
782 // Aborted due to thread limit being reached
783 if (state == ZOMBIE) {
784 thread->set_osthread(NULL);
785 delete osthread;
786 return false;
787 }
788
789 // The thread is returned suspended (in state INITIALIZED),
790 // and is started higher up in the call chain
791 assert(state == INITIALIZED, "race condition");
792 return true;
793 }
794
795 /////////////////////////////////////////////////////////////////////////////
796 // attach existing thread
797
798 // bootstrap the main thread
799 bool os::create_main_thread(JavaThread* thread) {
800 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
801 return create_attached_thread(thread);
802 }
803
804 bool os::create_attached_thread(JavaThread* thread) {
805 #ifdef ASSERT
806 thread->verify_not_published();
807 #endif
808
809 // Allocate the OSThread object
810 OSThread* osthread = new OSThread(NULL, NULL);
811
812 if (osthread == NULL) {
813 return false;
814 }
815
816 // Store pthread info into the OSThread
817 osthread->set_thread_id(os::Linux::gettid());
818 osthread->set_pthread_id(::pthread_self());
819
820 // initialize floating point control register
821 os::Linux::init_thread_fpu_state();
822
823 // Initial thread state is RUNNABLE
824 osthread->set_state(RUNNABLE);
825
826 thread->set_osthread(osthread);
827
828 if (UseNUMA) {
829 int lgrp_id = os::numa_get_group_id();
830 if (lgrp_id != -1) {
831 thread->set_lgrp_id(lgrp_id);
832 }
833 }
834
835 if (os::Linux::is_initial_thread()) {
836 // If current thread is initial thread, its stack is mapped on demand,
837 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
838 // the entire stack region to avoid SEGV in stack banging.
839 // It is also useful to get around the heap-stack-gap problem on SuSE
840 // kernel (see 4821821 for details). We first expand stack to the top
841 // of yellow zone, then enable stack yellow zone (order is significant,
842 // enabling yellow zone first will crash JVM on SuSE Linux), so there
843 // is no gap between the last two virtual memory regions.
844
845 JavaThread *jt = (JavaThread *)thread;
846 address addr = jt->stack_yellow_zone_base();
847 assert(addr != NULL, "initialization problem?");
848 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
849
850 osthread->set_expanding_stack();
851 os::Linux::manually_expand_stack(jt, addr);
852 osthread->clear_expanding_stack();
853 }
854
855 // initialize signal mask for this thread
856 // and save the caller's signal mask
857 os::Linux::hotspot_sigmask(thread);
858
859 return true;
860 }
861
862 void os::pd_start_thread(Thread* thread) {
863 OSThread * osthread = thread->osthread();
864 assert(osthread->get_state() != INITIALIZED, "just checking");
865 Monitor* sync_with_child = osthread->startThread_lock();
866 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
867 sync_with_child->notify();
868 }
869
870 // Free Linux resources related to the OSThread
871 void os::free_thread(OSThread* osthread) {
872 assert(osthread != NULL, "osthread not set");
873
874 if (Thread::current()->osthread() == osthread) {
875 // Restore caller's signal mask
876 sigset_t sigmask = osthread->caller_sigmask();
877 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
878 }
879
880 delete osthread;
881 }
882
883 //////////////////////////////////////////////////////////////////////////////
884 // thread local storage
885
886 // Restore the thread pointer if the destructor is called. This is in case
887 // someone from JNI code sets up a destructor with pthread_key_create to run
888 // detachCurrentThread on thread death. Unless we restore the thread pointer we
889 // will hang or crash. When detachCurrentThread is called the key will be set
890 // to null and we will not be called again. If detachCurrentThread is never
891 // called we could loop forever depending on the pthread implementation.
892 static void restore_thread_pointer(void* p) {
893 Thread* thread = (Thread*) p;
894 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
895 }
896
897 int os::allocate_thread_local_storage() {
898 pthread_key_t key;
899 int rslt = pthread_key_create(&key, restore_thread_pointer);
900 assert(rslt == 0, "cannot allocate thread local storage");
901 return (int)key;
902 }
903
904 // Note: This is currently not used by VM, as we don't destroy TLS key
905 // on VM exit.
906 void os::free_thread_local_storage(int index) {
907 int rslt = pthread_key_delete((pthread_key_t)index);
908 assert(rslt == 0, "invalid index");
909 }
910
911 void os::thread_local_storage_at_put(int index, void* value) {
912 int rslt = pthread_setspecific((pthread_key_t)index, value);
913 assert(rslt == 0, "pthread_setspecific failed");
914 }
915
916 extern "C" Thread* get_thread() {
917 return ThreadLocalStorage::thread();
918 }
919
920 //////////////////////////////////////////////////////////////////////////////
921 // initial thread
922
923 // Check if current thread is the initial thread, similar to Solaris thr_main.
924 bool os::Linux::is_initial_thread(void) {
925 char dummy;
926 // If called before init complete, thread stack bottom will be null.
927 // Can be called if fatal error occurs before initialization.
928 if (initial_thread_stack_bottom() == NULL) return false;
929 assert(initial_thread_stack_bottom() != NULL &&
930 initial_thread_stack_size() != 0,
931 "os::init did not locate initial thread's stack region");
932 if ((address)&dummy >= initial_thread_stack_bottom() &&
933 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
934 return true;
935 } else {
936 return false;
937 }
938 }
939
940 // Find the virtual memory area that contains addr
941 static bool find_vma(address addr, address* vma_low, address* vma_high) {
942 FILE *fp = fopen("/proc/self/maps", "r");
943 if (fp) {
944 address low, high;
945 while (!feof(fp)) {
946 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
947 if (low <= addr && addr < high) {
948 if (vma_low) *vma_low = low;
949 if (vma_high) *vma_high = high;
950 fclose(fp);
951 return true;
952 }
953 }
954 for (;;) {
955 int ch = fgetc(fp);
956 if (ch == EOF || ch == (int)'\n') break;
957 }
958 }
959 fclose(fp);
960 }
961 return false;
962 }
963
964 // Locate initial thread stack. This special handling of initial thread stack
965 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
966 // bogus value for initial thread.
967 void os::Linux::capture_initial_stack(size_t max_size) {
968 // stack size is the easy part, get it from RLIMIT_STACK
969 size_t stack_size;
970 struct rlimit rlim;
971 getrlimit(RLIMIT_STACK, &rlim);
972 stack_size = rlim.rlim_cur;
973
974 // 6308388: a bug in ld.so will relocate its own .data section to the
975 // lower end of primordial stack; reduce ulimit -s value a little bit
976 // so we won't install guard page on ld.so's data section.
977 stack_size -= 2 * page_size();
978
979 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
980 // 7.1, in both cases we will get 2G in return value.
981 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
982 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
983 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
984 // in case other parts in glibc still assumes 2M max stack size.
985 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
986 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
987 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
988 stack_size = 2 * K * K IA64_ONLY(*2);
989 }
990 // Try to figure out where the stack base (top) is. This is harder.
991 //
992 // When an application is started, glibc saves the initial stack pointer in
993 // a global variable "__libc_stack_end", which is then used by system
994 // libraries. __libc_stack_end should be pretty close to stack top. The
995 // variable is available since the very early days. However, because it is
996 // a private interface, it could disappear in the future.
997 //
998 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
999 // to __libc_stack_end, it is very close to stack top, but isn't the real
1000 // stack top. Note that /proc may not exist if VM is running as a chroot
1001 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1002 // /proc/<pid>/stat could change in the future (though unlikely).
1003 //
1004 // We try __libc_stack_end first. If that doesn't work, look for
1005 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1006 // as a hint, which should work well in most cases.
1007
1008 uintptr_t stack_start;
1009
1010 // try __libc_stack_end first
1011 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1012 if (p && *p) {
1013 stack_start = *p;
1014 } else {
1015 // see if we can get the start_stack field from /proc/self/stat
1016 FILE *fp;
1017 int pid;
1018 char state;
1019 int ppid;
1020 int pgrp;
1021 int session;
1022 int nr;
1023 int tpgrp;
1024 unsigned long flags;
1025 unsigned long minflt;
1026 unsigned long cminflt;
1027 unsigned long majflt;
1028 unsigned long cmajflt;
1029 unsigned long utime;
1030 unsigned long stime;
1031 long cutime;
1032 long cstime;
1033 long prio;
1034 long nice;
1035 long junk;
1036 long it_real;
1037 uintptr_t start;
1038 uintptr_t vsize;
1039 intptr_t rss;
1040 uintptr_t rsslim;
1041 uintptr_t scodes;
1042 uintptr_t ecode;
1043 int i;
1044
1045 // Figure what the primordial thread stack base is. Code is inspired
1046 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1047 // followed by command name surrounded by parentheses, state, etc.
1048 char stat[2048];
1049 int statlen;
1050
1051 fp = fopen("/proc/self/stat", "r");
1052 if (fp) {
1053 statlen = fread(stat, 1, 2047, fp);
1054 stat[statlen] = '\0';
1055 fclose(fp);
1056
1057 // Skip pid and the command string. Note that we could be dealing with
1058 // weird command names, e.g. user could decide to rename java launcher
1059 // to "java 1.4.2 :)", then the stat file would look like
1060 // 1234 (java 1.4.2 :)) R ... ...
1061 // We don't really need to know the command string, just find the last
1062 // occurrence of ")" and then start parsing from there. See bug 4726580.
1063 char * s = strrchr(stat, ')');
1064
1065 i = 0;
1066 if (s) {
1067 // Skip blank chars
1068 do { s++; } while (s && isspace(*s));
1069
1070 #define _UFM UINTX_FORMAT
1071 #define _DFM INTX_FORMAT
1072
1073 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1074 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1075 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1076 &state, // 3 %c
1077 &ppid, // 4 %d
1078 &pgrp, // 5 %d
1079 &session, // 6 %d
1080 &nr, // 7 %d
1081 &tpgrp, // 8 %d
1082 &flags, // 9 %lu
1083 &minflt, // 10 %lu
1084 &cminflt, // 11 %lu
1085 &majflt, // 12 %lu
1086 &cmajflt, // 13 %lu
1087 &utime, // 14 %lu
1088 &stime, // 15 %lu
1089 &cutime, // 16 %ld
1090 &cstime, // 17 %ld
1091 &prio, // 18 %ld
1092 &nice, // 19 %ld
1093 &junk, // 20 %ld
1094 &it_real, // 21 %ld
1095 &start, // 22 UINTX_FORMAT
1096 &vsize, // 23 UINTX_FORMAT
1097 &rss, // 24 INTX_FORMAT
1098 &rsslim, // 25 UINTX_FORMAT
1099 &scodes, // 26 UINTX_FORMAT
1100 &ecode, // 27 UINTX_FORMAT
1101 &stack_start); // 28 UINTX_FORMAT
1102 }
1103
1104 #undef _UFM
1105 #undef _DFM
1106
1107 if (i != 28 - 2) {
1108 assert(false, "Bad conversion from /proc/self/stat");
1109 // product mode - assume we are the initial thread, good luck in the
1110 // embedded case.
1111 warning("Can't detect initial thread stack location - bad conversion");
1112 stack_start = (uintptr_t) &rlim;
1113 }
1114 } else {
1115 // For some reason we can't open /proc/self/stat (for example, running on
1116 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1117 // most cases, so don't abort:
1118 warning("Can't detect initial thread stack location - no /proc/self/stat");
1119 stack_start = (uintptr_t) &rlim;
1120 }
1121 }
1122
1123 // Now we have a pointer (stack_start) very close to the stack top, the
1124 // next thing to do is to figure out the exact location of stack top. We
1125 // can find out the virtual memory area that contains stack_start by
1126 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1127 // and its upper limit is the real stack top. (again, this would fail if
1128 // running inside chroot, because /proc may not exist.)
1129
1130 uintptr_t stack_top;
1131 address low, high;
1132 if (find_vma((address)stack_start, &low, &high)) {
1133 // success, "high" is the true stack top. (ignore "low", because initial
1134 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1135 stack_top = (uintptr_t)high;
1136 } else {
1137 // failed, likely because /proc/self/maps does not exist
1138 warning("Can't detect initial thread stack location - find_vma failed");
1139 // best effort: stack_start is normally within a few pages below the real
1140 // stack top, use it as stack top, and reduce stack size so we won't put
1141 // guard page outside stack.
1142 stack_top = stack_start;
1143 stack_size -= 16 * page_size();
1144 }
1145
1146 // stack_top could be partially down the page so align it
1147 stack_top = align_size_up(stack_top, page_size());
1148
1149 if (max_size && stack_size > max_size) {
1150 _initial_thread_stack_size = max_size;
1151 } else {
1152 _initial_thread_stack_size = stack_size;
1153 }
1154
1155 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1156 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1157 }
1158
1159 ////////////////////////////////////////////////////////////////////////////////
1160 // time support
1161
1162 // Time since start-up in seconds to a fine granularity.
1163 // Used by VMSelfDestructTimer and the MemProfiler.
1164 double os::elapsedTime() {
1165
1166 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1167 }
1168
1169 jlong os::elapsed_counter() {
1170 return javaTimeNanos() - initial_time_count;
1171 }
1172
1173 jlong os::elapsed_frequency() {
1174 return NANOSECS_PER_SEC; // nanosecond resolution
1175 }
1176
1177 bool os::supports_vtime() { return true; }
1178 bool os::enable_vtime() { return false; }
1179 bool os::vtime_enabled() { return false; }
1180
1181 double os::elapsedVTime() {
1182 struct rusage usage;
1183 int retval = getrusage(RUSAGE_THREAD, &usage);
1184 if (retval == 0) {
1185 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1186 } else {
1187 // better than nothing, but not much
1188 return elapsedTime();
1189 }
1190 }
1191
1192 jlong os::javaTimeMillis() {
1193 timeval time;
1194 int status = gettimeofday(&time, NULL);
1195 assert(status != -1, "linux error");
1196 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1197 }
1198
1199 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1200 timeval time;
1201 int status = gettimeofday(&time, NULL);
1202 assert(status != -1, "linux error");
1203 seconds = jlong(time.tv_sec);
1204 nanos = jlong(time.tv_usec) * 1000;
1205 }
1206
1207
1208 #ifndef CLOCK_MONOTONIC
1209 #define CLOCK_MONOTONIC (1)
1210 #endif
1211
1212 void os::Linux::clock_init() {
1213 // we do dlopen's in this particular order due to bug in linux
1214 // dynamical loader (see 6348968) leading to crash on exit
1215 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1216 if (handle == NULL) {
1217 handle = dlopen("librt.so", RTLD_LAZY);
1218 }
1219
1220 if (handle) {
1221 int (*clock_getres_func)(clockid_t, struct timespec*) =
1222 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1223 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1224 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1225 if (clock_getres_func && clock_gettime_func) {
1226 // See if monotonic clock is supported by the kernel. Note that some
1227 // early implementations simply return kernel jiffies (updated every
1228 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1229 // for nano time (though the monotonic property is still nice to have).
1230 // It's fixed in newer kernels, however clock_getres() still returns
1231 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1232 // resolution for now. Hopefully as people move to new kernels, this
1233 // won't be a problem.
1234 struct timespec res;
1235 struct timespec tp;
1236 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1237 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1238 // yes, monotonic clock is supported
1239 _clock_gettime = clock_gettime_func;
1240 return;
1241 } else {
1242 // close librt if there is no monotonic clock
1243 dlclose(handle);
1244 }
1245 }
1246 }
1247 warning("No monotonic clock was available - timed services may " \
1248 "be adversely affected if the time-of-day clock changes");
1249 }
1250
1251 #ifndef SYS_clock_getres
1252 #if defined(IA32) || defined(AMD64)
1253 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1254 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1255 #else
1256 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1257 #define sys_clock_getres(x,y) -1
1258 #endif
1259 #else
1260 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1261 #endif
1262
1263 void os::Linux::fast_thread_clock_init() {
1264 if (!UseLinuxPosixThreadCPUClocks) {
1265 return;
1266 }
1267 clockid_t clockid;
1268 struct timespec tp;
1269 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1270 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1271
1272 // Switch to using fast clocks for thread cpu time if
1273 // the sys_clock_getres() returns 0 error code.
1274 // Note, that some kernels may support the current thread
1275 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1276 // returned by the pthread_getcpuclockid().
1277 // If the fast Posix clocks are supported then the sys_clock_getres()
1278 // must return at least tp.tv_sec == 0 which means a resolution
1279 // better than 1 sec. This is extra check for reliability.
1280
1281 if (pthread_getcpuclockid_func &&
1282 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1283 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1284 _supports_fast_thread_cpu_time = true;
1285 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1286 }
1287 }
1288
1289 jlong os::javaTimeNanos() {
1290 if (os::supports_monotonic_clock()) {
1291 struct timespec tp;
1292 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1293 assert(status == 0, "gettime error");
1294 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1295 return result;
1296 } else {
1297 timeval time;
1298 int status = gettimeofday(&time, NULL);
1299 assert(status != -1, "linux error");
1300 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1301 return 1000 * usecs;
1302 }
1303 }
1304
1305 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1306 if (os::supports_monotonic_clock()) {
1307 info_ptr->max_value = ALL_64_BITS;
1308
1309 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1310 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1311 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1312 } else {
1313 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1314 info_ptr->max_value = ALL_64_BITS;
1315
1316 // gettimeofday is a real time clock so it skips
1317 info_ptr->may_skip_backward = true;
1318 info_ptr->may_skip_forward = true;
1319 }
1320
1321 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1322 }
1323
1324 // Return the real, user, and system times in seconds from an
1325 // arbitrary fixed point in the past.
1326 bool os::getTimesSecs(double* process_real_time,
1327 double* process_user_time,
1328 double* process_system_time) {
1329 struct tms ticks;
1330 clock_t real_ticks = times(&ticks);
1331
1332 if (real_ticks == (clock_t) (-1)) {
1333 return false;
1334 } else {
1335 double ticks_per_second = (double) clock_tics_per_sec;
1336 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1337 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1338 *process_real_time = ((double) real_ticks) / ticks_per_second;
1339
1340 return true;
1341 }
1342 }
1343
1344
1345 char * os::local_time_string(char *buf, size_t buflen) {
1346 struct tm t;
1347 time_t long_time;
1348 time(&long_time);
1349 localtime_r(&long_time, &t);
1350 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1351 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1352 t.tm_hour, t.tm_min, t.tm_sec);
1353 return buf;
1354 }
1355
1356 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1357 return localtime_r(clock, res);
1358 }
1359
1360 ////////////////////////////////////////////////////////////////////////////////
1361 // runtime exit support
1362
1363 // Note: os::shutdown() might be called very early during initialization, or
1364 // called from signal handler. Before adding something to os::shutdown(), make
1365 // sure it is async-safe and can handle partially initialized VM.
1366 void os::shutdown() {
1367
1368 // allow PerfMemory to attempt cleanup of any persistent resources
1369 perfMemory_exit();
1370
1371 // needs to remove object in file system
1372 AttachListener::abort();
1373
1374 // flush buffered output, finish log files
1375 ostream_abort();
1376
1377 // Check for abort hook
1378 abort_hook_t abort_hook = Arguments::abort_hook();
1379 if (abort_hook != NULL) {
1380 abort_hook();
1381 }
1382
1383 }
1384
1385 // Note: os::abort() might be called very early during initialization, or
1386 // called from signal handler. Before adding something to os::abort(), make
1387 // sure it is async-safe and can handle partially initialized VM.
1388 void os::abort(bool dump_core, void* siginfo, void* context) {
1389 os::shutdown();
1390 if (dump_core) {
1391 #ifndef PRODUCT
1392 fdStream out(defaultStream::output_fd());
1393 out.print_raw("Current thread is ");
1394 char buf[16];
1395 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1396 out.print_raw_cr(buf);
1397 out.print_raw_cr("Dumping core ...");
1398 #endif
1399 ::abort(); // dump core
1400 }
1401
1402 ::exit(1);
1403 }
1404
1405 // Die immediately, no exit hook, no abort hook, no cleanup.
1406 void os::die() {
1407 ::abort();
1408 }
1409
1410
1411 // This method is a copy of JDK's sysGetLastErrorString
1412 // from src/solaris/hpi/src/system_md.c
1413
1414 size_t os::lasterror(char *buf, size_t len) {
1415 if (errno == 0) return 0;
1416
1417 const char *s = ::strerror(errno);
1418 size_t n = ::strlen(s);
1419 if (n >= len) {
1420 n = len - 1;
1421 }
1422 ::strncpy(buf, s, n);
1423 buf[n] = '\0';
1424 return n;
1425 }
1426
1427 intx os::current_thread_id() { return (intx)pthread_self(); }
1428 int os::current_process_id() {
1429 return ::getpid();
1430 }
1431
1432 // DLL functions
1433
1434 const char* os::dll_file_extension() { return ".so"; }
1435
1436 // This must be hard coded because it's the system's temporary
1437 // directory not the java application's temp directory, ala java.io.tmpdir.
1438 const char* os::get_temp_directory() { return "/tmp"; }
1439
1440 static bool file_exists(const char* filename) {
1441 struct stat statbuf;
1442 if (filename == NULL || strlen(filename) == 0) {
1443 return false;
1444 }
1445 return os::stat(filename, &statbuf) == 0;
1446 }
1447
1448 bool os::dll_build_name(char* buffer, size_t buflen,
1449 const char* pname, const char* fname) {
1450 bool retval = false;
1451 // Copied from libhpi
1452 const size_t pnamelen = pname ? strlen(pname) : 0;
1453
1454 // Return error on buffer overflow.
1455 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1456 return retval;
1457 }
1458
1459 if (pnamelen == 0) {
1460 snprintf(buffer, buflen, "lib%s.so", fname);
1461 retval = true;
1462 } else if (strchr(pname, *os::path_separator()) != NULL) {
1463 int n;
1464 char** pelements = split_path(pname, &n);
1465 if (pelements == NULL) {
1466 return false;
1467 }
1468 for (int i = 0; i < n; i++) {
1469 // Really shouldn't be NULL, but check can't hurt
1470 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1471 continue; // skip the empty path values
1472 }
1473 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1474 if (file_exists(buffer)) {
1475 retval = true;
1476 break;
1477 }
1478 }
1479 // release the storage
1480 for (int i = 0; i < n; i++) {
1481 if (pelements[i] != NULL) {
1482 FREE_C_HEAP_ARRAY(char, pelements[i]);
1483 }
1484 }
1485 if (pelements != NULL) {
1486 FREE_C_HEAP_ARRAY(char*, pelements);
1487 }
1488 } else {
1489 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1490 retval = true;
1491 }
1492 return retval;
1493 }
1494
1495 // check if addr is inside libjvm.so
1496 bool os::address_is_in_vm(address addr) {
1497 static address libjvm_base_addr;
1498 Dl_info dlinfo;
1499
1500 if (libjvm_base_addr == NULL) {
1501 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1502 libjvm_base_addr = (address)dlinfo.dli_fbase;
1503 }
1504 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1505 }
1506
1507 if (dladdr((void *)addr, &dlinfo) != 0) {
1508 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1509 }
1510
1511 return false;
1512 }
1513
1514 bool os::dll_address_to_function_name(address addr, char *buf,
1515 int buflen, int *offset,
1516 bool demangle) {
1517 // buf is not optional, but offset is optional
1518 assert(buf != NULL, "sanity check");
1519
1520 Dl_info dlinfo;
1521
1522 if (dladdr((void*)addr, &dlinfo) != 0) {
1523 // see if we have a matching symbol
1524 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1525 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1526 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1527 }
1528 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1529 return true;
1530 }
1531 // no matching symbol so try for just file info
1532 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1533 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1534 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1535 return true;
1536 }
1537 }
1538 }
1539
1540 buf[0] = '\0';
1541 if (offset != NULL) *offset = -1;
1542 return false;
1543 }
1544
1545 struct _address_to_library_name {
1546 address addr; // input : memory address
1547 size_t buflen; // size of fname
1548 char* fname; // output: library name
1549 address base; // library base addr
1550 };
1551
1552 static int address_to_library_name_callback(struct dl_phdr_info *info,
1553 size_t size, void *data) {
1554 int i;
1555 bool found = false;
1556 address libbase = NULL;
1557 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1558
1559 // iterate through all loadable segments
1560 for (i = 0; i < info->dlpi_phnum; i++) {
1561 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1562 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1563 // base address of a library is the lowest address of its loaded
1564 // segments.
1565 if (libbase == NULL || libbase > segbase) {
1566 libbase = segbase;
1567 }
1568 // see if 'addr' is within current segment
1569 if (segbase <= d->addr &&
1570 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1571 found = true;
1572 }
1573 }
1574 }
1575
1576 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1577 // so dll_address_to_library_name() can fall through to use dladdr() which
1578 // can figure out executable name from argv[0].
1579 if (found && info->dlpi_name && info->dlpi_name[0]) {
1580 d->base = libbase;
1581 if (d->fname) {
1582 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1583 }
1584 return 1;
1585 }
1586 return 0;
1587 }
1588
1589 bool os::dll_address_to_library_name(address addr, char* buf,
1590 int buflen, int* offset) {
1591 // buf is not optional, but offset is optional
1592 assert(buf != NULL, "sanity check");
1593
1594 Dl_info dlinfo;
1595 struct _address_to_library_name data;
1596
1597 // There is a bug in old glibc dladdr() implementation that it could resolve
1598 // to wrong library name if the .so file has a base address != NULL. Here
1599 // we iterate through the program headers of all loaded libraries to find
1600 // out which library 'addr' really belongs to. This workaround can be
1601 // removed once the minimum requirement for glibc is moved to 2.3.x.
1602 data.addr = addr;
1603 data.fname = buf;
1604 data.buflen = buflen;
1605 data.base = NULL;
1606 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1607
1608 if (rslt) {
1609 // buf already contains library name
1610 if (offset) *offset = addr - data.base;
1611 return true;
1612 }
1613 if (dladdr((void*)addr, &dlinfo) != 0) {
1614 if (dlinfo.dli_fname != NULL) {
1615 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1616 }
1617 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1618 *offset = addr - (address)dlinfo.dli_fbase;
1619 }
1620 return true;
1621 }
1622
1623 buf[0] = '\0';
1624 if (offset) *offset = -1;
1625 return false;
1626 }
1627
1628 // Loads .dll/.so and
1629 // in case of error it checks if .dll/.so was built for the
1630 // same architecture as Hotspot is running on
1631
1632
1633 // Remember the stack's state. The Linux dynamic linker will change
1634 // the stack to 'executable' at most once, so we must safepoint only once.
1635 bool os::Linux::_stack_is_executable = false;
1636
1637 // VM operation that loads a library. This is necessary if stack protection
1638 // of the Java stacks can be lost during loading the library. If we
1639 // do not stop the Java threads, they can stack overflow before the stacks
1640 // are protected again.
1641 class VM_LinuxDllLoad: public VM_Operation {
1642 private:
1643 const char *_filename;
1644 char *_ebuf;
1645 int _ebuflen;
1646 void *_lib;
1647 public:
1648 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1649 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1650 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1651 void doit() {
1652 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1653 os::Linux::_stack_is_executable = true;
1654 }
1655 void* loaded_library() { return _lib; }
1656 };
1657
1658 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1659 void * result = NULL;
1660 bool load_attempted = false;
1661
1662 // Check whether the library to load might change execution rights
1663 // of the stack. If they are changed, the protection of the stack
1664 // guard pages will be lost. We need a safepoint to fix this.
1665 //
1666 // See Linux man page execstack(8) for more info.
1667 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1668 ElfFile ef(filename);
1669 if (!ef.specifies_noexecstack()) {
1670 if (!is_init_completed()) {
1671 os::Linux::_stack_is_executable = true;
1672 // This is OK - No Java threads have been created yet, and hence no
1673 // stack guard pages to fix.
1674 //
1675 // This should happen only when you are building JDK7 using a very
1676 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1677 //
1678 // Dynamic loader will make all stacks executable after
1679 // this function returns, and will not do that again.
1680 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1681 } else {
1682 warning("You have loaded library %s which might have disabled stack guard. "
1683 "The VM will try to fix the stack guard now.\n"
1684 "It's highly recommended that you fix the library with "
1685 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1686 filename);
1687
1688 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1689 JavaThread *jt = JavaThread::current();
1690 if (jt->thread_state() != _thread_in_native) {
1691 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1692 // that requires ExecStack. Cannot enter safe point. Let's give up.
1693 warning("Unable to fix stack guard. Giving up.");
1694 } else {
1695 if (!LoadExecStackDllInVMThread) {
1696 // This is for the case where the DLL has an static
1697 // constructor function that executes JNI code. We cannot
1698 // load such DLLs in the VMThread.
1699 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1700 }
1701
1702 ThreadInVMfromNative tiv(jt);
1703 debug_only(VMNativeEntryWrapper vew;)
1704
1705 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1706 VMThread::execute(&op);
1707 if (LoadExecStackDllInVMThread) {
1708 result = op.loaded_library();
1709 }
1710 load_attempted = true;
1711 }
1712 }
1713 }
1714 }
1715
1716 if (!load_attempted) {
1717 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1718 }
1719
1720 if (result != NULL) {
1721 // Successful loading
1722 return result;
1723 }
1724
1725 Elf32_Ehdr elf_head;
1726 int diag_msg_max_length=ebuflen-strlen(ebuf);
1727 char* diag_msg_buf=ebuf+strlen(ebuf);
1728
1729 if (diag_msg_max_length==0) {
1730 // No more space in ebuf for additional diagnostics message
1731 return NULL;
1732 }
1733
1734
1735 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1736
1737 if (file_descriptor < 0) {
1738 // Can't open library, report dlerror() message
1739 return NULL;
1740 }
1741
1742 bool failed_to_read_elf_head=
1743 (sizeof(elf_head)!=
1744 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1745
1746 ::close(file_descriptor);
1747 if (failed_to_read_elf_head) {
1748 // file i/o error - report dlerror() msg
1749 return NULL;
1750 }
1751
1752 typedef struct {
1753 Elf32_Half code; // Actual value as defined in elf.h
1754 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1755 char elf_class; // 32 or 64 bit
1756 char endianess; // MSB or LSB
1757 char* name; // String representation
1758 } arch_t;
1759
1760 #ifndef EM_486
1761 #define EM_486 6 /* Intel 80486 */
1762 #endif
1763 #ifndef EM_AARCH64
1764 #define EM_AARCH64 183 /* ARM AARCH64 */
1765 #endif
1766
1767 static const arch_t arch_array[]={
1768 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1769 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1770 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1771 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1772 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1773 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1774 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1775 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1776 #if defined(VM_LITTLE_ENDIAN)
1777 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1778 #else
1779 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1780 #endif
1781 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1782 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1783 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1784 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1785 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1786 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1787 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1788 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1789 };
1790
1791 #if (defined IA32)
1792 static Elf32_Half running_arch_code=EM_386;
1793 #elif (defined AMD64)
1794 static Elf32_Half running_arch_code=EM_X86_64;
1795 #elif (defined IA64)
1796 static Elf32_Half running_arch_code=EM_IA_64;
1797 #elif (defined __sparc) && (defined _LP64)
1798 static Elf32_Half running_arch_code=EM_SPARCV9;
1799 #elif (defined __sparc) && (!defined _LP64)
1800 static Elf32_Half running_arch_code=EM_SPARC;
1801 #elif (defined __powerpc64__)
1802 static Elf32_Half running_arch_code=EM_PPC64;
1803 #elif (defined __powerpc__)
1804 static Elf32_Half running_arch_code=EM_PPC;
1805 #elif (defined ARM)
1806 static Elf32_Half running_arch_code=EM_ARM;
1807 #elif (defined S390)
1808 static Elf32_Half running_arch_code=EM_S390;
1809 #elif (defined ALPHA)
1810 static Elf32_Half running_arch_code=EM_ALPHA;
1811 #elif (defined MIPSEL)
1812 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1813 #elif (defined PARISC)
1814 static Elf32_Half running_arch_code=EM_PARISC;
1815 #elif (defined MIPS)
1816 static Elf32_Half running_arch_code=EM_MIPS;
1817 #elif (defined M68K)
1818 static Elf32_Half running_arch_code=EM_68K;
1819 #elif (defined AARCH64)
1820 static Elf32_Half running_arch_code=EM_AARCH64;
1821 #else
1822 #error Method os::dll_load requires that one of following is defined:\
1823 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1824 #endif
1825
1826 // Identify compatability class for VM's architecture and library's architecture
1827 // Obtain string descriptions for architectures
1828
1829 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1830 int running_arch_index=-1;
1831
1832 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1833 if (running_arch_code == arch_array[i].code) {
1834 running_arch_index = i;
1835 }
1836 if (lib_arch.code == arch_array[i].code) {
1837 lib_arch.compat_class = arch_array[i].compat_class;
1838 lib_arch.name = arch_array[i].name;
1839 }
1840 }
1841
1842 assert(running_arch_index != -1,
1843 "Didn't find running architecture code (running_arch_code) in arch_array");
1844 if (running_arch_index == -1) {
1845 // Even though running architecture detection failed
1846 // we may still continue with reporting dlerror() message
1847 return NULL;
1848 }
1849
1850 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1851 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1852 return NULL;
1853 }
1854
1855 #ifndef S390
1856 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1857 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1858 return NULL;
1859 }
1860 #endif // !S390
1861
1862 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1863 if (lib_arch.name!=NULL) {
1864 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1865 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1866 lib_arch.name, arch_array[running_arch_index].name);
1867 } else {
1868 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1869 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1870 lib_arch.code,
1871 arch_array[running_arch_index].name);
1872 }
1873 }
1874
1875 return NULL;
1876 }
1877
1878 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1879 int ebuflen) {
1880 void * result = ::dlopen(filename, RTLD_LAZY);
1881 if (result == NULL) {
1882 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1883 ebuf[ebuflen-1] = '\0';
1884 }
1885 return result;
1886 }
1887
1888 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1889 int ebuflen) {
1890 void * result = NULL;
1891 if (LoadExecStackDllInVMThread) {
1892 result = dlopen_helper(filename, ebuf, ebuflen);
1893 }
1894
1895 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1896 // library that requires an executable stack, or which does not have this
1897 // stack attribute set, dlopen changes the stack attribute to executable. The
1898 // read protection of the guard pages gets lost.
1899 //
1900 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1901 // may have been queued at the same time.
1902
1903 if (!_stack_is_executable) {
1904 JavaThread *jt = Threads::first();
1905
1906 while (jt) {
1907 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1908 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
1909 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1910 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1911 warning("Attempt to reguard stack yellow zone failed.");
1912 }
1913 }
1914 jt = jt->next();
1915 }
1916 }
1917
1918 return result;
1919 }
1920
1921 void* os::dll_lookup(void* handle, const char* name) {
1922 void* res = dlsym(handle, name);
1923 return res;
1924 }
1925
1926 void* os::get_default_process_handle() {
1927 return (void*)::dlopen(NULL, RTLD_LAZY);
1928 }
1929
1930 static bool _print_ascii_file(const char* filename, outputStream* st) {
1931 int fd = ::open(filename, O_RDONLY);
1932 if (fd == -1) {
1933 return false;
1934 }
1935
1936 char buf[32];
1937 int bytes;
1938 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1939 st->print_raw(buf, bytes);
1940 }
1941
1942 ::close(fd);
1943
1944 return true;
1945 }
1946
1947 void os::print_dll_info(outputStream *st) {
1948 st->print_cr("Dynamic libraries:");
1949
1950 char fname[32];
1951 pid_t pid = os::Linux::gettid();
1952
1953 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1954
1955 if (!_print_ascii_file(fname, st)) {
1956 st->print("Can not get library information for pid = %d\n", pid);
1957 }
1958 }
1959
1960 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1961 FILE *procmapsFile = NULL;
1962
1963 // Open the procfs maps file for the current process
1964 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1965 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1966 char line[PATH_MAX + 100];
1967
1968 // Read line by line from 'file'
1969 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1970 u8 base, top, offset, inode;
1971 char permissions[5];
1972 char device[6];
1973 char name[PATH_MAX + 1];
1974
1975 // Parse fields from line
1976 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
1977
1978 // Filter by device id '00:00' so that we only get file system mapped files.
1979 if (strcmp(device, "00:00") != 0) {
1980
1981 // Call callback with the fields of interest
1982 if(callback(name, (address)base, (address)top, param)) {
1983 // Oops abort, callback aborted
1984 fclose(procmapsFile);
1985 return 1;
1986 }
1987 }
1988 }
1989 fclose(procmapsFile);
1990 }
1991 return 0;
1992 }
1993
1994 void os::print_os_info_brief(outputStream* st) {
1995 os::Linux::print_distro_info(st);
1996
1997 os::Posix::print_uname_info(st);
1998
1999 os::Linux::print_libversion_info(st);
2000
2001 }
2002
2003 void os::print_os_info(outputStream* st) {
2004 st->print("OS:");
2005
2006 os::Linux::print_distro_info(st);
2007
2008 os::Posix::print_uname_info(st);
2009
2010 // Print warning if unsafe chroot environment detected
2011 if (unsafe_chroot_detected) {
2012 st->print("WARNING!! ");
2013 st->print_cr("%s", unstable_chroot_error);
2014 }
2015
2016 os::Linux::print_libversion_info(st);
2017
2018 os::Posix::print_rlimit_info(st);
2019
2020 os::Posix::print_load_average(st);
2021
2022 os::Linux::print_full_memory_info(st);
2023 }
2024
2025 // Try to identify popular distros.
2026 // Most Linux distributions have a /etc/XXX-release file, which contains
2027 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2028 // file that also contains the OS version string. Some have more than one
2029 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2030 // /etc/redhat-release.), so the order is important.
2031 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2032 // their own specific XXX-release file as well as a redhat-release file.
2033 // Because of this the XXX-release file needs to be searched for before the
2034 // redhat-release file.
2035 // Since Red Hat has a lsb-release file that is not very descriptive the
2036 // search for redhat-release needs to be before lsb-release.
2037 // Since the lsb-release file is the new standard it needs to be searched
2038 // before the older style release files.
2039 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2040 // next to last resort. The os-release file is a new standard that contains
2041 // distribution information and the system-release file seems to be an old
2042 // standard that has been replaced by the lsb-release and os-release files.
2043 // Searching for the debian_version file is the last resort. It contains
2044 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2045 // "Debian " is printed before the contents of the debian_version file.
2046
2047 const char* distro_files[] = {
2048 "/etc/oracle-release",
2049 "/etc/mandriva-release",
2050 "/etc/mandrake-release",
2051 "/etc/sun-release",
2052 "/etc/redhat-release",
2053 "/etc/lsb-release",
2054 "/etc/SuSE-release",
2055 "/etc/turbolinux-release",
2056 "/etc/gentoo-release",
2057 "/etc/ltib-release",
2058 "/etc/angstrom-version",
2059 "/etc/system-release",
2060 "/etc/os-release",
2061 NULL };
2062
2063 void os::Linux::print_distro_info(outputStream* st) {
2064 for (int i = 0;; i++) {
2065 const char* file = distro_files[i];
2066 if (file == NULL) break;
2067 // If file prints, we found it.
2068 if (_print_ascii_file(file, st)) return;
2069 }
2070 if (file_exists("/etc/debian_version")) {
2071 st->print("Debian ");
2072 _print_ascii_file("/etc/debian_version", st);
2073 } else {
2074 st->print("Linux");
2075 }
2076 st->cr();
2077 }
2078
2079 static void parse_os_info(char* distro, size_t length, const char* file) {
2080 FILE* fp = fopen(file, "r");
2081 if (fp != NULL) {
2082 char buf[257];
2083 // get last line of the file.
2084 while (!feof(fp)) {
2085 fgets(buf, sizeof(buf), fp);
2086 }
2087 // Edit out ubuntu things
2088 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2089 char* ptr = strstr(buf, "\"");
2090 if (ptr != NULL) {
2091 ptr++; // go beyond first "
2092 char* nl = strchr(ptr, '\"');
2093 if (nl != NULL) *nl = '\0';
2094 strncpy(distro, ptr, length);
2095 fclose(fp);
2096 return;
2097 }
2098 } else {
2099 // if not Ubuntu, print out whole line minus \n
2100 char* nl = strchr(buf, '\n');
2101 if (nl != NULL) *nl = '\0';
2102 strncpy(distro, buf, length);
2103 fclose(fp);
2104 return;
2105 }
2106 }
2107 }
2108
2109 void os::get_summary_os_info(char* buf, size_t buflen) {
2110 for (int i = 0;; i++) {
2111 const char* file = distro_files[i];
2112 if (file == NULL) break;
2113 if (file_exists(file)) {
2114 parse_os_info(buf, buflen, file);
2115 return;
2116 }
2117 }
2118 // special case for debian
2119 if (file_exists("/etc/debian_version")) {
2120 strncpy(buf, "Debian ", buflen);
2121 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2122 } else {
2123 strncpy(buf, "Linux", buflen);
2124 }
2125 }
2126
2127 void os::Linux::print_libversion_info(outputStream* st) {
2128 // libc, pthread
2129 st->print("libc:");
2130 st->print("%s ", os::Linux::glibc_version());
2131 st->print("%s ", os::Linux::libpthread_version());
2132 st->cr();
2133 }
2134
2135 void os::Linux::print_full_memory_info(outputStream* st) {
2136 st->print("\n/proc/meminfo:\n");
2137 _print_ascii_file("/proc/meminfo", st);
2138 st->cr();
2139 }
2140
2141 void os::print_memory_info(outputStream* st) {
2142
2143 st->print("Memory:");
2144 st->print(" %dk page", os::vm_page_size()>>10);
2145
2146 // values in struct sysinfo are "unsigned long"
2147 struct sysinfo si;
2148 sysinfo(&si);
2149
2150 st->print(", physical " UINT64_FORMAT "k",
2151 os::physical_memory() >> 10);
2152 st->print("(" UINT64_FORMAT "k free)",
2153 os::available_memory() >> 10);
2154 st->print(", swap " UINT64_FORMAT "k",
2155 ((jlong)si.totalswap * si.mem_unit) >> 10);
2156 st->print("(" UINT64_FORMAT "k free)",
2157 ((jlong)si.freeswap * si.mem_unit) >> 10);
2158 st->cr();
2159 }
2160
2161 // Print the first "model name" line and the first "flags" line
2162 // that we find and nothing more. We assume "model name" comes
2163 // before "flags" so if we find a second "model name", then the
2164 // "flags" field is considered missing.
2165 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2166 #if defined(IA32) || defined(AMD64)
2167 // Other platforms have less repetitive cpuinfo files
2168 FILE *fp = fopen("/proc/cpuinfo", "r");
2169 if (fp) {
2170 while (!feof(fp)) {
2171 if (fgets(buf, buflen, fp)) {
2172 // Assume model name comes before flags
2173 bool model_name_printed = false;
2174 if (strstr(buf, "model name") != NULL) {
2175 if (!model_name_printed) {
2176 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2177 st->print_raw(buf);
2178 model_name_printed = true;
2179 } else {
2180 // model name printed but not flags? Odd, just return
2181 fclose(fp);
2182 return true;
2183 }
2184 }
2185 // print the flags line too
2186 if (strstr(buf, "flags") != NULL) {
2187 st->print_raw(buf);
2188 fclose(fp);
2189 return true;
2190 }
2191 }
2192 }
2193 fclose(fp);
2194 }
2195 #endif // x86 platforms
2196 return false;
2197 }
2198
2199 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2200 // Only print the model name if the platform provides this as a summary
2201 if (!print_model_name_and_flags(st, buf, buflen)) {
2202 st->print("\n/proc/cpuinfo:\n");
2203 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2204 st->print_cr(" <Not Available>");
2205 }
2206 }
2207 }
2208
2209 const char* search_string = IA32_ONLY("model name") AMD64_ONLY("model name")
2210 IA64_ONLY("") SPARC_ONLY("cpu")
2211 ARM32_ONLY("Processor") PPC_ONLY("Processor") AARCH64_ONLY("Processor");
2212
2213 // Parses the cpuinfo file for string representing the model name.
2214 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2215 FILE* fp = fopen("/proc/cpuinfo", "r");
2216 if (fp != NULL) {
2217 while (!feof(fp)) {
2218 char buf[257];
2219 if (fgets(buf, sizeof(buf), fp)) {
2220 char* start = strstr(buf, search_string);
2221 if (start != NULL) {
2222 char *ptr = start + strlen(search_string);
2223 char *end = buf + strlen(buf);
2224 while (ptr != end) {
2225 // skip whitespace and colon for the rest of the name.
2226 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2227 break;
2228 }
2229 ptr++;
2230 }
2231 if (ptr != end) {
2232 // reasonable string, get rid of newline and keep the rest
2233 char* nl = strchr(buf, '\n');
2234 if (nl != NULL) *nl = '\0';
2235 strncpy(cpuinfo, ptr, length);
2236 fclose(fp);
2237 return;
2238 }
2239 }
2240 }
2241 }
2242 }
2243 // cpuinfo not found or parsing failed, just print generic string. The entire
2244 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2245 strncpy(cpuinfo, IA32_ONLY("x86 IA32") AMD64_ONLY("x86 IA64")
2246 IA64_ONLY("IA64") SPARC_ONLY("sparcv9")
2247 ARM32_ONLY("arm32") PPC_ONLY("PPC64") AARCH64_ONLY("AARCH64"), length);
2248 }
2249
2250 void os::print_siginfo(outputStream* st, void* siginfo) {
2251 const siginfo_t* si = (const siginfo_t*)siginfo;
2252
2253 os::Posix::print_siginfo_brief(st, si);
2254 #if INCLUDE_CDS
2255 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2256 UseSharedSpaces) {
2257 FileMapInfo* mapinfo = FileMapInfo::current_info();
2258 if (mapinfo->is_in_shared_space(si->si_addr)) {
2259 st->print("\n\nError accessing class data sharing archive." \
2260 " Mapped file inaccessible during execution, " \
2261 " possible disk/network problem.");
2262 }
2263 }
2264 #endif
2265 st->cr();
2266 }
2267
2268
2269 static void print_signal_handler(outputStream* st, int sig,
2270 char* buf, size_t buflen);
2271
2272 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2273 st->print_cr("Signal Handlers:");
2274 print_signal_handler(st, SIGSEGV, buf, buflen);
2275 print_signal_handler(st, SIGBUS , buf, buflen);
2276 print_signal_handler(st, SIGFPE , buf, buflen);
2277 print_signal_handler(st, SIGPIPE, buf, buflen);
2278 print_signal_handler(st, SIGXFSZ, buf, buflen);
2279 print_signal_handler(st, SIGILL , buf, buflen);
2280 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2281 print_signal_handler(st, SR_signum, buf, buflen);
2282 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2283 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2284 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2285 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2286 #if defined(PPC64)
2287 print_signal_handler(st, SIGTRAP, buf, buflen);
2288 #endif
2289 }
2290
2291 static char saved_jvm_path[MAXPATHLEN] = {0};
2292
2293 // Find the full path to the current module, libjvm.so
2294 void os::jvm_path(char *buf, jint buflen) {
2295 // Error checking.
2296 if (buflen < MAXPATHLEN) {
2297 assert(false, "must use a large-enough buffer");
2298 buf[0] = '\0';
2299 return;
2300 }
2301 // Lazy resolve the path to current module.
2302 if (saved_jvm_path[0] != 0) {
2303 strcpy(buf, saved_jvm_path);
2304 return;
2305 }
2306
2307 char dli_fname[MAXPATHLEN];
2308 bool ret = dll_address_to_library_name(
2309 CAST_FROM_FN_PTR(address, os::jvm_path),
2310 dli_fname, sizeof(dli_fname), NULL);
2311 assert(ret, "cannot locate libjvm");
2312 char *rp = NULL;
2313 if (ret && dli_fname[0] != '\0') {
2314 rp = realpath(dli_fname, buf);
2315 }
2316 if (rp == NULL) {
2317 return;
2318 }
2319
2320 if (Arguments::sun_java_launcher_is_altjvm()) {
2321 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2322 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2323 // If "/jre/lib/" appears at the right place in the string, then
2324 // assume we are installed in a JDK and we're done. Otherwise, check
2325 // for a JAVA_HOME environment variable and fix up the path so it
2326 // looks like libjvm.so is installed there (append a fake suffix
2327 // hotspot/libjvm.so).
2328 const char *p = buf + strlen(buf) - 1;
2329 for (int count = 0; p > buf && count < 5; ++count) {
2330 for (--p; p > buf && *p != '/'; --p)
2331 /* empty */ ;
2332 }
2333
2334 if (strncmp(p, "/jre/lib/", 9) != 0) {
2335 // Look for JAVA_HOME in the environment.
2336 char* java_home_var = ::getenv("JAVA_HOME");
2337 if (java_home_var != NULL && java_home_var[0] != 0) {
2338 char* jrelib_p;
2339 int len;
2340
2341 // Check the current module name "libjvm.so".
2342 p = strrchr(buf, '/');
2343 if (p == NULL) {
2344 return;
2345 }
2346 assert(strstr(p, "/libjvm") == p, "invalid library name");
2347
2348 rp = realpath(java_home_var, buf);
2349 if (rp == NULL) {
2350 return;
2351 }
2352
2353 // determine if this is a legacy image or modules image
2354 // modules image doesn't have "jre" subdirectory
2355 len = strlen(buf);
2356 assert(len < buflen, "Ran out of buffer room");
2357 jrelib_p = buf + len;
2358 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2359 if (0 != access(buf, F_OK)) {
2360 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2361 }
2362
2363 if (0 == access(buf, F_OK)) {
2364 // Use current module name "libjvm.so"
2365 len = strlen(buf);
2366 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2367 } else {
2368 // Go back to path of .so
2369 rp = realpath(dli_fname, buf);
2370 if (rp == NULL) {
2371 return;
2372 }
2373 }
2374 }
2375 }
2376 }
2377
2378 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2379 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2380 }
2381
2382 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2383 // no prefix required, not even "_"
2384 }
2385
2386 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2387 // no suffix required
2388 }
2389
2390 ////////////////////////////////////////////////////////////////////////////////
2391 // sun.misc.Signal support
2392
2393 static volatile jint sigint_count = 0;
2394
2395 static void UserHandler(int sig, void *siginfo, void *context) {
2396 // 4511530 - sem_post is serialized and handled by the manager thread. When
2397 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2398 // don't want to flood the manager thread with sem_post requests.
2399 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2400 return;
2401 }
2402
2403 // Ctrl-C is pressed during error reporting, likely because the error
2404 // handler fails to abort. Let VM die immediately.
2405 if (sig == SIGINT && is_error_reported()) {
2406 os::die();
2407 }
2408
2409 os::signal_notify(sig);
2410 }
2411
2412 void* os::user_handler() {
2413 return CAST_FROM_FN_PTR(void*, UserHandler);
2414 }
2415
2416 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2417 struct timespec ts;
2418 // Semaphore's are always associated with CLOCK_REALTIME
2419 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2420 // see unpackTime for discussion on overflow checking
2421 if (sec >= MAX_SECS) {
2422 ts.tv_sec += MAX_SECS;
2423 ts.tv_nsec = 0;
2424 } else {
2425 ts.tv_sec += sec;
2426 ts.tv_nsec += nsec;
2427 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2428 ts.tv_nsec -= NANOSECS_PER_SEC;
2429 ++ts.tv_sec; // note: this must be <= max_secs
2430 }
2431 }
2432
2433 return ts;
2434 }
2435
2436 extern "C" {
2437 typedef void (*sa_handler_t)(int);
2438 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2439 }
2440
2441 void* os::signal(int signal_number, void* handler) {
2442 struct sigaction sigAct, oldSigAct;
2443
2444 sigfillset(&(sigAct.sa_mask));
2445 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2446 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2447
2448 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2449 // -1 means registration failed
2450 return (void *)-1;
2451 }
2452
2453 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2454 }
2455
2456 void os::signal_raise(int signal_number) {
2457 ::raise(signal_number);
2458 }
2459
2460 // The following code is moved from os.cpp for making this
2461 // code platform specific, which it is by its very nature.
2462
2463 // Will be modified when max signal is changed to be dynamic
2464 int os::sigexitnum_pd() {
2465 return NSIG;
2466 }
2467
2468 // a counter for each possible signal value
2469 static volatile jint pending_signals[NSIG+1] = { 0 };
2470
2471 // Linux(POSIX) specific hand shaking semaphore.
2472 static sem_t sig_sem;
2473 static PosixSemaphore sr_semaphore;
2474
2475 void os::signal_init_pd() {
2476 // Initialize signal structures
2477 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2478
2479 // Initialize signal semaphore
2480 ::sem_init(&sig_sem, 0, 0);
2481 }
2482
2483 void os::signal_notify(int sig) {
2484 Atomic::inc(&pending_signals[sig]);
2485 ::sem_post(&sig_sem);
2486 }
2487
2488 static int check_pending_signals(bool wait) {
2489 Atomic::store(0, &sigint_count);
2490 for (;;) {
2491 for (int i = 0; i < NSIG + 1; i++) {
2492 jint n = pending_signals[i];
2493 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2494 return i;
2495 }
2496 }
2497 if (!wait) {
2498 return -1;
2499 }
2500 JavaThread *thread = JavaThread::current();
2501 ThreadBlockInVM tbivm(thread);
2502
2503 bool threadIsSuspended;
2504 do {
2505 thread->set_suspend_equivalent();
2506 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2507 ::sem_wait(&sig_sem);
2508
2509 // were we externally suspended while we were waiting?
2510 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2511 if (threadIsSuspended) {
2512 // The semaphore has been incremented, but while we were waiting
2513 // another thread suspended us. We don't want to continue running
2514 // while suspended because that would surprise the thread that
2515 // suspended us.
2516 ::sem_post(&sig_sem);
2517
2518 thread->java_suspend_self();
2519 }
2520 } while (threadIsSuspended);
2521 }
2522 }
2523
2524 int os::signal_lookup() {
2525 return check_pending_signals(false);
2526 }
2527
2528 int os::signal_wait() {
2529 return check_pending_signals(true);
2530 }
2531
2532 ////////////////////////////////////////////////////////////////////////////////
2533 // Virtual Memory
2534
2535 int os::vm_page_size() {
2536 // Seems redundant as all get out
2537 assert(os::Linux::page_size() != -1, "must call os::init");
2538 return os::Linux::page_size();
2539 }
2540
2541 // Solaris allocates memory by pages.
2542 int os::vm_allocation_granularity() {
2543 assert(os::Linux::page_size() != -1, "must call os::init");
2544 return os::Linux::page_size();
2545 }
2546
2547 // Rationale behind this function:
2548 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2549 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2550 // samples for JITted code. Here we create private executable mapping over the code cache
2551 // and then we can use standard (well, almost, as mapping can change) way to provide
2552 // info for the reporting script by storing timestamp and location of symbol
2553 void linux_wrap_code(char* base, size_t size) {
2554 static volatile jint cnt = 0;
2555
2556 if (!UseOprofile) {
2557 return;
2558 }
2559
2560 char buf[PATH_MAX+1];
2561 int num = Atomic::add(1, &cnt);
2562
2563 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2564 os::get_temp_directory(), os::current_process_id(), num);
2565 unlink(buf);
2566
2567 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2568
2569 if (fd != -1) {
2570 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2571 if (rv != (off_t)-1) {
2572 if (::write(fd, "", 1) == 1) {
2573 mmap(base, size,
2574 PROT_READ|PROT_WRITE|PROT_EXEC,
2575 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2576 }
2577 }
2578 ::close(fd);
2579 unlink(buf);
2580 }
2581 }
2582
2583 static bool recoverable_mmap_error(int err) {
2584 // See if the error is one we can let the caller handle. This
2585 // list of errno values comes from JBS-6843484. I can't find a
2586 // Linux man page that documents this specific set of errno
2587 // values so while this list currently matches Solaris, it may
2588 // change as we gain experience with this failure mode.
2589 switch (err) {
2590 case EBADF:
2591 case EINVAL:
2592 case ENOTSUP:
2593 // let the caller deal with these errors
2594 return true;
2595
2596 default:
2597 // Any remaining errors on this OS can cause our reserved mapping
2598 // to be lost. That can cause confusion where different data
2599 // structures think they have the same memory mapped. The worst
2600 // scenario is if both the VM and a library think they have the
2601 // same memory mapped.
2602 return false;
2603 }
2604 }
2605
2606 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2607 int err) {
2608 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2609 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2610 strerror(err), err);
2611 }
2612
2613 static void warn_fail_commit_memory(char* addr, size_t size,
2614 size_t alignment_hint, bool exec,
2615 int err) {
2616 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2617 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2618 alignment_hint, exec, strerror(err), err);
2619 }
2620
2621 // NOTE: Linux kernel does not really reserve the pages for us.
2622 // All it does is to check if there are enough free pages
2623 // left at the time of mmap(). This could be a potential
2624 // problem.
2625 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2626 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2627 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2628 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2629 if (res != (uintptr_t) MAP_FAILED) {
2630 if (UseNUMAInterleaving) {
2631 numa_make_global(addr, size);
2632 }
2633 return 0;
2634 }
2635
2636 int err = errno; // save errno from mmap() call above
2637
2638 if (!recoverable_mmap_error(err)) {
2639 warn_fail_commit_memory(addr, size, exec, err);
2640 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2641 }
2642
2643 return err;
2644 }
2645
2646 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2647 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2648 }
2649
2650 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2651 const char* mesg) {
2652 assert(mesg != NULL, "mesg must be specified");
2653 int err = os::Linux::commit_memory_impl(addr, size, exec);
2654 if (err != 0) {
2655 // the caller wants all commit errors to exit with the specified mesg:
2656 warn_fail_commit_memory(addr, size, exec, err);
2657 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2658 }
2659 }
2660
2661 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2662 #ifndef MAP_HUGETLB
2663 #define MAP_HUGETLB 0x40000
2664 #endif
2665
2666 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2667 #ifndef MADV_HUGEPAGE
2668 #define MADV_HUGEPAGE 14
2669 #endif
2670
2671 int os::Linux::commit_memory_impl(char* addr, size_t size,
2672 size_t alignment_hint, bool exec) {
2673 int err = os::Linux::commit_memory_impl(addr, size, exec);
2674 if (err == 0) {
2675 realign_memory(addr, size, alignment_hint);
2676 }
2677 return err;
2678 }
2679
2680 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2681 bool exec) {
2682 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2683 }
2684
2685 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2686 size_t alignment_hint, bool exec,
2687 const char* mesg) {
2688 assert(mesg != NULL, "mesg must be specified");
2689 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2690 if (err != 0) {
2691 // the caller wants all commit errors to exit with the specified mesg:
2692 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2693 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2694 }
2695 }
2696
2697 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2698 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2699 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2700 // be supported or the memory may already be backed by huge pages.
2701 ::madvise(addr, bytes, MADV_HUGEPAGE);
2702 }
2703 }
2704
2705 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2706 // This method works by doing an mmap over an existing mmaping and effectively discarding
2707 // the existing pages. However it won't work for SHM-based large pages that cannot be
2708 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2709 // small pages on top of the SHM segment. This method always works for small pages, so we
2710 // allow that in any case.
2711 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2712 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2713 }
2714 }
2715
2716 void os::numa_make_global(char *addr, size_t bytes) {
2717 Linux::numa_interleave_memory(addr, bytes);
2718 }
2719
2720 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2721 // bind policy to MPOL_PREFERRED for the current thread.
2722 #define USE_MPOL_PREFERRED 0
2723
2724 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2725 // To make NUMA and large pages more robust when both enabled, we need to ease
2726 // the requirements on where the memory should be allocated. MPOL_BIND is the
2727 // default policy and it will force memory to be allocated on the specified
2728 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2729 // the specified node, but will not force it. Using this policy will prevent
2730 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2731 // free large pages.
2732 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2733 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2734 }
2735
2736 bool os::numa_topology_changed() { return false; }
2737
2738 size_t os::numa_get_groups_num() {
2739 int max_node = Linux::numa_max_node();
2740 return max_node > 0 ? max_node + 1 : 1;
2741 }
2742
2743 int os::numa_get_group_id() {
2744 int cpu_id = Linux::sched_getcpu();
2745 if (cpu_id != -1) {
2746 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2747 if (lgrp_id != -1) {
2748 return lgrp_id;
2749 }
2750 }
2751 return 0;
2752 }
2753
2754 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2755 for (size_t i = 0; i < size; i++) {
2756 ids[i] = i;
2757 }
2758 return size;
2759 }
2760
2761 bool os::get_page_info(char *start, page_info* info) {
2762 return false;
2763 }
2764
2765 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2766 page_info* page_found) {
2767 return end;
2768 }
2769
2770
2771 int os::Linux::sched_getcpu_syscall(void) {
2772 unsigned int cpu;
2773 int retval = -1;
2774
2775 #if defined(IA32)
2776 #ifndef SYS_getcpu
2777 #define SYS_getcpu 318
2778 #endif
2779 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2780 #elif defined(AMD64)
2781 // Unfortunately we have to bring all these macros here from vsyscall.h
2782 // to be able to compile on old linuxes.
2783 #define __NR_vgetcpu 2
2784 #define VSYSCALL_START (-10UL << 20)
2785 #define VSYSCALL_SIZE 1024
2786 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2787 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2788 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2789 retval = vgetcpu(&cpu, NULL, NULL);
2790 #endif
2791
2792 return (retval == -1) ? retval : cpu;
2793 }
2794
2795 // Something to do with the numa-aware allocator needs these symbols
2796 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2797 extern "C" JNIEXPORT void numa_error(char *where) { }
2798 extern "C" JNIEXPORT int fork1() { return fork(); }
2799
2800
2801 // If we are running with libnuma version > 2, then we should
2802 // be trying to use symbols with versions 1.1
2803 // If we are running with earlier version, which did not have symbol versions,
2804 // we should use the base version.
2805 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2806 void *f = dlvsym(handle, name, "libnuma_1.1");
2807 if (f == NULL) {
2808 f = dlsym(handle, name);
2809 }
2810 return f;
2811 }
2812
2813 bool os::Linux::libnuma_init() {
2814 // sched_getcpu() should be in libc.
2815 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2816 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2817
2818 // If it's not, try a direct syscall.
2819 if (sched_getcpu() == -1) {
2820 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2821 (void*)&sched_getcpu_syscall));
2822 }
2823
2824 if (sched_getcpu() != -1) { // Does it work?
2825 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2826 if (handle != NULL) {
2827 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2828 libnuma_dlsym(handle, "numa_node_to_cpus")));
2829 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2830 libnuma_dlsym(handle, "numa_max_node")));
2831 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2832 libnuma_dlsym(handle, "numa_available")));
2833 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2834 libnuma_dlsym(handle, "numa_tonode_memory")));
2835 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2836 libnuma_dlsym(handle, "numa_interleave_memory")));
2837 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2838 libnuma_dlsym(handle, "numa_set_bind_policy")));
2839
2840
2841 if (numa_available() != -1) {
2842 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2843 // Create a cpu -> node mapping
2844 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2845 rebuild_cpu_to_node_map();
2846 return true;
2847 }
2848 }
2849 }
2850 return false;
2851 }
2852
2853 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2854 // The table is later used in get_node_by_cpu().
2855 void os::Linux::rebuild_cpu_to_node_map() {
2856 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2857 // in libnuma (possible values are starting from 16,
2858 // and continuing up with every other power of 2, but less
2859 // than the maximum number of CPUs supported by kernel), and
2860 // is a subject to change (in libnuma version 2 the requirements
2861 // are more reasonable) we'll just hardcode the number they use
2862 // in the library.
2863 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2864
2865 size_t cpu_num = os::active_processor_count();
2866 size_t cpu_map_size = NCPUS / BitsPerCLong;
2867 size_t cpu_map_valid_size =
2868 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2869
2870 cpu_to_node()->clear();
2871 cpu_to_node()->at_grow(cpu_num - 1);
2872 size_t node_num = numa_get_groups_num();
2873
2874 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2875 for (size_t i = 0; i < node_num; i++) {
2876 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2877 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2878 if (cpu_map[j] != 0) {
2879 for (size_t k = 0; k < BitsPerCLong; k++) {
2880 if (cpu_map[j] & (1UL << k)) {
2881 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2882 }
2883 }
2884 }
2885 }
2886 }
2887 }
2888 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2889 }
2890
2891 int os::Linux::get_node_by_cpu(int cpu_id) {
2892 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2893 return cpu_to_node()->at(cpu_id);
2894 }
2895 return -1;
2896 }
2897
2898 GrowableArray<int>* os::Linux::_cpu_to_node;
2899 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2900 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2901 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2902 os::Linux::numa_available_func_t os::Linux::_numa_available;
2903 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2904 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2905 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2906 unsigned long* os::Linux::_numa_all_nodes;
2907
2908 bool os::pd_uncommit_memory(char* addr, size_t size) {
2909 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2910 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2911 return res != (uintptr_t) MAP_FAILED;
2912 }
2913
2914 static address get_stack_commited_bottom(address bottom, size_t size) {
2915 address nbot = bottom;
2916 address ntop = bottom + size;
2917
2918 size_t page_sz = os::vm_page_size();
2919 unsigned pages = size / page_sz;
2920
2921 unsigned char vec[1];
2922 unsigned imin = 1, imax = pages + 1, imid;
2923 int mincore_return_value = 0;
2924
2925 assert(imin <= imax, "Unexpected page size");
2926
2927 while (imin < imax) {
2928 imid = (imax + imin) / 2;
2929 nbot = ntop - (imid * page_sz);
2930
2931 // Use a trick with mincore to check whether the page is mapped or not.
2932 // mincore sets vec to 1 if page resides in memory and to 0 if page
2933 // is swapped output but if page we are asking for is unmapped
2934 // it returns -1,ENOMEM
2935 mincore_return_value = mincore(nbot, page_sz, vec);
2936
2937 if (mincore_return_value == -1) {
2938 // Page is not mapped go up
2939 // to find first mapped page
2940 if (errno != EAGAIN) {
2941 assert(errno == ENOMEM, "Unexpected mincore errno");
2942 imax = imid;
2943 }
2944 } else {
2945 // Page is mapped go down
2946 // to find first not mapped page
2947 imin = imid + 1;
2948 }
2949 }
2950
2951 nbot = nbot + page_sz;
2952
2953 // Adjust stack bottom one page up if last checked page is not mapped
2954 if (mincore_return_value == -1) {
2955 nbot = nbot + page_sz;
2956 }
2957
2958 return nbot;
2959 }
2960
2961
2962 // Linux uses a growable mapping for the stack, and if the mapping for
2963 // the stack guard pages is not removed when we detach a thread the
2964 // stack cannot grow beyond the pages where the stack guard was
2965 // mapped. If at some point later in the process the stack expands to
2966 // that point, the Linux kernel cannot expand the stack any further
2967 // because the guard pages are in the way, and a segfault occurs.
2968 //
2969 // However, it's essential not to split the stack region by unmapping
2970 // a region (leaving a hole) that's already part of the stack mapping,
2971 // so if the stack mapping has already grown beyond the guard pages at
2972 // the time we create them, we have to truncate the stack mapping.
2973 // So, we need to know the extent of the stack mapping when
2974 // create_stack_guard_pages() is called.
2975
2976 // We only need this for stacks that are growable: at the time of
2977 // writing thread stacks don't use growable mappings (i.e. those
2978 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2979 // only applies to the main thread.
2980
2981 // If the (growable) stack mapping already extends beyond the point
2982 // where we're going to put our guard pages, truncate the mapping at
2983 // that point by munmap()ping it. This ensures that when we later
2984 // munmap() the guard pages we don't leave a hole in the stack
2985 // mapping. This only affects the main/initial thread
2986
2987 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2988 if (os::Linux::is_initial_thread()) {
2989 // As we manually grow stack up to bottom inside create_attached_thread(),
2990 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2991 // we don't need to do anything special.
2992 // Check it first, before calling heavy function.
2993 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2994 unsigned char vec[1];
2995
2996 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2997 // Fallback to slow path on all errors, including EAGAIN
2998 stack_extent = (uintptr_t) get_stack_commited_bottom(
2999 os::Linux::initial_thread_stack_bottom(),
3000 (size_t)addr - stack_extent);
3001 }
3002
3003 if (stack_extent < (uintptr_t)addr) {
3004 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3005 }
3006 }
3007
3008 return os::commit_memory(addr, size, !ExecMem);
3009 }
3010
3011 // If this is a growable mapping, remove the guard pages entirely by
3012 // munmap()ping them. If not, just call uncommit_memory(). This only
3013 // affects the main/initial thread, but guard against future OS changes
3014 // It's safe to always unmap guard pages for initial thread because we
3015 // always place it right after end of the mapped region
3016
3017 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3018 uintptr_t stack_extent, stack_base;
3019
3020 if (os::Linux::is_initial_thread()) {
3021 return ::munmap(addr, size) == 0;
3022 }
3023
3024 return os::uncommit_memory(addr, size);
3025 }
3026
3027 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3028 // at 'requested_addr'. If there are existing memory mappings at the same
3029 // location, however, they will be overwritten. If 'fixed' is false,
3030 // 'requested_addr' is only treated as a hint, the return value may or
3031 // may not start from the requested address. Unlike Linux mmap(), this
3032 // function returns NULL to indicate failure.
3033 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3034 char * addr;
3035 int flags;
3036
3037 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3038 if (fixed) {
3039 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3040 flags |= MAP_FIXED;
3041 }
3042
3043 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3044 // touch an uncommitted page. Otherwise, the read/write might
3045 // succeed if we have enough swap space to back the physical page.
3046 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3047 flags, -1, 0);
3048
3049 return addr == MAP_FAILED ? NULL : addr;
3050 }
3051
3052 static int anon_munmap(char * addr, size_t size) {
3053 return ::munmap(addr, size) == 0;
3054 }
3055
3056 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3057 size_t alignment_hint) {
3058 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3059 }
3060
3061 bool os::pd_release_memory(char* addr, size_t size) {
3062 return anon_munmap(addr, size);
3063 }
3064
3065 static bool linux_mprotect(char* addr, size_t size, int prot) {
3066 // Linux wants the mprotect address argument to be page aligned.
3067 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3068
3069 // According to SUSv3, mprotect() should only be used with mappings
3070 // established by mmap(), and mmap() always maps whole pages. Unaligned
3071 // 'addr' likely indicates problem in the VM (e.g. trying to change
3072 // protection of malloc'ed or statically allocated memory). Check the
3073 // caller if you hit this assert.
3074 assert(addr == bottom, "sanity check");
3075
3076 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3077 return ::mprotect(bottom, size, prot) == 0;
3078 }
3079
3080 // Set protections specified
3081 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3082 bool is_committed) {
3083 unsigned int p = 0;
3084 switch (prot) {
3085 case MEM_PROT_NONE: p = PROT_NONE; break;
3086 case MEM_PROT_READ: p = PROT_READ; break;
3087 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3088 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3089 default:
3090 ShouldNotReachHere();
3091 }
3092 // is_committed is unused.
3093 return linux_mprotect(addr, bytes, p);
3094 }
3095
3096 bool os::guard_memory(char* addr, size_t size) {
3097 return linux_mprotect(addr, size, PROT_NONE);
3098 }
3099
3100 bool os::unguard_memory(char* addr, size_t size) {
3101 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3102 }
3103
3104 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3105 size_t page_size) {
3106 bool result = false;
3107 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3108 MAP_ANONYMOUS|MAP_PRIVATE,
3109 -1, 0);
3110 if (p != MAP_FAILED) {
3111 void *aligned_p = align_ptr_up(p, page_size);
3112
3113 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3114
3115 munmap(p, page_size * 2);
3116 }
3117
3118 if (warn && !result) {
3119 warning("TransparentHugePages is not supported by the operating system.");
3120 }
3121
3122 return result;
3123 }
3124
3125 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3126 bool result = false;
3127 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3128 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3129 -1, 0);
3130
3131 if (p != MAP_FAILED) {
3132 // We don't know if this really is a huge page or not.
3133 FILE *fp = fopen("/proc/self/maps", "r");
3134 if (fp) {
3135 while (!feof(fp)) {
3136 char chars[257];
3137 long x = 0;
3138 if (fgets(chars, sizeof(chars), fp)) {
3139 if (sscanf(chars, "%lx-%*x", &x) == 1
3140 && x == (long)p) {
3141 if (strstr (chars, "hugepage")) {
3142 result = true;
3143 break;
3144 }
3145 }
3146 }
3147 }
3148 fclose(fp);
3149 }
3150 munmap(p, page_size);
3151 }
3152
3153 if (warn && !result) {
3154 warning("HugeTLBFS is not supported by the operating system.");
3155 }
3156
3157 return result;
3158 }
3159
3160 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3161 //
3162 // From the coredump_filter documentation:
3163 //
3164 // - (bit 0) anonymous private memory
3165 // - (bit 1) anonymous shared memory
3166 // - (bit 2) file-backed private memory
3167 // - (bit 3) file-backed shared memory
3168 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3169 // effective only if the bit 2 is cleared)
3170 // - (bit 5) hugetlb private memory
3171 // - (bit 6) hugetlb shared memory
3172 //
3173 static void set_coredump_filter(void) {
3174 FILE *f;
3175 long cdm;
3176
3177 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3178 return;
3179 }
3180
3181 if (fscanf(f, "%lx", &cdm) != 1) {
3182 fclose(f);
3183 return;
3184 }
3185
3186 rewind(f);
3187
3188 if ((cdm & LARGEPAGES_BIT) == 0) {
3189 cdm |= LARGEPAGES_BIT;
3190 fprintf(f, "%#lx", cdm);
3191 }
3192
3193 fclose(f);
3194 }
3195
3196 // Large page support
3197
3198 static size_t _large_page_size = 0;
3199
3200 size_t os::Linux::find_large_page_size() {
3201 size_t large_page_size = 0;
3202
3203 // large_page_size on Linux is used to round up heap size. x86 uses either
3204 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3205 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3206 // page as large as 256M.
3207 //
3208 // Here we try to figure out page size by parsing /proc/meminfo and looking
3209 // for a line with the following format:
3210 // Hugepagesize: 2048 kB
3211 //
3212 // If we can't determine the value (e.g. /proc is not mounted, or the text
3213 // format has been changed), we'll use the largest page size supported by
3214 // the processor.
3215
3216 #ifndef ZERO
3217 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3218 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3219 #endif // ZERO
3220
3221 FILE *fp = fopen("/proc/meminfo", "r");
3222 if (fp) {
3223 while (!feof(fp)) {
3224 int x = 0;
3225 char buf[16];
3226 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3227 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3228 large_page_size = x * K;
3229 break;
3230 }
3231 } else {
3232 // skip to next line
3233 for (;;) {
3234 int ch = fgetc(fp);
3235 if (ch == EOF || ch == (int)'\n') break;
3236 }
3237 }
3238 }
3239 fclose(fp);
3240 }
3241
3242 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3243 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3244 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3245 proper_unit_for_byte_size(large_page_size));
3246 }
3247
3248 return large_page_size;
3249 }
3250
3251 size_t os::Linux::setup_large_page_size() {
3252 _large_page_size = Linux::find_large_page_size();
3253 const size_t default_page_size = (size_t)Linux::page_size();
3254 if (_large_page_size > default_page_size) {
3255 _page_sizes[0] = _large_page_size;
3256 _page_sizes[1] = default_page_size;
3257 _page_sizes[2] = 0;
3258 }
3259
3260 return _large_page_size;
3261 }
3262
3263 bool os::Linux::setup_large_page_type(size_t page_size) {
3264 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3265 FLAG_IS_DEFAULT(UseSHM) &&
3266 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3267
3268 // The type of large pages has not been specified by the user.
3269
3270 // Try UseHugeTLBFS and then UseSHM.
3271 UseHugeTLBFS = UseSHM = true;
3272
3273 // Don't try UseTransparentHugePages since there are known
3274 // performance issues with it turned on. This might change in the future.
3275 UseTransparentHugePages = false;
3276 }
3277
3278 if (UseTransparentHugePages) {
3279 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3280 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3281 UseHugeTLBFS = false;
3282 UseSHM = false;
3283 return true;
3284 }
3285 UseTransparentHugePages = false;
3286 }
3287
3288 if (UseHugeTLBFS) {
3289 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3290 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3291 UseSHM = false;
3292 return true;
3293 }
3294 UseHugeTLBFS = false;
3295 }
3296
3297 return UseSHM;
3298 }
3299
3300 void os::large_page_init() {
3301 if (!UseLargePages &&
3302 !UseTransparentHugePages &&
3303 !UseHugeTLBFS &&
3304 !UseSHM) {
3305 // Not using large pages.
3306 return;
3307 }
3308
3309 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3310 // The user explicitly turned off large pages.
3311 // Ignore the rest of the large pages flags.
3312 UseTransparentHugePages = false;
3313 UseHugeTLBFS = false;
3314 UseSHM = false;
3315 return;
3316 }
3317
3318 size_t large_page_size = Linux::setup_large_page_size();
3319 UseLargePages = Linux::setup_large_page_type(large_page_size);
3320
3321 set_coredump_filter();
3322 }
3323
3324 #ifndef SHM_HUGETLB
3325 #define SHM_HUGETLB 04000
3326 #endif
3327
3328 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3329 char* req_addr, bool exec) {
3330 // "exec" is passed in but not used. Creating the shared image for
3331 // the code cache doesn't have an SHM_X executable permission to check.
3332 assert(UseLargePages && UseSHM, "only for SHM large pages");
3333 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3334
3335 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3336 return NULL; // Fallback to small pages.
3337 }
3338
3339 key_t key = IPC_PRIVATE;
3340 char *addr;
3341
3342 bool warn_on_failure = UseLargePages &&
3343 (!FLAG_IS_DEFAULT(UseLargePages) ||
3344 !FLAG_IS_DEFAULT(UseSHM) ||
3345 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3346 char msg[128];
3347
3348 // Create a large shared memory region to attach to based on size.
3349 // Currently, size is the total size of the heap
3350 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3351 if (shmid == -1) {
3352 // Possible reasons for shmget failure:
3353 // 1. shmmax is too small for Java heap.
3354 // > check shmmax value: cat /proc/sys/kernel/shmmax
3355 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3356 // 2. not enough large page memory.
3357 // > check available large pages: cat /proc/meminfo
3358 // > increase amount of large pages:
3359 // echo new_value > /proc/sys/vm/nr_hugepages
3360 // Note 1: different Linux may use different name for this property,
3361 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3362 // Note 2: it's possible there's enough physical memory available but
3363 // they are so fragmented after a long run that they can't
3364 // coalesce into large pages. Try to reserve large pages when
3365 // the system is still "fresh".
3366 if (warn_on_failure) {
3367 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3368 warning("%s", msg);
3369 }
3370 return NULL;
3371 }
3372
3373 // attach to the region
3374 addr = (char*)shmat(shmid, req_addr, 0);
3375 int err = errno;
3376
3377 // Remove shmid. If shmat() is successful, the actual shared memory segment
3378 // will be deleted when it's detached by shmdt() or when the process
3379 // terminates. If shmat() is not successful this will remove the shared
3380 // segment immediately.
3381 shmctl(shmid, IPC_RMID, NULL);
3382
3383 if ((intptr_t)addr == -1) {
3384 if (warn_on_failure) {
3385 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3386 warning("%s", msg);
3387 }
3388 return NULL;
3389 }
3390
3391 return addr;
3392 }
3393
3394 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3395 int error) {
3396 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3397
3398 bool warn_on_failure = UseLargePages &&
3399 (!FLAG_IS_DEFAULT(UseLargePages) ||
3400 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3401 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3402
3403 if (warn_on_failure) {
3404 char msg[128];
3405 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3406 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3407 warning("%s", msg);
3408 }
3409 }
3410
3411 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3412 char* req_addr,
3413 bool exec) {
3414 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3415 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3416 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3417
3418 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3419 char* addr = (char*)::mmap(req_addr, bytes, prot,
3420 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3421 -1, 0);
3422
3423 if (addr == MAP_FAILED) {
3424 warn_on_large_pages_failure(req_addr, bytes, errno);
3425 return NULL;
3426 }
3427
3428 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3429
3430 return addr;
3431 }
3432
3433 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3434 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3435 // (req_addr != NULL) or with a given alignment.
3436 // - bytes shall be a multiple of alignment.
3437 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3438 // - alignment sets the alignment at which memory shall be allocated.
3439 // It must be a multiple of allocation granularity.
3440 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3441 // req_addr or NULL.
3442 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3443
3444 size_t extra_size = bytes;
3445 if (req_addr == NULL && alignment > 0) {
3446 extra_size += alignment;
3447 }
3448
3449 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3450 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3451 -1, 0);
3452 if (start == MAP_FAILED) {
3453 start = NULL;
3454 } else {
3455 if (req_addr != NULL) {
3456 if (start != req_addr) {
3457 ::munmap(start, extra_size);
3458 start = NULL;
3459 }
3460 } else {
3461 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3462 char* const end_aligned = start_aligned + bytes;
3463 char* const end = start + extra_size;
3464 if (start_aligned > start) {
3465 ::munmap(start, start_aligned - start);
3466 }
3467 if (end_aligned < end) {
3468 ::munmap(end_aligned, end - end_aligned);
3469 }
3470 start = start_aligned;
3471 }
3472 }
3473 return start;
3474
3475 }
3476
3477 // Reserve memory using mmap(MAP_HUGETLB).
3478 // - bytes shall be a multiple of alignment.
3479 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3480 // - alignment sets the alignment at which memory shall be allocated.
3481 // It must be a multiple of allocation granularity.
3482 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3483 // req_addr or NULL.
3484 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3485 size_t alignment,
3486 char* req_addr,
3487 bool exec) {
3488 size_t large_page_size = os::large_page_size();
3489 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3490
3491 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3492 assert(is_size_aligned(bytes, alignment), "Must be");
3493
3494 // First reserve - but not commit - the address range in small pages.
3495 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3496
3497 if (start == NULL) {
3498 return NULL;
3499 }
3500
3501 assert(is_ptr_aligned(start, alignment), "Must be");
3502
3503 char* end = start + bytes;
3504
3505 // Find the regions of the allocated chunk that can be promoted to large pages.
3506 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3507 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3508
3509 size_t lp_bytes = lp_end - lp_start;
3510
3511 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3512
3513 if (lp_bytes == 0) {
3514 // The mapped region doesn't even span the start and the end of a large page.
3515 // Fall back to allocate a non-special area.
3516 ::munmap(start, end - start);
3517 return NULL;
3518 }
3519
3520 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3521
3522 void* result;
3523
3524 // Commit small-paged leading area.
3525 if (start != lp_start) {
3526 result = ::mmap(start, lp_start - start, prot,
3527 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3528 -1, 0);
3529 if (result == MAP_FAILED) {
3530 ::munmap(lp_start, end - lp_start);
3531 return NULL;
3532 }
3533 }
3534
3535 // Commit large-paged area.
3536 result = ::mmap(lp_start, lp_bytes, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3538 -1, 0);
3539 if (result == MAP_FAILED) {
3540 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3541 // If the mmap above fails, the large pages region will be unmapped and we
3542 // have regions before and after with small pages. Release these regions.
3543 //
3544 // | mapped | unmapped | mapped |
3545 // ^ ^ ^ ^
3546 // start lp_start lp_end end
3547 //
3548 ::munmap(start, lp_start - start);
3549 ::munmap(lp_end, end - lp_end);
3550 return NULL;
3551 }
3552
3553 // Commit small-paged trailing area.
3554 if (lp_end != end) {
3555 result = ::mmap(lp_end, end - lp_end, prot,
3556 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3557 -1, 0);
3558 if (result == MAP_FAILED) {
3559 ::munmap(start, lp_end - start);
3560 return NULL;
3561 }
3562 }
3563
3564 return start;
3565 }
3566
3567 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3568 size_t alignment,
3569 char* req_addr,
3570 bool exec) {
3571 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3572 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3573 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3574 assert(is_power_of_2(os::large_page_size()), "Must be");
3575 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3576
3577 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3578 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3579 } else {
3580 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3581 }
3582 }
3583
3584 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3585 char* req_addr, bool exec) {
3586 assert(UseLargePages, "only for large pages");
3587
3588 char* addr;
3589 if (UseSHM) {
3590 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3591 } else {
3592 assert(UseHugeTLBFS, "must be");
3593 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3594 }
3595
3596 if (addr != NULL) {
3597 if (UseNUMAInterleaving) {
3598 numa_make_global(addr, bytes);
3599 }
3600
3601 // The memory is committed
3602 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3603 }
3604
3605 return addr;
3606 }
3607
3608 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3609 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3610 return shmdt(base) == 0;
3611 }
3612
3613 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3614 return pd_release_memory(base, bytes);
3615 }
3616
3617 bool os::release_memory_special(char* base, size_t bytes) {
3618 bool res;
3619 if (MemTracker::tracking_level() > NMT_minimal) {
3620 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3621 res = os::Linux::release_memory_special_impl(base, bytes);
3622 if (res) {
3623 tkr.record((address)base, bytes);
3624 }
3625
3626 } else {
3627 res = os::Linux::release_memory_special_impl(base, bytes);
3628 }
3629 return res;
3630 }
3631
3632 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3633 assert(UseLargePages, "only for large pages");
3634 bool res;
3635
3636 if (UseSHM) {
3637 res = os::Linux::release_memory_special_shm(base, bytes);
3638 } else {
3639 assert(UseHugeTLBFS, "must be");
3640 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3641 }
3642 return res;
3643 }
3644
3645 size_t os::large_page_size() {
3646 return _large_page_size;
3647 }
3648
3649 // With SysV SHM the entire memory region must be allocated as shared
3650 // memory.
3651 // HugeTLBFS allows application to commit large page memory on demand.
3652 // However, when committing memory with HugeTLBFS fails, the region
3653 // that was supposed to be committed will lose the old reservation
3654 // and allow other threads to steal that memory region. Because of this
3655 // behavior we can't commit HugeTLBFS memory.
3656 bool os::can_commit_large_page_memory() {
3657 return UseTransparentHugePages;
3658 }
3659
3660 bool os::can_execute_large_page_memory() {
3661 return UseTransparentHugePages || UseHugeTLBFS;
3662 }
3663
3664 // Reserve memory at an arbitrary address, only if that area is
3665 // available (and not reserved for something else).
3666
3667 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3668 const int max_tries = 10;
3669 char* base[max_tries];
3670 size_t size[max_tries];
3671 const size_t gap = 0x000000;
3672
3673 // Assert only that the size is a multiple of the page size, since
3674 // that's all that mmap requires, and since that's all we really know
3675 // about at this low abstraction level. If we need higher alignment,
3676 // we can either pass an alignment to this method or verify alignment
3677 // in one of the methods further up the call chain. See bug 5044738.
3678 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3679
3680 // Repeatedly allocate blocks until the block is allocated at the
3681 // right spot.
3682
3683 // Linux mmap allows caller to pass an address as hint; give it a try first,
3684 // if kernel honors the hint then we can return immediately.
3685 char * addr = anon_mmap(requested_addr, bytes, false);
3686 if (addr == requested_addr) {
3687 return requested_addr;
3688 }
3689
3690 if (addr != NULL) {
3691 // mmap() is successful but it fails to reserve at the requested address
3692 anon_munmap(addr, bytes);
3693 }
3694
3695 int i;
3696 for (i = 0; i < max_tries; ++i) {
3697 base[i] = reserve_memory(bytes);
3698
3699 if (base[i] != NULL) {
3700 // Is this the block we wanted?
3701 if (base[i] == requested_addr) {
3702 size[i] = bytes;
3703 break;
3704 }
3705
3706 // Does this overlap the block we wanted? Give back the overlapped
3707 // parts and try again.
3708
3709 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3710 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3711 unmap_memory(base[i], top_overlap);
3712 base[i] += top_overlap;
3713 size[i] = bytes - top_overlap;
3714 } else {
3715 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3716 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3717 unmap_memory(requested_addr, bottom_overlap);
3718 size[i] = bytes - bottom_overlap;
3719 } else {
3720 size[i] = bytes;
3721 }
3722 }
3723 }
3724 }
3725
3726 // Give back the unused reserved pieces.
3727
3728 for (int j = 0; j < i; ++j) {
3729 if (base[j] != NULL) {
3730 unmap_memory(base[j], size[j]);
3731 }
3732 }
3733
3734 if (i < max_tries) {
3735 return requested_addr;
3736 } else {
3737 return NULL;
3738 }
3739 }
3740
3741 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3742 return ::read(fd, buf, nBytes);
3743 }
3744
3745 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3746 return ::pread(fd, buf, nBytes, offset);
3747 }
3748
3749 // Short sleep, direct OS call.
3750 //
3751 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3752 // sched_yield(2) will actually give up the CPU:
3753 //
3754 // * Alone on this pariticular CPU, keeps running.
3755 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3756 // (pre 2.6.39).
3757 //
3758 // So calling this with 0 is an alternative.
3759 //
3760 void os::naked_short_sleep(jlong ms) {
3761 struct timespec req;
3762
3763 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3764 req.tv_sec = 0;
3765 if (ms > 0) {
3766 req.tv_nsec = (ms % 1000) * 1000000;
3767 } else {
3768 req.tv_nsec = 1;
3769 }
3770
3771 nanosleep(&req, NULL);
3772
3773 return;
3774 }
3775
3776 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3777 void os::infinite_sleep() {
3778 while (true) { // sleep forever ...
3779 ::sleep(100); // ... 100 seconds at a time
3780 }
3781 }
3782
3783 // Used to convert frequent JVM_Yield() to nops
3784 bool os::dont_yield() {
3785 return DontYieldALot;
3786 }
3787
3788 void os::naked_yield() {
3789 sched_yield();
3790 }
3791
3792 ////////////////////////////////////////////////////////////////////////////////
3793 // thread priority support
3794
3795 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3796 // only supports dynamic priority, static priority must be zero. For real-time
3797 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3798 // However, for large multi-threaded applications, SCHED_RR is not only slower
3799 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3800 // of 5 runs - Sep 2005).
3801 //
3802 // The following code actually changes the niceness of kernel-thread/LWP. It
3803 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3804 // not the entire user process, and user level threads are 1:1 mapped to kernel
3805 // threads. It has always been the case, but could change in the future. For
3806 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3807 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3808
3809 int os::java_to_os_priority[CriticalPriority + 1] = {
3810 19, // 0 Entry should never be used
3811
3812 4, // 1 MinPriority
3813 3, // 2
3814 2, // 3
3815
3816 1, // 4
3817 0, // 5 NormPriority
3818 -1, // 6
3819
3820 -2, // 7
3821 -3, // 8
3822 -4, // 9 NearMaxPriority
3823
3824 -5, // 10 MaxPriority
3825
3826 -5 // 11 CriticalPriority
3827 };
3828
3829 static int prio_init() {
3830 if (ThreadPriorityPolicy == 1) {
3831 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3832 // if effective uid is not root. Perhaps, a more elegant way of doing
3833 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3834 if (geteuid() != 0) {
3835 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3836 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3837 }
3838 ThreadPriorityPolicy = 0;
3839 }
3840 }
3841 if (UseCriticalJavaThreadPriority) {
3842 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3843 }
3844 return 0;
3845 }
3846
3847 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3848 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3849
3850 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3851 return (ret == 0) ? OS_OK : OS_ERR;
3852 }
3853
3854 OSReturn os::get_native_priority(const Thread* const thread,
3855 int *priority_ptr) {
3856 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3857 *priority_ptr = java_to_os_priority[NormPriority];
3858 return OS_OK;
3859 }
3860
3861 errno = 0;
3862 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3863 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3864 }
3865
3866 // Hint to the underlying OS that a task switch would not be good.
3867 // Void return because it's a hint and can fail.
3868 void os::hint_no_preempt() {}
3869
3870 ////////////////////////////////////////////////////////////////////////////////
3871 // suspend/resume support
3872
3873 // the low-level signal-based suspend/resume support is a remnant from the
3874 // old VM-suspension that used to be for java-suspension, safepoints etc,
3875 // within hotspot. Now there is a single use-case for this:
3876 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3877 // that runs in the watcher thread.
3878 // The remaining code is greatly simplified from the more general suspension
3879 // code that used to be used.
3880 //
3881 // The protocol is quite simple:
3882 // - suspend:
3883 // - sends a signal to the target thread
3884 // - polls the suspend state of the osthread using a yield loop
3885 // - target thread signal handler (SR_handler) sets suspend state
3886 // and blocks in sigsuspend until continued
3887 // - resume:
3888 // - sets target osthread state to continue
3889 // - sends signal to end the sigsuspend loop in the SR_handler
3890 //
3891 // Note that the SR_lock plays no role in this suspend/resume protocol.
3892
3893 static void resume_clear_context(OSThread *osthread) {
3894 osthread->set_ucontext(NULL);
3895 osthread->set_siginfo(NULL);
3896 }
3897
3898 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3899 ucontext_t* context) {
3900 osthread->set_ucontext(context);
3901 osthread->set_siginfo(siginfo);
3902 }
3903
3904 // Handler function invoked when a thread's execution is suspended or
3905 // resumed. We have to be careful that only async-safe functions are
3906 // called here (Note: most pthread functions are not async safe and
3907 // should be avoided.)
3908 //
3909 // Note: sigwait() is a more natural fit than sigsuspend() from an
3910 // interface point of view, but sigwait() prevents the signal hander
3911 // from being run. libpthread would get very confused by not having
3912 // its signal handlers run and prevents sigwait()'s use with the
3913 // mutex granting granting signal.
3914 //
3915 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3916 //
3917 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3918 // Save and restore errno to avoid confusing native code with EINTR
3919 // after sigsuspend.
3920 int old_errno = errno;
3921
3922 Thread* thread = Thread::current();
3923 OSThread* osthread = thread->osthread();
3924 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3925
3926 os::SuspendResume::State current = osthread->sr.state();
3927 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3928 suspend_save_context(osthread, siginfo, context);
3929
3930 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3931 os::SuspendResume::State state = osthread->sr.suspended();
3932 if (state == os::SuspendResume::SR_SUSPENDED) {
3933 sigset_t suspend_set; // signals for sigsuspend()
3934
3935 // get current set of blocked signals and unblock resume signal
3936 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3937 sigdelset(&suspend_set, SR_signum);
3938
3939 sr_semaphore.signal();
3940 // wait here until we are resumed
3941 while (1) {
3942 sigsuspend(&suspend_set);
3943
3944 os::SuspendResume::State result = osthread->sr.running();
3945 if (result == os::SuspendResume::SR_RUNNING) {
3946 sr_semaphore.signal();
3947 break;
3948 }
3949 }
3950
3951 } else if (state == os::SuspendResume::SR_RUNNING) {
3952 // request was cancelled, continue
3953 } else {
3954 ShouldNotReachHere();
3955 }
3956
3957 resume_clear_context(osthread);
3958 } else if (current == os::SuspendResume::SR_RUNNING) {
3959 // request was cancelled, continue
3960 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3961 // ignore
3962 } else {
3963 // ignore
3964 }
3965
3966 errno = old_errno;
3967 }
3968
3969
3970 static int SR_initialize() {
3971 struct sigaction act;
3972 char *s;
3973 // Get signal number to use for suspend/resume
3974 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3975 int sig = ::strtol(s, 0, 10);
3976 if (sig > 0 || sig < _NSIG) {
3977 SR_signum = sig;
3978 }
3979 }
3980
3981 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3982 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3983
3984 sigemptyset(&SR_sigset);
3985 sigaddset(&SR_sigset, SR_signum);
3986
3987 // Set up signal handler for suspend/resume
3988 act.sa_flags = SA_RESTART|SA_SIGINFO;
3989 act.sa_handler = (void (*)(int)) SR_handler;
3990
3991 // SR_signum is blocked by default.
3992 // 4528190 - We also need to block pthread restart signal (32 on all
3993 // supported Linux platforms). Note that LinuxThreads need to block
3994 // this signal for all threads to work properly. So we don't have
3995 // to use hard-coded signal number when setting up the mask.
3996 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3997
3998 if (sigaction(SR_signum, &act, 0) == -1) {
3999 return -1;
4000 }
4001
4002 // Save signal flag
4003 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4004 return 0;
4005 }
4006
4007 static int sr_notify(OSThread* osthread) {
4008 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4009 assert_status(status == 0, status, "pthread_kill");
4010 return status;
4011 }
4012
4013 // "Randomly" selected value for how long we want to spin
4014 // before bailing out on suspending a thread, also how often
4015 // we send a signal to a thread we want to resume
4016 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4017 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4018
4019 // returns true on success and false on error - really an error is fatal
4020 // but this seems the normal response to library errors
4021 static bool do_suspend(OSThread* osthread) {
4022 assert(osthread->sr.is_running(), "thread should be running");
4023 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4024
4025 // mark as suspended and send signal
4026 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4027 // failed to switch, state wasn't running?
4028 ShouldNotReachHere();
4029 return false;
4030 }
4031
4032 if (sr_notify(osthread) != 0) {
4033 ShouldNotReachHere();
4034 }
4035
4036 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4037 while (true) {
4038 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4039 break;
4040 } else {
4041 // timeout
4042 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4043 if (cancelled == os::SuspendResume::SR_RUNNING) {
4044 return false;
4045 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4046 // make sure that we consume the signal on the semaphore as well
4047 sr_semaphore.wait();
4048 break;
4049 } else {
4050 ShouldNotReachHere();
4051 return false;
4052 }
4053 }
4054 }
4055
4056 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4057 return true;
4058 }
4059
4060 static void do_resume(OSThread* osthread) {
4061 assert(osthread->sr.is_suspended(), "thread should be suspended");
4062 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4063
4064 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4065 // failed to switch to WAKEUP_REQUEST
4066 ShouldNotReachHere();
4067 return;
4068 }
4069
4070 while (true) {
4071 if (sr_notify(osthread) == 0) {
4072 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4073 if (osthread->sr.is_running()) {
4074 return;
4075 }
4076 }
4077 } else {
4078 ShouldNotReachHere();
4079 }
4080 }
4081
4082 guarantee(osthread->sr.is_running(), "Must be running!");
4083 }
4084
4085 ///////////////////////////////////////////////////////////////////////////////////
4086 // signal handling (except suspend/resume)
4087
4088 // This routine may be used by user applications as a "hook" to catch signals.
4089 // The user-defined signal handler must pass unrecognized signals to this
4090 // routine, and if it returns true (non-zero), then the signal handler must
4091 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4092 // routine will never retun false (zero), but instead will execute a VM panic
4093 // routine kill the process.
4094 //
4095 // If this routine returns false, it is OK to call it again. This allows
4096 // the user-defined signal handler to perform checks either before or after
4097 // the VM performs its own checks. Naturally, the user code would be making
4098 // a serious error if it tried to handle an exception (such as a null check
4099 // or breakpoint) that the VM was generating for its own correct operation.
4100 //
4101 // This routine may recognize any of the following kinds of signals:
4102 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4103 // It should be consulted by handlers for any of those signals.
4104 //
4105 // The caller of this routine must pass in the three arguments supplied
4106 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4107 // field of the structure passed to sigaction(). This routine assumes that
4108 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4109 //
4110 // Note that the VM will print warnings if it detects conflicting signal
4111 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4112 //
4113 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4114 siginfo_t* siginfo,
4115 void* ucontext,
4116 int abort_if_unrecognized);
4117
4118 void signalHandler(int sig, siginfo_t* info, void* uc) {
4119 assert(info != NULL && uc != NULL, "it must be old kernel");
4120 int orig_errno = errno; // Preserve errno value over signal handler.
4121 JVM_handle_linux_signal(sig, info, uc, true);
4122 errno = orig_errno;
4123 }
4124
4125
4126 // This boolean allows users to forward their own non-matching signals
4127 // to JVM_handle_linux_signal, harmlessly.
4128 bool os::Linux::signal_handlers_are_installed = false;
4129
4130 // For signal-chaining
4131 struct sigaction os::Linux::sigact[MAXSIGNUM];
4132 unsigned int os::Linux::sigs = 0;
4133 bool os::Linux::libjsig_is_loaded = false;
4134 typedef struct sigaction *(*get_signal_t)(int);
4135 get_signal_t os::Linux::get_signal_action = NULL;
4136
4137 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4138 struct sigaction *actp = NULL;
4139
4140 if (libjsig_is_loaded) {
4141 // Retrieve the old signal handler from libjsig
4142 actp = (*get_signal_action)(sig);
4143 }
4144 if (actp == NULL) {
4145 // Retrieve the preinstalled signal handler from jvm
4146 actp = get_preinstalled_handler(sig);
4147 }
4148
4149 return actp;
4150 }
4151
4152 static bool call_chained_handler(struct sigaction *actp, int sig,
4153 siginfo_t *siginfo, void *context) {
4154 // Call the old signal handler
4155 if (actp->sa_handler == SIG_DFL) {
4156 // It's more reasonable to let jvm treat it as an unexpected exception
4157 // instead of taking the default action.
4158 return false;
4159 } else if (actp->sa_handler != SIG_IGN) {
4160 if ((actp->sa_flags & SA_NODEFER) == 0) {
4161 // automaticlly block the signal
4162 sigaddset(&(actp->sa_mask), sig);
4163 }
4164
4165 sa_handler_t hand;
4166 sa_sigaction_t sa;
4167 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4168 // retrieve the chained handler
4169 if (siginfo_flag_set) {
4170 sa = actp->sa_sigaction;
4171 } else {
4172 hand = actp->sa_handler;
4173 }
4174
4175 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4176 actp->sa_handler = SIG_DFL;
4177 }
4178
4179 // try to honor the signal mask
4180 sigset_t oset;
4181 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4182
4183 // call into the chained handler
4184 if (siginfo_flag_set) {
4185 (*sa)(sig, siginfo, context);
4186 } else {
4187 (*hand)(sig);
4188 }
4189
4190 // restore the signal mask
4191 pthread_sigmask(SIG_SETMASK, &oset, 0);
4192 }
4193 // Tell jvm's signal handler the signal is taken care of.
4194 return true;
4195 }
4196
4197 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4198 bool chained = false;
4199 // signal-chaining
4200 if (UseSignalChaining) {
4201 struct sigaction *actp = get_chained_signal_action(sig);
4202 if (actp != NULL) {
4203 chained = call_chained_handler(actp, sig, siginfo, context);
4204 }
4205 }
4206 return chained;
4207 }
4208
4209 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4210 if ((((unsigned int)1 << sig) & sigs) != 0) {
4211 return &sigact[sig];
4212 }
4213 return NULL;
4214 }
4215
4216 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4217 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4218 sigact[sig] = oldAct;
4219 sigs |= (unsigned int)1 << sig;
4220 }
4221
4222 // for diagnostic
4223 int os::Linux::sigflags[MAXSIGNUM];
4224
4225 int os::Linux::get_our_sigflags(int sig) {
4226 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4227 return sigflags[sig];
4228 }
4229
4230 void os::Linux::set_our_sigflags(int sig, int flags) {
4231 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4232 sigflags[sig] = flags;
4233 }
4234
4235 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4236 // Check for overwrite.
4237 struct sigaction oldAct;
4238 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4239
4240 void* oldhand = oldAct.sa_sigaction
4241 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4242 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4243 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4244 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4245 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4246 if (AllowUserSignalHandlers || !set_installed) {
4247 // Do not overwrite; user takes responsibility to forward to us.
4248 return;
4249 } else if (UseSignalChaining) {
4250 // save the old handler in jvm
4251 save_preinstalled_handler(sig, oldAct);
4252 // libjsig also interposes the sigaction() call below and saves the
4253 // old sigaction on it own.
4254 } else {
4255 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4256 "%#lx for signal %d.", (long)oldhand, sig));
4257 }
4258 }
4259
4260 struct sigaction sigAct;
4261 sigfillset(&(sigAct.sa_mask));
4262 sigAct.sa_handler = SIG_DFL;
4263 if (!set_installed) {
4264 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4265 } else {
4266 sigAct.sa_sigaction = signalHandler;
4267 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4268 }
4269 // Save flags, which are set by ours
4270 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4271 sigflags[sig] = sigAct.sa_flags;
4272
4273 int ret = sigaction(sig, &sigAct, &oldAct);
4274 assert(ret == 0, "check");
4275
4276 void* oldhand2 = oldAct.sa_sigaction
4277 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4278 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4279 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4280 }
4281
4282 // install signal handlers for signals that HotSpot needs to
4283 // handle in order to support Java-level exception handling.
4284
4285 void os::Linux::install_signal_handlers() {
4286 if (!signal_handlers_are_installed) {
4287 signal_handlers_are_installed = true;
4288
4289 // signal-chaining
4290 typedef void (*signal_setting_t)();
4291 signal_setting_t begin_signal_setting = NULL;
4292 signal_setting_t end_signal_setting = NULL;
4293 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4294 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4295 if (begin_signal_setting != NULL) {
4296 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4297 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4298 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4299 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4300 libjsig_is_loaded = true;
4301 assert(UseSignalChaining, "should enable signal-chaining");
4302 }
4303 if (libjsig_is_loaded) {
4304 // Tell libjsig jvm is setting signal handlers
4305 (*begin_signal_setting)();
4306 }
4307
4308 set_signal_handler(SIGSEGV, true);
4309 set_signal_handler(SIGPIPE, true);
4310 set_signal_handler(SIGBUS, true);
4311 set_signal_handler(SIGILL, true);
4312 set_signal_handler(SIGFPE, true);
4313 #if defined(PPC64)
4314 set_signal_handler(SIGTRAP, true);
4315 #endif
4316 set_signal_handler(SIGXFSZ, true);
4317
4318 if (libjsig_is_loaded) {
4319 // Tell libjsig jvm finishes setting signal handlers
4320 (*end_signal_setting)();
4321 }
4322
4323 // We don't activate signal checker if libjsig is in place, we trust ourselves
4324 // and if UserSignalHandler is installed all bets are off.
4325 // Log that signal checking is off only if -verbose:jni is specified.
4326 if (CheckJNICalls) {
4327 if (libjsig_is_loaded) {
4328 if (PrintJNIResolving) {
4329 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4330 }
4331 check_signals = false;
4332 }
4333 if (AllowUserSignalHandlers) {
4334 if (PrintJNIResolving) {
4335 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4336 }
4337 check_signals = false;
4338 }
4339 }
4340 }
4341 }
4342
4343 // This is the fastest way to get thread cpu time on Linux.
4344 // Returns cpu time (user+sys) for any thread, not only for current.
4345 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4346 // It might work on 2.6.10+ with a special kernel/glibc patch.
4347 // For reference, please, see IEEE Std 1003.1-2004:
4348 // http://www.unix.org/single_unix_specification
4349
4350 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4351 struct timespec tp;
4352 int rc = os::Linux::clock_gettime(clockid, &tp);
4353 assert(rc == 0, "clock_gettime is expected to return 0 code");
4354
4355 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4356 }
4357
4358 /////
4359 // glibc on Linux platform uses non-documented flag
4360 // to indicate, that some special sort of signal
4361 // trampoline is used.
4362 // We will never set this flag, and we should
4363 // ignore this flag in our diagnostic
4364 #ifdef SIGNIFICANT_SIGNAL_MASK
4365 #undef SIGNIFICANT_SIGNAL_MASK
4366 #endif
4367 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4368
4369 static const char* get_signal_handler_name(address handler,
4370 char* buf, int buflen) {
4371 int offset;
4372 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4373 if (found) {
4374 // skip directory names
4375 const char *p1, *p2;
4376 p1 = buf;
4377 size_t len = strlen(os::file_separator());
4378 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4379 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4380 } else {
4381 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4382 }
4383 return buf;
4384 }
4385
4386 static void print_signal_handler(outputStream* st, int sig,
4387 char* buf, size_t buflen) {
4388 struct sigaction sa;
4389
4390 sigaction(sig, NULL, &sa);
4391
4392 // See comment for SIGNIFICANT_SIGNAL_MASK define
4393 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4394
4395 st->print("%s: ", os::exception_name(sig, buf, buflen));
4396
4397 address handler = (sa.sa_flags & SA_SIGINFO)
4398 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4399 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4400
4401 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4402 st->print("SIG_DFL");
4403 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4404 st->print("SIG_IGN");
4405 } else {
4406 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4407 }
4408
4409 st->print(", sa_mask[0]=");
4410 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4411
4412 address rh = VMError::get_resetted_sighandler(sig);
4413 // May be, handler was resetted by VMError?
4414 if (rh != NULL) {
4415 handler = rh;
4416 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4417 }
4418
4419 st->print(", sa_flags=");
4420 os::Posix::print_sa_flags(st, sa.sa_flags);
4421
4422 // Check: is it our handler?
4423 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4424 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4425 // It is our signal handler
4426 // check for flags, reset system-used one!
4427 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4428 st->print(
4429 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4430 os::Linux::get_our_sigflags(sig));
4431 }
4432 }
4433 st->cr();
4434 }
4435
4436
4437 #define DO_SIGNAL_CHECK(sig) \
4438 do { \
4439 if (!sigismember(&check_signal_done, sig)) { \
4440 os::Linux::check_signal_handler(sig); \
4441 } \
4442 } while (0)
4443
4444 // This method is a periodic task to check for misbehaving JNI applications
4445 // under CheckJNI, we can add any periodic checks here
4446
4447 void os::run_periodic_checks() {
4448 if (check_signals == false) return;
4449
4450 // SEGV and BUS if overridden could potentially prevent
4451 // generation of hs*.log in the event of a crash, debugging
4452 // such a case can be very challenging, so we absolutely
4453 // check the following for a good measure:
4454 DO_SIGNAL_CHECK(SIGSEGV);
4455 DO_SIGNAL_CHECK(SIGILL);
4456 DO_SIGNAL_CHECK(SIGFPE);
4457 DO_SIGNAL_CHECK(SIGBUS);
4458 DO_SIGNAL_CHECK(SIGPIPE);
4459 DO_SIGNAL_CHECK(SIGXFSZ);
4460 #if defined(PPC64)
4461 DO_SIGNAL_CHECK(SIGTRAP);
4462 #endif
4463
4464 // ReduceSignalUsage allows the user to override these handlers
4465 // see comments at the very top and jvm_solaris.h
4466 if (!ReduceSignalUsage) {
4467 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4468 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4469 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4470 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4471 }
4472
4473 DO_SIGNAL_CHECK(SR_signum);
4474 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4475 }
4476
4477 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4478
4479 static os_sigaction_t os_sigaction = NULL;
4480
4481 void os::Linux::check_signal_handler(int sig) {
4482 char buf[O_BUFLEN];
4483 address jvmHandler = NULL;
4484
4485
4486 struct sigaction act;
4487 if (os_sigaction == NULL) {
4488 // only trust the default sigaction, in case it has been interposed
4489 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4490 if (os_sigaction == NULL) return;
4491 }
4492
4493 os_sigaction(sig, (struct sigaction*)NULL, &act);
4494
4495
4496 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4497
4498 address thisHandler = (act.sa_flags & SA_SIGINFO)
4499 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4500 : CAST_FROM_FN_PTR(address, act.sa_handler);
4501
4502
4503 switch (sig) {
4504 case SIGSEGV:
4505 case SIGBUS:
4506 case SIGFPE:
4507 case SIGPIPE:
4508 case SIGILL:
4509 case SIGXFSZ:
4510 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4511 break;
4512
4513 case SHUTDOWN1_SIGNAL:
4514 case SHUTDOWN2_SIGNAL:
4515 case SHUTDOWN3_SIGNAL:
4516 case BREAK_SIGNAL:
4517 jvmHandler = (address)user_handler();
4518 break;
4519
4520 case INTERRUPT_SIGNAL:
4521 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4522 break;
4523
4524 default:
4525 if (sig == SR_signum) {
4526 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4527 } else {
4528 return;
4529 }
4530 break;
4531 }
4532
4533 if (thisHandler != jvmHandler) {
4534 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4535 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4536 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4537 // No need to check this sig any longer
4538 sigaddset(&check_signal_done, sig);
4539 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4540 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4541 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4542 exception_name(sig, buf, O_BUFLEN));
4543 }
4544 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4545 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4546 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4547 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4548 // No need to check this sig any longer
4549 sigaddset(&check_signal_done, sig);
4550 }
4551
4552 // Dump all the signal
4553 if (sigismember(&check_signal_done, sig)) {
4554 print_signal_handlers(tty, buf, O_BUFLEN);
4555 }
4556 }
4557
4558 extern void report_error(char* file_name, int line_no, char* title,
4559 char* format, ...);
4560
4561 extern bool signal_name(int signo, char* buf, size_t len);
4562
4563 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4564 if (0 < exception_code && exception_code <= SIGRTMAX) {
4565 // signal
4566 if (!signal_name(exception_code, buf, size)) {
4567 jio_snprintf(buf, size, "SIG%d", exception_code);
4568 }
4569 return buf;
4570 } else {
4571 return NULL;
4572 }
4573 }
4574
4575 // this is called _before_ the most of global arguments have been parsed
4576 void os::init(void) {
4577 char dummy; // used to get a guess on initial stack address
4578 // first_hrtime = gethrtime();
4579
4580 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4581
4582 init_random(1234567);
4583
4584 ThreadCritical::initialize();
4585
4586 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4587 if (Linux::page_size() == -1) {
4588 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4589 strerror(errno)));
4590 }
4591 init_page_sizes((size_t) Linux::page_size());
4592
4593 Linux::initialize_system_info();
4594
4595 // main_thread points to the aboriginal thread
4596 Linux::_main_thread = pthread_self();
4597
4598 Linux::clock_init();
4599 initial_time_count = javaTimeNanos();
4600
4601 // pthread_condattr initialization for monotonic clock
4602 int status;
4603 pthread_condattr_t* _condattr = os::Linux::condAttr();
4604 if ((status = pthread_condattr_init(_condattr)) != 0) {
4605 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4606 }
4607 // Only set the clock if CLOCK_MONOTONIC is available
4608 if (os::supports_monotonic_clock()) {
4609 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4610 if (status == EINVAL) {
4611 warning("Unable to use monotonic clock with relative timed-waits" \
4612 " - changes to the time-of-day clock may have adverse affects");
4613 } else {
4614 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4615 }
4616 }
4617 }
4618 // else it defaults to CLOCK_REALTIME
4619
4620 // If the pagesize of the VM is greater than 8K determine the appropriate
4621 // number of initial guard pages. The user can change this with the
4622 // command line arguments, if needed.
4623 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4624 StackYellowPages = 1;
4625 StackRedPages = 1;
4626 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4627 }
4628
4629 // retrieve entry point for pthread_setname_np
4630 Linux::_pthread_setname_np =
4631 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4632
4633 }
4634
4635 // To install functions for atexit system call
4636 extern "C" {
4637 static void perfMemory_exit_helper() {
4638 perfMemory_exit();
4639 }
4640 }
4641
4642 // this is called _after_ the global arguments have been parsed
4643 jint os::init_2(void) {
4644 Linux::fast_thread_clock_init();
4645
4646 // Allocate a single page and mark it as readable for safepoint polling
4647 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4648 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4649
4650 os::set_polling_page(polling_page);
4651
4652 #ifndef PRODUCT
4653 if (Verbose && PrintMiscellaneous) {
4654 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4655 (intptr_t)polling_page);
4656 }
4657 #endif
4658
4659 if (!UseMembar) {
4660 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4661 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4662 os::set_memory_serialize_page(mem_serialize_page);
4663
4664 #ifndef PRODUCT
4665 if (Verbose && PrintMiscellaneous) {
4666 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4667 (intptr_t)mem_serialize_page);
4668 }
4669 #endif
4670 }
4671
4672 // initialize suspend/resume support - must do this before signal_sets_init()
4673 if (SR_initialize() != 0) {
4674 perror("SR_initialize failed");
4675 return JNI_ERR;
4676 }
4677
4678 Linux::signal_sets_init();
4679 Linux::install_signal_handlers();
4680
4681 // Check minimum allowable stack size for thread creation and to initialize
4682 // the java system classes, including StackOverflowError - depends on page
4683 // size. Add a page for compiler2 recursion in main thread.
4684 // Add in 2*BytesPerWord times page size to account for VM stack during
4685 // class initialization depending on 32 or 64 bit VM.
4686 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4687 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4688 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4689
4690 size_t threadStackSizeInBytes = ThreadStackSize * K;
4691 if (threadStackSizeInBytes != 0 &&
4692 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4693 tty->print_cr("\nThe stack size specified is too small, "
4694 "Specify at least %dk",
4695 os::Linux::min_stack_allowed/ K);
4696 return JNI_ERR;
4697 }
4698
4699 // Make the stack size a multiple of the page size so that
4700 // the yellow/red zones can be guarded.
4701 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4702 vm_page_size()));
4703
4704 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4705
4706 #if defined(IA32)
4707 workaround_expand_exec_shield_cs_limit();
4708 #endif
4709
4710 Linux::libpthread_init();
4711 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4712 tty->print_cr("[HotSpot is running with %s, %s]\n",
4713 Linux::glibc_version(), Linux::libpthread_version());
4714 }
4715
4716 if (UseNUMA) {
4717 if (!Linux::libnuma_init()) {
4718 UseNUMA = false;
4719 } else {
4720 if ((Linux::numa_max_node() < 1)) {
4721 // There's only one node(they start from 0), disable NUMA.
4722 UseNUMA = false;
4723 }
4724 }
4725 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4726 // we can make the adaptive lgrp chunk resizing work. If the user specified
4727 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4728 // disable adaptive resizing.
4729 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4730 if (FLAG_IS_DEFAULT(UseNUMA)) {
4731 UseNUMA = false;
4732 } else {
4733 if (FLAG_IS_DEFAULT(UseLargePages) &&
4734 FLAG_IS_DEFAULT(UseSHM) &&
4735 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4736 UseLargePages = false;
4737 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4738 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4739 UseAdaptiveSizePolicy = false;
4740 UseAdaptiveNUMAChunkSizing = false;
4741 }
4742 }
4743 }
4744 if (!UseNUMA && ForceNUMA) {
4745 UseNUMA = true;
4746 }
4747 }
4748
4749 if (MaxFDLimit) {
4750 // set the number of file descriptors to max. print out error
4751 // if getrlimit/setrlimit fails but continue regardless.
4752 struct rlimit nbr_files;
4753 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4754 if (status != 0) {
4755 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4756 perror("os::init_2 getrlimit failed");
4757 }
4758 } else {
4759 nbr_files.rlim_cur = nbr_files.rlim_max;
4760 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4761 if (status != 0) {
4762 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4763 perror("os::init_2 setrlimit failed");
4764 }
4765 }
4766 }
4767 }
4768
4769 // Initialize lock used to serialize thread creation (see os::create_thread)
4770 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4771
4772 // at-exit methods are called in the reverse order of their registration.
4773 // atexit functions are called on return from main or as a result of a
4774 // call to exit(3C). There can be only 32 of these functions registered
4775 // and atexit() does not set errno.
4776
4777 if (PerfAllowAtExitRegistration) {
4778 // only register atexit functions if PerfAllowAtExitRegistration is set.
4779 // atexit functions can be delayed until process exit time, which
4780 // can be problematic for embedded VM situations. Embedded VMs should
4781 // call DestroyJavaVM() to assure that VM resources are released.
4782
4783 // note: perfMemory_exit_helper atexit function may be removed in
4784 // the future if the appropriate cleanup code can be added to the
4785 // VM_Exit VMOperation's doit method.
4786 if (atexit(perfMemory_exit_helper) != 0) {
4787 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4788 }
4789 }
4790
4791 // initialize thread priority policy
4792 prio_init();
4793
4794 return JNI_OK;
4795 }
4796
4797 // Mark the polling page as unreadable
4798 void os::make_polling_page_unreadable(void) {
4799 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4800 fatal("Could not disable polling page");
4801 }
4802 }
4803
4804 // Mark the polling page as readable
4805 void os::make_polling_page_readable(void) {
4806 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4807 fatal("Could not enable polling page");
4808 }
4809 }
4810
4811 int os::active_processor_count() {
4812 // Linux doesn't yet have a (official) notion of processor sets,
4813 // so just return the number of online processors.
4814 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4815 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4816 return online_cpus;
4817 }
4818
4819 void os::set_native_thread_name(const char *name) {
4820 if (Linux::_pthread_setname_np) {
4821 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4822 snprintf(buf, sizeof(buf), "%s", name);
4823 buf[sizeof(buf) - 1] = '\0';
4824 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4825 // ERANGE should not happen; all other errors should just be ignored.
4826 assert(rc != ERANGE, "pthread_setname_np failed");
4827 }
4828 }
4829
4830 bool os::distribute_processes(uint length, uint* distribution) {
4831 // Not yet implemented.
4832 return false;
4833 }
4834
4835 bool os::bind_to_processor(uint processor_id) {
4836 // Not yet implemented.
4837 return false;
4838 }
4839
4840 ///
4841
4842 void os::SuspendedThreadTask::internal_do_task() {
4843 if (do_suspend(_thread->osthread())) {
4844 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4845 do_task(context);
4846 do_resume(_thread->osthread());
4847 }
4848 }
4849
4850 class PcFetcher : public os::SuspendedThreadTask {
4851 public:
4852 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4853 ExtendedPC result();
4854 protected:
4855 void do_task(const os::SuspendedThreadTaskContext& context);
4856 private:
4857 ExtendedPC _epc;
4858 };
4859
4860 ExtendedPC PcFetcher::result() {
4861 guarantee(is_done(), "task is not done yet.");
4862 return _epc;
4863 }
4864
4865 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4866 Thread* thread = context.thread();
4867 OSThread* osthread = thread->osthread();
4868 if (osthread->ucontext() != NULL) {
4869 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4870 } else {
4871 // NULL context is unexpected, double-check this is the VMThread
4872 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4873 }
4874 }
4875
4876 // Suspends the target using the signal mechanism and then grabs the PC before
4877 // resuming the target. Used by the flat-profiler only
4878 ExtendedPC os::get_thread_pc(Thread* thread) {
4879 // Make sure that it is called by the watcher for the VMThread
4880 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4881 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4882
4883 PcFetcher fetcher(thread);
4884 fetcher.run();
4885 return fetcher.result();
4886 }
4887
4888 ////////////////////////////////////////////////////////////////////////////////
4889 // debug support
4890
4891 bool os::find(address addr, outputStream* st) {
4892 Dl_info dlinfo;
4893 memset(&dlinfo, 0, sizeof(dlinfo));
4894 if (dladdr(addr, &dlinfo) != 0) {
4895 st->print(PTR_FORMAT ": ", addr);
4896 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4897 st->print("%s+%#x", dlinfo.dli_sname,
4898 addr - (intptr_t)dlinfo.dli_saddr);
4899 } else if (dlinfo.dli_fbase != NULL) {
4900 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4901 } else {
4902 st->print("<absolute address>");
4903 }
4904 if (dlinfo.dli_fname != NULL) {
4905 st->print(" in %s", dlinfo.dli_fname);
4906 }
4907 if (dlinfo.dli_fbase != NULL) {
4908 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4909 }
4910 st->cr();
4911
4912 if (Verbose) {
4913 // decode some bytes around the PC
4914 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4915 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4916 address lowest = (address) dlinfo.dli_sname;
4917 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4918 if (begin < lowest) begin = lowest;
4919 Dl_info dlinfo2;
4920 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4921 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4922 end = (address) dlinfo2.dli_saddr;
4923 }
4924 Disassembler::decode(begin, end, st);
4925 }
4926 return true;
4927 }
4928 return false;
4929 }
4930
4931 ////////////////////////////////////////////////////////////////////////////////
4932 // misc
4933
4934 // This does not do anything on Linux. This is basically a hook for being
4935 // able to use structured exception handling (thread-local exception filters)
4936 // on, e.g., Win32.
4937 void
4938 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4939 JavaCallArguments* args, Thread* thread) {
4940 f(value, method, args, thread);
4941 }
4942
4943 void os::print_statistics() {
4944 }
4945
4946 int os::message_box(const char* title, const char* message) {
4947 int i;
4948 fdStream err(defaultStream::error_fd());
4949 for (i = 0; i < 78; i++) err.print_raw("=");
4950 err.cr();
4951 err.print_raw_cr(title);
4952 for (i = 0; i < 78; i++) err.print_raw("-");
4953 err.cr();
4954 err.print_raw_cr(message);
4955 for (i = 0; i < 78; i++) err.print_raw("=");
4956 err.cr();
4957
4958 char buf[16];
4959 // Prevent process from exiting upon "read error" without consuming all CPU
4960 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4961
4962 return buf[0] == 'y' || buf[0] == 'Y';
4963 }
4964
4965 int os::stat(const char *path, struct stat *sbuf) {
4966 char pathbuf[MAX_PATH];
4967 if (strlen(path) > MAX_PATH - 1) {
4968 errno = ENAMETOOLONG;
4969 return -1;
4970 }
4971 os::native_path(strcpy(pathbuf, path));
4972 return ::stat(pathbuf, sbuf);
4973 }
4974
4975 bool os::check_heap(bool force) {
4976 return true;
4977 }
4978
4979 // Is a (classpath) directory empty?
4980 bool os::dir_is_empty(const char* path) {
4981 DIR *dir = NULL;
4982 struct dirent *ptr;
4983
4984 dir = opendir(path);
4985 if (dir == NULL) return true;
4986
4987 // Scan the directory
4988 bool result = true;
4989 char buf[sizeof(struct dirent) + MAX_PATH];
4990 while (result && (ptr = ::readdir(dir)) != NULL) {
4991 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4992 result = false;
4993 }
4994 }
4995 closedir(dir);
4996 return result;
4997 }
4998
4999 // This code originates from JDK's sysOpen and open64_w
5000 // from src/solaris/hpi/src/system_md.c
5001
5002 int os::open(const char *path, int oflag, int mode) {
5003 if (strlen(path) > MAX_PATH - 1) {
5004 errno = ENAMETOOLONG;
5005 return -1;
5006 }
5007
5008 // All file descriptors that are opened in the Java process and not
5009 // specifically destined for a subprocess should have the close-on-exec
5010 // flag set. If we don't set it, then careless 3rd party native code
5011 // might fork and exec without closing all appropriate file descriptors
5012 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5013 // turn might:
5014 //
5015 // - cause end-of-file to fail to be detected on some file
5016 // descriptors, resulting in mysterious hangs, or
5017 //
5018 // - might cause an fopen in the subprocess to fail on a system
5019 // suffering from bug 1085341.
5020 //
5021 // (Yes, the default setting of the close-on-exec flag is a Unix
5022 // design flaw)
5023 //
5024 // See:
5025 // 1085341: 32-bit stdio routines should support file descriptors >255
5026 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5027 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5028 //
5029 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5030 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5031 // because it saves a system call and removes a small window where the flag
5032 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5033 // and we fall back to using FD_CLOEXEC (see below).
5034 #ifdef O_CLOEXEC
5035 oflag |= O_CLOEXEC;
5036 #endif
5037
5038 int fd = ::open64(path, oflag, mode);
5039 if (fd == -1) return -1;
5040
5041 //If the open succeeded, the file might still be a directory
5042 {
5043 struct stat64 buf64;
5044 int ret = ::fstat64(fd, &buf64);
5045 int st_mode = buf64.st_mode;
5046
5047 if (ret != -1) {
5048 if ((st_mode & S_IFMT) == S_IFDIR) {
5049 errno = EISDIR;
5050 ::close(fd);
5051 return -1;
5052 }
5053 } else {
5054 ::close(fd);
5055 return -1;
5056 }
5057 }
5058
5059 #ifdef FD_CLOEXEC
5060 // Validate that the use of the O_CLOEXEC flag on open above worked.
5061 // With recent kernels, we will perform this check exactly once.
5062 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5063 if (!O_CLOEXEC_is_known_to_work) {
5064 int flags = ::fcntl(fd, F_GETFD);
5065 if (flags != -1) {
5066 if ((flags & FD_CLOEXEC) != 0)
5067 O_CLOEXEC_is_known_to_work = 1;
5068 else
5069 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5070 }
5071 }
5072 #endif
5073
5074 return fd;
5075 }
5076
5077
5078 // create binary file, rewriting existing file if required
5079 int os::create_binary_file(const char* path, bool rewrite_existing) {
5080 int oflags = O_WRONLY | O_CREAT;
5081 if (!rewrite_existing) {
5082 oflags |= O_EXCL;
5083 }
5084 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5085 }
5086
5087 // return current position of file pointer
5088 jlong os::current_file_offset(int fd) {
5089 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5090 }
5091
5092 // move file pointer to the specified offset
5093 jlong os::seek_to_file_offset(int fd, jlong offset) {
5094 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5095 }
5096
5097 // This code originates from JDK's sysAvailable
5098 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5099
5100 int os::available(int fd, jlong *bytes) {
5101 jlong cur, end;
5102 int mode;
5103 struct stat64 buf64;
5104
5105 if (::fstat64(fd, &buf64) >= 0) {
5106 mode = buf64.st_mode;
5107 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5108 // XXX: is the following call interruptible? If so, this might
5109 // need to go through the INTERRUPT_IO() wrapper as for other
5110 // blocking, interruptible calls in this file.
5111 int n;
5112 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5113 *bytes = n;
5114 return 1;
5115 }
5116 }
5117 }
5118 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5119 return 0;
5120 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5121 return 0;
5122 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5123 return 0;
5124 }
5125 *bytes = end - cur;
5126 return 1;
5127 }
5128
5129 // Map a block of memory.
5130 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5131 char *addr, size_t bytes, bool read_only,
5132 bool allow_exec) {
5133 int prot;
5134 int flags = MAP_PRIVATE;
5135
5136 if (read_only) {
5137 prot = PROT_READ;
5138 } else {
5139 prot = PROT_READ | PROT_WRITE;
5140 }
5141
5142 if (allow_exec) {
5143 prot |= PROT_EXEC;
5144 }
5145
5146 if (addr != NULL) {
5147 flags |= MAP_FIXED;
5148 }
5149
5150 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5151 fd, file_offset);
5152 if (mapped_address == MAP_FAILED) {
5153 return NULL;
5154 }
5155 return mapped_address;
5156 }
5157
5158
5159 // Remap a block of memory.
5160 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5161 char *addr, size_t bytes, bool read_only,
5162 bool allow_exec) {
5163 // same as map_memory() on this OS
5164 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5165 allow_exec);
5166 }
5167
5168
5169 // Unmap a block of memory.
5170 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5171 return munmap(addr, bytes) == 0;
5172 }
5173
5174 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5175
5176 static clockid_t thread_cpu_clockid(Thread* thread) {
5177 pthread_t tid = thread->osthread()->pthread_id();
5178 clockid_t clockid;
5179
5180 // Get thread clockid
5181 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5182 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5183 return clockid;
5184 }
5185
5186 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5187 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5188 // of a thread.
5189 //
5190 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5191 // the fast estimate available on the platform.
5192
5193 jlong os::current_thread_cpu_time() {
5194 if (os::Linux::supports_fast_thread_cpu_time()) {
5195 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5196 } else {
5197 // return user + sys since the cost is the same
5198 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5199 }
5200 }
5201
5202 jlong os::thread_cpu_time(Thread* thread) {
5203 // consistent with what current_thread_cpu_time() returns
5204 if (os::Linux::supports_fast_thread_cpu_time()) {
5205 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5206 } else {
5207 return slow_thread_cpu_time(thread, true /* user + sys */);
5208 }
5209 }
5210
5211 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5212 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5213 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5214 } else {
5215 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5216 }
5217 }
5218
5219 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5220 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5221 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5222 } else {
5223 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5224 }
5225 }
5226
5227 // -1 on error.
5228 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5229 pid_t tid = thread->osthread()->thread_id();
5230 char *s;
5231 char stat[2048];
5232 int statlen;
5233 char proc_name[64];
5234 int count;
5235 long sys_time, user_time;
5236 char cdummy;
5237 int idummy;
5238 long ldummy;
5239 FILE *fp;
5240
5241 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5242 fp = fopen(proc_name, "r");
5243 if (fp == NULL) return -1;
5244 statlen = fread(stat, 1, 2047, fp);
5245 stat[statlen] = '\0';
5246 fclose(fp);
5247
5248 // Skip pid and the command string. Note that we could be dealing with
5249 // weird command names, e.g. user could decide to rename java launcher
5250 // to "java 1.4.2 :)", then the stat file would look like
5251 // 1234 (java 1.4.2 :)) R ... ...
5252 // We don't really need to know the command string, just find the last
5253 // occurrence of ")" and then start parsing from there. See bug 4726580.
5254 s = strrchr(stat, ')');
5255 if (s == NULL) return -1;
5256
5257 // Skip blank chars
5258 do { s++; } while (s && isspace(*s));
5259
5260 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5261 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5262 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5263 &user_time, &sys_time);
5264 if (count != 13) return -1;
5265 if (user_sys_cpu_time) {
5266 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5267 } else {
5268 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5269 }
5270 }
5271
5272 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5273 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5274 info_ptr->may_skip_backward = false; // elapsed time not wall time
5275 info_ptr->may_skip_forward = false; // elapsed time not wall time
5276 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5277 }
5278
5279 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5280 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5281 info_ptr->may_skip_backward = false; // elapsed time not wall time
5282 info_ptr->may_skip_forward = false; // elapsed time not wall time
5283 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5284 }
5285
5286 bool os::is_thread_cpu_time_supported() {
5287 return true;
5288 }
5289
5290 // System loadavg support. Returns -1 if load average cannot be obtained.
5291 // Linux doesn't yet have a (official) notion of processor sets,
5292 // so just return the system wide load average.
5293 int os::loadavg(double loadavg[], int nelem) {
5294 return ::getloadavg(loadavg, nelem);
5295 }
5296
5297 void os::pause() {
5298 char filename[MAX_PATH];
5299 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5300 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5301 } else {
5302 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5303 }
5304
5305 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5306 if (fd != -1) {
5307 struct stat buf;
5308 ::close(fd);
5309 while (::stat(filename, &buf) == 0) {
5310 (void)::poll(NULL, 0, 100);
5311 }
5312 } else {
5313 jio_fprintf(stderr,
5314 "Could not open pause file '%s', continuing immediately.\n", filename);
5315 }
5316 }
5317
5318
5319 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5320 // comment paragraphs are worth repeating here:
5321 //
5322 // Assumption:
5323 // Only one parker can exist on an event, which is why we allocate
5324 // them per-thread. Multiple unparkers can coexist.
5325 //
5326 // _Event serves as a restricted-range semaphore.
5327 // -1 : thread is blocked, i.e. there is a waiter
5328 // 0 : neutral: thread is running or ready,
5329 // could have been signaled after a wait started
5330 // 1 : signaled - thread is running or ready
5331 //
5332 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5333 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5334 // For specifics regarding the bug see GLIBC BUGID 261237 :
5335 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5336 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5337 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5338 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5339 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5340 // and monitorenter when we're using 1-0 locking. All those operations may result in
5341 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5342 // of libpthread avoids the problem, but isn't practical.
5343 //
5344 // Possible remedies:
5345 //
5346 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5347 // This is palliative and probabilistic, however. If the thread is preempted
5348 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5349 // than the minimum period may have passed, and the abstime may be stale (in the
5350 // past) resultin in a hang. Using this technique reduces the odds of a hang
5351 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5352 //
5353 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5354 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5355 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5356 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5357 // thread.
5358 //
5359 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5360 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5361 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5362 // This also works well. In fact it avoids kernel-level scalability impediments
5363 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5364 // timers in a graceful fashion.
5365 //
5366 // 4. When the abstime value is in the past it appears that control returns
5367 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5368 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5369 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5370 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5371 // It may be possible to avoid reinitialization by checking the return
5372 // value from pthread_cond_timedwait(). In addition to reinitializing the
5373 // condvar we must establish the invariant that cond_signal() is only called
5374 // within critical sections protected by the adjunct mutex. This prevents
5375 // cond_signal() from "seeing" a condvar that's in the midst of being
5376 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5377 // desirable signal-after-unlock optimization that avoids futile context switching.
5378 //
5379 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5380 // structure when a condvar is used or initialized. cond_destroy() would
5381 // release the helper structure. Our reinitialize-after-timedwait fix
5382 // put excessive stress on malloc/free and locks protecting the c-heap.
5383 //
5384 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5385 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5386 // and only enabling the work-around for vulnerable environments.
5387
5388 // utility to compute the abstime argument to timedwait:
5389 // millis is the relative timeout time
5390 // abstime will be the absolute timeout time
5391 // TODO: replace compute_abstime() with unpackTime()
5392
5393 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5394 if (millis < 0) millis = 0;
5395
5396 jlong seconds = millis / 1000;
5397 millis %= 1000;
5398 if (seconds > 50000000) { // see man cond_timedwait(3T)
5399 seconds = 50000000;
5400 }
5401
5402 if (os::supports_monotonic_clock()) {
5403 struct timespec now;
5404 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5405 assert_status(status == 0, status, "clock_gettime");
5406 abstime->tv_sec = now.tv_sec + seconds;
5407 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5408 if (nanos >= NANOSECS_PER_SEC) {
5409 abstime->tv_sec += 1;
5410 nanos -= NANOSECS_PER_SEC;
5411 }
5412 abstime->tv_nsec = nanos;
5413 } else {
5414 struct timeval now;
5415 int status = gettimeofday(&now, NULL);
5416 assert(status == 0, "gettimeofday");
5417 abstime->tv_sec = now.tv_sec + seconds;
5418 long usec = now.tv_usec + millis * 1000;
5419 if (usec >= 1000000) {
5420 abstime->tv_sec += 1;
5421 usec -= 1000000;
5422 }
5423 abstime->tv_nsec = usec * 1000;
5424 }
5425 return abstime;
5426 }
5427
5428 void os::PlatformEvent::park() { // AKA "down()"
5429 // Transitions for _Event:
5430 // -1 => -1 : illegal
5431 // 1 => 0 : pass - return immediately
5432 // 0 => -1 : block; then set _Event to 0 before returning
5433
5434 // Invariant: Only the thread associated with the Event/PlatformEvent
5435 // may call park().
5436 // TODO: assert that _Assoc != NULL or _Assoc == Self
5437 assert(_nParked == 0, "invariant");
5438
5439 int v;
5440 for (;;) {
5441 v = _Event;
5442 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5443 }
5444 guarantee(v >= 0, "invariant");
5445 if (v == 0) {
5446 // Do this the hard way by blocking ...
5447 int status = pthread_mutex_lock(_mutex);
5448 assert_status(status == 0, status, "mutex_lock");
5449 guarantee(_nParked == 0, "invariant");
5450 ++_nParked;
5451 while (_Event < 0) {
5452 status = pthread_cond_wait(_cond, _mutex);
5453 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5454 // Treat this the same as if the wait was interrupted
5455 if (status == ETIME) { status = EINTR; }
5456 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5457 }
5458 --_nParked;
5459
5460 _Event = 0;
5461 status = pthread_mutex_unlock(_mutex);
5462 assert_status(status == 0, status, "mutex_unlock");
5463 // Paranoia to ensure our locked and lock-free paths interact
5464 // correctly with each other.
5465 OrderAccess::fence();
5466 }
5467 guarantee(_Event >= 0, "invariant");
5468 }
5469
5470 int os::PlatformEvent::park(jlong millis) {
5471 // Transitions for _Event:
5472 // -1 => -1 : illegal
5473 // 1 => 0 : pass - return immediately
5474 // 0 => -1 : block; then set _Event to 0 before returning
5475
5476 guarantee(_nParked == 0, "invariant");
5477
5478 int v;
5479 for (;;) {
5480 v = _Event;
5481 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5482 }
5483 guarantee(v >= 0, "invariant");
5484 if (v != 0) return OS_OK;
5485
5486 // We do this the hard way, by blocking the thread.
5487 // Consider enforcing a minimum timeout value.
5488 struct timespec abst;
5489 compute_abstime(&abst, millis);
5490
5491 int ret = OS_TIMEOUT;
5492 int status = pthread_mutex_lock(_mutex);
5493 assert_status(status == 0, status, "mutex_lock");
5494 guarantee(_nParked == 0, "invariant");
5495 ++_nParked;
5496
5497 // Object.wait(timo) will return because of
5498 // (a) notification
5499 // (b) timeout
5500 // (c) thread.interrupt
5501 //
5502 // Thread.interrupt and object.notify{All} both call Event::set.
5503 // That is, we treat thread.interrupt as a special case of notification.
5504 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5505 // We assume all ETIME returns are valid.
5506 //
5507 // TODO: properly differentiate simultaneous notify+interrupt.
5508 // In that case, we should propagate the notify to another waiter.
5509
5510 while (_Event < 0) {
5511 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5512 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5513 pthread_cond_destroy(_cond);
5514 pthread_cond_init(_cond, os::Linux::condAttr());
5515 }
5516 assert_status(status == 0 || status == EINTR ||
5517 status == ETIME || status == ETIMEDOUT,
5518 status, "cond_timedwait");
5519 if (!FilterSpuriousWakeups) break; // previous semantics
5520 if (status == ETIME || status == ETIMEDOUT) break;
5521 // We consume and ignore EINTR and spurious wakeups.
5522 }
5523 --_nParked;
5524 if (_Event >= 0) {
5525 ret = OS_OK;
5526 }
5527 _Event = 0;
5528 status = pthread_mutex_unlock(_mutex);
5529 assert_status(status == 0, status, "mutex_unlock");
5530 assert(_nParked == 0, "invariant");
5531 // Paranoia to ensure our locked and lock-free paths interact
5532 // correctly with each other.
5533 OrderAccess::fence();
5534 return ret;
5535 }
5536
5537 void os::PlatformEvent::unpark() {
5538 // Transitions for _Event:
5539 // 0 => 1 : just return
5540 // 1 => 1 : just return
5541 // -1 => either 0 or 1; must signal target thread
5542 // That is, we can safely transition _Event from -1 to either
5543 // 0 or 1.
5544 // See also: "Semaphores in Plan 9" by Mullender & Cox
5545 //
5546 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5547 // that it will take two back-to-back park() calls for the owning
5548 // thread to block. This has the benefit of forcing a spurious return
5549 // from the first park() call after an unpark() call which will help
5550 // shake out uses of park() and unpark() without condition variables.
5551
5552 if (Atomic::xchg(1, &_Event) >= 0) return;
5553
5554 // Wait for the thread associated with the event to vacate
5555 int status = pthread_mutex_lock(_mutex);
5556 assert_status(status == 0, status, "mutex_lock");
5557 int AnyWaiters = _nParked;
5558 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5559 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5560 AnyWaiters = 0;
5561 pthread_cond_signal(_cond);
5562 }
5563 status = pthread_mutex_unlock(_mutex);
5564 assert_status(status == 0, status, "mutex_unlock");
5565 if (AnyWaiters != 0) {
5566 // Note that we signal() *after* dropping the lock for "immortal" Events.
5567 // This is safe and avoids a common class of futile wakeups. In rare
5568 // circumstances this can cause a thread to return prematurely from
5569 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5570 // will simply re-test the condition and re-park itself.
5571 // This provides particular benefit if the underlying platform does not
5572 // provide wait morphing.
5573 status = pthread_cond_signal(_cond);
5574 assert_status(status == 0, status, "cond_signal");
5575 }
5576 }
5577
5578
5579 // JSR166
5580 // -------------------------------------------------------
5581
5582 // The solaris and linux implementations of park/unpark are fairly
5583 // conservative for now, but can be improved. They currently use a
5584 // mutex/condvar pair, plus a a count.
5585 // Park decrements count if > 0, else does a condvar wait. Unpark
5586 // sets count to 1 and signals condvar. Only one thread ever waits
5587 // on the condvar. Contention seen when trying to park implies that someone
5588 // is unparking you, so don't wait. And spurious returns are fine, so there
5589 // is no need to track notifications.
5590
5591 // This code is common to linux and solaris and will be moved to a
5592 // common place in dolphin.
5593 //
5594 // The passed in time value is either a relative time in nanoseconds
5595 // or an absolute time in milliseconds. Either way it has to be unpacked
5596 // into suitable seconds and nanoseconds components and stored in the
5597 // given timespec structure.
5598 // Given time is a 64-bit value and the time_t used in the timespec is only
5599 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5600 // overflow if times way in the future are given. Further on Solaris versions
5601 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5602 // number of seconds, in abstime, is less than current_time + 100,000,000.
5603 // As it will be 28 years before "now + 100000000" will overflow we can
5604 // ignore overflow and just impose a hard-limit on seconds using the value
5605 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5606 // years from "now".
5607
5608 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5609 assert(time > 0, "convertTime");
5610 time_t max_secs = 0;
5611
5612 if (!os::supports_monotonic_clock() || isAbsolute) {
5613 struct timeval now;
5614 int status = gettimeofday(&now, NULL);
5615 assert(status == 0, "gettimeofday");
5616
5617 max_secs = now.tv_sec + MAX_SECS;
5618
5619 if (isAbsolute) {
5620 jlong secs = time / 1000;
5621 if (secs > max_secs) {
5622 absTime->tv_sec = max_secs;
5623 } else {
5624 absTime->tv_sec = secs;
5625 }
5626 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5627 } else {
5628 jlong secs = time / NANOSECS_PER_SEC;
5629 if (secs >= MAX_SECS) {
5630 absTime->tv_sec = max_secs;
5631 absTime->tv_nsec = 0;
5632 } else {
5633 absTime->tv_sec = now.tv_sec + secs;
5634 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5635 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5636 absTime->tv_nsec -= NANOSECS_PER_SEC;
5637 ++absTime->tv_sec; // note: this must be <= max_secs
5638 }
5639 }
5640 }
5641 } else {
5642 // must be relative using monotonic clock
5643 struct timespec now;
5644 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5645 assert_status(status == 0, status, "clock_gettime");
5646 max_secs = now.tv_sec + MAX_SECS;
5647 jlong secs = time / NANOSECS_PER_SEC;
5648 if (secs >= MAX_SECS) {
5649 absTime->tv_sec = max_secs;
5650 absTime->tv_nsec = 0;
5651 } else {
5652 absTime->tv_sec = now.tv_sec + secs;
5653 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5654 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5655 absTime->tv_nsec -= NANOSECS_PER_SEC;
5656 ++absTime->tv_sec; // note: this must be <= max_secs
5657 }
5658 }
5659 }
5660 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5661 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5662 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5663 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5664 }
5665
5666 void Parker::park(bool isAbsolute, jlong time) {
5667 // Ideally we'd do something useful while spinning, such
5668 // as calling unpackTime().
5669
5670 // Optional fast-path check:
5671 // Return immediately if a permit is available.
5672 // We depend on Atomic::xchg() having full barrier semantics
5673 // since we are doing a lock-free update to _counter.
5674 if (Atomic::xchg(0, &_counter) > 0) return;
5675
5676 Thread* thread = Thread::current();
5677 assert(thread->is_Java_thread(), "Must be JavaThread");
5678 JavaThread *jt = (JavaThread *)thread;
5679
5680 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5681 // Check interrupt before trying to wait
5682 if (Thread::is_interrupted(thread, false)) {
5683 return;
5684 }
5685
5686 // Next, demultiplex/decode time arguments
5687 timespec absTime;
5688 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5689 return;
5690 }
5691 if (time > 0) {
5692 unpackTime(&absTime, isAbsolute, time);
5693 }
5694
5695
5696 // Enter safepoint region
5697 // Beware of deadlocks such as 6317397.
5698 // The per-thread Parker:: mutex is a classic leaf-lock.
5699 // In particular a thread must never block on the Threads_lock while
5700 // holding the Parker:: mutex. If safepoints are pending both the
5701 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5702 ThreadBlockInVM tbivm(jt);
5703
5704 // Don't wait if cannot get lock since interference arises from
5705 // unblocking. Also. check interrupt before trying wait
5706 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5707 return;
5708 }
5709
5710 int status;
5711 if (_counter > 0) { // no wait needed
5712 _counter = 0;
5713 status = pthread_mutex_unlock(_mutex);
5714 assert(status == 0, "invariant");
5715 // Paranoia to ensure our locked and lock-free paths interact
5716 // correctly with each other and Java-level accesses.
5717 OrderAccess::fence();
5718 return;
5719 }
5720
5721 #ifdef ASSERT
5722 // Don't catch signals while blocked; let the running threads have the signals.
5723 // (This allows a debugger to break into the running thread.)
5724 sigset_t oldsigs;
5725 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5726 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5727 #endif
5728
5729 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5730 jt->set_suspend_equivalent();
5731 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5732
5733 assert(_cur_index == -1, "invariant");
5734 if (time == 0) {
5735 _cur_index = REL_INDEX; // arbitrary choice when not timed
5736 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5737 } else {
5738 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5739 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5740 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5741 pthread_cond_destroy(&_cond[_cur_index]);
5742 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5743 }
5744 }
5745 _cur_index = -1;
5746 assert_status(status == 0 || status == EINTR ||
5747 status == ETIME || status == ETIMEDOUT,
5748 status, "cond_timedwait");
5749
5750 #ifdef ASSERT
5751 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5752 #endif
5753
5754 _counter = 0;
5755 status = pthread_mutex_unlock(_mutex);
5756 assert_status(status == 0, status, "invariant");
5757 // Paranoia to ensure our locked and lock-free paths interact
5758 // correctly with each other and Java-level accesses.
5759 OrderAccess::fence();
5760
5761 // If externally suspended while waiting, re-suspend
5762 if (jt->handle_special_suspend_equivalent_condition()) {
5763 jt->java_suspend_self();
5764 }
5765 }
5766
5767 void Parker::unpark() {
5768 int status = pthread_mutex_lock(_mutex);
5769 assert(status == 0, "invariant");
5770 const int s = _counter;
5771 _counter = 1;
5772 if (s < 1) {
5773 // thread might be parked
5774 if (_cur_index != -1) {
5775 // thread is definitely parked
5776 if (WorkAroundNPTLTimedWaitHang) {
5777 status = pthread_cond_signal(&_cond[_cur_index]);
5778 assert(status == 0, "invariant");
5779 status = pthread_mutex_unlock(_mutex);
5780 assert(status == 0, "invariant");
5781 } else {
5782 status = pthread_mutex_unlock(_mutex);
5783 assert(status == 0, "invariant");
5784 status = pthread_cond_signal(&_cond[_cur_index]);
5785 assert(status == 0, "invariant");
5786 }
5787 } else {
5788 pthread_mutex_unlock(_mutex);
5789 assert(status == 0, "invariant");
5790 }
5791 } else {
5792 pthread_mutex_unlock(_mutex);
5793 assert(status == 0, "invariant");
5794 }
5795 }
5796
5797
5798 extern char** environ;
5799
5800 // Run the specified command in a separate process. Return its exit value,
5801 // or -1 on failure (e.g. can't fork a new process).
5802 // Unlike system(), this function can be called from signal handler. It
5803 // doesn't block SIGINT et al.
5804 int os::fork_and_exec(char* cmd) {
5805 const char * argv[4] = {"sh", "-c", cmd, NULL};
5806
5807 pid_t pid = fork();
5808
5809 if (pid < 0) {
5810 // fork failed
5811 return -1;
5812
5813 } else if (pid == 0) {
5814 // child process
5815
5816 execve("/bin/sh", (char* const*)argv, environ);
5817
5818 // execve failed
5819 _exit(-1);
5820
5821 } else {
5822 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5823 // care about the actual exit code, for now.
5824
5825 int status;
5826
5827 // Wait for the child process to exit. This returns immediately if
5828 // the child has already exited. */
5829 while (waitpid(pid, &status, 0) < 0) {
5830 switch (errno) {
5831 case ECHILD: return 0;
5832 case EINTR: break;
5833 default: return -1;
5834 }
5835 }
5836
5837 if (WIFEXITED(status)) {
5838 // The child exited normally; get its exit code.
5839 return WEXITSTATUS(status);
5840 } else if (WIFSIGNALED(status)) {
5841 // The child exited because of a signal
5842 // The best value to return is 0x80 + signal number,
5843 // because that is what all Unix shells do, and because
5844 // it allows callers to distinguish between process exit and
5845 // process death by signal.
5846 return 0x80 + WTERMSIG(status);
5847 } else {
5848 // Unknown exit code; pass it through
5849 return status;
5850 }
5851 }
5852 }
5853
5854 // is_headless_jre()
5855 //
5856 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5857 // in order to report if we are running in a headless jre
5858 //
5859 // Since JDK8 xawt/libmawt.so was moved into the same directory
5860 // as libawt.so, and renamed libawt_xawt.so
5861 //
5862 bool os::is_headless_jre() {
5863 struct stat statbuf;
5864 char buf[MAXPATHLEN];
5865 char libmawtpath[MAXPATHLEN];
5866 const char *xawtstr = "/xawt/libmawt.so";
5867 const char *new_xawtstr = "/libawt_xawt.so";
5868 char *p;
5869
5870 // Get path to libjvm.so
5871 os::jvm_path(buf, sizeof(buf));
5872
5873 // Get rid of libjvm.so
5874 p = strrchr(buf, '/');
5875 if (p == NULL) {
5876 return false;
5877 } else {
5878 *p = '\0';
5879 }
5880
5881 // Get rid of client or server
5882 p = strrchr(buf, '/');
5883 if (p == NULL) {
5884 return false;
5885 } else {
5886 *p = '\0';
5887 }
5888
5889 // check xawt/libmawt.so
5890 strcpy(libmawtpath, buf);
5891 strcat(libmawtpath, xawtstr);
5892 if (::stat(libmawtpath, &statbuf) == 0) return false;
5893
5894 // check libawt_xawt.so
5895 strcpy(libmawtpath, buf);
5896 strcat(libmawtpath, new_xawtstr);
5897 if (::stat(libmawtpath, &statbuf) == 0) return false;
5898
5899 return true;
5900 }
5901
5902 // Get the default path to the core file
5903 // Returns the length of the string
5904 int os::get_core_path(char* buffer, size_t bufferSize) {
5905 /*
5906 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5907 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5908 */
5909 const int core_pattern_len = 129;
5910 char core_pattern[core_pattern_len] = {0};
5911
5912 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5913 if (core_pattern_file != -1) {
5914 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5915 ::close(core_pattern_file);
5916
5917 if (ret > 0) {
5918 char *last_char = core_pattern + strlen(core_pattern) - 1;
5919
5920 if (*last_char == '\n') {
5921 *last_char = '\0';
5922 }
5923 }
5924 }
5925
5926 if (strlen(core_pattern) == 0) {
5927 return -1;
5928 }
5929
5930 char *pid_pos = strstr(core_pattern, "%p");
5931 int written;
5932
5933 if (core_pattern[0] == '/') {
5934 written = jio_snprintf(buffer, bufferSize, core_pattern);
5935 } else {
5936 char cwd[PATH_MAX];
5937
5938 const char* p = get_current_directory(cwd, PATH_MAX);
5939 if (p == NULL) {
5940 return -1;
5941 }
5942
5943 if (core_pattern[0] == '|') {
5944 written = jio_snprintf(buffer, bufferSize,
5945 "\"%s\" (or dumping to %s/core.%d)",
5946 &core_pattern[1], p, current_process_id());
5947 } else {
5948 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5949 }
5950 }
5951
5952 if (written < 0) {
5953 return -1;
5954 }
5955
5956 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5957 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5958
5959 if (core_uses_pid_file != -1) {
5960 char core_uses_pid = 0;
5961 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5962 ::close(core_uses_pid_file);
5963
5964 if (core_uses_pid == '1') {
5965 jio_snprintf(buffer + written, bufferSize - written,
5966 ".%d", current_process_id());
5967 }
5968 }
5969 }
5970
5971 return strlen(buffer);
5972 }
5973
5974 /////////////// Unit tests ///////////////
5975
5976 #ifndef PRODUCT
5977
5978 #define test_log(...) \
5979 do { \
5980 if (VerboseInternalVMTests) { \
5981 tty->print_cr(__VA_ARGS__); \
5982 tty->flush(); \
5983 } \
5984 } while (false)
5985
5986 class TestReserveMemorySpecial : AllStatic {
5987 public:
5988 static void small_page_write(void* addr, size_t size) {
5989 size_t page_size = os::vm_page_size();
5990
5991 char* end = (char*)addr + size;
5992 for (char* p = (char*)addr; p < end; p += page_size) {
5993 *p = 1;
5994 }
5995 }
5996
5997 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5998 if (!UseHugeTLBFS) {
5999 return;
6000 }
6001
6002 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6003
6004 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6005
6006 if (addr != NULL) {
6007 small_page_write(addr, size);
6008
6009 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6010 }
6011 }
6012
6013 static void test_reserve_memory_special_huge_tlbfs_only() {
6014 if (!UseHugeTLBFS) {
6015 return;
6016 }
6017
6018 size_t lp = os::large_page_size();
6019
6020 for (size_t size = lp; size <= lp * 10; size += lp) {
6021 test_reserve_memory_special_huge_tlbfs_only(size);
6022 }
6023 }
6024
6025 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6026 size_t lp = os::large_page_size();
6027 size_t ag = os::vm_allocation_granularity();
6028
6029 // sizes to test
6030 const size_t sizes[] = {
6031 lp, lp + ag, lp + lp / 2, lp * 2,
6032 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6033 lp * 10, lp * 10 + lp / 2
6034 };
6035 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6036
6037 // For each size/alignment combination, we test three scenarios:
6038 // 1) with req_addr == NULL
6039 // 2) with a non-null req_addr at which we expect to successfully allocate
6040 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6041 // expect the allocation to either fail or to ignore req_addr
6042
6043 // Pre-allocate two areas; they shall be as large as the largest allocation
6044 // and aligned to the largest alignment we will be testing.
6045 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6046 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6047 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6048 -1, 0);
6049 assert(mapping1 != MAP_FAILED, "should work");
6050
6051 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6052 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6053 -1, 0);
6054 assert(mapping2 != MAP_FAILED, "should work");
6055
6056 // Unmap the first mapping, but leave the second mapping intact: the first
6057 // mapping will serve as a value for a "good" req_addr (case 2). The second
6058 // mapping, still intact, as "bad" req_addr (case 3).
6059 ::munmap(mapping1, mapping_size);
6060
6061 // Case 1
6062 test_log("%s, req_addr NULL:", __FUNCTION__);
6063 test_log("size align result");
6064
6065 for (int i = 0; i < num_sizes; i++) {
6066 const size_t size = sizes[i];
6067 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6068 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6069 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6070 size, alignment, p, (p != NULL ? "" : "(failed)"));
6071 if (p != NULL) {
6072 assert(is_ptr_aligned(p, alignment), "must be");
6073 small_page_write(p, size);
6074 os::Linux::release_memory_special_huge_tlbfs(p, size);
6075 }
6076 }
6077 }
6078
6079 // Case 2
6080 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6081 test_log("size align req_addr result");
6082
6083 for (int i = 0; i < num_sizes; i++) {
6084 const size_t size = sizes[i];
6085 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6086 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6087 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6088 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6089 size, alignment, req_addr, p,
6090 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6091 if (p != NULL) {
6092 assert(p == req_addr, "must be");
6093 small_page_write(p, size);
6094 os::Linux::release_memory_special_huge_tlbfs(p, size);
6095 }
6096 }
6097 }
6098
6099 // Case 3
6100 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6101 test_log("size align req_addr result");
6102
6103 for (int i = 0; i < num_sizes; i++) {
6104 const size_t size = sizes[i];
6105 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6106 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6107 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6108 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6109 size, alignment, req_addr, p,
6110 ((p != NULL ? "" : "(failed)")));
6111 // as the area around req_addr contains already existing mappings, the API should always
6112 // return NULL (as per contract, it cannot return another address)
6113 assert(p == NULL, "must be");
6114 }
6115 }
6116
6117 ::munmap(mapping2, mapping_size);
6118
6119 }
6120
6121 static void test_reserve_memory_special_huge_tlbfs() {
6122 if (!UseHugeTLBFS) {
6123 return;
6124 }
6125
6126 test_reserve_memory_special_huge_tlbfs_only();
6127 test_reserve_memory_special_huge_tlbfs_mixed();
6128 }
6129
6130 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6131 if (!UseSHM) {
6132 return;
6133 }
6134
6135 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6136
6137 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6138
6139 if (addr != NULL) {
6140 assert(is_ptr_aligned(addr, alignment), "Check");
6141 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6142
6143 small_page_write(addr, size);
6144
6145 os::Linux::release_memory_special_shm(addr, size);
6146 }
6147 }
6148
6149 static void test_reserve_memory_special_shm() {
6150 size_t lp = os::large_page_size();
6151 size_t ag = os::vm_allocation_granularity();
6152
6153 for (size_t size = ag; size < lp * 3; size += ag) {
6154 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6155 test_reserve_memory_special_shm(size, alignment);
6156 }
6157 }
6158 }
6159
6160 static void test() {
6161 test_reserve_memory_special_huge_tlbfs();
6162 test_reserve_memory_special_shm();
6163 }
6164 };
6165
6166 void TestReserveMemorySpecial_test() {
6167 TestReserveMemorySpecial::test();
6168 }
6169
6170 #endif