1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
110
111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
112 // getrusage() is prepared to handle the associated failure.
113 #ifndef RUSAGE_THREAD
114 #define RUSAGE_THREAD (1) /* only the calling thread */
115 #endif
116
117 #define MAX_PATH (2 * K)
118
119 #define MAX_SECS 100000000
120
121 // for timer info max values which include all bits
122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123
124 #define LARGEPAGES_BIT (1 << 6)
125 ////////////////////////////////////////////////////////////////////////////////
126 // global variables
127 julong os::Linux::_physical_memory = 0;
128
129 address os::Linux::_initial_thread_stack_bottom = NULL;
130 uintptr_t os::Linux::_initial_thread_stack_size = 0;
131
132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 const int os::Linux::_vm_default_page_size = (8 * K);
139 bool os::Linux::_supports_fast_thread_cpu_time = false;
140 const char * os::Linux::_glibc_version = NULL;
141 const char * os::Linux::_libpthread_version = NULL;
142 pthread_condattr_t os::Linux::_condattr[1];
143
144 static jlong initial_time_count=0;
145
146 static int clock_tics_per_sec = 100;
147
148 // For diagnostics to print a message once. see run_periodic_checks
149 static sigset_t check_signal_done;
150 static bool check_signals = true;
151
152 // Signal number used to suspend/resume a thread
153
154 // do not use any signal number less than SIGSEGV, see 4355769
155 static int SR_signum = SIGUSR2;
156 sigset_t SR_sigset;
157
158 // Declarations
159 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
160
161 // utility functions
162
163 static int SR_initialize();
164
165 julong os::available_memory() {
166 return Linux::available_memory();
167 }
168
169 julong os::Linux::available_memory() {
170 // values in struct sysinfo are "unsigned long"
171 struct sysinfo si;
172 sysinfo(&si);
173
174 return (julong)si.freeram * si.mem_unit;
175 }
176
177 julong os::physical_memory() {
178 return Linux::physical_memory();
179 }
180
181 // Return true if user is running as root.
182
183 bool os::have_special_privileges() {
184 static bool init = false;
185 static bool privileges = false;
186 if (!init) {
187 privileges = (getuid() != geteuid()) || (getgid() != getegid());
188 init = true;
189 }
190 return privileges;
191 }
192
193
194 #ifndef SYS_gettid
195 // i386: 224, ia64: 1105, amd64: 186, sparc 143
196 #ifdef __ia64__
197 #define SYS_gettid 1105
198 #else
199 #ifdef __i386__
200 #define SYS_gettid 224
201 #else
202 #ifdef __amd64__
203 #define SYS_gettid 186
204 #else
205 #ifdef __sparc__
206 #define SYS_gettid 143
207 #else
208 #error define gettid for the arch
209 #endif
210 #endif
211 #endif
212 #endif
213 #endif
214
215 // Cpu architecture string
216 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
217
218
219 // pid_t gettid()
220 //
221 // Returns the kernel thread id of the currently running thread. Kernel
222 // thread id is used to access /proc.
223 pid_t os::Linux::gettid() {
224 int rslt = syscall(SYS_gettid);
225 assert(rslt != -1, "must be."); // old linuxthreads implementation?
226 return (pid_t)rslt;
227 }
228
229 // Most versions of linux have a bug where the number of processors are
230 // determined by looking at the /proc file system. In a chroot environment,
231 // the system call returns 1. This causes the VM to act as if it is
232 // a single processor and elide locking (see is_MP() call).
233 static bool unsafe_chroot_detected = false;
234 static const char *unstable_chroot_error = "/proc file system not found.\n"
235 "Java may be unstable running multithreaded in a chroot "
236 "environment on Linux when /proc filesystem is not mounted.";
237
238 void os::Linux::initialize_system_info() {
239 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
240 if (processor_count() == 1) {
241 pid_t pid = os::Linux::gettid();
242 char fname[32];
243 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
244 FILE *fp = fopen(fname, "r");
245 if (fp == NULL) {
246 unsafe_chroot_detected = true;
247 } else {
248 fclose(fp);
249 }
250 }
251 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
252 assert(processor_count() > 0, "linux error");
253 }
254
255 void os::init_system_properties_values() {
256 // The next steps are taken in the product version:
257 //
258 // Obtain the JAVA_HOME value from the location of libjvm.so.
259 // This library should be located at:
260 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
261 //
262 // If "/jre/lib/" appears at the right place in the path, then we
263 // assume libjvm.so is installed in a JDK and we use this path.
264 //
265 // Otherwise exit with message: "Could not create the Java virtual machine."
266 //
267 // The following extra steps are taken in the debugging version:
268 //
269 // If "/jre/lib/" does NOT appear at the right place in the path
270 // instead of exit check for $JAVA_HOME environment variable.
271 //
272 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
273 // then we append a fake suffix "hotspot/libjvm.so" to this path so
274 // it looks like libjvm.so is installed there
275 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
276 //
277 // Otherwise exit.
278 //
279 // Important note: if the location of libjvm.so changes this
280 // code needs to be changed accordingly.
281
282 // See ld(1):
283 // The linker uses the following search paths to locate required
284 // shared libraries:
285 // 1: ...
286 // ...
287 // 7: The default directories, normally /lib and /usr/lib.
288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
289 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
290 #else
291 #define DEFAULT_LIBPATH "/lib:/usr/lib"
292 #endif
293
294 // Base path of extensions installed on the system.
295 #define SYS_EXT_DIR "/usr/java/packages"
296 #define EXTENSIONS_DIR "/lib/ext"
297
298 // Buffer that fits several sprintfs.
299 // Note that the space for the colon and the trailing null are provided
300 // by the nulls included by the sizeof operator.
301 const size_t bufsize =
302 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
303 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
304 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
305
306 // sysclasspath, java_home, dll_dir
307 {
308 char *pslash;
309 os::jvm_path(buf, bufsize);
310
311 // Found the full path to libjvm.so.
312 // Now cut the path to <java_home>/jre if we can.
313 pslash = strrchr(buf, '/');
314 if (pslash != NULL) {
315 *pslash = '\0'; // Get rid of /libjvm.so.
316 }
317 pslash = strrchr(buf, '/');
318 if (pslash != NULL) {
319 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
320 }
321 Arguments::set_dll_dir(buf);
322
323 if (pslash != NULL) {
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /<arch>.
327 pslash = strrchr(buf, '/');
328 if (pslash != NULL) {
329 *pslash = '\0'; // Get rid of /lib.
330 }
331 }
332 }
333 Arguments::set_java_home(buf);
334 set_boot_path('/', ':');
335 }
336
337 // Where to look for native libraries.
338 //
339 // Note: Due to a legacy implementation, most of the library path
340 // is set in the launcher. This was to accomodate linking restrictions
341 // on legacy Linux implementations (which are no longer supported).
342 // Eventually, all the library path setting will be done here.
343 //
344 // However, to prevent the proliferation of improperly built native
345 // libraries, the new path component /usr/java/packages is added here.
346 // Eventually, all the library path setting will be done here.
347 {
348 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
349 // should always exist (until the legacy problem cited above is
350 // addressed).
351 const char *v = ::getenv("LD_LIBRARY_PATH");
352 const char *v_colon = ":";
353 if (v == NULL) { v = ""; v_colon = ""; }
354 // That's +1 for the colon and +1 for the trailing '\0'.
355 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
356 strlen(v) + 1 +
357 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
358 mtInternal);
359 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
360 Arguments::set_library_path(ld_library_path);
361 FREE_C_HEAP_ARRAY(char, ld_library_path);
362 }
363
364 // Extensions directories.
365 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
366 Arguments::set_ext_dirs(buf);
367
368 FREE_C_HEAP_ARRAY(char, buf);
369
370 #undef DEFAULT_LIBPATH
371 #undef SYS_EXT_DIR
372 #undef EXTENSIONS_DIR
373 }
374
375 ////////////////////////////////////////////////////////////////////////////////
376 // breakpoint support
377
378 void os::breakpoint() {
379 BREAKPOINT;
380 }
381
382 extern "C" void breakpoint() {
383 // use debugger to set breakpoint here
384 }
385
386 ////////////////////////////////////////////////////////////////////////////////
387 // signal support
388
389 debug_only(static bool signal_sets_initialized = false);
390 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
391
392 bool os::Linux::is_sig_ignored(int sig) {
393 struct sigaction oact;
394 sigaction(sig, (struct sigaction*)NULL, &oact);
395 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
396 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
397 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
398 return true;
399 } else {
400 return false;
401 }
402 }
403
404 void os::Linux::signal_sets_init() {
405 // Should also have an assertion stating we are still single-threaded.
406 assert(!signal_sets_initialized, "Already initialized");
407 // Fill in signals that are necessarily unblocked for all threads in
408 // the VM. Currently, we unblock the following signals:
409 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
410 // by -Xrs (=ReduceSignalUsage));
411 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
412 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
413 // the dispositions or masks wrt these signals.
414 // Programs embedding the VM that want to use the above signals for their
415 // own purposes must, at this time, use the "-Xrs" option to prevent
416 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
417 // (See bug 4345157, and other related bugs).
418 // In reality, though, unblocking these signals is really a nop, since
419 // these signals are not blocked by default.
420 sigemptyset(&unblocked_sigs);
421 sigemptyset(&allowdebug_blocked_sigs);
422 sigaddset(&unblocked_sigs, SIGILL);
423 sigaddset(&unblocked_sigs, SIGSEGV);
424 sigaddset(&unblocked_sigs, SIGBUS);
425 sigaddset(&unblocked_sigs, SIGFPE);
426 #if defined(PPC64)
427 sigaddset(&unblocked_sigs, SIGTRAP);
428 #endif
429 sigaddset(&unblocked_sigs, SR_signum);
430
431 if (!ReduceSignalUsage) {
432 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
433 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
434 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
435 }
436 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
437 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
438 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
439 }
440 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
441 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
443 }
444 }
445 // Fill in signals that are blocked by all but the VM thread.
446 sigemptyset(&vm_sigs);
447 if (!ReduceSignalUsage) {
448 sigaddset(&vm_sigs, BREAK_SIGNAL);
449 }
450 debug_only(signal_sets_initialized = true);
451
452 }
453
454 // These are signals that are unblocked while a thread is running Java.
455 // (For some reason, they get blocked by default.)
456 sigset_t* os::Linux::unblocked_signals() {
457 assert(signal_sets_initialized, "Not initialized");
458 return &unblocked_sigs;
459 }
460
461 // These are the signals that are blocked while a (non-VM) thread is
462 // running Java. Only the VM thread handles these signals.
463 sigset_t* os::Linux::vm_signals() {
464 assert(signal_sets_initialized, "Not initialized");
465 return &vm_sigs;
466 }
467
468 // These are signals that are blocked during cond_wait to allow debugger in
469 sigset_t* os::Linux::allowdebug_blocked_signals() {
470 assert(signal_sets_initialized, "Not initialized");
471 return &allowdebug_blocked_sigs;
472 }
473
474 void os::Linux::hotspot_sigmask(Thread* thread) {
475
476 //Save caller's signal mask before setting VM signal mask
477 sigset_t caller_sigmask;
478 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
479
480 OSThread* osthread = thread->osthread();
481 osthread->set_caller_sigmask(caller_sigmask);
482
483 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
484
485 if (!ReduceSignalUsage) {
486 if (thread->is_VM_thread()) {
487 // Only the VM thread handles BREAK_SIGNAL ...
488 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
489 } else {
490 // ... all other threads block BREAK_SIGNAL
491 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
492 }
493 }
494 }
495
496 //////////////////////////////////////////////////////////////////////////////
497 // detecting pthread library
498
499 void os::Linux::libpthread_init() {
500 // Save glibc and pthread version strings.
501 #if !defined(_CS_GNU_LIBC_VERSION) || \
502 !defined(_CS_GNU_LIBPTHREAD_VERSION)
503 #error "glibc too old (< 2.3.2)"
504 #endif
505
506 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
507 assert(n > 0, "cannot retrieve glibc version");
508 char *str = (char *)malloc(n, mtInternal);
509 confstr(_CS_GNU_LIBC_VERSION, str, n);
510 os::Linux::set_glibc_version(str);
511
512 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
513 assert(n > 0, "cannot retrieve pthread version");
514 str = (char *)malloc(n, mtInternal);
515 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
516 os::Linux::set_libpthread_version(str);
517 }
518
519 /////////////////////////////////////////////////////////////////////////////
520 // thread stack expansion
521
522 // os::Linux::manually_expand_stack() takes care of expanding the thread
523 // stack. Note that this is normally not needed: pthread stacks allocate
524 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
525 // committed. Therefore it is not necessary to expand the stack manually.
526 //
527 // Manually expanding the stack was historically needed on LinuxThreads
528 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
529 // it is kept to deal with very rare corner cases:
530 //
531 // For one, user may run the VM on an own implementation of threads
532 // whose stacks are - like the old LinuxThreads - implemented using
533 // mmap(MAP_GROWSDOWN).
534 //
535 // Also, this coding may be needed if the VM is running on the primordial
536 // thread. Normally we avoid running on the primordial thread; however,
537 // user may still invoke the VM on the primordial thread.
538 //
539 // The following historical comment describes the details about running
540 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
541
542
543 // Force Linux kernel to expand current thread stack. If "bottom" is close
544 // to the stack guard, caller should block all signals.
545 //
546 // MAP_GROWSDOWN:
547 // A special mmap() flag that is used to implement thread stacks. It tells
548 // kernel that the memory region should extend downwards when needed. This
549 // allows early versions of LinuxThreads to only mmap the first few pages
550 // when creating a new thread. Linux kernel will automatically expand thread
551 // stack as needed (on page faults).
552 //
553 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
554 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
555 // region, it's hard to tell if the fault is due to a legitimate stack
556 // access or because of reading/writing non-exist memory (e.g. buffer
557 // overrun). As a rule, if the fault happens below current stack pointer,
558 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
559 // application (see Linux kernel fault.c).
560 //
561 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
562 // stack overflow detection.
563 //
564 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
565 // not use MAP_GROWSDOWN.
566 //
567 // To get around the problem and allow stack banging on Linux, we need to
568 // manually expand thread stack after receiving the SIGSEGV.
569 //
570 // There are two ways to expand thread stack to address "bottom", we used
571 // both of them in JVM before 1.5:
572 // 1. adjust stack pointer first so that it is below "bottom", and then
573 // touch "bottom"
574 // 2. mmap() the page in question
575 //
576 // Now alternate signal stack is gone, it's harder to use 2. For instance,
577 // if current sp is already near the lower end of page 101, and we need to
578 // call mmap() to map page 100, it is possible that part of the mmap() frame
579 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
580 // That will destroy the mmap() frame and cause VM to crash.
581 //
582 // The following code works by adjusting sp first, then accessing the "bottom"
583 // page to force a page fault. Linux kernel will then automatically expand the
584 // stack mapping.
585 //
586 // _expand_stack_to() assumes its frame size is less than page size, which
587 // should always be true if the function is not inlined.
588
589 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
590 #define NOINLINE
591 #else
592 #define NOINLINE __attribute__ ((noinline))
593 #endif
594
595 static void _expand_stack_to(address bottom) NOINLINE;
596
597 static void _expand_stack_to(address bottom) {
598 address sp;
599 size_t size;
600 volatile char *p;
601
602 // Adjust bottom to point to the largest address within the same page, it
603 // gives us a one-page buffer if alloca() allocates slightly more memory.
604 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
605 bottom += os::Linux::page_size() - 1;
606
607 // sp might be slightly above current stack pointer; if that's the case, we
608 // will alloca() a little more space than necessary, which is OK. Don't use
609 // os::current_stack_pointer(), as its result can be slightly below current
610 // stack pointer, causing us to not alloca enough to reach "bottom".
611 sp = (address)&sp;
612
613 if (sp > bottom) {
614 size = sp - bottom;
615 p = (volatile char *)alloca(size);
616 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
617 p[0] = '\0';
618 }
619 }
620
621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
622 assert(t!=NULL, "just checking");
623 assert(t->osthread()->expanding_stack(), "expand should be set");
624 assert(t->stack_base() != NULL, "stack_base was not initialized");
625
626 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
627 sigset_t mask_all, old_sigset;
628 sigfillset(&mask_all);
629 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
630 _expand_stack_to(addr);
631 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
632 return true;
633 }
634 return false;
635 }
636
637 //////////////////////////////////////////////////////////////////////////////
638 // create new thread
639
640 // Thread start routine for all newly created threads
641 static void *java_start(Thread *thread) {
642 // Try to randomize the cache line index of hot stack frames.
643 // This helps when threads of the same stack traces evict each other's
644 // cache lines. The threads can be either from the same JVM instance, or
645 // from different JVM instances. The benefit is especially true for
646 // processors with hyperthreading technology.
647 static int counter = 0;
648 int pid = os::current_process_id();
649 alloca(((pid ^ counter++) & 7) * 128);
650
651 ThreadLocalStorage::set_thread(thread);
652
653 OSThread* osthread = thread->osthread();
654 Monitor* sync = osthread->startThread_lock();
655
656 osthread->set_thread_id(os::current_thread_id());
657
658 if (UseNUMA) {
659 int lgrp_id = os::numa_get_group_id();
660 if (lgrp_id != -1) {
661 thread->set_lgrp_id(lgrp_id);
662 }
663 }
664 // initialize signal mask for this thread
665 os::Linux::hotspot_sigmask(thread);
666
667 // initialize floating point control register
668 os::Linux::init_thread_fpu_state();
669
670 // handshaking with parent thread
671 {
672 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
673
674 // notify parent thread
675 osthread->set_state(INITIALIZED);
676 sync->notify_all();
677
678 // wait until os::start_thread()
679 while (osthread->get_state() == INITIALIZED) {
680 sync->wait(Mutex::_no_safepoint_check_flag);
681 }
682 }
683
684 // call one more level start routine
685 thread->run();
686
687 return 0;
688 }
689
690 bool os::create_thread(Thread* thread, ThreadType thr_type,
691 size_t stack_size) {
692 assert(thread->osthread() == NULL, "caller responsible");
693
694 // Allocate the OSThread object
695 OSThread* osthread = new OSThread(NULL, NULL);
696 if (osthread == NULL) {
697 return false;
698 }
699
700 // set the correct thread state
701 osthread->set_thread_type(thr_type);
702
703 // Initial state is ALLOCATED but not INITIALIZED
704 osthread->set_state(ALLOCATED);
705
706 thread->set_osthread(osthread);
707
708 // init thread attributes
709 pthread_attr_t attr;
710 pthread_attr_init(&attr);
711 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
712
713 // stack size
714 if (os::Linux::supports_variable_stack_size()) {
715 // calculate stack size if it's not specified by caller
716 if (stack_size == 0) {
717 stack_size = os::Linux::default_stack_size(thr_type);
718
719 switch (thr_type) {
720 case os::java_thread:
721 // Java threads use ThreadStackSize which default value can be
722 // changed with the flag -Xss
723 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
724 stack_size = JavaThread::stack_size_at_create();
725 break;
726 case os::compiler_thread:
727 if (CompilerThreadStackSize > 0) {
728 stack_size = (size_t)(CompilerThreadStackSize * K);
729 break;
730 } // else fall through:
731 // use VMThreadStackSize if CompilerThreadStackSize is not defined
732 case os::vm_thread:
733 case os::pgc_thread:
734 case os::cgc_thread:
735 case os::watcher_thread:
736 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
737 break;
738 }
739 }
740
741 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
742 pthread_attr_setstacksize(&attr, stack_size);
743 } else {
744 // let pthread_create() pick the default value.
745 }
746
747 // glibc guard page
748 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
749
750 ThreadState state;
751
752 {
753 pthread_t tid;
754 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
755
756 pthread_attr_destroy(&attr);
757
758 if (ret != 0) {
759 if (PrintMiscellaneous && (Verbose || WizardMode)) {
760 perror("pthread_create()");
761 }
762 // Need to clean up stuff we've allocated so far
763 thread->set_osthread(NULL);
764 delete osthread;
765 return false;
766 }
767
768 // Store pthread info into the OSThread
769 osthread->set_pthread_id(tid);
770
771 // Wait until child thread is either initialized or aborted
772 {
773 Monitor* sync_with_child = osthread->startThread_lock();
774 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
775 while ((state = osthread->get_state()) == ALLOCATED) {
776 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
777 }
778 }
779 }
780
781 // Aborted due to thread limit being reached
782 if (state == ZOMBIE) {
783 thread->set_osthread(NULL);
784 delete osthread;
785 return false;
786 }
787
788 // The thread is returned suspended (in state INITIALIZED),
789 // and is started higher up in the call chain
790 assert(state == INITIALIZED, "race condition");
791 return true;
792 }
793
794 /////////////////////////////////////////////////////////////////////////////
795 // attach existing thread
796
797 // bootstrap the main thread
798 bool os::create_main_thread(JavaThread* thread) {
799 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
800 return create_attached_thread(thread);
801 }
802
803 bool os::create_attached_thread(JavaThread* thread) {
804 #ifdef ASSERT
805 thread->verify_not_published();
806 #endif
807
808 // Allocate the OSThread object
809 OSThread* osthread = new OSThread(NULL, NULL);
810
811 if (osthread == NULL) {
812 return false;
813 }
814
815 // Store pthread info into the OSThread
816 osthread->set_thread_id(os::Linux::gettid());
817 osthread->set_pthread_id(::pthread_self());
818
819 // initialize floating point control register
820 os::Linux::init_thread_fpu_state();
821
822 // Initial thread state is RUNNABLE
823 osthread->set_state(RUNNABLE);
824
825 thread->set_osthread(osthread);
826
827 if (UseNUMA) {
828 int lgrp_id = os::numa_get_group_id();
829 if (lgrp_id != -1) {
830 thread->set_lgrp_id(lgrp_id);
831 }
832 }
833
834 if (os::Linux::is_initial_thread()) {
835 // If current thread is initial thread, its stack is mapped on demand,
836 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
837 // the entire stack region to avoid SEGV in stack banging.
838 // It is also useful to get around the heap-stack-gap problem on SuSE
839 // kernel (see 4821821 for details). We first expand stack to the top
840 // of yellow zone, then enable stack yellow zone (order is significant,
841 // enabling yellow zone first will crash JVM on SuSE Linux), so there
842 // is no gap between the last two virtual memory regions.
843
844 JavaThread *jt = (JavaThread *)thread;
845 address addr = jt->stack_yellow_zone_base();
846 assert(addr != NULL, "initialization problem?");
847 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
848
849 osthread->set_expanding_stack();
850 os::Linux::manually_expand_stack(jt, addr);
851 osthread->clear_expanding_stack();
852 }
853
854 // initialize signal mask for this thread
855 // and save the caller's signal mask
856 os::Linux::hotspot_sigmask(thread);
857
858 return true;
859 }
860
861 void os::pd_start_thread(Thread* thread) {
862 OSThread * osthread = thread->osthread();
863 assert(osthread->get_state() != INITIALIZED, "just checking");
864 Monitor* sync_with_child = osthread->startThread_lock();
865 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
866 sync_with_child->notify();
867 }
868
869 // Free Linux resources related to the OSThread
870 void os::free_thread(OSThread* osthread) {
871 assert(osthread != NULL, "osthread not set");
872
873 if (Thread::current()->osthread() == osthread) {
874 // Restore caller's signal mask
875 sigset_t sigmask = osthread->caller_sigmask();
876 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
877 }
878
879 delete osthread;
880 }
881
882 //////////////////////////////////////////////////////////////////////////////
883 // thread local storage
884
885 // Restore the thread pointer if the destructor is called. This is in case
886 // someone from JNI code sets up a destructor with pthread_key_create to run
887 // detachCurrentThread on thread death. Unless we restore the thread pointer we
888 // will hang or crash. When detachCurrentThread is called the key will be set
889 // to null and we will not be called again. If detachCurrentThread is never
890 // called we could loop forever depending on the pthread implementation.
891 static void restore_thread_pointer(void* p) {
892 Thread* thread = (Thread*) p;
893 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
894 }
895
896 int os::allocate_thread_local_storage() {
897 pthread_key_t key;
898 int rslt = pthread_key_create(&key, restore_thread_pointer);
899 assert(rslt == 0, "cannot allocate thread local storage");
900 return (int)key;
901 }
902
903 // Note: This is currently not used by VM, as we don't destroy TLS key
904 // on VM exit.
905 void os::free_thread_local_storage(int index) {
906 int rslt = pthread_key_delete((pthread_key_t)index);
907 assert(rslt == 0, "invalid index");
908 }
909
910 void os::thread_local_storage_at_put(int index, void* value) {
911 int rslt = pthread_setspecific((pthread_key_t)index, value);
912 assert(rslt == 0, "pthread_setspecific failed");
913 }
914
915 extern "C" Thread* get_thread() {
916 return ThreadLocalStorage::thread();
917 }
918
919 //////////////////////////////////////////////////////////////////////////////
920 // initial thread
921
922 // Check if current thread is the initial thread, similar to Solaris thr_main.
923 bool os::Linux::is_initial_thread(void) {
924 char dummy;
925 // If called before init complete, thread stack bottom will be null.
926 // Can be called if fatal error occurs before initialization.
927 if (initial_thread_stack_bottom() == NULL) return false;
928 assert(initial_thread_stack_bottom() != NULL &&
929 initial_thread_stack_size() != 0,
930 "os::init did not locate initial thread's stack region");
931 if ((address)&dummy >= initial_thread_stack_bottom() &&
932 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
933 return true;
934 } else {
935 return false;
936 }
937 }
938
939 // Find the virtual memory area that contains addr
940 static bool find_vma(address addr, address* vma_low, address* vma_high) {
941 FILE *fp = fopen("/proc/self/maps", "r");
942 if (fp) {
943 address low, high;
944 while (!feof(fp)) {
945 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
946 if (low <= addr && addr < high) {
947 if (vma_low) *vma_low = low;
948 if (vma_high) *vma_high = high;
949 fclose(fp);
950 return true;
951 }
952 }
953 for (;;) {
954 int ch = fgetc(fp);
955 if (ch == EOF || ch == (int)'\n') break;
956 }
957 }
958 fclose(fp);
959 }
960 return false;
961 }
962
963 // Locate initial thread stack. This special handling of initial thread stack
964 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
965 // bogus value for initial thread.
966 void os::Linux::capture_initial_stack(size_t max_size) {
967 // stack size is the easy part, get it from RLIMIT_STACK
968 size_t stack_size;
969 struct rlimit rlim;
970 getrlimit(RLIMIT_STACK, &rlim);
971 stack_size = rlim.rlim_cur;
972
973 // 6308388: a bug in ld.so will relocate its own .data section to the
974 // lower end of primordial stack; reduce ulimit -s value a little bit
975 // so we won't install guard page on ld.so's data section.
976 stack_size -= 2 * page_size();
977
978 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
979 // 7.1, in both cases we will get 2G in return value.
980 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
981 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
982 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
983 // in case other parts in glibc still assumes 2M max stack size.
984 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
985 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
986 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
987 stack_size = 2 * K * K IA64_ONLY(*2);
988 }
989 // Try to figure out where the stack base (top) is. This is harder.
990 //
991 // When an application is started, glibc saves the initial stack pointer in
992 // a global variable "__libc_stack_end", which is then used by system
993 // libraries. __libc_stack_end should be pretty close to stack top. The
994 // variable is available since the very early days. However, because it is
995 // a private interface, it could disappear in the future.
996 //
997 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
998 // to __libc_stack_end, it is very close to stack top, but isn't the real
999 // stack top. Note that /proc may not exist if VM is running as a chroot
1000 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1001 // /proc/<pid>/stat could change in the future (though unlikely).
1002 //
1003 // We try __libc_stack_end first. If that doesn't work, look for
1004 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1005 // as a hint, which should work well in most cases.
1006
1007 uintptr_t stack_start;
1008
1009 // try __libc_stack_end first
1010 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1011 if (p && *p) {
1012 stack_start = *p;
1013 } else {
1014 // see if we can get the start_stack field from /proc/self/stat
1015 FILE *fp;
1016 int pid;
1017 char state;
1018 int ppid;
1019 int pgrp;
1020 int session;
1021 int nr;
1022 int tpgrp;
1023 unsigned long flags;
1024 unsigned long minflt;
1025 unsigned long cminflt;
1026 unsigned long majflt;
1027 unsigned long cmajflt;
1028 unsigned long utime;
1029 unsigned long stime;
1030 long cutime;
1031 long cstime;
1032 long prio;
1033 long nice;
1034 long junk;
1035 long it_real;
1036 uintptr_t start;
1037 uintptr_t vsize;
1038 intptr_t rss;
1039 uintptr_t rsslim;
1040 uintptr_t scodes;
1041 uintptr_t ecode;
1042 int i;
1043
1044 // Figure what the primordial thread stack base is. Code is inspired
1045 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1046 // followed by command name surrounded by parentheses, state, etc.
1047 char stat[2048];
1048 int statlen;
1049
1050 fp = fopen("/proc/self/stat", "r");
1051 if (fp) {
1052 statlen = fread(stat, 1, 2047, fp);
1053 stat[statlen] = '\0';
1054 fclose(fp);
1055
1056 // Skip pid and the command string. Note that we could be dealing with
1057 // weird command names, e.g. user could decide to rename java launcher
1058 // to "java 1.4.2 :)", then the stat file would look like
1059 // 1234 (java 1.4.2 :)) R ... ...
1060 // We don't really need to know the command string, just find the last
1061 // occurrence of ")" and then start parsing from there. See bug 4726580.
1062 char * s = strrchr(stat, ')');
1063
1064 i = 0;
1065 if (s) {
1066 // Skip blank chars
1067 do { s++; } while (s && isspace(*s));
1068
1069 #define _UFM UINTX_FORMAT
1070 #define _DFM INTX_FORMAT
1071
1072 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1073 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1074 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1075 &state, // 3 %c
1076 &ppid, // 4 %d
1077 &pgrp, // 5 %d
1078 &session, // 6 %d
1079 &nr, // 7 %d
1080 &tpgrp, // 8 %d
1081 &flags, // 9 %lu
1082 &minflt, // 10 %lu
1083 &cminflt, // 11 %lu
1084 &majflt, // 12 %lu
1085 &cmajflt, // 13 %lu
1086 &utime, // 14 %lu
1087 &stime, // 15 %lu
1088 &cutime, // 16 %ld
1089 &cstime, // 17 %ld
1090 &prio, // 18 %ld
1091 &nice, // 19 %ld
1092 &junk, // 20 %ld
1093 &it_real, // 21 %ld
1094 &start, // 22 UINTX_FORMAT
1095 &vsize, // 23 UINTX_FORMAT
1096 &rss, // 24 INTX_FORMAT
1097 &rsslim, // 25 UINTX_FORMAT
1098 &scodes, // 26 UINTX_FORMAT
1099 &ecode, // 27 UINTX_FORMAT
1100 &stack_start); // 28 UINTX_FORMAT
1101 }
1102
1103 #undef _UFM
1104 #undef _DFM
1105
1106 if (i != 28 - 2) {
1107 assert(false, "Bad conversion from /proc/self/stat");
1108 // product mode - assume we are the initial thread, good luck in the
1109 // embedded case.
1110 warning("Can't detect initial thread stack location - bad conversion");
1111 stack_start = (uintptr_t) &rlim;
1112 }
1113 } else {
1114 // For some reason we can't open /proc/self/stat (for example, running on
1115 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1116 // most cases, so don't abort:
1117 warning("Can't detect initial thread stack location - no /proc/self/stat");
1118 stack_start = (uintptr_t) &rlim;
1119 }
1120 }
1121
1122 // Now we have a pointer (stack_start) very close to the stack top, the
1123 // next thing to do is to figure out the exact location of stack top. We
1124 // can find out the virtual memory area that contains stack_start by
1125 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1126 // and its upper limit is the real stack top. (again, this would fail if
1127 // running inside chroot, because /proc may not exist.)
1128
1129 uintptr_t stack_top;
1130 address low, high;
1131 if (find_vma((address)stack_start, &low, &high)) {
1132 // success, "high" is the true stack top. (ignore "low", because initial
1133 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1134 stack_top = (uintptr_t)high;
1135 } else {
1136 // failed, likely because /proc/self/maps does not exist
1137 warning("Can't detect initial thread stack location - find_vma failed");
1138 // best effort: stack_start is normally within a few pages below the real
1139 // stack top, use it as stack top, and reduce stack size so we won't put
1140 // guard page outside stack.
1141 stack_top = stack_start;
1142 stack_size -= 16 * page_size();
1143 }
1144
1145 // stack_top could be partially down the page so align it
1146 stack_top = align_size_up(stack_top, page_size());
1147
1148 if (max_size && stack_size > max_size) {
1149 _initial_thread_stack_size = max_size;
1150 } else {
1151 _initial_thread_stack_size = stack_size;
1152 }
1153
1154 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1155 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156 }
1157
1158 ////////////////////////////////////////////////////////////////////////////////
1159 // time support
1160
1161 // Time since start-up in seconds to a fine granularity.
1162 // Used by VMSelfDestructTimer and the MemProfiler.
1163 double os::elapsedTime() {
1164
1165 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1166 }
1167
1168 jlong os::elapsed_counter() {
1169 return javaTimeNanos() - initial_time_count;
1170 }
1171
1172 jlong os::elapsed_frequency() {
1173 return NANOSECS_PER_SEC; // nanosecond resolution
1174 }
1175
1176 bool os::supports_vtime() { return true; }
1177 bool os::enable_vtime() { return false; }
1178 bool os::vtime_enabled() { return false; }
1179
1180 double os::elapsedVTime() {
1181 struct rusage usage;
1182 int retval = getrusage(RUSAGE_THREAD, &usage);
1183 if (retval == 0) {
1184 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1185 } else {
1186 // better than nothing, but not much
1187 return elapsedTime();
1188 }
1189 }
1190
1191 jlong os::javaTimeMillis() {
1192 timeval time;
1193 int status = gettimeofday(&time, NULL);
1194 assert(status != -1, "linux error");
1195 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1196 }
1197
1198 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1199 timeval time;
1200 int status = gettimeofday(&time, NULL);
1201 assert(status != -1, "linux error");
1202 seconds = jlong(time.tv_sec);
1203 nanos = jlong(time.tv_usec) * 1000;
1204 }
1205
1206
1207 #ifndef CLOCK_MONOTONIC
1208 #define CLOCK_MONOTONIC (1)
1209 #endif
1210
1211 void os::Linux::clock_init() {
1212 // we do dlopen's in this particular order due to bug in linux
1213 // dynamical loader (see 6348968) leading to crash on exit
1214 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1215 if (handle == NULL) {
1216 handle = dlopen("librt.so", RTLD_LAZY);
1217 }
1218
1219 if (handle) {
1220 int (*clock_getres_func)(clockid_t, struct timespec*) =
1221 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1222 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1223 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1224 if (clock_getres_func && clock_gettime_func) {
1225 // See if monotonic clock is supported by the kernel. Note that some
1226 // early implementations simply return kernel jiffies (updated every
1227 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1228 // for nano time (though the monotonic property is still nice to have).
1229 // It's fixed in newer kernels, however clock_getres() still returns
1230 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1231 // resolution for now. Hopefully as people move to new kernels, this
1232 // won't be a problem.
1233 struct timespec res;
1234 struct timespec tp;
1235 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1236 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1237 // yes, monotonic clock is supported
1238 _clock_gettime = clock_gettime_func;
1239 return;
1240 } else {
1241 // close librt if there is no monotonic clock
1242 dlclose(handle);
1243 }
1244 }
1245 }
1246 warning("No monotonic clock was available - timed services may " \
1247 "be adversely affected if the time-of-day clock changes");
1248 }
1249
1250 #ifndef SYS_clock_getres
1251 #if defined(IA32) || defined(AMD64)
1252 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1253 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1254 #else
1255 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1256 #define sys_clock_getres(x,y) -1
1257 #endif
1258 #else
1259 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1260 #endif
1261
1262 void os::Linux::fast_thread_clock_init() {
1263 if (!UseLinuxPosixThreadCPUClocks) {
1264 return;
1265 }
1266 clockid_t clockid;
1267 struct timespec tp;
1268 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1269 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1270
1271 // Switch to using fast clocks for thread cpu time if
1272 // the sys_clock_getres() returns 0 error code.
1273 // Note, that some kernels may support the current thread
1274 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1275 // returned by the pthread_getcpuclockid().
1276 // If the fast Posix clocks are supported then the sys_clock_getres()
1277 // must return at least tp.tv_sec == 0 which means a resolution
1278 // better than 1 sec. This is extra check for reliability.
1279
1280 if (pthread_getcpuclockid_func &&
1281 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1282 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1283 _supports_fast_thread_cpu_time = true;
1284 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1285 }
1286 }
1287
1288 jlong os::javaTimeNanos() {
1289 if (os::supports_monotonic_clock()) {
1290 struct timespec tp;
1291 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1292 assert(status == 0, "gettime error");
1293 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1294 return result;
1295 } else {
1296 timeval time;
1297 int status = gettimeofday(&time, NULL);
1298 assert(status != -1, "linux error");
1299 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1300 return 1000 * usecs;
1301 }
1302 }
1303
1304 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1305 if (os::supports_monotonic_clock()) {
1306 info_ptr->max_value = ALL_64_BITS;
1307
1308 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1309 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1310 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1311 } else {
1312 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1313 info_ptr->max_value = ALL_64_BITS;
1314
1315 // gettimeofday is a real time clock so it skips
1316 info_ptr->may_skip_backward = true;
1317 info_ptr->may_skip_forward = true;
1318 }
1319
1320 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1321 }
1322
1323 // Return the real, user, and system times in seconds from an
1324 // arbitrary fixed point in the past.
1325 bool os::getTimesSecs(double* process_real_time,
1326 double* process_user_time,
1327 double* process_system_time) {
1328 struct tms ticks;
1329 clock_t real_ticks = times(&ticks);
1330
1331 if (real_ticks == (clock_t) (-1)) {
1332 return false;
1333 } else {
1334 double ticks_per_second = (double) clock_tics_per_sec;
1335 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1336 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1337 *process_real_time = ((double) real_ticks) / ticks_per_second;
1338
1339 return true;
1340 }
1341 }
1342
1343
1344 char * os::local_time_string(char *buf, size_t buflen) {
1345 struct tm t;
1346 time_t long_time;
1347 time(&long_time);
1348 localtime_r(&long_time, &t);
1349 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1350 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1351 t.tm_hour, t.tm_min, t.tm_sec);
1352 return buf;
1353 }
1354
1355 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1356 return localtime_r(clock, res);
1357 }
1358
1359 ////////////////////////////////////////////////////////////////////////////////
1360 // runtime exit support
1361
1362 // Note: os::shutdown() might be called very early during initialization, or
1363 // called from signal handler. Before adding something to os::shutdown(), make
1364 // sure it is async-safe and can handle partially initialized VM.
1365 void os::shutdown() {
1366
1367 // allow PerfMemory to attempt cleanup of any persistent resources
1368 perfMemory_exit();
1369
1370 // needs to remove object in file system
1371 AttachListener::abort();
1372
1373 // flush buffered output, finish log files
1374 ostream_abort();
1375
1376 // Check for abort hook
1377 abort_hook_t abort_hook = Arguments::abort_hook();
1378 if (abort_hook != NULL) {
1379 abort_hook();
1380 }
1381
1382 }
1383
1384 // Note: os::abort() might be called very early during initialization, or
1385 // called from signal handler. Before adding something to os::abort(), make
1386 // sure it is async-safe and can handle partially initialized VM.
1387 void os::abort(bool dump_core, void* siginfo, void* context) {
1388 os::shutdown();
1389 if (dump_core) {
1390 #ifndef PRODUCT
1391 fdStream out(defaultStream::output_fd());
1392 out.print_raw("Current thread is ");
1393 char buf[16];
1394 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1395 out.print_raw_cr(buf);
1396 out.print_raw_cr("Dumping core ...");
1397 #endif
1398 ::abort(); // dump core
1399 }
1400
1401 ::exit(1);
1402 }
1403
1404 // Die immediately, no exit hook, no abort hook, no cleanup.
1405 void os::die() {
1406 ::abort();
1407 }
1408
1409
1410 // This method is a copy of JDK's sysGetLastErrorString
1411 // from src/solaris/hpi/src/system_md.c
1412
1413 size_t os::lasterror(char *buf, size_t len) {
1414 if (errno == 0) return 0;
1415
1416 const char *s = ::strerror(errno);
1417 size_t n = ::strlen(s);
1418 if (n >= len) {
1419 n = len - 1;
1420 }
1421 ::strncpy(buf, s, n);
1422 buf[n] = '\0';
1423 return n;
1424 }
1425
1426 // thread_id is kernel thread id (similar to Solaris LWP id)
1427 intx os::current_thread_id() { return os::Linux::gettid(); }
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) {
2067 break; // done
2068 }
2069 // If file prints, we found it.
2070 if (_print_ascii_file(file, st)) {
2071 return;
2072 }
2073 }
2074
2075 if (file_exists("/etc/debian_version")) {
2076 st->print("Debian ");
2077 _print_ascii_file("/etc/debian_version", st);
2078 } else {
2079 st->print("Linux");
2080 }
2081 st->cr();
2082 }
2083
2084 static void parse_os_info(char* distro, size_t length, const char* file) {
2085 FILE* fp = fopen(file, "r");
2086 if (fp != NULL) {
2087 char buf[256];
2088 // get last line of the file.
2089 while (fgets(buf, sizeof(buf), fp)) { }
2090 // Edit out extra stuff in expected ubuntu format
2091 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2092 char* ptr = strstr(buf, "\""); // the name is in quotes
2093 if (ptr != NULL) {
2094 ptr++; // go beyond first quote
2095 char* nl = strchr(ptr, '\"');
2096 if (nl != NULL) *nl = '\0';
2097 strncpy(distro, ptr, length);
2098 } else {
2099 ptr = strstr(buf, "=");
2100 ptr++; // go beyond equals then
2101 char* nl = strchr(ptr, '\n');
2102 if (nl != NULL) *nl = '\0';
2103 strncpy(distro, ptr, length);
2104 }
2105 } else {
2106 // if not in expected Ubuntu format, print out whole line minus \n
2107 char* nl = strchr(buf, '\n');
2108 if (nl != NULL) *nl = '\0';
2109 strncpy(distro, buf, length);
2110 }
2111 // close distro file
2112 fclose(fp);
2113 }
2114 }
2115
2116 void os::get_summary_os_info(char* buf, size_t buflen) {
2117 for (int i = 0;; i++) {
2118 const char* file = distro_files[i];
2119 if (file == NULL) {
2120 break; // ran out of distro_files
2121 }
2122 if (file_exists(file)) {
2123 parse_os_info(buf, buflen, file);
2124 return;
2125 }
2126 }
2127 // special case for debian
2128 if (file_exists("/etc/debian_version")) {
2129 strncpy(buf, "Debian ", buflen);
2130 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2131 } else {
2132 strncpy(buf, "Linux", buflen);
2133 }
2134 }
2135
2136 void os::Linux::print_libversion_info(outputStream* st) {
2137 // libc, pthread
2138 st->print("libc:");
2139 st->print("%s ", os::Linux::glibc_version());
2140 st->print("%s ", os::Linux::libpthread_version());
2141 st->cr();
2142 }
2143
2144 void os::Linux::print_full_memory_info(outputStream* st) {
2145 st->print("\n/proc/meminfo:\n");
2146 _print_ascii_file("/proc/meminfo", st);
2147 st->cr();
2148 }
2149
2150 void os::print_memory_info(outputStream* st) {
2151
2152 st->print("Memory:");
2153 st->print(" %dk page", os::vm_page_size()>>10);
2154
2155 // values in struct sysinfo are "unsigned long"
2156 struct sysinfo si;
2157 sysinfo(&si);
2158
2159 st->print(", physical " UINT64_FORMAT "k",
2160 os::physical_memory() >> 10);
2161 st->print("(" UINT64_FORMAT "k free)",
2162 os::available_memory() >> 10);
2163 st->print(", swap " UINT64_FORMAT "k",
2164 ((jlong)si.totalswap * si.mem_unit) >> 10);
2165 st->print("(" UINT64_FORMAT "k free)",
2166 ((jlong)si.freeswap * si.mem_unit) >> 10);
2167 st->cr();
2168 }
2169
2170 // Print the first "model name" line and the first "flags" line
2171 // that we find and nothing more. We assume "model name" comes
2172 // before "flags" so if we find a second "model name", then the
2173 // "flags" field is considered missing.
2174 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2175 #if defined(IA32) || defined(AMD64)
2176 // Other platforms have less repetitive cpuinfo files
2177 FILE *fp = fopen("/proc/cpuinfo", "r");
2178 if (fp) {
2179 while (!feof(fp)) {
2180 if (fgets(buf, buflen, fp)) {
2181 // Assume model name comes before flags
2182 bool model_name_printed = false;
2183 if (strstr(buf, "model name") != NULL) {
2184 if (!model_name_printed) {
2185 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2186 st->print_raw(buf);
2187 model_name_printed = true;
2188 } else {
2189 // model name printed but not flags? Odd, just return
2190 fclose(fp);
2191 return true;
2192 }
2193 }
2194 // print the flags line too
2195 if (strstr(buf, "flags") != NULL) {
2196 st->print_raw(buf);
2197 fclose(fp);
2198 return true;
2199 }
2200 }
2201 }
2202 fclose(fp);
2203 }
2204 #endif // x86 platforms
2205 return false;
2206 }
2207
2208 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2209 // Only print the model name if the platform provides this as a summary
2210 if (!print_model_name_and_flags(st, buf, buflen)) {
2211 st->print("\n/proc/cpuinfo:\n");
2212 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2213 st->print_cr(" <Not Available>");
2214 }
2215 }
2216 }
2217
2218 const char* search_string = IA32_ONLY("model name") AMD64_ONLY("model name")
2219 IA64_ONLY("") SPARC_ONLY("cpu")
2220 ARM32_ONLY("Processor") PPC_ONLY("Processor") AARCH64_ONLY("Processor");
2221
2222 // Parses the cpuinfo file for string representing the model name.
2223 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2224 FILE* fp = fopen("/proc/cpuinfo", "r");
2225 if (fp != NULL) {
2226 while (!feof(fp)) {
2227 char buf[256];
2228 if (fgets(buf, sizeof(buf), fp)) {
2229 char* start = strstr(buf, search_string);
2230 if (start != NULL) {
2231 char *ptr = start + strlen(search_string);
2232 char *end = buf + strlen(buf);
2233 while (ptr != end) {
2234 // skip whitespace and colon for the rest of the name.
2235 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2236 break;
2237 }
2238 ptr++;
2239 }
2240 if (ptr != end) {
2241 // reasonable string, get rid of newline and keep the rest
2242 char* nl = strchr(buf, '\n');
2243 if (nl != NULL) *nl = '\0';
2244 strncpy(cpuinfo, ptr, length);
2245 fclose(fp);
2246 return;
2247 }
2248 }
2249 }
2250 }
2251 fclose(fp);
2252 }
2253 // cpuinfo not found or parsing failed, just print generic string. The entire
2254 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2255 strncpy(cpuinfo, IA32_ONLY("x86_32") AMD64_ONLY("x86_32")
2256 IA64_ONLY("IA64") SPARC_ONLY("sparcv9")
2257 ARM32_ONLY("ARM") PPC_ONLY("PPC64") AARCH64_ONLY("AArch64"), length);
2258 }
2259
2260 void os::print_siginfo(outputStream* st, void* siginfo) {
2261 const siginfo_t* si = (const siginfo_t*)siginfo;
2262
2263 os::Posix::print_siginfo_brief(st, si);
2264 #if INCLUDE_CDS
2265 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2266 UseSharedSpaces) {
2267 FileMapInfo* mapinfo = FileMapInfo::current_info();
2268 if (mapinfo->is_in_shared_space(si->si_addr)) {
2269 st->print("\n\nError accessing class data sharing archive." \
2270 " Mapped file inaccessible during execution, " \
2271 " possible disk/network problem.");
2272 }
2273 }
2274 #endif
2275 st->cr();
2276 }
2277
2278
2279 static void print_signal_handler(outputStream* st, int sig,
2280 char* buf, size_t buflen);
2281
2282 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2283 st->print_cr("Signal Handlers:");
2284 print_signal_handler(st, SIGSEGV, buf, buflen);
2285 print_signal_handler(st, SIGBUS , buf, buflen);
2286 print_signal_handler(st, SIGFPE , buf, buflen);
2287 print_signal_handler(st, SIGPIPE, buf, buflen);
2288 print_signal_handler(st, SIGXFSZ, buf, buflen);
2289 print_signal_handler(st, SIGILL , buf, buflen);
2290 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2291 print_signal_handler(st, SR_signum, buf, buflen);
2292 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2293 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2294 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2295 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2296 #if defined(PPC64)
2297 print_signal_handler(st, SIGTRAP, buf, buflen);
2298 #endif
2299 }
2300
2301 static char saved_jvm_path[MAXPATHLEN] = {0};
2302
2303 // Find the full path to the current module, libjvm.so
2304 void os::jvm_path(char *buf, jint buflen) {
2305 // Error checking.
2306 if (buflen < MAXPATHLEN) {
2307 assert(false, "must use a large-enough buffer");
2308 buf[0] = '\0';
2309 return;
2310 }
2311 // Lazy resolve the path to current module.
2312 if (saved_jvm_path[0] != 0) {
2313 strcpy(buf, saved_jvm_path);
2314 return;
2315 }
2316
2317 char dli_fname[MAXPATHLEN];
2318 bool ret = dll_address_to_library_name(
2319 CAST_FROM_FN_PTR(address, os::jvm_path),
2320 dli_fname, sizeof(dli_fname), NULL);
2321 assert(ret, "cannot locate libjvm");
2322 char *rp = NULL;
2323 if (ret && dli_fname[0] != '\0') {
2324 rp = realpath(dli_fname, buf);
2325 }
2326 if (rp == NULL) {
2327 return;
2328 }
2329
2330 if (Arguments::sun_java_launcher_is_altjvm()) {
2331 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2332 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2333 // If "/jre/lib/" appears at the right place in the string, then
2334 // assume we are installed in a JDK and we're done. Otherwise, check
2335 // for a JAVA_HOME environment variable and fix up the path so it
2336 // looks like libjvm.so is installed there (append a fake suffix
2337 // hotspot/libjvm.so).
2338 const char *p = buf + strlen(buf) - 1;
2339 for (int count = 0; p > buf && count < 5; ++count) {
2340 for (--p; p > buf && *p != '/'; --p)
2341 /* empty */ ;
2342 }
2343
2344 if (strncmp(p, "/jre/lib/", 9) != 0) {
2345 // Look for JAVA_HOME in the environment.
2346 char* java_home_var = ::getenv("JAVA_HOME");
2347 if (java_home_var != NULL && java_home_var[0] != 0) {
2348 char* jrelib_p;
2349 int len;
2350
2351 // Check the current module name "libjvm.so".
2352 p = strrchr(buf, '/');
2353 if (p == NULL) {
2354 return;
2355 }
2356 assert(strstr(p, "/libjvm") == p, "invalid library name");
2357
2358 rp = realpath(java_home_var, buf);
2359 if (rp == NULL) {
2360 return;
2361 }
2362
2363 // determine if this is a legacy image or modules image
2364 // modules image doesn't have "jre" subdirectory
2365 len = strlen(buf);
2366 assert(len < buflen, "Ran out of buffer room");
2367 jrelib_p = buf + len;
2368 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2369 if (0 != access(buf, F_OK)) {
2370 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2371 }
2372
2373 if (0 == access(buf, F_OK)) {
2374 // Use current module name "libjvm.so"
2375 len = strlen(buf);
2376 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2377 } else {
2378 // Go back to path of .so
2379 rp = realpath(dli_fname, buf);
2380 if (rp == NULL) {
2381 return;
2382 }
2383 }
2384 }
2385 }
2386 }
2387
2388 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2389 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2390 }
2391
2392 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2393 // no prefix required, not even "_"
2394 }
2395
2396 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2397 // no suffix required
2398 }
2399
2400 ////////////////////////////////////////////////////////////////////////////////
2401 // sun.misc.Signal support
2402
2403 static volatile jint sigint_count = 0;
2404
2405 static void UserHandler(int sig, void *siginfo, void *context) {
2406 // 4511530 - sem_post is serialized and handled by the manager thread. When
2407 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2408 // don't want to flood the manager thread with sem_post requests.
2409 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2410 return;
2411 }
2412
2413 // Ctrl-C is pressed during error reporting, likely because the error
2414 // handler fails to abort. Let VM die immediately.
2415 if (sig == SIGINT && is_error_reported()) {
2416 os::die();
2417 }
2418
2419 os::signal_notify(sig);
2420 }
2421
2422 void* os::user_handler() {
2423 return CAST_FROM_FN_PTR(void*, UserHandler);
2424 }
2425
2426 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2427 struct timespec ts;
2428 // Semaphore's are always associated with CLOCK_REALTIME
2429 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2430 // see unpackTime for discussion on overflow checking
2431 if (sec >= MAX_SECS) {
2432 ts.tv_sec += MAX_SECS;
2433 ts.tv_nsec = 0;
2434 } else {
2435 ts.tv_sec += sec;
2436 ts.tv_nsec += nsec;
2437 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2438 ts.tv_nsec -= NANOSECS_PER_SEC;
2439 ++ts.tv_sec; // note: this must be <= max_secs
2440 }
2441 }
2442
2443 return ts;
2444 }
2445
2446 extern "C" {
2447 typedef void (*sa_handler_t)(int);
2448 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2449 }
2450
2451 void* os::signal(int signal_number, void* handler) {
2452 struct sigaction sigAct, oldSigAct;
2453
2454 sigfillset(&(sigAct.sa_mask));
2455 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2456 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2457
2458 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2459 // -1 means registration failed
2460 return (void *)-1;
2461 }
2462
2463 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2464 }
2465
2466 void os::signal_raise(int signal_number) {
2467 ::raise(signal_number);
2468 }
2469
2470 // The following code is moved from os.cpp for making this
2471 // code platform specific, which it is by its very nature.
2472
2473 // Will be modified when max signal is changed to be dynamic
2474 int os::sigexitnum_pd() {
2475 return NSIG;
2476 }
2477
2478 // a counter for each possible signal value
2479 static volatile jint pending_signals[NSIG+1] = { 0 };
2480
2481 // Linux(POSIX) specific hand shaking semaphore.
2482 static sem_t sig_sem;
2483 static PosixSemaphore sr_semaphore;
2484
2485 void os::signal_init_pd() {
2486 // Initialize signal structures
2487 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2488
2489 // Initialize signal semaphore
2490 ::sem_init(&sig_sem, 0, 0);
2491 }
2492
2493 void os::signal_notify(int sig) {
2494 Atomic::inc(&pending_signals[sig]);
2495 ::sem_post(&sig_sem);
2496 }
2497
2498 static int check_pending_signals(bool wait) {
2499 Atomic::store(0, &sigint_count);
2500 for (;;) {
2501 for (int i = 0; i < NSIG + 1; i++) {
2502 jint n = pending_signals[i];
2503 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2504 return i;
2505 }
2506 }
2507 if (!wait) {
2508 return -1;
2509 }
2510 JavaThread *thread = JavaThread::current();
2511 ThreadBlockInVM tbivm(thread);
2512
2513 bool threadIsSuspended;
2514 do {
2515 thread->set_suspend_equivalent();
2516 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2517 ::sem_wait(&sig_sem);
2518
2519 // were we externally suspended while we were waiting?
2520 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2521 if (threadIsSuspended) {
2522 // The semaphore has been incremented, but while we were waiting
2523 // another thread suspended us. We don't want to continue running
2524 // while suspended because that would surprise the thread that
2525 // suspended us.
2526 ::sem_post(&sig_sem);
2527
2528 thread->java_suspend_self();
2529 }
2530 } while (threadIsSuspended);
2531 }
2532 }
2533
2534 int os::signal_lookup() {
2535 return check_pending_signals(false);
2536 }
2537
2538 int os::signal_wait() {
2539 return check_pending_signals(true);
2540 }
2541
2542 ////////////////////////////////////////////////////////////////////////////////
2543 // Virtual Memory
2544
2545 int os::vm_page_size() {
2546 // Seems redundant as all get out
2547 assert(os::Linux::page_size() != -1, "must call os::init");
2548 return os::Linux::page_size();
2549 }
2550
2551 // Solaris allocates memory by pages.
2552 int os::vm_allocation_granularity() {
2553 assert(os::Linux::page_size() != -1, "must call os::init");
2554 return os::Linux::page_size();
2555 }
2556
2557 // Rationale behind this function:
2558 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2559 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2560 // samples for JITted code. Here we create private executable mapping over the code cache
2561 // and then we can use standard (well, almost, as mapping can change) way to provide
2562 // info for the reporting script by storing timestamp and location of symbol
2563 void linux_wrap_code(char* base, size_t size) {
2564 static volatile jint cnt = 0;
2565
2566 if (!UseOprofile) {
2567 return;
2568 }
2569
2570 char buf[PATH_MAX+1];
2571 int num = Atomic::add(1, &cnt);
2572
2573 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2574 os::get_temp_directory(), os::current_process_id(), num);
2575 unlink(buf);
2576
2577 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2578
2579 if (fd != -1) {
2580 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2581 if (rv != (off_t)-1) {
2582 if (::write(fd, "", 1) == 1) {
2583 mmap(base, size,
2584 PROT_READ|PROT_WRITE|PROT_EXEC,
2585 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2586 }
2587 }
2588 ::close(fd);
2589 unlink(buf);
2590 }
2591 }
2592
2593 static bool recoverable_mmap_error(int err) {
2594 // See if the error is one we can let the caller handle. This
2595 // list of errno values comes from JBS-6843484. I can't find a
2596 // Linux man page that documents this specific set of errno
2597 // values so while this list currently matches Solaris, it may
2598 // change as we gain experience with this failure mode.
2599 switch (err) {
2600 case EBADF:
2601 case EINVAL:
2602 case ENOTSUP:
2603 // let the caller deal with these errors
2604 return true;
2605
2606 default:
2607 // Any remaining errors on this OS can cause our reserved mapping
2608 // to be lost. That can cause confusion where different data
2609 // structures think they have the same memory mapped. The worst
2610 // scenario is if both the VM and a library think they have the
2611 // same memory mapped.
2612 return false;
2613 }
2614 }
2615
2616 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2617 int err) {
2618 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2619 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2620 strerror(err), err);
2621 }
2622
2623 static void warn_fail_commit_memory(char* addr, size_t size,
2624 size_t alignment_hint, bool exec,
2625 int err) {
2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2628 alignment_hint, exec, strerror(err), err);
2629 }
2630
2631 // NOTE: Linux kernel does not really reserve the pages for us.
2632 // All it does is to check if there are enough free pages
2633 // left at the time of mmap(). This could be a potential
2634 // problem.
2635 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2636 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2637 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2638 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2639 if (res != (uintptr_t) MAP_FAILED) {
2640 if (UseNUMAInterleaving) {
2641 numa_make_global(addr, size);
2642 }
2643 return 0;
2644 }
2645
2646 int err = errno; // save errno from mmap() call above
2647
2648 if (!recoverable_mmap_error(err)) {
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2651 }
2652
2653 return err;
2654 }
2655
2656 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2657 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2658 }
2659
2660 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2661 const char* mesg) {
2662 assert(mesg != NULL, "mesg must be specified");
2663 int err = os::Linux::commit_memory_impl(addr, size, exec);
2664 if (err != 0) {
2665 // the caller wants all commit errors to exit with the specified mesg:
2666 warn_fail_commit_memory(addr, size, exec, err);
2667 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2668 }
2669 }
2670
2671 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2672 #ifndef MAP_HUGETLB
2673 #define MAP_HUGETLB 0x40000
2674 #endif
2675
2676 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2677 #ifndef MADV_HUGEPAGE
2678 #define MADV_HUGEPAGE 14
2679 #endif
2680
2681 int os::Linux::commit_memory_impl(char* addr, size_t size,
2682 size_t alignment_hint, bool exec) {
2683 int err = os::Linux::commit_memory_impl(addr, size, exec);
2684 if (err == 0) {
2685 realign_memory(addr, size, alignment_hint);
2686 }
2687 return err;
2688 }
2689
2690 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2691 bool exec) {
2692 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2693 }
2694
2695 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2696 size_t alignment_hint, bool exec,
2697 const char* mesg) {
2698 assert(mesg != NULL, "mesg must be specified");
2699 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2700 if (err != 0) {
2701 // the caller wants all commit errors to exit with the specified mesg:
2702 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2703 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2704 }
2705 }
2706
2707 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2708 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2709 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2710 // be supported or the memory may already be backed by huge pages.
2711 ::madvise(addr, bytes, MADV_HUGEPAGE);
2712 }
2713 }
2714
2715 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716 // This method works by doing an mmap over an existing mmaping and effectively discarding
2717 // the existing pages. However it won't work for SHM-based large pages that cannot be
2718 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2719 // small pages on top of the SHM segment. This method always works for small pages, so we
2720 // allow that in any case.
2721 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2722 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2723 }
2724 }
2725
2726 void os::numa_make_global(char *addr, size_t bytes) {
2727 Linux::numa_interleave_memory(addr, bytes);
2728 }
2729
2730 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2731 // bind policy to MPOL_PREFERRED for the current thread.
2732 #define USE_MPOL_PREFERRED 0
2733
2734 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2735 // To make NUMA and large pages more robust when both enabled, we need to ease
2736 // the requirements on where the memory should be allocated. MPOL_BIND is the
2737 // default policy and it will force memory to be allocated on the specified
2738 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2739 // the specified node, but will not force it. Using this policy will prevent
2740 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2741 // free large pages.
2742 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2743 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2744 }
2745
2746 bool os::numa_topology_changed() { return false; }
2747
2748 size_t os::numa_get_groups_num() {
2749 int max_node = Linux::numa_max_node();
2750 return max_node > 0 ? max_node + 1 : 1;
2751 }
2752
2753 int os::numa_get_group_id() {
2754 int cpu_id = Linux::sched_getcpu();
2755 if (cpu_id != -1) {
2756 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2757 if (lgrp_id != -1) {
2758 return lgrp_id;
2759 }
2760 }
2761 return 0;
2762 }
2763
2764 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2765 for (size_t i = 0; i < size; i++) {
2766 ids[i] = i;
2767 }
2768 return size;
2769 }
2770
2771 bool os::get_page_info(char *start, page_info* info) {
2772 return false;
2773 }
2774
2775 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2776 page_info* page_found) {
2777 return end;
2778 }
2779
2780
2781 int os::Linux::sched_getcpu_syscall(void) {
2782 unsigned int cpu;
2783 int retval = -1;
2784
2785 #if defined(IA32)
2786 #ifndef SYS_getcpu
2787 #define SYS_getcpu 318
2788 #endif
2789 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2790 #elif defined(AMD64)
2791 // Unfortunately we have to bring all these macros here from vsyscall.h
2792 // to be able to compile on old linuxes.
2793 #define __NR_vgetcpu 2
2794 #define VSYSCALL_START (-10UL << 20)
2795 #define VSYSCALL_SIZE 1024
2796 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2797 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2798 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2799 retval = vgetcpu(&cpu, NULL, NULL);
2800 #endif
2801
2802 return (retval == -1) ? retval : cpu;
2803 }
2804
2805 // Something to do with the numa-aware allocator needs these symbols
2806 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2807 extern "C" JNIEXPORT void numa_error(char *where) { }
2808 extern "C" JNIEXPORT int fork1() { return fork(); }
2809
2810
2811 // If we are running with libnuma version > 2, then we should
2812 // be trying to use symbols with versions 1.1
2813 // If we are running with earlier version, which did not have symbol versions,
2814 // we should use the base version.
2815 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2816 void *f = dlvsym(handle, name, "libnuma_1.1");
2817 if (f == NULL) {
2818 f = dlsym(handle, name);
2819 }
2820 return f;
2821 }
2822
2823 bool os::Linux::libnuma_init() {
2824 // sched_getcpu() should be in libc.
2825 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2826 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2827
2828 // If it's not, try a direct syscall.
2829 if (sched_getcpu() == -1) {
2830 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2831 (void*)&sched_getcpu_syscall));
2832 }
2833
2834 if (sched_getcpu() != -1) { // Does it work?
2835 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2836 if (handle != NULL) {
2837 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2838 libnuma_dlsym(handle, "numa_node_to_cpus")));
2839 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2840 libnuma_dlsym(handle, "numa_max_node")));
2841 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2842 libnuma_dlsym(handle, "numa_available")));
2843 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2844 libnuma_dlsym(handle, "numa_tonode_memory")));
2845 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2846 libnuma_dlsym(handle, "numa_interleave_memory")));
2847 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2848 libnuma_dlsym(handle, "numa_set_bind_policy")));
2849
2850
2851 if (numa_available() != -1) {
2852 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2853 // Create a cpu -> node mapping
2854 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2855 rebuild_cpu_to_node_map();
2856 return true;
2857 }
2858 }
2859 }
2860 return false;
2861 }
2862
2863 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2864 // The table is later used in get_node_by_cpu().
2865 void os::Linux::rebuild_cpu_to_node_map() {
2866 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2867 // in libnuma (possible values are starting from 16,
2868 // and continuing up with every other power of 2, but less
2869 // than the maximum number of CPUs supported by kernel), and
2870 // is a subject to change (in libnuma version 2 the requirements
2871 // are more reasonable) we'll just hardcode the number they use
2872 // in the library.
2873 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2874
2875 size_t cpu_num = os::active_processor_count();
2876 size_t cpu_map_size = NCPUS / BitsPerCLong;
2877 size_t cpu_map_valid_size =
2878 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2879
2880 cpu_to_node()->clear();
2881 cpu_to_node()->at_grow(cpu_num - 1);
2882 size_t node_num = numa_get_groups_num();
2883
2884 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2885 for (size_t i = 0; i < node_num; i++) {
2886 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2887 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2888 if (cpu_map[j] != 0) {
2889 for (size_t k = 0; k < BitsPerCLong; k++) {
2890 if (cpu_map[j] & (1UL << k)) {
2891 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2892 }
2893 }
2894 }
2895 }
2896 }
2897 }
2898 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2899 }
2900
2901 int os::Linux::get_node_by_cpu(int cpu_id) {
2902 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2903 return cpu_to_node()->at(cpu_id);
2904 }
2905 return -1;
2906 }
2907
2908 GrowableArray<int>* os::Linux::_cpu_to_node;
2909 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2910 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2911 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2912 os::Linux::numa_available_func_t os::Linux::_numa_available;
2913 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2914 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2915 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2916 unsigned long* os::Linux::_numa_all_nodes;
2917
2918 bool os::pd_uncommit_memory(char* addr, size_t size) {
2919 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2920 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2921 return res != (uintptr_t) MAP_FAILED;
2922 }
2923
2924 static address get_stack_commited_bottom(address bottom, size_t size) {
2925 address nbot = bottom;
2926 address ntop = bottom + size;
2927
2928 size_t page_sz = os::vm_page_size();
2929 unsigned pages = size / page_sz;
2930
2931 unsigned char vec[1];
2932 unsigned imin = 1, imax = pages + 1, imid;
2933 int mincore_return_value = 0;
2934
2935 assert(imin <= imax, "Unexpected page size");
2936
2937 while (imin < imax) {
2938 imid = (imax + imin) / 2;
2939 nbot = ntop - (imid * page_sz);
2940
2941 // Use a trick with mincore to check whether the page is mapped or not.
2942 // mincore sets vec to 1 if page resides in memory and to 0 if page
2943 // is swapped output but if page we are asking for is unmapped
2944 // it returns -1,ENOMEM
2945 mincore_return_value = mincore(nbot, page_sz, vec);
2946
2947 if (mincore_return_value == -1) {
2948 // Page is not mapped go up
2949 // to find first mapped page
2950 if (errno != EAGAIN) {
2951 assert(errno == ENOMEM, "Unexpected mincore errno");
2952 imax = imid;
2953 }
2954 } else {
2955 // Page is mapped go down
2956 // to find first not mapped page
2957 imin = imid + 1;
2958 }
2959 }
2960
2961 nbot = nbot + page_sz;
2962
2963 // Adjust stack bottom one page up if last checked page is not mapped
2964 if (mincore_return_value == -1) {
2965 nbot = nbot + page_sz;
2966 }
2967
2968 return nbot;
2969 }
2970
2971
2972 // Linux uses a growable mapping for the stack, and if the mapping for
2973 // the stack guard pages is not removed when we detach a thread the
2974 // stack cannot grow beyond the pages where the stack guard was
2975 // mapped. If at some point later in the process the stack expands to
2976 // that point, the Linux kernel cannot expand the stack any further
2977 // because the guard pages are in the way, and a segfault occurs.
2978 //
2979 // However, it's essential not to split the stack region by unmapping
2980 // a region (leaving a hole) that's already part of the stack mapping,
2981 // so if the stack mapping has already grown beyond the guard pages at
2982 // the time we create them, we have to truncate the stack mapping.
2983 // So, we need to know the extent of the stack mapping when
2984 // create_stack_guard_pages() is called.
2985
2986 // We only need this for stacks that are growable: at the time of
2987 // writing thread stacks don't use growable mappings (i.e. those
2988 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2989 // only applies to the main thread.
2990
2991 // If the (growable) stack mapping already extends beyond the point
2992 // where we're going to put our guard pages, truncate the mapping at
2993 // that point by munmap()ping it. This ensures that when we later
2994 // munmap() the guard pages we don't leave a hole in the stack
2995 // mapping. This only affects the main/initial thread
2996
2997 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2998 if (os::Linux::is_initial_thread()) {
2999 // As we manually grow stack up to bottom inside create_attached_thread(),
3000 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3001 // we don't need to do anything special.
3002 // Check it first, before calling heavy function.
3003 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3004 unsigned char vec[1];
3005
3006 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3007 // Fallback to slow path on all errors, including EAGAIN
3008 stack_extent = (uintptr_t) get_stack_commited_bottom(
3009 os::Linux::initial_thread_stack_bottom(),
3010 (size_t)addr - stack_extent);
3011 }
3012
3013 if (stack_extent < (uintptr_t)addr) {
3014 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3015 }
3016 }
3017
3018 return os::commit_memory(addr, size, !ExecMem);
3019 }
3020
3021 // If this is a growable mapping, remove the guard pages entirely by
3022 // munmap()ping them. If not, just call uncommit_memory(). This only
3023 // affects the main/initial thread, but guard against future OS changes
3024 // It's safe to always unmap guard pages for initial thread because we
3025 // always place it right after end of the mapped region
3026
3027 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3028 uintptr_t stack_extent, stack_base;
3029
3030 if (os::Linux::is_initial_thread()) {
3031 return ::munmap(addr, size) == 0;
3032 }
3033
3034 return os::uncommit_memory(addr, size);
3035 }
3036
3037 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3038 // at 'requested_addr'. If there are existing memory mappings at the same
3039 // location, however, they will be overwritten. If 'fixed' is false,
3040 // 'requested_addr' is only treated as a hint, the return value may or
3041 // may not start from the requested address. Unlike Linux mmap(), this
3042 // function returns NULL to indicate failure.
3043 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3044 char * addr;
3045 int flags;
3046
3047 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3048 if (fixed) {
3049 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3050 flags |= MAP_FIXED;
3051 }
3052
3053 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3054 // touch an uncommitted page. Otherwise, the read/write might
3055 // succeed if we have enough swap space to back the physical page.
3056 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3057 flags, -1, 0);
3058
3059 return addr == MAP_FAILED ? NULL : addr;
3060 }
3061
3062 static int anon_munmap(char * addr, size_t size) {
3063 return ::munmap(addr, size) == 0;
3064 }
3065
3066 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3067 size_t alignment_hint) {
3068 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3069 }
3070
3071 bool os::pd_release_memory(char* addr, size_t size) {
3072 return anon_munmap(addr, size);
3073 }
3074
3075 static bool linux_mprotect(char* addr, size_t size, int prot) {
3076 // Linux wants the mprotect address argument to be page aligned.
3077 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3078
3079 // According to SUSv3, mprotect() should only be used with mappings
3080 // established by mmap(), and mmap() always maps whole pages. Unaligned
3081 // 'addr' likely indicates problem in the VM (e.g. trying to change
3082 // protection of malloc'ed or statically allocated memory). Check the
3083 // caller if you hit this assert.
3084 assert(addr == bottom, "sanity check");
3085
3086 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3087 return ::mprotect(bottom, size, prot) == 0;
3088 }
3089
3090 // Set protections specified
3091 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3092 bool is_committed) {
3093 unsigned int p = 0;
3094 switch (prot) {
3095 case MEM_PROT_NONE: p = PROT_NONE; break;
3096 case MEM_PROT_READ: p = PROT_READ; break;
3097 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3098 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3099 default:
3100 ShouldNotReachHere();
3101 }
3102 // is_committed is unused.
3103 return linux_mprotect(addr, bytes, p);
3104 }
3105
3106 bool os::guard_memory(char* addr, size_t size) {
3107 return linux_mprotect(addr, size, PROT_NONE);
3108 }
3109
3110 bool os::unguard_memory(char* addr, size_t size) {
3111 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3112 }
3113
3114 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3115 size_t page_size) {
3116 bool result = false;
3117 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3118 MAP_ANONYMOUS|MAP_PRIVATE,
3119 -1, 0);
3120 if (p != MAP_FAILED) {
3121 void *aligned_p = align_ptr_up(p, page_size);
3122
3123 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3124
3125 munmap(p, page_size * 2);
3126 }
3127
3128 if (warn && !result) {
3129 warning("TransparentHugePages is not supported by the operating system.");
3130 }
3131
3132 return result;
3133 }
3134
3135 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3136 bool result = false;
3137 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3138 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3139 -1, 0);
3140
3141 if (p != MAP_FAILED) {
3142 // We don't know if this really is a huge page or not.
3143 FILE *fp = fopen("/proc/self/maps", "r");
3144 if (fp) {
3145 while (!feof(fp)) {
3146 char chars[257];
3147 long x = 0;
3148 if (fgets(chars, sizeof(chars), fp)) {
3149 if (sscanf(chars, "%lx-%*x", &x) == 1
3150 && x == (long)p) {
3151 if (strstr (chars, "hugepage")) {
3152 result = true;
3153 break;
3154 }
3155 }
3156 }
3157 }
3158 fclose(fp);
3159 }
3160 munmap(p, page_size);
3161 }
3162
3163 if (warn && !result) {
3164 warning("HugeTLBFS is not supported by the operating system.");
3165 }
3166
3167 return result;
3168 }
3169
3170 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3171 //
3172 // From the coredump_filter documentation:
3173 //
3174 // - (bit 0) anonymous private memory
3175 // - (bit 1) anonymous shared memory
3176 // - (bit 2) file-backed private memory
3177 // - (bit 3) file-backed shared memory
3178 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3179 // effective only if the bit 2 is cleared)
3180 // - (bit 5) hugetlb private memory
3181 // - (bit 6) hugetlb shared memory
3182 //
3183 static void set_coredump_filter(void) {
3184 FILE *f;
3185 long cdm;
3186
3187 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3188 return;
3189 }
3190
3191 if (fscanf(f, "%lx", &cdm) != 1) {
3192 fclose(f);
3193 return;
3194 }
3195
3196 rewind(f);
3197
3198 if ((cdm & LARGEPAGES_BIT) == 0) {
3199 cdm |= LARGEPAGES_BIT;
3200 fprintf(f, "%#lx", cdm);
3201 }
3202
3203 fclose(f);
3204 }
3205
3206 // Large page support
3207
3208 static size_t _large_page_size = 0;
3209
3210 size_t os::Linux::find_large_page_size() {
3211 size_t large_page_size = 0;
3212
3213 // large_page_size on Linux is used to round up heap size. x86 uses either
3214 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3215 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3216 // page as large as 256M.
3217 //
3218 // Here we try to figure out page size by parsing /proc/meminfo and looking
3219 // for a line with the following format:
3220 // Hugepagesize: 2048 kB
3221 //
3222 // If we can't determine the value (e.g. /proc is not mounted, or the text
3223 // format has been changed), we'll use the largest page size supported by
3224 // the processor.
3225
3226 #ifndef ZERO
3227 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3228 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3229 #endif // ZERO
3230
3231 FILE *fp = fopen("/proc/meminfo", "r");
3232 if (fp) {
3233 while (!feof(fp)) {
3234 int x = 0;
3235 char buf[16];
3236 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3237 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3238 large_page_size = x * K;
3239 break;
3240 }
3241 } else {
3242 // skip to next line
3243 for (;;) {
3244 int ch = fgetc(fp);
3245 if (ch == EOF || ch == (int)'\n') break;
3246 }
3247 }
3248 }
3249 fclose(fp);
3250 }
3251
3252 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3253 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3254 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3255 proper_unit_for_byte_size(large_page_size));
3256 }
3257
3258 return large_page_size;
3259 }
3260
3261 size_t os::Linux::setup_large_page_size() {
3262 _large_page_size = Linux::find_large_page_size();
3263 const size_t default_page_size = (size_t)Linux::page_size();
3264 if (_large_page_size > default_page_size) {
3265 _page_sizes[0] = _large_page_size;
3266 _page_sizes[1] = default_page_size;
3267 _page_sizes[2] = 0;
3268 }
3269
3270 return _large_page_size;
3271 }
3272
3273 bool os::Linux::setup_large_page_type(size_t page_size) {
3274 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3275 FLAG_IS_DEFAULT(UseSHM) &&
3276 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3277
3278 // The type of large pages has not been specified by the user.
3279
3280 // Try UseHugeTLBFS and then UseSHM.
3281 UseHugeTLBFS = UseSHM = true;
3282
3283 // Don't try UseTransparentHugePages since there are known
3284 // performance issues with it turned on. This might change in the future.
3285 UseTransparentHugePages = false;
3286 }
3287
3288 if (UseTransparentHugePages) {
3289 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3290 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3291 UseHugeTLBFS = false;
3292 UseSHM = false;
3293 return true;
3294 }
3295 UseTransparentHugePages = false;
3296 }
3297
3298 if (UseHugeTLBFS) {
3299 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3300 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3301 UseSHM = false;
3302 return true;
3303 }
3304 UseHugeTLBFS = false;
3305 }
3306
3307 return UseSHM;
3308 }
3309
3310 void os::large_page_init() {
3311 if (!UseLargePages &&
3312 !UseTransparentHugePages &&
3313 !UseHugeTLBFS &&
3314 !UseSHM) {
3315 // Not using large pages.
3316 return;
3317 }
3318
3319 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3320 // The user explicitly turned off large pages.
3321 // Ignore the rest of the large pages flags.
3322 UseTransparentHugePages = false;
3323 UseHugeTLBFS = false;
3324 UseSHM = false;
3325 return;
3326 }
3327
3328 size_t large_page_size = Linux::setup_large_page_size();
3329 UseLargePages = Linux::setup_large_page_type(large_page_size);
3330
3331 set_coredump_filter();
3332 }
3333
3334 #ifndef SHM_HUGETLB
3335 #define SHM_HUGETLB 04000
3336 #endif
3337
3338 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3339 char* req_addr, bool exec) {
3340 // "exec" is passed in but not used. Creating the shared image for
3341 // the code cache doesn't have an SHM_X executable permission to check.
3342 assert(UseLargePages && UseSHM, "only for SHM large pages");
3343 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3344
3345 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3346 return NULL; // Fallback to small pages.
3347 }
3348
3349 key_t key = IPC_PRIVATE;
3350 char *addr;
3351
3352 bool warn_on_failure = UseLargePages &&
3353 (!FLAG_IS_DEFAULT(UseLargePages) ||
3354 !FLAG_IS_DEFAULT(UseSHM) ||
3355 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3356 char msg[128];
3357
3358 // Create a large shared memory region to attach to based on size.
3359 // Currently, size is the total size of the heap
3360 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3361 if (shmid == -1) {
3362 // Possible reasons for shmget failure:
3363 // 1. shmmax is too small for Java heap.
3364 // > check shmmax value: cat /proc/sys/kernel/shmmax
3365 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3366 // 2. not enough large page memory.
3367 // > check available large pages: cat /proc/meminfo
3368 // > increase amount of large pages:
3369 // echo new_value > /proc/sys/vm/nr_hugepages
3370 // Note 1: different Linux may use different name for this property,
3371 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3372 // Note 2: it's possible there's enough physical memory available but
3373 // they are so fragmented after a long run that they can't
3374 // coalesce into large pages. Try to reserve large pages when
3375 // the system is still "fresh".
3376 if (warn_on_failure) {
3377 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3378 warning("%s", msg);
3379 }
3380 return NULL;
3381 }
3382
3383 // attach to the region
3384 addr = (char*)shmat(shmid, req_addr, 0);
3385 int err = errno;
3386
3387 // Remove shmid. If shmat() is successful, the actual shared memory segment
3388 // will be deleted when it's detached by shmdt() or when the process
3389 // terminates. If shmat() is not successful this will remove the shared
3390 // segment immediately.
3391 shmctl(shmid, IPC_RMID, NULL);
3392
3393 if ((intptr_t)addr == -1) {
3394 if (warn_on_failure) {
3395 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3396 warning("%s", msg);
3397 }
3398 return NULL;
3399 }
3400
3401 return addr;
3402 }
3403
3404 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3405 int error) {
3406 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3407
3408 bool warn_on_failure = UseLargePages &&
3409 (!FLAG_IS_DEFAULT(UseLargePages) ||
3410 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3411 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3412
3413 if (warn_on_failure) {
3414 char msg[128];
3415 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3416 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3417 warning("%s", msg);
3418 }
3419 }
3420
3421 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3422 char* req_addr,
3423 bool exec) {
3424 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3425 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3426 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3427
3428 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3429 char* addr = (char*)::mmap(req_addr, bytes, prot,
3430 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3431 -1, 0);
3432
3433 if (addr == MAP_FAILED) {
3434 warn_on_large_pages_failure(req_addr, bytes, errno);
3435 return NULL;
3436 }
3437
3438 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3439
3440 return addr;
3441 }
3442
3443 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3444 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3445 // (req_addr != NULL) or with a given alignment.
3446 // - bytes shall be a multiple of alignment.
3447 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3448 // - alignment sets the alignment at which memory shall be allocated.
3449 // It must be a multiple of allocation granularity.
3450 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3451 // req_addr or NULL.
3452 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3453
3454 size_t extra_size = bytes;
3455 if (req_addr == NULL && alignment > 0) {
3456 extra_size += alignment;
3457 }
3458
3459 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3460 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3461 -1, 0);
3462 if (start == MAP_FAILED) {
3463 start = NULL;
3464 } else {
3465 if (req_addr != NULL) {
3466 if (start != req_addr) {
3467 ::munmap(start, extra_size);
3468 start = NULL;
3469 }
3470 } else {
3471 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3472 char* const end_aligned = start_aligned + bytes;
3473 char* const end = start + extra_size;
3474 if (start_aligned > start) {
3475 ::munmap(start, start_aligned - start);
3476 }
3477 if (end_aligned < end) {
3478 ::munmap(end_aligned, end - end_aligned);
3479 }
3480 start = start_aligned;
3481 }
3482 }
3483 return start;
3484
3485 }
3486
3487 // Reserve memory using mmap(MAP_HUGETLB).
3488 // - bytes shall be a multiple of alignment.
3489 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3490 // - alignment sets the alignment at which memory shall be allocated.
3491 // It must be a multiple of allocation granularity.
3492 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3493 // req_addr or NULL.
3494 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3495 size_t alignment,
3496 char* req_addr,
3497 bool exec) {
3498 size_t large_page_size = os::large_page_size();
3499 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3500
3501 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3502 assert(is_size_aligned(bytes, alignment), "Must be");
3503
3504 // First reserve - but not commit - the address range in small pages.
3505 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3506
3507 if (start == NULL) {
3508 return NULL;
3509 }
3510
3511 assert(is_ptr_aligned(start, alignment), "Must be");
3512
3513 char* end = start + bytes;
3514
3515 // Find the regions of the allocated chunk that can be promoted to large pages.
3516 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3517 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3518
3519 size_t lp_bytes = lp_end - lp_start;
3520
3521 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3522
3523 if (lp_bytes == 0) {
3524 // The mapped region doesn't even span the start and the end of a large page.
3525 // Fall back to allocate a non-special area.
3526 ::munmap(start, end - start);
3527 return NULL;
3528 }
3529
3530 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3531
3532 void* result;
3533
3534 // Commit small-paged leading area.
3535 if (start != lp_start) {
3536 result = ::mmap(start, lp_start - start, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3538 -1, 0);
3539 if (result == MAP_FAILED) {
3540 ::munmap(lp_start, end - lp_start);
3541 return NULL;
3542 }
3543 }
3544
3545 // Commit large-paged area.
3546 result = ::mmap(lp_start, lp_bytes, prot,
3547 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3548 -1, 0);
3549 if (result == MAP_FAILED) {
3550 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3551 // If the mmap above fails, the large pages region will be unmapped and we
3552 // have regions before and after with small pages. Release these regions.
3553 //
3554 // | mapped | unmapped | mapped |
3555 // ^ ^ ^ ^
3556 // start lp_start lp_end end
3557 //
3558 ::munmap(start, lp_start - start);
3559 ::munmap(lp_end, end - lp_end);
3560 return NULL;
3561 }
3562
3563 // Commit small-paged trailing area.
3564 if (lp_end != end) {
3565 result = ::mmap(lp_end, end - lp_end, prot,
3566 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3567 -1, 0);
3568 if (result == MAP_FAILED) {
3569 ::munmap(start, lp_end - start);
3570 return NULL;
3571 }
3572 }
3573
3574 return start;
3575 }
3576
3577 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3578 size_t alignment,
3579 char* req_addr,
3580 bool exec) {
3581 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3582 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3583 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3584 assert(is_power_of_2(os::large_page_size()), "Must be");
3585 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3586
3587 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3588 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3589 } else {
3590 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3591 }
3592 }
3593
3594 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3595 char* req_addr, bool exec) {
3596 assert(UseLargePages, "only for large pages");
3597
3598 char* addr;
3599 if (UseSHM) {
3600 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3601 } else {
3602 assert(UseHugeTLBFS, "must be");
3603 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3604 }
3605
3606 if (addr != NULL) {
3607 if (UseNUMAInterleaving) {
3608 numa_make_global(addr, bytes);
3609 }
3610
3611 // The memory is committed
3612 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3613 }
3614
3615 return addr;
3616 }
3617
3618 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3619 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3620 return shmdt(base) == 0;
3621 }
3622
3623 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3624 return pd_release_memory(base, bytes);
3625 }
3626
3627 bool os::release_memory_special(char* base, size_t bytes) {
3628 bool res;
3629 if (MemTracker::tracking_level() > NMT_minimal) {
3630 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3631 res = os::Linux::release_memory_special_impl(base, bytes);
3632 if (res) {
3633 tkr.record((address)base, bytes);
3634 }
3635
3636 } else {
3637 res = os::Linux::release_memory_special_impl(base, bytes);
3638 }
3639 return res;
3640 }
3641
3642 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3643 assert(UseLargePages, "only for large pages");
3644 bool res;
3645
3646 if (UseSHM) {
3647 res = os::Linux::release_memory_special_shm(base, bytes);
3648 } else {
3649 assert(UseHugeTLBFS, "must be");
3650 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3651 }
3652 return res;
3653 }
3654
3655 size_t os::large_page_size() {
3656 return _large_page_size;
3657 }
3658
3659 // With SysV SHM the entire memory region must be allocated as shared
3660 // memory.
3661 // HugeTLBFS allows application to commit large page memory on demand.
3662 // However, when committing memory with HugeTLBFS fails, the region
3663 // that was supposed to be committed will lose the old reservation
3664 // and allow other threads to steal that memory region. Because of this
3665 // behavior we can't commit HugeTLBFS memory.
3666 bool os::can_commit_large_page_memory() {
3667 return UseTransparentHugePages;
3668 }
3669
3670 bool os::can_execute_large_page_memory() {
3671 return UseTransparentHugePages || UseHugeTLBFS;
3672 }
3673
3674 // Reserve memory at an arbitrary address, only if that area is
3675 // available (and not reserved for something else).
3676
3677 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3678 const int max_tries = 10;
3679 char* base[max_tries];
3680 size_t size[max_tries];
3681 const size_t gap = 0x000000;
3682
3683 // Assert only that the size is a multiple of the page size, since
3684 // that's all that mmap requires, and since that's all we really know
3685 // about at this low abstraction level. If we need higher alignment,
3686 // we can either pass an alignment to this method or verify alignment
3687 // in one of the methods further up the call chain. See bug 5044738.
3688 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3689
3690 // Repeatedly allocate blocks until the block is allocated at the
3691 // right spot.
3692
3693 // Linux mmap allows caller to pass an address as hint; give it a try first,
3694 // if kernel honors the hint then we can return immediately.
3695 char * addr = anon_mmap(requested_addr, bytes, false);
3696 if (addr == requested_addr) {
3697 return requested_addr;
3698 }
3699
3700 if (addr != NULL) {
3701 // mmap() is successful but it fails to reserve at the requested address
3702 anon_munmap(addr, bytes);
3703 }
3704
3705 int i;
3706 for (i = 0; i < max_tries; ++i) {
3707 base[i] = reserve_memory(bytes);
3708
3709 if (base[i] != NULL) {
3710 // Is this the block we wanted?
3711 if (base[i] == requested_addr) {
3712 size[i] = bytes;
3713 break;
3714 }
3715
3716 // Does this overlap the block we wanted? Give back the overlapped
3717 // parts and try again.
3718
3719 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3720 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3721 unmap_memory(base[i], top_overlap);
3722 base[i] += top_overlap;
3723 size[i] = bytes - top_overlap;
3724 } else {
3725 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3726 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3727 unmap_memory(requested_addr, bottom_overlap);
3728 size[i] = bytes - bottom_overlap;
3729 } else {
3730 size[i] = bytes;
3731 }
3732 }
3733 }
3734 }
3735
3736 // Give back the unused reserved pieces.
3737
3738 for (int j = 0; j < i; ++j) {
3739 if (base[j] != NULL) {
3740 unmap_memory(base[j], size[j]);
3741 }
3742 }
3743
3744 if (i < max_tries) {
3745 return requested_addr;
3746 } else {
3747 return NULL;
3748 }
3749 }
3750
3751 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3752 return ::read(fd, buf, nBytes);
3753 }
3754
3755 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3756 return ::pread(fd, buf, nBytes, offset);
3757 }
3758
3759 // Short sleep, direct OS call.
3760 //
3761 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3762 // sched_yield(2) will actually give up the CPU:
3763 //
3764 // * Alone on this pariticular CPU, keeps running.
3765 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3766 // (pre 2.6.39).
3767 //
3768 // So calling this with 0 is an alternative.
3769 //
3770 void os::naked_short_sleep(jlong ms) {
3771 struct timespec req;
3772
3773 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3774 req.tv_sec = 0;
3775 if (ms > 0) {
3776 req.tv_nsec = (ms % 1000) * 1000000;
3777 } else {
3778 req.tv_nsec = 1;
3779 }
3780
3781 nanosleep(&req, NULL);
3782
3783 return;
3784 }
3785
3786 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3787 void os::infinite_sleep() {
3788 while (true) { // sleep forever ...
3789 ::sleep(100); // ... 100 seconds at a time
3790 }
3791 }
3792
3793 // Used to convert frequent JVM_Yield() to nops
3794 bool os::dont_yield() {
3795 return DontYieldALot;
3796 }
3797
3798 void os::naked_yield() {
3799 sched_yield();
3800 }
3801
3802 ////////////////////////////////////////////////////////////////////////////////
3803 // thread priority support
3804
3805 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3806 // only supports dynamic priority, static priority must be zero. For real-time
3807 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3808 // However, for large multi-threaded applications, SCHED_RR is not only slower
3809 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3810 // of 5 runs - Sep 2005).
3811 //
3812 // The following code actually changes the niceness of kernel-thread/LWP. It
3813 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3814 // not the entire user process, and user level threads are 1:1 mapped to kernel
3815 // threads. It has always been the case, but could change in the future. For
3816 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3817 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3818
3819 int os::java_to_os_priority[CriticalPriority + 1] = {
3820 19, // 0 Entry should never be used
3821
3822 4, // 1 MinPriority
3823 3, // 2
3824 2, // 3
3825
3826 1, // 4
3827 0, // 5 NormPriority
3828 -1, // 6
3829
3830 -2, // 7
3831 -3, // 8
3832 -4, // 9 NearMaxPriority
3833
3834 -5, // 10 MaxPriority
3835
3836 -5 // 11 CriticalPriority
3837 };
3838
3839 static int prio_init() {
3840 if (ThreadPriorityPolicy == 1) {
3841 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3842 // if effective uid is not root. Perhaps, a more elegant way of doing
3843 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3844 if (geteuid() != 0) {
3845 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3846 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3847 }
3848 ThreadPriorityPolicy = 0;
3849 }
3850 }
3851 if (UseCriticalJavaThreadPriority) {
3852 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3853 }
3854 return 0;
3855 }
3856
3857 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3858 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3859
3860 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3861 return (ret == 0) ? OS_OK : OS_ERR;
3862 }
3863
3864 OSReturn os::get_native_priority(const Thread* const thread,
3865 int *priority_ptr) {
3866 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3867 *priority_ptr = java_to_os_priority[NormPriority];
3868 return OS_OK;
3869 }
3870
3871 errno = 0;
3872 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3873 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3874 }
3875
3876 // Hint to the underlying OS that a task switch would not be good.
3877 // Void return because it's a hint and can fail.
3878 void os::hint_no_preempt() {}
3879
3880 ////////////////////////////////////////////////////////////////////////////////
3881 // suspend/resume support
3882
3883 // the low-level signal-based suspend/resume support is a remnant from the
3884 // old VM-suspension that used to be for java-suspension, safepoints etc,
3885 // within hotspot. Now there is a single use-case for this:
3886 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3887 // that runs in the watcher thread.
3888 // The remaining code is greatly simplified from the more general suspension
3889 // code that used to be used.
3890 //
3891 // The protocol is quite simple:
3892 // - suspend:
3893 // - sends a signal to the target thread
3894 // - polls the suspend state of the osthread using a yield loop
3895 // - target thread signal handler (SR_handler) sets suspend state
3896 // and blocks in sigsuspend until continued
3897 // - resume:
3898 // - sets target osthread state to continue
3899 // - sends signal to end the sigsuspend loop in the SR_handler
3900 //
3901 // Note that the SR_lock plays no role in this suspend/resume protocol.
3902
3903 static void resume_clear_context(OSThread *osthread) {
3904 osthread->set_ucontext(NULL);
3905 osthread->set_siginfo(NULL);
3906 }
3907
3908 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3909 ucontext_t* context) {
3910 osthread->set_ucontext(context);
3911 osthread->set_siginfo(siginfo);
3912 }
3913
3914 // Handler function invoked when a thread's execution is suspended or
3915 // resumed. We have to be careful that only async-safe functions are
3916 // called here (Note: most pthread functions are not async safe and
3917 // should be avoided.)
3918 //
3919 // Note: sigwait() is a more natural fit than sigsuspend() from an
3920 // interface point of view, but sigwait() prevents the signal hander
3921 // from being run. libpthread would get very confused by not having
3922 // its signal handlers run and prevents sigwait()'s use with the
3923 // mutex granting granting signal.
3924 //
3925 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3926 //
3927 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3928 // Save and restore errno to avoid confusing native code with EINTR
3929 // after sigsuspend.
3930 int old_errno = errno;
3931
3932 Thread* thread = Thread::current();
3933 OSThread* osthread = thread->osthread();
3934 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3935
3936 os::SuspendResume::State current = osthread->sr.state();
3937 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3938 suspend_save_context(osthread, siginfo, context);
3939
3940 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3941 os::SuspendResume::State state = osthread->sr.suspended();
3942 if (state == os::SuspendResume::SR_SUSPENDED) {
3943 sigset_t suspend_set; // signals for sigsuspend()
3944
3945 // get current set of blocked signals and unblock resume signal
3946 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3947 sigdelset(&suspend_set, SR_signum);
3948
3949 sr_semaphore.signal();
3950 // wait here until we are resumed
3951 while (1) {
3952 sigsuspend(&suspend_set);
3953
3954 os::SuspendResume::State result = osthread->sr.running();
3955 if (result == os::SuspendResume::SR_RUNNING) {
3956 sr_semaphore.signal();
3957 break;
3958 }
3959 }
3960
3961 } else if (state == os::SuspendResume::SR_RUNNING) {
3962 // request was cancelled, continue
3963 } else {
3964 ShouldNotReachHere();
3965 }
3966
3967 resume_clear_context(osthread);
3968 } else if (current == os::SuspendResume::SR_RUNNING) {
3969 // request was cancelled, continue
3970 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3971 // ignore
3972 } else {
3973 // ignore
3974 }
3975
3976 errno = old_errno;
3977 }
3978
3979
3980 static int SR_initialize() {
3981 struct sigaction act;
3982 char *s;
3983 // Get signal number to use for suspend/resume
3984 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3985 int sig = ::strtol(s, 0, 10);
3986 if (sig > 0 || sig < _NSIG) {
3987 SR_signum = sig;
3988 }
3989 }
3990
3991 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3992 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3993
3994 sigemptyset(&SR_sigset);
3995 sigaddset(&SR_sigset, SR_signum);
3996
3997 // Set up signal handler for suspend/resume
3998 act.sa_flags = SA_RESTART|SA_SIGINFO;
3999 act.sa_handler = (void (*)(int)) SR_handler;
4000
4001 // SR_signum is blocked by default.
4002 // 4528190 - We also need to block pthread restart signal (32 on all
4003 // supported Linux platforms). Note that LinuxThreads need to block
4004 // this signal for all threads to work properly. So we don't have
4005 // to use hard-coded signal number when setting up the mask.
4006 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4007
4008 if (sigaction(SR_signum, &act, 0) == -1) {
4009 return -1;
4010 }
4011
4012 // Save signal flag
4013 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4014 return 0;
4015 }
4016
4017 static int sr_notify(OSThread* osthread) {
4018 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4019 assert_status(status == 0, status, "pthread_kill");
4020 return status;
4021 }
4022
4023 // "Randomly" selected value for how long we want to spin
4024 // before bailing out on suspending a thread, also how often
4025 // we send a signal to a thread we want to resume
4026 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4027 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4028
4029 // returns true on success and false on error - really an error is fatal
4030 // but this seems the normal response to library errors
4031 static bool do_suspend(OSThread* osthread) {
4032 assert(osthread->sr.is_running(), "thread should be running");
4033 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4034
4035 // mark as suspended and send signal
4036 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4037 // failed to switch, state wasn't running?
4038 ShouldNotReachHere();
4039 return false;
4040 }
4041
4042 if (sr_notify(osthread) != 0) {
4043 ShouldNotReachHere();
4044 }
4045
4046 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4047 while (true) {
4048 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4049 break;
4050 } else {
4051 // timeout
4052 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4053 if (cancelled == os::SuspendResume::SR_RUNNING) {
4054 return false;
4055 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4056 // make sure that we consume the signal on the semaphore as well
4057 sr_semaphore.wait();
4058 break;
4059 } else {
4060 ShouldNotReachHere();
4061 return false;
4062 }
4063 }
4064 }
4065
4066 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4067 return true;
4068 }
4069
4070 static void do_resume(OSThread* osthread) {
4071 assert(osthread->sr.is_suspended(), "thread should be suspended");
4072 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4073
4074 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4075 // failed to switch to WAKEUP_REQUEST
4076 ShouldNotReachHere();
4077 return;
4078 }
4079
4080 while (true) {
4081 if (sr_notify(osthread) == 0) {
4082 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4083 if (osthread->sr.is_running()) {
4084 return;
4085 }
4086 }
4087 } else {
4088 ShouldNotReachHere();
4089 }
4090 }
4091
4092 guarantee(osthread->sr.is_running(), "Must be running!");
4093 }
4094
4095 ///////////////////////////////////////////////////////////////////////////////////
4096 // signal handling (except suspend/resume)
4097
4098 // This routine may be used by user applications as a "hook" to catch signals.
4099 // The user-defined signal handler must pass unrecognized signals to this
4100 // routine, and if it returns true (non-zero), then the signal handler must
4101 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4102 // routine will never retun false (zero), but instead will execute a VM panic
4103 // routine kill the process.
4104 //
4105 // If this routine returns false, it is OK to call it again. This allows
4106 // the user-defined signal handler to perform checks either before or after
4107 // the VM performs its own checks. Naturally, the user code would be making
4108 // a serious error if it tried to handle an exception (such as a null check
4109 // or breakpoint) that the VM was generating for its own correct operation.
4110 //
4111 // This routine may recognize any of the following kinds of signals:
4112 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4113 // It should be consulted by handlers for any of those signals.
4114 //
4115 // The caller of this routine must pass in the three arguments supplied
4116 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4117 // field of the structure passed to sigaction(). This routine assumes that
4118 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4119 //
4120 // Note that the VM will print warnings if it detects conflicting signal
4121 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4122 //
4123 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4124 siginfo_t* siginfo,
4125 void* ucontext,
4126 int abort_if_unrecognized);
4127
4128 void signalHandler(int sig, siginfo_t* info, void* uc) {
4129 assert(info != NULL && uc != NULL, "it must be old kernel");
4130 int orig_errno = errno; // Preserve errno value over signal handler.
4131 JVM_handle_linux_signal(sig, info, uc, true);
4132 errno = orig_errno;
4133 }
4134
4135
4136 // This boolean allows users to forward their own non-matching signals
4137 // to JVM_handle_linux_signal, harmlessly.
4138 bool os::Linux::signal_handlers_are_installed = false;
4139
4140 // For signal-chaining
4141 struct sigaction os::Linux::sigact[MAXSIGNUM];
4142 unsigned int os::Linux::sigs = 0;
4143 bool os::Linux::libjsig_is_loaded = false;
4144 typedef struct sigaction *(*get_signal_t)(int);
4145 get_signal_t os::Linux::get_signal_action = NULL;
4146
4147 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4148 struct sigaction *actp = NULL;
4149
4150 if (libjsig_is_loaded) {
4151 // Retrieve the old signal handler from libjsig
4152 actp = (*get_signal_action)(sig);
4153 }
4154 if (actp == NULL) {
4155 // Retrieve the preinstalled signal handler from jvm
4156 actp = get_preinstalled_handler(sig);
4157 }
4158
4159 return actp;
4160 }
4161
4162 static bool call_chained_handler(struct sigaction *actp, int sig,
4163 siginfo_t *siginfo, void *context) {
4164 // Call the old signal handler
4165 if (actp->sa_handler == SIG_DFL) {
4166 // It's more reasonable to let jvm treat it as an unexpected exception
4167 // instead of taking the default action.
4168 return false;
4169 } else if (actp->sa_handler != SIG_IGN) {
4170 if ((actp->sa_flags & SA_NODEFER) == 0) {
4171 // automaticlly block the signal
4172 sigaddset(&(actp->sa_mask), sig);
4173 }
4174
4175 sa_handler_t hand;
4176 sa_sigaction_t sa;
4177 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4178 // retrieve the chained handler
4179 if (siginfo_flag_set) {
4180 sa = actp->sa_sigaction;
4181 } else {
4182 hand = actp->sa_handler;
4183 }
4184
4185 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4186 actp->sa_handler = SIG_DFL;
4187 }
4188
4189 // try to honor the signal mask
4190 sigset_t oset;
4191 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4192
4193 // call into the chained handler
4194 if (siginfo_flag_set) {
4195 (*sa)(sig, siginfo, context);
4196 } else {
4197 (*hand)(sig);
4198 }
4199
4200 // restore the signal mask
4201 pthread_sigmask(SIG_SETMASK, &oset, 0);
4202 }
4203 // Tell jvm's signal handler the signal is taken care of.
4204 return true;
4205 }
4206
4207 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4208 bool chained = false;
4209 // signal-chaining
4210 if (UseSignalChaining) {
4211 struct sigaction *actp = get_chained_signal_action(sig);
4212 if (actp != NULL) {
4213 chained = call_chained_handler(actp, sig, siginfo, context);
4214 }
4215 }
4216 return chained;
4217 }
4218
4219 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4220 if ((((unsigned int)1 << sig) & sigs) != 0) {
4221 return &sigact[sig];
4222 }
4223 return NULL;
4224 }
4225
4226 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4227 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4228 sigact[sig] = oldAct;
4229 sigs |= (unsigned int)1 << sig;
4230 }
4231
4232 // for diagnostic
4233 int os::Linux::sigflags[MAXSIGNUM];
4234
4235 int os::Linux::get_our_sigflags(int sig) {
4236 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4237 return sigflags[sig];
4238 }
4239
4240 void os::Linux::set_our_sigflags(int sig, int flags) {
4241 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4242 sigflags[sig] = flags;
4243 }
4244
4245 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4246 // Check for overwrite.
4247 struct sigaction oldAct;
4248 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4249
4250 void* oldhand = oldAct.sa_sigaction
4251 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4252 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4253 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4254 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4255 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4256 if (AllowUserSignalHandlers || !set_installed) {
4257 // Do not overwrite; user takes responsibility to forward to us.
4258 return;
4259 } else if (UseSignalChaining) {
4260 // save the old handler in jvm
4261 save_preinstalled_handler(sig, oldAct);
4262 // libjsig also interposes the sigaction() call below and saves the
4263 // old sigaction on it own.
4264 } else {
4265 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4266 "%#lx for signal %d.", (long)oldhand, sig));
4267 }
4268 }
4269
4270 struct sigaction sigAct;
4271 sigfillset(&(sigAct.sa_mask));
4272 sigAct.sa_handler = SIG_DFL;
4273 if (!set_installed) {
4274 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4275 } else {
4276 sigAct.sa_sigaction = signalHandler;
4277 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4278 }
4279 // Save flags, which are set by ours
4280 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4281 sigflags[sig] = sigAct.sa_flags;
4282
4283 int ret = sigaction(sig, &sigAct, &oldAct);
4284 assert(ret == 0, "check");
4285
4286 void* oldhand2 = oldAct.sa_sigaction
4287 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4288 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4289 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4290 }
4291
4292 // install signal handlers for signals that HotSpot needs to
4293 // handle in order to support Java-level exception handling.
4294
4295 void os::Linux::install_signal_handlers() {
4296 if (!signal_handlers_are_installed) {
4297 signal_handlers_are_installed = true;
4298
4299 // signal-chaining
4300 typedef void (*signal_setting_t)();
4301 signal_setting_t begin_signal_setting = NULL;
4302 signal_setting_t end_signal_setting = NULL;
4303 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4304 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4305 if (begin_signal_setting != NULL) {
4306 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4307 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4308 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4309 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4310 libjsig_is_loaded = true;
4311 assert(UseSignalChaining, "should enable signal-chaining");
4312 }
4313 if (libjsig_is_loaded) {
4314 // Tell libjsig jvm is setting signal handlers
4315 (*begin_signal_setting)();
4316 }
4317
4318 set_signal_handler(SIGSEGV, true);
4319 set_signal_handler(SIGPIPE, true);
4320 set_signal_handler(SIGBUS, true);
4321 set_signal_handler(SIGILL, true);
4322 set_signal_handler(SIGFPE, true);
4323 #if defined(PPC64)
4324 set_signal_handler(SIGTRAP, true);
4325 #endif
4326 set_signal_handler(SIGXFSZ, true);
4327
4328 if (libjsig_is_loaded) {
4329 // Tell libjsig jvm finishes setting signal handlers
4330 (*end_signal_setting)();
4331 }
4332
4333 // We don't activate signal checker if libjsig is in place, we trust ourselves
4334 // and if UserSignalHandler is installed all bets are off.
4335 // Log that signal checking is off only if -verbose:jni is specified.
4336 if (CheckJNICalls) {
4337 if (libjsig_is_loaded) {
4338 if (PrintJNIResolving) {
4339 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4340 }
4341 check_signals = false;
4342 }
4343 if (AllowUserSignalHandlers) {
4344 if (PrintJNIResolving) {
4345 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4346 }
4347 check_signals = false;
4348 }
4349 }
4350 }
4351 }
4352
4353 // This is the fastest way to get thread cpu time on Linux.
4354 // Returns cpu time (user+sys) for any thread, not only for current.
4355 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4356 // It might work on 2.6.10+ with a special kernel/glibc patch.
4357 // For reference, please, see IEEE Std 1003.1-2004:
4358 // http://www.unix.org/single_unix_specification
4359
4360 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4361 struct timespec tp;
4362 int rc = os::Linux::clock_gettime(clockid, &tp);
4363 assert(rc == 0, "clock_gettime is expected to return 0 code");
4364
4365 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4366 }
4367
4368 /////
4369 // glibc on Linux platform uses non-documented flag
4370 // to indicate, that some special sort of signal
4371 // trampoline is used.
4372 // We will never set this flag, and we should
4373 // ignore this flag in our diagnostic
4374 #ifdef SIGNIFICANT_SIGNAL_MASK
4375 #undef SIGNIFICANT_SIGNAL_MASK
4376 #endif
4377 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4378
4379 static const char* get_signal_handler_name(address handler,
4380 char* buf, int buflen) {
4381 int offset;
4382 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4383 if (found) {
4384 // skip directory names
4385 const char *p1, *p2;
4386 p1 = buf;
4387 size_t len = strlen(os::file_separator());
4388 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4389 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4390 } else {
4391 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4392 }
4393 return buf;
4394 }
4395
4396 static void print_signal_handler(outputStream* st, int sig,
4397 char* buf, size_t buflen) {
4398 struct sigaction sa;
4399
4400 sigaction(sig, NULL, &sa);
4401
4402 // See comment for SIGNIFICANT_SIGNAL_MASK define
4403 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4404
4405 st->print("%s: ", os::exception_name(sig, buf, buflen));
4406
4407 address handler = (sa.sa_flags & SA_SIGINFO)
4408 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4409 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4410
4411 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4412 st->print("SIG_DFL");
4413 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4414 st->print("SIG_IGN");
4415 } else {
4416 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4417 }
4418
4419 st->print(", sa_mask[0]=");
4420 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4421
4422 address rh = VMError::get_resetted_sighandler(sig);
4423 // May be, handler was resetted by VMError?
4424 if (rh != NULL) {
4425 handler = rh;
4426 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4427 }
4428
4429 st->print(", sa_flags=");
4430 os::Posix::print_sa_flags(st, sa.sa_flags);
4431
4432 // Check: is it our handler?
4433 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4434 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4435 // It is our signal handler
4436 // check for flags, reset system-used one!
4437 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4438 st->print(
4439 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4440 os::Linux::get_our_sigflags(sig));
4441 }
4442 }
4443 st->cr();
4444 }
4445
4446
4447 #define DO_SIGNAL_CHECK(sig) \
4448 do { \
4449 if (!sigismember(&check_signal_done, sig)) { \
4450 os::Linux::check_signal_handler(sig); \
4451 } \
4452 } while (0)
4453
4454 // This method is a periodic task to check for misbehaving JNI applications
4455 // under CheckJNI, we can add any periodic checks here
4456
4457 void os::run_periodic_checks() {
4458 if (check_signals == false) return;
4459
4460 // SEGV and BUS if overridden could potentially prevent
4461 // generation of hs*.log in the event of a crash, debugging
4462 // such a case can be very challenging, so we absolutely
4463 // check the following for a good measure:
4464 DO_SIGNAL_CHECK(SIGSEGV);
4465 DO_SIGNAL_CHECK(SIGILL);
4466 DO_SIGNAL_CHECK(SIGFPE);
4467 DO_SIGNAL_CHECK(SIGBUS);
4468 DO_SIGNAL_CHECK(SIGPIPE);
4469 DO_SIGNAL_CHECK(SIGXFSZ);
4470 #if defined(PPC64)
4471 DO_SIGNAL_CHECK(SIGTRAP);
4472 #endif
4473
4474 // ReduceSignalUsage allows the user to override these handlers
4475 // see comments at the very top and jvm_solaris.h
4476 if (!ReduceSignalUsage) {
4477 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4478 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4479 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4480 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4481 }
4482
4483 DO_SIGNAL_CHECK(SR_signum);
4484 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4485 }
4486
4487 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4488
4489 static os_sigaction_t os_sigaction = NULL;
4490
4491 void os::Linux::check_signal_handler(int sig) {
4492 char buf[O_BUFLEN];
4493 address jvmHandler = NULL;
4494
4495
4496 struct sigaction act;
4497 if (os_sigaction == NULL) {
4498 // only trust the default sigaction, in case it has been interposed
4499 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4500 if (os_sigaction == NULL) return;
4501 }
4502
4503 os_sigaction(sig, (struct sigaction*)NULL, &act);
4504
4505
4506 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4507
4508 address thisHandler = (act.sa_flags & SA_SIGINFO)
4509 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4510 : CAST_FROM_FN_PTR(address, act.sa_handler);
4511
4512
4513 switch (sig) {
4514 case SIGSEGV:
4515 case SIGBUS:
4516 case SIGFPE:
4517 case SIGPIPE:
4518 case SIGILL:
4519 case SIGXFSZ:
4520 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4521 break;
4522
4523 case SHUTDOWN1_SIGNAL:
4524 case SHUTDOWN2_SIGNAL:
4525 case SHUTDOWN3_SIGNAL:
4526 case BREAK_SIGNAL:
4527 jvmHandler = (address)user_handler();
4528 break;
4529
4530 case INTERRUPT_SIGNAL:
4531 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4532 break;
4533
4534 default:
4535 if (sig == SR_signum) {
4536 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4537 } else {
4538 return;
4539 }
4540 break;
4541 }
4542
4543 if (thisHandler != jvmHandler) {
4544 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4545 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4546 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4547 // No need to check this sig any longer
4548 sigaddset(&check_signal_done, sig);
4549 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4550 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4551 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4552 exception_name(sig, buf, O_BUFLEN));
4553 }
4554 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4555 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4556 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4557 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4558 // No need to check this sig any longer
4559 sigaddset(&check_signal_done, sig);
4560 }
4561
4562 // Dump all the signal
4563 if (sigismember(&check_signal_done, sig)) {
4564 print_signal_handlers(tty, buf, O_BUFLEN);
4565 }
4566 }
4567
4568 extern void report_error(char* file_name, int line_no, char* title,
4569 char* format, ...);
4570
4571 extern bool signal_name(int signo, char* buf, size_t len);
4572
4573 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4574 if (0 < exception_code && exception_code <= SIGRTMAX) {
4575 // signal
4576 if (!signal_name(exception_code, buf, size)) {
4577 jio_snprintf(buf, size, "SIG%d", exception_code);
4578 }
4579 return buf;
4580 } else {
4581 return NULL;
4582 }
4583 }
4584
4585 // this is called _before_ the most of global arguments have been parsed
4586 void os::init(void) {
4587 char dummy; // used to get a guess on initial stack address
4588 // first_hrtime = gethrtime();
4589
4590 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4591
4592 init_random(1234567);
4593
4594 ThreadCritical::initialize();
4595
4596 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4597 if (Linux::page_size() == -1) {
4598 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4599 strerror(errno)));
4600 }
4601 init_page_sizes((size_t) Linux::page_size());
4602
4603 Linux::initialize_system_info();
4604
4605 // main_thread points to the aboriginal thread
4606 Linux::_main_thread = pthread_self();
4607
4608 Linux::clock_init();
4609 initial_time_count = javaTimeNanos();
4610
4611 // pthread_condattr initialization for monotonic clock
4612 int status;
4613 pthread_condattr_t* _condattr = os::Linux::condAttr();
4614 if ((status = pthread_condattr_init(_condattr)) != 0) {
4615 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4616 }
4617 // Only set the clock if CLOCK_MONOTONIC is available
4618 if (os::supports_monotonic_clock()) {
4619 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4620 if (status == EINVAL) {
4621 warning("Unable to use monotonic clock with relative timed-waits" \
4622 " - changes to the time-of-day clock may have adverse affects");
4623 } else {
4624 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4625 }
4626 }
4627 }
4628 // else it defaults to CLOCK_REALTIME
4629
4630 // If the pagesize of the VM is greater than 8K determine the appropriate
4631 // number of initial guard pages. The user can change this with the
4632 // command line arguments, if needed.
4633 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4634 StackYellowPages = 1;
4635 StackRedPages = 1;
4636 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4637 }
4638
4639 // retrieve entry point for pthread_setname_np
4640 Linux::_pthread_setname_np =
4641 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4642
4643 }
4644
4645 // To install functions for atexit system call
4646 extern "C" {
4647 static void perfMemory_exit_helper() {
4648 perfMemory_exit();
4649 }
4650 }
4651
4652 // this is called _after_ the global arguments have been parsed
4653 jint os::init_2(void) {
4654 Linux::fast_thread_clock_init();
4655
4656 // Allocate a single page and mark it as readable for safepoint polling
4657 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4658 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4659
4660 os::set_polling_page(polling_page);
4661
4662 #ifndef PRODUCT
4663 if (Verbose && PrintMiscellaneous) {
4664 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4665 (intptr_t)polling_page);
4666 }
4667 #endif
4668
4669 if (!UseMembar) {
4670 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4671 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4672 os::set_memory_serialize_page(mem_serialize_page);
4673
4674 #ifndef PRODUCT
4675 if (Verbose && PrintMiscellaneous) {
4676 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4677 (intptr_t)mem_serialize_page);
4678 }
4679 #endif
4680 }
4681
4682 // initialize suspend/resume support - must do this before signal_sets_init()
4683 if (SR_initialize() != 0) {
4684 perror("SR_initialize failed");
4685 return JNI_ERR;
4686 }
4687
4688 Linux::signal_sets_init();
4689 Linux::install_signal_handlers();
4690
4691 // Check minimum allowable stack size for thread creation and to initialize
4692 // the java system classes, including StackOverflowError - depends on page
4693 // size. Add a page for compiler2 recursion in main thread.
4694 // Add in 2*BytesPerWord times page size to account for VM stack during
4695 // class initialization depending on 32 or 64 bit VM.
4696 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4697 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4698 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4699
4700 size_t threadStackSizeInBytes = ThreadStackSize * K;
4701 if (threadStackSizeInBytes != 0 &&
4702 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4703 tty->print_cr("\nThe stack size specified is too small, "
4704 "Specify at least %dk",
4705 os::Linux::min_stack_allowed/ K);
4706 return JNI_ERR;
4707 }
4708
4709 // Make the stack size a multiple of the page size so that
4710 // the yellow/red zones can be guarded.
4711 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4712 vm_page_size()));
4713
4714 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4715
4716 #if defined(IA32)
4717 workaround_expand_exec_shield_cs_limit();
4718 #endif
4719
4720 Linux::libpthread_init();
4721 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4722 tty->print_cr("[HotSpot is running with %s, %s]\n",
4723 Linux::glibc_version(), Linux::libpthread_version());
4724 }
4725
4726 if (UseNUMA) {
4727 if (!Linux::libnuma_init()) {
4728 UseNUMA = false;
4729 } else {
4730 if ((Linux::numa_max_node() < 1)) {
4731 // There's only one node(they start from 0), disable NUMA.
4732 UseNUMA = false;
4733 }
4734 }
4735 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4736 // we can make the adaptive lgrp chunk resizing work. If the user specified
4737 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4738 // disable adaptive resizing.
4739 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4740 if (FLAG_IS_DEFAULT(UseNUMA)) {
4741 UseNUMA = false;
4742 } else {
4743 if (FLAG_IS_DEFAULT(UseLargePages) &&
4744 FLAG_IS_DEFAULT(UseSHM) &&
4745 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4746 UseLargePages = false;
4747 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4748 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4749 UseAdaptiveSizePolicy = false;
4750 UseAdaptiveNUMAChunkSizing = false;
4751 }
4752 }
4753 }
4754 if (!UseNUMA && ForceNUMA) {
4755 UseNUMA = true;
4756 }
4757 }
4758
4759 if (MaxFDLimit) {
4760 // set the number of file descriptors to max. print out error
4761 // if getrlimit/setrlimit fails but continue regardless.
4762 struct rlimit nbr_files;
4763 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4764 if (status != 0) {
4765 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4766 perror("os::init_2 getrlimit failed");
4767 }
4768 } else {
4769 nbr_files.rlim_cur = nbr_files.rlim_max;
4770 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4771 if (status != 0) {
4772 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4773 perror("os::init_2 setrlimit failed");
4774 }
4775 }
4776 }
4777 }
4778
4779 // Initialize lock used to serialize thread creation (see os::create_thread)
4780 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4781
4782 // at-exit methods are called in the reverse order of their registration.
4783 // atexit functions are called on return from main or as a result of a
4784 // call to exit(3C). There can be only 32 of these functions registered
4785 // and atexit() does not set errno.
4786
4787 if (PerfAllowAtExitRegistration) {
4788 // only register atexit functions if PerfAllowAtExitRegistration is set.
4789 // atexit functions can be delayed until process exit time, which
4790 // can be problematic for embedded VM situations. Embedded VMs should
4791 // call DestroyJavaVM() to assure that VM resources are released.
4792
4793 // note: perfMemory_exit_helper atexit function may be removed in
4794 // the future if the appropriate cleanup code can be added to the
4795 // VM_Exit VMOperation's doit method.
4796 if (atexit(perfMemory_exit_helper) != 0) {
4797 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4798 }
4799 }
4800
4801 // initialize thread priority policy
4802 prio_init();
4803
4804 return JNI_OK;
4805 }
4806
4807 // Mark the polling page as unreadable
4808 void os::make_polling_page_unreadable(void) {
4809 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4810 fatal("Could not disable polling page");
4811 }
4812 }
4813
4814 // Mark the polling page as readable
4815 void os::make_polling_page_readable(void) {
4816 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4817 fatal("Could not enable polling page");
4818 }
4819 }
4820
4821 int os::active_processor_count() {
4822 // Linux doesn't yet have a (official) notion of processor sets,
4823 // so just return the number of online processors.
4824 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4825 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4826 return online_cpus;
4827 }
4828
4829 void os::set_native_thread_name(const char *name) {
4830 if (Linux::_pthread_setname_np) {
4831 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4832 snprintf(buf, sizeof(buf), "%s", name);
4833 buf[sizeof(buf) - 1] = '\0';
4834 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4835 // ERANGE should not happen; all other errors should just be ignored.
4836 assert(rc != ERANGE, "pthread_setname_np failed");
4837 }
4838 }
4839
4840 bool os::distribute_processes(uint length, uint* distribution) {
4841 // Not yet implemented.
4842 return false;
4843 }
4844
4845 bool os::bind_to_processor(uint processor_id) {
4846 // Not yet implemented.
4847 return false;
4848 }
4849
4850 ///
4851
4852 void os::SuspendedThreadTask::internal_do_task() {
4853 if (do_suspend(_thread->osthread())) {
4854 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4855 do_task(context);
4856 do_resume(_thread->osthread());
4857 }
4858 }
4859
4860 class PcFetcher : public os::SuspendedThreadTask {
4861 public:
4862 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4863 ExtendedPC result();
4864 protected:
4865 void do_task(const os::SuspendedThreadTaskContext& context);
4866 private:
4867 ExtendedPC _epc;
4868 };
4869
4870 ExtendedPC PcFetcher::result() {
4871 guarantee(is_done(), "task is not done yet.");
4872 return _epc;
4873 }
4874
4875 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4876 Thread* thread = context.thread();
4877 OSThread* osthread = thread->osthread();
4878 if (osthread->ucontext() != NULL) {
4879 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4880 } else {
4881 // NULL context is unexpected, double-check this is the VMThread
4882 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4883 }
4884 }
4885
4886 // Suspends the target using the signal mechanism and then grabs the PC before
4887 // resuming the target. Used by the flat-profiler only
4888 ExtendedPC os::get_thread_pc(Thread* thread) {
4889 // Make sure that it is called by the watcher for the VMThread
4890 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4891 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4892
4893 PcFetcher fetcher(thread);
4894 fetcher.run();
4895 return fetcher.result();
4896 }
4897
4898 ////////////////////////////////////////////////////////////////////////////////
4899 // debug support
4900
4901 bool os::find(address addr, outputStream* st) {
4902 Dl_info dlinfo;
4903 memset(&dlinfo, 0, sizeof(dlinfo));
4904 if (dladdr(addr, &dlinfo) != 0) {
4905 st->print(PTR_FORMAT ": ", addr);
4906 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4907 st->print("%s+%#x", dlinfo.dli_sname,
4908 addr - (intptr_t)dlinfo.dli_saddr);
4909 } else if (dlinfo.dli_fbase != NULL) {
4910 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4911 } else {
4912 st->print("<absolute address>");
4913 }
4914 if (dlinfo.dli_fname != NULL) {
4915 st->print(" in %s", dlinfo.dli_fname);
4916 }
4917 if (dlinfo.dli_fbase != NULL) {
4918 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4919 }
4920 st->cr();
4921
4922 if (Verbose) {
4923 // decode some bytes around the PC
4924 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4925 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4926 address lowest = (address) dlinfo.dli_sname;
4927 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4928 if (begin < lowest) begin = lowest;
4929 Dl_info dlinfo2;
4930 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4931 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4932 end = (address) dlinfo2.dli_saddr;
4933 }
4934 Disassembler::decode(begin, end, st);
4935 }
4936 return true;
4937 }
4938 return false;
4939 }
4940
4941 ////////////////////////////////////////////////////////////////////////////////
4942 // misc
4943
4944 // This does not do anything on Linux. This is basically a hook for being
4945 // able to use structured exception handling (thread-local exception filters)
4946 // on, e.g., Win32.
4947 void
4948 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4949 JavaCallArguments* args, Thread* thread) {
4950 f(value, method, args, thread);
4951 }
4952
4953 void os::print_statistics() {
4954 }
4955
4956 int os::message_box(const char* title, const char* message) {
4957 int i;
4958 fdStream err(defaultStream::error_fd());
4959 for (i = 0; i < 78; i++) err.print_raw("=");
4960 err.cr();
4961 err.print_raw_cr(title);
4962 for (i = 0; i < 78; i++) err.print_raw("-");
4963 err.cr();
4964 err.print_raw_cr(message);
4965 for (i = 0; i < 78; i++) err.print_raw("=");
4966 err.cr();
4967
4968 char buf[16];
4969 // Prevent process from exiting upon "read error" without consuming all CPU
4970 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4971
4972 return buf[0] == 'y' || buf[0] == 'Y';
4973 }
4974
4975 int os::stat(const char *path, struct stat *sbuf) {
4976 char pathbuf[MAX_PATH];
4977 if (strlen(path) > MAX_PATH - 1) {
4978 errno = ENAMETOOLONG;
4979 return -1;
4980 }
4981 os::native_path(strcpy(pathbuf, path));
4982 return ::stat(pathbuf, sbuf);
4983 }
4984
4985 bool os::check_heap(bool force) {
4986 return true;
4987 }
4988
4989 // Is a (classpath) directory empty?
4990 bool os::dir_is_empty(const char* path) {
4991 DIR *dir = NULL;
4992 struct dirent *ptr;
4993
4994 dir = opendir(path);
4995 if (dir == NULL) return true;
4996
4997 // Scan the directory
4998 bool result = true;
4999 char buf[sizeof(struct dirent) + MAX_PATH];
5000 while (result && (ptr = ::readdir(dir)) != NULL) {
5001 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5002 result = false;
5003 }
5004 }
5005 closedir(dir);
5006 return result;
5007 }
5008
5009 // This code originates from JDK's sysOpen and open64_w
5010 // from src/solaris/hpi/src/system_md.c
5011
5012 int os::open(const char *path, int oflag, int mode) {
5013 if (strlen(path) > MAX_PATH - 1) {
5014 errno = ENAMETOOLONG;
5015 return -1;
5016 }
5017
5018 // All file descriptors that are opened in the Java process and not
5019 // specifically destined for a subprocess should have the close-on-exec
5020 // flag set. If we don't set it, then careless 3rd party native code
5021 // might fork and exec without closing all appropriate file descriptors
5022 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5023 // turn might:
5024 //
5025 // - cause end-of-file to fail to be detected on some file
5026 // descriptors, resulting in mysterious hangs, or
5027 //
5028 // - might cause an fopen in the subprocess to fail on a system
5029 // suffering from bug 1085341.
5030 //
5031 // (Yes, the default setting of the close-on-exec flag is a Unix
5032 // design flaw)
5033 //
5034 // See:
5035 // 1085341: 32-bit stdio routines should support file descriptors >255
5036 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5037 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5038 //
5039 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5040 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5041 // because it saves a system call and removes a small window where the flag
5042 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5043 // and we fall back to using FD_CLOEXEC (see below).
5044 #ifdef O_CLOEXEC
5045 oflag |= O_CLOEXEC;
5046 #endif
5047
5048 int fd = ::open64(path, oflag, mode);
5049 if (fd == -1) return -1;
5050
5051 //If the open succeeded, the file might still be a directory
5052 {
5053 struct stat64 buf64;
5054 int ret = ::fstat64(fd, &buf64);
5055 int st_mode = buf64.st_mode;
5056
5057 if (ret != -1) {
5058 if ((st_mode & S_IFMT) == S_IFDIR) {
5059 errno = EISDIR;
5060 ::close(fd);
5061 return -1;
5062 }
5063 } else {
5064 ::close(fd);
5065 return -1;
5066 }
5067 }
5068
5069 #ifdef FD_CLOEXEC
5070 // Validate that the use of the O_CLOEXEC flag on open above worked.
5071 // With recent kernels, we will perform this check exactly once.
5072 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5073 if (!O_CLOEXEC_is_known_to_work) {
5074 int flags = ::fcntl(fd, F_GETFD);
5075 if (flags != -1) {
5076 if ((flags & FD_CLOEXEC) != 0)
5077 O_CLOEXEC_is_known_to_work = 1;
5078 else
5079 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5080 }
5081 }
5082 #endif
5083
5084 return fd;
5085 }
5086
5087
5088 // create binary file, rewriting existing file if required
5089 int os::create_binary_file(const char* path, bool rewrite_existing) {
5090 int oflags = O_WRONLY | O_CREAT;
5091 if (!rewrite_existing) {
5092 oflags |= O_EXCL;
5093 }
5094 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5095 }
5096
5097 // return current position of file pointer
5098 jlong os::current_file_offset(int fd) {
5099 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5100 }
5101
5102 // move file pointer to the specified offset
5103 jlong os::seek_to_file_offset(int fd, jlong offset) {
5104 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5105 }
5106
5107 // This code originates from JDK's sysAvailable
5108 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5109
5110 int os::available(int fd, jlong *bytes) {
5111 jlong cur, end;
5112 int mode;
5113 struct stat64 buf64;
5114
5115 if (::fstat64(fd, &buf64) >= 0) {
5116 mode = buf64.st_mode;
5117 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5118 // XXX: is the following call interruptible? If so, this might
5119 // need to go through the INTERRUPT_IO() wrapper as for other
5120 // blocking, interruptible calls in this file.
5121 int n;
5122 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5123 *bytes = n;
5124 return 1;
5125 }
5126 }
5127 }
5128 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5129 return 0;
5130 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5131 return 0;
5132 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5133 return 0;
5134 }
5135 *bytes = end - cur;
5136 return 1;
5137 }
5138
5139 // Map a block of memory.
5140 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5141 char *addr, size_t bytes, bool read_only,
5142 bool allow_exec) {
5143 int prot;
5144 int flags = MAP_PRIVATE;
5145
5146 if (read_only) {
5147 prot = PROT_READ;
5148 } else {
5149 prot = PROT_READ | PROT_WRITE;
5150 }
5151
5152 if (allow_exec) {
5153 prot |= PROT_EXEC;
5154 }
5155
5156 if (addr != NULL) {
5157 flags |= MAP_FIXED;
5158 }
5159
5160 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5161 fd, file_offset);
5162 if (mapped_address == MAP_FAILED) {
5163 return NULL;
5164 }
5165 return mapped_address;
5166 }
5167
5168
5169 // Remap a block of memory.
5170 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5171 char *addr, size_t bytes, bool read_only,
5172 bool allow_exec) {
5173 // same as map_memory() on this OS
5174 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5175 allow_exec);
5176 }
5177
5178
5179 // Unmap a block of memory.
5180 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5181 return munmap(addr, bytes) == 0;
5182 }
5183
5184 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5185
5186 static clockid_t thread_cpu_clockid(Thread* thread) {
5187 pthread_t tid = thread->osthread()->pthread_id();
5188 clockid_t clockid;
5189
5190 // Get thread clockid
5191 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5192 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5193 return clockid;
5194 }
5195
5196 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5197 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5198 // of a thread.
5199 //
5200 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5201 // the fast estimate available on the platform.
5202
5203 jlong os::current_thread_cpu_time() {
5204 if (os::Linux::supports_fast_thread_cpu_time()) {
5205 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5206 } else {
5207 // return user + sys since the cost is the same
5208 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5209 }
5210 }
5211
5212 jlong os::thread_cpu_time(Thread* thread) {
5213 // consistent with what current_thread_cpu_time() returns
5214 if (os::Linux::supports_fast_thread_cpu_time()) {
5215 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5216 } else {
5217 return slow_thread_cpu_time(thread, true /* user + sys */);
5218 }
5219 }
5220
5221 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5222 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5223 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5224 } else {
5225 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5226 }
5227 }
5228
5229 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5230 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5231 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5232 } else {
5233 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5234 }
5235 }
5236
5237 // -1 on error.
5238 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5239 pid_t tid = thread->osthread()->thread_id();
5240 char *s;
5241 char stat[2048];
5242 int statlen;
5243 char proc_name[64];
5244 int count;
5245 long sys_time, user_time;
5246 char cdummy;
5247 int idummy;
5248 long ldummy;
5249 FILE *fp;
5250
5251 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5252 fp = fopen(proc_name, "r");
5253 if (fp == NULL) return -1;
5254 statlen = fread(stat, 1, 2047, fp);
5255 stat[statlen] = '\0';
5256 fclose(fp);
5257
5258 // Skip pid and the command string. Note that we could be dealing with
5259 // weird command names, e.g. user could decide to rename java launcher
5260 // to "java 1.4.2 :)", then the stat file would look like
5261 // 1234 (java 1.4.2 :)) R ... ...
5262 // We don't really need to know the command string, just find the last
5263 // occurrence of ")" and then start parsing from there. See bug 4726580.
5264 s = strrchr(stat, ')');
5265 if (s == NULL) return -1;
5266
5267 // Skip blank chars
5268 do { s++; } while (s && isspace(*s));
5269
5270 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5271 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5272 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5273 &user_time, &sys_time);
5274 if (count != 13) return -1;
5275 if (user_sys_cpu_time) {
5276 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5277 } else {
5278 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5279 }
5280 }
5281
5282 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5283 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5284 info_ptr->may_skip_backward = false; // elapsed time not wall time
5285 info_ptr->may_skip_forward = false; // elapsed time not wall time
5286 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5287 }
5288
5289 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5290 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5291 info_ptr->may_skip_backward = false; // elapsed time not wall time
5292 info_ptr->may_skip_forward = false; // elapsed time not wall time
5293 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5294 }
5295
5296 bool os::is_thread_cpu_time_supported() {
5297 return true;
5298 }
5299
5300 // System loadavg support. Returns -1 if load average cannot be obtained.
5301 // Linux doesn't yet have a (official) notion of processor sets,
5302 // so just return the system wide load average.
5303 int os::loadavg(double loadavg[], int nelem) {
5304 return ::getloadavg(loadavg, nelem);
5305 }
5306
5307 void os::pause() {
5308 char filename[MAX_PATH];
5309 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5310 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5311 } else {
5312 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5313 }
5314
5315 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5316 if (fd != -1) {
5317 struct stat buf;
5318 ::close(fd);
5319 while (::stat(filename, &buf) == 0) {
5320 (void)::poll(NULL, 0, 100);
5321 }
5322 } else {
5323 jio_fprintf(stderr,
5324 "Could not open pause file '%s', continuing immediately.\n", filename);
5325 }
5326 }
5327
5328
5329 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5330 // comment paragraphs are worth repeating here:
5331 //
5332 // Assumption:
5333 // Only one parker can exist on an event, which is why we allocate
5334 // them per-thread. Multiple unparkers can coexist.
5335 //
5336 // _Event serves as a restricted-range semaphore.
5337 // -1 : thread is blocked, i.e. there is a waiter
5338 // 0 : neutral: thread is running or ready,
5339 // could have been signaled after a wait started
5340 // 1 : signaled - thread is running or ready
5341 //
5342 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5343 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5344 // For specifics regarding the bug see GLIBC BUGID 261237 :
5345 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5346 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5347 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5348 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5349 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5350 // and monitorenter when we're using 1-0 locking. All those operations may result in
5351 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5352 // of libpthread avoids the problem, but isn't practical.
5353 //
5354 // Possible remedies:
5355 //
5356 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5357 // This is palliative and probabilistic, however. If the thread is preempted
5358 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5359 // than the minimum period may have passed, and the abstime may be stale (in the
5360 // past) resultin in a hang. Using this technique reduces the odds of a hang
5361 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5362 //
5363 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5364 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5365 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5366 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5367 // thread.
5368 //
5369 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5370 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5371 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5372 // This also works well. In fact it avoids kernel-level scalability impediments
5373 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5374 // timers in a graceful fashion.
5375 //
5376 // 4. When the abstime value is in the past it appears that control returns
5377 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5378 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5379 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5380 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5381 // It may be possible to avoid reinitialization by checking the return
5382 // value from pthread_cond_timedwait(). In addition to reinitializing the
5383 // condvar we must establish the invariant that cond_signal() is only called
5384 // within critical sections protected by the adjunct mutex. This prevents
5385 // cond_signal() from "seeing" a condvar that's in the midst of being
5386 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5387 // desirable signal-after-unlock optimization that avoids futile context switching.
5388 //
5389 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5390 // structure when a condvar is used or initialized. cond_destroy() would
5391 // release the helper structure. Our reinitialize-after-timedwait fix
5392 // put excessive stress on malloc/free and locks protecting the c-heap.
5393 //
5394 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5395 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5396 // and only enabling the work-around for vulnerable environments.
5397
5398 // utility to compute the abstime argument to timedwait:
5399 // millis is the relative timeout time
5400 // abstime will be the absolute timeout time
5401 // TODO: replace compute_abstime() with unpackTime()
5402
5403 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5404 if (millis < 0) millis = 0;
5405
5406 jlong seconds = millis / 1000;
5407 millis %= 1000;
5408 if (seconds > 50000000) { // see man cond_timedwait(3T)
5409 seconds = 50000000;
5410 }
5411
5412 if (os::supports_monotonic_clock()) {
5413 struct timespec now;
5414 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5415 assert_status(status == 0, status, "clock_gettime");
5416 abstime->tv_sec = now.tv_sec + seconds;
5417 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5418 if (nanos >= NANOSECS_PER_SEC) {
5419 abstime->tv_sec += 1;
5420 nanos -= NANOSECS_PER_SEC;
5421 }
5422 abstime->tv_nsec = nanos;
5423 } else {
5424 struct timeval now;
5425 int status = gettimeofday(&now, NULL);
5426 assert(status == 0, "gettimeofday");
5427 abstime->tv_sec = now.tv_sec + seconds;
5428 long usec = now.tv_usec + millis * 1000;
5429 if (usec >= 1000000) {
5430 abstime->tv_sec += 1;
5431 usec -= 1000000;
5432 }
5433 abstime->tv_nsec = usec * 1000;
5434 }
5435 return abstime;
5436 }
5437
5438 void os::PlatformEvent::park() { // AKA "down()"
5439 // Transitions for _Event:
5440 // -1 => -1 : illegal
5441 // 1 => 0 : pass - return immediately
5442 // 0 => -1 : block; then set _Event to 0 before returning
5443
5444 // Invariant: Only the thread associated with the Event/PlatformEvent
5445 // may call park().
5446 // TODO: assert that _Assoc != NULL or _Assoc == Self
5447 assert(_nParked == 0, "invariant");
5448
5449 int v;
5450 for (;;) {
5451 v = _Event;
5452 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5453 }
5454 guarantee(v >= 0, "invariant");
5455 if (v == 0) {
5456 // Do this the hard way by blocking ...
5457 int status = pthread_mutex_lock(_mutex);
5458 assert_status(status == 0, status, "mutex_lock");
5459 guarantee(_nParked == 0, "invariant");
5460 ++_nParked;
5461 while (_Event < 0) {
5462 status = pthread_cond_wait(_cond, _mutex);
5463 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5464 // Treat this the same as if the wait was interrupted
5465 if (status == ETIME) { status = EINTR; }
5466 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5467 }
5468 --_nParked;
5469
5470 _Event = 0;
5471 status = pthread_mutex_unlock(_mutex);
5472 assert_status(status == 0, status, "mutex_unlock");
5473 // Paranoia to ensure our locked and lock-free paths interact
5474 // correctly with each other.
5475 OrderAccess::fence();
5476 }
5477 guarantee(_Event >= 0, "invariant");
5478 }
5479
5480 int os::PlatformEvent::park(jlong millis) {
5481 // Transitions for _Event:
5482 // -1 => -1 : illegal
5483 // 1 => 0 : pass - return immediately
5484 // 0 => -1 : block; then set _Event to 0 before returning
5485
5486 guarantee(_nParked == 0, "invariant");
5487
5488 int v;
5489 for (;;) {
5490 v = _Event;
5491 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5492 }
5493 guarantee(v >= 0, "invariant");
5494 if (v != 0) return OS_OK;
5495
5496 // We do this the hard way, by blocking the thread.
5497 // Consider enforcing a minimum timeout value.
5498 struct timespec abst;
5499 compute_abstime(&abst, millis);
5500
5501 int ret = OS_TIMEOUT;
5502 int status = pthread_mutex_lock(_mutex);
5503 assert_status(status == 0, status, "mutex_lock");
5504 guarantee(_nParked == 0, "invariant");
5505 ++_nParked;
5506
5507 // Object.wait(timo) will return because of
5508 // (a) notification
5509 // (b) timeout
5510 // (c) thread.interrupt
5511 //
5512 // Thread.interrupt and object.notify{All} both call Event::set.
5513 // That is, we treat thread.interrupt as a special case of notification.
5514 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5515 // We assume all ETIME returns are valid.
5516 //
5517 // TODO: properly differentiate simultaneous notify+interrupt.
5518 // In that case, we should propagate the notify to another waiter.
5519
5520 while (_Event < 0) {
5521 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5522 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5523 pthread_cond_destroy(_cond);
5524 pthread_cond_init(_cond, os::Linux::condAttr());
5525 }
5526 assert_status(status == 0 || status == EINTR ||
5527 status == ETIME || status == ETIMEDOUT,
5528 status, "cond_timedwait");
5529 if (!FilterSpuriousWakeups) break; // previous semantics
5530 if (status == ETIME || status == ETIMEDOUT) break;
5531 // We consume and ignore EINTR and spurious wakeups.
5532 }
5533 --_nParked;
5534 if (_Event >= 0) {
5535 ret = OS_OK;
5536 }
5537 _Event = 0;
5538 status = pthread_mutex_unlock(_mutex);
5539 assert_status(status == 0, status, "mutex_unlock");
5540 assert(_nParked == 0, "invariant");
5541 // Paranoia to ensure our locked and lock-free paths interact
5542 // correctly with each other.
5543 OrderAccess::fence();
5544 return ret;
5545 }
5546
5547 void os::PlatformEvent::unpark() {
5548 // Transitions for _Event:
5549 // 0 => 1 : just return
5550 // 1 => 1 : just return
5551 // -1 => either 0 or 1; must signal target thread
5552 // That is, we can safely transition _Event from -1 to either
5553 // 0 or 1.
5554 // See also: "Semaphores in Plan 9" by Mullender & Cox
5555 //
5556 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5557 // that it will take two back-to-back park() calls for the owning
5558 // thread to block. This has the benefit of forcing a spurious return
5559 // from the first park() call after an unpark() call which will help
5560 // shake out uses of park() and unpark() without condition variables.
5561
5562 if (Atomic::xchg(1, &_Event) >= 0) return;
5563
5564 // Wait for the thread associated with the event to vacate
5565 int status = pthread_mutex_lock(_mutex);
5566 assert_status(status == 0, status, "mutex_lock");
5567 int AnyWaiters = _nParked;
5568 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5569 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5570 AnyWaiters = 0;
5571 pthread_cond_signal(_cond);
5572 }
5573 status = pthread_mutex_unlock(_mutex);
5574 assert_status(status == 0, status, "mutex_unlock");
5575 if (AnyWaiters != 0) {
5576 // Note that we signal() *after* dropping the lock for "immortal" Events.
5577 // This is safe and avoids a common class of futile wakeups. In rare
5578 // circumstances this can cause a thread to return prematurely from
5579 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5580 // will simply re-test the condition and re-park itself.
5581 // This provides particular benefit if the underlying platform does not
5582 // provide wait morphing.
5583 status = pthread_cond_signal(_cond);
5584 assert_status(status == 0, status, "cond_signal");
5585 }
5586 }
5587
5588
5589 // JSR166
5590 // -------------------------------------------------------
5591
5592 // The solaris and linux implementations of park/unpark are fairly
5593 // conservative for now, but can be improved. They currently use a
5594 // mutex/condvar pair, plus a a count.
5595 // Park decrements count if > 0, else does a condvar wait. Unpark
5596 // sets count to 1 and signals condvar. Only one thread ever waits
5597 // on the condvar. Contention seen when trying to park implies that someone
5598 // is unparking you, so don't wait. And spurious returns are fine, so there
5599 // is no need to track notifications.
5600
5601 // This code is common to linux and solaris and will be moved to a
5602 // common place in dolphin.
5603 //
5604 // The passed in time value is either a relative time in nanoseconds
5605 // or an absolute time in milliseconds. Either way it has to be unpacked
5606 // into suitable seconds and nanoseconds components and stored in the
5607 // given timespec structure.
5608 // Given time is a 64-bit value and the time_t used in the timespec is only
5609 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5610 // overflow if times way in the future are given. Further on Solaris versions
5611 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5612 // number of seconds, in abstime, is less than current_time + 100,000,000.
5613 // As it will be 28 years before "now + 100000000" will overflow we can
5614 // ignore overflow and just impose a hard-limit on seconds using the value
5615 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5616 // years from "now".
5617
5618 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5619 assert(time > 0, "convertTime");
5620 time_t max_secs = 0;
5621
5622 if (!os::supports_monotonic_clock() || isAbsolute) {
5623 struct timeval now;
5624 int status = gettimeofday(&now, NULL);
5625 assert(status == 0, "gettimeofday");
5626
5627 max_secs = now.tv_sec + MAX_SECS;
5628
5629 if (isAbsolute) {
5630 jlong secs = time / 1000;
5631 if (secs > max_secs) {
5632 absTime->tv_sec = max_secs;
5633 } else {
5634 absTime->tv_sec = secs;
5635 }
5636 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5637 } else {
5638 jlong secs = time / NANOSECS_PER_SEC;
5639 if (secs >= MAX_SECS) {
5640 absTime->tv_sec = max_secs;
5641 absTime->tv_nsec = 0;
5642 } else {
5643 absTime->tv_sec = now.tv_sec + secs;
5644 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5645 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5646 absTime->tv_nsec -= NANOSECS_PER_SEC;
5647 ++absTime->tv_sec; // note: this must be <= max_secs
5648 }
5649 }
5650 }
5651 } else {
5652 // must be relative using monotonic clock
5653 struct timespec now;
5654 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5655 assert_status(status == 0, status, "clock_gettime");
5656 max_secs = now.tv_sec + MAX_SECS;
5657 jlong secs = time / NANOSECS_PER_SEC;
5658 if (secs >= MAX_SECS) {
5659 absTime->tv_sec = max_secs;
5660 absTime->tv_nsec = 0;
5661 } else {
5662 absTime->tv_sec = now.tv_sec + secs;
5663 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5664 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5665 absTime->tv_nsec -= NANOSECS_PER_SEC;
5666 ++absTime->tv_sec; // note: this must be <= max_secs
5667 }
5668 }
5669 }
5670 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5671 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5672 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5673 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5674 }
5675
5676 void Parker::park(bool isAbsolute, jlong time) {
5677 // Ideally we'd do something useful while spinning, such
5678 // as calling unpackTime().
5679
5680 // Optional fast-path check:
5681 // Return immediately if a permit is available.
5682 // We depend on Atomic::xchg() having full barrier semantics
5683 // since we are doing a lock-free update to _counter.
5684 if (Atomic::xchg(0, &_counter) > 0) return;
5685
5686 Thread* thread = Thread::current();
5687 assert(thread->is_Java_thread(), "Must be JavaThread");
5688 JavaThread *jt = (JavaThread *)thread;
5689
5690 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5691 // Check interrupt before trying to wait
5692 if (Thread::is_interrupted(thread, false)) {
5693 return;
5694 }
5695
5696 // Next, demultiplex/decode time arguments
5697 timespec absTime;
5698 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5699 return;
5700 }
5701 if (time > 0) {
5702 unpackTime(&absTime, isAbsolute, time);
5703 }
5704
5705
5706 // Enter safepoint region
5707 // Beware of deadlocks such as 6317397.
5708 // The per-thread Parker:: mutex is a classic leaf-lock.
5709 // In particular a thread must never block on the Threads_lock while
5710 // holding the Parker:: mutex. If safepoints are pending both the
5711 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5712 ThreadBlockInVM tbivm(jt);
5713
5714 // Don't wait if cannot get lock since interference arises from
5715 // unblocking. Also. check interrupt before trying wait
5716 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5717 return;
5718 }
5719
5720 int status;
5721 if (_counter > 0) { // no wait needed
5722 _counter = 0;
5723 status = pthread_mutex_unlock(_mutex);
5724 assert(status == 0, "invariant");
5725 // Paranoia to ensure our locked and lock-free paths interact
5726 // correctly with each other and Java-level accesses.
5727 OrderAccess::fence();
5728 return;
5729 }
5730
5731 #ifdef ASSERT
5732 // Don't catch signals while blocked; let the running threads have the signals.
5733 // (This allows a debugger to break into the running thread.)
5734 sigset_t oldsigs;
5735 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5736 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5737 #endif
5738
5739 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5740 jt->set_suspend_equivalent();
5741 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5742
5743 assert(_cur_index == -1, "invariant");
5744 if (time == 0) {
5745 _cur_index = REL_INDEX; // arbitrary choice when not timed
5746 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5747 } else {
5748 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5749 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5750 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5751 pthread_cond_destroy(&_cond[_cur_index]);
5752 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5753 }
5754 }
5755 _cur_index = -1;
5756 assert_status(status == 0 || status == EINTR ||
5757 status == ETIME || status == ETIMEDOUT,
5758 status, "cond_timedwait");
5759
5760 #ifdef ASSERT
5761 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5762 #endif
5763
5764 _counter = 0;
5765 status = pthread_mutex_unlock(_mutex);
5766 assert_status(status == 0, status, "invariant");
5767 // Paranoia to ensure our locked and lock-free paths interact
5768 // correctly with each other and Java-level accesses.
5769 OrderAccess::fence();
5770
5771 // If externally suspended while waiting, re-suspend
5772 if (jt->handle_special_suspend_equivalent_condition()) {
5773 jt->java_suspend_self();
5774 }
5775 }
5776
5777 void Parker::unpark() {
5778 int status = pthread_mutex_lock(_mutex);
5779 assert(status == 0, "invariant");
5780 const int s = _counter;
5781 _counter = 1;
5782 if (s < 1) {
5783 // thread might be parked
5784 if (_cur_index != -1) {
5785 // thread is definitely parked
5786 if (WorkAroundNPTLTimedWaitHang) {
5787 status = pthread_cond_signal(&_cond[_cur_index]);
5788 assert(status == 0, "invariant");
5789 status = pthread_mutex_unlock(_mutex);
5790 assert(status == 0, "invariant");
5791 } else {
5792 status = pthread_mutex_unlock(_mutex);
5793 assert(status == 0, "invariant");
5794 status = pthread_cond_signal(&_cond[_cur_index]);
5795 assert(status == 0, "invariant");
5796 }
5797 } else {
5798 pthread_mutex_unlock(_mutex);
5799 assert(status == 0, "invariant");
5800 }
5801 } else {
5802 pthread_mutex_unlock(_mutex);
5803 assert(status == 0, "invariant");
5804 }
5805 }
5806
5807
5808 extern char** environ;
5809
5810 // Run the specified command in a separate process. Return its exit value,
5811 // or -1 on failure (e.g. can't fork a new process).
5812 // Unlike system(), this function can be called from signal handler. It
5813 // doesn't block SIGINT et al.
5814 int os::fork_and_exec(char* cmd) {
5815 const char * argv[4] = {"sh", "-c", cmd, NULL};
5816
5817 pid_t pid = fork();
5818
5819 if (pid < 0) {
5820 // fork failed
5821 return -1;
5822
5823 } else if (pid == 0) {
5824 // child process
5825
5826 execve("/bin/sh", (char* const*)argv, environ);
5827
5828 // execve failed
5829 _exit(-1);
5830
5831 } else {
5832 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5833 // care about the actual exit code, for now.
5834
5835 int status;
5836
5837 // Wait for the child process to exit. This returns immediately if
5838 // the child has already exited. */
5839 while (waitpid(pid, &status, 0) < 0) {
5840 switch (errno) {
5841 case ECHILD: return 0;
5842 case EINTR: break;
5843 default: return -1;
5844 }
5845 }
5846
5847 if (WIFEXITED(status)) {
5848 // The child exited normally; get its exit code.
5849 return WEXITSTATUS(status);
5850 } else if (WIFSIGNALED(status)) {
5851 // The child exited because of a signal
5852 // The best value to return is 0x80 + signal number,
5853 // because that is what all Unix shells do, and because
5854 // it allows callers to distinguish between process exit and
5855 // process death by signal.
5856 return 0x80 + WTERMSIG(status);
5857 } else {
5858 // Unknown exit code; pass it through
5859 return status;
5860 }
5861 }
5862 }
5863
5864 // is_headless_jre()
5865 //
5866 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5867 // in order to report if we are running in a headless jre
5868 //
5869 // Since JDK8 xawt/libmawt.so was moved into the same directory
5870 // as libawt.so, and renamed libawt_xawt.so
5871 //
5872 bool os::is_headless_jre() {
5873 struct stat statbuf;
5874 char buf[MAXPATHLEN];
5875 char libmawtpath[MAXPATHLEN];
5876 const char *xawtstr = "/xawt/libmawt.so";
5877 const char *new_xawtstr = "/libawt_xawt.so";
5878 char *p;
5879
5880 // Get path to libjvm.so
5881 os::jvm_path(buf, sizeof(buf));
5882
5883 // Get rid of libjvm.so
5884 p = strrchr(buf, '/');
5885 if (p == NULL) {
5886 return false;
5887 } else {
5888 *p = '\0';
5889 }
5890
5891 // Get rid of client or server
5892 p = strrchr(buf, '/');
5893 if (p == NULL) {
5894 return false;
5895 } else {
5896 *p = '\0';
5897 }
5898
5899 // check xawt/libmawt.so
5900 strcpy(libmawtpath, buf);
5901 strcat(libmawtpath, xawtstr);
5902 if (::stat(libmawtpath, &statbuf) == 0) return false;
5903
5904 // check libawt_xawt.so
5905 strcpy(libmawtpath, buf);
5906 strcat(libmawtpath, new_xawtstr);
5907 if (::stat(libmawtpath, &statbuf) == 0) return false;
5908
5909 return true;
5910 }
5911
5912 // Get the default path to the core file
5913 // Returns the length of the string
5914 int os::get_core_path(char* buffer, size_t bufferSize) {
5915 /*
5916 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5917 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5918 */
5919 const int core_pattern_len = 129;
5920 char core_pattern[core_pattern_len] = {0};
5921
5922 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5923 if (core_pattern_file != -1) {
5924 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5925 ::close(core_pattern_file);
5926
5927 if (ret > 0) {
5928 char *last_char = core_pattern + strlen(core_pattern) - 1;
5929
5930 if (*last_char == '\n') {
5931 *last_char = '\0';
5932 }
5933 }
5934 }
5935
5936 if (strlen(core_pattern) == 0) {
5937 return -1;
5938 }
5939
5940 char *pid_pos = strstr(core_pattern, "%p");
5941 int written;
5942
5943 if (core_pattern[0] == '/') {
5944 written = jio_snprintf(buffer, bufferSize, core_pattern);
5945 } else {
5946 char cwd[PATH_MAX];
5947
5948 const char* p = get_current_directory(cwd, PATH_MAX);
5949 if (p == NULL) {
5950 return -1;
5951 }
5952
5953 if (core_pattern[0] == '|') {
5954 written = jio_snprintf(buffer, bufferSize,
5955 "\"%s\" (or dumping to %s/core.%d)",
5956 &core_pattern[1], p, current_process_id());
5957 } else {
5958 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5959 }
5960 }
5961
5962 if (written < 0) {
5963 return -1;
5964 }
5965
5966 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5967 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5968
5969 if (core_uses_pid_file != -1) {
5970 char core_uses_pid = 0;
5971 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5972 ::close(core_uses_pid_file);
5973
5974 if (core_uses_pid == '1') {
5975 jio_snprintf(buffer + written, bufferSize - written,
5976 ".%d", current_process_id());
5977 }
5978 }
5979 }
5980
5981 return strlen(buffer);
5982 }
5983
5984 /////////////// Unit tests ///////////////
5985
5986 #ifndef PRODUCT
5987
5988 #define test_log(...) \
5989 do { \
5990 if (VerboseInternalVMTests) { \
5991 tty->print_cr(__VA_ARGS__); \
5992 tty->flush(); \
5993 } \
5994 } while (false)
5995
5996 class TestReserveMemorySpecial : AllStatic {
5997 public:
5998 static void small_page_write(void* addr, size_t size) {
5999 size_t page_size = os::vm_page_size();
6000
6001 char* end = (char*)addr + size;
6002 for (char* p = (char*)addr; p < end; p += page_size) {
6003 *p = 1;
6004 }
6005 }
6006
6007 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6008 if (!UseHugeTLBFS) {
6009 return;
6010 }
6011
6012 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6013
6014 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6015
6016 if (addr != NULL) {
6017 small_page_write(addr, size);
6018
6019 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6020 }
6021 }
6022
6023 static void test_reserve_memory_special_huge_tlbfs_only() {
6024 if (!UseHugeTLBFS) {
6025 return;
6026 }
6027
6028 size_t lp = os::large_page_size();
6029
6030 for (size_t size = lp; size <= lp * 10; size += lp) {
6031 test_reserve_memory_special_huge_tlbfs_only(size);
6032 }
6033 }
6034
6035 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6036 size_t lp = os::large_page_size();
6037 size_t ag = os::vm_allocation_granularity();
6038
6039 // sizes to test
6040 const size_t sizes[] = {
6041 lp, lp + ag, lp + lp / 2, lp * 2,
6042 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6043 lp * 10, lp * 10 + lp / 2
6044 };
6045 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6046
6047 // For each size/alignment combination, we test three scenarios:
6048 // 1) with req_addr == NULL
6049 // 2) with a non-null req_addr at which we expect to successfully allocate
6050 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6051 // expect the allocation to either fail or to ignore req_addr
6052
6053 // Pre-allocate two areas; they shall be as large as the largest allocation
6054 // and aligned to the largest alignment we will be testing.
6055 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6056 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6057 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6058 -1, 0);
6059 assert(mapping1 != MAP_FAILED, "should work");
6060
6061 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6062 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6063 -1, 0);
6064 assert(mapping2 != MAP_FAILED, "should work");
6065
6066 // Unmap the first mapping, but leave the second mapping intact: the first
6067 // mapping will serve as a value for a "good" req_addr (case 2). The second
6068 // mapping, still intact, as "bad" req_addr (case 3).
6069 ::munmap(mapping1, mapping_size);
6070
6071 // Case 1
6072 test_log("%s, req_addr NULL:", __FUNCTION__);
6073 test_log("size align result");
6074
6075 for (int i = 0; i < num_sizes; i++) {
6076 const size_t size = sizes[i];
6077 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6078 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6079 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6080 size, alignment, p, (p != NULL ? "" : "(failed)"));
6081 if (p != NULL) {
6082 assert(is_ptr_aligned(p, alignment), "must be");
6083 small_page_write(p, size);
6084 os::Linux::release_memory_special_huge_tlbfs(p, size);
6085 }
6086 }
6087 }
6088
6089 // Case 2
6090 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6091 test_log("size align req_addr result");
6092
6093 for (int i = 0; i < num_sizes; i++) {
6094 const size_t size = sizes[i];
6095 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6096 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6097 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6098 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6099 size, alignment, req_addr, p,
6100 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6101 if (p != NULL) {
6102 assert(p == req_addr, "must be");
6103 small_page_write(p, size);
6104 os::Linux::release_memory_special_huge_tlbfs(p, size);
6105 }
6106 }
6107 }
6108
6109 // Case 3
6110 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6111 test_log("size align req_addr result");
6112
6113 for (int i = 0; i < num_sizes; i++) {
6114 const size_t size = sizes[i];
6115 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6116 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6117 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6118 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6119 size, alignment, req_addr, p,
6120 ((p != NULL ? "" : "(failed)")));
6121 // as the area around req_addr contains already existing mappings, the API should always
6122 // return NULL (as per contract, it cannot return another address)
6123 assert(p == NULL, "must be");
6124 }
6125 }
6126
6127 ::munmap(mapping2, mapping_size);
6128
6129 }
6130
6131 static void test_reserve_memory_special_huge_tlbfs() {
6132 if (!UseHugeTLBFS) {
6133 return;
6134 }
6135
6136 test_reserve_memory_special_huge_tlbfs_only();
6137 test_reserve_memory_special_huge_tlbfs_mixed();
6138 }
6139
6140 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6141 if (!UseSHM) {
6142 return;
6143 }
6144
6145 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6146
6147 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6148
6149 if (addr != NULL) {
6150 assert(is_ptr_aligned(addr, alignment), "Check");
6151 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6152
6153 small_page_write(addr, size);
6154
6155 os::Linux::release_memory_special_shm(addr, size);
6156 }
6157 }
6158
6159 static void test_reserve_memory_special_shm() {
6160 size_t lp = os::large_page_size();
6161 size_t ag = os::vm_allocation_granularity();
6162
6163 for (size_t size = ag; size < lp * 3; size += ag) {
6164 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6165 test_reserve_memory_special_shm(size, alignment);
6166 }
6167 }
6168 }
6169
6170 static void test() {
6171 test_reserve_memory_special_huge_tlbfs();
6172 test_reserve_memory_special_shm();
6173 }
6174 };
6175
6176 void TestReserveMemorySpecial_test() {
6177 TestReserveMemorySpecial::test();
6178 }
6179
6180 #endif