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 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
110 // getrusage() is prepared to handle the associated failure.
111 #ifndef RUSAGE_THREAD
112 #define RUSAGE_THREAD (1) /* only the calling thread */
113 #endif
114
115 #define MAX_PATH (2 * K)
116
117 #define MAX_SECS 100000000
118
119 // for timer info max values which include all bits
120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
121
122 #define LARGEPAGES_BIT (1 << 6)
123 ////////////////////////////////////////////////////////////////////////////////
124 // global variables
125 julong os::Linux::_physical_memory = 0;
126
127 address os::Linux::_initial_thread_stack_bottom = NULL;
128 uintptr_t os::Linux::_initial_thread_stack_size = 0;
129
130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
132 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
133 Mutex* os::Linux::_createThread_lock = NULL;
134 pthread_t os::Linux::_main_thread;
135 int os::Linux::_page_size = -1;
136 const int os::Linux::_vm_default_page_size = (8 * K);
137 bool os::Linux::_supports_fast_thread_cpu_time = false;
138 const char * os::Linux::_glibc_version = NULL;
139 const char * os::Linux::_libpthread_version = NULL;
140 pthread_condattr_t os::Linux::_condattr[1];
141
142 static jlong initial_time_count=0;
143
144 static int clock_tics_per_sec = 100;
145
146 // For diagnostics to print a message once. see run_periodic_checks
147 static sigset_t check_signal_done;
148 static bool check_signals = true;
149
150 // Signal number used to suspend/resume a thread
151
152 // do not use any signal number less than SIGSEGV, see 4355769
153 static int SR_signum = SIGUSR2;
154 sigset_t SR_sigset;
155
156 // Declarations
157 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
158
159 // utility functions
160
161 static int SR_initialize();
162
163 julong os::available_memory() {
164 return Linux::available_memory();
165 }
166
167 julong os::Linux::available_memory() {
168 // values in struct sysinfo are "unsigned long"
169 struct sysinfo si;
170 sysinfo(&si);
171
172 return (julong)si.freeram * si.mem_unit;
173 }
174
175 julong os::physical_memory() {
176 return Linux::physical_memory();
177 }
178
179 // Return true if user is running as root.
180
181 bool os::have_special_privileges() {
182 static bool init = false;
183 static bool privileges = false;
184 if (!init) {
185 privileges = (getuid() != geteuid()) || (getgid() != getegid());
186 init = true;
187 }
188 return privileges;
189 }
190
191
192 #ifndef SYS_gettid
193 // i386: 224, ia64: 1105, amd64: 186, sparc 143
194 #ifdef __ia64__
195 #define SYS_gettid 1105
196 #else
197 #ifdef __i386__
198 #define SYS_gettid 224
199 #else
200 #ifdef __amd64__
201 #define SYS_gettid 186
202 #else
203 #ifdef __sparc__
204 #define SYS_gettid 143
205 #else
206 #error define gettid for the arch
207 #endif
208 #endif
209 #endif
210 #endif
211 #endif
212
213 // Cpu architecture string
214 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
215
216
217 // pid_t gettid()
218 //
219 // Returns the kernel thread id of the currently running thread. Kernel
220 // thread id is used to access /proc.
221 pid_t os::Linux::gettid() {
222 int rslt = syscall(SYS_gettid);
223 assert(rslt != -1, "must be."); // old linuxthreads implementation?
224 return (pid_t)rslt;
225 }
226
227 // Most versions of linux have a bug where the number of processors are
228 // determined by looking at the /proc file system. In a chroot environment,
229 // the system call returns 1. This causes the VM to act as if it is
230 // a single processor and elide locking (see is_MP() call).
231 static bool unsafe_chroot_detected = false;
232 static const char *unstable_chroot_error = "/proc file system not found.\n"
233 "Java may be unstable running multithreaded in a chroot "
234 "environment on Linux when /proc filesystem is not mounted.";
235
236 void os::Linux::initialize_system_info() {
237 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
238 if (processor_count() == 1) {
239 pid_t pid = os::Linux::gettid();
240 char fname[32];
241 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
242 FILE *fp = fopen(fname, "r");
243 if (fp == NULL) {
244 unsafe_chroot_detected = true;
245 } else {
246 fclose(fp);
247 }
248 }
249 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
250 assert(processor_count() > 0, "linux error");
251 }
252
253 void os::init_system_properties_values() {
254 // The next steps are taken in the product version:
255 //
256 // Obtain the JAVA_HOME value from the location of libjvm.so.
257 // This library should be located at:
258 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
259 //
260 // If "/jre/lib/" appears at the right place in the path, then we
261 // assume libjvm.so is installed in a JDK and we use this path.
262 //
263 // Otherwise exit with message: "Could not create the Java virtual machine."
264 //
265 // The following extra steps are taken in the debugging version:
266 //
267 // If "/jre/lib/" does NOT appear at the right place in the path
268 // instead of exit check for $JAVA_HOME environment variable.
269 //
270 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
271 // then we append a fake suffix "hotspot/libjvm.so" to this path so
272 // it looks like libjvm.so is installed there
273 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
274 //
275 // Otherwise exit.
276 //
277 // Important note: if the location of libjvm.so changes this
278 // code needs to be changed accordingly.
279
280 // See ld(1):
281 // The linker uses the following search paths to locate required
282 // shared libraries:
283 // 1: ...
284 // ...
285 // 7: The default directories, normally /lib and /usr/lib.
286 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
287 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
288 #else
289 #define DEFAULT_LIBPATH "/lib:/usr/lib"
290 #endif
291
292 // Base path of extensions installed on the system.
293 #define SYS_EXT_DIR "/usr/java/packages"
294 #define EXTENSIONS_DIR "/lib/ext"
295
296 // Buffer that fits several sprintfs.
297 // Note that the space for the colon and the trailing null are provided
298 // by the nulls included by the sizeof operator.
299 const size_t bufsize =
300 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
301 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
302 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
303
304 // sysclasspath, java_home, dll_dir
305 {
306 char *pslash;
307 os::jvm_path(buf, bufsize);
308
309 // Found the full path to libjvm.so.
310 // Now cut the path to <java_home>/jre if we can.
311 pslash = strrchr(buf, '/');
312 if (pslash != NULL) {
313 *pslash = '\0'; // Get rid of /libjvm.so.
314 }
315 pslash = strrchr(buf, '/');
316 if (pslash != NULL) {
317 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
318 }
319 Arguments::set_dll_dir(buf);
320
321 if (pslash != NULL) {
322 pslash = strrchr(buf, '/');
323 if (pslash != NULL) {
324 *pslash = '\0'; // Get rid of /<arch>.
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /lib.
328 }
329 }
330 }
331 Arguments::set_java_home(buf);
332 set_boot_path('/', ':');
333 }
334
335 // Where to look for native libraries.
336 //
337 // Note: Due to a legacy implementation, most of the library path
338 // is set in the launcher. This was to accomodate linking restrictions
339 // on legacy Linux implementations (which are no longer supported).
340 // Eventually, all the library path setting will be done here.
341 //
342 // However, to prevent the proliferation of improperly built native
343 // libraries, the new path component /usr/java/packages is added here.
344 // Eventually, all the library path setting will be done here.
345 {
346 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
347 // should always exist (until the legacy problem cited above is
348 // addressed).
349 const char *v = ::getenv("LD_LIBRARY_PATH");
350 const char *v_colon = ":";
351 if (v == NULL) { v = ""; v_colon = ""; }
352 // That's +1 for the colon and +1 for the trailing '\0'.
353 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
354 strlen(v) + 1 +
355 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
356 mtInternal);
357 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
358 Arguments::set_library_path(ld_library_path);
359 FREE_C_HEAP_ARRAY(char, ld_library_path);
360 }
361
362 // Extensions directories.
363 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
364 Arguments::set_ext_dirs(buf);
365
366 FREE_C_HEAP_ARRAY(char, buf);
367
368 #undef DEFAULT_LIBPATH
369 #undef SYS_EXT_DIR
370 #undef EXTENSIONS_DIR
371 }
372
373 ////////////////////////////////////////////////////////////////////////////////
374 // breakpoint support
375
376 void os::breakpoint() {
377 BREAKPOINT;
378 }
379
380 extern "C" void breakpoint() {
381 // use debugger to set breakpoint here
382 }
383
384 ////////////////////////////////////////////////////////////////////////////////
385 // signal support
386
387 debug_only(static bool signal_sets_initialized = false);
388 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
389
390 bool os::Linux::is_sig_ignored(int sig) {
391 struct sigaction oact;
392 sigaction(sig, (struct sigaction*)NULL, &oact);
393 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
394 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
395 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
396 return true;
397 } else {
398 return false;
399 }
400 }
401
402 void os::Linux::signal_sets_init() {
403 // Should also have an assertion stating we are still single-threaded.
404 assert(!signal_sets_initialized, "Already initialized");
405 // Fill in signals that are necessarily unblocked for all threads in
406 // the VM. Currently, we unblock the following signals:
407 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
408 // by -Xrs (=ReduceSignalUsage));
409 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
410 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
411 // the dispositions or masks wrt these signals.
412 // Programs embedding the VM that want to use the above signals for their
413 // own purposes must, at this time, use the "-Xrs" option to prevent
414 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
415 // (See bug 4345157, and other related bugs).
416 // In reality, though, unblocking these signals is really a nop, since
417 // these signals are not blocked by default.
418 sigemptyset(&unblocked_sigs);
419 sigemptyset(&allowdebug_blocked_sigs);
420 sigaddset(&unblocked_sigs, SIGILL);
421 sigaddset(&unblocked_sigs, SIGSEGV);
422 sigaddset(&unblocked_sigs, SIGBUS);
423 sigaddset(&unblocked_sigs, SIGFPE);
424 #if defined(PPC64)
425 sigaddset(&unblocked_sigs, SIGTRAP);
426 #endif
427 sigaddset(&unblocked_sigs, SR_signum);
428
429 if (!ReduceSignalUsage) {
430 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
431 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
432 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
433 }
434 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
435 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
436 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
437 }
438 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
439 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
441 }
442 }
443 // Fill in signals that are blocked by all but the VM thread.
444 sigemptyset(&vm_sigs);
445 if (!ReduceSignalUsage) {
446 sigaddset(&vm_sigs, BREAK_SIGNAL);
447 }
448 debug_only(signal_sets_initialized = true);
449
450 }
451
452 // These are signals that are unblocked while a thread is running Java.
453 // (For some reason, they get blocked by default.)
454 sigset_t* os::Linux::unblocked_signals() {
455 assert(signal_sets_initialized, "Not initialized");
456 return &unblocked_sigs;
457 }
458
459 // These are the signals that are blocked while a (non-VM) thread is
460 // running Java. Only the VM thread handles these signals.
461 sigset_t* os::Linux::vm_signals() {
462 assert(signal_sets_initialized, "Not initialized");
463 return &vm_sigs;
464 }
465
466 // These are signals that are blocked during cond_wait to allow debugger in
467 sigset_t* os::Linux::allowdebug_blocked_signals() {
468 assert(signal_sets_initialized, "Not initialized");
469 return &allowdebug_blocked_sigs;
470 }
471
472 void os::Linux::hotspot_sigmask(Thread* thread) {
473
474 //Save caller's signal mask before setting VM signal mask
475 sigset_t caller_sigmask;
476 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
477
478 OSThread* osthread = thread->osthread();
479 osthread->set_caller_sigmask(caller_sigmask);
480
481 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
482
483 if (!ReduceSignalUsage) {
484 if (thread->is_VM_thread()) {
485 // Only the VM thread handles BREAK_SIGNAL ...
486 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
487 } else {
488 // ... all other threads block BREAK_SIGNAL
489 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
490 }
491 }
492 }
493
494 //////////////////////////////////////////////////////////////////////////////
495 // detecting pthread library
496
497 void os::Linux::libpthread_init() {
498 // Save glibc and pthread version strings.
499 #if !defined(_CS_GNU_LIBC_VERSION) || \
500 !defined(_CS_GNU_LIBPTHREAD_VERSION)
501 #error "glibc too old (< 2.3.2)"
502 #endif
503
504 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
505 assert(n > 0, "cannot retrieve glibc version");
506 char *str = (char *)malloc(n, mtInternal);
507 confstr(_CS_GNU_LIBC_VERSION, str, n);
508 os::Linux::set_glibc_version(str);
509
510 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
511 assert(n > 0, "cannot retrieve pthread version");
512 str = (char *)malloc(n, mtInternal);
513 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
514 os::Linux::set_libpthread_version(str);
515 }
516
517 /////////////////////////////////////////////////////////////////////////////
518 // thread stack expansion
519
520 // os::Linux::manually_expand_stack() takes care of expanding the thread
521 // stack. Note that this is normally not needed: pthread stacks allocate
522 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
523 // committed. Therefore it is not necessary to expand the stack manually.
524 //
525 // Manually expanding the stack was historically needed on LinuxThreads
526 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
527 // it is kept to deal with very rare corner cases:
528 //
529 // For one, user may run the VM on an own implementation of threads
530 // whose stacks are - like the old LinuxThreads - implemented using
531 // mmap(MAP_GROWSDOWN).
532 //
533 // Also, this coding may be needed if the VM is running on the primordial
534 // thread. Normally we avoid running on the primordial thread; however,
535 // user may still invoke the VM on the primordial thread.
536 //
537 // The following historical comment describes the details about running
538 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
539
540
541 // Force Linux kernel to expand current thread stack. If "bottom" is close
542 // to the stack guard, caller should block all signals.
543 //
544 // MAP_GROWSDOWN:
545 // A special mmap() flag that is used to implement thread stacks. It tells
546 // kernel that the memory region should extend downwards when needed. This
547 // allows early versions of LinuxThreads to only mmap the first few pages
548 // when creating a new thread. Linux kernel will automatically expand thread
549 // stack as needed (on page faults).
550 //
551 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
552 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
553 // region, it's hard to tell if the fault is due to a legitimate stack
554 // access or because of reading/writing non-exist memory (e.g. buffer
555 // overrun). As a rule, if the fault happens below current stack pointer,
556 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
557 // application (see Linux kernel fault.c).
558 //
559 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
560 // stack overflow detection.
561 //
562 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
563 // not use MAP_GROWSDOWN.
564 //
565 // To get around the problem and allow stack banging on Linux, we need to
566 // manually expand thread stack after receiving the SIGSEGV.
567 //
568 // There are two ways to expand thread stack to address "bottom", we used
569 // both of them in JVM before 1.5:
570 // 1. adjust stack pointer first so that it is below "bottom", and then
571 // touch "bottom"
572 // 2. mmap() the page in question
573 //
574 // Now alternate signal stack is gone, it's harder to use 2. For instance,
575 // if current sp is already near the lower end of page 101, and we need to
576 // call mmap() to map page 100, it is possible that part of the mmap() frame
577 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
578 // That will destroy the mmap() frame and cause VM to crash.
579 //
580 // The following code works by adjusting sp first, then accessing the "bottom"
581 // page to force a page fault. Linux kernel will then automatically expand the
582 // stack mapping.
583 //
584 // _expand_stack_to() assumes its frame size is less than page size, which
585 // should always be true if the function is not inlined.
586
587 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
588 #define NOINLINE
589 #else
590 #define NOINLINE __attribute__ ((noinline))
591 #endif
592
593 static void _expand_stack_to(address bottom) NOINLINE;
594
595 static void _expand_stack_to(address bottom) {
596 address sp;
597 size_t size;
598 volatile char *p;
599
600 // Adjust bottom to point to the largest address within the same page, it
601 // gives us a one-page buffer if alloca() allocates slightly more memory.
602 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
603 bottom += os::Linux::page_size() - 1;
604
605 // sp might be slightly above current stack pointer; if that's the case, we
606 // will alloca() a little more space than necessary, which is OK. Don't use
607 // os::current_stack_pointer(), as its result can be slightly below current
608 // stack pointer, causing us to not alloca enough to reach "bottom".
609 sp = (address)&sp;
610
611 if (sp > bottom) {
612 size = sp - bottom;
613 p = (volatile char *)alloca(size);
614 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
615 p[0] = '\0';
616 }
617 }
618
619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
620 assert(t!=NULL, "just checking");
621 assert(t->osthread()->expanding_stack(), "expand should be set");
622 assert(t->stack_base() != NULL, "stack_base was not initialized");
623
624 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
625 sigset_t mask_all, old_sigset;
626 sigfillset(&mask_all);
627 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
628 _expand_stack_to(addr);
629 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
630 return true;
631 }
632 return false;
633 }
634
635 //////////////////////////////////////////////////////////////////////////////
636 // create new thread
637
638 // Thread start routine for all newly created threads
639 static void *java_start(Thread *thread) {
640 // Try to randomize the cache line index of hot stack frames.
641 // This helps when threads of the same stack traces evict each other's
642 // cache lines. The threads can be either from the same JVM instance, or
643 // from different JVM instances. The benefit is especially true for
644 // processors with hyperthreading technology.
645 static int counter = 0;
646 int pid = os::current_process_id();
647 alloca(((pid ^ counter++) & 7) * 128);
648
649 thread->initialize_thread_current();
650
651 OSThread* osthread = thread->osthread();
652 Monitor* sync = osthread->startThread_lock();
653
654 osthread->set_thread_id(os::current_thread_id());
655
656 if (UseNUMA) {
657 int lgrp_id = os::numa_get_group_id();
658 if (lgrp_id != -1) {
659 thread->set_lgrp_id(lgrp_id);
660 }
661 }
662 // initialize signal mask for this thread
663 os::Linux::hotspot_sigmask(thread);
664
665 // initialize floating point control register
666 os::Linux::init_thread_fpu_state();
667
668 // handshaking with parent thread
669 {
670 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
671
672 // notify parent thread
673 osthread->set_state(INITIALIZED);
674 sync->notify_all();
675
676 // wait until os::start_thread()
677 while (osthread->get_state() == INITIALIZED) {
678 sync->wait(Mutex::_no_safepoint_check_flag);
679 }
680 }
681
682 // call one more level start routine
683 thread->run();
684
685 return 0;
686 }
687
688 bool os::create_thread(Thread* thread, ThreadType thr_type,
689 size_t stack_size) {
690 assert(thread->osthread() == NULL, "caller responsible");
691
692 // Allocate the OSThread object
693 OSThread* osthread = new OSThread(NULL, NULL);
694 if (osthread == NULL) {
695 return false;
696 }
697
698 // set the correct thread state
699 osthread->set_thread_type(thr_type);
700
701 // Initial state is ALLOCATED but not INITIALIZED
702 osthread->set_state(ALLOCATED);
703
704 thread->set_osthread(osthread);
705
706 // init thread attributes
707 pthread_attr_t attr;
708 pthread_attr_init(&attr);
709 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
710
711 // stack size
712 // calculate stack size if it's not specified by caller
713 if (stack_size == 0) {
714 stack_size = os::Linux::default_stack_size(thr_type);
715
716 switch (thr_type) {
717 case os::java_thread:
718 // Java threads use ThreadStackSize which default value can be
719 // changed with the flag -Xss
720 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
721 stack_size = JavaThread::stack_size_at_create();
722 break;
723 case os::compiler_thread:
724 if (CompilerThreadStackSize > 0) {
725 stack_size = (size_t)(CompilerThreadStackSize * K);
726 break;
727 } // else fall through:
728 // use VMThreadStackSize if CompilerThreadStackSize is not defined
729 case os::vm_thread:
730 case os::pgc_thread:
731 case os::cgc_thread:
732 case os::watcher_thread:
733 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
734 break;
735 }
736 }
737
738 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
739 pthread_attr_setstacksize(&attr, stack_size);
740
741 // glibc guard page
742 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
743
744 ThreadState state;
745
746 {
747 pthread_t tid;
748 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
749
750 pthread_attr_destroy(&attr);
751
752 if (ret != 0) {
753 if (PrintMiscellaneous && (Verbose || WizardMode)) {
754 perror("pthread_create()");
755 }
756 // Need to clean up stuff we've allocated so far
757 thread->set_osthread(NULL);
758 delete osthread;
759 return false;
760 }
761
762 // Store pthread info into the OSThread
763 osthread->set_pthread_id(tid);
764
765 // Wait until child thread is either initialized or aborted
766 {
767 Monitor* sync_with_child = osthread->startThread_lock();
768 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
769 while ((state = osthread->get_state()) == ALLOCATED) {
770 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
771 }
772 }
773 }
774
775 // Aborted due to thread limit being reached
776 if (state == ZOMBIE) {
777 thread->set_osthread(NULL);
778 delete osthread;
779 return false;
780 }
781
782 // The thread is returned suspended (in state INITIALIZED),
783 // and is started higher up in the call chain
784 assert(state == INITIALIZED, "race condition");
785 return true;
786 }
787
788 /////////////////////////////////////////////////////////////////////////////
789 // attach existing thread
790
791 // bootstrap the main thread
792 bool os::create_main_thread(JavaThread* thread) {
793 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
794 return create_attached_thread(thread);
795 }
796
797 bool os::create_attached_thread(JavaThread* thread) {
798 #ifdef ASSERT
799 thread->verify_not_published();
800 #endif
801
802 // Allocate the OSThread object
803 OSThread* osthread = new OSThread(NULL, NULL);
804
805 if (osthread == NULL) {
806 return false;
807 }
808
809 // Store pthread info into the OSThread
810 osthread->set_thread_id(os::Linux::gettid());
811 osthread->set_pthread_id(::pthread_self());
812
813 // initialize floating point control register
814 os::Linux::init_thread_fpu_state();
815
816 // Initial thread state is RUNNABLE
817 osthread->set_state(RUNNABLE);
818
819 thread->set_osthread(osthread);
820
821 if (UseNUMA) {
822 int lgrp_id = os::numa_get_group_id();
823 if (lgrp_id != -1) {
824 thread->set_lgrp_id(lgrp_id);
825 }
826 }
827
828 if (os::Linux::is_initial_thread()) {
829 // If current thread is initial thread, its stack is mapped on demand,
830 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
831 // the entire stack region to avoid SEGV in stack banging.
832 // It is also useful to get around the heap-stack-gap problem on SuSE
833 // kernel (see 4821821 for details). We first expand stack to the top
834 // of yellow zone, then enable stack yellow zone (order is significant,
835 // enabling yellow zone first will crash JVM on SuSE Linux), so there
836 // is no gap between the last two virtual memory regions.
837
838 JavaThread *jt = (JavaThread *)thread;
839 address addr = jt->stack_reserved_zone_base();
840 assert(addr != NULL, "initialization problem?");
841 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
842
843 osthread->set_expanding_stack();
844 os::Linux::manually_expand_stack(jt, addr);
845 osthread->clear_expanding_stack();
846 }
847
848 // initialize signal mask for this thread
849 // and save the caller's signal mask
850 os::Linux::hotspot_sigmask(thread);
851
852 return true;
853 }
854
855 void os::pd_start_thread(Thread* thread) {
856 OSThread * osthread = thread->osthread();
857 assert(osthread->get_state() != INITIALIZED, "just checking");
858 Monitor* sync_with_child = osthread->startThread_lock();
859 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
860 sync_with_child->notify();
861 }
862
863 // Free Linux resources related to the OSThread
864 void os::free_thread(OSThread* osthread) {
865 assert(osthread != NULL, "osthread not set");
866
867 if (Thread::current()->osthread() == osthread) {
868 // Restore caller's signal mask
869 sigset_t sigmask = osthread->caller_sigmask();
870 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
871 }
872
873 delete osthread;
874 }
875
876 //////////////////////////////////////////////////////////////////////////////
877 // initial thread
878
879 // Check if current thread is the initial thread, similar to Solaris thr_main.
880 bool os::Linux::is_initial_thread(void) {
881 char dummy;
882 // If called before init complete, thread stack bottom will be null.
883 // Can be called if fatal error occurs before initialization.
884 if (initial_thread_stack_bottom() == NULL) return false;
885 assert(initial_thread_stack_bottom() != NULL &&
886 initial_thread_stack_size() != 0,
887 "os::init did not locate initial thread's stack region");
888 if ((address)&dummy >= initial_thread_stack_bottom() &&
889 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
890 return true;
891 } else {
892 return false;
893 }
894 }
895
896 // Find the virtual memory area that contains addr
897 static bool find_vma(address addr, address* vma_low, address* vma_high) {
898 FILE *fp = fopen("/proc/self/maps", "r");
899 if (fp) {
900 address low, high;
901 while (!feof(fp)) {
902 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
903 if (low <= addr && addr < high) {
904 if (vma_low) *vma_low = low;
905 if (vma_high) *vma_high = high;
906 fclose(fp);
907 return true;
908 }
909 }
910 for (;;) {
911 int ch = fgetc(fp);
912 if (ch == EOF || ch == (int)'\n') break;
913 }
914 }
915 fclose(fp);
916 }
917 return false;
918 }
919
920 // Locate initial thread stack. This special handling of initial thread stack
921 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
922 // bogus value for initial thread.
923 void os::Linux::capture_initial_stack(size_t max_size) {
924 // stack size is the easy part, get it from RLIMIT_STACK
925 size_t stack_size;
926 struct rlimit rlim;
927 getrlimit(RLIMIT_STACK, &rlim);
928 stack_size = rlim.rlim_cur;
929
930 // 6308388: a bug in ld.so will relocate its own .data section to the
931 // lower end of primordial stack; reduce ulimit -s value a little bit
932 // so we won't install guard page on ld.so's data section.
933 stack_size -= 2 * page_size();
934
935 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
936 // 7.1, in both cases we will get 2G in return value.
937 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
938 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
939 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
940 // in case other parts in glibc still assumes 2M max stack size.
941 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
942 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
943 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
944 stack_size = 2 * K * K IA64_ONLY(*2);
945 }
946 // Try to figure out where the stack base (top) is. This is harder.
947 //
948 // When an application is started, glibc saves the initial stack pointer in
949 // a global variable "__libc_stack_end", which is then used by system
950 // libraries. __libc_stack_end should be pretty close to stack top. The
951 // variable is available since the very early days. However, because it is
952 // a private interface, it could disappear in the future.
953 //
954 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
955 // to __libc_stack_end, it is very close to stack top, but isn't the real
956 // stack top. Note that /proc may not exist if VM is running as a chroot
957 // program, so reading /proc/<pid>/stat could fail. Also the contents of
958 // /proc/<pid>/stat could change in the future (though unlikely).
959 //
960 // We try __libc_stack_end first. If that doesn't work, look for
961 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
962 // as a hint, which should work well in most cases.
963
964 uintptr_t stack_start;
965
966 // try __libc_stack_end first
967 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
968 if (p && *p) {
969 stack_start = *p;
970 } else {
971 // see if we can get the start_stack field from /proc/self/stat
972 FILE *fp;
973 int pid;
974 char state;
975 int ppid;
976 int pgrp;
977 int session;
978 int nr;
979 int tpgrp;
980 unsigned long flags;
981 unsigned long minflt;
982 unsigned long cminflt;
983 unsigned long majflt;
984 unsigned long cmajflt;
985 unsigned long utime;
986 unsigned long stime;
987 long cutime;
988 long cstime;
989 long prio;
990 long nice;
991 long junk;
992 long it_real;
993 uintptr_t start;
994 uintptr_t vsize;
995 intptr_t rss;
996 uintptr_t rsslim;
997 uintptr_t scodes;
998 uintptr_t ecode;
999 int i;
1000
1001 // Figure what the primordial thread stack base is. Code is inspired
1002 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1003 // followed by command name surrounded by parentheses, state, etc.
1004 char stat[2048];
1005 int statlen;
1006
1007 fp = fopen("/proc/self/stat", "r");
1008 if (fp) {
1009 statlen = fread(stat, 1, 2047, fp);
1010 stat[statlen] = '\0';
1011 fclose(fp);
1012
1013 // Skip pid and the command string. Note that we could be dealing with
1014 // weird command names, e.g. user could decide to rename java launcher
1015 // to "java 1.4.2 :)", then the stat file would look like
1016 // 1234 (java 1.4.2 :)) R ... ...
1017 // We don't really need to know the command string, just find the last
1018 // occurrence of ")" and then start parsing from there. See bug 4726580.
1019 char * s = strrchr(stat, ')');
1020
1021 i = 0;
1022 if (s) {
1023 // Skip blank chars
1024 do { s++; } while (s && isspace(*s));
1025
1026 #define _UFM UINTX_FORMAT
1027 #define _DFM INTX_FORMAT
1028
1029 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1030 // 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
1031 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,
1032 &state, // 3 %c
1033 &ppid, // 4 %d
1034 &pgrp, // 5 %d
1035 &session, // 6 %d
1036 &nr, // 7 %d
1037 &tpgrp, // 8 %d
1038 &flags, // 9 %lu
1039 &minflt, // 10 %lu
1040 &cminflt, // 11 %lu
1041 &majflt, // 12 %lu
1042 &cmajflt, // 13 %lu
1043 &utime, // 14 %lu
1044 &stime, // 15 %lu
1045 &cutime, // 16 %ld
1046 &cstime, // 17 %ld
1047 &prio, // 18 %ld
1048 &nice, // 19 %ld
1049 &junk, // 20 %ld
1050 &it_real, // 21 %ld
1051 &start, // 22 UINTX_FORMAT
1052 &vsize, // 23 UINTX_FORMAT
1053 &rss, // 24 INTX_FORMAT
1054 &rsslim, // 25 UINTX_FORMAT
1055 &scodes, // 26 UINTX_FORMAT
1056 &ecode, // 27 UINTX_FORMAT
1057 &stack_start); // 28 UINTX_FORMAT
1058 }
1059
1060 #undef _UFM
1061 #undef _DFM
1062
1063 if (i != 28 - 2) {
1064 assert(false, "Bad conversion from /proc/self/stat");
1065 // product mode - assume we are the initial thread, good luck in the
1066 // embedded case.
1067 warning("Can't detect initial thread stack location - bad conversion");
1068 stack_start = (uintptr_t) &rlim;
1069 }
1070 } else {
1071 // For some reason we can't open /proc/self/stat (for example, running on
1072 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1073 // most cases, so don't abort:
1074 warning("Can't detect initial thread stack location - no /proc/self/stat");
1075 stack_start = (uintptr_t) &rlim;
1076 }
1077 }
1078
1079 // Now we have a pointer (stack_start) very close to the stack top, the
1080 // next thing to do is to figure out the exact location of stack top. We
1081 // can find out the virtual memory area that contains stack_start by
1082 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1083 // and its upper limit is the real stack top. (again, this would fail if
1084 // running inside chroot, because /proc may not exist.)
1085
1086 uintptr_t stack_top;
1087 address low, high;
1088 if (find_vma((address)stack_start, &low, &high)) {
1089 // success, "high" is the true stack top. (ignore "low", because initial
1090 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1091 stack_top = (uintptr_t)high;
1092 } else {
1093 // failed, likely because /proc/self/maps does not exist
1094 warning("Can't detect initial thread stack location - find_vma failed");
1095 // best effort: stack_start is normally within a few pages below the real
1096 // stack top, use it as stack top, and reduce stack size so we won't put
1097 // guard page outside stack.
1098 stack_top = stack_start;
1099 stack_size -= 16 * page_size();
1100 }
1101
1102 // stack_top could be partially down the page so align it
1103 stack_top = align_size_up(stack_top, page_size());
1104
1105 if (max_size && stack_size > max_size) {
1106 _initial_thread_stack_size = max_size;
1107 } else {
1108 _initial_thread_stack_size = stack_size;
1109 }
1110
1111 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1112 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1113 }
1114
1115 ////////////////////////////////////////////////////////////////////////////////
1116 // time support
1117
1118 // Time since start-up in seconds to a fine granularity.
1119 // Used by VMSelfDestructTimer and the MemProfiler.
1120 double os::elapsedTime() {
1121
1122 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1123 }
1124
1125 jlong os::elapsed_counter() {
1126 return javaTimeNanos() - initial_time_count;
1127 }
1128
1129 jlong os::elapsed_frequency() {
1130 return NANOSECS_PER_SEC; // nanosecond resolution
1131 }
1132
1133 bool os::supports_vtime() { return true; }
1134 bool os::enable_vtime() { return false; }
1135 bool os::vtime_enabled() { return false; }
1136
1137 double os::elapsedVTime() {
1138 struct rusage usage;
1139 int retval = getrusage(RUSAGE_THREAD, &usage);
1140 if (retval == 0) {
1141 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);
1142 } else {
1143 // better than nothing, but not much
1144 return elapsedTime();
1145 }
1146 }
1147
1148 jlong os::javaTimeMillis() {
1149 timeval time;
1150 int status = gettimeofday(&time, NULL);
1151 assert(status != -1, "linux error");
1152 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1153 }
1154
1155 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1156 timeval time;
1157 int status = gettimeofday(&time, NULL);
1158 assert(status != -1, "linux error");
1159 seconds = jlong(time.tv_sec);
1160 nanos = jlong(time.tv_usec) * 1000;
1161 }
1162
1163
1164 #ifndef CLOCK_MONOTONIC
1165 #define CLOCK_MONOTONIC (1)
1166 #endif
1167
1168 void os::Linux::clock_init() {
1169 // we do dlopen's in this particular order due to bug in linux
1170 // dynamical loader (see 6348968) leading to crash on exit
1171 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1172 if (handle == NULL) {
1173 handle = dlopen("librt.so", RTLD_LAZY);
1174 }
1175
1176 if (handle) {
1177 int (*clock_getres_func)(clockid_t, struct timespec*) =
1178 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1179 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1180 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1181 if (clock_getres_func && clock_gettime_func) {
1182 // See if monotonic clock is supported by the kernel. Note that some
1183 // early implementations simply return kernel jiffies (updated every
1184 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1185 // for nano time (though the monotonic property is still nice to have).
1186 // It's fixed in newer kernels, however clock_getres() still returns
1187 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1188 // resolution for now. Hopefully as people move to new kernels, this
1189 // won't be a problem.
1190 struct timespec res;
1191 struct timespec tp;
1192 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1193 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1194 // yes, monotonic clock is supported
1195 _clock_gettime = clock_gettime_func;
1196 return;
1197 } else {
1198 // close librt if there is no monotonic clock
1199 dlclose(handle);
1200 }
1201 }
1202 }
1203 warning("No monotonic clock was available - timed services may " \
1204 "be adversely affected if the time-of-day clock changes");
1205 }
1206
1207 #ifndef SYS_clock_getres
1208 #if defined(IA32) || defined(AMD64)
1209 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1210 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1211 #else
1212 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1213 #define sys_clock_getres(x,y) -1
1214 #endif
1215 #else
1216 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1217 #endif
1218
1219 void os::Linux::fast_thread_clock_init() {
1220 if (!UseLinuxPosixThreadCPUClocks) {
1221 return;
1222 }
1223 clockid_t clockid;
1224 struct timespec tp;
1225 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1226 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1227
1228 // Switch to using fast clocks for thread cpu time if
1229 // the sys_clock_getres() returns 0 error code.
1230 // Note, that some kernels may support the current thread
1231 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1232 // returned by the pthread_getcpuclockid().
1233 // If the fast Posix clocks are supported then the sys_clock_getres()
1234 // must return at least tp.tv_sec == 0 which means a resolution
1235 // better than 1 sec. This is extra check for reliability.
1236
1237 if (pthread_getcpuclockid_func &&
1238 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1239 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1240 _supports_fast_thread_cpu_time = true;
1241 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1242 }
1243 }
1244
1245 jlong os::javaTimeNanos() {
1246 if (os::supports_monotonic_clock()) {
1247 struct timespec tp;
1248 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1249 assert(status == 0, "gettime error");
1250 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1251 return result;
1252 } else {
1253 timeval time;
1254 int status = gettimeofday(&time, NULL);
1255 assert(status != -1, "linux error");
1256 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1257 return 1000 * usecs;
1258 }
1259 }
1260
1261 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1262 if (os::supports_monotonic_clock()) {
1263 info_ptr->max_value = ALL_64_BITS;
1264
1265 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1266 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1267 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1268 } else {
1269 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1270 info_ptr->max_value = ALL_64_BITS;
1271
1272 // gettimeofday is a real time clock so it skips
1273 info_ptr->may_skip_backward = true;
1274 info_ptr->may_skip_forward = true;
1275 }
1276
1277 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1278 }
1279
1280 // Return the real, user, and system times in seconds from an
1281 // arbitrary fixed point in the past.
1282 bool os::getTimesSecs(double* process_real_time,
1283 double* process_user_time,
1284 double* process_system_time) {
1285 struct tms ticks;
1286 clock_t real_ticks = times(&ticks);
1287
1288 if (real_ticks == (clock_t) (-1)) {
1289 return false;
1290 } else {
1291 double ticks_per_second = (double) clock_tics_per_sec;
1292 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1293 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1294 *process_real_time = ((double) real_ticks) / ticks_per_second;
1295
1296 return true;
1297 }
1298 }
1299
1300
1301 char * os::local_time_string(char *buf, size_t buflen) {
1302 struct tm t;
1303 time_t long_time;
1304 time(&long_time);
1305 localtime_r(&long_time, &t);
1306 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1307 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1308 t.tm_hour, t.tm_min, t.tm_sec);
1309 return buf;
1310 }
1311
1312 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1313 return localtime_r(clock, res);
1314 }
1315
1316 ////////////////////////////////////////////////////////////////////////////////
1317 // runtime exit support
1318
1319 // Note: os::shutdown() might be called very early during initialization, or
1320 // called from signal handler. Before adding something to os::shutdown(), make
1321 // sure it is async-safe and can handle partially initialized VM.
1322 void os::shutdown() {
1323
1324 // allow PerfMemory to attempt cleanup of any persistent resources
1325 perfMemory_exit();
1326
1327 // needs to remove object in file system
1328 AttachListener::abort();
1329
1330 // flush buffered output, finish log files
1331 ostream_abort();
1332
1333 // Check for abort hook
1334 abort_hook_t abort_hook = Arguments::abort_hook();
1335 if (abort_hook != NULL) {
1336 abort_hook();
1337 }
1338
1339 }
1340
1341 // Note: os::abort() might be called very early during initialization, or
1342 // called from signal handler. Before adding something to os::abort(), make
1343 // sure it is async-safe and can handle partially initialized VM.
1344 void os::abort(bool dump_core, void* siginfo, const void* context) {
1345 os::shutdown();
1346 if (dump_core) {
1347 #ifndef PRODUCT
1348 fdStream out(defaultStream::output_fd());
1349 out.print_raw("Current thread is ");
1350 char buf[16];
1351 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1352 out.print_raw_cr(buf);
1353 out.print_raw_cr("Dumping core ...");
1354 #endif
1355 ::abort(); // dump core
1356 }
1357
1358 ::exit(1);
1359 }
1360
1361 // Die immediately, no exit hook, no abort hook, no cleanup.
1362 void os::die() {
1363 ::abort();
1364 }
1365
1366
1367 // This method is a copy of JDK's sysGetLastErrorString
1368 // from src/solaris/hpi/src/system_md.c
1369
1370 size_t os::lasterror(char *buf, size_t len) {
1371 if (errno == 0) return 0;
1372
1373 const char *s = ::strerror(errno);
1374 size_t n = ::strlen(s);
1375 if (n >= len) {
1376 n = len - 1;
1377 }
1378 ::strncpy(buf, s, n);
1379 buf[n] = '\0';
1380 return n;
1381 }
1382
1383 // thread_id is kernel thread id (similar to Solaris LWP id)
1384 intx os::current_thread_id() { return os::Linux::gettid(); }
1385 int os::current_process_id() {
1386 return ::getpid();
1387 }
1388
1389 // DLL functions
1390
1391 const char* os::dll_file_extension() { return ".so"; }
1392
1393 // This must be hard coded because it's the system's temporary
1394 // directory not the java application's temp directory, ala java.io.tmpdir.
1395 const char* os::get_temp_directory() { return "/tmp"; }
1396
1397 static bool file_exists(const char* filename) {
1398 struct stat statbuf;
1399 if (filename == NULL || strlen(filename) == 0) {
1400 return false;
1401 }
1402 return os::stat(filename, &statbuf) == 0;
1403 }
1404
1405 bool os::dll_build_name(char* buffer, size_t buflen,
1406 const char* pname, const char* fname) {
1407 bool retval = false;
1408 // Copied from libhpi
1409 const size_t pnamelen = pname ? strlen(pname) : 0;
1410
1411 // Return error on buffer overflow.
1412 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1413 return retval;
1414 }
1415
1416 if (pnamelen == 0) {
1417 snprintf(buffer, buflen, "lib%s.so", fname);
1418 retval = true;
1419 } else if (strchr(pname, *os::path_separator()) != NULL) {
1420 int n;
1421 char** pelements = split_path(pname, &n);
1422 if (pelements == NULL) {
1423 return false;
1424 }
1425 for (int i = 0; i < n; i++) {
1426 // Really shouldn't be NULL, but check can't hurt
1427 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1428 continue; // skip the empty path values
1429 }
1430 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1431 if (file_exists(buffer)) {
1432 retval = true;
1433 break;
1434 }
1435 }
1436 // release the storage
1437 for (int i = 0; i < n; i++) {
1438 if (pelements[i] != NULL) {
1439 FREE_C_HEAP_ARRAY(char, pelements[i]);
1440 }
1441 }
1442 if (pelements != NULL) {
1443 FREE_C_HEAP_ARRAY(char*, pelements);
1444 }
1445 } else {
1446 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1447 retval = true;
1448 }
1449 return retval;
1450 }
1451
1452 // check if addr is inside libjvm.so
1453 bool os::address_is_in_vm(address addr) {
1454 static address libjvm_base_addr;
1455 Dl_info dlinfo;
1456
1457 if (libjvm_base_addr == NULL) {
1458 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1459 libjvm_base_addr = (address)dlinfo.dli_fbase;
1460 }
1461 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1462 }
1463
1464 if (dladdr((void *)addr, &dlinfo) != 0) {
1465 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1466 }
1467
1468 return false;
1469 }
1470
1471 bool os::dll_address_to_function_name(address addr, char *buf,
1472 int buflen, int *offset,
1473 bool demangle) {
1474 // buf is not optional, but offset is optional
1475 assert(buf != NULL, "sanity check");
1476
1477 Dl_info dlinfo;
1478
1479 if (dladdr((void*)addr, &dlinfo) != 0) {
1480 // see if we have a matching symbol
1481 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1482 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1483 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1484 }
1485 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1486 return true;
1487 }
1488 // no matching symbol so try for just file info
1489 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1490 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1491 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1492 return true;
1493 }
1494 }
1495 }
1496
1497 buf[0] = '\0';
1498 if (offset != NULL) *offset = -1;
1499 return false;
1500 }
1501
1502 struct _address_to_library_name {
1503 address addr; // input : memory address
1504 size_t buflen; // size of fname
1505 char* fname; // output: library name
1506 address base; // library base addr
1507 };
1508
1509 static int address_to_library_name_callback(struct dl_phdr_info *info,
1510 size_t size, void *data) {
1511 int i;
1512 bool found = false;
1513 address libbase = NULL;
1514 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1515
1516 // iterate through all loadable segments
1517 for (i = 0; i < info->dlpi_phnum; i++) {
1518 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1519 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1520 // base address of a library is the lowest address of its loaded
1521 // segments.
1522 if (libbase == NULL || libbase > segbase) {
1523 libbase = segbase;
1524 }
1525 // see if 'addr' is within current segment
1526 if (segbase <= d->addr &&
1527 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1528 found = true;
1529 }
1530 }
1531 }
1532
1533 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1534 // so dll_address_to_library_name() can fall through to use dladdr() which
1535 // can figure out executable name from argv[0].
1536 if (found && info->dlpi_name && info->dlpi_name[0]) {
1537 d->base = libbase;
1538 if (d->fname) {
1539 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1540 }
1541 return 1;
1542 }
1543 return 0;
1544 }
1545
1546 bool os::dll_address_to_library_name(address addr, char* buf,
1547 int buflen, int* offset) {
1548 // buf is not optional, but offset is optional
1549 assert(buf != NULL, "sanity check");
1550
1551 Dl_info dlinfo;
1552 struct _address_to_library_name data;
1553
1554 // There is a bug in old glibc dladdr() implementation that it could resolve
1555 // to wrong library name if the .so file has a base address != NULL. Here
1556 // we iterate through the program headers of all loaded libraries to find
1557 // out which library 'addr' really belongs to. This workaround can be
1558 // removed once the minimum requirement for glibc is moved to 2.3.x.
1559 data.addr = addr;
1560 data.fname = buf;
1561 data.buflen = buflen;
1562 data.base = NULL;
1563 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1564
1565 if (rslt) {
1566 // buf already contains library name
1567 if (offset) *offset = addr - data.base;
1568 return true;
1569 }
1570 if (dladdr((void*)addr, &dlinfo) != 0) {
1571 if (dlinfo.dli_fname != NULL) {
1572 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1573 }
1574 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1575 *offset = addr - (address)dlinfo.dli_fbase;
1576 }
1577 return true;
1578 }
1579
1580 buf[0] = '\0';
1581 if (offset) *offset = -1;
1582 return false;
1583 }
1584
1585 // Loads .dll/.so and
1586 // in case of error it checks if .dll/.so was built for the
1587 // same architecture as Hotspot is running on
1588
1589
1590 // Remember the stack's state. The Linux dynamic linker will change
1591 // the stack to 'executable' at most once, so we must safepoint only once.
1592 bool os::Linux::_stack_is_executable = false;
1593
1594 // VM operation that loads a library. This is necessary if stack protection
1595 // of the Java stacks can be lost during loading the library. If we
1596 // do not stop the Java threads, they can stack overflow before the stacks
1597 // are protected again.
1598 class VM_LinuxDllLoad: public VM_Operation {
1599 private:
1600 const char *_filename;
1601 char *_ebuf;
1602 int _ebuflen;
1603 void *_lib;
1604 public:
1605 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1606 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1607 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1608 void doit() {
1609 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1610 os::Linux::_stack_is_executable = true;
1611 }
1612 void* loaded_library() { return _lib; }
1613 };
1614
1615 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1616 void * result = NULL;
1617 bool load_attempted = false;
1618
1619 // Check whether the library to load might change execution rights
1620 // of the stack. If they are changed, the protection of the stack
1621 // guard pages will be lost. We need a safepoint to fix this.
1622 //
1623 // See Linux man page execstack(8) for more info.
1624 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1625 ElfFile ef(filename);
1626 if (!ef.specifies_noexecstack()) {
1627 if (!is_init_completed()) {
1628 os::Linux::_stack_is_executable = true;
1629 // This is OK - No Java threads have been created yet, and hence no
1630 // stack guard pages to fix.
1631 //
1632 // This should happen only when you are building JDK7 using a very
1633 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1634 //
1635 // Dynamic loader will make all stacks executable after
1636 // this function returns, and will not do that again.
1637 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1638 } else {
1639 warning("You have loaded library %s which might have disabled stack guard. "
1640 "The VM will try to fix the stack guard now.\n"
1641 "It's highly recommended that you fix the library with "
1642 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1643 filename);
1644
1645 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1646 JavaThread *jt = JavaThread::current();
1647 if (jt->thread_state() != _thread_in_native) {
1648 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1649 // that requires ExecStack. Cannot enter safe point. Let's give up.
1650 warning("Unable to fix stack guard. Giving up.");
1651 } else {
1652 if (!LoadExecStackDllInVMThread) {
1653 // This is for the case where the DLL has an static
1654 // constructor function that executes JNI code. We cannot
1655 // load such DLLs in the VMThread.
1656 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1657 }
1658
1659 ThreadInVMfromNative tiv(jt);
1660 debug_only(VMNativeEntryWrapper vew;)
1661
1662 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1663 VMThread::execute(&op);
1664 if (LoadExecStackDllInVMThread) {
1665 result = op.loaded_library();
1666 }
1667 load_attempted = true;
1668 }
1669 }
1670 }
1671 }
1672
1673 if (!load_attempted) {
1674 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1675 }
1676
1677 if (result != NULL) {
1678 // Successful loading
1679 return result;
1680 }
1681
1682 Elf32_Ehdr elf_head;
1683 int diag_msg_max_length=ebuflen-strlen(ebuf);
1684 char* diag_msg_buf=ebuf+strlen(ebuf);
1685
1686 if (diag_msg_max_length==0) {
1687 // No more space in ebuf for additional diagnostics message
1688 return NULL;
1689 }
1690
1691
1692 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1693
1694 if (file_descriptor < 0) {
1695 // Can't open library, report dlerror() message
1696 return NULL;
1697 }
1698
1699 bool failed_to_read_elf_head=
1700 (sizeof(elf_head)!=
1701 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1702
1703 ::close(file_descriptor);
1704 if (failed_to_read_elf_head) {
1705 // file i/o error - report dlerror() msg
1706 return NULL;
1707 }
1708
1709 typedef struct {
1710 Elf32_Half code; // Actual value as defined in elf.h
1711 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1712 char elf_class; // 32 or 64 bit
1713 char endianess; // MSB or LSB
1714 char* name; // String representation
1715 } arch_t;
1716
1717 #ifndef EM_486
1718 #define EM_486 6 /* Intel 80486 */
1719 #endif
1720 #ifndef EM_AARCH64
1721 #define EM_AARCH64 183 /* ARM AARCH64 */
1722 #endif
1723
1724 static const arch_t arch_array[]={
1725 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1726 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1727 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1728 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1729 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1730 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1731 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1732 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1733 #if defined(VM_LITTLE_ENDIAN)
1734 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1735 #else
1736 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1737 #endif
1738 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1739 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1740 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1741 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1742 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1743 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1744 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1745 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1746 };
1747
1748 #if (defined IA32)
1749 static Elf32_Half running_arch_code=EM_386;
1750 #elif (defined AMD64)
1751 static Elf32_Half running_arch_code=EM_X86_64;
1752 #elif (defined IA64)
1753 static Elf32_Half running_arch_code=EM_IA_64;
1754 #elif (defined __sparc) && (defined _LP64)
1755 static Elf32_Half running_arch_code=EM_SPARCV9;
1756 #elif (defined __sparc) && (!defined _LP64)
1757 static Elf32_Half running_arch_code=EM_SPARC;
1758 #elif (defined __powerpc64__)
1759 static Elf32_Half running_arch_code=EM_PPC64;
1760 #elif (defined __powerpc__)
1761 static Elf32_Half running_arch_code=EM_PPC;
1762 #elif (defined ARM)
1763 static Elf32_Half running_arch_code=EM_ARM;
1764 #elif (defined S390)
1765 static Elf32_Half running_arch_code=EM_S390;
1766 #elif (defined ALPHA)
1767 static Elf32_Half running_arch_code=EM_ALPHA;
1768 #elif (defined MIPSEL)
1769 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1770 #elif (defined PARISC)
1771 static Elf32_Half running_arch_code=EM_PARISC;
1772 #elif (defined MIPS)
1773 static Elf32_Half running_arch_code=EM_MIPS;
1774 #elif (defined M68K)
1775 static Elf32_Half running_arch_code=EM_68K;
1776 #elif (defined AARCH64)
1777 static Elf32_Half running_arch_code=EM_AARCH64;
1778 #else
1779 #error Method os::dll_load requires that one of following is defined:\
1780 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1781 #endif
1782
1783 // Identify compatability class for VM's architecture and library's architecture
1784 // Obtain string descriptions for architectures
1785
1786 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1787 int running_arch_index=-1;
1788
1789 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1790 if (running_arch_code == arch_array[i].code) {
1791 running_arch_index = i;
1792 }
1793 if (lib_arch.code == arch_array[i].code) {
1794 lib_arch.compat_class = arch_array[i].compat_class;
1795 lib_arch.name = arch_array[i].name;
1796 }
1797 }
1798
1799 assert(running_arch_index != -1,
1800 "Didn't find running architecture code (running_arch_code) in arch_array");
1801 if (running_arch_index == -1) {
1802 // Even though running architecture detection failed
1803 // we may still continue with reporting dlerror() message
1804 return NULL;
1805 }
1806
1807 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1808 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1809 return NULL;
1810 }
1811
1812 #ifndef S390
1813 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1814 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1815 return NULL;
1816 }
1817 #endif // !S390
1818
1819 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1820 if (lib_arch.name!=NULL) {
1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1822 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1823 lib_arch.name, arch_array[running_arch_index].name);
1824 } else {
1825 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1826 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1827 lib_arch.code,
1828 arch_array[running_arch_index].name);
1829 }
1830 }
1831
1832 return NULL;
1833 }
1834
1835 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1836 int ebuflen) {
1837 void * result = ::dlopen(filename, RTLD_LAZY);
1838 if (result == NULL) {
1839 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1840 ebuf[ebuflen-1] = '\0';
1841 }
1842 return result;
1843 }
1844
1845 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1846 int ebuflen) {
1847 void * result = NULL;
1848 if (LoadExecStackDllInVMThread) {
1849 result = dlopen_helper(filename, ebuf, ebuflen);
1850 }
1851
1852 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1853 // library that requires an executable stack, or which does not have this
1854 // stack attribute set, dlopen changes the stack attribute to executable. The
1855 // read protection of the guard pages gets lost.
1856 //
1857 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1858 // may have been queued at the same time.
1859
1860 if (!_stack_is_executable) {
1861 JavaThread *jt = Threads::first();
1862
1863 while (jt) {
1864 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1865 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1866 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1867 warning("Attempt to reguard stack yellow zone failed.");
1868 }
1869 }
1870 jt = jt->next();
1871 }
1872 }
1873
1874 return result;
1875 }
1876
1877 void* os::dll_lookup(void* handle, const char* name) {
1878 void* res = dlsym(handle, name);
1879 return res;
1880 }
1881
1882 void* os::get_default_process_handle() {
1883 return (void*)::dlopen(NULL, RTLD_LAZY);
1884 }
1885
1886 static bool _print_ascii_file(const char* filename, outputStream* st) {
1887 int fd = ::open(filename, O_RDONLY);
1888 if (fd == -1) {
1889 return false;
1890 }
1891
1892 char buf[32];
1893 int bytes;
1894 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1895 st->print_raw(buf, bytes);
1896 }
1897
1898 ::close(fd);
1899
1900 return true;
1901 }
1902
1903 void os::print_dll_info(outputStream *st) {
1904 st->print_cr("Dynamic libraries:");
1905
1906 char fname[32];
1907 pid_t pid = os::Linux::gettid();
1908
1909 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1910
1911 if (!_print_ascii_file(fname, st)) {
1912 st->print("Can not get library information for pid = %d\n", pid);
1913 }
1914 }
1915
1916 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1917 FILE *procmapsFile = NULL;
1918
1919 // Open the procfs maps file for the current process
1920 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1921 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1922 char line[PATH_MAX + 100];
1923
1924 // Read line by line from 'file'
1925 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1926 u8 base, top, offset, inode;
1927 char permissions[5];
1928 char device[6];
1929 char name[PATH_MAX + 1];
1930
1931 // Parse fields from line
1932 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1933 &base, &top, permissions, &offset, device, &inode, name);
1934
1935 // Filter by device id '00:00' so that we only get file system mapped files.
1936 if (strcmp(device, "00:00") != 0) {
1937
1938 // Call callback with the fields of interest
1939 if(callback(name, (address)base, (address)top, param)) {
1940 // Oops abort, callback aborted
1941 fclose(procmapsFile);
1942 return 1;
1943 }
1944 }
1945 }
1946 fclose(procmapsFile);
1947 }
1948 return 0;
1949 }
1950
1951 void os::print_os_info_brief(outputStream* st) {
1952 os::Linux::print_distro_info(st);
1953
1954 os::Posix::print_uname_info(st);
1955
1956 os::Linux::print_libversion_info(st);
1957
1958 }
1959
1960 void os::print_os_info(outputStream* st) {
1961 st->print("OS:");
1962
1963 os::Linux::print_distro_info(st);
1964
1965 os::Posix::print_uname_info(st);
1966
1967 // Print warning if unsafe chroot environment detected
1968 if (unsafe_chroot_detected) {
1969 st->print("WARNING!! ");
1970 st->print_cr("%s", unstable_chroot_error);
1971 }
1972
1973 os::Linux::print_libversion_info(st);
1974
1975 os::Posix::print_rlimit_info(st);
1976
1977 os::Posix::print_load_average(st);
1978
1979 os::Linux::print_full_memory_info(st);
1980 }
1981
1982 // Try to identify popular distros.
1983 // Most Linux distributions have a /etc/XXX-release file, which contains
1984 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1985 // file that also contains the OS version string. Some have more than one
1986 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1987 // /etc/redhat-release.), so the order is important.
1988 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1989 // their own specific XXX-release file as well as a redhat-release file.
1990 // Because of this the XXX-release file needs to be searched for before the
1991 // redhat-release file.
1992 // Since Red Hat has a lsb-release file that is not very descriptive the
1993 // search for redhat-release needs to be before lsb-release.
1994 // Since the lsb-release file is the new standard it needs to be searched
1995 // before the older style release files.
1996 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1997 // next to last resort. The os-release file is a new standard that contains
1998 // distribution information and the system-release file seems to be an old
1999 // standard that has been replaced by the lsb-release and os-release files.
2000 // Searching for the debian_version file is the last resort. It contains
2001 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2002 // "Debian " is printed before the contents of the debian_version file.
2003
2004 const char* distro_files[] = {
2005 "/etc/oracle-release",
2006 "/etc/mandriva-release",
2007 "/etc/mandrake-release",
2008 "/etc/sun-release",
2009 "/etc/redhat-release",
2010 "/etc/lsb-release",
2011 "/etc/SuSE-release",
2012 "/etc/turbolinux-release",
2013 "/etc/gentoo-release",
2014 "/etc/ltib-release",
2015 "/etc/angstrom-version",
2016 "/etc/system-release",
2017 "/etc/os-release",
2018 NULL };
2019
2020 void os::Linux::print_distro_info(outputStream* st) {
2021 for (int i = 0;; i++) {
2022 const char* file = distro_files[i];
2023 if (file == NULL) {
2024 break; // done
2025 }
2026 // If file prints, we found it.
2027 if (_print_ascii_file(file, st)) {
2028 return;
2029 }
2030 }
2031
2032 if (file_exists("/etc/debian_version")) {
2033 st->print("Debian ");
2034 _print_ascii_file("/etc/debian_version", st);
2035 } else {
2036 st->print("Linux");
2037 }
2038 st->cr();
2039 }
2040
2041 static void parse_os_info(char* distro, size_t length, const char* file) {
2042 FILE* fp = fopen(file, "r");
2043 if (fp != NULL) {
2044 char buf[256];
2045 // get last line of the file.
2046 while (fgets(buf, sizeof(buf), fp)) { }
2047 // Edit out extra stuff in expected ubuntu format
2048 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2049 char* ptr = strstr(buf, "\""); // the name is in quotes
2050 if (ptr != NULL) {
2051 ptr++; // go beyond first quote
2052 char* nl = strchr(ptr, '\"');
2053 if (nl != NULL) *nl = '\0';
2054 strncpy(distro, ptr, length);
2055 } else {
2056 ptr = strstr(buf, "=");
2057 ptr++; // go beyond equals then
2058 char* nl = strchr(ptr, '\n');
2059 if (nl != NULL) *nl = '\0';
2060 strncpy(distro, ptr, length);
2061 }
2062 } else {
2063 // if not in expected Ubuntu format, print out whole line minus \n
2064 char* nl = strchr(buf, '\n');
2065 if (nl != NULL) *nl = '\0';
2066 strncpy(distro, buf, length);
2067 }
2068 // close distro file
2069 fclose(fp);
2070 }
2071 }
2072
2073 void os::get_summary_os_info(char* buf, size_t buflen) {
2074 for (int i = 0;; i++) {
2075 const char* file = distro_files[i];
2076 if (file == NULL) {
2077 break; // ran out of distro_files
2078 }
2079 if (file_exists(file)) {
2080 parse_os_info(buf, buflen, file);
2081 return;
2082 }
2083 }
2084 // special case for debian
2085 if (file_exists("/etc/debian_version")) {
2086 strncpy(buf, "Debian ", buflen);
2087 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2088 } else {
2089 strncpy(buf, "Linux", buflen);
2090 }
2091 }
2092
2093 void os::Linux::print_libversion_info(outputStream* st) {
2094 // libc, pthread
2095 st->print("libc:");
2096 st->print("%s ", os::Linux::glibc_version());
2097 st->print("%s ", os::Linux::libpthread_version());
2098 st->cr();
2099 }
2100
2101 void os::Linux::print_full_memory_info(outputStream* st) {
2102 st->print("\n/proc/meminfo:\n");
2103 _print_ascii_file("/proc/meminfo", st);
2104 st->cr();
2105 }
2106
2107 void os::print_memory_info(outputStream* st) {
2108
2109 st->print("Memory:");
2110 st->print(" %dk page", os::vm_page_size()>>10);
2111
2112 // values in struct sysinfo are "unsigned long"
2113 struct sysinfo si;
2114 sysinfo(&si);
2115
2116 st->print(", physical " UINT64_FORMAT "k",
2117 os::physical_memory() >> 10);
2118 st->print("(" UINT64_FORMAT "k free)",
2119 os::available_memory() >> 10);
2120 st->print(", swap " UINT64_FORMAT "k",
2121 ((jlong)si.totalswap * si.mem_unit) >> 10);
2122 st->print("(" UINT64_FORMAT "k free)",
2123 ((jlong)si.freeswap * si.mem_unit) >> 10);
2124 st->cr();
2125 }
2126
2127 // Print the first "model name" line and the first "flags" line
2128 // that we find and nothing more. We assume "model name" comes
2129 // before "flags" so if we find a second "model name", then the
2130 // "flags" field is considered missing.
2131 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2132 #if defined(IA32) || defined(AMD64)
2133 // Other platforms have less repetitive cpuinfo files
2134 FILE *fp = fopen("/proc/cpuinfo", "r");
2135 if (fp) {
2136 while (!feof(fp)) {
2137 if (fgets(buf, buflen, fp)) {
2138 // Assume model name comes before flags
2139 bool model_name_printed = false;
2140 if (strstr(buf, "model name") != NULL) {
2141 if (!model_name_printed) {
2142 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2143 st->print_raw(buf);
2144 model_name_printed = true;
2145 } else {
2146 // model name printed but not flags? Odd, just return
2147 fclose(fp);
2148 return true;
2149 }
2150 }
2151 // print the flags line too
2152 if (strstr(buf, "flags") != NULL) {
2153 st->print_raw(buf);
2154 fclose(fp);
2155 return true;
2156 }
2157 }
2158 }
2159 fclose(fp);
2160 }
2161 #endif // x86 platforms
2162 return false;
2163 }
2164
2165 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2166 // Only print the model name if the platform provides this as a summary
2167 if (!print_model_name_and_flags(st, buf, buflen)) {
2168 st->print("\n/proc/cpuinfo:\n");
2169 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2170 st->print_cr(" <Not Available>");
2171 }
2172 }
2173 }
2174
2175 #if defined(AMD64) || defined(IA32) || defined(X32)
2176 const char* search_string = "model name";
2177 #elif defined(SPARC)
2178 const char* search_string = "cpu";
2179 #elif defined(PPC64)
2180 const char* search_string = "cpu";
2181 #else
2182 const char* search_string = "Processor";
2183 #endif
2184
2185 // Parses the cpuinfo file for string representing the model name.
2186 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2187 FILE* fp = fopen("/proc/cpuinfo", "r");
2188 if (fp != NULL) {
2189 while (!feof(fp)) {
2190 char buf[256];
2191 if (fgets(buf, sizeof(buf), fp)) {
2192 char* start = strstr(buf, search_string);
2193 if (start != NULL) {
2194 char *ptr = start + strlen(search_string);
2195 char *end = buf + strlen(buf);
2196 while (ptr != end) {
2197 // skip whitespace and colon for the rest of the name.
2198 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2199 break;
2200 }
2201 ptr++;
2202 }
2203 if (ptr != end) {
2204 // reasonable string, get rid of newline and keep the rest
2205 char* nl = strchr(buf, '\n');
2206 if (nl != NULL) *nl = '\0';
2207 strncpy(cpuinfo, ptr, length);
2208 fclose(fp);
2209 return;
2210 }
2211 }
2212 }
2213 }
2214 fclose(fp);
2215 }
2216 // cpuinfo not found or parsing failed, just print generic string. The entire
2217 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2218 #if defined(AMD64)
2219 strncpy(cpuinfo, "x86_64", length);
2220 #elif defined(IA32)
2221 strncpy(cpuinfo, "x86_32", length);
2222 #elif defined(IA64)
2223 strncpy(cpuinfo, "IA64", length);
2224 #elif defined(SPARC)
2225 strncpy(cpuinfo, "sparcv9", length);
2226 #elif defined(AARCH64)
2227 strncpy(cpuinfo, "AArch64", length);
2228 #elif defined(ARM)
2229 strncpy(cpuinfo, "ARM", length);
2230 #elif defined(PPC)
2231 strncpy(cpuinfo, "PPC64", length);
2232 #elif defined(ZERO_LIBARCH)
2233 strncpy(cpuinfo, ZERO_LIBARCH, length);
2234 #else
2235 strncpy(cpuinfo, "unknown", length);
2236 #endif
2237 }
2238
2239 static void print_signal_handler(outputStream* st, int sig,
2240 char* buf, size_t buflen);
2241
2242 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2243 st->print_cr("Signal Handlers:");
2244 print_signal_handler(st, SIGSEGV, buf, buflen);
2245 print_signal_handler(st, SIGBUS , buf, buflen);
2246 print_signal_handler(st, SIGFPE , buf, buflen);
2247 print_signal_handler(st, SIGPIPE, buf, buflen);
2248 print_signal_handler(st, SIGXFSZ, buf, buflen);
2249 print_signal_handler(st, SIGILL , buf, buflen);
2250 print_signal_handler(st, SR_signum, buf, buflen);
2251 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2252 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2253 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2254 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2255 #if defined(PPC64)
2256 print_signal_handler(st, SIGTRAP, buf, buflen);
2257 #endif
2258 }
2259
2260 static char saved_jvm_path[MAXPATHLEN] = {0};
2261
2262 // Find the full path to the current module, libjvm.so
2263 void os::jvm_path(char *buf, jint buflen) {
2264 // Error checking.
2265 if (buflen < MAXPATHLEN) {
2266 assert(false, "must use a large-enough buffer");
2267 buf[0] = '\0';
2268 return;
2269 }
2270 // Lazy resolve the path to current module.
2271 if (saved_jvm_path[0] != 0) {
2272 strcpy(buf, saved_jvm_path);
2273 return;
2274 }
2275
2276 char dli_fname[MAXPATHLEN];
2277 bool ret = dll_address_to_library_name(
2278 CAST_FROM_FN_PTR(address, os::jvm_path),
2279 dli_fname, sizeof(dli_fname), NULL);
2280 assert(ret, "cannot locate libjvm");
2281 char *rp = NULL;
2282 if (ret && dli_fname[0] != '\0') {
2283 rp = realpath(dli_fname, buf);
2284 }
2285 if (rp == NULL) {
2286 return;
2287 }
2288
2289 if (Arguments::sun_java_launcher_is_altjvm()) {
2290 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2291 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2292 // If "/jre/lib/" appears at the right place in the string, then
2293 // assume we are installed in a JDK and we're done. Otherwise, check
2294 // for a JAVA_HOME environment variable and fix up the path so it
2295 // looks like libjvm.so is installed there (append a fake suffix
2296 // hotspot/libjvm.so).
2297 const char *p = buf + strlen(buf) - 1;
2298 for (int count = 0; p > buf && count < 5; ++count) {
2299 for (--p; p > buf && *p != '/'; --p)
2300 /* empty */ ;
2301 }
2302
2303 if (strncmp(p, "/jre/lib/", 9) != 0) {
2304 // Look for JAVA_HOME in the environment.
2305 char* java_home_var = ::getenv("JAVA_HOME");
2306 if (java_home_var != NULL && java_home_var[0] != 0) {
2307 char* jrelib_p;
2308 int len;
2309
2310 // Check the current module name "libjvm.so".
2311 p = strrchr(buf, '/');
2312 if (p == NULL) {
2313 return;
2314 }
2315 assert(strstr(p, "/libjvm") == p, "invalid library name");
2316
2317 rp = realpath(java_home_var, buf);
2318 if (rp == NULL) {
2319 return;
2320 }
2321
2322 // determine if this is a legacy image or modules image
2323 // modules image doesn't have "jre" subdirectory
2324 len = strlen(buf);
2325 assert(len < buflen, "Ran out of buffer room");
2326 jrelib_p = buf + len;
2327 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2328 if (0 != access(buf, F_OK)) {
2329 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2330 }
2331
2332 if (0 == access(buf, F_OK)) {
2333 // Use current module name "libjvm.so"
2334 len = strlen(buf);
2335 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2336 } else {
2337 // Go back to path of .so
2338 rp = realpath(dli_fname, buf);
2339 if (rp == NULL) {
2340 return;
2341 }
2342 }
2343 }
2344 }
2345 }
2346
2347 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2348 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2349 }
2350
2351 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2352 // no prefix required, not even "_"
2353 }
2354
2355 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2356 // no suffix required
2357 }
2358
2359 ////////////////////////////////////////////////////////////////////////////////
2360 // sun.misc.Signal support
2361
2362 static volatile jint sigint_count = 0;
2363
2364 static void UserHandler(int sig, void *siginfo, void *context) {
2365 // 4511530 - sem_post is serialized and handled by the manager thread. When
2366 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2367 // don't want to flood the manager thread with sem_post requests.
2368 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2369 return;
2370 }
2371
2372 // Ctrl-C is pressed during error reporting, likely because the error
2373 // handler fails to abort. Let VM die immediately.
2374 if (sig == SIGINT && is_error_reported()) {
2375 os::die();
2376 }
2377
2378 os::signal_notify(sig);
2379 }
2380
2381 void* os::user_handler() {
2382 return CAST_FROM_FN_PTR(void*, UserHandler);
2383 }
2384
2385 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2386 struct timespec ts;
2387 // Semaphore's are always associated with CLOCK_REALTIME
2388 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2389 // see unpackTime for discussion on overflow checking
2390 if (sec >= MAX_SECS) {
2391 ts.tv_sec += MAX_SECS;
2392 ts.tv_nsec = 0;
2393 } else {
2394 ts.tv_sec += sec;
2395 ts.tv_nsec += nsec;
2396 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2397 ts.tv_nsec -= NANOSECS_PER_SEC;
2398 ++ts.tv_sec; // note: this must be <= max_secs
2399 }
2400 }
2401
2402 return ts;
2403 }
2404
2405 extern "C" {
2406 typedef void (*sa_handler_t)(int);
2407 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2408 }
2409
2410 void* os::signal(int signal_number, void* handler) {
2411 struct sigaction sigAct, oldSigAct;
2412
2413 sigfillset(&(sigAct.sa_mask));
2414 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2415 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2416
2417 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2418 // -1 means registration failed
2419 return (void *)-1;
2420 }
2421
2422 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2423 }
2424
2425 void os::signal_raise(int signal_number) {
2426 ::raise(signal_number);
2427 }
2428
2429 // The following code is moved from os.cpp for making this
2430 // code platform specific, which it is by its very nature.
2431
2432 // Will be modified when max signal is changed to be dynamic
2433 int os::sigexitnum_pd() {
2434 return NSIG;
2435 }
2436
2437 // a counter for each possible signal value
2438 static volatile jint pending_signals[NSIG+1] = { 0 };
2439
2440 // Linux(POSIX) specific hand shaking semaphore.
2441 static sem_t sig_sem;
2442 static PosixSemaphore sr_semaphore;
2443
2444 void os::signal_init_pd() {
2445 // Initialize signal structures
2446 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2447
2448 // Initialize signal semaphore
2449 ::sem_init(&sig_sem, 0, 0);
2450 }
2451
2452 void os::signal_notify(int sig) {
2453 Atomic::inc(&pending_signals[sig]);
2454 ::sem_post(&sig_sem);
2455 }
2456
2457 static int check_pending_signals(bool wait) {
2458 Atomic::store(0, &sigint_count);
2459 for (;;) {
2460 for (int i = 0; i < NSIG + 1; i++) {
2461 jint n = pending_signals[i];
2462 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2463 return i;
2464 }
2465 }
2466 if (!wait) {
2467 return -1;
2468 }
2469 JavaThread *thread = JavaThread::current();
2470 ThreadBlockInVM tbivm(thread);
2471
2472 bool threadIsSuspended;
2473 do {
2474 thread->set_suspend_equivalent();
2475 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2476 ::sem_wait(&sig_sem);
2477
2478 // were we externally suspended while we were waiting?
2479 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2480 if (threadIsSuspended) {
2481 // The semaphore has been incremented, but while we were waiting
2482 // another thread suspended us. We don't want to continue running
2483 // while suspended because that would surprise the thread that
2484 // suspended us.
2485 ::sem_post(&sig_sem);
2486
2487 thread->java_suspend_self();
2488 }
2489 } while (threadIsSuspended);
2490 }
2491 }
2492
2493 int os::signal_lookup() {
2494 return check_pending_signals(false);
2495 }
2496
2497 int os::signal_wait() {
2498 return check_pending_signals(true);
2499 }
2500
2501 ////////////////////////////////////////////////////////////////////////////////
2502 // Virtual Memory
2503
2504 int os::vm_page_size() {
2505 // Seems redundant as all get out
2506 assert(os::Linux::page_size() != -1, "must call os::init");
2507 return os::Linux::page_size();
2508 }
2509
2510 // Solaris allocates memory by pages.
2511 int os::vm_allocation_granularity() {
2512 assert(os::Linux::page_size() != -1, "must call os::init");
2513 return os::Linux::page_size();
2514 }
2515
2516 // Rationale behind this function:
2517 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2518 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2519 // samples for JITted code. Here we create private executable mapping over the code cache
2520 // and then we can use standard (well, almost, as mapping can change) way to provide
2521 // info for the reporting script by storing timestamp and location of symbol
2522 void linux_wrap_code(char* base, size_t size) {
2523 static volatile jint cnt = 0;
2524
2525 if (!UseOprofile) {
2526 return;
2527 }
2528
2529 char buf[PATH_MAX+1];
2530 int num = Atomic::add(1, &cnt);
2531
2532 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2533 os::get_temp_directory(), os::current_process_id(), num);
2534 unlink(buf);
2535
2536 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2537
2538 if (fd != -1) {
2539 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2540 if (rv != (off_t)-1) {
2541 if (::write(fd, "", 1) == 1) {
2542 mmap(base, size,
2543 PROT_READ|PROT_WRITE|PROT_EXEC,
2544 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2545 }
2546 }
2547 ::close(fd);
2548 unlink(buf);
2549 }
2550 }
2551
2552 static bool recoverable_mmap_error(int err) {
2553 // See if the error is one we can let the caller handle. This
2554 // list of errno values comes from JBS-6843484. I can't find a
2555 // Linux man page that documents this specific set of errno
2556 // values so while this list currently matches Solaris, it may
2557 // change as we gain experience with this failure mode.
2558 switch (err) {
2559 case EBADF:
2560 case EINVAL:
2561 case ENOTSUP:
2562 // let the caller deal with these errors
2563 return true;
2564
2565 default:
2566 // Any remaining errors on this OS can cause our reserved mapping
2567 // to be lost. That can cause confusion where different data
2568 // structures think they have the same memory mapped. The worst
2569 // scenario is if both the VM and a library think they have the
2570 // same memory mapped.
2571 return false;
2572 }
2573 }
2574
2575 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2576 int err) {
2577 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2578 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2579 strerror(err), err);
2580 }
2581
2582 static void warn_fail_commit_memory(char* addr, size_t size,
2583 size_t alignment_hint, bool exec,
2584 int err) {
2585 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2586 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2587 alignment_hint, exec, strerror(err), err);
2588 }
2589
2590 // NOTE: Linux kernel does not really reserve the pages for us.
2591 // All it does is to check if there are enough free pages
2592 // left at the time of mmap(). This could be a potential
2593 // problem.
2594 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2595 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2596 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2597 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2598 if (res != (uintptr_t) MAP_FAILED) {
2599 if (UseNUMAInterleaving) {
2600 numa_make_global(addr, size);
2601 }
2602 return 0;
2603 }
2604
2605 int err = errno; // save errno from mmap() call above
2606
2607 if (!recoverable_mmap_error(err)) {
2608 warn_fail_commit_memory(addr, size, exec, err);
2609 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2610 }
2611
2612 return err;
2613 }
2614
2615 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2616 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2617 }
2618
2619 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2620 const char* mesg) {
2621 assert(mesg != NULL, "mesg must be specified");
2622 int err = os::Linux::commit_memory_impl(addr, size, exec);
2623 if (err != 0) {
2624 // the caller wants all commit errors to exit with the specified mesg:
2625 warn_fail_commit_memory(addr, size, exec, err);
2626 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2627 }
2628 }
2629
2630 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2631 #ifndef MAP_HUGETLB
2632 #define MAP_HUGETLB 0x40000
2633 #endif
2634
2635 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2636 #ifndef MADV_HUGEPAGE
2637 #define MADV_HUGEPAGE 14
2638 #endif
2639
2640 int os::Linux::commit_memory_impl(char* addr, size_t size,
2641 size_t alignment_hint, bool exec) {
2642 int err = os::Linux::commit_memory_impl(addr, size, exec);
2643 if (err == 0) {
2644 realign_memory(addr, size, alignment_hint);
2645 }
2646 return err;
2647 }
2648
2649 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2650 bool exec) {
2651 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2652 }
2653
2654 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2655 size_t alignment_hint, bool exec,
2656 const char* mesg) {
2657 assert(mesg != NULL, "mesg must be specified");
2658 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2659 if (err != 0) {
2660 // the caller wants all commit errors to exit with the specified mesg:
2661 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2662 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2663 }
2664 }
2665
2666 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2667 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2668 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2669 // be supported or the memory may already be backed by huge pages.
2670 ::madvise(addr, bytes, MADV_HUGEPAGE);
2671 }
2672 }
2673
2674 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2675 // This method works by doing an mmap over an existing mmaping and effectively discarding
2676 // the existing pages. However it won't work for SHM-based large pages that cannot be
2677 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2678 // small pages on top of the SHM segment. This method always works for small pages, so we
2679 // allow that in any case.
2680 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2681 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2682 }
2683 }
2684
2685 void os::numa_make_global(char *addr, size_t bytes) {
2686 Linux::numa_interleave_memory(addr, bytes);
2687 }
2688
2689 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2690 // bind policy to MPOL_PREFERRED for the current thread.
2691 #define USE_MPOL_PREFERRED 0
2692
2693 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2694 // To make NUMA and large pages more robust when both enabled, we need to ease
2695 // the requirements on where the memory should be allocated. MPOL_BIND is the
2696 // default policy and it will force memory to be allocated on the specified
2697 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2698 // the specified node, but will not force it. Using this policy will prevent
2699 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2700 // free large pages.
2701 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2702 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2703 }
2704
2705 bool os::numa_topology_changed() { return false; }
2706
2707 size_t os::numa_get_groups_num() {
2708 int max_node = Linux::numa_max_node();
2709 return max_node > 0 ? max_node + 1 : 1;
2710 }
2711
2712 int os::numa_get_group_id() {
2713 int cpu_id = Linux::sched_getcpu();
2714 if (cpu_id != -1) {
2715 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2716 if (lgrp_id != -1) {
2717 return lgrp_id;
2718 }
2719 }
2720 return 0;
2721 }
2722
2723 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2724 for (size_t i = 0; i < size; i++) {
2725 ids[i] = i;
2726 }
2727 return size;
2728 }
2729
2730 bool os::get_page_info(char *start, page_info* info) {
2731 return false;
2732 }
2733
2734 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2735 page_info* page_found) {
2736 return end;
2737 }
2738
2739
2740 int os::Linux::sched_getcpu_syscall(void) {
2741 unsigned int cpu = 0;
2742 int retval = -1;
2743
2744 #if defined(IA32)
2745 #ifndef SYS_getcpu
2746 #define SYS_getcpu 318
2747 #endif
2748 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2749 #elif defined(AMD64)
2750 // Unfortunately we have to bring all these macros here from vsyscall.h
2751 // to be able to compile on old linuxes.
2752 #define __NR_vgetcpu 2
2753 #define VSYSCALL_START (-10UL << 20)
2754 #define VSYSCALL_SIZE 1024
2755 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2756 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2757 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2758 retval = vgetcpu(&cpu, NULL, NULL);
2759 #endif
2760
2761 return (retval == -1) ? retval : cpu;
2762 }
2763
2764 // Something to do with the numa-aware allocator needs these symbols
2765 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2766 extern "C" JNIEXPORT void numa_error(char *where) { }
2767
2768
2769 // If we are running with libnuma version > 2, then we should
2770 // be trying to use symbols with versions 1.1
2771 // If we are running with earlier version, which did not have symbol versions,
2772 // we should use the base version.
2773 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2774 void *f = dlvsym(handle, name, "libnuma_1.1");
2775 if (f == NULL) {
2776 f = dlsym(handle, name);
2777 }
2778 return f;
2779 }
2780
2781 bool os::Linux::libnuma_init() {
2782 // sched_getcpu() should be in libc.
2783 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2784 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2785
2786 // If it's not, try a direct syscall.
2787 if (sched_getcpu() == -1) {
2788 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2789 (void*)&sched_getcpu_syscall));
2790 }
2791
2792 if (sched_getcpu() != -1) { // Does it work?
2793 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2794 if (handle != NULL) {
2795 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2796 libnuma_dlsym(handle, "numa_node_to_cpus")));
2797 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2798 libnuma_dlsym(handle, "numa_max_node")));
2799 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2800 libnuma_dlsym(handle, "numa_available")));
2801 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2802 libnuma_dlsym(handle, "numa_tonode_memory")));
2803 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2804 libnuma_dlsym(handle, "numa_interleave_memory")));
2805 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2806 libnuma_dlsym(handle, "numa_set_bind_policy")));
2807
2808
2809 if (numa_available() != -1) {
2810 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2811 // Create a cpu -> node mapping
2812 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2813 rebuild_cpu_to_node_map();
2814 return true;
2815 }
2816 }
2817 }
2818 return false;
2819 }
2820
2821 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2822 // The table is later used in get_node_by_cpu().
2823 void os::Linux::rebuild_cpu_to_node_map() {
2824 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2825 // in libnuma (possible values are starting from 16,
2826 // and continuing up with every other power of 2, but less
2827 // than the maximum number of CPUs supported by kernel), and
2828 // is a subject to change (in libnuma version 2 the requirements
2829 // are more reasonable) we'll just hardcode the number they use
2830 // in the library.
2831 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2832
2833 size_t cpu_num = os::active_processor_count();
2834 size_t cpu_map_size = NCPUS / BitsPerCLong;
2835 size_t cpu_map_valid_size =
2836 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2837
2838 cpu_to_node()->clear();
2839 cpu_to_node()->at_grow(cpu_num - 1);
2840 size_t node_num = numa_get_groups_num();
2841
2842 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2843 for (size_t i = 0; i < node_num; i++) {
2844 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2845 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2846 if (cpu_map[j] != 0) {
2847 for (size_t k = 0; k < BitsPerCLong; k++) {
2848 if (cpu_map[j] & (1UL << k)) {
2849 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2850 }
2851 }
2852 }
2853 }
2854 }
2855 }
2856 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2857 }
2858
2859 int os::Linux::get_node_by_cpu(int cpu_id) {
2860 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2861 return cpu_to_node()->at(cpu_id);
2862 }
2863 return -1;
2864 }
2865
2866 GrowableArray<int>* os::Linux::_cpu_to_node;
2867 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2868 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2869 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2870 os::Linux::numa_available_func_t os::Linux::_numa_available;
2871 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2872 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2873 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2874 unsigned long* os::Linux::_numa_all_nodes;
2875
2876 bool os::pd_uncommit_memory(char* addr, size_t size) {
2877 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2878 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2879 return res != (uintptr_t) MAP_FAILED;
2880 }
2881
2882 static address get_stack_commited_bottom(address bottom, size_t size) {
2883 address nbot = bottom;
2884 address ntop = bottom + size;
2885
2886 size_t page_sz = os::vm_page_size();
2887 unsigned pages = size / page_sz;
2888
2889 unsigned char vec[1];
2890 unsigned imin = 1, imax = pages + 1, imid;
2891 int mincore_return_value = 0;
2892
2893 assert(imin <= imax, "Unexpected page size");
2894
2895 while (imin < imax) {
2896 imid = (imax + imin) / 2;
2897 nbot = ntop - (imid * page_sz);
2898
2899 // Use a trick with mincore to check whether the page is mapped or not.
2900 // mincore sets vec to 1 if page resides in memory and to 0 if page
2901 // is swapped output but if page we are asking for is unmapped
2902 // it returns -1,ENOMEM
2903 mincore_return_value = mincore(nbot, page_sz, vec);
2904
2905 if (mincore_return_value == -1) {
2906 // Page is not mapped go up
2907 // to find first mapped page
2908 if (errno != EAGAIN) {
2909 assert(errno == ENOMEM, "Unexpected mincore errno");
2910 imax = imid;
2911 }
2912 } else {
2913 // Page is mapped go down
2914 // to find first not mapped page
2915 imin = imid + 1;
2916 }
2917 }
2918
2919 nbot = nbot + page_sz;
2920
2921 // Adjust stack bottom one page up if last checked page is not mapped
2922 if (mincore_return_value == -1) {
2923 nbot = nbot + page_sz;
2924 }
2925
2926 return nbot;
2927 }
2928
2929
2930 // Linux uses a growable mapping for the stack, and if the mapping for
2931 // the stack guard pages is not removed when we detach a thread the
2932 // stack cannot grow beyond the pages where the stack guard was
2933 // mapped. If at some point later in the process the stack expands to
2934 // that point, the Linux kernel cannot expand the stack any further
2935 // because the guard pages are in the way, and a segfault occurs.
2936 //
2937 // However, it's essential not to split the stack region by unmapping
2938 // a region (leaving a hole) that's already part of the stack mapping,
2939 // so if the stack mapping has already grown beyond the guard pages at
2940 // the time we create them, we have to truncate the stack mapping.
2941 // So, we need to know the extent of the stack mapping when
2942 // create_stack_guard_pages() is called.
2943
2944 // We only need this for stacks that are growable: at the time of
2945 // writing thread stacks don't use growable mappings (i.e. those
2946 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2947 // only applies to the main thread.
2948
2949 // If the (growable) stack mapping already extends beyond the point
2950 // where we're going to put our guard pages, truncate the mapping at
2951 // that point by munmap()ping it. This ensures that when we later
2952 // munmap() the guard pages we don't leave a hole in the stack
2953 // mapping. This only affects the main/initial thread
2954
2955 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2956 if (os::Linux::is_initial_thread()) {
2957 // As we manually grow stack up to bottom inside create_attached_thread(),
2958 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2959 // we don't need to do anything special.
2960 // Check it first, before calling heavy function.
2961 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2962 unsigned char vec[1];
2963
2964 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2965 // Fallback to slow path on all errors, including EAGAIN
2966 stack_extent = (uintptr_t) get_stack_commited_bottom(
2967 os::Linux::initial_thread_stack_bottom(),
2968 (size_t)addr - stack_extent);
2969 }
2970
2971 if (stack_extent < (uintptr_t)addr) {
2972 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2973 }
2974 }
2975
2976 return os::commit_memory(addr, size, !ExecMem);
2977 }
2978
2979 // If this is a growable mapping, remove the guard pages entirely by
2980 // munmap()ping them. If not, just call uncommit_memory(). This only
2981 // affects the main/initial thread, but guard against future OS changes
2982 // It's safe to always unmap guard pages for initial thread because we
2983 // always place it right after end of the mapped region
2984
2985 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2986 uintptr_t stack_extent, stack_base;
2987
2988 if (os::Linux::is_initial_thread()) {
2989 return ::munmap(addr, size) == 0;
2990 }
2991
2992 return os::uncommit_memory(addr, size);
2993 }
2994
2995 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2996 // at 'requested_addr'. If there are existing memory mappings at the same
2997 // location, however, they will be overwritten. If 'fixed' is false,
2998 // 'requested_addr' is only treated as a hint, the return value may or
2999 // may not start from the requested address. Unlike Linux mmap(), this
3000 // function returns NULL to indicate failure.
3001 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3002 char * addr;
3003 int flags;
3004
3005 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3006 if (fixed) {
3007 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3008 flags |= MAP_FIXED;
3009 }
3010
3011 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3012 // touch an uncommitted page. Otherwise, the read/write might
3013 // succeed if we have enough swap space to back the physical page.
3014 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3015 flags, -1, 0);
3016
3017 return addr == MAP_FAILED ? NULL : addr;
3018 }
3019
3020 static int anon_munmap(char * addr, size_t size) {
3021 return ::munmap(addr, size) == 0;
3022 }
3023
3024 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3025 size_t alignment_hint) {
3026 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3027 }
3028
3029 bool os::pd_release_memory(char* addr, size_t size) {
3030 return anon_munmap(addr, size);
3031 }
3032
3033 static bool linux_mprotect(char* addr, size_t size, int prot) {
3034 // Linux wants the mprotect address argument to be page aligned.
3035 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3036
3037 // According to SUSv3, mprotect() should only be used with mappings
3038 // established by mmap(), and mmap() always maps whole pages. Unaligned
3039 // 'addr' likely indicates problem in the VM (e.g. trying to change
3040 // protection of malloc'ed or statically allocated memory). Check the
3041 // caller if you hit this assert.
3042 assert(addr == bottom, "sanity check");
3043
3044 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3045 return ::mprotect(bottom, size, prot) == 0;
3046 }
3047
3048 // Set protections specified
3049 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3050 bool is_committed) {
3051 unsigned int p = 0;
3052 switch (prot) {
3053 case MEM_PROT_NONE: p = PROT_NONE; break;
3054 case MEM_PROT_READ: p = PROT_READ; break;
3055 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3056 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3057 default:
3058 ShouldNotReachHere();
3059 }
3060 // is_committed is unused.
3061 return linux_mprotect(addr, bytes, p);
3062 }
3063
3064 bool os::guard_memory(char* addr, size_t size) {
3065 return linux_mprotect(addr, size, PROT_NONE);
3066 }
3067
3068 bool os::unguard_memory(char* addr, size_t size) {
3069 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3070 }
3071
3072 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3073 size_t page_size) {
3074 bool result = false;
3075 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3076 MAP_ANONYMOUS|MAP_PRIVATE,
3077 -1, 0);
3078 if (p != MAP_FAILED) {
3079 void *aligned_p = align_ptr_up(p, page_size);
3080
3081 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3082
3083 munmap(p, page_size * 2);
3084 }
3085
3086 if (warn && !result) {
3087 warning("TransparentHugePages is not supported by the operating system.");
3088 }
3089
3090 return result;
3091 }
3092
3093 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3094 bool result = false;
3095 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3096 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3097 -1, 0);
3098
3099 if (p != MAP_FAILED) {
3100 // We don't know if this really is a huge page or not.
3101 FILE *fp = fopen("/proc/self/maps", "r");
3102 if (fp) {
3103 while (!feof(fp)) {
3104 char chars[257];
3105 long x = 0;
3106 if (fgets(chars, sizeof(chars), fp)) {
3107 if (sscanf(chars, "%lx-%*x", &x) == 1
3108 && x == (long)p) {
3109 if (strstr (chars, "hugepage")) {
3110 result = true;
3111 break;
3112 }
3113 }
3114 }
3115 }
3116 fclose(fp);
3117 }
3118 munmap(p, page_size);
3119 }
3120
3121 if (warn && !result) {
3122 warning("HugeTLBFS is not supported by the operating system.");
3123 }
3124
3125 return result;
3126 }
3127
3128 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3129 //
3130 // From the coredump_filter documentation:
3131 //
3132 // - (bit 0) anonymous private memory
3133 // - (bit 1) anonymous shared memory
3134 // - (bit 2) file-backed private memory
3135 // - (bit 3) file-backed shared memory
3136 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3137 // effective only if the bit 2 is cleared)
3138 // - (bit 5) hugetlb private memory
3139 // - (bit 6) hugetlb shared memory
3140 //
3141 static void set_coredump_filter(void) {
3142 FILE *f;
3143 long cdm;
3144
3145 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3146 return;
3147 }
3148
3149 if (fscanf(f, "%lx", &cdm) != 1) {
3150 fclose(f);
3151 return;
3152 }
3153
3154 rewind(f);
3155
3156 if ((cdm & LARGEPAGES_BIT) == 0) {
3157 cdm |= LARGEPAGES_BIT;
3158 fprintf(f, "%#lx", cdm);
3159 }
3160
3161 fclose(f);
3162 }
3163
3164 // Large page support
3165
3166 static size_t _large_page_size = 0;
3167
3168 size_t os::Linux::find_large_page_size() {
3169 size_t large_page_size = 0;
3170
3171 // large_page_size on Linux is used to round up heap size. x86 uses either
3172 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3173 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3174 // page as large as 256M.
3175 //
3176 // Here we try to figure out page size by parsing /proc/meminfo and looking
3177 // for a line with the following format:
3178 // Hugepagesize: 2048 kB
3179 //
3180 // If we can't determine the value (e.g. /proc is not mounted, or the text
3181 // format has been changed), we'll use the largest page size supported by
3182 // the processor.
3183
3184 #ifndef ZERO
3185 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3186 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3187 #endif // ZERO
3188
3189 FILE *fp = fopen("/proc/meminfo", "r");
3190 if (fp) {
3191 while (!feof(fp)) {
3192 int x = 0;
3193 char buf[16];
3194 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3195 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3196 large_page_size = x * K;
3197 break;
3198 }
3199 } else {
3200 // skip to next line
3201 for (;;) {
3202 int ch = fgetc(fp);
3203 if (ch == EOF || ch == (int)'\n') break;
3204 }
3205 }
3206 }
3207 fclose(fp);
3208 }
3209
3210 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3211 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3212 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3213 proper_unit_for_byte_size(large_page_size));
3214 }
3215
3216 return large_page_size;
3217 }
3218
3219 size_t os::Linux::setup_large_page_size() {
3220 _large_page_size = Linux::find_large_page_size();
3221 const size_t default_page_size = (size_t)Linux::page_size();
3222 if (_large_page_size > default_page_size) {
3223 _page_sizes[0] = _large_page_size;
3224 _page_sizes[1] = default_page_size;
3225 _page_sizes[2] = 0;
3226 }
3227
3228 return _large_page_size;
3229 }
3230
3231 bool os::Linux::setup_large_page_type(size_t page_size) {
3232 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3233 FLAG_IS_DEFAULT(UseSHM) &&
3234 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3235
3236 // The type of large pages has not been specified by the user.
3237
3238 // Try UseHugeTLBFS and then UseSHM.
3239 UseHugeTLBFS = UseSHM = true;
3240
3241 // Don't try UseTransparentHugePages since there are known
3242 // performance issues with it turned on. This might change in the future.
3243 UseTransparentHugePages = false;
3244 }
3245
3246 if (UseTransparentHugePages) {
3247 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3248 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3249 UseHugeTLBFS = false;
3250 UseSHM = false;
3251 return true;
3252 }
3253 UseTransparentHugePages = false;
3254 }
3255
3256 if (UseHugeTLBFS) {
3257 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3258 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3259 UseSHM = false;
3260 return true;
3261 }
3262 UseHugeTLBFS = false;
3263 }
3264
3265 return UseSHM;
3266 }
3267
3268 void os::large_page_init() {
3269 if (!UseLargePages &&
3270 !UseTransparentHugePages &&
3271 !UseHugeTLBFS &&
3272 !UseSHM) {
3273 // Not using large pages.
3274 return;
3275 }
3276
3277 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3278 // The user explicitly turned off large pages.
3279 // Ignore the rest of the large pages flags.
3280 UseTransparentHugePages = false;
3281 UseHugeTLBFS = false;
3282 UseSHM = false;
3283 return;
3284 }
3285
3286 size_t large_page_size = Linux::setup_large_page_size();
3287 UseLargePages = Linux::setup_large_page_type(large_page_size);
3288
3289 set_coredump_filter();
3290 }
3291
3292 #ifndef SHM_HUGETLB
3293 #define SHM_HUGETLB 04000
3294 #endif
3295
3296 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3297 char* req_addr, bool exec) {
3298 // "exec" is passed in but not used. Creating the shared image for
3299 // the code cache doesn't have an SHM_X executable permission to check.
3300 assert(UseLargePages && UseSHM, "only for SHM large pages");
3301 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3302
3303 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3304 return NULL; // Fallback to small pages.
3305 }
3306
3307 key_t key = IPC_PRIVATE;
3308 char *addr;
3309
3310 bool warn_on_failure = UseLargePages &&
3311 (!FLAG_IS_DEFAULT(UseLargePages) ||
3312 !FLAG_IS_DEFAULT(UseSHM) ||
3313 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3314 char msg[128];
3315
3316 // Create a large shared memory region to attach to based on size.
3317 // Currently, size is the total size of the heap
3318 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3319 if (shmid == -1) {
3320 // Possible reasons for shmget failure:
3321 // 1. shmmax is too small for Java heap.
3322 // > check shmmax value: cat /proc/sys/kernel/shmmax
3323 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3324 // 2. not enough large page memory.
3325 // > check available large pages: cat /proc/meminfo
3326 // > increase amount of large pages:
3327 // echo new_value > /proc/sys/vm/nr_hugepages
3328 // Note 1: different Linux may use different name for this property,
3329 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3330 // Note 2: it's possible there's enough physical memory available but
3331 // they are so fragmented after a long run that they can't
3332 // coalesce into large pages. Try to reserve large pages when
3333 // the system is still "fresh".
3334 if (warn_on_failure) {
3335 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3336 warning("%s", msg);
3337 }
3338 return NULL;
3339 }
3340
3341 // attach to the region
3342 addr = (char*)shmat(shmid, req_addr, 0);
3343 int err = errno;
3344
3345 // Remove shmid. If shmat() is successful, the actual shared memory segment
3346 // will be deleted when it's detached by shmdt() or when the process
3347 // terminates. If shmat() is not successful this will remove the shared
3348 // segment immediately.
3349 shmctl(shmid, IPC_RMID, NULL);
3350
3351 if ((intptr_t)addr == -1) {
3352 if (warn_on_failure) {
3353 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3354 warning("%s", msg);
3355 }
3356 return NULL;
3357 }
3358
3359 return addr;
3360 }
3361
3362 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3363 int error) {
3364 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3365
3366 bool warn_on_failure = UseLargePages &&
3367 (!FLAG_IS_DEFAULT(UseLargePages) ||
3368 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3369 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3370
3371 if (warn_on_failure) {
3372 char msg[128];
3373 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3374 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3375 warning("%s", msg);
3376 }
3377 }
3378
3379 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3380 char* req_addr,
3381 bool exec) {
3382 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3383 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3384 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3385
3386 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3387 char* addr = (char*)::mmap(req_addr, bytes, prot,
3388 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3389 -1, 0);
3390
3391 if (addr == MAP_FAILED) {
3392 warn_on_large_pages_failure(req_addr, bytes, errno);
3393 return NULL;
3394 }
3395
3396 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3397
3398 return addr;
3399 }
3400
3401 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3402 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3403 // (req_addr != NULL) or with a given alignment.
3404 // - bytes shall be a multiple of alignment.
3405 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3406 // - alignment sets the alignment at which memory shall be allocated.
3407 // It must be a multiple of allocation granularity.
3408 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3409 // req_addr or NULL.
3410 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3411
3412 size_t extra_size = bytes;
3413 if (req_addr == NULL && alignment > 0) {
3414 extra_size += alignment;
3415 }
3416
3417 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3418 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3419 -1, 0);
3420 if (start == MAP_FAILED) {
3421 start = NULL;
3422 } else {
3423 if (req_addr != NULL) {
3424 if (start != req_addr) {
3425 ::munmap(start, extra_size);
3426 start = NULL;
3427 }
3428 } else {
3429 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3430 char* const end_aligned = start_aligned + bytes;
3431 char* const end = start + extra_size;
3432 if (start_aligned > start) {
3433 ::munmap(start, start_aligned - start);
3434 }
3435 if (end_aligned < end) {
3436 ::munmap(end_aligned, end - end_aligned);
3437 }
3438 start = start_aligned;
3439 }
3440 }
3441 return start;
3442
3443 }
3444
3445 // Reserve memory using mmap(MAP_HUGETLB).
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 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3453 size_t alignment,
3454 char* req_addr,
3455 bool exec) {
3456 size_t large_page_size = os::large_page_size();
3457 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3458
3459 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3460 assert(is_size_aligned(bytes, alignment), "Must be");
3461
3462 // First reserve - but not commit - the address range in small pages.
3463 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3464
3465 if (start == NULL) {
3466 return NULL;
3467 }
3468
3469 assert(is_ptr_aligned(start, alignment), "Must be");
3470
3471 char* end = start + bytes;
3472
3473 // Find the regions of the allocated chunk that can be promoted to large pages.
3474 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3475 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3476
3477 size_t lp_bytes = lp_end - lp_start;
3478
3479 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3480
3481 if (lp_bytes == 0) {
3482 // The mapped region doesn't even span the start and the end of a large page.
3483 // Fall back to allocate a non-special area.
3484 ::munmap(start, end - start);
3485 return NULL;
3486 }
3487
3488 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3489
3490 void* result;
3491
3492 // Commit small-paged leading area.
3493 if (start != lp_start) {
3494 result = ::mmap(start, lp_start - start, prot,
3495 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3496 -1, 0);
3497 if (result == MAP_FAILED) {
3498 ::munmap(lp_start, end - lp_start);
3499 return NULL;
3500 }
3501 }
3502
3503 // Commit large-paged area.
3504 result = ::mmap(lp_start, lp_bytes, prot,
3505 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3506 -1, 0);
3507 if (result == MAP_FAILED) {
3508 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3509 // If the mmap above fails, the large pages region will be unmapped and we
3510 // have regions before and after with small pages. Release these regions.
3511 //
3512 // | mapped | unmapped | mapped |
3513 // ^ ^ ^ ^
3514 // start lp_start lp_end end
3515 //
3516 ::munmap(start, lp_start - start);
3517 ::munmap(lp_end, end - lp_end);
3518 return NULL;
3519 }
3520
3521 // Commit small-paged trailing area.
3522 if (lp_end != end) {
3523 result = ::mmap(lp_end, end - lp_end, prot,
3524 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3525 -1, 0);
3526 if (result == MAP_FAILED) {
3527 ::munmap(start, lp_end - start);
3528 return NULL;
3529 }
3530 }
3531
3532 return start;
3533 }
3534
3535 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3536 size_t alignment,
3537 char* req_addr,
3538 bool exec) {
3539 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3540 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3541 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3542 assert(is_power_of_2(os::large_page_size()), "Must be");
3543 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3544
3545 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3546 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3547 } else {
3548 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3549 }
3550 }
3551
3552 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3553 char* req_addr, bool exec) {
3554 assert(UseLargePages, "only for large pages");
3555
3556 char* addr;
3557 if (UseSHM) {
3558 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3559 } else {
3560 assert(UseHugeTLBFS, "must be");
3561 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3562 }
3563
3564 if (addr != NULL) {
3565 if (UseNUMAInterleaving) {
3566 numa_make_global(addr, bytes);
3567 }
3568
3569 // The memory is committed
3570 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3571 }
3572
3573 return addr;
3574 }
3575
3576 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3577 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3578 return shmdt(base) == 0;
3579 }
3580
3581 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3582 return pd_release_memory(base, bytes);
3583 }
3584
3585 bool os::release_memory_special(char* base, size_t bytes) {
3586 bool res;
3587 if (MemTracker::tracking_level() > NMT_minimal) {
3588 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3589 res = os::Linux::release_memory_special_impl(base, bytes);
3590 if (res) {
3591 tkr.record((address)base, bytes);
3592 }
3593
3594 } else {
3595 res = os::Linux::release_memory_special_impl(base, bytes);
3596 }
3597 return res;
3598 }
3599
3600 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3601 assert(UseLargePages, "only for large pages");
3602 bool res;
3603
3604 if (UseSHM) {
3605 res = os::Linux::release_memory_special_shm(base, bytes);
3606 } else {
3607 assert(UseHugeTLBFS, "must be");
3608 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3609 }
3610 return res;
3611 }
3612
3613 size_t os::large_page_size() {
3614 return _large_page_size;
3615 }
3616
3617 // With SysV SHM the entire memory region must be allocated as shared
3618 // memory.
3619 // HugeTLBFS allows application to commit large page memory on demand.
3620 // However, when committing memory with HugeTLBFS fails, the region
3621 // that was supposed to be committed will lose the old reservation
3622 // and allow other threads to steal that memory region. Because of this
3623 // behavior we can't commit HugeTLBFS memory.
3624 bool os::can_commit_large_page_memory() {
3625 return UseTransparentHugePages;
3626 }
3627
3628 bool os::can_execute_large_page_memory() {
3629 return UseTransparentHugePages || UseHugeTLBFS;
3630 }
3631
3632 // Reserve memory at an arbitrary address, only if that area is
3633 // available (and not reserved for something else).
3634
3635 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3636 const int max_tries = 10;
3637 char* base[max_tries];
3638 size_t size[max_tries];
3639 const size_t gap = 0x000000;
3640
3641 // Assert only that the size is a multiple of the page size, since
3642 // that's all that mmap requires, and since that's all we really know
3643 // about at this low abstraction level. If we need higher alignment,
3644 // we can either pass an alignment to this method or verify alignment
3645 // in one of the methods further up the call chain. See bug 5044738.
3646 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3647
3648 // Repeatedly allocate blocks until the block is allocated at the
3649 // right spot.
3650
3651 // Linux mmap allows caller to pass an address as hint; give it a try first,
3652 // if kernel honors the hint then we can return immediately.
3653 char * addr = anon_mmap(requested_addr, bytes, false);
3654 if (addr == requested_addr) {
3655 return requested_addr;
3656 }
3657
3658 if (addr != NULL) {
3659 // mmap() is successful but it fails to reserve at the requested address
3660 anon_munmap(addr, bytes);
3661 }
3662
3663 int i;
3664 for (i = 0; i < max_tries; ++i) {
3665 base[i] = reserve_memory(bytes);
3666
3667 if (base[i] != NULL) {
3668 // Is this the block we wanted?
3669 if (base[i] == requested_addr) {
3670 size[i] = bytes;
3671 break;
3672 }
3673
3674 // Does this overlap the block we wanted? Give back the overlapped
3675 // parts and try again.
3676
3677 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3678 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3679 unmap_memory(base[i], top_overlap);
3680 base[i] += top_overlap;
3681 size[i] = bytes - top_overlap;
3682 } else {
3683 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3684 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3685 unmap_memory(requested_addr, bottom_overlap);
3686 size[i] = bytes - bottom_overlap;
3687 } else {
3688 size[i] = bytes;
3689 }
3690 }
3691 }
3692 }
3693
3694 // Give back the unused reserved pieces.
3695
3696 for (int j = 0; j < i; ++j) {
3697 if (base[j] != NULL) {
3698 unmap_memory(base[j], size[j]);
3699 }
3700 }
3701
3702 if (i < max_tries) {
3703 return requested_addr;
3704 } else {
3705 return NULL;
3706 }
3707 }
3708
3709 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3710 return ::read(fd, buf, nBytes);
3711 }
3712
3713 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3714 return ::pread(fd, buf, nBytes, offset);
3715 }
3716
3717 // Short sleep, direct OS call.
3718 //
3719 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3720 // sched_yield(2) will actually give up the CPU:
3721 //
3722 // * Alone on this pariticular CPU, keeps running.
3723 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3724 // (pre 2.6.39).
3725 //
3726 // So calling this with 0 is an alternative.
3727 //
3728 void os::naked_short_sleep(jlong ms) {
3729 struct timespec req;
3730
3731 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3732 req.tv_sec = 0;
3733 if (ms > 0) {
3734 req.tv_nsec = (ms % 1000) * 1000000;
3735 } else {
3736 req.tv_nsec = 1;
3737 }
3738
3739 nanosleep(&req, NULL);
3740
3741 return;
3742 }
3743
3744 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3745 void os::infinite_sleep() {
3746 while (true) { // sleep forever ...
3747 ::sleep(100); // ... 100 seconds at a time
3748 }
3749 }
3750
3751 // Used to convert frequent JVM_Yield() to nops
3752 bool os::dont_yield() {
3753 return DontYieldALot;
3754 }
3755
3756 void os::naked_yield() {
3757 sched_yield();
3758 }
3759
3760 ////////////////////////////////////////////////////////////////////////////////
3761 // thread priority support
3762
3763 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3764 // only supports dynamic priority, static priority must be zero. For real-time
3765 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3766 // However, for large multi-threaded applications, SCHED_RR is not only slower
3767 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3768 // of 5 runs - Sep 2005).
3769 //
3770 // The following code actually changes the niceness of kernel-thread/LWP. It
3771 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3772 // not the entire user process, and user level threads are 1:1 mapped to kernel
3773 // threads. It has always been the case, but could change in the future. For
3774 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3775 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3776
3777 int os::java_to_os_priority[CriticalPriority + 1] = {
3778 19, // 0 Entry should never be used
3779
3780 4, // 1 MinPriority
3781 3, // 2
3782 2, // 3
3783
3784 1, // 4
3785 0, // 5 NormPriority
3786 -1, // 6
3787
3788 -2, // 7
3789 -3, // 8
3790 -4, // 9 NearMaxPriority
3791
3792 -5, // 10 MaxPriority
3793
3794 -5 // 11 CriticalPriority
3795 };
3796
3797 static int prio_init() {
3798 if (ThreadPriorityPolicy == 1) {
3799 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3800 // if effective uid is not root. Perhaps, a more elegant way of doing
3801 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3802 if (geteuid() != 0) {
3803 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3804 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3805 }
3806 ThreadPriorityPolicy = 0;
3807 }
3808 }
3809 if (UseCriticalJavaThreadPriority) {
3810 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3811 }
3812 return 0;
3813 }
3814
3815 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3816 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3817
3818 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3819 return (ret == 0) ? OS_OK : OS_ERR;
3820 }
3821
3822 OSReturn os::get_native_priority(const Thread* const thread,
3823 int *priority_ptr) {
3824 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3825 *priority_ptr = java_to_os_priority[NormPriority];
3826 return OS_OK;
3827 }
3828
3829 errno = 0;
3830 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3831 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3832 }
3833
3834 // Hint to the underlying OS that a task switch would not be good.
3835 // Void return because it's a hint and can fail.
3836 void os::hint_no_preempt() {}
3837
3838 ////////////////////////////////////////////////////////////////////////////////
3839 // suspend/resume support
3840
3841 // the low-level signal-based suspend/resume support is a remnant from the
3842 // old VM-suspension that used to be for java-suspension, safepoints etc,
3843 // within hotspot. Now there is a single use-case for this:
3844 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3845 // that runs in the watcher thread.
3846 // The remaining code is greatly simplified from the more general suspension
3847 // code that used to be used.
3848 //
3849 // The protocol is quite simple:
3850 // - suspend:
3851 // - sends a signal to the target thread
3852 // - polls the suspend state of the osthread using a yield loop
3853 // - target thread signal handler (SR_handler) sets suspend state
3854 // and blocks in sigsuspend until continued
3855 // - resume:
3856 // - sets target osthread state to continue
3857 // - sends signal to end the sigsuspend loop in the SR_handler
3858 //
3859 // Note that the SR_lock plays no role in this suspend/resume protocol.
3860
3861 static void resume_clear_context(OSThread *osthread) {
3862 osthread->set_ucontext(NULL);
3863 osthread->set_siginfo(NULL);
3864 }
3865
3866 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3867 ucontext_t* context) {
3868 osthread->set_ucontext(context);
3869 osthread->set_siginfo(siginfo);
3870 }
3871
3872 // Handler function invoked when a thread's execution is suspended or
3873 // resumed. We have to be careful that only async-safe functions are
3874 // called here (Note: most pthread functions are not async safe and
3875 // should be avoided.)
3876 //
3877 // Note: sigwait() is a more natural fit than sigsuspend() from an
3878 // interface point of view, but sigwait() prevents the signal hander
3879 // from being run. libpthread would get very confused by not having
3880 // its signal handlers run and prevents sigwait()'s use with the
3881 // mutex granting granting signal.
3882 //
3883 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3884 //
3885 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3886 // Save and restore errno to avoid confusing native code with EINTR
3887 // after sigsuspend.
3888 int old_errno = errno;
3889
3890 Thread* thread = Thread::current();
3891 OSThread* osthread = thread->osthread();
3892 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3893
3894 os::SuspendResume::State current = osthread->sr.state();
3895 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3896 suspend_save_context(osthread, siginfo, context);
3897
3898 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3899 os::SuspendResume::State state = osthread->sr.suspended();
3900 if (state == os::SuspendResume::SR_SUSPENDED) {
3901 sigset_t suspend_set; // signals for sigsuspend()
3902
3903 // get current set of blocked signals and unblock resume signal
3904 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3905 sigdelset(&suspend_set, SR_signum);
3906
3907 sr_semaphore.signal();
3908 // wait here until we are resumed
3909 while (1) {
3910 sigsuspend(&suspend_set);
3911
3912 os::SuspendResume::State result = osthread->sr.running();
3913 if (result == os::SuspendResume::SR_RUNNING) {
3914 sr_semaphore.signal();
3915 break;
3916 }
3917 }
3918
3919 } else if (state == os::SuspendResume::SR_RUNNING) {
3920 // request was cancelled, continue
3921 } else {
3922 ShouldNotReachHere();
3923 }
3924
3925 resume_clear_context(osthread);
3926 } else if (current == os::SuspendResume::SR_RUNNING) {
3927 // request was cancelled, continue
3928 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3929 // ignore
3930 } else {
3931 // ignore
3932 }
3933
3934 errno = old_errno;
3935 }
3936
3937 static int SR_initialize() {
3938 struct sigaction act;
3939 char *s;
3940
3941 // Get signal number to use for suspend/resume
3942 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3943 int sig = ::strtol(s, 0, 10);
3944 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
3945 sig < NSIG) { // Must be legal signal and fit into sigflags[].
3946 SR_signum = sig;
3947 } else {
3948 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3949 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3950 }
3951 }
3952
3953 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3954 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3955
3956 sigemptyset(&SR_sigset);
3957 sigaddset(&SR_sigset, SR_signum);
3958
3959 // Set up signal handler for suspend/resume
3960 act.sa_flags = SA_RESTART|SA_SIGINFO;
3961 act.sa_handler = (void (*)(int)) SR_handler;
3962
3963 // SR_signum is blocked by default.
3964 // 4528190 - We also need to block pthread restart signal (32 on all
3965 // supported Linux platforms). Note that LinuxThreads need to block
3966 // this signal for all threads to work properly. So we don't have
3967 // to use hard-coded signal number when setting up the mask.
3968 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3969
3970 if (sigaction(SR_signum, &act, 0) == -1) {
3971 return -1;
3972 }
3973
3974 // Save signal flag
3975 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3976 return 0;
3977 }
3978
3979 static int sr_notify(OSThread* osthread) {
3980 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3981 assert_status(status == 0, status, "pthread_kill");
3982 return status;
3983 }
3984
3985 // "Randomly" selected value for how long we want to spin
3986 // before bailing out on suspending a thread, also how often
3987 // we send a signal to a thread we want to resume
3988 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3989 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3990
3991 // returns true on success and false on error - really an error is fatal
3992 // but this seems the normal response to library errors
3993 static bool do_suspend(OSThread* osthread) {
3994 assert(osthread->sr.is_running(), "thread should be running");
3995 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3996
3997 // mark as suspended and send signal
3998 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3999 // failed to switch, state wasn't running?
4000 ShouldNotReachHere();
4001 return false;
4002 }
4003
4004 if (sr_notify(osthread) != 0) {
4005 ShouldNotReachHere();
4006 }
4007
4008 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4009 while (true) {
4010 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4011 break;
4012 } else {
4013 // timeout
4014 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4015 if (cancelled == os::SuspendResume::SR_RUNNING) {
4016 return false;
4017 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4018 // make sure that we consume the signal on the semaphore as well
4019 sr_semaphore.wait();
4020 break;
4021 } else {
4022 ShouldNotReachHere();
4023 return false;
4024 }
4025 }
4026 }
4027
4028 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4029 return true;
4030 }
4031
4032 static void do_resume(OSThread* osthread) {
4033 assert(osthread->sr.is_suspended(), "thread should be suspended");
4034 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4035
4036 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4037 // failed to switch to WAKEUP_REQUEST
4038 ShouldNotReachHere();
4039 return;
4040 }
4041
4042 while (true) {
4043 if (sr_notify(osthread) == 0) {
4044 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4045 if (osthread->sr.is_running()) {
4046 return;
4047 }
4048 }
4049 } else {
4050 ShouldNotReachHere();
4051 }
4052 }
4053
4054 guarantee(osthread->sr.is_running(), "Must be running!");
4055 }
4056
4057 ///////////////////////////////////////////////////////////////////////////////////
4058 // signal handling (except suspend/resume)
4059
4060 // This routine may be used by user applications as a "hook" to catch signals.
4061 // The user-defined signal handler must pass unrecognized signals to this
4062 // routine, and if it returns true (non-zero), then the signal handler must
4063 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4064 // routine will never retun false (zero), but instead will execute a VM panic
4065 // routine kill the process.
4066 //
4067 // If this routine returns false, it is OK to call it again. This allows
4068 // the user-defined signal handler to perform checks either before or after
4069 // the VM performs its own checks. Naturally, the user code would be making
4070 // a serious error if it tried to handle an exception (such as a null check
4071 // or breakpoint) that the VM was generating for its own correct operation.
4072 //
4073 // This routine may recognize any of the following kinds of signals:
4074 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4075 // It should be consulted by handlers for any of those signals.
4076 //
4077 // The caller of this routine must pass in the three arguments supplied
4078 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4079 // field of the structure passed to sigaction(). This routine assumes that
4080 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4081 //
4082 // Note that the VM will print warnings if it detects conflicting signal
4083 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4084 //
4085 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4086 siginfo_t* siginfo,
4087 void* ucontext,
4088 int abort_if_unrecognized);
4089
4090 void signalHandler(int sig, siginfo_t* info, void* uc) {
4091 assert(info != NULL && uc != NULL, "it must be old kernel");
4092 int orig_errno = errno; // Preserve errno value over signal handler.
4093 JVM_handle_linux_signal(sig, info, uc, true);
4094 errno = orig_errno;
4095 }
4096
4097
4098 // This boolean allows users to forward their own non-matching signals
4099 // to JVM_handle_linux_signal, harmlessly.
4100 bool os::Linux::signal_handlers_are_installed = false;
4101
4102 // For signal-chaining
4103 struct sigaction sigact[NSIG];
4104 uint64_t sigs = 0;
4105 #if (64 < NSIG-1)
4106 #error "Not all signals can be encoded in sigs. Adapt its type!"
4107 #endif
4108 bool os::Linux::libjsig_is_loaded = false;
4109 typedef struct sigaction *(*get_signal_t)(int);
4110 get_signal_t os::Linux::get_signal_action = NULL;
4111
4112 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4113 struct sigaction *actp = NULL;
4114
4115 if (libjsig_is_loaded) {
4116 // Retrieve the old signal handler from libjsig
4117 actp = (*get_signal_action)(sig);
4118 }
4119 if (actp == NULL) {
4120 // Retrieve the preinstalled signal handler from jvm
4121 actp = get_preinstalled_handler(sig);
4122 }
4123
4124 return actp;
4125 }
4126
4127 static bool call_chained_handler(struct sigaction *actp, int sig,
4128 siginfo_t *siginfo, void *context) {
4129 // Call the old signal handler
4130 if (actp->sa_handler == SIG_DFL) {
4131 // It's more reasonable to let jvm treat it as an unexpected exception
4132 // instead of taking the default action.
4133 return false;
4134 } else if (actp->sa_handler != SIG_IGN) {
4135 if ((actp->sa_flags & SA_NODEFER) == 0) {
4136 // automaticlly block the signal
4137 sigaddset(&(actp->sa_mask), sig);
4138 }
4139
4140 sa_handler_t hand = NULL;
4141 sa_sigaction_t sa = NULL;
4142 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4143 // retrieve the chained handler
4144 if (siginfo_flag_set) {
4145 sa = actp->sa_sigaction;
4146 } else {
4147 hand = actp->sa_handler;
4148 }
4149
4150 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4151 actp->sa_handler = SIG_DFL;
4152 }
4153
4154 // try to honor the signal mask
4155 sigset_t oset;
4156 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4157
4158 // call into the chained handler
4159 if (siginfo_flag_set) {
4160 (*sa)(sig, siginfo, context);
4161 } else {
4162 (*hand)(sig);
4163 }
4164
4165 // restore the signal mask
4166 pthread_sigmask(SIG_SETMASK, &oset, 0);
4167 }
4168 // Tell jvm's signal handler the signal is taken care of.
4169 return true;
4170 }
4171
4172 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4173 bool chained = false;
4174 // signal-chaining
4175 if (UseSignalChaining) {
4176 struct sigaction *actp = get_chained_signal_action(sig);
4177 if (actp != NULL) {
4178 chained = call_chained_handler(actp, sig, siginfo, context);
4179 }
4180 }
4181 return chained;
4182 }
4183
4184 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4185 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4186 return &sigact[sig];
4187 }
4188 return NULL;
4189 }
4190
4191 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4192 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4193 sigact[sig] = oldAct;
4194 sigs |= (uint64_t)1 << (sig-1);
4195 }
4196
4197 // for diagnostic
4198 int sigflags[NSIG];
4199
4200 int os::Linux::get_our_sigflags(int sig) {
4201 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4202 return sigflags[sig];
4203 }
4204
4205 void os::Linux::set_our_sigflags(int sig, int flags) {
4206 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4207 if (sig > 0 && sig < NSIG) {
4208 sigflags[sig] = flags;
4209 }
4210 }
4211
4212 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4213 // Check for overwrite.
4214 struct sigaction oldAct;
4215 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4216
4217 void* oldhand = oldAct.sa_sigaction
4218 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4219 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4220 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4221 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4222 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4223 if (AllowUserSignalHandlers || !set_installed) {
4224 // Do not overwrite; user takes responsibility to forward to us.
4225 return;
4226 } else if (UseSignalChaining) {
4227 // save the old handler in jvm
4228 save_preinstalled_handler(sig, oldAct);
4229 // libjsig also interposes the sigaction() call below and saves the
4230 // old sigaction on it own.
4231 } else {
4232 fatal("Encountered unexpected pre-existing sigaction handler "
4233 "%#lx for signal %d.", (long)oldhand, sig);
4234 }
4235 }
4236
4237 struct sigaction sigAct;
4238 sigfillset(&(sigAct.sa_mask));
4239 sigAct.sa_handler = SIG_DFL;
4240 if (!set_installed) {
4241 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4242 } else {
4243 sigAct.sa_sigaction = signalHandler;
4244 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4245 }
4246 // Save flags, which are set by ours
4247 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4248 sigflags[sig] = sigAct.sa_flags;
4249
4250 int ret = sigaction(sig, &sigAct, &oldAct);
4251 assert(ret == 0, "check");
4252
4253 void* oldhand2 = oldAct.sa_sigaction
4254 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4255 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4256 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4257 }
4258
4259 // install signal handlers for signals that HotSpot needs to
4260 // handle in order to support Java-level exception handling.
4261
4262 void os::Linux::install_signal_handlers() {
4263 if (!signal_handlers_are_installed) {
4264 signal_handlers_are_installed = true;
4265
4266 // signal-chaining
4267 typedef void (*signal_setting_t)();
4268 signal_setting_t begin_signal_setting = NULL;
4269 signal_setting_t end_signal_setting = NULL;
4270 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4271 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4272 if (begin_signal_setting != NULL) {
4273 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4274 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4275 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4276 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4277 libjsig_is_loaded = true;
4278 assert(UseSignalChaining, "should enable signal-chaining");
4279 }
4280 if (libjsig_is_loaded) {
4281 // Tell libjsig jvm is setting signal handlers
4282 (*begin_signal_setting)();
4283 }
4284
4285 set_signal_handler(SIGSEGV, true);
4286 set_signal_handler(SIGPIPE, true);
4287 set_signal_handler(SIGBUS, true);
4288 set_signal_handler(SIGILL, true);
4289 set_signal_handler(SIGFPE, true);
4290 #if defined(PPC64)
4291 set_signal_handler(SIGTRAP, true);
4292 #endif
4293 set_signal_handler(SIGXFSZ, true);
4294
4295 if (libjsig_is_loaded) {
4296 // Tell libjsig jvm finishes setting signal handlers
4297 (*end_signal_setting)();
4298 }
4299
4300 // We don't activate signal checker if libjsig is in place, we trust ourselves
4301 // and if UserSignalHandler is installed all bets are off.
4302 // Log that signal checking is off only if -verbose:jni is specified.
4303 if (CheckJNICalls) {
4304 if (libjsig_is_loaded) {
4305 if (PrintJNIResolving) {
4306 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4307 }
4308 check_signals = false;
4309 }
4310 if (AllowUserSignalHandlers) {
4311 if (PrintJNIResolving) {
4312 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4313 }
4314 check_signals = false;
4315 }
4316 }
4317 }
4318 }
4319
4320 // This is the fastest way to get thread cpu time on Linux.
4321 // Returns cpu time (user+sys) for any thread, not only for current.
4322 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4323 // It might work on 2.6.10+ with a special kernel/glibc patch.
4324 // For reference, please, see IEEE Std 1003.1-2004:
4325 // http://www.unix.org/single_unix_specification
4326
4327 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4328 struct timespec tp;
4329 int rc = os::Linux::clock_gettime(clockid, &tp);
4330 assert(rc == 0, "clock_gettime is expected to return 0 code");
4331
4332 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4333 }
4334
4335 /////
4336 // glibc on Linux platform uses non-documented flag
4337 // to indicate, that some special sort of signal
4338 // trampoline is used.
4339 // We will never set this flag, and we should
4340 // ignore this flag in our diagnostic
4341 #ifdef SIGNIFICANT_SIGNAL_MASK
4342 #undef SIGNIFICANT_SIGNAL_MASK
4343 #endif
4344 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4345
4346 static const char* get_signal_handler_name(address handler,
4347 char* buf, int buflen) {
4348 int offset = 0;
4349 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4350 if (found) {
4351 // skip directory names
4352 const char *p1, *p2;
4353 p1 = buf;
4354 size_t len = strlen(os::file_separator());
4355 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4356 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4357 } else {
4358 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4359 }
4360 return buf;
4361 }
4362
4363 static void print_signal_handler(outputStream* st, int sig,
4364 char* buf, size_t buflen) {
4365 struct sigaction sa;
4366
4367 sigaction(sig, NULL, &sa);
4368
4369 // See comment for SIGNIFICANT_SIGNAL_MASK define
4370 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4371
4372 st->print("%s: ", os::exception_name(sig, buf, buflen));
4373
4374 address handler = (sa.sa_flags & SA_SIGINFO)
4375 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4376 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4377
4378 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4379 st->print("SIG_DFL");
4380 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4381 st->print("SIG_IGN");
4382 } else {
4383 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4384 }
4385
4386 st->print(", sa_mask[0]=");
4387 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4388
4389 address rh = VMError::get_resetted_sighandler(sig);
4390 // May be, handler was resetted by VMError?
4391 if (rh != NULL) {
4392 handler = rh;
4393 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4394 }
4395
4396 st->print(", sa_flags=");
4397 os::Posix::print_sa_flags(st, sa.sa_flags);
4398
4399 // Check: is it our handler?
4400 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4401 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4402 // It is our signal handler
4403 // check for flags, reset system-used one!
4404 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4405 st->print(
4406 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4407 os::Linux::get_our_sigflags(sig));
4408 }
4409 }
4410 st->cr();
4411 }
4412
4413
4414 #define DO_SIGNAL_CHECK(sig) \
4415 do { \
4416 if (!sigismember(&check_signal_done, sig)) { \
4417 os::Linux::check_signal_handler(sig); \
4418 } \
4419 } while (0)
4420
4421 // This method is a periodic task to check for misbehaving JNI applications
4422 // under CheckJNI, we can add any periodic checks here
4423
4424 void os::run_periodic_checks() {
4425 if (check_signals == false) return;
4426
4427 // SEGV and BUS if overridden could potentially prevent
4428 // generation of hs*.log in the event of a crash, debugging
4429 // such a case can be very challenging, so we absolutely
4430 // check the following for a good measure:
4431 DO_SIGNAL_CHECK(SIGSEGV);
4432 DO_SIGNAL_CHECK(SIGILL);
4433 DO_SIGNAL_CHECK(SIGFPE);
4434 DO_SIGNAL_CHECK(SIGBUS);
4435 DO_SIGNAL_CHECK(SIGPIPE);
4436 DO_SIGNAL_CHECK(SIGXFSZ);
4437 #if defined(PPC64)
4438 DO_SIGNAL_CHECK(SIGTRAP);
4439 #endif
4440
4441 // ReduceSignalUsage allows the user to override these handlers
4442 // see comments at the very top and jvm_solaris.h
4443 if (!ReduceSignalUsage) {
4444 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4445 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4446 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4447 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4448 }
4449
4450 DO_SIGNAL_CHECK(SR_signum);
4451 }
4452
4453 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4454
4455 static os_sigaction_t os_sigaction = NULL;
4456
4457 void os::Linux::check_signal_handler(int sig) {
4458 char buf[O_BUFLEN];
4459 address jvmHandler = NULL;
4460
4461
4462 struct sigaction act;
4463 if (os_sigaction == NULL) {
4464 // only trust the default sigaction, in case it has been interposed
4465 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4466 if (os_sigaction == NULL) return;
4467 }
4468
4469 os_sigaction(sig, (struct sigaction*)NULL, &act);
4470
4471
4472 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4473
4474 address thisHandler = (act.sa_flags & SA_SIGINFO)
4475 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4476 : CAST_FROM_FN_PTR(address, act.sa_handler);
4477
4478
4479 switch (sig) {
4480 case SIGSEGV:
4481 case SIGBUS:
4482 case SIGFPE:
4483 case SIGPIPE:
4484 case SIGILL:
4485 case SIGXFSZ:
4486 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4487 break;
4488
4489 case SHUTDOWN1_SIGNAL:
4490 case SHUTDOWN2_SIGNAL:
4491 case SHUTDOWN3_SIGNAL:
4492 case BREAK_SIGNAL:
4493 jvmHandler = (address)user_handler();
4494 break;
4495
4496 default:
4497 if (sig == SR_signum) {
4498 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4499 } else {
4500 return;
4501 }
4502 break;
4503 }
4504
4505 if (thisHandler != jvmHandler) {
4506 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4507 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4508 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4509 // No need to check this sig any longer
4510 sigaddset(&check_signal_done, sig);
4511 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4512 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4513 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4514 exception_name(sig, buf, O_BUFLEN));
4515 }
4516 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4517 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4518 tty->print("expected:");
4519 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4520 tty->cr();
4521 tty->print(" found:");
4522 os::Posix::print_sa_flags(tty, act.sa_flags);
4523 tty->cr();
4524 // No need to check this sig any longer
4525 sigaddset(&check_signal_done, sig);
4526 }
4527
4528 // Dump all the signal
4529 if (sigismember(&check_signal_done, sig)) {
4530 print_signal_handlers(tty, buf, O_BUFLEN);
4531 }
4532 }
4533
4534 extern void report_error(char* file_name, int line_no, char* title,
4535 char* format, ...);
4536
4537 // this is called _before_ the most of global arguments have been parsed
4538 void os::init(void) {
4539 char dummy; // used to get a guess on initial stack address
4540 // first_hrtime = gethrtime();
4541
4542 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4543
4544 init_random(1234567);
4545
4546 ThreadCritical::initialize();
4547
4548 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4549 if (Linux::page_size() == -1) {
4550 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4551 strerror(errno));
4552 }
4553 init_page_sizes((size_t) Linux::page_size());
4554
4555 Linux::initialize_system_info();
4556
4557 // main_thread points to the aboriginal thread
4558 Linux::_main_thread = pthread_self();
4559
4560 Linux::clock_init();
4561 initial_time_count = javaTimeNanos();
4562
4563 // pthread_condattr initialization for monotonic clock
4564 int status;
4565 pthread_condattr_t* _condattr = os::Linux::condAttr();
4566 if ((status = pthread_condattr_init(_condattr)) != 0) {
4567 fatal("pthread_condattr_init: %s", strerror(status));
4568 }
4569 // Only set the clock if CLOCK_MONOTONIC is available
4570 if (os::supports_monotonic_clock()) {
4571 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4572 if (status == EINVAL) {
4573 warning("Unable to use monotonic clock with relative timed-waits" \
4574 " - changes to the time-of-day clock may have adverse affects");
4575 } else {
4576 fatal("pthread_condattr_setclock: %s", strerror(status));
4577 }
4578 }
4579 }
4580 // else it defaults to CLOCK_REALTIME
4581
4582 // retrieve entry point for pthread_setname_np
4583 Linux::_pthread_setname_np =
4584 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4585
4586 }
4587
4588 // To install functions for atexit system call
4589 extern "C" {
4590 static void perfMemory_exit_helper() {
4591 perfMemory_exit();
4592 }
4593 }
4594
4595 // this is called _after_ the global arguments have been parsed
4596 jint os::init_2(void) {
4597 Linux::fast_thread_clock_init();
4598
4599 // Allocate a single page and mark it as readable for safepoint polling
4600 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4601 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4602
4603 os::set_polling_page(polling_page);
4604
4605 #ifndef PRODUCT
4606 if (Verbose && PrintMiscellaneous) {
4607 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4608 (intptr_t)polling_page);
4609 }
4610 #endif
4611
4612 if (!UseMembar) {
4613 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4614 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4615 os::set_memory_serialize_page(mem_serialize_page);
4616
4617 #ifndef PRODUCT
4618 if (Verbose && PrintMiscellaneous) {
4619 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4620 (intptr_t)mem_serialize_page);
4621 }
4622 #endif
4623 }
4624
4625 // initialize suspend/resume support - must do this before signal_sets_init()
4626 if (SR_initialize() != 0) {
4627 perror("SR_initialize failed");
4628 return JNI_ERR;
4629 }
4630
4631 Linux::signal_sets_init();
4632 Linux::install_signal_handlers();
4633
4634 // Check minimum allowable stack size for thread creation and to initialize
4635 // the java system classes, including StackOverflowError - depends on page
4636 // size. Add a page for compiler2 recursion in main thread.
4637 // Add in 2*BytesPerWord times page size to account for VM stack during
4638 // class initialization depending on 32 or 64 bit VM.
4639 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4640 JavaThread::stack_guard_zone_size() +
4641 JavaThread::stack_shadow_zone_size() +
4642 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4643
4644 size_t threadStackSizeInBytes = ThreadStackSize * K;
4645 if (threadStackSizeInBytes != 0 &&
4646 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4647 tty->print_cr("\nThe stack size specified is too small, "
4648 "Specify at least " SIZE_FORMAT "k",
4649 os::Linux::min_stack_allowed/ K);
4650 return JNI_ERR;
4651 }
4652
4653 // Make the stack size a multiple of the page size so that
4654 // the yellow/red zones can be guarded.
4655 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4656 vm_page_size()));
4657
4658 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4659
4660 #if defined(IA32)
4661 workaround_expand_exec_shield_cs_limit();
4662 #endif
4663
4664 Linux::libpthread_init();
4665 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4666 tty->print_cr("[HotSpot is running with %s, %s]\n",
4667 Linux::glibc_version(), Linux::libpthread_version());
4668 }
4669
4670 if (UseNUMA) {
4671 if (!Linux::libnuma_init()) {
4672 UseNUMA = false;
4673 } else {
4674 if ((Linux::numa_max_node() < 1)) {
4675 // There's only one node(they start from 0), disable NUMA.
4676 UseNUMA = false;
4677 }
4678 }
4679 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4680 // we can make the adaptive lgrp chunk resizing work. If the user specified
4681 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4682 // disable adaptive resizing.
4683 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4684 if (FLAG_IS_DEFAULT(UseNUMA)) {
4685 UseNUMA = false;
4686 } else {
4687 if (FLAG_IS_DEFAULT(UseLargePages) &&
4688 FLAG_IS_DEFAULT(UseSHM) &&
4689 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4690 UseLargePages = false;
4691 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4692 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4693 UseAdaptiveSizePolicy = false;
4694 UseAdaptiveNUMAChunkSizing = false;
4695 }
4696 }
4697 }
4698 if (!UseNUMA && ForceNUMA) {
4699 UseNUMA = true;
4700 }
4701 }
4702
4703 if (MaxFDLimit) {
4704 // set the number of file descriptors to max. print out error
4705 // if getrlimit/setrlimit fails but continue regardless.
4706 struct rlimit nbr_files;
4707 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4708 if (status != 0) {
4709 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4710 perror("os::init_2 getrlimit failed");
4711 }
4712 } else {
4713 nbr_files.rlim_cur = nbr_files.rlim_max;
4714 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4715 if (status != 0) {
4716 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4717 perror("os::init_2 setrlimit failed");
4718 }
4719 }
4720 }
4721 }
4722
4723 // Initialize lock used to serialize thread creation (see os::create_thread)
4724 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4725
4726 // at-exit methods are called in the reverse order of their registration.
4727 // atexit functions are called on return from main or as a result of a
4728 // call to exit(3C). There can be only 32 of these functions registered
4729 // and atexit() does not set errno.
4730
4731 if (PerfAllowAtExitRegistration) {
4732 // only register atexit functions if PerfAllowAtExitRegistration is set.
4733 // atexit functions can be delayed until process exit time, which
4734 // can be problematic for embedded VM situations. Embedded VMs should
4735 // call DestroyJavaVM() to assure that VM resources are released.
4736
4737 // note: perfMemory_exit_helper atexit function may be removed in
4738 // the future if the appropriate cleanup code can be added to the
4739 // VM_Exit VMOperation's doit method.
4740 if (atexit(perfMemory_exit_helper) != 0) {
4741 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4742 }
4743 }
4744
4745 // initialize thread priority policy
4746 prio_init();
4747
4748 return JNI_OK;
4749 }
4750
4751 // Mark the polling page as unreadable
4752 void os::make_polling_page_unreadable(void) {
4753 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4754 fatal("Could not disable polling page");
4755 }
4756 }
4757
4758 // Mark the polling page as readable
4759 void os::make_polling_page_readable(void) {
4760 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4761 fatal("Could not enable polling page");
4762 }
4763 }
4764
4765 int os::active_processor_count() {
4766 // Linux doesn't yet have a (official) notion of processor sets,
4767 // so just return the number of online processors.
4768 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4769 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4770 return online_cpus;
4771 }
4772
4773 void os::set_native_thread_name(const char *name) {
4774 if (Linux::_pthread_setname_np) {
4775 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4776 snprintf(buf, sizeof(buf), "%s", name);
4777 buf[sizeof(buf) - 1] = '\0';
4778 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4779 // ERANGE should not happen; all other errors should just be ignored.
4780 assert(rc != ERANGE, "pthread_setname_np failed");
4781 }
4782 }
4783
4784 bool os::distribute_processes(uint length, uint* distribution) {
4785 // Not yet implemented.
4786 return false;
4787 }
4788
4789 bool os::bind_to_processor(uint processor_id) {
4790 // Not yet implemented.
4791 return false;
4792 }
4793
4794 ///
4795
4796 void os::SuspendedThreadTask::internal_do_task() {
4797 if (do_suspend(_thread->osthread())) {
4798 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4799 do_task(context);
4800 do_resume(_thread->osthread());
4801 }
4802 }
4803
4804 class PcFetcher : public os::SuspendedThreadTask {
4805 public:
4806 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4807 ExtendedPC result();
4808 protected:
4809 void do_task(const os::SuspendedThreadTaskContext& context);
4810 private:
4811 ExtendedPC _epc;
4812 };
4813
4814 ExtendedPC PcFetcher::result() {
4815 guarantee(is_done(), "task is not done yet.");
4816 return _epc;
4817 }
4818
4819 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4820 Thread* thread = context.thread();
4821 OSThread* osthread = thread->osthread();
4822 if (osthread->ucontext() != NULL) {
4823 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4824 } else {
4825 // NULL context is unexpected, double-check this is the VMThread
4826 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4827 }
4828 }
4829
4830 // Suspends the target using the signal mechanism and then grabs the PC before
4831 // resuming the target. Used by the flat-profiler only
4832 ExtendedPC os::get_thread_pc(Thread* thread) {
4833 // Make sure that it is called by the watcher for the VMThread
4834 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4835 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4836
4837 PcFetcher fetcher(thread);
4838 fetcher.run();
4839 return fetcher.result();
4840 }
4841
4842 ////////////////////////////////////////////////////////////////////////////////
4843 // debug support
4844
4845 bool os::find(address addr, outputStream* st) {
4846 Dl_info dlinfo;
4847 memset(&dlinfo, 0, sizeof(dlinfo));
4848 if (dladdr(addr, &dlinfo) != 0) {
4849 st->print(PTR_FORMAT ": ", p2i(addr));
4850 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4851 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4852 p2i(addr) - p2i(dlinfo.dli_saddr));
4853 } else if (dlinfo.dli_fbase != NULL) {
4854 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4855 } else {
4856 st->print("<absolute address>");
4857 }
4858 if (dlinfo.dli_fname != NULL) {
4859 st->print(" in %s", dlinfo.dli_fname);
4860 }
4861 if (dlinfo.dli_fbase != NULL) {
4862 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4863 }
4864 st->cr();
4865
4866 if (Verbose) {
4867 // decode some bytes around the PC
4868 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4869 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4870 address lowest = (address) dlinfo.dli_sname;
4871 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4872 if (begin < lowest) begin = lowest;
4873 Dl_info dlinfo2;
4874 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4875 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4876 end = (address) dlinfo2.dli_saddr;
4877 }
4878 Disassembler::decode(begin, end, st);
4879 }
4880 return true;
4881 }
4882 return false;
4883 }
4884
4885 ////////////////////////////////////////////////////////////////////////////////
4886 // misc
4887
4888 // This does not do anything on Linux. This is basically a hook for being
4889 // able to use structured exception handling (thread-local exception filters)
4890 // on, e.g., Win32.
4891 void
4892 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
4893 JavaCallArguments* args, Thread* thread) {
4894 f(value, method, args, thread);
4895 }
4896
4897 void os::print_statistics() {
4898 }
4899
4900 bool os::message_box(const char* title, const char* message) {
4901 int i;
4902 fdStream err(defaultStream::error_fd());
4903 for (i = 0; i < 78; i++) err.print_raw("=");
4904 err.cr();
4905 err.print_raw_cr(title);
4906 for (i = 0; i < 78; i++) err.print_raw("-");
4907 err.cr();
4908 err.print_raw_cr(message);
4909 for (i = 0; i < 78; i++) err.print_raw("=");
4910 err.cr();
4911
4912 char buf[16];
4913 // Prevent process from exiting upon "read error" without consuming all CPU
4914 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4915
4916 return buf[0] == 'y' || buf[0] == 'Y';
4917 }
4918
4919 int os::stat(const char *path, struct stat *sbuf) {
4920 char pathbuf[MAX_PATH];
4921 if (strlen(path) > MAX_PATH - 1) {
4922 errno = ENAMETOOLONG;
4923 return -1;
4924 }
4925 os::native_path(strcpy(pathbuf, path));
4926 return ::stat(pathbuf, sbuf);
4927 }
4928
4929 bool os::check_heap(bool force) {
4930 return true;
4931 }
4932
4933 // Is a (classpath) directory empty?
4934 bool os::dir_is_empty(const char* path) {
4935 DIR *dir = NULL;
4936 struct dirent *ptr;
4937
4938 dir = opendir(path);
4939 if (dir == NULL) return true;
4940
4941 // Scan the directory
4942 bool result = true;
4943 char buf[sizeof(struct dirent) + MAX_PATH];
4944 while (result && (ptr = ::readdir(dir)) != NULL) {
4945 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4946 result = false;
4947 }
4948 }
4949 closedir(dir);
4950 return result;
4951 }
4952
4953 // This code originates from JDK's sysOpen and open64_w
4954 // from src/solaris/hpi/src/system_md.c
4955
4956 int os::open(const char *path, int oflag, int mode) {
4957 if (strlen(path) > MAX_PATH - 1) {
4958 errno = ENAMETOOLONG;
4959 return -1;
4960 }
4961
4962 // All file descriptors that are opened in the Java process and not
4963 // specifically destined for a subprocess should have the close-on-exec
4964 // flag set. If we don't set it, then careless 3rd party native code
4965 // might fork and exec without closing all appropriate file descriptors
4966 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
4967 // turn might:
4968 //
4969 // - cause end-of-file to fail to be detected on some file
4970 // descriptors, resulting in mysterious hangs, or
4971 //
4972 // - might cause an fopen in the subprocess to fail on a system
4973 // suffering from bug 1085341.
4974 //
4975 // (Yes, the default setting of the close-on-exec flag is a Unix
4976 // design flaw)
4977 //
4978 // See:
4979 // 1085341: 32-bit stdio routines should support file descriptors >255
4980 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4981 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4982 //
4983 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
4984 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
4985 // because it saves a system call and removes a small window where the flag
4986 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
4987 // and we fall back to using FD_CLOEXEC (see below).
4988 #ifdef O_CLOEXEC
4989 oflag |= O_CLOEXEC;
4990 #endif
4991
4992 int fd = ::open64(path, oflag, mode);
4993 if (fd == -1) return -1;
4994
4995 //If the open succeeded, the file might still be a directory
4996 {
4997 struct stat64 buf64;
4998 int ret = ::fstat64(fd, &buf64);
4999 int st_mode = buf64.st_mode;
5000
5001 if (ret != -1) {
5002 if ((st_mode & S_IFMT) == S_IFDIR) {
5003 errno = EISDIR;
5004 ::close(fd);
5005 return -1;
5006 }
5007 } else {
5008 ::close(fd);
5009 return -1;
5010 }
5011 }
5012
5013 #ifdef FD_CLOEXEC
5014 // Validate that the use of the O_CLOEXEC flag on open above worked.
5015 // With recent kernels, we will perform this check exactly once.
5016 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5017 if (!O_CLOEXEC_is_known_to_work) {
5018 int flags = ::fcntl(fd, F_GETFD);
5019 if (flags != -1) {
5020 if ((flags & FD_CLOEXEC) != 0)
5021 O_CLOEXEC_is_known_to_work = 1;
5022 else
5023 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5024 }
5025 }
5026 #endif
5027
5028 return fd;
5029 }
5030
5031
5032 // create binary file, rewriting existing file if required
5033 int os::create_binary_file(const char* path, bool rewrite_existing) {
5034 int oflags = O_WRONLY | O_CREAT;
5035 if (!rewrite_existing) {
5036 oflags |= O_EXCL;
5037 }
5038 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5039 }
5040
5041 // return current position of file pointer
5042 jlong os::current_file_offset(int fd) {
5043 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5044 }
5045
5046 // move file pointer to the specified offset
5047 jlong os::seek_to_file_offset(int fd, jlong offset) {
5048 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5049 }
5050
5051 // This code originates from JDK's sysAvailable
5052 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5053
5054 int os::available(int fd, jlong *bytes) {
5055 jlong cur, end;
5056 int mode;
5057 struct stat64 buf64;
5058
5059 if (::fstat64(fd, &buf64) >= 0) {
5060 mode = buf64.st_mode;
5061 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5062 int n;
5063 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5064 *bytes = n;
5065 return 1;
5066 }
5067 }
5068 }
5069 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5070 return 0;
5071 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5072 return 0;
5073 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5074 return 0;
5075 }
5076 *bytes = end - cur;
5077 return 1;
5078 }
5079
5080 // Map a block of memory.
5081 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5082 char *addr, size_t bytes, bool read_only,
5083 bool allow_exec) {
5084 int prot;
5085 int flags = MAP_PRIVATE;
5086
5087 if (read_only) {
5088 prot = PROT_READ;
5089 } else {
5090 prot = PROT_READ | PROT_WRITE;
5091 }
5092
5093 if (allow_exec) {
5094 prot |= PROT_EXEC;
5095 }
5096
5097 if (addr != NULL) {
5098 flags |= MAP_FIXED;
5099 }
5100
5101 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5102 fd, file_offset);
5103 if (mapped_address == MAP_FAILED) {
5104 return NULL;
5105 }
5106 return mapped_address;
5107 }
5108
5109
5110 // Remap a block of memory.
5111 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5112 char *addr, size_t bytes, bool read_only,
5113 bool allow_exec) {
5114 // same as map_memory() on this OS
5115 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5116 allow_exec);
5117 }
5118
5119
5120 // Unmap a block of memory.
5121 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5122 return munmap(addr, bytes) == 0;
5123 }
5124
5125 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5126
5127 static clockid_t thread_cpu_clockid(Thread* thread) {
5128 pthread_t tid = thread->osthread()->pthread_id();
5129 clockid_t clockid;
5130
5131 // Get thread clockid
5132 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5133 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5134 return clockid;
5135 }
5136
5137 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5138 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5139 // of a thread.
5140 //
5141 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5142 // the fast estimate available on the platform.
5143
5144 jlong os::current_thread_cpu_time() {
5145 if (os::Linux::supports_fast_thread_cpu_time()) {
5146 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5147 } else {
5148 // return user + sys since the cost is the same
5149 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5150 }
5151 }
5152
5153 jlong os::thread_cpu_time(Thread* thread) {
5154 // consistent with what current_thread_cpu_time() returns
5155 if (os::Linux::supports_fast_thread_cpu_time()) {
5156 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5157 } else {
5158 return slow_thread_cpu_time(thread, true /* user + sys */);
5159 }
5160 }
5161
5162 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5163 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5164 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5165 } else {
5166 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5167 }
5168 }
5169
5170 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5171 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5172 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5173 } else {
5174 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5175 }
5176 }
5177
5178 // -1 on error.
5179 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5180 pid_t tid = thread->osthread()->thread_id();
5181 char *s;
5182 char stat[2048];
5183 int statlen;
5184 char proc_name[64];
5185 int count;
5186 long sys_time, user_time;
5187 char cdummy;
5188 int idummy;
5189 long ldummy;
5190 FILE *fp;
5191
5192 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5193 fp = fopen(proc_name, "r");
5194 if (fp == NULL) return -1;
5195 statlen = fread(stat, 1, 2047, fp);
5196 stat[statlen] = '\0';
5197 fclose(fp);
5198
5199 // Skip pid and the command string. Note that we could be dealing with
5200 // weird command names, e.g. user could decide to rename java launcher
5201 // to "java 1.4.2 :)", then the stat file would look like
5202 // 1234 (java 1.4.2 :)) R ... ...
5203 // We don't really need to know the command string, just find the last
5204 // occurrence of ")" and then start parsing from there. See bug 4726580.
5205 s = strrchr(stat, ')');
5206 if (s == NULL) return -1;
5207
5208 // Skip blank chars
5209 do { s++; } while (s && isspace(*s));
5210
5211 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5212 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5213 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5214 &user_time, &sys_time);
5215 if (count != 13) return -1;
5216 if (user_sys_cpu_time) {
5217 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5218 } else {
5219 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5220 }
5221 }
5222
5223 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5224 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5225 info_ptr->may_skip_backward = false; // elapsed time not wall time
5226 info_ptr->may_skip_forward = false; // elapsed time not wall time
5227 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5228 }
5229
5230 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5231 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5232 info_ptr->may_skip_backward = false; // elapsed time not wall time
5233 info_ptr->may_skip_forward = false; // elapsed time not wall time
5234 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5235 }
5236
5237 bool os::is_thread_cpu_time_supported() {
5238 return true;
5239 }
5240
5241 // System loadavg support. Returns -1 if load average cannot be obtained.
5242 // Linux doesn't yet have a (official) notion of processor sets,
5243 // so just return the system wide load average.
5244 int os::loadavg(double loadavg[], int nelem) {
5245 return ::getloadavg(loadavg, nelem);
5246 }
5247
5248 void os::pause() {
5249 char filename[MAX_PATH];
5250 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5251 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5252 } else {
5253 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5254 }
5255
5256 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5257 if (fd != -1) {
5258 struct stat buf;
5259 ::close(fd);
5260 while (::stat(filename, &buf) == 0) {
5261 (void)::poll(NULL, 0, 100);
5262 }
5263 } else {
5264 jio_fprintf(stderr,
5265 "Could not open pause file '%s', continuing immediately.\n", filename);
5266 }
5267 }
5268
5269
5270 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5271 // comment paragraphs are worth repeating here:
5272 //
5273 // Assumption:
5274 // Only one parker can exist on an event, which is why we allocate
5275 // them per-thread. Multiple unparkers can coexist.
5276 //
5277 // _Event serves as a restricted-range semaphore.
5278 // -1 : thread is blocked, i.e. there is a waiter
5279 // 0 : neutral: thread is running or ready,
5280 // could have been signaled after a wait started
5281 // 1 : signaled - thread is running or ready
5282 //
5283 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5284 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5285 // For specifics regarding the bug see GLIBC BUGID 261237 :
5286 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5287 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5288 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5289 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5290 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5291 // and monitorenter when we're using 1-0 locking. All those operations may result in
5292 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5293 // of libpthread avoids the problem, but isn't practical.
5294 //
5295 // Possible remedies:
5296 //
5297 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5298 // This is palliative and probabilistic, however. If the thread is preempted
5299 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5300 // than the minimum period may have passed, and the abstime may be stale (in the
5301 // past) resultin in a hang. Using this technique reduces the odds of a hang
5302 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5303 //
5304 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5305 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5306 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5307 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5308 // thread.
5309 //
5310 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5311 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5312 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5313 // This also works well. In fact it avoids kernel-level scalability impediments
5314 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5315 // timers in a graceful fashion.
5316 //
5317 // 4. When the abstime value is in the past it appears that control returns
5318 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5319 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5320 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5321 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5322 // It may be possible to avoid reinitialization by checking the return
5323 // value from pthread_cond_timedwait(). In addition to reinitializing the
5324 // condvar we must establish the invariant that cond_signal() is only called
5325 // within critical sections protected by the adjunct mutex. This prevents
5326 // cond_signal() from "seeing" a condvar that's in the midst of being
5327 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5328 // desirable signal-after-unlock optimization that avoids futile context switching.
5329 //
5330 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5331 // structure when a condvar is used or initialized. cond_destroy() would
5332 // release the helper structure. Our reinitialize-after-timedwait fix
5333 // put excessive stress on malloc/free and locks protecting the c-heap.
5334 //
5335 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5336 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5337 // and only enabling the work-around for vulnerable environments.
5338
5339 // utility to compute the abstime argument to timedwait:
5340 // millis is the relative timeout time
5341 // abstime will be the absolute timeout time
5342 // TODO: replace compute_abstime() with unpackTime()
5343
5344 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5345 if (millis < 0) millis = 0;
5346
5347 jlong seconds = millis / 1000;
5348 millis %= 1000;
5349 if (seconds > 50000000) { // see man cond_timedwait(3T)
5350 seconds = 50000000;
5351 }
5352
5353 if (os::supports_monotonic_clock()) {
5354 struct timespec now;
5355 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5356 assert_status(status == 0, status, "clock_gettime");
5357 abstime->tv_sec = now.tv_sec + seconds;
5358 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5359 if (nanos >= NANOSECS_PER_SEC) {
5360 abstime->tv_sec += 1;
5361 nanos -= NANOSECS_PER_SEC;
5362 }
5363 abstime->tv_nsec = nanos;
5364 } else {
5365 struct timeval now;
5366 int status = gettimeofday(&now, NULL);
5367 assert(status == 0, "gettimeofday");
5368 abstime->tv_sec = now.tv_sec + seconds;
5369 long usec = now.tv_usec + millis * 1000;
5370 if (usec >= 1000000) {
5371 abstime->tv_sec += 1;
5372 usec -= 1000000;
5373 }
5374 abstime->tv_nsec = usec * 1000;
5375 }
5376 return abstime;
5377 }
5378
5379 void os::PlatformEvent::park() { // AKA "down()"
5380 // Transitions for _Event:
5381 // -1 => -1 : illegal
5382 // 1 => 0 : pass - return immediately
5383 // 0 => -1 : block; then set _Event to 0 before returning
5384
5385 // Invariant: Only the thread associated with the Event/PlatformEvent
5386 // may call park().
5387 // TODO: assert that _Assoc != NULL or _Assoc == Self
5388 assert(_nParked == 0, "invariant");
5389
5390 int v;
5391 for (;;) {
5392 v = _Event;
5393 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5394 }
5395 guarantee(v >= 0, "invariant");
5396 if (v == 0) {
5397 // Do this the hard way by blocking ...
5398 int status = pthread_mutex_lock(_mutex);
5399 assert_status(status == 0, status, "mutex_lock");
5400 guarantee(_nParked == 0, "invariant");
5401 ++_nParked;
5402 while (_Event < 0) {
5403 status = pthread_cond_wait(_cond, _mutex);
5404 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5405 // Treat this the same as if the wait was interrupted
5406 if (status == ETIME) { status = EINTR; }
5407 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5408 }
5409 --_nParked;
5410
5411 _Event = 0;
5412 status = pthread_mutex_unlock(_mutex);
5413 assert_status(status == 0, status, "mutex_unlock");
5414 // Paranoia to ensure our locked and lock-free paths interact
5415 // correctly with each other.
5416 OrderAccess::fence();
5417 }
5418 guarantee(_Event >= 0, "invariant");
5419 }
5420
5421 int os::PlatformEvent::park(jlong millis) {
5422 // Transitions for _Event:
5423 // -1 => -1 : illegal
5424 // 1 => 0 : pass - return immediately
5425 // 0 => -1 : block; then set _Event to 0 before returning
5426
5427 guarantee(_nParked == 0, "invariant");
5428
5429 int v;
5430 for (;;) {
5431 v = _Event;
5432 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5433 }
5434 guarantee(v >= 0, "invariant");
5435 if (v != 0) return OS_OK;
5436
5437 // We do this the hard way, by blocking the thread.
5438 // Consider enforcing a minimum timeout value.
5439 struct timespec abst;
5440 compute_abstime(&abst, millis);
5441
5442 int ret = OS_TIMEOUT;
5443 int status = pthread_mutex_lock(_mutex);
5444 assert_status(status == 0, status, "mutex_lock");
5445 guarantee(_nParked == 0, "invariant");
5446 ++_nParked;
5447
5448 // Object.wait(timo) will return because of
5449 // (a) notification
5450 // (b) timeout
5451 // (c) thread.interrupt
5452 //
5453 // Thread.interrupt and object.notify{All} both call Event::set.
5454 // That is, we treat thread.interrupt as a special case of notification.
5455 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5456 // We assume all ETIME returns are valid.
5457 //
5458 // TODO: properly differentiate simultaneous notify+interrupt.
5459 // In that case, we should propagate the notify to another waiter.
5460
5461 while (_Event < 0) {
5462 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5463 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5464 pthread_cond_destroy(_cond);
5465 pthread_cond_init(_cond, os::Linux::condAttr());
5466 }
5467 assert_status(status == 0 || status == EINTR ||
5468 status == ETIME || status == ETIMEDOUT,
5469 status, "cond_timedwait");
5470 if (!FilterSpuriousWakeups) break; // previous semantics
5471 if (status == ETIME || status == ETIMEDOUT) break;
5472 // We consume and ignore EINTR and spurious wakeups.
5473 }
5474 --_nParked;
5475 if (_Event >= 0) {
5476 ret = OS_OK;
5477 }
5478 _Event = 0;
5479 status = pthread_mutex_unlock(_mutex);
5480 assert_status(status == 0, status, "mutex_unlock");
5481 assert(_nParked == 0, "invariant");
5482 // Paranoia to ensure our locked and lock-free paths interact
5483 // correctly with each other.
5484 OrderAccess::fence();
5485 return ret;
5486 }
5487
5488 void os::PlatformEvent::unpark() {
5489 // Transitions for _Event:
5490 // 0 => 1 : just return
5491 // 1 => 1 : just return
5492 // -1 => either 0 or 1; must signal target thread
5493 // That is, we can safely transition _Event from -1 to either
5494 // 0 or 1.
5495 // See also: "Semaphores in Plan 9" by Mullender & Cox
5496 //
5497 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5498 // that it will take two back-to-back park() calls for the owning
5499 // thread to block. This has the benefit of forcing a spurious return
5500 // from the first park() call after an unpark() call which will help
5501 // shake out uses of park() and unpark() without condition variables.
5502
5503 if (Atomic::xchg(1, &_Event) >= 0) return;
5504
5505 // Wait for the thread associated with the event to vacate
5506 int status = pthread_mutex_lock(_mutex);
5507 assert_status(status == 0, status, "mutex_lock");
5508 int AnyWaiters = _nParked;
5509 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5510 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5511 AnyWaiters = 0;
5512 pthread_cond_signal(_cond);
5513 }
5514 status = pthread_mutex_unlock(_mutex);
5515 assert_status(status == 0, status, "mutex_unlock");
5516 if (AnyWaiters != 0) {
5517 // Note that we signal() *after* dropping the lock for "immortal" Events.
5518 // This is safe and avoids a common class of futile wakeups. In rare
5519 // circumstances this can cause a thread to return prematurely from
5520 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5521 // will simply re-test the condition and re-park itself.
5522 // This provides particular benefit if the underlying platform does not
5523 // provide wait morphing.
5524 status = pthread_cond_signal(_cond);
5525 assert_status(status == 0, status, "cond_signal");
5526 }
5527 }
5528
5529
5530 // JSR166
5531 // -------------------------------------------------------
5532
5533 // The solaris and linux implementations of park/unpark are fairly
5534 // conservative for now, but can be improved. They currently use a
5535 // mutex/condvar pair, plus a a count.
5536 // Park decrements count if > 0, else does a condvar wait. Unpark
5537 // sets count to 1 and signals condvar. Only one thread ever waits
5538 // on the condvar. Contention seen when trying to park implies that someone
5539 // is unparking you, so don't wait. And spurious returns are fine, so there
5540 // is no need to track notifications.
5541
5542 // This code is common to linux and solaris and will be moved to a
5543 // common place in dolphin.
5544 //
5545 // The passed in time value is either a relative time in nanoseconds
5546 // or an absolute time in milliseconds. Either way it has to be unpacked
5547 // into suitable seconds and nanoseconds components and stored in the
5548 // given timespec structure.
5549 // Given time is a 64-bit value and the time_t used in the timespec is only
5550 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5551 // overflow if times way in the future are given. Further on Solaris versions
5552 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5553 // number of seconds, in abstime, is less than current_time + 100,000,000.
5554 // As it will be 28 years before "now + 100000000" will overflow we can
5555 // ignore overflow and just impose a hard-limit on seconds using the value
5556 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5557 // years from "now".
5558
5559 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5560 assert(time > 0, "convertTime");
5561 time_t max_secs = 0;
5562
5563 if (!os::supports_monotonic_clock() || isAbsolute) {
5564 struct timeval now;
5565 int status = gettimeofday(&now, NULL);
5566 assert(status == 0, "gettimeofday");
5567
5568 max_secs = now.tv_sec + MAX_SECS;
5569
5570 if (isAbsolute) {
5571 jlong secs = time / 1000;
5572 if (secs > max_secs) {
5573 absTime->tv_sec = max_secs;
5574 } else {
5575 absTime->tv_sec = secs;
5576 }
5577 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5578 } else {
5579 jlong secs = time / NANOSECS_PER_SEC;
5580 if (secs >= MAX_SECS) {
5581 absTime->tv_sec = max_secs;
5582 absTime->tv_nsec = 0;
5583 } else {
5584 absTime->tv_sec = now.tv_sec + secs;
5585 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5586 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5587 absTime->tv_nsec -= NANOSECS_PER_SEC;
5588 ++absTime->tv_sec; // note: this must be <= max_secs
5589 }
5590 }
5591 }
5592 } else {
5593 // must be relative using monotonic clock
5594 struct timespec now;
5595 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5596 assert_status(status == 0, status, "clock_gettime");
5597 max_secs = now.tv_sec + MAX_SECS;
5598 jlong secs = time / NANOSECS_PER_SEC;
5599 if (secs >= MAX_SECS) {
5600 absTime->tv_sec = max_secs;
5601 absTime->tv_nsec = 0;
5602 } else {
5603 absTime->tv_sec = now.tv_sec + secs;
5604 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5605 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5606 absTime->tv_nsec -= NANOSECS_PER_SEC;
5607 ++absTime->tv_sec; // note: this must be <= max_secs
5608 }
5609 }
5610 }
5611 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5612 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5613 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5614 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5615 }
5616
5617 void Parker::park(bool isAbsolute, jlong time) {
5618 // Ideally we'd do something useful while spinning, such
5619 // as calling unpackTime().
5620
5621 // Optional fast-path check:
5622 // Return immediately if a permit is available.
5623 // We depend on Atomic::xchg() having full barrier semantics
5624 // since we are doing a lock-free update to _counter.
5625 if (Atomic::xchg(0, &_counter) > 0) return;
5626
5627 Thread* thread = Thread::current();
5628 assert(thread->is_Java_thread(), "Must be JavaThread");
5629 JavaThread *jt = (JavaThread *)thread;
5630
5631 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5632 // Check interrupt before trying to wait
5633 if (Thread::is_interrupted(thread, false)) {
5634 return;
5635 }
5636
5637 // Next, demultiplex/decode time arguments
5638 timespec absTime;
5639 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5640 return;
5641 }
5642 if (time > 0) {
5643 unpackTime(&absTime, isAbsolute, time);
5644 }
5645
5646
5647 // Enter safepoint region
5648 // Beware of deadlocks such as 6317397.
5649 // The per-thread Parker:: mutex is a classic leaf-lock.
5650 // In particular a thread must never block on the Threads_lock while
5651 // holding the Parker:: mutex. If safepoints are pending both the
5652 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5653 ThreadBlockInVM tbivm(jt);
5654
5655 // Don't wait if cannot get lock since interference arises from
5656 // unblocking. Also. check interrupt before trying wait
5657 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5658 return;
5659 }
5660
5661 int status;
5662 if (_counter > 0) { // no wait needed
5663 _counter = 0;
5664 status = pthread_mutex_unlock(_mutex);
5665 assert(status == 0, "invariant");
5666 // Paranoia to ensure our locked and lock-free paths interact
5667 // correctly with each other and Java-level accesses.
5668 OrderAccess::fence();
5669 return;
5670 }
5671
5672 #ifdef ASSERT
5673 // Don't catch signals while blocked; let the running threads have the signals.
5674 // (This allows a debugger to break into the running thread.)
5675 sigset_t oldsigs;
5676 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5677 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5678 #endif
5679
5680 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5681 jt->set_suspend_equivalent();
5682 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5683
5684 assert(_cur_index == -1, "invariant");
5685 if (time == 0) {
5686 _cur_index = REL_INDEX; // arbitrary choice when not timed
5687 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5688 } else {
5689 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5690 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5691 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5692 pthread_cond_destroy(&_cond[_cur_index]);
5693 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5694 }
5695 }
5696 _cur_index = -1;
5697 assert_status(status == 0 || status == EINTR ||
5698 status == ETIME || status == ETIMEDOUT,
5699 status, "cond_timedwait");
5700
5701 #ifdef ASSERT
5702 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5703 #endif
5704
5705 _counter = 0;
5706 status = pthread_mutex_unlock(_mutex);
5707 assert_status(status == 0, status, "invariant");
5708 // Paranoia to ensure our locked and lock-free paths interact
5709 // correctly with each other and Java-level accesses.
5710 OrderAccess::fence();
5711
5712 // If externally suspended while waiting, re-suspend
5713 if (jt->handle_special_suspend_equivalent_condition()) {
5714 jt->java_suspend_self();
5715 }
5716 }
5717
5718 void Parker::unpark() {
5719 int status = pthread_mutex_lock(_mutex);
5720 assert(status == 0, "invariant");
5721 const int s = _counter;
5722 _counter = 1;
5723 if (s < 1) {
5724 // thread might be parked
5725 if (_cur_index != -1) {
5726 // thread is definitely parked
5727 if (WorkAroundNPTLTimedWaitHang) {
5728 status = pthread_cond_signal(&_cond[_cur_index]);
5729 assert(status == 0, "invariant");
5730 status = pthread_mutex_unlock(_mutex);
5731 assert(status == 0, "invariant");
5732 } else {
5733 // must capture correct index before unlocking
5734 int index = _cur_index;
5735 status = pthread_mutex_unlock(_mutex);
5736 assert(status == 0, "invariant");
5737 status = pthread_cond_signal(&_cond[index]);
5738 assert(status == 0, "invariant");
5739 }
5740 } else {
5741 pthread_mutex_unlock(_mutex);
5742 assert(status == 0, "invariant");
5743 }
5744 } else {
5745 pthread_mutex_unlock(_mutex);
5746 assert(status == 0, "invariant");
5747 }
5748 }
5749
5750
5751 extern char** environ;
5752
5753 // Run the specified command in a separate process. Return its exit value,
5754 // or -1 on failure (e.g. can't fork a new process).
5755 // Unlike system(), this function can be called from signal handler. It
5756 // doesn't block SIGINT et al.
5757 int os::fork_and_exec(char* cmd) {
5758 const char * argv[4] = {"sh", "-c", cmd, NULL};
5759
5760 pid_t pid = fork();
5761
5762 if (pid < 0) {
5763 // fork failed
5764 return -1;
5765
5766 } else if (pid == 0) {
5767 // child process
5768
5769 execve("/bin/sh", (char* const*)argv, environ);
5770
5771 // execve failed
5772 _exit(-1);
5773
5774 } else {
5775 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5776 // care about the actual exit code, for now.
5777
5778 int status;
5779
5780 // Wait for the child process to exit. This returns immediately if
5781 // the child has already exited. */
5782 while (waitpid(pid, &status, 0) < 0) {
5783 switch (errno) {
5784 case ECHILD: return 0;
5785 case EINTR: break;
5786 default: return -1;
5787 }
5788 }
5789
5790 if (WIFEXITED(status)) {
5791 // The child exited normally; get its exit code.
5792 return WEXITSTATUS(status);
5793 } else if (WIFSIGNALED(status)) {
5794 // The child exited because of a signal
5795 // The best value to return is 0x80 + signal number,
5796 // because that is what all Unix shells do, and because
5797 // it allows callers to distinguish between process exit and
5798 // process death by signal.
5799 return 0x80 + WTERMSIG(status);
5800 } else {
5801 // Unknown exit code; pass it through
5802 return status;
5803 }
5804 }
5805 }
5806
5807 // is_headless_jre()
5808 //
5809 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5810 // in order to report if we are running in a headless jre
5811 //
5812 // Since JDK8 xawt/libmawt.so was moved into the same directory
5813 // as libawt.so, and renamed libawt_xawt.so
5814 //
5815 bool os::is_headless_jre() {
5816 struct stat statbuf;
5817 char buf[MAXPATHLEN];
5818 char libmawtpath[MAXPATHLEN];
5819 const char *xawtstr = "/xawt/libmawt.so";
5820 const char *new_xawtstr = "/libawt_xawt.so";
5821 char *p;
5822
5823 // Get path to libjvm.so
5824 os::jvm_path(buf, sizeof(buf));
5825
5826 // Get rid of libjvm.so
5827 p = strrchr(buf, '/');
5828 if (p == NULL) {
5829 return false;
5830 } else {
5831 *p = '\0';
5832 }
5833
5834 // Get rid of client or server
5835 p = strrchr(buf, '/');
5836 if (p == NULL) {
5837 return false;
5838 } else {
5839 *p = '\0';
5840 }
5841
5842 // check xawt/libmawt.so
5843 strcpy(libmawtpath, buf);
5844 strcat(libmawtpath, xawtstr);
5845 if (::stat(libmawtpath, &statbuf) == 0) return false;
5846
5847 // check libawt_xawt.so
5848 strcpy(libmawtpath, buf);
5849 strcat(libmawtpath, new_xawtstr);
5850 if (::stat(libmawtpath, &statbuf) == 0) return false;
5851
5852 return true;
5853 }
5854
5855 // Get the default path to the core file
5856 // Returns the length of the string
5857 int os::get_core_path(char* buffer, size_t bufferSize) {
5858 /*
5859 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5860 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5861 */
5862 const int core_pattern_len = 129;
5863 char core_pattern[core_pattern_len] = {0};
5864
5865 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5866 if (core_pattern_file == -1) {
5867 return -1;
5868 }
5869
5870 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5871 ::close(core_pattern_file);
5872 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5873 return -1;
5874 }
5875 if (core_pattern[ret-1] == '\n') {
5876 core_pattern[ret-1] = '\0';
5877 } else {
5878 core_pattern[ret] = '\0';
5879 }
5880
5881 char *pid_pos = strstr(core_pattern, "%p");
5882 int written;
5883
5884 if (core_pattern[0] == '/') {
5885 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5886 } else {
5887 char cwd[PATH_MAX];
5888
5889 const char* p = get_current_directory(cwd, PATH_MAX);
5890 if (p == NULL) {
5891 return -1;
5892 }
5893
5894 if (core_pattern[0] == '|') {
5895 written = jio_snprintf(buffer, bufferSize,
5896 "\"%s\" (or dumping to %s/core.%d)",
5897 &core_pattern[1], p, current_process_id());
5898 } else {
5899 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5900 }
5901 }
5902
5903 if (written < 0) {
5904 return -1;
5905 }
5906
5907 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5908 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5909
5910 if (core_uses_pid_file != -1) {
5911 char core_uses_pid = 0;
5912 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5913 ::close(core_uses_pid_file);
5914
5915 if (core_uses_pid == '1') {
5916 jio_snprintf(buffer + written, bufferSize - written,
5917 ".%d", current_process_id());
5918 }
5919 }
5920 }
5921
5922 return strlen(buffer);
5923 }
5924
5925 bool os::start_debugging(char *buf, int buflen) {
5926 int len = (int)strlen(buf);
5927 char *p = &buf[len];
5928
5929 jio_snprintf(p, buflen-len,
5930 "\n\n"
5931 "Do you want to debug the problem?\n\n"
5932 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5933 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5934 "Otherwise, press RETURN to abort...",
5935 os::current_process_id(), os::current_process_id(),
5936 os::current_thread_id(), os::current_thread_id());
5937
5938 bool yes = os::message_box("Unexpected Error", buf);
5939
5940 if (yes) {
5941 // yes, user asked VM to launch debugger
5942 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
5943 os::current_process_id(), os::current_process_id());
5944
5945 os::fork_and_exec(buf);
5946 yes = false;
5947 }
5948 return yes;
5949 }
5950
5951
5952
5953 /////////////// Unit tests ///////////////
5954
5955 #ifndef PRODUCT
5956
5957 #define test_log(...) \
5958 do { \
5959 if (VerboseInternalVMTests) { \
5960 tty->print_cr(__VA_ARGS__); \
5961 tty->flush(); \
5962 } \
5963 } while (false)
5964
5965 class TestReserveMemorySpecial : AllStatic {
5966 public:
5967 static void small_page_write(void* addr, size_t size) {
5968 size_t page_size = os::vm_page_size();
5969
5970 char* end = (char*)addr + size;
5971 for (char* p = (char*)addr; p < end; p += page_size) {
5972 *p = 1;
5973 }
5974 }
5975
5976 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5977 if (!UseHugeTLBFS) {
5978 return;
5979 }
5980
5981 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5982
5983 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5984
5985 if (addr != NULL) {
5986 small_page_write(addr, size);
5987
5988 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5989 }
5990 }
5991
5992 static void test_reserve_memory_special_huge_tlbfs_only() {
5993 if (!UseHugeTLBFS) {
5994 return;
5995 }
5996
5997 size_t lp = os::large_page_size();
5998
5999 for (size_t size = lp; size <= lp * 10; size += lp) {
6000 test_reserve_memory_special_huge_tlbfs_only(size);
6001 }
6002 }
6003
6004 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6005 size_t lp = os::large_page_size();
6006 size_t ag = os::vm_allocation_granularity();
6007
6008 // sizes to test
6009 const size_t sizes[] = {
6010 lp, lp + ag, lp + lp / 2, lp * 2,
6011 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6012 lp * 10, lp * 10 + lp / 2
6013 };
6014 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6015
6016 // For each size/alignment combination, we test three scenarios:
6017 // 1) with req_addr == NULL
6018 // 2) with a non-null req_addr at which we expect to successfully allocate
6019 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6020 // expect the allocation to either fail or to ignore req_addr
6021
6022 // Pre-allocate two areas; they shall be as large as the largest allocation
6023 // and aligned to the largest alignment we will be testing.
6024 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6025 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6026 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6027 -1, 0);
6028 assert(mapping1 != MAP_FAILED, "should work");
6029
6030 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6031 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6032 -1, 0);
6033 assert(mapping2 != MAP_FAILED, "should work");
6034
6035 // Unmap the first mapping, but leave the second mapping intact: the first
6036 // mapping will serve as a value for a "good" req_addr (case 2). The second
6037 // mapping, still intact, as "bad" req_addr (case 3).
6038 ::munmap(mapping1, mapping_size);
6039
6040 // Case 1
6041 test_log("%s, req_addr NULL:", __FUNCTION__);
6042 test_log("size align result");
6043
6044 for (int i = 0; i < num_sizes; i++) {
6045 const size_t size = sizes[i];
6046 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6047 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6048 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6049 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6050 if (p != NULL) {
6051 assert(is_ptr_aligned(p, alignment), "must be");
6052 small_page_write(p, size);
6053 os::Linux::release_memory_special_huge_tlbfs(p, size);
6054 }
6055 }
6056 }
6057
6058 // Case 2
6059 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6060 test_log("size align req_addr result");
6061
6062 for (int i = 0; i < num_sizes; i++) {
6063 const size_t size = sizes[i];
6064 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6065 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6066 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6067 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6068 size, alignment, p2i(req_addr), p2i(p),
6069 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6070 if (p != NULL) {
6071 assert(p == req_addr, "must be");
6072 small_page_write(p, size);
6073 os::Linux::release_memory_special_huge_tlbfs(p, size);
6074 }
6075 }
6076 }
6077
6078 // Case 3
6079 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6080 test_log("size align req_addr result");
6081
6082 for (int i = 0; i < num_sizes; i++) {
6083 const size_t size = sizes[i];
6084 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6085 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6086 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6087 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6088 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6089 // as the area around req_addr contains already existing mappings, the API should always
6090 // return NULL (as per contract, it cannot return another address)
6091 assert(p == NULL, "must be");
6092 }
6093 }
6094
6095 ::munmap(mapping2, mapping_size);
6096
6097 }
6098
6099 static void test_reserve_memory_special_huge_tlbfs() {
6100 if (!UseHugeTLBFS) {
6101 return;
6102 }
6103
6104 test_reserve_memory_special_huge_tlbfs_only();
6105 test_reserve_memory_special_huge_tlbfs_mixed();
6106 }
6107
6108 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6109 if (!UseSHM) {
6110 return;
6111 }
6112
6113 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6114
6115 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6116
6117 if (addr != NULL) {
6118 assert(is_ptr_aligned(addr, alignment), "Check");
6119 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6120
6121 small_page_write(addr, size);
6122
6123 os::Linux::release_memory_special_shm(addr, size);
6124 }
6125 }
6126
6127 static void test_reserve_memory_special_shm() {
6128 size_t lp = os::large_page_size();
6129 size_t ag = os::vm_allocation_granularity();
6130
6131 for (size_t size = ag; size < lp * 3; size += ag) {
6132 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6133 test_reserve_memory_special_shm(size, alignment);
6134 }
6135 }
6136 }
6137
6138 static void test() {
6139 test_reserve_memory_special_huge_tlbfs();
6140 test_reserve_memory_special_shm();
6141 }
6142 };
6143
6144 void TestReserveMemorySpecial_test() {
6145 TestReserveMemorySpecial::test();
6146 }
6147
6148 #endif