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