1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "semaphore_posix.hpp"
64 #include "services/attachListener.hpp"
65 #include "services/memTracker.hpp"
66 #include "services/runtimeService.hpp"
67 #include "utilities/decoder.hpp"
68 #include "utilities/defaultStream.hpp"
69 #include "utilities/events.hpp"
70 #include "utilities/elfFile.hpp"
71 #include "utilities/growableArray.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 #ifndef _GNU_SOURCE
110 #define _GNU_SOURCE
111 #include <sched.h>
112 #undef _GNU_SOURCE
113 #else
114 #include <sched.h>
115 #endif
116
117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
118 // getrusage() is prepared to handle the associated failure.
119 #ifndef RUSAGE_THREAD
120 #define RUSAGE_THREAD (1) /* only the calling thread */
121 #endif
122
123 #define MAX_PATH (2 * K)
124
125 #define MAX_SECS 100000000
126
127 // for timer info max values which include all bits
128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129
130 #define LARGEPAGES_BIT (1 << 6)
131 ////////////////////////////////////////////////////////////////////////////////
132 // global variables
133 julong os::Linux::_physical_memory = 0;
134
135 address os::Linux::_initial_thread_stack_bottom = NULL;
136 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137
138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
141 Mutex* os::Linux::_createThread_lock = NULL;
142 pthread_t os::Linux::_main_thread;
143 int os::Linux::_page_size = -1;
144 bool os::Linux::_supports_fast_thread_cpu_time = false;
145 uint32_t os::Linux::_os_version = 0;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148 pthread_condattr_t os::Linux::_condattr[1];
149
150 static jlong initial_time_count=0;
151
152 static int clock_tics_per_sec = 100;
153
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
157
158 // Signal number used to suspend/resume a thread
159
160 // do not use any signal number less than SIGSEGV, see 4355769
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 // Return true if user is running as root.
188
189 bool os::have_special_privileges() {
190 static bool init = false;
191 static bool privileges = false;
192 if (!init) {
193 privileges = (getuid() != geteuid()) || (getgid() != getegid());
194 init = true;
195 }
196 return privileges;
197 }
198
199
200 #ifndef SYS_gettid
201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
202 #ifdef __ia64__
203 #define SYS_gettid 1105
204 #else
205 #ifdef __i386__
206 #define SYS_gettid 224
207 #else
208 #ifdef __amd64__
209 #define SYS_gettid 186
210 #else
211 #ifdef __sparc__
212 #define SYS_gettid 143
213 #else
214 #error define gettid for the arch
215 #endif
216 #endif
217 #endif
218 #endif
219 #endif
220
221 // Cpu architecture string
222 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
223
224
225 // pid_t gettid()
226 //
227 // Returns the kernel thread id of the currently running thread. Kernel
228 // thread id is used to access /proc.
229 pid_t os::Linux::gettid() {
230 int rslt = syscall(SYS_gettid);
231 assert(rslt != -1, "must be."); // old linuxthreads implementation?
232 return (pid_t)rslt;
233 }
234
235 // Most versions of linux have a bug where the number of processors are
236 // determined by looking at the /proc file system. In a chroot environment,
237 // the system call returns 1. This causes the VM to act as if it is
238 // a single processor and elide locking (see is_MP() call).
239 static bool unsafe_chroot_detected = false;
240 static const char *unstable_chroot_error = "/proc file system not found.\n"
241 "Java may be unstable running multithreaded in a chroot "
242 "environment on Linux when /proc filesystem is not mounted.";
243
244 void os::Linux::initialize_system_info() {
245 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
246 if (processor_count() == 1) {
247 pid_t pid = os::Linux::gettid();
248 char fname[32];
249 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
250 FILE *fp = fopen(fname, "r");
251 if (fp == NULL) {
252 unsafe_chroot_detected = true;
253 } else {
254 fclose(fp);
255 }
256 }
257 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
258 assert(processor_count() > 0, "linux error");
259 }
260
261 void os::init_system_properties_values() {
262 // The next steps are taken in the product version:
263 //
264 // Obtain the JAVA_HOME value from the location of libjvm.so.
265 // This library should be located at:
266 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
267 //
268 // If "/jre/lib/" appears at the right place in the path, then we
269 // assume libjvm.so is installed in a JDK and we use this path.
270 //
271 // Otherwise exit with message: "Could not create the Java virtual machine."
272 //
273 // The following extra steps are taken in the debugging version:
274 //
275 // If "/jre/lib/" does NOT appear at the right place in the path
276 // instead of exit check for $JAVA_HOME environment variable.
277 //
278 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
279 // then we append a fake suffix "hotspot/libjvm.so" to this path so
280 // it looks like libjvm.so is installed there
281 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
282 //
283 // Otherwise exit.
284 //
285 // Important note: if the location of libjvm.so changes this
286 // code needs to be changed accordingly.
287
288 // See ld(1):
289 // The linker uses the following search paths to locate required
290 // shared libraries:
291 // 1: ...
292 // ...
293 // 7: The default directories, normally /lib and /usr/lib.
294 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
295 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
296 #else
297 #define DEFAULT_LIBPATH "/lib:/usr/lib"
298 #endif
299
300 // Base path of extensions installed on the system.
301 #define SYS_EXT_DIR "/usr/java/packages"
302 #define EXTENSIONS_DIR "/lib/ext"
303
304 // Buffer that fits several sprintfs.
305 // Note that the space for the colon and the trailing null are provided
306 // by the nulls included by the sizeof operator.
307 const size_t bufsize =
308 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
309 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
310 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
311
312 // sysclasspath, java_home, dll_dir
313 {
314 char *pslash;
315 os::jvm_path(buf, bufsize);
316
317 // Found the full path to libjvm.so.
318 // Now cut the path to <java_home>/jre if we can.
319 pslash = strrchr(buf, '/');
320 if (pslash != NULL) {
321 *pslash = '\0'; // Get rid of /libjvm.so.
322 }
323 pslash = strrchr(buf, '/');
324 if (pslash != NULL) {
325 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
326 }
327 Arguments::set_dll_dir(buf);
328
329 if (pslash != NULL) {
330 pslash = strrchr(buf, '/');
331 if (pslash != NULL) {
332 *pslash = '\0'; // Get rid of /<arch>.
333 pslash = strrchr(buf, '/');
334 if (pslash != NULL) {
335 *pslash = '\0'; // Get rid of /lib.
336 }
337 }
338 }
339 Arguments::set_java_home(buf);
340 set_boot_path('/', ':');
341 }
342
343 // Where to look for native libraries.
344 //
345 // Note: Due to a legacy implementation, most of the library path
346 // is set in the launcher. This was to accomodate linking restrictions
347 // on legacy Linux implementations (which are no longer supported).
348 // Eventually, all the library path setting will be done here.
349 //
350 // However, to prevent the proliferation of improperly built native
351 // libraries, the new path component /usr/java/packages is added here.
352 // Eventually, all the library path setting will be done here.
353 {
354 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
355 // should always exist (until the legacy problem cited above is
356 // addressed).
357 const char *v = ::getenv("LD_LIBRARY_PATH");
358 const char *v_colon = ":";
359 if (v == NULL) { v = ""; v_colon = ""; }
360 // That's +1 for the colon and +1 for the trailing '\0'.
361 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
362 strlen(v) + 1 +
363 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
364 mtInternal);
365 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
366 Arguments::set_library_path(ld_library_path);
367 FREE_C_HEAP_ARRAY(char, ld_library_path);
368 }
369
370 // Extensions directories.
371 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
372 Arguments::set_ext_dirs(buf);
373
374 FREE_C_HEAP_ARRAY(char, buf);
375
376 #undef DEFAULT_LIBPATH
377 #undef SYS_EXT_DIR
378 #undef EXTENSIONS_DIR
379 }
380
381 ////////////////////////////////////////////////////////////////////////////////
382 // breakpoint support
383
384 void os::breakpoint() {
385 BREAKPOINT;
386 }
387
388 extern "C" void breakpoint() {
389 // use debugger to set breakpoint here
390 }
391
392 ////////////////////////////////////////////////////////////////////////////////
393 // signal support
394
395 debug_only(static bool signal_sets_initialized = false);
396 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
397
398 bool os::Linux::is_sig_ignored(int sig) {
399 struct sigaction oact;
400 sigaction(sig, (struct sigaction*)NULL, &oact);
401 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
402 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
403 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
404 return true;
405 } else {
406 return false;
407 }
408 }
409
410 void os::Linux::signal_sets_init() {
411 // Should also have an assertion stating we are still single-threaded.
412 assert(!signal_sets_initialized, "Already initialized");
413 // Fill in signals that are necessarily unblocked for all threads in
414 // the VM. Currently, we unblock the following signals:
415 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
416 // by -Xrs (=ReduceSignalUsage));
417 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
418 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
419 // the dispositions or masks wrt these signals.
420 // Programs embedding the VM that want to use the above signals for their
421 // own purposes must, at this time, use the "-Xrs" option to prevent
422 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
423 // (See bug 4345157, and other related bugs).
424 // In reality, though, unblocking these signals is really a nop, since
425 // these signals are not blocked by default.
426 sigemptyset(&unblocked_sigs);
427 sigemptyset(&allowdebug_blocked_sigs);
428 sigaddset(&unblocked_sigs, SIGILL);
429 sigaddset(&unblocked_sigs, SIGSEGV);
430 sigaddset(&unblocked_sigs, SIGBUS);
431 sigaddset(&unblocked_sigs, SIGFPE);
432 #if defined(PPC64)
433 sigaddset(&unblocked_sigs, SIGTRAP);
434 #endif
435 sigaddset(&unblocked_sigs, SR_signum);
436
437 if (!ReduceSignalUsage) {
438 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
439 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
441 }
442 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
443 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
444 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
445 }
446 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
447 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
448 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
449 }
450 }
451 // Fill in signals that are blocked by all but the VM thread.
452 sigemptyset(&vm_sigs);
453 if (!ReduceSignalUsage) {
454 sigaddset(&vm_sigs, BREAK_SIGNAL);
455 }
456 debug_only(signal_sets_initialized = true);
457
458 }
459
460 // These are signals that are unblocked while a thread is running Java.
461 // (For some reason, they get blocked by default.)
462 sigset_t* os::Linux::unblocked_signals() {
463 assert(signal_sets_initialized, "Not initialized");
464 return &unblocked_sigs;
465 }
466
467 // These are the signals that are blocked while a (non-VM) thread is
468 // running Java. Only the VM thread handles these signals.
469 sigset_t* os::Linux::vm_signals() {
470 assert(signal_sets_initialized, "Not initialized");
471 return &vm_sigs;
472 }
473
474 // These are signals that are blocked during cond_wait to allow debugger in
475 sigset_t* os::Linux::allowdebug_blocked_signals() {
476 assert(signal_sets_initialized, "Not initialized");
477 return &allowdebug_blocked_sigs;
478 }
479
480 void os::Linux::hotspot_sigmask(Thread* thread) {
481
482 //Save caller's signal mask before setting VM signal mask
483 sigset_t caller_sigmask;
484 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
485
486 OSThread* osthread = thread->osthread();
487 osthread->set_caller_sigmask(caller_sigmask);
488
489 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
490
491 if (!ReduceSignalUsage) {
492 if (thread->is_VM_thread()) {
493 // Only the VM thread handles BREAK_SIGNAL ...
494 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
495 } else {
496 // ... all other threads block BREAK_SIGNAL
497 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
498 }
499 }
500 }
501
502 //////////////////////////////////////////////////////////////////////////////
503 // detecting pthread library
504
505 void os::Linux::libpthread_init() {
506 // Save glibc and pthread version strings.
507 #if !defined(_CS_GNU_LIBC_VERSION) || \
508 !defined(_CS_GNU_LIBPTHREAD_VERSION)
509 #error "glibc too old (< 2.3.2)"
510 #endif
511
512 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
513 assert(n > 0, "cannot retrieve glibc version");
514 char *str = (char *)malloc(n, mtInternal);
515 confstr(_CS_GNU_LIBC_VERSION, str, n);
516 os::Linux::set_glibc_version(str);
517
518 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
519 assert(n > 0, "cannot retrieve pthread version");
520 str = (char *)malloc(n, mtInternal);
521 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
522 os::Linux::set_libpthread_version(str);
523 }
524
525 /////////////////////////////////////////////////////////////////////////////
526 // thread stack expansion
527
528 // os::Linux::manually_expand_stack() takes care of expanding the thread
529 // stack. Note that this is normally not needed: pthread stacks allocate
530 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
531 // committed. Therefore it is not necessary to expand the stack manually.
532 //
533 // Manually expanding the stack was historically needed on LinuxThreads
534 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
535 // it is kept to deal with very rare corner cases:
536 //
537 // For one, user may run the VM on an own implementation of threads
538 // whose stacks are - like the old LinuxThreads - implemented using
539 // mmap(MAP_GROWSDOWN).
540 //
541 // Also, this coding may be needed if the VM is running on the primordial
542 // thread. Normally we avoid running on the primordial thread; however,
543 // user may still invoke the VM on the primordial thread.
544 //
545 // The following historical comment describes the details about running
546 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
547
548
549 // Force Linux kernel to expand current thread stack. If "bottom" is close
550 // to the stack guard, caller should block all signals.
551 //
552 // MAP_GROWSDOWN:
553 // A special mmap() flag that is used to implement thread stacks. It tells
554 // kernel that the memory region should extend downwards when needed. This
555 // allows early versions of LinuxThreads to only mmap the first few pages
556 // when creating a new thread. Linux kernel will automatically expand thread
557 // stack as needed (on page faults).
558 //
559 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
560 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
561 // region, it's hard to tell if the fault is due to a legitimate stack
562 // access or because of reading/writing non-exist memory (e.g. buffer
563 // overrun). As a rule, if the fault happens below current stack pointer,
564 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
565 // application (see Linux kernel fault.c).
566 //
567 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
568 // stack overflow detection.
569 //
570 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
571 // not use MAP_GROWSDOWN.
572 //
573 // To get around the problem and allow stack banging on Linux, we need to
574 // manually expand thread stack after receiving the SIGSEGV.
575 //
576 // There are two ways to expand thread stack to address "bottom", we used
577 // both of them in JVM before 1.5:
578 // 1. adjust stack pointer first so that it is below "bottom", and then
579 // touch "bottom"
580 // 2. mmap() the page in question
581 //
582 // Now alternate signal stack is gone, it's harder to use 2. For instance,
583 // if current sp is already near the lower end of page 101, and we need to
584 // call mmap() to map page 100, it is possible that part of the mmap() frame
585 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
586 // That will destroy the mmap() frame and cause VM to crash.
587 //
588 // The following code works by adjusting sp first, then accessing the "bottom"
589 // page to force a page fault. Linux kernel will then automatically expand the
590 // stack mapping.
591 //
592 // _expand_stack_to() assumes its frame size is less than page size, which
593 // should always be true if the function is not inlined.
594
595 static void NOINLINE _expand_stack_to(address bottom) {
596 address sp;
597 size_t size;
598 volatile char *p;
599
600 // Adjust bottom to point to the largest address within the same page, it
601 // gives us a one-page buffer if alloca() allocates slightly more memory.
602 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
603 bottom += os::Linux::page_size() - 1;
604
605 // sp might be slightly above current stack pointer; if that's the case, we
606 // will alloca() a little more space than necessary, which is OK. Don't use
607 // os::current_stack_pointer(), as its result can be slightly below current
608 // stack pointer, causing us to not alloca enough to reach "bottom".
609 sp = (address)&sp;
610
611 if (sp > bottom) {
612 size = sp - bottom;
613 p = (volatile char *)alloca(size);
614 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
615 p[0] = '\0';
616 }
617 }
618
619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
620 assert(t!=NULL, "just checking");
621 assert(t->osthread()->expanding_stack(), "expand should be set");
622 assert(t->stack_base() != NULL, "stack_base was not initialized");
623
624 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
625 sigset_t mask_all, old_sigset;
626 sigfillset(&mask_all);
627 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
628 _expand_stack_to(addr);
629 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
630 return true;
631 }
632 return false;
633 }
634
635 //////////////////////////////////////////////////////////////////////////////
636 // create new thread
637
638 // Thread start routine for all newly created threads
639 static void *thread_native_entry(Thread *thread) {
640 // Try to randomize the cache line index of hot stack frames.
641 // This helps when threads of the same stack traces evict each other's
642 // cache lines. The threads can be either from the same JVM instance, or
643 // from different JVM instances. The benefit is especially true for
644 // processors with hyperthreading technology.
645 static int counter = 0;
646 int pid = os::current_process_id();
647 alloca(((pid ^ counter++) & 7) * 128);
648
649 thread->initialize_thread_current();
650
651 OSThread* osthread = thread->osthread();
652 Monitor* sync = osthread->startThread_lock();
653
654 osthread->set_thread_id(os::current_thread_id());
655
656 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
657 os::current_thread_id(), (uintx) pthread_self());
658
659 if (UseNUMA) {
660 int lgrp_id = os::numa_get_group_id();
661 if (lgrp_id != -1) {
662 thread->set_lgrp_id(lgrp_id);
663 }
664 }
665 // initialize signal mask for this thread
666 os::Linux::hotspot_sigmask(thread);
667
668 // initialize floating point control register
669 os::Linux::init_thread_fpu_state();
670
671 // handshaking with parent thread
672 {
673 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
674
675 // notify parent thread
676 osthread->set_state(INITIALIZED);
677 sync->notify_all();
678
679 // wait until os::start_thread()
680 while (osthread->get_state() == INITIALIZED) {
681 sync->wait(Mutex::_no_safepoint_check_flag);
682 }
683 }
684
685 // call one more level start routine
686 thread->run();
687
688 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
689 os::current_thread_id(), (uintx) pthread_self());
690
691 // If a thread has not deleted itself ("delete this") as part of its
692 // termination sequence, we have to ensure thread-local-storage is
693 // cleared before we actually terminate. No threads should ever be
694 // deleted asynchronously with respect to their termination.
695 if (Thread::current_or_null_safe() != NULL) {
696 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
697 thread->clear_thread_current();
698 }
699
700 return 0;
701 }
702
703 bool os::create_thread(Thread* thread, ThreadType thr_type,
704 size_t req_stack_size) {
705 assert(thread->osthread() == NULL, "caller responsible");
706
707 // Allocate the OSThread object
708 OSThread* osthread = new OSThread(NULL, NULL);
709 if (osthread == NULL) {
710 return false;
711 }
712
713 // set the correct thread state
714 osthread->set_thread_type(thr_type);
715
716 // Initial state is ALLOCATED but not INITIALIZED
717 osthread->set_state(ALLOCATED);
718
719 thread->set_osthread(osthread);
720
721 // init thread attributes
722 pthread_attr_t attr;
723 pthread_attr_init(&attr);
724 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
725
726 // Calculate stack size if it's not specified by caller.
727 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
728 // In the Linux NPTL pthread implementation the guard size mechanism
729 // is not implemented properly. The posix standard requires adding
730 // the size of the guard pages to the stack size, instead Linux
731 // takes the space out of 'stacksize'. Thus we adapt the requested
732 // stack_size by the size of the guard pages to mimick proper
733 // behaviour.
734 stack_size = align_size_up(stack_size + os::Linux::default_guard_size(thr_type), vm_page_size());
735 pthread_attr_setstacksize(&attr, stack_size);
736
737 // glibc guard page
738 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
739
740 ThreadState state;
741
742 {
743 pthread_t tid;
744 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
745
746 char buf[64];
747 if (ret == 0) {
748 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
749 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
750 } else {
751 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
752 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
753 }
754
755 pthread_attr_destroy(&attr);
756
757 if (ret != 0) {
758 // Need to clean up stuff we've allocated so far
759 thread->set_osthread(NULL);
760 delete osthread;
761 return false;
762 }
763
764 // Store pthread info into the OSThread
765 osthread->set_pthread_id(tid);
766
767 // Wait until child thread is either initialized or aborted
768 {
769 Monitor* sync_with_child = osthread->startThread_lock();
770 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
771 while ((state = osthread->get_state()) == ALLOCATED) {
772 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
773 }
774 }
775 }
776
777 // Aborted due to thread limit being reached
778 if (state == ZOMBIE) {
779 thread->set_osthread(NULL);
780 delete osthread;
781 return false;
782 }
783
784 // The thread is returned suspended (in state INITIALIZED),
785 // and is started higher up in the call chain
786 assert(state == INITIALIZED, "race condition");
787 return true;
788 }
789
790 /////////////////////////////////////////////////////////////////////////////
791 // attach existing thread
792
793 // bootstrap the main thread
794 bool os::create_main_thread(JavaThread* thread) {
795 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
796 return create_attached_thread(thread);
797 }
798
799 bool os::create_attached_thread(JavaThread* thread) {
800 #ifdef ASSERT
801 thread->verify_not_published();
802 #endif
803
804 // Allocate the OSThread object
805 OSThread* osthread = new OSThread(NULL, NULL);
806
807 if (osthread == NULL) {
808 return false;
809 }
810
811 // Store pthread info into the OSThread
812 osthread->set_thread_id(os::Linux::gettid());
813 osthread->set_pthread_id(::pthread_self());
814
815 // initialize floating point control register
816 os::Linux::init_thread_fpu_state();
817
818 // Initial thread state is RUNNABLE
819 osthread->set_state(RUNNABLE);
820
821 thread->set_osthread(osthread);
822
823 if (UseNUMA) {
824 int lgrp_id = os::numa_get_group_id();
825 if (lgrp_id != -1) {
826 thread->set_lgrp_id(lgrp_id);
827 }
828 }
829
830 if (os::Linux::is_initial_thread()) {
831 // If current thread is initial thread, its stack is mapped on demand,
832 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
833 // the entire stack region to avoid SEGV in stack banging.
834 // It is also useful to get around the heap-stack-gap problem on SuSE
835 // kernel (see 4821821 for details). We first expand stack to the top
836 // of yellow zone, then enable stack yellow zone (order is significant,
837 // enabling yellow zone first will crash JVM on SuSE Linux), so there
838 // is no gap between the last two virtual memory regions.
839
840 JavaThread *jt = (JavaThread *)thread;
841 address addr = jt->stack_reserved_zone_base();
842 assert(addr != NULL, "initialization problem?");
843 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
844
845 osthread->set_expanding_stack();
846 os::Linux::manually_expand_stack(jt, addr);
847 osthread->clear_expanding_stack();
848 }
849
850 // initialize signal mask for this thread
851 // and save the caller's signal mask
852 os::Linux::hotspot_sigmask(thread);
853
854 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
855 os::current_thread_id(), (uintx) pthread_self());
856
857 return true;
858 }
859
860 void os::pd_start_thread(Thread* thread) {
861 OSThread * osthread = thread->osthread();
862 assert(osthread->get_state() != INITIALIZED, "just checking");
863 Monitor* sync_with_child = osthread->startThread_lock();
864 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
865 sync_with_child->notify();
866 }
867
868 // Free Linux resources related to the OSThread
869 void os::free_thread(OSThread* osthread) {
870 assert(osthread != NULL, "osthread not set");
871
872 // We are told to free resources of the argument thread,
873 // but we can only really operate on the current thread.
874 assert(Thread::current()->osthread() == osthread,
875 "os::free_thread but not current thread");
876
877 #ifdef ASSERT
878 sigset_t current;
879 sigemptyset(¤t);
880 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
881 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
882 #endif
883
884 // Restore caller's signal mask
885 sigset_t sigmask = osthread->caller_sigmask();
886 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
887
888 delete osthread;
889 }
890
891 //////////////////////////////////////////////////////////////////////////////
892 // initial thread
893
894 // Check if current thread is the initial thread, similar to Solaris thr_main.
895 bool os::Linux::is_initial_thread(void) {
896 char dummy;
897 // If called before init complete, thread stack bottom will be null.
898 // Can be called if fatal error occurs before initialization.
899 if (initial_thread_stack_bottom() == NULL) return false;
900 assert(initial_thread_stack_bottom() != NULL &&
901 initial_thread_stack_size() != 0,
902 "os::init did not locate initial thread's stack region");
903 if ((address)&dummy >= initial_thread_stack_bottom() &&
904 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
905 return true;
906 } else {
907 return false;
908 }
909 }
910
911 // Find the virtual memory area that contains addr
912 static bool find_vma(address addr, address* vma_low, address* vma_high) {
913 FILE *fp = fopen("/proc/self/maps", "r");
914 if (fp) {
915 address low, high;
916 while (!feof(fp)) {
917 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
918 if (low <= addr && addr < high) {
919 if (vma_low) *vma_low = low;
920 if (vma_high) *vma_high = high;
921 fclose(fp);
922 return true;
923 }
924 }
925 for (;;) {
926 int ch = fgetc(fp);
927 if (ch == EOF || ch == (int)'\n') break;
928 }
929 }
930 fclose(fp);
931 }
932 return false;
933 }
934
935 // Locate initial thread stack. This special handling of initial thread stack
936 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
937 // bogus value for initial thread.
938 void os::Linux::capture_initial_stack(size_t max_size) {
939 // stack size is the easy part, get it from RLIMIT_STACK
940 struct rlimit rlim;
941 getrlimit(RLIMIT_STACK, &rlim);
942 size_t stack_size = rlim.rlim_cur;
943
944 // 6308388: a bug in ld.so will relocate its own .data section to the
945 // lower end of primordial stack; reduce ulimit -s value a little bit
946 // so we won't install guard page on ld.so's data section.
947 stack_size -= 2 * page_size();
948
949 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
950 // 7.1, in both cases we will get 2G in return value.
951 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
952 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
953 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
954 // in case other parts in glibc still assumes 2M max stack size.
955 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
956 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
957 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
958 stack_size = 2 * K * K IA64_ONLY(*2);
959 }
960 // Try to figure out where the stack base (top) is. This is harder.
961 //
962 // When an application is started, glibc saves the initial stack pointer in
963 // a global variable "__libc_stack_end", which is then used by system
964 // libraries. __libc_stack_end should be pretty close to stack top. The
965 // variable is available since the very early days. However, because it is
966 // a private interface, it could disappear in the future.
967 //
968 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
969 // to __libc_stack_end, it is very close to stack top, but isn't the real
970 // stack top. Note that /proc may not exist if VM is running as a chroot
971 // program, so reading /proc/<pid>/stat could fail. Also the contents of
972 // /proc/<pid>/stat could change in the future (though unlikely).
973 //
974 // We try __libc_stack_end first. If that doesn't work, look for
975 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
976 // as a hint, which should work well in most cases.
977
978 uintptr_t stack_start;
979
980 // try __libc_stack_end first
981 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
982 if (p && *p) {
983 stack_start = *p;
984 } else {
985 // see if we can get the start_stack field from /proc/self/stat
986 FILE *fp;
987 int pid;
988 char state;
989 int ppid;
990 int pgrp;
991 int session;
992 int nr;
993 int tpgrp;
994 unsigned long flags;
995 unsigned long minflt;
996 unsigned long cminflt;
997 unsigned long majflt;
998 unsigned long cmajflt;
999 unsigned long utime;
1000 unsigned long stime;
1001 long cutime;
1002 long cstime;
1003 long prio;
1004 long nice;
1005 long junk;
1006 long it_real;
1007 uintptr_t start;
1008 uintptr_t vsize;
1009 intptr_t rss;
1010 uintptr_t rsslim;
1011 uintptr_t scodes;
1012 uintptr_t ecode;
1013 int i;
1014
1015 // Figure what the primordial thread stack base is. Code is inspired
1016 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1017 // followed by command name surrounded by parentheses, state, etc.
1018 char stat[2048];
1019 int statlen;
1020
1021 fp = fopen("/proc/self/stat", "r");
1022 if (fp) {
1023 statlen = fread(stat, 1, 2047, fp);
1024 stat[statlen] = '\0';
1025 fclose(fp);
1026
1027 // Skip pid and the command string. Note that we could be dealing with
1028 // weird command names, e.g. user could decide to rename java launcher
1029 // to "java 1.4.2 :)", then the stat file would look like
1030 // 1234 (java 1.4.2 :)) R ... ...
1031 // We don't really need to know the command string, just find the last
1032 // occurrence of ")" and then start parsing from there. See bug 4726580.
1033 char * s = strrchr(stat, ')');
1034
1035 i = 0;
1036 if (s) {
1037 // Skip blank chars
1038 do { s++; } while (s && isspace(*s));
1039
1040 #define _UFM UINTX_FORMAT
1041 #define _DFM INTX_FORMAT
1042
1043 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1044 // 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
1045 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,
1046 &state, // 3 %c
1047 &ppid, // 4 %d
1048 &pgrp, // 5 %d
1049 &session, // 6 %d
1050 &nr, // 7 %d
1051 &tpgrp, // 8 %d
1052 &flags, // 9 %lu
1053 &minflt, // 10 %lu
1054 &cminflt, // 11 %lu
1055 &majflt, // 12 %lu
1056 &cmajflt, // 13 %lu
1057 &utime, // 14 %lu
1058 &stime, // 15 %lu
1059 &cutime, // 16 %ld
1060 &cstime, // 17 %ld
1061 &prio, // 18 %ld
1062 &nice, // 19 %ld
1063 &junk, // 20 %ld
1064 &it_real, // 21 %ld
1065 &start, // 22 UINTX_FORMAT
1066 &vsize, // 23 UINTX_FORMAT
1067 &rss, // 24 INTX_FORMAT
1068 &rsslim, // 25 UINTX_FORMAT
1069 &scodes, // 26 UINTX_FORMAT
1070 &ecode, // 27 UINTX_FORMAT
1071 &stack_start); // 28 UINTX_FORMAT
1072 }
1073
1074 #undef _UFM
1075 #undef _DFM
1076
1077 if (i != 28 - 2) {
1078 assert(false, "Bad conversion from /proc/self/stat");
1079 // product mode - assume we are the initial thread, good luck in the
1080 // embedded case.
1081 warning("Can't detect initial thread stack location - bad conversion");
1082 stack_start = (uintptr_t) &rlim;
1083 }
1084 } else {
1085 // For some reason we can't open /proc/self/stat (for example, running on
1086 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1087 // most cases, so don't abort:
1088 warning("Can't detect initial thread stack location - no /proc/self/stat");
1089 stack_start = (uintptr_t) &rlim;
1090 }
1091 }
1092
1093 // Now we have a pointer (stack_start) very close to the stack top, the
1094 // next thing to do is to figure out the exact location of stack top. We
1095 // can find out the virtual memory area that contains stack_start by
1096 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1097 // and its upper limit is the real stack top. (again, this would fail if
1098 // running inside chroot, because /proc may not exist.)
1099
1100 uintptr_t stack_top;
1101 address low, high;
1102 if (find_vma((address)stack_start, &low, &high)) {
1103 // success, "high" is the true stack top. (ignore "low", because initial
1104 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1105 stack_top = (uintptr_t)high;
1106 } else {
1107 // failed, likely because /proc/self/maps does not exist
1108 warning("Can't detect initial thread stack location - find_vma failed");
1109 // best effort: stack_start is normally within a few pages below the real
1110 // stack top, use it as stack top, and reduce stack size so we won't put
1111 // guard page outside stack.
1112 stack_top = stack_start;
1113 stack_size -= 16 * page_size();
1114 }
1115
1116 // stack_top could be partially down the page so align it
1117 stack_top = align_size_up(stack_top, page_size());
1118
1119 if (max_size && stack_size > max_size) {
1120 _initial_thread_stack_size = max_size;
1121 } else {
1122 _initial_thread_stack_size = stack_size;
1123 }
1124
1125 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1126 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1127 }
1128
1129 ////////////////////////////////////////////////////////////////////////////////
1130 // time support
1131
1132 // Time since start-up in seconds to a fine granularity.
1133 // Used by VMSelfDestructTimer and the MemProfiler.
1134 double os::elapsedTime() {
1135
1136 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1137 }
1138
1139 jlong os::elapsed_counter() {
1140 return javaTimeNanos() - initial_time_count;
1141 }
1142
1143 jlong os::elapsed_frequency() {
1144 return NANOSECS_PER_SEC; // nanosecond resolution
1145 }
1146
1147 bool os::supports_vtime() { return true; }
1148 bool os::enable_vtime() { return false; }
1149 bool os::vtime_enabled() { return false; }
1150
1151 double os::elapsedVTime() {
1152 struct rusage usage;
1153 int retval = getrusage(RUSAGE_THREAD, &usage);
1154 if (retval == 0) {
1155 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);
1156 } else {
1157 // better than nothing, but not much
1158 return elapsedTime();
1159 }
1160 }
1161
1162 jlong os::javaTimeMillis() {
1163 timeval time;
1164 int status = gettimeofday(&time, NULL);
1165 assert(status != -1, "linux error");
1166 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1167 }
1168
1169 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1170 timeval time;
1171 int status = gettimeofday(&time, NULL);
1172 assert(status != -1, "linux error");
1173 seconds = jlong(time.tv_sec);
1174 nanos = jlong(time.tv_usec) * 1000;
1175 }
1176
1177
1178 #ifndef CLOCK_MONOTONIC
1179 #define CLOCK_MONOTONIC (1)
1180 #endif
1181
1182 void os::Linux::clock_init() {
1183 // we do dlopen's in this particular order due to bug in linux
1184 // dynamical loader (see 6348968) leading to crash on exit
1185 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1186 if (handle == NULL) {
1187 handle = dlopen("librt.so", RTLD_LAZY);
1188 }
1189
1190 if (handle) {
1191 int (*clock_getres_func)(clockid_t, struct timespec*) =
1192 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1193 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1194 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1195 if (clock_getres_func && clock_gettime_func) {
1196 // See if monotonic clock is supported by the kernel. Note that some
1197 // early implementations simply return kernel jiffies (updated every
1198 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1199 // for nano time (though the monotonic property is still nice to have).
1200 // It's fixed in newer kernels, however clock_getres() still returns
1201 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1202 // resolution for now. Hopefully as people move to new kernels, this
1203 // won't be a problem.
1204 struct timespec res;
1205 struct timespec tp;
1206 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1207 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1208 // yes, monotonic clock is supported
1209 _clock_gettime = clock_gettime_func;
1210 return;
1211 } else {
1212 // close librt if there is no monotonic clock
1213 dlclose(handle);
1214 }
1215 }
1216 }
1217 warning("No monotonic clock was available - timed services may " \
1218 "be adversely affected if the time-of-day clock changes");
1219 }
1220
1221 #ifndef SYS_clock_getres
1222 #if defined(X86) || defined(PPC64) || defined(S390)
1223 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1224 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1225 #else
1226 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1227 #define sys_clock_getres(x,y) -1
1228 #endif
1229 #else
1230 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1231 #endif
1232
1233 void os::Linux::fast_thread_clock_init() {
1234 if (!UseLinuxPosixThreadCPUClocks) {
1235 return;
1236 }
1237 clockid_t clockid;
1238 struct timespec tp;
1239 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1240 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1241
1242 // Switch to using fast clocks for thread cpu time if
1243 // the sys_clock_getres() returns 0 error code.
1244 // Note, that some kernels may support the current thread
1245 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1246 // returned by the pthread_getcpuclockid().
1247 // If the fast Posix clocks are supported then the sys_clock_getres()
1248 // must return at least tp.tv_sec == 0 which means a resolution
1249 // better than 1 sec. This is extra check for reliability.
1250
1251 if (pthread_getcpuclockid_func &&
1252 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1253 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1254 _supports_fast_thread_cpu_time = true;
1255 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1256 }
1257 }
1258
1259 jlong os::javaTimeNanos() {
1260 if (os::supports_monotonic_clock()) {
1261 struct timespec tp;
1262 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1263 assert(status == 0, "gettime error");
1264 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1265 return result;
1266 } else {
1267 timeval time;
1268 int status = gettimeofday(&time, NULL);
1269 assert(status != -1, "linux error");
1270 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1271 return 1000 * usecs;
1272 }
1273 }
1274
1275 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1276 if (os::supports_monotonic_clock()) {
1277 info_ptr->max_value = ALL_64_BITS;
1278
1279 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1280 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1281 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1282 } else {
1283 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1284 info_ptr->max_value = ALL_64_BITS;
1285
1286 // gettimeofday is a real time clock so it skips
1287 info_ptr->may_skip_backward = true;
1288 info_ptr->may_skip_forward = true;
1289 }
1290
1291 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1292 }
1293
1294 // Return the real, user, and system times in seconds from an
1295 // arbitrary fixed point in the past.
1296 bool os::getTimesSecs(double* process_real_time,
1297 double* process_user_time,
1298 double* process_system_time) {
1299 struct tms ticks;
1300 clock_t real_ticks = times(&ticks);
1301
1302 if (real_ticks == (clock_t) (-1)) {
1303 return false;
1304 } else {
1305 double ticks_per_second = (double) clock_tics_per_sec;
1306 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1307 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1308 *process_real_time = ((double) real_ticks) / ticks_per_second;
1309
1310 return true;
1311 }
1312 }
1313
1314
1315 char * os::local_time_string(char *buf, size_t buflen) {
1316 struct tm t;
1317 time_t long_time;
1318 time(&long_time);
1319 localtime_r(&long_time, &t);
1320 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1321 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1322 t.tm_hour, t.tm_min, t.tm_sec);
1323 return buf;
1324 }
1325
1326 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1327 return localtime_r(clock, res);
1328 }
1329
1330 ////////////////////////////////////////////////////////////////////////////////
1331 // runtime exit support
1332
1333 // Note: os::shutdown() might be called very early during initialization, or
1334 // called from signal handler. Before adding something to os::shutdown(), make
1335 // sure it is async-safe and can handle partially initialized VM.
1336 void os::shutdown() {
1337
1338 // allow PerfMemory to attempt cleanup of any persistent resources
1339 perfMemory_exit();
1340
1341 // needs to remove object in file system
1342 AttachListener::abort();
1343
1344 // flush buffered output, finish log files
1345 ostream_abort();
1346
1347 // Check for abort hook
1348 abort_hook_t abort_hook = Arguments::abort_hook();
1349 if (abort_hook != NULL) {
1350 abort_hook();
1351 }
1352
1353 }
1354
1355 // Note: os::abort() might be called very early during initialization, or
1356 // called from signal handler. Before adding something to os::abort(), make
1357 // sure it is async-safe and can handle partially initialized VM.
1358 void os::abort(bool dump_core, void* siginfo, const void* context) {
1359 os::shutdown();
1360 if (dump_core) {
1361 #ifndef PRODUCT
1362 fdStream out(defaultStream::output_fd());
1363 out.print_raw("Current thread is ");
1364 char buf[16];
1365 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1366 out.print_raw_cr(buf);
1367 out.print_raw_cr("Dumping core ...");
1368 #endif
1369 ::abort(); // dump core
1370 }
1371
1372 ::exit(1);
1373 }
1374
1375 // Die immediately, no exit hook, no abort hook, no cleanup.
1376 void os::die() {
1377 ::abort();
1378 }
1379
1380
1381 // This method is a copy of JDK's sysGetLastErrorString
1382 // from src/solaris/hpi/src/system_md.c
1383
1384 size_t os::lasterror(char *buf, size_t len) {
1385 if (errno == 0) return 0;
1386
1387 const char *s = os::strerror(errno);
1388 size_t n = ::strlen(s);
1389 if (n >= len) {
1390 n = len - 1;
1391 }
1392 ::strncpy(buf, s, n);
1393 buf[n] = '\0';
1394 return n;
1395 }
1396
1397 // thread_id is kernel thread id (similar to Solaris LWP id)
1398 intx os::current_thread_id() { return os::Linux::gettid(); }
1399 int os::current_process_id() {
1400 return ::getpid();
1401 }
1402
1403 // DLL functions
1404
1405 const char* os::dll_file_extension() { return ".so"; }
1406
1407 // This must be hard coded because it's the system's temporary
1408 // directory not the java application's temp directory, ala java.io.tmpdir.
1409 const char* os::get_temp_directory() { return "/tmp"; }
1410
1411 static bool file_exists(const char* filename) {
1412 struct stat statbuf;
1413 if (filename == NULL || strlen(filename) == 0) {
1414 return false;
1415 }
1416 return os::stat(filename, &statbuf) == 0;
1417 }
1418
1419 bool os::dll_build_name(char* buffer, size_t buflen,
1420 const char* pname, const char* fname) {
1421 bool retval = false;
1422 // Copied from libhpi
1423 const size_t pnamelen = pname ? strlen(pname) : 0;
1424
1425 // Return error on buffer overflow.
1426 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1427 return retval;
1428 }
1429
1430 if (pnamelen == 0) {
1431 snprintf(buffer, buflen, "lib%s.so", fname);
1432 retval = true;
1433 } else if (strchr(pname, *os::path_separator()) != NULL) {
1434 int n;
1435 char** pelements = split_path(pname, &n);
1436 if (pelements == NULL) {
1437 return false;
1438 }
1439 for (int i = 0; i < n; i++) {
1440 // Really shouldn't be NULL, but check can't hurt
1441 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1442 continue; // skip the empty path values
1443 }
1444 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1445 if (file_exists(buffer)) {
1446 retval = true;
1447 break;
1448 }
1449 }
1450 // release the storage
1451 for (int i = 0; i < n; i++) {
1452 if (pelements[i] != NULL) {
1453 FREE_C_HEAP_ARRAY(char, pelements[i]);
1454 }
1455 }
1456 if (pelements != NULL) {
1457 FREE_C_HEAP_ARRAY(char*, pelements);
1458 }
1459 } else {
1460 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1461 retval = true;
1462 }
1463 return retval;
1464 }
1465
1466 // check if addr is inside libjvm.so
1467 bool os::address_is_in_vm(address addr) {
1468 static address libjvm_base_addr;
1469 Dl_info dlinfo;
1470
1471 if (libjvm_base_addr == NULL) {
1472 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1473 libjvm_base_addr = (address)dlinfo.dli_fbase;
1474 }
1475 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1476 }
1477
1478 if (dladdr((void *)addr, &dlinfo) != 0) {
1479 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1480 }
1481
1482 return false;
1483 }
1484
1485 bool os::dll_address_to_function_name(address addr, char *buf,
1486 int buflen, int *offset,
1487 bool demangle) {
1488 // buf is not optional, but offset is optional
1489 assert(buf != NULL, "sanity check");
1490
1491 Dl_info dlinfo;
1492
1493 if (dladdr((void*)addr, &dlinfo) != 0) {
1494 // see if we have a matching symbol
1495 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1496 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1497 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1498 }
1499 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1500 return true;
1501 }
1502 // no matching symbol so try for just file info
1503 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1504 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1505 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1506 return true;
1507 }
1508 }
1509 }
1510
1511 buf[0] = '\0';
1512 if (offset != NULL) *offset = -1;
1513 return false;
1514 }
1515
1516 struct _address_to_library_name {
1517 address addr; // input : memory address
1518 size_t buflen; // size of fname
1519 char* fname; // output: library name
1520 address base; // library base addr
1521 };
1522
1523 static int address_to_library_name_callback(struct dl_phdr_info *info,
1524 size_t size, void *data) {
1525 int i;
1526 bool found = false;
1527 address libbase = NULL;
1528 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1529
1530 // iterate through all loadable segments
1531 for (i = 0; i < info->dlpi_phnum; i++) {
1532 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1533 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1534 // base address of a library is the lowest address of its loaded
1535 // segments.
1536 if (libbase == NULL || libbase > segbase) {
1537 libbase = segbase;
1538 }
1539 // see if 'addr' is within current segment
1540 if (segbase <= d->addr &&
1541 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1542 found = true;
1543 }
1544 }
1545 }
1546
1547 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1548 // so dll_address_to_library_name() can fall through to use dladdr() which
1549 // can figure out executable name from argv[0].
1550 if (found && info->dlpi_name && info->dlpi_name[0]) {
1551 d->base = libbase;
1552 if (d->fname) {
1553 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1554 }
1555 return 1;
1556 }
1557 return 0;
1558 }
1559
1560 bool os::dll_address_to_library_name(address addr, char* buf,
1561 int buflen, int* offset) {
1562 // buf is not optional, but offset is optional
1563 assert(buf != NULL, "sanity check");
1564
1565 Dl_info dlinfo;
1566 struct _address_to_library_name data;
1567
1568 // There is a bug in old glibc dladdr() implementation that it could resolve
1569 // to wrong library name if the .so file has a base address != NULL. Here
1570 // we iterate through the program headers of all loaded libraries to find
1571 // out which library 'addr' really belongs to. This workaround can be
1572 // removed once the minimum requirement for glibc is moved to 2.3.x.
1573 data.addr = addr;
1574 data.fname = buf;
1575 data.buflen = buflen;
1576 data.base = NULL;
1577 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1578
1579 if (rslt) {
1580 // buf already contains library name
1581 if (offset) *offset = addr - data.base;
1582 return true;
1583 }
1584 if (dladdr((void*)addr, &dlinfo) != 0) {
1585 if (dlinfo.dli_fname != NULL) {
1586 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1587 }
1588 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1589 *offset = addr - (address)dlinfo.dli_fbase;
1590 }
1591 return true;
1592 }
1593
1594 buf[0] = '\0';
1595 if (offset) *offset = -1;
1596 return false;
1597 }
1598
1599 // Loads .dll/.so and
1600 // in case of error it checks if .dll/.so was built for the
1601 // same architecture as Hotspot is running on
1602
1603
1604 // Remember the stack's state. The Linux dynamic linker will change
1605 // the stack to 'executable' at most once, so we must safepoint only once.
1606 bool os::Linux::_stack_is_executable = false;
1607
1608 // VM operation that loads a library. This is necessary if stack protection
1609 // of the Java stacks can be lost during loading the library. If we
1610 // do not stop the Java threads, they can stack overflow before the stacks
1611 // are protected again.
1612 class VM_LinuxDllLoad: public VM_Operation {
1613 private:
1614 const char *_filename;
1615 char *_ebuf;
1616 int _ebuflen;
1617 void *_lib;
1618 public:
1619 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1620 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1621 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1622 void doit() {
1623 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1624 os::Linux::_stack_is_executable = true;
1625 }
1626 void* loaded_library() { return _lib; }
1627 };
1628
1629 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1630 void * result = NULL;
1631 bool load_attempted = false;
1632
1633 // Check whether the library to load might change execution rights
1634 // of the stack. If they are changed, the protection of the stack
1635 // guard pages will be lost. We need a safepoint to fix this.
1636 //
1637 // See Linux man page execstack(8) for more info.
1638 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1639 ElfFile ef(filename);
1640 if (!ef.specifies_noexecstack()) {
1641 if (!is_init_completed()) {
1642 os::Linux::_stack_is_executable = true;
1643 // This is OK - No Java threads have been created yet, and hence no
1644 // stack guard pages to fix.
1645 //
1646 // This should happen only when you are building JDK7 using a very
1647 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1648 //
1649 // Dynamic loader will make all stacks executable after
1650 // this function returns, and will not do that again.
1651 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1652 } else {
1653 warning("You have loaded library %s which might have disabled stack guard. "
1654 "The VM will try to fix the stack guard now.\n"
1655 "It's highly recommended that you fix the library with "
1656 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1657 filename);
1658
1659 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1660 JavaThread *jt = JavaThread::current();
1661 if (jt->thread_state() != _thread_in_native) {
1662 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1663 // that requires ExecStack. Cannot enter safe point. Let's give up.
1664 warning("Unable to fix stack guard. Giving up.");
1665 } else {
1666 if (!LoadExecStackDllInVMThread) {
1667 // This is for the case where the DLL has an static
1668 // constructor function that executes JNI code. We cannot
1669 // load such DLLs in the VMThread.
1670 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1671 }
1672
1673 ThreadInVMfromNative tiv(jt);
1674 debug_only(VMNativeEntryWrapper vew;)
1675
1676 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1677 VMThread::execute(&op);
1678 if (LoadExecStackDllInVMThread) {
1679 result = op.loaded_library();
1680 }
1681 load_attempted = true;
1682 }
1683 }
1684 }
1685 }
1686
1687 if (!load_attempted) {
1688 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1689 }
1690
1691 if (result != NULL) {
1692 // Successful loading
1693 return result;
1694 }
1695
1696 Elf32_Ehdr elf_head;
1697 int diag_msg_max_length=ebuflen-strlen(ebuf);
1698 char* diag_msg_buf=ebuf+strlen(ebuf);
1699
1700 if (diag_msg_max_length==0) {
1701 // No more space in ebuf for additional diagnostics message
1702 return NULL;
1703 }
1704
1705
1706 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1707
1708 if (file_descriptor < 0) {
1709 // Can't open library, report dlerror() message
1710 return NULL;
1711 }
1712
1713 bool failed_to_read_elf_head=
1714 (sizeof(elf_head)!=
1715 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1716
1717 ::close(file_descriptor);
1718 if (failed_to_read_elf_head) {
1719 // file i/o error - report dlerror() msg
1720 return NULL;
1721 }
1722
1723 typedef struct {
1724 Elf32_Half code; // Actual value as defined in elf.h
1725 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1726 unsigned char elf_class; // 32 or 64 bit
1727 unsigned char endianess; // MSB or LSB
1728 char* name; // String representation
1729 } arch_t;
1730
1731 #ifndef EM_486
1732 #define EM_486 6 /* Intel 80486 */
1733 #endif
1734 #ifndef EM_AARCH64
1735 #define EM_AARCH64 183 /* ARM AARCH64 */
1736 #endif
1737
1738 static const arch_t arch_array[]={
1739 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1740 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1741 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1742 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1743 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1744 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1745 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1746 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1747 #if defined(VM_LITTLE_ENDIAN)
1748 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1749 #else
1750 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1751 #endif
1752 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1753 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1754 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1755 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1756 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1757 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1758 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1759 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1760 };
1761
1762 #if (defined IA32)
1763 static Elf32_Half running_arch_code=EM_386;
1764 #elif (defined AMD64)
1765 static Elf32_Half running_arch_code=EM_X86_64;
1766 #elif (defined IA64)
1767 static Elf32_Half running_arch_code=EM_IA_64;
1768 #elif (defined __sparc) && (defined _LP64)
1769 static Elf32_Half running_arch_code=EM_SPARCV9;
1770 #elif (defined __sparc) && (!defined _LP64)
1771 static Elf32_Half running_arch_code=EM_SPARC;
1772 #elif (defined __powerpc64__)
1773 static Elf32_Half running_arch_code=EM_PPC64;
1774 #elif (defined __powerpc__)
1775 static Elf32_Half running_arch_code=EM_PPC;
1776 #elif (defined AARCH64)
1777 static Elf32_Half running_arch_code=EM_AARCH64;
1778 #elif (defined ARM)
1779 static Elf32_Half running_arch_code=EM_ARM;
1780 #elif (defined S390)
1781 static Elf32_Half running_arch_code=EM_S390;
1782 #elif (defined ALPHA)
1783 static Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif (defined MIPSEL)
1785 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif (defined PARISC)
1787 static Elf32_Half running_arch_code=EM_PARISC;
1788 #elif (defined MIPS)
1789 static Elf32_Half running_arch_code=EM_MIPS;
1790 #elif (defined M68K)
1791 static Elf32_Half running_arch_code=EM_68K;
1792 #else
1793 #error Method os::dll_load requires that one of following is defined:\
1794 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1795 #endif
1796
1797 // Identify compatability class for VM's architecture and library's architecture
1798 // Obtain string descriptions for architectures
1799
1800 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1801 int running_arch_index=-1;
1802
1803 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1804 if (running_arch_code == arch_array[i].code) {
1805 running_arch_index = i;
1806 }
1807 if (lib_arch.code == arch_array[i].code) {
1808 lib_arch.compat_class = arch_array[i].compat_class;
1809 lib_arch.name = arch_array[i].name;
1810 }
1811 }
1812
1813 assert(running_arch_index != -1,
1814 "Didn't find running architecture code (running_arch_code) in arch_array");
1815 if (running_arch_index == -1) {
1816 // Even though running architecture detection failed
1817 // we may still continue with reporting dlerror() message
1818 return NULL;
1819 }
1820
1821 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1822 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1823 return NULL;
1824 }
1825
1826 #ifndef S390
1827 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1828 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1829 return NULL;
1830 }
1831 #endif // !S390
1832
1833 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1834 if (lib_arch.name!=NULL) {
1835 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1836 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1837 lib_arch.name, arch_array[running_arch_index].name);
1838 } else {
1839 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1840 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1841 lib_arch.code,
1842 arch_array[running_arch_index].name);
1843 }
1844 }
1845
1846 return NULL;
1847 }
1848
1849 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1850 int ebuflen) {
1851 void * result = ::dlopen(filename, RTLD_LAZY);
1852 if (result == NULL) {
1853 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1854 ebuf[ebuflen-1] = '\0';
1855 }
1856 return result;
1857 }
1858
1859 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1860 int ebuflen) {
1861 void * result = NULL;
1862 if (LoadExecStackDllInVMThread) {
1863 result = dlopen_helper(filename, ebuf, ebuflen);
1864 }
1865
1866 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1867 // library that requires an executable stack, or which does not have this
1868 // stack attribute set, dlopen changes the stack attribute to executable. The
1869 // read protection of the guard pages gets lost.
1870 //
1871 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1872 // may have been queued at the same time.
1873
1874 if (!_stack_is_executable) {
1875 JavaThread *jt = Threads::first();
1876
1877 while (jt) {
1878 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1879 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1880 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1881 warning("Attempt to reguard stack yellow zone failed.");
1882 }
1883 }
1884 jt = jt->next();
1885 }
1886 }
1887
1888 return result;
1889 }
1890
1891 void* os::dll_lookup(void* handle, const char* name) {
1892 void* res = dlsym(handle, name);
1893 return res;
1894 }
1895
1896 void* os::get_default_process_handle() {
1897 return (void*)::dlopen(NULL, RTLD_LAZY);
1898 }
1899
1900 static bool _print_ascii_file(const char* filename, outputStream* st) {
1901 int fd = ::open(filename, O_RDONLY);
1902 if (fd == -1) {
1903 return false;
1904 }
1905
1906 char buf[33];
1907 int bytes;
1908 buf[32] = '\0';
1909 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1910 st->print_raw(buf, bytes);
1911 }
1912
1913 ::close(fd);
1914
1915 return true;
1916 }
1917
1918 void os::print_dll_info(outputStream *st) {
1919 st->print_cr("Dynamic libraries:");
1920
1921 char fname[32];
1922 pid_t pid = os::Linux::gettid();
1923
1924 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1925
1926 if (!_print_ascii_file(fname, st)) {
1927 st->print("Can not get library information for pid = %d\n", pid);
1928 }
1929 }
1930
1931 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1932 FILE *procmapsFile = NULL;
1933
1934 // Open the procfs maps file for the current process
1935 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1936 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1937 char line[PATH_MAX + 100];
1938
1939 // Read line by line from 'file'
1940 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1941 u8 base, top, offset, inode;
1942 char permissions[5];
1943 char device[6];
1944 char name[PATH_MAX + 1];
1945
1946 // Parse fields from line
1947 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1948 &base, &top, permissions, &offset, device, &inode, name);
1949
1950 // Filter by device id '00:00' so that we only get file system mapped files.
1951 if (strcmp(device, "00:00") != 0) {
1952
1953 // Call callback with the fields of interest
1954 if(callback(name, (address)base, (address)top, param)) {
1955 // Oops abort, callback aborted
1956 fclose(procmapsFile);
1957 return 1;
1958 }
1959 }
1960 }
1961 fclose(procmapsFile);
1962 }
1963 return 0;
1964 }
1965
1966 void os::print_os_info_brief(outputStream* st) {
1967 os::Linux::print_distro_info(st);
1968
1969 os::Posix::print_uname_info(st);
1970
1971 os::Linux::print_libversion_info(st);
1972
1973 }
1974
1975 void os::print_os_info(outputStream* st) {
1976 st->print("OS:");
1977
1978 os::Linux::print_distro_info(st);
1979
1980 os::Posix::print_uname_info(st);
1981
1982 // Print warning if unsafe chroot environment detected
1983 if (unsafe_chroot_detected) {
1984 st->print("WARNING!! ");
1985 st->print_cr("%s", unstable_chroot_error);
1986 }
1987
1988 os::Linux::print_libversion_info(st);
1989
1990 os::Posix::print_rlimit_info(st);
1991
1992 os::Posix::print_load_average(st);
1993
1994 os::Linux::print_full_memory_info(st);
1995 }
1996
1997 // Try to identify popular distros.
1998 // Most Linux distributions have a /etc/XXX-release file, which contains
1999 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2000 // file that also contains the OS version string. Some have more than one
2001 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2002 // /etc/redhat-release.), so the order is important.
2003 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2004 // their own specific XXX-release file as well as a redhat-release file.
2005 // Because of this the XXX-release file needs to be searched for before the
2006 // redhat-release file.
2007 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2008 // search for redhat-release / SuSE-release needs to be before lsb-release.
2009 // Since the lsb-release file is the new standard it needs to be searched
2010 // before the older style release files.
2011 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2012 // next to last resort. The os-release file is a new standard that contains
2013 // distribution information and the system-release file seems to be an old
2014 // standard that has been replaced by the lsb-release and os-release files.
2015 // Searching for the debian_version file is the last resort. It contains
2016 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2017 // "Debian " is printed before the contents of the debian_version file.
2018
2019 const char* distro_files[] = {
2020 "/etc/oracle-release",
2021 "/etc/mandriva-release",
2022 "/etc/mandrake-release",
2023 "/etc/sun-release",
2024 "/etc/redhat-release",
2025 "/etc/SuSE-release",
2026 "/etc/lsb-release",
2027 "/etc/turbolinux-release",
2028 "/etc/gentoo-release",
2029 "/etc/ltib-release",
2030 "/etc/angstrom-version",
2031 "/etc/system-release",
2032 "/etc/os-release",
2033 NULL };
2034
2035 void os::Linux::print_distro_info(outputStream* st) {
2036 for (int i = 0;; i++) {
2037 const char* file = distro_files[i];
2038 if (file == NULL) {
2039 break; // done
2040 }
2041 // If file prints, we found it.
2042 if (_print_ascii_file(file, st)) {
2043 return;
2044 }
2045 }
2046
2047 if (file_exists("/etc/debian_version")) {
2048 st->print("Debian ");
2049 _print_ascii_file("/etc/debian_version", st);
2050 } else {
2051 st->print("Linux");
2052 }
2053 st->cr();
2054 }
2055
2056 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2057 char buf[256];
2058 while (fgets(buf, sizeof(buf), fp)) {
2059 // Edit out extra stuff in expected format
2060 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2061 char* ptr = strstr(buf, "\""); // the name is in quotes
2062 if (ptr != NULL) {
2063 ptr++; // go beyond first quote
2064 char* nl = strchr(ptr, '\"');
2065 if (nl != NULL) *nl = '\0';
2066 strncpy(distro, ptr, length);
2067 } else {
2068 ptr = strstr(buf, "=");
2069 ptr++; // go beyond equals then
2070 char* nl = strchr(ptr, '\n');
2071 if (nl != NULL) *nl = '\0';
2072 strncpy(distro, ptr, length);
2073 }
2074 return;
2075 } else if (get_first_line) {
2076 char* nl = strchr(buf, '\n');
2077 if (nl != NULL) *nl = '\0';
2078 strncpy(distro, buf, length);
2079 return;
2080 }
2081 }
2082 // print last line and close
2083 char* nl = strchr(buf, '\n');
2084 if (nl != NULL) *nl = '\0';
2085 strncpy(distro, buf, length);
2086 }
2087
2088 static void parse_os_info(char* distro, size_t length, const char* file) {
2089 FILE* fp = fopen(file, "r");
2090 if (fp != NULL) {
2091 // if suse format, print out first line
2092 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2093 parse_os_info_helper(fp, distro, length, get_first_line);
2094 fclose(fp);
2095 }
2096 }
2097
2098 void os::get_summary_os_info(char* buf, size_t buflen) {
2099 for (int i = 0;; i++) {
2100 const char* file = distro_files[i];
2101 if (file == NULL) {
2102 break; // ran out of distro_files
2103 }
2104 if (file_exists(file)) {
2105 parse_os_info(buf, buflen, file);
2106 return;
2107 }
2108 }
2109 // special case for debian
2110 if (file_exists("/etc/debian_version")) {
2111 strncpy(buf, "Debian ", buflen);
2112 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2113 } else {
2114 strncpy(buf, "Linux", buflen);
2115 }
2116 }
2117
2118 void os::Linux::print_libversion_info(outputStream* st) {
2119 // libc, pthread
2120 st->print("libc:");
2121 st->print("%s ", os::Linux::glibc_version());
2122 st->print("%s ", os::Linux::libpthread_version());
2123 st->cr();
2124 }
2125
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127 st->print("\n/proc/meminfo:\n");
2128 _print_ascii_file("/proc/meminfo", st);
2129 st->cr();
2130 }
2131
2132 void os::print_memory_info(outputStream* st) {
2133
2134 st->print("Memory:");
2135 st->print(" %dk page", os::vm_page_size()>>10);
2136
2137 // values in struct sysinfo are "unsigned long"
2138 struct sysinfo si;
2139 sysinfo(&si);
2140
2141 st->print(", physical " UINT64_FORMAT "k",
2142 os::physical_memory() >> 10);
2143 st->print("(" UINT64_FORMAT "k free)",
2144 os::available_memory() >> 10);
2145 st->print(", swap " UINT64_FORMAT "k",
2146 ((jlong)si.totalswap * si.mem_unit) >> 10);
2147 st->print("(" UINT64_FORMAT "k free)",
2148 ((jlong)si.freeswap * si.mem_unit) >> 10);
2149 st->cr();
2150 }
2151
2152 // Print the first "model name" line and the first "flags" line
2153 // that we find and nothing more. We assume "model name" comes
2154 // before "flags" so if we find a second "model name", then the
2155 // "flags" field is considered missing.
2156 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2157 #if defined(IA32) || defined(AMD64)
2158 // Other platforms have less repetitive cpuinfo files
2159 FILE *fp = fopen("/proc/cpuinfo", "r");
2160 if (fp) {
2161 while (!feof(fp)) {
2162 if (fgets(buf, buflen, fp)) {
2163 // Assume model name comes before flags
2164 bool model_name_printed = false;
2165 if (strstr(buf, "model name") != NULL) {
2166 if (!model_name_printed) {
2167 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2168 st->print_raw(buf);
2169 model_name_printed = true;
2170 } else {
2171 // model name printed but not flags? Odd, just return
2172 fclose(fp);
2173 return true;
2174 }
2175 }
2176 // print the flags line too
2177 if (strstr(buf, "flags") != NULL) {
2178 st->print_raw(buf);
2179 fclose(fp);
2180 return true;
2181 }
2182 }
2183 }
2184 fclose(fp);
2185 }
2186 #endif // x86 platforms
2187 return false;
2188 }
2189
2190 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2191 // Only print the model name if the platform provides this as a summary
2192 if (!print_model_name_and_flags(st, buf, buflen)) {
2193 st->print("\n/proc/cpuinfo:\n");
2194 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2195 st->print_cr(" <Not Available>");
2196 }
2197 }
2198 }
2199
2200 #if defined(AMD64) || defined(IA32) || defined(X32)
2201 const char* search_string = "model name";
2202 #elif defined(PPC64)
2203 const char* search_string = "cpu";
2204 #elif defined(S390)
2205 const char* search_string = "processor";
2206 #elif defined(SPARC)
2207 const char* search_string = "cpu";
2208 #else
2209 const char* search_string = "Processor";
2210 #endif
2211
2212 // Parses the cpuinfo file for string representing the model name.
2213 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2214 FILE* fp = fopen("/proc/cpuinfo", "r");
2215 if (fp != NULL) {
2216 while (!feof(fp)) {
2217 char buf[256];
2218 if (fgets(buf, sizeof(buf), fp)) {
2219 char* start = strstr(buf, search_string);
2220 if (start != NULL) {
2221 char *ptr = start + strlen(search_string);
2222 char *end = buf + strlen(buf);
2223 while (ptr != end) {
2224 // skip whitespace and colon for the rest of the name.
2225 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2226 break;
2227 }
2228 ptr++;
2229 }
2230 if (ptr != end) {
2231 // reasonable string, get rid of newline and keep the rest
2232 char* nl = strchr(buf, '\n');
2233 if (nl != NULL) *nl = '\0';
2234 strncpy(cpuinfo, ptr, length);
2235 fclose(fp);
2236 return;
2237 }
2238 }
2239 }
2240 }
2241 fclose(fp);
2242 }
2243 // cpuinfo not found or parsing failed, just print generic string. The entire
2244 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2245 #if defined(AARCH64)
2246 strncpy(cpuinfo, "AArch64", length);
2247 #elif defined(AMD64)
2248 strncpy(cpuinfo, "x86_64", length);
2249 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2250 strncpy(cpuinfo, "ARM", length);
2251 #elif defined(IA32)
2252 strncpy(cpuinfo, "x86_32", length);
2253 #elif defined(IA64)
2254 strncpy(cpuinfo, "IA64", length);
2255 #elif defined(PPC)
2256 strncpy(cpuinfo, "PPC64", length);
2257 #elif defined(S390)
2258 strncpy(cpuinfo, "S390", length);
2259 #elif defined(SPARC)
2260 strncpy(cpuinfo, "sparcv9", length);
2261 #elif defined(ZERO_LIBARCH)
2262 strncpy(cpuinfo, ZERO_LIBARCH, length);
2263 #else
2264 strncpy(cpuinfo, "unknown", length);
2265 #endif
2266 }
2267
2268 static void print_signal_handler(outputStream* st, int sig,
2269 char* buf, size_t buflen);
2270
2271 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2272 st->print_cr("Signal Handlers:");
2273 print_signal_handler(st, SIGSEGV, buf, buflen);
2274 print_signal_handler(st, SIGBUS , buf, buflen);
2275 print_signal_handler(st, SIGFPE , buf, buflen);
2276 print_signal_handler(st, SIGPIPE, buf, buflen);
2277 print_signal_handler(st, SIGXFSZ, buf, buflen);
2278 print_signal_handler(st, SIGILL , buf, buflen);
2279 print_signal_handler(st, SR_signum, buf, buflen);
2280 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2281 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2282 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2283 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2284 #if defined(PPC64)
2285 print_signal_handler(st, SIGTRAP, buf, buflen);
2286 #endif
2287 }
2288
2289 static char saved_jvm_path[MAXPATHLEN] = {0};
2290
2291 // Find the full path to the current module, libjvm.so
2292 void os::jvm_path(char *buf, jint buflen) {
2293 // Error checking.
2294 if (buflen < MAXPATHLEN) {
2295 assert(false, "must use a large-enough buffer");
2296 buf[0] = '\0';
2297 return;
2298 }
2299 // Lazy resolve the path to current module.
2300 if (saved_jvm_path[0] != 0) {
2301 strcpy(buf, saved_jvm_path);
2302 return;
2303 }
2304
2305 char dli_fname[MAXPATHLEN];
2306 bool ret = dll_address_to_library_name(
2307 CAST_FROM_FN_PTR(address, os::jvm_path),
2308 dli_fname, sizeof(dli_fname), NULL);
2309 assert(ret, "cannot locate libjvm");
2310 char *rp = NULL;
2311 if (ret && dli_fname[0] != '\0') {
2312 rp = realpath(dli_fname, buf);
2313 }
2314 if (rp == NULL) {
2315 return;
2316 }
2317
2318 if (Arguments::sun_java_launcher_is_altjvm()) {
2319 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2320 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2321 // If "/jre/lib/" appears at the right place in the string, then
2322 // assume we are installed in a JDK and we're done. Otherwise, check
2323 // for a JAVA_HOME environment variable and fix up the path so it
2324 // looks like libjvm.so is installed there (append a fake suffix
2325 // hotspot/libjvm.so).
2326 const char *p = buf + strlen(buf) - 1;
2327 for (int count = 0; p > buf && count < 5; ++count) {
2328 for (--p; p > buf && *p != '/'; --p)
2329 /* empty */ ;
2330 }
2331
2332 if (strncmp(p, "/jre/lib/", 9) != 0) {
2333 // Look for JAVA_HOME in the environment.
2334 char* java_home_var = ::getenv("JAVA_HOME");
2335 if (java_home_var != NULL && java_home_var[0] != 0) {
2336 char* jrelib_p;
2337 int len;
2338
2339 // Check the current module name "libjvm.so".
2340 p = strrchr(buf, '/');
2341 if (p == NULL) {
2342 return;
2343 }
2344 assert(strstr(p, "/libjvm") == p, "invalid library name");
2345
2346 rp = realpath(java_home_var, buf);
2347 if (rp == NULL) {
2348 return;
2349 }
2350
2351 // determine if this is a legacy image or modules image
2352 // modules image doesn't have "jre" subdirectory
2353 len = strlen(buf);
2354 assert(len < buflen, "Ran out of buffer room");
2355 jrelib_p = buf + len;
2356 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2357 if (0 != access(buf, F_OK)) {
2358 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2359 }
2360
2361 if (0 == access(buf, F_OK)) {
2362 // Use current module name "libjvm.so"
2363 len = strlen(buf);
2364 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2365 } else {
2366 // Go back to path of .so
2367 rp = realpath(dli_fname, buf);
2368 if (rp == NULL) {
2369 return;
2370 }
2371 }
2372 }
2373 }
2374 }
2375
2376 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2377 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2378 }
2379
2380 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2381 // no prefix required, not even "_"
2382 }
2383
2384 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2385 // no suffix required
2386 }
2387
2388 ////////////////////////////////////////////////////////////////////////////////
2389 // sun.misc.Signal support
2390
2391 static volatile jint sigint_count = 0;
2392
2393 static void UserHandler(int sig, void *siginfo, void *context) {
2394 // 4511530 - sem_post is serialized and handled by the manager thread. When
2395 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2396 // don't want to flood the manager thread with sem_post requests.
2397 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2398 return;
2399 }
2400
2401 // Ctrl-C is pressed during error reporting, likely because the error
2402 // handler fails to abort. Let VM die immediately.
2403 if (sig == SIGINT && is_error_reported()) {
2404 os::die();
2405 }
2406
2407 os::signal_notify(sig);
2408 }
2409
2410 void* os::user_handler() {
2411 return CAST_FROM_FN_PTR(void*, UserHandler);
2412 }
2413
2414 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2415 struct timespec ts;
2416 // Semaphore's are always associated with CLOCK_REALTIME
2417 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2418 // see unpackTime for discussion on overflow checking
2419 if (sec >= MAX_SECS) {
2420 ts.tv_sec += MAX_SECS;
2421 ts.tv_nsec = 0;
2422 } else {
2423 ts.tv_sec += sec;
2424 ts.tv_nsec += nsec;
2425 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2426 ts.tv_nsec -= NANOSECS_PER_SEC;
2427 ++ts.tv_sec; // note: this must be <= max_secs
2428 }
2429 }
2430
2431 return ts;
2432 }
2433
2434 extern "C" {
2435 typedef void (*sa_handler_t)(int);
2436 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2437 }
2438
2439 void* os::signal(int signal_number, void* handler) {
2440 struct sigaction sigAct, oldSigAct;
2441
2442 sigfillset(&(sigAct.sa_mask));
2443 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2444 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2445
2446 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2447 // -1 means registration failed
2448 return (void *)-1;
2449 }
2450
2451 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2452 }
2453
2454 void os::signal_raise(int signal_number) {
2455 ::raise(signal_number);
2456 }
2457
2458 // The following code is moved from os.cpp for making this
2459 // code platform specific, which it is by its very nature.
2460
2461 // Will be modified when max signal is changed to be dynamic
2462 int os::sigexitnum_pd() {
2463 return NSIG;
2464 }
2465
2466 // a counter for each possible signal value
2467 static volatile jint pending_signals[NSIG+1] = { 0 };
2468
2469 // Linux(POSIX) specific hand shaking semaphore.
2470 static sem_t sig_sem;
2471 static PosixSemaphore sr_semaphore;
2472
2473 void os::signal_init_pd() {
2474 // Initialize signal structures
2475 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2476
2477 // Initialize signal semaphore
2478 ::sem_init(&sig_sem, 0, 0);
2479 }
2480
2481 void os::signal_notify(int sig) {
2482 Atomic::inc(&pending_signals[sig]);
2483 ::sem_post(&sig_sem);
2484 }
2485
2486 static int check_pending_signals(bool wait) {
2487 Atomic::store(0, &sigint_count);
2488 for (;;) {
2489 for (int i = 0; i < NSIG + 1; i++) {
2490 jint n = pending_signals[i];
2491 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2492 return i;
2493 }
2494 }
2495 if (!wait) {
2496 return -1;
2497 }
2498 JavaThread *thread = JavaThread::current();
2499 ThreadBlockInVM tbivm(thread);
2500
2501 bool threadIsSuspended;
2502 do {
2503 thread->set_suspend_equivalent();
2504 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2505 ::sem_wait(&sig_sem);
2506
2507 // were we externally suspended while we were waiting?
2508 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2509 if (threadIsSuspended) {
2510 // The semaphore has been incremented, but while we were waiting
2511 // another thread suspended us. We don't want to continue running
2512 // while suspended because that would surprise the thread that
2513 // suspended us.
2514 ::sem_post(&sig_sem);
2515
2516 thread->java_suspend_self();
2517 }
2518 } while (threadIsSuspended);
2519 }
2520 }
2521
2522 int os::signal_lookup() {
2523 return check_pending_signals(false);
2524 }
2525
2526 int os::signal_wait() {
2527 return check_pending_signals(true);
2528 }
2529
2530 ////////////////////////////////////////////////////////////////////////////////
2531 // Virtual Memory
2532
2533 int os::vm_page_size() {
2534 // Seems redundant as all get out
2535 assert(os::Linux::page_size() != -1, "must call os::init");
2536 return os::Linux::page_size();
2537 }
2538
2539 // Solaris allocates memory by pages.
2540 int os::vm_allocation_granularity() {
2541 assert(os::Linux::page_size() != -1, "must call os::init");
2542 return os::Linux::page_size();
2543 }
2544
2545 // Rationale behind this function:
2546 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2547 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2548 // samples for JITted code. Here we create private executable mapping over the code cache
2549 // and then we can use standard (well, almost, as mapping can change) way to provide
2550 // info for the reporting script by storing timestamp and location of symbol
2551 void linux_wrap_code(char* base, size_t size) {
2552 static volatile jint cnt = 0;
2553
2554 if (!UseOprofile) {
2555 return;
2556 }
2557
2558 char buf[PATH_MAX+1];
2559 int num = Atomic::add(1, &cnt);
2560
2561 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2562 os::get_temp_directory(), os::current_process_id(), num);
2563 unlink(buf);
2564
2565 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2566
2567 if (fd != -1) {
2568 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2569 if (rv != (off_t)-1) {
2570 if (::write(fd, "", 1) == 1) {
2571 mmap(base, size,
2572 PROT_READ|PROT_WRITE|PROT_EXEC,
2573 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2574 }
2575 }
2576 ::close(fd);
2577 unlink(buf);
2578 }
2579 }
2580
2581 static bool recoverable_mmap_error(int err) {
2582 // See if the error is one we can let the caller handle. This
2583 // list of errno values comes from JBS-6843484. I can't find a
2584 // Linux man page that documents this specific set of errno
2585 // values so while this list currently matches Solaris, it may
2586 // change as we gain experience with this failure mode.
2587 switch (err) {
2588 case EBADF:
2589 case EINVAL:
2590 case ENOTSUP:
2591 // let the caller deal with these errors
2592 return true;
2593
2594 default:
2595 // Any remaining errors on this OS can cause our reserved mapping
2596 // to be lost. That can cause confusion where different data
2597 // structures think they have the same memory mapped. The worst
2598 // scenario is if both the VM and a library think they have the
2599 // same memory mapped.
2600 return false;
2601 }
2602 }
2603
2604 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2605 int err) {
2606 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2607 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2608 os::strerror(err), err);
2609 }
2610
2611 static void warn_fail_commit_memory(char* addr, size_t size,
2612 size_t alignment_hint, bool exec,
2613 int err) {
2614 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2616 alignment_hint, exec, os::strerror(err), err);
2617 }
2618
2619 // NOTE: Linux kernel does not really reserve the pages for us.
2620 // All it does is to check if there are enough free pages
2621 // left at the time of mmap(). This could be a potential
2622 // problem.
2623 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2624 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2625 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2626 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2627 if (res != (uintptr_t) MAP_FAILED) {
2628 if (UseNUMAInterleaving) {
2629 numa_make_global(addr, size);
2630 }
2631 return 0;
2632 }
2633
2634 int err = errno; // save errno from mmap() call above
2635
2636 if (!recoverable_mmap_error(err)) {
2637 warn_fail_commit_memory(addr, size, exec, err);
2638 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2639 }
2640
2641 return err;
2642 }
2643
2644 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2645 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2646 }
2647
2648 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2649 const char* mesg) {
2650 assert(mesg != NULL, "mesg must be specified");
2651 int err = os::Linux::commit_memory_impl(addr, size, exec);
2652 if (err != 0) {
2653 // the caller wants all commit errors to exit with the specified mesg:
2654 warn_fail_commit_memory(addr, size, exec, err);
2655 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2656 }
2657 }
2658
2659 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2660 #ifndef MAP_HUGETLB
2661 #define MAP_HUGETLB 0x40000
2662 #endif
2663
2664 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2665 #ifndef MADV_HUGEPAGE
2666 #define MADV_HUGEPAGE 14
2667 #endif
2668
2669 int os::Linux::commit_memory_impl(char* addr, size_t size,
2670 size_t alignment_hint, bool exec) {
2671 int err = os::Linux::commit_memory_impl(addr, size, exec);
2672 if (err == 0) {
2673 realign_memory(addr, size, alignment_hint);
2674 }
2675 return err;
2676 }
2677
2678 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2679 bool exec) {
2680 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2681 }
2682
2683 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2684 size_t alignment_hint, bool exec,
2685 const char* mesg) {
2686 assert(mesg != NULL, "mesg must be specified");
2687 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2688 if (err != 0) {
2689 // the caller wants all commit errors to exit with the specified mesg:
2690 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2691 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2692 }
2693 }
2694
2695 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2696 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2697 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2698 // be supported or the memory may already be backed by huge pages.
2699 ::madvise(addr, bytes, MADV_HUGEPAGE);
2700 }
2701 }
2702
2703 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2704 // This method works by doing an mmap over an existing mmaping and effectively discarding
2705 // the existing pages. However it won't work for SHM-based large pages that cannot be
2706 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2707 // small pages on top of the SHM segment. This method always works for small pages, so we
2708 // allow that in any case.
2709 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2710 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2711 }
2712 }
2713
2714 void os::numa_make_global(char *addr, size_t bytes) {
2715 Linux::numa_interleave_memory(addr, bytes);
2716 }
2717
2718 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2719 // bind policy to MPOL_PREFERRED for the current thread.
2720 #define USE_MPOL_PREFERRED 0
2721
2722 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2723 // To make NUMA and large pages more robust when both enabled, we need to ease
2724 // the requirements on where the memory should be allocated. MPOL_BIND is the
2725 // default policy and it will force memory to be allocated on the specified
2726 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2727 // the specified node, but will not force it. Using this policy will prevent
2728 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2729 // free large pages.
2730 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2731 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2732 }
2733
2734 bool os::numa_topology_changed() { return false; }
2735
2736 size_t os::numa_get_groups_num() {
2737 int max_node = Linux::numa_max_node();
2738 return max_node > 0 ? max_node + 1 : 1;
2739 }
2740
2741 int os::numa_get_group_id() {
2742 int cpu_id = Linux::sched_getcpu();
2743 if (cpu_id != -1) {
2744 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2745 if (lgrp_id != -1) {
2746 return lgrp_id;
2747 }
2748 }
2749 return 0;
2750 }
2751
2752 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2753 for (size_t i = 0; i < size; i++) {
2754 ids[i] = i;
2755 }
2756 return size;
2757 }
2758
2759 bool os::get_page_info(char *start, page_info* info) {
2760 return false;
2761 }
2762
2763 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2764 page_info* page_found) {
2765 return end;
2766 }
2767
2768
2769 int os::Linux::sched_getcpu_syscall(void) {
2770 unsigned int cpu = 0;
2771 int retval = -1;
2772
2773 #if defined(IA32)
2774 #ifndef SYS_getcpu
2775 #define SYS_getcpu 318
2776 #endif
2777 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2778 #elif defined(AMD64)
2779 // Unfortunately we have to bring all these macros here from vsyscall.h
2780 // to be able to compile on old linuxes.
2781 #define __NR_vgetcpu 2
2782 #define VSYSCALL_START (-10UL << 20)
2783 #define VSYSCALL_SIZE 1024
2784 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2785 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2786 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2787 retval = vgetcpu(&cpu, NULL, NULL);
2788 #endif
2789
2790 return (retval == -1) ? retval : cpu;
2791 }
2792
2793 // Something to do with the numa-aware allocator needs these symbols
2794 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2795 extern "C" JNIEXPORT void numa_error(char *where) { }
2796
2797
2798 // If we are running with libnuma version > 2, then we should
2799 // be trying to use symbols with versions 1.1
2800 // If we are running with earlier version, which did not have symbol versions,
2801 // we should use the base version.
2802 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2803 void *f = dlvsym(handle, name, "libnuma_1.1");
2804 if (f == NULL) {
2805 f = dlsym(handle, name);
2806 }
2807 return f;
2808 }
2809
2810 bool os::Linux::libnuma_init() {
2811 // sched_getcpu() should be in libc.
2812 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2814
2815 // If it's not, try a direct syscall.
2816 if (sched_getcpu() == -1) {
2817 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2818 (void*)&sched_getcpu_syscall));
2819 }
2820
2821 if (sched_getcpu() != -1) { // Does it work?
2822 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2823 if (handle != NULL) {
2824 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2825 libnuma_dlsym(handle, "numa_node_to_cpus")));
2826 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2827 libnuma_dlsym(handle, "numa_max_node")));
2828 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2829 libnuma_dlsym(handle, "numa_available")));
2830 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2831 libnuma_dlsym(handle, "numa_tonode_memory")));
2832 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2833 libnuma_dlsym(handle, "numa_interleave_memory")));
2834 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2835 libnuma_dlsym(handle, "numa_set_bind_policy")));
2836
2837
2838 if (numa_available() != -1) {
2839 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2840 // Create a cpu -> node mapping
2841 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2842 rebuild_cpu_to_node_map();
2843 return true;
2844 }
2845 }
2846 }
2847 return false;
2848 }
2849
2850 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2851 // Creating guard page is very expensive. Java thread has HotSpot
2852 // guard pages, only enable glibc guard page for non-Java threads.
2853 // (Remember: compiler thread is a Java thread, too!)
2854 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2855 }
2856
2857 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2858 // The table is later used in get_node_by_cpu().
2859 void os::Linux::rebuild_cpu_to_node_map() {
2860 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2861 // in libnuma (possible values are starting from 16,
2862 // and continuing up with every other power of 2, but less
2863 // than the maximum number of CPUs supported by kernel), and
2864 // is a subject to change (in libnuma version 2 the requirements
2865 // are more reasonable) we'll just hardcode the number they use
2866 // in the library.
2867 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2868
2869 size_t cpu_num = processor_count();
2870 size_t cpu_map_size = NCPUS / BitsPerCLong;
2871 size_t cpu_map_valid_size =
2872 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2873
2874 cpu_to_node()->clear();
2875 cpu_to_node()->at_grow(cpu_num - 1);
2876 size_t node_num = numa_get_groups_num();
2877
2878 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2879 for (size_t i = 0; i < node_num; i++) {
2880 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2881 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2882 if (cpu_map[j] != 0) {
2883 for (size_t k = 0; k < BitsPerCLong; k++) {
2884 if (cpu_map[j] & (1UL << k)) {
2885 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2886 }
2887 }
2888 }
2889 }
2890 }
2891 }
2892 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2893 }
2894
2895 int os::Linux::get_node_by_cpu(int cpu_id) {
2896 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2897 return cpu_to_node()->at(cpu_id);
2898 }
2899 return -1;
2900 }
2901
2902 GrowableArray<int>* os::Linux::_cpu_to_node;
2903 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2904 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2905 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2906 os::Linux::numa_available_func_t os::Linux::_numa_available;
2907 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2908 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2909 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2910 unsigned long* os::Linux::_numa_all_nodes;
2911
2912 bool os::pd_uncommit_memory(char* addr, size_t size) {
2913 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2914 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2915 return res != (uintptr_t) MAP_FAILED;
2916 }
2917
2918 static address get_stack_commited_bottom(address bottom, size_t size) {
2919 address nbot = bottom;
2920 address ntop = bottom + size;
2921
2922 size_t page_sz = os::vm_page_size();
2923 unsigned pages = size / page_sz;
2924
2925 unsigned char vec[1];
2926 unsigned imin = 1, imax = pages + 1, imid;
2927 int mincore_return_value = 0;
2928
2929 assert(imin <= imax, "Unexpected page size");
2930
2931 while (imin < imax) {
2932 imid = (imax + imin) / 2;
2933 nbot = ntop - (imid * page_sz);
2934
2935 // Use a trick with mincore to check whether the page is mapped or not.
2936 // mincore sets vec to 1 if page resides in memory and to 0 if page
2937 // is swapped output but if page we are asking for is unmapped
2938 // it returns -1,ENOMEM
2939 mincore_return_value = mincore(nbot, page_sz, vec);
2940
2941 if (mincore_return_value == -1) {
2942 // Page is not mapped go up
2943 // to find first mapped page
2944 if (errno != EAGAIN) {
2945 assert(errno == ENOMEM, "Unexpected mincore errno");
2946 imax = imid;
2947 }
2948 } else {
2949 // Page is mapped go down
2950 // to find first not mapped page
2951 imin = imid + 1;
2952 }
2953 }
2954
2955 nbot = nbot + page_sz;
2956
2957 // Adjust stack bottom one page up if last checked page is not mapped
2958 if (mincore_return_value == -1) {
2959 nbot = nbot + page_sz;
2960 }
2961
2962 return nbot;
2963 }
2964
2965
2966 // Linux uses a growable mapping for the stack, and if the mapping for
2967 // the stack guard pages is not removed when we detach a thread the
2968 // stack cannot grow beyond the pages where the stack guard was
2969 // mapped. If at some point later in the process the stack expands to
2970 // that point, the Linux kernel cannot expand the stack any further
2971 // because the guard pages are in the way, and a segfault occurs.
2972 //
2973 // However, it's essential not to split the stack region by unmapping
2974 // a region (leaving a hole) that's already part of the stack mapping,
2975 // so if the stack mapping has already grown beyond the guard pages at
2976 // the time we create them, we have to truncate the stack mapping.
2977 // So, we need to know the extent of the stack mapping when
2978 // create_stack_guard_pages() is called.
2979
2980 // We only need this for stacks that are growable: at the time of
2981 // writing thread stacks don't use growable mappings (i.e. those
2982 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2983 // only applies to the main thread.
2984
2985 // If the (growable) stack mapping already extends beyond the point
2986 // where we're going to put our guard pages, truncate the mapping at
2987 // that point by munmap()ping it. This ensures that when we later
2988 // munmap() the guard pages we don't leave a hole in the stack
2989 // mapping. This only affects the main/initial thread
2990
2991 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2992 if (os::Linux::is_initial_thread()) {
2993 // As we manually grow stack up to bottom inside create_attached_thread(),
2994 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2995 // we don't need to do anything special.
2996 // Check it first, before calling heavy function.
2997 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2998 unsigned char vec[1];
2999
3000 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3001 // Fallback to slow path on all errors, including EAGAIN
3002 stack_extent = (uintptr_t) get_stack_commited_bottom(
3003 os::Linux::initial_thread_stack_bottom(),
3004 (size_t)addr - stack_extent);
3005 }
3006
3007 if (stack_extent < (uintptr_t)addr) {
3008 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3009 }
3010 }
3011
3012 return os::commit_memory(addr, size, !ExecMem);
3013 }
3014
3015 // If this is a growable mapping, remove the guard pages entirely by
3016 // munmap()ping them. If not, just call uncommit_memory(). This only
3017 // affects the main/initial thread, but guard against future OS changes
3018 // It's safe to always unmap guard pages for initial thread because we
3019 // always place it right after end of the mapped region
3020
3021 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3022 uintptr_t stack_extent, stack_base;
3023
3024 if (os::Linux::is_initial_thread()) {
3025 return ::munmap(addr, size) == 0;
3026 }
3027
3028 return os::uncommit_memory(addr, size);
3029 }
3030
3031 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3032 // at 'requested_addr'. If there are existing memory mappings at the same
3033 // location, however, they will be overwritten. If 'fixed' is false,
3034 // 'requested_addr' is only treated as a hint, the return value may or
3035 // may not start from the requested address. Unlike Linux mmap(), this
3036 // function returns NULL to indicate failure.
3037 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3038 char * addr;
3039 int flags;
3040
3041 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3042 if (fixed) {
3043 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3044 flags |= MAP_FIXED;
3045 }
3046
3047 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3048 // touch an uncommitted page. Otherwise, the read/write might
3049 // succeed if we have enough swap space to back the physical page.
3050 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3051 flags, -1, 0);
3052
3053 return addr == MAP_FAILED ? NULL : addr;
3054 }
3055
3056 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3057 // (req_addr != NULL) or with a given alignment.
3058 // - bytes shall be a multiple of alignment.
3059 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3060 // - alignment sets the alignment at which memory shall be allocated.
3061 // It must be a multiple of allocation granularity.
3062 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3063 // req_addr or NULL.
3064 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3065
3066 size_t extra_size = bytes;
3067 if (req_addr == NULL && alignment > 0) {
3068 extra_size += alignment;
3069 }
3070
3071 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3072 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3073 -1, 0);
3074 if (start == MAP_FAILED) {
3075 start = NULL;
3076 } else {
3077 if (req_addr != NULL) {
3078 if (start != req_addr) {
3079 ::munmap(start, extra_size);
3080 start = NULL;
3081 }
3082 } else {
3083 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3084 char* const end_aligned = start_aligned + bytes;
3085 char* const end = start + extra_size;
3086 if (start_aligned > start) {
3087 ::munmap(start, start_aligned - start);
3088 }
3089 if (end_aligned < end) {
3090 ::munmap(end_aligned, end - end_aligned);
3091 }
3092 start = start_aligned;
3093 }
3094 }
3095 return start;
3096 }
3097
3098 static int anon_munmap(char * addr, size_t size) {
3099 return ::munmap(addr, size) == 0;
3100 }
3101
3102 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3103 size_t alignment_hint) {
3104 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3105 }
3106
3107 bool os::pd_release_memory(char* addr, size_t size) {
3108 return anon_munmap(addr, size);
3109 }
3110
3111 static bool linux_mprotect(char* addr, size_t size, int prot) {
3112 // Linux wants the mprotect address argument to be page aligned.
3113 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3114
3115 // According to SUSv3, mprotect() should only be used with mappings
3116 // established by mmap(), and mmap() always maps whole pages. Unaligned
3117 // 'addr' likely indicates problem in the VM (e.g. trying to change
3118 // protection of malloc'ed or statically allocated memory). Check the
3119 // caller if you hit this assert.
3120 assert(addr == bottom, "sanity check");
3121
3122 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3123 return ::mprotect(bottom, size, prot) == 0;
3124 }
3125
3126 // Set protections specified
3127 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3128 bool is_committed) {
3129 unsigned int p = 0;
3130 switch (prot) {
3131 case MEM_PROT_NONE: p = PROT_NONE; break;
3132 case MEM_PROT_READ: p = PROT_READ; break;
3133 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3134 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3135 default:
3136 ShouldNotReachHere();
3137 }
3138 // is_committed is unused.
3139 return linux_mprotect(addr, bytes, p);
3140 }
3141
3142 bool os::guard_memory(char* addr, size_t size) {
3143 return linux_mprotect(addr, size, PROT_NONE);
3144 }
3145
3146 bool os::unguard_memory(char* addr, size_t size) {
3147 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3148 }
3149
3150 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3151 size_t page_size) {
3152 bool result = false;
3153 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3154 MAP_ANONYMOUS|MAP_PRIVATE,
3155 -1, 0);
3156 if (p != MAP_FAILED) {
3157 void *aligned_p = align_ptr_up(p, page_size);
3158
3159 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3160
3161 munmap(p, page_size * 2);
3162 }
3163
3164 if (warn && !result) {
3165 warning("TransparentHugePages is not supported by the operating system.");
3166 }
3167
3168 return result;
3169 }
3170
3171 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3172 bool result = false;
3173 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3174 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3175 -1, 0);
3176
3177 if (p != MAP_FAILED) {
3178 // We don't know if this really is a huge page or not.
3179 FILE *fp = fopen("/proc/self/maps", "r");
3180 if (fp) {
3181 while (!feof(fp)) {
3182 char chars[257];
3183 long x = 0;
3184 if (fgets(chars, sizeof(chars), fp)) {
3185 if (sscanf(chars, "%lx-%*x", &x) == 1
3186 && x == (long)p) {
3187 if (strstr (chars, "hugepage")) {
3188 result = true;
3189 break;
3190 }
3191 }
3192 }
3193 }
3194 fclose(fp);
3195 }
3196 munmap(p, page_size);
3197 }
3198
3199 if (warn && !result) {
3200 warning("HugeTLBFS is not supported by the operating system.");
3201 }
3202
3203 return result;
3204 }
3205
3206 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3207 //
3208 // From the coredump_filter documentation:
3209 //
3210 // - (bit 0) anonymous private memory
3211 // - (bit 1) anonymous shared memory
3212 // - (bit 2) file-backed private memory
3213 // - (bit 3) file-backed shared memory
3214 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3215 // effective only if the bit 2 is cleared)
3216 // - (bit 5) hugetlb private memory
3217 // - (bit 6) hugetlb shared memory
3218 //
3219 static void set_coredump_filter(void) {
3220 FILE *f;
3221 long cdm;
3222
3223 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3224 return;
3225 }
3226
3227 if (fscanf(f, "%lx", &cdm) != 1) {
3228 fclose(f);
3229 return;
3230 }
3231
3232 rewind(f);
3233
3234 if ((cdm & LARGEPAGES_BIT) == 0) {
3235 cdm |= LARGEPAGES_BIT;
3236 fprintf(f, "%#lx", cdm);
3237 }
3238
3239 fclose(f);
3240 }
3241
3242 // Large page support
3243
3244 static size_t _large_page_size = 0;
3245
3246 size_t os::Linux::find_large_page_size() {
3247 size_t large_page_size = 0;
3248
3249 // large_page_size on Linux is used to round up heap size. x86 uses either
3250 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3251 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3252 // page as large as 256M.
3253 //
3254 // Here we try to figure out page size by parsing /proc/meminfo and looking
3255 // for a line with the following format:
3256 // Hugepagesize: 2048 kB
3257 //
3258 // If we can't determine the value (e.g. /proc is not mounted, or the text
3259 // format has been changed), we'll use the largest page size supported by
3260 // the processor.
3261
3262 #ifndef ZERO
3263 large_page_size =
3264 AARCH64_ONLY(2 * M)
3265 AMD64_ONLY(2 * M)
3266 ARM32_ONLY(2 * M)
3267 IA32_ONLY(4 * M)
3268 IA64_ONLY(256 * M)
3269 PPC_ONLY(4 * M)
3270 S390_ONLY(1 * M)
3271 SPARC_ONLY(4 * M);
3272 #endif // ZERO
3273
3274 FILE *fp = fopen("/proc/meminfo", "r");
3275 if (fp) {
3276 while (!feof(fp)) {
3277 int x = 0;
3278 char buf[16];
3279 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3280 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3281 large_page_size = x * K;
3282 break;
3283 }
3284 } else {
3285 // skip to next line
3286 for (;;) {
3287 int ch = fgetc(fp);
3288 if (ch == EOF || ch == (int)'\n') break;
3289 }
3290 }
3291 }
3292 fclose(fp);
3293 }
3294
3295 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3296 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3297 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3298 proper_unit_for_byte_size(large_page_size));
3299 }
3300
3301 return large_page_size;
3302 }
3303
3304 size_t os::Linux::setup_large_page_size() {
3305 _large_page_size = Linux::find_large_page_size();
3306 const size_t default_page_size = (size_t)Linux::page_size();
3307 if (_large_page_size > default_page_size) {
3308 _page_sizes[0] = _large_page_size;
3309 _page_sizes[1] = default_page_size;
3310 _page_sizes[2] = 0;
3311 }
3312
3313 return _large_page_size;
3314 }
3315
3316 bool os::Linux::setup_large_page_type(size_t page_size) {
3317 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3318 FLAG_IS_DEFAULT(UseSHM) &&
3319 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3320
3321 // The type of large pages has not been specified by the user.
3322
3323 // Try UseHugeTLBFS and then UseSHM.
3324 UseHugeTLBFS = UseSHM = true;
3325
3326 // Don't try UseTransparentHugePages since there are known
3327 // performance issues with it turned on. This might change in the future.
3328 UseTransparentHugePages = false;
3329 }
3330
3331 if (UseTransparentHugePages) {
3332 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3333 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3334 UseHugeTLBFS = false;
3335 UseSHM = false;
3336 return true;
3337 }
3338 UseTransparentHugePages = false;
3339 }
3340
3341 if (UseHugeTLBFS) {
3342 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3343 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3344 UseSHM = false;
3345 return true;
3346 }
3347 UseHugeTLBFS = false;
3348 }
3349
3350 return UseSHM;
3351 }
3352
3353 void os::large_page_init() {
3354 if (!UseLargePages &&
3355 !UseTransparentHugePages &&
3356 !UseHugeTLBFS &&
3357 !UseSHM) {
3358 // Not using large pages.
3359 return;
3360 }
3361
3362 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3363 // The user explicitly turned off large pages.
3364 // Ignore the rest of the large pages flags.
3365 UseTransparentHugePages = false;
3366 UseHugeTLBFS = false;
3367 UseSHM = false;
3368 return;
3369 }
3370
3371 size_t large_page_size = Linux::setup_large_page_size();
3372 UseLargePages = Linux::setup_large_page_type(large_page_size);
3373
3374 set_coredump_filter();
3375 }
3376
3377 #ifndef SHM_HUGETLB
3378 #define SHM_HUGETLB 04000
3379 #endif
3380
3381 #define shm_warning_format(format, ...) \
3382 do { \
3383 if (UseLargePages && \
3384 (!FLAG_IS_DEFAULT(UseLargePages) || \
3385 !FLAG_IS_DEFAULT(UseSHM) || \
3386 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3387 warning(format, __VA_ARGS__); \
3388 } \
3389 } while (0)
3390
3391 #define shm_warning(str) shm_warning_format("%s", str)
3392
3393 #define shm_warning_with_errno(str) \
3394 do { \
3395 int err = errno; \
3396 shm_warning_format(str " (error = %d)", err); \
3397 } while (0)
3398
3399 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3400 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3401
3402 if (!is_size_aligned(alignment, SHMLBA)) {
3403 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3404 return NULL;
3405 }
3406
3407 // To ensure that we get 'alignment' aligned memory from shmat,
3408 // we pre-reserve aligned virtual memory and then attach to that.
3409
3410 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3411 if (pre_reserved_addr == NULL) {
3412 // Couldn't pre-reserve aligned memory.
3413 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3414 return NULL;
3415 }
3416
3417 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3418 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3419
3420 if ((intptr_t)addr == -1) {
3421 int err = errno;
3422 shm_warning_with_errno("Failed to attach shared memory.");
3423
3424 assert(err != EACCES, "Unexpected error");
3425 assert(err != EIDRM, "Unexpected error");
3426 assert(err != EINVAL, "Unexpected error");
3427
3428 // Since we don't know if the kernel unmapped the pre-reserved memory area
3429 // we can't unmap it, since that would potentially unmap memory that was
3430 // mapped from other threads.
3431 return NULL;
3432 }
3433
3434 return addr;
3435 }
3436
3437 static char* shmat_at_address(int shmid, char* req_addr) {
3438 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3439 assert(false, "Requested address needs to be SHMLBA aligned");
3440 return NULL;
3441 }
3442
3443 char* addr = (char*)shmat(shmid, req_addr, 0);
3444
3445 if ((intptr_t)addr == -1) {
3446 shm_warning_with_errno("Failed to attach shared memory.");
3447 return NULL;
3448 }
3449
3450 return addr;
3451 }
3452
3453 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3454 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3455 if (req_addr != NULL) {
3456 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3457 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3458 return shmat_at_address(shmid, req_addr);
3459 }
3460
3461 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3462 // return large page size aligned memory addresses when req_addr == NULL.
3463 // However, if the alignment is larger than the large page size, we have
3464 // to manually ensure that the memory returned is 'alignment' aligned.
3465 if (alignment > os::large_page_size()) {
3466 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3467 return shmat_with_alignment(shmid, bytes, alignment);
3468 } else {
3469 return shmat_at_address(shmid, NULL);
3470 }
3471 }
3472
3473 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3474 char* req_addr, bool exec) {
3475 // "exec" is passed in but not used. Creating the shared image for
3476 // the code cache doesn't have an SHM_X executable permission to check.
3477 assert(UseLargePages && UseSHM, "only for SHM large pages");
3478 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3479 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3480
3481 if (!is_size_aligned(bytes, os::large_page_size())) {
3482 return NULL; // Fallback to small pages.
3483 }
3484
3485 // Create a large shared memory region to attach to based on size.
3486 // Currently, size is the total size of the heap.
3487 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3488 if (shmid == -1) {
3489 // Possible reasons for shmget failure:
3490 // 1. shmmax is too small for Java heap.
3491 // > check shmmax value: cat /proc/sys/kernel/shmmax
3492 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3493 // 2. not enough large page memory.
3494 // > check available large pages: cat /proc/meminfo
3495 // > increase amount of large pages:
3496 // echo new_value > /proc/sys/vm/nr_hugepages
3497 // Note 1: different Linux may use different name for this property,
3498 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3499 // Note 2: it's possible there's enough physical memory available but
3500 // they are so fragmented after a long run that they can't
3501 // coalesce into large pages. Try to reserve large pages when
3502 // the system is still "fresh".
3503 shm_warning_with_errno("Failed to reserve shared memory.");
3504 return NULL;
3505 }
3506
3507 // Attach to the region.
3508 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3509
3510 // Remove shmid. If shmat() is successful, the actual shared memory segment
3511 // will be deleted when it's detached by shmdt() or when the process
3512 // terminates. If shmat() is not successful this will remove the shared
3513 // segment immediately.
3514 shmctl(shmid, IPC_RMID, NULL);
3515
3516 return addr;
3517 }
3518
3519 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3520 int error) {
3521 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3522
3523 bool warn_on_failure = UseLargePages &&
3524 (!FLAG_IS_DEFAULT(UseLargePages) ||
3525 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3526 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3527
3528 if (warn_on_failure) {
3529 char msg[128];
3530 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3531 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3532 warning("%s", msg);
3533 }
3534 }
3535
3536 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3537 char* req_addr,
3538 bool exec) {
3539 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3540 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3541 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3542
3543 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3544 char* addr = (char*)::mmap(req_addr, bytes, prot,
3545 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3546 -1, 0);
3547
3548 if (addr == MAP_FAILED) {
3549 warn_on_large_pages_failure(req_addr, bytes, errno);
3550 return NULL;
3551 }
3552
3553 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3554
3555 return addr;
3556 }
3557
3558 // Reserve memory using mmap(MAP_HUGETLB).
3559 // - bytes shall be a multiple of alignment.
3560 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3561 // - alignment sets the alignment at which memory shall be allocated.
3562 // It must be a multiple of allocation granularity.
3563 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3564 // req_addr or NULL.
3565 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3566 size_t alignment,
3567 char* req_addr,
3568 bool exec) {
3569 size_t large_page_size = os::large_page_size();
3570 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3571
3572 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3573 assert(is_size_aligned(bytes, alignment), "Must be");
3574
3575 // First reserve - but not commit - the address range in small pages.
3576 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3577
3578 if (start == NULL) {
3579 return NULL;
3580 }
3581
3582 assert(is_ptr_aligned(start, alignment), "Must be");
3583
3584 char* end = start + bytes;
3585
3586 // Find the regions of the allocated chunk that can be promoted to large pages.
3587 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3588 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3589
3590 size_t lp_bytes = lp_end - lp_start;
3591
3592 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3593
3594 if (lp_bytes == 0) {
3595 // The mapped region doesn't even span the start and the end of a large page.
3596 // Fall back to allocate a non-special area.
3597 ::munmap(start, end - start);
3598 return NULL;
3599 }
3600
3601 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3602
3603 void* result;
3604
3605 // Commit small-paged leading area.
3606 if (start != lp_start) {
3607 result = ::mmap(start, lp_start - start, prot,
3608 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3609 -1, 0);
3610 if (result == MAP_FAILED) {
3611 ::munmap(lp_start, end - lp_start);
3612 return NULL;
3613 }
3614 }
3615
3616 // Commit large-paged area.
3617 result = ::mmap(lp_start, lp_bytes, prot,
3618 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3619 -1, 0);
3620 if (result == MAP_FAILED) {
3621 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3622 // If the mmap above fails, the large pages region will be unmapped and we
3623 // have regions before and after with small pages. Release these regions.
3624 //
3625 // | mapped | unmapped | mapped |
3626 // ^ ^ ^ ^
3627 // start lp_start lp_end end
3628 //
3629 ::munmap(start, lp_start - start);
3630 ::munmap(lp_end, end - lp_end);
3631 return NULL;
3632 }
3633
3634 // Commit small-paged trailing area.
3635 if (lp_end != end) {
3636 result = ::mmap(lp_end, end - lp_end, prot,
3637 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3638 -1, 0);
3639 if (result == MAP_FAILED) {
3640 ::munmap(start, lp_end - start);
3641 return NULL;
3642 }
3643 }
3644
3645 return start;
3646 }
3647
3648 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3649 size_t alignment,
3650 char* req_addr,
3651 bool exec) {
3652 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3653 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3654 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3655 assert(is_power_of_2(os::large_page_size()), "Must be");
3656 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3657
3658 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3659 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3660 } else {
3661 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3662 }
3663 }
3664
3665 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3666 char* req_addr, bool exec) {
3667 assert(UseLargePages, "only for large pages");
3668
3669 char* addr;
3670 if (UseSHM) {
3671 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3672 } else {
3673 assert(UseHugeTLBFS, "must be");
3674 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3675 }
3676
3677 if (addr != NULL) {
3678 if (UseNUMAInterleaving) {
3679 numa_make_global(addr, bytes);
3680 }
3681
3682 // The memory is committed
3683 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3684 }
3685
3686 return addr;
3687 }
3688
3689 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3690 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3691 return shmdt(base) == 0;
3692 }
3693
3694 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3695 return pd_release_memory(base, bytes);
3696 }
3697
3698 bool os::release_memory_special(char* base, size_t bytes) {
3699 bool res;
3700 if (MemTracker::tracking_level() > NMT_minimal) {
3701 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3702 res = os::Linux::release_memory_special_impl(base, bytes);
3703 if (res) {
3704 tkr.record((address)base, bytes);
3705 }
3706
3707 } else {
3708 res = os::Linux::release_memory_special_impl(base, bytes);
3709 }
3710 return res;
3711 }
3712
3713 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3714 assert(UseLargePages, "only for large pages");
3715 bool res;
3716
3717 if (UseSHM) {
3718 res = os::Linux::release_memory_special_shm(base, bytes);
3719 } else {
3720 assert(UseHugeTLBFS, "must be");
3721 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3722 }
3723 return res;
3724 }
3725
3726 size_t os::large_page_size() {
3727 return _large_page_size;
3728 }
3729
3730 // With SysV SHM the entire memory region must be allocated as shared
3731 // memory.
3732 // HugeTLBFS allows application to commit large page memory on demand.
3733 // However, when committing memory with HugeTLBFS fails, the region
3734 // that was supposed to be committed will lose the old reservation
3735 // and allow other threads to steal that memory region. Because of this
3736 // behavior we can't commit HugeTLBFS memory.
3737 bool os::can_commit_large_page_memory() {
3738 return UseTransparentHugePages;
3739 }
3740
3741 bool os::can_execute_large_page_memory() {
3742 return UseTransparentHugePages || UseHugeTLBFS;
3743 }
3744
3745 // Reserve memory at an arbitrary address, only if that area is
3746 // available (and not reserved for something else).
3747
3748 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3749 const int max_tries = 10;
3750 char* base[max_tries];
3751 size_t size[max_tries];
3752 const size_t gap = 0x000000;
3753
3754 // Assert only that the size is a multiple of the page size, since
3755 // that's all that mmap requires, and since that's all we really know
3756 // about at this low abstraction level. If we need higher alignment,
3757 // we can either pass an alignment to this method or verify alignment
3758 // in one of the methods further up the call chain. See bug 5044738.
3759 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3760
3761 // Repeatedly allocate blocks until the block is allocated at the
3762 // right spot.
3763
3764 // Linux mmap allows caller to pass an address as hint; give it a try first,
3765 // if kernel honors the hint then we can return immediately.
3766 char * addr = anon_mmap(requested_addr, bytes, false);
3767 if (addr == requested_addr) {
3768 return requested_addr;
3769 }
3770
3771 if (addr != NULL) {
3772 // mmap() is successful but it fails to reserve at the requested address
3773 anon_munmap(addr, bytes);
3774 }
3775
3776 int i;
3777 for (i = 0; i < max_tries; ++i) {
3778 base[i] = reserve_memory(bytes);
3779
3780 if (base[i] != NULL) {
3781 // Is this the block we wanted?
3782 if (base[i] == requested_addr) {
3783 size[i] = bytes;
3784 break;
3785 }
3786
3787 // Does this overlap the block we wanted? Give back the overlapped
3788 // parts and try again.
3789
3790 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3791 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3792 unmap_memory(base[i], top_overlap);
3793 base[i] += top_overlap;
3794 size[i] = bytes - top_overlap;
3795 } else {
3796 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3797 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3798 unmap_memory(requested_addr, bottom_overlap);
3799 size[i] = bytes - bottom_overlap;
3800 } else {
3801 size[i] = bytes;
3802 }
3803 }
3804 }
3805 }
3806
3807 // Give back the unused reserved pieces.
3808
3809 for (int j = 0; j < i; ++j) {
3810 if (base[j] != NULL) {
3811 unmap_memory(base[j], size[j]);
3812 }
3813 }
3814
3815 if (i < max_tries) {
3816 return requested_addr;
3817 } else {
3818 return NULL;
3819 }
3820 }
3821
3822 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3823 return ::read(fd, buf, nBytes);
3824 }
3825
3826 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3827 return ::pread(fd, buf, nBytes, offset);
3828 }
3829
3830 // Short sleep, direct OS call.
3831 //
3832 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3833 // sched_yield(2) will actually give up the CPU:
3834 //
3835 // * Alone on this pariticular CPU, keeps running.
3836 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3837 // (pre 2.6.39).
3838 //
3839 // So calling this with 0 is an alternative.
3840 //
3841 void os::naked_short_sleep(jlong ms) {
3842 struct timespec req;
3843
3844 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3845 req.tv_sec = 0;
3846 if (ms > 0) {
3847 req.tv_nsec = (ms % 1000) * 1000000;
3848 } else {
3849 req.tv_nsec = 1;
3850 }
3851
3852 nanosleep(&req, NULL);
3853
3854 return;
3855 }
3856
3857 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3858 void os::infinite_sleep() {
3859 while (true) { // sleep forever ...
3860 ::sleep(100); // ... 100 seconds at a time
3861 }
3862 }
3863
3864 // Used to convert frequent JVM_Yield() to nops
3865 bool os::dont_yield() {
3866 return DontYieldALot;
3867 }
3868
3869 void os::naked_yield() {
3870 sched_yield();
3871 }
3872
3873 ////////////////////////////////////////////////////////////////////////////////
3874 // thread priority support
3875
3876 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3877 // only supports dynamic priority, static priority must be zero. For real-time
3878 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3879 // However, for large multi-threaded applications, SCHED_RR is not only slower
3880 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3881 // of 5 runs - Sep 2005).
3882 //
3883 // The following code actually changes the niceness of kernel-thread/LWP. It
3884 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3885 // not the entire user process, and user level threads are 1:1 mapped to kernel
3886 // threads. It has always been the case, but could change in the future. For
3887 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3888 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3889
3890 int os::java_to_os_priority[CriticalPriority + 1] = {
3891 19, // 0 Entry should never be used
3892
3893 4, // 1 MinPriority
3894 3, // 2
3895 2, // 3
3896
3897 1, // 4
3898 0, // 5 NormPriority
3899 -1, // 6
3900
3901 -2, // 7
3902 -3, // 8
3903 -4, // 9 NearMaxPriority
3904
3905 -5, // 10 MaxPriority
3906
3907 -5 // 11 CriticalPriority
3908 };
3909
3910 static int prio_init() {
3911 if (ThreadPriorityPolicy == 1) {
3912 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3913 // if effective uid is not root. Perhaps, a more elegant way of doing
3914 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3915 if (geteuid() != 0) {
3916 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3917 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3918 }
3919 ThreadPriorityPolicy = 0;
3920 }
3921 }
3922 if (UseCriticalJavaThreadPriority) {
3923 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3924 }
3925 return 0;
3926 }
3927
3928 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3929 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3930
3931 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3932 return (ret == 0) ? OS_OK : OS_ERR;
3933 }
3934
3935 OSReturn os::get_native_priority(const Thread* const thread,
3936 int *priority_ptr) {
3937 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3938 *priority_ptr = java_to_os_priority[NormPriority];
3939 return OS_OK;
3940 }
3941
3942 errno = 0;
3943 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3944 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3945 }
3946
3947 // Hint to the underlying OS that a task switch would not be good.
3948 // Void return because it's a hint and can fail.
3949 void os::hint_no_preempt() {}
3950
3951 ////////////////////////////////////////////////////////////////////////////////
3952 // suspend/resume support
3953
3954 // the low-level signal-based suspend/resume support is a remnant from the
3955 // old VM-suspension that used to be for java-suspension, safepoints etc,
3956 // within hotspot. Now there is a single use-case for this:
3957 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3958 // that runs in the watcher thread.
3959 // The remaining code is greatly simplified from the more general suspension
3960 // code that used to be used.
3961 //
3962 // The protocol is quite simple:
3963 // - suspend:
3964 // - sends a signal to the target thread
3965 // - polls the suspend state of the osthread using a yield loop
3966 // - target thread signal handler (SR_handler) sets suspend state
3967 // and blocks in sigsuspend until continued
3968 // - resume:
3969 // - sets target osthread state to continue
3970 // - sends signal to end the sigsuspend loop in the SR_handler
3971 //
3972 // Note that the SR_lock plays no role in this suspend/resume protocol,
3973 // but is checked for NULL in SR_handler as a thread termination indicator.
3974
3975 static void resume_clear_context(OSThread *osthread) {
3976 osthread->set_ucontext(NULL);
3977 osthread->set_siginfo(NULL);
3978 }
3979
3980 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3981 ucontext_t* context) {
3982 osthread->set_ucontext(context);
3983 osthread->set_siginfo(siginfo);
3984 }
3985
3986 // Handler function invoked when a thread's execution is suspended or
3987 // resumed. We have to be careful that only async-safe functions are
3988 // called here (Note: most pthread functions are not async safe and
3989 // should be avoided.)
3990 //
3991 // Note: sigwait() is a more natural fit than sigsuspend() from an
3992 // interface point of view, but sigwait() prevents the signal hander
3993 // from being run. libpthread would get very confused by not having
3994 // its signal handlers run and prevents sigwait()'s use with the
3995 // mutex granting granting signal.
3996 //
3997 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3998 //
3999 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4000 // Save and restore errno to avoid confusing native code with EINTR
4001 // after sigsuspend.
4002 int old_errno = errno;
4003
4004 Thread* thread = Thread::current_or_null_safe();
4005 assert(thread != NULL, "Missing current thread in SR_handler");
4006
4007 // On some systems we have seen signal delivery get "stuck" until the signal
4008 // mask is changed as part of thread termination. Check that the current thread
4009 // has not already terminated (via SR_lock()) - else the following assertion
4010 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4011 // destructor has completed.
4012
4013 if (thread->SR_lock() == NULL) {
4014 return;
4015 }
4016
4017 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4018
4019 OSThread* osthread = thread->osthread();
4020
4021 os::SuspendResume::State current = osthread->sr.state();
4022 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4023 suspend_save_context(osthread, siginfo, context);
4024
4025 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4026 os::SuspendResume::State state = osthread->sr.suspended();
4027 if (state == os::SuspendResume::SR_SUSPENDED) {
4028 sigset_t suspend_set; // signals for sigsuspend()
4029 sigemptyset(&suspend_set);
4030 // get current set of blocked signals and unblock resume signal
4031 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4032 sigdelset(&suspend_set, SR_signum);
4033
4034 sr_semaphore.signal();
4035 // wait here until we are resumed
4036 while (1) {
4037 sigsuspend(&suspend_set);
4038
4039 os::SuspendResume::State result = osthread->sr.running();
4040 if (result == os::SuspendResume::SR_RUNNING) {
4041 sr_semaphore.signal();
4042 break;
4043 }
4044 }
4045
4046 } else if (state == os::SuspendResume::SR_RUNNING) {
4047 // request was cancelled, continue
4048 } else {
4049 ShouldNotReachHere();
4050 }
4051
4052 resume_clear_context(osthread);
4053 } else if (current == os::SuspendResume::SR_RUNNING) {
4054 // request was cancelled, continue
4055 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4056 // ignore
4057 } else {
4058 // ignore
4059 }
4060
4061 errno = old_errno;
4062 }
4063
4064 static int SR_initialize() {
4065 struct sigaction act;
4066 char *s;
4067
4068 // Get signal number to use for suspend/resume
4069 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4070 int sig = ::strtol(s, 0, 10);
4071 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4072 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4073 SR_signum = sig;
4074 } else {
4075 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4076 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4077 }
4078 }
4079
4080 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4081 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4082
4083 sigemptyset(&SR_sigset);
4084 sigaddset(&SR_sigset, SR_signum);
4085
4086 // Set up signal handler for suspend/resume
4087 act.sa_flags = SA_RESTART|SA_SIGINFO;
4088 act.sa_handler = (void (*)(int)) SR_handler;
4089
4090 // SR_signum is blocked by default.
4091 // 4528190 - We also need to block pthread restart signal (32 on all
4092 // supported Linux platforms). Note that LinuxThreads need to block
4093 // this signal for all threads to work properly. So we don't have
4094 // to use hard-coded signal number when setting up the mask.
4095 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4096
4097 if (sigaction(SR_signum, &act, 0) == -1) {
4098 return -1;
4099 }
4100
4101 // Save signal flag
4102 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4103 return 0;
4104 }
4105
4106 static int sr_notify(OSThread* osthread) {
4107 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4108 assert_status(status == 0, status, "pthread_kill");
4109 return status;
4110 }
4111
4112 // "Randomly" selected value for how long we want to spin
4113 // before bailing out on suspending a thread, also how often
4114 // we send a signal to a thread we want to resume
4115 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4116 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4117
4118 // returns true on success and false on error - really an error is fatal
4119 // but this seems the normal response to library errors
4120 static bool do_suspend(OSThread* osthread) {
4121 assert(osthread->sr.is_running(), "thread should be running");
4122 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4123
4124 // mark as suspended and send signal
4125 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4126 // failed to switch, state wasn't running?
4127 ShouldNotReachHere();
4128 return false;
4129 }
4130
4131 if (sr_notify(osthread) != 0) {
4132 ShouldNotReachHere();
4133 }
4134
4135 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4136 while (true) {
4137 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4138 break;
4139 } else {
4140 // timeout
4141 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4142 if (cancelled == os::SuspendResume::SR_RUNNING) {
4143 return false;
4144 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4145 // make sure that we consume the signal on the semaphore as well
4146 sr_semaphore.wait();
4147 break;
4148 } else {
4149 ShouldNotReachHere();
4150 return false;
4151 }
4152 }
4153 }
4154
4155 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4156 return true;
4157 }
4158
4159 static void do_resume(OSThread* osthread) {
4160 assert(osthread->sr.is_suspended(), "thread should be suspended");
4161 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4162
4163 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4164 // failed to switch to WAKEUP_REQUEST
4165 ShouldNotReachHere();
4166 return;
4167 }
4168
4169 while (true) {
4170 if (sr_notify(osthread) == 0) {
4171 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4172 if (osthread->sr.is_running()) {
4173 return;
4174 }
4175 }
4176 } else {
4177 ShouldNotReachHere();
4178 }
4179 }
4180
4181 guarantee(osthread->sr.is_running(), "Must be running!");
4182 }
4183
4184 ///////////////////////////////////////////////////////////////////////////////////
4185 // signal handling (except suspend/resume)
4186
4187 // This routine may be used by user applications as a "hook" to catch signals.
4188 // The user-defined signal handler must pass unrecognized signals to this
4189 // routine, and if it returns true (non-zero), then the signal handler must
4190 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4191 // routine will never retun false (zero), but instead will execute a VM panic
4192 // routine kill the process.
4193 //
4194 // If this routine returns false, it is OK to call it again. This allows
4195 // the user-defined signal handler to perform checks either before or after
4196 // the VM performs its own checks. Naturally, the user code would be making
4197 // a serious error if it tried to handle an exception (such as a null check
4198 // or breakpoint) that the VM was generating for its own correct operation.
4199 //
4200 // This routine may recognize any of the following kinds of signals:
4201 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4202 // It should be consulted by handlers for any of those signals.
4203 //
4204 // The caller of this routine must pass in the three arguments supplied
4205 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4206 // field of the structure passed to sigaction(). This routine assumes that
4207 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4208 //
4209 // Note that the VM will print warnings if it detects conflicting signal
4210 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4211 //
4212 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4213 siginfo_t* siginfo,
4214 void* ucontext,
4215 int abort_if_unrecognized);
4216
4217 void signalHandler(int sig, siginfo_t* info, void* uc) {
4218 assert(info != NULL && uc != NULL, "it must be old kernel");
4219 int orig_errno = errno; // Preserve errno value over signal handler.
4220 JVM_handle_linux_signal(sig, info, uc, true);
4221 errno = orig_errno;
4222 }
4223
4224
4225 // This boolean allows users to forward their own non-matching signals
4226 // to JVM_handle_linux_signal, harmlessly.
4227 bool os::Linux::signal_handlers_are_installed = false;
4228
4229 // For signal-chaining
4230 struct sigaction sigact[NSIG];
4231 uint64_t sigs = 0;
4232 #if (64 < NSIG-1)
4233 #error "Not all signals can be encoded in sigs. Adapt its type!"
4234 #endif
4235 bool os::Linux::libjsig_is_loaded = false;
4236 typedef struct sigaction *(*get_signal_t)(int);
4237 get_signal_t os::Linux::get_signal_action = NULL;
4238
4239 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4240 struct sigaction *actp = NULL;
4241
4242 if (libjsig_is_loaded) {
4243 // Retrieve the old signal handler from libjsig
4244 actp = (*get_signal_action)(sig);
4245 }
4246 if (actp == NULL) {
4247 // Retrieve the preinstalled signal handler from jvm
4248 actp = get_preinstalled_handler(sig);
4249 }
4250
4251 return actp;
4252 }
4253
4254 static bool call_chained_handler(struct sigaction *actp, int sig,
4255 siginfo_t *siginfo, void *context) {
4256 // Call the old signal handler
4257 if (actp->sa_handler == SIG_DFL) {
4258 // It's more reasonable to let jvm treat it as an unexpected exception
4259 // instead of taking the default action.
4260 return false;
4261 } else if (actp->sa_handler != SIG_IGN) {
4262 if ((actp->sa_flags & SA_NODEFER) == 0) {
4263 // automaticlly block the signal
4264 sigaddset(&(actp->sa_mask), sig);
4265 }
4266
4267 sa_handler_t hand = NULL;
4268 sa_sigaction_t sa = NULL;
4269 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4270 // retrieve the chained handler
4271 if (siginfo_flag_set) {
4272 sa = actp->sa_sigaction;
4273 } else {
4274 hand = actp->sa_handler;
4275 }
4276
4277 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4278 actp->sa_handler = SIG_DFL;
4279 }
4280
4281 // try to honor the signal mask
4282 sigset_t oset;
4283 sigemptyset(&oset);
4284 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4285
4286 // call into the chained handler
4287 if (siginfo_flag_set) {
4288 (*sa)(sig, siginfo, context);
4289 } else {
4290 (*hand)(sig);
4291 }
4292
4293 // restore the signal mask
4294 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4295 }
4296 // Tell jvm's signal handler the signal is taken care of.
4297 return true;
4298 }
4299
4300 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4301 bool chained = false;
4302 // signal-chaining
4303 if (UseSignalChaining) {
4304 struct sigaction *actp = get_chained_signal_action(sig);
4305 if (actp != NULL) {
4306 chained = call_chained_handler(actp, sig, siginfo, context);
4307 }
4308 }
4309 return chained;
4310 }
4311
4312 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4313 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4314 return &sigact[sig];
4315 }
4316 return NULL;
4317 }
4318
4319 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4320 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4321 sigact[sig] = oldAct;
4322 sigs |= (uint64_t)1 << (sig-1);
4323 }
4324
4325 // for diagnostic
4326 int sigflags[NSIG];
4327
4328 int os::Linux::get_our_sigflags(int sig) {
4329 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4330 return sigflags[sig];
4331 }
4332
4333 void os::Linux::set_our_sigflags(int sig, int flags) {
4334 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4335 if (sig > 0 && sig < NSIG) {
4336 sigflags[sig] = flags;
4337 }
4338 }
4339
4340 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4341 // Check for overwrite.
4342 struct sigaction oldAct;
4343 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4344
4345 void* oldhand = oldAct.sa_sigaction
4346 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4347 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4348 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4349 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4350 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4351 if (AllowUserSignalHandlers || !set_installed) {
4352 // Do not overwrite; user takes responsibility to forward to us.
4353 return;
4354 } else if (UseSignalChaining) {
4355 // save the old handler in jvm
4356 save_preinstalled_handler(sig, oldAct);
4357 // libjsig also interposes the sigaction() call below and saves the
4358 // old sigaction on it own.
4359 } else {
4360 fatal("Encountered unexpected pre-existing sigaction handler "
4361 "%#lx for signal %d.", (long)oldhand, sig);
4362 }
4363 }
4364
4365 struct sigaction sigAct;
4366 sigfillset(&(sigAct.sa_mask));
4367 sigAct.sa_handler = SIG_DFL;
4368 if (!set_installed) {
4369 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4370 } else {
4371 sigAct.sa_sigaction = signalHandler;
4372 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4373 }
4374 // Save flags, which are set by ours
4375 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4376 sigflags[sig] = sigAct.sa_flags;
4377
4378 int ret = sigaction(sig, &sigAct, &oldAct);
4379 assert(ret == 0, "check");
4380
4381 void* oldhand2 = oldAct.sa_sigaction
4382 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4383 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4384 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4385 }
4386
4387 // install signal handlers for signals that HotSpot needs to
4388 // handle in order to support Java-level exception handling.
4389
4390 void os::Linux::install_signal_handlers() {
4391 if (!signal_handlers_are_installed) {
4392 signal_handlers_are_installed = true;
4393
4394 // signal-chaining
4395 typedef void (*signal_setting_t)();
4396 signal_setting_t begin_signal_setting = NULL;
4397 signal_setting_t end_signal_setting = NULL;
4398 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4399 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4400 if (begin_signal_setting != NULL) {
4401 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4402 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4403 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4404 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4405 libjsig_is_loaded = true;
4406 assert(UseSignalChaining, "should enable signal-chaining");
4407 }
4408 if (libjsig_is_loaded) {
4409 // Tell libjsig jvm is setting signal handlers
4410 (*begin_signal_setting)();
4411 }
4412
4413 set_signal_handler(SIGSEGV, true);
4414 set_signal_handler(SIGPIPE, true);
4415 set_signal_handler(SIGBUS, true);
4416 set_signal_handler(SIGILL, true);
4417 set_signal_handler(SIGFPE, true);
4418 #if defined(PPC64)
4419 set_signal_handler(SIGTRAP, true);
4420 #endif
4421 set_signal_handler(SIGXFSZ, true);
4422
4423 if (libjsig_is_loaded) {
4424 // Tell libjsig jvm finishes setting signal handlers
4425 (*end_signal_setting)();
4426 }
4427
4428 // We don't activate signal checker if libjsig is in place, we trust ourselves
4429 // and if UserSignalHandler is installed all bets are off.
4430 // Log that signal checking is off only if -verbose:jni is specified.
4431 if (CheckJNICalls) {
4432 if (libjsig_is_loaded) {
4433 if (PrintJNIResolving) {
4434 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4435 }
4436 check_signals = false;
4437 }
4438 if (AllowUserSignalHandlers) {
4439 if (PrintJNIResolving) {
4440 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4441 }
4442 check_signals = false;
4443 }
4444 }
4445 }
4446 }
4447
4448 // This is the fastest way to get thread cpu time on Linux.
4449 // Returns cpu time (user+sys) for any thread, not only for current.
4450 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4451 // It might work on 2.6.10+ with a special kernel/glibc patch.
4452 // For reference, please, see IEEE Std 1003.1-2004:
4453 // http://www.unix.org/single_unix_specification
4454
4455 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4456 struct timespec tp;
4457 int rc = os::Linux::clock_gettime(clockid, &tp);
4458 assert(rc == 0, "clock_gettime is expected to return 0 code");
4459
4460 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4461 }
4462
4463 void os::Linux::initialize_os_info() {
4464 assert(_os_version == 0, "OS info already initialized");
4465
4466 struct utsname _uname;
4467
4468 uint32_t major;
4469 uint32_t minor;
4470 uint32_t fix;
4471
4472 int rc;
4473
4474 // Kernel version is unknown if
4475 // verification below fails.
4476 _os_version = 0x01000000;
4477
4478 rc = uname(&_uname);
4479 if (rc != -1) {
4480
4481 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4482 if (rc == 3) {
4483
4484 if (major < 256 && minor < 256 && fix < 256) {
4485 // Kernel version format is as expected,
4486 // set it overriding unknown state.
4487 _os_version = (major << 16) |
4488 (minor << 8 ) |
4489 (fix << 0 ) ;
4490 }
4491 }
4492 }
4493 }
4494
4495 uint32_t os::Linux::os_version() {
4496 assert(_os_version != 0, "not initialized");
4497 return _os_version & 0x00FFFFFF;
4498 }
4499
4500 bool os::Linux::os_version_is_known() {
4501 assert(_os_version != 0, "not initialized");
4502 return _os_version & 0x01000000 ? false : true;
4503 }
4504
4505 /////
4506 // glibc on Linux platform uses non-documented flag
4507 // to indicate, that some special sort of signal
4508 // trampoline is used.
4509 // We will never set this flag, and we should
4510 // ignore this flag in our diagnostic
4511 #ifdef SIGNIFICANT_SIGNAL_MASK
4512 #undef SIGNIFICANT_SIGNAL_MASK
4513 #endif
4514 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4515
4516 static const char* get_signal_handler_name(address handler,
4517 char* buf, int buflen) {
4518 int offset = 0;
4519 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4520 if (found) {
4521 // skip directory names
4522 const char *p1, *p2;
4523 p1 = buf;
4524 size_t len = strlen(os::file_separator());
4525 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4526 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4527 } else {
4528 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4529 }
4530 return buf;
4531 }
4532
4533 static void print_signal_handler(outputStream* st, int sig,
4534 char* buf, size_t buflen) {
4535 struct sigaction sa;
4536
4537 sigaction(sig, NULL, &sa);
4538
4539 // See comment for SIGNIFICANT_SIGNAL_MASK define
4540 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4541
4542 st->print("%s: ", os::exception_name(sig, buf, buflen));
4543
4544 address handler = (sa.sa_flags & SA_SIGINFO)
4545 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4546 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4547
4548 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4549 st->print("SIG_DFL");
4550 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4551 st->print("SIG_IGN");
4552 } else {
4553 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4554 }
4555
4556 st->print(", sa_mask[0]=");
4557 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4558
4559 address rh = VMError::get_resetted_sighandler(sig);
4560 // May be, handler was resetted by VMError?
4561 if (rh != NULL) {
4562 handler = rh;
4563 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4564 }
4565
4566 st->print(", sa_flags=");
4567 os::Posix::print_sa_flags(st, sa.sa_flags);
4568
4569 // Check: is it our handler?
4570 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4571 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4572 // It is our signal handler
4573 // check for flags, reset system-used one!
4574 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4575 st->print(
4576 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4577 os::Linux::get_our_sigflags(sig));
4578 }
4579 }
4580 st->cr();
4581 }
4582
4583
4584 #define DO_SIGNAL_CHECK(sig) \
4585 do { \
4586 if (!sigismember(&check_signal_done, sig)) { \
4587 os::Linux::check_signal_handler(sig); \
4588 } \
4589 } while (0)
4590
4591 // This method is a periodic task to check for misbehaving JNI applications
4592 // under CheckJNI, we can add any periodic checks here
4593
4594 void os::run_periodic_checks() {
4595 if (check_signals == false) return;
4596
4597 // SEGV and BUS if overridden could potentially prevent
4598 // generation of hs*.log in the event of a crash, debugging
4599 // such a case can be very challenging, so we absolutely
4600 // check the following for a good measure:
4601 DO_SIGNAL_CHECK(SIGSEGV);
4602 DO_SIGNAL_CHECK(SIGILL);
4603 DO_SIGNAL_CHECK(SIGFPE);
4604 DO_SIGNAL_CHECK(SIGBUS);
4605 DO_SIGNAL_CHECK(SIGPIPE);
4606 DO_SIGNAL_CHECK(SIGXFSZ);
4607 #if defined(PPC64)
4608 DO_SIGNAL_CHECK(SIGTRAP);
4609 #endif
4610
4611 // ReduceSignalUsage allows the user to override these handlers
4612 // see comments at the very top and jvm_solaris.h
4613 if (!ReduceSignalUsage) {
4614 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4615 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4616 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4617 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4618 }
4619
4620 DO_SIGNAL_CHECK(SR_signum);
4621 }
4622
4623 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4624
4625 static os_sigaction_t os_sigaction = NULL;
4626
4627 void os::Linux::check_signal_handler(int sig) {
4628 char buf[O_BUFLEN];
4629 address jvmHandler = NULL;
4630
4631
4632 struct sigaction act;
4633 if (os_sigaction == NULL) {
4634 // only trust the default sigaction, in case it has been interposed
4635 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4636 if (os_sigaction == NULL) return;
4637 }
4638
4639 os_sigaction(sig, (struct sigaction*)NULL, &act);
4640
4641
4642 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4643
4644 address thisHandler = (act.sa_flags & SA_SIGINFO)
4645 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4646 : CAST_FROM_FN_PTR(address, act.sa_handler);
4647
4648
4649 switch (sig) {
4650 case SIGSEGV:
4651 case SIGBUS:
4652 case SIGFPE:
4653 case SIGPIPE:
4654 case SIGILL:
4655 case SIGXFSZ:
4656 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4657 break;
4658
4659 case SHUTDOWN1_SIGNAL:
4660 case SHUTDOWN2_SIGNAL:
4661 case SHUTDOWN3_SIGNAL:
4662 case BREAK_SIGNAL:
4663 jvmHandler = (address)user_handler();
4664 break;
4665
4666 default:
4667 if (sig == SR_signum) {
4668 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4669 } else {
4670 return;
4671 }
4672 break;
4673 }
4674
4675 if (thisHandler != jvmHandler) {
4676 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4677 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4678 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4679 // No need to check this sig any longer
4680 sigaddset(&check_signal_done, sig);
4681 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4682 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4683 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4684 exception_name(sig, buf, O_BUFLEN));
4685 }
4686 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4687 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4688 tty->print("expected:");
4689 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4690 tty->cr();
4691 tty->print(" found:");
4692 os::Posix::print_sa_flags(tty, act.sa_flags);
4693 tty->cr();
4694 // No need to check this sig any longer
4695 sigaddset(&check_signal_done, sig);
4696 }
4697
4698 // Dump all the signal
4699 if (sigismember(&check_signal_done, sig)) {
4700 print_signal_handlers(tty, buf, O_BUFLEN);
4701 }
4702 }
4703
4704 extern void report_error(char* file_name, int line_no, char* title,
4705 char* format, ...);
4706
4707 // this is called _before_ the most of global arguments have been parsed
4708 void os::init(void) {
4709 char dummy; // used to get a guess on initial stack address
4710 // first_hrtime = gethrtime();
4711
4712 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4713
4714 init_random(1234567);
4715
4716 ThreadCritical::initialize();
4717
4718 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4719 if (Linux::page_size() == -1) {
4720 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4721 os::strerror(errno));
4722 }
4723 init_page_sizes((size_t) Linux::page_size());
4724
4725 Linux::initialize_system_info();
4726
4727 Linux::initialize_os_info();
4728
4729 // main_thread points to the aboriginal thread
4730 Linux::_main_thread = pthread_self();
4731
4732 Linux::clock_init();
4733 initial_time_count = javaTimeNanos();
4734
4735 // pthread_condattr initialization for monotonic clock
4736 int status;
4737 pthread_condattr_t* _condattr = os::Linux::condAttr();
4738 if ((status = pthread_condattr_init(_condattr)) != 0) {
4739 fatal("pthread_condattr_init: %s", os::strerror(status));
4740 }
4741 // Only set the clock if CLOCK_MONOTONIC is available
4742 if (os::supports_monotonic_clock()) {
4743 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4744 if (status == EINVAL) {
4745 warning("Unable to use monotonic clock with relative timed-waits" \
4746 " - changes to the time-of-day clock may have adverse affects");
4747 } else {
4748 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4749 }
4750 }
4751 }
4752 // else it defaults to CLOCK_REALTIME
4753
4754 // retrieve entry point for pthread_setname_np
4755 Linux::_pthread_setname_np =
4756 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4757
4758 }
4759
4760 // To install functions for atexit system call
4761 extern "C" {
4762 static void perfMemory_exit_helper() {
4763 perfMemory_exit();
4764 }
4765 }
4766
4767 // this is called _after_ the global arguments have been parsed
4768 jint os::init_2(void) {
4769 Linux::fast_thread_clock_init();
4770
4771 // Allocate a single page and mark it as readable for safepoint polling
4772 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4773 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4774
4775 os::set_polling_page(polling_page);
4776 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4777
4778 if (!UseMembar) {
4779 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4780 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4781 os::set_memory_serialize_page(mem_serialize_page);
4782 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4783 }
4784
4785 // initialize suspend/resume support - must do this before signal_sets_init()
4786 if (SR_initialize() != 0) {
4787 perror("SR_initialize failed");
4788 return JNI_ERR;
4789 }
4790
4791 Linux::signal_sets_init();
4792 Linux::install_signal_handlers();
4793
4794 // Check and sets minimum stack sizes against command line options
4795 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4796 return JNI_ERR;
4797 }
4798 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4799
4800 #if defined(IA32)
4801 workaround_expand_exec_shield_cs_limit();
4802 #endif
4803
4804 Linux::libpthread_init();
4805 log_info(os)("HotSpot is running with %s, %s",
4806 Linux::glibc_version(), Linux::libpthread_version());
4807
4808 if (UseNUMA) {
4809 if (!Linux::libnuma_init()) {
4810 UseNUMA = false;
4811 } else {
4812 if ((Linux::numa_max_node() < 1)) {
4813 // There's only one node(they start from 0), disable NUMA.
4814 UseNUMA = false;
4815 }
4816 }
4817 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4818 // we can make the adaptive lgrp chunk resizing work. If the user specified
4819 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4820 // disable adaptive resizing.
4821 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4822 if (FLAG_IS_DEFAULT(UseNUMA)) {
4823 UseNUMA = false;
4824 } else {
4825 if (FLAG_IS_DEFAULT(UseLargePages) &&
4826 FLAG_IS_DEFAULT(UseSHM) &&
4827 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4828 UseLargePages = false;
4829 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4830 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4831 UseAdaptiveSizePolicy = false;
4832 UseAdaptiveNUMAChunkSizing = false;
4833 }
4834 }
4835 }
4836 if (!UseNUMA && ForceNUMA) {
4837 UseNUMA = true;
4838 }
4839 }
4840
4841 if (MaxFDLimit) {
4842 // set the number of file descriptors to max. print out error
4843 // if getrlimit/setrlimit fails but continue regardless.
4844 struct rlimit nbr_files;
4845 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4846 if (status != 0) {
4847 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4848 } else {
4849 nbr_files.rlim_cur = nbr_files.rlim_max;
4850 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4851 if (status != 0) {
4852 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4853 }
4854 }
4855 }
4856
4857 // Initialize lock used to serialize thread creation (see os::create_thread)
4858 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4859
4860 // at-exit methods are called in the reverse order of their registration.
4861 // atexit functions are called on return from main or as a result of a
4862 // call to exit(3C). There can be only 32 of these functions registered
4863 // and atexit() does not set errno.
4864
4865 if (PerfAllowAtExitRegistration) {
4866 // only register atexit functions if PerfAllowAtExitRegistration is set.
4867 // atexit functions can be delayed until process exit time, which
4868 // can be problematic for embedded VM situations. Embedded VMs should
4869 // call DestroyJavaVM() to assure that VM resources are released.
4870
4871 // note: perfMemory_exit_helper atexit function may be removed in
4872 // the future if the appropriate cleanup code can be added to the
4873 // VM_Exit VMOperation's doit method.
4874 if (atexit(perfMemory_exit_helper) != 0) {
4875 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4876 }
4877 }
4878
4879 // initialize thread priority policy
4880 prio_init();
4881
4882 return JNI_OK;
4883 }
4884
4885 // Mark the polling page as unreadable
4886 void os::make_polling_page_unreadable(void) {
4887 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4888 fatal("Could not disable polling page");
4889 }
4890 }
4891
4892 // Mark the polling page as readable
4893 void os::make_polling_page_readable(void) {
4894 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4895 fatal("Could not enable polling page");
4896 }
4897 }
4898
4899 // older glibc versions don't have this macro (which expands to
4900 // an optimized bit-counting function) so we have to roll our own
4901 #ifndef CPU_COUNT
4902
4903 static int _cpu_count(const cpu_set_t* cpus) {
4904 int count = 0;
4905 // only look up to the number of configured processors
4906 for (int i = 0; i < os::processor_count(); i++) {
4907 if (CPU_ISSET(i, cpus)) {
4908 count++;
4909 }
4910 }
4911 return count;
4912 }
4913
4914 #define CPU_COUNT(cpus) _cpu_count(cpus)
4915
4916 #endif // CPU_COUNT
4917
4918 // Get the current number of available processors for this process.
4919 // This value can change at any time during a process's lifetime.
4920 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4921 // If it appears there may be more than 1024 processors then we do a
4922 // dynamic check - see 6515172 for details.
4923 // If anything goes wrong we fallback to returning the number of online
4924 // processors - which can be greater than the number available to the process.
4925 int os::active_processor_count() {
4926 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4927 cpu_set_t* cpus_p = &cpus;
4928 int cpus_size = sizeof(cpu_set_t);
4929
4930 int configured_cpus = processor_count(); // upper bound on available cpus
4931 int cpu_count = 0;
4932
4933 // old build platforms may not support dynamic cpu sets
4934 #ifdef CPU_ALLOC
4935
4936 // To enable easy testing of the dynamic path on different platforms we
4937 // introduce a diagnostic flag: UseCpuAllocPath
4938 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4939 // kernel may use a mask bigger than cpu_set_t
4940 log_trace(os)("active_processor_count: using dynamic path %s"
4941 "- configured processors: %d",
4942 UseCpuAllocPath ? "(forced) " : "",
4943 configured_cpus);
4944 cpus_p = CPU_ALLOC(configured_cpus);
4945 if (cpus_p != NULL) {
4946 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4947 // zero it just to be safe
4948 CPU_ZERO_S(cpus_size, cpus_p);
4949 }
4950 else {
4951 // failed to allocate so fallback to online cpus
4952 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4953 log_trace(os)("active_processor_count: "
4954 "CPU_ALLOC failed (%s) - using "
4955 "online processor count: %d",
4956 os::strerror(errno), online_cpus);
4957 return online_cpus;
4958 }
4959 }
4960 else {
4961 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4962 configured_cpus);
4963 }
4964 #else // CPU_ALLOC
4965 // these stubs won't be executed
4966 #define CPU_COUNT_S(size, cpus) -1
4967 #define CPU_FREE(cpus)
4968
4969 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4970 configured_cpus);
4971 #endif // CPU_ALLOC
4972
4973 // pid 0 means the current thread - which we have to assume represents the process
4974 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4975 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4976 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4977 }
4978 else {
4979 cpu_count = CPU_COUNT(cpus_p);
4980 }
4981 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4982 }
4983 else {
4984 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4985 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4986 "which may exceed available processors", os::strerror(errno), cpu_count);
4987 }
4988
4989 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4990 CPU_FREE(cpus_p);
4991 }
4992
4993 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4994 return cpu_count;
4995 }
4996
4997 void os::set_native_thread_name(const char *name) {
4998 if (Linux::_pthread_setname_np) {
4999 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5000 snprintf(buf, sizeof(buf), "%s", name);
5001 buf[sizeof(buf) - 1] = '\0';
5002 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5003 // ERANGE should not happen; all other errors should just be ignored.
5004 assert(rc != ERANGE, "pthread_setname_np failed");
5005 }
5006 }
5007
5008 bool os::distribute_processes(uint length, uint* distribution) {
5009 // Not yet implemented.
5010 return false;
5011 }
5012
5013 bool os::bind_to_processor(uint processor_id) {
5014 // Not yet implemented.
5015 return false;
5016 }
5017
5018 ///
5019
5020 void os::SuspendedThreadTask::internal_do_task() {
5021 if (do_suspend(_thread->osthread())) {
5022 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5023 do_task(context);
5024 do_resume(_thread->osthread());
5025 }
5026 }
5027
5028 class PcFetcher : public os::SuspendedThreadTask {
5029 public:
5030 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5031 ExtendedPC result();
5032 protected:
5033 void do_task(const os::SuspendedThreadTaskContext& context);
5034 private:
5035 ExtendedPC _epc;
5036 };
5037
5038 ExtendedPC PcFetcher::result() {
5039 guarantee(is_done(), "task is not done yet.");
5040 return _epc;
5041 }
5042
5043 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5044 Thread* thread = context.thread();
5045 OSThread* osthread = thread->osthread();
5046 if (osthread->ucontext() != NULL) {
5047 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5048 } else {
5049 // NULL context is unexpected, double-check this is the VMThread
5050 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5051 }
5052 }
5053
5054 // Suspends the target using the signal mechanism and then grabs the PC before
5055 // resuming the target. Used by the flat-profiler only
5056 ExtendedPC os::get_thread_pc(Thread* thread) {
5057 // Make sure that it is called by the watcher for the VMThread
5058 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5059 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5060
5061 PcFetcher fetcher(thread);
5062 fetcher.run();
5063 return fetcher.result();
5064 }
5065
5066 ////////////////////////////////////////////////////////////////////////////////
5067 // debug support
5068
5069 bool os::find(address addr, outputStream* st) {
5070 Dl_info dlinfo;
5071 memset(&dlinfo, 0, sizeof(dlinfo));
5072 if (dladdr(addr, &dlinfo) != 0) {
5073 st->print(PTR_FORMAT ": ", p2i(addr));
5074 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5075 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5076 p2i(addr) - p2i(dlinfo.dli_saddr));
5077 } else if (dlinfo.dli_fbase != NULL) {
5078 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5079 } else {
5080 st->print("<absolute address>");
5081 }
5082 if (dlinfo.dli_fname != NULL) {
5083 st->print(" in %s", dlinfo.dli_fname);
5084 }
5085 if (dlinfo.dli_fbase != NULL) {
5086 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5087 }
5088 st->cr();
5089
5090 if (Verbose) {
5091 // decode some bytes around the PC
5092 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5093 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5094 address lowest = (address) dlinfo.dli_sname;
5095 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5096 if (begin < lowest) begin = lowest;
5097 Dl_info dlinfo2;
5098 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5099 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5100 end = (address) dlinfo2.dli_saddr;
5101 }
5102 Disassembler::decode(begin, end, st);
5103 }
5104 return true;
5105 }
5106 return false;
5107 }
5108
5109 ////////////////////////////////////////////////////////////////////////////////
5110 // misc
5111
5112 // This does not do anything on Linux. This is basically a hook for being
5113 // able to use structured exception handling (thread-local exception filters)
5114 // on, e.g., Win32.
5115 void
5116 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5117 JavaCallArguments* args, Thread* thread) {
5118 f(value, method, args, thread);
5119 }
5120
5121 void os::print_statistics() {
5122 }
5123
5124 bool os::message_box(const char* title, const char* message) {
5125 int i;
5126 fdStream err(defaultStream::error_fd());
5127 for (i = 0; i < 78; i++) err.print_raw("=");
5128 err.cr();
5129 err.print_raw_cr(title);
5130 for (i = 0; i < 78; i++) err.print_raw("-");
5131 err.cr();
5132 err.print_raw_cr(message);
5133 for (i = 0; i < 78; i++) err.print_raw("=");
5134 err.cr();
5135
5136 char buf[16];
5137 // Prevent process from exiting upon "read error" without consuming all CPU
5138 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5139
5140 return buf[0] == 'y' || buf[0] == 'Y';
5141 }
5142
5143 int os::stat(const char *path, struct stat *sbuf) {
5144 char pathbuf[MAX_PATH];
5145 if (strlen(path) > MAX_PATH - 1) {
5146 errno = ENAMETOOLONG;
5147 return -1;
5148 }
5149 os::native_path(strcpy(pathbuf, path));
5150 return ::stat(pathbuf, sbuf);
5151 }
5152
5153 // Is a (classpath) directory empty?
5154 bool os::dir_is_empty(const char* path) {
5155 DIR *dir = NULL;
5156 struct dirent *ptr;
5157
5158 dir = opendir(path);
5159 if (dir == NULL) return true;
5160
5161 // Scan the directory
5162 bool result = true;
5163 char buf[sizeof(struct dirent) + MAX_PATH];
5164 while (result && (ptr = ::readdir(dir)) != NULL) {
5165 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5166 result = false;
5167 }
5168 }
5169 closedir(dir);
5170 return result;
5171 }
5172
5173 // This code originates from JDK's sysOpen and open64_w
5174 // from src/solaris/hpi/src/system_md.c
5175
5176 int os::open(const char *path, int oflag, int mode) {
5177 if (strlen(path) > MAX_PATH - 1) {
5178 errno = ENAMETOOLONG;
5179 return -1;
5180 }
5181
5182 // All file descriptors that are opened in the Java process and not
5183 // specifically destined for a subprocess should have the close-on-exec
5184 // flag set. If we don't set it, then careless 3rd party native code
5185 // might fork and exec without closing all appropriate file descriptors
5186 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5187 // turn might:
5188 //
5189 // - cause end-of-file to fail to be detected on some file
5190 // descriptors, resulting in mysterious hangs, or
5191 //
5192 // - might cause an fopen in the subprocess to fail on a system
5193 // suffering from bug 1085341.
5194 //
5195 // (Yes, the default setting of the close-on-exec flag is a Unix
5196 // design flaw)
5197 //
5198 // See:
5199 // 1085341: 32-bit stdio routines should support file descriptors >255
5200 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5201 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5202 //
5203 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5204 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5205 // because it saves a system call and removes a small window where the flag
5206 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5207 // and we fall back to using FD_CLOEXEC (see below).
5208 #ifdef O_CLOEXEC
5209 oflag |= O_CLOEXEC;
5210 #endif
5211
5212 int fd = ::open64(path, oflag, mode);
5213 if (fd == -1) return -1;
5214
5215 //If the open succeeded, the file might still be a directory
5216 {
5217 struct stat64 buf64;
5218 int ret = ::fstat64(fd, &buf64);
5219 int st_mode = buf64.st_mode;
5220
5221 if (ret != -1) {
5222 if ((st_mode & S_IFMT) == S_IFDIR) {
5223 errno = EISDIR;
5224 ::close(fd);
5225 return -1;
5226 }
5227 } else {
5228 ::close(fd);
5229 return -1;
5230 }
5231 }
5232
5233 #ifdef FD_CLOEXEC
5234 // Validate that the use of the O_CLOEXEC flag on open above worked.
5235 // With recent kernels, we will perform this check exactly once.
5236 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5237 if (!O_CLOEXEC_is_known_to_work) {
5238 int flags = ::fcntl(fd, F_GETFD);
5239 if (flags != -1) {
5240 if ((flags & FD_CLOEXEC) != 0)
5241 O_CLOEXEC_is_known_to_work = 1;
5242 else
5243 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5244 }
5245 }
5246 #endif
5247
5248 return fd;
5249 }
5250
5251
5252 // create binary file, rewriting existing file if required
5253 int os::create_binary_file(const char* path, bool rewrite_existing) {
5254 int oflags = O_WRONLY | O_CREAT;
5255 if (!rewrite_existing) {
5256 oflags |= O_EXCL;
5257 }
5258 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5259 }
5260
5261 // return current position of file pointer
5262 jlong os::current_file_offset(int fd) {
5263 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5264 }
5265
5266 // move file pointer to the specified offset
5267 jlong os::seek_to_file_offset(int fd, jlong offset) {
5268 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5269 }
5270
5271 // This code originates from JDK's sysAvailable
5272 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5273
5274 int os::available(int fd, jlong *bytes) {
5275 jlong cur, end;
5276 int mode;
5277 struct stat64 buf64;
5278
5279 if (::fstat64(fd, &buf64) >= 0) {
5280 mode = buf64.st_mode;
5281 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5282 int n;
5283 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5284 *bytes = n;
5285 return 1;
5286 }
5287 }
5288 }
5289 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5290 return 0;
5291 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5292 return 0;
5293 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5294 return 0;
5295 }
5296 *bytes = end - cur;
5297 return 1;
5298 }
5299
5300 // Map a block of memory.
5301 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5302 char *addr, size_t bytes, bool read_only,
5303 bool allow_exec) {
5304 int prot;
5305 int flags = MAP_PRIVATE;
5306
5307 if (read_only) {
5308 prot = PROT_READ;
5309 } else {
5310 prot = PROT_READ | PROT_WRITE;
5311 }
5312
5313 if (allow_exec) {
5314 prot |= PROT_EXEC;
5315 }
5316
5317 if (addr != NULL) {
5318 flags |= MAP_FIXED;
5319 }
5320
5321 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5322 fd, file_offset);
5323 if (mapped_address == MAP_FAILED) {
5324 return NULL;
5325 }
5326 return mapped_address;
5327 }
5328
5329
5330 // Remap a block of memory.
5331 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5332 char *addr, size_t bytes, bool read_only,
5333 bool allow_exec) {
5334 // same as map_memory() on this OS
5335 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5336 allow_exec);
5337 }
5338
5339
5340 // Unmap a block of memory.
5341 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5342 return munmap(addr, bytes) == 0;
5343 }
5344
5345 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5346
5347 static clockid_t thread_cpu_clockid(Thread* thread) {
5348 pthread_t tid = thread->osthread()->pthread_id();
5349 clockid_t clockid;
5350
5351 // Get thread clockid
5352 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5353 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5354 return clockid;
5355 }
5356
5357 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5358 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5359 // of a thread.
5360 //
5361 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5362 // the fast estimate available on the platform.
5363
5364 jlong os::current_thread_cpu_time() {
5365 if (os::Linux::supports_fast_thread_cpu_time()) {
5366 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5367 } else {
5368 // return user + sys since the cost is the same
5369 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5370 }
5371 }
5372
5373 jlong os::thread_cpu_time(Thread* thread) {
5374 // consistent with what current_thread_cpu_time() returns
5375 if (os::Linux::supports_fast_thread_cpu_time()) {
5376 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5377 } else {
5378 return slow_thread_cpu_time(thread, true /* user + sys */);
5379 }
5380 }
5381
5382 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5383 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5384 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5385 } else {
5386 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5387 }
5388 }
5389
5390 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5391 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5392 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5393 } else {
5394 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5395 }
5396 }
5397
5398 // -1 on error.
5399 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5400 pid_t tid = thread->osthread()->thread_id();
5401 char *s;
5402 char stat[2048];
5403 int statlen;
5404 char proc_name[64];
5405 int count;
5406 long sys_time, user_time;
5407 char cdummy;
5408 int idummy;
5409 long ldummy;
5410 FILE *fp;
5411
5412 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5413 fp = fopen(proc_name, "r");
5414 if (fp == NULL) return -1;
5415 statlen = fread(stat, 1, 2047, fp);
5416 stat[statlen] = '\0';
5417 fclose(fp);
5418
5419 // Skip pid and the command string. Note that we could be dealing with
5420 // weird command names, e.g. user could decide to rename java launcher
5421 // to "java 1.4.2 :)", then the stat file would look like
5422 // 1234 (java 1.4.2 :)) R ... ...
5423 // We don't really need to know the command string, just find the last
5424 // occurrence of ")" and then start parsing from there. See bug 4726580.
5425 s = strrchr(stat, ')');
5426 if (s == NULL) return -1;
5427
5428 // Skip blank chars
5429 do { s++; } while (s && isspace(*s));
5430
5431 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5432 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5433 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5434 &user_time, &sys_time);
5435 if (count != 13) return -1;
5436 if (user_sys_cpu_time) {
5437 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5438 } else {
5439 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5440 }
5441 }
5442
5443 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5444 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5445 info_ptr->may_skip_backward = false; // elapsed time not wall time
5446 info_ptr->may_skip_forward = false; // elapsed time not wall time
5447 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5448 }
5449
5450 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5451 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5452 info_ptr->may_skip_backward = false; // elapsed time not wall time
5453 info_ptr->may_skip_forward = false; // elapsed time not wall time
5454 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5455 }
5456
5457 bool os::is_thread_cpu_time_supported() {
5458 return true;
5459 }
5460
5461 // System loadavg support. Returns -1 if load average cannot be obtained.
5462 // Linux doesn't yet have a (official) notion of processor sets,
5463 // so just return the system wide load average.
5464 int os::loadavg(double loadavg[], int nelem) {
5465 return ::getloadavg(loadavg, nelem);
5466 }
5467
5468 void os::pause() {
5469 char filename[MAX_PATH];
5470 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5471 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5472 } else {
5473 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5474 }
5475
5476 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5477 if (fd != -1) {
5478 struct stat buf;
5479 ::close(fd);
5480 while (::stat(filename, &buf) == 0) {
5481 (void)::poll(NULL, 0, 100);
5482 }
5483 } else {
5484 jio_fprintf(stderr,
5485 "Could not open pause file '%s', continuing immediately.\n", filename);
5486 }
5487 }
5488
5489
5490 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5491 // comment paragraphs are worth repeating here:
5492 //
5493 // Assumption:
5494 // Only one parker can exist on an event, which is why we allocate
5495 // them per-thread. Multiple unparkers can coexist.
5496 //
5497 // _Event serves as a restricted-range semaphore.
5498 // -1 : thread is blocked, i.e. there is a waiter
5499 // 0 : neutral: thread is running or ready,
5500 // could have been signaled after a wait started
5501 // 1 : signaled - thread is running or ready
5502 //
5503
5504 // utility to compute the abstime argument to timedwait:
5505 // millis is the relative timeout time
5506 // abstime will be the absolute timeout time
5507 // TODO: replace compute_abstime() with unpackTime()
5508
5509 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5510 if (millis < 0) millis = 0;
5511
5512 jlong seconds = millis / 1000;
5513 millis %= 1000;
5514 if (seconds > 50000000) { // see man cond_timedwait(3T)
5515 seconds = 50000000;
5516 }
5517
5518 if (os::supports_monotonic_clock()) {
5519 struct timespec now;
5520 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5521 assert_status(status == 0, status, "clock_gettime");
5522 abstime->tv_sec = now.tv_sec + seconds;
5523 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5524 if (nanos >= NANOSECS_PER_SEC) {
5525 abstime->tv_sec += 1;
5526 nanos -= NANOSECS_PER_SEC;
5527 }
5528 abstime->tv_nsec = nanos;
5529 } else {
5530 struct timeval now;
5531 int status = gettimeofday(&now, NULL);
5532 assert(status == 0, "gettimeofday");
5533 abstime->tv_sec = now.tv_sec + seconds;
5534 long usec = now.tv_usec + millis * 1000;
5535 if (usec >= 1000000) {
5536 abstime->tv_sec += 1;
5537 usec -= 1000000;
5538 }
5539 abstime->tv_nsec = usec * 1000;
5540 }
5541 return abstime;
5542 }
5543
5544 void os::PlatformEvent::park() { // AKA "down()"
5545 // Transitions for _Event:
5546 // -1 => -1 : illegal
5547 // 1 => 0 : pass - return immediately
5548 // 0 => -1 : block; then set _Event to 0 before returning
5549
5550 // Invariant: Only the thread associated with the Event/PlatformEvent
5551 // may call park().
5552 // TODO: assert that _Assoc != NULL or _Assoc == Self
5553 assert(_nParked == 0, "invariant");
5554
5555 int v;
5556 for (;;) {
5557 v = _Event;
5558 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5559 }
5560 guarantee(v >= 0, "invariant");
5561 if (v == 0) {
5562 // Do this the hard way by blocking ...
5563 int status = pthread_mutex_lock(_mutex);
5564 assert_status(status == 0, status, "mutex_lock");
5565 guarantee(_nParked == 0, "invariant");
5566 ++_nParked;
5567 while (_Event < 0) {
5568 status = pthread_cond_wait(_cond, _mutex);
5569 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5570 // Treat this the same as if the wait was interrupted
5571 if (status == ETIME) { status = EINTR; }
5572 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5573 }
5574 --_nParked;
5575
5576 _Event = 0;
5577 status = pthread_mutex_unlock(_mutex);
5578 assert_status(status == 0, status, "mutex_unlock");
5579 // Paranoia to ensure our locked and lock-free paths interact
5580 // correctly with each other.
5581 OrderAccess::fence();
5582 }
5583 guarantee(_Event >= 0, "invariant");
5584 }
5585
5586 int os::PlatformEvent::park(jlong millis) {
5587 // Transitions for _Event:
5588 // -1 => -1 : illegal
5589 // 1 => 0 : pass - return immediately
5590 // 0 => -1 : block; then set _Event to 0 before returning
5591
5592 guarantee(_nParked == 0, "invariant");
5593
5594 int v;
5595 for (;;) {
5596 v = _Event;
5597 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5598 }
5599 guarantee(v >= 0, "invariant");
5600 if (v != 0) return OS_OK;
5601
5602 // We do this the hard way, by blocking the thread.
5603 // Consider enforcing a minimum timeout value.
5604 struct timespec abst;
5605 compute_abstime(&abst, millis);
5606
5607 int ret = OS_TIMEOUT;
5608 int status = pthread_mutex_lock(_mutex);
5609 assert_status(status == 0, status, "mutex_lock");
5610 guarantee(_nParked == 0, "invariant");
5611 ++_nParked;
5612
5613 // Object.wait(timo) will return because of
5614 // (a) notification
5615 // (b) timeout
5616 // (c) thread.interrupt
5617 //
5618 // Thread.interrupt and object.notify{All} both call Event::set.
5619 // That is, we treat thread.interrupt as a special case of notification.
5620 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5621 // We assume all ETIME returns are valid.
5622 //
5623 // TODO: properly differentiate simultaneous notify+interrupt.
5624 // In that case, we should propagate the notify to another waiter.
5625
5626 while (_Event < 0) {
5627 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5628 assert_status(status == 0 || status == EINTR ||
5629 status == ETIME || status == ETIMEDOUT,
5630 status, "cond_timedwait");
5631 if (!FilterSpuriousWakeups) break; // previous semantics
5632 if (status == ETIME || status == ETIMEDOUT) break;
5633 // We consume and ignore EINTR and spurious wakeups.
5634 }
5635 --_nParked;
5636 if (_Event >= 0) {
5637 ret = OS_OK;
5638 }
5639 _Event = 0;
5640 status = pthread_mutex_unlock(_mutex);
5641 assert_status(status == 0, status, "mutex_unlock");
5642 assert(_nParked == 0, "invariant");
5643 // Paranoia to ensure our locked and lock-free paths interact
5644 // correctly with each other.
5645 OrderAccess::fence();
5646 return ret;
5647 }
5648
5649 void os::PlatformEvent::unpark() {
5650 // Transitions for _Event:
5651 // 0 => 1 : just return
5652 // 1 => 1 : just return
5653 // -1 => either 0 or 1; must signal target thread
5654 // That is, we can safely transition _Event from -1 to either
5655 // 0 or 1.
5656 // See also: "Semaphores in Plan 9" by Mullender & Cox
5657 //
5658 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5659 // that it will take two back-to-back park() calls for the owning
5660 // thread to block. This has the benefit of forcing a spurious return
5661 // from the first park() call after an unpark() call which will help
5662 // shake out uses of park() and unpark() without condition variables.
5663
5664 if (Atomic::xchg(1, &_Event) >= 0) return;
5665
5666 // Wait for the thread associated with the event to vacate
5667 int status = pthread_mutex_lock(_mutex);
5668 assert_status(status == 0, status, "mutex_lock");
5669 int AnyWaiters = _nParked;
5670 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5671 status = pthread_mutex_unlock(_mutex);
5672 assert_status(status == 0, status, "mutex_unlock");
5673 if (AnyWaiters != 0) {
5674 // Note that we signal() *after* dropping the lock for "immortal" Events.
5675 // This is safe and avoids a common class of futile wakeups. In rare
5676 // circumstances this can cause a thread to return prematurely from
5677 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5678 // will simply re-test the condition and re-park itself.
5679 // This provides particular benefit if the underlying platform does not
5680 // provide wait morphing.
5681 status = pthread_cond_signal(_cond);
5682 assert_status(status == 0, status, "cond_signal");
5683 }
5684 }
5685
5686
5687 // JSR166
5688 // -------------------------------------------------------
5689
5690 // The solaris and linux implementations of park/unpark are fairly
5691 // conservative for now, but can be improved. They currently use a
5692 // mutex/condvar pair, plus a a count.
5693 // Park decrements count if > 0, else does a condvar wait. Unpark
5694 // sets count to 1 and signals condvar. Only one thread ever waits
5695 // on the condvar. Contention seen when trying to park implies that someone
5696 // is unparking you, so don't wait. And spurious returns are fine, so there
5697 // is no need to track notifications.
5698
5699 // This code is common to linux and solaris and will be moved to a
5700 // common place in dolphin.
5701 //
5702 // The passed in time value is either a relative time in nanoseconds
5703 // or an absolute time in milliseconds. Either way it has to be unpacked
5704 // into suitable seconds and nanoseconds components and stored in the
5705 // given timespec structure.
5706 // Given time is a 64-bit value and the time_t used in the timespec is only
5707 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5708 // overflow if times way in the future are given. Further on Solaris versions
5709 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5710 // number of seconds, in abstime, is less than current_time + 100,000,000.
5711 // As it will be 28 years before "now + 100000000" will overflow we can
5712 // ignore overflow and just impose a hard-limit on seconds using the value
5713 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5714 // years from "now".
5715
5716 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5717 assert(time > 0, "convertTime");
5718 time_t max_secs = 0;
5719
5720 if (!os::supports_monotonic_clock() || isAbsolute) {
5721 struct timeval now;
5722 int status = gettimeofday(&now, NULL);
5723 assert(status == 0, "gettimeofday");
5724
5725 max_secs = now.tv_sec + MAX_SECS;
5726
5727 if (isAbsolute) {
5728 jlong secs = time / 1000;
5729 if (secs > max_secs) {
5730 absTime->tv_sec = max_secs;
5731 } else {
5732 absTime->tv_sec = secs;
5733 }
5734 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5735 } else {
5736 jlong secs = time / NANOSECS_PER_SEC;
5737 if (secs >= MAX_SECS) {
5738 absTime->tv_sec = max_secs;
5739 absTime->tv_nsec = 0;
5740 } else {
5741 absTime->tv_sec = now.tv_sec + secs;
5742 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5743 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5744 absTime->tv_nsec -= NANOSECS_PER_SEC;
5745 ++absTime->tv_sec; // note: this must be <= max_secs
5746 }
5747 }
5748 }
5749 } else {
5750 // must be relative using monotonic clock
5751 struct timespec now;
5752 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5753 assert_status(status == 0, status, "clock_gettime");
5754 max_secs = now.tv_sec + MAX_SECS;
5755 jlong secs = time / NANOSECS_PER_SEC;
5756 if (secs >= MAX_SECS) {
5757 absTime->tv_sec = max_secs;
5758 absTime->tv_nsec = 0;
5759 } else {
5760 absTime->tv_sec = now.tv_sec + secs;
5761 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5762 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5763 absTime->tv_nsec -= NANOSECS_PER_SEC;
5764 ++absTime->tv_sec; // note: this must be <= max_secs
5765 }
5766 }
5767 }
5768 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5769 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5770 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5771 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5772 }
5773
5774 void Parker::park(bool isAbsolute, jlong time) {
5775 // Ideally we'd do something useful while spinning, such
5776 // as calling unpackTime().
5777
5778 // Optional fast-path check:
5779 // Return immediately if a permit is available.
5780 // We depend on Atomic::xchg() having full barrier semantics
5781 // since we are doing a lock-free update to _counter.
5782 if (Atomic::xchg(0, &_counter) > 0) return;
5783
5784 Thread* thread = Thread::current();
5785 assert(thread->is_Java_thread(), "Must be JavaThread");
5786 JavaThread *jt = (JavaThread *)thread;
5787
5788 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5789 // Check interrupt before trying to wait
5790 if (Thread::is_interrupted(thread, false)) {
5791 return;
5792 }
5793
5794 // Next, demultiplex/decode time arguments
5795 timespec absTime;
5796 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5797 return;
5798 }
5799 if (time > 0) {
5800 unpackTime(&absTime, isAbsolute, time);
5801 }
5802
5803
5804 // Enter safepoint region
5805 // Beware of deadlocks such as 6317397.
5806 // The per-thread Parker:: mutex is a classic leaf-lock.
5807 // In particular a thread must never block on the Threads_lock while
5808 // holding the Parker:: mutex. If safepoints are pending both the
5809 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5810 ThreadBlockInVM tbivm(jt);
5811
5812 // Don't wait if cannot get lock since interference arises from
5813 // unblocking. Also. check interrupt before trying wait
5814 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5815 return;
5816 }
5817
5818 int status;
5819 if (_counter > 0) { // no wait needed
5820 _counter = 0;
5821 status = pthread_mutex_unlock(_mutex);
5822 assert_status(status == 0, status, "invariant");
5823 // Paranoia to ensure our locked and lock-free paths interact
5824 // correctly with each other and Java-level accesses.
5825 OrderAccess::fence();
5826 return;
5827 }
5828
5829 #ifdef ASSERT
5830 // Don't catch signals while blocked; let the running threads have the signals.
5831 // (This allows a debugger to break into the running thread.)
5832 sigset_t oldsigs;
5833 sigemptyset(&oldsigs);
5834 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5835 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5836 #endif
5837
5838 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5839 jt->set_suspend_equivalent();
5840 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5841
5842 assert(_cur_index == -1, "invariant");
5843 if (time == 0) {
5844 _cur_index = REL_INDEX; // arbitrary choice when not timed
5845 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5846 } else {
5847 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5848 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5849 }
5850 _cur_index = -1;
5851 assert_status(status == 0 || status == EINTR ||
5852 status == ETIME || status == ETIMEDOUT,
5853 status, "cond_timedwait");
5854
5855 #ifdef ASSERT
5856 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5857 #endif
5858
5859 _counter = 0;
5860 status = pthread_mutex_unlock(_mutex);
5861 assert_status(status == 0, status, "invariant");
5862 // Paranoia to ensure our locked and lock-free paths interact
5863 // correctly with each other and Java-level accesses.
5864 OrderAccess::fence();
5865
5866 // If externally suspended while waiting, re-suspend
5867 if (jt->handle_special_suspend_equivalent_condition()) {
5868 jt->java_suspend_self();
5869 }
5870 }
5871
5872 void Parker::unpark() {
5873 int status = pthread_mutex_lock(_mutex);
5874 assert_status(status == 0, status, "invariant");
5875 const int s = _counter;
5876 _counter = 1;
5877 // must capture correct index before unlocking
5878 int index = _cur_index;
5879 status = pthread_mutex_unlock(_mutex);
5880 assert_status(status == 0, status, "invariant");
5881 if (s < 1 && index != -1) {
5882 // thread is definitely parked
5883 status = pthread_cond_signal(&_cond[index]);
5884 assert_status(status == 0, status, "invariant");
5885 }
5886 }
5887
5888
5889 extern char** environ;
5890
5891 // Run the specified command in a separate process. Return its exit value,
5892 // or -1 on failure (e.g. can't fork a new process).
5893 // Unlike system(), this function can be called from signal handler. It
5894 // doesn't block SIGINT et al.
5895 int os::fork_and_exec(char* cmd) {
5896 const char * argv[4] = {"sh", "-c", cmd, NULL};
5897
5898 pid_t pid = fork();
5899
5900 if (pid < 0) {
5901 // fork failed
5902 return -1;
5903
5904 } else if (pid == 0) {
5905 // child process
5906
5907 execve("/bin/sh", (char* const*)argv, environ);
5908
5909 // execve failed
5910 _exit(-1);
5911
5912 } else {
5913 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5914 // care about the actual exit code, for now.
5915
5916 int status;
5917
5918 // Wait for the child process to exit. This returns immediately if
5919 // the child has already exited. */
5920 while (waitpid(pid, &status, 0) < 0) {
5921 switch (errno) {
5922 case ECHILD: return 0;
5923 case EINTR: break;
5924 default: return -1;
5925 }
5926 }
5927
5928 if (WIFEXITED(status)) {
5929 // The child exited normally; get its exit code.
5930 return WEXITSTATUS(status);
5931 } else if (WIFSIGNALED(status)) {
5932 // The child exited because of a signal
5933 // The best value to return is 0x80 + signal number,
5934 // because that is what all Unix shells do, and because
5935 // it allows callers to distinguish between process exit and
5936 // process death by signal.
5937 return 0x80 + WTERMSIG(status);
5938 } else {
5939 // Unknown exit code; pass it through
5940 return status;
5941 }
5942 }
5943 }
5944
5945 // is_headless_jre()
5946 //
5947 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5948 // in order to report if we are running in a headless jre
5949 //
5950 // Since JDK8 xawt/libmawt.so was moved into the same directory
5951 // as libawt.so, and renamed libawt_xawt.so
5952 //
5953 bool os::is_headless_jre() {
5954 struct stat statbuf;
5955 char buf[MAXPATHLEN];
5956 char libmawtpath[MAXPATHLEN];
5957 const char *xawtstr = "/xawt/libmawt.so";
5958 const char *new_xawtstr = "/libawt_xawt.so";
5959 char *p;
5960
5961 // Get path to libjvm.so
5962 os::jvm_path(buf, sizeof(buf));
5963
5964 // Get rid of libjvm.so
5965 p = strrchr(buf, '/');
5966 if (p == NULL) {
5967 return false;
5968 } else {
5969 *p = '\0';
5970 }
5971
5972 // Get rid of client or server
5973 p = strrchr(buf, '/');
5974 if (p == NULL) {
5975 return false;
5976 } else {
5977 *p = '\0';
5978 }
5979
5980 // check xawt/libmawt.so
5981 strcpy(libmawtpath, buf);
5982 strcat(libmawtpath, xawtstr);
5983 if (::stat(libmawtpath, &statbuf) == 0) return false;
5984
5985 // check libawt_xawt.so
5986 strcpy(libmawtpath, buf);
5987 strcat(libmawtpath, new_xawtstr);
5988 if (::stat(libmawtpath, &statbuf) == 0) return false;
5989
5990 return true;
5991 }
5992
5993 // Get the default path to the core file
5994 // Returns the length of the string
5995 int os::get_core_path(char* buffer, size_t bufferSize) {
5996 /*
5997 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5998 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5999 */
6000 const int core_pattern_len = 129;
6001 char core_pattern[core_pattern_len] = {0};
6002
6003 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6004 if (core_pattern_file == -1) {
6005 return -1;
6006 }
6007
6008 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6009 ::close(core_pattern_file);
6010 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6011 return -1;
6012 }
6013 if (core_pattern[ret-1] == '\n') {
6014 core_pattern[ret-1] = '\0';
6015 } else {
6016 core_pattern[ret] = '\0';
6017 }
6018
6019 char *pid_pos = strstr(core_pattern, "%p");
6020 int written;
6021
6022 if (core_pattern[0] == '/') {
6023 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6024 } else {
6025 char cwd[PATH_MAX];
6026
6027 const char* p = get_current_directory(cwd, PATH_MAX);
6028 if (p == NULL) {
6029 return -1;
6030 }
6031
6032 if (core_pattern[0] == '|') {
6033 written = jio_snprintf(buffer, bufferSize,
6034 "\"%s\" (or dumping to %s/core.%d)",
6035 &core_pattern[1], p, current_process_id());
6036 } else {
6037 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6038 }
6039 }
6040
6041 if (written < 0) {
6042 return -1;
6043 }
6044
6045 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6046 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6047
6048 if (core_uses_pid_file != -1) {
6049 char core_uses_pid = 0;
6050 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6051 ::close(core_uses_pid_file);
6052
6053 if (core_uses_pid == '1') {
6054 jio_snprintf(buffer + written, bufferSize - written,
6055 ".%d", current_process_id());
6056 }
6057 }
6058 }
6059
6060 return strlen(buffer);
6061 }
6062
6063 bool os::start_debugging(char *buf, int buflen) {
6064 int len = (int)strlen(buf);
6065 char *p = &buf[len];
6066
6067 jio_snprintf(p, buflen-len,
6068 "\n\n"
6069 "Do you want to debug the problem?\n\n"
6070 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6071 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6072 "Otherwise, press RETURN to abort...",
6073 os::current_process_id(), os::current_process_id(),
6074 os::current_thread_id(), os::current_thread_id());
6075
6076 bool yes = os::message_box("Unexpected Error", buf);
6077
6078 if (yes) {
6079 // yes, user asked VM to launch debugger
6080 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6081 os::current_process_id(), os::current_process_id());
6082
6083 os::fork_and_exec(buf);
6084 yes = false;
6085 }
6086 return yes;
6087 }
6088
6089
6090 // Java/Compiler thread:
6091 //
6092 // Low memory addresses
6093 // P0 +------------------------+
6094 // | |\ Java thread created by VM does not have glibc
6095 // | glibc guard page | - guard page, attached Java thread usually has
6096 // | |/ 1 page glibc guard.
6097 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6098 // | |\
6099 // | HotSpot Guard Pages | - red, yellow and reserved pages
6100 // | |/
6101 // +------------------------+ JavaThread::stack_reserved_zone_base()
6102 // | |\
6103 // | Normal Stack | -
6104 // | |/
6105 // P2 +------------------------+ Thread::stack_base()
6106 //
6107 // Non-Java thread:
6108 //
6109 // Low memory addresses
6110 // P0 +------------------------+
6111 // | |\
6112 // | glibc guard page | - usually 1 page
6113 // | |/
6114 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6115 // | |\
6116 // | Normal Stack | -
6117 // | |/
6118 // P2 +------------------------+ Thread::stack_base()
6119 //
6120 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6121 // returned from pthread_attr_getstack().
6122 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6123 // of the stack size given in pthread_attr. We work around this for
6124 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6125 //
6126 #ifndef ZERO
6127 static void current_stack_region(address * bottom, size_t * size) {
6128 if (os::Linux::is_initial_thread()) {
6129 // initial thread needs special handling because pthread_getattr_np()
6130 // may return bogus value.
6131 *bottom = os::Linux::initial_thread_stack_bottom();
6132 *size = os::Linux::initial_thread_stack_size();
6133 } else {
6134 pthread_attr_t attr;
6135
6136 int rslt = pthread_getattr_np(pthread_self(), &attr);
6137
6138 // JVM needs to know exact stack location, abort if it fails
6139 if (rslt != 0) {
6140 if (rslt == ENOMEM) {
6141 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6142 } else {
6143 fatal("pthread_getattr_np failed with errno = %d", rslt);
6144 }
6145 }
6146
6147 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6148 fatal("Cannot locate current stack attributes!");
6149 }
6150
6151 // Work around NPTL stack guard error.
6152 size_t guard_size = 0;
6153 rslt = pthread_attr_getguardsize(&attr, &guard_size);
6154 if (rslt != 0) {
6155 fatal("pthread_attr_getguardsize failed with errno = %d", rslt);
6156 }
6157 *bottom += guard_size;
6158 *size -= guard_size;
6159
6160 pthread_attr_destroy(&attr);
6161
6162 }
6163 assert(os::current_stack_pointer() >= *bottom &&
6164 os::current_stack_pointer() < *bottom + *size, "just checking");
6165 }
6166
6167 address os::current_stack_base() {
6168 address bottom;
6169 size_t size;
6170 current_stack_region(&bottom, &size);
6171 return (bottom + size);
6172 }
6173
6174 size_t os::current_stack_size() {
6175 // This stack size includes the usable stack and HotSpot guard pages
6176 // (for the threads that have Hotspot guard pages).
6177 address bottom;
6178 size_t size;
6179 current_stack_region(&bottom, &size);
6180 return size;
6181 }
6182 #endif
6183
6184 static inline struct timespec get_mtime(const char* filename) {
6185 struct stat st;
6186 int ret = os::stat(filename, &st);
6187 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6188 return st.st_mtim;
6189 }
6190
6191 int os::compare_file_modified_times(const char* file1, const char* file2) {
6192 struct timespec filetime1 = get_mtime(file1);
6193 struct timespec filetime2 = get_mtime(file2);
6194 int diff = filetime1.tv_sec - filetime2.tv_sec;
6195 if (diff == 0) {
6196 return filetime1.tv_nsec - filetime2.tv_nsec;
6197 }
6198 return diff;
6199 }
6200
6201 /////////////// Unit tests ///////////////
6202
6203 #ifndef PRODUCT
6204
6205 #define test_log(...) \
6206 do { \
6207 if (VerboseInternalVMTests) { \
6208 tty->print_cr(__VA_ARGS__); \
6209 tty->flush(); \
6210 } \
6211 } while (false)
6212
6213 class TestReserveMemorySpecial : AllStatic {
6214 public:
6215 static void small_page_write(void* addr, size_t size) {
6216 size_t page_size = os::vm_page_size();
6217
6218 char* end = (char*)addr + size;
6219 for (char* p = (char*)addr; p < end; p += page_size) {
6220 *p = 1;
6221 }
6222 }
6223
6224 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6225 if (!UseHugeTLBFS) {
6226 return;
6227 }
6228
6229 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6230
6231 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6232
6233 if (addr != NULL) {
6234 small_page_write(addr, size);
6235
6236 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6237 }
6238 }
6239
6240 static void test_reserve_memory_special_huge_tlbfs_only() {
6241 if (!UseHugeTLBFS) {
6242 return;
6243 }
6244
6245 size_t lp = os::large_page_size();
6246
6247 for (size_t size = lp; size <= lp * 10; size += lp) {
6248 test_reserve_memory_special_huge_tlbfs_only(size);
6249 }
6250 }
6251
6252 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6253 size_t lp = os::large_page_size();
6254 size_t ag = os::vm_allocation_granularity();
6255
6256 // sizes to test
6257 const size_t sizes[] = {
6258 lp, lp + ag, lp + lp / 2, lp * 2,
6259 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6260 lp * 10, lp * 10 + lp / 2
6261 };
6262 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6263
6264 // For each size/alignment combination, we test three scenarios:
6265 // 1) with req_addr == NULL
6266 // 2) with a non-null req_addr at which we expect to successfully allocate
6267 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6268 // expect the allocation to either fail or to ignore req_addr
6269
6270 // Pre-allocate two areas; they shall be as large as the largest allocation
6271 // and aligned to the largest alignment we will be testing.
6272 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6273 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6274 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6275 -1, 0);
6276 assert(mapping1 != MAP_FAILED, "should work");
6277
6278 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6279 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6280 -1, 0);
6281 assert(mapping2 != MAP_FAILED, "should work");
6282
6283 // Unmap the first mapping, but leave the second mapping intact: the first
6284 // mapping will serve as a value for a "good" req_addr (case 2). The second
6285 // mapping, still intact, as "bad" req_addr (case 3).
6286 ::munmap(mapping1, mapping_size);
6287
6288 // Case 1
6289 test_log("%s, req_addr NULL:", __FUNCTION__);
6290 test_log("size align result");
6291
6292 for (int i = 0; i < num_sizes; i++) {
6293 const size_t size = sizes[i];
6294 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6295 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6296 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6297 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6298 if (p != NULL) {
6299 assert(is_ptr_aligned(p, alignment), "must be");
6300 small_page_write(p, size);
6301 os::Linux::release_memory_special_huge_tlbfs(p, size);
6302 }
6303 }
6304 }
6305
6306 // Case 2
6307 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6308 test_log("size align req_addr result");
6309
6310 for (int i = 0; i < num_sizes; i++) {
6311 const size_t size = sizes[i];
6312 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6313 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6314 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6315 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6316 size, alignment, p2i(req_addr), p2i(p),
6317 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6318 if (p != NULL) {
6319 assert(p == req_addr, "must be");
6320 small_page_write(p, size);
6321 os::Linux::release_memory_special_huge_tlbfs(p, size);
6322 }
6323 }
6324 }
6325
6326 // Case 3
6327 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6328 test_log("size align req_addr result");
6329
6330 for (int i = 0; i < num_sizes; i++) {
6331 const size_t size = sizes[i];
6332 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6333 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6334 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6335 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6336 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6337 // as the area around req_addr contains already existing mappings, the API should always
6338 // return NULL (as per contract, it cannot return another address)
6339 assert(p == NULL, "must be");
6340 }
6341 }
6342
6343 ::munmap(mapping2, mapping_size);
6344
6345 }
6346
6347 static void test_reserve_memory_special_huge_tlbfs() {
6348 if (!UseHugeTLBFS) {
6349 return;
6350 }
6351
6352 test_reserve_memory_special_huge_tlbfs_only();
6353 test_reserve_memory_special_huge_tlbfs_mixed();
6354 }
6355
6356 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6357 if (!UseSHM) {
6358 return;
6359 }
6360
6361 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6362
6363 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6364
6365 if (addr != NULL) {
6366 assert(is_ptr_aligned(addr, alignment), "Check");
6367 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6368
6369 small_page_write(addr, size);
6370
6371 os::Linux::release_memory_special_shm(addr, size);
6372 }
6373 }
6374
6375 static void test_reserve_memory_special_shm() {
6376 size_t lp = os::large_page_size();
6377 size_t ag = os::vm_allocation_granularity();
6378
6379 for (size_t size = ag; size < lp * 3; size += ag) {
6380 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6381 test_reserve_memory_special_shm(size, alignment);
6382 }
6383 }
6384 }
6385
6386 static void test() {
6387 test_reserve_memory_special_huge_tlbfs();
6388 test_reserve_memory_special_shm();
6389 }
6390 };
6391
6392 void TestReserveMemorySpecial_test() {
6393 TestReserveMemorySpecial::test();
6394 }
6395
6396 #endif