1 /*
2 * Copyright (c) 1999, 2016, 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 "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_share_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm.h"
44 #include "prims/jvm_misc.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/atomic.inline.hpp"
47 #include "runtime/extendedPC.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/interfaceSupport.hpp"
50 #include "runtime/init.hpp"
51 #include "runtime/java.hpp"
52 #include "runtime/javaCalls.hpp"
53 #include "runtime/mutexLocker.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/orderAccess.inline.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 const int os::Linux::_vm_default_page_size = (8 * K);
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149 pthread_condattr_t os::Linux::_condattr[1];
150
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // Declarations
166 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
167
168 // utility functions
169
170 static int SR_initialize();
171
172 julong os::available_memory() {
173 return Linux::available_memory();
174 }
175
176 julong os::Linux::available_memory() {
177 // values in struct sysinfo are "unsigned long"
178 struct sysinfo si;
179 sysinfo(&si);
180
181 return (julong)si.freeram * si.mem_unit;
182 }
183
184 julong os::physical_memory() {
185 return Linux::physical_memory();
186 }
187
188 // Return true if user is running as root.
189
190 bool os::have_special_privileges() {
191 static bool init = false;
192 static bool privileges = false;
193 if (!init) {
194 privileges = (getuid() != geteuid()) || (getgid() != getegid());
195 init = true;
196 }
197 return privileges;
198 }
199
200
201 #ifndef SYS_gettid
202 // i386: 224, ia64: 1105, amd64: 186, sparc 143
203 #ifdef __ia64__
204 #define SYS_gettid 1105
205 #else
206 #ifdef __i386__
207 #define SYS_gettid 224
208 #else
209 #ifdef __amd64__
210 #define SYS_gettid 186
211 #else
212 #ifdef __sparc__
213 #define SYS_gettid 143
214 #else
215 #error define gettid for the arch
216 #endif
217 #endif
218 #endif
219 #endif
220 #endif
221
222 // Cpu architecture string
223 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
224
225
226 // pid_t gettid()
227 //
228 // Returns the kernel thread id of the currently running thread. Kernel
229 // thread id is used to access /proc.
230 pid_t os::Linux::gettid() {
231 int rslt = syscall(SYS_gettid);
232 assert(rslt != -1, "must be."); // old linuxthreads implementation?
233 return (pid_t)rslt;
234 }
235
236 // Most versions of linux have a bug where the number of processors are
237 // determined by looking at the /proc file system. In a chroot environment,
238 // the system call returns 1. This causes the VM to act as if it is
239 // a single processor and elide locking (see is_MP() call).
240 static bool unsafe_chroot_detected = false;
241 static const char *unstable_chroot_error = "/proc file system not found.\n"
242 "Java may be unstable running multithreaded in a chroot "
243 "environment on Linux when /proc filesystem is not mounted.";
244
245 void os::Linux::initialize_system_info() {
246 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
247 if (processor_count() == 1) {
248 pid_t pid = os::Linux::gettid();
249 char fname[32];
250 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
251 FILE *fp = fopen(fname, "r");
252 if (fp == NULL) {
253 unsafe_chroot_detected = true;
254 } else {
255 fclose(fp);
256 }
257 }
258 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
259 assert(processor_count() > 0, "linux error");
260 }
261
262 void os::init_system_properties_values() {
263 // The next steps are taken in the product version:
264 //
265 // Obtain the JAVA_HOME value from the location of libjvm.so.
266 // This library should be located at:
267 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
268 //
269 // If "/jre/lib/" appears at the right place in the path, then we
270 // assume libjvm.so is installed in a JDK and we use this path.
271 //
272 // Otherwise exit with message: "Could not create the Java virtual machine."
273 //
274 // The following extra steps are taken in the debugging version:
275 //
276 // If "/jre/lib/" does NOT appear at the right place in the path
277 // instead of exit check for $JAVA_HOME environment variable.
278 //
279 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
280 // then we append a fake suffix "hotspot/libjvm.so" to this path so
281 // it looks like libjvm.so is installed there
282 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
283 //
284 // Otherwise exit.
285 //
286 // Important note: if the location of libjvm.so changes this
287 // code needs to be changed accordingly.
288
289 // See ld(1):
290 // The linker uses the following search paths to locate required
291 // shared libraries:
292 // 1: ...
293 // ...
294 // 7: The default directories, normally /lib and /usr/lib.
295 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
296 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
297 #else
298 #define DEFAULT_LIBPATH "/lib:/usr/lib"
299 #endif
300
301 // Base path of extensions installed on the system.
302 #define SYS_EXT_DIR "/usr/java/packages"
303 #define EXTENSIONS_DIR "/lib/ext"
304
305 // Buffer that fits several sprintfs.
306 // Note that the space for the colon and the trailing null are provided
307 // by the nulls included by the sizeof operator.
308 const size_t bufsize =
309 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
310 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
311 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
312
313 // sysclasspath, java_home, dll_dir
314 {
315 char *pslash;
316 os::jvm_path(buf, bufsize);
317
318 // Found the full path to libjvm.so.
319 // Now cut the path to <java_home>/jre if we can.
320 pslash = strrchr(buf, '/');
321 if (pslash != NULL) {
322 *pslash = '\0'; // Get rid of /libjvm.so.
323 }
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
327 }
328 Arguments::set_dll_dir(buf);
329
330 if (pslash != NULL) {
331 pslash = strrchr(buf, '/');
332 if (pslash != NULL) {
333 *pslash = '\0'; // Get rid of /<arch>.
334 pslash = strrchr(buf, '/');
335 if (pslash != NULL) {
336 *pslash = '\0'; // Get rid of /lib.
337 }
338 }
339 }
340 Arguments::set_java_home(buf);
341 set_boot_path('/', ':');
342 }
343
344 // Where to look for native libraries.
345 //
346 // Note: Due to a legacy implementation, most of the library path
347 // is set in the launcher. This was to accomodate linking restrictions
348 // on legacy Linux implementations (which are no longer supported).
349 // Eventually, all the library path setting will be done here.
350 //
351 // However, to prevent the proliferation of improperly built native
352 // libraries, the new path component /usr/java/packages is added here.
353 // Eventually, all the library path setting will be done here.
354 {
355 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
356 // should always exist (until the legacy problem cited above is
357 // addressed).
358 const char *v = ::getenv("LD_LIBRARY_PATH");
359 const char *v_colon = ":";
360 if (v == NULL) { v = ""; v_colon = ""; }
361 // That's +1 for the colon and +1 for the trailing '\0'.
362 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
363 strlen(v) + 1 +
364 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
365 mtInternal);
366 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
367 Arguments::set_library_path(ld_library_path);
368 FREE_C_HEAP_ARRAY(char, ld_library_path);
369 }
370
371 // Extensions directories.
372 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
373 Arguments::set_ext_dirs(buf);
374
375 FREE_C_HEAP_ARRAY(char, buf);
376
377 #undef DEFAULT_LIBPATH
378 #undef SYS_EXT_DIR
379 #undef EXTENSIONS_DIR
380 }
381
382 ////////////////////////////////////////////////////////////////////////////////
383 // breakpoint support
384
385 void os::breakpoint() {
386 BREAKPOINT;
387 }
388
389 extern "C" void breakpoint() {
390 // use debugger to set breakpoint here
391 }
392
393 ////////////////////////////////////////////////////////////////////////////////
394 // signal support
395
396 debug_only(static bool signal_sets_initialized = false);
397 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
398
399 bool os::Linux::is_sig_ignored(int sig) {
400 struct sigaction oact;
401 sigaction(sig, (struct sigaction*)NULL, &oact);
402 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
403 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
404 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
405 return true;
406 } else {
407 return false;
408 }
409 }
410
411 void os::Linux::signal_sets_init() {
412 // Should also have an assertion stating we are still single-threaded.
413 assert(!signal_sets_initialized, "Already initialized");
414 // Fill in signals that are necessarily unblocked for all threads in
415 // the VM. Currently, we unblock the following signals:
416 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
417 // by -Xrs (=ReduceSignalUsage));
418 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
419 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
420 // the dispositions or masks wrt these signals.
421 // Programs embedding the VM that want to use the above signals for their
422 // own purposes must, at this time, use the "-Xrs" option to prevent
423 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
424 // (See bug 4345157, and other related bugs).
425 // In reality, though, unblocking these signals is really a nop, since
426 // these signals are not blocked by default.
427 sigemptyset(&unblocked_sigs);
428 sigemptyset(&allowdebug_blocked_sigs);
429 sigaddset(&unblocked_sigs, SIGILL);
430 sigaddset(&unblocked_sigs, SIGSEGV);
431 sigaddset(&unblocked_sigs, SIGBUS);
432 sigaddset(&unblocked_sigs, SIGFPE);
433 #if defined(PPC64)
434 sigaddset(&unblocked_sigs, SIGTRAP);
435 #endif
436 sigaddset(&unblocked_sigs, SR_signum);
437
438 if (!ReduceSignalUsage) {
439 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
440 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
441 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
442 }
443 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
444 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
445 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
446 }
447 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
448 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
449 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
450 }
451 }
452 // Fill in signals that are blocked by all but the VM thread.
453 sigemptyset(&vm_sigs);
454 if (!ReduceSignalUsage) {
455 sigaddset(&vm_sigs, BREAK_SIGNAL);
456 }
457 debug_only(signal_sets_initialized = true);
458
459 }
460
461 // These are signals that are unblocked while a thread is running Java.
462 // (For some reason, they get blocked by default.)
463 sigset_t* os::Linux::unblocked_signals() {
464 assert(signal_sets_initialized, "Not initialized");
465 return &unblocked_sigs;
466 }
467
468 // These are the signals that are blocked while a (non-VM) thread is
469 // running Java. Only the VM thread handles these signals.
470 sigset_t* os::Linux::vm_signals() {
471 assert(signal_sets_initialized, "Not initialized");
472 return &vm_sigs;
473 }
474
475 // These are signals that are blocked during cond_wait to allow debugger in
476 sigset_t* os::Linux::allowdebug_blocked_signals() {
477 assert(signal_sets_initialized, "Not initialized");
478 return &allowdebug_blocked_sigs;
479 }
480
481 void os::Linux::hotspot_sigmask(Thread* thread) {
482
483 //Save caller's signal mask before setting VM signal mask
484 sigset_t caller_sigmask;
485 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
486
487 OSThread* osthread = thread->osthread();
488 osthread->set_caller_sigmask(caller_sigmask);
489
490 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
491
492 if (!ReduceSignalUsage) {
493 if (thread->is_VM_thread()) {
494 // Only the VM thread handles BREAK_SIGNAL ...
495 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
496 } else {
497 // ... all other threads block BREAK_SIGNAL
498 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
499 }
500 }
501 }
502
503 //////////////////////////////////////////////////////////////////////////////
504 // detecting pthread library
505
506 void os::Linux::libpthread_init() {
507 // Save glibc and pthread version strings.
508 #if !defined(_CS_GNU_LIBC_VERSION) || \
509 !defined(_CS_GNU_LIBPTHREAD_VERSION)
510 #error "glibc too old (< 2.3.2)"
511 #endif
512
513 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
514 assert(n > 0, "cannot retrieve glibc version");
515 char *str = (char *)malloc(n, mtInternal);
516 confstr(_CS_GNU_LIBC_VERSION, str, n);
517 os::Linux::set_glibc_version(str);
518
519 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
520 assert(n > 0, "cannot retrieve pthread version");
521 str = (char *)malloc(n, mtInternal);
522 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
523 os::Linux::set_libpthread_version(str);
524 }
525
526 /////////////////////////////////////////////////////////////////////////////
527 // thread stack expansion
528
529 // os::Linux::manually_expand_stack() takes care of expanding the thread
530 // stack. Note that this is normally not needed: pthread stacks allocate
531 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
532 // committed. Therefore it is not necessary to expand the stack manually.
533 //
534 // Manually expanding the stack was historically needed on LinuxThreads
535 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
536 // it is kept to deal with very rare corner cases:
537 //
538 // For one, user may run the VM on an own implementation of threads
539 // whose stacks are - like the old LinuxThreads - implemented using
540 // mmap(MAP_GROWSDOWN).
541 //
542 // Also, this coding may be needed if the VM is running on the primordial
543 // thread. Normally we avoid running on the primordial thread; however,
544 // user may still invoke the VM on the primordial thread.
545 //
546 // The following historical comment describes the details about running
547 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
548
549
550 // Force Linux kernel to expand current thread stack. If "bottom" is close
551 // to the stack guard, caller should block all signals.
552 //
553 // MAP_GROWSDOWN:
554 // A special mmap() flag that is used to implement thread stacks. It tells
555 // kernel that the memory region should extend downwards when needed. This
556 // allows early versions of LinuxThreads to only mmap the first few pages
557 // when creating a new thread. Linux kernel will automatically expand thread
558 // stack as needed (on page faults).
559 //
560 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
561 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
562 // region, it's hard to tell if the fault is due to a legitimate stack
563 // access or because of reading/writing non-exist memory (e.g. buffer
564 // overrun). As a rule, if the fault happens below current stack pointer,
565 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
566 // application (see Linux kernel fault.c).
567 //
568 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
569 // stack overflow detection.
570 //
571 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
572 // not use MAP_GROWSDOWN.
573 //
574 // To get around the problem and allow stack banging on Linux, we need to
575 // manually expand thread stack after receiving the SIGSEGV.
576 //
577 // There are two ways to expand thread stack to address "bottom", we used
578 // both of them in JVM before 1.5:
579 // 1. adjust stack pointer first so that it is below "bottom", and then
580 // touch "bottom"
581 // 2. mmap() the page in question
582 //
583 // Now alternate signal stack is gone, it's harder to use 2. For instance,
584 // if current sp is already near the lower end of page 101, and we need to
585 // call mmap() to map page 100, it is possible that part of the mmap() frame
586 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
587 // That will destroy the mmap() frame and cause VM to crash.
588 //
589 // The following code works by adjusting sp first, then accessing the "bottom"
590 // page to force a page fault. Linux kernel will then automatically expand the
591 // stack mapping.
592 //
593 // _expand_stack_to() assumes its frame size is less than page size, which
594 // should always be true if the function is not inlined.
595
596 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
597 #define NOINLINE
598 #else
599 #define NOINLINE __attribute__ ((noinline))
600 #endif
601
602 static void _expand_stack_to(address bottom) NOINLINE;
603
604 static void _expand_stack_to(address bottom) {
605 address sp;
606 size_t size;
607 volatile char *p;
608
609 // Adjust bottom to point to the largest address within the same page, it
610 // gives us a one-page buffer if alloca() allocates slightly more memory.
611 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
612 bottom += os::Linux::page_size() - 1;
613
614 // sp might be slightly above current stack pointer; if that's the case, we
615 // will alloca() a little more space than necessary, which is OK. Don't use
616 // os::current_stack_pointer(), as its result can be slightly below current
617 // stack pointer, causing us to not alloca enough to reach "bottom".
618 sp = (address)&sp;
619
620 if (sp > bottom) {
621 size = sp - bottom;
622 p = (volatile char *)alloca(size);
623 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
624 p[0] = '\0';
625 }
626 }
627
628 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
629 assert(t!=NULL, "just checking");
630 assert(t->osthread()->expanding_stack(), "expand should be set");
631 assert(t->stack_base() != NULL, "stack_base was not initialized");
632
633 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
634 sigset_t mask_all, old_sigset;
635 sigfillset(&mask_all);
636 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
637 _expand_stack_to(addr);
638 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
639 return true;
640 }
641 return false;
642 }
643
644 //////////////////////////////////////////////////////////////////////////////
645 // create new thread
646
647 // Thread start routine for all newly created threads
648 static void *java_start(Thread *thread) {
649 // Try to randomize the cache line index of hot stack frames.
650 // This helps when threads of the same stack traces evict each other's
651 // cache lines. The threads can be either from the same JVM instance, or
652 // from different JVM instances. The benefit is especially true for
653 // processors with hyperthreading technology.
654 static int counter = 0;
655 int pid = os::current_process_id();
656 alloca(((pid ^ counter++) & 7) * 128);
657
658 thread->initialize_thread_current();
659
660 OSThread* osthread = thread->osthread();
661 Monitor* sync = osthread->startThread_lock();
662
663 osthread->set_thread_id(os::current_thread_id());
664
665 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
666 os::current_thread_id(), (uintx) pthread_self());
667
668 if (UseNUMA) {
669 int lgrp_id = os::numa_get_group_id();
670 if (lgrp_id != -1) {
671 thread->set_lgrp_id(lgrp_id);
672 }
673 }
674 // initialize signal mask for this thread
675 os::Linux::hotspot_sigmask(thread);
676
677 // initialize floating point control register
678 os::Linux::init_thread_fpu_state();
679
680 // handshaking with parent thread
681 {
682 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
683
684 // notify parent thread
685 osthread->set_state(INITIALIZED);
686 sync->notify_all();
687
688 // wait until os::start_thread()
689 while (osthread->get_state() == INITIALIZED) {
690 sync->wait(Mutex::_no_safepoint_check_flag);
691 }
692 }
693
694 // call one more level start routine
695 thread->run();
696
697 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
698 os::current_thread_id(), (uintx) pthread_self());
699
700 return 0;
701 }
702
703 bool os::create_thread(Thread* thread, ThreadType thr_type,
704 size_t stack_size) {
705 assert(thread->osthread() == NULL, "caller responsible");
706
707 // Allocate the OSThread object
708 OSThread* osthread = new OSThread(NULL, NULL);
709 if (osthread == NULL) {
710 return false;
711 }
712
713 // set the correct thread state
714 osthread->set_thread_type(thr_type);
715
716 // Initial state is ALLOCATED but not INITIALIZED
717 osthread->set_state(ALLOCATED);
718
719 thread->set_osthread(osthread);
720
721 // init thread attributes
722 pthread_attr_t attr;
723 pthread_attr_init(&attr);
724 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
725
726 // stack size
727 // calculate stack size if it's not specified by caller
728 if (stack_size == 0) {
729 stack_size = os::Linux::default_stack_size(thr_type);
730
731 switch (thr_type) {
732 case os::java_thread:
733 // Java threads use ThreadStackSize which default value can be
734 // changed with the flag -Xss
735 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
736 stack_size = JavaThread::stack_size_at_create();
737 break;
738 case os::compiler_thread:
739 if (CompilerThreadStackSize > 0) {
740 stack_size = (size_t)(CompilerThreadStackSize * K);
741 break;
742 } // else fall through:
743 // use VMThreadStackSize if CompilerThreadStackSize is not defined
744 case os::vm_thread:
745 case os::pgc_thread:
746 case os::cgc_thread:
747 case os::watcher_thread:
748 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
749 break;
750 }
751 }
752
753 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
754 pthread_attr_setstacksize(&attr, stack_size);
755
756 // glibc guard page
757 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
758
759 ThreadState state;
760
761 {
762 pthread_t tid;
763 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
764
765 char buf[64];
766 if (ret == 0) {
767 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
768 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
769 } else {
770 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
771 strerror(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
772 }
773
774 pthread_attr_destroy(&attr);
775
776 if (ret != 0) {
777 // Need to clean up stuff we've allocated so far
778 thread->set_osthread(NULL);
779 delete osthread;
780 return false;
781 }
782
783 // Store pthread info into the OSThread
784 osthread->set_pthread_id(tid);
785
786 // Wait until child thread is either initialized or aborted
787 {
788 Monitor* sync_with_child = osthread->startThread_lock();
789 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
790 while ((state = osthread->get_state()) == ALLOCATED) {
791 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
792 }
793 }
794 }
795
796 // Aborted due to thread limit being reached
797 if (state == ZOMBIE) {
798 thread->set_osthread(NULL);
799 delete osthread;
800 return false;
801 }
802
803 // The thread is returned suspended (in state INITIALIZED),
804 // and is started higher up in the call chain
805 assert(state == INITIALIZED, "race condition");
806 return true;
807 }
808
809 /////////////////////////////////////////////////////////////////////////////
810 // attach existing thread
811
812 // bootstrap the main thread
813 bool os::create_main_thread(JavaThread* thread) {
814 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
815 return create_attached_thread(thread);
816 }
817
818 bool os::create_attached_thread(JavaThread* thread) {
819 #ifdef ASSERT
820 thread->verify_not_published();
821 #endif
822
823 // Allocate the OSThread object
824 OSThread* osthread = new OSThread(NULL, NULL);
825
826 if (osthread == NULL) {
827 return false;
828 }
829
830 // Store pthread info into the OSThread
831 osthread->set_thread_id(os::Linux::gettid());
832 osthread->set_pthread_id(::pthread_self());
833
834 // initialize floating point control register
835 os::Linux::init_thread_fpu_state();
836
837 // Initial thread state is RUNNABLE
838 osthread->set_state(RUNNABLE);
839
840 thread->set_osthread(osthread);
841
842 if (UseNUMA) {
843 int lgrp_id = os::numa_get_group_id();
844 if (lgrp_id != -1) {
845 thread->set_lgrp_id(lgrp_id);
846 }
847 }
848
849 if (os::Linux::is_initial_thread()) {
850 // If current thread is initial thread, its stack is mapped on demand,
851 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
852 // the entire stack region to avoid SEGV in stack banging.
853 // It is also useful to get around the heap-stack-gap problem on SuSE
854 // kernel (see 4821821 for details). We first expand stack to the top
855 // of yellow zone, then enable stack yellow zone (order is significant,
856 // enabling yellow zone first will crash JVM on SuSE Linux), so there
857 // is no gap between the last two virtual memory regions.
858
859 JavaThread *jt = (JavaThread *)thread;
860 address addr = jt->stack_reserved_zone_base();
861 assert(addr != NULL, "initialization problem?");
862 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
863
864 osthread->set_expanding_stack();
865 os::Linux::manually_expand_stack(jt, addr);
866 osthread->clear_expanding_stack();
867 }
868
869 // initialize signal mask for this thread
870 // and save the caller's signal mask
871 os::Linux::hotspot_sigmask(thread);
872
873 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
874 os::current_thread_id(), (uintx) pthread_self());
875
876 return true;
877 }
878
879 void os::pd_start_thread(Thread* thread) {
880 OSThread * osthread = thread->osthread();
881 assert(osthread->get_state() != INITIALIZED, "just checking");
882 Monitor* sync_with_child = osthread->startThread_lock();
883 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
884 sync_with_child->notify();
885 }
886
887 // Free Linux resources related to the OSThread
888 void os::free_thread(OSThread* osthread) {
889 assert(osthread != NULL, "osthread not set");
890
891 if (Thread::current()->osthread() == osthread) {
892 // Restore caller's signal mask
893 sigset_t sigmask = osthread->caller_sigmask();
894 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
895 }
896
897 delete osthread;
898 }
899
900 //////////////////////////////////////////////////////////////////////////////
901 // initial thread
902
903 // Check if current thread is the initial thread, similar to Solaris thr_main.
904 bool os::Linux::is_initial_thread(void) {
905 char dummy;
906 // If called before init complete, thread stack bottom will be null.
907 // Can be called if fatal error occurs before initialization.
908 if (initial_thread_stack_bottom() == NULL) return false;
909 assert(initial_thread_stack_bottom() != NULL &&
910 initial_thread_stack_size() != 0,
911 "os::init did not locate initial thread's stack region");
912 if ((address)&dummy >= initial_thread_stack_bottom() &&
913 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
914 return true;
915 } else {
916 return false;
917 }
918 }
919
920 // Find the virtual memory area that contains addr
921 static bool find_vma(address addr, address* vma_low, address* vma_high) {
922 FILE *fp = fopen("/proc/self/maps", "r");
923 if (fp) {
924 address low, high;
925 while (!feof(fp)) {
926 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
927 if (low <= addr && addr < high) {
928 if (vma_low) *vma_low = low;
929 if (vma_high) *vma_high = high;
930 fclose(fp);
931 return true;
932 }
933 }
934 for (;;) {
935 int ch = fgetc(fp);
936 if (ch == EOF || ch == (int)'\n') break;
937 }
938 }
939 fclose(fp);
940 }
941 return false;
942 }
943
944 // Locate initial thread stack. This special handling of initial thread stack
945 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
946 // bogus value for initial thread.
947 void os::Linux::capture_initial_stack(size_t max_size) {
948 // stack size is the easy part, get it from RLIMIT_STACK
949 size_t stack_size;
950 struct rlimit rlim;
951 getrlimit(RLIMIT_STACK, &rlim);
952 stack_size = rlim.rlim_cur;
953
954 // 6308388: a bug in ld.so will relocate its own .data section to the
955 // lower end of primordial stack; reduce ulimit -s value a little bit
956 // so we won't install guard page on ld.so's data section.
957 stack_size -= 2 * page_size();
958
959 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
960 // 7.1, in both cases we will get 2G in return value.
961 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
962 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
963 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
964 // in case other parts in glibc still assumes 2M max stack size.
965 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
966 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
967 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
968 stack_size = 2 * K * K IA64_ONLY(*2);
969 }
970 // Try to figure out where the stack base (top) is. This is harder.
971 //
972 // When an application is started, glibc saves the initial stack pointer in
973 // a global variable "__libc_stack_end", which is then used by system
974 // libraries. __libc_stack_end should be pretty close to stack top. The
975 // variable is available since the very early days. However, because it is
976 // a private interface, it could disappear in the future.
977 //
978 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
979 // to __libc_stack_end, it is very close to stack top, but isn't the real
980 // stack top. Note that /proc may not exist if VM is running as a chroot
981 // program, so reading /proc/<pid>/stat could fail. Also the contents of
982 // /proc/<pid>/stat could change in the future (though unlikely).
983 //
984 // We try __libc_stack_end first. If that doesn't work, look for
985 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
986 // as a hint, which should work well in most cases.
987
988 uintptr_t stack_start;
989
990 // try __libc_stack_end first
991 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
992 if (p && *p) {
993 stack_start = *p;
994 } else {
995 // see if we can get the start_stack field from /proc/self/stat
996 FILE *fp;
997 int pid;
998 char state;
999 int ppid;
1000 int pgrp;
1001 int session;
1002 int nr;
1003 int tpgrp;
1004 unsigned long flags;
1005 unsigned long minflt;
1006 unsigned long cminflt;
1007 unsigned long majflt;
1008 unsigned long cmajflt;
1009 unsigned long utime;
1010 unsigned long stime;
1011 long cutime;
1012 long cstime;
1013 long prio;
1014 long nice;
1015 long junk;
1016 long it_real;
1017 uintptr_t start;
1018 uintptr_t vsize;
1019 intptr_t rss;
1020 uintptr_t rsslim;
1021 uintptr_t scodes;
1022 uintptr_t ecode;
1023 int i;
1024
1025 // Figure what the primordial thread stack base is. Code is inspired
1026 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1027 // followed by command name surrounded by parentheses, state, etc.
1028 char stat[2048];
1029 int statlen;
1030
1031 fp = fopen("/proc/self/stat", "r");
1032 if (fp) {
1033 statlen = fread(stat, 1, 2047, fp);
1034 stat[statlen] = '\0';
1035 fclose(fp);
1036
1037 // Skip pid and the command string. Note that we could be dealing with
1038 // weird command names, e.g. user could decide to rename java launcher
1039 // to "java 1.4.2 :)", then the stat file would look like
1040 // 1234 (java 1.4.2 :)) R ... ...
1041 // We don't really need to know the command string, just find the last
1042 // occurrence of ")" and then start parsing from there. See bug 4726580.
1043 char * s = strrchr(stat, ')');
1044
1045 i = 0;
1046 if (s) {
1047 // Skip blank chars
1048 do { s++; } while (s && isspace(*s));
1049
1050 #define _UFM UINTX_FORMAT
1051 #define _DFM INTX_FORMAT
1052
1053 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1054 // 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
1055 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,
1056 &state, // 3 %c
1057 &ppid, // 4 %d
1058 &pgrp, // 5 %d
1059 &session, // 6 %d
1060 &nr, // 7 %d
1061 &tpgrp, // 8 %d
1062 &flags, // 9 %lu
1063 &minflt, // 10 %lu
1064 &cminflt, // 11 %lu
1065 &majflt, // 12 %lu
1066 &cmajflt, // 13 %lu
1067 &utime, // 14 %lu
1068 &stime, // 15 %lu
1069 &cutime, // 16 %ld
1070 &cstime, // 17 %ld
1071 &prio, // 18 %ld
1072 &nice, // 19 %ld
1073 &junk, // 20 %ld
1074 &it_real, // 21 %ld
1075 &start, // 22 UINTX_FORMAT
1076 &vsize, // 23 UINTX_FORMAT
1077 &rss, // 24 INTX_FORMAT
1078 &rsslim, // 25 UINTX_FORMAT
1079 &scodes, // 26 UINTX_FORMAT
1080 &ecode, // 27 UINTX_FORMAT
1081 &stack_start); // 28 UINTX_FORMAT
1082 }
1083
1084 #undef _UFM
1085 #undef _DFM
1086
1087 if (i != 28 - 2) {
1088 assert(false, "Bad conversion from /proc/self/stat");
1089 // product mode - assume we are the initial thread, good luck in the
1090 // embedded case.
1091 warning("Can't detect initial thread stack location - bad conversion");
1092 stack_start = (uintptr_t) &rlim;
1093 }
1094 } else {
1095 // For some reason we can't open /proc/self/stat (for example, running on
1096 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1097 // most cases, so don't abort:
1098 warning("Can't detect initial thread stack location - no /proc/self/stat");
1099 stack_start = (uintptr_t) &rlim;
1100 }
1101 }
1102
1103 // Now we have a pointer (stack_start) very close to the stack top, the
1104 // next thing to do is to figure out the exact location of stack top. We
1105 // can find out the virtual memory area that contains stack_start by
1106 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1107 // and its upper limit is the real stack top. (again, this would fail if
1108 // running inside chroot, because /proc may not exist.)
1109
1110 uintptr_t stack_top;
1111 address low, high;
1112 if (find_vma((address)stack_start, &low, &high)) {
1113 // success, "high" is the true stack top. (ignore "low", because initial
1114 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1115 stack_top = (uintptr_t)high;
1116 } else {
1117 // failed, likely because /proc/self/maps does not exist
1118 warning("Can't detect initial thread stack location - find_vma failed");
1119 // best effort: stack_start is normally within a few pages below the real
1120 // stack top, use it as stack top, and reduce stack size so we won't put
1121 // guard page outside stack.
1122 stack_top = stack_start;
1123 stack_size -= 16 * page_size();
1124 }
1125
1126 // stack_top could be partially down the page so align it
1127 stack_top = align_size_up(stack_top, page_size());
1128
1129 if (max_size && stack_size > max_size) {
1130 _initial_thread_stack_size = max_size;
1131 } else {
1132 _initial_thread_stack_size = stack_size;
1133 }
1134
1135 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1136 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1137 }
1138
1139 ////////////////////////////////////////////////////////////////////////////////
1140 // time support
1141
1142 // Time since start-up in seconds to a fine granularity.
1143 // Used by VMSelfDestructTimer and the MemProfiler.
1144 double os::elapsedTime() {
1145
1146 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1147 }
1148
1149 jlong os::elapsed_counter() {
1150 return javaTimeNanos() - initial_time_count;
1151 }
1152
1153 jlong os::elapsed_frequency() {
1154 return NANOSECS_PER_SEC; // nanosecond resolution
1155 }
1156
1157 bool os::supports_vtime() { return true; }
1158 bool os::enable_vtime() { return false; }
1159 bool os::vtime_enabled() { return false; }
1160
1161 double os::elapsedVTime() {
1162 struct rusage usage;
1163 int retval = getrusage(RUSAGE_THREAD, &usage);
1164 if (retval == 0) {
1165 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);
1166 } else {
1167 // better than nothing, but not much
1168 return elapsedTime();
1169 }
1170 }
1171
1172 jlong os::javaTimeMillis() {
1173 timeval time;
1174 int status = gettimeofday(&time, NULL);
1175 assert(status != -1, "linux error");
1176 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1177 }
1178
1179 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1180 timeval time;
1181 int status = gettimeofday(&time, NULL);
1182 assert(status != -1, "linux error");
1183 seconds = jlong(time.tv_sec);
1184 nanos = jlong(time.tv_usec) * 1000;
1185 }
1186
1187
1188 #ifndef CLOCK_MONOTONIC
1189 #define CLOCK_MONOTONIC (1)
1190 #endif
1191
1192 void os::Linux::clock_init() {
1193 // we do dlopen's in this particular order due to bug in linux
1194 // dynamical loader (see 6348968) leading to crash on exit
1195 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1196 if (handle == NULL) {
1197 handle = dlopen("librt.so", RTLD_LAZY);
1198 }
1199
1200 if (handle) {
1201 int (*clock_getres_func)(clockid_t, struct timespec*) =
1202 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1203 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1204 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1205 if (clock_getres_func && clock_gettime_func) {
1206 // See if monotonic clock is supported by the kernel. Note that some
1207 // early implementations simply return kernel jiffies (updated every
1208 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1209 // for nano time (though the monotonic property is still nice to have).
1210 // It's fixed in newer kernels, however clock_getres() still returns
1211 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1212 // resolution for now. Hopefully as people move to new kernels, this
1213 // won't be a problem.
1214 struct timespec res;
1215 struct timespec tp;
1216 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1217 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1218 // yes, monotonic clock is supported
1219 _clock_gettime = clock_gettime_func;
1220 return;
1221 } else {
1222 // close librt if there is no monotonic clock
1223 dlclose(handle);
1224 }
1225 }
1226 }
1227 warning("No monotonic clock was available - timed services may " \
1228 "be adversely affected if the time-of-day clock changes");
1229 }
1230
1231 #ifndef SYS_clock_getres
1232 #if defined(IA32) || defined(AMD64)
1233 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1234 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1235 #else
1236 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1237 #define sys_clock_getres(x,y) -1
1238 #endif
1239 #else
1240 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1241 #endif
1242
1243 void os::Linux::fast_thread_clock_init() {
1244 if (!UseLinuxPosixThreadCPUClocks) {
1245 return;
1246 }
1247 clockid_t clockid;
1248 struct timespec tp;
1249 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1250 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1251
1252 // Switch to using fast clocks for thread cpu time if
1253 // the sys_clock_getres() returns 0 error code.
1254 // Note, that some kernels may support the current thread
1255 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1256 // returned by the pthread_getcpuclockid().
1257 // If the fast Posix clocks are supported then the sys_clock_getres()
1258 // must return at least tp.tv_sec == 0 which means a resolution
1259 // better than 1 sec. This is extra check for reliability.
1260
1261 if (pthread_getcpuclockid_func &&
1262 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1263 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1264 _supports_fast_thread_cpu_time = true;
1265 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1266 }
1267 }
1268
1269 jlong os::javaTimeNanos() {
1270 if (os::supports_monotonic_clock()) {
1271 struct timespec tp;
1272 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1273 assert(status == 0, "gettime error");
1274 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1275 return result;
1276 } else {
1277 timeval time;
1278 int status = gettimeofday(&time, NULL);
1279 assert(status != -1, "linux error");
1280 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1281 return 1000 * usecs;
1282 }
1283 }
1284
1285 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1286 if (os::supports_monotonic_clock()) {
1287 info_ptr->max_value = ALL_64_BITS;
1288
1289 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1290 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1291 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1292 } else {
1293 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1294 info_ptr->max_value = ALL_64_BITS;
1295
1296 // gettimeofday is a real time clock so it skips
1297 info_ptr->may_skip_backward = true;
1298 info_ptr->may_skip_forward = true;
1299 }
1300
1301 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1302 }
1303
1304 // Return the real, user, and system times in seconds from an
1305 // arbitrary fixed point in the past.
1306 bool os::getTimesSecs(double* process_real_time,
1307 double* process_user_time,
1308 double* process_system_time) {
1309 struct tms ticks;
1310 clock_t real_ticks = times(&ticks);
1311
1312 if (real_ticks == (clock_t) (-1)) {
1313 return false;
1314 } else {
1315 double ticks_per_second = (double) clock_tics_per_sec;
1316 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1317 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1318 *process_real_time = ((double) real_ticks) / ticks_per_second;
1319
1320 return true;
1321 }
1322 }
1323
1324
1325 char * os::local_time_string(char *buf, size_t buflen) {
1326 struct tm t;
1327 time_t long_time;
1328 time(&long_time);
1329 localtime_r(&long_time, &t);
1330 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1331 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1332 t.tm_hour, t.tm_min, t.tm_sec);
1333 return buf;
1334 }
1335
1336 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1337 return localtime_r(clock, res);
1338 }
1339
1340 ////////////////////////////////////////////////////////////////////////////////
1341 // runtime exit support
1342
1343 // Note: os::shutdown() might be called very early during initialization, or
1344 // called from signal handler. Before adding something to os::shutdown(), make
1345 // sure it is async-safe and can handle partially initialized VM.
1346 void os::shutdown() {
1347
1348 // allow PerfMemory to attempt cleanup of any persistent resources
1349 perfMemory_exit();
1350
1351 // needs to remove object in file system
1352 AttachListener::abort();
1353
1354 // flush buffered output, finish log files
1355 ostream_abort();
1356
1357 // Check for abort hook
1358 abort_hook_t abort_hook = Arguments::abort_hook();
1359 if (abort_hook != NULL) {
1360 abort_hook();
1361 }
1362
1363 }
1364
1365 // Note: os::abort() might be called very early during initialization, or
1366 // called from signal handler. Before adding something to os::abort(), make
1367 // sure it is async-safe and can handle partially initialized VM.
1368 void os::abort(bool dump_core, void* siginfo, const void* context) {
1369 os::shutdown();
1370 if (dump_core) {
1371 #ifndef PRODUCT
1372 fdStream out(defaultStream::output_fd());
1373 out.print_raw("Current thread is ");
1374 char buf[16];
1375 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1376 out.print_raw_cr(buf);
1377 out.print_raw_cr("Dumping core ...");
1378 #endif
1379 ::abort(); // dump core
1380 }
1381
1382 ::exit(1);
1383 }
1384
1385 // Die immediately, no exit hook, no abort hook, no cleanup.
1386 void os::die() {
1387 ::abort();
1388 }
1389
1390
1391 // This method is a copy of JDK's sysGetLastErrorString
1392 // from src/solaris/hpi/src/system_md.c
1393
1394 size_t os::lasterror(char *buf, size_t len) {
1395 if (errno == 0) return 0;
1396
1397 const char *s = ::strerror(errno);
1398 size_t n = ::strlen(s);
1399 if (n >= len) {
1400 n = len - 1;
1401 }
1402 ::strncpy(buf, s, n);
1403 buf[n] = '\0';
1404 return n;
1405 }
1406
1407 // thread_id is kernel thread id (similar to Solaris LWP id)
1408 intx os::current_thread_id() { return os::Linux::gettid(); }
1409 int os::current_process_id() {
1410 return ::getpid();
1411 }
1412
1413 // DLL functions
1414
1415 const char* os::dll_file_extension() { return ".so"; }
1416
1417 // This must be hard coded because it's the system's temporary
1418 // directory not the java application's temp directory, ala java.io.tmpdir.
1419 const char* os::get_temp_directory() { return "/tmp"; }
1420
1421 static bool file_exists(const char* filename) {
1422 struct stat statbuf;
1423 if (filename == NULL || strlen(filename) == 0) {
1424 return false;
1425 }
1426 return os::stat(filename, &statbuf) == 0;
1427 }
1428
1429 bool os::dll_build_name(char* buffer, size_t buflen,
1430 const char* pname, const char* fname) {
1431 bool retval = false;
1432 // Copied from libhpi
1433 const size_t pnamelen = pname ? strlen(pname) : 0;
1434
1435 // Return error on buffer overflow.
1436 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1437 return retval;
1438 }
1439
1440 if (pnamelen == 0) {
1441 snprintf(buffer, buflen, "lib%s.so", fname);
1442 retval = true;
1443 } else if (strchr(pname, *os::path_separator()) != NULL) {
1444 int n;
1445 char** pelements = split_path(pname, &n);
1446 if (pelements == NULL) {
1447 return false;
1448 }
1449 for (int i = 0; i < n; i++) {
1450 // Really shouldn't be NULL, but check can't hurt
1451 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1452 continue; // skip the empty path values
1453 }
1454 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1455 if (file_exists(buffer)) {
1456 retval = true;
1457 break;
1458 }
1459 }
1460 // release the storage
1461 for (int i = 0; i < n; i++) {
1462 if (pelements[i] != NULL) {
1463 FREE_C_HEAP_ARRAY(char, pelements[i]);
1464 }
1465 }
1466 if (pelements != NULL) {
1467 FREE_C_HEAP_ARRAY(char*, pelements);
1468 }
1469 } else {
1470 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1471 retval = true;
1472 }
1473 return retval;
1474 }
1475
1476 // check if addr is inside libjvm.so
1477 bool os::address_is_in_vm(address addr) {
1478 static address libjvm_base_addr;
1479 Dl_info dlinfo;
1480
1481 if (libjvm_base_addr == NULL) {
1482 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1483 libjvm_base_addr = (address)dlinfo.dli_fbase;
1484 }
1485 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1486 }
1487
1488 if (dladdr((void *)addr, &dlinfo) != 0) {
1489 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1490 }
1491
1492 return false;
1493 }
1494
1495 bool os::dll_address_to_function_name(address addr, char *buf,
1496 int buflen, int *offset,
1497 bool demangle) {
1498 // buf is not optional, but offset is optional
1499 assert(buf != NULL, "sanity check");
1500
1501 Dl_info dlinfo;
1502
1503 if (dladdr((void*)addr, &dlinfo) != 0) {
1504 // see if we have a matching symbol
1505 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1506 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1507 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1508 }
1509 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1510 return true;
1511 }
1512 // no matching symbol so try for just file info
1513 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1514 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1515 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1516 return true;
1517 }
1518 }
1519 }
1520
1521 buf[0] = '\0';
1522 if (offset != NULL) *offset = -1;
1523 return false;
1524 }
1525
1526 struct _address_to_library_name {
1527 address addr; // input : memory address
1528 size_t buflen; // size of fname
1529 char* fname; // output: library name
1530 address base; // library base addr
1531 };
1532
1533 static int address_to_library_name_callback(struct dl_phdr_info *info,
1534 size_t size, void *data) {
1535 int i;
1536 bool found = false;
1537 address libbase = NULL;
1538 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1539
1540 // iterate through all loadable segments
1541 for (i = 0; i < info->dlpi_phnum; i++) {
1542 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1543 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1544 // base address of a library is the lowest address of its loaded
1545 // segments.
1546 if (libbase == NULL || libbase > segbase) {
1547 libbase = segbase;
1548 }
1549 // see if 'addr' is within current segment
1550 if (segbase <= d->addr &&
1551 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1552 found = true;
1553 }
1554 }
1555 }
1556
1557 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1558 // so dll_address_to_library_name() can fall through to use dladdr() which
1559 // can figure out executable name from argv[0].
1560 if (found && info->dlpi_name && info->dlpi_name[0]) {
1561 d->base = libbase;
1562 if (d->fname) {
1563 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1564 }
1565 return 1;
1566 }
1567 return 0;
1568 }
1569
1570 bool os::dll_address_to_library_name(address addr, char* buf,
1571 int buflen, int* offset) {
1572 // buf is not optional, but offset is optional
1573 assert(buf != NULL, "sanity check");
1574
1575 Dl_info dlinfo;
1576 struct _address_to_library_name data;
1577
1578 // There is a bug in old glibc dladdr() implementation that it could resolve
1579 // to wrong library name if the .so file has a base address != NULL. Here
1580 // we iterate through the program headers of all loaded libraries to find
1581 // out which library 'addr' really belongs to. This workaround can be
1582 // removed once the minimum requirement for glibc is moved to 2.3.x.
1583 data.addr = addr;
1584 data.fname = buf;
1585 data.buflen = buflen;
1586 data.base = NULL;
1587 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1588
1589 if (rslt) {
1590 // buf already contains library name
1591 if (offset) *offset = addr - data.base;
1592 return true;
1593 }
1594 if (dladdr((void*)addr, &dlinfo) != 0) {
1595 if (dlinfo.dli_fname != NULL) {
1596 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1597 }
1598 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1599 *offset = addr - (address)dlinfo.dli_fbase;
1600 }
1601 return true;
1602 }
1603
1604 buf[0] = '\0';
1605 if (offset) *offset = -1;
1606 return false;
1607 }
1608
1609 // Loads .dll/.so and
1610 // in case of error it checks if .dll/.so was built for the
1611 // same architecture as Hotspot is running on
1612
1613
1614 // Remember the stack's state. The Linux dynamic linker will change
1615 // the stack to 'executable' at most once, so we must safepoint only once.
1616 bool os::Linux::_stack_is_executable = false;
1617
1618 // VM operation that loads a library. This is necessary if stack protection
1619 // of the Java stacks can be lost during loading the library. If we
1620 // do not stop the Java threads, they can stack overflow before the stacks
1621 // are protected again.
1622 class VM_LinuxDllLoad: public VM_Operation {
1623 private:
1624 const char *_filename;
1625 char *_ebuf;
1626 int _ebuflen;
1627 void *_lib;
1628 public:
1629 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1630 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1631 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1632 void doit() {
1633 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1634 os::Linux::_stack_is_executable = true;
1635 }
1636 void* loaded_library() { return _lib; }
1637 };
1638
1639 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1640 void * result = NULL;
1641 bool load_attempted = false;
1642
1643 // Check whether the library to load might change execution rights
1644 // of the stack. If they are changed, the protection of the stack
1645 // guard pages will be lost. We need a safepoint to fix this.
1646 //
1647 // See Linux man page execstack(8) for more info.
1648 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1649 ElfFile ef(filename);
1650 if (!ef.specifies_noexecstack()) {
1651 if (!is_init_completed()) {
1652 os::Linux::_stack_is_executable = true;
1653 // This is OK - No Java threads have been created yet, and hence no
1654 // stack guard pages to fix.
1655 //
1656 // This should happen only when you are building JDK7 using a very
1657 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1658 //
1659 // Dynamic loader will make all stacks executable after
1660 // this function returns, and will not do that again.
1661 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1662 } else {
1663 warning("You have loaded library %s which might have disabled stack guard. "
1664 "The VM will try to fix the stack guard now.\n"
1665 "It's highly recommended that you fix the library with "
1666 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1667 filename);
1668
1669 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1670 JavaThread *jt = JavaThread::current();
1671 if (jt->thread_state() != _thread_in_native) {
1672 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1673 // that requires ExecStack. Cannot enter safe point. Let's give up.
1674 warning("Unable to fix stack guard. Giving up.");
1675 } else {
1676 if (!LoadExecStackDllInVMThread) {
1677 // This is for the case where the DLL has an static
1678 // constructor function that executes JNI code. We cannot
1679 // load such DLLs in the VMThread.
1680 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1681 }
1682
1683 ThreadInVMfromNative tiv(jt);
1684 debug_only(VMNativeEntryWrapper vew;)
1685
1686 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1687 VMThread::execute(&op);
1688 if (LoadExecStackDllInVMThread) {
1689 result = op.loaded_library();
1690 }
1691 load_attempted = true;
1692 }
1693 }
1694 }
1695 }
1696
1697 if (!load_attempted) {
1698 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1699 }
1700
1701 if (result != NULL) {
1702 // Successful loading
1703 return result;
1704 }
1705
1706 Elf32_Ehdr elf_head;
1707 int diag_msg_max_length=ebuflen-strlen(ebuf);
1708 char* diag_msg_buf=ebuf+strlen(ebuf);
1709
1710 if (diag_msg_max_length==0) {
1711 // No more space in ebuf for additional diagnostics message
1712 return NULL;
1713 }
1714
1715
1716 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1717
1718 if (file_descriptor < 0) {
1719 // Can't open library, report dlerror() message
1720 return NULL;
1721 }
1722
1723 bool failed_to_read_elf_head=
1724 (sizeof(elf_head)!=
1725 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1726
1727 ::close(file_descriptor);
1728 if (failed_to_read_elf_head) {
1729 // file i/o error - report dlerror() msg
1730 return NULL;
1731 }
1732
1733 typedef struct {
1734 Elf32_Half code; // Actual value as defined in elf.h
1735 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1736 char elf_class; // 32 or 64 bit
1737 char endianess; // MSB or LSB
1738 char* name; // String representation
1739 } arch_t;
1740
1741 #ifndef EM_486
1742 #define EM_486 6 /* Intel 80486 */
1743 #endif
1744 #ifndef EM_AARCH64
1745 #define EM_AARCH64 183 /* ARM AARCH64 */
1746 #endif
1747
1748 static const arch_t arch_array[]={
1749 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1750 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1751 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1752 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1753 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1754 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1755 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1756 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1757 #if defined(VM_LITTLE_ENDIAN)
1758 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1759 #else
1760 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1761 #endif
1762 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1763 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1764 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1765 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1766 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1767 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1768 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1769 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1770 };
1771
1772 #if (defined IA32)
1773 static Elf32_Half running_arch_code=EM_386;
1774 #elif (defined AMD64)
1775 static Elf32_Half running_arch_code=EM_X86_64;
1776 #elif (defined IA64)
1777 static Elf32_Half running_arch_code=EM_IA_64;
1778 #elif (defined __sparc) && (defined _LP64)
1779 static Elf32_Half running_arch_code=EM_SPARCV9;
1780 #elif (defined __sparc) && (!defined _LP64)
1781 static Elf32_Half running_arch_code=EM_SPARC;
1782 #elif (defined __powerpc64__)
1783 static Elf32_Half running_arch_code=EM_PPC64;
1784 #elif (defined __powerpc__)
1785 static Elf32_Half running_arch_code=EM_PPC;
1786 #elif (defined ARM)
1787 static Elf32_Half running_arch_code=EM_ARM;
1788 #elif (defined S390)
1789 static Elf32_Half running_arch_code=EM_S390;
1790 #elif (defined ALPHA)
1791 static Elf32_Half running_arch_code=EM_ALPHA;
1792 #elif (defined MIPSEL)
1793 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1794 #elif (defined PARISC)
1795 static Elf32_Half running_arch_code=EM_PARISC;
1796 #elif (defined MIPS)
1797 static Elf32_Half running_arch_code=EM_MIPS;
1798 #elif (defined M68K)
1799 static Elf32_Half running_arch_code=EM_68K;
1800 #elif (defined AARCH64)
1801 static Elf32_Half running_arch_code=EM_AARCH64;
1802 #else
1803 #error Method os::dll_load requires that one of following is defined:\
1804 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1805 #endif
1806
1807 // Identify compatability class for VM's architecture and library's architecture
1808 // Obtain string descriptions for architectures
1809
1810 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1811 int running_arch_index=-1;
1812
1813 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1814 if (running_arch_code == arch_array[i].code) {
1815 running_arch_index = i;
1816 }
1817 if (lib_arch.code == arch_array[i].code) {
1818 lib_arch.compat_class = arch_array[i].compat_class;
1819 lib_arch.name = arch_array[i].name;
1820 }
1821 }
1822
1823 assert(running_arch_index != -1,
1824 "Didn't find running architecture code (running_arch_code) in arch_array");
1825 if (running_arch_index == -1) {
1826 // Even though running architecture detection failed
1827 // we may still continue with reporting dlerror() message
1828 return NULL;
1829 }
1830
1831 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1832 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1833 return NULL;
1834 }
1835
1836 #ifndef S390
1837 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1839 return NULL;
1840 }
1841 #endif // !S390
1842
1843 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1844 if (lib_arch.name!=NULL) {
1845 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1846 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1847 lib_arch.name, arch_array[running_arch_index].name);
1848 } else {
1849 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1850 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1851 lib_arch.code,
1852 arch_array[running_arch_index].name);
1853 }
1854 }
1855
1856 return NULL;
1857 }
1858
1859 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1860 int ebuflen) {
1861 void * result = ::dlopen(filename, RTLD_LAZY);
1862 if (result == NULL) {
1863 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1864 ebuf[ebuflen-1] = '\0';
1865 }
1866 return result;
1867 }
1868
1869 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1870 int ebuflen) {
1871 void * result = NULL;
1872 if (LoadExecStackDllInVMThread) {
1873 result = dlopen_helper(filename, ebuf, ebuflen);
1874 }
1875
1876 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1877 // library that requires an executable stack, or which does not have this
1878 // stack attribute set, dlopen changes the stack attribute to executable. The
1879 // read protection of the guard pages gets lost.
1880 //
1881 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1882 // may have been queued at the same time.
1883
1884 if (!_stack_is_executable) {
1885 JavaThread *jt = Threads::first();
1886
1887 while (jt) {
1888 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1889 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1890 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1891 warning("Attempt to reguard stack yellow zone failed.");
1892 }
1893 }
1894 jt = jt->next();
1895 }
1896 }
1897
1898 return result;
1899 }
1900
1901 void* os::dll_lookup(void* handle, const char* name) {
1902 void* res = dlsym(handle, name);
1903 return res;
1904 }
1905
1906 void* os::get_default_process_handle() {
1907 return (void*)::dlopen(NULL, RTLD_LAZY);
1908 }
1909
1910 static bool _print_ascii_file(const char* filename, outputStream* st) {
1911 int fd = ::open(filename, O_RDONLY);
1912 if (fd == -1) {
1913 return false;
1914 }
1915
1916 char buf[32];
1917 int bytes;
1918 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1919 st->print_raw(buf, bytes);
1920 }
1921
1922 ::close(fd);
1923
1924 return true;
1925 }
1926
1927 void os::print_dll_info(outputStream *st) {
1928 st->print_cr("Dynamic libraries:");
1929
1930 char fname[32];
1931 pid_t pid = os::Linux::gettid();
1932
1933 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1934
1935 if (!_print_ascii_file(fname, st)) {
1936 st->print("Can not get library information for pid = %d\n", pid);
1937 }
1938 }
1939
1940 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1941 FILE *procmapsFile = NULL;
1942
1943 // Open the procfs maps file for the current process
1944 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1945 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1946 char line[PATH_MAX + 100];
1947
1948 // Read line by line from 'file'
1949 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1950 u8 base, top, offset, inode;
1951 char permissions[5];
1952 char device[6];
1953 char name[PATH_MAX + 1];
1954
1955 // Parse fields from line
1956 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1957 &base, &top, permissions, &offset, device, &inode, name);
1958
1959 // Filter by device id '00:00' so that we only get file system mapped files.
1960 if (strcmp(device, "00:00") != 0) {
1961
1962 // Call callback with the fields of interest
1963 if(callback(name, (address)base, (address)top, param)) {
1964 // Oops abort, callback aborted
1965 fclose(procmapsFile);
1966 return 1;
1967 }
1968 }
1969 }
1970 fclose(procmapsFile);
1971 }
1972 return 0;
1973 }
1974
1975 void os::print_os_info_brief(outputStream* st) {
1976 os::Linux::print_distro_info(st);
1977
1978 os::Posix::print_uname_info(st);
1979
1980 os::Linux::print_libversion_info(st);
1981
1982 }
1983
1984 void os::print_os_info(outputStream* st) {
1985 st->print("OS:");
1986
1987 os::Linux::print_distro_info(st);
1988
1989 os::Posix::print_uname_info(st);
1990
1991 // Print warning if unsafe chroot environment detected
1992 if (unsafe_chroot_detected) {
1993 st->print("WARNING!! ");
1994 st->print_cr("%s", unstable_chroot_error);
1995 }
1996
1997 os::Linux::print_libversion_info(st);
1998
1999 os::Posix::print_rlimit_info(st);
2000
2001 os::Posix::print_load_average(st);
2002
2003 os::Linux::print_full_memory_info(st);
2004 }
2005
2006 // Try to identify popular distros.
2007 // Most Linux distributions have a /etc/XXX-release file, which contains
2008 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2009 // file that also contains the OS version string. Some have more than one
2010 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2011 // /etc/redhat-release.), so the order is important.
2012 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2013 // their own specific XXX-release file as well as a redhat-release file.
2014 // Because of this the XXX-release file needs to be searched for before the
2015 // redhat-release file.
2016 // Since Red Hat has a lsb-release file that is not very descriptive the
2017 // search for redhat-release needs to be before lsb-release.
2018 // Since the lsb-release file is the new standard it needs to be searched
2019 // before the older style release files.
2020 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2021 // next to last resort. The os-release file is a new standard that contains
2022 // distribution information and the system-release file seems to be an old
2023 // standard that has been replaced by the lsb-release and os-release files.
2024 // Searching for the debian_version file is the last resort. It contains
2025 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2026 // "Debian " is printed before the contents of the debian_version file.
2027
2028 const char* distro_files[] = {
2029 "/etc/oracle-release",
2030 "/etc/mandriva-release",
2031 "/etc/mandrake-release",
2032 "/etc/sun-release",
2033 "/etc/redhat-release",
2034 "/etc/lsb-release",
2035 "/etc/SuSE-release",
2036 "/etc/turbolinux-release",
2037 "/etc/gentoo-release",
2038 "/etc/ltib-release",
2039 "/etc/angstrom-version",
2040 "/etc/system-release",
2041 "/etc/os-release",
2042 NULL };
2043
2044 void os::Linux::print_distro_info(outputStream* st) {
2045 for (int i = 0;; i++) {
2046 const char* file = distro_files[i];
2047 if (file == NULL) {
2048 break; // done
2049 }
2050 // If file prints, we found it.
2051 if (_print_ascii_file(file, st)) {
2052 return;
2053 }
2054 }
2055
2056 if (file_exists("/etc/debian_version")) {
2057 st->print("Debian ");
2058 _print_ascii_file("/etc/debian_version", st);
2059 } else {
2060 st->print("Linux");
2061 }
2062 st->cr();
2063 }
2064
2065 static void parse_os_info(char* distro, size_t length, const char* file) {
2066 FILE* fp = fopen(file, "r");
2067 if (fp != NULL) {
2068 char buf[256];
2069 // get last line of the file.
2070 while (fgets(buf, sizeof(buf), fp)) { }
2071 // Edit out extra stuff in expected ubuntu format
2072 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2073 char* ptr = strstr(buf, "\""); // the name is in quotes
2074 if (ptr != NULL) {
2075 ptr++; // go beyond first quote
2076 char* nl = strchr(ptr, '\"');
2077 if (nl != NULL) *nl = '\0';
2078 strncpy(distro, ptr, length);
2079 } else {
2080 ptr = strstr(buf, "=");
2081 ptr++; // go beyond equals then
2082 char* nl = strchr(ptr, '\n');
2083 if (nl != NULL) *nl = '\0';
2084 strncpy(distro, ptr, length);
2085 }
2086 } else {
2087 // if not in expected Ubuntu format, print out whole line minus \n
2088 char* nl = strchr(buf, '\n');
2089 if (nl != NULL) *nl = '\0';
2090 strncpy(distro, buf, length);
2091 }
2092 // close distro file
2093 fclose(fp);
2094 }
2095 }
2096
2097 void os::get_summary_os_info(char* buf, size_t buflen) {
2098 for (int i = 0;; i++) {
2099 const char* file = distro_files[i];
2100 if (file == NULL) {
2101 break; // ran out of distro_files
2102 }
2103 if (file_exists(file)) {
2104 parse_os_info(buf, buflen, file);
2105 return;
2106 }
2107 }
2108 // special case for debian
2109 if (file_exists("/etc/debian_version")) {
2110 strncpy(buf, "Debian ", buflen);
2111 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2112 } else {
2113 strncpy(buf, "Linux", buflen);
2114 }
2115 }
2116
2117 void os::Linux::print_libversion_info(outputStream* st) {
2118 // libc, pthread
2119 st->print("libc:");
2120 st->print("%s ", os::Linux::glibc_version());
2121 st->print("%s ", os::Linux::libpthread_version());
2122 st->cr();
2123 }
2124
2125 void os::Linux::print_full_memory_info(outputStream* st) {
2126 st->print("\n/proc/meminfo:\n");
2127 _print_ascii_file("/proc/meminfo", st);
2128 st->cr();
2129 }
2130
2131 void os::print_memory_info(outputStream* st) {
2132
2133 st->print("Memory:");
2134 st->print(" %dk page", os::vm_page_size()>>10);
2135
2136 // values in struct sysinfo are "unsigned long"
2137 struct sysinfo si;
2138 sysinfo(&si);
2139
2140 st->print(", physical " UINT64_FORMAT "k",
2141 os::physical_memory() >> 10);
2142 st->print("(" UINT64_FORMAT "k free)",
2143 os::available_memory() >> 10);
2144 st->print(", swap " UINT64_FORMAT "k",
2145 ((jlong)si.totalswap * si.mem_unit) >> 10);
2146 st->print("(" UINT64_FORMAT "k free)",
2147 ((jlong)si.freeswap * si.mem_unit) >> 10);
2148 st->cr();
2149 }
2150
2151 // Print the first "model name" line and the first "flags" line
2152 // that we find and nothing more. We assume "model name" comes
2153 // before "flags" so if we find a second "model name", then the
2154 // "flags" field is considered missing.
2155 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2156 #if defined(IA32) || defined(AMD64)
2157 // Other platforms have less repetitive cpuinfo files
2158 FILE *fp = fopen("/proc/cpuinfo", "r");
2159 if (fp) {
2160 while (!feof(fp)) {
2161 if (fgets(buf, buflen, fp)) {
2162 // Assume model name comes before flags
2163 bool model_name_printed = false;
2164 if (strstr(buf, "model name") != NULL) {
2165 if (!model_name_printed) {
2166 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2167 st->print_raw(buf);
2168 model_name_printed = true;
2169 } else {
2170 // model name printed but not flags? Odd, just return
2171 fclose(fp);
2172 return true;
2173 }
2174 }
2175 // print the flags line too
2176 if (strstr(buf, "flags") != NULL) {
2177 st->print_raw(buf);
2178 fclose(fp);
2179 return true;
2180 }
2181 }
2182 }
2183 fclose(fp);
2184 }
2185 #endif // x86 platforms
2186 return false;
2187 }
2188
2189 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2190 // Only print the model name if the platform provides this as a summary
2191 if (!print_model_name_and_flags(st, buf, buflen)) {
2192 st->print("\n/proc/cpuinfo:\n");
2193 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2194 st->print_cr(" <Not Available>");
2195 }
2196 }
2197 }
2198
2199 #if defined(AMD64) || defined(IA32) || defined(X32)
2200 const char* search_string = "model name";
2201 #elif defined(SPARC)
2202 const char* search_string = "cpu";
2203 #elif defined(PPC64)
2204 const char* search_string = "cpu";
2205 #else
2206 const char* search_string = "Processor";
2207 #endif
2208
2209 // Parses the cpuinfo file for string representing the model name.
2210 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2211 FILE* fp = fopen("/proc/cpuinfo", "r");
2212 if (fp != NULL) {
2213 while (!feof(fp)) {
2214 char buf[256];
2215 if (fgets(buf, sizeof(buf), fp)) {
2216 char* start = strstr(buf, search_string);
2217 if (start != NULL) {
2218 char *ptr = start + strlen(search_string);
2219 char *end = buf + strlen(buf);
2220 while (ptr != end) {
2221 // skip whitespace and colon for the rest of the name.
2222 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2223 break;
2224 }
2225 ptr++;
2226 }
2227 if (ptr != end) {
2228 // reasonable string, get rid of newline and keep the rest
2229 char* nl = strchr(buf, '\n');
2230 if (nl != NULL) *nl = '\0';
2231 strncpy(cpuinfo, ptr, length);
2232 fclose(fp);
2233 return;
2234 }
2235 }
2236 }
2237 }
2238 fclose(fp);
2239 }
2240 // cpuinfo not found or parsing failed, just print generic string. The entire
2241 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2242 #if defined(AMD64)
2243 strncpy(cpuinfo, "x86_64", length);
2244 #elif defined(IA32)
2245 strncpy(cpuinfo, "x86_32", length);
2246 #elif defined(IA64)
2247 strncpy(cpuinfo, "IA64", length);
2248 #elif defined(SPARC)
2249 strncpy(cpuinfo, "sparcv9", length);
2250 #elif defined(AARCH64)
2251 strncpy(cpuinfo, "AArch64", length);
2252 #elif defined(ARM)
2253 strncpy(cpuinfo, "ARM", length);
2254 #elif defined(PPC)
2255 strncpy(cpuinfo, "PPC64", length);
2256 #elif defined(ZERO_LIBARCH)
2257 strncpy(cpuinfo, ZERO_LIBARCH, length);
2258 #else
2259 strncpy(cpuinfo, "unknown", length);
2260 #endif
2261 }
2262
2263 static void print_signal_handler(outputStream* st, int sig,
2264 char* buf, size_t buflen);
2265
2266 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2267 st->print_cr("Signal Handlers:");
2268 print_signal_handler(st, SIGSEGV, buf, buflen);
2269 print_signal_handler(st, SIGBUS , buf, buflen);
2270 print_signal_handler(st, SIGFPE , buf, buflen);
2271 print_signal_handler(st, SIGPIPE, buf, buflen);
2272 print_signal_handler(st, SIGXFSZ, buf, buflen);
2273 print_signal_handler(st, SIGILL , buf, buflen);
2274 print_signal_handler(st, SR_signum, buf, buflen);
2275 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2276 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2277 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2278 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2279 #if defined(PPC64)
2280 print_signal_handler(st, SIGTRAP, buf, buflen);
2281 #endif
2282 }
2283
2284 static char saved_jvm_path[MAXPATHLEN] = {0};
2285
2286 // Find the full path to the current module, libjvm.so
2287 void os::jvm_path(char *buf, jint buflen) {
2288 // Error checking.
2289 if (buflen < MAXPATHLEN) {
2290 assert(false, "must use a large-enough buffer");
2291 buf[0] = '\0';
2292 return;
2293 }
2294 // Lazy resolve the path to current module.
2295 if (saved_jvm_path[0] != 0) {
2296 strcpy(buf, saved_jvm_path);
2297 return;
2298 }
2299
2300 char dli_fname[MAXPATHLEN];
2301 bool ret = dll_address_to_library_name(
2302 CAST_FROM_FN_PTR(address, os::jvm_path),
2303 dli_fname, sizeof(dli_fname), NULL);
2304 assert(ret, "cannot locate libjvm");
2305 char *rp = NULL;
2306 if (ret && dli_fname[0] != '\0') {
2307 rp = realpath(dli_fname, buf);
2308 }
2309 if (rp == NULL) {
2310 return;
2311 }
2312
2313 if (Arguments::sun_java_launcher_is_altjvm()) {
2314 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2315 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2316 // If "/jre/lib/" appears at the right place in the string, then
2317 // assume we are installed in a JDK and we're done. Otherwise, check
2318 // for a JAVA_HOME environment variable and fix up the path so it
2319 // looks like libjvm.so is installed there (append a fake suffix
2320 // hotspot/libjvm.so).
2321 const char *p = buf + strlen(buf) - 1;
2322 for (int count = 0; p > buf && count < 5; ++count) {
2323 for (--p; p > buf && *p != '/'; --p)
2324 /* empty */ ;
2325 }
2326
2327 if (strncmp(p, "/jre/lib/", 9) != 0) {
2328 // Look for JAVA_HOME in the environment.
2329 char* java_home_var = ::getenv("JAVA_HOME");
2330 if (java_home_var != NULL && java_home_var[0] != 0) {
2331 char* jrelib_p;
2332 int len;
2333
2334 // Check the current module name "libjvm.so".
2335 p = strrchr(buf, '/');
2336 if (p == NULL) {
2337 return;
2338 }
2339 assert(strstr(p, "/libjvm") == p, "invalid library name");
2340
2341 rp = realpath(java_home_var, buf);
2342 if (rp == NULL) {
2343 return;
2344 }
2345
2346 // determine if this is a legacy image or modules image
2347 // modules image doesn't have "jre" subdirectory
2348 len = strlen(buf);
2349 assert(len < buflen, "Ran out of buffer room");
2350 jrelib_p = buf + len;
2351 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2352 if (0 != access(buf, F_OK)) {
2353 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2354 }
2355
2356 if (0 == access(buf, F_OK)) {
2357 // Use current module name "libjvm.so"
2358 len = strlen(buf);
2359 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2360 } else {
2361 // Go back to path of .so
2362 rp = realpath(dli_fname, buf);
2363 if (rp == NULL) {
2364 return;
2365 }
2366 }
2367 }
2368 }
2369 }
2370
2371 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2372 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2373 }
2374
2375 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2376 // no prefix required, not even "_"
2377 }
2378
2379 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2380 // no suffix required
2381 }
2382
2383 ////////////////////////////////////////////////////////////////////////////////
2384 // sun.misc.Signal support
2385
2386 static volatile jint sigint_count = 0;
2387
2388 static void UserHandler(int sig, void *siginfo, void *context) {
2389 // 4511530 - sem_post is serialized and handled by the manager thread. When
2390 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2391 // don't want to flood the manager thread with sem_post requests.
2392 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2393 return;
2394 }
2395
2396 // Ctrl-C is pressed during error reporting, likely because the error
2397 // handler fails to abort. Let VM die immediately.
2398 if (sig == SIGINT && is_error_reported()) {
2399 os::die();
2400 }
2401
2402 os::signal_notify(sig);
2403 }
2404
2405 void* os::user_handler() {
2406 return CAST_FROM_FN_PTR(void*, UserHandler);
2407 }
2408
2409 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2410 struct timespec ts;
2411 // Semaphore's are always associated with CLOCK_REALTIME
2412 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2413 // see unpackTime for discussion on overflow checking
2414 if (sec >= MAX_SECS) {
2415 ts.tv_sec += MAX_SECS;
2416 ts.tv_nsec = 0;
2417 } else {
2418 ts.tv_sec += sec;
2419 ts.tv_nsec += nsec;
2420 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2421 ts.tv_nsec -= NANOSECS_PER_SEC;
2422 ++ts.tv_sec; // note: this must be <= max_secs
2423 }
2424 }
2425
2426 return ts;
2427 }
2428
2429 extern "C" {
2430 typedef void (*sa_handler_t)(int);
2431 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2432 }
2433
2434 void* os::signal(int signal_number, void* handler) {
2435 struct sigaction sigAct, oldSigAct;
2436
2437 sigfillset(&(sigAct.sa_mask));
2438 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2439 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2440
2441 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2442 // -1 means registration failed
2443 return (void *)-1;
2444 }
2445
2446 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2447 }
2448
2449 void os::signal_raise(int signal_number) {
2450 ::raise(signal_number);
2451 }
2452
2453 // The following code is moved from os.cpp for making this
2454 // code platform specific, which it is by its very nature.
2455
2456 // Will be modified when max signal is changed to be dynamic
2457 int os::sigexitnum_pd() {
2458 return NSIG;
2459 }
2460
2461 // a counter for each possible signal value
2462 static volatile jint pending_signals[NSIG+1] = { 0 };
2463
2464 // Linux(POSIX) specific hand shaking semaphore.
2465 static sem_t sig_sem;
2466 static PosixSemaphore sr_semaphore;
2467
2468 void os::signal_init_pd() {
2469 // Initialize signal structures
2470 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2471
2472 // Initialize signal semaphore
2473 ::sem_init(&sig_sem, 0, 0);
2474 }
2475
2476 void os::signal_notify(int sig) {
2477 Atomic::inc(&pending_signals[sig]);
2478 ::sem_post(&sig_sem);
2479 }
2480
2481 static int check_pending_signals(bool wait) {
2482 Atomic::store(0, &sigint_count);
2483 for (;;) {
2484 for (int i = 0; i < NSIG + 1; i++) {
2485 jint n = pending_signals[i];
2486 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2487 return i;
2488 }
2489 }
2490 if (!wait) {
2491 return -1;
2492 }
2493 JavaThread *thread = JavaThread::current();
2494 ThreadBlockInVM tbivm(thread);
2495
2496 bool threadIsSuspended;
2497 do {
2498 thread->set_suspend_equivalent();
2499 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2500 ::sem_wait(&sig_sem);
2501
2502 // were we externally suspended while we were waiting?
2503 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2504 if (threadIsSuspended) {
2505 // The semaphore has been incremented, but while we were waiting
2506 // another thread suspended us. We don't want to continue running
2507 // while suspended because that would surprise the thread that
2508 // suspended us.
2509 ::sem_post(&sig_sem);
2510
2511 thread->java_suspend_self();
2512 }
2513 } while (threadIsSuspended);
2514 }
2515 }
2516
2517 int os::signal_lookup() {
2518 return check_pending_signals(false);
2519 }
2520
2521 int os::signal_wait() {
2522 return check_pending_signals(true);
2523 }
2524
2525 ////////////////////////////////////////////////////////////////////////////////
2526 // Virtual Memory
2527
2528 int os::vm_page_size() {
2529 // Seems redundant as all get out
2530 assert(os::Linux::page_size() != -1, "must call os::init");
2531 return os::Linux::page_size();
2532 }
2533
2534 // Solaris allocates memory by pages.
2535 int os::vm_allocation_granularity() {
2536 assert(os::Linux::page_size() != -1, "must call os::init");
2537 return os::Linux::page_size();
2538 }
2539
2540 // Rationale behind this function:
2541 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2542 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2543 // samples for JITted code. Here we create private executable mapping over the code cache
2544 // and then we can use standard (well, almost, as mapping can change) way to provide
2545 // info for the reporting script by storing timestamp and location of symbol
2546 void linux_wrap_code(char* base, size_t size) {
2547 static volatile jint cnt = 0;
2548
2549 if (!UseOprofile) {
2550 return;
2551 }
2552
2553 char buf[PATH_MAX+1];
2554 int num = Atomic::add(1, &cnt);
2555
2556 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2557 os::get_temp_directory(), os::current_process_id(), num);
2558 unlink(buf);
2559
2560 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2561
2562 if (fd != -1) {
2563 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2564 if (rv != (off_t)-1) {
2565 if (::write(fd, "", 1) == 1) {
2566 mmap(base, size,
2567 PROT_READ|PROT_WRITE|PROT_EXEC,
2568 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2569 }
2570 }
2571 ::close(fd);
2572 unlink(buf);
2573 }
2574 }
2575
2576 static bool recoverable_mmap_error(int err) {
2577 // See if the error is one we can let the caller handle. This
2578 // list of errno values comes from JBS-6843484. I can't find a
2579 // Linux man page that documents this specific set of errno
2580 // values so while this list currently matches Solaris, it may
2581 // change as we gain experience with this failure mode.
2582 switch (err) {
2583 case EBADF:
2584 case EINVAL:
2585 case ENOTSUP:
2586 // let the caller deal with these errors
2587 return true;
2588
2589 default:
2590 // Any remaining errors on this OS can cause our reserved mapping
2591 // to be lost. That can cause confusion where different data
2592 // structures think they have the same memory mapped. The worst
2593 // scenario is if both the VM and a library think they have the
2594 // same memory mapped.
2595 return false;
2596 }
2597 }
2598
2599 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2600 int err) {
2601 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2602 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2603 strerror(err), err);
2604 }
2605
2606 static void warn_fail_commit_memory(char* addr, size_t size,
2607 size_t alignment_hint, bool exec,
2608 int err) {
2609 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2610 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2611 alignment_hint, exec, strerror(err), err);
2612 }
2613
2614 // NOTE: Linux kernel does not really reserve the pages for us.
2615 // All it does is to check if there are enough free pages
2616 // left at the time of mmap(). This could be a potential
2617 // problem.
2618 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2619 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2620 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2621 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2622 if (res != (uintptr_t) MAP_FAILED) {
2623 if (UseNUMAInterleaving) {
2624 numa_make_global(addr, size);
2625 }
2626 return 0;
2627 }
2628
2629 int err = errno; // save errno from mmap() call above
2630
2631 if (!recoverable_mmap_error(err)) {
2632 warn_fail_commit_memory(addr, size, exec, err);
2633 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2634 }
2635
2636 return err;
2637 }
2638
2639 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2640 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2641 }
2642
2643 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2644 const char* mesg) {
2645 assert(mesg != NULL, "mesg must be specified");
2646 int err = os::Linux::commit_memory_impl(addr, size, exec);
2647 if (err != 0) {
2648 // the caller wants all commit errors to exit with the specified mesg:
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2651 }
2652 }
2653
2654 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2655 #ifndef MAP_HUGETLB
2656 #define MAP_HUGETLB 0x40000
2657 #endif
2658
2659 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2660 #ifndef MADV_HUGEPAGE
2661 #define MADV_HUGEPAGE 14
2662 #endif
2663
2664 int os::Linux::commit_memory_impl(char* addr, size_t size,
2665 size_t alignment_hint, bool exec) {
2666 int err = os::Linux::commit_memory_impl(addr, size, exec);
2667 if (err == 0) {
2668 realign_memory(addr, size, alignment_hint);
2669 }
2670 return err;
2671 }
2672
2673 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2674 bool exec) {
2675 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2676 }
2677
2678 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2679 size_t alignment_hint, bool exec,
2680 const char* mesg) {
2681 assert(mesg != NULL, "mesg must be specified");
2682 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2683 if (err != 0) {
2684 // the caller wants all commit errors to exit with the specified mesg:
2685 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2686 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2687 }
2688 }
2689
2690 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2691 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2692 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2693 // be supported or the memory may already be backed by huge pages.
2694 ::madvise(addr, bytes, MADV_HUGEPAGE);
2695 }
2696 }
2697
2698 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2699 // This method works by doing an mmap over an existing mmaping and effectively discarding
2700 // the existing pages. However it won't work for SHM-based large pages that cannot be
2701 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2702 // small pages on top of the SHM segment. This method always works for small pages, so we
2703 // allow that in any case.
2704 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2705 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2706 }
2707 }
2708
2709 void os::numa_make_global(char *addr, size_t bytes) {
2710 Linux::numa_interleave_memory(addr, bytes);
2711 }
2712
2713 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2714 // bind policy to MPOL_PREFERRED for the current thread.
2715 #define USE_MPOL_PREFERRED 0
2716
2717 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2718 // To make NUMA and large pages more robust when both enabled, we need to ease
2719 // the requirements on where the memory should be allocated. MPOL_BIND is the
2720 // default policy and it will force memory to be allocated on the specified
2721 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2722 // the specified node, but will not force it. Using this policy will prevent
2723 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2724 // free large pages.
2725 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2726 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2727 }
2728
2729 bool os::numa_topology_changed() { return false; }
2730
2731 size_t os::numa_get_groups_num() {
2732 int max_node = Linux::numa_max_node();
2733 return max_node > 0 ? max_node + 1 : 1;
2734 }
2735
2736 int os::numa_get_group_id() {
2737 int cpu_id = Linux::sched_getcpu();
2738 if (cpu_id != -1) {
2739 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2740 if (lgrp_id != -1) {
2741 return lgrp_id;
2742 }
2743 }
2744 return 0;
2745 }
2746
2747 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2748 for (size_t i = 0; i < size; i++) {
2749 ids[i] = i;
2750 }
2751 return size;
2752 }
2753
2754 bool os::get_page_info(char *start, page_info* info) {
2755 return false;
2756 }
2757
2758 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2759 page_info* page_found) {
2760 return end;
2761 }
2762
2763
2764 int os::Linux::sched_getcpu_syscall(void) {
2765 unsigned int cpu = 0;
2766 int retval = -1;
2767
2768 #if defined(IA32)
2769 #ifndef SYS_getcpu
2770 #define SYS_getcpu 318
2771 #endif
2772 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2773 #elif defined(AMD64)
2774 // Unfortunately we have to bring all these macros here from vsyscall.h
2775 // to be able to compile on old linuxes.
2776 #define __NR_vgetcpu 2
2777 #define VSYSCALL_START (-10UL << 20)
2778 #define VSYSCALL_SIZE 1024
2779 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2780 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2781 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2782 retval = vgetcpu(&cpu, NULL, NULL);
2783 #endif
2784
2785 return (retval == -1) ? retval : cpu;
2786 }
2787
2788 // Something to do with the numa-aware allocator needs these symbols
2789 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2790 extern "C" JNIEXPORT void numa_error(char *where) { }
2791
2792
2793 // If we are running with libnuma version > 2, then we should
2794 // be trying to use symbols with versions 1.1
2795 // If we are running with earlier version, which did not have symbol versions,
2796 // we should use the base version.
2797 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2798 void *f = dlvsym(handle, name, "libnuma_1.1");
2799 if (f == NULL) {
2800 f = dlsym(handle, name);
2801 }
2802 return f;
2803 }
2804
2805 bool os::Linux::libnuma_init() {
2806 // sched_getcpu() should be in libc.
2807 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2808 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2809
2810 // If it's not, try a direct syscall.
2811 if (sched_getcpu() == -1) {
2812 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813 (void*)&sched_getcpu_syscall));
2814 }
2815
2816 if (sched_getcpu() != -1) { // Does it work?
2817 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2818 if (handle != NULL) {
2819 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2820 libnuma_dlsym(handle, "numa_node_to_cpus")));
2821 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2822 libnuma_dlsym(handle, "numa_max_node")));
2823 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2824 libnuma_dlsym(handle, "numa_available")));
2825 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2826 libnuma_dlsym(handle, "numa_tonode_memory")));
2827 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2828 libnuma_dlsym(handle, "numa_interleave_memory")));
2829 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2830 libnuma_dlsym(handle, "numa_set_bind_policy")));
2831
2832
2833 if (numa_available() != -1) {
2834 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2835 // Create a cpu -> node mapping
2836 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2837 rebuild_cpu_to_node_map();
2838 return true;
2839 }
2840 }
2841 }
2842 return false;
2843 }
2844
2845 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2846 // The table is later used in get_node_by_cpu().
2847 void os::Linux::rebuild_cpu_to_node_map() {
2848 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2849 // in libnuma (possible values are starting from 16,
2850 // and continuing up with every other power of 2, but less
2851 // than the maximum number of CPUs supported by kernel), and
2852 // is a subject to change (in libnuma version 2 the requirements
2853 // are more reasonable) we'll just hardcode the number they use
2854 // in the library.
2855 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2856
2857 size_t cpu_num = os::active_processor_count();
2858 size_t cpu_map_size = NCPUS / BitsPerCLong;
2859 size_t cpu_map_valid_size =
2860 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2861
2862 cpu_to_node()->clear();
2863 cpu_to_node()->at_grow(cpu_num - 1);
2864 size_t node_num = numa_get_groups_num();
2865
2866 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2867 for (size_t i = 0; i < node_num; i++) {
2868 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2869 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2870 if (cpu_map[j] != 0) {
2871 for (size_t k = 0; k < BitsPerCLong; k++) {
2872 if (cpu_map[j] & (1UL << k)) {
2873 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2874 }
2875 }
2876 }
2877 }
2878 }
2879 }
2880 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2881 }
2882
2883 int os::Linux::get_node_by_cpu(int cpu_id) {
2884 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2885 return cpu_to_node()->at(cpu_id);
2886 }
2887 return -1;
2888 }
2889
2890 GrowableArray<int>* os::Linux::_cpu_to_node;
2891 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2892 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2893 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2894 os::Linux::numa_available_func_t os::Linux::_numa_available;
2895 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2896 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2897 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2898 unsigned long* os::Linux::_numa_all_nodes;
2899
2900 bool os::pd_uncommit_memory(char* addr, size_t size) {
2901 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2902 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2903 return res != (uintptr_t) MAP_FAILED;
2904 }
2905
2906 static address get_stack_commited_bottom(address bottom, size_t size) {
2907 address nbot = bottom;
2908 address ntop = bottom + size;
2909
2910 size_t page_sz = os::vm_page_size();
2911 unsigned pages = size / page_sz;
2912
2913 unsigned char vec[1];
2914 unsigned imin = 1, imax = pages + 1, imid;
2915 int mincore_return_value = 0;
2916
2917 assert(imin <= imax, "Unexpected page size");
2918
2919 while (imin < imax) {
2920 imid = (imax + imin) / 2;
2921 nbot = ntop - (imid * page_sz);
2922
2923 // Use a trick with mincore to check whether the page is mapped or not.
2924 // mincore sets vec to 1 if page resides in memory and to 0 if page
2925 // is swapped output but if page we are asking for is unmapped
2926 // it returns -1,ENOMEM
2927 mincore_return_value = mincore(nbot, page_sz, vec);
2928
2929 if (mincore_return_value == -1) {
2930 // Page is not mapped go up
2931 // to find first mapped page
2932 if (errno != EAGAIN) {
2933 assert(errno == ENOMEM, "Unexpected mincore errno");
2934 imax = imid;
2935 }
2936 } else {
2937 // Page is mapped go down
2938 // to find first not mapped page
2939 imin = imid + 1;
2940 }
2941 }
2942
2943 nbot = nbot + page_sz;
2944
2945 // Adjust stack bottom one page up if last checked page is not mapped
2946 if (mincore_return_value == -1) {
2947 nbot = nbot + page_sz;
2948 }
2949
2950 return nbot;
2951 }
2952
2953
2954 // Linux uses a growable mapping for the stack, and if the mapping for
2955 // the stack guard pages is not removed when we detach a thread the
2956 // stack cannot grow beyond the pages where the stack guard was
2957 // mapped. If at some point later in the process the stack expands to
2958 // that point, the Linux kernel cannot expand the stack any further
2959 // because the guard pages are in the way, and a segfault occurs.
2960 //
2961 // However, it's essential not to split the stack region by unmapping
2962 // a region (leaving a hole) that's already part of the stack mapping,
2963 // so if the stack mapping has already grown beyond the guard pages at
2964 // the time we create them, we have to truncate the stack mapping.
2965 // So, we need to know the extent of the stack mapping when
2966 // create_stack_guard_pages() is called.
2967
2968 // We only need this for stacks that are growable: at the time of
2969 // writing thread stacks don't use growable mappings (i.e. those
2970 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2971 // only applies to the main thread.
2972
2973 // If the (growable) stack mapping already extends beyond the point
2974 // where we're going to put our guard pages, truncate the mapping at
2975 // that point by munmap()ping it. This ensures that when we later
2976 // munmap() the guard pages we don't leave a hole in the stack
2977 // mapping. This only affects the main/initial thread
2978
2979 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2980 if (os::Linux::is_initial_thread()) {
2981 // As we manually grow stack up to bottom inside create_attached_thread(),
2982 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2983 // we don't need to do anything special.
2984 // Check it first, before calling heavy function.
2985 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2986 unsigned char vec[1];
2987
2988 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2989 // Fallback to slow path on all errors, including EAGAIN
2990 stack_extent = (uintptr_t) get_stack_commited_bottom(
2991 os::Linux::initial_thread_stack_bottom(),
2992 (size_t)addr - stack_extent);
2993 }
2994
2995 if (stack_extent < (uintptr_t)addr) {
2996 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2997 }
2998 }
2999
3000 return os::commit_memory(addr, size, !ExecMem);
3001 }
3002
3003 // If this is a growable mapping, remove the guard pages entirely by
3004 // munmap()ping them. If not, just call uncommit_memory(). This only
3005 // affects the main/initial thread, but guard against future OS changes
3006 // It's safe to always unmap guard pages for initial thread because we
3007 // always place it right after end of the mapped region
3008
3009 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3010 uintptr_t stack_extent, stack_base;
3011
3012 if (os::Linux::is_initial_thread()) {
3013 return ::munmap(addr, size) == 0;
3014 }
3015
3016 return os::uncommit_memory(addr, size);
3017 }
3018
3019 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3020 // at 'requested_addr'. If there are existing memory mappings at the same
3021 // location, however, they will be overwritten. If 'fixed' is false,
3022 // 'requested_addr' is only treated as a hint, the return value may or
3023 // may not start from the requested address. Unlike Linux mmap(), this
3024 // function returns NULL to indicate failure.
3025 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3026 char * addr;
3027 int flags;
3028
3029 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3030 if (fixed) {
3031 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3032 flags |= MAP_FIXED;
3033 }
3034
3035 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3036 // touch an uncommitted page. Otherwise, the read/write might
3037 // succeed if we have enough swap space to back the physical page.
3038 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3039 flags, -1, 0);
3040
3041 return addr == MAP_FAILED ? NULL : addr;
3042 }
3043
3044 static int anon_munmap(char * addr, size_t size) {
3045 return ::munmap(addr, size) == 0;
3046 }
3047
3048 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3049 size_t alignment_hint) {
3050 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3051 }
3052
3053 bool os::pd_release_memory(char* addr, size_t size) {
3054 return anon_munmap(addr, size);
3055 }
3056
3057 static bool linux_mprotect(char* addr, size_t size, int prot) {
3058 // Linux wants the mprotect address argument to be page aligned.
3059 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3060
3061 // According to SUSv3, mprotect() should only be used with mappings
3062 // established by mmap(), and mmap() always maps whole pages. Unaligned
3063 // 'addr' likely indicates problem in the VM (e.g. trying to change
3064 // protection of malloc'ed or statically allocated memory). Check the
3065 // caller if you hit this assert.
3066 assert(addr == bottom, "sanity check");
3067
3068 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3069 return ::mprotect(bottom, size, prot) == 0;
3070 }
3071
3072 // Set protections specified
3073 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3074 bool is_committed) {
3075 unsigned int p = 0;
3076 switch (prot) {
3077 case MEM_PROT_NONE: p = PROT_NONE; break;
3078 case MEM_PROT_READ: p = PROT_READ; break;
3079 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3080 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3081 default:
3082 ShouldNotReachHere();
3083 }
3084 // is_committed is unused.
3085 return linux_mprotect(addr, bytes, p);
3086 }
3087
3088 bool os::guard_memory(char* addr, size_t size) {
3089 return linux_mprotect(addr, size, PROT_NONE);
3090 }
3091
3092 bool os::unguard_memory(char* addr, size_t size) {
3093 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3094 }
3095
3096 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3097 size_t page_size) {
3098 bool result = false;
3099 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3100 MAP_ANONYMOUS|MAP_PRIVATE,
3101 -1, 0);
3102 if (p != MAP_FAILED) {
3103 void *aligned_p = align_ptr_up(p, page_size);
3104
3105 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3106
3107 munmap(p, page_size * 2);
3108 }
3109
3110 if (warn && !result) {
3111 warning("TransparentHugePages is not supported by the operating system.");
3112 }
3113
3114 return result;
3115 }
3116
3117 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3118 bool result = false;
3119 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3120 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3121 -1, 0);
3122
3123 if (p != MAP_FAILED) {
3124 // We don't know if this really is a huge page or not.
3125 FILE *fp = fopen("/proc/self/maps", "r");
3126 if (fp) {
3127 while (!feof(fp)) {
3128 char chars[257];
3129 long x = 0;
3130 if (fgets(chars, sizeof(chars), fp)) {
3131 if (sscanf(chars, "%lx-%*x", &x) == 1
3132 && x == (long)p) {
3133 if (strstr (chars, "hugepage")) {
3134 result = true;
3135 break;
3136 }
3137 }
3138 }
3139 }
3140 fclose(fp);
3141 }
3142 munmap(p, page_size);
3143 }
3144
3145 if (warn && !result) {
3146 warning("HugeTLBFS is not supported by the operating system.");
3147 }
3148
3149 return result;
3150 }
3151
3152 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3153 //
3154 // From the coredump_filter documentation:
3155 //
3156 // - (bit 0) anonymous private memory
3157 // - (bit 1) anonymous shared memory
3158 // - (bit 2) file-backed private memory
3159 // - (bit 3) file-backed shared memory
3160 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3161 // effective only if the bit 2 is cleared)
3162 // - (bit 5) hugetlb private memory
3163 // - (bit 6) hugetlb shared memory
3164 //
3165 static void set_coredump_filter(void) {
3166 FILE *f;
3167 long cdm;
3168
3169 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3170 return;
3171 }
3172
3173 if (fscanf(f, "%lx", &cdm) != 1) {
3174 fclose(f);
3175 return;
3176 }
3177
3178 rewind(f);
3179
3180 if ((cdm & LARGEPAGES_BIT) == 0) {
3181 cdm |= LARGEPAGES_BIT;
3182 fprintf(f, "%#lx", cdm);
3183 }
3184
3185 fclose(f);
3186 }
3187
3188 // Large page support
3189
3190 static size_t _large_page_size = 0;
3191
3192 size_t os::Linux::find_large_page_size() {
3193 size_t large_page_size = 0;
3194
3195 // large_page_size on Linux is used to round up heap size. x86 uses either
3196 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3197 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3198 // page as large as 256M.
3199 //
3200 // Here we try to figure out page size by parsing /proc/meminfo and looking
3201 // for a line with the following format:
3202 // Hugepagesize: 2048 kB
3203 //
3204 // If we can't determine the value (e.g. /proc is not mounted, or the text
3205 // format has been changed), we'll use the largest page size supported by
3206 // the processor.
3207
3208 #ifndef ZERO
3209 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3210 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3211 #endif // ZERO
3212
3213 FILE *fp = fopen("/proc/meminfo", "r");
3214 if (fp) {
3215 while (!feof(fp)) {
3216 int x = 0;
3217 char buf[16];
3218 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3219 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3220 large_page_size = x * K;
3221 break;
3222 }
3223 } else {
3224 // skip to next line
3225 for (;;) {
3226 int ch = fgetc(fp);
3227 if (ch == EOF || ch == (int)'\n') break;
3228 }
3229 }
3230 }
3231 fclose(fp);
3232 }
3233
3234 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3235 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3236 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3237 proper_unit_for_byte_size(large_page_size));
3238 }
3239
3240 return large_page_size;
3241 }
3242
3243 size_t os::Linux::setup_large_page_size() {
3244 _large_page_size = Linux::find_large_page_size();
3245 const size_t default_page_size = (size_t)Linux::page_size();
3246 if (_large_page_size > default_page_size) {
3247 _page_sizes[0] = _large_page_size;
3248 _page_sizes[1] = default_page_size;
3249 _page_sizes[2] = 0;
3250 }
3251
3252 return _large_page_size;
3253 }
3254
3255 bool os::Linux::setup_large_page_type(size_t page_size) {
3256 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3257 FLAG_IS_DEFAULT(UseSHM) &&
3258 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3259
3260 // The type of large pages has not been specified by the user.
3261
3262 // Try UseHugeTLBFS and then UseSHM.
3263 UseHugeTLBFS = UseSHM = true;
3264
3265 // Don't try UseTransparentHugePages since there are known
3266 // performance issues with it turned on. This might change in the future.
3267 UseTransparentHugePages = false;
3268 }
3269
3270 if (UseTransparentHugePages) {
3271 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3272 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3273 UseHugeTLBFS = false;
3274 UseSHM = false;
3275 return true;
3276 }
3277 UseTransparentHugePages = false;
3278 }
3279
3280 if (UseHugeTLBFS) {
3281 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3282 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3283 UseSHM = false;
3284 return true;
3285 }
3286 UseHugeTLBFS = false;
3287 }
3288
3289 return UseSHM;
3290 }
3291
3292 void os::large_page_init() {
3293 if (!UseLargePages &&
3294 !UseTransparentHugePages &&
3295 !UseHugeTLBFS &&
3296 !UseSHM) {
3297 // Not using large pages.
3298 return;
3299 }
3300
3301 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3302 // The user explicitly turned off large pages.
3303 // Ignore the rest of the large pages flags.
3304 UseTransparentHugePages = false;
3305 UseHugeTLBFS = false;
3306 UseSHM = false;
3307 return;
3308 }
3309
3310 size_t large_page_size = Linux::setup_large_page_size();
3311 UseLargePages = Linux::setup_large_page_type(large_page_size);
3312
3313 set_coredump_filter();
3314 }
3315
3316 #ifndef SHM_HUGETLB
3317 #define SHM_HUGETLB 04000
3318 #endif
3319
3320 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3321 char* req_addr, bool exec) {
3322 // "exec" is passed in but not used. Creating the shared image for
3323 // the code cache doesn't have an SHM_X executable permission to check.
3324 assert(UseLargePages && UseSHM, "only for SHM large pages");
3325 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3326
3327 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3328 return NULL; // Fallback to small pages.
3329 }
3330
3331 key_t key = IPC_PRIVATE;
3332 char *addr;
3333
3334 bool warn_on_failure = UseLargePages &&
3335 (!FLAG_IS_DEFAULT(UseLargePages) ||
3336 !FLAG_IS_DEFAULT(UseSHM) ||
3337 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3338 char msg[128];
3339
3340 // Create a large shared memory region to attach to based on size.
3341 // Currently, size is the total size of the heap
3342 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3343 if (shmid == -1) {
3344 // Possible reasons for shmget failure:
3345 // 1. shmmax is too small for Java heap.
3346 // > check shmmax value: cat /proc/sys/kernel/shmmax
3347 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3348 // 2. not enough large page memory.
3349 // > check available large pages: cat /proc/meminfo
3350 // > increase amount of large pages:
3351 // echo new_value > /proc/sys/vm/nr_hugepages
3352 // Note 1: different Linux may use different name for this property,
3353 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3354 // Note 2: it's possible there's enough physical memory available but
3355 // they are so fragmented after a long run that they can't
3356 // coalesce into large pages. Try to reserve large pages when
3357 // the system is still "fresh".
3358 if (warn_on_failure) {
3359 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3360 warning("%s", msg);
3361 }
3362 return NULL;
3363 }
3364
3365 // attach to the region
3366 addr = (char*)shmat(shmid, req_addr, 0);
3367 int err = errno;
3368
3369 // Remove shmid. If shmat() is successful, the actual shared memory segment
3370 // will be deleted when it's detached by shmdt() or when the process
3371 // terminates. If shmat() is not successful this will remove the shared
3372 // segment immediately.
3373 shmctl(shmid, IPC_RMID, NULL);
3374
3375 if ((intptr_t)addr == -1) {
3376 if (warn_on_failure) {
3377 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3378 warning("%s", msg);
3379 }
3380 return NULL;
3381 }
3382
3383 return addr;
3384 }
3385
3386 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3387 int error) {
3388 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3389
3390 bool warn_on_failure = UseLargePages &&
3391 (!FLAG_IS_DEFAULT(UseLargePages) ||
3392 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3393 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3394
3395 if (warn_on_failure) {
3396 char msg[128];
3397 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3398 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3399 warning("%s", msg);
3400 }
3401 }
3402
3403 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3404 char* req_addr,
3405 bool exec) {
3406 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3407 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3408 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3409
3410 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3411 char* addr = (char*)::mmap(req_addr, bytes, prot,
3412 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3413 -1, 0);
3414
3415 if (addr == MAP_FAILED) {
3416 warn_on_large_pages_failure(req_addr, bytes, errno);
3417 return NULL;
3418 }
3419
3420 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3421
3422 return addr;
3423 }
3424
3425 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3426 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3427 // (req_addr != NULL) or with a given alignment.
3428 // - bytes shall be a multiple of alignment.
3429 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3430 // - alignment sets the alignment at which memory shall be allocated.
3431 // It must be a multiple of allocation granularity.
3432 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3433 // req_addr or NULL.
3434 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3435
3436 size_t extra_size = bytes;
3437 if (req_addr == NULL && alignment > 0) {
3438 extra_size += alignment;
3439 }
3440
3441 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3442 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3443 -1, 0);
3444 if (start == MAP_FAILED) {
3445 start = NULL;
3446 } else {
3447 if (req_addr != NULL) {
3448 if (start != req_addr) {
3449 ::munmap(start, extra_size);
3450 start = NULL;
3451 }
3452 } else {
3453 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3454 char* const end_aligned = start_aligned + bytes;
3455 char* const end = start + extra_size;
3456 if (start_aligned > start) {
3457 ::munmap(start, start_aligned - start);
3458 }
3459 if (end_aligned < end) {
3460 ::munmap(end_aligned, end - end_aligned);
3461 }
3462 start = start_aligned;
3463 }
3464 }
3465 return start;
3466
3467 }
3468
3469 // Reserve memory using mmap(MAP_HUGETLB).
3470 // - bytes shall be a multiple of alignment.
3471 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3472 // - alignment sets the alignment at which memory shall be allocated.
3473 // It must be a multiple of allocation granularity.
3474 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3475 // req_addr or NULL.
3476 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3477 size_t alignment,
3478 char* req_addr,
3479 bool exec) {
3480 size_t large_page_size = os::large_page_size();
3481 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3482
3483 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3484 assert(is_size_aligned(bytes, alignment), "Must be");
3485
3486 // First reserve - but not commit - the address range in small pages.
3487 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3488
3489 if (start == NULL) {
3490 return NULL;
3491 }
3492
3493 assert(is_ptr_aligned(start, alignment), "Must be");
3494
3495 char* end = start + bytes;
3496
3497 // Find the regions of the allocated chunk that can be promoted to large pages.
3498 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3499 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3500
3501 size_t lp_bytes = lp_end - lp_start;
3502
3503 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3504
3505 if (lp_bytes == 0) {
3506 // The mapped region doesn't even span the start and the end of a large page.
3507 // Fall back to allocate a non-special area.
3508 ::munmap(start, end - start);
3509 return NULL;
3510 }
3511
3512 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3513
3514 void* result;
3515
3516 // Commit small-paged leading area.
3517 if (start != lp_start) {
3518 result = ::mmap(start, lp_start - start, prot,
3519 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3520 -1, 0);
3521 if (result == MAP_FAILED) {
3522 ::munmap(lp_start, end - lp_start);
3523 return NULL;
3524 }
3525 }
3526
3527 // Commit large-paged area.
3528 result = ::mmap(lp_start, lp_bytes, prot,
3529 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3530 -1, 0);
3531 if (result == MAP_FAILED) {
3532 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3533 // If the mmap above fails, the large pages region will be unmapped and we
3534 // have regions before and after with small pages. Release these regions.
3535 //
3536 // | mapped | unmapped | mapped |
3537 // ^ ^ ^ ^
3538 // start lp_start lp_end end
3539 //
3540 ::munmap(start, lp_start - start);
3541 ::munmap(lp_end, end - lp_end);
3542 return NULL;
3543 }
3544
3545 // Commit small-paged trailing area.
3546 if (lp_end != end) {
3547 result = ::mmap(lp_end, end - lp_end, prot,
3548 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3549 -1, 0);
3550 if (result == MAP_FAILED) {
3551 ::munmap(start, lp_end - start);
3552 return NULL;
3553 }
3554 }
3555
3556 return start;
3557 }
3558
3559 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3560 size_t alignment,
3561 char* req_addr,
3562 bool exec) {
3563 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3564 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3565 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3566 assert(is_power_of_2(os::large_page_size()), "Must be");
3567 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3568
3569 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3570 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3571 } else {
3572 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3573 }
3574 }
3575
3576 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3577 char* req_addr, bool exec) {
3578 assert(UseLargePages, "only for large pages");
3579
3580 char* addr;
3581 if (UseSHM) {
3582 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3583 } else {
3584 assert(UseHugeTLBFS, "must be");
3585 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3586 }
3587
3588 if (addr != NULL) {
3589 if (UseNUMAInterleaving) {
3590 numa_make_global(addr, bytes);
3591 }
3592
3593 // The memory is committed
3594 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3595 }
3596
3597 return addr;
3598 }
3599
3600 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3601 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3602 return shmdt(base) == 0;
3603 }
3604
3605 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3606 return pd_release_memory(base, bytes);
3607 }
3608
3609 bool os::release_memory_special(char* base, size_t bytes) {
3610 bool res;
3611 if (MemTracker::tracking_level() > NMT_minimal) {
3612 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3613 res = os::Linux::release_memory_special_impl(base, bytes);
3614 if (res) {
3615 tkr.record((address)base, bytes);
3616 }
3617
3618 } else {
3619 res = os::Linux::release_memory_special_impl(base, bytes);
3620 }
3621 return res;
3622 }
3623
3624 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3625 assert(UseLargePages, "only for large pages");
3626 bool res;
3627
3628 if (UseSHM) {
3629 res = os::Linux::release_memory_special_shm(base, bytes);
3630 } else {
3631 assert(UseHugeTLBFS, "must be");
3632 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3633 }
3634 return res;
3635 }
3636
3637 size_t os::large_page_size() {
3638 return _large_page_size;
3639 }
3640
3641 // With SysV SHM the entire memory region must be allocated as shared
3642 // memory.
3643 // HugeTLBFS allows application to commit large page memory on demand.
3644 // However, when committing memory with HugeTLBFS fails, the region
3645 // that was supposed to be committed will lose the old reservation
3646 // and allow other threads to steal that memory region. Because of this
3647 // behavior we can't commit HugeTLBFS memory.
3648 bool os::can_commit_large_page_memory() {
3649 return UseTransparentHugePages;
3650 }
3651
3652 bool os::can_execute_large_page_memory() {
3653 return UseTransparentHugePages || UseHugeTLBFS;
3654 }
3655
3656 // Reserve memory at an arbitrary address, only if that area is
3657 // available (and not reserved for something else).
3658
3659 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3660 const int max_tries = 10;
3661 char* base[max_tries];
3662 size_t size[max_tries];
3663 const size_t gap = 0x000000;
3664
3665 // Assert only that the size is a multiple of the page size, since
3666 // that's all that mmap requires, and since that's all we really know
3667 // about at this low abstraction level. If we need higher alignment,
3668 // we can either pass an alignment to this method or verify alignment
3669 // in one of the methods further up the call chain. See bug 5044738.
3670 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3671
3672 // Repeatedly allocate blocks until the block is allocated at the
3673 // right spot.
3674
3675 // Linux mmap allows caller to pass an address as hint; give it a try first,
3676 // if kernel honors the hint then we can return immediately.
3677 char * addr = anon_mmap(requested_addr, bytes, false);
3678 if (addr == requested_addr) {
3679 return requested_addr;
3680 }
3681
3682 if (addr != NULL) {
3683 // mmap() is successful but it fails to reserve at the requested address
3684 anon_munmap(addr, bytes);
3685 }
3686
3687 int i;
3688 for (i = 0; i < max_tries; ++i) {
3689 base[i] = reserve_memory(bytes);
3690
3691 if (base[i] != NULL) {
3692 // Is this the block we wanted?
3693 if (base[i] == requested_addr) {
3694 size[i] = bytes;
3695 break;
3696 }
3697
3698 // Does this overlap the block we wanted? Give back the overlapped
3699 // parts and try again.
3700
3701 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3702 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3703 unmap_memory(base[i], top_overlap);
3704 base[i] += top_overlap;
3705 size[i] = bytes - top_overlap;
3706 } else {
3707 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3708 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3709 unmap_memory(requested_addr, bottom_overlap);
3710 size[i] = bytes - bottom_overlap;
3711 } else {
3712 size[i] = bytes;
3713 }
3714 }
3715 }
3716 }
3717
3718 // Give back the unused reserved pieces.
3719
3720 for (int j = 0; j < i; ++j) {
3721 if (base[j] != NULL) {
3722 unmap_memory(base[j], size[j]);
3723 }
3724 }
3725
3726 if (i < max_tries) {
3727 return requested_addr;
3728 } else {
3729 return NULL;
3730 }
3731 }
3732
3733 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3734 return ::read(fd, buf, nBytes);
3735 }
3736
3737 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3738 return ::pread(fd, buf, nBytes, offset);
3739 }
3740
3741 // Short sleep, direct OS call.
3742 //
3743 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3744 // sched_yield(2) will actually give up the CPU:
3745 //
3746 // * Alone on this pariticular CPU, keeps running.
3747 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3748 // (pre 2.6.39).
3749 //
3750 // So calling this with 0 is an alternative.
3751 //
3752 void os::naked_short_sleep(jlong ms) {
3753 struct timespec req;
3754
3755 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3756 req.tv_sec = 0;
3757 if (ms > 0) {
3758 req.tv_nsec = (ms % 1000) * 1000000;
3759 } else {
3760 req.tv_nsec = 1;
3761 }
3762
3763 nanosleep(&req, NULL);
3764
3765 return;
3766 }
3767
3768 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3769 void os::infinite_sleep() {
3770 while (true) { // sleep forever ...
3771 ::sleep(100); // ... 100 seconds at a time
3772 }
3773 }
3774
3775 // Used to convert frequent JVM_Yield() to nops
3776 bool os::dont_yield() {
3777 return DontYieldALot;
3778 }
3779
3780 void os::naked_yield() {
3781 sched_yield();
3782 }
3783
3784 ////////////////////////////////////////////////////////////////////////////////
3785 // thread priority support
3786
3787 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3788 // only supports dynamic priority, static priority must be zero. For real-time
3789 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3790 // However, for large multi-threaded applications, SCHED_RR is not only slower
3791 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3792 // of 5 runs - Sep 2005).
3793 //
3794 // The following code actually changes the niceness of kernel-thread/LWP. It
3795 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3796 // not the entire user process, and user level threads are 1:1 mapped to kernel
3797 // threads. It has always been the case, but could change in the future. For
3798 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3799 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3800
3801 int os::java_to_os_priority[CriticalPriority + 1] = {
3802 19, // 0 Entry should never be used
3803
3804 4, // 1 MinPriority
3805 3, // 2
3806 2, // 3
3807
3808 1, // 4
3809 0, // 5 NormPriority
3810 -1, // 6
3811
3812 -2, // 7
3813 -3, // 8
3814 -4, // 9 NearMaxPriority
3815
3816 -5, // 10 MaxPriority
3817
3818 -5 // 11 CriticalPriority
3819 };
3820
3821 static int prio_init() {
3822 if (ThreadPriorityPolicy == 1) {
3823 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3824 // if effective uid is not root. Perhaps, a more elegant way of doing
3825 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3826 if (geteuid() != 0) {
3827 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3828 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3829 }
3830 ThreadPriorityPolicy = 0;
3831 }
3832 }
3833 if (UseCriticalJavaThreadPriority) {
3834 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3835 }
3836 return 0;
3837 }
3838
3839 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3840 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3841
3842 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3843 return (ret == 0) ? OS_OK : OS_ERR;
3844 }
3845
3846 OSReturn os::get_native_priority(const Thread* const thread,
3847 int *priority_ptr) {
3848 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3849 *priority_ptr = java_to_os_priority[NormPriority];
3850 return OS_OK;
3851 }
3852
3853 errno = 0;
3854 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3855 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3856 }
3857
3858 // Hint to the underlying OS that a task switch would not be good.
3859 // Void return because it's a hint and can fail.
3860 void os::hint_no_preempt() {}
3861
3862 ////////////////////////////////////////////////////////////////////////////////
3863 // suspend/resume support
3864
3865 // the low-level signal-based suspend/resume support is a remnant from the
3866 // old VM-suspension that used to be for java-suspension, safepoints etc,
3867 // within hotspot. Now there is a single use-case for this:
3868 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3869 // that runs in the watcher thread.
3870 // The remaining code is greatly simplified from the more general suspension
3871 // code that used to be used.
3872 //
3873 // The protocol is quite simple:
3874 // - suspend:
3875 // - sends a signal to the target thread
3876 // - polls the suspend state of the osthread using a yield loop
3877 // - target thread signal handler (SR_handler) sets suspend state
3878 // and blocks in sigsuspend until continued
3879 // - resume:
3880 // - sets target osthread state to continue
3881 // - sends signal to end the sigsuspend loop in the SR_handler
3882 //
3883 // Note that the SR_lock plays no role in this suspend/resume protocol.
3884
3885 static void resume_clear_context(OSThread *osthread) {
3886 osthread->set_ucontext(NULL);
3887 osthread->set_siginfo(NULL);
3888 }
3889
3890 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3891 ucontext_t* context) {
3892 osthread->set_ucontext(context);
3893 osthread->set_siginfo(siginfo);
3894 }
3895
3896 // Handler function invoked when a thread's execution is suspended or
3897 // resumed. We have to be careful that only async-safe functions are
3898 // called here (Note: most pthread functions are not async safe and
3899 // should be avoided.)
3900 //
3901 // Note: sigwait() is a more natural fit than sigsuspend() from an
3902 // interface point of view, but sigwait() prevents the signal hander
3903 // from being run. libpthread would get very confused by not having
3904 // its signal handlers run and prevents sigwait()'s use with the
3905 // mutex granting granting signal.
3906 //
3907 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3908 //
3909 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3910 // Save and restore errno to avoid confusing native code with EINTR
3911 // after sigsuspend.
3912 int old_errno = errno;
3913
3914 Thread* thread = Thread::current();
3915 OSThread* osthread = thread->osthread();
3916 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3917
3918 os::SuspendResume::State current = osthread->sr.state();
3919 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3920 suspend_save_context(osthread, siginfo, context);
3921
3922 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3923 os::SuspendResume::State state = osthread->sr.suspended();
3924 if (state == os::SuspendResume::SR_SUSPENDED) {
3925 sigset_t suspend_set; // signals for sigsuspend()
3926
3927 // get current set of blocked signals and unblock resume signal
3928 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3929 sigdelset(&suspend_set, SR_signum);
3930
3931 sr_semaphore.signal();
3932 // wait here until we are resumed
3933 while (1) {
3934 sigsuspend(&suspend_set);
3935
3936 os::SuspendResume::State result = osthread->sr.running();
3937 if (result == os::SuspendResume::SR_RUNNING) {
3938 sr_semaphore.signal();
3939 break;
3940 }
3941 }
3942
3943 } else if (state == os::SuspendResume::SR_RUNNING) {
3944 // request was cancelled, continue
3945 } else {
3946 ShouldNotReachHere();
3947 }
3948
3949 resume_clear_context(osthread);
3950 } else if (current == os::SuspendResume::SR_RUNNING) {
3951 // request was cancelled, continue
3952 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3953 // ignore
3954 } else {
3955 // ignore
3956 }
3957
3958 errno = old_errno;
3959 }
3960
3961 static int SR_initialize() {
3962 struct sigaction act;
3963 char *s;
3964
3965 // Get signal number to use for suspend/resume
3966 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3967 int sig = ::strtol(s, 0, 10);
3968 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
3969 sig < NSIG) { // Must be legal signal and fit into sigflags[].
3970 SR_signum = sig;
3971 } else {
3972 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3973 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3974 }
3975 }
3976
3977 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3978 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3979
3980 sigemptyset(&SR_sigset);
3981 sigaddset(&SR_sigset, SR_signum);
3982
3983 // Set up signal handler for suspend/resume
3984 act.sa_flags = SA_RESTART|SA_SIGINFO;
3985 act.sa_handler = (void (*)(int)) SR_handler;
3986
3987 // SR_signum is blocked by default.
3988 // 4528190 - We also need to block pthread restart signal (32 on all
3989 // supported Linux platforms). Note that LinuxThreads need to block
3990 // this signal for all threads to work properly. So we don't have
3991 // to use hard-coded signal number when setting up the mask.
3992 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3993
3994 if (sigaction(SR_signum, &act, 0) == -1) {
3995 return -1;
3996 }
3997
3998 // Save signal flag
3999 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4000 return 0;
4001 }
4002
4003 static int sr_notify(OSThread* osthread) {
4004 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4005 assert_status(status == 0, status, "pthread_kill");
4006 return status;
4007 }
4008
4009 // "Randomly" selected value for how long we want to spin
4010 // before bailing out on suspending a thread, also how often
4011 // we send a signal to a thread we want to resume
4012 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4013 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4014
4015 // returns true on success and false on error - really an error is fatal
4016 // but this seems the normal response to library errors
4017 static bool do_suspend(OSThread* osthread) {
4018 assert(osthread->sr.is_running(), "thread should be running");
4019 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4020
4021 // mark as suspended and send signal
4022 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4023 // failed to switch, state wasn't running?
4024 ShouldNotReachHere();
4025 return false;
4026 }
4027
4028 if (sr_notify(osthread) != 0) {
4029 ShouldNotReachHere();
4030 }
4031
4032 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4033 while (true) {
4034 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4035 break;
4036 } else {
4037 // timeout
4038 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4039 if (cancelled == os::SuspendResume::SR_RUNNING) {
4040 return false;
4041 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4042 // make sure that we consume the signal on the semaphore as well
4043 sr_semaphore.wait();
4044 break;
4045 } else {
4046 ShouldNotReachHere();
4047 return false;
4048 }
4049 }
4050 }
4051
4052 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4053 return true;
4054 }
4055
4056 static void do_resume(OSThread* osthread) {
4057 assert(osthread->sr.is_suspended(), "thread should be suspended");
4058 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4059
4060 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4061 // failed to switch to WAKEUP_REQUEST
4062 ShouldNotReachHere();
4063 return;
4064 }
4065
4066 while (true) {
4067 if (sr_notify(osthread) == 0) {
4068 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4069 if (osthread->sr.is_running()) {
4070 return;
4071 }
4072 }
4073 } else {
4074 ShouldNotReachHere();
4075 }
4076 }
4077
4078 guarantee(osthread->sr.is_running(), "Must be running!");
4079 }
4080
4081 ///////////////////////////////////////////////////////////////////////////////////
4082 // signal handling (except suspend/resume)
4083
4084 // This routine may be used by user applications as a "hook" to catch signals.
4085 // The user-defined signal handler must pass unrecognized signals to this
4086 // routine, and if it returns true (non-zero), then the signal handler must
4087 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4088 // routine will never retun false (zero), but instead will execute a VM panic
4089 // routine kill the process.
4090 //
4091 // If this routine returns false, it is OK to call it again. This allows
4092 // the user-defined signal handler to perform checks either before or after
4093 // the VM performs its own checks. Naturally, the user code would be making
4094 // a serious error if it tried to handle an exception (such as a null check
4095 // or breakpoint) that the VM was generating for its own correct operation.
4096 //
4097 // This routine may recognize any of the following kinds of signals:
4098 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4099 // It should be consulted by handlers for any of those signals.
4100 //
4101 // The caller of this routine must pass in the three arguments supplied
4102 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4103 // field of the structure passed to sigaction(). This routine assumes that
4104 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4105 //
4106 // Note that the VM will print warnings if it detects conflicting signal
4107 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4108 //
4109 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4110 siginfo_t* siginfo,
4111 void* ucontext,
4112 int abort_if_unrecognized);
4113
4114 void signalHandler(int sig, siginfo_t* info, void* uc) {
4115 assert(info != NULL && uc != NULL, "it must be old kernel");
4116 int orig_errno = errno; // Preserve errno value over signal handler.
4117 JVM_handle_linux_signal(sig, info, uc, true);
4118 errno = orig_errno;
4119 }
4120
4121
4122 // This boolean allows users to forward their own non-matching signals
4123 // to JVM_handle_linux_signal, harmlessly.
4124 bool os::Linux::signal_handlers_are_installed = false;
4125
4126 // For signal-chaining
4127 struct sigaction sigact[NSIG];
4128 uint64_t sigs = 0;
4129 #if (64 < NSIG-1)
4130 #error "Not all signals can be encoded in sigs. Adapt its type!"
4131 #endif
4132 bool os::Linux::libjsig_is_loaded = false;
4133 typedef struct sigaction *(*get_signal_t)(int);
4134 get_signal_t os::Linux::get_signal_action = NULL;
4135
4136 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4137 struct sigaction *actp = NULL;
4138
4139 if (libjsig_is_loaded) {
4140 // Retrieve the old signal handler from libjsig
4141 actp = (*get_signal_action)(sig);
4142 }
4143 if (actp == NULL) {
4144 // Retrieve the preinstalled signal handler from jvm
4145 actp = get_preinstalled_handler(sig);
4146 }
4147
4148 return actp;
4149 }
4150
4151 static bool call_chained_handler(struct sigaction *actp, int sig,
4152 siginfo_t *siginfo, void *context) {
4153 // Call the old signal handler
4154 if (actp->sa_handler == SIG_DFL) {
4155 // It's more reasonable to let jvm treat it as an unexpected exception
4156 // instead of taking the default action.
4157 return false;
4158 } else if (actp->sa_handler != SIG_IGN) {
4159 if ((actp->sa_flags & SA_NODEFER) == 0) {
4160 // automaticlly block the signal
4161 sigaddset(&(actp->sa_mask), sig);
4162 }
4163
4164 sa_handler_t hand = NULL;
4165 sa_sigaction_t sa = NULL;
4166 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4167 // retrieve the chained handler
4168 if (siginfo_flag_set) {
4169 sa = actp->sa_sigaction;
4170 } else {
4171 hand = actp->sa_handler;
4172 }
4173
4174 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4175 actp->sa_handler = SIG_DFL;
4176 }
4177
4178 // try to honor the signal mask
4179 sigset_t oset;
4180 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4181
4182 // call into the chained handler
4183 if (siginfo_flag_set) {
4184 (*sa)(sig, siginfo, context);
4185 } else {
4186 (*hand)(sig);
4187 }
4188
4189 // restore the signal mask
4190 pthread_sigmask(SIG_SETMASK, &oset, 0);
4191 }
4192 // Tell jvm's signal handler the signal is taken care of.
4193 return true;
4194 }
4195
4196 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4197 bool chained = false;
4198 // signal-chaining
4199 if (UseSignalChaining) {
4200 struct sigaction *actp = get_chained_signal_action(sig);
4201 if (actp != NULL) {
4202 chained = call_chained_handler(actp, sig, siginfo, context);
4203 }
4204 }
4205 return chained;
4206 }
4207
4208 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4209 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4210 return &sigact[sig];
4211 }
4212 return NULL;
4213 }
4214
4215 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4216 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4217 sigact[sig] = oldAct;
4218 sigs |= (uint64_t)1 << (sig-1);
4219 }
4220
4221 // for diagnostic
4222 int sigflags[NSIG];
4223
4224 int os::Linux::get_our_sigflags(int sig) {
4225 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4226 return sigflags[sig];
4227 }
4228
4229 void os::Linux::set_our_sigflags(int sig, int flags) {
4230 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4231 if (sig > 0 && sig < NSIG) {
4232 sigflags[sig] = flags;
4233 }
4234 }
4235
4236 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4237 // Check for overwrite.
4238 struct sigaction oldAct;
4239 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4240
4241 void* oldhand = oldAct.sa_sigaction
4242 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4243 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4244 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4245 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4246 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4247 if (AllowUserSignalHandlers || !set_installed) {
4248 // Do not overwrite; user takes responsibility to forward to us.
4249 return;
4250 } else if (UseSignalChaining) {
4251 // save the old handler in jvm
4252 save_preinstalled_handler(sig, oldAct);
4253 // libjsig also interposes the sigaction() call below and saves the
4254 // old sigaction on it own.
4255 } else {
4256 fatal("Encountered unexpected pre-existing sigaction handler "
4257 "%#lx for signal %d.", (long)oldhand, sig);
4258 }
4259 }
4260
4261 struct sigaction sigAct;
4262 sigfillset(&(sigAct.sa_mask));
4263 sigAct.sa_handler = SIG_DFL;
4264 if (!set_installed) {
4265 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4266 } else {
4267 sigAct.sa_sigaction = signalHandler;
4268 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4269 }
4270 // Save flags, which are set by ours
4271 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4272 sigflags[sig] = sigAct.sa_flags;
4273
4274 int ret = sigaction(sig, &sigAct, &oldAct);
4275 assert(ret == 0, "check");
4276
4277 void* oldhand2 = oldAct.sa_sigaction
4278 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4279 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4280 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4281 }
4282
4283 // install signal handlers for signals that HotSpot needs to
4284 // handle in order to support Java-level exception handling.
4285
4286 void os::Linux::install_signal_handlers() {
4287 if (!signal_handlers_are_installed) {
4288 signal_handlers_are_installed = true;
4289
4290 // signal-chaining
4291 typedef void (*signal_setting_t)();
4292 signal_setting_t begin_signal_setting = NULL;
4293 signal_setting_t end_signal_setting = NULL;
4294 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4295 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4296 if (begin_signal_setting != NULL) {
4297 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4298 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4299 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4300 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4301 libjsig_is_loaded = true;
4302 assert(UseSignalChaining, "should enable signal-chaining");
4303 }
4304 if (libjsig_is_loaded) {
4305 // Tell libjsig jvm is setting signal handlers
4306 (*begin_signal_setting)();
4307 }
4308
4309 set_signal_handler(SIGSEGV, true);
4310 set_signal_handler(SIGPIPE, true);
4311 set_signal_handler(SIGBUS, true);
4312 set_signal_handler(SIGILL, true);
4313 set_signal_handler(SIGFPE, true);
4314 #if defined(PPC64)
4315 set_signal_handler(SIGTRAP, true);
4316 #endif
4317 set_signal_handler(SIGXFSZ, true);
4318
4319 if (libjsig_is_loaded) {
4320 // Tell libjsig jvm finishes setting signal handlers
4321 (*end_signal_setting)();
4322 }
4323
4324 // We don't activate signal checker if libjsig is in place, we trust ourselves
4325 // and if UserSignalHandler is installed all bets are off.
4326 // Log that signal checking is off only if -verbose:jni is specified.
4327 if (CheckJNICalls) {
4328 if (libjsig_is_loaded) {
4329 if (PrintJNIResolving) {
4330 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4331 }
4332 check_signals = false;
4333 }
4334 if (AllowUserSignalHandlers) {
4335 if (PrintJNIResolving) {
4336 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4337 }
4338 check_signals = false;
4339 }
4340 }
4341 }
4342 }
4343
4344 // This is the fastest way to get thread cpu time on Linux.
4345 // Returns cpu time (user+sys) for any thread, not only for current.
4346 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4347 // It might work on 2.6.10+ with a special kernel/glibc patch.
4348 // For reference, please, see IEEE Std 1003.1-2004:
4349 // http://www.unix.org/single_unix_specification
4350
4351 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4352 struct timespec tp;
4353 int rc = os::Linux::clock_gettime(clockid, &tp);
4354 assert(rc == 0, "clock_gettime is expected to return 0 code");
4355
4356 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4357 }
4358
4359 /////
4360 // glibc on Linux platform uses non-documented flag
4361 // to indicate, that some special sort of signal
4362 // trampoline is used.
4363 // We will never set this flag, and we should
4364 // ignore this flag in our diagnostic
4365 #ifdef SIGNIFICANT_SIGNAL_MASK
4366 #undef SIGNIFICANT_SIGNAL_MASK
4367 #endif
4368 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4369
4370 static const char* get_signal_handler_name(address handler,
4371 char* buf, int buflen) {
4372 int offset = 0;
4373 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4374 if (found) {
4375 // skip directory names
4376 const char *p1, *p2;
4377 p1 = buf;
4378 size_t len = strlen(os::file_separator());
4379 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4380 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4381 } else {
4382 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4383 }
4384 return buf;
4385 }
4386
4387 static void print_signal_handler(outputStream* st, int sig,
4388 char* buf, size_t buflen) {
4389 struct sigaction sa;
4390
4391 sigaction(sig, NULL, &sa);
4392
4393 // See comment for SIGNIFICANT_SIGNAL_MASK define
4394 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4395
4396 st->print("%s: ", os::exception_name(sig, buf, buflen));
4397
4398 address handler = (sa.sa_flags & SA_SIGINFO)
4399 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4400 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4401
4402 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4403 st->print("SIG_DFL");
4404 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4405 st->print("SIG_IGN");
4406 } else {
4407 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4408 }
4409
4410 st->print(", sa_mask[0]=");
4411 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4412
4413 address rh = VMError::get_resetted_sighandler(sig);
4414 // May be, handler was resetted by VMError?
4415 if (rh != NULL) {
4416 handler = rh;
4417 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4418 }
4419
4420 st->print(", sa_flags=");
4421 os::Posix::print_sa_flags(st, sa.sa_flags);
4422
4423 // Check: is it our handler?
4424 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4425 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4426 // It is our signal handler
4427 // check for flags, reset system-used one!
4428 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4429 st->print(
4430 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4431 os::Linux::get_our_sigflags(sig));
4432 }
4433 }
4434 st->cr();
4435 }
4436
4437
4438 #define DO_SIGNAL_CHECK(sig) \
4439 do { \
4440 if (!sigismember(&check_signal_done, sig)) { \
4441 os::Linux::check_signal_handler(sig); \
4442 } \
4443 } while (0)
4444
4445 // This method is a periodic task to check for misbehaving JNI applications
4446 // under CheckJNI, we can add any periodic checks here
4447
4448 void os::run_periodic_checks() {
4449 if (check_signals == false) return;
4450
4451 // SEGV and BUS if overridden could potentially prevent
4452 // generation of hs*.log in the event of a crash, debugging
4453 // such a case can be very challenging, so we absolutely
4454 // check the following for a good measure:
4455 DO_SIGNAL_CHECK(SIGSEGV);
4456 DO_SIGNAL_CHECK(SIGILL);
4457 DO_SIGNAL_CHECK(SIGFPE);
4458 DO_SIGNAL_CHECK(SIGBUS);
4459 DO_SIGNAL_CHECK(SIGPIPE);
4460 DO_SIGNAL_CHECK(SIGXFSZ);
4461 #if defined(PPC64)
4462 DO_SIGNAL_CHECK(SIGTRAP);
4463 #endif
4464
4465 // ReduceSignalUsage allows the user to override these handlers
4466 // see comments at the very top and jvm_solaris.h
4467 if (!ReduceSignalUsage) {
4468 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4469 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4470 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4471 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4472 }
4473
4474 DO_SIGNAL_CHECK(SR_signum);
4475 }
4476
4477 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4478
4479 static os_sigaction_t os_sigaction = NULL;
4480
4481 void os::Linux::check_signal_handler(int sig) {
4482 char buf[O_BUFLEN];
4483 address jvmHandler = NULL;
4484
4485
4486 struct sigaction act;
4487 if (os_sigaction == NULL) {
4488 // only trust the default sigaction, in case it has been interposed
4489 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4490 if (os_sigaction == NULL) return;
4491 }
4492
4493 os_sigaction(sig, (struct sigaction*)NULL, &act);
4494
4495
4496 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4497
4498 address thisHandler = (act.sa_flags & SA_SIGINFO)
4499 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4500 : CAST_FROM_FN_PTR(address, act.sa_handler);
4501
4502
4503 switch (sig) {
4504 case SIGSEGV:
4505 case SIGBUS:
4506 case SIGFPE:
4507 case SIGPIPE:
4508 case SIGILL:
4509 case SIGXFSZ:
4510 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4511 break;
4512
4513 case SHUTDOWN1_SIGNAL:
4514 case SHUTDOWN2_SIGNAL:
4515 case SHUTDOWN3_SIGNAL:
4516 case BREAK_SIGNAL:
4517 jvmHandler = (address)user_handler();
4518 break;
4519
4520 default:
4521 if (sig == SR_signum) {
4522 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4523 } else {
4524 return;
4525 }
4526 break;
4527 }
4528
4529 if (thisHandler != jvmHandler) {
4530 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4531 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4532 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4533 // No need to check this sig any longer
4534 sigaddset(&check_signal_done, sig);
4535 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4536 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4537 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4538 exception_name(sig, buf, O_BUFLEN));
4539 }
4540 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4541 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4542 tty->print("expected:");
4543 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4544 tty->cr();
4545 tty->print(" found:");
4546 os::Posix::print_sa_flags(tty, act.sa_flags);
4547 tty->cr();
4548 // No need to check this sig any longer
4549 sigaddset(&check_signal_done, sig);
4550 }
4551
4552 // Dump all the signal
4553 if (sigismember(&check_signal_done, sig)) {
4554 print_signal_handlers(tty, buf, O_BUFLEN);
4555 }
4556 }
4557
4558 extern void report_error(char* file_name, int line_no, char* title,
4559 char* format, ...);
4560
4561 // this is called _before_ the most of global arguments have been parsed
4562 void os::init(void) {
4563 char dummy; // used to get a guess on initial stack address
4564 // first_hrtime = gethrtime();
4565
4566 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4567
4568 init_random(1234567);
4569
4570 ThreadCritical::initialize();
4571
4572 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4573 if (Linux::page_size() == -1) {
4574 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4575 strerror(errno));
4576 }
4577 init_page_sizes((size_t) Linux::page_size());
4578
4579 Linux::initialize_system_info();
4580
4581 // main_thread points to the aboriginal thread
4582 Linux::_main_thread = pthread_self();
4583
4584 Linux::clock_init();
4585 initial_time_count = javaTimeNanos();
4586
4587 // pthread_condattr initialization for monotonic clock
4588 int status;
4589 pthread_condattr_t* _condattr = os::Linux::condAttr();
4590 if ((status = pthread_condattr_init(_condattr)) != 0) {
4591 fatal("pthread_condattr_init: %s", strerror(status));
4592 }
4593 // Only set the clock if CLOCK_MONOTONIC is available
4594 if (os::supports_monotonic_clock()) {
4595 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4596 if (status == EINVAL) {
4597 warning("Unable to use monotonic clock with relative timed-waits" \
4598 " - changes to the time-of-day clock may have adverse affects");
4599 } else {
4600 fatal("pthread_condattr_setclock: %s", strerror(status));
4601 }
4602 }
4603 }
4604 // else it defaults to CLOCK_REALTIME
4605
4606 // retrieve entry point for pthread_setname_np
4607 Linux::_pthread_setname_np =
4608 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4609
4610 }
4611
4612 // To install functions for atexit system call
4613 extern "C" {
4614 static void perfMemory_exit_helper() {
4615 perfMemory_exit();
4616 }
4617 }
4618
4619 // this is called _after_ the global arguments have been parsed
4620 jint os::init_2(void) {
4621 Linux::fast_thread_clock_init();
4622
4623 // Allocate a single page and mark it as readable for safepoint polling
4624 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4625 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4626
4627 os::set_polling_page(polling_page);
4628
4629 #ifndef PRODUCT
4630 if (Verbose && PrintMiscellaneous) {
4631 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4632 (intptr_t)polling_page);
4633 }
4634 #endif
4635
4636 if (!UseMembar) {
4637 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4638 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4639 os::set_memory_serialize_page(mem_serialize_page);
4640
4641 #ifndef PRODUCT
4642 if (Verbose && PrintMiscellaneous) {
4643 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4644 (intptr_t)mem_serialize_page);
4645 }
4646 #endif
4647 }
4648
4649 // initialize suspend/resume support - must do this before signal_sets_init()
4650 if (SR_initialize() != 0) {
4651 perror("SR_initialize failed");
4652 return JNI_ERR;
4653 }
4654
4655 Linux::signal_sets_init();
4656 Linux::install_signal_handlers();
4657
4658 // Check minimum allowable stack size for thread creation and to initialize
4659 // the java system classes, including StackOverflowError - depends on page
4660 // size. Add a page for compiler2 recursion in main thread.
4661 // Add in 2*BytesPerWord times page size to account for VM stack during
4662 // class initialization depending on 32 or 64 bit VM.
4663 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4664 JavaThread::stack_guard_zone_size() +
4665 JavaThread::stack_shadow_zone_size() +
4666 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4667
4668 size_t threadStackSizeInBytes = ThreadStackSize * K;
4669 if (threadStackSizeInBytes != 0 &&
4670 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4671 tty->print_cr("\nThe stack size specified is too small, "
4672 "Specify at least " SIZE_FORMAT "k",
4673 os::Linux::min_stack_allowed/ K);
4674 return JNI_ERR;
4675 }
4676
4677 // Make the stack size a multiple of the page size so that
4678 // the yellow/red zones can be guarded.
4679 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4680 vm_page_size()));
4681
4682 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4683
4684 #if defined(IA32)
4685 workaround_expand_exec_shield_cs_limit();
4686 #endif
4687
4688 Linux::libpthread_init();
4689 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4690 tty->print_cr("[HotSpot is running with %s, %s]\n",
4691 Linux::glibc_version(), Linux::libpthread_version());
4692 }
4693
4694 if (UseNUMA) {
4695 if (!Linux::libnuma_init()) {
4696 UseNUMA = false;
4697 } else {
4698 if ((Linux::numa_max_node() < 1)) {
4699 // There's only one node(they start from 0), disable NUMA.
4700 UseNUMA = false;
4701 }
4702 }
4703 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4704 // we can make the adaptive lgrp chunk resizing work. If the user specified
4705 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4706 // disable adaptive resizing.
4707 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4708 if (FLAG_IS_DEFAULT(UseNUMA)) {
4709 UseNUMA = false;
4710 } else {
4711 if (FLAG_IS_DEFAULT(UseLargePages) &&
4712 FLAG_IS_DEFAULT(UseSHM) &&
4713 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4714 UseLargePages = false;
4715 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4716 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4717 UseAdaptiveSizePolicy = false;
4718 UseAdaptiveNUMAChunkSizing = false;
4719 }
4720 }
4721 }
4722 if (!UseNUMA && ForceNUMA) {
4723 UseNUMA = true;
4724 }
4725 }
4726
4727 if (MaxFDLimit) {
4728 // set the number of file descriptors to max. print out error
4729 // if getrlimit/setrlimit fails but continue regardless.
4730 struct rlimit nbr_files;
4731 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4732 if (status != 0) {
4733 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4734 perror("os::init_2 getrlimit failed");
4735 }
4736 } else {
4737 nbr_files.rlim_cur = nbr_files.rlim_max;
4738 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4739 if (status != 0) {
4740 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4741 perror("os::init_2 setrlimit failed");
4742 }
4743 }
4744 }
4745 }
4746
4747 // Initialize lock used to serialize thread creation (see os::create_thread)
4748 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4749
4750 // at-exit methods are called in the reverse order of their registration.
4751 // atexit functions are called on return from main or as a result of a
4752 // call to exit(3C). There can be only 32 of these functions registered
4753 // and atexit() does not set errno.
4754
4755 if (PerfAllowAtExitRegistration) {
4756 // only register atexit functions if PerfAllowAtExitRegistration is set.
4757 // atexit functions can be delayed until process exit time, which
4758 // can be problematic for embedded VM situations. Embedded VMs should
4759 // call DestroyJavaVM() to assure that VM resources are released.
4760
4761 // note: perfMemory_exit_helper atexit function may be removed in
4762 // the future if the appropriate cleanup code can be added to the
4763 // VM_Exit VMOperation's doit method.
4764 if (atexit(perfMemory_exit_helper) != 0) {
4765 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4766 }
4767 }
4768
4769 // initialize thread priority policy
4770 prio_init();
4771
4772 return JNI_OK;
4773 }
4774
4775 // Mark the polling page as unreadable
4776 void os::make_polling_page_unreadable(void) {
4777 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4778 fatal("Could not disable polling page");
4779 }
4780 }
4781
4782 // Mark the polling page as readable
4783 void os::make_polling_page_readable(void) {
4784 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4785 fatal("Could not enable polling page");
4786 }
4787 }
4788
4789 // older glibc versions don't have this macro (which expands to
4790 // an optimized bit-counting function) so we have to roll our own
4791 #ifndef CPU_COUNT
4792
4793 static int _cpu_count(const cpu_set_t* cpus) {
4794 int count = 0;
4795 // only look up to the number of configured processors
4796 for (int i = 0; i < os::processor_count(); i++) {
4797 if (CPU_ISSET(i, cpus)) {
4798 count++;
4799 }
4800 }
4801 return count;
4802 }
4803
4804 #define CPU_COUNT(cpus) _cpu_count(cpus)
4805
4806 #endif // CPU_COUNT
4807
4808 // Get the current number of available processors for this process.
4809 // This value can change at any time during a process's lifetime.
4810 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4811 // If it appears there may be more than 1024 processors then we do a
4812 // dynamic check - see 6515172 for details.
4813 // If anything goes wrong we fallback to returning the number of online
4814 // processors - which can be greater than the number available to the process.
4815 int os::active_processor_count() {
4816 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4817 cpu_set_t* cpus_p = &cpus;
4818 int cpus_size = sizeof(cpu_set_t);
4819
4820 int configured_cpus = processor_count(); // upper bound on available cpus
4821 int cpu_count = 0;
4822
4823 // old build platforms may not support dynamic cpu sets
4824 #ifdef CPU_ALLOC
4825
4826 // To enable easy testing of the dynamic path on different platforms we
4827 // introduce a diagnostic flag: UseCpuAllocPath
4828 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4829 // kernel may use a mask bigger than cpu_set_t
4830 log_trace(os)("active_processor_count: using dynamic path %s"
4831 "- configured processors: %d",
4832 UseCpuAllocPath ? "(forced) " : "",
4833 configured_cpus);
4834 cpus_p = CPU_ALLOC(configured_cpus);
4835 if (cpus_p != NULL) {
4836 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4837 // zero it just to be safe
4838 CPU_ZERO_S(cpus_size, cpus_p);
4839 }
4840 else {
4841 // failed to allocate so fallback to online cpus
4842 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4843 log_trace(os)("active_processor_count: "
4844 "CPU_ALLOC failed (%s) - using "
4845 "online processor count: %d",
4846 strerror(errno), online_cpus);
4847 return online_cpus;
4848 }
4849 }
4850 else {
4851 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4852 configured_cpus);
4853 }
4854 #else // CPU_ALLOC
4855 // these stubs won't be executed
4856 #define CPU_COUNT_S(size, cpus) -1
4857 #define CPU_FREE(cpus)
4858
4859 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4860 configured_cpus);
4861 #endif // CPU_ALLOC
4862
4863 // pid 0 means the current thread - which we have to assume represents the process
4864 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4865 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4866 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4867 }
4868 else {
4869 cpu_count = CPU_COUNT(cpus_p);
4870 }
4871 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4872 }
4873 else {
4874 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4875 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4876 "which may exceed available processors", strerror(errno), cpu_count);
4877 }
4878
4879 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4880 CPU_FREE(cpus_p);
4881 }
4882
4883 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4884 return cpu_count;
4885 }
4886
4887 void os::set_native_thread_name(const char *name) {
4888 if (Linux::_pthread_setname_np) {
4889 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4890 snprintf(buf, sizeof(buf), "%s", name);
4891 buf[sizeof(buf) - 1] = '\0';
4892 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4893 // ERANGE should not happen; all other errors should just be ignored.
4894 assert(rc != ERANGE, "pthread_setname_np failed");
4895 }
4896 }
4897
4898 bool os::distribute_processes(uint length, uint* distribution) {
4899 // Not yet implemented.
4900 return false;
4901 }
4902
4903 bool os::bind_to_processor(uint processor_id) {
4904 // Not yet implemented.
4905 return false;
4906 }
4907
4908 ///
4909
4910 void os::SuspendedThreadTask::internal_do_task() {
4911 if (do_suspend(_thread->osthread())) {
4912 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4913 do_task(context);
4914 do_resume(_thread->osthread());
4915 }
4916 }
4917
4918 class PcFetcher : public os::SuspendedThreadTask {
4919 public:
4920 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4921 ExtendedPC result();
4922 protected:
4923 void do_task(const os::SuspendedThreadTaskContext& context);
4924 private:
4925 ExtendedPC _epc;
4926 };
4927
4928 ExtendedPC PcFetcher::result() {
4929 guarantee(is_done(), "task is not done yet.");
4930 return _epc;
4931 }
4932
4933 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4934 Thread* thread = context.thread();
4935 OSThread* osthread = thread->osthread();
4936 if (osthread->ucontext() != NULL) {
4937 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4938 } else {
4939 // NULL context is unexpected, double-check this is the VMThread
4940 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4941 }
4942 }
4943
4944 // Suspends the target using the signal mechanism and then grabs the PC before
4945 // resuming the target. Used by the flat-profiler only
4946 ExtendedPC os::get_thread_pc(Thread* thread) {
4947 // Make sure that it is called by the watcher for the VMThread
4948 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4949 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4950
4951 PcFetcher fetcher(thread);
4952 fetcher.run();
4953 return fetcher.result();
4954 }
4955
4956 ////////////////////////////////////////////////////////////////////////////////
4957 // debug support
4958
4959 bool os::find(address addr, outputStream* st) {
4960 Dl_info dlinfo;
4961 memset(&dlinfo, 0, sizeof(dlinfo));
4962 if (dladdr(addr, &dlinfo) != 0) {
4963 st->print(PTR_FORMAT ": ", p2i(addr));
4964 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4965 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4966 p2i(addr) - p2i(dlinfo.dli_saddr));
4967 } else if (dlinfo.dli_fbase != NULL) {
4968 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4969 } else {
4970 st->print("<absolute address>");
4971 }
4972 if (dlinfo.dli_fname != NULL) {
4973 st->print(" in %s", dlinfo.dli_fname);
4974 }
4975 if (dlinfo.dli_fbase != NULL) {
4976 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4977 }
4978 st->cr();
4979
4980 if (Verbose) {
4981 // decode some bytes around the PC
4982 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4983 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4984 address lowest = (address) dlinfo.dli_sname;
4985 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4986 if (begin < lowest) begin = lowest;
4987 Dl_info dlinfo2;
4988 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4989 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4990 end = (address) dlinfo2.dli_saddr;
4991 }
4992 Disassembler::decode(begin, end, st);
4993 }
4994 return true;
4995 }
4996 return false;
4997 }
4998
4999 ////////////////////////////////////////////////////////////////////////////////
5000 // misc
5001
5002 // This does not do anything on Linux. This is basically a hook for being
5003 // able to use structured exception handling (thread-local exception filters)
5004 // on, e.g., Win32.
5005 void
5006 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5007 JavaCallArguments* args, Thread* thread) {
5008 f(value, method, args, thread);
5009 }
5010
5011 void os::print_statistics() {
5012 }
5013
5014 bool os::message_box(const char* title, const char* message) {
5015 int i;
5016 fdStream err(defaultStream::error_fd());
5017 for (i = 0; i < 78; i++) err.print_raw("=");
5018 err.cr();
5019 err.print_raw_cr(title);
5020 for (i = 0; i < 78; i++) err.print_raw("-");
5021 err.cr();
5022 err.print_raw_cr(message);
5023 for (i = 0; i < 78; i++) err.print_raw("=");
5024 err.cr();
5025
5026 char buf[16];
5027 // Prevent process from exiting upon "read error" without consuming all CPU
5028 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5029
5030 return buf[0] == 'y' || buf[0] == 'Y';
5031 }
5032
5033 int os::stat(const char *path, struct stat *sbuf) {
5034 char pathbuf[MAX_PATH];
5035 if (strlen(path) > MAX_PATH - 1) {
5036 errno = ENAMETOOLONG;
5037 return -1;
5038 }
5039 os::native_path(strcpy(pathbuf, path));
5040 return ::stat(pathbuf, sbuf);
5041 }
5042
5043 bool os::check_heap(bool force) {
5044 return true;
5045 }
5046
5047 // Is a (classpath) directory empty?
5048 bool os::dir_is_empty(const char* path) {
5049 DIR *dir = NULL;
5050 struct dirent *ptr;
5051
5052 dir = opendir(path);
5053 if (dir == NULL) return true;
5054
5055 // Scan the directory
5056 bool result = true;
5057 char buf[sizeof(struct dirent) + MAX_PATH];
5058 while (result && (ptr = ::readdir(dir)) != NULL) {
5059 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5060 result = false;
5061 }
5062 }
5063 closedir(dir);
5064 return result;
5065 }
5066
5067 // This code originates from JDK's sysOpen and open64_w
5068 // from src/solaris/hpi/src/system_md.c
5069
5070 int os::open(const char *path, int oflag, int mode) {
5071 if (strlen(path) > MAX_PATH - 1) {
5072 errno = ENAMETOOLONG;
5073 return -1;
5074 }
5075
5076 // All file descriptors that are opened in the Java process and not
5077 // specifically destined for a subprocess should have the close-on-exec
5078 // flag set. If we don't set it, then careless 3rd party native code
5079 // might fork and exec without closing all appropriate file descriptors
5080 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5081 // turn might:
5082 //
5083 // - cause end-of-file to fail to be detected on some file
5084 // descriptors, resulting in mysterious hangs, or
5085 //
5086 // - might cause an fopen in the subprocess to fail on a system
5087 // suffering from bug 1085341.
5088 //
5089 // (Yes, the default setting of the close-on-exec flag is a Unix
5090 // design flaw)
5091 //
5092 // See:
5093 // 1085341: 32-bit stdio routines should support file descriptors >255
5094 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5095 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5096 //
5097 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5098 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5099 // because it saves a system call and removes a small window where the flag
5100 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5101 // and we fall back to using FD_CLOEXEC (see below).
5102 #ifdef O_CLOEXEC
5103 oflag |= O_CLOEXEC;
5104 #endif
5105
5106 int fd = ::open64(path, oflag, mode);
5107 if (fd == -1) return -1;
5108
5109 //If the open succeeded, the file might still be a directory
5110 {
5111 struct stat64 buf64;
5112 int ret = ::fstat64(fd, &buf64);
5113 int st_mode = buf64.st_mode;
5114
5115 if (ret != -1) {
5116 if ((st_mode & S_IFMT) == S_IFDIR) {
5117 errno = EISDIR;
5118 ::close(fd);
5119 return -1;
5120 }
5121 } else {
5122 ::close(fd);
5123 return -1;
5124 }
5125 }
5126
5127 #ifdef FD_CLOEXEC
5128 // Validate that the use of the O_CLOEXEC flag on open above worked.
5129 // With recent kernels, we will perform this check exactly once.
5130 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5131 if (!O_CLOEXEC_is_known_to_work) {
5132 int flags = ::fcntl(fd, F_GETFD);
5133 if (flags != -1) {
5134 if ((flags & FD_CLOEXEC) != 0)
5135 O_CLOEXEC_is_known_to_work = 1;
5136 else
5137 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5138 }
5139 }
5140 #endif
5141
5142 return fd;
5143 }
5144
5145
5146 // create binary file, rewriting existing file if required
5147 int os::create_binary_file(const char* path, bool rewrite_existing) {
5148 int oflags = O_WRONLY | O_CREAT;
5149 if (!rewrite_existing) {
5150 oflags |= O_EXCL;
5151 }
5152 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5153 }
5154
5155 // return current position of file pointer
5156 jlong os::current_file_offset(int fd) {
5157 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5158 }
5159
5160 // move file pointer to the specified offset
5161 jlong os::seek_to_file_offset(int fd, jlong offset) {
5162 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5163 }
5164
5165 // This code originates from JDK's sysAvailable
5166 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5167
5168 int os::available(int fd, jlong *bytes) {
5169 jlong cur, end;
5170 int mode;
5171 struct stat64 buf64;
5172
5173 if (::fstat64(fd, &buf64) >= 0) {
5174 mode = buf64.st_mode;
5175 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5176 int n;
5177 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5178 *bytes = n;
5179 return 1;
5180 }
5181 }
5182 }
5183 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5184 return 0;
5185 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5186 return 0;
5187 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5188 return 0;
5189 }
5190 *bytes = end - cur;
5191 return 1;
5192 }
5193
5194 // Map a block of memory.
5195 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5196 char *addr, size_t bytes, bool read_only,
5197 bool allow_exec) {
5198 int prot;
5199 int flags = MAP_PRIVATE;
5200
5201 if (read_only) {
5202 prot = PROT_READ;
5203 } else {
5204 prot = PROT_READ | PROT_WRITE;
5205 }
5206
5207 if (allow_exec) {
5208 prot |= PROT_EXEC;
5209 }
5210
5211 if (addr != NULL) {
5212 flags |= MAP_FIXED;
5213 }
5214
5215 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5216 fd, file_offset);
5217 if (mapped_address == MAP_FAILED) {
5218 return NULL;
5219 }
5220 return mapped_address;
5221 }
5222
5223
5224 // Remap a block of memory.
5225 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5226 char *addr, size_t bytes, bool read_only,
5227 bool allow_exec) {
5228 // same as map_memory() on this OS
5229 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5230 allow_exec);
5231 }
5232
5233
5234 // Unmap a block of memory.
5235 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5236 return munmap(addr, bytes) == 0;
5237 }
5238
5239 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5240
5241 static clockid_t thread_cpu_clockid(Thread* thread) {
5242 pthread_t tid = thread->osthread()->pthread_id();
5243 clockid_t clockid;
5244
5245 // Get thread clockid
5246 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5247 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5248 return clockid;
5249 }
5250
5251 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5252 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5253 // of a thread.
5254 //
5255 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5256 // the fast estimate available on the platform.
5257
5258 jlong os::current_thread_cpu_time() {
5259 if (os::Linux::supports_fast_thread_cpu_time()) {
5260 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5261 } else {
5262 // return user + sys since the cost is the same
5263 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5264 }
5265 }
5266
5267 jlong os::thread_cpu_time(Thread* thread) {
5268 // consistent with what current_thread_cpu_time() returns
5269 if (os::Linux::supports_fast_thread_cpu_time()) {
5270 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5271 } else {
5272 return slow_thread_cpu_time(thread, true /* user + sys */);
5273 }
5274 }
5275
5276 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5277 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5278 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5279 } else {
5280 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5281 }
5282 }
5283
5284 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5285 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5286 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5287 } else {
5288 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5289 }
5290 }
5291
5292 // -1 on error.
5293 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5294 pid_t tid = thread->osthread()->thread_id();
5295 char *s;
5296 char stat[2048];
5297 int statlen;
5298 char proc_name[64];
5299 int count;
5300 long sys_time, user_time;
5301 char cdummy;
5302 int idummy;
5303 long ldummy;
5304 FILE *fp;
5305
5306 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5307 fp = fopen(proc_name, "r");
5308 if (fp == NULL) return -1;
5309 statlen = fread(stat, 1, 2047, fp);
5310 stat[statlen] = '\0';
5311 fclose(fp);
5312
5313 // Skip pid and the command string. Note that we could be dealing with
5314 // weird command names, e.g. user could decide to rename java launcher
5315 // to "java 1.4.2 :)", then the stat file would look like
5316 // 1234 (java 1.4.2 :)) R ... ...
5317 // We don't really need to know the command string, just find the last
5318 // occurrence of ")" and then start parsing from there. See bug 4726580.
5319 s = strrchr(stat, ')');
5320 if (s == NULL) return -1;
5321
5322 // Skip blank chars
5323 do { s++; } while (s && isspace(*s));
5324
5325 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5326 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5327 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5328 &user_time, &sys_time);
5329 if (count != 13) return -1;
5330 if (user_sys_cpu_time) {
5331 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5332 } else {
5333 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5334 }
5335 }
5336
5337 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5338 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5339 info_ptr->may_skip_backward = false; // elapsed time not wall time
5340 info_ptr->may_skip_forward = false; // elapsed time not wall time
5341 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5342 }
5343
5344 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5345 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5346 info_ptr->may_skip_backward = false; // elapsed time not wall time
5347 info_ptr->may_skip_forward = false; // elapsed time not wall time
5348 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5349 }
5350
5351 bool os::is_thread_cpu_time_supported() {
5352 return true;
5353 }
5354
5355 // System loadavg support. Returns -1 if load average cannot be obtained.
5356 // Linux doesn't yet have a (official) notion of processor sets,
5357 // so just return the system wide load average.
5358 int os::loadavg(double loadavg[], int nelem) {
5359 return ::getloadavg(loadavg, nelem);
5360 }
5361
5362 void os::pause() {
5363 char filename[MAX_PATH];
5364 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5365 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5366 } else {
5367 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5368 }
5369
5370 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5371 if (fd != -1) {
5372 struct stat buf;
5373 ::close(fd);
5374 while (::stat(filename, &buf) == 0) {
5375 (void)::poll(NULL, 0, 100);
5376 }
5377 } else {
5378 jio_fprintf(stderr,
5379 "Could not open pause file '%s', continuing immediately.\n", filename);
5380 }
5381 }
5382
5383
5384 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5385 // comment paragraphs are worth repeating here:
5386 //
5387 // Assumption:
5388 // Only one parker can exist on an event, which is why we allocate
5389 // them per-thread. Multiple unparkers can coexist.
5390 //
5391 // _Event serves as a restricted-range semaphore.
5392 // -1 : thread is blocked, i.e. there is a waiter
5393 // 0 : neutral: thread is running or ready,
5394 // could have been signaled after a wait started
5395 // 1 : signaled - thread is running or ready
5396 //
5397
5398 // utility to compute the abstime argument to timedwait:
5399 // millis is the relative timeout time
5400 // abstime will be the absolute timeout time
5401 // TODO: replace compute_abstime() with unpackTime()
5402
5403 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5404 if (millis < 0) millis = 0;
5405
5406 jlong seconds = millis / 1000;
5407 millis %= 1000;
5408 if (seconds > 50000000) { // see man cond_timedwait(3T)
5409 seconds = 50000000;
5410 }
5411
5412 if (os::supports_monotonic_clock()) {
5413 struct timespec now;
5414 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5415 assert_status(status == 0, status, "clock_gettime");
5416 abstime->tv_sec = now.tv_sec + seconds;
5417 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5418 if (nanos >= NANOSECS_PER_SEC) {
5419 abstime->tv_sec += 1;
5420 nanos -= NANOSECS_PER_SEC;
5421 }
5422 abstime->tv_nsec = nanos;
5423 } else {
5424 struct timeval now;
5425 int status = gettimeofday(&now, NULL);
5426 assert(status == 0, "gettimeofday");
5427 abstime->tv_sec = now.tv_sec + seconds;
5428 long usec = now.tv_usec + millis * 1000;
5429 if (usec >= 1000000) {
5430 abstime->tv_sec += 1;
5431 usec -= 1000000;
5432 }
5433 abstime->tv_nsec = usec * 1000;
5434 }
5435 return abstime;
5436 }
5437
5438 void os::PlatformEvent::park() { // AKA "down()"
5439 // Transitions for _Event:
5440 // -1 => -1 : illegal
5441 // 1 => 0 : pass - return immediately
5442 // 0 => -1 : block; then set _Event to 0 before returning
5443
5444 // Invariant: Only the thread associated with the Event/PlatformEvent
5445 // may call park().
5446 // TODO: assert that _Assoc != NULL or _Assoc == Self
5447 assert(_nParked == 0, "invariant");
5448
5449 int v;
5450 for (;;) {
5451 v = _Event;
5452 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5453 }
5454 guarantee(v >= 0, "invariant");
5455 if (v == 0) {
5456 // Do this the hard way by blocking ...
5457 int status = pthread_mutex_lock(_mutex);
5458 assert_status(status == 0, status, "mutex_lock");
5459 guarantee(_nParked == 0, "invariant");
5460 ++_nParked;
5461 while (_Event < 0) {
5462 status = pthread_cond_wait(_cond, _mutex);
5463 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5464 // Treat this the same as if the wait was interrupted
5465 if (status == ETIME) { status = EINTR; }
5466 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5467 }
5468 --_nParked;
5469
5470 _Event = 0;
5471 status = pthread_mutex_unlock(_mutex);
5472 assert_status(status == 0, status, "mutex_unlock");
5473 // Paranoia to ensure our locked and lock-free paths interact
5474 // correctly with each other.
5475 OrderAccess::fence();
5476 }
5477 guarantee(_Event >= 0, "invariant");
5478 }
5479
5480 int os::PlatformEvent::park(jlong millis) {
5481 // Transitions for _Event:
5482 // -1 => -1 : illegal
5483 // 1 => 0 : pass - return immediately
5484 // 0 => -1 : block; then set _Event to 0 before returning
5485
5486 guarantee(_nParked == 0, "invariant");
5487
5488 int v;
5489 for (;;) {
5490 v = _Event;
5491 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5492 }
5493 guarantee(v >= 0, "invariant");
5494 if (v != 0) return OS_OK;
5495
5496 // We do this the hard way, by blocking the thread.
5497 // Consider enforcing a minimum timeout value.
5498 struct timespec abst;
5499 compute_abstime(&abst, millis);
5500
5501 int ret = OS_TIMEOUT;
5502 int status = pthread_mutex_lock(_mutex);
5503 assert_status(status == 0, status, "mutex_lock");
5504 guarantee(_nParked == 0, "invariant");
5505 ++_nParked;
5506
5507 // Object.wait(timo) will return because of
5508 // (a) notification
5509 // (b) timeout
5510 // (c) thread.interrupt
5511 //
5512 // Thread.interrupt and object.notify{All} both call Event::set.
5513 // That is, we treat thread.interrupt as a special case of notification.
5514 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5515 // We assume all ETIME returns are valid.
5516 //
5517 // TODO: properly differentiate simultaneous notify+interrupt.
5518 // In that case, we should propagate the notify to another waiter.
5519
5520 while (_Event < 0) {
5521 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5522 assert_status(status == 0 || status == EINTR ||
5523 status == ETIME || status == ETIMEDOUT,
5524 status, "cond_timedwait");
5525 if (!FilterSpuriousWakeups) break; // previous semantics
5526 if (status == ETIME || status == ETIMEDOUT) break;
5527 // We consume and ignore EINTR and spurious wakeups.
5528 }
5529 --_nParked;
5530 if (_Event >= 0) {
5531 ret = OS_OK;
5532 }
5533 _Event = 0;
5534 status = pthread_mutex_unlock(_mutex);
5535 assert_status(status == 0, status, "mutex_unlock");
5536 assert(_nParked == 0, "invariant");
5537 // Paranoia to ensure our locked and lock-free paths interact
5538 // correctly with each other.
5539 OrderAccess::fence();
5540 return ret;
5541 }
5542
5543 void os::PlatformEvent::unpark() {
5544 // Transitions for _Event:
5545 // 0 => 1 : just return
5546 // 1 => 1 : just return
5547 // -1 => either 0 or 1; must signal target thread
5548 // That is, we can safely transition _Event from -1 to either
5549 // 0 or 1.
5550 // See also: "Semaphores in Plan 9" by Mullender & Cox
5551 //
5552 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5553 // that it will take two back-to-back park() calls for the owning
5554 // thread to block. This has the benefit of forcing a spurious return
5555 // from the first park() call after an unpark() call which will help
5556 // shake out uses of park() and unpark() without condition variables.
5557
5558 if (Atomic::xchg(1, &_Event) >= 0) return;
5559
5560 // Wait for the thread associated with the event to vacate
5561 int status = pthread_mutex_lock(_mutex);
5562 assert_status(status == 0, status, "mutex_lock");
5563 int AnyWaiters = _nParked;
5564 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5565 status = pthread_mutex_unlock(_mutex);
5566 assert_status(status == 0, status, "mutex_unlock");
5567 if (AnyWaiters != 0) {
5568 // Note that we signal() *after* dropping the lock for "immortal" Events.
5569 // This is safe and avoids a common class of futile wakeups. In rare
5570 // circumstances this can cause a thread to return prematurely from
5571 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5572 // will simply re-test the condition and re-park itself.
5573 // This provides particular benefit if the underlying platform does not
5574 // provide wait morphing.
5575 status = pthread_cond_signal(_cond);
5576 assert_status(status == 0, status, "cond_signal");
5577 }
5578 }
5579
5580
5581 // JSR166
5582 // -------------------------------------------------------
5583
5584 // The solaris and linux implementations of park/unpark are fairly
5585 // conservative for now, but can be improved. They currently use a
5586 // mutex/condvar pair, plus a a count.
5587 // Park decrements count if > 0, else does a condvar wait. Unpark
5588 // sets count to 1 and signals condvar. Only one thread ever waits
5589 // on the condvar. Contention seen when trying to park implies that someone
5590 // is unparking you, so don't wait. And spurious returns are fine, so there
5591 // is no need to track notifications.
5592
5593 // This code is common to linux and solaris and will be moved to a
5594 // common place in dolphin.
5595 //
5596 // The passed in time value is either a relative time in nanoseconds
5597 // or an absolute time in milliseconds. Either way it has to be unpacked
5598 // into suitable seconds and nanoseconds components and stored in the
5599 // given timespec structure.
5600 // Given time is a 64-bit value and the time_t used in the timespec is only
5601 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5602 // overflow if times way in the future are given. Further on Solaris versions
5603 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5604 // number of seconds, in abstime, is less than current_time + 100,000,000.
5605 // As it will be 28 years before "now + 100000000" will overflow we can
5606 // ignore overflow and just impose a hard-limit on seconds using the value
5607 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5608 // years from "now".
5609
5610 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5611 assert(time > 0, "convertTime");
5612 time_t max_secs = 0;
5613
5614 if (!os::supports_monotonic_clock() || isAbsolute) {
5615 struct timeval now;
5616 int status = gettimeofday(&now, NULL);
5617 assert(status == 0, "gettimeofday");
5618
5619 max_secs = now.tv_sec + MAX_SECS;
5620
5621 if (isAbsolute) {
5622 jlong secs = time / 1000;
5623 if (secs > max_secs) {
5624 absTime->tv_sec = max_secs;
5625 } else {
5626 absTime->tv_sec = secs;
5627 }
5628 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5629 } else {
5630 jlong secs = time / NANOSECS_PER_SEC;
5631 if (secs >= MAX_SECS) {
5632 absTime->tv_sec = max_secs;
5633 absTime->tv_nsec = 0;
5634 } else {
5635 absTime->tv_sec = now.tv_sec + secs;
5636 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5637 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5638 absTime->tv_nsec -= NANOSECS_PER_SEC;
5639 ++absTime->tv_sec; // note: this must be <= max_secs
5640 }
5641 }
5642 }
5643 } else {
5644 // must be relative using monotonic clock
5645 struct timespec now;
5646 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5647 assert_status(status == 0, status, "clock_gettime");
5648 max_secs = now.tv_sec + MAX_SECS;
5649 jlong secs = time / NANOSECS_PER_SEC;
5650 if (secs >= MAX_SECS) {
5651 absTime->tv_sec = max_secs;
5652 absTime->tv_nsec = 0;
5653 } else {
5654 absTime->tv_sec = now.tv_sec + secs;
5655 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5656 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5657 absTime->tv_nsec -= NANOSECS_PER_SEC;
5658 ++absTime->tv_sec; // note: this must be <= max_secs
5659 }
5660 }
5661 }
5662 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5663 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5664 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5665 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5666 }
5667
5668 void Parker::park(bool isAbsolute, jlong time) {
5669 // Ideally we'd do something useful while spinning, such
5670 // as calling unpackTime().
5671
5672 // Optional fast-path check:
5673 // Return immediately if a permit is available.
5674 // We depend on Atomic::xchg() having full barrier semantics
5675 // since we are doing a lock-free update to _counter.
5676 if (Atomic::xchg(0, &_counter) > 0) return;
5677
5678 Thread* thread = Thread::current();
5679 assert(thread->is_Java_thread(), "Must be JavaThread");
5680 JavaThread *jt = (JavaThread *)thread;
5681
5682 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5683 // Check interrupt before trying to wait
5684 if (Thread::is_interrupted(thread, false)) {
5685 return;
5686 }
5687
5688 // Next, demultiplex/decode time arguments
5689 timespec absTime;
5690 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5691 return;
5692 }
5693 if (time > 0) {
5694 unpackTime(&absTime, isAbsolute, time);
5695 }
5696
5697
5698 // Enter safepoint region
5699 // Beware of deadlocks such as 6317397.
5700 // The per-thread Parker:: mutex is a classic leaf-lock.
5701 // In particular a thread must never block on the Threads_lock while
5702 // holding the Parker:: mutex. If safepoints are pending both the
5703 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5704 ThreadBlockInVM tbivm(jt);
5705
5706 // Don't wait if cannot get lock since interference arises from
5707 // unblocking. Also. check interrupt before trying wait
5708 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5709 return;
5710 }
5711
5712 int status;
5713 if (_counter > 0) { // no wait needed
5714 _counter = 0;
5715 status = pthread_mutex_unlock(_mutex);
5716 assert_status(status == 0, status, "invariant");
5717 // Paranoia to ensure our locked and lock-free paths interact
5718 // correctly with each other and Java-level accesses.
5719 OrderAccess::fence();
5720 return;
5721 }
5722
5723 #ifdef ASSERT
5724 // Don't catch signals while blocked; let the running threads have the signals.
5725 // (This allows a debugger to break into the running thread.)
5726 sigset_t oldsigs;
5727 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5728 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5729 #endif
5730
5731 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5732 jt->set_suspend_equivalent();
5733 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5734
5735 assert(_cur_index == -1, "invariant");
5736 if (time == 0) {
5737 _cur_index = REL_INDEX; // arbitrary choice when not timed
5738 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5739 } else {
5740 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5741 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5742 }
5743 _cur_index = -1;
5744 assert_status(status == 0 || status == EINTR ||
5745 status == ETIME || status == ETIMEDOUT,
5746 status, "cond_timedwait");
5747
5748 #ifdef ASSERT
5749 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5750 #endif
5751
5752 _counter = 0;
5753 status = pthread_mutex_unlock(_mutex);
5754 assert_status(status == 0, status, "invariant");
5755 // Paranoia to ensure our locked and lock-free paths interact
5756 // correctly with each other and Java-level accesses.
5757 OrderAccess::fence();
5758
5759 // If externally suspended while waiting, re-suspend
5760 if (jt->handle_special_suspend_equivalent_condition()) {
5761 jt->java_suspend_self();
5762 }
5763 }
5764
5765 void Parker::unpark() {
5766 int status = pthread_mutex_lock(_mutex);
5767 assert_status(status == 0, status, "invariant");
5768 const int s = _counter;
5769 _counter = 1;
5770 // must capture correct index before unlocking
5771 int index = _cur_index;
5772 status = pthread_mutex_unlock(_mutex);
5773 assert_status(status == 0, status, "invariant");
5774 if (s < 1 && index != -1) {
5775 // thread is definitely parked
5776 status = pthread_cond_signal(&_cond[index]);
5777 assert_status(status == 0, status, "invariant");
5778 }
5779 }
5780
5781
5782 extern char** environ;
5783
5784 // Run the specified command in a separate process. Return its exit value,
5785 // or -1 on failure (e.g. can't fork a new process).
5786 // Unlike system(), this function can be called from signal handler. It
5787 // doesn't block SIGINT et al.
5788 int os::fork_and_exec(char* cmd) {
5789 const char * argv[4] = {"sh", "-c", cmd, NULL};
5790
5791 pid_t pid = fork();
5792
5793 if (pid < 0) {
5794 // fork failed
5795 return -1;
5796
5797 } else if (pid == 0) {
5798 // child process
5799
5800 execve("/bin/sh", (char* const*)argv, environ);
5801
5802 // execve failed
5803 _exit(-1);
5804
5805 } else {
5806 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5807 // care about the actual exit code, for now.
5808
5809 int status;
5810
5811 // Wait for the child process to exit. This returns immediately if
5812 // the child has already exited. */
5813 while (waitpid(pid, &status, 0) < 0) {
5814 switch (errno) {
5815 case ECHILD: return 0;
5816 case EINTR: break;
5817 default: return -1;
5818 }
5819 }
5820
5821 if (WIFEXITED(status)) {
5822 // The child exited normally; get its exit code.
5823 return WEXITSTATUS(status);
5824 } else if (WIFSIGNALED(status)) {
5825 // The child exited because of a signal
5826 // The best value to return is 0x80 + signal number,
5827 // because that is what all Unix shells do, and because
5828 // it allows callers to distinguish between process exit and
5829 // process death by signal.
5830 return 0x80 + WTERMSIG(status);
5831 } else {
5832 // Unknown exit code; pass it through
5833 return status;
5834 }
5835 }
5836 }
5837
5838 // is_headless_jre()
5839 //
5840 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5841 // in order to report if we are running in a headless jre
5842 //
5843 // Since JDK8 xawt/libmawt.so was moved into the same directory
5844 // as libawt.so, and renamed libawt_xawt.so
5845 //
5846 bool os::is_headless_jre() {
5847 struct stat statbuf;
5848 char buf[MAXPATHLEN];
5849 char libmawtpath[MAXPATHLEN];
5850 const char *xawtstr = "/xawt/libmawt.so";
5851 const char *new_xawtstr = "/libawt_xawt.so";
5852 char *p;
5853
5854 // Get path to libjvm.so
5855 os::jvm_path(buf, sizeof(buf));
5856
5857 // Get rid of libjvm.so
5858 p = strrchr(buf, '/');
5859 if (p == NULL) {
5860 return false;
5861 } else {
5862 *p = '\0';
5863 }
5864
5865 // Get rid of client or server
5866 p = strrchr(buf, '/');
5867 if (p == NULL) {
5868 return false;
5869 } else {
5870 *p = '\0';
5871 }
5872
5873 // check xawt/libmawt.so
5874 strcpy(libmawtpath, buf);
5875 strcat(libmawtpath, xawtstr);
5876 if (::stat(libmawtpath, &statbuf) == 0) return false;
5877
5878 // check libawt_xawt.so
5879 strcpy(libmawtpath, buf);
5880 strcat(libmawtpath, new_xawtstr);
5881 if (::stat(libmawtpath, &statbuf) == 0) return false;
5882
5883 return true;
5884 }
5885
5886 // Get the default path to the core file
5887 // Returns the length of the string
5888 int os::get_core_path(char* buffer, size_t bufferSize) {
5889 /*
5890 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5891 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5892 */
5893 const int core_pattern_len = 129;
5894 char core_pattern[core_pattern_len] = {0};
5895
5896 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5897 if (core_pattern_file == -1) {
5898 return -1;
5899 }
5900
5901 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5902 ::close(core_pattern_file);
5903 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5904 return -1;
5905 }
5906 if (core_pattern[ret-1] == '\n') {
5907 core_pattern[ret-1] = '\0';
5908 } else {
5909 core_pattern[ret] = '\0';
5910 }
5911
5912 char *pid_pos = strstr(core_pattern, "%p");
5913 int written;
5914
5915 if (core_pattern[0] == '/') {
5916 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5917 } else {
5918 char cwd[PATH_MAX];
5919
5920 const char* p = get_current_directory(cwd, PATH_MAX);
5921 if (p == NULL) {
5922 return -1;
5923 }
5924
5925 if (core_pattern[0] == '|') {
5926 written = jio_snprintf(buffer, bufferSize,
5927 "\"%s\" (or dumping to %s/core.%d)",
5928 &core_pattern[1], p, current_process_id());
5929 } else {
5930 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5931 }
5932 }
5933
5934 if (written < 0) {
5935 return -1;
5936 }
5937
5938 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5939 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5940
5941 if (core_uses_pid_file != -1) {
5942 char core_uses_pid = 0;
5943 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5944 ::close(core_uses_pid_file);
5945
5946 if (core_uses_pid == '1') {
5947 jio_snprintf(buffer + written, bufferSize - written,
5948 ".%d", current_process_id());
5949 }
5950 }
5951 }
5952
5953 return strlen(buffer);
5954 }
5955
5956 bool os::start_debugging(char *buf, int buflen) {
5957 int len = (int)strlen(buf);
5958 char *p = &buf[len];
5959
5960 jio_snprintf(p, buflen-len,
5961 "\n\n"
5962 "Do you want to debug the problem?\n\n"
5963 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5964 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5965 "Otherwise, press RETURN to abort...",
5966 os::current_process_id(), os::current_process_id(),
5967 os::current_thread_id(), os::current_thread_id());
5968
5969 bool yes = os::message_box("Unexpected Error", buf);
5970
5971 if (yes) {
5972 // yes, user asked VM to launch debugger
5973 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
5974 os::current_process_id(), os::current_process_id());
5975
5976 os::fork_and_exec(buf);
5977 yes = false;
5978 }
5979 return yes;
5980 }
5981
5982
5983
5984 /////////////// Unit tests ///////////////
5985
5986 #ifndef PRODUCT
5987
5988 #define test_log(...) \
5989 do { \
5990 if (VerboseInternalVMTests) { \
5991 tty->print_cr(__VA_ARGS__); \
5992 tty->flush(); \
5993 } \
5994 } while (false)
5995
5996 class TestReserveMemorySpecial : AllStatic {
5997 public:
5998 static void small_page_write(void* addr, size_t size) {
5999 size_t page_size = os::vm_page_size();
6000
6001 char* end = (char*)addr + size;
6002 for (char* p = (char*)addr; p < end; p += page_size) {
6003 *p = 1;
6004 }
6005 }
6006
6007 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6008 if (!UseHugeTLBFS) {
6009 return;
6010 }
6011
6012 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6013
6014 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6015
6016 if (addr != NULL) {
6017 small_page_write(addr, size);
6018
6019 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6020 }
6021 }
6022
6023 static void test_reserve_memory_special_huge_tlbfs_only() {
6024 if (!UseHugeTLBFS) {
6025 return;
6026 }
6027
6028 size_t lp = os::large_page_size();
6029
6030 for (size_t size = lp; size <= lp * 10; size += lp) {
6031 test_reserve_memory_special_huge_tlbfs_only(size);
6032 }
6033 }
6034
6035 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6036 size_t lp = os::large_page_size();
6037 size_t ag = os::vm_allocation_granularity();
6038
6039 // sizes to test
6040 const size_t sizes[] = {
6041 lp, lp + ag, lp + lp / 2, lp * 2,
6042 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6043 lp * 10, lp * 10 + lp / 2
6044 };
6045 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6046
6047 // For each size/alignment combination, we test three scenarios:
6048 // 1) with req_addr == NULL
6049 // 2) with a non-null req_addr at which we expect to successfully allocate
6050 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6051 // expect the allocation to either fail or to ignore req_addr
6052
6053 // Pre-allocate two areas; they shall be as large as the largest allocation
6054 // and aligned to the largest alignment we will be testing.
6055 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6056 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6057 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6058 -1, 0);
6059 assert(mapping1 != MAP_FAILED, "should work");
6060
6061 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6062 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6063 -1, 0);
6064 assert(mapping2 != MAP_FAILED, "should work");
6065
6066 // Unmap the first mapping, but leave the second mapping intact: the first
6067 // mapping will serve as a value for a "good" req_addr (case 2). The second
6068 // mapping, still intact, as "bad" req_addr (case 3).
6069 ::munmap(mapping1, mapping_size);
6070
6071 // Case 1
6072 test_log("%s, req_addr NULL:", __FUNCTION__);
6073 test_log("size align result");
6074
6075 for (int i = 0; i < num_sizes; i++) {
6076 const size_t size = sizes[i];
6077 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6078 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6079 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6080 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6081 if (p != NULL) {
6082 assert(is_ptr_aligned(p, alignment), "must be");
6083 small_page_write(p, size);
6084 os::Linux::release_memory_special_huge_tlbfs(p, size);
6085 }
6086 }
6087 }
6088
6089 // Case 2
6090 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6091 test_log("size align req_addr result");
6092
6093 for (int i = 0; i < num_sizes; i++) {
6094 const size_t size = sizes[i];
6095 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6096 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6097 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6098 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6099 size, alignment, p2i(req_addr), p2i(p),
6100 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6101 if (p != NULL) {
6102 assert(p == req_addr, "must be");
6103 small_page_write(p, size);
6104 os::Linux::release_memory_special_huge_tlbfs(p, size);
6105 }
6106 }
6107 }
6108
6109 // Case 3
6110 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6111 test_log("size align req_addr result");
6112
6113 for (int i = 0; i < num_sizes; i++) {
6114 const size_t size = sizes[i];
6115 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6116 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6117 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6118 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6119 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6120 // as the area around req_addr contains already existing mappings, the API should always
6121 // return NULL (as per contract, it cannot return another address)
6122 assert(p == NULL, "must be");
6123 }
6124 }
6125
6126 ::munmap(mapping2, mapping_size);
6127
6128 }
6129
6130 static void test_reserve_memory_special_huge_tlbfs() {
6131 if (!UseHugeTLBFS) {
6132 return;
6133 }
6134
6135 test_reserve_memory_special_huge_tlbfs_only();
6136 test_reserve_memory_special_huge_tlbfs_mixed();
6137 }
6138
6139 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6140 if (!UseSHM) {
6141 return;
6142 }
6143
6144 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6145
6146 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6147
6148 if (addr != NULL) {
6149 assert(is_ptr_aligned(addr, alignment), "Check");
6150 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6151
6152 small_page_write(addr, size);
6153
6154 os::Linux::release_memory_special_shm(addr, size);
6155 }
6156 }
6157
6158 static void test_reserve_memory_special_shm() {
6159 size_t lp = os::large_page_size();
6160 size_t ag = os::vm_allocation_granularity();
6161
6162 for (size_t size = ag; size < lp * 3; size += ag) {
6163 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6164 test_reserve_memory_special_shm(size, alignment);
6165 }
6166 }
6167 }
6168
6169 static void test() {
6170 test_reserve_memory_special_huge_tlbfs();
6171 test_reserve_memory_special_shm();
6172 }
6173 };
6174
6175 void TestReserveMemorySpecial_test() {
6176 TestReserveMemorySpecial::test();
6177 }
6178
6179 #endif