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 "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
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_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/atomic.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "semaphore_posix.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/align.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 #ifndef _GNU_SOURCE
110 #define _GNU_SOURCE
111 #include <sched.h>
112 #undef _GNU_SOURCE
113 #else
114 #include <sched.h>
115 #endif
116
117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
118 // getrusage() is prepared to handle the associated failure.
119 #ifndef RUSAGE_THREAD
120 #define RUSAGE_THREAD (1) /* only the calling thread */
121 #endif
122
123 #define MAX_PATH (2 * K)
124
125 #define MAX_SECS 100000000
126
127 // for timer info max values which include all bits
128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129
130 #define LARGEPAGES_BIT (1 << 6)
131 ////////////////////////////////////////////////////////////////////////////////
132 // global variables
133 julong os::Linux::_physical_memory = 0;
134
135 address os::Linux::_initial_thread_stack_bottom = NULL;
136 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137
138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
141 Mutex* os::Linux::_createThread_lock = NULL;
142 pthread_t os::Linux::_main_thread;
143 int os::Linux::_page_size = -1;
144 bool os::Linux::_supports_fast_thread_cpu_time = false;
145 uint32_t os::Linux::_os_version = 0;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148
149 static jlong initial_time_count=0;
150
151 static int clock_tics_per_sec = 100;
152
153 // For diagnostics to print a message once. see run_periodic_checks
154 static sigset_t check_signal_done;
155 static bool check_signals = true;
156
157 // Signal number used to suspend/resume a thread
158
159 // do not use any signal number less than SIGSEGV, see 4355769
160 static int SR_signum = SIGUSR2;
161 sigset_t SR_sigset;
162
163 // utility functions
164
165 static int SR_initialize();
166
167 julong os::available_memory() {
168 return Linux::available_memory();
169 }
170
171 julong os::Linux::available_memory() {
172 // values in struct sysinfo are "unsigned long"
173 struct sysinfo si;
174 sysinfo(&si);
175
176 return (julong)si.freeram * si.mem_unit;
177 }
178
179 julong os::physical_memory() {
180 return Linux::physical_memory();
181 }
182
183 // Return true if user is running as root.
184
185 bool os::have_special_privileges() {
186 static bool init = false;
187 static bool privileges = false;
188 if (!init) {
189 privileges = (getuid() != geteuid()) || (getgid() != getegid());
190 init = true;
191 }
192 return privileges;
193 }
194
195
196 #ifndef SYS_gettid
197 // i386: 224, ia64: 1105, amd64: 186, sparc 143
198 #ifdef __ia64__
199 #define SYS_gettid 1105
200 #else
201 #ifdef __i386__
202 #define SYS_gettid 224
203 #else
204 #ifdef __amd64__
205 #define SYS_gettid 186
206 #else
207 #ifdef __sparc__
208 #define SYS_gettid 143
209 #else
210 #error define gettid for the arch
211 #endif
212 #endif
213 #endif
214 #endif
215 #endif
216
217
218 // pid_t gettid()
219 //
220 // Returns the kernel thread id of the currently running thread. Kernel
221 // thread id is used to access /proc.
222 pid_t os::Linux::gettid() {
223 int rslt = syscall(SYS_gettid);
224 assert(rslt != -1, "must be."); // old linuxthreads implementation?
225 return (pid_t)rslt;
226 }
227
228 // Most versions of linux have a bug where the number of processors are
229 // determined by looking at the /proc file system. In a chroot environment,
230 // the system call returns 1. This causes the VM to act as if it is
231 // a single processor and elide locking (see is_MP() call).
232 static bool unsafe_chroot_detected = false;
233 static const char *unstable_chroot_error = "/proc file system not found.\n"
234 "Java may be unstable running multithreaded in a chroot "
235 "environment on Linux when /proc filesystem is not mounted.";
236
237 void os::Linux::initialize_system_info() {
238 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
239 if (processor_count() == 1) {
240 pid_t pid = os::Linux::gettid();
241 char fname[32];
242 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
243 FILE *fp = fopen(fname, "r");
244 if (fp == NULL) {
245 unsafe_chroot_detected = true;
246 } else {
247 fclose(fp);
248 }
249 }
250 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
251 assert(processor_count() > 0, "linux error");
252 }
253
254 void os::init_system_properties_values() {
255 // The next steps are taken in the product version:
256 //
257 // Obtain the JAVA_HOME value from the location of libjvm.so.
258 // This library should be located at:
259 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
260 //
261 // If "/jre/lib/" appears at the right place in the path, then we
262 // assume libjvm.so is installed in a JDK and we use this path.
263 //
264 // Otherwise exit with message: "Could not create the Java virtual machine."
265 //
266 // The following extra steps are taken in the debugging version:
267 //
268 // If "/jre/lib/" does NOT appear at the right place in the path
269 // instead of exit check for $JAVA_HOME environment variable.
270 //
271 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
272 // then we append a fake suffix "hotspot/libjvm.so" to this path so
273 // it looks like libjvm.so is installed there
274 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
275 //
276 // Otherwise exit.
277 //
278 // Important note: if the location of libjvm.so changes this
279 // code needs to be changed accordingly.
280
281 // See ld(1):
282 // The linker uses the following search paths to locate required
283 // shared libraries:
284 // 1: ...
285 // ...
286 // 7: The default directories, normally /lib and /usr/lib.
287 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
288 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
289 #else
290 #define DEFAULT_LIBPATH "/lib:/usr/lib"
291 #endif
292
293 // Base path of extensions installed on the system.
294 #define SYS_EXT_DIR "/usr/java/packages"
295 #define EXTENSIONS_DIR "/lib/ext"
296
297 // Buffer that fits several sprintfs.
298 // Note that the space for the colon and the trailing null are provided
299 // by the nulls included by the sizeof operator.
300 const size_t bufsize =
301 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
302 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
303 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
304
305 // sysclasspath, java_home, dll_dir
306 {
307 char *pslash;
308 os::jvm_path(buf, bufsize);
309
310 // Found the full path to libjvm.so.
311 // Now cut the path to <java_home>/jre if we can.
312 pslash = strrchr(buf, '/');
313 if (pslash != NULL) {
314 *pslash = '\0'; // Get rid of /libjvm.so.
315 }
316 pslash = strrchr(buf, '/');
317 if (pslash != NULL) {
318 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
319 }
320 Arguments::set_dll_dir(buf);
321
322 if (pslash != NULL) {
323 pslash = strrchr(buf, '/');
324 if (pslash != NULL) {
325 *pslash = '\0'; // Get rid of /lib.
326 }
327 }
328 Arguments::set_java_home(buf);
329 set_boot_path('/', ':');
330 }
331
332 // Where to look for native libraries.
333 //
334 // Note: Due to a legacy implementation, most of the library path
335 // is set in the launcher. This was to accomodate linking restrictions
336 // on legacy Linux implementations (which are no longer supported).
337 // Eventually, all the library path setting will be done here.
338 //
339 // However, to prevent the proliferation of improperly built native
340 // libraries, the new path component /usr/java/packages is added here.
341 // Eventually, all the library path setting will be done here.
342 {
343 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
344 // should always exist (until the legacy problem cited above is
345 // addressed).
346 const char *v = ::getenv("LD_LIBRARY_PATH");
347 const char *v_colon = ":";
348 if (v == NULL) { v = ""; v_colon = ""; }
349 // That's +1 for the colon and +1 for the trailing '\0'.
350 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
351 strlen(v) + 1 +
352 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
353 mtInternal);
354 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
355 Arguments::set_library_path(ld_library_path);
356 FREE_C_HEAP_ARRAY(char, ld_library_path);
357 }
358
359 // Extensions directories.
360 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
361 Arguments::set_ext_dirs(buf);
362
363 FREE_C_HEAP_ARRAY(char, buf);
364
365 #undef DEFAULT_LIBPATH
366 #undef SYS_EXT_DIR
367 #undef EXTENSIONS_DIR
368 }
369
370 ////////////////////////////////////////////////////////////////////////////////
371 // breakpoint support
372
373 void os::breakpoint() {
374 BREAKPOINT;
375 }
376
377 extern "C" void breakpoint() {
378 // use debugger to set breakpoint here
379 }
380
381 ////////////////////////////////////////////////////////////////////////////////
382 // signal support
383
384 debug_only(static bool signal_sets_initialized = false);
385 static sigset_t unblocked_sigs, vm_sigs;
386
387 bool os::Linux::is_sig_ignored(int sig) {
388 struct sigaction oact;
389 sigaction(sig, (struct sigaction*)NULL, &oact);
390 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
391 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
392 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
393 return true;
394 } else {
395 return false;
396 }
397 }
398
399 void os::Linux::signal_sets_init() {
400 // Should also have an assertion stating we are still single-threaded.
401 assert(!signal_sets_initialized, "Already initialized");
402 // Fill in signals that are necessarily unblocked for all threads in
403 // the VM. Currently, we unblock the following signals:
404 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
405 // by -Xrs (=ReduceSignalUsage));
406 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
407 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
408 // the dispositions or masks wrt these signals.
409 // Programs embedding the VM that want to use the above signals for their
410 // own purposes must, at this time, use the "-Xrs" option to prevent
411 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
412 // (See bug 4345157, and other related bugs).
413 // In reality, though, unblocking these signals is really a nop, since
414 // these signals are not blocked by default.
415 sigemptyset(&unblocked_sigs);
416 sigaddset(&unblocked_sigs, SIGILL);
417 sigaddset(&unblocked_sigs, SIGSEGV);
418 sigaddset(&unblocked_sigs, SIGBUS);
419 sigaddset(&unblocked_sigs, SIGFPE);
420 #if defined(PPC64)
421 sigaddset(&unblocked_sigs, SIGTRAP);
422 #endif
423 sigaddset(&unblocked_sigs, SR_signum);
424
425 if (!ReduceSignalUsage) {
426 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
427 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
428 }
429 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
430 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
431 }
432 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
433 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
434 }
435 }
436 // Fill in signals that are blocked by all but the VM thread.
437 sigemptyset(&vm_sigs);
438 if (!ReduceSignalUsage) {
439 sigaddset(&vm_sigs, BREAK_SIGNAL);
440 }
441 debug_only(signal_sets_initialized = true);
442
443 }
444
445 // These are signals that are unblocked while a thread is running Java.
446 // (For some reason, they get blocked by default.)
447 sigset_t* os::Linux::unblocked_signals() {
448 assert(signal_sets_initialized, "Not initialized");
449 return &unblocked_sigs;
450 }
451
452 // These are the signals that are blocked while a (non-VM) thread is
453 // running Java. Only the VM thread handles these signals.
454 sigset_t* os::Linux::vm_signals() {
455 assert(signal_sets_initialized, "Not initialized");
456 return &vm_sigs;
457 }
458
459 void os::Linux::hotspot_sigmask(Thread* thread) {
460
461 //Save caller's signal mask before setting VM signal mask
462 sigset_t caller_sigmask;
463 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
464
465 OSThread* osthread = thread->osthread();
466 osthread->set_caller_sigmask(caller_sigmask);
467
468 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
469
470 if (!ReduceSignalUsage) {
471 if (thread->is_VM_thread()) {
472 // Only the VM thread handles BREAK_SIGNAL ...
473 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
474 } else {
475 // ... all other threads block BREAK_SIGNAL
476 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
477 }
478 }
479 }
480
481 //////////////////////////////////////////////////////////////////////////////
482 // detecting pthread library
483
484 void os::Linux::libpthread_init() {
485 // Save glibc and pthread version strings.
486 #if !defined(_CS_GNU_LIBC_VERSION) || \
487 !defined(_CS_GNU_LIBPTHREAD_VERSION)
488 #error "glibc too old (< 2.3.2)"
489 #endif
490
491 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
492 assert(n > 0, "cannot retrieve glibc version");
493 char *str = (char *)malloc(n, mtInternal);
494 confstr(_CS_GNU_LIBC_VERSION, str, n);
495 os::Linux::set_glibc_version(str);
496
497 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
498 assert(n > 0, "cannot retrieve pthread version");
499 str = (char *)malloc(n, mtInternal);
500 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
501 os::Linux::set_libpthread_version(str);
502 }
503
504 /////////////////////////////////////////////////////////////////////////////
505 // thread stack expansion
506
507 // os::Linux::manually_expand_stack() takes care of expanding the thread
508 // stack. Note that this is normally not needed: pthread stacks allocate
509 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
510 // committed. Therefore it is not necessary to expand the stack manually.
511 //
512 // Manually expanding the stack was historically needed on LinuxThreads
513 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
514 // it is kept to deal with very rare corner cases:
515 //
516 // For one, user may run the VM on an own implementation of threads
517 // whose stacks are - like the old LinuxThreads - implemented using
518 // mmap(MAP_GROWSDOWN).
519 //
520 // Also, this coding may be needed if the VM is running on the primordial
521 // thread. Normally we avoid running on the primordial thread; however,
522 // user may still invoke the VM on the primordial thread.
523 //
524 // The following historical comment describes the details about running
525 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
526
527
528 // Force Linux kernel to expand current thread stack. If "bottom" is close
529 // to the stack guard, caller should block all signals.
530 //
531 // MAP_GROWSDOWN:
532 // A special mmap() flag that is used to implement thread stacks. It tells
533 // kernel that the memory region should extend downwards when needed. This
534 // allows early versions of LinuxThreads to only mmap the first few pages
535 // when creating a new thread. Linux kernel will automatically expand thread
536 // stack as needed (on page faults).
537 //
538 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
539 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
540 // region, it's hard to tell if the fault is due to a legitimate stack
541 // access or because of reading/writing non-exist memory (e.g. buffer
542 // overrun). As a rule, if the fault happens below current stack pointer,
543 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
544 // application (see Linux kernel fault.c).
545 //
546 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
547 // stack overflow detection.
548 //
549 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
550 // not use MAP_GROWSDOWN.
551 //
552 // To get around the problem and allow stack banging on Linux, we need to
553 // manually expand thread stack after receiving the SIGSEGV.
554 //
555 // There are two ways to expand thread stack to address "bottom", we used
556 // both of them in JVM before 1.5:
557 // 1. adjust stack pointer first so that it is below "bottom", and then
558 // touch "bottom"
559 // 2. mmap() the page in question
560 //
561 // Now alternate signal stack is gone, it's harder to use 2. For instance,
562 // if current sp is already near the lower end of page 101, and we need to
563 // call mmap() to map page 100, it is possible that part of the mmap() frame
564 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
565 // That will destroy the mmap() frame and cause VM to crash.
566 //
567 // The following code works by adjusting sp first, then accessing the "bottom"
568 // page to force a page fault. Linux kernel will then automatically expand the
569 // stack mapping.
570 //
571 // _expand_stack_to() assumes its frame size is less than page size, which
572 // should always be true if the function is not inlined.
573
574 static void NOINLINE _expand_stack_to(address bottom) {
575 address sp;
576 size_t size;
577 volatile char *p;
578
579 // Adjust bottom to point to the largest address within the same page, it
580 // gives us a one-page buffer if alloca() allocates slightly more memory.
581 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
582 bottom += os::Linux::page_size() - 1;
583
584 // sp might be slightly above current stack pointer; if that's the case, we
585 // will alloca() a little more space than necessary, which is OK. Don't use
586 // os::current_stack_pointer(), as its result can be slightly below current
587 // stack pointer, causing us to not alloca enough to reach "bottom".
588 sp = (address)&sp;
589
590 if (sp > bottom) {
591 size = sp - bottom;
592 p = (volatile char *)alloca(size);
593 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
594 p[0] = '\0';
595 }
596 }
597
598 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
599 assert(t!=NULL, "just checking");
600 assert(t->osthread()->expanding_stack(), "expand should be set");
601 assert(t->stack_base() != NULL, "stack_base was not initialized");
602
603 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
604 sigset_t mask_all, old_sigset;
605 sigfillset(&mask_all);
606 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
607 _expand_stack_to(addr);
608 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
609 return true;
610 }
611 return false;
612 }
613
614 //////////////////////////////////////////////////////////////////////////////
615 // create new thread
616
617 // Thread start routine for all newly created threads
618 static void *thread_native_entry(Thread *thread) {
619 // Try to randomize the cache line index of hot stack frames.
620 // This helps when threads of the same stack traces evict each other's
621 // cache lines. The threads can be either from the same JVM instance, or
622 // from different JVM instances. The benefit is especially true for
623 // processors with hyperthreading technology.
624 static int counter = 0;
625 int pid = os::current_process_id();
626 alloca(((pid ^ counter++) & 7) * 128);
627
628 thread->initialize_thread_current();
629
630 OSThread* osthread = thread->osthread();
631 Monitor* sync = osthread->startThread_lock();
632
633 osthread->set_thread_id(os::current_thread_id());
634
635 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
636 os::current_thread_id(), (uintx) pthread_self());
637
638 if (UseNUMA) {
639 int lgrp_id = os::numa_get_group_id();
640 if (lgrp_id != -1) {
641 thread->set_lgrp_id(lgrp_id);
642 }
643 }
644 // initialize signal mask for this thread
645 os::Linux::hotspot_sigmask(thread);
646
647 // initialize floating point control register
648 os::Linux::init_thread_fpu_state();
649
650 // handshaking with parent thread
651 {
652 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
653
654 // notify parent thread
655 osthread->set_state(INITIALIZED);
656 sync->notify_all();
657
658 // wait until os::start_thread()
659 while (osthread->get_state() == INITIALIZED) {
660 sync->wait(Mutex::_no_safepoint_check_flag);
661 }
662 }
663
664 // call one more level start routine
665 thread->run();
666
667 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
668 os::current_thread_id(), (uintx) pthread_self());
669
670 // If a thread has not deleted itself ("delete this") as part of its
671 // termination sequence, we have to ensure thread-local-storage is
672 // cleared before we actually terminate. No threads should ever be
673 // deleted asynchronously with respect to their termination.
674 if (Thread::current_or_null_safe() != NULL) {
675 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
676 thread->clear_thread_current();
677 }
678
679 return 0;
680 }
681
682 bool os::create_thread(Thread* thread, ThreadType thr_type,
683 size_t req_stack_size) {
684 assert(thread->osthread() == NULL, "caller responsible");
685
686 // Allocate the OSThread object
687 OSThread* osthread = new OSThread(NULL, NULL);
688 if (osthread == NULL) {
689 return false;
690 }
691
692 // set the correct thread state
693 osthread->set_thread_type(thr_type);
694
695 // Initial state is ALLOCATED but not INITIALIZED
696 osthread->set_state(ALLOCATED);
697
698 thread->set_osthread(osthread);
699
700 // init thread attributes
701 pthread_attr_t attr;
702 pthread_attr_init(&attr);
703 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
704
705 // Calculate stack size if it's not specified by caller.
706 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
707 // In the Linux NPTL pthread implementation the guard size mechanism
708 // is not implemented properly. The posix standard requires adding
709 // the size of the guard pages to the stack size, instead Linux
710 // takes the space out of 'stacksize'. Thus we adapt the requested
711 // stack_size by the size of the guard pages to mimick proper
712 // behaviour. However, be careful not to end up with a size
713 // of zero due to overflow. Don't add the guard page in that case.
714 size_t guard_size = os::Linux::default_guard_size(thr_type);
715 if (stack_size <= SIZE_MAX - guard_size) {
716 stack_size += guard_size;
717 }
718 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
719
720 int status = pthread_attr_setstacksize(&attr, stack_size);
721 assert_status(status == 0, status, "pthread_attr_setstacksize");
722
723 // Configure glibc guard page.
724 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
725
726 ThreadState state;
727
728 {
729 pthread_t tid;
730 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
731
732 char buf[64];
733 if (ret == 0) {
734 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
735 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
736 } else {
737 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
738 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
739 }
740
741 pthread_attr_destroy(&attr);
742
743 if (ret != 0) {
744 // Need to clean up stuff we've allocated so far
745 thread->set_osthread(NULL);
746 delete osthread;
747 return false;
748 }
749
750 // Store pthread info into the OSThread
751 osthread->set_pthread_id(tid);
752
753 // Wait until child thread is either initialized or aborted
754 {
755 Monitor* sync_with_child = osthread->startThread_lock();
756 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
757 while ((state = osthread->get_state()) == ALLOCATED) {
758 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
759 }
760 }
761 }
762
763 // Aborted due to thread limit being reached
764 if (state == ZOMBIE) {
765 thread->set_osthread(NULL);
766 delete osthread;
767 return false;
768 }
769
770 // The thread is returned suspended (in state INITIALIZED),
771 // and is started higher up in the call chain
772 assert(state == INITIALIZED, "race condition");
773 return true;
774 }
775
776 /////////////////////////////////////////////////////////////////////////////
777 // attach existing thread
778
779 // bootstrap the main thread
780 bool os::create_main_thread(JavaThread* thread) {
781 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
782 return create_attached_thread(thread);
783 }
784
785 bool os::create_attached_thread(JavaThread* thread) {
786 #ifdef ASSERT
787 thread->verify_not_published();
788 #endif
789
790 // Allocate the OSThread object
791 OSThread* osthread = new OSThread(NULL, NULL);
792
793 if (osthread == NULL) {
794 return false;
795 }
796
797 // Store pthread info into the OSThread
798 osthread->set_thread_id(os::Linux::gettid());
799 osthread->set_pthread_id(::pthread_self());
800
801 // initialize floating point control register
802 os::Linux::init_thread_fpu_state();
803
804 // Initial thread state is RUNNABLE
805 osthread->set_state(RUNNABLE);
806
807 thread->set_osthread(osthread);
808
809 if (UseNUMA) {
810 int lgrp_id = os::numa_get_group_id();
811 if (lgrp_id != -1) {
812 thread->set_lgrp_id(lgrp_id);
813 }
814 }
815
816 if (os::Linux::is_initial_thread()) {
817 // If current thread is initial thread, its stack is mapped on demand,
818 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
819 // the entire stack region to avoid SEGV in stack banging.
820 // It is also useful to get around the heap-stack-gap problem on SuSE
821 // kernel (see 4821821 for details). We first expand stack to the top
822 // of yellow zone, then enable stack yellow zone (order is significant,
823 // enabling yellow zone first will crash JVM on SuSE Linux), so there
824 // is no gap between the last two virtual memory regions.
825
826 JavaThread *jt = (JavaThread *)thread;
827 address addr = jt->stack_reserved_zone_base();
828 assert(addr != NULL, "initialization problem?");
829 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
830
831 osthread->set_expanding_stack();
832 os::Linux::manually_expand_stack(jt, addr);
833 osthread->clear_expanding_stack();
834 }
835
836 // initialize signal mask for this thread
837 // and save the caller's signal mask
838 os::Linux::hotspot_sigmask(thread);
839
840 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
841 os::current_thread_id(), (uintx) pthread_self());
842
843 return true;
844 }
845
846 void os::pd_start_thread(Thread* thread) {
847 OSThread * osthread = thread->osthread();
848 assert(osthread->get_state() != INITIALIZED, "just checking");
849 Monitor* sync_with_child = osthread->startThread_lock();
850 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
851 sync_with_child->notify();
852 }
853
854 // Free Linux resources related to the OSThread
855 void os::free_thread(OSThread* osthread) {
856 assert(osthread != NULL, "osthread not set");
857
858 // We are told to free resources of the argument thread,
859 // but we can only really operate on the current thread.
860 assert(Thread::current()->osthread() == osthread,
861 "os::free_thread but not current thread");
862
863 #ifdef ASSERT
864 sigset_t current;
865 sigemptyset(¤t);
866 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
867 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
868 #endif
869
870 // Restore caller's signal mask
871 sigset_t sigmask = osthread->caller_sigmask();
872 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
873
874 delete osthread;
875 }
876
877 //////////////////////////////////////////////////////////////////////////////
878 // initial thread
879
880 // Check if current thread is the initial thread, similar to Solaris thr_main.
881 bool os::Linux::is_initial_thread(void) {
882 char dummy;
883 // If called before init complete, thread stack bottom will be null.
884 // Can be called if fatal error occurs before initialization.
885 if (initial_thread_stack_bottom() == NULL) return false;
886 assert(initial_thread_stack_bottom() != NULL &&
887 initial_thread_stack_size() != 0,
888 "os::init did not locate initial thread's stack region");
889 if ((address)&dummy >= initial_thread_stack_bottom() &&
890 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
891 return true;
892 } else {
893 return false;
894 }
895 }
896
897 // Find the virtual memory area that contains addr
898 static bool find_vma(address addr, address* vma_low, address* vma_high) {
899 FILE *fp = fopen("/proc/self/maps", "r");
900 if (fp) {
901 address low, high;
902 while (!feof(fp)) {
903 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
904 if (low <= addr && addr < high) {
905 if (vma_low) *vma_low = low;
906 if (vma_high) *vma_high = high;
907 fclose(fp);
908 return true;
909 }
910 }
911 for (;;) {
912 int ch = fgetc(fp);
913 if (ch == EOF || ch == (int)'\n') break;
914 }
915 }
916 fclose(fp);
917 }
918 return false;
919 }
920
921 // Locate initial thread stack. This special handling of initial thread stack
922 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
923 // bogus value for the primordial process thread. While the launcher has created
924 // the VM in a new thread since JDK 6, we still have to allow for the use of the
925 // JNI invocation API from a primordial thread.
926 void os::Linux::capture_initial_stack(size_t max_size) {
927
928 // max_size is either 0 (which means accept OS default for thread stacks) or
929 // a user-specified value known to be at least the minimum needed. If we
930 // are actually on the primordial thread we can make it appear that we have a
931 // smaller max_size stack by inserting the guard pages at that location. But we
932 // cannot do anything to emulate a larger stack than what has been provided by
933 // the OS or threading library. In fact if we try to use a stack greater than
934 // what is set by rlimit then we will crash the hosting process.
935
936 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
937 // If this is "unlimited" then it will be a huge value.
938 struct rlimit rlim;
939 getrlimit(RLIMIT_STACK, &rlim);
940 size_t stack_size = rlim.rlim_cur;
941
942 // 6308388: a bug in ld.so will relocate its own .data section to the
943 // lower end of primordial stack; reduce ulimit -s value a little bit
944 // so we won't install guard page on ld.so's data section.
945 stack_size -= 2 * page_size();
946
947 // Try to figure out where the stack base (top) is. This is harder.
948 //
949 // When an application is started, glibc saves the initial stack pointer in
950 // a global variable "__libc_stack_end", which is then used by system
951 // libraries. __libc_stack_end should be pretty close to stack top. The
952 // variable is available since the very early days. However, because it is
953 // a private interface, it could disappear in the future.
954 //
955 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
956 // to __libc_stack_end, it is very close to stack top, but isn't the real
957 // stack top. Note that /proc may not exist if VM is running as a chroot
958 // program, so reading /proc/<pid>/stat could fail. Also the contents of
959 // /proc/<pid>/stat could change in the future (though unlikely).
960 //
961 // We try __libc_stack_end first. If that doesn't work, look for
962 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
963 // as a hint, which should work well in most cases.
964
965 uintptr_t stack_start;
966
967 // try __libc_stack_end first
968 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
969 if (p && *p) {
970 stack_start = *p;
971 } else {
972 // see if we can get the start_stack field from /proc/self/stat
973 FILE *fp;
974 int pid;
975 char state;
976 int ppid;
977 int pgrp;
978 int session;
979 int nr;
980 int tpgrp;
981 unsigned long flags;
982 unsigned long minflt;
983 unsigned long cminflt;
984 unsigned long majflt;
985 unsigned long cmajflt;
986 unsigned long utime;
987 unsigned long stime;
988 long cutime;
989 long cstime;
990 long prio;
991 long nice;
992 long junk;
993 long it_real;
994 uintptr_t start;
995 uintptr_t vsize;
996 intptr_t rss;
997 uintptr_t rsslim;
998 uintptr_t scodes;
999 uintptr_t ecode;
1000 int i;
1001
1002 // Figure what the primordial thread stack base is. Code is inspired
1003 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1004 // followed by command name surrounded by parentheses, state, etc.
1005 char stat[2048];
1006 int statlen;
1007
1008 fp = fopen("/proc/self/stat", "r");
1009 if (fp) {
1010 statlen = fread(stat, 1, 2047, fp);
1011 stat[statlen] = '\0';
1012 fclose(fp);
1013
1014 // Skip pid and the command string. Note that we could be dealing with
1015 // weird command names, e.g. user could decide to rename java launcher
1016 // to "java 1.4.2 :)", then the stat file would look like
1017 // 1234 (java 1.4.2 :)) R ... ...
1018 // We don't really need to know the command string, just find the last
1019 // occurrence of ")" and then start parsing from there. See bug 4726580.
1020 char * s = strrchr(stat, ')');
1021
1022 i = 0;
1023 if (s) {
1024 // Skip blank chars
1025 do { s++; } while (s && isspace(*s));
1026
1027 #define _UFM UINTX_FORMAT
1028 #define _DFM INTX_FORMAT
1029
1030 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1031 // 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
1032 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,
1033 &state, // 3 %c
1034 &ppid, // 4 %d
1035 &pgrp, // 5 %d
1036 &session, // 6 %d
1037 &nr, // 7 %d
1038 &tpgrp, // 8 %d
1039 &flags, // 9 %lu
1040 &minflt, // 10 %lu
1041 &cminflt, // 11 %lu
1042 &majflt, // 12 %lu
1043 &cmajflt, // 13 %lu
1044 &utime, // 14 %lu
1045 &stime, // 15 %lu
1046 &cutime, // 16 %ld
1047 &cstime, // 17 %ld
1048 &prio, // 18 %ld
1049 &nice, // 19 %ld
1050 &junk, // 20 %ld
1051 &it_real, // 21 %ld
1052 &start, // 22 UINTX_FORMAT
1053 &vsize, // 23 UINTX_FORMAT
1054 &rss, // 24 INTX_FORMAT
1055 &rsslim, // 25 UINTX_FORMAT
1056 &scodes, // 26 UINTX_FORMAT
1057 &ecode, // 27 UINTX_FORMAT
1058 &stack_start); // 28 UINTX_FORMAT
1059 }
1060
1061 #undef _UFM
1062 #undef _DFM
1063
1064 if (i != 28 - 2) {
1065 assert(false, "Bad conversion from /proc/self/stat");
1066 // product mode - assume we are the initial thread, good luck in the
1067 // embedded case.
1068 warning("Can't detect initial thread stack location - bad conversion");
1069 stack_start = (uintptr_t) &rlim;
1070 }
1071 } else {
1072 // For some reason we can't open /proc/self/stat (for example, running on
1073 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1074 // most cases, so don't abort:
1075 warning("Can't detect initial thread stack location - no /proc/self/stat");
1076 stack_start = (uintptr_t) &rlim;
1077 }
1078 }
1079
1080 // Now we have a pointer (stack_start) very close to the stack top, the
1081 // next thing to do is to figure out the exact location of stack top. We
1082 // can find out the virtual memory area that contains stack_start by
1083 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1084 // and its upper limit is the real stack top. (again, this would fail if
1085 // running inside chroot, because /proc may not exist.)
1086
1087 uintptr_t stack_top;
1088 address low, high;
1089 if (find_vma((address)stack_start, &low, &high)) {
1090 // success, "high" is the true stack top. (ignore "low", because initial
1091 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1092 stack_top = (uintptr_t)high;
1093 } else {
1094 // failed, likely because /proc/self/maps does not exist
1095 warning("Can't detect initial thread stack location - find_vma failed");
1096 // best effort: stack_start is normally within a few pages below the real
1097 // stack top, use it as stack top, and reduce stack size so we won't put
1098 // guard page outside stack.
1099 stack_top = stack_start;
1100 stack_size -= 16 * page_size();
1101 }
1102
1103 // stack_top could be partially down the page so align it
1104 stack_top = align_up(stack_top, page_size());
1105
1106 // Allowed stack value is minimum of max_size and what we derived from rlimit
1107 if (max_size > 0) {
1108 _initial_thread_stack_size = MIN2(max_size, stack_size);
1109 } else {
1110 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1111 // clamp it at 8MB as we do on Solaris
1112 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1113 }
1114 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1115 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1116
1117 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1118
1119 if (log_is_enabled(Info, os, thread)) {
1120 // See if we seem to be on primordial process thread
1121 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1122 uintptr_t(&rlim) < stack_top;
1123
1124 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1125 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1126 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1127 stack_top, intptr_t(_initial_thread_stack_bottom));
1128 }
1129 }
1130
1131 ////////////////////////////////////////////////////////////////////////////////
1132 // time support
1133
1134 // Time since start-up in seconds to a fine granularity.
1135 // Used by VMSelfDestructTimer and the MemProfiler.
1136 double os::elapsedTime() {
1137
1138 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1139 }
1140
1141 jlong os::elapsed_counter() {
1142 return javaTimeNanos() - initial_time_count;
1143 }
1144
1145 jlong os::elapsed_frequency() {
1146 return NANOSECS_PER_SEC; // nanosecond resolution
1147 }
1148
1149 bool os::supports_vtime() { return true; }
1150 bool os::enable_vtime() { return false; }
1151 bool os::vtime_enabled() { return false; }
1152
1153 double os::elapsedVTime() {
1154 struct rusage usage;
1155 int retval = getrusage(RUSAGE_THREAD, &usage);
1156 if (retval == 0) {
1157 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);
1158 } else {
1159 // better than nothing, but not much
1160 return elapsedTime();
1161 }
1162 }
1163
1164 jlong os::javaTimeMillis() {
1165 timeval time;
1166 int status = gettimeofday(&time, NULL);
1167 assert(status != -1, "linux error");
1168 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1169 }
1170
1171 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1172 timeval time;
1173 int status = gettimeofday(&time, NULL);
1174 assert(status != -1, "linux error");
1175 seconds = jlong(time.tv_sec);
1176 nanos = jlong(time.tv_usec) * 1000;
1177 }
1178
1179
1180 #ifndef CLOCK_MONOTONIC
1181 #define CLOCK_MONOTONIC (1)
1182 #endif
1183
1184 void os::Linux::clock_init() {
1185 // we do dlopen's in this particular order due to bug in linux
1186 // dynamical loader (see 6348968) leading to crash on exit
1187 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1188 if (handle == NULL) {
1189 handle = dlopen("librt.so", RTLD_LAZY);
1190 }
1191
1192 if (handle) {
1193 int (*clock_getres_func)(clockid_t, struct timespec*) =
1194 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1195 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1196 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1197 if (clock_getres_func && clock_gettime_func) {
1198 // See if monotonic clock is supported by the kernel. Note that some
1199 // early implementations simply return kernel jiffies (updated every
1200 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1201 // for nano time (though the monotonic property is still nice to have).
1202 // It's fixed in newer kernels, however clock_getres() still returns
1203 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1204 // resolution for now. Hopefully as people move to new kernels, this
1205 // won't be a problem.
1206 struct timespec res;
1207 struct timespec tp;
1208 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1209 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1210 // yes, monotonic clock is supported
1211 _clock_gettime = clock_gettime_func;
1212 return;
1213 } else {
1214 // close librt if there is no monotonic clock
1215 dlclose(handle);
1216 }
1217 }
1218 }
1219 warning("No monotonic clock was available - timed services may " \
1220 "be adversely affected if the time-of-day clock changes");
1221 }
1222
1223 #ifndef SYS_clock_getres
1224 #if defined(X86) || defined(PPC64) || defined(S390)
1225 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1226 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1227 #else
1228 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1229 #define sys_clock_getres(x,y) -1
1230 #endif
1231 #else
1232 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1233 #endif
1234
1235 void os::Linux::fast_thread_clock_init() {
1236 if (!UseLinuxPosixThreadCPUClocks) {
1237 return;
1238 }
1239 clockid_t clockid;
1240 struct timespec tp;
1241 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1242 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1243
1244 // Switch to using fast clocks for thread cpu time if
1245 // the sys_clock_getres() returns 0 error code.
1246 // Note, that some kernels may support the current thread
1247 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1248 // returned by the pthread_getcpuclockid().
1249 // If the fast Posix clocks are supported then the sys_clock_getres()
1250 // must return at least tp.tv_sec == 0 which means a resolution
1251 // better than 1 sec. This is extra check for reliability.
1252
1253 if (pthread_getcpuclockid_func &&
1254 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1255 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1256 _supports_fast_thread_cpu_time = true;
1257 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1258 }
1259 }
1260
1261 jlong os::javaTimeNanos() {
1262 if (os::supports_monotonic_clock()) {
1263 struct timespec tp;
1264 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1265 assert(status == 0, "gettime error");
1266 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1267 return result;
1268 } else {
1269 timeval time;
1270 int status = gettimeofday(&time, NULL);
1271 assert(status != -1, "linux error");
1272 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1273 return 1000 * usecs;
1274 }
1275 }
1276
1277 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1278 if (os::supports_monotonic_clock()) {
1279 info_ptr->max_value = ALL_64_BITS;
1280
1281 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1282 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1283 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1284 } else {
1285 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1286 info_ptr->max_value = ALL_64_BITS;
1287
1288 // gettimeofday is a real time clock so it skips
1289 info_ptr->may_skip_backward = true;
1290 info_ptr->may_skip_forward = true;
1291 }
1292
1293 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1294 }
1295
1296 // Return the real, user, and system times in seconds from an
1297 // arbitrary fixed point in the past.
1298 bool os::getTimesSecs(double* process_real_time,
1299 double* process_user_time,
1300 double* process_system_time) {
1301 struct tms ticks;
1302 clock_t real_ticks = times(&ticks);
1303
1304 if (real_ticks == (clock_t) (-1)) {
1305 return false;
1306 } else {
1307 double ticks_per_second = (double) clock_tics_per_sec;
1308 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1309 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1310 *process_real_time = ((double) real_ticks) / ticks_per_second;
1311
1312 return true;
1313 }
1314 }
1315
1316
1317 char * os::local_time_string(char *buf, size_t buflen) {
1318 struct tm t;
1319 time_t long_time;
1320 time(&long_time);
1321 localtime_r(&long_time, &t);
1322 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1323 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1324 t.tm_hour, t.tm_min, t.tm_sec);
1325 return buf;
1326 }
1327
1328 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1329 return localtime_r(clock, res);
1330 }
1331
1332 ////////////////////////////////////////////////////////////////////////////////
1333 // runtime exit support
1334
1335 // Note: os::shutdown() might be called very early during initialization, or
1336 // called from signal handler. Before adding something to os::shutdown(), make
1337 // sure it is async-safe and can handle partially initialized VM.
1338 void os::shutdown() {
1339
1340 // allow PerfMemory to attempt cleanup of any persistent resources
1341 perfMemory_exit();
1342
1343 // needs to remove object in file system
1344 AttachListener::abort();
1345
1346 // flush buffered output, finish log files
1347 ostream_abort();
1348
1349 // Check for abort hook
1350 abort_hook_t abort_hook = Arguments::abort_hook();
1351 if (abort_hook != NULL) {
1352 abort_hook();
1353 }
1354
1355 }
1356
1357 // Note: os::abort() might be called very early during initialization, or
1358 // called from signal handler. Before adding something to os::abort(), make
1359 // sure it is async-safe and can handle partially initialized VM.
1360 void os::abort(bool dump_core, void* siginfo, const void* context) {
1361 os::shutdown();
1362 if (dump_core) {
1363 #ifndef PRODUCT
1364 fdStream out(defaultStream::output_fd());
1365 out.print_raw("Current thread is ");
1366 char buf[16];
1367 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1368 out.print_raw_cr(buf);
1369 out.print_raw_cr("Dumping core ...");
1370 #endif
1371 ::abort(); // dump core
1372 }
1373
1374 ::exit(1);
1375 }
1376
1377 // Die immediately, no exit hook, no abort hook, no cleanup.
1378 void os::die() {
1379 ::abort();
1380 }
1381
1382
1383 // This method is a copy of JDK's sysGetLastErrorString
1384 // from src/solaris/hpi/src/system_md.c
1385
1386 size_t os::lasterror(char *buf, size_t len) {
1387 if (errno == 0) return 0;
1388
1389 const char *s = os::strerror(errno);
1390 size_t n = ::strlen(s);
1391 if (n >= len) {
1392 n = len - 1;
1393 }
1394 ::strncpy(buf, s, n);
1395 buf[n] = '\0';
1396 return n;
1397 }
1398
1399 // thread_id is kernel thread id (similar to Solaris LWP id)
1400 intx os::current_thread_id() { return os::Linux::gettid(); }
1401 int os::current_process_id() {
1402 return ::getpid();
1403 }
1404
1405 // DLL functions
1406
1407 const char* os::dll_file_extension() { return ".so"; }
1408
1409 // This must be hard coded because it's the system's temporary
1410 // directory not the java application's temp directory, ala java.io.tmpdir.
1411 const char* os::get_temp_directory() { return "/tmp"; }
1412
1413 static bool file_exists(const char* filename) {
1414 struct stat statbuf;
1415 if (filename == NULL || strlen(filename) == 0) {
1416 return false;
1417 }
1418 return os::stat(filename, &statbuf) == 0;
1419 }
1420
1421 // check if addr is inside libjvm.so
1422 bool os::address_is_in_vm(address addr) {
1423 static address libjvm_base_addr;
1424 Dl_info dlinfo;
1425
1426 if (libjvm_base_addr == NULL) {
1427 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1428 libjvm_base_addr = (address)dlinfo.dli_fbase;
1429 }
1430 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1431 }
1432
1433 if (dladdr((void *)addr, &dlinfo) != 0) {
1434 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1435 }
1436
1437 return false;
1438 }
1439
1440 bool os::dll_address_to_function_name(address addr, char *buf,
1441 int buflen, int *offset,
1442 bool demangle) {
1443 // buf is not optional, but offset is optional
1444 assert(buf != NULL, "sanity check");
1445
1446 Dl_info dlinfo;
1447
1448 if (dladdr((void*)addr, &dlinfo) != 0) {
1449 // see if we have a matching symbol
1450 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1451 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1452 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1453 }
1454 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1455 return true;
1456 }
1457 // no matching symbol so try for just file info
1458 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1459 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1460 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1461 return true;
1462 }
1463 }
1464 }
1465
1466 buf[0] = '\0';
1467 if (offset != NULL) *offset = -1;
1468 return false;
1469 }
1470
1471 struct _address_to_library_name {
1472 address addr; // input : memory address
1473 size_t buflen; // size of fname
1474 char* fname; // output: library name
1475 address base; // library base addr
1476 };
1477
1478 static int address_to_library_name_callback(struct dl_phdr_info *info,
1479 size_t size, void *data) {
1480 int i;
1481 bool found = false;
1482 address libbase = NULL;
1483 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1484
1485 // iterate through all loadable segments
1486 for (i = 0; i < info->dlpi_phnum; i++) {
1487 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1488 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1489 // base address of a library is the lowest address of its loaded
1490 // segments.
1491 if (libbase == NULL || libbase > segbase) {
1492 libbase = segbase;
1493 }
1494 // see if 'addr' is within current segment
1495 if (segbase <= d->addr &&
1496 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1497 found = true;
1498 }
1499 }
1500 }
1501
1502 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1503 // so dll_address_to_library_name() can fall through to use dladdr() which
1504 // can figure out executable name from argv[0].
1505 if (found && info->dlpi_name && info->dlpi_name[0]) {
1506 d->base = libbase;
1507 if (d->fname) {
1508 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1509 }
1510 return 1;
1511 }
1512 return 0;
1513 }
1514
1515 bool os::dll_address_to_library_name(address addr, char* buf,
1516 int buflen, int* offset) {
1517 // buf is not optional, but offset is optional
1518 assert(buf != NULL, "sanity check");
1519
1520 Dl_info dlinfo;
1521 struct _address_to_library_name data;
1522
1523 // There is a bug in old glibc dladdr() implementation that it could resolve
1524 // to wrong library name if the .so file has a base address != NULL. Here
1525 // we iterate through the program headers of all loaded libraries to find
1526 // out which library 'addr' really belongs to. This workaround can be
1527 // removed once the minimum requirement for glibc is moved to 2.3.x.
1528 data.addr = addr;
1529 data.fname = buf;
1530 data.buflen = buflen;
1531 data.base = NULL;
1532 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1533
1534 if (rslt) {
1535 // buf already contains library name
1536 if (offset) *offset = addr - data.base;
1537 return true;
1538 }
1539 if (dladdr((void*)addr, &dlinfo) != 0) {
1540 if (dlinfo.dli_fname != NULL) {
1541 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1542 }
1543 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1544 *offset = addr - (address)dlinfo.dli_fbase;
1545 }
1546 return true;
1547 }
1548
1549 buf[0] = '\0';
1550 if (offset) *offset = -1;
1551 return false;
1552 }
1553
1554 // Loads .dll/.so and
1555 // in case of error it checks if .dll/.so was built for the
1556 // same architecture as Hotspot is running on
1557
1558
1559 // Remember the stack's state. The Linux dynamic linker will change
1560 // the stack to 'executable' at most once, so we must safepoint only once.
1561 bool os::Linux::_stack_is_executable = false;
1562
1563 // VM operation that loads a library. This is necessary if stack protection
1564 // of the Java stacks can be lost during loading the library. If we
1565 // do not stop the Java threads, they can stack overflow before the stacks
1566 // are protected again.
1567 class VM_LinuxDllLoad: public VM_Operation {
1568 private:
1569 const char *_filename;
1570 char *_ebuf;
1571 int _ebuflen;
1572 void *_lib;
1573 public:
1574 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1575 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1576 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1577 void doit() {
1578 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1579 os::Linux::_stack_is_executable = true;
1580 }
1581 void* loaded_library() { return _lib; }
1582 };
1583
1584 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1585 void * result = NULL;
1586 bool load_attempted = false;
1587
1588 // Check whether the library to load might change execution rights
1589 // of the stack. If they are changed, the protection of the stack
1590 // guard pages will be lost. We need a safepoint to fix this.
1591 //
1592 // See Linux man page execstack(8) for more info.
1593 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1594 if (!ElfFile::specifies_noexecstack(filename)) {
1595 if (!is_init_completed()) {
1596 os::Linux::_stack_is_executable = true;
1597 // This is OK - No Java threads have been created yet, and hence no
1598 // stack guard pages to fix.
1599 //
1600 // This should happen only when you are building JDK7 using a very
1601 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1602 //
1603 // Dynamic loader will make all stacks executable after
1604 // this function returns, and will not do that again.
1605 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1606 } else {
1607 warning("You have loaded library %s which might have disabled stack guard. "
1608 "The VM will try to fix the stack guard now.\n"
1609 "It's highly recommended that you fix the library with "
1610 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1611 filename);
1612
1613 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1614 JavaThread *jt = JavaThread::current();
1615 if (jt->thread_state() != _thread_in_native) {
1616 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1617 // that requires ExecStack. Cannot enter safe point. Let's give up.
1618 warning("Unable to fix stack guard. Giving up.");
1619 } else {
1620 if (!LoadExecStackDllInVMThread) {
1621 // This is for the case where the DLL has an static
1622 // constructor function that executes JNI code. We cannot
1623 // load such DLLs in the VMThread.
1624 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1625 }
1626
1627 ThreadInVMfromNative tiv(jt);
1628 debug_only(VMNativeEntryWrapper vew;)
1629
1630 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1631 VMThread::execute(&op);
1632 if (LoadExecStackDllInVMThread) {
1633 result = op.loaded_library();
1634 }
1635 load_attempted = true;
1636 }
1637 }
1638 }
1639 }
1640
1641 if (!load_attempted) {
1642 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1643 }
1644
1645 if (result != NULL) {
1646 // Successful loading
1647 return result;
1648 }
1649
1650 Elf32_Ehdr elf_head;
1651 int diag_msg_max_length=ebuflen-strlen(ebuf);
1652 char* diag_msg_buf=ebuf+strlen(ebuf);
1653
1654 if (diag_msg_max_length==0) {
1655 // No more space in ebuf for additional diagnostics message
1656 return NULL;
1657 }
1658
1659
1660 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1661
1662 if (file_descriptor < 0) {
1663 // Can't open library, report dlerror() message
1664 return NULL;
1665 }
1666
1667 bool failed_to_read_elf_head=
1668 (sizeof(elf_head)!=
1669 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1670
1671 ::close(file_descriptor);
1672 if (failed_to_read_elf_head) {
1673 // file i/o error - report dlerror() msg
1674 return NULL;
1675 }
1676
1677 typedef struct {
1678 Elf32_Half code; // Actual value as defined in elf.h
1679 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1680 unsigned char elf_class; // 32 or 64 bit
1681 unsigned char endianess; // MSB or LSB
1682 char* name; // String representation
1683 } arch_t;
1684
1685 #ifndef EM_486
1686 #define EM_486 6 /* Intel 80486 */
1687 #endif
1688 #ifndef EM_AARCH64
1689 #define EM_AARCH64 183 /* ARM AARCH64 */
1690 #endif
1691
1692 static const arch_t arch_array[]={
1693 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1694 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1695 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1696 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1697 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1698 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1699 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1700 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1701 #if defined(VM_LITTLE_ENDIAN)
1702 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1703 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1704 #else
1705 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1706 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1707 #endif
1708 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1709 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1710 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1711 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1712 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1713 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1714 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1715 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1716 };
1717
1718 #if (defined IA32)
1719 static Elf32_Half running_arch_code=EM_386;
1720 #elif (defined AMD64)
1721 static Elf32_Half running_arch_code=EM_X86_64;
1722 #elif (defined IA64)
1723 static Elf32_Half running_arch_code=EM_IA_64;
1724 #elif (defined __sparc) && (defined _LP64)
1725 static Elf32_Half running_arch_code=EM_SPARCV9;
1726 #elif (defined __sparc) && (!defined _LP64)
1727 static Elf32_Half running_arch_code=EM_SPARC;
1728 #elif (defined __powerpc64__)
1729 static Elf32_Half running_arch_code=EM_PPC64;
1730 #elif (defined __powerpc__)
1731 static Elf32_Half running_arch_code=EM_PPC;
1732 #elif (defined AARCH64)
1733 static Elf32_Half running_arch_code=EM_AARCH64;
1734 #elif (defined ARM)
1735 static Elf32_Half running_arch_code=EM_ARM;
1736 #elif (defined S390)
1737 static Elf32_Half running_arch_code=EM_S390;
1738 #elif (defined ALPHA)
1739 static Elf32_Half running_arch_code=EM_ALPHA;
1740 #elif (defined MIPSEL)
1741 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1742 #elif (defined PARISC)
1743 static Elf32_Half running_arch_code=EM_PARISC;
1744 #elif (defined MIPS)
1745 static Elf32_Half running_arch_code=EM_MIPS;
1746 #elif (defined M68K)
1747 static Elf32_Half running_arch_code=EM_68K;
1748 #elif (defined SH)
1749 static Elf32_Half running_arch_code=EM_SH;
1750 #else
1751 #error Method os::dll_load requires that one of following is defined:\
1752 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1753 #endif
1754
1755 // Identify compatability class for VM's architecture and library's architecture
1756 // Obtain string descriptions for architectures
1757
1758 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1759 int running_arch_index=-1;
1760
1761 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1762 if (running_arch_code == arch_array[i].code) {
1763 running_arch_index = i;
1764 }
1765 if (lib_arch.code == arch_array[i].code) {
1766 lib_arch.compat_class = arch_array[i].compat_class;
1767 lib_arch.name = arch_array[i].name;
1768 }
1769 }
1770
1771 assert(running_arch_index != -1,
1772 "Didn't find running architecture code (running_arch_code) in arch_array");
1773 if (running_arch_index == -1) {
1774 // Even though running architecture detection failed
1775 // we may still continue with reporting dlerror() message
1776 return NULL;
1777 }
1778
1779 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1780 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1781 return NULL;
1782 }
1783
1784 #ifndef S390
1785 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1786 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1787 return NULL;
1788 }
1789 #endif // !S390
1790
1791 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1792 if (lib_arch.name!=NULL) {
1793 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1794 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1795 lib_arch.name, arch_array[running_arch_index].name);
1796 } else {
1797 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1798 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1799 lib_arch.code,
1800 arch_array[running_arch_index].name);
1801 }
1802 }
1803
1804 return NULL;
1805 }
1806
1807 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1808 int ebuflen) {
1809 void * result = ::dlopen(filename, RTLD_LAZY);
1810 if (result == NULL) {
1811 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1812 ebuf[ebuflen-1] = '\0';
1813 }
1814 return result;
1815 }
1816
1817 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1818 int ebuflen) {
1819 void * result = NULL;
1820 if (LoadExecStackDllInVMThread) {
1821 result = dlopen_helper(filename, ebuf, ebuflen);
1822 }
1823
1824 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1825 // library that requires an executable stack, or which does not have this
1826 // stack attribute set, dlopen changes the stack attribute to executable. The
1827 // read protection of the guard pages gets lost.
1828 //
1829 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1830 // may have been queued at the same time.
1831
1832 if (!_stack_is_executable) {
1833 JavaThread *jt = Threads::first();
1834
1835 while (jt) {
1836 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1837 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1838 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1839 warning("Attempt to reguard stack yellow zone failed.");
1840 }
1841 }
1842 jt = jt->next();
1843 }
1844 }
1845
1846 return result;
1847 }
1848
1849 void* os::dll_lookup(void* handle, const char* name) {
1850 void* res = dlsym(handle, name);
1851 return res;
1852 }
1853
1854 void* os::get_default_process_handle() {
1855 return (void*)::dlopen(NULL, RTLD_LAZY);
1856 }
1857
1858 static bool _print_ascii_file(const char* filename, outputStream* st) {
1859 int fd = ::open(filename, O_RDONLY);
1860 if (fd == -1) {
1861 return false;
1862 }
1863
1864 char buf[33];
1865 int bytes;
1866 buf[32] = '\0';
1867 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1868 st->print_raw(buf, bytes);
1869 }
1870
1871 ::close(fd);
1872
1873 return true;
1874 }
1875
1876 void os::print_dll_info(outputStream *st) {
1877 st->print_cr("Dynamic libraries:");
1878
1879 char fname[32];
1880 pid_t pid = os::Linux::gettid();
1881
1882 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1883
1884 if (!_print_ascii_file(fname, st)) {
1885 st->print("Can not get library information for pid = %d\n", pid);
1886 }
1887 }
1888
1889 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1890 FILE *procmapsFile = NULL;
1891
1892 // Open the procfs maps file for the current process
1893 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1894 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1895 char line[PATH_MAX + 100];
1896
1897 // Read line by line from 'file'
1898 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1899 u8 base, top, offset, inode;
1900 char permissions[5];
1901 char device[6];
1902 char name[PATH_MAX + 1];
1903
1904 // Parse fields from line
1905 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1906 &base, &top, permissions, &offset, device, &inode, name);
1907
1908 // Filter by device id '00:00' so that we only get file system mapped files.
1909 if (strcmp(device, "00:00") != 0) {
1910
1911 // Call callback with the fields of interest
1912 if(callback(name, (address)base, (address)top, param)) {
1913 // Oops abort, callback aborted
1914 fclose(procmapsFile);
1915 return 1;
1916 }
1917 }
1918 }
1919 fclose(procmapsFile);
1920 }
1921 return 0;
1922 }
1923
1924 void os::print_os_info_brief(outputStream* st) {
1925 os::Linux::print_distro_info(st);
1926
1927 os::Posix::print_uname_info(st);
1928
1929 os::Linux::print_libversion_info(st);
1930
1931 }
1932
1933 void os::print_os_info(outputStream* st) {
1934 st->print("OS:");
1935
1936 os::Linux::print_distro_info(st);
1937
1938 os::Posix::print_uname_info(st);
1939
1940 // Print warning if unsafe chroot environment detected
1941 if (unsafe_chroot_detected) {
1942 st->print("WARNING!! ");
1943 st->print_cr("%s", unstable_chroot_error);
1944 }
1945
1946 os::Linux::print_libversion_info(st);
1947
1948 os::Posix::print_rlimit_info(st);
1949
1950 os::Posix::print_load_average(st);
1951
1952 os::Linux::print_full_memory_info(st);
1953 }
1954
1955 // Try to identify popular distros.
1956 // Most Linux distributions have a /etc/XXX-release file, which contains
1957 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1958 // file that also contains the OS version string. Some have more than one
1959 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1960 // /etc/redhat-release.), so the order is important.
1961 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1962 // their own specific XXX-release file as well as a redhat-release file.
1963 // Because of this the XXX-release file needs to be searched for before the
1964 // redhat-release file.
1965 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1966 // search for redhat-release / SuSE-release needs to be before lsb-release.
1967 // Since the lsb-release file is the new standard it needs to be searched
1968 // before the older style release files.
1969 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1970 // next to last resort. The os-release file is a new standard that contains
1971 // distribution information and the system-release file seems to be an old
1972 // standard that has been replaced by the lsb-release and os-release files.
1973 // Searching for the debian_version file is the last resort. It contains
1974 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1975 // "Debian " is printed before the contents of the debian_version file.
1976
1977 const char* distro_files[] = {
1978 "/etc/oracle-release",
1979 "/etc/mandriva-release",
1980 "/etc/mandrake-release",
1981 "/etc/sun-release",
1982 "/etc/redhat-release",
1983 "/etc/SuSE-release",
1984 "/etc/lsb-release",
1985 "/etc/turbolinux-release",
1986 "/etc/gentoo-release",
1987 "/etc/ltib-release",
1988 "/etc/angstrom-version",
1989 "/etc/system-release",
1990 "/etc/os-release",
1991 NULL };
1992
1993 void os::Linux::print_distro_info(outputStream* st) {
1994 for (int i = 0;; i++) {
1995 const char* file = distro_files[i];
1996 if (file == NULL) {
1997 break; // done
1998 }
1999 // If file prints, we found it.
2000 if (_print_ascii_file(file, st)) {
2001 return;
2002 }
2003 }
2004
2005 if (file_exists("/etc/debian_version")) {
2006 st->print("Debian ");
2007 _print_ascii_file("/etc/debian_version", st);
2008 } else {
2009 st->print("Linux");
2010 }
2011 st->cr();
2012 }
2013
2014 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2015 char buf[256];
2016 while (fgets(buf, sizeof(buf), fp)) {
2017 // Edit out extra stuff in expected format
2018 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2019 char* ptr = strstr(buf, "\""); // the name is in quotes
2020 if (ptr != NULL) {
2021 ptr++; // go beyond first quote
2022 char* nl = strchr(ptr, '\"');
2023 if (nl != NULL) *nl = '\0';
2024 strncpy(distro, ptr, length);
2025 } else {
2026 ptr = strstr(buf, "=");
2027 ptr++; // go beyond equals then
2028 char* nl = strchr(ptr, '\n');
2029 if (nl != NULL) *nl = '\0';
2030 strncpy(distro, ptr, length);
2031 }
2032 return;
2033 } else if (get_first_line) {
2034 char* nl = strchr(buf, '\n');
2035 if (nl != NULL) *nl = '\0';
2036 strncpy(distro, buf, length);
2037 return;
2038 }
2039 }
2040 // print last line and close
2041 char* nl = strchr(buf, '\n');
2042 if (nl != NULL) *nl = '\0';
2043 strncpy(distro, buf, length);
2044 }
2045
2046 static void parse_os_info(char* distro, size_t length, const char* file) {
2047 FILE* fp = fopen(file, "r");
2048 if (fp != NULL) {
2049 // if suse format, print out first line
2050 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2051 parse_os_info_helper(fp, distro, length, get_first_line);
2052 fclose(fp);
2053 }
2054 }
2055
2056 void os::get_summary_os_info(char* buf, size_t buflen) {
2057 for (int i = 0;; i++) {
2058 const char* file = distro_files[i];
2059 if (file == NULL) {
2060 break; // ran out of distro_files
2061 }
2062 if (file_exists(file)) {
2063 parse_os_info(buf, buflen, file);
2064 return;
2065 }
2066 }
2067 // special case for debian
2068 if (file_exists("/etc/debian_version")) {
2069 strncpy(buf, "Debian ", buflen);
2070 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2071 } else {
2072 strncpy(buf, "Linux", buflen);
2073 }
2074 }
2075
2076 void os::Linux::print_libversion_info(outputStream* st) {
2077 // libc, pthread
2078 st->print("libc:");
2079 st->print("%s ", os::Linux::glibc_version());
2080 st->print("%s ", os::Linux::libpthread_version());
2081 st->cr();
2082 }
2083
2084 void os::Linux::print_full_memory_info(outputStream* st) {
2085 st->print("\n/proc/meminfo:\n");
2086 _print_ascii_file("/proc/meminfo", st);
2087 st->cr();
2088 }
2089
2090 void os::print_memory_info(outputStream* st) {
2091
2092 st->print("Memory:");
2093 st->print(" %dk page", os::vm_page_size()>>10);
2094
2095 // values in struct sysinfo are "unsigned long"
2096 struct sysinfo si;
2097 sysinfo(&si);
2098
2099 st->print(", physical " UINT64_FORMAT "k",
2100 os::physical_memory() >> 10);
2101 st->print("(" UINT64_FORMAT "k free)",
2102 os::available_memory() >> 10);
2103 st->print(", swap " UINT64_FORMAT "k",
2104 ((jlong)si.totalswap * si.mem_unit) >> 10);
2105 st->print("(" UINT64_FORMAT "k free)",
2106 ((jlong)si.freeswap * si.mem_unit) >> 10);
2107 st->cr();
2108 }
2109
2110 // Print the first "model name" line and the first "flags" line
2111 // that we find and nothing more. We assume "model name" comes
2112 // before "flags" so if we find a second "model name", then the
2113 // "flags" field is considered missing.
2114 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2115 #if defined(IA32) || defined(AMD64)
2116 // Other platforms have less repetitive cpuinfo files
2117 FILE *fp = fopen("/proc/cpuinfo", "r");
2118 if (fp) {
2119 while (!feof(fp)) {
2120 if (fgets(buf, buflen, fp)) {
2121 // Assume model name comes before flags
2122 bool model_name_printed = false;
2123 if (strstr(buf, "model name") != NULL) {
2124 if (!model_name_printed) {
2125 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2126 st->print_raw(buf);
2127 model_name_printed = true;
2128 } else {
2129 // model name printed but not flags? Odd, just return
2130 fclose(fp);
2131 return true;
2132 }
2133 }
2134 // print the flags line too
2135 if (strstr(buf, "flags") != NULL) {
2136 st->print_raw(buf);
2137 fclose(fp);
2138 return true;
2139 }
2140 }
2141 }
2142 fclose(fp);
2143 }
2144 #endif // x86 platforms
2145 return false;
2146 }
2147
2148 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2149 // Only print the model name if the platform provides this as a summary
2150 if (!print_model_name_and_flags(st, buf, buflen)) {
2151 st->print("\n/proc/cpuinfo:\n");
2152 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2153 st->print_cr(" <Not Available>");
2154 }
2155 }
2156 }
2157
2158 #if defined(AMD64) || defined(IA32) || defined(X32)
2159 const char* search_string = "model name";
2160 #elif defined(M68K)
2161 const char* search_string = "CPU";
2162 #elif defined(PPC64)
2163 const char* search_string = "cpu";
2164 #elif defined(S390)
2165 const char* search_string = "processor";
2166 #elif defined(SPARC)
2167 const char* search_string = "cpu";
2168 #else
2169 const char* search_string = "Processor";
2170 #endif
2171
2172 // Parses the cpuinfo file for string representing the model name.
2173 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2174 FILE* fp = fopen("/proc/cpuinfo", "r");
2175 if (fp != NULL) {
2176 while (!feof(fp)) {
2177 char buf[256];
2178 if (fgets(buf, sizeof(buf), fp)) {
2179 char* start = strstr(buf, search_string);
2180 if (start != NULL) {
2181 char *ptr = start + strlen(search_string);
2182 char *end = buf + strlen(buf);
2183 while (ptr != end) {
2184 // skip whitespace and colon for the rest of the name.
2185 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2186 break;
2187 }
2188 ptr++;
2189 }
2190 if (ptr != end) {
2191 // reasonable string, get rid of newline and keep the rest
2192 char* nl = strchr(buf, '\n');
2193 if (nl != NULL) *nl = '\0';
2194 strncpy(cpuinfo, ptr, length);
2195 fclose(fp);
2196 return;
2197 }
2198 }
2199 }
2200 }
2201 fclose(fp);
2202 }
2203 // cpuinfo not found or parsing failed, just print generic string. The entire
2204 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2205 #if defined(AARCH64)
2206 strncpy(cpuinfo, "AArch64", length);
2207 #elif defined(AMD64)
2208 strncpy(cpuinfo, "x86_64", length);
2209 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2210 strncpy(cpuinfo, "ARM", length);
2211 #elif defined(IA32)
2212 strncpy(cpuinfo, "x86_32", length);
2213 #elif defined(IA64)
2214 strncpy(cpuinfo, "IA64", length);
2215 #elif defined(PPC)
2216 strncpy(cpuinfo, "PPC64", length);
2217 #elif defined(S390)
2218 strncpy(cpuinfo, "S390", length);
2219 #elif defined(SPARC)
2220 strncpy(cpuinfo, "sparcv9", length);
2221 #elif defined(ZERO_LIBARCH)
2222 strncpy(cpuinfo, ZERO_LIBARCH, length);
2223 #else
2224 strncpy(cpuinfo, "unknown", length);
2225 #endif
2226 }
2227
2228 static void print_signal_handler(outputStream* st, int sig,
2229 char* buf, size_t buflen);
2230
2231 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2232 st->print_cr("Signal Handlers:");
2233 print_signal_handler(st, SIGSEGV, buf, buflen);
2234 print_signal_handler(st, SIGBUS , buf, buflen);
2235 print_signal_handler(st, SIGFPE , buf, buflen);
2236 print_signal_handler(st, SIGPIPE, buf, buflen);
2237 print_signal_handler(st, SIGXFSZ, buf, buflen);
2238 print_signal_handler(st, SIGILL , buf, buflen);
2239 print_signal_handler(st, SR_signum, buf, buflen);
2240 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2241 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2242 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2243 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2244 #if defined(PPC64)
2245 print_signal_handler(st, SIGTRAP, buf, buflen);
2246 #endif
2247 }
2248
2249 static char saved_jvm_path[MAXPATHLEN] = {0};
2250
2251 // Find the full path to the current module, libjvm.so
2252 void os::jvm_path(char *buf, jint buflen) {
2253 // Error checking.
2254 if (buflen < MAXPATHLEN) {
2255 assert(false, "must use a large-enough buffer");
2256 buf[0] = '\0';
2257 return;
2258 }
2259 // Lazy resolve the path to current module.
2260 if (saved_jvm_path[0] != 0) {
2261 strcpy(buf, saved_jvm_path);
2262 return;
2263 }
2264
2265 char dli_fname[MAXPATHLEN];
2266 bool ret = dll_address_to_library_name(
2267 CAST_FROM_FN_PTR(address, os::jvm_path),
2268 dli_fname, sizeof(dli_fname), NULL);
2269 assert(ret, "cannot locate libjvm");
2270 char *rp = NULL;
2271 if (ret && dli_fname[0] != '\0') {
2272 rp = os::Posix::realpath(dli_fname, buf, buflen);
2273 }
2274 if (rp == NULL) {
2275 return;
2276 }
2277
2278 if (Arguments::sun_java_launcher_is_altjvm()) {
2279 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2280 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2281 // If "/jre/lib/" appears at the right place in the string, then
2282 // assume we are installed in a JDK and we're done. Otherwise, check
2283 // for a JAVA_HOME environment variable and fix up the path so it
2284 // looks like libjvm.so is installed there (append a fake suffix
2285 // hotspot/libjvm.so).
2286 const char *p = buf + strlen(buf) - 1;
2287 for (int count = 0; p > buf && count < 5; ++count) {
2288 for (--p; p > buf && *p != '/'; --p)
2289 /* empty */ ;
2290 }
2291
2292 if (strncmp(p, "/jre/lib/", 9) != 0) {
2293 // Look for JAVA_HOME in the environment.
2294 char* java_home_var = ::getenv("JAVA_HOME");
2295 if (java_home_var != NULL && java_home_var[0] != 0) {
2296 char* jrelib_p;
2297 int len;
2298
2299 // Check the current module name "libjvm.so".
2300 p = strrchr(buf, '/');
2301 if (p == NULL) {
2302 return;
2303 }
2304 assert(strstr(p, "/libjvm") == p, "invalid library name");
2305
2306 rp = os::Posix::realpath(java_home_var, buf, buflen);
2307 if (rp == NULL) {
2308 return;
2309 }
2310
2311 // determine if this is a legacy image or modules image
2312 // modules image doesn't have "jre" subdirectory
2313 len = strlen(buf);
2314 assert(len < buflen, "Ran out of buffer room");
2315 jrelib_p = buf + len;
2316 snprintf(jrelib_p, buflen-len, "/jre/lib");
2317 if (0 != access(buf, F_OK)) {
2318 snprintf(jrelib_p, buflen-len, "/lib");
2319 }
2320
2321 if (0 == access(buf, F_OK)) {
2322 // Use current module name "libjvm.so"
2323 len = strlen(buf);
2324 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2325 } else {
2326 // Go back to path of .so
2327 rp = os::Posix::realpath(dli_fname, buf, buflen);
2328 if (rp == NULL) {
2329 return;
2330 }
2331 }
2332 }
2333 }
2334 }
2335
2336 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2337 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2338 }
2339
2340 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2341 // no prefix required, not even "_"
2342 }
2343
2344 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2345 // no suffix required
2346 }
2347
2348 ////////////////////////////////////////////////////////////////////////////////
2349 // sun.misc.Signal support
2350
2351 static volatile jint sigint_count = 0;
2352
2353 static void UserHandler(int sig, void *siginfo, void *context) {
2354 // 4511530 - sem_post is serialized and handled by the manager thread. When
2355 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2356 // don't want to flood the manager thread with sem_post requests.
2357 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2358 return;
2359 }
2360
2361 // Ctrl-C is pressed during error reporting, likely because the error
2362 // handler fails to abort. Let VM die immediately.
2363 if (sig == SIGINT && VMError::is_error_reported()) {
2364 os::die();
2365 }
2366
2367 os::signal_notify(sig);
2368 }
2369
2370 void* os::user_handler() {
2371 return CAST_FROM_FN_PTR(void*, UserHandler);
2372 }
2373
2374 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2375 struct timespec ts;
2376 // Semaphore's are always associated with CLOCK_REALTIME
2377 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2378 // see unpackTime for discussion on overflow checking
2379 if (sec >= MAX_SECS) {
2380 ts.tv_sec += MAX_SECS;
2381 ts.tv_nsec = 0;
2382 } else {
2383 ts.tv_sec += sec;
2384 ts.tv_nsec += nsec;
2385 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2386 ts.tv_nsec -= NANOSECS_PER_SEC;
2387 ++ts.tv_sec; // note: this must be <= max_secs
2388 }
2389 }
2390
2391 return ts;
2392 }
2393
2394 extern "C" {
2395 typedef void (*sa_handler_t)(int);
2396 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2397 }
2398
2399 void* os::signal(int signal_number, void* handler) {
2400 struct sigaction sigAct, oldSigAct;
2401
2402 sigfillset(&(sigAct.sa_mask));
2403 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2404 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2405
2406 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2407 // -1 means registration failed
2408 return (void *)-1;
2409 }
2410
2411 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2412 }
2413
2414 void os::signal_raise(int signal_number) {
2415 ::raise(signal_number);
2416 }
2417
2418 // The following code is moved from os.cpp for making this
2419 // code platform specific, which it is by its very nature.
2420
2421 // Will be modified when max signal is changed to be dynamic
2422 int os::sigexitnum_pd() {
2423 return NSIG;
2424 }
2425
2426 // a counter for each possible signal value
2427 static volatile jint pending_signals[NSIG+1] = { 0 };
2428
2429 // Linux(POSIX) specific hand shaking semaphore.
2430 static sem_t sig_sem;
2431 static PosixSemaphore sr_semaphore;
2432
2433 void os::signal_init_pd() {
2434 // Initialize signal structures
2435 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2436
2437 // Initialize signal semaphore
2438 ::sem_init(&sig_sem, 0, 0);
2439 }
2440
2441 void os::signal_notify(int sig) {
2442 Atomic::inc(&pending_signals[sig]);
2443 ::sem_post(&sig_sem);
2444 }
2445
2446 static int check_pending_signals(bool wait) {
2447 Atomic::store(0, &sigint_count);
2448 for (;;) {
2449 for (int i = 0; i < NSIG + 1; i++) {
2450 jint n = pending_signals[i];
2451 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2452 return i;
2453 }
2454 }
2455 if (!wait) {
2456 return -1;
2457 }
2458 JavaThread *thread = JavaThread::current();
2459 ThreadBlockInVM tbivm(thread);
2460
2461 bool threadIsSuspended;
2462 do {
2463 thread->set_suspend_equivalent();
2464 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2465 ::sem_wait(&sig_sem);
2466
2467 // were we externally suspended while we were waiting?
2468 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2469 if (threadIsSuspended) {
2470 // The semaphore has been incremented, but while we were waiting
2471 // another thread suspended us. We don't want to continue running
2472 // while suspended because that would surprise the thread that
2473 // suspended us.
2474 ::sem_post(&sig_sem);
2475
2476 thread->java_suspend_self();
2477 }
2478 } while (threadIsSuspended);
2479 }
2480 }
2481
2482 int os::signal_lookup() {
2483 return check_pending_signals(false);
2484 }
2485
2486 int os::signal_wait() {
2487 return check_pending_signals(true);
2488 }
2489
2490 ////////////////////////////////////////////////////////////////////////////////
2491 // Virtual Memory
2492
2493 int os::vm_page_size() {
2494 // Seems redundant as all get out
2495 assert(os::Linux::page_size() != -1, "must call os::init");
2496 return os::Linux::page_size();
2497 }
2498
2499 // Solaris allocates memory by pages.
2500 int os::vm_allocation_granularity() {
2501 assert(os::Linux::page_size() != -1, "must call os::init");
2502 return os::Linux::page_size();
2503 }
2504
2505 // Rationale behind this function:
2506 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2507 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2508 // samples for JITted code. Here we create private executable mapping over the code cache
2509 // and then we can use standard (well, almost, as mapping can change) way to provide
2510 // info for the reporting script by storing timestamp and location of symbol
2511 void linux_wrap_code(char* base, size_t size) {
2512 static volatile jint cnt = 0;
2513
2514 if (!UseOprofile) {
2515 return;
2516 }
2517
2518 char buf[PATH_MAX+1];
2519 int num = Atomic::add(1, &cnt);
2520
2521 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2522 os::get_temp_directory(), os::current_process_id(), num);
2523 unlink(buf);
2524
2525 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2526
2527 if (fd != -1) {
2528 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2529 if (rv != (off_t)-1) {
2530 if (::write(fd, "", 1) == 1) {
2531 mmap(base, size,
2532 PROT_READ|PROT_WRITE|PROT_EXEC,
2533 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2534 }
2535 }
2536 ::close(fd);
2537 unlink(buf);
2538 }
2539 }
2540
2541 static bool recoverable_mmap_error(int err) {
2542 // See if the error is one we can let the caller handle. This
2543 // list of errno values comes from JBS-6843484. I can't find a
2544 // Linux man page that documents this specific set of errno
2545 // values so while this list currently matches Solaris, it may
2546 // change as we gain experience with this failure mode.
2547 switch (err) {
2548 case EBADF:
2549 case EINVAL:
2550 case ENOTSUP:
2551 // let the caller deal with these errors
2552 return true;
2553
2554 default:
2555 // Any remaining errors on this OS can cause our reserved mapping
2556 // to be lost. That can cause confusion where different data
2557 // structures think they have the same memory mapped. The worst
2558 // scenario is if both the VM and a library think they have the
2559 // same memory mapped.
2560 return false;
2561 }
2562 }
2563
2564 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2565 int err) {
2566 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2567 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2568 os::strerror(err), err);
2569 }
2570
2571 static void warn_fail_commit_memory(char* addr, size_t size,
2572 size_t alignment_hint, bool exec,
2573 int err) {
2574 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2575 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2576 alignment_hint, exec, os::strerror(err), err);
2577 }
2578
2579 // NOTE: Linux kernel does not really reserve the pages for us.
2580 // All it does is to check if there are enough free pages
2581 // left at the time of mmap(). This could be a potential
2582 // problem.
2583 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2584 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2585 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2586 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2587 if (res != (uintptr_t) MAP_FAILED) {
2588 if (UseNUMAInterleaving) {
2589 numa_make_global(addr, size);
2590 }
2591 return 0;
2592 }
2593
2594 int err = errno; // save errno from mmap() call above
2595
2596 if (!recoverable_mmap_error(err)) {
2597 warn_fail_commit_memory(addr, size, exec, err);
2598 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2599 }
2600
2601 return err;
2602 }
2603
2604 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2605 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2606 }
2607
2608 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2609 const char* mesg) {
2610 assert(mesg != NULL, "mesg must be specified");
2611 int err = os::Linux::commit_memory_impl(addr, size, exec);
2612 if (err != 0) {
2613 // the caller wants all commit errors to exit with the specified mesg:
2614 warn_fail_commit_memory(addr, size, exec, err);
2615 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2616 }
2617 }
2618
2619 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2620 #ifndef MAP_HUGETLB
2621 #define MAP_HUGETLB 0x40000
2622 #endif
2623
2624 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2625 #ifndef MADV_HUGEPAGE
2626 #define MADV_HUGEPAGE 14
2627 #endif
2628
2629 int os::Linux::commit_memory_impl(char* addr, size_t size,
2630 size_t alignment_hint, bool exec) {
2631 int err = os::Linux::commit_memory_impl(addr, size, exec);
2632 if (err == 0) {
2633 realign_memory(addr, size, alignment_hint);
2634 }
2635 return err;
2636 }
2637
2638 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2639 bool exec) {
2640 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2641 }
2642
2643 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2644 size_t alignment_hint, bool exec,
2645 const char* mesg) {
2646 assert(mesg != NULL, "mesg must be specified");
2647 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2648 if (err != 0) {
2649 // the caller wants all commit errors to exit with the specified mesg:
2650 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2651 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2652 }
2653 }
2654
2655 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2656 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2657 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2658 // be supported or the memory may already be backed by huge pages.
2659 ::madvise(addr, bytes, MADV_HUGEPAGE);
2660 }
2661 }
2662
2663 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2664 // This method works by doing an mmap over an existing mmaping and effectively discarding
2665 // the existing pages. However it won't work for SHM-based large pages that cannot be
2666 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2667 // small pages on top of the SHM segment. This method always works for small pages, so we
2668 // allow that in any case.
2669 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2670 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2671 }
2672 }
2673
2674 void os::numa_make_global(char *addr, size_t bytes) {
2675 Linux::numa_interleave_memory(addr, bytes);
2676 }
2677
2678 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2679 // bind policy to MPOL_PREFERRED for the current thread.
2680 #define USE_MPOL_PREFERRED 0
2681
2682 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2683 // To make NUMA and large pages more robust when both enabled, we need to ease
2684 // the requirements on where the memory should be allocated. MPOL_BIND is the
2685 // default policy and it will force memory to be allocated on the specified
2686 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2687 // the specified node, but will not force it. Using this policy will prevent
2688 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2689 // free large pages.
2690 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2691 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2692 }
2693
2694 bool os::numa_topology_changed() { return false; }
2695
2696 size_t os::numa_get_groups_num() {
2697 // Return just the number of nodes in which it's possible to allocate memory
2698 // (in numa terminology, configured nodes).
2699 return Linux::numa_num_configured_nodes();
2700 }
2701
2702 int os::numa_get_group_id() {
2703 int cpu_id = Linux::sched_getcpu();
2704 if (cpu_id != -1) {
2705 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2706 if (lgrp_id != -1) {
2707 return lgrp_id;
2708 }
2709 }
2710 return 0;
2711 }
2712
2713 int os::Linux::get_existing_num_nodes() {
2714 size_t node;
2715 size_t highest_node_number = Linux::numa_max_node();
2716 int num_nodes = 0;
2717
2718 // Get the total number of nodes in the system including nodes without memory.
2719 for (node = 0; node <= highest_node_number; node++) {
2720 if (isnode_in_existing_nodes(node)) {
2721 num_nodes++;
2722 }
2723 }
2724 return num_nodes;
2725 }
2726
2727 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2728 size_t highest_node_number = Linux::numa_max_node();
2729 size_t i = 0;
2730
2731 // Map all node ids in which is possible to allocate memory. Also nodes are
2732 // not always consecutively available, i.e. available from 0 to the highest
2733 // node number.
2734 for (size_t node = 0; node <= highest_node_number; node++) {
2735 if (Linux::isnode_in_configured_nodes(node)) {
2736 ids[i++] = node;
2737 }
2738 }
2739 return i;
2740 }
2741
2742 bool os::get_page_info(char *start, page_info* info) {
2743 return false;
2744 }
2745
2746 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2747 page_info* page_found) {
2748 return end;
2749 }
2750
2751
2752 int os::Linux::sched_getcpu_syscall(void) {
2753 unsigned int cpu = 0;
2754 int retval = -1;
2755
2756 #if defined(IA32)
2757 #ifndef SYS_getcpu
2758 #define SYS_getcpu 318
2759 #endif
2760 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2761 #elif defined(AMD64)
2762 // Unfortunately we have to bring all these macros here from vsyscall.h
2763 // to be able to compile on old linuxes.
2764 #define __NR_vgetcpu 2
2765 #define VSYSCALL_START (-10UL << 20)
2766 #define VSYSCALL_SIZE 1024
2767 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2768 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2769 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2770 retval = vgetcpu(&cpu, NULL, NULL);
2771 #endif
2772
2773 return (retval == -1) ? retval : cpu;
2774 }
2775
2776 void os::Linux::sched_getcpu_init() {
2777 // sched_getcpu() should be in libc.
2778 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2779 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2780
2781 // If it's not, try a direct syscall.
2782 if (sched_getcpu() == -1) {
2783 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2784 (void*)&sched_getcpu_syscall));
2785 }
2786 }
2787
2788 // Something to do with the numa-aware allocator needs these symbols
2789 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2790 extern "C" JNIEXPORT void numa_error(char *where) { }
2791
2792 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2793 // load symbol from base version instead.
2794 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2795 void *f = dlvsym(handle, name, "libnuma_1.1");
2796 if (f == NULL) {
2797 f = dlsym(handle, name);
2798 }
2799 return f;
2800 }
2801
2802 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2803 // Return NULL if the symbol is not defined in this particular version.
2804 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2805 return dlvsym(handle, name, "libnuma_1.2");
2806 }
2807
2808 bool os::Linux::libnuma_init() {
2809 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2810 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2811 if (handle != NULL) {
2812 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2813 libnuma_dlsym(handle, "numa_node_to_cpus")));
2814 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2815 libnuma_dlsym(handle, "numa_max_node")));
2816 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2817 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2818 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2819 libnuma_dlsym(handle, "numa_available")));
2820 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2821 libnuma_dlsym(handle, "numa_tonode_memory")));
2822 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2823 libnuma_dlsym(handle, "numa_interleave_memory")));
2824 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2825 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2826 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2827 libnuma_dlsym(handle, "numa_set_bind_policy")));
2828 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2829 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2830 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2831 libnuma_dlsym(handle, "numa_distance")));
2832
2833 if (numa_available() != -1) {
2834 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2835 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2836 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2837 // Create an index -> node mapping, since nodes are not always consecutive
2838 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2839 rebuild_nindex_to_node_map();
2840 // Create a cpu -> node mapping
2841 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2842 rebuild_cpu_to_node_map();
2843 return true;
2844 }
2845 }
2846 }
2847 return false;
2848 }
2849
2850 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2851 // Creating guard page is very expensive. Java thread has HotSpot
2852 // guard pages, only enable glibc guard page for non-Java threads.
2853 // (Remember: compiler thread is a Java thread, too!)
2854 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2855 }
2856
2857 void os::Linux::rebuild_nindex_to_node_map() {
2858 int highest_node_number = Linux::numa_max_node();
2859
2860 nindex_to_node()->clear();
2861 for (int node = 0; node <= highest_node_number; node++) {
2862 if (Linux::isnode_in_existing_nodes(node)) {
2863 nindex_to_node()->append(node);
2864 }
2865 }
2866 }
2867
2868 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2869 // The table is later used in get_node_by_cpu().
2870 void os::Linux::rebuild_cpu_to_node_map() {
2871 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2872 // in libnuma (possible values are starting from 16,
2873 // and continuing up with every other power of 2, but less
2874 // than the maximum number of CPUs supported by kernel), and
2875 // is a subject to change (in libnuma version 2 the requirements
2876 // are more reasonable) we'll just hardcode the number they use
2877 // in the library.
2878 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2879
2880 size_t cpu_num = processor_count();
2881 size_t cpu_map_size = NCPUS / BitsPerCLong;
2882 size_t cpu_map_valid_size =
2883 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2884
2885 cpu_to_node()->clear();
2886 cpu_to_node()->at_grow(cpu_num - 1);
2887
2888 size_t node_num = get_existing_num_nodes();
2889
2890 int distance = 0;
2891 int closest_distance = INT_MAX;
2892 int closest_node = 0;
2893 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2894 for (size_t i = 0; i < node_num; i++) {
2895 // Check if node is configured (not a memory-less node). If it is not, find
2896 // the closest configured node.
2897 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2898 closest_distance = INT_MAX;
2899 // Check distance from all remaining nodes in the system. Ignore distance
2900 // from itself and from another non-configured node.
2901 for (size_t m = 0; m < node_num; m++) {
2902 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2903 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2904 // If a closest node is found, update. There is always at least one
2905 // configured node in the system so there is always at least one node
2906 // close.
2907 if (distance != 0 && distance < closest_distance) {
2908 closest_distance = distance;
2909 closest_node = nindex_to_node()->at(m);
2910 }
2911 }
2912 }
2913 } else {
2914 // Current node is already a configured node.
2915 closest_node = nindex_to_node()->at(i);
2916 }
2917
2918 // Get cpus from the original node and map them to the closest node. If node
2919 // is a configured node (not a memory-less node), then original node and
2920 // closest node are the same.
2921 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2922 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2923 if (cpu_map[j] != 0) {
2924 for (size_t k = 0; k < BitsPerCLong; k++) {
2925 if (cpu_map[j] & (1UL << k)) {
2926 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2927 }
2928 }
2929 }
2930 }
2931 }
2932 }
2933 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2934 }
2935
2936 int os::Linux::get_node_by_cpu(int cpu_id) {
2937 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2938 return cpu_to_node()->at(cpu_id);
2939 }
2940 return -1;
2941 }
2942
2943 GrowableArray<int>* os::Linux::_cpu_to_node;
2944 GrowableArray<int>* os::Linux::_nindex_to_node;
2945 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2946 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2947 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2948 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
2949 os::Linux::numa_available_func_t os::Linux::_numa_available;
2950 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2951 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2952 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
2953 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2954 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
2955 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
2956 unsigned long* os::Linux::_numa_all_nodes;
2957 struct bitmask* os::Linux::_numa_all_nodes_ptr;
2958 struct bitmask* os::Linux::_numa_nodes_ptr;
2959
2960 bool os::pd_uncommit_memory(char* addr, size_t size) {
2961 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2962 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2963 return res != (uintptr_t) MAP_FAILED;
2964 }
2965
2966 // If there is no page mapped/committed, top (bottom + size) is returned
2967 static address get_stack_mapped_bottom(address bottom,
2968 size_t size,
2969 bool committed_only /* must have backing pages */) {
2970 // address used to test if the page is mapped/committed
2971 address test_addr = bottom + size;
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 test_addr = bottom + (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(test_addr, page_sz, vec);
2990
2991 if (mincore_return_value == -1 || (committed_only && (vec[0] & 0x01) == 0)) {
2992 // Page is not mapped/committed go up
2993 // to find first mapped/committed page
2994 if (errno != EAGAIN || (committed_only && (vec[0] & 0x01) == 0)) {
2995 assert(mincore_return_value != -1 || errno == ENOMEM, "Unexpected mincore errno");
2996
2997 imin = imid + 1;
2998 }
2999 } else {
3000 // mapped/committed, go down
3001 imax= imid;
3002 }
3003 }
3004
3005 // Adjust stack bottom one page up if last checked page is not mapped/committed
3006 if (mincore_return_value == -1 || (committed_only && (vec[0] & 0x01) == 0)) {
3007 test_addr = test_addr + page_sz;
3008 }
3009
3010 return test_addr;
3011 }
3012
3013 // Linux uses a growable mapping for the stack, and if the mapping for
3014 // the stack guard pages is not removed when we detach a thread the
3015 // stack cannot grow beyond the pages where the stack guard was
3016 // mapped. If at some point later in the process the stack expands to
3017 // that point, the Linux kernel cannot expand the stack any further
3018 // because the guard pages are in the way, and a segfault occurs.
3019 //
3020 // However, it's essential not to split the stack region by unmapping
3021 // a region (leaving a hole) that's already part of the stack mapping,
3022 // so if the stack mapping has already grown beyond the guard pages at
3023 // the time we create them, we have to truncate the stack mapping.
3024 // So, we need to know the extent of the stack mapping when
3025 // create_stack_guard_pages() is called.
3026
3027 // We only need this for stacks that are growable: at the time of
3028 // writing thread stacks don't use growable mappings (i.e. those
3029 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3030 // only applies to the main thread.
3031
3032 // If the (growable) stack mapping already extends beyond the point
3033 // where we're going to put our guard pages, truncate the mapping at
3034 // that point by munmap()ping it. This ensures that when we later
3035 // munmap() the guard pages we don't leave a hole in the stack
3036 // mapping. This only affects the main/initial thread
3037
3038 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3039 if (os::Linux::is_initial_thread()) {
3040 // As we manually grow stack up to bottom inside create_attached_thread(),
3041 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3042 // we don't need to do anything special.
3043 // Check it first, before calling heavy function.
3044 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3045 unsigned char vec[1];
3046
3047 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3048 // Fallback to slow path on all errors, including EAGAIN
3049 stack_extent = (uintptr_t) get_stack_mapped_bottom(os::Linux::initial_thread_stack_bottom(),
3050 (size_t)addr - stack_extent,
3051 false /* committed_only */);
3052 }
3053
3054 if (stack_extent < (uintptr_t)addr) {
3055 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3056 }
3057 }
3058
3059 return os::commit_memory(addr, size, !ExecMem);
3060 }
3061
3062 // If this is a growable mapping, remove the guard pages entirely by
3063 // munmap()ping them. If not, just call uncommit_memory(). This only
3064 // affects the main/initial thread, but guard against future OS changes
3065 // It's safe to always unmap guard pages for initial thread because we
3066 // always place it right after end of the mapped region
3067
3068 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3069 uintptr_t stack_extent, stack_base;
3070
3071 if (os::Linux::is_initial_thread()) {
3072 return ::munmap(addr, size) == 0;
3073 }
3074
3075 return os::uncommit_memory(addr, size);
3076 }
3077
3078 size_t os::pd_committed_stack_size(address bottom, size_t size) {
3079 address bot = get_stack_mapped_bottom(bottom, size, true /* committed_only */);
3080 return size_t(bottom + size - bot);
3081 }
3082
3083 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3084 // at 'requested_addr'. If there are existing memory mappings at the same
3085 // location, however, they will be overwritten. If 'fixed' is false,
3086 // 'requested_addr' is only treated as a hint, the return value may or
3087 // may not start from the requested address. Unlike Linux mmap(), this
3088 // function returns NULL to indicate failure.
3089 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3090 char * addr;
3091 int flags;
3092
3093 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3094 if (fixed) {
3095 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3096 flags |= MAP_FIXED;
3097 }
3098
3099 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3100 // touch an uncommitted page. Otherwise, the read/write might
3101 // succeed if we have enough swap space to back the physical page.
3102 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3103 flags, -1, 0);
3104
3105 return addr == MAP_FAILED ? NULL : addr;
3106 }
3107
3108 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3109 // (req_addr != NULL) or with a given alignment.
3110 // - bytes shall be a multiple of alignment.
3111 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3112 // - alignment sets the alignment at which memory shall be allocated.
3113 // It must be a multiple of allocation granularity.
3114 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3115 // req_addr or NULL.
3116 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3117
3118 size_t extra_size = bytes;
3119 if (req_addr == NULL && alignment > 0) {
3120 extra_size += alignment;
3121 }
3122
3123 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3124 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3125 -1, 0);
3126 if (start == MAP_FAILED) {
3127 start = NULL;
3128 } else {
3129 if (req_addr != NULL) {
3130 if (start != req_addr) {
3131 ::munmap(start, extra_size);
3132 start = NULL;
3133 }
3134 } else {
3135 char* const start_aligned = align_up(start, alignment);
3136 char* const end_aligned = start_aligned + bytes;
3137 char* const end = start + extra_size;
3138 if (start_aligned > start) {
3139 ::munmap(start, start_aligned - start);
3140 }
3141 if (end_aligned < end) {
3142 ::munmap(end_aligned, end - end_aligned);
3143 }
3144 start = start_aligned;
3145 }
3146 }
3147 return start;
3148 }
3149
3150 static int anon_munmap(char * addr, size_t size) {
3151 return ::munmap(addr, size) == 0;
3152 }
3153
3154 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3155 size_t alignment_hint) {
3156 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3157 }
3158
3159 bool os::pd_release_memory(char* addr, size_t size) {
3160 return anon_munmap(addr, size);
3161 }
3162
3163 static bool linux_mprotect(char* addr, size_t size, int prot) {
3164 // Linux wants the mprotect address argument to be page aligned.
3165 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3166
3167 // According to SUSv3, mprotect() should only be used with mappings
3168 // established by mmap(), and mmap() always maps whole pages. Unaligned
3169 // 'addr' likely indicates problem in the VM (e.g. trying to change
3170 // protection of malloc'ed or statically allocated memory). Check the
3171 // caller if you hit this assert.
3172 assert(addr == bottom, "sanity check");
3173
3174 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3175 return ::mprotect(bottom, size, prot) == 0;
3176 }
3177
3178 // Set protections specified
3179 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3180 bool is_committed) {
3181 unsigned int p = 0;
3182 switch (prot) {
3183 case MEM_PROT_NONE: p = PROT_NONE; break;
3184 case MEM_PROT_READ: p = PROT_READ; break;
3185 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3186 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3187 default:
3188 ShouldNotReachHere();
3189 }
3190 // is_committed is unused.
3191 return linux_mprotect(addr, bytes, p);
3192 }
3193
3194 bool os::guard_memory(char* addr, size_t size) {
3195 return linux_mprotect(addr, size, PROT_NONE);
3196 }
3197
3198 bool os::unguard_memory(char* addr, size_t size) {
3199 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3200 }
3201
3202 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3203 size_t page_size) {
3204 bool result = false;
3205 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3206 MAP_ANONYMOUS|MAP_PRIVATE,
3207 -1, 0);
3208 if (p != MAP_FAILED) {
3209 void *aligned_p = align_up(p, page_size);
3210
3211 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3212
3213 munmap(p, page_size * 2);
3214 }
3215
3216 if (warn && !result) {
3217 warning("TransparentHugePages is not supported by the operating system.");
3218 }
3219
3220 return result;
3221 }
3222
3223 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3224 bool result = false;
3225 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3226 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3227 -1, 0);
3228
3229 if (p != MAP_FAILED) {
3230 // We don't know if this really is a huge page or not.
3231 FILE *fp = fopen("/proc/self/maps", "r");
3232 if (fp) {
3233 while (!feof(fp)) {
3234 char chars[257];
3235 long x = 0;
3236 if (fgets(chars, sizeof(chars), fp)) {
3237 if (sscanf(chars, "%lx-%*x", &x) == 1
3238 && x == (long)p) {
3239 if (strstr (chars, "hugepage")) {
3240 result = true;
3241 break;
3242 }
3243 }
3244 }
3245 }
3246 fclose(fp);
3247 }
3248 munmap(p, page_size);
3249 }
3250
3251 if (warn && !result) {
3252 warning("HugeTLBFS is not supported by the operating system.");
3253 }
3254
3255 return result;
3256 }
3257
3258 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3259 //
3260 // From the coredump_filter documentation:
3261 //
3262 // - (bit 0) anonymous private memory
3263 // - (bit 1) anonymous shared memory
3264 // - (bit 2) file-backed private memory
3265 // - (bit 3) file-backed shared memory
3266 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3267 // effective only if the bit 2 is cleared)
3268 // - (bit 5) hugetlb private memory
3269 // - (bit 6) hugetlb shared memory
3270 //
3271 static void set_coredump_filter(void) {
3272 FILE *f;
3273 long cdm;
3274
3275 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3276 return;
3277 }
3278
3279 if (fscanf(f, "%lx", &cdm) != 1) {
3280 fclose(f);
3281 return;
3282 }
3283
3284 rewind(f);
3285
3286 if ((cdm & LARGEPAGES_BIT) == 0) {
3287 cdm |= LARGEPAGES_BIT;
3288 fprintf(f, "%#lx", cdm);
3289 }
3290
3291 fclose(f);
3292 }
3293
3294 // Large page support
3295
3296 static size_t _large_page_size = 0;
3297
3298 size_t os::Linux::find_large_page_size() {
3299 size_t large_page_size = 0;
3300
3301 // large_page_size on Linux is used to round up heap size. x86 uses either
3302 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3303 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3304 // page as large as 256M.
3305 //
3306 // Here we try to figure out page size by parsing /proc/meminfo and looking
3307 // for a line with the following format:
3308 // Hugepagesize: 2048 kB
3309 //
3310 // If we can't determine the value (e.g. /proc is not mounted, or the text
3311 // format has been changed), we'll use the largest page size supported by
3312 // the processor.
3313
3314 #ifndef ZERO
3315 large_page_size =
3316 AARCH64_ONLY(2 * M)
3317 AMD64_ONLY(2 * M)
3318 ARM32_ONLY(2 * M)
3319 IA32_ONLY(4 * M)
3320 IA64_ONLY(256 * M)
3321 PPC_ONLY(4 * M)
3322 S390_ONLY(1 * M)
3323 SPARC_ONLY(4 * M);
3324 #endif // ZERO
3325
3326 FILE *fp = fopen("/proc/meminfo", "r");
3327 if (fp) {
3328 while (!feof(fp)) {
3329 int x = 0;
3330 char buf[16];
3331 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3332 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3333 large_page_size = x * K;
3334 break;
3335 }
3336 } else {
3337 // skip to next line
3338 for (;;) {
3339 int ch = fgetc(fp);
3340 if (ch == EOF || ch == (int)'\n') break;
3341 }
3342 }
3343 }
3344 fclose(fp);
3345 }
3346
3347 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3348 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3349 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3350 proper_unit_for_byte_size(large_page_size));
3351 }
3352
3353 return large_page_size;
3354 }
3355
3356 size_t os::Linux::setup_large_page_size() {
3357 _large_page_size = Linux::find_large_page_size();
3358 const size_t default_page_size = (size_t)Linux::page_size();
3359 if (_large_page_size > default_page_size) {
3360 _page_sizes[0] = _large_page_size;
3361 _page_sizes[1] = default_page_size;
3362 _page_sizes[2] = 0;
3363 }
3364
3365 return _large_page_size;
3366 }
3367
3368 bool os::Linux::setup_large_page_type(size_t page_size) {
3369 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3370 FLAG_IS_DEFAULT(UseSHM) &&
3371 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3372
3373 // The type of large pages has not been specified by the user.
3374
3375 // Try UseHugeTLBFS and then UseSHM.
3376 UseHugeTLBFS = UseSHM = true;
3377
3378 // Don't try UseTransparentHugePages since there are known
3379 // performance issues with it turned on. This might change in the future.
3380 UseTransparentHugePages = false;
3381 }
3382
3383 if (UseTransparentHugePages) {
3384 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3385 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3386 UseHugeTLBFS = false;
3387 UseSHM = false;
3388 return true;
3389 }
3390 UseTransparentHugePages = false;
3391 }
3392
3393 if (UseHugeTLBFS) {
3394 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3395 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3396 UseSHM = false;
3397 return true;
3398 }
3399 UseHugeTLBFS = false;
3400 }
3401
3402 return UseSHM;
3403 }
3404
3405 void os::large_page_init() {
3406 if (!UseLargePages &&
3407 !UseTransparentHugePages &&
3408 !UseHugeTLBFS &&
3409 !UseSHM) {
3410 // Not using large pages.
3411 return;
3412 }
3413
3414 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3415 // The user explicitly turned off large pages.
3416 // Ignore the rest of the large pages flags.
3417 UseTransparentHugePages = false;
3418 UseHugeTLBFS = false;
3419 UseSHM = false;
3420 return;
3421 }
3422
3423 size_t large_page_size = Linux::setup_large_page_size();
3424 UseLargePages = Linux::setup_large_page_type(large_page_size);
3425
3426 set_coredump_filter();
3427 }
3428
3429 #ifndef SHM_HUGETLB
3430 #define SHM_HUGETLB 04000
3431 #endif
3432
3433 #define shm_warning_format(format, ...) \
3434 do { \
3435 if (UseLargePages && \
3436 (!FLAG_IS_DEFAULT(UseLargePages) || \
3437 !FLAG_IS_DEFAULT(UseSHM) || \
3438 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3439 warning(format, __VA_ARGS__); \
3440 } \
3441 } while (0)
3442
3443 #define shm_warning(str) shm_warning_format("%s", str)
3444
3445 #define shm_warning_with_errno(str) \
3446 do { \
3447 int err = errno; \
3448 shm_warning_format(str " (error = %d)", err); \
3449 } while (0)
3450
3451 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3452 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3453
3454 if (!is_aligned(alignment, SHMLBA)) {
3455 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3456 return NULL;
3457 }
3458
3459 // To ensure that we get 'alignment' aligned memory from shmat,
3460 // we pre-reserve aligned virtual memory and then attach to that.
3461
3462 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3463 if (pre_reserved_addr == NULL) {
3464 // Couldn't pre-reserve aligned memory.
3465 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3466 return NULL;
3467 }
3468
3469 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3470 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3471
3472 if ((intptr_t)addr == -1) {
3473 int err = errno;
3474 shm_warning_with_errno("Failed to attach shared memory.");
3475
3476 assert(err != EACCES, "Unexpected error");
3477 assert(err != EIDRM, "Unexpected error");
3478 assert(err != EINVAL, "Unexpected error");
3479
3480 // Since we don't know if the kernel unmapped the pre-reserved memory area
3481 // we can't unmap it, since that would potentially unmap memory that was
3482 // mapped from other threads.
3483 return NULL;
3484 }
3485
3486 return addr;
3487 }
3488
3489 static char* shmat_at_address(int shmid, char* req_addr) {
3490 if (!is_aligned(req_addr, SHMLBA)) {
3491 assert(false, "Requested address needs to be SHMLBA aligned");
3492 return NULL;
3493 }
3494
3495 char* addr = (char*)shmat(shmid, req_addr, 0);
3496
3497 if ((intptr_t)addr == -1) {
3498 shm_warning_with_errno("Failed to attach shared memory.");
3499 return NULL;
3500 }
3501
3502 return addr;
3503 }
3504
3505 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3506 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3507 if (req_addr != NULL) {
3508 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3509 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3510 return shmat_at_address(shmid, req_addr);
3511 }
3512
3513 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3514 // return large page size aligned memory addresses when req_addr == NULL.
3515 // However, if the alignment is larger than the large page size, we have
3516 // to manually ensure that the memory returned is 'alignment' aligned.
3517 if (alignment > os::large_page_size()) {
3518 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3519 return shmat_with_alignment(shmid, bytes, alignment);
3520 } else {
3521 return shmat_at_address(shmid, NULL);
3522 }
3523 }
3524
3525 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3526 char* req_addr, bool exec) {
3527 // "exec" is passed in but not used. Creating the shared image for
3528 // the code cache doesn't have an SHM_X executable permission to check.
3529 assert(UseLargePages && UseSHM, "only for SHM large pages");
3530 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3531 assert(is_aligned(req_addr, alignment), "Unaligned address");
3532
3533 if (!is_aligned(bytes, os::large_page_size())) {
3534 return NULL; // Fallback to small pages.
3535 }
3536
3537 // Create a large shared memory region to attach to based on size.
3538 // Currently, size is the total size of the heap.
3539 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3540 if (shmid == -1) {
3541 // Possible reasons for shmget failure:
3542 // 1. shmmax is too small for Java heap.
3543 // > check shmmax value: cat /proc/sys/kernel/shmmax
3544 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3545 // 2. not enough large page memory.
3546 // > check available large pages: cat /proc/meminfo
3547 // > increase amount of large pages:
3548 // echo new_value > /proc/sys/vm/nr_hugepages
3549 // Note 1: different Linux may use different name for this property,
3550 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3551 // Note 2: it's possible there's enough physical memory available but
3552 // they are so fragmented after a long run that they can't
3553 // coalesce into large pages. Try to reserve large pages when
3554 // the system is still "fresh".
3555 shm_warning_with_errno("Failed to reserve shared memory.");
3556 return NULL;
3557 }
3558
3559 // Attach to the region.
3560 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3561
3562 // Remove shmid. If shmat() is successful, the actual shared memory segment
3563 // will be deleted when it's detached by shmdt() or when the process
3564 // terminates. If shmat() is not successful this will remove the shared
3565 // segment immediately.
3566 shmctl(shmid, IPC_RMID, NULL);
3567
3568 return addr;
3569 }
3570
3571 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3572 int error) {
3573 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3574
3575 bool warn_on_failure = UseLargePages &&
3576 (!FLAG_IS_DEFAULT(UseLargePages) ||
3577 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3578 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3579
3580 if (warn_on_failure) {
3581 char msg[128];
3582 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3583 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3584 warning("%s", msg);
3585 }
3586 }
3587
3588 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3589 char* req_addr,
3590 bool exec) {
3591 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3592 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3593 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3594
3595 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3596 char* addr = (char*)::mmap(req_addr, bytes, prot,
3597 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3598 -1, 0);
3599
3600 if (addr == MAP_FAILED) {
3601 warn_on_large_pages_failure(req_addr, bytes, errno);
3602 return NULL;
3603 }
3604
3605 assert(is_aligned(addr, os::large_page_size()), "Must be");
3606
3607 return addr;
3608 }
3609
3610 // Reserve memory using mmap(MAP_HUGETLB).
3611 // - bytes shall be a multiple of alignment.
3612 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3613 // - alignment sets the alignment at which memory shall be allocated.
3614 // It must be a multiple of allocation granularity.
3615 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3616 // req_addr or NULL.
3617 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3618 size_t alignment,
3619 char* req_addr,
3620 bool exec) {
3621 size_t large_page_size = os::large_page_size();
3622 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3623
3624 assert(is_aligned(req_addr, alignment), "Must be");
3625 assert(is_aligned(bytes, alignment), "Must be");
3626
3627 // First reserve - but not commit - the address range in small pages.
3628 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3629
3630 if (start == NULL) {
3631 return NULL;
3632 }
3633
3634 assert(is_aligned(start, alignment), "Must be");
3635
3636 char* end = start + bytes;
3637
3638 // Find the regions of the allocated chunk that can be promoted to large pages.
3639 char* lp_start = align_up(start, large_page_size);
3640 char* lp_end = align_down(end, large_page_size);
3641
3642 size_t lp_bytes = lp_end - lp_start;
3643
3644 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3645
3646 if (lp_bytes == 0) {
3647 // The mapped region doesn't even span the start and the end of a large page.
3648 // Fall back to allocate a non-special area.
3649 ::munmap(start, end - start);
3650 return NULL;
3651 }
3652
3653 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3654
3655 void* result;
3656
3657 // Commit small-paged leading area.
3658 if (start != lp_start) {
3659 result = ::mmap(start, lp_start - start, prot,
3660 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3661 -1, 0);
3662 if (result == MAP_FAILED) {
3663 ::munmap(lp_start, end - lp_start);
3664 return NULL;
3665 }
3666 }
3667
3668 // Commit large-paged area.
3669 result = ::mmap(lp_start, lp_bytes, prot,
3670 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3671 -1, 0);
3672 if (result == MAP_FAILED) {
3673 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3674 // If the mmap above fails, the large pages region will be unmapped and we
3675 // have regions before and after with small pages. Release these regions.
3676 //
3677 // | mapped | unmapped | mapped |
3678 // ^ ^ ^ ^
3679 // start lp_start lp_end end
3680 //
3681 ::munmap(start, lp_start - start);
3682 ::munmap(lp_end, end - lp_end);
3683 return NULL;
3684 }
3685
3686 // Commit small-paged trailing area.
3687 if (lp_end != end) {
3688 result = ::mmap(lp_end, end - lp_end, prot,
3689 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3690 -1, 0);
3691 if (result == MAP_FAILED) {
3692 ::munmap(start, lp_end - start);
3693 return NULL;
3694 }
3695 }
3696
3697 return start;
3698 }
3699
3700 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3701 size_t alignment,
3702 char* req_addr,
3703 bool exec) {
3704 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3705 assert(is_aligned(req_addr, alignment), "Must be");
3706 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3707 assert(is_power_of_2(os::large_page_size()), "Must be");
3708 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3709
3710 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3711 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3712 } else {
3713 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3714 }
3715 }
3716
3717 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3718 char* req_addr, bool exec) {
3719 assert(UseLargePages, "only for large pages");
3720
3721 char* addr;
3722 if (UseSHM) {
3723 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3724 } else {
3725 assert(UseHugeTLBFS, "must be");
3726 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3727 }
3728
3729 if (addr != NULL) {
3730 if (UseNUMAInterleaving) {
3731 numa_make_global(addr, bytes);
3732 }
3733
3734 // The memory is committed
3735 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3736 }
3737
3738 return addr;
3739 }
3740
3741 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3742 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3743 return shmdt(base) == 0;
3744 }
3745
3746 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3747 return pd_release_memory(base, bytes);
3748 }
3749
3750 bool os::release_memory_special(char* base, size_t bytes) {
3751 bool res;
3752 if (MemTracker::tracking_level() > NMT_minimal) {
3753 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3754 res = os::Linux::release_memory_special_impl(base, bytes);
3755 if (res) {
3756 tkr.record((address)base, bytes);
3757 }
3758
3759 } else {
3760 res = os::Linux::release_memory_special_impl(base, bytes);
3761 }
3762 return res;
3763 }
3764
3765 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3766 assert(UseLargePages, "only for large pages");
3767 bool res;
3768
3769 if (UseSHM) {
3770 res = os::Linux::release_memory_special_shm(base, bytes);
3771 } else {
3772 assert(UseHugeTLBFS, "must be");
3773 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3774 }
3775 return res;
3776 }
3777
3778 size_t os::large_page_size() {
3779 return _large_page_size;
3780 }
3781
3782 // With SysV SHM the entire memory region must be allocated as shared
3783 // memory.
3784 // HugeTLBFS allows application to commit large page memory on demand.
3785 // However, when committing memory with HugeTLBFS fails, the region
3786 // that was supposed to be committed will lose the old reservation
3787 // and allow other threads to steal that memory region. Because of this
3788 // behavior we can't commit HugeTLBFS memory.
3789 bool os::can_commit_large_page_memory() {
3790 return UseTransparentHugePages;
3791 }
3792
3793 bool os::can_execute_large_page_memory() {
3794 return UseTransparentHugePages || UseHugeTLBFS;
3795 }
3796
3797 // Reserve memory at an arbitrary address, only if that area is
3798 // available (and not reserved for something else).
3799
3800 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3801 const int max_tries = 10;
3802 char* base[max_tries];
3803 size_t size[max_tries];
3804 const size_t gap = 0x000000;
3805
3806 // Assert only that the size is a multiple of the page size, since
3807 // that's all that mmap requires, and since that's all we really know
3808 // about at this low abstraction level. If we need higher alignment,
3809 // we can either pass an alignment to this method or verify alignment
3810 // in one of the methods further up the call chain. See bug 5044738.
3811 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3812
3813 // Repeatedly allocate blocks until the block is allocated at the
3814 // right spot.
3815
3816 // Linux mmap allows caller to pass an address as hint; give it a try first,
3817 // if kernel honors the hint then we can return immediately.
3818 char * addr = anon_mmap(requested_addr, bytes, false);
3819 if (addr == requested_addr) {
3820 return requested_addr;
3821 }
3822
3823 if (addr != NULL) {
3824 // mmap() is successful but it fails to reserve at the requested address
3825 anon_munmap(addr, bytes);
3826 }
3827
3828 int i;
3829 for (i = 0; i < max_tries; ++i) {
3830 base[i] = reserve_memory(bytes);
3831
3832 if (base[i] != NULL) {
3833 // Is this the block we wanted?
3834 if (base[i] == requested_addr) {
3835 size[i] = bytes;
3836 break;
3837 }
3838
3839 // Does this overlap the block we wanted? Give back the overlapped
3840 // parts and try again.
3841
3842 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3843 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3844 unmap_memory(base[i], top_overlap);
3845 base[i] += top_overlap;
3846 size[i] = bytes - top_overlap;
3847 } else {
3848 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3849 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3850 unmap_memory(requested_addr, bottom_overlap);
3851 size[i] = bytes - bottom_overlap;
3852 } else {
3853 size[i] = bytes;
3854 }
3855 }
3856 }
3857 }
3858
3859 // Give back the unused reserved pieces.
3860
3861 for (int j = 0; j < i; ++j) {
3862 if (base[j] != NULL) {
3863 unmap_memory(base[j], size[j]);
3864 }
3865 }
3866
3867 if (i < max_tries) {
3868 return requested_addr;
3869 } else {
3870 return NULL;
3871 }
3872 }
3873
3874 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3875 return ::read(fd, buf, nBytes);
3876 }
3877
3878 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3879 return ::pread(fd, buf, nBytes, offset);
3880 }
3881
3882 // Short sleep, direct OS call.
3883 //
3884 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3885 // sched_yield(2) will actually give up the CPU:
3886 //
3887 // * Alone on this pariticular CPU, keeps running.
3888 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3889 // (pre 2.6.39).
3890 //
3891 // So calling this with 0 is an alternative.
3892 //
3893 void os::naked_short_sleep(jlong ms) {
3894 struct timespec req;
3895
3896 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3897 req.tv_sec = 0;
3898 if (ms > 0) {
3899 req.tv_nsec = (ms % 1000) * 1000000;
3900 } else {
3901 req.tv_nsec = 1;
3902 }
3903
3904 nanosleep(&req, NULL);
3905
3906 return;
3907 }
3908
3909 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3910 void os::infinite_sleep() {
3911 while (true) { // sleep forever ...
3912 ::sleep(100); // ... 100 seconds at a time
3913 }
3914 }
3915
3916 // Used to convert frequent JVM_Yield() to nops
3917 bool os::dont_yield() {
3918 return DontYieldALot;
3919 }
3920
3921 void os::naked_yield() {
3922 sched_yield();
3923 }
3924
3925 ////////////////////////////////////////////////////////////////////////////////
3926 // thread priority support
3927
3928 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3929 // only supports dynamic priority, static priority must be zero. For real-time
3930 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3931 // However, for large multi-threaded applications, SCHED_RR is not only slower
3932 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3933 // of 5 runs - Sep 2005).
3934 //
3935 // The following code actually changes the niceness of kernel-thread/LWP. It
3936 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3937 // not the entire user process, and user level threads are 1:1 mapped to kernel
3938 // threads. It has always been the case, but could change in the future. For
3939 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3940 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3941
3942 int os::java_to_os_priority[CriticalPriority + 1] = {
3943 19, // 0 Entry should never be used
3944
3945 4, // 1 MinPriority
3946 3, // 2
3947 2, // 3
3948
3949 1, // 4
3950 0, // 5 NormPriority
3951 -1, // 6
3952
3953 -2, // 7
3954 -3, // 8
3955 -4, // 9 NearMaxPriority
3956
3957 -5, // 10 MaxPriority
3958
3959 -5 // 11 CriticalPriority
3960 };
3961
3962 static int prio_init() {
3963 if (ThreadPriorityPolicy == 1) {
3964 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3965 // if effective uid is not root. Perhaps, a more elegant way of doing
3966 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3967 if (geteuid() != 0) {
3968 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3969 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3970 }
3971 ThreadPriorityPolicy = 0;
3972 }
3973 }
3974 if (UseCriticalJavaThreadPriority) {
3975 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3976 }
3977 return 0;
3978 }
3979
3980 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3981 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3982
3983 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3984 return (ret == 0) ? OS_OK : OS_ERR;
3985 }
3986
3987 OSReturn os::get_native_priority(const Thread* const thread,
3988 int *priority_ptr) {
3989 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3990 *priority_ptr = java_to_os_priority[NormPriority];
3991 return OS_OK;
3992 }
3993
3994 errno = 0;
3995 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3996 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3997 }
3998
3999 // Hint to the underlying OS that a task switch would not be good.
4000 // Void return because it's a hint and can fail.
4001 void os::hint_no_preempt() {}
4002
4003 ////////////////////////////////////////////////////////////////////////////////
4004 // suspend/resume support
4005
4006 // The low-level signal-based suspend/resume support is a remnant from the
4007 // old VM-suspension that used to be for java-suspension, safepoints etc,
4008 // within hotspot. Currently used by JFR's OSThreadSampler
4009 //
4010 // The remaining code is greatly simplified from the more general suspension
4011 // code that used to be used.
4012 //
4013 // The protocol is quite simple:
4014 // - suspend:
4015 // - sends a signal to the target thread
4016 // - polls the suspend state of the osthread using a yield loop
4017 // - target thread signal handler (SR_handler) sets suspend state
4018 // and blocks in sigsuspend until continued
4019 // - resume:
4020 // - sets target osthread state to continue
4021 // - sends signal to end the sigsuspend loop in the SR_handler
4022 //
4023 // Note that the SR_lock plays no role in this suspend/resume protocol,
4024 // but is checked for NULL in SR_handler as a thread termination indicator.
4025 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4026 //
4027 // Note that resume_clear_context() and suspend_save_context() are needed
4028 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4029 // which in part is used by:
4030 // - Forte Analyzer: AsyncGetCallTrace()
4031 // - StackBanging: get_frame_at_stack_banging_point()
4032
4033 static void resume_clear_context(OSThread *osthread) {
4034 osthread->set_ucontext(NULL);
4035 osthread->set_siginfo(NULL);
4036 }
4037
4038 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4039 ucontext_t* context) {
4040 osthread->set_ucontext(context);
4041 osthread->set_siginfo(siginfo);
4042 }
4043
4044 // Handler function invoked when a thread's execution is suspended or
4045 // resumed. We have to be careful that only async-safe functions are
4046 // called here (Note: most pthread functions are not async safe and
4047 // should be avoided.)
4048 //
4049 // Note: sigwait() is a more natural fit than sigsuspend() from an
4050 // interface point of view, but sigwait() prevents the signal hander
4051 // from being run. libpthread would get very confused by not having
4052 // its signal handlers run and prevents sigwait()'s use with the
4053 // mutex granting granting signal.
4054 //
4055 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4056 //
4057 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4058 // Save and restore errno to avoid confusing native code with EINTR
4059 // after sigsuspend.
4060 int old_errno = errno;
4061
4062 Thread* thread = Thread::current_or_null_safe();
4063 assert(thread != NULL, "Missing current thread in SR_handler");
4064
4065 // On some systems we have seen signal delivery get "stuck" until the signal
4066 // mask is changed as part of thread termination. Check that the current thread
4067 // has not already terminated (via SR_lock()) - else the following assertion
4068 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4069 // destructor has completed.
4070
4071 if (thread->SR_lock() == NULL) {
4072 return;
4073 }
4074
4075 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4076
4077 OSThread* osthread = thread->osthread();
4078
4079 os::SuspendResume::State current = osthread->sr.state();
4080 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4081 suspend_save_context(osthread, siginfo, context);
4082
4083 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4084 os::SuspendResume::State state = osthread->sr.suspended();
4085 if (state == os::SuspendResume::SR_SUSPENDED) {
4086 sigset_t suspend_set; // signals for sigsuspend()
4087 sigemptyset(&suspend_set);
4088 // get current set of blocked signals and unblock resume signal
4089 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4090 sigdelset(&suspend_set, SR_signum);
4091
4092 sr_semaphore.signal();
4093 // wait here until we are resumed
4094 while (1) {
4095 sigsuspend(&suspend_set);
4096
4097 os::SuspendResume::State result = osthread->sr.running();
4098 if (result == os::SuspendResume::SR_RUNNING) {
4099 sr_semaphore.signal();
4100 break;
4101 }
4102 }
4103
4104 } else if (state == os::SuspendResume::SR_RUNNING) {
4105 // request was cancelled, continue
4106 } else {
4107 ShouldNotReachHere();
4108 }
4109
4110 resume_clear_context(osthread);
4111 } else if (current == os::SuspendResume::SR_RUNNING) {
4112 // request was cancelled, continue
4113 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4114 // ignore
4115 } else {
4116 // ignore
4117 }
4118
4119 errno = old_errno;
4120 }
4121
4122 static int SR_initialize() {
4123 struct sigaction act;
4124 char *s;
4125
4126 // Get signal number to use for suspend/resume
4127 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4128 int sig = ::strtol(s, 0, 10);
4129 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4130 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4131 SR_signum = sig;
4132 } else {
4133 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4134 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4135 }
4136 }
4137
4138 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4139 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4140
4141 sigemptyset(&SR_sigset);
4142 sigaddset(&SR_sigset, SR_signum);
4143
4144 // Set up signal handler for suspend/resume
4145 act.sa_flags = SA_RESTART|SA_SIGINFO;
4146 act.sa_handler = (void (*)(int)) SR_handler;
4147
4148 // SR_signum is blocked by default.
4149 // 4528190 - We also need to block pthread restart signal (32 on all
4150 // supported Linux platforms). Note that LinuxThreads need to block
4151 // this signal for all threads to work properly. So we don't have
4152 // to use hard-coded signal number when setting up the mask.
4153 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4154
4155 if (sigaction(SR_signum, &act, 0) == -1) {
4156 return -1;
4157 }
4158
4159 // Save signal flag
4160 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4161 return 0;
4162 }
4163
4164 static int sr_notify(OSThread* osthread) {
4165 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4166 assert_status(status == 0, status, "pthread_kill");
4167 return status;
4168 }
4169
4170 // "Randomly" selected value for how long we want to spin
4171 // before bailing out on suspending a thread, also how often
4172 // we send a signal to a thread we want to resume
4173 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4174 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4175
4176 // returns true on success and false on error - really an error is fatal
4177 // but this seems the normal response to library errors
4178 static bool do_suspend(OSThread* osthread) {
4179 assert(osthread->sr.is_running(), "thread should be running");
4180 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4181
4182 // mark as suspended and send signal
4183 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4184 // failed to switch, state wasn't running?
4185 ShouldNotReachHere();
4186 return false;
4187 }
4188
4189 if (sr_notify(osthread) != 0) {
4190 ShouldNotReachHere();
4191 }
4192
4193 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4194 while (true) {
4195 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4196 break;
4197 } else {
4198 // timeout
4199 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4200 if (cancelled == os::SuspendResume::SR_RUNNING) {
4201 return false;
4202 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4203 // make sure that we consume the signal on the semaphore as well
4204 sr_semaphore.wait();
4205 break;
4206 } else {
4207 ShouldNotReachHere();
4208 return false;
4209 }
4210 }
4211 }
4212
4213 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4214 return true;
4215 }
4216
4217 static void do_resume(OSThread* osthread) {
4218 assert(osthread->sr.is_suspended(), "thread should be suspended");
4219 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4220
4221 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4222 // failed to switch to WAKEUP_REQUEST
4223 ShouldNotReachHere();
4224 return;
4225 }
4226
4227 while (true) {
4228 if (sr_notify(osthread) == 0) {
4229 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4230 if (osthread->sr.is_running()) {
4231 return;
4232 }
4233 }
4234 } else {
4235 ShouldNotReachHere();
4236 }
4237 }
4238
4239 guarantee(osthread->sr.is_running(), "Must be running!");
4240 }
4241
4242 ///////////////////////////////////////////////////////////////////////////////////
4243 // signal handling (except suspend/resume)
4244
4245 // This routine may be used by user applications as a "hook" to catch signals.
4246 // The user-defined signal handler must pass unrecognized signals to this
4247 // routine, and if it returns true (non-zero), then the signal handler must
4248 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4249 // routine will never retun false (zero), but instead will execute a VM panic
4250 // routine kill the process.
4251 //
4252 // If this routine returns false, it is OK to call it again. This allows
4253 // the user-defined signal handler to perform checks either before or after
4254 // the VM performs its own checks. Naturally, the user code would be making
4255 // a serious error if it tried to handle an exception (such as a null check
4256 // or breakpoint) that the VM was generating for its own correct operation.
4257 //
4258 // This routine may recognize any of the following kinds of signals:
4259 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4260 // It should be consulted by handlers for any of those signals.
4261 //
4262 // The caller of this routine must pass in the three arguments supplied
4263 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4264 // field of the structure passed to sigaction(). This routine assumes that
4265 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4266 //
4267 // Note that the VM will print warnings if it detects conflicting signal
4268 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4269 //
4270 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4271 siginfo_t* siginfo,
4272 void* ucontext,
4273 int abort_if_unrecognized);
4274
4275 void signalHandler(int sig, siginfo_t* info, void* uc) {
4276 assert(info != NULL && uc != NULL, "it must be old kernel");
4277 int orig_errno = errno; // Preserve errno value over signal handler.
4278 JVM_handle_linux_signal(sig, info, uc, true);
4279 errno = orig_errno;
4280 }
4281
4282
4283 // This boolean allows users to forward their own non-matching signals
4284 // to JVM_handle_linux_signal, harmlessly.
4285 bool os::Linux::signal_handlers_are_installed = false;
4286
4287 // For signal-chaining
4288 struct sigaction sigact[NSIG];
4289 uint64_t sigs = 0;
4290 #if (64 < NSIG-1)
4291 #error "Not all signals can be encoded in sigs. Adapt its type!"
4292 #endif
4293 bool os::Linux::libjsig_is_loaded = false;
4294 typedef struct sigaction *(*get_signal_t)(int);
4295 get_signal_t os::Linux::get_signal_action = NULL;
4296
4297 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4298 struct sigaction *actp = NULL;
4299
4300 if (libjsig_is_loaded) {
4301 // Retrieve the old signal handler from libjsig
4302 actp = (*get_signal_action)(sig);
4303 }
4304 if (actp == NULL) {
4305 // Retrieve the preinstalled signal handler from jvm
4306 actp = get_preinstalled_handler(sig);
4307 }
4308
4309 return actp;
4310 }
4311
4312 static bool call_chained_handler(struct sigaction *actp, int sig,
4313 siginfo_t *siginfo, void *context) {
4314 // Call the old signal handler
4315 if (actp->sa_handler == SIG_DFL) {
4316 // It's more reasonable to let jvm treat it as an unexpected exception
4317 // instead of taking the default action.
4318 return false;
4319 } else if (actp->sa_handler != SIG_IGN) {
4320 if ((actp->sa_flags & SA_NODEFER) == 0) {
4321 // automaticlly block the signal
4322 sigaddset(&(actp->sa_mask), sig);
4323 }
4324
4325 sa_handler_t hand = NULL;
4326 sa_sigaction_t sa = NULL;
4327 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4328 // retrieve the chained handler
4329 if (siginfo_flag_set) {
4330 sa = actp->sa_sigaction;
4331 } else {
4332 hand = actp->sa_handler;
4333 }
4334
4335 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4336 actp->sa_handler = SIG_DFL;
4337 }
4338
4339 // try to honor the signal mask
4340 sigset_t oset;
4341 sigemptyset(&oset);
4342 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4343
4344 // call into the chained handler
4345 if (siginfo_flag_set) {
4346 (*sa)(sig, siginfo, context);
4347 } else {
4348 (*hand)(sig);
4349 }
4350
4351 // restore the signal mask
4352 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4353 }
4354 // Tell jvm's signal handler the signal is taken care of.
4355 return true;
4356 }
4357
4358 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4359 bool chained = false;
4360 // signal-chaining
4361 if (UseSignalChaining) {
4362 struct sigaction *actp = get_chained_signal_action(sig);
4363 if (actp != NULL) {
4364 chained = call_chained_handler(actp, sig, siginfo, context);
4365 }
4366 }
4367 return chained;
4368 }
4369
4370 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4371 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4372 return &sigact[sig];
4373 }
4374 return NULL;
4375 }
4376
4377 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4378 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4379 sigact[sig] = oldAct;
4380 sigs |= (uint64_t)1 << (sig-1);
4381 }
4382
4383 // for diagnostic
4384 int sigflags[NSIG];
4385
4386 int os::Linux::get_our_sigflags(int sig) {
4387 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4388 return sigflags[sig];
4389 }
4390
4391 void os::Linux::set_our_sigflags(int sig, int flags) {
4392 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4393 if (sig > 0 && sig < NSIG) {
4394 sigflags[sig] = flags;
4395 }
4396 }
4397
4398 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4399 // Check for overwrite.
4400 struct sigaction oldAct;
4401 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4402
4403 void* oldhand = oldAct.sa_sigaction
4404 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4405 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4406 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4407 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4408 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4409 if (AllowUserSignalHandlers || !set_installed) {
4410 // Do not overwrite; user takes responsibility to forward to us.
4411 return;
4412 } else if (UseSignalChaining) {
4413 // save the old handler in jvm
4414 save_preinstalled_handler(sig, oldAct);
4415 // libjsig also interposes the sigaction() call below and saves the
4416 // old sigaction on it own.
4417 } else {
4418 fatal("Encountered unexpected pre-existing sigaction handler "
4419 "%#lx for signal %d.", (long)oldhand, sig);
4420 }
4421 }
4422
4423 struct sigaction sigAct;
4424 sigfillset(&(sigAct.sa_mask));
4425 sigAct.sa_handler = SIG_DFL;
4426 if (!set_installed) {
4427 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4428 } else {
4429 sigAct.sa_sigaction = signalHandler;
4430 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4431 }
4432 // Save flags, which are set by ours
4433 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4434 sigflags[sig] = sigAct.sa_flags;
4435
4436 int ret = sigaction(sig, &sigAct, &oldAct);
4437 assert(ret == 0, "check");
4438
4439 void* oldhand2 = oldAct.sa_sigaction
4440 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4441 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4442 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4443 }
4444
4445 // install signal handlers for signals that HotSpot needs to
4446 // handle in order to support Java-level exception handling.
4447
4448 void os::Linux::install_signal_handlers() {
4449 if (!signal_handlers_are_installed) {
4450 signal_handlers_are_installed = true;
4451
4452 // signal-chaining
4453 typedef void (*signal_setting_t)();
4454 signal_setting_t begin_signal_setting = NULL;
4455 signal_setting_t end_signal_setting = NULL;
4456 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4457 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4458 if (begin_signal_setting != NULL) {
4459 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4460 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4461 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4462 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4463 libjsig_is_loaded = true;
4464 assert(UseSignalChaining, "should enable signal-chaining");
4465 }
4466 if (libjsig_is_loaded) {
4467 // Tell libjsig jvm is setting signal handlers
4468 (*begin_signal_setting)();
4469 }
4470
4471 set_signal_handler(SIGSEGV, true);
4472 set_signal_handler(SIGPIPE, true);
4473 set_signal_handler(SIGBUS, true);
4474 set_signal_handler(SIGILL, true);
4475 set_signal_handler(SIGFPE, true);
4476 #if defined(PPC64)
4477 set_signal_handler(SIGTRAP, true);
4478 #endif
4479 set_signal_handler(SIGXFSZ, true);
4480
4481 if (libjsig_is_loaded) {
4482 // Tell libjsig jvm finishes setting signal handlers
4483 (*end_signal_setting)();
4484 }
4485
4486 // We don't activate signal checker if libjsig is in place, we trust ourselves
4487 // and if UserSignalHandler is installed all bets are off.
4488 // Log that signal checking is off only if -verbose:jni is specified.
4489 if (CheckJNICalls) {
4490 if (libjsig_is_loaded) {
4491 if (PrintJNIResolving) {
4492 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4493 }
4494 check_signals = false;
4495 }
4496 if (AllowUserSignalHandlers) {
4497 if (PrintJNIResolving) {
4498 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4499 }
4500 check_signals = false;
4501 }
4502 }
4503 }
4504 }
4505
4506 // This is the fastest way to get thread cpu time on Linux.
4507 // Returns cpu time (user+sys) for any thread, not only for current.
4508 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4509 // It might work on 2.6.10+ with a special kernel/glibc patch.
4510 // For reference, please, see IEEE Std 1003.1-2004:
4511 // http://www.unix.org/single_unix_specification
4512
4513 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4514 struct timespec tp;
4515 int rc = os::Linux::clock_gettime(clockid, &tp);
4516 assert(rc == 0, "clock_gettime is expected to return 0 code");
4517
4518 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4519 }
4520
4521 void os::Linux::initialize_os_info() {
4522 assert(_os_version == 0, "OS info already initialized");
4523
4524 struct utsname _uname;
4525
4526 uint32_t major;
4527 uint32_t minor;
4528 uint32_t fix;
4529
4530 int rc;
4531
4532 // Kernel version is unknown if
4533 // verification below fails.
4534 _os_version = 0x01000000;
4535
4536 rc = uname(&_uname);
4537 if (rc != -1) {
4538
4539 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4540 if (rc == 3) {
4541
4542 if (major < 256 && minor < 256 && fix < 256) {
4543 // Kernel version format is as expected,
4544 // set it overriding unknown state.
4545 _os_version = (major << 16) |
4546 (minor << 8 ) |
4547 (fix << 0 ) ;
4548 }
4549 }
4550 }
4551 }
4552
4553 uint32_t os::Linux::os_version() {
4554 assert(_os_version != 0, "not initialized");
4555 return _os_version & 0x00FFFFFF;
4556 }
4557
4558 bool os::Linux::os_version_is_known() {
4559 assert(_os_version != 0, "not initialized");
4560 return _os_version & 0x01000000 ? false : true;
4561 }
4562
4563 /////
4564 // glibc on Linux platform uses non-documented flag
4565 // to indicate, that some special sort of signal
4566 // trampoline is used.
4567 // We will never set this flag, and we should
4568 // ignore this flag in our diagnostic
4569 #ifdef SIGNIFICANT_SIGNAL_MASK
4570 #undef SIGNIFICANT_SIGNAL_MASK
4571 #endif
4572 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4573
4574 static const char* get_signal_handler_name(address handler,
4575 char* buf, int buflen) {
4576 int offset = 0;
4577 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4578 if (found) {
4579 // skip directory names
4580 const char *p1, *p2;
4581 p1 = buf;
4582 size_t len = strlen(os::file_separator());
4583 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4584 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4585 } else {
4586 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4587 }
4588 return buf;
4589 }
4590
4591 static void print_signal_handler(outputStream* st, int sig,
4592 char* buf, size_t buflen) {
4593 struct sigaction sa;
4594
4595 sigaction(sig, NULL, &sa);
4596
4597 // See comment for SIGNIFICANT_SIGNAL_MASK define
4598 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4599
4600 st->print("%s: ", os::exception_name(sig, buf, buflen));
4601
4602 address handler = (sa.sa_flags & SA_SIGINFO)
4603 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4604 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4605
4606 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4607 st->print("SIG_DFL");
4608 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4609 st->print("SIG_IGN");
4610 } else {
4611 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4612 }
4613
4614 st->print(", sa_mask[0]=");
4615 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4616
4617 address rh = VMError::get_resetted_sighandler(sig);
4618 // May be, handler was resetted by VMError?
4619 if (rh != NULL) {
4620 handler = rh;
4621 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4622 }
4623
4624 st->print(", sa_flags=");
4625 os::Posix::print_sa_flags(st, sa.sa_flags);
4626
4627 // Check: is it our handler?
4628 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4629 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4630 // It is our signal handler
4631 // check for flags, reset system-used one!
4632 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4633 st->print(
4634 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4635 os::Linux::get_our_sigflags(sig));
4636 }
4637 }
4638 st->cr();
4639 }
4640
4641
4642 #define DO_SIGNAL_CHECK(sig) \
4643 do { \
4644 if (!sigismember(&check_signal_done, sig)) { \
4645 os::Linux::check_signal_handler(sig); \
4646 } \
4647 } while (0)
4648
4649 // This method is a periodic task to check for misbehaving JNI applications
4650 // under CheckJNI, we can add any periodic checks here
4651
4652 void os::run_periodic_checks() {
4653 if (check_signals == false) return;
4654
4655 // SEGV and BUS if overridden could potentially prevent
4656 // generation of hs*.log in the event of a crash, debugging
4657 // such a case can be very challenging, so we absolutely
4658 // check the following for a good measure:
4659 DO_SIGNAL_CHECK(SIGSEGV);
4660 DO_SIGNAL_CHECK(SIGILL);
4661 DO_SIGNAL_CHECK(SIGFPE);
4662 DO_SIGNAL_CHECK(SIGBUS);
4663 DO_SIGNAL_CHECK(SIGPIPE);
4664 DO_SIGNAL_CHECK(SIGXFSZ);
4665 #if defined(PPC64)
4666 DO_SIGNAL_CHECK(SIGTRAP);
4667 #endif
4668
4669 // ReduceSignalUsage allows the user to override these handlers
4670 // see comments at the very top and jvm_md.h
4671 if (!ReduceSignalUsage) {
4672 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4673 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4674 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4675 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4676 }
4677
4678 DO_SIGNAL_CHECK(SR_signum);
4679 }
4680
4681 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4682
4683 static os_sigaction_t os_sigaction = NULL;
4684
4685 void os::Linux::check_signal_handler(int sig) {
4686 char buf[O_BUFLEN];
4687 address jvmHandler = NULL;
4688
4689
4690 struct sigaction act;
4691 if (os_sigaction == NULL) {
4692 // only trust the default sigaction, in case it has been interposed
4693 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4694 if (os_sigaction == NULL) return;
4695 }
4696
4697 os_sigaction(sig, (struct sigaction*)NULL, &act);
4698
4699
4700 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4701
4702 address thisHandler = (act.sa_flags & SA_SIGINFO)
4703 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4704 : CAST_FROM_FN_PTR(address, act.sa_handler);
4705
4706
4707 switch (sig) {
4708 case SIGSEGV:
4709 case SIGBUS:
4710 case SIGFPE:
4711 case SIGPIPE:
4712 case SIGILL:
4713 case SIGXFSZ:
4714 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4715 break;
4716
4717 case SHUTDOWN1_SIGNAL:
4718 case SHUTDOWN2_SIGNAL:
4719 case SHUTDOWN3_SIGNAL:
4720 case BREAK_SIGNAL:
4721 jvmHandler = (address)user_handler();
4722 break;
4723
4724 default:
4725 if (sig == SR_signum) {
4726 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4727 } else {
4728 return;
4729 }
4730 break;
4731 }
4732
4733 if (thisHandler != jvmHandler) {
4734 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4735 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4736 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4737 // No need to check this sig any longer
4738 sigaddset(&check_signal_done, sig);
4739 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4740 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4741 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4742 exception_name(sig, buf, O_BUFLEN));
4743 }
4744 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4745 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4746 tty->print("expected:");
4747 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4748 tty->cr();
4749 tty->print(" found:");
4750 os::Posix::print_sa_flags(tty, act.sa_flags);
4751 tty->cr();
4752 // No need to check this sig any longer
4753 sigaddset(&check_signal_done, sig);
4754 }
4755
4756 // Dump all the signal
4757 if (sigismember(&check_signal_done, sig)) {
4758 print_signal_handlers(tty, buf, O_BUFLEN);
4759 }
4760 }
4761
4762 extern void report_error(char* file_name, int line_no, char* title,
4763 char* format, ...);
4764
4765 // this is called _before_ the most of global arguments have been parsed
4766 void os::init(void) {
4767 char dummy; // used to get a guess on initial stack address
4768 // first_hrtime = gethrtime();
4769
4770 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4771
4772 init_random(1234567);
4773
4774 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4775 if (Linux::page_size() == -1) {
4776 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4777 os::strerror(errno));
4778 }
4779 init_page_sizes((size_t) Linux::page_size());
4780
4781 Linux::initialize_system_info();
4782
4783 Linux::initialize_os_info();
4784
4785 // main_thread points to the aboriginal thread
4786 Linux::_main_thread = pthread_self();
4787
4788 Linux::clock_init();
4789 initial_time_count = javaTimeNanos();
4790
4791 // retrieve entry point for pthread_setname_np
4792 Linux::_pthread_setname_np =
4793 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4794
4795 os::Posix::init();
4796 }
4797
4798 // To install functions for atexit system call
4799 extern "C" {
4800 static void perfMemory_exit_helper() {
4801 perfMemory_exit();
4802 }
4803 }
4804
4805 // this is called _after_ the global arguments have been parsed
4806 jint os::init_2(void) {
4807
4808 os::Posix::init_2();
4809
4810 Linux::fast_thread_clock_init();
4811
4812 // initialize suspend/resume support - must do this before signal_sets_init()
4813 if (SR_initialize() != 0) {
4814 perror("SR_initialize failed");
4815 return JNI_ERR;
4816 }
4817
4818 Linux::signal_sets_init();
4819 Linux::install_signal_handlers();
4820
4821 // Check and sets minimum stack sizes against command line options
4822 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4823 return JNI_ERR;
4824 }
4825 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4826
4827 #if defined(IA32)
4828 workaround_expand_exec_shield_cs_limit();
4829 #endif
4830
4831 Linux::libpthread_init();
4832 Linux::sched_getcpu_init();
4833 log_info(os)("HotSpot is running with %s, %s",
4834 Linux::glibc_version(), Linux::libpthread_version());
4835
4836 if (UseNUMA) {
4837 if (!Linux::libnuma_init()) {
4838 UseNUMA = false;
4839 } else {
4840 if ((Linux::numa_max_node() < 1)) {
4841 // There's only one node(they start from 0), disable NUMA.
4842 UseNUMA = false;
4843 }
4844 }
4845 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4846 // we can make the adaptive lgrp chunk resizing work. If the user specified
4847 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4848 // disable adaptive resizing.
4849 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4850 if (FLAG_IS_DEFAULT(UseNUMA)) {
4851 UseNUMA = false;
4852 } else {
4853 if (FLAG_IS_DEFAULT(UseLargePages) &&
4854 FLAG_IS_DEFAULT(UseSHM) &&
4855 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4856 UseLargePages = false;
4857 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4858 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4859 UseAdaptiveSizePolicy = false;
4860 UseAdaptiveNUMAChunkSizing = false;
4861 }
4862 }
4863 }
4864 if (!UseNUMA && ForceNUMA) {
4865 UseNUMA = true;
4866 }
4867 }
4868
4869 if (MaxFDLimit) {
4870 // set the number of file descriptors to max. print out error
4871 // if getrlimit/setrlimit fails but continue regardless.
4872 struct rlimit nbr_files;
4873 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4874 if (status != 0) {
4875 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4876 } else {
4877 nbr_files.rlim_cur = nbr_files.rlim_max;
4878 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4879 if (status != 0) {
4880 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4881 }
4882 }
4883 }
4884
4885 // Initialize lock used to serialize thread creation (see os::create_thread)
4886 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4887
4888 // at-exit methods are called in the reverse order of their registration.
4889 // atexit functions are called on return from main or as a result of a
4890 // call to exit(3C). There can be only 32 of these functions registered
4891 // and atexit() does not set errno.
4892
4893 if (PerfAllowAtExitRegistration) {
4894 // only register atexit functions if PerfAllowAtExitRegistration is set.
4895 // atexit functions can be delayed until process exit time, which
4896 // can be problematic for embedded VM situations. Embedded VMs should
4897 // call DestroyJavaVM() to assure that VM resources are released.
4898
4899 // note: perfMemory_exit_helper atexit function may be removed in
4900 // the future if the appropriate cleanup code can be added to the
4901 // VM_Exit VMOperation's doit method.
4902 if (atexit(perfMemory_exit_helper) != 0) {
4903 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4904 }
4905 }
4906
4907 // initialize thread priority policy
4908 prio_init();
4909
4910 return JNI_OK;
4911 }
4912
4913 // Mark the polling page as unreadable
4914 void os::make_polling_page_unreadable(void) {
4915 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4916 fatal("Could not disable polling page");
4917 }
4918 }
4919
4920 // Mark the polling page as readable
4921 void os::make_polling_page_readable(void) {
4922 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4923 fatal("Could not enable polling page");
4924 }
4925 }
4926
4927 // older glibc versions don't have this macro (which expands to
4928 // an optimized bit-counting function) so we have to roll our own
4929 #ifndef CPU_COUNT
4930
4931 static int _cpu_count(const cpu_set_t* cpus) {
4932 int count = 0;
4933 // only look up to the number of configured processors
4934 for (int i = 0; i < os::processor_count(); i++) {
4935 if (CPU_ISSET(i, cpus)) {
4936 count++;
4937 }
4938 }
4939 return count;
4940 }
4941
4942 #define CPU_COUNT(cpus) _cpu_count(cpus)
4943
4944 #endif // CPU_COUNT
4945
4946 // Get the current number of available processors for this process.
4947 // This value can change at any time during a process's lifetime.
4948 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4949 // If it appears there may be more than 1024 processors then we do a
4950 // dynamic check - see 6515172 for details.
4951 // If anything goes wrong we fallback to returning the number of online
4952 // processors - which can be greater than the number available to the process.
4953 int os::active_processor_count() {
4954 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4955 cpu_set_t* cpus_p = &cpus;
4956 int cpus_size = sizeof(cpu_set_t);
4957
4958 int configured_cpus = processor_count(); // upper bound on available cpus
4959 int cpu_count = 0;
4960
4961 // old build platforms may not support dynamic cpu sets
4962 #ifdef CPU_ALLOC
4963
4964 // To enable easy testing of the dynamic path on different platforms we
4965 // introduce a diagnostic flag: UseCpuAllocPath
4966 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4967 // kernel may use a mask bigger than cpu_set_t
4968 log_trace(os)("active_processor_count: using dynamic path %s"
4969 "- configured processors: %d",
4970 UseCpuAllocPath ? "(forced) " : "",
4971 configured_cpus);
4972 cpus_p = CPU_ALLOC(configured_cpus);
4973 if (cpus_p != NULL) {
4974 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4975 // zero it just to be safe
4976 CPU_ZERO_S(cpus_size, cpus_p);
4977 }
4978 else {
4979 // failed to allocate so fallback to online cpus
4980 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4981 log_trace(os)("active_processor_count: "
4982 "CPU_ALLOC failed (%s) - using "
4983 "online processor count: %d",
4984 os::strerror(errno), online_cpus);
4985 return online_cpus;
4986 }
4987 }
4988 else {
4989 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4990 configured_cpus);
4991 }
4992 #else // CPU_ALLOC
4993 // these stubs won't be executed
4994 #define CPU_COUNT_S(size, cpus) -1
4995 #define CPU_FREE(cpus)
4996
4997 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4998 configured_cpus);
4999 #endif // CPU_ALLOC
5000
5001 // pid 0 means the current thread - which we have to assume represents the process
5002 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5003 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5004 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5005 }
5006 else {
5007 cpu_count = CPU_COUNT(cpus_p);
5008 }
5009 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5010 }
5011 else {
5012 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5013 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5014 "which may exceed available processors", os::strerror(errno), cpu_count);
5015 }
5016
5017 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5018 CPU_FREE(cpus_p);
5019 }
5020
5021 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5022 return cpu_count;
5023 }
5024
5025 void os::set_native_thread_name(const char *name) {
5026 if (Linux::_pthread_setname_np) {
5027 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5028 snprintf(buf, sizeof(buf), "%s", name);
5029 buf[sizeof(buf) - 1] = '\0';
5030 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5031 // ERANGE should not happen; all other errors should just be ignored.
5032 assert(rc != ERANGE, "pthread_setname_np failed");
5033 }
5034 }
5035
5036 bool os::distribute_processes(uint length, uint* distribution) {
5037 // Not yet implemented.
5038 return false;
5039 }
5040
5041 bool os::bind_to_processor(uint processor_id) {
5042 // Not yet implemented.
5043 return false;
5044 }
5045
5046 ///
5047
5048 void os::SuspendedThreadTask::internal_do_task() {
5049 if (do_suspend(_thread->osthread())) {
5050 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5051 do_task(context);
5052 do_resume(_thread->osthread());
5053 }
5054 }
5055
5056 ////////////////////////////////////////////////////////////////////////////////
5057 // debug support
5058
5059 bool os::find(address addr, outputStream* st) {
5060 Dl_info dlinfo;
5061 memset(&dlinfo, 0, sizeof(dlinfo));
5062 if (dladdr(addr, &dlinfo) != 0) {
5063 st->print(PTR_FORMAT ": ", p2i(addr));
5064 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5065 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5066 p2i(addr) - p2i(dlinfo.dli_saddr));
5067 } else if (dlinfo.dli_fbase != NULL) {
5068 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5069 } else {
5070 st->print("<absolute address>");
5071 }
5072 if (dlinfo.dli_fname != NULL) {
5073 st->print(" in %s", dlinfo.dli_fname);
5074 }
5075 if (dlinfo.dli_fbase != NULL) {
5076 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5077 }
5078 st->cr();
5079
5080 if (Verbose) {
5081 // decode some bytes around the PC
5082 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5083 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5084 address lowest = (address) dlinfo.dli_sname;
5085 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5086 if (begin < lowest) begin = lowest;
5087 Dl_info dlinfo2;
5088 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5089 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5090 end = (address) dlinfo2.dli_saddr;
5091 }
5092 Disassembler::decode(begin, end, st);
5093 }
5094 return true;
5095 }
5096 return false;
5097 }
5098
5099 ////////////////////////////////////////////////////////////////////////////////
5100 // misc
5101
5102 // This does not do anything on Linux. This is basically a hook for being
5103 // able to use structured exception handling (thread-local exception filters)
5104 // on, e.g., Win32.
5105 void
5106 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5107 JavaCallArguments* args, Thread* thread) {
5108 f(value, method, args, thread);
5109 }
5110
5111 void os::print_statistics() {
5112 }
5113
5114 bool os::message_box(const char* title, const char* message) {
5115 int i;
5116 fdStream err(defaultStream::error_fd());
5117 for (i = 0; i < 78; i++) err.print_raw("=");
5118 err.cr();
5119 err.print_raw_cr(title);
5120 for (i = 0; i < 78; i++) err.print_raw("-");
5121 err.cr();
5122 err.print_raw_cr(message);
5123 for (i = 0; i < 78; i++) err.print_raw("=");
5124 err.cr();
5125
5126 char buf[16];
5127 // Prevent process from exiting upon "read error" without consuming all CPU
5128 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5129
5130 return buf[0] == 'y' || buf[0] == 'Y';
5131 }
5132
5133 int os::stat(const char *path, struct stat *sbuf) {
5134 char pathbuf[MAX_PATH];
5135 if (strlen(path) > MAX_PATH - 1) {
5136 errno = ENAMETOOLONG;
5137 return -1;
5138 }
5139 os::native_path(strcpy(pathbuf, path));
5140 return ::stat(pathbuf, sbuf);
5141 }
5142
5143 // Is a (classpath) directory empty?
5144 bool os::dir_is_empty(const char* path) {
5145 DIR *dir = NULL;
5146 struct dirent *ptr;
5147
5148 dir = opendir(path);
5149 if (dir == NULL) return true;
5150
5151 // Scan the directory
5152 bool result = true;
5153 char buf[sizeof(struct dirent) + MAX_PATH];
5154 while (result && (ptr = ::readdir(dir)) != NULL) {
5155 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5156 result = false;
5157 }
5158 }
5159 closedir(dir);
5160 return result;
5161 }
5162
5163 // This code originates from JDK's sysOpen and open64_w
5164 // from src/solaris/hpi/src/system_md.c
5165
5166 int os::open(const char *path, int oflag, int mode) {
5167 if (strlen(path) > MAX_PATH - 1) {
5168 errno = ENAMETOOLONG;
5169 return -1;
5170 }
5171
5172 // All file descriptors that are opened in the Java process and not
5173 // specifically destined for a subprocess should have the close-on-exec
5174 // flag set. If we don't set it, then careless 3rd party native code
5175 // might fork and exec without closing all appropriate file descriptors
5176 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5177 // turn might:
5178 //
5179 // - cause end-of-file to fail to be detected on some file
5180 // descriptors, resulting in mysterious hangs, or
5181 //
5182 // - might cause an fopen in the subprocess to fail on a system
5183 // suffering from bug 1085341.
5184 //
5185 // (Yes, the default setting of the close-on-exec flag is a Unix
5186 // design flaw)
5187 //
5188 // See:
5189 // 1085341: 32-bit stdio routines should support file descriptors >255
5190 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5191 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5192 //
5193 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5194 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5195 // because it saves a system call and removes a small window where the flag
5196 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5197 // and we fall back to using FD_CLOEXEC (see below).
5198 #ifdef O_CLOEXEC
5199 oflag |= O_CLOEXEC;
5200 #endif
5201
5202 int fd = ::open64(path, oflag, mode);
5203 if (fd == -1) return -1;
5204
5205 //If the open succeeded, the file might still be a directory
5206 {
5207 struct stat64 buf64;
5208 int ret = ::fstat64(fd, &buf64);
5209 int st_mode = buf64.st_mode;
5210
5211 if (ret != -1) {
5212 if ((st_mode & S_IFMT) == S_IFDIR) {
5213 errno = EISDIR;
5214 ::close(fd);
5215 return -1;
5216 }
5217 } else {
5218 ::close(fd);
5219 return -1;
5220 }
5221 }
5222
5223 #ifdef FD_CLOEXEC
5224 // Validate that the use of the O_CLOEXEC flag on open above worked.
5225 // With recent kernels, we will perform this check exactly once.
5226 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5227 if (!O_CLOEXEC_is_known_to_work) {
5228 int flags = ::fcntl(fd, F_GETFD);
5229 if (flags != -1) {
5230 if ((flags & FD_CLOEXEC) != 0)
5231 O_CLOEXEC_is_known_to_work = 1;
5232 else
5233 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5234 }
5235 }
5236 #endif
5237
5238 return fd;
5239 }
5240
5241
5242 // create binary file, rewriting existing file if required
5243 int os::create_binary_file(const char* path, bool rewrite_existing) {
5244 int oflags = O_WRONLY | O_CREAT;
5245 if (!rewrite_existing) {
5246 oflags |= O_EXCL;
5247 }
5248 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5249 }
5250
5251 // return current position of file pointer
5252 jlong os::current_file_offset(int fd) {
5253 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5254 }
5255
5256 // move file pointer to the specified offset
5257 jlong os::seek_to_file_offset(int fd, jlong offset) {
5258 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5259 }
5260
5261 // This code originates from JDK's sysAvailable
5262 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5263
5264 int os::available(int fd, jlong *bytes) {
5265 jlong cur, end;
5266 int mode;
5267 struct stat64 buf64;
5268
5269 if (::fstat64(fd, &buf64) >= 0) {
5270 mode = buf64.st_mode;
5271 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5272 int n;
5273 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5274 *bytes = n;
5275 return 1;
5276 }
5277 }
5278 }
5279 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5280 return 0;
5281 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5282 return 0;
5283 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5284 return 0;
5285 }
5286 *bytes = end - cur;
5287 return 1;
5288 }
5289
5290 // Map a block of memory.
5291 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5292 char *addr, size_t bytes, bool read_only,
5293 bool allow_exec) {
5294 int prot;
5295 int flags = MAP_PRIVATE;
5296
5297 if (read_only) {
5298 prot = PROT_READ;
5299 } else {
5300 prot = PROT_READ | PROT_WRITE;
5301 }
5302
5303 if (allow_exec) {
5304 prot |= PROT_EXEC;
5305 }
5306
5307 if (addr != NULL) {
5308 flags |= MAP_FIXED;
5309 }
5310
5311 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5312 fd, file_offset);
5313 if (mapped_address == MAP_FAILED) {
5314 return NULL;
5315 }
5316 return mapped_address;
5317 }
5318
5319
5320 // Remap a block of memory.
5321 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5322 char *addr, size_t bytes, bool read_only,
5323 bool allow_exec) {
5324 // same as map_memory() on this OS
5325 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5326 allow_exec);
5327 }
5328
5329
5330 // Unmap a block of memory.
5331 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5332 return munmap(addr, bytes) == 0;
5333 }
5334
5335 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5336
5337 static clockid_t thread_cpu_clockid(Thread* thread) {
5338 pthread_t tid = thread->osthread()->pthread_id();
5339 clockid_t clockid;
5340
5341 // Get thread clockid
5342 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5343 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5344 return clockid;
5345 }
5346
5347 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5348 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5349 // of a thread.
5350 //
5351 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5352 // the fast estimate available on the platform.
5353
5354 jlong os::current_thread_cpu_time() {
5355 if (os::Linux::supports_fast_thread_cpu_time()) {
5356 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5357 } else {
5358 // return user + sys since the cost is the same
5359 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5360 }
5361 }
5362
5363 jlong os::thread_cpu_time(Thread* thread) {
5364 // consistent with what current_thread_cpu_time() returns
5365 if (os::Linux::supports_fast_thread_cpu_time()) {
5366 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5367 } else {
5368 return slow_thread_cpu_time(thread, true /* user + sys */);
5369 }
5370 }
5371
5372 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5373 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5374 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5375 } else {
5376 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5377 }
5378 }
5379
5380 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5381 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5382 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5383 } else {
5384 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5385 }
5386 }
5387
5388 // -1 on error.
5389 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5390 pid_t tid = thread->osthread()->thread_id();
5391 char *s;
5392 char stat[2048];
5393 int statlen;
5394 char proc_name[64];
5395 int count;
5396 long sys_time, user_time;
5397 char cdummy;
5398 int idummy;
5399 long ldummy;
5400 FILE *fp;
5401
5402 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5403 fp = fopen(proc_name, "r");
5404 if (fp == NULL) return -1;
5405 statlen = fread(stat, 1, 2047, fp);
5406 stat[statlen] = '\0';
5407 fclose(fp);
5408
5409 // Skip pid and the command string. Note that we could be dealing with
5410 // weird command names, e.g. user could decide to rename java launcher
5411 // to "java 1.4.2 :)", then the stat file would look like
5412 // 1234 (java 1.4.2 :)) R ... ...
5413 // We don't really need to know the command string, just find the last
5414 // occurrence of ")" and then start parsing from there. See bug 4726580.
5415 s = strrchr(stat, ')');
5416 if (s == NULL) return -1;
5417
5418 // Skip blank chars
5419 do { s++; } while (s && isspace(*s));
5420
5421 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5422 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5423 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5424 &user_time, &sys_time);
5425 if (count != 13) return -1;
5426 if (user_sys_cpu_time) {
5427 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5428 } else {
5429 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5430 }
5431 }
5432
5433 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5434 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5435 info_ptr->may_skip_backward = false; // elapsed time not wall time
5436 info_ptr->may_skip_forward = false; // elapsed time not wall time
5437 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5438 }
5439
5440 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5441 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5442 info_ptr->may_skip_backward = false; // elapsed time not wall time
5443 info_ptr->may_skip_forward = false; // elapsed time not wall time
5444 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5445 }
5446
5447 bool os::is_thread_cpu_time_supported() {
5448 return true;
5449 }
5450
5451 // System loadavg support. Returns -1 if load average cannot be obtained.
5452 // Linux doesn't yet have a (official) notion of processor sets,
5453 // so just return the system wide load average.
5454 int os::loadavg(double loadavg[], int nelem) {
5455 return ::getloadavg(loadavg, nelem);
5456 }
5457
5458 void os::pause() {
5459 char filename[MAX_PATH];
5460 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5461 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5462 } else {
5463 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5464 }
5465
5466 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5467 if (fd != -1) {
5468 struct stat buf;
5469 ::close(fd);
5470 while (::stat(filename, &buf) == 0) {
5471 (void)::poll(NULL, 0, 100);
5472 }
5473 } else {
5474 jio_fprintf(stderr,
5475 "Could not open pause file '%s', continuing immediately.\n", filename);
5476 }
5477 }
5478
5479 extern char** environ;
5480
5481 // Run the specified command in a separate process. Return its exit value,
5482 // or -1 on failure (e.g. can't fork a new process).
5483 // Unlike system(), this function can be called from signal handler. It
5484 // doesn't block SIGINT et al.
5485 int os::fork_and_exec(char* cmd) {
5486 const char * argv[4] = {"sh", "-c", cmd, NULL};
5487
5488 pid_t pid = fork();
5489
5490 if (pid < 0) {
5491 // fork failed
5492 return -1;
5493
5494 } else if (pid == 0) {
5495 // child process
5496
5497 execve("/bin/sh", (char* const*)argv, environ);
5498
5499 // execve failed
5500 _exit(-1);
5501
5502 } else {
5503 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5504 // care about the actual exit code, for now.
5505
5506 int status;
5507
5508 // Wait for the child process to exit. This returns immediately if
5509 // the child has already exited. */
5510 while (waitpid(pid, &status, 0) < 0) {
5511 switch (errno) {
5512 case ECHILD: return 0;
5513 case EINTR: break;
5514 default: return -1;
5515 }
5516 }
5517
5518 if (WIFEXITED(status)) {
5519 // The child exited normally; get its exit code.
5520 return WEXITSTATUS(status);
5521 } else if (WIFSIGNALED(status)) {
5522 // The child exited because of a signal
5523 // The best value to return is 0x80 + signal number,
5524 // because that is what all Unix shells do, and because
5525 // it allows callers to distinguish between process exit and
5526 // process death by signal.
5527 return 0x80 + WTERMSIG(status);
5528 } else {
5529 // Unknown exit code; pass it through
5530 return status;
5531 }
5532 }
5533 }
5534
5535 // is_headless_jre()
5536 //
5537 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5538 // in order to report if we are running in a headless jre
5539 //
5540 // Since JDK8 xawt/libmawt.so was moved into the same directory
5541 // as libawt.so, and renamed libawt_xawt.so
5542 //
5543 bool os::is_headless_jre() {
5544 struct stat statbuf;
5545 char buf[MAXPATHLEN];
5546 char libmawtpath[MAXPATHLEN];
5547 const char *xawtstr = "/xawt/libmawt.so";
5548 const char *new_xawtstr = "/libawt_xawt.so";
5549 char *p;
5550
5551 // Get path to libjvm.so
5552 os::jvm_path(buf, sizeof(buf));
5553
5554 // Get rid of libjvm.so
5555 p = strrchr(buf, '/');
5556 if (p == NULL) {
5557 return false;
5558 } else {
5559 *p = '\0';
5560 }
5561
5562 // Get rid of client or server
5563 p = strrchr(buf, '/');
5564 if (p == NULL) {
5565 return false;
5566 } else {
5567 *p = '\0';
5568 }
5569
5570 // check xawt/libmawt.so
5571 strcpy(libmawtpath, buf);
5572 strcat(libmawtpath, xawtstr);
5573 if (::stat(libmawtpath, &statbuf) == 0) return false;
5574
5575 // check libawt_xawt.so
5576 strcpy(libmawtpath, buf);
5577 strcat(libmawtpath, new_xawtstr);
5578 if (::stat(libmawtpath, &statbuf) == 0) return false;
5579
5580 return true;
5581 }
5582
5583 // Get the default path to the core file
5584 // Returns the length of the string
5585 int os::get_core_path(char* buffer, size_t bufferSize) {
5586 /*
5587 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5588 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5589 */
5590 const int core_pattern_len = 129;
5591 char core_pattern[core_pattern_len] = {0};
5592
5593 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5594 if (core_pattern_file == -1) {
5595 return -1;
5596 }
5597
5598 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5599 ::close(core_pattern_file);
5600 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5601 return -1;
5602 }
5603 if (core_pattern[ret-1] == '\n') {
5604 core_pattern[ret-1] = '\0';
5605 } else {
5606 core_pattern[ret] = '\0';
5607 }
5608
5609 char *pid_pos = strstr(core_pattern, "%p");
5610 int written;
5611
5612 if (core_pattern[0] == '/') {
5613 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5614 } else {
5615 char cwd[PATH_MAX];
5616
5617 const char* p = get_current_directory(cwd, PATH_MAX);
5618 if (p == NULL) {
5619 return -1;
5620 }
5621
5622 if (core_pattern[0] == '|') {
5623 written = jio_snprintf(buffer, bufferSize,
5624 "\"%s\" (or dumping to %s/core.%d)",
5625 &core_pattern[1], p, current_process_id());
5626 } else {
5627 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5628 }
5629 }
5630
5631 if (written < 0) {
5632 return -1;
5633 }
5634
5635 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5636 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5637
5638 if (core_uses_pid_file != -1) {
5639 char core_uses_pid = 0;
5640 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5641 ::close(core_uses_pid_file);
5642
5643 if (core_uses_pid == '1') {
5644 jio_snprintf(buffer + written, bufferSize - written,
5645 ".%d", current_process_id());
5646 }
5647 }
5648 }
5649
5650 return strlen(buffer);
5651 }
5652
5653 bool os::start_debugging(char *buf, int buflen) {
5654 int len = (int)strlen(buf);
5655 char *p = &buf[len];
5656
5657 jio_snprintf(p, buflen-len,
5658 "\n\n"
5659 "Do you want to debug the problem?\n\n"
5660 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5661 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5662 "Otherwise, press RETURN to abort...",
5663 os::current_process_id(), os::current_process_id(),
5664 os::current_thread_id(), os::current_thread_id());
5665
5666 bool yes = os::message_box("Unexpected Error", buf);
5667
5668 if (yes) {
5669 // yes, user asked VM to launch debugger
5670 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5671 os::current_process_id(), os::current_process_id());
5672
5673 os::fork_and_exec(buf);
5674 yes = false;
5675 }
5676 return yes;
5677 }
5678
5679
5680 // Java/Compiler thread:
5681 //
5682 // Low memory addresses
5683 // P0 +------------------------+
5684 // | |\ Java thread created by VM does not have glibc
5685 // | glibc guard page | - guard page, attached Java thread usually has
5686 // | |/ 1 glibc guard page.
5687 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5688 // | |\
5689 // | HotSpot Guard Pages | - red, yellow and reserved pages
5690 // | |/
5691 // +------------------------+ JavaThread::stack_reserved_zone_base()
5692 // | |\
5693 // | Normal Stack | -
5694 // | |/
5695 // P2 +------------------------+ Thread::stack_base()
5696 //
5697 // Non-Java thread:
5698 //
5699 // Low memory addresses
5700 // P0 +------------------------+
5701 // | |\
5702 // | glibc guard page | - usually 1 page
5703 // | |/
5704 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5705 // | |\
5706 // | Normal Stack | -
5707 // | |/
5708 // P2 +------------------------+ Thread::stack_base()
5709 //
5710 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5711 // returned from pthread_attr_getstack().
5712 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5713 // of the stack size given in pthread_attr. We work around this for
5714 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5715 //
5716 #ifndef ZERO
5717 static void current_stack_region(address * bottom, size_t * size) {
5718 if (os::Linux::is_initial_thread()) {
5719 // initial thread needs special handling because pthread_getattr_np()
5720 // may return bogus value.
5721 *bottom = os::Linux::initial_thread_stack_bottom();
5722 *size = os::Linux::initial_thread_stack_size();
5723 } else {
5724 pthread_attr_t attr;
5725
5726 int rslt = pthread_getattr_np(pthread_self(), &attr);
5727
5728 // JVM needs to know exact stack location, abort if it fails
5729 if (rslt != 0) {
5730 if (rslt == ENOMEM) {
5731 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5732 } else {
5733 fatal("pthread_getattr_np failed with error = %d", rslt);
5734 }
5735 }
5736
5737 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5738 fatal("Cannot locate current stack attributes!");
5739 }
5740
5741 // Work around NPTL stack guard error.
5742 size_t guard_size = 0;
5743 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5744 if (rslt != 0) {
5745 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5746 }
5747 *bottom += guard_size;
5748 *size -= guard_size;
5749
5750 pthread_attr_destroy(&attr);
5751
5752 }
5753 assert(os::current_stack_pointer() >= *bottom &&
5754 os::current_stack_pointer() < *bottom + *size, "just checking");
5755 }
5756
5757 address os::current_stack_base() {
5758 address bottom;
5759 size_t size;
5760 current_stack_region(&bottom, &size);
5761 return (bottom + size);
5762 }
5763
5764 size_t os::current_stack_size() {
5765 // This stack size includes the usable stack and HotSpot guard pages
5766 // (for the threads that have Hotspot guard pages).
5767 address bottom;
5768 size_t size;
5769 current_stack_region(&bottom, &size);
5770 return size;
5771 }
5772 #endif
5773
5774 static inline struct timespec get_mtime(const char* filename) {
5775 struct stat st;
5776 int ret = os::stat(filename, &st);
5777 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5778 return st.st_mtim;
5779 }
5780
5781 int os::compare_file_modified_times(const char* file1, const char* file2) {
5782 struct timespec filetime1 = get_mtime(file1);
5783 struct timespec filetime2 = get_mtime(file2);
5784 int diff = filetime1.tv_sec - filetime2.tv_sec;
5785 if (diff == 0) {
5786 return filetime1.tv_nsec - filetime2.tv_nsec;
5787 }
5788 return diff;
5789 }
5790
5791 /////////////// Unit tests ///////////////
5792
5793 #ifndef PRODUCT
5794
5795 #define test_log(...) \
5796 do { \
5797 if (VerboseInternalVMTests) { \
5798 tty->print_cr(__VA_ARGS__); \
5799 tty->flush(); \
5800 } \
5801 } while (false)
5802
5803 class TestReserveMemorySpecial : AllStatic {
5804 public:
5805 static void small_page_write(void* addr, size_t size) {
5806 size_t page_size = os::vm_page_size();
5807
5808 char* end = (char*)addr + size;
5809 for (char* p = (char*)addr; p < end; p += page_size) {
5810 *p = 1;
5811 }
5812 }
5813
5814 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5815 if (!UseHugeTLBFS) {
5816 return;
5817 }
5818
5819 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5820
5821 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5822
5823 if (addr != NULL) {
5824 small_page_write(addr, size);
5825
5826 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5827 }
5828 }
5829
5830 static void test_reserve_memory_special_huge_tlbfs_only() {
5831 if (!UseHugeTLBFS) {
5832 return;
5833 }
5834
5835 size_t lp = os::large_page_size();
5836
5837 for (size_t size = lp; size <= lp * 10; size += lp) {
5838 test_reserve_memory_special_huge_tlbfs_only(size);
5839 }
5840 }
5841
5842 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5843 size_t lp = os::large_page_size();
5844 size_t ag = os::vm_allocation_granularity();
5845
5846 // sizes to test
5847 const size_t sizes[] = {
5848 lp, lp + ag, lp + lp / 2, lp * 2,
5849 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5850 lp * 10, lp * 10 + lp / 2
5851 };
5852 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5853
5854 // For each size/alignment combination, we test three scenarios:
5855 // 1) with req_addr == NULL
5856 // 2) with a non-null req_addr at which we expect to successfully allocate
5857 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5858 // expect the allocation to either fail or to ignore req_addr
5859
5860 // Pre-allocate two areas; they shall be as large as the largest allocation
5861 // and aligned to the largest alignment we will be testing.
5862 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5863 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5864 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5865 -1, 0);
5866 assert(mapping1 != MAP_FAILED, "should work");
5867
5868 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5869 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5870 -1, 0);
5871 assert(mapping2 != MAP_FAILED, "should work");
5872
5873 // Unmap the first mapping, but leave the second mapping intact: the first
5874 // mapping will serve as a value for a "good" req_addr (case 2). The second
5875 // mapping, still intact, as "bad" req_addr (case 3).
5876 ::munmap(mapping1, mapping_size);
5877
5878 // Case 1
5879 test_log("%s, req_addr NULL:", __FUNCTION__);
5880 test_log("size align result");
5881
5882 for (int i = 0; i < num_sizes; i++) {
5883 const size_t size = sizes[i];
5884 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5885 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5886 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5887 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5888 if (p != NULL) {
5889 assert(is_aligned(p, alignment), "must be");
5890 small_page_write(p, size);
5891 os::Linux::release_memory_special_huge_tlbfs(p, size);
5892 }
5893 }
5894 }
5895
5896 // Case 2
5897 test_log("%s, req_addr non-NULL:", __FUNCTION__);
5898 test_log("size align req_addr result");
5899
5900 for (int i = 0; i < num_sizes; i++) {
5901 const size_t size = sizes[i];
5902 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5903 char* const req_addr = align_up(mapping1, alignment);
5904 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5905 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
5906 size, alignment, p2i(req_addr), p2i(p),
5907 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
5908 if (p != NULL) {
5909 assert(p == req_addr, "must be");
5910 small_page_write(p, size);
5911 os::Linux::release_memory_special_huge_tlbfs(p, size);
5912 }
5913 }
5914 }
5915
5916 // Case 3
5917 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
5918 test_log("size align req_addr result");
5919
5920 for (int i = 0; i < num_sizes; i++) {
5921 const size_t size = sizes[i];
5922 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5923 char* const req_addr = align_up(mapping2, alignment);
5924 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5925 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
5926 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
5927 // as the area around req_addr contains already existing mappings, the API should always
5928 // return NULL (as per contract, it cannot return another address)
5929 assert(p == NULL, "must be");
5930 }
5931 }
5932
5933 ::munmap(mapping2, mapping_size);
5934
5935 }
5936
5937 static void test_reserve_memory_special_huge_tlbfs() {
5938 if (!UseHugeTLBFS) {
5939 return;
5940 }
5941
5942 test_reserve_memory_special_huge_tlbfs_only();
5943 test_reserve_memory_special_huge_tlbfs_mixed();
5944 }
5945
5946 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
5947 if (!UseSHM) {
5948 return;
5949 }
5950
5951 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
5952
5953 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
5954
5955 if (addr != NULL) {
5956 assert(is_aligned(addr, alignment), "Check");
5957 assert(is_aligned(addr, os::large_page_size()), "Check");
5958
5959 small_page_write(addr, size);
5960
5961 os::Linux::release_memory_special_shm(addr, size);
5962 }
5963 }
5964
5965 static void test_reserve_memory_special_shm() {
5966 size_t lp = os::large_page_size();
5967 size_t ag = os::vm_allocation_granularity();
5968
5969 for (size_t size = ag; size < lp * 3; size += ag) {
5970 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5971 test_reserve_memory_special_shm(size, alignment);
5972 }
5973 }
5974 }
5975
5976 static void test() {
5977 test_reserve_memory_special_huge_tlbfs();
5978 test_reserve_memory_special_shm();
5979 }
5980 };
5981
5982 void TestReserveMemorySpecial_test() {
5983 TestReserveMemorySpecial::test();
5984 }
5985
5986 #endif