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