1 /*
2 * Copyright (c) 1999, 2017, 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 "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 #ifndef _GNU_SOURCE
110 #define _GNU_SOURCE
111 #include <sched.h>
112 #undef _GNU_SOURCE
113 #else
114 #include <sched.h>
115 #endif
116
117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
118 // getrusage() is prepared to handle the associated failure.
119 #ifndef RUSAGE_THREAD
120 #define RUSAGE_THREAD (1) /* only the calling thread */
121 #endif
122
123 #define MAX_PATH (2 * K)
124
125 #define MAX_SECS 100000000
126
127 // for timer info max values which include all bits
128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129
130 #define LARGEPAGES_BIT (1 << 6)
131 ////////////////////////////////////////////////////////////////////////////////
132 // global variables
133 julong os::Linux::_physical_memory = 0;
134
135 address os::Linux::_initial_thread_stack_bottom = NULL;
136 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137
138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
141 Mutex* os::Linux::_createThread_lock = NULL;
142 pthread_t os::Linux::_main_thread;
143 int os::Linux::_page_size = -1;
144 bool os::Linux::_supports_fast_thread_cpu_time = false;
145 uint32_t os::Linux::_os_version = 0;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148 pthread_condattr_t os::Linux::_condattr[1];
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 // Signal number used to suspend/resume a thread
159
160 // do not use any signal number less than SIGSEGV, see 4355769
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 // Return true if user is running as root.
188
189 bool os::have_special_privileges() {
190 static bool init = false;
191 static bool privileges = false;
192 if (!init) {
193 privileges = (getuid() != geteuid()) || (getgid() != getegid());
194 init = true;
195 }
196 return privileges;
197 }
198
199
200 #ifndef SYS_gettid
201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
202 #ifdef __ia64__
203 #define SYS_gettid 1105
204 #else
205 #ifdef __i386__
206 #define SYS_gettid 224
207 #else
208 #ifdef __amd64__
209 #define SYS_gettid 186
210 #else
211 #ifdef __sparc__
212 #define SYS_gettid 143
213 #else
214 #error define gettid for the arch
215 #endif
216 #endif
217 #endif
218 #endif
219 #endif
220
221
222 // pid_t gettid()
223 //
224 // Returns the kernel thread id of the currently running thread. Kernel
225 // thread id is used to access /proc.
226 pid_t os::Linux::gettid() {
227 int rslt = syscall(SYS_gettid);
228 assert(rslt != -1, "must be."); // old linuxthreads implementation?
229 return (pid_t)rslt;
230 }
231
232 // Most versions of linux have a bug where the number of processors are
233 // determined by looking at the /proc file system. In a chroot environment,
234 // the system call returns 1. This causes the VM to act as if it is
235 // a single processor and elide locking (see is_MP() call).
236 static bool unsafe_chroot_detected = false;
237 static const char *unstable_chroot_error = "/proc file system not found.\n"
238 "Java may be unstable running multithreaded in a chroot "
239 "environment on Linux when /proc filesystem is not mounted.";
240
241 void os::Linux::initialize_system_info() {
242 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
243 if (processor_count() == 1) {
244 pid_t pid = os::Linux::gettid();
245 char fname[32];
246 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
247 FILE *fp = fopen(fname, "r");
248 if (fp == NULL) {
249 unsafe_chroot_detected = true;
250 } else {
251 fclose(fp);
252 }
253 }
254 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
255 assert(processor_count() > 0, "linux error");
256 }
257
258 void os::init_system_properties_values() {
259 // The next steps are taken in the product version:
260 //
261 // Obtain the JAVA_HOME value from the location of libjvm.so.
262 // This library should be located at:
263 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
264 //
265 // If "/jre/lib/" appears at the right place in the path, then we
266 // assume libjvm.so is installed in a JDK and we use this path.
267 //
268 // Otherwise exit with message: "Could not create the Java virtual machine."
269 //
270 // The following extra steps are taken in the debugging version:
271 //
272 // If "/jre/lib/" does NOT appear at the right place in the path
273 // instead of exit check for $JAVA_HOME environment variable.
274 //
275 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
276 // then we append a fake suffix "hotspot/libjvm.so" to this path so
277 // it looks like libjvm.so is installed there
278 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
279 //
280 // Otherwise exit.
281 //
282 // Important note: if the location of libjvm.so changes this
283 // code needs to be changed accordingly.
284
285 // See ld(1):
286 // The linker uses the following search paths to locate required
287 // shared libraries:
288 // 1: ...
289 // ...
290 // 7: The default directories, normally /lib and /usr/lib.
291 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
292 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
293 #else
294 #define DEFAULT_LIBPATH "/lib:/usr/lib"
295 #endif
296
297 // Base path of extensions installed on the system.
298 #define SYS_EXT_DIR "/usr/java/packages"
299 #define EXTENSIONS_DIR "/lib/ext"
300
301 // Buffer that fits several sprintfs.
302 // Note that the space for the colon and the trailing null are provided
303 // by the nulls included by the sizeof operator.
304 const size_t bufsize =
305 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
306 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
307 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
308
309 // sysclasspath, java_home, dll_dir
310 {
311 char *pslash;
312 os::jvm_path(buf, bufsize);
313
314 // Found the full path to libjvm.so.
315 // Now cut the path to <java_home>/jre if we can.
316 pslash = strrchr(buf, '/');
317 if (pslash != NULL) {
318 *pslash = '\0'; // Get rid of /libjvm.so.
319 }
320 pslash = strrchr(buf, '/');
321 if (pslash != NULL) {
322 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
323 }
324 Arguments::set_dll_dir(buf);
325
326 if (pslash != NULL) {
327 pslash = strrchr(buf, '/');
328 if (pslash != NULL) {
329 *pslash = '\0'; // Get rid of /lib.
330 }
331 }
332 Arguments::set_java_home(buf);
333 set_boot_path('/', ':');
334 }
335
336 // Where to look for native libraries.
337 //
338 // Note: Due to a legacy implementation, most of the library path
339 // is set in the launcher. This was to accomodate linking restrictions
340 // on legacy Linux implementations (which are no longer supported).
341 // Eventually, all the library path setting will be done here.
342 //
343 // However, to prevent the proliferation of improperly built native
344 // libraries, the new path component /usr/java/packages is added here.
345 // Eventually, all the library path setting will be done here.
346 {
347 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
348 // should always exist (until the legacy problem cited above is
349 // addressed).
350 const char *v = ::getenv("LD_LIBRARY_PATH");
351 const char *v_colon = ":";
352 if (v == NULL) { v = ""; v_colon = ""; }
353 // That's +1 for the colon and +1 for the trailing '\0'.
354 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
355 strlen(v) + 1 +
356 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
357 mtInternal);
358 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
359 Arguments::set_library_path(ld_library_path);
360 FREE_C_HEAP_ARRAY(char, ld_library_path);
361 }
362
363 // Extensions directories.
364 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
365 Arguments::set_ext_dirs(buf);
366
367 FREE_C_HEAP_ARRAY(char, buf);
368
369 #undef DEFAULT_LIBPATH
370 #undef SYS_EXT_DIR
371 #undef EXTENSIONS_DIR
372 }
373
374 ////////////////////////////////////////////////////////////////////////////////
375 // breakpoint support
376
377 void os::breakpoint() {
378 BREAKPOINT;
379 }
380
381 extern "C" void breakpoint() {
382 // use debugger to set breakpoint here
383 }
384
385 ////////////////////////////////////////////////////////////////////////////////
386 // signal support
387
388 debug_only(static bool signal_sets_initialized = false);
389 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
390
391 bool os::Linux::is_sig_ignored(int sig) {
392 struct sigaction oact;
393 sigaction(sig, (struct sigaction*)NULL, &oact);
394 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
395 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
396 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
397 return true;
398 } else {
399 return false;
400 }
401 }
402
403 void os::Linux::signal_sets_init() {
404 // Should also have an assertion stating we are still single-threaded.
405 assert(!signal_sets_initialized, "Already initialized");
406 // Fill in signals that are necessarily unblocked for all threads in
407 // the VM. Currently, we unblock the following signals:
408 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
409 // by -Xrs (=ReduceSignalUsage));
410 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
411 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
412 // the dispositions or masks wrt these signals.
413 // Programs embedding the VM that want to use the above signals for their
414 // own purposes must, at this time, use the "-Xrs" option to prevent
415 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
416 // (See bug 4345157, and other related bugs).
417 // In reality, though, unblocking these signals is really a nop, since
418 // these signals are not blocked by default.
419 sigemptyset(&unblocked_sigs);
420 sigemptyset(&allowdebug_blocked_sigs);
421 sigaddset(&unblocked_sigs, SIGILL);
422 sigaddset(&unblocked_sigs, SIGSEGV);
423 sigaddset(&unblocked_sigs, SIGBUS);
424 sigaddset(&unblocked_sigs, SIGFPE);
425 #if defined(PPC64)
426 sigaddset(&unblocked_sigs, SIGTRAP);
427 #endif
428 sigaddset(&unblocked_sigs, SR_signum);
429
430 if (!ReduceSignalUsage) {
431 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
432 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
433 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
434 }
435 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
436 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
437 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
438 }
439 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
440 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
441 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
442 }
443 }
444 // Fill in signals that are blocked by all but the VM thread.
445 sigemptyset(&vm_sigs);
446 if (!ReduceSignalUsage) {
447 sigaddset(&vm_sigs, BREAK_SIGNAL);
448 }
449 debug_only(signal_sets_initialized = true);
450
451 }
452
453 // These are signals that are unblocked while a thread is running Java.
454 // (For some reason, they get blocked by default.)
455 sigset_t* os::Linux::unblocked_signals() {
456 assert(signal_sets_initialized, "Not initialized");
457 return &unblocked_sigs;
458 }
459
460 // These are the signals that are blocked while a (non-VM) thread is
461 // running Java. Only the VM thread handles these signals.
462 sigset_t* os::Linux::vm_signals() {
463 assert(signal_sets_initialized, "Not initialized");
464 return &vm_sigs;
465 }
466
467 // These are signals that are blocked during cond_wait to allow debugger in
468 sigset_t* os::Linux::allowdebug_blocked_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &allowdebug_blocked_sigs;
471 }
472
473 void os::Linux::hotspot_sigmask(Thread* thread) {
474
475 //Save caller's signal mask before setting VM signal mask
476 sigset_t caller_sigmask;
477 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
478
479 OSThread* osthread = thread->osthread();
480 osthread->set_caller_sigmask(caller_sigmask);
481
482 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
483
484 if (!ReduceSignalUsage) {
485 if (thread->is_VM_thread()) {
486 // Only the VM thread handles BREAK_SIGNAL ...
487 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
488 } else {
489 // ... all other threads block BREAK_SIGNAL
490 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
491 }
492 }
493 }
494
495 //////////////////////////////////////////////////////////////////////////////
496 // detecting pthread library
497
498 void os::Linux::libpthread_init() {
499 // Save glibc and pthread version strings.
500 #if !defined(_CS_GNU_LIBC_VERSION) || \
501 !defined(_CS_GNU_LIBPTHREAD_VERSION)
502 #error "glibc too old (< 2.3.2)"
503 #endif
504
505 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
506 assert(n > 0, "cannot retrieve glibc version");
507 char *str = (char *)malloc(n, mtInternal);
508 confstr(_CS_GNU_LIBC_VERSION, str, n);
509 os::Linux::set_glibc_version(str);
510
511 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
512 assert(n > 0, "cannot retrieve pthread version");
513 str = (char *)malloc(n, mtInternal);
514 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
515 os::Linux::set_libpthread_version(str);
516 }
517
518 /////////////////////////////////////////////////////////////////////////////
519 // thread stack expansion
520
521 // os::Linux::manually_expand_stack() takes care of expanding the thread
522 // stack. Note that this is normally not needed: pthread stacks allocate
523 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
524 // committed. Therefore it is not necessary to expand the stack manually.
525 //
526 // Manually expanding the stack was historically needed on LinuxThreads
527 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
528 // it is kept to deal with very rare corner cases:
529 //
530 // For one, user may run the VM on an own implementation of threads
531 // whose stacks are - like the old LinuxThreads - implemented using
532 // mmap(MAP_GROWSDOWN).
533 //
534 // Also, this coding may be needed if the VM is running on the primordial
535 // thread. Normally we avoid running on the primordial thread; however,
536 // user may still invoke the VM on the primordial thread.
537 //
538 // The following historical comment describes the details about running
539 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
540
541
542 // Force Linux kernel to expand current thread stack. If "bottom" is close
543 // to the stack guard, caller should block all signals.
544 //
545 // MAP_GROWSDOWN:
546 // A special mmap() flag that is used to implement thread stacks. It tells
547 // kernel that the memory region should extend downwards when needed. This
548 // allows early versions of LinuxThreads to only mmap the first few pages
549 // when creating a new thread. Linux kernel will automatically expand thread
550 // stack as needed (on page faults).
551 //
552 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
553 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
554 // region, it's hard to tell if the fault is due to a legitimate stack
555 // access or because of reading/writing non-exist memory (e.g. buffer
556 // overrun). As a rule, if the fault happens below current stack pointer,
557 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
558 // application (see Linux kernel fault.c).
559 //
560 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
561 // stack overflow detection.
562 //
563 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
564 // not use MAP_GROWSDOWN.
565 //
566 // To get around the problem and allow stack banging on Linux, we need to
567 // manually expand thread stack after receiving the SIGSEGV.
568 //
569 // There are two ways to expand thread stack to address "bottom", we used
570 // both of them in JVM before 1.5:
571 // 1. adjust stack pointer first so that it is below "bottom", and then
572 // touch "bottom"
573 // 2. mmap() the page in question
574 //
575 // Now alternate signal stack is gone, it's harder to use 2. For instance,
576 // if current sp is already near the lower end of page 101, and we need to
577 // call mmap() to map page 100, it is possible that part of the mmap() frame
578 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
579 // That will destroy the mmap() frame and cause VM to crash.
580 //
581 // The following code works by adjusting sp first, then accessing the "bottom"
582 // page to force a page fault. Linux kernel will then automatically expand the
583 // stack mapping.
584 //
585 // _expand_stack_to() assumes its frame size is less than page size, which
586 // should always be true if the function is not inlined.
587
588 static void NOINLINE _expand_stack_to(address bottom) {
589 address sp;
590 size_t size;
591 volatile char *p;
592
593 // Adjust bottom to point to the largest address within the same page, it
594 // gives us a one-page buffer if alloca() allocates slightly more memory.
595 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
596 bottom += os::Linux::page_size() - 1;
597
598 // sp might be slightly above current stack pointer; if that's the case, we
599 // will alloca() a little more space than necessary, which is OK. Don't use
600 // os::current_stack_pointer(), as its result can be slightly below current
601 // stack pointer, causing us to not alloca enough to reach "bottom".
602 sp = (address)&sp;
603
604 if (sp > bottom) {
605 size = sp - bottom;
606 p = (volatile char *)alloca(size);
607 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
608 p[0] = '\0';
609 }
610 }
611
612 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
613 assert(t!=NULL, "just checking");
614 assert(t->osthread()->expanding_stack(), "expand should be set");
615 assert(t->stack_base() != NULL, "stack_base was not initialized");
616
617 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
618 sigset_t mask_all, old_sigset;
619 sigfillset(&mask_all);
620 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
621 _expand_stack_to(addr);
622 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
623 return true;
624 }
625 return false;
626 }
627
628 //////////////////////////////////////////////////////////////////////////////
629 // create new thread
630
631 // Thread start routine for all newly created threads
632 static void *thread_native_entry(Thread *thread) {
633 // Try to randomize the cache line index of hot stack frames.
634 // This helps when threads of the same stack traces evict each other's
635 // cache lines. The threads can be either from the same JVM instance, or
636 // from different JVM instances. The benefit is especially true for
637 // processors with hyperthreading technology.
638 static int counter = 0;
639 int pid = os::current_process_id();
640 alloca(((pid ^ counter++) & 7) * 128);
641
642 thread->initialize_thread_current();
643
644 OSThread* osthread = thread->osthread();
645 Monitor* sync = osthread->startThread_lock();
646
647 osthread->set_thread_id(os::current_thread_id());
648
649 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
650 os::current_thread_id(), (uintx) pthread_self());
651
652 if (UseNUMA) {
653 int lgrp_id = os::numa_get_group_id();
654 if (lgrp_id != -1) {
655 thread->set_lgrp_id(lgrp_id);
656 }
657 }
658 // initialize signal mask for this thread
659 os::Linux::hotspot_sigmask(thread);
660
661 // initialize floating point control register
662 os::Linux::init_thread_fpu_state();
663
664 // handshaking with parent thread
665 {
666 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
667
668 // notify parent thread
669 osthread->set_state(INITIALIZED);
670 sync->notify_all();
671
672 // wait until os::start_thread()
673 while (osthread->get_state() == INITIALIZED) {
674 sync->wait(Mutex::_no_safepoint_check_flag);
675 }
676 }
677
678 // call one more level start routine
679 thread->run();
680
681 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
682 os::current_thread_id(), (uintx) pthread_self());
683
684 // If a thread has not deleted itself ("delete this") as part of its
685 // termination sequence, we have to ensure thread-local-storage is
686 // cleared before we actually terminate. No threads should ever be
687 // deleted asynchronously with respect to their termination.
688 if (Thread::current_or_null_safe() != NULL) {
689 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
690 thread->clear_thread_current();
691 }
692
693 return 0;
694 }
695
696 bool os::create_thread(Thread* thread, ThreadType thr_type,
697 size_t req_stack_size) {
698 assert(thread->osthread() == NULL, "caller responsible");
699
700 // Allocate the OSThread object
701 OSThread* osthread = new OSThread(NULL, NULL);
702 if (osthread == NULL) {
703 return false;
704 }
705
706 // set the correct thread state
707 osthread->set_thread_type(thr_type);
708
709 // Initial state is ALLOCATED but not INITIALIZED
710 osthread->set_state(ALLOCATED);
711
712 thread->set_osthread(osthread);
713
714 // init thread attributes
715 pthread_attr_t attr;
716 pthread_attr_init(&attr);
717 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
718
719 // Calculate stack size if it's not specified by caller.
720 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
721 // In the Linux NPTL pthread implementation the guard size mechanism
722 // is not implemented properly. The posix standard requires adding
723 // the size of the guard pages to the stack size, instead Linux
724 // takes the space out of 'stacksize'. Thus we adapt the requested
725 // stack_size by the size of the guard pages to mimick proper
726 // behaviour. However, be careful not to end up with a size
727 // of zero due to overflow. Don't add the guard page in that case.
728 size_t guard_size = os::Linux::default_guard_size(thr_type);
729 if (stack_size <= SIZE_MAX - guard_size) {
730 stack_size += guard_size;
731 }
732 assert(is_size_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
733
734 int status = pthread_attr_setstacksize(&attr, stack_size);
735 assert_status(status == 0, status, "pthread_attr_setstacksize");
736
737 // Configure glibc guard page.
738 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
739
740 ThreadState state;
741
742 {
743 pthread_t tid;
744 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
745
746 char buf[64];
747 if (ret == 0) {
748 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
749 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
750 } else {
751 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
752 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
753 }
754
755 pthread_attr_destroy(&attr);
756
757 if (ret != 0) {
758 // Need to clean up stuff we've allocated so far
759 thread->set_osthread(NULL);
760 delete osthread;
761 return false;
762 }
763
764 // Store pthread info into the OSThread
765 osthread->set_pthread_id(tid);
766
767 // Wait until child thread is either initialized or aborted
768 {
769 Monitor* sync_with_child = osthread->startThread_lock();
770 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
771 while ((state = osthread->get_state()) == ALLOCATED) {
772 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
773 }
774 }
775 }
776
777 // Aborted due to thread limit being reached
778 if (state == ZOMBIE) {
779 thread->set_osthread(NULL);
780 delete osthread;
781 return false;
782 }
783
784 // The thread is returned suspended (in state INITIALIZED),
785 // and is started higher up in the call chain
786 assert(state == INITIALIZED, "race condition");
787 return true;
788 }
789
790 /////////////////////////////////////////////////////////////////////////////
791 // attach existing thread
792
793 // bootstrap the main thread
794 bool os::create_main_thread(JavaThread* thread) {
795 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
796 return create_attached_thread(thread);
797 }
798
799 bool os::create_attached_thread(JavaThread* thread) {
800 #ifdef ASSERT
801 thread->verify_not_published();
802 #endif
803
804 // Allocate the OSThread object
805 OSThread* osthread = new OSThread(NULL, NULL);
806
807 if (osthread == NULL) {
808 return false;
809 }
810
811 // Store pthread info into the OSThread
812 osthread->set_thread_id(os::Linux::gettid());
813 osthread->set_pthread_id(::pthread_self());
814
815 // initialize floating point control register
816 os::Linux::init_thread_fpu_state();
817
818 // Initial thread state is RUNNABLE
819 osthread->set_state(RUNNABLE);
820
821 thread->set_osthread(osthread);
822
823 if (UseNUMA) {
824 int lgrp_id = os::numa_get_group_id();
825 if (lgrp_id != -1) {
826 thread->set_lgrp_id(lgrp_id);
827 }
828 }
829
830 if (os::Linux::is_initial_thread()) {
831 // If current thread is initial thread, its stack is mapped on demand,
832 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
833 // the entire stack region to avoid SEGV in stack banging.
834 // It is also useful to get around the heap-stack-gap problem on SuSE
835 // kernel (see 4821821 for details). We first expand stack to the top
836 // of yellow zone, then enable stack yellow zone (order is significant,
837 // enabling yellow zone first will crash JVM on SuSE Linux), so there
838 // is no gap between the last two virtual memory regions.
839
840 JavaThread *jt = (JavaThread *)thread;
841 address addr = jt->stack_reserved_zone_base();
842 assert(addr != NULL, "initialization problem?");
843 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
844
845 osthread->set_expanding_stack();
846 os::Linux::manually_expand_stack(jt, addr);
847 osthread->clear_expanding_stack();
848 }
849
850 // initialize signal mask for this thread
851 // and save the caller's signal mask
852 os::Linux::hotspot_sigmask(thread);
853
854 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
855 os::current_thread_id(), (uintx) pthread_self());
856
857 return true;
858 }
859
860 void os::pd_start_thread(Thread* thread) {
861 OSThread * osthread = thread->osthread();
862 assert(osthread->get_state() != INITIALIZED, "just checking");
863 Monitor* sync_with_child = osthread->startThread_lock();
864 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
865 sync_with_child->notify();
866 }
867
868 // Free Linux resources related to the OSThread
869 void os::free_thread(OSThread* osthread) {
870 assert(osthread != NULL, "osthread not set");
871
872 // We are told to free resources of the argument thread,
873 // but we can only really operate on the current thread.
874 assert(Thread::current()->osthread() == osthread,
875 "os::free_thread but not current thread");
876
877 #ifdef ASSERT
878 sigset_t current;
879 sigemptyset(¤t);
880 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
881 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
882 #endif
883
884 // Restore caller's signal mask
885 sigset_t sigmask = osthread->caller_sigmask();
886 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
887
888 delete osthread;
889 }
890
891 //////////////////////////////////////////////////////////////////////////////
892 // initial thread
893
894 // Check if current thread is the initial thread, similar to Solaris thr_main.
895 bool os::Linux::is_initial_thread(void) {
896 char dummy;
897 // If called before init complete, thread stack bottom will be null.
898 // Can be called if fatal error occurs before initialization.
899 if (initial_thread_stack_bottom() == NULL) return false;
900 assert(initial_thread_stack_bottom() != NULL &&
901 initial_thread_stack_size() != 0,
902 "os::init did not locate initial thread's stack region");
903 if ((address)&dummy >= initial_thread_stack_bottom() &&
904 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
905 return true;
906 } else {
907 return false;
908 }
909 }
910
911 // Find the virtual memory area that contains addr
912 static bool find_vma(address addr, address* vma_low, address* vma_high) {
913 FILE *fp = fopen("/proc/self/maps", "r");
914 if (fp) {
915 address low, high;
916 while (!feof(fp)) {
917 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
918 if (low <= addr && addr < high) {
919 if (vma_low) *vma_low = low;
920 if (vma_high) *vma_high = high;
921 fclose(fp);
922 return true;
923 }
924 }
925 for (;;) {
926 int ch = fgetc(fp);
927 if (ch == EOF || ch == (int)'\n') break;
928 }
929 }
930 fclose(fp);
931 }
932 return false;
933 }
934
935 // Locate initial thread stack. This special handling of initial thread stack
936 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
937 // bogus value for the primordial process thread. While the launcher has created
938 // the VM in a new thread since JDK 6, we still have to allow for the use of the
939 // JNI invocation API from a primordial thread.
940 void os::Linux::capture_initial_stack(size_t max_size) {
941
942 // max_size is either 0 (which means accept OS default for thread stacks) or
943 // a user-specified value known to be at least the minimum needed. If we
944 // are actually on the primordial thread we can make it appear that we have a
945 // smaller max_size stack by inserting the guard pages at that location. But we
946 // cannot do anything to emulate a larger stack than what has been provided by
947 // the OS or threading library. In fact if we try to use a stack greater than
948 // what is set by rlimit then we will crash the hosting process.
949
950 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
951 // If this is "unlimited" then it will be a huge value.
952 struct rlimit rlim;
953 getrlimit(RLIMIT_STACK, &rlim);
954 size_t stack_size = rlim.rlim_cur;
955
956 // 6308388: a bug in ld.so will relocate its own .data section to the
957 // lower end of primordial stack; reduce ulimit -s value a little bit
958 // so we won't install guard page on ld.so's data section.
959 stack_size -= 2 * page_size();
960
961 // Try to figure out where the stack base (top) is. This is harder.
962 //
963 // When an application is started, glibc saves the initial stack pointer in
964 // a global variable "__libc_stack_end", which is then used by system
965 // libraries. __libc_stack_end should be pretty close to stack top. The
966 // variable is available since the very early days. However, because it is
967 // a private interface, it could disappear in the future.
968 //
969 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
970 // to __libc_stack_end, it is very close to stack top, but isn't the real
971 // stack top. Note that /proc may not exist if VM is running as a chroot
972 // program, so reading /proc/<pid>/stat could fail. Also the contents of
973 // /proc/<pid>/stat could change in the future (though unlikely).
974 //
975 // We try __libc_stack_end first. If that doesn't work, look for
976 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
977 // as a hint, which should work well in most cases.
978
979 uintptr_t stack_start;
980
981 // try __libc_stack_end first
982 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
983 if (p && *p) {
984 stack_start = *p;
985 } else {
986 // see if we can get the start_stack field from /proc/self/stat
987 FILE *fp;
988 int pid;
989 char state;
990 int ppid;
991 int pgrp;
992 int session;
993 int nr;
994 int tpgrp;
995 unsigned long flags;
996 unsigned long minflt;
997 unsigned long cminflt;
998 unsigned long majflt;
999 unsigned long cmajflt;
1000 unsigned long utime;
1001 unsigned long stime;
1002 long cutime;
1003 long cstime;
1004 long prio;
1005 long nice;
1006 long junk;
1007 long it_real;
1008 uintptr_t start;
1009 uintptr_t vsize;
1010 intptr_t rss;
1011 uintptr_t rsslim;
1012 uintptr_t scodes;
1013 uintptr_t ecode;
1014 int i;
1015
1016 // Figure what the primordial thread stack base is. Code is inspired
1017 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1018 // followed by command name surrounded by parentheses, state, etc.
1019 char stat[2048];
1020 int statlen;
1021
1022 fp = fopen("/proc/self/stat", "r");
1023 if (fp) {
1024 statlen = fread(stat, 1, 2047, fp);
1025 stat[statlen] = '\0';
1026 fclose(fp);
1027
1028 // Skip pid and the command string. Note that we could be dealing with
1029 // weird command names, e.g. user could decide to rename java launcher
1030 // to "java 1.4.2 :)", then the stat file would look like
1031 // 1234 (java 1.4.2 :)) R ... ...
1032 // We don't really need to know the command string, just find the last
1033 // occurrence of ")" and then start parsing from there. See bug 4726580.
1034 char * s = strrchr(stat, ')');
1035
1036 i = 0;
1037 if (s) {
1038 // Skip blank chars
1039 do { s++; } while (s && isspace(*s));
1040
1041 #define _UFM UINTX_FORMAT
1042 #define _DFM INTX_FORMAT
1043
1044 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1045 // 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
1046 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,
1047 &state, // 3 %c
1048 &ppid, // 4 %d
1049 &pgrp, // 5 %d
1050 &session, // 6 %d
1051 &nr, // 7 %d
1052 &tpgrp, // 8 %d
1053 &flags, // 9 %lu
1054 &minflt, // 10 %lu
1055 &cminflt, // 11 %lu
1056 &majflt, // 12 %lu
1057 &cmajflt, // 13 %lu
1058 &utime, // 14 %lu
1059 &stime, // 15 %lu
1060 &cutime, // 16 %ld
1061 &cstime, // 17 %ld
1062 &prio, // 18 %ld
1063 &nice, // 19 %ld
1064 &junk, // 20 %ld
1065 &it_real, // 21 %ld
1066 &start, // 22 UINTX_FORMAT
1067 &vsize, // 23 UINTX_FORMAT
1068 &rss, // 24 INTX_FORMAT
1069 &rsslim, // 25 UINTX_FORMAT
1070 &scodes, // 26 UINTX_FORMAT
1071 &ecode, // 27 UINTX_FORMAT
1072 &stack_start); // 28 UINTX_FORMAT
1073 }
1074
1075 #undef _UFM
1076 #undef _DFM
1077
1078 if (i != 28 - 2) {
1079 assert(false, "Bad conversion from /proc/self/stat");
1080 // product mode - assume we are the initial thread, good luck in the
1081 // embedded case.
1082 warning("Can't detect initial thread stack location - bad conversion");
1083 stack_start = (uintptr_t) &rlim;
1084 }
1085 } else {
1086 // For some reason we can't open /proc/self/stat (for example, running on
1087 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1088 // most cases, so don't abort:
1089 warning("Can't detect initial thread stack location - no /proc/self/stat");
1090 stack_start = (uintptr_t) &rlim;
1091 }
1092 }
1093
1094 // Now we have a pointer (stack_start) very close to the stack top, the
1095 // next thing to do is to figure out the exact location of stack top. We
1096 // can find out the virtual memory area that contains stack_start by
1097 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1098 // and its upper limit is the real stack top. (again, this would fail if
1099 // running inside chroot, because /proc may not exist.)
1100
1101 uintptr_t stack_top;
1102 address low, high;
1103 if (find_vma((address)stack_start, &low, &high)) {
1104 // success, "high" is the true stack top. (ignore "low", because initial
1105 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1106 stack_top = (uintptr_t)high;
1107 } else {
1108 // failed, likely because /proc/self/maps does not exist
1109 warning("Can't detect initial thread stack location - find_vma failed");
1110 // best effort: stack_start is normally within a few pages below the real
1111 // stack top, use it as stack top, and reduce stack size so we won't put
1112 // guard page outside stack.
1113 stack_top = stack_start;
1114 stack_size -= 16 * page_size();
1115 }
1116
1117 // stack_top could be partially down the page so align it
1118 stack_top = align_size_up(stack_top, page_size());
1119
1120 // Allowed stack value is minimum of max_size and what we derived from rlimit
1121 if (max_size > 0) {
1122 _initial_thread_stack_size = MIN2(max_size, stack_size);
1123 } else {
1124 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1125 // clamp it at 8MB as we do on Solaris
1126 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1127 }
1128 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1129 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1130
1131 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1132
1133 if (log_is_enabled(Info, os, thread)) {
1134 // See if we seem to be on primordial process thread
1135 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1136 uintptr_t(&rlim) < stack_top;
1137
1138 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1139 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1140 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1141 stack_top, intptr_t(_initial_thread_stack_bottom));
1142 }
1143 }
1144
1145 ////////////////////////////////////////////////////////////////////////////////
1146 // time support
1147
1148 // Time since start-up in seconds to a fine granularity.
1149 // Used by VMSelfDestructTimer and the MemProfiler.
1150 double os::elapsedTime() {
1151
1152 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1153 }
1154
1155 jlong os::elapsed_counter() {
1156 return javaTimeNanos() - initial_time_count;
1157 }
1158
1159 jlong os::elapsed_frequency() {
1160 return NANOSECS_PER_SEC; // nanosecond resolution
1161 }
1162
1163 bool os::supports_vtime() { return true; }
1164 bool os::enable_vtime() { return false; }
1165 bool os::vtime_enabled() { return false; }
1166
1167 double os::elapsedVTime() {
1168 struct rusage usage;
1169 int retval = getrusage(RUSAGE_THREAD, &usage);
1170 if (retval == 0) {
1171 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);
1172 } else {
1173 // better than nothing, but not much
1174 return elapsedTime();
1175 }
1176 }
1177
1178 jlong os::javaTimeMillis() {
1179 timeval time;
1180 int status = gettimeofday(&time, NULL);
1181 assert(status != -1, "linux error");
1182 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1183 }
1184
1185 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1186 timeval time;
1187 int status = gettimeofday(&time, NULL);
1188 assert(status != -1, "linux error");
1189 seconds = jlong(time.tv_sec);
1190 nanos = jlong(time.tv_usec) * 1000;
1191 }
1192
1193
1194 #ifndef CLOCK_MONOTONIC
1195 #define CLOCK_MONOTONIC (1)
1196 #endif
1197
1198 void os::Linux::clock_init() {
1199 // we do dlopen's in this particular order due to bug in linux
1200 // dynamical loader (see 6348968) leading to crash on exit
1201 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1202 if (handle == NULL) {
1203 handle = dlopen("librt.so", RTLD_LAZY);
1204 }
1205
1206 if (handle) {
1207 int (*clock_getres_func)(clockid_t, struct timespec*) =
1208 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1209 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1210 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1211 if (clock_getres_func && clock_gettime_func) {
1212 // See if monotonic clock is supported by the kernel. Note that some
1213 // early implementations simply return kernel jiffies (updated every
1214 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1215 // for nano time (though the monotonic property is still nice to have).
1216 // It's fixed in newer kernels, however clock_getres() still returns
1217 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1218 // resolution for now. Hopefully as people move to new kernels, this
1219 // won't be a problem.
1220 struct timespec res;
1221 struct timespec tp;
1222 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1223 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1224 // yes, monotonic clock is supported
1225 _clock_gettime = clock_gettime_func;
1226 return;
1227 } else {
1228 // close librt if there is no monotonic clock
1229 dlclose(handle);
1230 }
1231 }
1232 }
1233 warning("No monotonic clock was available - timed services may " \
1234 "be adversely affected if the time-of-day clock changes");
1235 }
1236
1237 #ifndef SYS_clock_getres
1238 #if defined(X86) || defined(PPC64) || defined(S390)
1239 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1240 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1241 #else
1242 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1243 #define sys_clock_getres(x,y) -1
1244 #endif
1245 #else
1246 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1247 #endif
1248
1249 void os::Linux::fast_thread_clock_init() {
1250 if (!UseLinuxPosixThreadCPUClocks) {
1251 return;
1252 }
1253 clockid_t clockid;
1254 struct timespec tp;
1255 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1256 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1257
1258 // Switch to using fast clocks for thread cpu time if
1259 // the sys_clock_getres() returns 0 error code.
1260 // Note, that some kernels may support the current thread
1261 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1262 // returned by the pthread_getcpuclockid().
1263 // If the fast Posix clocks are supported then the sys_clock_getres()
1264 // must return at least tp.tv_sec == 0 which means a resolution
1265 // better than 1 sec. This is extra check for reliability.
1266
1267 if (pthread_getcpuclockid_func &&
1268 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1269 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1270 _supports_fast_thread_cpu_time = true;
1271 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1272 }
1273 }
1274
1275 jlong os::javaTimeNanos() {
1276 if (os::supports_monotonic_clock()) {
1277 struct timespec tp;
1278 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1279 assert(status == 0, "gettime error");
1280 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1281 return result;
1282 } else {
1283 timeval time;
1284 int status = gettimeofday(&time, NULL);
1285 assert(status != -1, "linux error");
1286 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1287 return 1000 * usecs;
1288 }
1289 }
1290
1291 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1292 if (os::supports_monotonic_clock()) {
1293 info_ptr->max_value = ALL_64_BITS;
1294
1295 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1296 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1297 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1298 } else {
1299 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1300 info_ptr->max_value = ALL_64_BITS;
1301
1302 // gettimeofday is a real time clock so it skips
1303 info_ptr->may_skip_backward = true;
1304 info_ptr->may_skip_forward = true;
1305 }
1306
1307 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1308 }
1309
1310 // Return the real, user, and system times in seconds from an
1311 // arbitrary fixed point in the past.
1312 bool os::getTimesSecs(double* process_real_time,
1313 double* process_user_time,
1314 double* process_system_time) {
1315 struct tms ticks;
1316 clock_t real_ticks = times(&ticks);
1317
1318 if (real_ticks == (clock_t) (-1)) {
1319 return false;
1320 } else {
1321 double ticks_per_second = (double) clock_tics_per_sec;
1322 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1323 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1324 *process_real_time = ((double) real_ticks) / ticks_per_second;
1325
1326 return true;
1327 }
1328 }
1329
1330
1331 char * os::local_time_string(char *buf, size_t buflen) {
1332 struct tm t;
1333 time_t long_time;
1334 time(&long_time);
1335 localtime_r(&long_time, &t);
1336 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1337 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1338 t.tm_hour, t.tm_min, t.tm_sec);
1339 return buf;
1340 }
1341
1342 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1343 return localtime_r(clock, res);
1344 }
1345
1346 ////////////////////////////////////////////////////////////////////////////////
1347 // runtime exit support
1348
1349 // Note: os::shutdown() might be called very early during initialization, or
1350 // called from signal handler. Before adding something to os::shutdown(), make
1351 // sure it is async-safe and can handle partially initialized VM.
1352 void os::shutdown() {
1353
1354 // allow PerfMemory to attempt cleanup of any persistent resources
1355 perfMemory_exit();
1356
1357 // needs to remove object in file system
1358 AttachListener::abort();
1359
1360 // flush buffered output, finish log files
1361 ostream_abort();
1362
1363 // Check for abort hook
1364 abort_hook_t abort_hook = Arguments::abort_hook();
1365 if (abort_hook != NULL) {
1366 abort_hook();
1367 }
1368
1369 }
1370
1371 // Note: os::abort() might be called very early during initialization, or
1372 // called from signal handler. Before adding something to os::abort(), make
1373 // sure it is async-safe and can handle partially initialized VM.
1374 void os::abort(bool dump_core, void* siginfo, const void* context) {
1375 os::shutdown();
1376 if (dump_core) {
1377 #ifndef PRODUCT
1378 fdStream out(defaultStream::output_fd());
1379 out.print_raw("Current thread is ");
1380 char buf[16];
1381 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1382 out.print_raw_cr(buf);
1383 out.print_raw_cr("Dumping core ...");
1384 #endif
1385 ::abort(); // dump core
1386 }
1387
1388 ::exit(1);
1389 }
1390
1391 // Die immediately, no exit hook, no abort hook, no cleanup.
1392 void os::die() {
1393 ::abort();
1394 }
1395
1396
1397 // This method is a copy of JDK's sysGetLastErrorString
1398 // from src/solaris/hpi/src/system_md.c
1399
1400 size_t os::lasterror(char *buf, size_t len) {
1401 if (errno == 0) return 0;
1402
1403 const char *s = os::strerror(errno);
1404 size_t n = ::strlen(s);
1405 if (n >= len) {
1406 n = len - 1;
1407 }
1408 ::strncpy(buf, s, n);
1409 buf[n] = '\0';
1410 return n;
1411 }
1412
1413 // thread_id is kernel thread id (similar to Solaris LWP id)
1414 intx os::current_thread_id() { return os::Linux::gettid(); }
1415 int os::current_process_id() {
1416 return ::getpid();
1417 }
1418
1419 // DLL functions
1420
1421 const char* os::dll_file_extension() { return ".so"; }
1422
1423 // This must be hard coded because it's the system's temporary
1424 // directory not the java application's temp directory, ala java.io.tmpdir.
1425 const char* os::get_temp_directory() { return "/tmp"; }
1426
1427 static bool file_exists(const char* filename) {
1428 struct stat statbuf;
1429 if (filename == NULL || strlen(filename) == 0) {
1430 return false;
1431 }
1432 return os::stat(filename, &statbuf) == 0;
1433 }
1434
1435 bool os::dll_build_name(char* buffer, size_t buflen,
1436 const char* pname, const char* fname) {
1437 bool retval = false;
1438 // Copied from libhpi
1439 const size_t pnamelen = pname ? strlen(pname) : 0;
1440
1441 // Return error on buffer overflow.
1442 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1443 return retval;
1444 }
1445
1446 if (pnamelen == 0) {
1447 snprintf(buffer, buflen, "lib%s.so", fname);
1448 retval = true;
1449 } else if (strchr(pname, *os::path_separator()) != NULL) {
1450 int n;
1451 char** pelements = split_path(pname, &n);
1452 if (pelements == NULL) {
1453 return false;
1454 }
1455 for (int i = 0; i < n; i++) {
1456 // Really shouldn't be NULL, but check can't hurt
1457 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1458 continue; // skip the empty path values
1459 }
1460 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1461 if (file_exists(buffer)) {
1462 retval = true;
1463 break;
1464 }
1465 }
1466 // release the storage
1467 for (int i = 0; i < n; i++) {
1468 if (pelements[i] != NULL) {
1469 FREE_C_HEAP_ARRAY(char, pelements[i]);
1470 }
1471 }
1472 if (pelements != NULL) {
1473 FREE_C_HEAP_ARRAY(char*, pelements);
1474 }
1475 } else {
1476 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1477 retval = true;
1478 }
1479 return retval;
1480 }
1481
1482 // check if addr is inside libjvm.so
1483 bool os::address_is_in_vm(address addr) {
1484 static address libjvm_base_addr;
1485 Dl_info dlinfo;
1486
1487 if (libjvm_base_addr == NULL) {
1488 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1489 libjvm_base_addr = (address)dlinfo.dli_fbase;
1490 }
1491 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1492 }
1493
1494 if (dladdr((void *)addr, &dlinfo) != 0) {
1495 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1496 }
1497
1498 return false;
1499 }
1500
1501 bool os::dll_address_to_function_name(address addr, char *buf,
1502 int buflen, int *offset,
1503 bool demangle) {
1504 // buf is not optional, but offset is optional
1505 assert(buf != NULL, "sanity check");
1506
1507 Dl_info dlinfo;
1508
1509 if (dladdr((void*)addr, &dlinfo) != 0) {
1510 // see if we have a matching symbol
1511 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1512 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1513 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1514 }
1515 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1516 return true;
1517 }
1518 // no matching symbol so try for just file info
1519 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1520 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1521 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1522 return true;
1523 }
1524 }
1525 }
1526
1527 buf[0] = '\0';
1528 if (offset != NULL) *offset = -1;
1529 return false;
1530 }
1531
1532 struct _address_to_library_name {
1533 address addr; // input : memory address
1534 size_t buflen; // size of fname
1535 char* fname; // output: library name
1536 address base; // library base addr
1537 };
1538
1539 static int address_to_library_name_callback(struct dl_phdr_info *info,
1540 size_t size, void *data) {
1541 int i;
1542 bool found = false;
1543 address libbase = NULL;
1544 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1545
1546 // iterate through all loadable segments
1547 for (i = 0; i < info->dlpi_phnum; i++) {
1548 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1549 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1550 // base address of a library is the lowest address of its loaded
1551 // segments.
1552 if (libbase == NULL || libbase > segbase) {
1553 libbase = segbase;
1554 }
1555 // see if 'addr' is within current segment
1556 if (segbase <= d->addr &&
1557 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1558 found = true;
1559 }
1560 }
1561 }
1562
1563 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1564 // so dll_address_to_library_name() can fall through to use dladdr() which
1565 // can figure out executable name from argv[0].
1566 if (found && info->dlpi_name && info->dlpi_name[0]) {
1567 d->base = libbase;
1568 if (d->fname) {
1569 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1570 }
1571 return 1;
1572 }
1573 return 0;
1574 }
1575
1576 bool os::dll_address_to_library_name(address addr, char* buf,
1577 int buflen, int* offset) {
1578 // buf is not optional, but offset is optional
1579 assert(buf != NULL, "sanity check");
1580
1581 Dl_info dlinfo;
1582 struct _address_to_library_name data;
1583
1584 // There is a bug in old glibc dladdr() implementation that it could resolve
1585 // to wrong library name if the .so file has a base address != NULL. Here
1586 // we iterate through the program headers of all loaded libraries to find
1587 // out which library 'addr' really belongs to. This workaround can be
1588 // removed once the minimum requirement for glibc is moved to 2.3.x.
1589 data.addr = addr;
1590 data.fname = buf;
1591 data.buflen = buflen;
1592 data.base = NULL;
1593 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1594
1595 if (rslt) {
1596 // buf already contains library name
1597 if (offset) *offset = addr - data.base;
1598 return true;
1599 }
1600 if (dladdr((void*)addr, &dlinfo) != 0) {
1601 if (dlinfo.dli_fname != NULL) {
1602 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1603 }
1604 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1605 *offset = addr - (address)dlinfo.dli_fbase;
1606 }
1607 return true;
1608 }
1609
1610 buf[0] = '\0';
1611 if (offset) *offset = -1;
1612 return false;
1613 }
1614
1615 // Loads .dll/.so and
1616 // in case of error it checks if .dll/.so was built for the
1617 // same architecture as Hotspot is running on
1618
1619
1620 // Remember the stack's state. The Linux dynamic linker will change
1621 // the stack to 'executable' at most once, so we must safepoint only once.
1622 bool os::Linux::_stack_is_executable = false;
1623
1624 // VM operation that loads a library. This is necessary if stack protection
1625 // of the Java stacks can be lost during loading the library. If we
1626 // do not stop the Java threads, they can stack overflow before the stacks
1627 // are protected again.
1628 class VM_LinuxDllLoad: public VM_Operation {
1629 private:
1630 const char *_filename;
1631 char *_ebuf;
1632 int _ebuflen;
1633 void *_lib;
1634 public:
1635 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1636 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1637 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1638 void doit() {
1639 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1640 os::Linux::_stack_is_executable = true;
1641 }
1642 void* loaded_library() { return _lib; }
1643 };
1644
1645 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1646 void * result = NULL;
1647 bool load_attempted = false;
1648
1649 // Check whether the library to load might change execution rights
1650 // of the stack. If they are changed, the protection of the stack
1651 // guard pages will be lost. We need a safepoint to fix this.
1652 //
1653 // See Linux man page execstack(8) for more info.
1654 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1655 ElfFile ef(filename);
1656 if (!ef.specifies_noexecstack()) {
1657 if (!is_init_completed()) {
1658 os::Linux::_stack_is_executable = true;
1659 // This is OK - No Java threads have been created yet, and hence no
1660 // stack guard pages to fix.
1661 //
1662 // This should happen only when you are building JDK7 using a very
1663 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1664 //
1665 // Dynamic loader will make all stacks executable after
1666 // this function returns, and will not do that again.
1667 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1668 } else {
1669 warning("You have loaded library %s which might have disabled stack guard. "
1670 "The VM will try to fix the stack guard now.\n"
1671 "It's highly recommended that you fix the library with "
1672 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1673 filename);
1674
1675 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1676 JavaThread *jt = JavaThread::current();
1677 if (jt->thread_state() != _thread_in_native) {
1678 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1679 // that requires ExecStack. Cannot enter safe point. Let's give up.
1680 warning("Unable to fix stack guard. Giving up.");
1681 } else {
1682 if (!LoadExecStackDllInVMThread) {
1683 // This is for the case where the DLL has an static
1684 // constructor function that executes JNI code. We cannot
1685 // load such DLLs in the VMThread.
1686 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1687 }
1688
1689 ThreadInVMfromNative tiv(jt);
1690 debug_only(VMNativeEntryWrapper vew;)
1691
1692 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1693 VMThread::execute(&op);
1694 if (LoadExecStackDllInVMThread) {
1695 result = op.loaded_library();
1696 }
1697 load_attempted = true;
1698 }
1699 }
1700 }
1701 }
1702
1703 if (!load_attempted) {
1704 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1705 }
1706
1707 if (result != NULL) {
1708 // Successful loading
1709 return result;
1710 }
1711
1712 Elf32_Ehdr elf_head;
1713 int diag_msg_max_length=ebuflen-strlen(ebuf);
1714 char* diag_msg_buf=ebuf+strlen(ebuf);
1715
1716 if (diag_msg_max_length==0) {
1717 // No more space in ebuf for additional diagnostics message
1718 return NULL;
1719 }
1720
1721
1722 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1723
1724 if (file_descriptor < 0) {
1725 // Can't open library, report dlerror() message
1726 return NULL;
1727 }
1728
1729 bool failed_to_read_elf_head=
1730 (sizeof(elf_head)!=
1731 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1732
1733 ::close(file_descriptor);
1734 if (failed_to_read_elf_head) {
1735 // file i/o error - report dlerror() msg
1736 return NULL;
1737 }
1738
1739 typedef struct {
1740 Elf32_Half code; // Actual value as defined in elf.h
1741 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1742 unsigned char elf_class; // 32 or 64 bit
1743 unsigned char endianess; // MSB or LSB
1744 char* name; // String representation
1745 } arch_t;
1746
1747 #ifndef EM_486
1748 #define EM_486 6 /* Intel 80486 */
1749 #endif
1750 #ifndef EM_AARCH64
1751 #define EM_AARCH64 183 /* ARM AARCH64 */
1752 #endif
1753
1754 static const arch_t arch_array[]={
1755 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1756 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1757 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1758 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1759 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1760 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1761 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1762 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1763 #if defined(VM_LITTLE_ENDIAN)
1764 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1765 #else
1766 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1767 #endif
1768 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1769 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1770 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1771 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1772 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1773 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1774 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1775 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1776 };
1777
1778 #if (defined IA32)
1779 static Elf32_Half running_arch_code=EM_386;
1780 #elif (defined AMD64)
1781 static Elf32_Half running_arch_code=EM_X86_64;
1782 #elif (defined IA64)
1783 static Elf32_Half running_arch_code=EM_IA_64;
1784 #elif (defined __sparc) && (defined _LP64)
1785 static Elf32_Half running_arch_code=EM_SPARCV9;
1786 #elif (defined __sparc) && (!defined _LP64)
1787 static Elf32_Half running_arch_code=EM_SPARC;
1788 #elif (defined __powerpc64__)
1789 static Elf32_Half running_arch_code=EM_PPC64;
1790 #elif (defined __powerpc__)
1791 static Elf32_Half running_arch_code=EM_PPC;
1792 #elif (defined AARCH64)
1793 static Elf32_Half running_arch_code=EM_AARCH64;
1794 #elif (defined ARM)
1795 static Elf32_Half running_arch_code=EM_ARM;
1796 #elif (defined S390)
1797 static Elf32_Half running_arch_code=EM_S390;
1798 #elif (defined ALPHA)
1799 static Elf32_Half running_arch_code=EM_ALPHA;
1800 #elif (defined MIPSEL)
1801 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1802 #elif (defined PARISC)
1803 static Elf32_Half running_arch_code=EM_PARISC;
1804 #elif (defined MIPS)
1805 static Elf32_Half running_arch_code=EM_MIPS;
1806 #elif (defined M68K)
1807 static Elf32_Half running_arch_code=EM_68K;
1808 #else
1809 #error Method os::dll_load requires that one of following is defined:\
1810 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1811 #endif
1812
1813 // Identify compatability class for VM's architecture and library's architecture
1814 // Obtain string descriptions for architectures
1815
1816 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1817 int running_arch_index=-1;
1818
1819 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1820 if (running_arch_code == arch_array[i].code) {
1821 running_arch_index = i;
1822 }
1823 if (lib_arch.code == arch_array[i].code) {
1824 lib_arch.compat_class = arch_array[i].compat_class;
1825 lib_arch.name = arch_array[i].name;
1826 }
1827 }
1828
1829 assert(running_arch_index != -1,
1830 "Didn't find running architecture code (running_arch_code) in arch_array");
1831 if (running_arch_index == -1) {
1832 // Even though running architecture detection failed
1833 // we may still continue with reporting dlerror() message
1834 return NULL;
1835 }
1836
1837 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1839 return NULL;
1840 }
1841
1842 #ifndef S390
1843 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1844 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1845 return NULL;
1846 }
1847 #endif // !S390
1848
1849 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1850 if (lib_arch.name!=NULL) {
1851 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1852 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1853 lib_arch.name, arch_array[running_arch_index].name);
1854 } else {
1855 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1856 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1857 lib_arch.code,
1858 arch_array[running_arch_index].name);
1859 }
1860 }
1861
1862 return NULL;
1863 }
1864
1865 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1866 int ebuflen) {
1867 void * result = ::dlopen(filename, RTLD_LAZY);
1868 if (result == NULL) {
1869 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1870 ebuf[ebuflen-1] = '\0';
1871 }
1872 return result;
1873 }
1874
1875 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1876 int ebuflen) {
1877 void * result = NULL;
1878 if (LoadExecStackDllInVMThread) {
1879 result = dlopen_helper(filename, ebuf, ebuflen);
1880 }
1881
1882 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1883 // library that requires an executable stack, or which does not have this
1884 // stack attribute set, dlopen changes the stack attribute to executable. The
1885 // read protection of the guard pages gets lost.
1886 //
1887 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1888 // may have been queued at the same time.
1889
1890 if (!_stack_is_executable) {
1891 JavaThread *jt = Threads::first();
1892
1893 while (jt) {
1894 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1895 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1896 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1897 warning("Attempt to reguard stack yellow zone failed.");
1898 }
1899 }
1900 jt = jt->next();
1901 }
1902 }
1903
1904 return result;
1905 }
1906
1907 void* os::dll_lookup(void* handle, const char* name) {
1908 void* res = dlsym(handle, name);
1909 return res;
1910 }
1911
1912 void* os::get_default_process_handle() {
1913 return (void*)::dlopen(NULL, RTLD_LAZY);
1914 }
1915
1916 static bool _print_ascii_file(const char* filename, outputStream* st) {
1917 int fd = ::open(filename, O_RDONLY);
1918 if (fd == -1) {
1919 return false;
1920 }
1921
1922 char buf[33];
1923 int bytes;
1924 buf[32] = '\0';
1925 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1926 st->print_raw(buf, bytes);
1927 }
1928
1929 ::close(fd);
1930
1931 return true;
1932 }
1933
1934 void os::print_dll_info(outputStream *st) {
1935 st->print_cr("Dynamic libraries:");
1936
1937 char fname[32];
1938 pid_t pid = os::Linux::gettid();
1939
1940 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1941
1942 if (!_print_ascii_file(fname, st)) {
1943 st->print("Can not get library information for pid = %d\n", pid);
1944 }
1945 }
1946
1947 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1948 FILE *procmapsFile = NULL;
1949
1950 // Open the procfs maps file for the current process
1951 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1952 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1953 char line[PATH_MAX + 100];
1954
1955 // Read line by line from 'file'
1956 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1957 u8 base, top, offset, inode;
1958 char permissions[5];
1959 char device[6];
1960 char name[PATH_MAX + 1];
1961
1962 // Parse fields from line
1963 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1964 &base, &top, permissions, &offset, device, &inode, name);
1965
1966 // Filter by device id '00:00' so that we only get file system mapped files.
1967 if (strcmp(device, "00:00") != 0) {
1968
1969 // Call callback with the fields of interest
1970 if(callback(name, (address)base, (address)top, param)) {
1971 // Oops abort, callback aborted
1972 fclose(procmapsFile);
1973 return 1;
1974 }
1975 }
1976 }
1977 fclose(procmapsFile);
1978 }
1979 return 0;
1980 }
1981
1982 void os::print_os_info_brief(outputStream* st) {
1983 os::Linux::print_distro_info(st);
1984
1985 os::Posix::print_uname_info(st);
1986
1987 os::Linux::print_libversion_info(st);
1988
1989 }
1990
1991 void os::print_os_info(outputStream* st) {
1992 st->print("OS:");
1993
1994 os::Linux::print_distro_info(st);
1995
1996 os::Posix::print_uname_info(st);
1997
1998 // Print warning if unsafe chroot environment detected
1999 if (unsafe_chroot_detected) {
2000 st->print("WARNING!! ");
2001 st->print_cr("%s", unstable_chroot_error);
2002 }
2003
2004 os::Linux::print_libversion_info(st);
2005
2006 os::Posix::print_rlimit_info(st);
2007
2008 os::Posix::print_load_average(st);
2009
2010 os::Linux::print_full_memory_info(st);
2011 }
2012
2013 // Try to identify popular distros.
2014 // Most Linux distributions have a /etc/XXX-release file, which contains
2015 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2016 // file that also contains the OS version string. Some have more than one
2017 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2018 // /etc/redhat-release.), so the order is important.
2019 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2020 // their own specific XXX-release file as well as a redhat-release file.
2021 // Because of this the XXX-release file needs to be searched for before the
2022 // redhat-release file.
2023 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2024 // search for redhat-release / SuSE-release needs to be before lsb-release.
2025 // Since the lsb-release file is the new standard it needs to be searched
2026 // before the older style release files.
2027 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2028 // next to last resort. The os-release file is a new standard that contains
2029 // distribution information and the system-release file seems to be an old
2030 // standard that has been replaced by the lsb-release and os-release files.
2031 // Searching for the debian_version file is the last resort. It contains
2032 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2033 // "Debian " is printed before the contents of the debian_version file.
2034
2035 const char* distro_files[] = {
2036 "/etc/oracle-release",
2037 "/etc/mandriva-release",
2038 "/etc/mandrake-release",
2039 "/etc/sun-release",
2040 "/etc/redhat-release",
2041 "/etc/SuSE-release",
2042 "/etc/lsb-release",
2043 "/etc/turbolinux-release",
2044 "/etc/gentoo-release",
2045 "/etc/ltib-release",
2046 "/etc/angstrom-version",
2047 "/etc/system-release",
2048 "/etc/os-release",
2049 NULL };
2050
2051 void os::Linux::print_distro_info(outputStream* st) {
2052 for (int i = 0;; i++) {
2053 const char* file = distro_files[i];
2054 if (file == NULL) {
2055 break; // done
2056 }
2057 // If file prints, we found it.
2058 if (_print_ascii_file(file, st)) {
2059 return;
2060 }
2061 }
2062
2063 if (file_exists("/etc/debian_version")) {
2064 st->print("Debian ");
2065 _print_ascii_file("/etc/debian_version", st);
2066 } else {
2067 st->print("Linux");
2068 }
2069 st->cr();
2070 }
2071
2072 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2073 char buf[256];
2074 while (fgets(buf, sizeof(buf), fp)) {
2075 // Edit out extra stuff in expected format
2076 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2077 char* ptr = strstr(buf, "\""); // the name is in quotes
2078 if (ptr != NULL) {
2079 ptr++; // go beyond first quote
2080 char* nl = strchr(ptr, '\"');
2081 if (nl != NULL) *nl = '\0';
2082 strncpy(distro, ptr, length);
2083 } else {
2084 ptr = strstr(buf, "=");
2085 ptr++; // go beyond equals then
2086 char* nl = strchr(ptr, '\n');
2087 if (nl != NULL) *nl = '\0';
2088 strncpy(distro, ptr, length);
2089 }
2090 return;
2091 } else if (get_first_line) {
2092 char* nl = strchr(buf, '\n');
2093 if (nl != NULL) *nl = '\0';
2094 strncpy(distro, buf, length);
2095 return;
2096 }
2097 }
2098 // print last line and close
2099 char* nl = strchr(buf, '\n');
2100 if (nl != NULL) *nl = '\0';
2101 strncpy(distro, buf, length);
2102 }
2103
2104 static void parse_os_info(char* distro, size_t length, const char* file) {
2105 FILE* fp = fopen(file, "r");
2106 if (fp != NULL) {
2107 // if suse format, print out first line
2108 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2109 parse_os_info_helper(fp, distro, length, get_first_line);
2110 fclose(fp);
2111 }
2112 }
2113
2114 void os::get_summary_os_info(char* buf, size_t buflen) {
2115 for (int i = 0;; i++) {
2116 const char* file = distro_files[i];
2117 if (file == NULL) {
2118 break; // ran out of distro_files
2119 }
2120 if (file_exists(file)) {
2121 parse_os_info(buf, buflen, file);
2122 return;
2123 }
2124 }
2125 // special case for debian
2126 if (file_exists("/etc/debian_version")) {
2127 strncpy(buf, "Debian ", buflen);
2128 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2129 } else {
2130 strncpy(buf, "Linux", buflen);
2131 }
2132 }
2133
2134 void os::Linux::print_libversion_info(outputStream* st) {
2135 // libc, pthread
2136 st->print("libc:");
2137 st->print("%s ", os::Linux::glibc_version());
2138 st->print("%s ", os::Linux::libpthread_version());
2139 st->cr();
2140 }
2141
2142 void os::Linux::print_full_memory_info(outputStream* st) {
2143 st->print("\n/proc/meminfo:\n");
2144 _print_ascii_file("/proc/meminfo", st);
2145 st->cr();
2146 }
2147
2148 void os::print_memory_info(outputStream* st) {
2149
2150 st->print("Memory:");
2151 st->print(" %dk page", os::vm_page_size()>>10);
2152
2153 // values in struct sysinfo are "unsigned long"
2154 struct sysinfo si;
2155 sysinfo(&si);
2156
2157 st->print(", physical " UINT64_FORMAT "k",
2158 os::physical_memory() >> 10);
2159 st->print("(" UINT64_FORMAT "k free)",
2160 os::available_memory() >> 10);
2161 st->print(", swap " UINT64_FORMAT "k",
2162 ((jlong)si.totalswap * si.mem_unit) >> 10);
2163 st->print("(" UINT64_FORMAT "k free)",
2164 ((jlong)si.freeswap * si.mem_unit) >> 10);
2165 st->cr();
2166 }
2167
2168 // Print the first "model name" line and the first "flags" line
2169 // that we find and nothing more. We assume "model name" comes
2170 // before "flags" so if we find a second "model name", then the
2171 // "flags" field is considered missing.
2172 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2173 #if defined(IA32) || defined(AMD64)
2174 // Other platforms have less repetitive cpuinfo files
2175 FILE *fp = fopen("/proc/cpuinfo", "r");
2176 if (fp) {
2177 while (!feof(fp)) {
2178 if (fgets(buf, buflen, fp)) {
2179 // Assume model name comes before flags
2180 bool model_name_printed = false;
2181 if (strstr(buf, "model name") != NULL) {
2182 if (!model_name_printed) {
2183 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2184 st->print_raw(buf);
2185 model_name_printed = true;
2186 } else {
2187 // model name printed but not flags? Odd, just return
2188 fclose(fp);
2189 return true;
2190 }
2191 }
2192 // print the flags line too
2193 if (strstr(buf, "flags") != NULL) {
2194 st->print_raw(buf);
2195 fclose(fp);
2196 return true;
2197 }
2198 }
2199 }
2200 fclose(fp);
2201 }
2202 #endif // x86 platforms
2203 return false;
2204 }
2205
2206 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2207 // Only print the model name if the platform provides this as a summary
2208 if (!print_model_name_and_flags(st, buf, buflen)) {
2209 st->print("\n/proc/cpuinfo:\n");
2210 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2211 st->print_cr(" <Not Available>");
2212 }
2213 }
2214 }
2215
2216 #if defined(AMD64) || defined(IA32) || defined(X32)
2217 const char* search_string = "model name";
2218 #elif defined(PPC64)
2219 const char* search_string = "cpu";
2220 #elif defined(S390)
2221 const char* search_string = "processor";
2222 #elif defined(SPARC)
2223 const char* search_string = "cpu";
2224 #else
2225 const char* search_string = "Processor";
2226 #endif
2227
2228 // Parses the cpuinfo file for string representing the model name.
2229 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2230 FILE* fp = fopen("/proc/cpuinfo", "r");
2231 if (fp != NULL) {
2232 while (!feof(fp)) {
2233 char buf[256];
2234 if (fgets(buf, sizeof(buf), fp)) {
2235 char* start = strstr(buf, search_string);
2236 if (start != NULL) {
2237 char *ptr = start + strlen(search_string);
2238 char *end = buf + strlen(buf);
2239 while (ptr != end) {
2240 // skip whitespace and colon for the rest of the name.
2241 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2242 break;
2243 }
2244 ptr++;
2245 }
2246 if (ptr != end) {
2247 // reasonable string, get rid of newline and keep the rest
2248 char* nl = strchr(buf, '\n');
2249 if (nl != NULL) *nl = '\0';
2250 strncpy(cpuinfo, ptr, length);
2251 fclose(fp);
2252 return;
2253 }
2254 }
2255 }
2256 }
2257 fclose(fp);
2258 }
2259 // cpuinfo not found or parsing failed, just print generic string. The entire
2260 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2261 #if defined(AARCH64)
2262 strncpy(cpuinfo, "AArch64", length);
2263 #elif defined(AMD64)
2264 strncpy(cpuinfo, "x86_64", length);
2265 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2266 strncpy(cpuinfo, "ARM", length);
2267 #elif defined(IA32)
2268 strncpy(cpuinfo, "x86_32", length);
2269 #elif defined(IA64)
2270 strncpy(cpuinfo, "IA64", length);
2271 #elif defined(PPC)
2272 strncpy(cpuinfo, "PPC64", length);
2273 #elif defined(S390)
2274 strncpy(cpuinfo, "S390", length);
2275 #elif defined(SPARC)
2276 strncpy(cpuinfo, "sparcv9", length);
2277 #elif defined(ZERO_LIBARCH)
2278 strncpy(cpuinfo, ZERO_LIBARCH, length);
2279 #else
2280 strncpy(cpuinfo, "unknown", length);
2281 #endif
2282 }
2283
2284 static void print_signal_handler(outputStream* st, int sig,
2285 char* buf, size_t buflen);
2286
2287 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2288 st->print_cr("Signal Handlers:");
2289 print_signal_handler(st, SIGSEGV, buf, buflen);
2290 print_signal_handler(st, SIGBUS , buf, buflen);
2291 print_signal_handler(st, SIGFPE , buf, buflen);
2292 print_signal_handler(st, SIGPIPE, buf, buflen);
2293 print_signal_handler(st, SIGXFSZ, buf, buflen);
2294 print_signal_handler(st, SIGILL , buf, buflen);
2295 print_signal_handler(st, SR_signum, buf, buflen);
2296 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2297 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2298 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2299 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2300 #if defined(PPC64)
2301 print_signal_handler(st, SIGTRAP, buf, buflen);
2302 #endif
2303 }
2304
2305 static char saved_jvm_path[MAXPATHLEN] = {0};
2306
2307 // Find the full path to the current module, libjvm.so
2308 void os::jvm_path(char *buf, jint buflen) {
2309 // Error checking.
2310 if (buflen < MAXPATHLEN) {
2311 assert(false, "must use a large-enough buffer");
2312 buf[0] = '\0';
2313 return;
2314 }
2315 // Lazy resolve the path to current module.
2316 if (saved_jvm_path[0] != 0) {
2317 strcpy(buf, saved_jvm_path);
2318 return;
2319 }
2320
2321 char dli_fname[MAXPATHLEN];
2322 bool ret = dll_address_to_library_name(
2323 CAST_FROM_FN_PTR(address, os::jvm_path),
2324 dli_fname, sizeof(dli_fname), NULL);
2325 assert(ret, "cannot locate libjvm");
2326 char *rp = NULL;
2327 if (ret && dli_fname[0] != '\0') {
2328 rp = os::Posix::realpath(dli_fname, buf, buflen);
2329 }
2330 if (rp == NULL) {
2331 return;
2332 }
2333
2334 if (Arguments::sun_java_launcher_is_altjvm()) {
2335 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2336 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2337 // If "/jre/lib/" appears at the right place in the string, then
2338 // assume we are installed in a JDK and we're done. Otherwise, check
2339 // for a JAVA_HOME environment variable and fix up the path so it
2340 // looks like libjvm.so is installed there (append a fake suffix
2341 // hotspot/libjvm.so).
2342 const char *p = buf + strlen(buf) - 1;
2343 for (int count = 0; p > buf && count < 5; ++count) {
2344 for (--p; p > buf && *p != '/'; --p)
2345 /* empty */ ;
2346 }
2347
2348 if (strncmp(p, "/jre/lib/", 9) != 0) {
2349 // Look for JAVA_HOME in the environment.
2350 char* java_home_var = ::getenv("JAVA_HOME");
2351 if (java_home_var != NULL && java_home_var[0] != 0) {
2352 char* jrelib_p;
2353 int len;
2354
2355 // Check the current module name "libjvm.so".
2356 p = strrchr(buf, '/');
2357 if (p == NULL) {
2358 return;
2359 }
2360 assert(strstr(p, "/libjvm") == p, "invalid library name");
2361
2362 rp = os::Posix::realpath(java_home_var, buf, buflen);
2363 if (rp == NULL) {
2364 return;
2365 }
2366
2367 // determine if this is a legacy image or modules image
2368 // modules image doesn't have "jre" subdirectory
2369 len = strlen(buf);
2370 assert(len < buflen, "Ran out of buffer room");
2371 jrelib_p = buf + len;
2372 snprintf(jrelib_p, buflen-len, "/jre/lib");
2373 if (0 != access(buf, F_OK)) {
2374 snprintf(jrelib_p, buflen-len, "/lib");
2375 }
2376
2377 if (0 == access(buf, F_OK)) {
2378 // Use current module name "libjvm.so"
2379 len = strlen(buf);
2380 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2381 } else {
2382 // Go back to path of .so
2383 rp = os::Posix::realpath(dli_fname, buf, buflen);
2384 if (rp == NULL) {
2385 return;
2386 }
2387 }
2388 }
2389 }
2390 }
2391
2392 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2393 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2394 }
2395
2396 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2397 // no prefix required, not even "_"
2398 }
2399
2400 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2401 // no suffix required
2402 }
2403
2404 ////////////////////////////////////////////////////////////////////////////////
2405 // sun.misc.Signal support
2406
2407 static volatile jint sigint_count = 0;
2408
2409 static void UserHandler(int sig, void *siginfo, void *context) {
2410 // 4511530 - sem_post is serialized and handled by the manager thread. When
2411 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2412 // don't want to flood the manager thread with sem_post requests.
2413 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2414 return;
2415 }
2416
2417 // Ctrl-C is pressed during error reporting, likely because the error
2418 // handler fails to abort. Let VM die immediately.
2419 if (sig == SIGINT && is_error_reported()) {
2420 os::die();
2421 }
2422
2423 os::signal_notify(sig);
2424 }
2425
2426 void* os::user_handler() {
2427 return CAST_FROM_FN_PTR(void*, UserHandler);
2428 }
2429
2430 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2431 struct timespec ts;
2432 // Semaphore's are always associated with CLOCK_REALTIME
2433 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2434 // see unpackTime for discussion on overflow checking
2435 if (sec >= MAX_SECS) {
2436 ts.tv_sec += MAX_SECS;
2437 ts.tv_nsec = 0;
2438 } else {
2439 ts.tv_sec += sec;
2440 ts.tv_nsec += nsec;
2441 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2442 ts.tv_nsec -= NANOSECS_PER_SEC;
2443 ++ts.tv_sec; // note: this must be <= max_secs
2444 }
2445 }
2446
2447 return ts;
2448 }
2449
2450 extern "C" {
2451 typedef void (*sa_handler_t)(int);
2452 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2453 }
2454
2455 void* os::signal(int signal_number, void* handler) {
2456 struct sigaction sigAct, oldSigAct;
2457
2458 sigfillset(&(sigAct.sa_mask));
2459 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2460 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2461
2462 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2463 // -1 means registration failed
2464 return (void *)-1;
2465 }
2466
2467 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2468 }
2469
2470 void os::signal_raise(int signal_number) {
2471 ::raise(signal_number);
2472 }
2473
2474 // The following code is moved from os.cpp for making this
2475 // code platform specific, which it is by its very nature.
2476
2477 // Will be modified when max signal is changed to be dynamic
2478 int os::sigexitnum_pd() {
2479 return NSIG;
2480 }
2481
2482 // a counter for each possible signal value
2483 static volatile jint pending_signals[NSIG+1] = { 0 };
2484
2485 // Linux(POSIX) specific hand shaking semaphore.
2486 static sem_t sig_sem;
2487 static PosixSemaphore sr_semaphore;
2488
2489 void os::signal_init_pd() {
2490 // Initialize signal structures
2491 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2492
2493 // Initialize signal semaphore
2494 ::sem_init(&sig_sem, 0, 0);
2495 }
2496
2497 void os::signal_notify(int sig) {
2498 Atomic::inc(&pending_signals[sig]);
2499 ::sem_post(&sig_sem);
2500 }
2501
2502 static int check_pending_signals(bool wait) {
2503 Atomic::store(0, &sigint_count);
2504 for (;;) {
2505 for (int i = 0; i < NSIG + 1; i++) {
2506 jint n = pending_signals[i];
2507 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2508 return i;
2509 }
2510 }
2511 if (!wait) {
2512 return -1;
2513 }
2514 JavaThread *thread = JavaThread::current();
2515 ThreadBlockInVM tbivm(thread);
2516
2517 bool threadIsSuspended;
2518 do {
2519 thread->set_suspend_equivalent();
2520 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2521 ::sem_wait(&sig_sem);
2522
2523 // were we externally suspended while we were waiting?
2524 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2525 if (threadIsSuspended) {
2526 // The semaphore has been incremented, but while we were waiting
2527 // another thread suspended us. We don't want to continue running
2528 // while suspended because that would surprise the thread that
2529 // suspended us.
2530 ::sem_post(&sig_sem);
2531
2532 thread->java_suspend_self();
2533 }
2534 } while (threadIsSuspended);
2535 }
2536 }
2537
2538 int os::signal_lookup() {
2539 return check_pending_signals(false);
2540 }
2541
2542 int os::signal_wait() {
2543 return check_pending_signals(true);
2544 }
2545
2546 ////////////////////////////////////////////////////////////////////////////////
2547 // Virtual Memory
2548
2549 int os::vm_page_size() {
2550 // Seems redundant as all get out
2551 assert(os::Linux::page_size() != -1, "must call os::init");
2552 return os::Linux::page_size();
2553 }
2554
2555 // Solaris allocates memory by pages.
2556 int os::vm_allocation_granularity() {
2557 assert(os::Linux::page_size() != -1, "must call os::init");
2558 return os::Linux::page_size();
2559 }
2560
2561 // Rationale behind this function:
2562 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2563 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2564 // samples for JITted code. Here we create private executable mapping over the code cache
2565 // and then we can use standard (well, almost, as mapping can change) way to provide
2566 // info for the reporting script by storing timestamp and location of symbol
2567 void linux_wrap_code(char* base, size_t size) {
2568 static volatile jint cnt = 0;
2569
2570 if (!UseOprofile) {
2571 return;
2572 }
2573
2574 char buf[PATH_MAX+1];
2575 int num = Atomic::add(1, &cnt);
2576
2577 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2578 os::get_temp_directory(), os::current_process_id(), num);
2579 unlink(buf);
2580
2581 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2582
2583 if (fd != -1) {
2584 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2585 if (rv != (off_t)-1) {
2586 if (::write(fd, "", 1) == 1) {
2587 mmap(base, size,
2588 PROT_READ|PROT_WRITE|PROT_EXEC,
2589 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2590 }
2591 }
2592 ::close(fd);
2593 unlink(buf);
2594 }
2595 }
2596
2597 static bool recoverable_mmap_error(int err) {
2598 // See if the error is one we can let the caller handle. This
2599 // list of errno values comes from JBS-6843484. I can't find a
2600 // Linux man page that documents this specific set of errno
2601 // values so while this list currently matches Solaris, it may
2602 // change as we gain experience with this failure mode.
2603 switch (err) {
2604 case EBADF:
2605 case EINVAL:
2606 case ENOTSUP:
2607 // let the caller deal with these errors
2608 return true;
2609
2610 default:
2611 // Any remaining errors on this OS can cause our reserved mapping
2612 // to be lost. That can cause confusion where different data
2613 // structures think they have the same memory mapped. The worst
2614 // scenario is if both the VM and a library think they have the
2615 // same memory mapped.
2616 return false;
2617 }
2618 }
2619
2620 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2621 int err) {
2622 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2624 os::strerror(err), err);
2625 }
2626
2627 static void warn_fail_commit_memory(char* addr, size_t size,
2628 size_t alignment_hint, bool exec,
2629 int err) {
2630 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2631 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2632 alignment_hint, exec, os::strerror(err), err);
2633 }
2634
2635 // NOTE: Linux kernel does not really reserve the pages for us.
2636 // All it does is to check if there are enough free pages
2637 // left at the time of mmap(). This could be a potential
2638 // problem.
2639 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2640 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2641 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2642 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2643 if (res != (uintptr_t) MAP_FAILED) {
2644 if (UseNUMAInterleaving) {
2645 numa_make_global(addr, size);
2646 }
2647 return 0;
2648 }
2649
2650 int err = errno; // save errno from mmap() call above
2651
2652 if (!recoverable_mmap_error(err)) {
2653 warn_fail_commit_memory(addr, size, exec, err);
2654 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2655 }
2656
2657 return err;
2658 }
2659
2660 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2661 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2662 }
2663
2664 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2665 const char* mesg) {
2666 assert(mesg != NULL, "mesg must be specified");
2667 int err = os::Linux::commit_memory_impl(addr, size, exec);
2668 if (err != 0) {
2669 // the caller wants all commit errors to exit with the specified mesg:
2670 warn_fail_commit_memory(addr, size, exec, err);
2671 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2672 }
2673 }
2674
2675 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2676 #ifndef MAP_HUGETLB
2677 #define MAP_HUGETLB 0x40000
2678 #endif
2679
2680 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2681 #ifndef MADV_HUGEPAGE
2682 #define MADV_HUGEPAGE 14
2683 #endif
2684
2685 int os::Linux::commit_memory_impl(char* addr, size_t size,
2686 size_t alignment_hint, bool exec) {
2687 int err = os::Linux::commit_memory_impl(addr, size, exec);
2688 if (err == 0) {
2689 realign_memory(addr, size, alignment_hint);
2690 }
2691 return err;
2692 }
2693
2694 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2695 bool exec) {
2696 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2697 }
2698
2699 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2700 size_t alignment_hint, bool exec,
2701 const char* mesg) {
2702 assert(mesg != NULL, "mesg must be specified");
2703 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2704 if (err != 0) {
2705 // the caller wants all commit errors to exit with the specified mesg:
2706 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2708 }
2709 }
2710
2711 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2712 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2713 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2714 // be supported or the memory may already be backed by huge pages.
2715 ::madvise(addr, bytes, MADV_HUGEPAGE);
2716 }
2717 }
2718
2719 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2720 // This method works by doing an mmap over an existing mmaping and effectively discarding
2721 // the existing pages. However it won't work for SHM-based large pages that cannot be
2722 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2723 // small pages on top of the SHM segment. This method always works for small pages, so we
2724 // allow that in any case.
2725 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2726 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2727 }
2728 }
2729
2730 void os::numa_make_global(char *addr, size_t bytes) {
2731 Linux::numa_interleave_memory(addr, bytes);
2732 }
2733
2734 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2735 // bind policy to MPOL_PREFERRED for the current thread.
2736 #define USE_MPOL_PREFERRED 0
2737
2738 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2739 // To make NUMA and large pages more robust when both enabled, we need to ease
2740 // the requirements on where the memory should be allocated. MPOL_BIND is the
2741 // default policy and it will force memory to be allocated on the specified
2742 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2743 // the specified node, but will not force it. Using this policy will prevent
2744 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2745 // free large pages.
2746 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2747 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2748 }
2749
2750 bool os::numa_topology_changed() { return false; }
2751
2752 size_t os::numa_get_groups_num() {
2753 // Return just the number of nodes in which it's possible to allocate memory
2754 // (in numa terminology, configured nodes).
2755 return Linux::numa_num_configured_nodes();
2756 }
2757
2758 int os::numa_get_group_id() {
2759 int cpu_id = Linux::sched_getcpu();
2760 if (cpu_id != -1) {
2761 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2762 if (lgrp_id != -1) {
2763 return lgrp_id;
2764 }
2765 }
2766 return 0;
2767 }
2768
2769 int os::Linux::get_existing_num_nodes() {
2770 size_t node;
2771 size_t highest_node_number = Linux::numa_max_node();
2772 int num_nodes = 0;
2773
2774 // Get the total number of nodes in the system including nodes without memory.
2775 for (node = 0; node <= highest_node_number; node++) {
2776 if (isnode_in_existing_nodes(node)) {
2777 num_nodes++;
2778 }
2779 }
2780 return num_nodes;
2781 }
2782
2783 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2784 size_t highest_node_number = Linux::numa_max_node();
2785 size_t i = 0;
2786
2787 // Map all node ids in which is possible to allocate memory. Also nodes are
2788 // not always consecutively available, i.e. available from 0 to the highest
2789 // node number.
2790 for (size_t node = 0; node <= highest_node_number; node++) {
2791 if (Linux::isnode_in_configured_nodes(node)) {
2792 ids[i++] = node;
2793 }
2794 }
2795 return i;
2796 }
2797
2798 bool os::get_page_info(char *start, page_info* info) {
2799 return false;
2800 }
2801
2802 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2803 page_info* page_found) {
2804 return end;
2805 }
2806
2807
2808 int os::Linux::sched_getcpu_syscall(void) {
2809 unsigned int cpu = 0;
2810 int retval = -1;
2811
2812 #if defined(IA32)
2813 #ifndef SYS_getcpu
2814 #define SYS_getcpu 318
2815 #endif
2816 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2817 #elif defined(AMD64)
2818 // Unfortunately we have to bring all these macros here from vsyscall.h
2819 // to be able to compile on old linuxes.
2820 #define __NR_vgetcpu 2
2821 #define VSYSCALL_START (-10UL << 20)
2822 #define VSYSCALL_SIZE 1024
2823 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2824 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2825 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2826 retval = vgetcpu(&cpu, NULL, NULL);
2827 #endif
2828
2829 return (retval == -1) ? retval : cpu;
2830 }
2831
2832 // Something to do with the numa-aware allocator needs these symbols
2833 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2834 extern "C" JNIEXPORT void numa_error(char *where) { }
2835
2836 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2837 // load symbol from base version instead.
2838 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2839 void *f = dlvsym(handle, name, "libnuma_1.1");
2840 if (f == NULL) {
2841 f = dlsym(handle, name);
2842 }
2843 return f;
2844 }
2845
2846 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2847 // Return NULL if the symbol is not defined in this particular version.
2848 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2849 return dlvsym(handle, name, "libnuma_1.2");
2850 }
2851
2852
2853 bool os::Linux::libnuma_init() {
2854 // sched_getcpu() should be in libc.
2855 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2856 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2857
2858 // If it's not, try a direct syscall.
2859 if (sched_getcpu() == -1) {
2860 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2861 (void*)&sched_getcpu_syscall));
2862 }
2863
2864 if (sched_getcpu() != -1) { // Does it work?
2865 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2866 if (handle != NULL) {
2867 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2868 libnuma_dlsym(handle, "numa_node_to_cpus")));
2869 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2870 libnuma_dlsym(handle, "numa_max_node")));
2871 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2872 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2873 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2874 libnuma_dlsym(handle, "numa_available")));
2875 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2876 libnuma_dlsym(handle, "numa_tonode_memory")));
2877 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2878 libnuma_dlsym(handle, "numa_interleave_memory")));
2879 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2880 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2881 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2882 libnuma_dlsym(handle, "numa_set_bind_policy")));
2883 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2884 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2885 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2886 libnuma_dlsym(handle, "numa_distance")));
2887
2888 if (numa_available() != -1) {
2889 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2890 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2891 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2892 // Create an index -> node mapping, since nodes are not always consecutive
2893 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2894 rebuild_nindex_to_node_map();
2895 // Create a cpu -> node mapping
2896 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2897 rebuild_cpu_to_node_map();
2898 return true;
2899 }
2900 }
2901 }
2902 return false;
2903 }
2904
2905 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2906 // Creating guard page is very expensive. Java thread has HotSpot
2907 // guard pages, only enable glibc guard page for non-Java threads.
2908 // (Remember: compiler thread is a Java thread, too!)
2909 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2910 }
2911
2912 void os::Linux::rebuild_nindex_to_node_map() {
2913 int highest_node_number = Linux::numa_max_node();
2914
2915 nindex_to_node()->clear();
2916 for (int node = 0; node <= highest_node_number; node++) {
2917 if (Linux::isnode_in_existing_nodes(node)) {
2918 nindex_to_node()->append(node);
2919 }
2920 }
2921 }
2922
2923 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2924 // The table is later used in get_node_by_cpu().
2925 void os::Linux::rebuild_cpu_to_node_map() {
2926 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2927 // in libnuma (possible values are starting from 16,
2928 // and continuing up with every other power of 2, but less
2929 // than the maximum number of CPUs supported by kernel), and
2930 // is a subject to change (in libnuma version 2 the requirements
2931 // are more reasonable) we'll just hardcode the number they use
2932 // in the library.
2933 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2934
2935 size_t cpu_num = processor_count();
2936 size_t cpu_map_size = NCPUS / BitsPerCLong;
2937 size_t cpu_map_valid_size =
2938 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2939
2940 cpu_to_node()->clear();
2941 cpu_to_node()->at_grow(cpu_num - 1);
2942
2943 size_t node_num = get_existing_num_nodes();
2944
2945 int distance = 0;
2946 int closest_distance = INT_MAX;
2947 int closest_node = 0;
2948 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2949 for (size_t i = 0; i < node_num; i++) {
2950 // Check if node is configured (not a memory-less node). If it is not, find
2951 // the closest configured node.
2952 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2953 closest_distance = INT_MAX;
2954 // Check distance from all remaining nodes in the system. Ignore distance
2955 // from itself and from another non-configured node.
2956 for (size_t m = 0; m < node_num; m++) {
2957 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2958 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2959 // If a closest node is found, update. There is always at least one
2960 // configured node in the system so there is always at least one node
2961 // close.
2962 if (distance != 0 && distance < closest_distance) {
2963 closest_distance = distance;
2964 closest_node = nindex_to_node()->at(m);
2965 }
2966 }
2967 }
2968 } else {
2969 // Current node is already a configured node.
2970 closest_node = nindex_to_node()->at(i);
2971 }
2972
2973 // Get cpus from the original node and map them to the closest node. If node
2974 // is a configured node (not a memory-less node), then original node and
2975 // closest node are the same.
2976 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2977 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2978 if (cpu_map[j] != 0) {
2979 for (size_t k = 0; k < BitsPerCLong; k++) {
2980 if (cpu_map[j] & (1UL << k)) {
2981 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2982 }
2983 }
2984 }
2985 }
2986 }
2987 }
2988 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2989 }
2990
2991 int os::Linux::get_node_by_cpu(int cpu_id) {
2992 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2993 return cpu_to_node()->at(cpu_id);
2994 }
2995 return -1;
2996 }
2997
2998 GrowableArray<int>* os::Linux::_cpu_to_node;
2999 GrowableArray<int>* os::Linux::_nindex_to_node;
3000 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3001 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3002 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3003 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3004 os::Linux::numa_available_func_t os::Linux::_numa_available;
3005 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3006 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3007 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3008 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3009 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3010 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3011 unsigned long* os::Linux::_numa_all_nodes;
3012 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3013 struct bitmask* os::Linux::_numa_nodes_ptr;
3014
3015 bool os::pd_uncommit_memory(char* addr, size_t size) {
3016 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3017 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3018 return res != (uintptr_t) MAP_FAILED;
3019 }
3020
3021 static address get_stack_commited_bottom(address bottom, size_t size) {
3022 address nbot = bottom;
3023 address ntop = bottom + size;
3024
3025 size_t page_sz = os::vm_page_size();
3026 unsigned pages = size / page_sz;
3027
3028 unsigned char vec[1];
3029 unsigned imin = 1, imax = pages + 1, imid;
3030 int mincore_return_value = 0;
3031
3032 assert(imin <= imax, "Unexpected page size");
3033
3034 while (imin < imax) {
3035 imid = (imax + imin) / 2;
3036 nbot = ntop - (imid * page_sz);
3037
3038 // Use a trick with mincore to check whether the page is mapped or not.
3039 // mincore sets vec to 1 if page resides in memory and to 0 if page
3040 // is swapped output but if page we are asking for is unmapped
3041 // it returns -1,ENOMEM
3042 mincore_return_value = mincore(nbot, page_sz, vec);
3043
3044 if (mincore_return_value == -1) {
3045 // Page is not mapped go up
3046 // to find first mapped page
3047 if (errno != EAGAIN) {
3048 assert(errno == ENOMEM, "Unexpected mincore errno");
3049 imax = imid;
3050 }
3051 } else {
3052 // Page is mapped go down
3053 // to find first not mapped page
3054 imin = imid + 1;
3055 }
3056 }
3057
3058 nbot = nbot + page_sz;
3059
3060 // Adjust stack bottom one page up if last checked page is not mapped
3061 if (mincore_return_value == -1) {
3062 nbot = nbot + page_sz;
3063 }
3064
3065 return nbot;
3066 }
3067
3068
3069 // Linux uses a growable mapping for the stack, and if the mapping for
3070 // the stack guard pages is not removed when we detach a thread the
3071 // stack cannot grow beyond the pages where the stack guard was
3072 // mapped. If at some point later in the process the stack expands to
3073 // that point, the Linux kernel cannot expand the stack any further
3074 // because the guard pages are in the way, and a segfault occurs.
3075 //
3076 // However, it's essential not to split the stack region by unmapping
3077 // a region (leaving a hole) that's already part of the stack mapping,
3078 // so if the stack mapping has already grown beyond the guard pages at
3079 // the time we create them, we have to truncate the stack mapping.
3080 // So, we need to know the extent of the stack mapping when
3081 // create_stack_guard_pages() is called.
3082
3083 // We only need this for stacks that are growable: at the time of
3084 // writing thread stacks don't use growable mappings (i.e. those
3085 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3086 // only applies to the main thread.
3087
3088 // If the (growable) stack mapping already extends beyond the point
3089 // where we're going to put our guard pages, truncate the mapping at
3090 // that point by munmap()ping it. This ensures that when we later
3091 // munmap() the guard pages we don't leave a hole in the stack
3092 // mapping. This only affects the main/initial thread
3093
3094 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3095 if (os::Linux::is_initial_thread()) {
3096 // As we manually grow stack up to bottom inside create_attached_thread(),
3097 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3098 // we don't need to do anything special.
3099 // Check it first, before calling heavy function.
3100 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3101 unsigned char vec[1];
3102
3103 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3104 // Fallback to slow path on all errors, including EAGAIN
3105 stack_extent = (uintptr_t) get_stack_commited_bottom(
3106 os::Linux::initial_thread_stack_bottom(),
3107 (size_t)addr - stack_extent);
3108 }
3109
3110 if (stack_extent < (uintptr_t)addr) {
3111 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3112 }
3113 }
3114
3115 return os::commit_memory(addr, size, !ExecMem);
3116 }
3117
3118 // If this is a growable mapping, remove the guard pages entirely by
3119 // munmap()ping them. If not, just call uncommit_memory(). This only
3120 // affects the main/initial thread, but guard against future OS changes
3121 // It's safe to always unmap guard pages for initial thread because we
3122 // always place it right after end of the mapped region
3123
3124 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3125 uintptr_t stack_extent, stack_base;
3126
3127 if (os::Linux::is_initial_thread()) {
3128 return ::munmap(addr, size) == 0;
3129 }
3130
3131 return os::uncommit_memory(addr, size);
3132 }
3133
3134 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3135 // at 'requested_addr'. If there are existing memory mappings at the same
3136 // location, however, they will be overwritten. If 'fixed' is false,
3137 // 'requested_addr' is only treated as a hint, the return value may or
3138 // may not start from the requested address. Unlike Linux mmap(), this
3139 // function returns NULL to indicate failure.
3140 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3141 char * addr;
3142 int flags;
3143
3144 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3145 if (fixed) {
3146 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3147 flags |= MAP_FIXED;
3148 }
3149
3150 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3151 // touch an uncommitted page. Otherwise, the read/write might
3152 // succeed if we have enough swap space to back the physical page.
3153 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3154 flags, -1, 0);
3155
3156 return addr == MAP_FAILED ? NULL : addr;
3157 }
3158
3159 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3160 // (req_addr != NULL) or with a given alignment.
3161 // - bytes shall be a multiple of alignment.
3162 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3163 // - alignment sets the alignment at which memory shall be allocated.
3164 // It must be a multiple of allocation granularity.
3165 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3166 // req_addr or NULL.
3167 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3168
3169 size_t extra_size = bytes;
3170 if (req_addr == NULL && alignment > 0) {
3171 extra_size += alignment;
3172 }
3173
3174 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3175 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3176 -1, 0);
3177 if (start == MAP_FAILED) {
3178 start = NULL;
3179 } else {
3180 if (req_addr != NULL) {
3181 if (start != req_addr) {
3182 ::munmap(start, extra_size);
3183 start = NULL;
3184 }
3185 } else {
3186 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3187 char* const end_aligned = start_aligned + bytes;
3188 char* const end = start + extra_size;
3189 if (start_aligned > start) {
3190 ::munmap(start, start_aligned - start);
3191 }
3192 if (end_aligned < end) {
3193 ::munmap(end_aligned, end - end_aligned);
3194 }
3195 start = start_aligned;
3196 }
3197 }
3198 return start;
3199 }
3200
3201 static int anon_munmap(char * addr, size_t size) {
3202 return ::munmap(addr, size) == 0;
3203 }
3204
3205 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3206 size_t alignment_hint) {
3207 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3208 }
3209
3210 bool os::pd_release_memory(char* addr, size_t size) {
3211 return anon_munmap(addr, size);
3212 }
3213
3214 static bool linux_mprotect(char* addr, size_t size, int prot) {
3215 // Linux wants the mprotect address argument to be page aligned.
3216 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3217
3218 // According to SUSv3, mprotect() should only be used with mappings
3219 // established by mmap(), and mmap() always maps whole pages. Unaligned
3220 // 'addr' likely indicates problem in the VM (e.g. trying to change
3221 // protection of malloc'ed or statically allocated memory). Check the
3222 // caller if you hit this assert.
3223 assert(addr == bottom, "sanity check");
3224
3225 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3226 return ::mprotect(bottom, size, prot) == 0;
3227 }
3228
3229 // Set protections specified
3230 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3231 bool is_committed) {
3232 unsigned int p = 0;
3233 switch (prot) {
3234 case MEM_PROT_NONE: p = PROT_NONE; break;
3235 case MEM_PROT_READ: p = PROT_READ; break;
3236 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3237 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3238 default:
3239 ShouldNotReachHere();
3240 }
3241 // is_committed is unused.
3242 return linux_mprotect(addr, bytes, p);
3243 }
3244
3245 bool os::guard_memory(char* addr, size_t size) {
3246 return linux_mprotect(addr, size, PROT_NONE);
3247 }
3248
3249 bool os::unguard_memory(char* addr, size_t size) {
3250 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3251 }
3252
3253 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3254 size_t page_size) {
3255 bool result = false;
3256 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3257 MAP_ANONYMOUS|MAP_PRIVATE,
3258 -1, 0);
3259 if (p != MAP_FAILED) {
3260 void *aligned_p = align_ptr_up(p, page_size);
3261
3262 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3263
3264 munmap(p, page_size * 2);
3265 }
3266
3267 if (warn && !result) {
3268 warning("TransparentHugePages is not supported by the operating system.");
3269 }
3270
3271 return result;
3272 }
3273
3274 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3275 bool result = false;
3276 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3277 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3278 -1, 0);
3279
3280 if (p != MAP_FAILED) {
3281 // We don't know if this really is a huge page or not.
3282 FILE *fp = fopen("/proc/self/maps", "r");
3283 if (fp) {
3284 while (!feof(fp)) {
3285 char chars[257];
3286 long x = 0;
3287 if (fgets(chars, sizeof(chars), fp)) {
3288 if (sscanf(chars, "%lx-%*x", &x) == 1
3289 && x == (long)p) {
3290 if (strstr (chars, "hugepage")) {
3291 result = true;
3292 break;
3293 }
3294 }
3295 }
3296 }
3297 fclose(fp);
3298 }
3299 munmap(p, page_size);
3300 }
3301
3302 if (warn && !result) {
3303 warning("HugeTLBFS is not supported by the operating system.");
3304 }
3305
3306 return result;
3307 }
3308
3309 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3310 //
3311 // From the coredump_filter documentation:
3312 //
3313 // - (bit 0) anonymous private memory
3314 // - (bit 1) anonymous shared memory
3315 // - (bit 2) file-backed private memory
3316 // - (bit 3) file-backed shared memory
3317 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3318 // effective only if the bit 2 is cleared)
3319 // - (bit 5) hugetlb private memory
3320 // - (bit 6) hugetlb shared memory
3321 //
3322 static void set_coredump_filter(void) {
3323 FILE *f;
3324 long cdm;
3325
3326 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3327 return;
3328 }
3329
3330 if (fscanf(f, "%lx", &cdm) != 1) {
3331 fclose(f);
3332 return;
3333 }
3334
3335 rewind(f);
3336
3337 if ((cdm & LARGEPAGES_BIT) == 0) {
3338 cdm |= LARGEPAGES_BIT;
3339 fprintf(f, "%#lx", cdm);
3340 }
3341
3342 fclose(f);
3343 }
3344
3345 // Large page support
3346
3347 static size_t _large_page_size = 0;
3348
3349 size_t os::Linux::find_large_page_size() {
3350 size_t large_page_size = 0;
3351
3352 // large_page_size on Linux is used to round up heap size. x86 uses either
3353 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3354 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3355 // page as large as 256M.
3356 //
3357 // Here we try to figure out page size by parsing /proc/meminfo and looking
3358 // for a line with the following format:
3359 // Hugepagesize: 2048 kB
3360 //
3361 // If we can't determine the value (e.g. /proc is not mounted, or the text
3362 // format has been changed), we'll use the largest page size supported by
3363 // the processor.
3364
3365 #ifndef ZERO
3366 large_page_size =
3367 AARCH64_ONLY(2 * M)
3368 AMD64_ONLY(2 * M)
3369 ARM32_ONLY(2 * M)
3370 IA32_ONLY(4 * M)
3371 IA64_ONLY(256 * M)
3372 PPC_ONLY(4 * M)
3373 S390_ONLY(1 * M)
3374 SPARC_ONLY(4 * M);
3375 #endif // ZERO
3376
3377 FILE *fp = fopen("/proc/meminfo", "r");
3378 if (fp) {
3379 while (!feof(fp)) {
3380 int x = 0;
3381 char buf[16];
3382 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3383 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3384 large_page_size = x * K;
3385 break;
3386 }
3387 } else {
3388 // skip to next line
3389 for (;;) {
3390 int ch = fgetc(fp);
3391 if (ch == EOF || ch == (int)'\n') break;
3392 }
3393 }
3394 }
3395 fclose(fp);
3396 }
3397
3398 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3399 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3400 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3401 proper_unit_for_byte_size(large_page_size));
3402 }
3403
3404 return large_page_size;
3405 }
3406
3407 size_t os::Linux::setup_large_page_size() {
3408 _large_page_size = Linux::find_large_page_size();
3409 const size_t default_page_size = (size_t)Linux::page_size();
3410 if (_large_page_size > default_page_size) {
3411 _page_sizes[0] = _large_page_size;
3412 _page_sizes[1] = default_page_size;
3413 _page_sizes[2] = 0;
3414 }
3415
3416 return _large_page_size;
3417 }
3418
3419 bool os::Linux::setup_large_page_type(size_t page_size) {
3420 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3421 FLAG_IS_DEFAULT(UseSHM) &&
3422 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3423
3424 // The type of large pages has not been specified by the user.
3425
3426 // Try UseHugeTLBFS and then UseSHM.
3427 UseHugeTLBFS = UseSHM = true;
3428
3429 // Don't try UseTransparentHugePages since there are known
3430 // performance issues with it turned on. This might change in the future.
3431 UseTransparentHugePages = false;
3432 }
3433
3434 if (UseTransparentHugePages) {
3435 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3436 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3437 UseHugeTLBFS = false;
3438 UseSHM = false;
3439 return true;
3440 }
3441 UseTransparentHugePages = false;
3442 }
3443
3444 if (UseHugeTLBFS) {
3445 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3446 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3447 UseSHM = false;
3448 return true;
3449 }
3450 UseHugeTLBFS = false;
3451 }
3452
3453 return UseSHM;
3454 }
3455
3456 void os::large_page_init() {
3457 if (!UseLargePages &&
3458 !UseTransparentHugePages &&
3459 !UseHugeTLBFS &&
3460 !UseSHM) {
3461 // Not using large pages.
3462 return;
3463 }
3464
3465 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3466 // The user explicitly turned off large pages.
3467 // Ignore the rest of the large pages flags.
3468 UseTransparentHugePages = false;
3469 UseHugeTLBFS = false;
3470 UseSHM = false;
3471 return;
3472 }
3473
3474 size_t large_page_size = Linux::setup_large_page_size();
3475 UseLargePages = Linux::setup_large_page_type(large_page_size);
3476
3477 set_coredump_filter();
3478 }
3479
3480 #ifndef SHM_HUGETLB
3481 #define SHM_HUGETLB 04000
3482 #endif
3483
3484 #define shm_warning_format(format, ...) \
3485 do { \
3486 if (UseLargePages && \
3487 (!FLAG_IS_DEFAULT(UseLargePages) || \
3488 !FLAG_IS_DEFAULT(UseSHM) || \
3489 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3490 warning(format, __VA_ARGS__); \
3491 } \
3492 } while (0)
3493
3494 #define shm_warning(str) shm_warning_format("%s", str)
3495
3496 #define shm_warning_with_errno(str) \
3497 do { \
3498 int err = errno; \
3499 shm_warning_format(str " (error = %d)", err); \
3500 } while (0)
3501
3502 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3503 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3504
3505 if (!is_size_aligned(alignment, SHMLBA)) {
3506 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3507 return NULL;
3508 }
3509
3510 // To ensure that we get 'alignment' aligned memory from shmat,
3511 // we pre-reserve aligned virtual memory and then attach to that.
3512
3513 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3514 if (pre_reserved_addr == NULL) {
3515 // Couldn't pre-reserve aligned memory.
3516 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3517 return NULL;
3518 }
3519
3520 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3521 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3522
3523 if ((intptr_t)addr == -1) {
3524 int err = errno;
3525 shm_warning_with_errno("Failed to attach shared memory.");
3526
3527 assert(err != EACCES, "Unexpected error");
3528 assert(err != EIDRM, "Unexpected error");
3529 assert(err != EINVAL, "Unexpected error");
3530
3531 // Since we don't know if the kernel unmapped the pre-reserved memory area
3532 // we can't unmap it, since that would potentially unmap memory that was
3533 // mapped from other threads.
3534 return NULL;
3535 }
3536
3537 return addr;
3538 }
3539
3540 static char* shmat_at_address(int shmid, char* req_addr) {
3541 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3542 assert(false, "Requested address needs to be SHMLBA aligned");
3543 return NULL;
3544 }
3545
3546 char* addr = (char*)shmat(shmid, req_addr, 0);
3547
3548 if ((intptr_t)addr == -1) {
3549 shm_warning_with_errno("Failed to attach shared memory.");
3550 return NULL;
3551 }
3552
3553 return addr;
3554 }
3555
3556 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3557 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3558 if (req_addr != NULL) {
3559 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3560 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3561 return shmat_at_address(shmid, req_addr);
3562 }
3563
3564 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3565 // return large page size aligned memory addresses when req_addr == NULL.
3566 // However, if the alignment is larger than the large page size, we have
3567 // to manually ensure that the memory returned is 'alignment' aligned.
3568 if (alignment > os::large_page_size()) {
3569 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3570 return shmat_with_alignment(shmid, bytes, alignment);
3571 } else {
3572 return shmat_at_address(shmid, NULL);
3573 }
3574 }
3575
3576 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3577 char* req_addr, bool exec) {
3578 // "exec" is passed in but not used. Creating the shared image for
3579 // the code cache doesn't have an SHM_X executable permission to check.
3580 assert(UseLargePages && UseSHM, "only for SHM large pages");
3581 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3582 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3583
3584 if (!is_size_aligned(bytes, os::large_page_size())) {
3585 return NULL; // Fallback to small pages.
3586 }
3587
3588 // Create a large shared memory region to attach to based on size.
3589 // Currently, size is the total size of the heap.
3590 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3591 if (shmid == -1) {
3592 // Possible reasons for shmget failure:
3593 // 1. shmmax is too small for Java heap.
3594 // > check shmmax value: cat /proc/sys/kernel/shmmax
3595 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3596 // 2. not enough large page memory.
3597 // > check available large pages: cat /proc/meminfo
3598 // > increase amount of large pages:
3599 // echo new_value > /proc/sys/vm/nr_hugepages
3600 // Note 1: different Linux may use different name for this property,
3601 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3602 // Note 2: it's possible there's enough physical memory available but
3603 // they are so fragmented after a long run that they can't
3604 // coalesce into large pages. Try to reserve large pages when
3605 // the system is still "fresh".
3606 shm_warning_with_errno("Failed to reserve shared memory.");
3607 return NULL;
3608 }
3609
3610 // Attach to the region.
3611 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3612
3613 // Remove shmid. If shmat() is successful, the actual shared memory segment
3614 // will be deleted when it's detached by shmdt() or when the process
3615 // terminates. If shmat() is not successful this will remove the shared
3616 // segment immediately.
3617 shmctl(shmid, IPC_RMID, NULL);
3618
3619 return addr;
3620 }
3621
3622 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3623 int error) {
3624 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3625
3626 bool warn_on_failure = UseLargePages &&
3627 (!FLAG_IS_DEFAULT(UseLargePages) ||
3628 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3629 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3630
3631 if (warn_on_failure) {
3632 char msg[128];
3633 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3634 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3635 warning("%s", msg);
3636 }
3637 }
3638
3639 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3640 char* req_addr,
3641 bool exec) {
3642 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3643 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3644 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3645
3646 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3647 char* addr = (char*)::mmap(req_addr, bytes, prot,
3648 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3649 -1, 0);
3650
3651 if (addr == MAP_FAILED) {
3652 warn_on_large_pages_failure(req_addr, bytes, errno);
3653 return NULL;
3654 }
3655
3656 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3657
3658 return addr;
3659 }
3660
3661 // Reserve memory using mmap(MAP_HUGETLB).
3662 // - bytes shall be a multiple of alignment.
3663 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3664 // - alignment sets the alignment at which memory shall be allocated.
3665 // It must be a multiple of allocation granularity.
3666 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3667 // req_addr or NULL.
3668 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3669 size_t alignment,
3670 char* req_addr,
3671 bool exec) {
3672 size_t large_page_size = os::large_page_size();
3673 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3674
3675 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3676 assert(is_size_aligned(bytes, alignment), "Must be");
3677
3678 // First reserve - but not commit - the address range in small pages.
3679 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3680
3681 if (start == NULL) {
3682 return NULL;
3683 }
3684
3685 assert(is_ptr_aligned(start, alignment), "Must be");
3686
3687 char* end = start + bytes;
3688
3689 // Find the regions of the allocated chunk that can be promoted to large pages.
3690 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3691 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3692
3693 size_t lp_bytes = lp_end - lp_start;
3694
3695 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3696
3697 if (lp_bytes == 0) {
3698 // The mapped region doesn't even span the start and the end of a large page.
3699 // Fall back to allocate a non-special area.
3700 ::munmap(start, end - start);
3701 return NULL;
3702 }
3703
3704 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3705
3706 void* result;
3707
3708 // Commit small-paged leading area.
3709 if (start != lp_start) {
3710 result = ::mmap(start, lp_start - start, prot,
3711 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3712 -1, 0);
3713 if (result == MAP_FAILED) {
3714 ::munmap(lp_start, end - lp_start);
3715 return NULL;
3716 }
3717 }
3718
3719 // Commit large-paged area.
3720 result = ::mmap(lp_start, lp_bytes, prot,
3721 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3722 -1, 0);
3723 if (result == MAP_FAILED) {
3724 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3725 // If the mmap above fails, the large pages region will be unmapped and we
3726 // have regions before and after with small pages. Release these regions.
3727 //
3728 // | mapped | unmapped | mapped |
3729 // ^ ^ ^ ^
3730 // start lp_start lp_end end
3731 //
3732 ::munmap(start, lp_start - start);
3733 ::munmap(lp_end, end - lp_end);
3734 return NULL;
3735 }
3736
3737 // Commit small-paged trailing area.
3738 if (lp_end != end) {
3739 result = ::mmap(lp_end, end - lp_end, prot,
3740 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3741 -1, 0);
3742 if (result == MAP_FAILED) {
3743 ::munmap(start, lp_end - start);
3744 return NULL;
3745 }
3746 }
3747
3748 return start;
3749 }
3750
3751 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3752 size_t alignment,
3753 char* req_addr,
3754 bool exec) {
3755 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3756 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3757 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3758 assert(is_power_of_2(os::large_page_size()), "Must be");
3759 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3760
3761 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3762 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3763 } else {
3764 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3765 }
3766 }
3767
3768 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3769 char* req_addr, bool exec) {
3770 assert(UseLargePages, "only for large pages");
3771
3772 char* addr;
3773 if (UseSHM) {
3774 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3775 } else {
3776 assert(UseHugeTLBFS, "must be");
3777 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3778 }
3779
3780 if (addr != NULL) {
3781 if (UseNUMAInterleaving) {
3782 numa_make_global(addr, bytes);
3783 }
3784
3785 // The memory is committed
3786 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3787 }
3788
3789 return addr;
3790 }
3791
3792 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3793 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3794 return shmdt(base) == 0;
3795 }
3796
3797 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3798 return pd_release_memory(base, bytes);
3799 }
3800
3801 bool os::release_memory_special(char* base, size_t bytes) {
3802 bool res;
3803 if (MemTracker::tracking_level() > NMT_minimal) {
3804 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3805 res = os::Linux::release_memory_special_impl(base, bytes);
3806 if (res) {
3807 tkr.record((address)base, bytes);
3808 }
3809
3810 } else {
3811 res = os::Linux::release_memory_special_impl(base, bytes);
3812 }
3813 return res;
3814 }
3815
3816 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3817 assert(UseLargePages, "only for large pages");
3818 bool res;
3819
3820 if (UseSHM) {
3821 res = os::Linux::release_memory_special_shm(base, bytes);
3822 } else {
3823 assert(UseHugeTLBFS, "must be");
3824 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3825 }
3826 return res;
3827 }
3828
3829 size_t os::large_page_size() {
3830 return _large_page_size;
3831 }
3832
3833 // With SysV SHM the entire memory region must be allocated as shared
3834 // memory.
3835 // HugeTLBFS allows application to commit large page memory on demand.
3836 // However, when committing memory with HugeTLBFS fails, the region
3837 // that was supposed to be committed will lose the old reservation
3838 // and allow other threads to steal that memory region. Because of this
3839 // behavior we can't commit HugeTLBFS memory.
3840 bool os::can_commit_large_page_memory() {
3841 return UseTransparentHugePages;
3842 }
3843
3844 bool os::can_execute_large_page_memory() {
3845 return UseTransparentHugePages || UseHugeTLBFS;
3846 }
3847
3848 // Reserve memory at an arbitrary address, only if that area is
3849 // available (and not reserved for something else).
3850
3851 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3852 const int max_tries = 10;
3853 char* base[max_tries];
3854 size_t size[max_tries];
3855 const size_t gap = 0x000000;
3856
3857 // Assert only that the size is a multiple of the page size, since
3858 // that's all that mmap requires, and since that's all we really know
3859 // about at this low abstraction level. If we need higher alignment,
3860 // we can either pass an alignment to this method or verify alignment
3861 // in one of the methods further up the call chain. See bug 5044738.
3862 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3863
3864 // Repeatedly allocate blocks until the block is allocated at the
3865 // right spot.
3866
3867 // Linux mmap allows caller to pass an address as hint; give it a try first,
3868 // if kernel honors the hint then we can return immediately.
3869 char * addr = anon_mmap(requested_addr, bytes, false);
3870 if (addr == requested_addr) {
3871 return requested_addr;
3872 }
3873
3874 if (addr != NULL) {
3875 // mmap() is successful but it fails to reserve at the requested address
3876 anon_munmap(addr, bytes);
3877 }
3878
3879 int i;
3880 for (i = 0; i < max_tries; ++i) {
3881 base[i] = reserve_memory(bytes);
3882
3883 if (base[i] != NULL) {
3884 // Is this the block we wanted?
3885 if (base[i] == requested_addr) {
3886 size[i] = bytes;
3887 break;
3888 }
3889
3890 // Does this overlap the block we wanted? Give back the overlapped
3891 // parts and try again.
3892
3893 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3894 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3895 unmap_memory(base[i], top_overlap);
3896 base[i] += top_overlap;
3897 size[i] = bytes - top_overlap;
3898 } else {
3899 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3900 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3901 unmap_memory(requested_addr, bottom_overlap);
3902 size[i] = bytes - bottom_overlap;
3903 } else {
3904 size[i] = bytes;
3905 }
3906 }
3907 }
3908 }
3909
3910 // Give back the unused reserved pieces.
3911
3912 for (int j = 0; j < i; ++j) {
3913 if (base[j] != NULL) {
3914 unmap_memory(base[j], size[j]);
3915 }
3916 }
3917
3918 if (i < max_tries) {
3919 return requested_addr;
3920 } else {
3921 return NULL;
3922 }
3923 }
3924
3925 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3926 return ::read(fd, buf, nBytes);
3927 }
3928
3929 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3930 return ::pread(fd, buf, nBytes, offset);
3931 }
3932
3933 // Short sleep, direct OS call.
3934 //
3935 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3936 // sched_yield(2) will actually give up the CPU:
3937 //
3938 // * Alone on this pariticular CPU, keeps running.
3939 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3940 // (pre 2.6.39).
3941 //
3942 // So calling this with 0 is an alternative.
3943 //
3944 void os::naked_short_sleep(jlong ms) {
3945 struct timespec req;
3946
3947 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3948 req.tv_sec = 0;
3949 if (ms > 0) {
3950 req.tv_nsec = (ms % 1000) * 1000000;
3951 } else {
3952 req.tv_nsec = 1;
3953 }
3954
3955 nanosleep(&req, NULL);
3956
3957 return;
3958 }
3959
3960 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3961 void os::infinite_sleep() {
3962 while (true) { // sleep forever ...
3963 ::sleep(100); // ... 100 seconds at a time
3964 }
3965 }
3966
3967 // Used to convert frequent JVM_Yield() to nops
3968 bool os::dont_yield() {
3969 return DontYieldALot;
3970 }
3971
3972 void os::naked_yield() {
3973 sched_yield();
3974 }
3975
3976 ////////////////////////////////////////////////////////////////////////////////
3977 // thread priority support
3978
3979 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3980 // only supports dynamic priority, static priority must be zero. For real-time
3981 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3982 // However, for large multi-threaded applications, SCHED_RR is not only slower
3983 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3984 // of 5 runs - Sep 2005).
3985 //
3986 // The following code actually changes the niceness of kernel-thread/LWP. It
3987 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3988 // not the entire user process, and user level threads are 1:1 mapped to kernel
3989 // threads. It has always been the case, but could change in the future. For
3990 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3991 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3992
3993 int os::java_to_os_priority[CriticalPriority + 1] = {
3994 19, // 0 Entry should never be used
3995
3996 4, // 1 MinPriority
3997 3, // 2
3998 2, // 3
3999
4000 1, // 4
4001 0, // 5 NormPriority
4002 -1, // 6
4003
4004 -2, // 7
4005 -3, // 8
4006 -4, // 9 NearMaxPriority
4007
4008 -5, // 10 MaxPriority
4009
4010 -5 // 11 CriticalPriority
4011 };
4012
4013 static int prio_init() {
4014 if (ThreadPriorityPolicy == 1) {
4015 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4016 // if effective uid is not root. Perhaps, a more elegant way of doing
4017 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4018 if (geteuid() != 0) {
4019 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4020 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4021 }
4022 ThreadPriorityPolicy = 0;
4023 }
4024 }
4025 if (UseCriticalJavaThreadPriority) {
4026 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4027 }
4028 return 0;
4029 }
4030
4031 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4032 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4033
4034 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4035 return (ret == 0) ? OS_OK : OS_ERR;
4036 }
4037
4038 OSReturn os::get_native_priority(const Thread* const thread,
4039 int *priority_ptr) {
4040 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4041 *priority_ptr = java_to_os_priority[NormPriority];
4042 return OS_OK;
4043 }
4044
4045 errno = 0;
4046 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4047 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4048 }
4049
4050 // Hint to the underlying OS that a task switch would not be good.
4051 // Void return because it's a hint and can fail.
4052 void os::hint_no_preempt() {}
4053
4054 ////////////////////////////////////////////////////////////////////////////////
4055 // suspend/resume support
4056
4057 // the low-level signal-based suspend/resume support is a remnant from the
4058 // old VM-suspension that used to be for java-suspension, safepoints etc,
4059 // within hotspot. Now there is a single use-case for this:
4060 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4061 // that runs in the watcher thread.
4062 // The remaining code is greatly simplified from the more general suspension
4063 // code that used to be used.
4064 //
4065 // The protocol is quite simple:
4066 // - suspend:
4067 // - sends a signal to the target thread
4068 // - polls the suspend state of the osthread using a yield loop
4069 // - target thread signal handler (SR_handler) sets suspend state
4070 // and blocks in sigsuspend until continued
4071 // - resume:
4072 // - sets target osthread state to continue
4073 // - sends signal to end the sigsuspend loop in the SR_handler
4074 //
4075 // Note that the SR_lock plays no role in this suspend/resume protocol,
4076 // but is checked for NULL in SR_handler as a thread termination indicator.
4077
4078 static void resume_clear_context(OSThread *osthread) {
4079 osthread->set_ucontext(NULL);
4080 osthread->set_siginfo(NULL);
4081 }
4082
4083 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4084 ucontext_t* context) {
4085 osthread->set_ucontext(context);
4086 osthread->set_siginfo(siginfo);
4087 }
4088
4089 // Handler function invoked when a thread's execution is suspended or
4090 // resumed. We have to be careful that only async-safe functions are
4091 // called here (Note: most pthread functions are not async safe and
4092 // should be avoided.)
4093 //
4094 // Note: sigwait() is a more natural fit than sigsuspend() from an
4095 // interface point of view, but sigwait() prevents the signal hander
4096 // from being run. libpthread would get very confused by not having
4097 // its signal handlers run and prevents sigwait()'s use with the
4098 // mutex granting granting signal.
4099 //
4100 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4101 //
4102 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4103 // Save and restore errno to avoid confusing native code with EINTR
4104 // after sigsuspend.
4105 int old_errno = errno;
4106
4107 Thread* thread = Thread::current_or_null_safe();
4108 assert(thread != NULL, "Missing current thread in SR_handler");
4109
4110 // On some systems we have seen signal delivery get "stuck" until the signal
4111 // mask is changed as part of thread termination. Check that the current thread
4112 // has not already terminated (via SR_lock()) - else the following assertion
4113 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4114 // destructor has completed.
4115
4116 if (thread->SR_lock() == NULL) {
4117 return;
4118 }
4119
4120 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4121
4122 OSThread* osthread = thread->osthread();
4123
4124 os::SuspendResume::State current = osthread->sr.state();
4125 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4126 suspend_save_context(osthread, siginfo, context);
4127
4128 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4129 os::SuspendResume::State state = osthread->sr.suspended();
4130 if (state == os::SuspendResume::SR_SUSPENDED) {
4131 sigset_t suspend_set; // signals for sigsuspend()
4132 sigemptyset(&suspend_set);
4133 // get current set of blocked signals and unblock resume signal
4134 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4135 sigdelset(&suspend_set, SR_signum);
4136
4137 sr_semaphore.signal();
4138 // wait here until we are resumed
4139 while (1) {
4140 sigsuspend(&suspend_set);
4141
4142 os::SuspendResume::State result = osthread->sr.running();
4143 if (result == os::SuspendResume::SR_RUNNING) {
4144 sr_semaphore.signal();
4145 break;
4146 }
4147 }
4148
4149 } else if (state == os::SuspendResume::SR_RUNNING) {
4150 // request was cancelled, continue
4151 } else {
4152 ShouldNotReachHere();
4153 }
4154
4155 resume_clear_context(osthread);
4156 } else if (current == os::SuspendResume::SR_RUNNING) {
4157 // request was cancelled, continue
4158 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4159 // ignore
4160 } else {
4161 // ignore
4162 }
4163
4164 errno = old_errno;
4165 }
4166
4167 static int SR_initialize() {
4168 struct sigaction act;
4169 char *s;
4170
4171 // Get signal number to use for suspend/resume
4172 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4173 int sig = ::strtol(s, 0, 10);
4174 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4175 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4176 SR_signum = sig;
4177 } else {
4178 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4179 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4180 }
4181 }
4182
4183 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4184 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4185
4186 sigemptyset(&SR_sigset);
4187 sigaddset(&SR_sigset, SR_signum);
4188
4189 // Set up signal handler for suspend/resume
4190 act.sa_flags = SA_RESTART|SA_SIGINFO;
4191 act.sa_handler = (void (*)(int)) SR_handler;
4192
4193 // SR_signum is blocked by default.
4194 // 4528190 - We also need to block pthread restart signal (32 on all
4195 // supported Linux platforms). Note that LinuxThreads need to block
4196 // this signal for all threads to work properly. So we don't have
4197 // to use hard-coded signal number when setting up the mask.
4198 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4199
4200 if (sigaction(SR_signum, &act, 0) == -1) {
4201 return -1;
4202 }
4203
4204 // Save signal flag
4205 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4206 return 0;
4207 }
4208
4209 static int sr_notify(OSThread* osthread) {
4210 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4211 assert_status(status == 0, status, "pthread_kill");
4212 return status;
4213 }
4214
4215 // "Randomly" selected value for how long we want to spin
4216 // before bailing out on suspending a thread, also how often
4217 // we send a signal to a thread we want to resume
4218 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4219 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4220
4221 // returns true on success and false on error - really an error is fatal
4222 // but this seems the normal response to library errors
4223 static bool do_suspend(OSThread* osthread) {
4224 assert(osthread->sr.is_running(), "thread should be running");
4225 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4226
4227 // mark as suspended and send signal
4228 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4229 // failed to switch, state wasn't running?
4230 ShouldNotReachHere();
4231 return false;
4232 }
4233
4234 if (sr_notify(osthread) != 0) {
4235 ShouldNotReachHere();
4236 }
4237
4238 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4239 while (true) {
4240 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4241 break;
4242 } else {
4243 // timeout
4244 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4245 if (cancelled == os::SuspendResume::SR_RUNNING) {
4246 return false;
4247 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4248 // make sure that we consume the signal on the semaphore as well
4249 sr_semaphore.wait();
4250 break;
4251 } else {
4252 ShouldNotReachHere();
4253 return false;
4254 }
4255 }
4256 }
4257
4258 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4259 return true;
4260 }
4261
4262 static void do_resume(OSThread* osthread) {
4263 assert(osthread->sr.is_suspended(), "thread should be suspended");
4264 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4265
4266 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4267 // failed to switch to WAKEUP_REQUEST
4268 ShouldNotReachHere();
4269 return;
4270 }
4271
4272 while (true) {
4273 if (sr_notify(osthread) == 0) {
4274 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4275 if (osthread->sr.is_running()) {
4276 return;
4277 }
4278 }
4279 } else {
4280 ShouldNotReachHere();
4281 }
4282 }
4283
4284 guarantee(osthread->sr.is_running(), "Must be running!");
4285 }
4286
4287 ///////////////////////////////////////////////////////////////////////////////////
4288 // signal handling (except suspend/resume)
4289
4290 // This routine may be used by user applications as a "hook" to catch signals.
4291 // The user-defined signal handler must pass unrecognized signals to this
4292 // routine, and if it returns true (non-zero), then the signal handler must
4293 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4294 // routine will never retun false (zero), but instead will execute a VM panic
4295 // routine kill the process.
4296 //
4297 // If this routine returns false, it is OK to call it again. This allows
4298 // the user-defined signal handler to perform checks either before or after
4299 // the VM performs its own checks. Naturally, the user code would be making
4300 // a serious error if it tried to handle an exception (such as a null check
4301 // or breakpoint) that the VM was generating for its own correct operation.
4302 //
4303 // This routine may recognize any of the following kinds of signals:
4304 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4305 // It should be consulted by handlers for any of those signals.
4306 //
4307 // The caller of this routine must pass in the three arguments supplied
4308 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4309 // field of the structure passed to sigaction(). This routine assumes that
4310 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4311 //
4312 // Note that the VM will print warnings if it detects conflicting signal
4313 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4314 //
4315 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4316 siginfo_t* siginfo,
4317 void* ucontext,
4318 int abort_if_unrecognized);
4319
4320 void signalHandler(int sig, siginfo_t* info, void* uc) {
4321 assert(info != NULL && uc != NULL, "it must be old kernel");
4322 int orig_errno = errno; // Preserve errno value over signal handler.
4323 JVM_handle_linux_signal(sig, info, uc, true);
4324 errno = orig_errno;
4325 }
4326
4327
4328 // This boolean allows users to forward their own non-matching signals
4329 // to JVM_handle_linux_signal, harmlessly.
4330 bool os::Linux::signal_handlers_are_installed = false;
4331
4332 // For signal-chaining
4333 struct sigaction sigact[NSIG];
4334 uint64_t sigs = 0;
4335 #if (64 < NSIG-1)
4336 #error "Not all signals can be encoded in sigs. Adapt its type!"
4337 #endif
4338 bool os::Linux::libjsig_is_loaded = false;
4339 typedef struct sigaction *(*get_signal_t)(int);
4340 get_signal_t os::Linux::get_signal_action = NULL;
4341
4342 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4343 struct sigaction *actp = NULL;
4344
4345 if (libjsig_is_loaded) {
4346 // Retrieve the old signal handler from libjsig
4347 actp = (*get_signal_action)(sig);
4348 }
4349 if (actp == NULL) {
4350 // Retrieve the preinstalled signal handler from jvm
4351 actp = get_preinstalled_handler(sig);
4352 }
4353
4354 return actp;
4355 }
4356
4357 static bool call_chained_handler(struct sigaction *actp, int sig,
4358 siginfo_t *siginfo, void *context) {
4359 // Call the old signal handler
4360 if (actp->sa_handler == SIG_DFL) {
4361 // It's more reasonable to let jvm treat it as an unexpected exception
4362 // instead of taking the default action.
4363 return false;
4364 } else if (actp->sa_handler != SIG_IGN) {
4365 if ((actp->sa_flags & SA_NODEFER) == 0) {
4366 // automaticlly block the signal
4367 sigaddset(&(actp->sa_mask), sig);
4368 }
4369
4370 sa_handler_t hand = NULL;
4371 sa_sigaction_t sa = NULL;
4372 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4373 // retrieve the chained handler
4374 if (siginfo_flag_set) {
4375 sa = actp->sa_sigaction;
4376 } else {
4377 hand = actp->sa_handler;
4378 }
4379
4380 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4381 actp->sa_handler = SIG_DFL;
4382 }
4383
4384 // try to honor the signal mask
4385 sigset_t oset;
4386 sigemptyset(&oset);
4387 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4388
4389 // call into the chained handler
4390 if (siginfo_flag_set) {
4391 (*sa)(sig, siginfo, context);
4392 } else {
4393 (*hand)(sig);
4394 }
4395
4396 // restore the signal mask
4397 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4398 }
4399 // Tell jvm's signal handler the signal is taken care of.
4400 return true;
4401 }
4402
4403 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4404 bool chained = false;
4405 // signal-chaining
4406 if (UseSignalChaining) {
4407 struct sigaction *actp = get_chained_signal_action(sig);
4408 if (actp != NULL) {
4409 chained = call_chained_handler(actp, sig, siginfo, context);
4410 }
4411 }
4412 return chained;
4413 }
4414
4415 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4416 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4417 return &sigact[sig];
4418 }
4419 return NULL;
4420 }
4421
4422 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4423 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4424 sigact[sig] = oldAct;
4425 sigs |= (uint64_t)1 << (sig-1);
4426 }
4427
4428 // for diagnostic
4429 int sigflags[NSIG];
4430
4431 int os::Linux::get_our_sigflags(int sig) {
4432 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4433 return sigflags[sig];
4434 }
4435
4436 void os::Linux::set_our_sigflags(int sig, int flags) {
4437 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4438 if (sig > 0 && sig < NSIG) {
4439 sigflags[sig] = flags;
4440 }
4441 }
4442
4443 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4444 // Check for overwrite.
4445 struct sigaction oldAct;
4446 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4447
4448 void* oldhand = oldAct.sa_sigaction
4449 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4450 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4451 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4452 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4453 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4454 if (AllowUserSignalHandlers || !set_installed) {
4455 // Do not overwrite; user takes responsibility to forward to us.
4456 return;
4457 } else if (UseSignalChaining) {
4458 // save the old handler in jvm
4459 save_preinstalled_handler(sig, oldAct);
4460 // libjsig also interposes the sigaction() call below and saves the
4461 // old sigaction on it own.
4462 } else {
4463 fatal("Encountered unexpected pre-existing sigaction handler "
4464 "%#lx for signal %d.", (long)oldhand, sig);
4465 }
4466 }
4467
4468 struct sigaction sigAct;
4469 sigfillset(&(sigAct.sa_mask));
4470 sigAct.sa_handler = SIG_DFL;
4471 if (!set_installed) {
4472 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4473 } else {
4474 sigAct.sa_sigaction = signalHandler;
4475 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4476 }
4477 // Save flags, which are set by ours
4478 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4479 sigflags[sig] = sigAct.sa_flags;
4480
4481 int ret = sigaction(sig, &sigAct, &oldAct);
4482 assert(ret == 0, "check");
4483
4484 void* oldhand2 = oldAct.sa_sigaction
4485 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4486 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4487 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4488 }
4489
4490 // install signal handlers for signals that HotSpot needs to
4491 // handle in order to support Java-level exception handling.
4492
4493 void os::Linux::install_signal_handlers() {
4494 if (!signal_handlers_are_installed) {
4495 signal_handlers_are_installed = true;
4496
4497 // signal-chaining
4498 typedef void (*signal_setting_t)();
4499 signal_setting_t begin_signal_setting = NULL;
4500 signal_setting_t end_signal_setting = NULL;
4501 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4502 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4503 if (begin_signal_setting != NULL) {
4504 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4505 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4506 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4507 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4508 libjsig_is_loaded = true;
4509 assert(UseSignalChaining, "should enable signal-chaining");
4510 }
4511 if (libjsig_is_loaded) {
4512 // Tell libjsig jvm is setting signal handlers
4513 (*begin_signal_setting)();
4514 }
4515
4516 set_signal_handler(SIGSEGV, true);
4517 set_signal_handler(SIGPIPE, true);
4518 set_signal_handler(SIGBUS, true);
4519 set_signal_handler(SIGILL, true);
4520 set_signal_handler(SIGFPE, true);
4521 #if defined(PPC64)
4522 set_signal_handler(SIGTRAP, true);
4523 #endif
4524 set_signal_handler(SIGXFSZ, true);
4525
4526 if (libjsig_is_loaded) {
4527 // Tell libjsig jvm finishes setting signal handlers
4528 (*end_signal_setting)();
4529 }
4530
4531 // We don't activate signal checker if libjsig is in place, we trust ourselves
4532 // and if UserSignalHandler is installed all bets are off.
4533 // Log that signal checking is off only if -verbose:jni is specified.
4534 if (CheckJNICalls) {
4535 if (libjsig_is_loaded) {
4536 if (PrintJNIResolving) {
4537 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4538 }
4539 check_signals = false;
4540 }
4541 if (AllowUserSignalHandlers) {
4542 if (PrintJNIResolving) {
4543 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4544 }
4545 check_signals = false;
4546 }
4547 }
4548 }
4549 }
4550
4551 // This is the fastest way to get thread cpu time on Linux.
4552 // Returns cpu time (user+sys) for any thread, not only for current.
4553 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4554 // It might work on 2.6.10+ with a special kernel/glibc patch.
4555 // For reference, please, see IEEE Std 1003.1-2004:
4556 // http://www.unix.org/single_unix_specification
4557
4558 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4559 struct timespec tp;
4560 int rc = os::Linux::clock_gettime(clockid, &tp);
4561 assert(rc == 0, "clock_gettime is expected to return 0 code");
4562
4563 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4564 }
4565
4566 void os::Linux::initialize_os_info() {
4567 assert(_os_version == 0, "OS info already initialized");
4568
4569 struct utsname _uname;
4570
4571 uint32_t major;
4572 uint32_t minor;
4573 uint32_t fix;
4574
4575 int rc;
4576
4577 // Kernel version is unknown if
4578 // verification below fails.
4579 _os_version = 0x01000000;
4580
4581 rc = uname(&_uname);
4582 if (rc != -1) {
4583
4584 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4585 if (rc == 3) {
4586
4587 if (major < 256 && minor < 256 && fix < 256) {
4588 // Kernel version format is as expected,
4589 // set it overriding unknown state.
4590 _os_version = (major << 16) |
4591 (minor << 8 ) |
4592 (fix << 0 ) ;
4593 }
4594 }
4595 }
4596 }
4597
4598 uint32_t os::Linux::os_version() {
4599 assert(_os_version != 0, "not initialized");
4600 return _os_version & 0x00FFFFFF;
4601 }
4602
4603 bool os::Linux::os_version_is_known() {
4604 assert(_os_version != 0, "not initialized");
4605 return _os_version & 0x01000000 ? false : true;
4606 }
4607
4608 /////
4609 // glibc on Linux platform uses non-documented flag
4610 // to indicate, that some special sort of signal
4611 // trampoline is used.
4612 // We will never set this flag, and we should
4613 // ignore this flag in our diagnostic
4614 #ifdef SIGNIFICANT_SIGNAL_MASK
4615 #undef SIGNIFICANT_SIGNAL_MASK
4616 #endif
4617 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4618
4619 static const char* get_signal_handler_name(address handler,
4620 char* buf, int buflen) {
4621 int offset = 0;
4622 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4623 if (found) {
4624 // skip directory names
4625 const char *p1, *p2;
4626 p1 = buf;
4627 size_t len = strlen(os::file_separator());
4628 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4629 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4630 } else {
4631 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4632 }
4633 return buf;
4634 }
4635
4636 static void print_signal_handler(outputStream* st, int sig,
4637 char* buf, size_t buflen) {
4638 struct sigaction sa;
4639
4640 sigaction(sig, NULL, &sa);
4641
4642 // See comment for SIGNIFICANT_SIGNAL_MASK define
4643 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4644
4645 st->print("%s: ", os::exception_name(sig, buf, buflen));
4646
4647 address handler = (sa.sa_flags & SA_SIGINFO)
4648 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4649 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4650
4651 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4652 st->print("SIG_DFL");
4653 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4654 st->print("SIG_IGN");
4655 } else {
4656 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4657 }
4658
4659 st->print(", sa_mask[0]=");
4660 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4661
4662 address rh = VMError::get_resetted_sighandler(sig);
4663 // May be, handler was resetted by VMError?
4664 if (rh != NULL) {
4665 handler = rh;
4666 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4667 }
4668
4669 st->print(", sa_flags=");
4670 os::Posix::print_sa_flags(st, sa.sa_flags);
4671
4672 // Check: is it our handler?
4673 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4674 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4675 // It is our signal handler
4676 // check for flags, reset system-used one!
4677 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4678 st->print(
4679 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4680 os::Linux::get_our_sigflags(sig));
4681 }
4682 }
4683 st->cr();
4684 }
4685
4686
4687 #define DO_SIGNAL_CHECK(sig) \
4688 do { \
4689 if (!sigismember(&check_signal_done, sig)) { \
4690 os::Linux::check_signal_handler(sig); \
4691 } \
4692 } while (0)
4693
4694 // This method is a periodic task to check for misbehaving JNI applications
4695 // under CheckJNI, we can add any periodic checks here
4696
4697 void os::run_periodic_checks() {
4698 if (check_signals == false) return;
4699
4700 // SEGV and BUS if overridden could potentially prevent
4701 // generation of hs*.log in the event of a crash, debugging
4702 // such a case can be very challenging, so we absolutely
4703 // check the following for a good measure:
4704 DO_SIGNAL_CHECK(SIGSEGV);
4705 DO_SIGNAL_CHECK(SIGILL);
4706 DO_SIGNAL_CHECK(SIGFPE);
4707 DO_SIGNAL_CHECK(SIGBUS);
4708 DO_SIGNAL_CHECK(SIGPIPE);
4709 DO_SIGNAL_CHECK(SIGXFSZ);
4710 #if defined(PPC64)
4711 DO_SIGNAL_CHECK(SIGTRAP);
4712 #endif
4713
4714 // ReduceSignalUsage allows the user to override these handlers
4715 // see comments at the very top and jvm_solaris.h
4716 if (!ReduceSignalUsage) {
4717 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4718 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4719 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4720 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4721 }
4722
4723 DO_SIGNAL_CHECK(SR_signum);
4724 }
4725
4726 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4727
4728 static os_sigaction_t os_sigaction = NULL;
4729
4730 void os::Linux::check_signal_handler(int sig) {
4731 char buf[O_BUFLEN];
4732 address jvmHandler = NULL;
4733
4734
4735 struct sigaction act;
4736 if (os_sigaction == NULL) {
4737 // only trust the default sigaction, in case it has been interposed
4738 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4739 if (os_sigaction == NULL) return;
4740 }
4741
4742 os_sigaction(sig, (struct sigaction*)NULL, &act);
4743
4744
4745 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4746
4747 address thisHandler = (act.sa_flags & SA_SIGINFO)
4748 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4749 : CAST_FROM_FN_PTR(address, act.sa_handler);
4750
4751
4752 switch (sig) {
4753 case SIGSEGV:
4754 case SIGBUS:
4755 case SIGFPE:
4756 case SIGPIPE:
4757 case SIGILL:
4758 case SIGXFSZ:
4759 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4760 break;
4761
4762 case SHUTDOWN1_SIGNAL:
4763 case SHUTDOWN2_SIGNAL:
4764 case SHUTDOWN3_SIGNAL:
4765 case BREAK_SIGNAL:
4766 jvmHandler = (address)user_handler();
4767 break;
4768
4769 default:
4770 if (sig == SR_signum) {
4771 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4772 } else {
4773 return;
4774 }
4775 break;
4776 }
4777
4778 if (thisHandler != jvmHandler) {
4779 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4780 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4781 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4782 // No need to check this sig any longer
4783 sigaddset(&check_signal_done, sig);
4784 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4785 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4786 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4787 exception_name(sig, buf, O_BUFLEN));
4788 }
4789 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4790 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4791 tty->print("expected:");
4792 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4793 tty->cr();
4794 tty->print(" found:");
4795 os::Posix::print_sa_flags(tty, act.sa_flags);
4796 tty->cr();
4797 // No need to check this sig any longer
4798 sigaddset(&check_signal_done, sig);
4799 }
4800
4801 // Dump all the signal
4802 if (sigismember(&check_signal_done, sig)) {
4803 print_signal_handlers(tty, buf, O_BUFLEN);
4804 }
4805 }
4806
4807 extern void report_error(char* file_name, int line_no, char* title,
4808 char* format, ...);
4809
4810 // this is called _before_ the most of global arguments have been parsed
4811 void os::init(void) {
4812 char dummy; // used to get a guess on initial stack address
4813 // first_hrtime = gethrtime();
4814
4815 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4816
4817 init_random(1234567);
4818
4819 ThreadCritical::initialize();
4820
4821 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4822 if (Linux::page_size() == -1) {
4823 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4824 os::strerror(errno));
4825 }
4826 init_page_sizes((size_t) Linux::page_size());
4827
4828 Linux::initialize_system_info();
4829
4830 Linux::initialize_os_info();
4831
4832 // main_thread points to the aboriginal thread
4833 Linux::_main_thread = pthread_self();
4834
4835 Linux::clock_init();
4836 initial_time_count = javaTimeNanos();
4837
4838 // pthread_condattr initialization for monotonic clock
4839 int status;
4840 pthread_condattr_t* _condattr = os::Linux::condAttr();
4841 if ((status = pthread_condattr_init(_condattr)) != 0) {
4842 fatal("pthread_condattr_init: %s", os::strerror(status));
4843 }
4844 // Only set the clock if CLOCK_MONOTONIC is available
4845 if (os::supports_monotonic_clock()) {
4846 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4847 if (status == EINVAL) {
4848 warning("Unable to use monotonic clock with relative timed-waits" \
4849 " - changes to the time-of-day clock may have adverse affects");
4850 } else {
4851 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4852 }
4853 }
4854 }
4855 // else it defaults to CLOCK_REALTIME
4856
4857 // retrieve entry point for pthread_setname_np
4858 Linux::_pthread_setname_np =
4859 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4860
4861 }
4862
4863 // To install functions for atexit system call
4864 extern "C" {
4865 static void perfMemory_exit_helper() {
4866 perfMemory_exit();
4867 }
4868 }
4869
4870 // this is called _after_ the global arguments have been parsed
4871 jint os::init_2(void) {
4872 Linux::fast_thread_clock_init();
4873
4874 // Allocate a single page and mark it as readable for safepoint polling
4875 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4876 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4877
4878 os::set_polling_page(polling_page);
4879 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4880
4881 if (!UseMembar) {
4882 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4883 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4884 os::set_memory_serialize_page(mem_serialize_page);
4885 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4886 }
4887
4888 // initialize suspend/resume support - must do this before signal_sets_init()
4889 if (SR_initialize() != 0) {
4890 perror("SR_initialize failed");
4891 return JNI_ERR;
4892 }
4893
4894 Linux::signal_sets_init();
4895 Linux::install_signal_handlers();
4896
4897 // Check and sets minimum stack sizes against command line options
4898 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4899 return JNI_ERR;
4900 }
4901 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4902
4903 #if defined(IA32)
4904 workaround_expand_exec_shield_cs_limit();
4905 #endif
4906
4907 Linux::libpthread_init();
4908 log_info(os)("HotSpot is running with %s, %s",
4909 Linux::glibc_version(), Linux::libpthread_version());
4910
4911 if (UseNUMA) {
4912 if (!Linux::libnuma_init()) {
4913 UseNUMA = false;
4914 } else {
4915 if ((Linux::numa_max_node() < 1)) {
4916 // There's only one node(they start from 0), disable NUMA.
4917 UseNUMA = false;
4918 }
4919 }
4920 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4921 // we can make the adaptive lgrp chunk resizing work. If the user specified
4922 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4923 // disable adaptive resizing.
4924 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4925 if (FLAG_IS_DEFAULT(UseNUMA)) {
4926 UseNUMA = false;
4927 } else {
4928 if (FLAG_IS_DEFAULT(UseLargePages) &&
4929 FLAG_IS_DEFAULT(UseSHM) &&
4930 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4931 UseLargePages = false;
4932 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4933 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4934 UseAdaptiveSizePolicy = false;
4935 UseAdaptiveNUMAChunkSizing = false;
4936 }
4937 }
4938 }
4939 if (!UseNUMA && ForceNUMA) {
4940 UseNUMA = true;
4941 }
4942 }
4943
4944 if (MaxFDLimit) {
4945 // set the number of file descriptors to max. print out error
4946 // if getrlimit/setrlimit fails but continue regardless.
4947 struct rlimit nbr_files;
4948 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4949 if (status != 0) {
4950 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4951 } else {
4952 nbr_files.rlim_cur = nbr_files.rlim_max;
4953 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4954 if (status != 0) {
4955 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4956 }
4957 }
4958 }
4959
4960 // Initialize lock used to serialize thread creation (see os::create_thread)
4961 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4962
4963 // at-exit methods are called in the reverse order of their registration.
4964 // atexit functions are called on return from main or as a result of a
4965 // call to exit(3C). There can be only 32 of these functions registered
4966 // and atexit() does not set errno.
4967
4968 if (PerfAllowAtExitRegistration) {
4969 // only register atexit functions if PerfAllowAtExitRegistration is set.
4970 // atexit functions can be delayed until process exit time, which
4971 // can be problematic for embedded VM situations. Embedded VMs should
4972 // call DestroyJavaVM() to assure that VM resources are released.
4973
4974 // note: perfMemory_exit_helper atexit function may be removed in
4975 // the future if the appropriate cleanup code can be added to the
4976 // VM_Exit VMOperation's doit method.
4977 if (atexit(perfMemory_exit_helper) != 0) {
4978 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4979 }
4980 }
4981
4982 // initialize thread priority policy
4983 prio_init();
4984
4985 return JNI_OK;
4986 }
4987
4988 // Mark the polling page as unreadable
4989 void os::make_polling_page_unreadable(void) {
4990 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4991 fatal("Could not disable polling page");
4992 }
4993 }
4994
4995 // Mark the polling page as readable
4996 void os::make_polling_page_readable(void) {
4997 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4998 fatal("Could not enable polling page");
4999 }
5000 }
5001
5002 // older glibc versions don't have this macro (which expands to
5003 // an optimized bit-counting function) so we have to roll our own
5004 #ifndef CPU_COUNT
5005
5006 static int _cpu_count(const cpu_set_t* cpus) {
5007 int count = 0;
5008 // only look up to the number of configured processors
5009 for (int i = 0; i < os::processor_count(); i++) {
5010 if (CPU_ISSET(i, cpus)) {
5011 count++;
5012 }
5013 }
5014 return count;
5015 }
5016
5017 #define CPU_COUNT(cpus) _cpu_count(cpus)
5018
5019 #endif // CPU_COUNT
5020
5021 // Get the current number of available processors for this process.
5022 // This value can change at any time during a process's lifetime.
5023 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5024 // If it appears there may be more than 1024 processors then we do a
5025 // dynamic check - see 6515172 for details.
5026 // If anything goes wrong we fallback to returning the number of online
5027 // processors - which can be greater than the number available to the process.
5028 int os::active_processor_count() {
5029 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5030 cpu_set_t* cpus_p = &cpus;
5031 int cpus_size = sizeof(cpu_set_t);
5032
5033 int configured_cpus = processor_count(); // upper bound on available cpus
5034 int cpu_count = 0;
5035
5036 // old build platforms may not support dynamic cpu sets
5037 #ifdef CPU_ALLOC
5038
5039 // To enable easy testing of the dynamic path on different platforms we
5040 // introduce a diagnostic flag: UseCpuAllocPath
5041 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5042 // kernel may use a mask bigger than cpu_set_t
5043 log_trace(os)("active_processor_count: using dynamic path %s"
5044 "- configured processors: %d",
5045 UseCpuAllocPath ? "(forced) " : "",
5046 configured_cpus);
5047 cpus_p = CPU_ALLOC(configured_cpus);
5048 if (cpus_p != NULL) {
5049 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5050 // zero it just to be safe
5051 CPU_ZERO_S(cpus_size, cpus_p);
5052 }
5053 else {
5054 // failed to allocate so fallback to online cpus
5055 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5056 log_trace(os)("active_processor_count: "
5057 "CPU_ALLOC failed (%s) - using "
5058 "online processor count: %d",
5059 os::strerror(errno), online_cpus);
5060 return online_cpus;
5061 }
5062 }
5063 else {
5064 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5065 configured_cpus);
5066 }
5067 #else // CPU_ALLOC
5068 // these stubs won't be executed
5069 #define CPU_COUNT_S(size, cpus) -1
5070 #define CPU_FREE(cpus)
5071
5072 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5073 configured_cpus);
5074 #endif // CPU_ALLOC
5075
5076 // pid 0 means the current thread - which we have to assume represents the process
5077 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5078 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5079 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5080 }
5081 else {
5082 cpu_count = CPU_COUNT(cpus_p);
5083 }
5084 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5085 }
5086 else {
5087 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5088 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5089 "which may exceed available processors", os::strerror(errno), cpu_count);
5090 }
5091
5092 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5093 CPU_FREE(cpus_p);
5094 }
5095
5096 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5097 return cpu_count;
5098 }
5099
5100 void os::set_native_thread_name(const char *name) {
5101 if (Linux::_pthread_setname_np) {
5102 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5103 snprintf(buf, sizeof(buf), "%s", name);
5104 buf[sizeof(buf) - 1] = '\0';
5105 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5106 // ERANGE should not happen; all other errors should just be ignored.
5107 assert(rc != ERANGE, "pthread_setname_np failed");
5108 }
5109 }
5110
5111 bool os::distribute_processes(uint length, uint* distribution) {
5112 // Not yet implemented.
5113 return false;
5114 }
5115
5116 bool os::bind_to_processor(uint processor_id) {
5117 // Not yet implemented.
5118 return false;
5119 }
5120
5121 ///
5122
5123 void os::SuspendedThreadTask::internal_do_task() {
5124 if (do_suspend(_thread->osthread())) {
5125 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5126 do_task(context);
5127 do_resume(_thread->osthread());
5128 }
5129 }
5130
5131 class PcFetcher : public os::SuspendedThreadTask {
5132 public:
5133 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5134 ExtendedPC result();
5135 protected:
5136 void do_task(const os::SuspendedThreadTaskContext& context);
5137 private:
5138 ExtendedPC _epc;
5139 };
5140
5141 ExtendedPC PcFetcher::result() {
5142 guarantee(is_done(), "task is not done yet.");
5143 return _epc;
5144 }
5145
5146 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5147 Thread* thread = context.thread();
5148 OSThread* osthread = thread->osthread();
5149 if (osthread->ucontext() != NULL) {
5150 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5151 } else {
5152 // NULL context is unexpected, double-check this is the VMThread
5153 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5154 }
5155 }
5156
5157 // Suspends the target using the signal mechanism and then grabs the PC before
5158 // resuming the target. Used by the flat-profiler only
5159 ExtendedPC os::get_thread_pc(Thread* thread) {
5160 // Make sure that it is called by the watcher for the VMThread
5161 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5162 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5163
5164 PcFetcher fetcher(thread);
5165 fetcher.run();
5166 return fetcher.result();
5167 }
5168
5169 ////////////////////////////////////////////////////////////////////////////////
5170 // debug support
5171
5172 bool os::find(address addr, outputStream* st) {
5173 Dl_info dlinfo;
5174 memset(&dlinfo, 0, sizeof(dlinfo));
5175 if (dladdr(addr, &dlinfo) != 0) {
5176 st->print(PTR_FORMAT ": ", p2i(addr));
5177 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5178 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5179 p2i(addr) - p2i(dlinfo.dli_saddr));
5180 } else if (dlinfo.dli_fbase != NULL) {
5181 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5182 } else {
5183 st->print("<absolute address>");
5184 }
5185 if (dlinfo.dli_fname != NULL) {
5186 st->print(" in %s", dlinfo.dli_fname);
5187 }
5188 if (dlinfo.dli_fbase != NULL) {
5189 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5190 }
5191 st->cr();
5192
5193 if (Verbose) {
5194 // decode some bytes around the PC
5195 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5196 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5197 address lowest = (address) dlinfo.dli_sname;
5198 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5199 if (begin < lowest) begin = lowest;
5200 Dl_info dlinfo2;
5201 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5202 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5203 end = (address) dlinfo2.dli_saddr;
5204 }
5205 Disassembler::decode(begin, end, st);
5206 }
5207 return true;
5208 }
5209 return false;
5210 }
5211
5212 ////////////////////////////////////////////////////////////////////////////////
5213 // misc
5214
5215 // This does not do anything on Linux. This is basically a hook for being
5216 // able to use structured exception handling (thread-local exception filters)
5217 // on, e.g., Win32.
5218 void
5219 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5220 JavaCallArguments* args, Thread* thread) {
5221 f(value, method, args, thread);
5222 }
5223
5224 void os::print_statistics() {
5225 }
5226
5227 bool os::message_box(const char* title, const char* message) {
5228 int i;
5229 fdStream err(defaultStream::error_fd());
5230 for (i = 0; i < 78; i++) err.print_raw("=");
5231 err.cr();
5232 err.print_raw_cr(title);
5233 for (i = 0; i < 78; i++) err.print_raw("-");
5234 err.cr();
5235 err.print_raw_cr(message);
5236 for (i = 0; i < 78; i++) err.print_raw("=");
5237 err.cr();
5238
5239 char buf[16];
5240 // Prevent process from exiting upon "read error" without consuming all CPU
5241 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5242
5243 return buf[0] == 'y' || buf[0] == 'Y';
5244 }
5245
5246 int os::stat(const char *path, struct stat *sbuf) {
5247 char pathbuf[MAX_PATH];
5248 if (strlen(path) > MAX_PATH - 1) {
5249 errno = ENAMETOOLONG;
5250 return -1;
5251 }
5252 os::native_path(strcpy(pathbuf, path));
5253 return ::stat(pathbuf, sbuf);
5254 }
5255
5256 // Is a (classpath) directory empty?
5257 bool os::dir_is_empty(const char* path) {
5258 DIR *dir = NULL;
5259 struct dirent *ptr;
5260
5261 dir = opendir(path);
5262 if (dir == NULL) return true;
5263
5264 // Scan the directory
5265 bool result = true;
5266 char buf[sizeof(struct dirent) + MAX_PATH];
5267 while (result && (ptr = ::readdir(dir)) != NULL) {
5268 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5269 result = false;
5270 }
5271 }
5272 closedir(dir);
5273 return result;
5274 }
5275
5276 // This code originates from JDK's sysOpen and open64_w
5277 // from src/solaris/hpi/src/system_md.c
5278
5279 int os::open(const char *path, int oflag, int mode) {
5280 if (strlen(path) > MAX_PATH - 1) {
5281 errno = ENAMETOOLONG;
5282 return -1;
5283 }
5284
5285 // All file descriptors that are opened in the Java process and not
5286 // specifically destined for a subprocess should have the close-on-exec
5287 // flag set. If we don't set it, then careless 3rd party native code
5288 // might fork and exec without closing all appropriate file descriptors
5289 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5290 // turn might:
5291 //
5292 // - cause end-of-file to fail to be detected on some file
5293 // descriptors, resulting in mysterious hangs, or
5294 //
5295 // - might cause an fopen in the subprocess to fail on a system
5296 // suffering from bug 1085341.
5297 //
5298 // (Yes, the default setting of the close-on-exec flag is a Unix
5299 // design flaw)
5300 //
5301 // See:
5302 // 1085341: 32-bit stdio routines should support file descriptors >255
5303 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5304 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5305 //
5306 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5307 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5308 // because it saves a system call and removes a small window where the flag
5309 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5310 // and we fall back to using FD_CLOEXEC (see below).
5311 #ifdef O_CLOEXEC
5312 oflag |= O_CLOEXEC;
5313 #endif
5314
5315 int fd = ::open64(path, oflag, mode);
5316 if (fd == -1) return -1;
5317
5318 //If the open succeeded, the file might still be a directory
5319 {
5320 struct stat64 buf64;
5321 int ret = ::fstat64(fd, &buf64);
5322 int st_mode = buf64.st_mode;
5323
5324 if (ret != -1) {
5325 if ((st_mode & S_IFMT) == S_IFDIR) {
5326 errno = EISDIR;
5327 ::close(fd);
5328 return -1;
5329 }
5330 } else {
5331 ::close(fd);
5332 return -1;
5333 }
5334 }
5335
5336 #ifdef FD_CLOEXEC
5337 // Validate that the use of the O_CLOEXEC flag on open above worked.
5338 // With recent kernels, we will perform this check exactly once.
5339 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5340 if (!O_CLOEXEC_is_known_to_work) {
5341 int flags = ::fcntl(fd, F_GETFD);
5342 if (flags != -1) {
5343 if ((flags & FD_CLOEXEC) != 0)
5344 O_CLOEXEC_is_known_to_work = 1;
5345 else
5346 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5347 }
5348 }
5349 #endif
5350
5351 return fd;
5352 }
5353
5354
5355 // create binary file, rewriting existing file if required
5356 int os::create_binary_file(const char* path, bool rewrite_existing) {
5357 int oflags = O_WRONLY | O_CREAT;
5358 if (!rewrite_existing) {
5359 oflags |= O_EXCL;
5360 }
5361 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5362 }
5363
5364 // return current position of file pointer
5365 jlong os::current_file_offset(int fd) {
5366 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5367 }
5368
5369 // move file pointer to the specified offset
5370 jlong os::seek_to_file_offset(int fd, jlong offset) {
5371 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5372 }
5373
5374 // This code originates from JDK's sysAvailable
5375 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5376
5377 int os::available(int fd, jlong *bytes) {
5378 jlong cur, end;
5379 int mode;
5380 struct stat64 buf64;
5381
5382 if (::fstat64(fd, &buf64) >= 0) {
5383 mode = buf64.st_mode;
5384 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5385 int n;
5386 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5387 *bytes = n;
5388 return 1;
5389 }
5390 }
5391 }
5392 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5393 return 0;
5394 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5395 return 0;
5396 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5397 return 0;
5398 }
5399 *bytes = end - cur;
5400 return 1;
5401 }
5402
5403 // Map a block of memory.
5404 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5405 char *addr, size_t bytes, bool read_only,
5406 bool allow_exec) {
5407 int prot;
5408 int flags = MAP_PRIVATE;
5409
5410 if (read_only) {
5411 prot = PROT_READ;
5412 } else {
5413 prot = PROT_READ | PROT_WRITE;
5414 }
5415
5416 if (allow_exec) {
5417 prot |= PROT_EXEC;
5418 }
5419
5420 if (addr != NULL) {
5421 flags |= MAP_FIXED;
5422 }
5423
5424 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5425 fd, file_offset);
5426 if (mapped_address == MAP_FAILED) {
5427 return NULL;
5428 }
5429 return mapped_address;
5430 }
5431
5432
5433 // Remap a block of memory.
5434 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5435 char *addr, size_t bytes, bool read_only,
5436 bool allow_exec) {
5437 // same as map_memory() on this OS
5438 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5439 allow_exec);
5440 }
5441
5442
5443 // Unmap a block of memory.
5444 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5445 return munmap(addr, bytes) == 0;
5446 }
5447
5448 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5449
5450 static clockid_t thread_cpu_clockid(Thread* thread) {
5451 pthread_t tid = thread->osthread()->pthread_id();
5452 clockid_t clockid;
5453
5454 // Get thread clockid
5455 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5456 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5457 return clockid;
5458 }
5459
5460 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5461 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5462 // of a thread.
5463 //
5464 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5465 // the fast estimate available on the platform.
5466
5467 jlong os::current_thread_cpu_time() {
5468 if (os::Linux::supports_fast_thread_cpu_time()) {
5469 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5470 } else {
5471 // return user + sys since the cost is the same
5472 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5473 }
5474 }
5475
5476 jlong os::thread_cpu_time(Thread* thread) {
5477 // consistent with what current_thread_cpu_time() returns
5478 if (os::Linux::supports_fast_thread_cpu_time()) {
5479 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5480 } else {
5481 return slow_thread_cpu_time(thread, true /* user + sys */);
5482 }
5483 }
5484
5485 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5486 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5487 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5488 } else {
5489 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5490 }
5491 }
5492
5493 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5494 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5495 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5496 } else {
5497 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5498 }
5499 }
5500
5501 // -1 on error.
5502 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5503 pid_t tid = thread->osthread()->thread_id();
5504 char *s;
5505 char stat[2048];
5506 int statlen;
5507 char proc_name[64];
5508 int count;
5509 long sys_time, user_time;
5510 char cdummy;
5511 int idummy;
5512 long ldummy;
5513 FILE *fp;
5514
5515 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5516 fp = fopen(proc_name, "r");
5517 if (fp == NULL) return -1;
5518 statlen = fread(stat, 1, 2047, fp);
5519 stat[statlen] = '\0';
5520 fclose(fp);
5521
5522 // Skip pid and the command string. Note that we could be dealing with
5523 // weird command names, e.g. user could decide to rename java launcher
5524 // to "java 1.4.2 :)", then the stat file would look like
5525 // 1234 (java 1.4.2 :)) R ... ...
5526 // We don't really need to know the command string, just find the last
5527 // occurrence of ")" and then start parsing from there. See bug 4726580.
5528 s = strrchr(stat, ')');
5529 if (s == NULL) return -1;
5530
5531 // Skip blank chars
5532 do { s++; } while (s && isspace(*s));
5533
5534 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5535 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5536 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5537 &user_time, &sys_time);
5538 if (count != 13) return -1;
5539 if (user_sys_cpu_time) {
5540 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5541 } else {
5542 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5543 }
5544 }
5545
5546 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5547 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5548 info_ptr->may_skip_backward = false; // elapsed time not wall time
5549 info_ptr->may_skip_forward = false; // elapsed time not wall time
5550 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5551 }
5552
5553 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5554 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5555 info_ptr->may_skip_backward = false; // elapsed time not wall time
5556 info_ptr->may_skip_forward = false; // elapsed time not wall time
5557 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5558 }
5559
5560 bool os::is_thread_cpu_time_supported() {
5561 return true;
5562 }
5563
5564 // System loadavg support. Returns -1 if load average cannot be obtained.
5565 // Linux doesn't yet have a (official) notion of processor sets,
5566 // so just return the system wide load average.
5567 int os::loadavg(double loadavg[], int nelem) {
5568 return ::getloadavg(loadavg, nelem);
5569 }
5570
5571 void os::pause() {
5572 char filename[MAX_PATH];
5573 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5574 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5575 } else {
5576 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5577 }
5578
5579 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5580 if (fd != -1) {
5581 struct stat buf;
5582 ::close(fd);
5583 while (::stat(filename, &buf) == 0) {
5584 (void)::poll(NULL, 0, 100);
5585 }
5586 } else {
5587 jio_fprintf(stderr,
5588 "Could not open pause file '%s', continuing immediately.\n", filename);
5589 }
5590 }
5591
5592
5593 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5594 // comment paragraphs are worth repeating here:
5595 //
5596 // Assumption:
5597 // Only one parker can exist on an event, which is why we allocate
5598 // them per-thread. Multiple unparkers can coexist.
5599 //
5600 // _Event serves as a restricted-range semaphore.
5601 // -1 : thread is blocked, i.e. there is a waiter
5602 // 0 : neutral: thread is running or ready,
5603 // could have been signaled after a wait started
5604 // 1 : signaled - thread is running or ready
5605 //
5606
5607 // utility to compute the abstime argument to timedwait:
5608 // millis is the relative timeout time
5609 // abstime will be the absolute timeout time
5610 // TODO: replace compute_abstime() with unpackTime()
5611
5612 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5613 if (millis < 0) millis = 0;
5614
5615 jlong seconds = millis / 1000;
5616 millis %= 1000;
5617 if (seconds > 50000000) { // see man cond_timedwait(3T)
5618 seconds = 50000000;
5619 }
5620
5621 if (os::supports_monotonic_clock()) {
5622 struct timespec now;
5623 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5624 assert_status(status == 0, status, "clock_gettime");
5625 abstime->tv_sec = now.tv_sec + seconds;
5626 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5627 if (nanos >= NANOSECS_PER_SEC) {
5628 abstime->tv_sec += 1;
5629 nanos -= NANOSECS_PER_SEC;
5630 }
5631 abstime->tv_nsec = nanos;
5632 } else {
5633 struct timeval now;
5634 int status = gettimeofday(&now, NULL);
5635 assert(status == 0, "gettimeofday");
5636 abstime->tv_sec = now.tv_sec + seconds;
5637 long usec = now.tv_usec + millis * 1000;
5638 if (usec >= 1000000) {
5639 abstime->tv_sec += 1;
5640 usec -= 1000000;
5641 }
5642 abstime->tv_nsec = usec * 1000;
5643 }
5644 return abstime;
5645 }
5646
5647 void os::PlatformEvent::park() { // AKA "down()"
5648 // Transitions for _Event:
5649 // -1 => -1 : illegal
5650 // 1 => 0 : pass - return immediately
5651 // 0 => -1 : block; then set _Event to 0 before returning
5652
5653 // Invariant: Only the thread associated with the Event/PlatformEvent
5654 // may call park().
5655 // TODO: assert that _Assoc != NULL or _Assoc == Self
5656 assert(_nParked == 0, "invariant");
5657
5658 int v;
5659 for (;;) {
5660 v = _Event;
5661 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5662 }
5663 guarantee(v >= 0, "invariant");
5664 if (v == 0) {
5665 // Do this the hard way by blocking ...
5666 int status = pthread_mutex_lock(_mutex);
5667 assert_status(status == 0, status, "mutex_lock");
5668 guarantee(_nParked == 0, "invariant");
5669 ++_nParked;
5670 while (_Event < 0) {
5671 status = pthread_cond_wait(_cond, _mutex);
5672 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5673 // Treat this the same as if the wait was interrupted
5674 if (status == ETIME) { status = EINTR; }
5675 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5676 }
5677 --_nParked;
5678
5679 _Event = 0;
5680 status = pthread_mutex_unlock(_mutex);
5681 assert_status(status == 0, status, "mutex_unlock");
5682 // Paranoia to ensure our locked and lock-free paths interact
5683 // correctly with each other.
5684 OrderAccess::fence();
5685 }
5686 guarantee(_Event >= 0, "invariant");
5687 }
5688
5689 int os::PlatformEvent::park(jlong millis) {
5690 // Transitions for _Event:
5691 // -1 => -1 : illegal
5692 // 1 => 0 : pass - return immediately
5693 // 0 => -1 : block; then set _Event to 0 before returning
5694
5695 guarantee(_nParked == 0, "invariant");
5696
5697 int v;
5698 for (;;) {
5699 v = _Event;
5700 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5701 }
5702 guarantee(v >= 0, "invariant");
5703 if (v != 0) return OS_OK;
5704
5705 // We do this the hard way, by blocking the thread.
5706 // Consider enforcing a minimum timeout value.
5707 struct timespec abst;
5708 compute_abstime(&abst, millis);
5709
5710 int ret = OS_TIMEOUT;
5711 int status = pthread_mutex_lock(_mutex);
5712 assert_status(status == 0, status, "mutex_lock");
5713 guarantee(_nParked == 0, "invariant");
5714 ++_nParked;
5715
5716 // Object.wait(timo) will return because of
5717 // (a) notification
5718 // (b) timeout
5719 // (c) thread.interrupt
5720 //
5721 // Thread.interrupt and object.notify{All} both call Event::set.
5722 // That is, we treat thread.interrupt as a special case of notification.
5723 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5724 // We assume all ETIME returns are valid.
5725 //
5726 // TODO: properly differentiate simultaneous notify+interrupt.
5727 // In that case, we should propagate the notify to another waiter.
5728
5729 while (_Event < 0) {
5730 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5731 assert_status(status == 0 || status == EINTR ||
5732 status == ETIME || status == ETIMEDOUT,
5733 status, "cond_timedwait");
5734 if (!FilterSpuriousWakeups) break; // previous semantics
5735 if (status == ETIME || status == ETIMEDOUT) break;
5736 // We consume and ignore EINTR and spurious wakeups.
5737 }
5738 --_nParked;
5739 if (_Event >= 0) {
5740 ret = OS_OK;
5741 }
5742 _Event = 0;
5743 status = pthread_mutex_unlock(_mutex);
5744 assert_status(status == 0, status, "mutex_unlock");
5745 assert(_nParked == 0, "invariant");
5746 // Paranoia to ensure our locked and lock-free paths interact
5747 // correctly with each other.
5748 OrderAccess::fence();
5749 return ret;
5750 }
5751
5752 void os::PlatformEvent::unpark() {
5753 // Transitions for _Event:
5754 // 0 => 1 : just return
5755 // 1 => 1 : just return
5756 // -1 => either 0 or 1; must signal target thread
5757 // That is, we can safely transition _Event from -1 to either
5758 // 0 or 1.
5759 // See also: "Semaphores in Plan 9" by Mullender & Cox
5760 //
5761 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5762 // that it will take two back-to-back park() calls for the owning
5763 // thread to block. This has the benefit of forcing a spurious return
5764 // from the first park() call after an unpark() call which will help
5765 // shake out uses of park() and unpark() without condition variables.
5766
5767 if (Atomic::xchg(1, &_Event) >= 0) return;
5768
5769 // Wait for the thread associated with the event to vacate
5770 int status = pthread_mutex_lock(_mutex);
5771 assert_status(status == 0, status, "mutex_lock");
5772 int AnyWaiters = _nParked;
5773 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5774 status = pthread_mutex_unlock(_mutex);
5775 assert_status(status == 0, status, "mutex_unlock");
5776 if (AnyWaiters != 0) {
5777 // Note that we signal() *after* dropping the lock for "immortal" Events.
5778 // This is safe and avoids a common class of futile wakeups. In rare
5779 // circumstances this can cause a thread to return prematurely from
5780 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5781 // will simply re-test the condition and re-park itself.
5782 // This provides particular benefit if the underlying platform does not
5783 // provide wait morphing.
5784 status = pthread_cond_signal(_cond);
5785 assert_status(status == 0, status, "cond_signal");
5786 }
5787 }
5788
5789
5790 // JSR166
5791 // -------------------------------------------------------
5792
5793 // The solaris and linux implementations of park/unpark are fairly
5794 // conservative for now, but can be improved. They currently use a
5795 // mutex/condvar pair, plus a a count.
5796 // Park decrements count if > 0, else does a condvar wait. Unpark
5797 // sets count to 1 and signals condvar. Only one thread ever waits
5798 // on the condvar. Contention seen when trying to park implies that someone
5799 // is unparking you, so don't wait. And spurious returns are fine, so there
5800 // is no need to track notifications.
5801
5802 // This code is common to linux and solaris and will be moved to a
5803 // common place in dolphin.
5804 //
5805 // The passed in time value is either a relative time in nanoseconds
5806 // or an absolute time in milliseconds. Either way it has to be unpacked
5807 // into suitable seconds and nanoseconds components and stored in the
5808 // given timespec structure.
5809 // Given time is a 64-bit value and the time_t used in the timespec is only
5810 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5811 // overflow if times way in the future are given. Further on Solaris versions
5812 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5813 // number of seconds, in abstime, is less than current_time + 100,000,000.
5814 // As it will be 28 years before "now + 100000000" will overflow we can
5815 // ignore overflow and just impose a hard-limit on seconds using the value
5816 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5817 // years from "now".
5818
5819 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5820 assert(time > 0, "convertTime");
5821 time_t max_secs = 0;
5822
5823 if (!os::supports_monotonic_clock() || isAbsolute) {
5824 struct timeval now;
5825 int status = gettimeofday(&now, NULL);
5826 assert(status == 0, "gettimeofday");
5827
5828 max_secs = now.tv_sec + MAX_SECS;
5829
5830 if (isAbsolute) {
5831 jlong secs = time / 1000;
5832 if (secs > max_secs) {
5833 absTime->tv_sec = max_secs;
5834 } else {
5835 absTime->tv_sec = secs;
5836 }
5837 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5838 } else {
5839 jlong secs = time / NANOSECS_PER_SEC;
5840 if (secs >= MAX_SECS) {
5841 absTime->tv_sec = max_secs;
5842 absTime->tv_nsec = 0;
5843 } else {
5844 absTime->tv_sec = now.tv_sec + secs;
5845 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5846 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5847 absTime->tv_nsec -= NANOSECS_PER_SEC;
5848 ++absTime->tv_sec; // note: this must be <= max_secs
5849 }
5850 }
5851 }
5852 } else {
5853 // must be relative using monotonic clock
5854 struct timespec now;
5855 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5856 assert_status(status == 0, status, "clock_gettime");
5857 max_secs = now.tv_sec + MAX_SECS;
5858 jlong secs = time / NANOSECS_PER_SEC;
5859 if (secs >= MAX_SECS) {
5860 absTime->tv_sec = max_secs;
5861 absTime->tv_nsec = 0;
5862 } else {
5863 absTime->tv_sec = now.tv_sec + secs;
5864 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5865 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5866 absTime->tv_nsec -= NANOSECS_PER_SEC;
5867 ++absTime->tv_sec; // note: this must be <= max_secs
5868 }
5869 }
5870 }
5871 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5872 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5873 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5874 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5875 }
5876
5877 void Parker::park(bool isAbsolute, jlong time) {
5878 // Ideally we'd do something useful while spinning, such
5879 // as calling unpackTime().
5880
5881 // Optional fast-path check:
5882 // Return immediately if a permit is available.
5883 // We depend on Atomic::xchg() having full barrier semantics
5884 // since we are doing a lock-free update to _counter.
5885 if (Atomic::xchg(0, &_counter) > 0) return;
5886
5887 Thread* thread = Thread::current();
5888 assert(thread->is_Java_thread(), "Must be JavaThread");
5889 JavaThread *jt = (JavaThread *)thread;
5890
5891 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5892 // Check interrupt before trying to wait
5893 if (Thread::is_interrupted(thread, false)) {
5894 return;
5895 }
5896
5897 // Next, demultiplex/decode time arguments
5898 timespec absTime;
5899 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5900 return;
5901 }
5902 if (time > 0) {
5903 unpackTime(&absTime, isAbsolute, time);
5904 }
5905
5906
5907 // Enter safepoint region
5908 // Beware of deadlocks such as 6317397.
5909 // The per-thread Parker:: mutex is a classic leaf-lock.
5910 // In particular a thread must never block on the Threads_lock while
5911 // holding the Parker:: mutex. If safepoints are pending both the
5912 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5913 ThreadBlockInVM tbivm(jt);
5914
5915 // Don't wait if cannot get lock since interference arises from
5916 // unblocking. Also. check interrupt before trying wait
5917 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5918 return;
5919 }
5920
5921 int status;
5922 if (_counter > 0) { // no wait needed
5923 _counter = 0;
5924 status = pthread_mutex_unlock(_mutex);
5925 assert_status(status == 0, status, "invariant");
5926 // Paranoia to ensure our locked and lock-free paths interact
5927 // correctly with each other and Java-level accesses.
5928 OrderAccess::fence();
5929 return;
5930 }
5931
5932 #ifdef ASSERT
5933 // Don't catch signals while blocked; let the running threads have the signals.
5934 // (This allows a debugger to break into the running thread.)
5935 sigset_t oldsigs;
5936 sigemptyset(&oldsigs);
5937 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5938 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5939 #endif
5940
5941 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5942 jt->set_suspend_equivalent();
5943 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5944
5945 assert(_cur_index == -1, "invariant");
5946 if (time == 0) {
5947 _cur_index = REL_INDEX; // arbitrary choice when not timed
5948 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5949 } else {
5950 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5951 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5952 }
5953 _cur_index = -1;
5954 assert_status(status == 0 || status == EINTR ||
5955 status == ETIME || status == ETIMEDOUT,
5956 status, "cond_timedwait");
5957
5958 #ifdef ASSERT
5959 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5960 #endif
5961
5962 _counter = 0;
5963 status = pthread_mutex_unlock(_mutex);
5964 assert_status(status == 0, status, "invariant");
5965 // Paranoia to ensure our locked and lock-free paths interact
5966 // correctly with each other and Java-level accesses.
5967 OrderAccess::fence();
5968
5969 // If externally suspended while waiting, re-suspend
5970 if (jt->handle_special_suspend_equivalent_condition()) {
5971 jt->java_suspend_self();
5972 }
5973 }
5974
5975 void Parker::unpark() {
5976 int status = pthread_mutex_lock(_mutex);
5977 assert_status(status == 0, status, "invariant");
5978 const int s = _counter;
5979 _counter = 1;
5980 // must capture correct index before unlocking
5981 int index = _cur_index;
5982 status = pthread_mutex_unlock(_mutex);
5983 assert_status(status == 0, status, "invariant");
5984 if (s < 1 && index != -1) {
5985 // thread is definitely parked
5986 status = pthread_cond_signal(&_cond[index]);
5987 assert_status(status == 0, status, "invariant");
5988 }
5989 }
5990
5991
5992 extern char** environ;
5993
5994 // Run the specified command in a separate process. Return its exit value,
5995 // or -1 on failure (e.g. can't fork a new process).
5996 // Unlike system(), this function can be called from signal handler. It
5997 // doesn't block SIGINT et al.
5998 int os::fork_and_exec(char* cmd) {
5999 const char * argv[4] = {"sh", "-c", cmd, NULL};
6000
6001 pid_t pid = fork();
6002
6003 if (pid < 0) {
6004 // fork failed
6005 return -1;
6006
6007 } else if (pid == 0) {
6008 // child process
6009
6010 execve("/bin/sh", (char* const*)argv, environ);
6011
6012 // execve failed
6013 _exit(-1);
6014
6015 } else {
6016 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6017 // care about the actual exit code, for now.
6018
6019 int status;
6020
6021 // Wait for the child process to exit. This returns immediately if
6022 // the child has already exited. */
6023 while (waitpid(pid, &status, 0) < 0) {
6024 switch (errno) {
6025 case ECHILD: return 0;
6026 case EINTR: break;
6027 default: return -1;
6028 }
6029 }
6030
6031 if (WIFEXITED(status)) {
6032 // The child exited normally; get its exit code.
6033 return WEXITSTATUS(status);
6034 } else if (WIFSIGNALED(status)) {
6035 // The child exited because of a signal
6036 // The best value to return is 0x80 + signal number,
6037 // because that is what all Unix shells do, and because
6038 // it allows callers to distinguish between process exit and
6039 // process death by signal.
6040 return 0x80 + WTERMSIG(status);
6041 } else {
6042 // Unknown exit code; pass it through
6043 return status;
6044 }
6045 }
6046 }
6047
6048 // is_headless_jre()
6049 //
6050 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6051 // in order to report if we are running in a headless jre
6052 //
6053 // Since JDK8 xawt/libmawt.so was moved into the same directory
6054 // as libawt.so, and renamed libawt_xawt.so
6055 //
6056 bool os::is_headless_jre() {
6057 struct stat statbuf;
6058 char buf[MAXPATHLEN];
6059 char libmawtpath[MAXPATHLEN];
6060 const char *xawtstr = "/xawt/libmawt.so";
6061 const char *new_xawtstr = "/libawt_xawt.so";
6062 char *p;
6063
6064 // Get path to libjvm.so
6065 os::jvm_path(buf, sizeof(buf));
6066
6067 // Get rid of libjvm.so
6068 p = strrchr(buf, '/');
6069 if (p == NULL) {
6070 return false;
6071 } else {
6072 *p = '\0';
6073 }
6074
6075 // Get rid of client or server
6076 p = strrchr(buf, '/');
6077 if (p == NULL) {
6078 return false;
6079 } else {
6080 *p = '\0';
6081 }
6082
6083 // check xawt/libmawt.so
6084 strcpy(libmawtpath, buf);
6085 strcat(libmawtpath, xawtstr);
6086 if (::stat(libmawtpath, &statbuf) == 0) return false;
6087
6088 // check libawt_xawt.so
6089 strcpy(libmawtpath, buf);
6090 strcat(libmawtpath, new_xawtstr);
6091 if (::stat(libmawtpath, &statbuf) == 0) return false;
6092
6093 return true;
6094 }
6095
6096 // Get the default path to the core file
6097 // Returns the length of the string
6098 int os::get_core_path(char* buffer, size_t bufferSize) {
6099 /*
6100 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6101 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6102 */
6103 const int core_pattern_len = 129;
6104 char core_pattern[core_pattern_len] = {0};
6105
6106 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6107 if (core_pattern_file == -1) {
6108 return -1;
6109 }
6110
6111 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6112 ::close(core_pattern_file);
6113 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6114 return -1;
6115 }
6116 if (core_pattern[ret-1] == '\n') {
6117 core_pattern[ret-1] = '\0';
6118 } else {
6119 core_pattern[ret] = '\0';
6120 }
6121
6122 char *pid_pos = strstr(core_pattern, "%p");
6123 int written;
6124
6125 if (core_pattern[0] == '/') {
6126 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6127 } else {
6128 char cwd[PATH_MAX];
6129
6130 const char* p = get_current_directory(cwd, PATH_MAX);
6131 if (p == NULL) {
6132 return -1;
6133 }
6134
6135 if (core_pattern[0] == '|') {
6136 written = jio_snprintf(buffer, bufferSize,
6137 "\"%s\" (or dumping to %s/core.%d)",
6138 &core_pattern[1], p, current_process_id());
6139 } else {
6140 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6141 }
6142 }
6143
6144 if (written < 0) {
6145 return -1;
6146 }
6147
6148 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6149 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6150
6151 if (core_uses_pid_file != -1) {
6152 char core_uses_pid = 0;
6153 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6154 ::close(core_uses_pid_file);
6155
6156 if (core_uses_pid == '1') {
6157 jio_snprintf(buffer + written, bufferSize - written,
6158 ".%d", current_process_id());
6159 }
6160 }
6161 }
6162
6163 return strlen(buffer);
6164 }
6165
6166 bool os::start_debugging(char *buf, int buflen) {
6167 int len = (int)strlen(buf);
6168 char *p = &buf[len];
6169
6170 jio_snprintf(p, buflen-len,
6171 "\n\n"
6172 "Do you want to debug the problem?\n\n"
6173 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6174 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6175 "Otherwise, press RETURN to abort...",
6176 os::current_process_id(), os::current_process_id(),
6177 os::current_thread_id(), os::current_thread_id());
6178
6179 bool yes = os::message_box("Unexpected Error", buf);
6180
6181 if (yes) {
6182 // yes, user asked VM to launch debugger
6183 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6184 os::current_process_id(), os::current_process_id());
6185
6186 os::fork_and_exec(buf);
6187 yes = false;
6188 }
6189 return yes;
6190 }
6191
6192
6193 // Java/Compiler thread:
6194 //
6195 // Low memory addresses
6196 // P0 +------------------------+
6197 // | |\ Java thread created by VM does not have glibc
6198 // | glibc guard page | - guard page, attached Java thread usually has
6199 // | |/ 1 glibc guard page.
6200 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6201 // | |\
6202 // | HotSpot Guard Pages | - red, yellow and reserved pages
6203 // | |/
6204 // +------------------------+ JavaThread::stack_reserved_zone_base()
6205 // | |\
6206 // | Normal Stack | -
6207 // | |/
6208 // P2 +------------------------+ Thread::stack_base()
6209 //
6210 // Non-Java thread:
6211 //
6212 // Low memory addresses
6213 // P0 +------------------------+
6214 // | |\
6215 // | glibc guard page | - usually 1 page
6216 // | |/
6217 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6218 // | |\
6219 // | Normal Stack | -
6220 // | |/
6221 // P2 +------------------------+ Thread::stack_base()
6222 //
6223 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6224 // returned from pthread_attr_getstack().
6225 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6226 // of the stack size given in pthread_attr. We work around this for
6227 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6228 //
6229 #ifndef ZERO
6230 static void current_stack_region(address * bottom, size_t * size) {
6231 if (os::Linux::is_initial_thread()) {
6232 // initial thread needs special handling because pthread_getattr_np()
6233 // may return bogus value.
6234 *bottom = os::Linux::initial_thread_stack_bottom();
6235 *size = os::Linux::initial_thread_stack_size();
6236 } else {
6237 pthread_attr_t attr;
6238
6239 int rslt = pthread_getattr_np(pthread_self(), &attr);
6240
6241 // JVM needs to know exact stack location, abort if it fails
6242 if (rslt != 0) {
6243 if (rslt == ENOMEM) {
6244 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6245 } else {
6246 fatal("pthread_getattr_np failed with error = %d", rslt);
6247 }
6248 }
6249
6250 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6251 fatal("Cannot locate current stack attributes!");
6252 }
6253
6254 // Work around NPTL stack guard error.
6255 size_t guard_size = 0;
6256 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6257 if (rslt != 0) {
6258 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6259 }
6260 *bottom += guard_size;
6261 *size -= guard_size;
6262
6263 pthread_attr_destroy(&attr);
6264
6265 }
6266 assert(os::current_stack_pointer() >= *bottom &&
6267 os::current_stack_pointer() < *bottom + *size, "just checking");
6268 }
6269
6270 address os::current_stack_base() {
6271 address bottom;
6272 size_t size;
6273 current_stack_region(&bottom, &size);
6274 return (bottom + size);
6275 }
6276
6277 size_t os::current_stack_size() {
6278 // This stack size includes the usable stack and HotSpot guard pages
6279 // (for the threads that have Hotspot guard pages).
6280 address bottom;
6281 size_t size;
6282 current_stack_region(&bottom, &size);
6283 return size;
6284 }
6285 #endif
6286
6287 static inline struct timespec get_mtime(const char* filename) {
6288 struct stat st;
6289 int ret = os::stat(filename, &st);
6290 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6291 return st.st_mtim;
6292 }
6293
6294 int os::compare_file_modified_times(const char* file1, const char* file2) {
6295 struct timespec filetime1 = get_mtime(file1);
6296 struct timespec filetime2 = get_mtime(file2);
6297 int diff = filetime1.tv_sec - filetime2.tv_sec;
6298 if (diff == 0) {
6299 return filetime1.tv_nsec - filetime2.tv_nsec;
6300 }
6301 return diff;
6302 }
6303
6304 /////////////// Unit tests ///////////////
6305
6306 #ifndef PRODUCT
6307
6308 #define test_log(...) \
6309 do { \
6310 if (VerboseInternalVMTests) { \
6311 tty->print_cr(__VA_ARGS__); \
6312 tty->flush(); \
6313 } \
6314 } while (false)
6315
6316 class TestReserveMemorySpecial : AllStatic {
6317 public:
6318 static void small_page_write(void* addr, size_t size) {
6319 size_t page_size = os::vm_page_size();
6320
6321 char* end = (char*)addr + size;
6322 for (char* p = (char*)addr; p < end; p += page_size) {
6323 *p = 1;
6324 }
6325 }
6326
6327 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6328 if (!UseHugeTLBFS) {
6329 return;
6330 }
6331
6332 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6333
6334 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6335
6336 if (addr != NULL) {
6337 small_page_write(addr, size);
6338
6339 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6340 }
6341 }
6342
6343 static void test_reserve_memory_special_huge_tlbfs_only() {
6344 if (!UseHugeTLBFS) {
6345 return;
6346 }
6347
6348 size_t lp = os::large_page_size();
6349
6350 for (size_t size = lp; size <= lp * 10; size += lp) {
6351 test_reserve_memory_special_huge_tlbfs_only(size);
6352 }
6353 }
6354
6355 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6356 size_t lp = os::large_page_size();
6357 size_t ag = os::vm_allocation_granularity();
6358
6359 // sizes to test
6360 const size_t sizes[] = {
6361 lp, lp + ag, lp + lp / 2, lp * 2,
6362 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6363 lp * 10, lp * 10 + lp / 2
6364 };
6365 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6366
6367 // For each size/alignment combination, we test three scenarios:
6368 // 1) with req_addr == NULL
6369 // 2) with a non-null req_addr at which we expect to successfully allocate
6370 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6371 // expect the allocation to either fail or to ignore req_addr
6372
6373 // Pre-allocate two areas; they shall be as large as the largest allocation
6374 // and aligned to the largest alignment we will be testing.
6375 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6376 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6377 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6378 -1, 0);
6379 assert(mapping1 != MAP_FAILED, "should work");
6380
6381 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6382 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6383 -1, 0);
6384 assert(mapping2 != MAP_FAILED, "should work");
6385
6386 // Unmap the first mapping, but leave the second mapping intact: the first
6387 // mapping will serve as a value for a "good" req_addr (case 2). The second
6388 // mapping, still intact, as "bad" req_addr (case 3).
6389 ::munmap(mapping1, mapping_size);
6390
6391 // Case 1
6392 test_log("%s, req_addr NULL:", __FUNCTION__);
6393 test_log("size align result");
6394
6395 for (int i = 0; i < num_sizes; i++) {
6396 const size_t size = sizes[i];
6397 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6398 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6399 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6400 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6401 if (p != NULL) {
6402 assert(is_ptr_aligned(p, alignment), "must be");
6403 small_page_write(p, size);
6404 os::Linux::release_memory_special_huge_tlbfs(p, size);
6405 }
6406 }
6407 }
6408
6409 // Case 2
6410 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6411 test_log("size align req_addr result");
6412
6413 for (int i = 0; i < num_sizes; i++) {
6414 const size_t size = sizes[i];
6415 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6416 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6417 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6418 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6419 size, alignment, p2i(req_addr), p2i(p),
6420 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6421 if (p != NULL) {
6422 assert(p == req_addr, "must be");
6423 small_page_write(p, size);
6424 os::Linux::release_memory_special_huge_tlbfs(p, size);
6425 }
6426 }
6427 }
6428
6429 // Case 3
6430 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6431 test_log("size align req_addr result");
6432
6433 for (int i = 0; i < num_sizes; i++) {
6434 const size_t size = sizes[i];
6435 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6436 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6437 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6438 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6439 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6440 // as the area around req_addr contains already existing mappings, the API should always
6441 // return NULL (as per contract, it cannot return another address)
6442 assert(p == NULL, "must be");
6443 }
6444 }
6445
6446 ::munmap(mapping2, mapping_size);
6447
6448 }
6449
6450 static void test_reserve_memory_special_huge_tlbfs() {
6451 if (!UseHugeTLBFS) {
6452 return;
6453 }
6454
6455 test_reserve_memory_special_huge_tlbfs_only();
6456 test_reserve_memory_special_huge_tlbfs_mixed();
6457 }
6458
6459 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6460 if (!UseSHM) {
6461 return;
6462 }
6463
6464 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6465
6466 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6467
6468 if (addr != NULL) {
6469 assert(is_ptr_aligned(addr, alignment), "Check");
6470 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6471
6472 small_page_write(addr, size);
6473
6474 os::Linux::release_memory_special_shm(addr, size);
6475 }
6476 }
6477
6478 static void test_reserve_memory_special_shm() {
6479 size_t lp = os::large_page_size();
6480 size_t ag = os::vm_allocation_granularity();
6481
6482 for (size_t size = ag; size < lp * 3; size += ag) {
6483 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6484 test_reserve_memory_special_shm(size, alignment);
6485 }
6486 }
6487 }
6488
6489 static void test() {
6490 test_reserve_memory_special_huge_tlbfs();
6491 test_reserve_memory_special_shm();
6492 }
6493 };
6494
6495 void TestReserveMemorySpecial_test() {
6496 TestReserveMemorySpecial::test();
6497 }
6498
6499 #endif