1 /*
2 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/align.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 uint32_t os::Linux::_os_version = 0;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 // Signal number used to suspend/resume a thread
159
160 // do not use any signal number less than SIGSEGV, see 4355769
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 // utility functions
165
166 static int SR_initialize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 sysinfo(&si);
176
177 return (julong)si.freeram * si.mem_unit;
178 }
179
180 julong os::physical_memory() {
181 return Linux::physical_memory();
182 }
183
184 // Return true if user is running as root.
185
186 bool os::have_special_privileges() {
187 static bool init = false;
188 static bool privileges = false;
189 if (!init) {
190 privileges = (getuid() != geteuid()) || (getgid() != getegid());
191 init = true;
192 }
193 return privileges;
194 }
195
196
197 #ifndef SYS_gettid
198 // i386: 224, ia64: 1105, amd64: 186, sparc 143
199 #ifdef __ia64__
200 #define SYS_gettid 1105
201 #else
202 #ifdef __i386__
203 #define SYS_gettid 224
204 #else
205 #ifdef __amd64__
206 #define SYS_gettid 186
207 #else
208 #ifdef __sparc__
209 #define SYS_gettid 143
210 #else
211 #error define gettid for the arch
212 #endif
213 #endif
214 #endif
215 #endif
216 #endif
217
218
219 // pid_t gettid()
220 //
221 // Returns the kernel thread id of the currently running thread. Kernel
222 // thread id is used to access /proc.
223 pid_t os::Linux::gettid() {
224 int rslt = syscall(SYS_gettid);
225 assert(rslt != -1, "must be."); // old linuxthreads implementation?
226 return (pid_t)rslt;
227 }
228
229 // Most versions of linux have a bug where the number of processors are
230 // determined by looking at the /proc file system. In a chroot environment,
231 // the system call returns 1. This causes the VM to act as if it is
232 // a single processor and elide locking (see is_MP() call).
233 static bool unsafe_chroot_detected = false;
234 static const char *unstable_chroot_error = "/proc file system not found.\n"
235 "Java may be unstable running multithreaded in a chroot "
236 "environment on Linux when /proc filesystem is not mounted.";
237
238 void os::Linux::initialize_system_info() {
239 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
240 if (processor_count() == 1) {
241 pid_t pid = os::Linux::gettid();
242 char fname[32];
243 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
244 FILE *fp = fopen(fname, "r");
245 if (fp == NULL) {
246 unsafe_chroot_detected = true;
247 } else {
248 fclose(fp);
249 }
250 }
251 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
252 assert(processor_count() > 0, "linux error");
253 }
254
255 void os::init_system_properties_values() {
256 // The next steps are taken in the product version:
257 //
258 // Obtain the JAVA_HOME value from the location of libjvm.so.
259 // This library should be located at:
260 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
261 //
262 // If "/jre/lib/" appears at the right place in the path, then we
263 // assume libjvm.so is installed in a JDK and we use this path.
264 //
265 // Otherwise exit with message: "Could not create the Java virtual machine."
266 //
267 // The following extra steps are taken in the debugging version:
268 //
269 // If "/jre/lib/" does NOT appear at the right place in the path
270 // instead of exit check for $JAVA_HOME environment variable.
271 //
272 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
273 // then we append a fake suffix "hotspot/libjvm.so" to this path so
274 // it looks like libjvm.so is installed there
275 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
276 //
277 // Otherwise exit.
278 //
279 // Important note: if the location of libjvm.so changes this
280 // code needs to be changed accordingly.
281
282 // See ld(1):
283 // The linker uses the following search paths to locate required
284 // shared libraries:
285 // 1: ...
286 // ...
287 // 7: The default directories, normally /lib and /usr/lib.
288 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
289 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
290 #else
291 #define DEFAULT_LIBPATH "/lib:/usr/lib"
292 #endif
293
294 // Base path of extensions installed on the system.
295 #define SYS_EXT_DIR "/usr/java/packages"
296 #define EXTENSIONS_DIR "/lib/ext"
297
298 // Buffer that fits several sprintfs.
299 // Note that the space for the colon and the trailing null are provided
300 // by the nulls included by the sizeof operator.
301 const size_t bufsize =
302 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
303 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
304 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
305
306 // sysclasspath, java_home, dll_dir
307 {
308 char *pslash;
309 os::jvm_path(buf, bufsize);
310
311 // Found the full path to libjvm.so.
312 // Now cut the path to <java_home>/jre if we can.
313 pslash = strrchr(buf, '/');
314 if (pslash != NULL) {
315 *pslash = '\0'; // Get rid of /libjvm.so.
316 }
317 pslash = strrchr(buf, '/');
318 if (pslash != NULL) {
319 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
320 }
321 Arguments::set_dll_dir(buf);
322
323 if (pslash != NULL) {
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /lib.
327 }
328 }
329 Arguments::set_java_home(buf);
330 set_boot_path('/', ':');
331 }
332
333 // Where to look for native libraries.
334 //
335 // Note: Due to a legacy implementation, most of the library path
336 // is set in the launcher. This was to accomodate linking restrictions
337 // on legacy Linux implementations (which are no longer supported).
338 // Eventually, all the library path setting will be done here.
339 //
340 // However, to prevent the proliferation of improperly built native
341 // libraries, the new path component /usr/java/packages is added here.
342 // Eventually, all the library path setting will be done here.
343 {
344 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
345 // should always exist (until the legacy problem cited above is
346 // addressed).
347 const char *v = ::getenv("LD_LIBRARY_PATH");
348 const char *v_colon = ":";
349 if (v == NULL) { v = ""; v_colon = ""; }
350 // That's +1 for the colon and +1 for the trailing '\0'.
351 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
352 strlen(v) + 1 +
353 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
354 mtInternal);
355 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
356 Arguments::set_library_path(ld_library_path);
357 FREE_C_HEAP_ARRAY(char, ld_library_path);
358 }
359
360 // Extensions directories.
361 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
362 Arguments::set_ext_dirs(buf);
363
364 FREE_C_HEAP_ARRAY(char, buf);
365
366 #undef DEFAULT_LIBPATH
367 #undef SYS_EXT_DIR
368 #undef EXTENSIONS_DIR
369 }
370
371 ////////////////////////////////////////////////////////////////////////////////
372 // breakpoint support
373
374 void os::breakpoint() {
375 BREAKPOINT;
376 }
377
378 extern "C" void breakpoint() {
379 // use debugger to set breakpoint here
380 }
381
382 ////////////////////////////////////////////////////////////////////////////////
383 // signal support
384
385 debug_only(static bool signal_sets_initialized = false);
386 static sigset_t unblocked_sigs, vm_sigs;
387
388 bool os::Linux::is_sig_ignored(int sig) {
389 struct sigaction oact;
390 sigaction(sig, (struct sigaction*)NULL, &oact);
391 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
392 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
393 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
394 return true;
395 } else {
396 return false;
397 }
398 }
399
400 void os::Linux::signal_sets_init() {
401 // Should also have an assertion stating we are still single-threaded.
402 assert(!signal_sets_initialized, "Already initialized");
403 // Fill in signals that are necessarily unblocked for all threads in
404 // the VM. Currently, we unblock the following signals:
405 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
406 // by -Xrs (=ReduceSignalUsage));
407 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
408 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
409 // the dispositions or masks wrt these signals.
410 // Programs embedding the VM that want to use the above signals for their
411 // own purposes must, at this time, use the "-Xrs" option to prevent
412 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
413 // (See bug 4345157, and other related bugs).
414 // In reality, though, unblocking these signals is really a nop, since
415 // these signals are not blocked by default.
416 sigemptyset(&unblocked_sigs);
417 sigaddset(&unblocked_sigs, SIGILL);
418 sigaddset(&unblocked_sigs, SIGSEGV);
419 sigaddset(&unblocked_sigs, SIGBUS);
420 sigaddset(&unblocked_sigs, SIGFPE);
421 #if defined(PPC64)
422 sigaddset(&unblocked_sigs, SIGTRAP);
423 #endif
424 sigaddset(&unblocked_sigs, SR_signum);
425
426 if (!ReduceSignalUsage) {
427 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
428 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
429 }
430 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
431 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
432 }
433 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
434 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
435 }
436 }
437 // Fill in signals that are blocked by all but the VM thread.
438 sigemptyset(&vm_sigs);
439 if (!ReduceSignalUsage) {
440 sigaddset(&vm_sigs, BREAK_SIGNAL);
441 }
442 debug_only(signal_sets_initialized = true);
443
444 }
445
446 // These are signals that are unblocked while a thread is running Java.
447 // (For some reason, they get blocked by default.)
448 sigset_t* os::Linux::unblocked_signals() {
449 assert(signal_sets_initialized, "Not initialized");
450 return &unblocked_sigs;
451 }
452
453 // These are the signals that are blocked while a (non-VM) thread is
454 // running Java. Only the VM thread handles these signals.
455 sigset_t* os::Linux::vm_signals() {
456 assert(signal_sets_initialized, "Not initialized");
457 return &vm_sigs;
458 }
459
460 void os::Linux::hotspot_sigmask(Thread* thread) {
461
462 //Save caller's signal mask before setting VM signal mask
463 sigset_t caller_sigmask;
464 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
465
466 OSThread* osthread = thread->osthread();
467 osthread->set_caller_sigmask(caller_sigmask);
468
469 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
470
471 if (!ReduceSignalUsage) {
472 if (thread->is_VM_thread()) {
473 // Only the VM thread handles BREAK_SIGNAL ...
474 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
475 } else {
476 // ... all other threads block BREAK_SIGNAL
477 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
478 }
479 }
480 }
481
482 //////////////////////////////////////////////////////////////////////////////
483 // detecting pthread library
484
485 void os::Linux::libpthread_init() {
486 // Save glibc and pthread version strings.
487 #if !defined(_CS_GNU_LIBC_VERSION) || \
488 !defined(_CS_GNU_LIBPTHREAD_VERSION)
489 #error "glibc too old (< 2.3.2)"
490 #endif
491
492 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
493 assert(n > 0, "cannot retrieve glibc version");
494 char *str = (char *)malloc(n, mtInternal);
495 confstr(_CS_GNU_LIBC_VERSION, str, n);
496 os::Linux::set_glibc_version(str);
497
498 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
499 assert(n > 0, "cannot retrieve pthread version");
500 str = (char *)malloc(n, mtInternal);
501 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
502 os::Linux::set_libpthread_version(str);
503 }
504
505 /////////////////////////////////////////////////////////////////////////////
506 // thread stack expansion
507
508 // os::Linux::manually_expand_stack() takes care of expanding the thread
509 // stack. Note that this is normally not needed: pthread stacks allocate
510 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
511 // committed. Therefore it is not necessary to expand the stack manually.
512 //
513 // Manually expanding the stack was historically needed on LinuxThreads
514 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
515 // it is kept to deal with very rare corner cases:
516 //
517 // For one, user may run the VM on an own implementation of threads
518 // whose stacks are - like the old LinuxThreads - implemented using
519 // mmap(MAP_GROWSDOWN).
520 //
521 // Also, this coding may be needed if the VM is running on the primordial
522 // thread. Normally we avoid running on the primordial thread; however,
523 // user may still invoke the VM on the primordial thread.
524 //
525 // The following historical comment describes the details about running
526 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
527
528
529 // Force Linux kernel to expand current thread stack. If "bottom" is close
530 // to the stack guard, caller should block all signals.
531 //
532 // MAP_GROWSDOWN:
533 // A special mmap() flag that is used to implement thread stacks. It tells
534 // kernel that the memory region should extend downwards when needed. This
535 // allows early versions of LinuxThreads to only mmap the first few pages
536 // when creating a new thread. Linux kernel will automatically expand thread
537 // stack as needed (on page faults).
538 //
539 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
540 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
541 // region, it's hard to tell if the fault is due to a legitimate stack
542 // access or because of reading/writing non-exist memory (e.g. buffer
543 // overrun). As a rule, if the fault happens below current stack pointer,
544 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
545 // application (see Linux kernel fault.c).
546 //
547 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
548 // stack overflow detection.
549 //
550 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
551 // not use MAP_GROWSDOWN.
552 //
553 // To get around the problem and allow stack banging on Linux, we need to
554 // manually expand thread stack after receiving the SIGSEGV.
555 //
556 // There are two ways to expand thread stack to address "bottom", we used
557 // both of them in JVM before 1.5:
558 // 1. adjust stack pointer first so that it is below "bottom", and then
559 // touch "bottom"
560 // 2. mmap() the page in question
561 //
562 // Now alternate signal stack is gone, it's harder to use 2. For instance,
563 // if current sp is already near the lower end of page 101, and we need to
564 // call mmap() to map page 100, it is possible that part of the mmap() frame
565 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
566 // That will destroy the mmap() frame and cause VM to crash.
567 //
568 // The following code works by adjusting sp first, then accessing the "bottom"
569 // page to force a page fault. Linux kernel will then automatically expand the
570 // stack mapping.
571 //
572 // _expand_stack_to() assumes its frame size is less than page size, which
573 // should always be true if the function is not inlined.
574
575 static void NOINLINE _expand_stack_to(address bottom) {
576 address sp;
577 size_t size;
578 volatile char *p;
579
580 // Adjust bottom to point to the largest address within the same page, it
581 // gives us a one-page buffer if alloca() allocates slightly more memory.
582 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
583 bottom += os::Linux::page_size() - 1;
584
585 // sp might be slightly above current stack pointer; if that's the case, we
586 // will alloca() a little more space than necessary, which is OK. Don't use
587 // os::current_stack_pointer(), as its result can be slightly below current
588 // stack pointer, causing us to not alloca enough to reach "bottom".
589 sp = (address)&sp;
590
591 if (sp > bottom) {
592 size = sp - bottom;
593 p = (volatile char *)alloca(size);
594 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
595 p[0] = '\0';
596 }
597 }
598
599 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
600 assert(t!=NULL, "just checking");
601 assert(t->osthread()->expanding_stack(), "expand should be set");
602 assert(t->stack_base() != NULL, "stack_base was not initialized");
603
604 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
605 sigset_t mask_all, old_sigset;
606 sigfillset(&mask_all);
607 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
608 _expand_stack_to(addr);
609 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
610 return true;
611 }
612 return false;
613 }
614
615 //////////////////////////////////////////////////////////////////////////////
616 // create new thread
617
618 // Thread start routine for all newly created threads
619 static void *thread_native_entry(Thread *thread) {
620 // Try to randomize the cache line index of hot stack frames.
621 // This helps when threads of the same stack traces evict each other's
622 // cache lines. The threads can be either from the same JVM instance, or
623 // from different JVM instances. The benefit is especially true for
624 // processors with hyperthreading technology.
625 static int counter = 0;
626 int pid = os::current_process_id();
627 alloca(((pid ^ counter++) & 7) * 128);
628
629 thread->initialize_thread_current();
630
631 OSThread* osthread = thread->osthread();
632 Monitor* sync = osthread->startThread_lock();
633
634 osthread->set_thread_id(os::current_thread_id());
635
636 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
637 os::current_thread_id(), (uintx) pthread_self());
638
639 if (UseNUMA) {
640 int lgrp_id = os::numa_get_group_id();
641 if (lgrp_id != -1) {
642 thread->set_lgrp_id(lgrp_id);
643 }
644 }
645 // initialize signal mask for this thread
646 os::Linux::hotspot_sigmask(thread);
647
648 // initialize floating point control register
649 os::Linux::init_thread_fpu_state();
650
651 // handshaking with parent thread
652 {
653 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
654
655 // notify parent thread
656 osthread->set_state(INITIALIZED);
657 sync->notify_all();
658
659 // wait until os::start_thread()
660 while (osthread->get_state() == INITIALIZED) {
661 sync->wait(Mutex::_no_safepoint_check_flag);
662 }
663 }
664
665 // call one more level start routine
666 thread->run();
667
668 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
669 os::current_thread_id(), (uintx) pthread_self());
670
671 // If a thread has not deleted itself ("delete this") as part of its
672 // termination sequence, we have to ensure thread-local-storage is
673 // cleared before we actually terminate. No threads should ever be
674 // deleted asynchronously with respect to their termination.
675 if (Thread::current_or_null_safe() != NULL) {
676 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
677 thread->clear_thread_current();
678 }
679
680 return 0;
681 }
682
683 bool os::create_thread(Thread* thread, ThreadType thr_type,
684 size_t req_stack_size) {
685 assert(thread->osthread() == NULL, "caller responsible");
686
687 // Allocate the OSThread object
688 OSThread* osthread = new OSThread(NULL, NULL);
689 if (osthread == NULL) {
690 return false;
691 }
692
693 // set the correct thread state
694 osthread->set_thread_type(thr_type);
695
696 // Initial state is ALLOCATED but not INITIALIZED
697 osthread->set_state(ALLOCATED);
698
699 thread->set_osthread(osthread);
700
701 // init thread attributes
702 pthread_attr_t attr;
703 pthread_attr_init(&attr);
704 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
705
706 // Calculate stack size if it's not specified by caller.
707 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
708 // In the Linux NPTL pthread implementation the guard size mechanism
709 // is not implemented properly. The posix standard requires adding
710 // the size of the guard pages to the stack size, instead Linux
711 // takes the space out of 'stacksize'. Thus we adapt the requested
712 // stack_size by the size of the guard pages to mimick proper
713 // behaviour. However, be careful not to end up with a size
714 // of zero due to overflow. Don't add the guard page in that case.
715 size_t guard_size = os::Linux::default_guard_size(thr_type);
716 if (stack_size <= SIZE_MAX - guard_size) {
717 stack_size += guard_size;
718 }
719 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
720
721 int status = pthread_attr_setstacksize(&attr, stack_size);
722 assert_status(status == 0, status, "pthread_attr_setstacksize");
723
724 // Configure glibc guard page.
725 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
726
727 ThreadState state;
728
729 {
730 pthread_t tid;
731 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
732
733 char buf[64];
734 if (ret == 0) {
735 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
736 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
737 } else {
738 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
739 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
740 }
741
742 pthread_attr_destroy(&attr);
743
744 if (ret != 0) {
745 // Need to clean up stuff we've allocated so far
746 thread->set_osthread(NULL);
747 delete osthread;
748 return false;
749 }
750
751 // Store pthread info into the OSThread
752 osthread->set_pthread_id(tid);
753
754 // Wait until child thread is either initialized or aborted
755 {
756 Monitor* sync_with_child = osthread->startThread_lock();
757 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
758 while ((state = osthread->get_state()) == ALLOCATED) {
759 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
760 }
761 }
762 }
763
764 // Aborted due to thread limit being reached
765 if (state == ZOMBIE) {
766 thread->set_osthread(NULL);
767 delete osthread;
768 return false;
769 }
770
771 // The thread is returned suspended (in state INITIALIZED),
772 // and is started higher up in the call chain
773 assert(state == INITIALIZED, "race condition");
774 return true;
775 }
776
777 /////////////////////////////////////////////////////////////////////////////
778 // attach existing thread
779
780 // bootstrap the main thread
781 bool os::create_main_thread(JavaThread* thread) {
782 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
783 return create_attached_thread(thread);
784 }
785
786 bool os::create_attached_thread(JavaThread* thread) {
787 #ifdef ASSERT
788 thread->verify_not_published();
789 #endif
790
791 // Allocate the OSThread object
792 OSThread* osthread = new OSThread(NULL, NULL);
793
794 if (osthread == NULL) {
795 return false;
796 }
797
798 // Store pthread info into the OSThread
799 osthread->set_thread_id(os::Linux::gettid());
800 osthread->set_pthread_id(::pthread_self());
801
802 // initialize floating point control register
803 os::Linux::init_thread_fpu_state();
804
805 // Initial thread state is RUNNABLE
806 osthread->set_state(RUNNABLE);
807
808 thread->set_osthread(osthread);
809
810 if (UseNUMA) {
811 int lgrp_id = os::numa_get_group_id();
812 if (lgrp_id != -1) {
813 thread->set_lgrp_id(lgrp_id);
814 }
815 }
816
817 if (os::Linux::is_initial_thread()) {
818 // If current thread is initial thread, its stack is mapped on demand,
819 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
820 // the entire stack region to avoid SEGV in stack banging.
821 // It is also useful to get around the heap-stack-gap problem on SuSE
822 // kernel (see 4821821 for details). We first expand stack to the top
823 // of yellow zone, then enable stack yellow zone (order is significant,
824 // enabling yellow zone first will crash JVM on SuSE Linux), so there
825 // is no gap between the last two virtual memory regions.
826
827 JavaThread *jt = (JavaThread *)thread;
828 address addr = jt->stack_reserved_zone_base();
829 assert(addr != NULL, "initialization problem?");
830 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
831
832 osthread->set_expanding_stack();
833 os::Linux::manually_expand_stack(jt, addr);
834 osthread->clear_expanding_stack();
835 }
836
837 // initialize signal mask for this thread
838 // and save the caller's signal mask
839 os::Linux::hotspot_sigmask(thread);
840
841 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
842 os::current_thread_id(), (uintx) pthread_self());
843
844 return true;
845 }
846
847 void os::pd_start_thread(Thread* thread) {
848 OSThread * osthread = thread->osthread();
849 assert(osthread->get_state() != INITIALIZED, "just checking");
850 Monitor* sync_with_child = osthread->startThread_lock();
851 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
852 sync_with_child->notify();
853 }
854
855 // Free Linux resources related to the OSThread
856 void os::free_thread(OSThread* osthread) {
857 assert(osthread != NULL, "osthread not set");
858
859 // We are told to free resources of the argument thread,
860 // but we can only really operate on the current thread.
861 assert(Thread::current()->osthread() == osthread,
862 "os::free_thread but not current thread");
863
864 #ifdef ASSERT
865 sigset_t current;
866 sigemptyset(¤t);
867 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
868 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
869 #endif
870
871 // Restore caller's signal mask
872 sigset_t sigmask = osthread->caller_sigmask();
873 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
874
875 delete osthread;
876 }
877
878 //////////////////////////////////////////////////////////////////////////////
879 // initial thread
880
881 // Check if current thread is the initial thread, similar to Solaris thr_main.
882 bool os::Linux::is_initial_thread(void) {
883 char dummy;
884 // If called before init complete, thread stack bottom will be null.
885 // Can be called if fatal error occurs before initialization.
886 if (initial_thread_stack_bottom() == NULL) return false;
887 assert(initial_thread_stack_bottom() != NULL &&
888 initial_thread_stack_size() != 0,
889 "os::init did not locate initial thread's stack region");
890 if ((address)&dummy >= initial_thread_stack_bottom() &&
891 (address)&dummy < initial_thread_stack_bottom() + 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 initial thread stack. This special handling of initial 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 stack_size -= 2 * page_size();
947
948 // Try to figure out where the stack base (top) is. This is harder.
949 //
950 // When an application is started, glibc saves the initial stack pointer in
951 // a global variable "__libc_stack_end", which is then used by system
952 // libraries. __libc_stack_end should be pretty close to stack top. The
953 // variable is available since the very early days. However, because it is
954 // a private interface, it could disappear in the future.
955 //
956 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
957 // to __libc_stack_end, it is very close to stack top, but isn't the real
958 // stack top. Note that /proc may not exist if VM is running as a chroot
959 // program, so reading /proc/<pid>/stat could fail. Also the contents of
960 // /proc/<pid>/stat could change in the future (though unlikely).
961 //
962 // We try __libc_stack_end first. If that doesn't work, look for
963 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
964 // as a hint, which should work well in most cases.
965
966 uintptr_t stack_start;
967
968 // try __libc_stack_end first
969 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
970 if (p && *p) {
971 stack_start = *p;
972 } else {
973 // see if we can get the start_stack field from /proc/self/stat
974 FILE *fp;
975 int pid;
976 char state;
977 int ppid;
978 int pgrp;
979 int session;
980 int nr;
981 int tpgrp;
982 unsigned long flags;
983 unsigned long minflt;
984 unsigned long cminflt;
985 unsigned long majflt;
986 unsigned long cmajflt;
987 unsigned long utime;
988 unsigned long stime;
989 long cutime;
990 long cstime;
991 long prio;
992 long nice;
993 long junk;
994 long it_real;
995 uintptr_t start;
996 uintptr_t vsize;
997 intptr_t rss;
998 uintptr_t rsslim;
999 uintptr_t scodes;
1000 uintptr_t ecode;
1001 int i;
1002
1003 // Figure what the primordial thread stack base is. Code is inspired
1004 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1005 // followed by command name surrounded by parentheses, state, etc.
1006 char stat[2048];
1007 int statlen;
1008
1009 fp = fopen("/proc/self/stat", "r");
1010 if (fp) {
1011 statlen = fread(stat, 1, 2047, fp);
1012 stat[statlen] = '\0';
1013 fclose(fp);
1014
1015 // Skip pid and the command string. Note that we could be dealing with
1016 // weird command names, e.g. user could decide to rename java launcher
1017 // to "java 1.4.2 :)", then the stat file would look like
1018 // 1234 (java 1.4.2 :)) R ... ...
1019 // We don't really need to know the command string, just find the last
1020 // occurrence of ")" and then start parsing from there. See bug 4726580.
1021 char * s = strrchr(stat, ')');
1022
1023 i = 0;
1024 if (s) {
1025 // Skip blank chars
1026 do { s++; } while (s && isspace(*s));
1027
1028 #define _UFM UINTX_FORMAT
1029 #define _DFM INTX_FORMAT
1030
1031 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1032 // 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
1033 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,
1034 &state, // 3 %c
1035 &ppid, // 4 %d
1036 &pgrp, // 5 %d
1037 &session, // 6 %d
1038 &nr, // 7 %d
1039 &tpgrp, // 8 %d
1040 &flags, // 9 %lu
1041 &minflt, // 10 %lu
1042 &cminflt, // 11 %lu
1043 &majflt, // 12 %lu
1044 &cmajflt, // 13 %lu
1045 &utime, // 14 %lu
1046 &stime, // 15 %lu
1047 &cutime, // 16 %ld
1048 &cstime, // 17 %ld
1049 &prio, // 18 %ld
1050 &nice, // 19 %ld
1051 &junk, // 20 %ld
1052 &it_real, // 21 %ld
1053 &start, // 22 UINTX_FORMAT
1054 &vsize, // 23 UINTX_FORMAT
1055 &rss, // 24 INTX_FORMAT
1056 &rsslim, // 25 UINTX_FORMAT
1057 &scodes, // 26 UINTX_FORMAT
1058 &ecode, // 27 UINTX_FORMAT
1059 &stack_start); // 28 UINTX_FORMAT
1060 }
1061
1062 #undef _UFM
1063 #undef _DFM
1064
1065 if (i != 28 - 2) {
1066 assert(false, "Bad conversion from /proc/self/stat");
1067 // product mode - assume we are the initial thread, good luck in the
1068 // embedded case.
1069 warning("Can't detect initial thread stack location - bad conversion");
1070 stack_start = (uintptr_t) &rlim;
1071 }
1072 } else {
1073 // For some reason we can't open /proc/self/stat (for example, running on
1074 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1075 // most cases, so don't abort:
1076 warning("Can't detect initial thread stack location - no /proc/self/stat");
1077 stack_start = (uintptr_t) &rlim;
1078 }
1079 }
1080
1081 // Now we have a pointer (stack_start) very close to the stack top, the
1082 // next thing to do is to figure out the exact location of stack top. We
1083 // can find out the virtual memory area that contains stack_start by
1084 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1085 // and its upper limit is the real stack top. (again, this would fail if
1086 // running inside chroot, because /proc may not exist.)
1087
1088 uintptr_t stack_top;
1089 address low, high;
1090 if (find_vma((address)stack_start, &low, &high)) {
1091 // success, "high" is the true stack top. (ignore "low", because initial
1092 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1093 stack_top = (uintptr_t)high;
1094 } else {
1095 // failed, likely because /proc/self/maps does not exist
1096 warning("Can't detect initial thread stack location - find_vma failed");
1097 // best effort: stack_start is normally within a few pages below the real
1098 // stack top, use it as stack top, and reduce stack size so we won't put
1099 // guard page outside stack.
1100 stack_top = stack_start;
1101 stack_size -= 16 * page_size();
1102 }
1103
1104 // stack_top could be partially down the page so align it
1105 stack_top = align_up(stack_top, page_size());
1106
1107 // Allowed stack value is minimum of max_size and what we derived from rlimit
1108 if (max_size > 0) {
1109 _initial_thread_stack_size = MIN2(max_size, stack_size);
1110 } else {
1111 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1112 // clamp it at 8MB as we do on Solaris
1113 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1114 }
1115 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1116 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1117
1118 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1119
1120 if (log_is_enabled(Info, os, thread)) {
1121 // See if we seem to be on primordial process thread
1122 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1123 uintptr_t(&rlim) < stack_top;
1124
1125 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1126 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1127 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1128 stack_top, intptr_t(_initial_thread_stack_bottom));
1129 }
1130 }
1131
1132 ////////////////////////////////////////////////////////////////////////////////
1133 // time support
1134
1135 // Time since start-up in seconds to a fine granularity.
1136 // Used by VMSelfDestructTimer and the MemProfiler.
1137 double os::elapsedTime() {
1138
1139 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1140 }
1141
1142 jlong os::elapsed_counter() {
1143 return javaTimeNanos() - initial_time_count;
1144 }
1145
1146 jlong os::elapsed_frequency() {
1147 return NANOSECS_PER_SEC; // nanosecond resolution
1148 }
1149
1150 bool os::supports_vtime() { return true; }
1151 bool os::enable_vtime() { return false; }
1152 bool os::vtime_enabled() { return false; }
1153
1154 double os::elapsedVTime() {
1155 struct rusage usage;
1156 int retval = getrusage(RUSAGE_THREAD, &usage);
1157 if (retval == 0) {
1158 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);
1159 } else {
1160 // better than nothing, but not much
1161 return elapsedTime();
1162 }
1163 }
1164
1165 jlong os::javaTimeMillis() {
1166 timeval time;
1167 int status = gettimeofday(&time, NULL);
1168 assert(status != -1, "linux error");
1169 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1170 }
1171
1172 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1173 timeval time;
1174 int status = gettimeofday(&time, NULL);
1175 assert(status != -1, "linux error");
1176 seconds = jlong(time.tv_sec);
1177 nanos = jlong(time.tv_usec) * 1000;
1178 }
1179
1180
1181 #ifndef CLOCK_MONOTONIC
1182 #define CLOCK_MONOTONIC (1)
1183 #endif
1184
1185 void os::Linux::clock_init() {
1186 // we do dlopen's in this particular order due to bug in linux
1187 // dynamical loader (see 6348968) leading to crash on exit
1188 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1189 if (handle == NULL) {
1190 handle = dlopen("librt.so", RTLD_LAZY);
1191 }
1192
1193 if (handle) {
1194 int (*clock_getres_func)(clockid_t, struct timespec*) =
1195 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1196 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1197 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1198 if (clock_getres_func && clock_gettime_func) {
1199 // See if monotonic clock is supported by the kernel. Note that some
1200 // early implementations simply return kernel jiffies (updated every
1201 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1202 // for nano time (though the monotonic property is still nice to have).
1203 // It's fixed in newer kernels, however clock_getres() still returns
1204 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1205 // resolution for now. Hopefully as people move to new kernels, this
1206 // won't be a problem.
1207 struct timespec res;
1208 struct timespec tp;
1209 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1210 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1211 // yes, monotonic clock is supported
1212 _clock_gettime = clock_gettime_func;
1213 return;
1214 } else {
1215 // close librt if there is no monotonic clock
1216 dlclose(handle);
1217 }
1218 }
1219 }
1220 warning("No monotonic clock was available - timed services may " \
1221 "be adversely affected if the time-of-day clock changes");
1222 }
1223
1224 #ifndef SYS_clock_getres
1225 #if defined(X86) || defined(PPC64) || defined(S390)
1226 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1227 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1228 #else
1229 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1230 #define sys_clock_getres(x,y) -1
1231 #endif
1232 #else
1233 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1234 #endif
1235
1236 void os::Linux::fast_thread_clock_init() {
1237 if (!UseLinuxPosixThreadCPUClocks) {
1238 return;
1239 }
1240 clockid_t clockid;
1241 struct timespec tp;
1242 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1243 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1244
1245 // Switch to using fast clocks for thread cpu time if
1246 // the sys_clock_getres() returns 0 error code.
1247 // Note, that some kernels may support the current thread
1248 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1249 // returned by the pthread_getcpuclockid().
1250 // If the fast Posix clocks are supported then the sys_clock_getres()
1251 // must return at least tp.tv_sec == 0 which means a resolution
1252 // better than 1 sec. This is extra check for reliability.
1253
1254 if (pthread_getcpuclockid_func &&
1255 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1256 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1257 _supports_fast_thread_cpu_time = true;
1258 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1259 }
1260 }
1261
1262 jlong os::javaTimeNanos() {
1263 if (os::supports_monotonic_clock()) {
1264 struct timespec tp;
1265 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1266 assert(status == 0, "gettime error");
1267 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1268 return result;
1269 } else {
1270 timeval time;
1271 int status = gettimeofday(&time, NULL);
1272 assert(status != -1, "linux error");
1273 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1274 return 1000 * usecs;
1275 }
1276 }
1277
1278 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1279 if (os::supports_monotonic_clock()) {
1280 info_ptr->max_value = ALL_64_BITS;
1281
1282 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1283 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1284 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1285 } else {
1286 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1287 info_ptr->max_value = ALL_64_BITS;
1288
1289 // gettimeofday is a real time clock so it skips
1290 info_ptr->may_skip_backward = true;
1291 info_ptr->may_skip_forward = true;
1292 }
1293
1294 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1295 }
1296
1297 // Return the real, user, and system times in seconds from an
1298 // arbitrary fixed point in the past.
1299 bool os::getTimesSecs(double* process_real_time,
1300 double* process_user_time,
1301 double* process_system_time) {
1302 struct tms ticks;
1303 clock_t real_ticks = times(&ticks);
1304
1305 if (real_ticks == (clock_t) (-1)) {
1306 return false;
1307 } else {
1308 double ticks_per_second = (double) clock_tics_per_sec;
1309 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1310 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1311 *process_real_time = ((double) real_ticks) / ticks_per_second;
1312
1313 return true;
1314 }
1315 }
1316
1317
1318 char * os::local_time_string(char *buf, size_t buflen) {
1319 struct tm t;
1320 time_t long_time;
1321 time(&long_time);
1322 localtime_r(&long_time, &t);
1323 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1324 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1325 t.tm_hour, t.tm_min, t.tm_sec);
1326 return buf;
1327 }
1328
1329 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1330 return localtime_r(clock, res);
1331 }
1332
1333 ////////////////////////////////////////////////////////////////////////////////
1334 // runtime exit support
1335
1336 // Note: os::shutdown() might be called very early during initialization, or
1337 // called from signal handler. Before adding something to os::shutdown(), make
1338 // sure it is async-safe and can handle partially initialized VM.
1339 void os::shutdown() {
1340
1341 // allow PerfMemory to attempt cleanup of any persistent resources
1342 perfMemory_exit();
1343
1344 // needs to remove object in file system
1345 AttachListener::abort();
1346
1347 // flush buffered output, finish log files
1348 ostream_abort();
1349
1350 // Check for abort hook
1351 abort_hook_t abort_hook = Arguments::abort_hook();
1352 if (abort_hook != NULL) {
1353 abort_hook();
1354 }
1355
1356 }
1357
1358 // Note: os::abort() might be called very early during initialization, or
1359 // called from signal handler. Before adding something to os::abort(), make
1360 // sure it is async-safe and can handle partially initialized VM.
1361 void os::abort(bool dump_core, void* siginfo, const void* context) {
1362 os::shutdown();
1363 if (dump_core) {
1364 #ifndef PRODUCT
1365 fdStream out(defaultStream::output_fd());
1366 out.print_raw("Current thread is ");
1367 char buf[16];
1368 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1369 out.print_raw_cr(buf);
1370 out.print_raw_cr("Dumping core ...");
1371 #endif
1372 ::abort(); // dump core
1373 }
1374
1375 ::exit(1);
1376 }
1377
1378 // Die immediately, no exit hook, no abort hook, no cleanup.
1379 void os::die() {
1380 ::abort();
1381 }
1382
1383
1384 // This method is a copy of JDK's sysGetLastErrorString
1385 // from src/solaris/hpi/src/system_md.c
1386
1387 size_t os::lasterror(char *buf, size_t len) {
1388 if (errno == 0) return 0;
1389
1390 const char *s = os::strerror(errno);
1391 size_t n = ::strlen(s);
1392 if (n >= len) {
1393 n = len - 1;
1394 }
1395 ::strncpy(buf, s, n);
1396 buf[n] = '\0';
1397 return n;
1398 }
1399
1400 // thread_id is kernel thread id (similar to Solaris LWP id)
1401 intx os::current_thread_id() { return os::Linux::gettid(); }
1402 int os::current_process_id() {
1403 return ::getpid();
1404 }
1405
1406 // DLL functions
1407
1408 const char* os::dll_file_extension() { return ".so"; }
1409
1410 // This must be hard coded because it's the system's temporary
1411 // directory not the java application's temp directory, ala java.io.tmpdir.
1412 const char* os::get_temp_directory() { return "/tmp"; }
1413
1414 static bool file_exists(const char* filename) {
1415 struct stat statbuf;
1416 if (filename == NULL || strlen(filename) == 0) {
1417 return false;
1418 }
1419 return os::stat(filename, &statbuf) == 0;
1420 }
1421
1422 bool os::dll_build_name(char* buffer, size_t buflen,
1423 const char* pname, const char* fname) {
1424 bool retval = false;
1425 // Copied from libhpi
1426 const size_t pnamelen = pname ? strlen(pname) : 0;
1427
1428 // Return error on buffer overflow.
1429 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1430 return retval;
1431 }
1432
1433 if (pnamelen == 0) {
1434 snprintf(buffer, buflen, "lib%s.so", fname);
1435 retval = true;
1436 } else if (strchr(pname, *os::path_separator()) != NULL) {
1437 int n;
1438 char** pelements = split_path(pname, &n);
1439 if (pelements == NULL) {
1440 return false;
1441 }
1442 for (int i = 0; i < n; i++) {
1443 // Really shouldn't be NULL, but check can't hurt
1444 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1445 continue; // skip the empty path values
1446 }
1447 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1448 if (file_exists(buffer)) {
1449 retval = true;
1450 break;
1451 }
1452 }
1453 // release the storage
1454 for (int i = 0; i < n; i++) {
1455 if (pelements[i] != NULL) {
1456 FREE_C_HEAP_ARRAY(char, pelements[i]);
1457 }
1458 }
1459 if (pelements != NULL) {
1460 FREE_C_HEAP_ARRAY(char*, pelements);
1461 }
1462 } else {
1463 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1464 retval = true;
1465 }
1466 return retval;
1467 }
1468
1469 // check if addr is inside libjvm.so
1470 bool os::address_is_in_vm(address addr) {
1471 static address libjvm_base_addr;
1472 Dl_info dlinfo;
1473
1474 if (libjvm_base_addr == NULL) {
1475 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1476 libjvm_base_addr = (address)dlinfo.dli_fbase;
1477 }
1478 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1479 }
1480
1481 if (dladdr((void *)addr, &dlinfo) != 0) {
1482 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1483 }
1484
1485 return false;
1486 }
1487
1488 bool os::dll_address_to_function_name(address addr, char *buf,
1489 int buflen, int *offset,
1490 bool demangle) {
1491 // buf is not optional, but offset is optional
1492 assert(buf != NULL, "sanity check");
1493
1494 Dl_info dlinfo;
1495
1496 if (dladdr((void*)addr, &dlinfo) != 0) {
1497 // see if we have a matching symbol
1498 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1499 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1500 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1501 }
1502 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1503 return true;
1504 }
1505 // no matching symbol so try for just file info
1506 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1507 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1508 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1509 return true;
1510 }
1511 }
1512 }
1513
1514 buf[0] = '\0';
1515 if (offset != NULL) *offset = -1;
1516 return false;
1517 }
1518
1519 struct _address_to_library_name {
1520 address addr; // input : memory address
1521 size_t buflen; // size of fname
1522 char* fname; // output: library name
1523 address base; // library base addr
1524 };
1525
1526 static int address_to_library_name_callback(struct dl_phdr_info *info,
1527 size_t size, void *data) {
1528 int i;
1529 bool found = false;
1530 address libbase = NULL;
1531 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1532
1533 // iterate through all loadable segments
1534 for (i = 0; i < info->dlpi_phnum; i++) {
1535 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1536 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1537 // base address of a library is the lowest address of its loaded
1538 // segments.
1539 if (libbase == NULL || libbase > segbase) {
1540 libbase = segbase;
1541 }
1542 // see if 'addr' is within current segment
1543 if (segbase <= d->addr &&
1544 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1545 found = true;
1546 }
1547 }
1548 }
1549
1550 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1551 // so dll_address_to_library_name() can fall through to use dladdr() which
1552 // can figure out executable name from argv[0].
1553 if (found && info->dlpi_name && info->dlpi_name[0]) {
1554 d->base = libbase;
1555 if (d->fname) {
1556 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1557 }
1558 return 1;
1559 }
1560 return 0;
1561 }
1562
1563 bool os::dll_address_to_library_name(address addr, char* buf,
1564 int buflen, int* offset) {
1565 // buf is not optional, but offset is optional
1566 assert(buf != NULL, "sanity check");
1567
1568 Dl_info dlinfo;
1569 struct _address_to_library_name data;
1570
1571 // There is a bug in old glibc dladdr() implementation that it could resolve
1572 // to wrong library name if the .so file has a base address != NULL. Here
1573 // we iterate through the program headers of all loaded libraries to find
1574 // out which library 'addr' really belongs to. This workaround can be
1575 // removed once the minimum requirement for glibc is moved to 2.3.x.
1576 data.addr = addr;
1577 data.fname = buf;
1578 data.buflen = buflen;
1579 data.base = NULL;
1580 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1581
1582 if (rslt) {
1583 // buf already contains library name
1584 if (offset) *offset = addr - data.base;
1585 return true;
1586 }
1587 if (dladdr((void*)addr, &dlinfo) != 0) {
1588 if (dlinfo.dli_fname != NULL) {
1589 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1590 }
1591 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1592 *offset = addr - (address)dlinfo.dli_fbase;
1593 }
1594 return true;
1595 }
1596
1597 buf[0] = '\0';
1598 if (offset) *offset = -1;
1599 return false;
1600 }
1601
1602 // Loads .dll/.so and
1603 // in case of error it checks if .dll/.so was built for the
1604 // same architecture as Hotspot is running on
1605
1606
1607 // Remember the stack's state. The Linux dynamic linker will change
1608 // the stack to 'executable' at most once, so we must safepoint only once.
1609 bool os::Linux::_stack_is_executable = false;
1610
1611 // VM operation that loads a library. This is necessary if stack protection
1612 // of the Java stacks can be lost during loading the library. If we
1613 // do not stop the Java threads, they can stack overflow before the stacks
1614 // are protected again.
1615 class VM_LinuxDllLoad: public VM_Operation {
1616 private:
1617 const char *_filename;
1618 char *_ebuf;
1619 int _ebuflen;
1620 void *_lib;
1621 public:
1622 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1623 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1624 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1625 void doit() {
1626 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1627 os::Linux::_stack_is_executable = true;
1628 }
1629 void* loaded_library() { return _lib; }
1630 };
1631
1632 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1633 void * result = NULL;
1634 bool load_attempted = false;
1635
1636 // Check whether the library to load might change execution rights
1637 // of the stack. If they are changed, the protection of the stack
1638 // guard pages will be lost. We need a safepoint to fix this.
1639 //
1640 // See Linux man page execstack(8) for more info.
1641 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1642 if (!ElfFile::specifies_noexecstack(filename)) {
1643 if (!is_init_completed()) {
1644 os::Linux::_stack_is_executable = true;
1645 // This is OK - No Java threads have been created yet, and hence no
1646 // stack guard pages to fix.
1647 //
1648 // This should happen only when you are building JDK7 using a very
1649 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1650 //
1651 // Dynamic loader will make all stacks executable after
1652 // this function returns, and will not do that again.
1653 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1654 } else {
1655 warning("You have loaded library %s which might have disabled stack guard. "
1656 "The VM will try to fix the stack guard now.\n"
1657 "It's highly recommended that you fix the library with "
1658 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1659 filename);
1660
1661 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1662 JavaThread *jt = JavaThread::current();
1663 if (jt->thread_state() != _thread_in_native) {
1664 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1665 // that requires ExecStack. Cannot enter safe point. Let's give up.
1666 warning("Unable to fix stack guard. Giving up.");
1667 } else {
1668 if (!LoadExecStackDllInVMThread) {
1669 // This is for the case where the DLL has an static
1670 // constructor function that executes JNI code. We cannot
1671 // load such DLLs in the VMThread.
1672 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1673 }
1674
1675 ThreadInVMfromNative tiv(jt);
1676 debug_only(VMNativeEntryWrapper vew;)
1677
1678 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1679 VMThread::execute(&op);
1680 if (LoadExecStackDllInVMThread) {
1681 result = op.loaded_library();
1682 }
1683 load_attempted = true;
1684 }
1685 }
1686 }
1687 }
1688
1689 if (!load_attempted) {
1690 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691 }
1692
1693 if (result != NULL) {
1694 // Successful loading
1695 return result;
1696 }
1697
1698 Elf32_Ehdr elf_head;
1699 int diag_msg_max_length=ebuflen-strlen(ebuf);
1700 char* diag_msg_buf=ebuf+strlen(ebuf);
1701
1702 if (diag_msg_max_length==0) {
1703 // No more space in ebuf for additional diagnostics message
1704 return NULL;
1705 }
1706
1707
1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1709
1710 if (file_descriptor < 0) {
1711 // Can't open library, report dlerror() message
1712 return NULL;
1713 }
1714
1715 bool failed_to_read_elf_head=
1716 (sizeof(elf_head)!=
1717 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1718
1719 ::close(file_descriptor);
1720 if (failed_to_read_elf_head) {
1721 // file i/o error - report dlerror() msg
1722 return NULL;
1723 }
1724
1725 typedef struct {
1726 Elf32_Half code; // Actual value as defined in elf.h
1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1728 unsigned char elf_class; // 32 or 64 bit
1729 unsigned char endianess; // MSB or LSB
1730 char* name; // String representation
1731 } arch_t;
1732
1733 #ifndef EM_486
1734 #define EM_486 6 /* Intel 80486 */
1735 #endif
1736 #ifndef EM_AARCH64
1737 #define EM_AARCH64 183 /* ARM AARCH64 */
1738 #endif
1739
1740 static const arch_t arch_array[]={
1741 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1742 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1743 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1744 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1745 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1746 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1747 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1748 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1749 #if defined(VM_LITTLE_ENDIAN)
1750 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1751 #else
1752 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1753 #endif
1754 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1755 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1756 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1757 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1758 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1759 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1760 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1761 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1762 };
1763
1764 #if (defined IA32)
1765 static Elf32_Half running_arch_code=EM_386;
1766 #elif (defined AMD64)
1767 static Elf32_Half running_arch_code=EM_X86_64;
1768 #elif (defined IA64)
1769 static Elf32_Half running_arch_code=EM_IA_64;
1770 #elif (defined __sparc) && (defined _LP64)
1771 static Elf32_Half running_arch_code=EM_SPARCV9;
1772 #elif (defined __sparc) && (!defined _LP64)
1773 static Elf32_Half running_arch_code=EM_SPARC;
1774 #elif (defined __powerpc64__)
1775 static Elf32_Half running_arch_code=EM_PPC64;
1776 #elif (defined __powerpc__)
1777 static Elf32_Half running_arch_code=EM_PPC;
1778 #elif (defined AARCH64)
1779 static Elf32_Half running_arch_code=EM_AARCH64;
1780 #elif (defined ARM)
1781 static Elf32_Half running_arch_code=EM_ARM;
1782 #elif (defined S390)
1783 static Elf32_Half running_arch_code=EM_S390;
1784 #elif (defined ALPHA)
1785 static Elf32_Half running_arch_code=EM_ALPHA;
1786 #elif (defined MIPSEL)
1787 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1788 #elif (defined PARISC)
1789 static Elf32_Half running_arch_code=EM_PARISC;
1790 #elif (defined MIPS)
1791 static Elf32_Half running_arch_code=EM_MIPS;
1792 #elif (defined M68K)
1793 static Elf32_Half running_arch_code=EM_68K;
1794 #else
1795 #error Method os::dll_load requires that one of following is defined:\
1796 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1797 #endif
1798
1799 // Identify compatability class for VM's architecture and library's architecture
1800 // Obtain string descriptions for architectures
1801
1802 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803 int running_arch_index=-1;
1804
1805 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806 if (running_arch_code == arch_array[i].code) {
1807 running_arch_index = i;
1808 }
1809 if (lib_arch.code == arch_array[i].code) {
1810 lib_arch.compat_class = arch_array[i].compat_class;
1811 lib_arch.name = arch_array[i].name;
1812 }
1813 }
1814
1815 assert(running_arch_index != -1,
1816 "Didn't find running architecture code (running_arch_code) in arch_array");
1817 if (running_arch_index == -1) {
1818 // Even though running architecture detection failed
1819 // we may still continue with reporting dlerror() message
1820 return NULL;
1821 }
1822
1823 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825 return NULL;
1826 }
1827
1828 #ifndef S390
1829 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831 return NULL;
1832 }
1833 #endif // !S390
1834
1835 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836 if (lib_arch.name!=NULL) {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839 lib_arch.name, arch_array[running_arch_index].name);
1840 } else {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843 lib_arch.code,
1844 arch_array[running_arch_index].name);
1845 }
1846 }
1847
1848 return NULL;
1849 }
1850
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852 int ebuflen) {
1853 void * result = ::dlopen(filename, RTLD_LAZY);
1854 if (result == NULL) {
1855 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856 ebuf[ebuflen-1] = '\0';
1857 }
1858 return result;
1859 }
1860
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862 int ebuflen) {
1863 void * result = NULL;
1864 if (LoadExecStackDllInVMThread) {
1865 result = dlopen_helper(filename, ebuf, ebuflen);
1866 }
1867
1868 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869 // library that requires an executable stack, or which does not have this
1870 // stack attribute set, dlopen changes the stack attribute to executable. The
1871 // read protection of the guard pages gets lost.
1872 //
1873 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874 // may have been queued at the same time.
1875
1876 if (!_stack_is_executable) {
1877 JavaThread *jt = Threads::first();
1878
1879 while (jt) {
1880 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1881 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1882 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1883 warning("Attempt to reguard stack yellow zone failed.");
1884 }
1885 }
1886 jt = jt->next();
1887 }
1888 }
1889
1890 return result;
1891 }
1892
1893 void* os::dll_lookup(void* handle, const char* name) {
1894 void* res = dlsym(handle, name);
1895 return res;
1896 }
1897
1898 void* os::get_default_process_handle() {
1899 return (void*)::dlopen(NULL, RTLD_LAZY);
1900 }
1901
1902 static bool _print_ascii_file(const char* filename, outputStream* st) {
1903 int fd = ::open(filename, O_RDONLY);
1904 if (fd == -1) {
1905 return false;
1906 }
1907
1908 char buf[33];
1909 int bytes;
1910 buf[32] = '\0';
1911 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1912 st->print_raw(buf, bytes);
1913 }
1914
1915 ::close(fd);
1916
1917 return true;
1918 }
1919
1920 void os::print_dll_info(outputStream *st) {
1921 st->print_cr("Dynamic libraries:");
1922
1923 char fname[32];
1924 pid_t pid = os::Linux::gettid();
1925
1926 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1927
1928 if (!_print_ascii_file(fname, st)) {
1929 st->print("Can not get library information for pid = %d\n", pid);
1930 }
1931 }
1932
1933 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1934 FILE *procmapsFile = NULL;
1935
1936 // Open the procfs maps file for the current process
1937 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1938 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1939 char line[PATH_MAX + 100];
1940
1941 // Read line by line from 'file'
1942 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1943 u8 base, top, offset, inode;
1944 char permissions[5];
1945 char device[6];
1946 char name[PATH_MAX + 1];
1947
1948 // Parse fields from line
1949 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1950 &base, &top, permissions, &offset, device, &inode, name);
1951
1952 // Filter by device id '00:00' so that we only get file system mapped files.
1953 if (strcmp(device, "00:00") != 0) {
1954
1955 // Call callback with the fields of interest
1956 if(callback(name, (address)base, (address)top, param)) {
1957 // Oops abort, callback aborted
1958 fclose(procmapsFile);
1959 return 1;
1960 }
1961 }
1962 }
1963 fclose(procmapsFile);
1964 }
1965 return 0;
1966 }
1967
1968 void os::print_os_info_brief(outputStream* st) {
1969 os::Linux::print_distro_info(st);
1970
1971 os::Posix::print_uname_info(st);
1972
1973 os::Linux::print_libversion_info(st);
1974
1975 }
1976
1977 void os::print_os_info(outputStream* st) {
1978 st->print("OS:");
1979
1980 os::Linux::print_distro_info(st);
1981
1982 os::Posix::print_uname_info(st);
1983
1984 // Print warning if unsafe chroot environment detected
1985 if (unsafe_chroot_detected) {
1986 st->print("WARNING!! ");
1987 st->print_cr("%s", unstable_chroot_error);
1988 }
1989
1990 os::Linux::print_libversion_info(st);
1991
1992 os::Posix::print_rlimit_info(st);
1993
1994 os::Posix::print_load_average(st);
1995
1996 os::Linux::print_full_memory_info(st);
1997 }
1998
1999 // Try to identify popular distros.
2000 // Most Linux distributions have a /etc/XXX-release file, which contains
2001 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2002 // file that also contains the OS version string. Some have more than one
2003 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2004 // /etc/redhat-release.), so the order is important.
2005 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2006 // their own specific XXX-release file as well as a redhat-release file.
2007 // Because of this the XXX-release file needs to be searched for before the
2008 // redhat-release file.
2009 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2010 // search for redhat-release / SuSE-release needs to be before lsb-release.
2011 // Since the lsb-release file is the new standard it needs to be searched
2012 // before the older style release files.
2013 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2014 // next to last resort. The os-release file is a new standard that contains
2015 // distribution information and the system-release file seems to be an old
2016 // standard that has been replaced by the lsb-release and os-release files.
2017 // Searching for the debian_version file is the last resort. It contains
2018 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2019 // "Debian " is printed before the contents of the debian_version file.
2020
2021 const char* distro_files[] = {
2022 "/etc/oracle-release",
2023 "/etc/mandriva-release",
2024 "/etc/mandrake-release",
2025 "/etc/sun-release",
2026 "/etc/redhat-release",
2027 "/etc/SuSE-release",
2028 "/etc/lsb-release",
2029 "/etc/turbolinux-release",
2030 "/etc/gentoo-release",
2031 "/etc/ltib-release",
2032 "/etc/angstrom-version",
2033 "/etc/system-release",
2034 "/etc/os-release",
2035 NULL };
2036
2037 void os::Linux::print_distro_info(outputStream* st) {
2038 for (int i = 0;; i++) {
2039 const char* file = distro_files[i];
2040 if (file == NULL) {
2041 break; // done
2042 }
2043 // If file prints, we found it.
2044 if (_print_ascii_file(file, st)) {
2045 return;
2046 }
2047 }
2048
2049 if (file_exists("/etc/debian_version")) {
2050 st->print("Debian ");
2051 _print_ascii_file("/etc/debian_version", st);
2052 } else {
2053 st->print("Linux");
2054 }
2055 st->cr();
2056 }
2057
2058 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2059 char buf[256];
2060 while (fgets(buf, sizeof(buf), fp)) {
2061 // Edit out extra stuff in expected format
2062 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2063 char* ptr = strstr(buf, "\""); // the name is in quotes
2064 if (ptr != NULL) {
2065 ptr++; // go beyond first quote
2066 char* nl = strchr(ptr, '\"');
2067 if (nl != NULL) *nl = '\0';
2068 strncpy(distro, ptr, length);
2069 } else {
2070 ptr = strstr(buf, "=");
2071 ptr++; // go beyond equals then
2072 char* nl = strchr(ptr, '\n');
2073 if (nl != NULL) *nl = '\0';
2074 strncpy(distro, ptr, length);
2075 }
2076 return;
2077 } else if (get_first_line) {
2078 char* nl = strchr(buf, '\n');
2079 if (nl != NULL) *nl = '\0';
2080 strncpy(distro, buf, length);
2081 return;
2082 }
2083 }
2084 // print last line and close
2085 char* nl = strchr(buf, '\n');
2086 if (nl != NULL) *nl = '\0';
2087 strncpy(distro, buf, length);
2088 }
2089
2090 static void parse_os_info(char* distro, size_t length, const char* file) {
2091 FILE* fp = fopen(file, "r");
2092 if (fp != NULL) {
2093 // if suse format, print out first line
2094 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2095 parse_os_info_helper(fp, distro, length, get_first_line);
2096 fclose(fp);
2097 }
2098 }
2099
2100 void os::get_summary_os_info(char* buf, size_t buflen) {
2101 for (int i = 0;; i++) {
2102 const char* file = distro_files[i];
2103 if (file == NULL) {
2104 break; // ran out of distro_files
2105 }
2106 if (file_exists(file)) {
2107 parse_os_info(buf, buflen, file);
2108 return;
2109 }
2110 }
2111 // special case for debian
2112 if (file_exists("/etc/debian_version")) {
2113 strncpy(buf, "Debian ", buflen);
2114 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2115 } else {
2116 strncpy(buf, "Linux", buflen);
2117 }
2118 }
2119
2120 void os::Linux::print_libversion_info(outputStream* st) {
2121 // libc, pthread
2122 st->print("libc:");
2123 st->print("%s ", os::Linux::glibc_version());
2124 st->print("%s ", os::Linux::libpthread_version());
2125 st->cr();
2126 }
2127
2128 void os::Linux::print_full_memory_info(outputStream* st) {
2129 st->print("\n/proc/meminfo:\n");
2130 _print_ascii_file("/proc/meminfo", st);
2131 st->cr();
2132 }
2133
2134 void os::print_memory_info(outputStream* st) {
2135
2136 st->print("Memory:");
2137 st->print(" %dk page", os::vm_page_size()>>10);
2138
2139 // values in struct sysinfo are "unsigned long"
2140 struct sysinfo si;
2141 sysinfo(&si);
2142
2143 st->print(", physical " UINT64_FORMAT "k",
2144 os::physical_memory() >> 10);
2145 st->print("(" UINT64_FORMAT "k free)",
2146 os::available_memory() >> 10);
2147 st->print(", swap " UINT64_FORMAT "k",
2148 ((jlong)si.totalswap * si.mem_unit) >> 10);
2149 st->print("(" UINT64_FORMAT "k free)",
2150 ((jlong)si.freeswap * si.mem_unit) >> 10);
2151 st->cr();
2152 }
2153
2154 // Print the first "model name" line and the first "flags" line
2155 // that we find and nothing more. We assume "model name" comes
2156 // before "flags" so if we find a second "model name", then the
2157 // "flags" field is considered missing.
2158 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2159 #if defined(IA32) || defined(AMD64)
2160 // Other platforms have less repetitive cpuinfo files
2161 FILE *fp = fopen("/proc/cpuinfo", "r");
2162 if (fp) {
2163 while (!feof(fp)) {
2164 if (fgets(buf, buflen, fp)) {
2165 // Assume model name comes before flags
2166 bool model_name_printed = false;
2167 if (strstr(buf, "model name") != NULL) {
2168 if (!model_name_printed) {
2169 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2170 st->print_raw(buf);
2171 model_name_printed = true;
2172 } else {
2173 // model name printed but not flags? Odd, just return
2174 fclose(fp);
2175 return true;
2176 }
2177 }
2178 // print the flags line too
2179 if (strstr(buf, "flags") != NULL) {
2180 st->print_raw(buf);
2181 fclose(fp);
2182 return true;
2183 }
2184 }
2185 }
2186 fclose(fp);
2187 }
2188 #endif // x86 platforms
2189 return false;
2190 }
2191
2192 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2193 // Only print the model name if the platform provides this as a summary
2194 if (!print_model_name_and_flags(st, buf, buflen)) {
2195 st->print("\n/proc/cpuinfo:\n");
2196 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2197 st->print_cr(" <Not Available>");
2198 }
2199 }
2200 }
2201
2202 #if defined(AMD64) || defined(IA32) || defined(X32)
2203 const char* search_string = "model name";
2204 #elif defined(PPC64)
2205 const char* search_string = "cpu";
2206 #elif defined(S390)
2207 const char* search_string = "processor";
2208 #elif defined(SPARC)
2209 const char* search_string = "cpu";
2210 #else
2211 const char* search_string = "Processor";
2212 #endif
2213
2214 // Parses the cpuinfo file for string representing the model name.
2215 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2216 FILE* fp = fopen("/proc/cpuinfo", "r");
2217 if (fp != NULL) {
2218 while (!feof(fp)) {
2219 char buf[256];
2220 if (fgets(buf, sizeof(buf), fp)) {
2221 char* start = strstr(buf, search_string);
2222 if (start != NULL) {
2223 char *ptr = start + strlen(search_string);
2224 char *end = buf + strlen(buf);
2225 while (ptr != end) {
2226 // skip whitespace and colon for the rest of the name.
2227 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2228 break;
2229 }
2230 ptr++;
2231 }
2232 if (ptr != end) {
2233 // reasonable string, get rid of newline and keep the rest
2234 char* nl = strchr(buf, '\n');
2235 if (nl != NULL) *nl = '\0';
2236 strncpy(cpuinfo, ptr, length);
2237 fclose(fp);
2238 return;
2239 }
2240 }
2241 }
2242 }
2243 fclose(fp);
2244 }
2245 // cpuinfo not found or parsing failed, just print generic string. The entire
2246 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2247 #if defined(AARCH64)
2248 strncpy(cpuinfo, "AArch64", length);
2249 #elif defined(AMD64)
2250 strncpy(cpuinfo, "x86_64", length);
2251 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2252 strncpy(cpuinfo, "ARM", length);
2253 #elif defined(IA32)
2254 strncpy(cpuinfo, "x86_32", length);
2255 #elif defined(IA64)
2256 strncpy(cpuinfo, "IA64", length);
2257 #elif defined(PPC)
2258 strncpy(cpuinfo, "PPC64", length);
2259 #elif defined(S390)
2260 strncpy(cpuinfo, "S390", length);
2261 #elif defined(SPARC)
2262 strncpy(cpuinfo, "sparcv9", length);
2263 #elif defined(ZERO_LIBARCH)
2264 strncpy(cpuinfo, ZERO_LIBARCH, length);
2265 #else
2266 strncpy(cpuinfo, "unknown", length);
2267 #endif
2268 }
2269
2270 static void print_signal_handler(outputStream* st, int sig,
2271 char* buf, size_t buflen);
2272
2273 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2274 st->print_cr("Signal Handlers:");
2275 print_signal_handler(st, SIGSEGV, buf, buflen);
2276 print_signal_handler(st, SIGBUS , buf, buflen);
2277 print_signal_handler(st, SIGFPE , buf, buflen);
2278 print_signal_handler(st, SIGPIPE, buf, buflen);
2279 print_signal_handler(st, SIGXFSZ, buf, buflen);
2280 print_signal_handler(st, SIGILL , buf, buflen);
2281 print_signal_handler(st, SR_signum, buf, buflen);
2282 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2283 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2284 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2285 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2286 #if defined(PPC64)
2287 print_signal_handler(st, SIGTRAP, buf, buflen);
2288 #endif
2289 }
2290
2291 static char saved_jvm_path[MAXPATHLEN] = {0};
2292
2293 // Find the full path to the current module, libjvm.so
2294 void os::jvm_path(char *buf, jint buflen) {
2295 // Error checking.
2296 if (buflen < MAXPATHLEN) {
2297 assert(false, "must use a large-enough buffer");
2298 buf[0] = '\0';
2299 return;
2300 }
2301 // Lazy resolve the path to current module.
2302 if (saved_jvm_path[0] != 0) {
2303 strcpy(buf, saved_jvm_path);
2304 return;
2305 }
2306
2307 char dli_fname[MAXPATHLEN];
2308 bool ret = dll_address_to_library_name(
2309 CAST_FROM_FN_PTR(address, os::jvm_path),
2310 dli_fname, sizeof(dli_fname), NULL);
2311 assert(ret, "cannot locate libjvm");
2312 char *rp = NULL;
2313 if (ret && dli_fname[0] != '\0') {
2314 rp = os::Posix::realpath(dli_fname, buf, buflen);
2315 }
2316 if (rp == NULL) {
2317 return;
2318 }
2319
2320 if (Arguments::sun_java_launcher_is_altjvm()) {
2321 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2322 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2323 // If "/jre/lib/" appears at the right place in the string, then
2324 // assume we are installed in a JDK and we're done. Otherwise, check
2325 // for a JAVA_HOME environment variable and fix up the path so it
2326 // looks like libjvm.so is installed there (append a fake suffix
2327 // hotspot/libjvm.so).
2328 const char *p = buf + strlen(buf) - 1;
2329 for (int count = 0; p > buf && count < 5; ++count) {
2330 for (--p; p > buf && *p != '/'; --p)
2331 /* empty */ ;
2332 }
2333
2334 if (strncmp(p, "/jre/lib/", 9) != 0) {
2335 // Look for JAVA_HOME in the environment.
2336 char* java_home_var = ::getenv("JAVA_HOME");
2337 if (java_home_var != NULL && java_home_var[0] != 0) {
2338 char* jrelib_p;
2339 int len;
2340
2341 // Check the current module name "libjvm.so".
2342 p = strrchr(buf, '/');
2343 if (p == NULL) {
2344 return;
2345 }
2346 assert(strstr(p, "/libjvm") == p, "invalid library name");
2347
2348 rp = os::Posix::realpath(java_home_var, buf, buflen);
2349 if (rp == NULL) {
2350 return;
2351 }
2352
2353 // determine if this is a legacy image or modules image
2354 // modules image doesn't have "jre" subdirectory
2355 len = strlen(buf);
2356 assert(len < buflen, "Ran out of buffer room");
2357 jrelib_p = buf + len;
2358 snprintf(jrelib_p, buflen-len, "/jre/lib");
2359 if (0 != access(buf, F_OK)) {
2360 snprintf(jrelib_p, buflen-len, "/lib");
2361 }
2362
2363 if (0 == access(buf, F_OK)) {
2364 // Use current module name "libjvm.so"
2365 len = strlen(buf);
2366 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2367 } else {
2368 // Go back to path of .so
2369 rp = os::Posix::realpath(dli_fname, buf, buflen);
2370 if (rp == NULL) {
2371 return;
2372 }
2373 }
2374 }
2375 }
2376 }
2377
2378 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2379 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2380 }
2381
2382 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2383 // no prefix required, not even "_"
2384 }
2385
2386 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2387 // no suffix required
2388 }
2389
2390 ////////////////////////////////////////////////////////////////////////////////
2391 // sun.misc.Signal support
2392
2393 static volatile jint sigint_count = 0;
2394
2395 static void UserHandler(int sig, void *siginfo, void *context) {
2396 // 4511530 - sem_post is serialized and handled by the manager thread. When
2397 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2398 // don't want to flood the manager thread with sem_post requests.
2399 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2400 return;
2401 }
2402
2403 // Ctrl-C is pressed during error reporting, likely because the error
2404 // handler fails to abort. Let VM die immediately.
2405 if (sig == SIGINT && VMError::is_error_reported()) {
2406 os::die();
2407 }
2408
2409 os::signal_notify(sig);
2410 }
2411
2412 void* os::user_handler() {
2413 return CAST_FROM_FN_PTR(void*, UserHandler);
2414 }
2415
2416 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2417 struct timespec ts;
2418 // Semaphore's are always associated with CLOCK_REALTIME
2419 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2420 // see unpackTime for discussion on overflow checking
2421 if (sec >= MAX_SECS) {
2422 ts.tv_sec += MAX_SECS;
2423 ts.tv_nsec = 0;
2424 } else {
2425 ts.tv_sec += sec;
2426 ts.tv_nsec += nsec;
2427 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2428 ts.tv_nsec -= NANOSECS_PER_SEC;
2429 ++ts.tv_sec; // note: this must be <= max_secs
2430 }
2431 }
2432
2433 return ts;
2434 }
2435
2436 extern "C" {
2437 typedef void (*sa_handler_t)(int);
2438 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2439 }
2440
2441 void* os::signal(int signal_number, void* handler) {
2442 struct sigaction sigAct, oldSigAct;
2443
2444 sigfillset(&(sigAct.sa_mask));
2445 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2446 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2447
2448 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2449 // -1 means registration failed
2450 return (void *)-1;
2451 }
2452
2453 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2454 }
2455
2456 void os::signal_raise(int signal_number) {
2457 ::raise(signal_number);
2458 }
2459
2460 // The following code is moved from os.cpp for making this
2461 // code platform specific, which it is by its very nature.
2462
2463 // Will be modified when max signal is changed to be dynamic
2464 int os::sigexitnum_pd() {
2465 return NSIG;
2466 }
2467
2468 // a counter for each possible signal value
2469 static volatile jint pending_signals[NSIG+1] = { 0 };
2470
2471 // Linux(POSIX) specific hand shaking semaphore.
2472 static sem_t sig_sem;
2473 static PosixSemaphore sr_semaphore;
2474
2475 void os::signal_init_pd() {
2476 // Initialize signal structures
2477 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2478
2479 // Initialize signal semaphore
2480 ::sem_init(&sig_sem, 0, 0);
2481 }
2482
2483 void os::signal_notify(int sig) {
2484 Atomic::inc(&pending_signals[sig]);
2485 ::sem_post(&sig_sem);
2486 }
2487
2488 static int check_pending_signals(bool wait) {
2489 Atomic::store(0, &sigint_count);
2490 for (;;) {
2491 for (int i = 0; i < NSIG + 1; i++) {
2492 jint n = pending_signals[i];
2493 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2494 return i;
2495 }
2496 }
2497 if (!wait) {
2498 return -1;
2499 }
2500 JavaThread *thread = JavaThread::current();
2501 ThreadBlockInVM tbivm(thread);
2502
2503 bool threadIsSuspended;
2504 do {
2505 thread->set_suspend_equivalent();
2506 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2507 ::sem_wait(&sig_sem);
2508
2509 // were we externally suspended while we were waiting?
2510 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2511 if (threadIsSuspended) {
2512 // The semaphore has been incremented, but while we were waiting
2513 // another thread suspended us. We don't want to continue running
2514 // while suspended because that would surprise the thread that
2515 // suspended us.
2516 ::sem_post(&sig_sem);
2517
2518 thread->java_suspend_self();
2519 }
2520 } while (threadIsSuspended);
2521 }
2522 }
2523
2524 int os::signal_lookup() {
2525 return check_pending_signals(false);
2526 }
2527
2528 int os::signal_wait() {
2529 return check_pending_signals(true);
2530 }
2531
2532 ////////////////////////////////////////////////////////////////////////////////
2533 // Virtual Memory
2534
2535 int os::vm_page_size() {
2536 // Seems redundant as all get out
2537 assert(os::Linux::page_size() != -1, "must call os::init");
2538 return os::Linux::page_size();
2539 }
2540
2541 // Solaris allocates memory by pages.
2542 int os::vm_allocation_granularity() {
2543 assert(os::Linux::page_size() != -1, "must call os::init");
2544 return os::Linux::page_size();
2545 }
2546
2547 // Rationale behind this function:
2548 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2549 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2550 // samples for JITted code. Here we create private executable mapping over the code cache
2551 // and then we can use standard (well, almost, as mapping can change) way to provide
2552 // info for the reporting script by storing timestamp and location of symbol
2553 void linux_wrap_code(char* base, size_t size) {
2554 static volatile jint cnt = 0;
2555
2556 if (!UseOprofile) {
2557 return;
2558 }
2559
2560 char buf[PATH_MAX+1];
2561 int num = Atomic::add(1, &cnt);
2562
2563 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2564 os::get_temp_directory(), os::current_process_id(), num);
2565 unlink(buf);
2566
2567 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2568
2569 if (fd != -1) {
2570 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2571 if (rv != (off_t)-1) {
2572 if (::write(fd, "", 1) == 1) {
2573 mmap(base, size,
2574 PROT_READ|PROT_WRITE|PROT_EXEC,
2575 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2576 }
2577 }
2578 ::close(fd);
2579 unlink(buf);
2580 }
2581 }
2582
2583 static bool recoverable_mmap_error(int err) {
2584 // See if the error is one we can let the caller handle. This
2585 // list of errno values comes from JBS-6843484. I can't find a
2586 // Linux man page that documents this specific set of errno
2587 // values so while this list currently matches Solaris, it may
2588 // change as we gain experience with this failure mode.
2589 switch (err) {
2590 case EBADF:
2591 case EINVAL:
2592 case ENOTSUP:
2593 // let the caller deal with these errors
2594 return true;
2595
2596 default:
2597 // Any remaining errors on this OS can cause our reserved mapping
2598 // to be lost. That can cause confusion where different data
2599 // structures think they have the same memory mapped. The worst
2600 // scenario is if both the VM and a library think they have the
2601 // same memory mapped.
2602 return false;
2603 }
2604 }
2605
2606 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2607 int err) {
2608 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2609 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2610 os::strerror(err), err);
2611 }
2612
2613 static void warn_fail_commit_memory(char* addr, size_t size,
2614 size_t alignment_hint, bool exec,
2615 int err) {
2616 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2617 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2618 alignment_hint, exec, os::strerror(err), err);
2619 }
2620
2621 // NOTE: Linux kernel does not really reserve the pages for us.
2622 // All it does is to check if there are enough free pages
2623 // left at the time of mmap(). This could be a potential
2624 // problem.
2625 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2626 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2627 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2628 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2629 if (res != (uintptr_t) MAP_FAILED) {
2630 if (UseNUMAInterleaving) {
2631 numa_make_global(addr, size);
2632 }
2633 return 0;
2634 }
2635
2636 int err = errno; // save errno from mmap() call above
2637
2638 if (!recoverable_mmap_error(err)) {
2639 warn_fail_commit_memory(addr, size, exec, err);
2640 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2641 }
2642
2643 return err;
2644 }
2645
2646 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2647 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2648 }
2649
2650 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2651 const char* mesg) {
2652 assert(mesg != NULL, "mesg must be specified");
2653 int err = os::Linux::commit_memory_impl(addr, size, exec);
2654 if (err != 0) {
2655 // the caller wants all commit errors to exit with the specified mesg:
2656 warn_fail_commit_memory(addr, size, exec, err);
2657 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2658 }
2659 }
2660
2661 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2662 #ifndef MAP_HUGETLB
2663 #define MAP_HUGETLB 0x40000
2664 #endif
2665
2666 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2667 #ifndef MADV_HUGEPAGE
2668 #define MADV_HUGEPAGE 14
2669 #endif
2670
2671 int os::Linux::commit_memory_impl(char* addr, size_t size,
2672 size_t alignment_hint, bool exec) {
2673 int err = os::Linux::commit_memory_impl(addr, size, exec);
2674 if (err == 0) {
2675 realign_memory(addr, size, alignment_hint);
2676 }
2677 return err;
2678 }
2679
2680 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2681 bool exec) {
2682 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2683 }
2684
2685 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2686 size_t alignment_hint, bool exec,
2687 const char* mesg) {
2688 assert(mesg != NULL, "mesg must be specified");
2689 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2690 if (err != 0) {
2691 // the caller wants all commit errors to exit with the specified mesg:
2692 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2693 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2694 }
2695 }
2696
2697 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2698 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2699 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2700 // be supported or the memory may already be backed by huge pages.
2701 ::madvise(addr, bytes, MADV_HUGEPAGE);
2702 }
2703 }
2704
2705 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2706 // This method works by doing an mmap over an existing mmaping and effectively discarding
2707 // the existing pages. However it won't work for SHM-based large pages that cannot be
2708 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2709 // small pages on top of the SHM segment. This method always works for small pages, so we
2710 // allow that in any case.
2711 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2712 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2713 }
2714 }
2715
2716 void os::numa_make_global(char *addr, size_t bytes) {
2717 Linux::numa_interleave_memory(addr, bytes);
2718 }
2719
2720 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2721 // bind policy to MPOL_PREFERRED for the current thread.
2722 #define USE_MPOL_PREFERRED 0
2723
2724 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2725 // To make NUMA and large pages more robust when both enabled, we need to ease
2726 // the requirements on where the memory should be allocated. MPOL_BIND is the
2727 // default policy and it will force memory to be allocated on the specified
2728 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2729 // the specified node, but will not force it. Using this policy will prevent
2730 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2731 // free large pages.
2732 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2733 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2734 }
2735
2736 bool os::numa_topology_changed() { return false; }
2737
2738 size_t os::numa_get_groups_num() {
2739 // Return just the number of nodes in which it's possible to allocate memory
2740 // (in numa terminology, configured nodes).
2741 return Linux::numa_num_configured_nodes();
2742 }
2743
2744 int os::numa_get_group_id() {
2745 int cpu_id = Linux::sched_getcpu();
2746 if (cpu_id != -1) {
2747 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2748 if (lgrp_id != -1) {
2749 return lgrp_id;
2750 }
2751 }
2752 return 0;
2753 }
2754
2755 int os::Linux::get_existing_num_nodes() {
2756 size_t node;
2757 size_t highest_node_number = Linux::numa_max_node();
2758 int num_nodes = 0;
2759
2760 // Get the total number of nodes in the system including nodes without memory.
2761 for (node = 0; node <= highest_node_number; node++) {
2762 if (isnode_in_existing_nodes(node)) {
2763 num_nodes++;
2764 }
2765 }
2766 return num_nodes;
2767 }
2768
2769 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2770 size_t highest_node_number = Linux::numa_max_node();
2771 size_t i = 0;
2772
2773 // Map all node ids in which is possible to allocate memory. Also nodes are
2774 // not always consecutively available, i.e. available from 0 to the highest
2775 // node number.
2776 for (size_t node = 0; node <= highest_node_number; node++) {
2777 if (Linux::isnode_in_configured_nodes(node)) {
2778 ids[i++] = node;
2779 }
2780 }
2781 return i;
2782 }
2783
2784 bool os::get_page_info(char *start, page_info* info) {
2785 return false;
2786 }
2787
2788 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2789 page_info* page_found) {
2790 return end;
2791 }
2792
2793
2794 int os::Linux::sched_getcpu_syscall(void) {
2795 unsigned int cpu = 0;
2796 int retval = -1;
2797
2798 #if defined(IA32)
2799 #ifndef SYS_getcpu
2800 #define SYS_getcpu 318
2801 #endif
2802 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2803 #elif defined(AMD64)
2804 // Unfortunately we have to bring all these macros here from vsyscall.h
2805 // to be able to compile on old linuxes.
2806 #define __NR_vgetcpu 2
2807 #define VSYSCALL_START (-10UL << 20)
2808 #define VSYSCALL_SIZE 1024
2809 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2810 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2811 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2812 retval = vgetcpu(&cpu, NULL, NULL);
2813 #endif
2814
2815 return (retval == -1) ? retval : cpu;
2816 }
2817
2818 void os::Linux::sched_getcpu_init() {
2819 // sched_getcpu() should be in libc.
2820 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2821 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2822
2823 // If it's not, try a direct syscall.
2824 if (sched_getcpu() == -1) {
2825 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2826 (void*)&sched_getcpu_syscall));
2827 }
2828 }
2829
2830 // Something to do with the numa-aware allocator needs these symbols
2831 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2832 extern "C" JNIEXPORT void numa_error(char *where) { }
2833
2834 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2835 // load symbol from base version instead.
2836 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2837 void *f = dlvsym(handle, name, "libnuma_1.1");
2838 if (f == NULL) {
2839 f = dlsym(handle, name);
2840 }
2841 return f;
2842 }
2843
2844 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2845 // Return NULL if the symbol is not defined in this particular version.
2846 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2847 return dlvsym(handle, name, "libnuma_1.2");
2848 }
2849
2850 bool os::Linux::libnuma_init() {
2851 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2852 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2853 if (handle != NULL) {
2854 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2855 libnuma_dlsym(handle, "numa_node_to_cpus")));
2856 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2857 libnuma_dlsym(handle, "numa_max_node")));
2858 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2859 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2860 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2861 libnuma_dlsym(handle, "numa_available")));
2862 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2863 libnuma_dlsym(handle, "numa_tonode_memory")));
2864 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2865 libnuma_dlsym(handle, "numa_interleave_memory")));
2866 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2867 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2868 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2869 libnuma_dlsym(handle, "numa_set_bind_policy")));
2870 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2871 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2872 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2873 libnuma_dlsym(handle, "numa_distance")));
2874
2875 if (numa_available() != -1) {
2876 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2877 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2878 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2879 // Create an index -> node mapping, since nodes are not always consecutive
2880 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2881 rebuild_nindex_to_node_map();
2882 // Create a cpu -> node mapping
2883 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2884 rebuild_cpu_to_node_map();
2885 return true;
2886 }
2887 }
2888 }
2889 return false;
2890 }
2891
2892 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2893 // Creating guard page is very expensive. Java thread has HotSpot
2894 // guard pages, only enable glibc guard page for non-Java threads.
2895 // (Remember: compiler thread is a Java thread, too!)
2896 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2897 }
2898
2899 void os::Linux::rebuild_nindex_to_node_map() {
2900 int highest_node_number = Linux::numa_max_node();
2901
2902 nindex_to_node()->clear();
2903 for (int node = 0; node <= highest_node_number; node++) {
2904 if (Linux::isnode_in_existing_nodes(node)) {
2905 nindex_to_node()->append(node);
2906 }
2907 }
2908 }
2909
2910 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2911 // The table is later used in get_node_by_cpu().
2912 void os::Linux::rebuild_cpu_to_node_map() {
2913 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2914 // in libnuma (possible values are starting from 16,
2915 // and continuing up with every other power of 2, but less
2916 // than the maximum number of CPUs supported by kernel), and
2917 // is a subject to change (in libnuma version 2 the requirements
2918 // are more reasonable) we'll just hardcode the number they use
2919 // in the library.
2920 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2921
2922 size_t cpu_num = processor_count();
2923 size_t cpu_map_size = NCPUS / BitsPerCLong;
2924 size_t cpu_map_valid_size =
2925 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2926
2927 cpu_to_node()->clear();
2928 cpu_to_node()->at_grow(cpu_num - 1);
2929
2930 size_t node_num = get_existing_num_nodes();
2931
2932 int distance = 0;
2933 int closest_distance = INT_MAX;
2934 int closest_node = 0;
2935 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2936 for (size_t i = 0; i < node_num; i++) {
2937 // Check if node is configured (not a memory-less node). If it is not, find
2938 // the closest configured node.
2939 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2940 closest_distance = INT_MAX;
2941 // Check distance from all remaining nodes in the system. Ignore distance
2942 // from itself and from another non-configured node.
2943 for (size_t m = 0; m < node_num; m++) {
2944 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2945 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2946 // If a closest node is found, update. There is always at least one
2947 // configured node in the system so there is always at least one node
2948 // close.
2949 if (distance != 0 && distance < closest_distance) {
2950 closest_distance = distance;
2951 closest_node = nindex_to_node()->at(m);
2952 }
2953 }
2954 }
2955 } else {
2956 // Current node is already a configured node.
2957 closest_node = nindex_to_node()->at(i);
2958 }
2959
2960 // Get cpus from the original node and map them to the closest node. If node
2961 // is a configured node (not a memory-less node), then original node and
2962 // closest node are the same.
2963 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2964 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2965 if (cpu_map[j] != 0) {
2966 for (size_t k = 0; k < BitsPerCLong; k++) {
2967 if (cpu_map[j] & (1UL << k)) {
2968 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2969 }
2970 }
2971 }
2972 }
2973 }
2974 }
2975 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2976 }
2977
2978 int os::Linux::get_node_by_cpu(int cpu_id) {
2979 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2980 return cpu_to_node()->at(cpu_id);
2981 }
2982 return -1;
2983 }
2984
2985 GrowableArray<int>* os::Linux::_cpu_to_node;
2986 GrowableArray<int>* os::Linux::_nindex_to_node;
2987 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2988 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2989 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2990 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
2991 os::Linux::numa_available_func_t os::Linux::_numa_available;
2992 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2993 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2994 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
2995 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2996 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
2997 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
2998 unsigned long* os::Linux::_numa_all_nodes;
2999 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3000 struct bitmask* os::Linux::_numa_nodes_ptr;
3001
3002 bool os::pd_uncommit_memory(char* addr, size_t size) {
3003 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3004 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3005 return res != (uintptr_t) MAP_FAILED;
3006 }
3007
3008 static address get_stack_commited_bottom(address bottom, size_t size) {
3009 address nbot = bottom;
3010 address ntop = bottom + size;
3011
3012 size_t page_sz = os::vm_page_size();
3013 unsigned pages = size / page_sz;
3014
3015 unsigned char vec[1];
3016 unsigned imin = 1, imax = pages + 1, imid;
3017 int mincore_return_value = 0;
3018
3019 assert(imin <= imax, "Unexpected page size");
3020
3021 while (imin < imax) {
3022 imid = (imax + imin) / 2;
3023 nbot = ntop - (imid * page_sz);
3024
3025 // Use a trick with mincore to check whether the page is mapped or not.
3026 // mincore sets vec to 1 if page resides in memory and to 0 if page
3027 // is swapped output but if page we are asking for is unmapped
3028 // it returns -1,ENOMEM
3029 mincore_return_value = mincore(nbot, page_sz, vec);
3030
3031 if (mincore_return_value == -1) {
3032 // Page is not mapped go up
3033 // to find first mapped page
3034 if (errno != EAGAIN) {
3035 assert(errno == ENOMEM, "Unexpected mincore errno");
3036 imax = imid;
3037 }
3038 } else {
3039 // Page is mapped go down
3040 // to find first not mapped page
3041 imin = imid + 1;
3042 }
3043 }
3044
3045 nbot = nbot + page_sz;
3046
3047 // Adjust stack bottom one page up if last checked page is not mapped
3048 if (mincore_return_value == -1) {
3049 nbot = nbot + page_sz;
3050 }
3051
3052 return nbot;
3053 }
3054
3055
3056 // Linux uses a growable mapping for the stack, and if the mapping for
3057 // the stack guard pages is not removed when we detach a thread the
3058 // stack cannot grow beyond the pages where the stack guard was
3059 // mapped. If at some point later in the process the stack expands to
3060 // that point, the Linux kernel cannot expand the stack any further
3061 // because the guard pages are in the way, and a segfault occurs.
3062 //
3063 // However, it's essential not to split the stack region by unmapping
3064 // a region (leaving a hole) that's already part of the stack mapping,
3065 // so if the stack mapping has already grown beyond the guard pages at
3066 // the time we create them, we have to truncate the stack mapping.
3067 // So, we need to know the extent of the stack mapping when
3068 // create_stack_guard_pages() is called.
3069
3070 // We only need this for stacks that are growable: at the time of
3071 // writing thread stacks don't use growable mappings (i.e. those
3072 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3073 // only applies to the main thread.
3074
3075 // If the (growable) stack mapping already extends beyond the point
3076 // where we're going to put our guard pages, truncate the mapping at
3077 // that point by munmap()ping it. This ensures that when we later
3078 // munmap() the guard pages we don't leave a hole in the stack
3079 // mapping. This only affects the main/initial thread
3080
3081 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3082 if (os::Linux::is_initial_thread()) {
3083 // As we manually grow stack up to bottom inside create_attached_thread(),
3084 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3085 // we don't need to do anything special.
3086 // Check it first, before calling heavy function.
3087 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3088 unsigned char vec[1];
3089
3090 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3091 // Fallback to slow path on all errors, including EAGAIN
3092 stack_extent = (uintptr_t) get_stack_commited_bottom(
3093 os::Linux::initial_thread_stack_bottom(),
3094 (size_t)addr - stack_extent);
3095 }
3096
3097 if (stack_extent < (uintptr_t)addr) {
3098 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3099 }
3100 }
3101
3102 return os::commit_memory(addr, size, !ExecMem);
3103 }
3104
3105 // If this is a growable mapping, remove the guard pages entirely by
3106 // munmap()ping them. If not, just call uncommit_memory(). This only
3107 // affects the main/initial thread, but guard against future OS changes
3108 // It's safe to always unmap guard pages for initial thread because we
3109 // always place it right after end of the mapped region
3110
3111 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3112 uintptr_t stack_extent, stack_base;
3113
3114 if (os::Linux::is_initial_thread()) {
3115 return ::munmap(addr, size) == 0;
3116 }
3117
3118 return os::uncommit_memory(addr, size);
3119 }
3120
3121 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3122 // at 'requested_addr'. If there are existing memory mappings at the same
3123 // location, however, they will be overwritten. If 'fixed' is false,
3124 // 'requested_addr' is only treated as a hint, the return value may or
3125 // may not start from the requested address. Unlike Linux mmap(), this
3126 // function returns NULL to indicate failure.
3127 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3128 char * addr;
3129 int flags;
3130
3131 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3132 if (fixed) {
3133 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3134 flags |= MAP_FIXED;
3135 }
3136
3137 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3138 // touch an uncommitted page. Otherwise, the read/write might
3139 // succeed if we have enough swap space to back the physical page.
3140 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3141 flags, -1, 0);
3142
3143 return addr == MAP_FAILED ? NULL : addr;
3144 }
3145
3146 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3147 // (req_addr != NULL) or with a given alignment.
3148 // - bytes shall be a multiple of alignment.
3149 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3150 // - alignment sets the alignment at which memory shall be allocated.
3151 // It must be a multiple of allocation granularity.
3152 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3153 // req_addr or NULL.
3154 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3155
3156 size_t extra_size = bytes;
3157 if (req_addr == NULL && alignment > 0) {
3158 extra_size += alignment;
3159 }
3160
3161 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3162 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3163 -1, 0);
3164 if (start == MAP_FAILED) {
3165 start = NULL;
3166 } else {
3167 if (req_addr != NULL) {
3168 if (start != req_addr) {
3169 ::munmap(start, extra_size);
3170 start = NULL;
3171 }
3172 } else {
3173 char* const start_aligned = align_up(start, alignment);
3174 char* const end_aligned = start_aligned + bytes;
3175 char* const end = start + extra_size;
3176 if (start_aligned > start) {
3177 ::munmap(start, start_aligned - start);
3178 }
3179 if (end_aligned < end) {
3180 ::munmap(end_aligned, end - end_aligned);
3181 }
3182 start = start_aligned;
3183 }
3184 }
3185 return start;
3186 }
3187
3188 static int anon_munmap(char * addr, size_t size) {
3189 return ::munmap(addr, size) == 0;
3190 }
3191
3192 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3193 size_t alignment_hint) {
3194 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3195 }
3196
3197 bool os::pd_release_memory(char* addr, size_t size) {
3198 return anon_munmap(addr, size);
3199 }
3200
3201 static bool linux_mprotect(char* addr, size_t size, int prot) {
3202 // Linux wants the mprotect address argument to be page aligned.
3203 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3204
3205 // According to SUSv3, mprotect() should only be used with mappings
3206 // established by mmap(), and mmap() always maps whole pages. Unaligned
3207 // 'addr' likely indicates problem in the VM (e.g. trying to change
3208 // protection of malloc'ed or statically allocated memory). Check the
3209 // caller if you hit this assert.
3210 assert(addr == bottom, "sanity check");
3211
3212 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3213 return ::mprotect(bottom, size, prot) == 0;
3214 }
3215
3216 // Set protections specified
3217 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3218 bool is_committed) {
3219 unsigned int p = 0;
3220 switch (prot) {
3221 case MEM_PROT_NONE: p = PROT_NONE; break;
3222 case MEM_PROT_READ: p = PROT_READ; break;
3223 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3224 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3225 default:
3226 ShouldNotReachHere();
3227 }
3228 // is_committed is unused.
3229 return linux_mprotect(addr, bytes, p);
3230 }
3231
3232 bool os::guard_memory(char* addr, size_t size) {
3233 return linux_mprotect(addr, size, PROT_NONE);
3234 }
3235
3236 bool os::unguard_memory(char* addr, size_t size) {
3237 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3238 }
3239
3240 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3241 size_t page_size) {
3242 bool result = false;
3243 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3244 MAP_ANONYMOUS|MAP_PRIVATE,
3245 -1, 0);
3246 if (p != MAP_FAILED) {
3247 void *aligned_p = align_up(p, page_size);
3248
3249 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3250
3251 munmap(p, page_size * 2);
3252 }
3253
3254 if (warn && !result) {
3255 warning("TransparentHugePages is not supported by the operating system.");
3256 }
3257
3258 return result;
3259 }
3260
3261 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3262 bool result = false;
3263 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3264 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3265 -1, 0);
3266
3267 if (p != MAP_FAILED) {
3268 // We don't know if this really is a huge page or not.
3269 FILE *fp = fopen("/proc/self/maps", "r");
3270 if (fp) {
3271 while (!feof(fp)) {
3272 char chars[257];
3273 long x = 0;
3274 if (fgets(chars, sizeof(chars), fp)) {
3275 if (sscanf(chars, "%lx-%*x", &x) == 1
3276 && x == (long)p) {
3277 if (strstr (chars, "hugepage")) {
3278 result = true;
3279 break;
3280 }
3281 }
3282 }
3283 }
3284 fclose(fp);
3285 }
3286 munmap(p, page_size);
3287 }
3288
3289 if (warn && !result) {
3290 warning("HugeTLBFS is not supported by the operating system.");
3291 }
3292
3293 return result;
3294 }
3295
3296 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3297 //
3298 // From the coredump_filter documentation:
3299 //
3300 // - (bit 0) anonymous private memory
3301 // - (bit 1) anonymous shared memory
3302 // - (bit 2) file-backed private memory
3303 // - (bit 3) file-backed shared memory
3304 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3305 // effective only if the bit 2 is cleared)
3306 // - (bit 5) hugetlb private memory
3307 // - (bit 6) hugetlb shared memory
3308 //
3309 static void set_coredump_filter(void) {
3310 FILE *f;
3311 long cdm;
3312
3313 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3314 return;
3315 }
3316
3317 if (fscanf(f, "%lx", &cdm) != 1) {
3318 fclose(f);
3319 return;
3320 }
3321
3322 rewind(f);
3323
3324 if ((cdm & LARGEPAGES_BIT) == 0) {
3325 cdm |= LARGEPAGES_BIT;
3326 fprintf(f, "%#lx", cdm);
3327 }
3328
3329 fclose(f);
3330 }
3331
3332 // Large page support
3333
3334 static size_t _large_page_size = 0;
3335
3336 size_t os::Linux::find_large_page_size() {
3337 size_t large_page_size = 0;
3338
3339 // large_page_size on Linux is used to round up heap size. x86 uses either
3340 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3341 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3342 // page as large as 256M.
3343 //
3344 // Here we try to figure out page size by parsing /proc/meminfo and looking
3345 // for a line with the following format:
3346 // Hugepagesize: 2048 kB
3347 //
3348 // If we can't determine the value (e.g. /proc is not mounted, or the text
3349 // format has been changed), we'll use the largest page size supported by
3350 // the processor.
3351
3352 #ifndef ZERO
3353 large_page_size =
3354 AARCH64_ONLY(2 * M)
3355 AMD64_ONLY(2 * M)
3356 ARM32_ONLY(2 * M)
3357 IA32_ONLY(4 * M)
3358 IA64_ONLY(256 * M)
3359 PPC_ONLY(4 * M)
3360 S390_ONLY(1 * M)
3361 SPARC_ONLY(4 * M);
3362 #endif // ZERO
3363
3364 FILE *fp = fopen("/proc/meminfo", "r");
3365 if (fp) {
3366 while (!feof(fp)) {
3367 int x = 0;
3368 char buf[16];
3369 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3370 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3371 large_page_size = x * K;
3372 break;
3373 }
3374 } else {
3375 // skip to next line
3376 for (;;) {
3377 int ch = fgetc(fp);
3378 if (ch == EOF || ch == (int)'\n') break;
3379 }
3380 }
3381 }
3382 fclose(fp);
3383 }
3384
3385 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3386 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3387 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3388 proper_unit_for_byte_size(large_page_size));
3389 }
3390
3391 return large_page_size;
3392 }
3393
3394 size_t os::Linux::setup_large_page_size() {
3395 _large_page_size = Linux::find_large_page_size();
3396 const size_t default_page_size = (size_t)Linux::page_size();
3397 if (_large_page_size > default_page_size) {
3398 _page_sizes[0] = _large_page_size;
3399 _page_sizes[1] = default_page_size;
3400 _page_sizes[2] = 0;
3401 }
3402
3403 return _large_page_size;
3404 }
3405
3406 bool os::Linux::setup_large_page_type(size_t page_size) {
3407 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3408 FLAG_IS_DEFAULT(UseSHM) &&
3409 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3410
3411 // The type of large pages has not been specified by the user.
3412
3413 // Try UseHugeTLBFS and then UseSHM.
3414 UseHugeTLBFS = UseSHM = true;
3415
3416 // Don't try UseTransparentHugePages since there are known
3417 // performance issues with it turned on. This might change in the future.
3418 UseTransparentHugePages = false;
3419 }
3420
3421 if (UseTransparentHugePages) {
3422 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3423 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3424 UseHugeTLBFS = false;
3425 UseSHM = false;
3426 return true;
3427 }
3428 UseTransparentHugePages = false;
3429 }
3430
3431 if (UseHugeTLBFS) {
3432 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3433 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3434 UseSHM = false;
3435 return true;
3436 }
3437 UseHugeTLBFS = false;
3438 }
3439
3440 return UseSHM;
3441 }
3442
3443 void os::large_page_init() {
3444 if (!UseLargePages &&
3445 !UseTransparentHugePages &&
3446 !UseHugeTLBFS &&
3447 !UseSHM) {
3448 // Not using large pages.
3449 return;
3450 }
3451
3452 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3453 // The user explicitly turned off large pages.
3454 // Ignore the rest of the large pages flags.
3455 UseTransparentHugePages = false;
3456 UseHugeTLBFS = false;
3457 UseSHM = false;
3458 return;
3459 }
3460
3461 size_t large_page_size = Linux::setup_large_page_size();
3462 UseLargePages = Linux::setup_large_page_type(large_page_size);
3463
3464 set_coredump_filter();
3465 }
3466
3467 #ifndef SHM_HUGETLB
3468 #define SHM_HUGETLB 04000
3469 #endif
3470
3471 #define shm_warning_format(format, ...) \
3472 do { \
3473 if (UseLargePages && \
3474 (!FLAG_IS_DEFAULT(UseLargePages) || \
3475 !FLAG_IS_DEFAULT(UseSHM) || \
3476 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3477 warning(format, __VA_ARGS__); \
3478 } \
3479 } while (0)
3480
3481 #define shm_warning(str) shm_warning_format("%s", str)
3482
3483 #define shm_warning_with_errno(str) \
3484 do { \
3485 int err = errno; \
3486 shm_warning_format(str " (error = %d)", err); \
3487 } while (0)
3488
3489 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3490 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3491
3492 if (!is_aligned(alignment, SHMLBA)) {
3493 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3494 return NULL;
3495 }
3496
3497 // To ensure that we get 'alignment' aligned memory from shmat,
3498 // we pre-reserve aligned virtual memory and then attach to that.
3499
3500 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3501 if (pre_reserved_addr == NULL) {
3502 // Couldn't pre-reserve aligned memory.
3503 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3504 return NULL;
3505 }
3506
3507 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3508 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3509
3510 if ((intptr_t)addr == -1) {
3511 int err = errno;
3512 shm_warning_with_errno("Failed to attach shared memory.");
3513
3514 assert(err != EACCES, "Unexpected error");
3515 assert(err != EIDRM, "Unexpected error");
3516 assert(err != EINVAL, "Unexpected error");
3517
3518 // Since we don't know if the kernel unmapped the pre-reserved memory area
3519 // we can't unmap it, since that would potentially unmap memory that was
3520 // mapped from other threads.
3521 return NULL;
3522 }
3523
3524 return addr;
3525 }
3526
3527 static char* shmat_at_address(int shmid, char* req_addr) {
3528 if (!is_aligned(req_addr, SHMLBA)) {
3529 assert(false, "Requested address needs to be SHMLBA aligned");
3530 return NULL;
3531 }
3532
3533 char* addr = (char*)shmat(shmid, req_addr, 0);
3534
3535 if ((intptr_t)addr == -1) {
3536 shm_warning_with_errno("Failed to attach shared memory.");
3537 return NULL;
3538 }
3539
3540 return addr;
3541 }
3542
3543 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3544 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3545 if (req_addr != NULL) {
3546 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3547 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3548 return shmat_at_address(shmid, req_addr);
3549 }
3550
3551 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3552 // return large page size aligned memory addresses when req_addr == NULL.
3553 // However, if the alignment is larger than the large page size, we have
3554 // to manually ensure that the memory returned is 'alignment' aligned.
3555 if (alignment > os::large_page_size()) {
3556 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3557 return shmat_with_alignment(shmid, bytes, alignment);
3558 } else {
3559 return shmat_at_address(shmid, NULL);
3560 }
3561 }
3562
3563 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3564 char* req_addr, bool exec) {
3565 // "exec" is passed in but not used. Creating the shared image for
3566 // the code cache doesn't have an SHM_X executable permission to check.
3567 assert(UseLargePages && UseSHM, "only for SHM large pages");
3568 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3569 assert(is_aligned(req_addr, alignment), "Unaligned address");
3570
3571 if (!is_aligned(bytes, os::large_page_size())) {
3572 return NULL; // Fallback to small pages.
3573 }
3574
3575 // Create a large shared memory region to attach to based on size.
3576 // Currently, size is the total size of the heap.
3577 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3578 if (shmid == -1) {
3579 // Possible reasons for shmget failure:
3580 // 1. shmmax is too small for Java heap.
3581 // > check shmmax value: cat /proc/sys/kernel/shmmax
3582 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3583 // 2. not enough large page memory.
3584 // > check available large pages: cat /proc/meminfo
3585 // > increase amount of large pages:
3586 // echo new_value > /proc/sys/vm/nr_hugepages
3587 // Note 1: different Linux may use different name for this property,
3588 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3589 // Note 2: it's possible there's enough physical memory available but
3590 // they are so fragmented after a long run that they can't
3591 // coalesce into large pages. Try to reserve large pages when
3592 // the system is still "fresh".
3593 shm_warning_with_errno("Failed to reserve shared memory.");
3594 return NULL;
3595 }
3596
3597 // Attach to the region.
3598 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3599
3600 // Remove shmid. If shmat() is successful, the actual shared memory segment
3601 // will be deleted when it's detached by shmdt() or when the process
3602 // terminates. If shmat() is not successful this will remove the shared
3603 // segment immediately.
3604 shmctl(shmid, IPC_RMID, NULL);
3605
3606 return addr;
3607 }
3608
3609 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3610 int error) {
3611 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3612
3613 bool warn_on_failure = UseLargePages &&
3614 (!FLAG_IS_DEFAULT(UseLargePages) ||
3615 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3616 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3617
3618 if (warn_on_failure) {
3619 char msg[128];
3620 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3621 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3622 warning("%s", msg);
3623 }
3624 }
3625
3626 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3627 char* req_addr,
3628 bool exec) {
3629 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3630 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3631 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3632
3633 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3634 char* addr = (char*)::mmap(req_addr, bytes, prot,
3635 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3636 -1, 0);
3637
3638 if (addr == MAP_FAILED) {
3639 warn_on_large_pages_failure(req_addr, bytes, errno);
3640 return NULL;
3641 }
3642
3643 assert(is_aligned(addr, os::large_page_size()), "Must be");
3644
3645 return addr;
3646 }
3647
3648 // Reserve memory using mmap(MAP_HUGETLB).
3649 // - bytes shall be a multiple of alignment.
3650 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3651 // - alignment sets the alignment at which memory shall be allocated.
3652 // It must be a multiple of allocation granularity.
3653 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3654 // req_addr or NULL.
3655 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3656 size_t alignment,
3657 char* req_addr,
3658 bool exec) {
3659 size_t large_page_size = os::large_page_size();
3660 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3661
3662 assert(is_aligned(req_addr, alignment), "Must be");
3663 assert(is_aligned(bytes, alignment), "Must be");
3664
3665 // First reserve - but not commit - the address range in small pages.
3666 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3667
3668 if (start == NULL) {
3669 return NULL;
3670 }
3671
3672 assert(is_aligned(start, alignment), "Must be");
3673
3674 char* end = start + bytes;
3675
3676 // Find the regions of the allocated chunk that can be promoted to large pages.
3677 char* lp_start = align_up(start, large_page_size);
3678 char* lp_end = align_down(end, large_page_size);
3679
3680 size_t lp_bytes = lp_end - lp_start;
3681
3682 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3683
3684 if (lp_bytes == 0) {
3685 // The mapped region doesn't even span the start and the end of a large page.
3686 // Fall back to allocate a non-special area.
3687 ::munmap(start, end - start);
3688 return NULL;
3689 }
3690
3691 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3692
3693 void* result;
3694
3695 // Commit small-paged leading area.
3696 if (start != lp_start) {
3697 result = ::mmap(start, lp_start - start, prot,
3698 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3699 -1, 0);
3700 if (result == MAP_FAILED) {
3701 ::munmap(lp_start, end - lp_start);
3702 return NULL;
3703 }
3704 }
3705
3706 // Commit large-paged area.
3707 result = ::mmap(lp_start, lp_bytes, prot,
3708 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3709 -1, 0);
3710 if (result == MAP_FAILED) {
3711 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3712 // If the mmap above fails, the large pages region will be unmapped and we
3713 // have regions before and after with small pages. Release these regions.
3714 //
3715 // | mapped | unmapped | mapped |
3716 // ^ ^ ^ ^
3717 // start lp_start lp_end end
3718 //
3719 ::munmap(start, lp_start - start);
3720 ::munmap(lp_end, end - lp_end);
3721 return NULL;
3722 }
3723
3724 // Commit small-paged trailing area.
3725 if (lp_end != end) {
3726 result = ::mmap(lp_end, end - lp_end, prot,
3727 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3728 -1, 0);
3729 if (result == MAP_FAILED) {
3730 ::munmap(start, lp_end - start);
3731 return NULL;
3732 }
3733 }
3734
3735 return start;
3736 }
3737
3738 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3739 size_t alignment,
3740 char* req_addr,
3741 bool exec) {
3742 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3743 assert(is_aligned(req_addr, alignment), "Must be");
3744 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3745 assert(is_power_of_2(os::large_page_size()), "Must be");
3746 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3747
3748 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3749 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3750 } else {
3751 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3752 }
3753 }
3754
3755 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3756 char* req_addr, bool exec) {
3757 assert(UseLargePages, "only for large pages");
3758
3759 char* addr;
3760 if (UseSHM) {
3761 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3762 } else {
3763 assert(UseHugeTLBFS, "must be");
3764 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3765 }
3766
3767 if (addr != NULL) {
3768 if (UseNUMAInterleaving) {
3769 numa_make_global(addr, bytes);
3770 }
3771
3772 // The memory is committed
3773 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3774 }
3775
3776 return addr;
3777 }
3778
3779 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3780 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3781 return shmdt(base) == 0;
3782 }
3783
3784 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3785 return pd_release_memory(base, bytes);
3786 }
3787
3788 bool os::release_memory_special(char* base, size_t bytes) {
3789 bool res;
3790 if (MemTracker::tracking_level() > NMT_minimal) {
3791 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3792 res = os::Linux::release_memory_special_impl(base, bytes);
3793 if (res) {
3794 tkr.record((address)base, bytes);
3795 }
3796
3797 } else {
3798 res = os::Linux::release_memory_special_impl(base, bytes);
3799 }
3800 return res;
3801 }
3802
3803 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3804 assert(UseLargePages, "only for large pages");
3805 bool res;
3806
3807 if (UseSHM) {
3808 res = os::Linux::release_memory_special_shm(base, bytes);
3809 } else {
3810 assert(UseHugeTLBFS, "must be");
3811 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3812 }
3813 return res;
3814 }
3815
3816 size_t os::large_page_size() {
3817 return _large_page_size;
3818 }
3819
3820 // With SysV SHM the entire memory region must be allocated as shared
3821 // memory.
3822 // HugeTLBFS allows application to commit large page memory on demand.
3823 // However, when committing memory with HugeTLBFS fails, the region
3824 // that was supposed to be committed will lose the old reservation
3825 // and allow other threads to steal that memory region. Because of this
3826 // behavior we can't commit HugeTLBFS memory.
3827 bool os::can_commit_large_page_memory() {
3828 return UseTransparentHugePages;
3829 }
3830
3831 bool os::can_execute_large_page_memory() {
3832 return UseTransparentHugePages || UseHugeTLBFS;
3833 }
3834
3835 // Reserve memory at an arbitrary address, only if that area is
3836 // available (and not reserved for something else).
3837
3838 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3839 const int max_tries = 10;
3840 char* base[max_tries];
3841 size_t size[max_tries];
3842 const size_t gap = 0x000000;
3843
3844 // Assert only that the size is a multiple of the page size, since
3845 // that's all that mmap requires, and since that's all we really know
3846 // about at this low abstraction level. If we need higher alignment,
3847 // we can either pass an alignment to this method or verify alignment
3848 // in one of the methods further up the call chain. See bug 5044738.
3849 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3850
3851 // Repeatedly allocate blocks until the block is allocated at the
3852 // right spot.
3853
3854 // Linux mmap allows caller to pass an address as hint; give it a try first,
3855 // if kernel honors the hint then we can return immediately.
3856 char * addr = anon_mmap(requested_addr, bytes, false);
3857 if (addr == requested_addr) {
3858 return requested_addr;
3859 }
3860
3861 if (addr != NULL) {
3862 // mmap() is successful but it fails to reserve at the requested address
3863 anon_munmap(addr, bytes);
3864 }
3865
3866 int i;
3867 for (i = 0; i < max_tries; ++i) {
3868 base[i] = reserve_memory(bytes);
3869
3870 if (base[i] != NULL) {
3871 // Is this the block we wanted?
3872 if (base[i] == requested_addr) {
3873 size[i] = bytes;
3874 break;
3875 }
3876
3877 // Does this overlap the block we wanted? Give back the overlapped
3878 // parts and try again.
3879
3880 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3881 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3882 unmap_memory(base[i], top_overlap);
3883 base[i] += top_overlap;
3884 size[i] = bytes - top_overlap;
3885 } else {
3886 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3887 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3888 unmap_memory(requested_addr, bottom_overlap);
3889 size[i] = bytes - bottom_overlap;
3890 } else {
3891 size[i] = bytes;
3892 }
3893 }
3894 }
3895 }
3896
3897 // Give back the unused reserved pieces.
3898
3899 for (int j = 0; j < i; ++j) {
3900 if (base[j] != NULL) {
3901 unmap_memory(base[j], size[j]);
3902 }
3903 }
3904
3905 if (i < max_tries) {
3906 return requested_addr;
3907 } else {
3908 return NULL;
3909 }
3910 }
3911
3912 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3913 return ::read(fd, buf, nBytes);
3914 }
3915
3916 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3917 return ::pread(fd, buf, nBytes, offset);
3918 }
3919
3920 // Short sleep, direct OS call.
3921 //
3922 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3923 // sched_yield(2) will actually give up the CPU:
3924 //
3925 // * Alone on this pariticular CPU, keeps running.
3926 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3927 // (pre 2.6.39).
3928 //
3929 // So calling this with 0 is an alternative.
3930 //
3931 void os::naked_short_sleep(jlong ms) {
3932 struct timespec req;
3933
3934 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3935 req.tv_sec = 0;
3936 if (ms > 0) {
3937 req.tv_nsec = (ms % 1000) * 1000000;
3938 } else {
3939 req.tv_nsec = 1;
3940 }
3941
3942 nanosleep(&req, NULL);
3943
3944 return;
3945 }
3946
3947 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3948 void os::infinite_sleep() {
3949 while (true) { // sleep forever ...
3950 ::sleep(100); // ... 100 seconds at a time
3951 }
3952 }
3953
3954 // Used to convert frequent JVM_Yield() to nops
3955 bool os::dont_yield() {
3956 return DontYieldALot;
3957 }
3958
3959 void os::naked_yield() {
3960 sched_yield();
3961 }
3962
3963 ////////////////////////////////////////////////////////////////////////////////
3964 // thread priority support
3965
3966 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3967 // only supports dynamic priority, static priority must be zero. For real-time
3968 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3969 // However, for large multi-threaded applications, SCHED_RR is not only slower
3970 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3971 // of 5 runs - Sep 2005).
3972 //
3973 // The following code actually changes the niceness of kernel-thread/LWP. It
3974 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3975 // not the entire user process, and user level threads are 1:1 mapped to kernel
3976 // threads. It has always been the case, but could change in the future. For
3977 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3978 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3979
3980 int os::java_to_os_priority[CriticalPriority + 1] = {
3981 19, // 0 Entry should never be used
3982
3983 4, // 1 MinPriority
3984 3, // 2
3985 2, // 3
3986
3987 1, // 4
3988 0, // 5 NormPriority
3989 -1, // 6
3990
3991 -2, // 7
3992 -3, // 8
3993 -4, // 9 NearMaxPriority
3994
3995 -5, // 10 MaxPriority
3996
3997 -5 // 11 CriticalPriority
3998 };
3999
4000 static int prio_init() {
4001 if (ThreadPriorityPolicy == 1) {
4002 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4003 // if effective uid is not root. Perhaps, a more elegant way of doing
4004 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4005 if (geteuid() != 0) {
4006 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4007 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4008 }
4009 ThreadPriorityPolicy = 0;
4010 }
4011 }
4012 if (UseCriticalJavaThreadPriority) {
4013 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4014 }
4015 return 0;
4016 }
4017
4018 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4019 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4020
4021 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4022 return (ret == 0) ? OS_OK : OS_ERR;
4023 }
4024
4025 OSReturn os::get_native_priority(const Thread* const thread,
4026 int *priority_ptr) {
4027 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4028 *priority_ptr = java_to_os_priority[NormPriority];
4029 return OS_OK;
4030 }
4031
4032 errno = 0;
4033 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4034 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4035 }
4036
4037 // Hint to the underlying OS that a task switch would not be good.
4038 // Void return because it's a hint and can fail.
4039 void os::hint_no_preempt() {}
4040
4041 ////////////////////////////////////////////////////////////////////////////////
4042 // suspend/resume support
4043
4044 // The low-level signal-based suspend/resume support is a remnant from the
4045 // old VM-suspension that used to be for java-suspension, safepoints etc,
4046 // within hotspot. Needed for fetch_frame_from_ucontext(), which is used by:
4047 // - Forte Analyzer: AsyncGetCallTrace()
4048 // - StackBanging: get_frame_at_stack_banging_point()
4049 //
4050 // The remaining code is greatly simplified from the more general suspension
4051 // code that used to be used.
4052 //
4053 // The protocol is quite simple:
4054 // - suspend:
4055 // - sends a signal to the target thread
4056 // - polls the suspend state of the osthread using a yield loop
4057 // - target thread signal handler (SR_handler) sets suspend state
4058 // and blocks in sigsuspend until continued
4059 // - resume:
4060 // - sets target osthread state to continue
4061 // - sends signal to end the sigsuspend loop in the SR_handler
4062 //
4063 // Note that the SR_lock plays no role in this suspend/resume protocol,
4064 // but is checked for NULL in SR_handler as a thread termination indicator.
4065
4066 static void resume_clear_context(OSThread *osthread) {
4067 osthread->set_ucontext(NULL);
4068 osthread->set_siginfo(NULL);
4069 }
4070
4071 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4072 ucontext_t* context) {
4073 osthread->set_ucontext(context);
4074 osthread->set_siginfo(siginfo);
4075 }
4076
4077 // Handler function invoked when a thread's execution is suspended or
4078 // resumed. We have to be careful that only async-safe functions are
4079 // called here (Note: most pthread functions are not async safe and
4080 // should be avoided.)
4081 //
4082 // Note: sigwait() is a more natural fit than sigsuspend() from an
4083 // interface point of view, but sigwait() prevents the signal hander
4084 // from being run. libpthread would get very confused by not having
4085 // its signal handlers run and prevents sigwait()'s use with the
4086 // mutex granting granting signal.
4087 //
4088 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4089 //
4090 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4091 // Save and restore errno to avoid confusing native code with EINTR
4092 // after sigsuspend.
4093 int old_errno = errno;
4094
4095 Thread* thread = Thread::current_or_null_safe();
4096 assert(thread != NULL, "Missing current thread in SR_handler");
4097
4098 // On some systems we have seen signal delivery get "stuck" until the signal
4099 // mask is changed as part of thread termination. Check that the current thread
4100 // has not already terminated (via SR_lock()) - else the following assertion
4101 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4102 // destructor has completed.
4103
4104 if (thread->SR_lock() == NULL) {
4105 return;
4106 }
4107
4108 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4109
4110 OSThread* osthread = thread->osthread();
4111
4112 os::SuspendResume::State current = osthread->sr.state();
4113 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4114 suspend_save_context(osthread, siginfo, context);
4115
4116 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4117 os::SuspendResume::State state = osthread->sr.suspended();
4118 if (state == os::SuspendResume::SR_SUSPENDED) {
4119 sigset_t suspend_set; // signals for sigsuspend()
4120 sigemptyset(&suspend_set);
4121 // get current set of blocked signals and unblock resume signal
4122 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4123 sigdelset(&suspend_set, SR_signum);
4124
4125 sr_semaphore.signal();
4126 // wait here until we are resumed
4127 while (1) {
4128 sigsuspend(&suspend_set);
4129
4130 os::SuspendResume::State result = osthread->sr.running();
4131 if (result == os::SuspendResume::SR_RUNNING) {
4132 sr_semaphore.signal();
4133 break;
4134 }
4135 }
4136
4137 } else if (state == os::SuspendResume::SR_RUNNING) {
4138 // request was cancelled, continue
4139 } else {
4140 ShouldNotReachHere();
4141 }
4142
4143 resume_clear_context(osthread);
4144 } else if (current == os::SuspendResume::SR_RUNNING) {
4145 // request was cancelled, continue
4146 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4147 // ignore
4148 } else {
4149 // ignore
4150 }
4151
4152 errno = old_errno;
4153 }
4154
4155 static int SR_initialize() {
4156 struct sigaction act;
4157 char *s;
4158
4159 // Get signal number to use for suspend/resume
4160 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4161 int sig = ::strtol(s, 0, 10);
4162 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4163 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4164 SR_signum = sig;
4165 } else {
4166 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4167 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4168 }
4169 }
4170
4171 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4172 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4173
4174 sigemptyset(&SR_sigset);
4175 sigaddset(&SR_sigset, SR_signum);
4176
4177 // Set up signal handler for suspend/resume
4178 act.sa_flags = SA_RESTART|SA_SIGINFO;
4179 act.sa_handler = (void (*)(int)) SR_handler;
4180
4181 // SR_signum is blocked by default.
4182 // 4528190 - We also need to block pthread restart signal (32 on all
4183 // supported Linux platforms). Note that LinuxThreads need to block
4184 // this signal for all threads to work properly. So we don't have
4185 // to use hard-coded signal number when setting up the mask.
4186 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4187
4188 if (sigaction(SR_signum, &act, 0) == -1) {
4189 return -1;
4190 }
4191
4192 // Save signal flag
4193 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4194 return 0;
4195 }
4196
4197 static int sr_notify(OSThread* osthread) {
4198 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4199 assert_status(status == 0, status, "pthread_kill");
4200 return status;
4201 }
4202
4203 // "Randomly" selected value for how long we want to spin
4204 // before bailing out on suspending a thread, also how often
4205 // we send a signal to a thread we want to resume
4206 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4207 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4208
4209 // returns true on success and false on error - really an error is fatal
4210 // but this seems the normal response to library errors
4211 static bool do_suspend(OSThread* osthread) {
4212 assert(osthread->sr.is_running(), "thread should be running");
4213 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4214
4215 // mark as suspended and send signal
4216 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4217 // failed to switch, state wasn't running?
4218 ShouldNotReachHere();
4219 return false;
4220 }
4221
4222 if (sr_notify(osthread) != 0) {
4223 ShouldNotReachHere();
4224 }
4225
4226 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4227 while (true) {
4228 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4229 break;
4230 } else {
4231 // timeout
4232 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4233 if (cancelled == os::SuspendResume::SR_RUNNING) {
4234 return false;
4235 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4236 // make sure that we consume the signal on the semaphore as well
4237 sr_semaphore.wait();
4238 break;
4239 } else {
4240 ShouldNotReachHere();
4241 return false;
4242 }
4243 }
4244 }
4245
4246 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4247 return true;
4248 }
4249
4250 static void do_resume(OSThread* osthread) {
4251 assert(osthread->sr.is_suspended(), "thread should be suspended");
4252 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4253
4254 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4255 // failed to switch to WAKEUP_REQUEST
4256 ShouldNotReachHere();
4257 return;
4258 }
4259
4260 while (true) {
4261 if (sr_notify(osthread) == 0) {
4262 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4263 if (osthread->sr.is_running()) {
4264 return;
4265 }
4266 }
4267 } else {
4268 ShouldNotReachHere();
4269 }
4270 }
4271
4272 guarantee(osthread->sr.is_running(), "Must be running!");
4273 }
4274
4275 ///////////////////////////////////////////////////////////////////////////////////
4276 // signal handling (except suspend/resume)
4277
4278 // This routine may be used by user applications as a "hook" to catch signals.
4279 // The user-defined signal handler must pass unrecognized signals to this
4280 // routine, and if it returns true (non-zero), then the signal handler must
4281 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4282 // routine will never retun false (zero), but instead will execute a VM panic
4283 // routine kill the process.
4284 //
4285 // If this routine returns false, it is OK to call it again. This allows
4286 // the user-defined signal handler to perform checks either before or after
4287 // the VM performs its own checks. Naturally, the user code would be making
4288 // a serious error if it tried to handle an exception (such as a null check
4289 // or breakpoint) that the VM was generating for its own correct operation.
4290 //
4291 // This routine may recognize any of the following kinds of signals:
4292 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4293 // It should be consulted by handlers for any of those signals.
4294 //
4295 // The caller of this routine must pass in the three arguments supplied
4296 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4297 // field of the structure passed to sigaction(). This routine assumes that
4298 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4299 //
4300 // Note that the VM will print warnings if it detects conflicting signal
4301 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4302 //
4303 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4304 siginfo_t* siginfo,
4305 void* ucontext,
4306 int abort_if_unrecognized);
4307
4308 void signalHandler(int sig, siginfo_t* info, void* uc) {
4309 assert(info != NULL && uc != NULL, "it must be old kernel");
4310 int orig_errno = errno; // Preserve errno value over signal handler.
4311 JVM_handle_linux_signal(sig, info, uc, true);
4312 errno = orig_errno;
4313 }
4314
4315
4316 // This boolean allows users to forward their own non-matching signals
4317 // to JVM_handle_linux_signal, harmlessly.
4318 bool os::Linux::signal_handlers_are_installed = false;
4319
4320 // For signal-chaining
4321 struct sigaction sigact[NSIG];
4322 uint64_t sigs = 0;
4323 #if (64 < NSIG-1)
4324 #error "Not all signals can be encoded in sigs. Adapt its type!"
4325 #endif
4326 bool os::Linux::libjsig_is_loaded = false;
4327 typedef struct sigaction *(*get_signal_t)(int);
4328 get_signal_t os::Linux::get_signal_action = NULL;
4329
4330 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4331 struct sigaction *actp = NULL;
4332
4333 if (libjsig_is_loaded) {
4334 // Retrieve the old signal handler from libjsig
4335 actp = (*get_signal_action)(sig);
4336 }
4337 if (actp == NULL) {
4338 // Retrieve the preinstalled signal handler from jvm
4339 actp = get_preinstalled_handler(sig);
4340 }
4341
4342 return actp;
4343 }
4344
4345 static bool call_chained_handler(struct sigaction *actp, int sig,
4346 siginfo_t *siginfo, void *context) {
4347 // Call the old signal handler
4348 if (actp->sa_handler == SIG_DFL) {
4349 // It's more reasonable to let jvm treat it as an unexpected exception
4350 // instead of taking the default action.
4351 return false;
4352 } else if (actp->sa_handler != SIG_IGN) {
4353 if ((actp->sa_flags & SA_NODEFER) == 0) {
4354 // automaticlly block the signal
4355 sigaddset(&(actp->sa_mask), sig);
4356 }
4357
4358 sa_handler_t hand = NULL;
4359 sa_sigaction_t sa = NULL;
4360 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4361 // retrieve the chained handler
4362 if (siginfo_flag_set) {
4363 sa = actp->sa_sigaction;
4364 } else {
4365 hand = actp->sa_handler;
4366 }
4367
4368 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4369 actp->sa_handler = SIG_DFL;
4370 }
4371
4372 // try to honor the signal mask
4373 sigset_t oset;
4374 sigemptyset(&oset);
4375 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4376
4377 // call into the chained handler
4378 if (siginfo_flag_set) {
4379 (*sa)(sig, siginfo, context);
4380 } else {
4381 (*hand)(sig);
4382 }
4383
4384 // restore the signal mask
4385 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4386 }
4387 // Tell jvm's signal handler the signal is taken care of.
4388 return true;
4389 }
4390
4391 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4392 bool chained = false;
4393 // signal-chaining
4394 if (UseSignalChaining) {
4395 struct sigaction *actp = get_chained_signal_action(sig);
4396 if (actp != NULL) {
4397 chained = call_chained_handler(actp, sig, siginfo, context);
4398 }
4399 }
4400 return chained;
4401 }
4402
4403 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4404 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4405 return &sigact[sig];
4406 }
4407 return NULL;
4408 }
4409
4410 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4411 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4412 sigact[sig] = oldAct;
4413 sigs |= (uint64_t)1 << (sig-1);
4414 }
4415
4416 // for diagnostic
4417 int sigflags[NSIG];
4418
4419 int os::Linux::get_our_sigflags(int sig) {
4420 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4421 return sigflags[sig];
4422 }
4423
4424 void os::Linux::set_our_sigflags(int sig, int flags) {
4425 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4426 if (sig > 0 && sig < NSIG) {
4427 sigflags[sig] = flags;
4428 }
4429 }
4430
4431 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4432 // Check for overwrite.
4433 struct sigaction oldAct;
4434 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4435
4436 void* oldhand = oldAct.sa_sigaction
4437 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4438 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4439 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4440 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4441 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4442 if (AllowUserSignalHandlers || !set_installed) {
4443 // Do not overwrite; user takes responsibility to forward to us.
4444 return;
4445 } else if (UseSignalChaining) {
4446 // save the old handler in jvm
4447 save_preinstalled_handler(sig, oldAct);
4448 // libjsig also interposes the sigaction() call below and saves the
4449 // old sigaction on it own.
4450 } else {
4451 fatal("Encountered unexpected pre-existing sigaction handler "
4452 "%#lx for signal %d.", (long)oldhand, sig);
4453 }
4454 }
4455
4456 struct sigaction sigAct;
4457 sigfillset(&(sigAct.sa_mask));
4458 sigAct.sa_handler = SIG_DFL;
4459 if (!set_installed) {
4460 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4461 } else {
4462 sigAct.sa_sigaction = signalHandler;
4463 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4464 }
4465 // Save flags, which are set by ours
4466 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4467 sigflags[sig] = sigAct.sa_flags;
4468
4469 int ret = sigaction(sig, &sigAct, &oldAct);
4470 assert(ret == 0, "check");
4471
4472 void* oldhand2 = oldAct.sa_sigaction
4473 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4474 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4475 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4476 }
4477
4478 // install signal handlers for signals that HotSpot needs to
4479 // handle in order to support Java-level exception handling.
4480
4481 void os::Linux::install_signal_handlers() {
4482 if (!signal_handlers_are_installed) {
4483 signal_handlers_are_installed = true;
4484
4485 // signal-chaining
4486 typedef void (*signal_setting_t)();
4487 signal_setting_t begin_signal_setting = NULL;
4488 signal_setting_t end_signal_setting = NULL;
4489 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4490 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4491 if (begin_signal_setting != NULL) {
4492 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4493 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4494 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4495 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4496 libjsig_is_loaded = true;
4497 assert(UseSignalChaining, "should enable signal-chaining");
4498 }
4499 if (libjsig_is_loaded) {
4500 // Tell libjsig jvm is setting signal handlers
4501 (*begin_signal_setting)();
4502 }
4503
4504 set_signal_handler(SIGSEGV, true);
4505 set_signal_handler(SIGPIPE, true);
4506 set_signal_handler(SIGBUS, true);
4507 set_signal_handler(SIGILL, true);
4508 set_signal_handler(SIGFPE, true);
4509 #if defined(PPC64)
4510 set_signal_handler(SIGTRAP, true);
4511 #endif
4512 set_signal_handler(SIGXFSZ, true);
4513
4514 if (libjsig_is_loaded) {
4515 // Tell libjsig jvm finishes setting signal handlers
4516 (*end_signal_setting)();
4517 }
4518
4519 // We don't activate signal checker if libjsig is in place, we trust ourselves
4520 // and if UserSignalHandler is installed all bets are off.
4521 // Log that signal checking is off only if -verbose:jni is specified.
4522 if (CheckJNICalls) {
4523 if (libjsig_is_loaded) {
4524 if (PrintJNIResolving) {
4525 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4526 }
4527 check_signals = false;
4528 }
4529 if (AllowUserSignalHandlers) {
4530 if (PrintJNIResolving) {
4531 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4532 }
4533 check_signals = false;
4534 }
4535 }
4536 }
4537 }
4538
4539 // This is the fastest way to get thread cpu time on Linux.
4540 // Returns cpu time (user+sys) for any thread, not only for current.
4541 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4542 // It might work on 2.6.10+ with a special kernel/glibc patch.
4543 // For reference, please, see IEEE Std 1003.1-2004:
4544 // http://www.unix.org/single_unix_specification
4545
4546 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4547 struct timespec tp;
4548 int rc = os::Linux::clock_gettime(clockid, &tp);
4549 assert(rc == 0, "clock_gettime is expected to return 0 code");
4550
4551 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4552 }
4553
4554 void os::Linux::initialize_os_info() {
4555 assert(_os_version == 0, "OS info already initialized");
4556
4557 struct utsname _uname;
4558
4559 uint32_t major;
4560 uint32_t minor;
4561 uint32_t fix;
4562
4563 int rc;
4564
4565 // Kernel version is unknown if
4566 // verification below fails.
4567 _os_version = 0x01000000;
4568
4569 rc = uname(&_uname);
4570 if (rc != -1) {
4571
4572 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4573 if (rc == 3) {
4574
4575 if (major < 256 && minor < 256 && fix < 256) {
4576 // Kernel version format is as expected,
4577 // set it overriding unknown state.
4578 _os_version = (major << 16) |
4579 (minor << 8 ) |
4580 (fix << 0 ) ;
4581 }
4582 }
4583 }
4584 }
4585
4586 uint32_t os::Linux::os_version() {
4587 assert(_os_version != 0, "not initialized");
4588 return _os_version & 0x00FFFFFF;
4589 }
4590
4591 bool os::Linux::os_version_is_known() {
4592 assert(_os_version != 0, "not initialized");
4593 return _os_version & 0x01000000 ? false : true;
4594 }
4595
4596 /////
4597 // glibc on Linux platform uses non-documented flag
4598 // to indicate, that some special sort of signal
4599 // trampoline is used.
4600 // We will never set this flag, and we should
4601 // ignore this flag in our diagnostic
4602 #ifdef SIGNIFICANT_SIGNAL_MASK
4603 #undef SIGNIFICANT_SIGNAL_MASK
4604 #endif
4605 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4606
4607 static const char* get_signal_handler_name(address handler,
4608 char* buf, int buflen) {
4609 int offset = 0;
4610 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4611 if (found) {
4612 // skip directory names
4613 const char *p1, *p2;
4614 p1 = buf;
4615 size_t len = strlen(os::file_separator());
4616 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4617 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4618 } else {
4619 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4620 }
4621 return buf;
4622 }
4623
4624 static void print_signal_handler(outputStream* st, int sig,
4625 char* buf, size_t buflen) {
4626 struct sigaction sa;
4627
4628 sigaction(sig, NULL, &sa);
4629
4630 // See comment for SIGNIFICANT_SIGNAL_MASK define
4631 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4632
4633 st->print("%s: ", os::exception_name(sig, buf, buflen));
4634
4635 address handler = (sa.sa_flags & SA_SIGINFO)
4636 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4637 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4638
4639 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4640 st->print("SIG_DFL");
4641 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4642 st->print("SIG_IGN");
4643 } else {
4644 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4645 }
4646
4647 st->print(", sa_mask[0]=");
4648 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4649
4650 address rh = VMError::get_resetted_sighandler(sig);
4651 // May be, handler was resetted by VMError?
4652 if (rh != NULL) {
4653 handler = rh;
4654 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4655 }
4656
4657 st->print(", sa_flags=");
4658 os::Posix::print_sa_flags(st, sa.sa_flags);
4659
4660 // Check: is it our handler?
4661 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4662 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4663 // It is our signal handler
4664 // check for flags, reset system-used one!
4665 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4666 st->print(
4667 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4668 os::Linux::get_our_sigflags(sig));
4669 }
4670 }
4671 st->cr();
4672 }
4673
4674
4675 #define DO_SIGNAL_CHECK(sig) \
4676 do { \
4677 if (!sigismember(&check_signal_done, sig)) { \
4678 os::Linux::check_signal_handler(sig); \
4679 } \
4680 } while (0)
4681
4682 // This method is a periodic task to check for misbehaving JNI applications
4683 // under CheckJNI, we can add any periodic checks here
4684
4685 void os::run_periodic_checks() {
4686 if (check_signals == false) return;
4687
4688 // SEGV and BUS if overridden could potentially prevent
4689 // generation of hs*.log in the event of a crash, debugging
4690 // such a case can be very challenging, so we absolutely
4691 // check the following for a good measure:
4692 DO_SIGNAL_CHECK(SIGSEGV);
4693 DO_SIGNAL_CHECK(SIGILL);
4694 DO_SIGNAL_CHECK(SIGFPE);
4695 DO_SIGNAL_CHECK(SIGBUS);
4696 DO_SIGNAL_CHECK(SIGPIPE);
4697 DO_SIGNAL_CHECK(SIGXFSZ);
4698 #if defined(PPC64)
4699 DO_SIGNAL_CHECK(SIGTRAP);
4700 #endif
4701
4702 // ReduceSignalUsage allows the user to override these handlers
4703 // see comments at the very top and jvm_solaris.h
4704 if (!ReduceSignalUsage) {
4705 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4706 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4707 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4708 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4709 }
4710
4711 DO_SIGNAL_CHECK(SR_signum);
4712 }
4713
4714 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4715
4716 static os_sigaction_t os_sigaction = NULL;
4717
4718 void os::Linux::check_signal_handler(int sig) {
4719 char buf[O_BUFLEN];
4720 address jvmHandler = NULL;
4721
4722
4723 struct sigaction act;
4724 if (os_sigaction == NULL) {
4725 // only trust the default sigaction, in case it has been interposed
4726 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4727 if (os_sigaction == NULL) return;
4728 }
4729
4730 os_sigaction(sig, (struct sigaction*)NULL, &act);
4731
4732
4733 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4734
4735 address thisHandler = (act.sa_flags & SA_SIGINFO)
4736 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4737 : CAST_FROM_FN_PTR(address, act.sa_handler);
4738
4739
4740 switch (sig) {
4741 case SIGSEGV:
4742 case SIGBUS:
4743 case SIGFPE:
4744 case SIGPIPE:
4745 case SIGILL:
4746 case SIGXFSZ:
4747 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4748 break;
4749
4750 case SHUTDOWN1_SIGNAL:
4751 case SHUTDOWN2_SIGNAL:
4752 case SHUTDOWN3_SIGNAL:
4753 case BREAK_SIGNAL:
4754 jvmHandler = (address)user_handler();
4755 break;
4756
4757 default:
4758 if (sig == SR_signum) {
4759 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4760 } else {
4761 return;
4762 }
4763 break;
4764 }
4765
4766 if (thisHandler != jvmHandler) {
4767 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4768 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4769 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4770 // No need to check this sig any longer
4771 sigaddset(&check_signal_done, sig);
4772 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4773 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4774 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4775 exception_name(sig, buf, O_BUFLEN));
4776 }
4777 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4778 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4779 tty->print("expected:");
4780 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4781 tty->cr();
4782 tty->print(" found:");
4783 os::Posix::print_sa_flags(tty, act.sa_flags);
4784 tty->cr();
4785 // No need to check this sig any longer
4786 sigaddset(&check_signal_done, sig);
4787 }
4788
4789 // Dump all the signal
4790 if (sigismember(&check_signal_done, sig)) {
4791 print_signal_handlers(tty, buf, O_BUFLEN);
4792 }
4793 }
4794
4795 extern void report_error(char* file_name, int line_no, char* title,
4796 char* format, ...);
4797
4798 // this is called _before_ the most of global arguments have been parsed
4799 void os::init(void) {
4800 char dummy; // used to get a guess on initial stack address
4801 // first_hrtime = gethrtime();
4802
4803 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4804
4805 init_random(1234567);
4806
4807 ThreadCritical::initialize();
4808
4809 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4810 if (Linux::page_size() == -1) {
4811 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4812 os::strerror(errno));
4813 }
4814 init_page_sizes((size_t) Linux::page_size());
4815
4816 Linux::initialize_system_info();
4817
4818 Linux::initialize_os_info();
4819
4820 // main_thread points to the aboriginal thread
4821 Linux::_main_thread = pthread_self();
4822
4823 Linux::clock_init();
4824 initial_time_count = javaTimeNanos();
4825
4826 // retrieve entry point for pthread_setname_np
4827 Linux::_pthread_setname_np =
4828 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4829
4830 os::Posix::init();
4831 }
4832
4833 // To install functions for atexit system call
4834 extern "C" {
4835 static void perfMemory_exit_helper() {
4836 perfMemory_exit();
4837 }
4838 }
4839
4840 // this is called _after_ the global arguments have been parsed
4841 jint os::init_2(void) {
4842
4843 os::Posix::init_2();
4844
4845 Linux::fast_thread_clock_init();
4846
4847 // Allocate a single page and mark it as readable for safepoint polling
4848 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4849 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4850
4851 os::set_polling_page(polling_page);
4852 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4853
4854 if (!UseMembar) {
4855 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4856 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4857 os::set_memory_serialize_page(mem_serialize_page);
4858 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4859 }
4860
4861 // initialize suspend/resume support - must do this before signal_sets_init()
4862 if (SR_initialize() != 0) {
4863 perror("SR_initialize failed");
4864 return JNI_ERR;
4865 }
4866
4867 Linux::signal_sets_init();
4868 Linux::install_signal_handlers();
4869
4870 // Check and sets minimum stack sizes against command line options
4871 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4872 return JNI_ERR;
4873 }
4874 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4875
4876 #if defined(IA32)
4877 workaround_expand_exec_shield_cs_limit();
4878 #endif
4879
4880 Linux::libpthread_init();
4881 Linux::sched_getcpu_init();
4882 log_info(os)("HotSpot is running with %s, %s",
4883 Linux::glibc_version(), Linux::libpthread_version());
4884
4885 if (UseNUMA) {
4886 if (!Linux::libnuma_init()) {
4887 UseNUMA = false;
4888 } else {
4889 if ((Linux::numa_max_node() < 1)) {
4890 // There's only one node(they start from 0), disable NUMA.
4891 UseNUMA = false;
4892 }
4893 }
4894 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4895 // we can make the adaptive lgrp chunk resizing work. If the user specified
4896 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4897 // disable adaptive resizing.
4898 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4899 if (FLAG_IS_DEFAULT(UseNUMA)) {
4900 UseNUMA = false;
4901 } else {
4902 if (FLAG_IS_DEFAULT(UseLargePages) &&
4903 FLAG_IS_DEFAULT(UseSHM) &&
4904 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4905 UseLargePages = false;
4906 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4907 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4908 UseAdaptiveSizePolicy = false;
4909 UseAdaptiveNUMAChunkSizing = false;
4910 }
4911 }
4912 }
4913 if (!UseNUMA && ForceNUMA) {
4914 UseNUMA = true;
4915 }
4916 }
4917
4918 if (MaxFDLimit) {
4919 // set the number of file descriptors to max. print out error
4920 // if getrlimit/setrlimit fails but continue regardless.
4921 struct rlimit nbr_files;
4922 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4923 if (status != 0) {
4924 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4925 } else {
4926 nbr_files.rlim_cur = nbr_files.rlim_max;
4927 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4928 if (status != 0) {
4929 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4930 }
4931 }
4932 }
4933
4934 // Initialize lock used to serialize thread creation (see os::create_thread)
4935 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4936
4937 // at-exit methods are called in the reverse order of their registration.
4938 // atexit functions are called on return from main or as a result of a
4939 // call to exit(3C). There can be only 32 of these functions registered
4940 // and atexit() does not set errno.
4941
4942 if (PerfAllowAtExitRegistration) {
4943 // only register atexit functions if PerfAllowAtExitRegistration is set.
4944 // atexit functions can be delayed until process exit time, which
4945 // can be problematic for embedded VM situations. Embedded VMs should
4946 // call DestroyJavaVM() to assure that VM resources are released.
4947
4948 // note: perfMemory_exit_helper atexit function may be removed in
4949 // the future if the appropriate cleanup code can be added to the
4950 // VM_Exit VMOperation's doit method.
4951 if (atexit(perfMemory_exit_helper) != 0) {
4952 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4953 }
4954 }
4955
4956 // initialize thread priority policy
4957 prio_init();
4958
4959 return JNI_OK;
4960 }
4961
4962 // Mark the polling page as unreadable
4963 void os::make_polling_page_unreadable(void) {
4964 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4965 fatal("Could not disable polling page");
4966 }
4967 }
4968
4969 // Mark the polling page as readable
4970 void os::make_polling_page_readable(void) {
4971 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4972 fatal("Could not enable polling page");
4973 }
4974 }
4975
4976 // older glibc versions don't have this macro (which expands to
4977 // an optimized bit-counting function) so we have to roll our own
4978 #ifndef CPU_COUNT
4979
4980 static int _cpu_count(const cpu_set_t* cpus) {
4981 int count = 0;
4982 // only look up to the number of configured processors
4983 for (int i = 0; i < os::processor_count(); i++) {
4984 if (CPU_ISSET(i, cpus)) {
4985 count++;
4986 }
4987 }
4988 return count;
4989 }
4990
4991 #define CPU_COUNT(cpus) _cpu_count(cpus)
4992
4993 #endif // CPU_COUNT
4994
4995 // Get the current number of available processors for this process.
4996 // This value can change at any time during a process's lifetime.
4997 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4998 // If it appears there may be more than 1024 processors then we do a
4999 // dynamic check - see 6515172 for details.
5000 // If anything goes wrong we fallback to returning the number of online
5001 // processors - which can be greater than the number available to the process.
5002 int os::active_processor_count() {
5003 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5004 cpu_set_t* cpus_p = &cpus;
5005 int cpus_size = sizeof(cpu_set_t);
5006
5007 int configured_cpus = processor_count(); // upper bound on available cpus
5008 int cpu_count = 0;
5009
5010 // old build platforms may not support dynamic cpu sets
5011 #ifdef CPU_ALLOC
5012
5013 // To enable easy testing of the dynamic path on different platforms we
5014 // introduce a diagnostic flag: UseCpuAllocPath
5015 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5016 // kernel may use a mask bigger than cpu_set_t
5017 log_trace(os)("active_processor_count: using dynamic path %s"
5018 "- configured processors: %d",
5019 UseCpuAllocPath ? "(forced) " : "",
5020 configured_cpus);
5021 cpus_p = CPU_ALLOC(configured_cpus);
5022 if (cpus_p != NULL) {
5023 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5024 // zero it just to be safe
5025 CPU_ZERO_S(cpus_size, cpus_p);
5026 }
5027 else {
5028 // failed to allocate so fallback to online cpus
5029 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5030 log_trace(os)("active_processor_count: "
5031 "CPU_ALLOC failed (%s) - using "
5032 "online processor count: %d",
5033 os::strerror(errno), online_cpus);
5034 return online_cpus;
5035 }
5036 }
5037 else {
5038 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5039 configured_cpus);
5040 }
5041 #else // CPU_ALLOC
5042 // these stubs won't be executed
5043 #define CPU_COUNT_S(size, cpus) -1
5044 #define CPU_FREE(cpus)
5045
5046 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5047 configured_cpus);
5048 #endif // CPU_ALLOC
5049
5050 // pid 0 means the current thread - which we have to assume represents the process
5051 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5052 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5053 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5054 }
5055 else {
5056 cpu_count = CPU_COUNT(cpus_p);
5057 }
5058 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5059 }
5060 else {
5061 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5062 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5063 "which may exceed available processors", os::strerror(errno), cpu_count);
5064 }
5065
5066 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5067 CPU_FREE(cpus_p);
5068 }
5069
5070 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5071 return cpu_count;
5072 }
5073
5074 void os::set_native_thread_name(const char *name) {
5075 if (Linux::_pthread_setname_np) {
5076 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5077 snprintf(buf, sizeof(buf), "%s", name);
5078 buf[sizeof(buf) - 1] = '\0';
5079 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5080 // ERANGE should not happen; all other errors should just be ignored.
5081 assert(rc != ERANGE, "pthread_setname_np failed");
5082 }
5083 }
5084
5085 bool os::distribute_processes(uint length, uint* distribution) {
5086 // Not yet implemented.
5087 return false;
5088 }
5089
5090 bool os::bind_to_processor(uint processor_id) {
5091 // Not yet implemented.
5092 return false;
5093 }
5094
5095 ///
5096
5097 void os::SuspendedThreadTask::internal_do_task() {
5098 if (do_suspend(_thread->osthread())) {
5099 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5100 do_task(context);
5101 do_resume(_thread->osthread());
5102 }
5103 }
5104
5105 ////////////////////////////////////////////////////////////////////////////////
5106 // debug support
5107
5108 bool os::find(address addr, outputStream* st) {
5109 Dl_info dlinfo;
5110 memset(&dlinfo, 0, sizeof(dlinfo));
5111 if (dladdr(addr, &dlinfo) != 0) {
5112 st->print(PTR_FORMAT ": ", p2i(addr));
5113 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5114 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5115 p2i(addr) - p2i(dlinfo.dli_saddr));
5116 } else if (dlinfo.dli_fbase != NULL) {
5117 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5118 } else {
5119 st->print("<absolute address>");
5120 }
5121 if (dlinfo.dli_fname != NULL) {
5122 st->print(" in %s", dlinfo.dli_fname);
5123 }
5124 if (dlinfo.dli_fbase != NULL) {
5125 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5126 }
5127 st->cr();
5128
5129 if (Verbose) {
5130 // decode some bytes around the PC
5131 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5132 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5133 address lowest = (address) dlinfo.dli_sname;
5134 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5135 if (begin < lowest) begin = lowest;
5136 Dl_info dlinfo2;
5137 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5138 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5139 end = (address) dlinfo2.dli_saddr;
5140 }
5141 Disassembler::decode(begin, end, st);
5142 }
5143 return true;
5144 }
5145 return false;
5146 }
5147
5148 ////////////////////////////////////////////////////////////////////////////////
5149 // misc
5150
5151 // This does not do anything on Linux. This is basically a hook for being
5152 // able to use structured exception handling (thread-local exception filters)
5153 // on, e.g., Win32.
5154 void
5155 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5156 JavaCallArguments* args, Thread* thread) {
5157 f(value, method, args, thread);
5158 }
5159
5160 void os::print_statistics() {
5161 }
5162
5163 bool os::message_box(const char* title, const char* message) {
5164 int i;
5165 fdStream err(defaultStream::error_fd());
5166 for (i = 0; i < 78; i++) err.print_raw("=");
5167 err.cr();
5168 err.print_raw_cr(title);
5169 for (i = 0; i < 78; i++) err.print_raw("-");
5170 err.cr();
5171 err.print_raw_cr(message);
5172 for (i = 0; i < 78; i++) err.print_raw("=");
5173 err.cr();
5174
5175 char buf[16];
5176 // Prevent process from exiting upon "read error" without consuming all CPU
5177 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5178
5179 return buf[0] == 'y' || buf[0] == 'Y';
5180 }
5181
5182 int os::stat(const char *path, struct stat *sbuf) {
5183 char pathbuf[MAX_PATH];
5184 if (strlen(path) > MAX_PATH - 1) {
5185 errno = ENAMETOOLONG;
5186 return -1;
5187 }
5188 os::native_path(strcpy(pathbuf, path));
5189 return ::stat(pathbuf, sbuf);
5190 }
5191
5192 // Is a (classpath) directory empty?
5193 bool os::dir_is_empty(const char* path) {
5194 DIR *dir = NULL;
5195 struct dirent *ptr;
5196
5197 dir = opendir(path);
5198 if (dir == NULL) return true;
5199
5200 // Scan the directory
5201 bool result = true;
5202 char buf[sizeof(struct dirent) + MAX_PATH];
5203 while (result && (ptr = ::readdir(dir)) != NULL) {
5204 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5205 result = false;
5206 }
5207 }
5208 closedir(dir);
5209 return result;
5210 }
5211
5212 // This code originates from JDK's sysOpen and open64_w
5213 // from src/solaris/hpi/src/system_md.c
5214
5215 int os::open(const char *path, int oflag, int mode) {
5216 if (strlen(path) > MAX_PATH - 1) {
5217 errno = ENAMETOOLONG;
5218 return -1;
5219 }
5220
5221 // All file descriptors that are opened in the Java process and not
5222 // specifically destined for a subprocess should have the close-on-exec
5223 // flag set. If we don't set it, then careless 3rd party native code
5224 // might fork and exec without closing all appropriate file descriptors
5225 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5226 // turn might:
5227 //
5228 // - cause end-of-file to fail to be detected on some file
5229 // descriptors, resulting in mysterious hangs, or
5230 //
5231 // - might cause an fopen in the subprocess to fail on a system
5232 // suffering from bug 1085341.
5233 //
5234 // (Yes, the default setting of the close-on-exec flag is a Unix
5235 // design flaw)
5236 //
5237 // See:
5238 // 1085341: 32-bit stdio routines should support file descriptors >255
5239 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5240 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5241 //
5242 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5243 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5244 // because it saves a system call and removes a small window where the flag
5245 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5246 // and we fall back to using FD_CLOEXEC (see below).
5247 #ifdef O_CLOEXEC
5248 oflag |= O_CLOEXEC;
5249 #endif
5250
5251 int fd = ::open64(path, oflag, mode);
5252 if (fd == -1) return -1;
5253
5254 //If the open succeeded, the file might still be a directory
5255 {
5256 struct stat64 buf64;
5257 int ret = ::fstat64(fd, &buf64);
5258 int st_mode = buf64.st_mode;
5259
5260 if (ret != -1) {
5261 if ((st_mode & S_IFMT) == S_IFDIR) {
5262 errno = EISDIR;
5263 ::close(fd);
5264 return -1;
5265 }
5266 } else {
5267 ::close(fd);
5268 return -1;
5269 }
5270 }
5271
5272 #ifdef FD_CLOEXEC
5273 // Validate that the use of the O_CLOEXEC flag on open above worked.
5274 // With recent kernels, we will perform this check exactly once.
5275 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5276 if (!O_CLOEXEC_is_known_to_work) {
5277 int flags = ::fcntl(fd, F_GETFD);
5278 if (flags != -1) {
5279 if ((flags & FD_CLOEXEC) != 0)
5280 O_CLOEXEC_is_known_to_work = 1;
5281 else
5282 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5283 }
5284 }
5285 #endif
5286
5287 return fd;
5288 }
5289
5290
5291 // create binary file, rewriting existing file if required
5292 int os::create_binary_file(const char* path, bool rewrite_existing) {
5293 int oflags = O_WRONLY | O_CREAT;
5294 if (!rewrite_existing) {
5295 oflags |= O_EXCL;
5296 }
5297 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5298 }
5299
5300 // return current position of file pointer
5301 jlong os::current_file_offset(int fd) {
5302 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5303 }
5304
5305 // move file pointer to the specified offset
5306 jlong os::seek_to_file_offset(int fd, jlong offset) {
5307 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5308 }
5309
5310 // This code originates from JDK's sysAvailable
5311 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5312
5313 int os::available(int fd, jlong *bytes) {
5314 jlong cur, end;
5315 int mode;
5316 struct stat64 buf64;
5317
5318 if (::fstat64(fd, &buf64) >= 0) {
5319 mode = buf64.st_mode;
5320 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5321 int n;
5322 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5323 *bytes = n;
5324 return 1;
5325 }
5326 }
5327 }
5328 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5329 return 0;
5330 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5331 return 0;
5332 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5333 return 0;
5334 }
5335 *bytes = end - cur;
5336 return 1;
5337 }
5338
5339 // Map a block of memory.
5340 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5341 char *addr, size_t bytes, bool read_only,
5342 bool allow_exec) {
5343 int prot;
5344 int flags = MAP_PRIVATE;
5345
5346 if (read_only) {
5347 prot = PROT_READ;
5348 } else {
5349 prot = PROT_READ | PROT_WRITE;
5350 }
5351
5352 if (allow_exec) {
5353 prot |= PROT_EXEC;
5354 }
5355
5356 if (addr != NULL) {
5357 flags |= MAP_FIXED;
5358 }
5359
5360 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5361 fd, file_offset);
5362 if (mapped_address == MAP_FAILED) {
5363 return NULL;
5364 }
5365 return mapped_address;
5366 }
5367
5368
5369 // Remap a block of memory.
5370 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5371 char *addr, size_t bytes, bool read_only,
5372 bool allow_exec) {
5373 // same as map_memory() on this OS
5374 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5375 allow_exec);
5376 }
5377
5378
5379 // Unmap a block of memory.
5380 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5381 return munmap(addr, bytes) == 0;
5382 }
5383
5384 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5385
5386 static clockid_t thread_cpu_clockid(Thread* thread) {
5387 pthread_t tid = thread->osthread()->pthread_id();
5388 clockid_t clockid;
5389
5390 // Get thread clockid
5391 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5392 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5393 return clockid;
5394 }
5395
5396 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5397 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5398 // of a thread.
5399 //
5400 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5401 // the fast estimate available on the platform.
5402
5403 jlong os::current_thread_cpu_time() {
5404 if (os::Linux::supports_fast_thread_cpu_time()) {
5405 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5406 } else {
5407 // return user + sys since the cost is the same
5408 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5409 }
5410 }
5411
5412 jlong os::thread_cpu_time(Thread* thread) {
5413 // consistent with what current_thread_cpu_time() returns
5414 if (os::Linux::supports_fast_thread_cpu_time()) {
5415 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5416 } else {
5417 return slow_thread_cpu_time(thread, true /* user + sys */);
5418 }
5419 }
5420
5421 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5422 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5423 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5424 } else {
5425 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5426 }
5427 }
5428
5429 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5430 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5431 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5432 } else {
5433 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5434 }
5435 }
5436
5437 // -1 on error.
5438 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5439 pid_t tid = thread->osthread()->thread_id();
5440 char *s;
5441 char stat[2048];
5442 int statlen;
5443 char proc_name[64];
5444 int count;
5445 long sys_time, user_time;
5446 char cdummy;
5447 int idummy;
5448 long ldummy;
5449 FILE *fp;
5450
5451 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5452 fp = fopen(proc_name, "r");
5453 if (fp == NULL) return -1;
5454 statlen = fread(stat, 1, 2047, fp);
5455 stat[statlen] = '\0';
5456 fclose(fp);
5457
5458 // Skip pid and the command string. Note that we could be dealing with
5459 // weird command names, e.g. user could decide to rename java launcher
5460 // to "java 1.4.2 :)", then the stat file would look like
5461 // 1234 (java 1.4.2 :)) R ... ...
5462 // We don't really need to know the command string, just find the last
5463 // occurrence of ")" and then start parsing from there. See bug 4726580.
5464 s = strrchr(stat, ')');
5465 if (s == NULL) return -1;
5466
5467 // Skip blank chars
5468 do { s++; } while (s && isspace(*s));
5469
5470 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5471 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5472 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5473 &user_time, &sys_time);
5474 if (count != 13) return -1;
5475 if (user_sys_cpu_time) {
5476 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5477 } else {
5478 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5479 }
5480 }
5481
5482 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5483 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5484 info_ptr->may_skip_backward = false; // elapsed time not wall time
5485 info_ptr->may_skip_forward = false; // elapsed time not wall time
5486 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5487 }
5488
5489 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5490 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5491 info_ptr->may_skip_backward = false; // elapsed time not wall time
5492 info_ptr->may_skip_forward = false; // elapsed time not wall time
5493 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5494 }
5495
5496 bool os::is_thread_cpu_time_supported() {
5497 return true;
5498 }
5499
5500 // System loadavg support. Returns -1 if load average cannot be obtained.
5501 // Linux doesn't yet have a (official) notion of processor sets,
5502 // so just return the system wide load average.
5503 int os::loadavg(double loadavg[], int nelem) {
5504 return ::getloadavg(loadavg, nelem);
5505 }
5506
5507 void os::pause() {
5508 char filename[MAX_PATH];
5509 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5510 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5511 } else {
5512 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5513 }
5514
5515 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5516 if (fd != -1) {
5517 struct stat buf;
5518 ::close(fd);
5519 while (::stat(filename, &buf) == 0) {
5520 (void)::poll(NULL, 0, 100);
5521 }
5522 } else {
5523 jio_fprintf(stderr,
5524 "Could not open pause file '%s', continuing immediately.\n", filename);
5525 }
5526 }
5527
5528 extern char** environ;
5529
5530 // Run the specified command in a separate process. Return its exit value,
5531 // or -1 on failure (e.g. can't fork a new process).
5532 // Unlike system(), this function can be called from signal handler. It
5533 // doesn't block SIGINT et al.
5534 int os::fork_and_exec(char* cmd) {
5535 const char * argv[4] = {"sh", "-c", cmd, NULL};
5536
5537 pid_t pid = fork();
5538
5539 if (pid < 0) {
5540 // fork failed
5541 return -1;
5542
5543 } else if (pid == 0) {
5544 // child process
5545
5546 execve("/bin/sh", (char* const*)argv, environ);
5547
5548 // execve failed
5549 _exit(-1);
5550
5551 } else {
5552 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5553 // care about the actual exit code, for now.
5554
5555 int status;
5556
5557 // Wait for the child process to exit. This returns immediately if
5558 // the child has already exited. */
5559 while (waitpid(pid, &status, 0) < 0) {
5560 switch (errno) {
5561 case ECHILD: return 0;
5562 case EINTR: break;
5563 default: return -1;
5564 }
5565 }
5566
5567 if (WIFEXITED(status)) {
5568 // The child exited normally; get its exit code.
5569 return WEXITSTATUS(status);
5570 } else if (WIFSIGNALED(status)) {
5571 // The child exited because of a signal
5572 // The best value to return is 0x80 + signal number,
5573 // because that is what all Unix shells do, and because
5574 // it allows callers to distinguish between process exit and
5575 // process death by signal.
5576 return 0x80 + WTERMSIG(status);
5577 } else {
5578 // Unknown exit code; pass it through
5579 return status;
5580 }
5581 }
5582 }
5583
5584 // is_headless_jre()
5585 //
5586 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5587 // in order to report if we are running in a headless jre
5588 //
5589 // Since JDK8 xawt/libmawt.so was moved into the same directory
5590 // as libawt.so, and renamed libawt_xawt.so
5591 //
5592 bool os::is_headless_jre() {
5593 struct stat statbuf;
5594 char buf[MAXPATHLEN];
5595 char libmawtpath[MAXPATHLEN];
5596 const char *xawtstr = "/xawt/libmawt.so";
5597 const char *new_xawtstr = "/libawt_xawt.so";
5598 char *p;
5599
5600 // Get path to libjvm.so
5601 os::jvm_path(buf, sizeof(buf));
5602
5603 // Get rid of libjvm.so
5604 p = strrchr(buf, '/');
5605 if (p == NULL) {
5606 return false;
5607 } else {
5608 *p = '\0';
5609 }
5610
5611 // Get rid of client or server
5612 p = strrchr(buf, '/');
5613 if (p == NULL) {
5614 return false;
5615 } else {
5616 *p = '\0';
5617 }
5618
5619 // check xawt/libmawt.so
5620 strcpy(libmawtpath, buf);
5621 strcat(libmawtpath, xawtstr);
5622 if (::stat(libmawtpath, &statbuf) == 0) return false;
5623
5624 // check libawt_xawt.so
5625 strcpy(libmawtpath, buf);
5626 strcat(libmawtpath, new_xawtstr);
5627 if (::stat(libmawtpath, &statbuf) == 0) return false;
5628
5629 return true;
5630 }
5631
5632 // Get the default path to the core file
5633 // Returns the length of the string
5634 int os::get_core_path(char* buffer, size_t bufferSize) {
5635 /*
5636 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5637 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5638 */
5639 const int core_pattern_len = 129;
5640 char core_pattern[core_pattern_len] = {0};
5641
5642 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5643 if (core_pattern_file == -1) {
5644 return -1;
5645 }
5646
5647 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5648 ::close(core_pattern_file);
5649 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5650 return -1;
5651 }
5652 if (core_pattern[ret-1] == '\n') {
5653 core_pattern[ret-1] = '\0';
5654 } else {
5655 core_pattern[ret] = '\0';
5656 }
5657
5658 char *pid_pos = strstr(core_pattern, "%p");
5659 int written;
5660
5661 if (core_pattern[0] == '/') {
5662 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5663 } else {
5664 char cwd[PATH_MAX];
5665
5666 const char* p = get_current_directory(cwd, PATH_MAX);
5667 if (p == NULL) {
5668 return -1;
5669 }
5670
5671 if (core_pattern[0] == '|') {
5672 written = jio_snprintf(buffer, bufferSize,
5673 "\"%s\" (or dumping to %s/core.%d)",
5674 &core_pattern[1], p, current_process_id());
5675 } else {
5676 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5677 }
5678 }
5679
5680 if (written < 0) {
5681 return -1;
5682 }
5683
5684 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5685 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5686
5687 if (core_uses_pid_file != -1) {
5688 char core_uses_pid = 0;
5689 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5690 ::close(core_uses_pid_file);
5691
5692 if (core_uses_pid == '1') {
5693 jio_snprintf(buffer + written, bufferSize - written,
5694 ".%d", current_process_id());
5695 }
5696 }
5697 }
5698
5699 return strlen(buffer);
5700 }
5701
5702 bool os::start_debugging(char *buf, int buflen) {
5703 int len = (int)strlen(buf);
5704 char *p = &buf[len];
5705
5706 jio_snprintf(p, buflen-len,
5707 "\n\n"
5708 "Do you want to debug the problem?\n\n"
5709 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5710 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5711 "Otherwise, press RETURN to abort...",
5712 os::current_process_id(), os::current_process_id(),
5713 os::current_thread_id(), os::current_thread_id());
5714
5715 bool yes = os::message_box("Unexpected Error", buf);
5716
5717 if (yes) {
5718 // yes, user asked VM to launch debugger
5719 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5720 os::current_process_id(), os::current_process_id());
5721
5722 os::fork_and_exec(buf);
5723 yes = false;
5724 }
5725 return yes;
5726 }
5727
5728
5729 // Java/Compiler thread:
5730 //
5731 // Low memory addresses
5732 // P0 +------------------------+
5733 // | |\ Java thread created by VM does not have glibc
5734 // | glibc guard page | - guard page, attached Java thread usually has
5735 // | |/ 1 glibc guard page.
5736 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5737 // | |\
5738 // | HotSpot Guard Pages | - red, yellow and reserved pages
5739 // | |/
5740 // +------------------------+ JavaThread::stack_reserved_zone_base()
5741 // | |\
5742 // | Normal Stack | -
5743 // | |/
5744 // P2 +------------------------+ Thread::stack_base()
5745 //
5746 // Non-Java thread:
5747 //
5748 // Low memory addresses
5749 // P0 +------------------------+
5750 // | |\
5751 // | glibc guard page | - usually 1 page
5752 // | |/
5753 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5754 // | |\
5755 // | Normal Stack | -
5756 // | |/
5757 // P2 +------------------------+ Thread::stack_base()
5758 //
5759 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5760 // returned from pthread_attr_getstack().
5761 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5762 // of the stack size given in pthread_attr. We work around this for
5763 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5764 //
5765 #ifndef ZERO
5766 static void current_stack_region(address * bottom, size_t * size) {
5767 if (os::Linux::is_initial_thread()) {
5768 // initial thread needs special handling because pthread_getattr_np()
5769 // may return bogus value.
5770 *bottom = os::Linux::initial_thread_stack_bottom();
5771 *size = os::Linux::initial_thread_stack_size();
5772 } else {
5773 pthread_attr_t attr;
5774
5775 int rslt = pthread_getattr_np(pthread_self(), &attr);
5776
5777 // JVM needs to know exact stack location, abort if it fails
5778 if (rslt != 0) {
5779 if (rslt == ENOMEM) {
5780 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5781 } else {
5782 fatal("pthread_getattr_np failed with error = %d", rslt);
5783 }
5784 }
5785
5786 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5787 fatal("Cannot locate current stack attributes!");
5788 }
5789
5790 // Work around NPTL stack guard error.
5791 size_t guard_size = 0;
5792 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5793 if (rslt != 0) {
5794 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5795 }
5796 *bottom += guard_size;
5797 *size -= guard_size;
5798
5799 pthread_attr_destroy(&attr);
5800
5801 }
5802 assert(os::current_stack_pointer() >= *bottom &&
5803 os::current_stack_pointer() < *bottom + *size, "just checking");
5804 }
5805
5806 address os::current_stack_base() {
5807 address bottom;
5808 size_t size;
5809 current_stack_region(&bottom, &size);
5810 return (bottom + size);
5811 }
5812
5813 size_t os::current_stack_size() {
5814 // This stack size includes the usable stack and HotSpot guard pages
5815 // (for the threads that have Hotspot guard pages).
5816 address bottom;
5817 size_t size;
5818 current_stack_region(&bottom, &size);
5819 return size;
5820 }
5821 #endif
5822
5823 static inline struct timespec get_mtime(const char* filename) {
5824 struct stat st;
5825 int ret = os::stat(filename, &st);
5826 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5827 return st.st_mtim;
5828 }
5829
5830 int os::compare_file_modified_times(const char* file1, const char* file2) {
5831 struct timespec filetime1 = get_mtime(file1);
5832 struct timespec filetime2 = get_mtime(file2);
5833 int diff = filetime1.tv_sec - filetime2.tv_sec;
5834 if (diff == 0) {
5835 return filetime1.tv_nsec - filetime2.tv_nsec;
5836 }
5837 return diff;
5838 }
5839
5840 /////////////// Unit tests ///////////////
5841
5842 #ifndef PRODUCT
5843
5844 #define test_log(...) \
5845 do { \
5846 if (VerboseInternalVMTests) { \
5847 tty->print_cr(__VA_ARGS__); \
5848 tty->flush(); \
5849 } \
5850 } while (false)
5851
5852 class TestReserveMemorySpecial : AllStatic {
5853 public:
5854 static void small_page_write(void* addr, size_t size) {
5855 size_t page_size = os::vm_page_size();
5856
5857 char* end = (char*)addr + size;
5858 for (char* p = (char*)addr; p < end; p += page_size) {
5859 *p = 1;
5860 }
5861 }
5862
5863 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5864 if (!UseHugeTLBFS) {
5865 return;
5866 }
5867
5868 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5869
5870 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5871
5872 if (addr != NULL) {
5873 small_page_write(addr, size);
5874
5875 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5876 }
5877 }
5878
5879 static void test_reserve_memory_special_huge_tlbfs_only() {
5880 if (!UseHugeTLBFS) {
5881 return;
5882 }
5883
5884 size_t lp = os::large_page_size();
5885
5886 for (size_t size = lp; size <= lp * 10; size += lp) {
5887 test_reserve_memory_special_huge_tlbfs_only(size);
5888 }
5889 }
5890
5891 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5892 size_t lp = os::large_page_size();
5893 size_t ag = os::vm_allocation_granularity();
5894
5895 // sizes to test
5896 const size_t sizes[] = {
5897 lp, lp + ag, lp + lp / 2, lp * 2,
5898 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5899 lp * 10, lp * 10 + lp / 2
5900 };
5901 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5902
5903 // For each size/alignment combination, we test three scenarios:
5904 // 1) with req_addr == NULL
5905 // 2) with a non-null req_addr at which we expect to successfully allocate
5906 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5907 // expect the allocation to either fail or to ignore req_addr
5908
5909 // Pre-allocate two areas; they shall be as large as the largest allocation
5910 // and aligned to the largest alignment we will be testing.
5911 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5912 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5913 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5914 -1, 0);
5915 assert(mapping1 != MAP_FAILED, "should work");
5916
5917 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5918 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5919 -1, 0);
5920 assert(mapping2 != MAP_FAILED, "should work");
5921
5922 // Unmap the first mapping, but leave the second mapping intact: the first
5923 // mapping will serve as a value for a "good" req_addr (case 2). The second
5924 // mapping, still intact, as "bad" req_addr (case 3).
5925 ::munmap(mapping1, mapping_size);
5926
5927 // Case 1
5928 test_log("%s, req_addr NULL:", __FUNCTION__);
5929 test_log("size align result");
5930
5931 for (int i = 0; i < num_sizes; i++) {
5932 const size_t size = sizes[i];
5933 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5934 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5935 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5936 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5937 if (p != NULL) {
5938 assert(is_aligned(p, alignment), "must be");
5939 small_page_write(p, size);
5940 os::Linux::release_memory_special_huge_tlbfs(p, size);
5941 }
5942 }
5943 }
5944
5945 // Case 2
5946 test_log("%s, req_addr non-NULL:", __FUNCTION__);
5947 test_log("size align req_addr result");
5948
5949 for (int i = 0; i < num_sizes; i++) {
5950 const size_t size = sizes[i];
5951 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5952 char* const req_addr = align_up(mapping1, alignment);
5953 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5954 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
5955 size, alignment, p2i(req_addr), p2i(p),
5956 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
5957 if (p != NULL) {
5958 assert(p == req_addr, "must be");
5959 small_page_write(p, size);
5960 os::Linux::release_memory_special_huge_tlbfs(p, size);
5961 }
5962 }
5963 }
5964
5965 // Case 3
5966 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
5967 test_log("size align req_addr result");
5968
5969 for (int i = 0; i < num_sizes; i++) {
5970 const size_t size = sizes[i];
5971 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5972 char* const req_addr = align_up(mapping2, alignment);
5973 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5974 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
5975 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
5976 // as the area around req_addr contains already existing mappings, the API should always
5977 // return NULL (as per contract, it cannot return another address)
5978 assert(p == NULL, "must be");
5979 }
5980 }
5981
5982 ::munmap(mapping2, mapping_size);
5983
5984 }
5985
5986 static void test_reserve_memory_special_huge_tlbfs() {
5987 if (!UseHugeTLBFS) {
5988 return;
5989 }
5990
5991 test_reserve_memory_special_huge_tlbfs_only();
5992 test_reserve_memory_special_huge_tlbfs_mixed();
5993 }
5994
5995 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
5996 if (!UseSHM) {
5997 return;
5998 }
5999
6000 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6001
6002 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6003
6004 if (addr != NULL) {
6005 assert(is_aligned(addr, alignment), "Check");
6006 assert(is_aligned(addr, os::large_page_size()), "Check");
6007
6008 small_page_write(addr, size);
6009
6010 os::Linux::release_memory_special_shm(addr, size);
6011 }
6012 }
6013
6014 static void test_reserve_memory_special_shm() {
6015 size_t lp = os::large_page_size();
6016 size_t ag = os::vm_allocation_granularity();
6017
6018 for (size_t size = ag; size < lp * 3; size += ag) {
6019 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6020 test_reserve_memory_special_shm(size, alignment);
6021 }
6022 }
6023 }
6024
6025 static void test() {
6026 test_reserve_memory_special_huge_tlbfs();
6027 test_reserve_memory_special_shm();
6028 }
6029 };
6030
6031 void TestReserveMemorySpecial_test() {
6032 TestReserveMemorySpecial::test();
6033 }
6034
6035 #endif