1 /*
2 * Copyright (c) 1999, 2015, 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 "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.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.inline.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 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
110 // getrusage() is prepared to handle the associated failure.
111 #ifndef RUSAGE_THREAD
112 #define RUSAGE_THREAD (1) /* only the calling thread */
113 #endif
114
115 #define MAX_PATH (2 * K)
116
117 #define MAX_SECS 100000000
118
119 // for timer info max values which include all bits
120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
121
122 #define LARGEPAGES_BIT (1 << 6)
123 ////////////////////////////////////////////////////////////////////////////////
124 // global variables
125 julong os::Linux::_physical_memory = 0;
126
127 address os::Linux::_initial_thread_stack_bottom = NULL;
128 uintptr_t os::Linux::_initial_thread_stack_size = 0;
129
130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
132 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
133 Mutex* os::Linux::_createThread_lock = NULL;
134 pthread_t os::Linux::_main_thread;
135 int os::Linux::_page_size = -1;
136 const int os::Linux::_vm_default_page_size = (8 * K);
137 bool os::Linux::_supports_fast_thread_cpu_time = false;
138 const char * os::Linux::_glibc_version = NULL;
139 const char * os::Linux::_libpthread_version = NULL;
140 pthread_condattr_t os::Linux::_condattr[1];
141
142 static jlong initial_time_count=0;
143
144 static int clock_tics_per_sec = 100;
145
146 // For diagnostics to print a message once. see run_periodic_checks
147 static sigset_t check_signal_done;
148 static bool check_signals = true;
149
150 // Signal number used to suspend/resume a thread
151
152 // do not use any signal number less than SIGSEGV, see 4355769
153 static int SR_signum = SIGUSR2;
154 sigset_t SR_sigset;
155
156 // Declarations
157 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
158
159 // utility functions
160
161 static int SR_initialize();
162
163 julong os::available_memory() {
164 return Linux::available_memory();
165 }
166
167 julong os::Linux::available_memory() {
168 // values in struct sysinfo are "unsigned long"
169 struct sysinfo si;
170 sysinfo(&si);
171
172 return (julong)si.freeram * si.mem_unit;
173 }
174
175 julong os::physical_memory() {
176 return Linux::physical_memory();
177 }
178
179 // Return true if user is running as root.
180
181 bool os::have_special_privileges() {
182 static bool init = false;
183 static bool privileges = false;
184 if (!init) {
185 privileges = (getuid() != geteuid()) || (getgid() != getegid());
186 init = true;
187 }
188 return privileges;
189 }
190
191
192 #ifndef SYS_gettid
193 // i386: 224, ia64: 1105, amd64: 186, sparc 143
194 #ifdef __ia64__
195 #define SYS_gettid 1105
196 #else
197 #ifdef __i386__
198 #define SYS_gettid 224
199 #else
200 #ifdef __amd64__
201 #define SYS_gettid 186
202 #else
203 #ifdef __sparc__
204 #define SYS_gettid 143
205 #else
206 #error define gettid for the arch
207 #endif
208 #endif
209 #endif
210 #endif
211 #endif
212
213 // Cpu architecture string
214 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
215
216
217 // pid_t gettid()
218 //
219 // Returns the kernel thread id of the currently running thread. Kernel
220 // thread id is used to access /proc.
221 pid_t os::Linux::gettid() {
222 int rslt = syscall(SYS_gettid);
223 assert(rslt != -1, "must be."); // old linuxthreads implementation?
224 return (pid_t)rslt;
225 }
226
227 // Most versions of linux have a bug where the number of processors are
228 // determined by looking at the /proc file system. In a chroot environment,
229 // the system call returns 1. This causes the VM to act as if it is
230 // a single processor and elide locking (see is_MP() call).
231 static bool unsafe_chroot_detected = false;
232 static const char *unstable_chroot_error = "/proc file system not found.\n"
233 "Java may be unstable running multithreaded in a chroot "
234 "environment on Linux when /proc filesystem is not mounted.";
235
236 void os::Linux::initialize_system_info() {
237 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
238 if (processor_count() == 1) {
239 pid_t pid = os::Linux::gettid();
240 char fname[32];
241 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
242 FILE *fp = fopen(fname, "r");
243 if (fp == NULL) {
244 unsafe_chroot_detected = true;
245 } else {
246 fclose(fp);
247 }
248 }
249 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
250 assert(processor_count() > 0, "linux error");
251 }
252
253 void os::init_system_properties_values() {
254 // The next steps are taken in the product version:
255 //
256 // Obtain the JAVA_HOME value from the location of libjvm.so.
257 // This library should be located at:
258 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
259 //
260 // If "/jre/lib/" appears at the right place in the path, then we
261 // assume libjvm.so is installed in a JDK and we use this path.
262 //
263 // Otherwise exit with message: "Could not create the Java virtual machine."
264 //
265 // The following extra steps are taken in the debugging version:
266 //
267 // If "/jre/lib/" does NOT appear at the right place in the path
268 // instead of exit check for $JAVA_HOME environment variable.
269 //
270 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
271 // then we append a fake suffix "hotspot/libjvm.so" to this path so
272 // it looks like libjvm.so is installed there
273 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
274 //
275 // Otherwise exit.
276 //
277 // Important note: if the location of libjvm.so changes this
278 // code needs to be changed accordingly.
279
280 // See ld(1):
281 // The linker uses the following search paths to locate required
282 // shared libraries:
283 // 1: ...
284 // ...
285 // 7: The default directories, normally /lib and /usr/lib.
286 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
287 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
288 #else
289 #define DEFAULT_LIBPATH "/lib:/usr/lib"
290 #endif
291
292 // Base path of extensions installed on the system.
293 #define SYS_EXT_DIR "/usr/java/packages"
294 #define EXTENSIONS_DIR "/lib/ext"
295
296 // Buffer that fits several sprintfs.
297 // Note that the space for the colon and the trailing null are provided
298 // by the nulls included by the sizeof operator.
299 const size_t bufsize =
300 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
301 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
302 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
303
304 // sysclasspath, java_home, dll_dir
305 {
306 char *pslash;
307 os::jvm_path(buf, bufsize);
308
309 // Found the full path to libjvm.so.
310 // Now cut the path to <java_home>/jre if we can.
311 pslash = strrchr(buf, '/');
312 if (pslash != NULL) {
313 *pslash = '\0'; // Get rid of /libjvm.so.
314 }
315 pslash = strrchr(buf, '/');
316 if (pslash != NULL) {
317 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
318 }
319 Arguments::set_dll_dir(buf);
320
321 if (pslash != NULL) {
322 pslash = strrchr(buf, '/');
323 if (pslash != NULL) {
324 *pslash = '\0'; // Get rid of /<arch>.
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /lib.
328 }
329 }
330 }
331 Arguments::set_java_home(buf);
332 set_boot_path('/', ':');
333 }
334
335 // Where to look for native libraries.
336 //
337 // Note: Due to a legacy implementation, most of the library path
338 // is set in the launcher. This was to accomodate linking restrictions
339 // on legacy Linux implementations (which are no longer supported).
340 // Eventually, all the library path setting will be done here.
341 //
342 // However, to prevent the proliferation of improperly built native
343 // libraries, the new path component /usr/java/packages is added here.
344 // Eventually, all the library path setting will be done here.
345 {
346 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
347 // should always exist (until the legacy problem cited above is
348 // addressed).
349 const char *v = ::getenv("LD_LIBRARY_PATH");
350 const char *v_colon = ":";
351 if (v == NULL) { v = ""; v_colon = ""; }
352 // That's +1 for the colon and +1 for the trailing '\0'.
353 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
354 strlen(v) + 1 +
355 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
356 mtInternal);
357 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
358 Arguments::set_library_path(ld_library_path);
359 FREE_C_HEAP_ARRAY(char, ld_library_path);
360 }
361
362 // Extensions directories.
363 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
364 Arguments::set_ext_dirs(buf);
365
366 FREE_C_HEAP_ARRAY(char, buf);
367
368 #undef DEFAULT_LIBPATH
369 #undef SYS_EXT_DIR
370 #undef EXTENSIONS_DIR
371 }
372
373 ////////////////////////////////////////////////////////////////////////////////
374 // breakpoint support
375
376 void os::breakpoint() {
377 BREAKPOINT;
378 }
379
380 extern "C" void breakpoint() {
381 // use debugger to set breakpoint here
382 }
383
384 ////////////////////////////////////////////////////////////////////////////////
385 // signal support
386
387 debug_only(static bool signal_sets_initialized = false);
388 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
389
390 bool os::Linux::is_sig_ignored(int sig) {
391 struct sigaction oact;
392 sigaction(sig, (struct sigaction*)NULL, &oact);
393 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
394 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
395 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
396 return true;
397 } else {
398 return false;
399 }
400 }
401
402 void os::Linux::signal_sets_init() {
403 // Should also have an assertion stating we are still single-threaded.
404 assert(!signal_sets_initialized, "Already initialized");
405 // Fill in signals that are necessarily unblocked for all threads in
406 // the VM. Currently, we unblock the following signals:
407 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
408 // by -Xrs (=ReduceSignalUsage));
409 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
410 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
411 // the dispositions or masks wrt these signals.
412 // Programs embedding the VM that want to use the above signals for their
413 // own purposes must, at this time, use the "-Xrs" option to prevent
414 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
415 // (See bug 4345157, and other related bugs).
416 // In reality, though, unblocking these signals is really a nop, since
417 // these signals are not blocked by default.
418 sigemptyset(&unblocked_sigs);
419 sigemptyset(&allowdebug_blocked_sigs);
420 sigaddset(&unblocked_sigs, SIGILL);
421 sigaddset(&unblocked_sigs, SIGSEGV);
422 sigaddset(&unblocked_sigs, SIGBUS);
423 sigaddset(&unblocked_sigs, SIGFPE);
424 #if defined(PPC64)
425 sigaddset(&unblocked_sigs, SIGTRAP);
426 #endif
427 sigaddset(&unblocked_sigs, SR_signum);
428
429 if (!ReduceSignalUsage) {
430 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
431 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
432 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
433 }
434 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
435 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
436 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
437 }
438 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
439 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
441 }
442 }
443 // Fill in signals that are blocked by all but the VM thread.
444 sigemptyset(&vm_sigs);
445 if (!ReduceSignalUsage) {
446 sigaddset(&vm_sigs, BREAK_SIGNAL);
447 }
448 debug_only(signal_sets_initialized = true);
449
450 }
451
452 // These are signals that are unblocked while a thread is running Java.
453 // (For some reason, they get blocked by default.)
454 sigset_t* os::Linux::unblocked_signals() {
455 assert(signal_sets_initialized, "Not initialized");
456 return &unblocked_sigs;
457 }
458
459 // These are the signals that are blocked while a (non-VM) thread is
460 // running Java. Only the VM thread handles these signals.
461 sigset_t* os::Linux::vm_signals() {
462 assert(signal_sets_initialized, "Not initialized");
463 return &vm_sigs;
464 }
465
466 // These are signals that are blocked during cond_wait to allow debugger in
467 sigset_t* os::Linux::allowdebug_blocked_signals() {
468 assert(signal_sets_initialized, "Not initialized");
469 return &allowdebug_blocked_sigs;
470 }
471
472 void os::Linux::hotspot_sigmask(Thread* thread) {
473
474 //Save caller's signal mask before setting VM signal mask
475 sigset_t caller_sigmask;
476 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
477
478 OSThread* osthread = thread->osthread();
479 osthread->set_caller_sigmask(caller_sigmask);
480
481 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
482
483 if (!ReduceSignalUsage) {
484 if (thread->is_VM_thread()) {
485 // Only the VM thread handles BREAK_SIGNAL ...
486 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
487 } else {
488 // ... all other threads block BREAK_SIGNAL
489 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
490 }
491 }
492 }
493
494 //////////////////////////////////////////////////////////////////////////////
495 // detecting pthread library
496
497 void os::Linux::libpthread_init() {
498 // Save glibc and pthread version strings.
499 #if !defined(_CS_GNU_LIBC_VERSION) || \
500 !defined(_CS_GNU_LIBPTHREAD_VERSION)
501 #error "glibc too old (< 2.3.2)"
502 #endif
503
504 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
505 assert(n > 0, "cannot retrieve glibc version");
506 char *str = (char *)malloc(n, mtInternal);
507 confstr(_CS_GNU_LIBC_VERSION, str, n);
508 os::Linux::set_glibc_version(str);
509
510 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
511 assert(n > 0, "cannot retrieve pthread version");
512 str = (char *)malloc(n, mtInternal);
513 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
514 os::Linux::set_libpthread_version(str);
515 }
516
517 /////////////////////////////////////////////////////////////////////////////
518 // thread stack expansion
519
520 // os::Linux::manually_expand_stack() takes care of expanding the thread
521 // stack. Note that this is normally not needed: pthread stacks allocate
522 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
523 // committed. Therefore it is not necessary to expand the stack manually.
524 //
525 // Manually expanding the stack was historically needed on LinuxThreads
526 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
527 // it is kept to deal with very rare corner cases:
528 //
529 // For one, user may run the VM on an own implementation of threads
530 // whose stacks are - like the old LinuxThreads - implemented using
531 // mmap(MAP_GROWSDOWN).
532 //
533 // Also, this coding may be needed if the VM is running on the primordial
534 // thread. Normally we avoid running on the primordial thread; however,
535 // user may still invoke the VM on the primordial thread.
536 //
537 // The following historical comment describes the details about running
538 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
539
540
541 // Force Linux kernel to expand current thread stack. If "bottom" is close
542 // to the stack guard, caller should block all signals.
543 //
544 // MAP_GROWSDOWN:
545 // A special mmap() flag that is used to implement thread stacks. It tells
546 // kernel that the memory region should extend downwards when needed. This
547 // allows early versions of LinuxThreads to only mmap the first few pages
548 // when creating a new thread. Linux kernel will automatically expand thread
549 // stack as needed (on page faults).
550 //
551 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
552 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
553 // region, it's hard to tell if the fault is due to a legitimate stack
554 // access or because of reading/writing non-exist memory (e.g. buffer
555 // overrun). As a rule, if the fault happens below current stack pointer,
556 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
557 // application (see Linux kernel fault.c).
558 //
559 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
560 // stack overflow detection.
561 //
562 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
563 // not use MAP_GROWSDOWN.
564 //
565 // To get around the problem and allow stack banging on Linux, we need to
566 // manually expand thread stack after receiving the SIGSEGV.
567 //
568 // There are two ways to expand thread stack to address "bottom", we used
569 // both of them in JVM before 1.5:
570 // 1. adjust stack pointer first so that it is below "bottom", and then
571 // touch "bottom"
572 // 2. mmap() the page in question
573 //
574 // Now alternate signal stack is gone, it's harder to use 2. For instance,
575 // if current sp is already near the lower end of page 101, and we need to
576 // call mmap() to map page 100, it is possible that part of the mmap() frame
577 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
578 // That will destroy the mmap() frame and cause VM to crash.
579 //
580 // The following code works by adjusting sp first, then accessing the "bottom"
581 // page to force a page fault. Linux kernel will then automatically expand the
582 // stack mapping.
583 //
584 // _expand_stack_to() assumes its frame size is less than page size, which
585 // should always be true if the function is not inlined.
586
587 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
588 #define NOINLINE
589 #else
590 #define NOINLINE __attribute__ ((noinline))
591 #endif
592
593 static void _expand_stack_to(address bottom) NOINLINE;
594
595 static void _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_yellow_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 *java_start(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 ThreadLocalStorage::set_thread(thread);
650
651 OSThread* osthread = thread->osthread();
652 Monitor* sync = osthread->startThread_lock();
653
654 osthread->set_thread_id(os::current_thread_id());
655
656 if (UseNUMA) {
657 int lgrp_id = os::numa_get_group_id();
658 if (lgrp_id != -1) {
659 thread->set_lgrp_id(lgrp_id);
660 }
661 }
662 // initialize signal mask for this thread
663 os::Linux::hotspot_sigmask(thread);
664
665 // initialize floating point control register
666 os::Linux::init_thread_fpu_state();
667
668 // handshaking with parent thread
669 {
670 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
671
672 // notify parent thread
673 osthread->set_state(INITIALIZED);
674 sync->notify_all();
675
676 // wait until os::start_thread()
677 while (osthread->get_state() == INITIALIZED) {
678 sync->wait(Mutex::_no_safepoint_check_flag);
679 }
680 }
681
682 // call one more level start routine
683 thread->run();
684
685 return 0;
686 }
687
688 bool os::create_thread(Thread* thread, ThreadType thr_type,
689 size_t stack_size) {
690 assert(thread->osthread() == NULL, "caller responsible");
691
692 // Allocate the OSThread object
693 OSThread* osthread = new OSThread(NULL, NULL);
694 if (osthread == NULL) {
695 return false;
696 }
697
698 // set the correct thread state
699 osthread->set_thread_type(thr_type);
700
701 // Initial state is ALLOCATED but not INITIALIZED
702 osthread->set_state(ALLOCATED);
703
704 thread->set_osthread(osthread);
705
706 // init thread attributes
707 pthread_attr_t attr;
708 pthread_attr_init(&attr);
709 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
710
711 // stack size
712 // calculate stack size if it's not specified by caller
713 if (stack_size == 0) {
714 stack_size = os::Linux::default_stack_size(thr_type);
715
716 switch (thr_type) {
717 case os::java_thread:
718 // Java threads use ThreadStackSize which default value can be
719 // changed with the flag -Xss
720 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
721 stack_size = JavaThread::stack_size_at_create();
722 break;
723 case os::compiler_thread:
724 if (CompilerThreadStackSize > 0) {
725 stack_size = (size_t)(CompilerThreadStackSize * K);
726 break;
727 } // else fall through:
728 // use VMThreadStackSize if CompilerThreadStackSize is not defined
729 case os::vm_thread:
730 case os::pgc_thread:
731 case os::cgc_thread:
732 case os::watcher_thread:
733 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
734 break;
735 }
736 }
737
738 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
739 pthread_attr_setstacksize(&attr, stack_size);
740
741 // glibc guard page
742 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
743
744 ThreadState state;
745
746 {
747 pthread_t tid;
748 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
749
750 pthread_attr_destroy(&attr);
751
752 if (ret != 0) {
753 if (PrintMiscellaneous && (Verbose || WizardMode)) {
754 perror("pthread_create()");
755 }
756 // Need to clean up stuff we've allocated so far
757 thread->set_osthread(NULL);
758 delete osthread;
759 return false;
760 }
761
762 // Store pthread info into the OSThread
763 osthread->set_pthread_id(tid);
764
765 // Wait until child thread is either initialized or aborted
766 {
767 Monitor* sync_with_child = osthread->startThread_lock();
768 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
769 while ((state = osthread->get_state()) == ALLOCATED) {
770 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
771 }
772 }
773 }
774
775 // Aborted due to thread limit being reached
776 if (state == ZOMBIE) {
777 thread->set_osthread(NULL);
778 delete osthread;
779 return false;
780 }
781
782 // The thread is returned suspended (in state INITIALIZED),
783 // and is started higher up in the call chain
784 assert(state == INITIALIZED, "race condition");
785 return true;
786 }
787
788 /////////////////////////////////////////////////////////////////////////////
789 // attach existing thread
790
791 // bootstrap the main thread
792 bool os::create_main_thread(JavaThread* thread) {
793 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
794 return create_attached_thread(thread);
795 }
796
797 bool os::create_attached_thread(JavaThread* thread) {
798 #ifdef ASSERT
799 thread->verify_not_published();
800 #endif
801
802 // Allocate the OSThread object
803 OSThread* osthread = new OSThread(NULL, NULL);
804
805 if (osthread == NULL) {
806 return false;
807 }
808
809 // Store pthread info into the OSThread
810 osthread->set_thread_id(os::Linux::gettid());
811 osthread->set_pthread_id(::pthread_self());
812
813 // initialize floating point control register
814 os::Linux::init_thread_fpu_state();
815
816 // Initial thread state is RUNNABLE
817 osthread->set_state(RUNNABLE);
818
819 thread->set_osthread(osthread);
820
821 if (UseNUMA) {
822 int lgrp_id = os::numa_get_group_id();
823 if (lgrp_id != -1) {
824 thread->set_lgrp_id(lgrp_id);
825 }
826 }
827
828 if (os::Linux::is_initial_thread()) {
829 // If current thread is initial thread, its stack is mapped on demand,
830 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
831 // the entire stack region to avoid SEGV in stack banging.
832 // It is also useful to get around the heap-stack-gap problem on SuSE
833 // kernel (see 4821821 for details). We first expand stack to the top
834 // of yellow zone, then enable stack yellow zone (order is significant,
835 // enabling yellow zone first will crash JVM on SuSE Linux), so there
836 // is no gap between the last two virtual memory regions.
837
838 JavaThread *jt = (JavaThread *)thread;
839 address addr = jt->stack_yellow_zone_base();
840 assert(addr != NULL, "initialization problem?");
841 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
842
843 osthread->set_expanding_stack();
844 os::Linux::manually_expand_stack(jt, addr);
845 osthread->clear_expanding_stack();
846 }
847
848 // initialize signal mask for this thread
849 // and save the caller's signal mask
850 os::Linux::hotspot_sigmask(thread);
851
852 return true;
853 }
854
855 void os::pd_start_thread(Thread* thread) {
856 OSThread * osthread = thread->osthread();
857 assert(osthread->get_state() != INITIALIZED, "just checking");
858 Monitor* sync_with_child = osthread->startThread_lock();
859 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
860 sync_with_child->notify();
861 }
862
863 // Free Linux resources related to the OSThread
864 void os::free_thread(OSThread* osthread) {
865 assert(osthread != NULL, "osthread not set");
866
867 if (Thread::current()->osthread() == osthread) {
868 // Restore caller's signal mask
869 sigset_t sigmask = osthread->caller_sigmask();
870 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
871 }
872
873 delete osthread;
874 }
875
876 //////////////////////////////////////////////////////////////////////////////
877 // thread local storage
878
879 // Restore the thread pointer if the destructor is called. This is in case
880 // someone from JNI code sets up a destructor with pthread_key_create to run
881 // detachCurrentThread on thread death. Unless we restore the thread pointer we
882 // will hang or crash. When detachCurrentThread is called the key will be set
883 // to null and we will not be called again. If detachCurrentThread is never
884 // called we could loop forever depending on the pthread implementation.
885 static void restore_thread_pointer(void* p) {
886 Thread* thread = (Thread*) p;
887 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
888 }
889
890 int os::allocate_thread_local_storage() {
891 pthread_key_t key;
892 int rslt = pthread_key_create(&key, restore_thread_pointer);
893 assert(rslt == 0, "cannot allocate thread local storage");
894 return (int)key;
895 }
896
897 // Note: This is currently not used by VM, as we don't destroy TLS key
898 // on VM exit.
899 void os::free_thread_local_storage(int index) {
900 int rslt = pthread_key_delete((pthread_key_t)index);
901 assert(rslt == 0, "invalid index");
902 }
903
904 void os::thread_local_storage_at_put(int index, void* value) {
905 int rslt = pthread_setspecific((pthread_key_t)index, value);
906 assert(rslt == 0, "pthread_setspecific failed");
907 }
908
909 extern "C" Thread* get_thread() {
910 return ThreadLocalStorage::thread();
911 }
912
913 //////////////////////////////////////////////////////////////////////////////
914 // initial thread
915
916 // Check if current thread is the initial thread, similar to Solaris thr_main.
917 bool os::Linux::is_initial_thread(void) {
918 char dummy;
919 // If called before init complete, thread stack bottom will be null.
920 // Can be called if fatal error occurs before initialization.
921 if (initial_thread_stack_bottom() == NULL) return false;
922 assert(initial_thread_stack_bottom() != NULL &&
923 initial_thread_stack_size() != 0,
924 "os::init did not locate initial thread's stack region");
925 if ((address)&dummy >= initial_thread_stack_bottom() &&
926 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
927 return true;
928 } else {
929 return false;
930 }
931 }
932
933 // Find the virtual memory area that contains addr
934 static bool find_vma(address addr, address* vma_low, address* vma_high) {
935 FILE *fp = fopen("/proc/self/maps", "r");
936 if (fp) {
937 address low, high;
938 while (!feof(fp)) {
939 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
940 if (low <= addr && addr < high) {
941 if (vma_low) *vma_low = low;
942 if (vma_high) *vma_high = high;
943 fclose(fp);
944 return true;
945 }
946 }
947 for (;;) {
948 int ch = fgetc(fp);
949 if (ch == EOF || ch == (int)'\n') break;
950 }
951 }
952 fclose(fp);
953 }
954 return false;
955 }
956
957 // Locate initial thread stack. This special handling of initial thread stack
958 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
959 // bogus value for initial thread.
960 void os::Linux::capture_initial_stack(size_t max_size) {
961 // stack size is the easy part, get it from RLIMIT_STACK
962 size_t stack_size;
963 struct rlimit rlim;
964 getrlimit(RLIMIT_STACK, &rlim);
965 stack_size = rlim.rlim_cur;
966
967 // 6308388: a bug in ld.so will relocate its own .data section to the
968 // lower end of primordial stack; reduce ulimit -s value a little bit
969 // so we won't install guard page on ld.so's data section.
970 stack_size -= 2 * page_size();
971
972 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
973 // 7.1, in both cases we will get 2G in return value.
974 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
975 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
976 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
977 // in case other parts in glibc still assumes 2M max stack size.
978 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
979 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
980 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
981 stack_size = 2 * K * K IA64_ONLY(*2);
982 }
983 // Try to figure out where the stack base (top) is. This is harder.
984 //
985 // When an application is started, glibc saves the initial stack pointer in
986 // a global variable "__libc_stack_end", which is then used by system
987 // libraries. __libc_stack_end should be pretty close to stack top. The
988 // variable is available since the very early days. However, because it is
989 // a private interface, it could disappear in the future.
990 //
991 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
992 // to __libc_stack_end, it is very close to stack top, but isn't the real
993 // stack top. Note that /proc may not exist if VM is running as a chroot
994 // program, so reading /proc/<pid>/stat could fail. Also the contents of
995 // /proc/<pid>/stat could change in the future (though unlikely).
996 //
997 // We try __libc_stack_end first. If that doesn't work, look for
998 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
999 // as a hint, which should work well in most cases.
1000
1001 uintptr_t stack_start;
1002
1003 // try __libc_stack_end first
1004 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1005 if (p && *p) {
1006 stack_start = *p;
1007 } else {
1008 // see if we can get the start_stack field from /proc/self/stat
1009 FILE *fp;
1010 int pid;
1011 char state;
1012 int ppid;
1013 int pgrp;
1014 int session;
1015 int nr;
1016 int tpgrp;
1017 unsigned long flags;
1018 unsigned long minflt;
1019 unsigned long cminflt;
1020 unsigned long majflt;
1021 unsigned long cmajflt;
1022 unsigned long utime;
1023 unsigned long stime;
1024 long cutime;
1025 long cstime;
1026 long prio;
1027 long nice;
1028 long junk;
1029 long it_real;
1030 uintptr_t start;
1031 uintptr_t vsize;
1032 intptr_t rss;
1033 uintptr_t rsslim;
1034 uintptr_t scodes;
1035 uintptr_t ecode;
1036 int i;
1037
1038 // Figure what the primordial thread stack base is. Code is inspired
1039 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1040 // followed by command name surrounded by parentheses, state, etc.
1041 char stat[2048];
1042 int statlen;
1043
1044 fp = fopen("/proc/self/stat", "r");
1045 if (fp) {
1046 statlen = fread(stat, 1, 2047, fp);
1047 stat[statlen] = '\0';
1048 fclose(fp);
1049
1050 // Skip pid and the command string. Note that we could be dealing with
1051 // weird command names, e.g. user could decide to rename java launcher
1052 // to "java 1.4.2 :)", then the stat file would look like
1053 // 1234 (java 1.4.2 :)) R ... ...
1054 // We don't really need to know the command string, just find the last
1055 // occurrence of ")" and then start parsing from there. See bug 4726580.
1056 char * s = strrchr(stat, ')');
1057
1058 i = 0;
1059 if (s) {
1060 // Skip blank chars
1061 do { s++; } while (s && isspace(*s));
1062
1063 #define _UFM UINTX_FORMAT
1064 #define _DFM INTX_FORMAT
1065
1066 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1067 // 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
1068 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,
1069 &state, // 3 %c
1070 &ppid, // 4 %d
1071 &pgrp, // 5 %d
1072 &session, // 6 %d
1073 &nr, // 7 %d
1074 &tpgrp, // 8 %d
1075 &flags, // 9 %lu
1076 &minflt, // 10 %lu
1077 &cminflt, // 11 %lu
1078 &majflt, // 12 %lu
1079 &cmajflt, // 13 %lu
1080 &utime, // 14 %lu
1081 &stime, // 15 %lu
1082 &cutime, // 16 %ld
1083 &cstime, // 17 %ld
1084 &prio, // 18 %ld
1085 &nice, // 19 %ld
1086 &junk, // 20 %ld
1087 &it_real, // 21 %ld
1088 &start, // 22 UINTX_FORMAT
1089 &vsize, // 23 UINTX_FORMAT
1090 &rss, // 24 INTX_FORMAT
1091 &rsslim, // 25 UINTX_FORMAT
1092 &scodes, // 26 UINTX_FORMAT
1093 &ecode, // 27 UINTX_FORMAT
1094 &stack_start); // 28 UINTX_FORMAT
1095 }
1096
1097 #undef _UFM
1098 #undef _DFM
1099
1100 if (i != 28 - 2) {
1101 assert(false, "Bad conversion from /proc/self/stat");
1102 // product mode - assume we are the initial thread, good luck in the
1103 // embedded case.
1104 warning("Can't detect initial thread stack location - bad conversion");
1105 stack_start = (uintptr_t) &rlim;
1106 }
1107 } else {
1108 // For some reason we can't open /proc/self/stat (for example, running on
1109 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1110 // most cases, so don't abort:
1111 warning("Can't detect initial thread stack location - no /proc/self/stat");
1112 stack_start = (uintptr_t) &rlim;
1113 }
1114 }
1115
1116 // Now we have a pointer (stack_start) very close to the stack top, the
1117 // next thing to do is to figure out the exact location of stack top. We
1118 // can find out the virtual memory area that contains stack_start by
1119 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1120 // and its upper limit is the real stack top. (again, this would fail if
1121 // running inside chroot, because /proc may not exist.)
1122
1123 uintptr_t stack_top;
1124 address low, high;
1125 if (find_vma((address)stack_start, &low, &high)) {
1126 // success, "high" is the true stack top. (ignore "low", because initial
1127 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1128 stack_top = (uintptr_t)high;
1129 } else {
1130 // failed, likely because /proc/self/maps does not exist
1131 warning("Can't detect initial thread stack location - find_vma failed");
1132 // best effort: stack_start is normally within a few pages below the real
1133 // stack top, use it as stack top, and reduce stack size so we won't put
1134 // guard page outside stack.
1135 stack_top = stack_start;
1136 stack_size -= 16 * page_size();
1137 }
1138
1139 // stack_top could be partially down the page so align it
1140 stack_top = align_size_up(stack_top, page_size());
1141
1142 if (max_size && stack_size > max_size) {
1143 _initial_thread_stack_size = max_size;
1144 } else {
1145 _initial_thread_stack_size = stack_size;
1146 }
1147
1148 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1149 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1150 }
1151
1152 ////////////////////////////////////////////////////////////////////////////////
1153 // time support
1154
1155 // Time since start-up in seconds to a fine granularity.
1156 // Used by VMSelfDestructTimer and the MemProfiler.
1157 double os::elapsedTime() {
1158
1159 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1160 }
1161
1162 jlong os::elapsed_counter() {
1163 return javaTimeNanos() - initial_time_count;
1164 }
1165
1166 jlong os::elapsed_frequency() {
1167 return NANOSECS_PER_SEC; // nanosecond resolution
1168 }
1169
1170 bool os::supports_vtime() { return true; }
1171 bool os::enable_vtime() { return false; }
1172 bool os::vtime_enabled() { return false; }
1173
1174 double os::elapsedVTime() {
1175 struct rusage usage;
1176 int retval = getrusage(RUSAGE_THREAD, &usage);
1177 if (retval == 0) {
1178 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);
1179 } else {
1180 // better than nothing, but not much
1181 return elapsedTime();
1182 }
1183 }
1184
1185 jlong os::javaTimeMillis() {
1186 timeval time;
1187 int status = gettimeofday(&time, NULL);
1188 assert(status != -1, "linux error");
1189 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1190 }
1191
1192 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1193 timeval time;
1194 int status = gettimeofday(&time, NULL);
1195 assert(status != -1, "linux error");
1196 seconds = jlong(time.tv_sec);
1197 nanos = jlong(time.tv_usec) * 1000;
1198 }
1199
1200
1201 #ifndef CLOCK_MONOTONIC
1202 #define CLOCK_MONOTONIC (1)
1203 #endif
1204
1205 void os::Linux::clock_init() {
1206 // we do dlopen's in this particular order due to bug in linux
1207 // dynamical loader (see 6348968) leading to crash on exit
1208 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1209 if (handle == NULL) {
1210 handle = dlopen("librt.so", RTLD_LAZY);
1211 }
1212
1213 if (handle) {
1214 int (*clock_getres_func)(clockid_t, struct timespec*) =
1215 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1216 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1217 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1218 if (clock_getres_func && clock_gettime_func) {
1219 // See if monotonic clock is supported by the kernel. Note that some
1220 // early implementations simply return kernel jiffies (updated every
1221 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1222 // for nano time (though the monotonic property is still nice to have).
1223 // It's fixed in newer kernels, however clock_getres() still returns
1224 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1225 // resolution for now. Hopefully as people move to new kernels, this
1226 // won't be a problem.
1227 struct timespec res;
1228 struct timespec tp;
1229 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1230 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1231 // yes, monotonic clock is supported
1232 _clock_gettime = clock_gettime_func;
1233 return;
1234 } else {
1235 // close librt if there is no monotonic clock
1236 dlclose(handle);
1237 }
1238 }
1239 }
1240 warning("No monotonic clock was available - timed services may " \
1241 "be adversely affected if the time-of-day clock changes");
1242 }
1243
1244 #ifndef SYS_clock_getres
1245 #if defined(IA32) || defined(AMD64)
1246 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1247 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1248 #else
1249 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1250 #define sys_clock_getres(x,y) -1
1251 #endif
1252 #else
1253 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1254 #endif
1255
1256 void os::Linux::fast_thread_clock_init() {
1257 if (!UseLinuxPosixThreadCPUClocks) {
1258 return;
1259 }
1260 clockid_t clockid;
1261 struct timespec tp;
1262 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1263 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1264
1265 // Switch to using fast clocks for thread cpu time if
1266 // the sys_clock_getres() returns 0 error code.
1267 // Note, that some kernels may support the current thread
1268 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1269 // returned by the pthread_getcpuclockid().
1270 // If the fast Posix clocks are supported then the sys_clock_getres()
1271 // must return at least tp.tv_sec == 0 which means a resolution
1272 // better than 1 sec. This is extra check for reliability.
1273
1274 if (pthread_getcpuclockid_func &&
1275 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1276 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1277 _supports_fast_thread_cpu_time = true;
1278 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1279 }
1280 }
1281
1282 jlong os::javaTimeNanos() {
1283 if (os::supports_monotonic_clock()) {
1284 struct timespec tp;
1285 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1286 assert(status == 0, "gettime error");
1287 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1288 return result;
1289 } else {
1290 timeval time;
1291 int status = gettimeofday(&time, NULL);
1292 assert(status != -1, "linux error");
1293 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1294 return 1000 * usecs;
1295 }
1296 }
1297
1298 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1299 if (os::supports_monotonic_clock()) {
1300 info_ptr->max_value = ALL_64_BITS;
1301
1302 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1303 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1304 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1305 } else {
1306 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1307 info_ptr->max_value = ALL_64_BITS;
1308
1309 // gettimeofday is a real time clock so it skips
1310 info_ptr->may_skip_backward = true;
1311 info_ptr->may_skip_forward = true;
1312 }
1313
1314 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1315 }
1316
1317 // Return the real, user, and system times in seconds from an
1318 // arbitrary fixed point in the past.
1319 bool os::getTimesSecs(double* process_real_time,
1320 double* process_user_time,
1321 double* process_system_time) {
1322 struct tms ticks;
1323 clock_t real_ticks = times(&ticks);
1324
1325 if (real_ticks == (clock_t) (-1)) {
1326 return false;
1327 } else {
1328 double ticks_per_second = (double) clock_tics_per_sec;
1329 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1330 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1331 *process_real_time = ((double) real_ticks) / ticks_per_second;
1332
1333 return true;
1334 }
1335 }
1336
1337
1338 char * os::local_time_string(char *buf, size_t buflen) {
1339 struct tm t;
1340 time_t long_time;
1341 time(&long_time);
1342 localtime_r(&long_time, &t);
1343 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1344 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1345 t.tm_hour, t.tm_min, t.tm_sec);
1346 return buf;
1347 }
1348
1349 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1350 return localtime_r(clock, res);
1351 }
1352
1353 ////////////////////////////////////////////////////////////////////////////////
1354 // runtime exit support
1355
1356 // Note: os::shutdown() might be called very early during initialization, or
1357 // called from signal handler. Before adding something to os::shutdown(), make
1358 // sure it is async-safe and can handle partially initialized VM.
1359 void os::shutdown() {
1360
1361 // allow PerfMemory to attempt cleanup of any persistent resources
1362 perfMemory_exit();
1363
1364 // needs to remove object in file system
1365 AttachListener::abort();
1366
1367 // flush buffered output, finish log files
1368 ostream_abort();
1369
1370 // Check for abort hook
1371 abort_hook_t abort_hook = Arguments::abort_hook();
1372 if (abort_hook != NULL) {
1373 abort_hook();
1374 }
1375
1376 }
1377
1378 // Note: os::abort() might be called very early during initialization, or
1379 // called from signal handler. Before adding something to os::abort(), make
1380 // sure it is async-safe and can handle partially initialized VM.
1381 void os::abort(bool dump_core, void* siginfo, void* context) {
1382 os::shutdown();
1383 if (dump_core) {
1384 #ifndef PRODUCT
1385 fdStream out(defaultStream::output_fd());
1386 out.print_raw("Current thread is ");
1387 char buf[16];
1388 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1389 out.print_raw_cr(buf);
1390 out.print_raw_cr("Dumping core ...");
1391 #endif
1392 ::abort(); // dump core
1393 }
1394
1395 ::exit(1);
1396 }
1397
1398 // Die immediately, no exit hook, no abort hook, no cleanup.
1399 void os::die() {
1400 ::abort();
1401 }
1402
1403
1404 // This method is a copy of JDK's sysGetLastErrorString
1405 // from src/solaris/hpi/src/system_md.c
1406
1407 size_t os::lasterror(char *buf, size_t len) {
1408 if (errno == 0) return 0;
1409
1410 const char *s = ::strerror(errno);
1411 size_t n = ::strlen(s);
1412 if (n >= len) {
1413 n = len - 1;
1414 }
1415 ::strncpy(buf, s, n);
1416 buf[n] = '\0';
1417 return n;
1418 }
1419
1420 // thread_id is kernel thread id (similar to Solaris LWP id)
1421 intx os::current_thread_id() { return os::Linux::gettid(); }
1422 int os::current_process_id() {
1423 return ::getpid();
1424 }
1425
1426 // DLL functions
1427
1428 const char* os::dll_file_extension() { return ".so"; }
1429
1430 // This must be hard coded because it's the system's temporary
1431 // directory not the java application's temp directory, ala java.io.tmpdir.
1432 const char* os::get_temp_directory() { return "/tmp"; }
1433
1434 static bool file_exists(const char* filename) {
1435 struct stat statbuf;
1436 if (filename == NULL || strlen(filename) == 0) {
1437 return false;
1438 }
1439 return os::stat(filename, &statbuf) == 0;
1440 }
1441
1442 bool os::dll_build_name(char* buffer, size_t buflen,
1443 const char* pname, const char* fname) {
1444 bool retval = false;
1445 // Copied from libhpi
1446 const size_t pnamelen = pname ? strlen(pname) : 0;
1447
1448 // Return error on buffer overflow.
1449 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1450 return retval;
1451 }
1452
1453 if (pnamelen == 0) {
1454 snprintf(buffer, buflen, "lib%s.so", fname);
1455 retval = true;
1456 } else if (strchr(pname, *os::path_separator()) != NULL) {
1457 int n;
1458 char** pelements = split_path(pname, &n);
1459 if (pelements == NULL) {
1460 return false;
1461 }
1462 for (int i = 0; i < n; i++) {
1463 // Really shouldn't be NULL, but check can't hurt
1464 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1465 continue; // skip the empty path values
1466 }
1467 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1468 if (file_exists(buffer)) {
1469 retval = true;
1470 break;
1471 }
1472 }
1473 // release the storage
1474 for (int i = 0; i < n; i++) {
1475 if (pelements[i] != NULL) {
1476 FREE_C_HEAP_ARRAY(char, pelements[i]);
1477 }
1478 }
1479 if (pelements != NULL) {
1480 FREE_C_HEAP_ARRAY(char*, pelements);
1481 }
1482 } else {
1483 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1484 retval = true;
1485 }
1486 return retval;
1487 }
1488
1489 // check if addr is inside libjvm.so
1490 bool os::address_is_in_vm(address addr) {
1491 static address libjvm_base_addr;
1492 Dl_info dlinfo;
1493
1494 if (libjvm_base_addr == NULL) {
1495 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1496 libjvm_base_addr = (address)dlinfo.dli_fbase;
1497 }
1498 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1499 }
1500
1501 if (dladdr((void *)addr, &dlinfo) != 0) {
1502 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1503 }
1504
1505 return false;
1506 }
1507
1508 bool os::dll_address_to_function_name(address addr, char *buf,
1509 int buflen, int *offset,
1510 bool demangle) {
1511 // buf is not optional, but offset is optional
1512 assert(buf != NULL, "sanity check");
1513
1514 Dl_info dlinfo;
1515
1516 if (dladdr((void*)addr, &dlinfo) != 0) {
1517 // see if we have a matching symbol
1518 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1519 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1520 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1521 }
1522 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1523 return true;
1524 }
1525 // no matching symbol so try for just file info
1526 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1527 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1528 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1529 return true;
1530 }
1531 }
1532 }
1533
1534 buf[0] = '\0';
1535 if (offset != NULL) *offset = -1;
1536 return false;
1537 }
1538
1539 struct _address_to_library_name {
1540 address addr; // input : memory address
1541 size_t buflen; // size of fname
1542 char* fname; // output: library name
1543 address base; // library base addr
1544 };
1545
1546 static int address_to_library_name_callback(struct dl_phdr_info *info,
1547 size_t size, void *data) {
1548 int i;
1549 bool found = false;
1550 address libbase = NULL;
1551 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1552
1553 // iterate through all loadable segments
1554 for (i = 0; i < info->dlpi_phnum; i++) {
1555 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1556 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1557 // base address of a library is the lowest address of its loaded
1558 // segments.
1559 if (libbase == NULL || libbase > segbase) {
1560 libbase = segbase;
1561 }
1562 // see if 'addr' is within current segment
1563 if (segbase <= d->addr &&
1564 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1565 found = true;
1566 }
1567 }
1568 }
1569
1570 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1571 // so dll_address_to_library_name() can fall through to use dladdr() which
1572 // can figure out executable name from argv[0].
1573 if (found && info->dlpi_name && info->dlpi_name[0]) {
1574 d->base = libbase;
1575 if (d->fname) {
1576 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1577 }
1578 return 1;
1579 }
1580 return 0;
1581 }
1582
1583 bool os::dll_address_to_library_name(address addr, char* buf,
1584 int buflen, int* offset) {
1585 // buf is not optional, but offset is optional
1586 assert(buf != NULL, "sanity check");
1587
1588 Dl_info dlinfo;
1589 struct _address_to_library_name data;
1590
1591 // There is a bug in old glibc dladdr() implementation that it could resolve
1592 // to wrong library name if the .so file has a base address != NULL. Here
1593 // we iterate through the program headers of all loaded libraries to find
1594 // out which library 'addr' really belongs to. This workaround can be
1595 // removed once the minimum requirement for glibc is moved to 2.3.x.
1596 data.addr = addr;
1597 data.fname = buf;
1598 data.buflen = buflen;
1599 data.base = NULL;
1600 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1601
1602 if (rslt) {
1603 // buf already contains library name
1604 if (offset) *offset = addr - data.base;
1605 return true;
1606 }
1607 if (dladdr((void*)addr, &dlinfo) != 0) {
1608 if (dlinfo.dli_fname != NULL) {
1609 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1610 }
1611 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1612 *offset = addr - (address)dlinfo.dli_fbase;
1613 }
1614 return true;
1615 }
1616
1617 buf[0] = '\0';
1618 if (offset) *offset = -1;
1619 return false;
1620 }
1621
1622 // Loads .dll/.so and
1623 // in case of error it checks if .dll/.so was built for the
1624 // same architecture as Hotspot is running on
1625
1626
1627 // Remember the stack's state. The Linux dynamic linker will change
1628 // the stack to 'executable' at most once, so we must safepoint only once.
1629 bool os::Linux::_stack_is_executable = false;
1630
1631 // VM operation that loads a library. This is necessary if stack protection
1632 // of the Java stacks can be lost during loading the library. If we
1633 // do not stop the Java threads, they can stack overflow before the stacks
1634 // are protected again.
1635 class VM_LinuxDllLoad: public VM_Operation {
1636 private:
1637 const char *_filename;
1638 char *_ebuf;
1639 int _ebuflen;
1640 void *_lib;
1641 public:
1642 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1643 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1644 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1645 void doit() {
1646 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1647 os::Linux::_stack_is_executable = true;
1648 }
1649 void* loaded_library() { return _lib; }
1650 };
1651
1652 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1653 void * result = NULL;
1654 bool load_attempted = false;
1655
1656 // Check whether the library to load might change execution rights
1657 // of the stack. If they are changed, the protection of the stack
1658 // guard pages will be lost. We need a safepoint to fix this.
1659 //
1660 // See Linux man page execstack(8) for more info.
1661 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1662 ElfFile ef(filename);
1663 if (!ef.specifies_noexecstack()) {
1664 if (!is_init_completed()) {
1665 os::Linux::_stack_is_executable = true;
1666 // This is OK - No Java threads have been created yet, and hence no
1667 // stack guard pages to fix.
1668 //
1669 // This should happen only when you are building JDK7 using a very
1670 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1671 //
1672 // Dynamic loader will make all stacks executable after
1673 // this function returns, and will not do that again.
1674 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1675 } else {
1676 warning("You have loaded library %s which might have disabled stack guard. "
1677 "The VM will try to fix the stack guard now.\n"
1678 "It's highly recommended that you fix the library with "
1679 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1680 filename);
1681
1682 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1683 JavaThread *jt = JavaThread::current();
1684 if (jt->thread_state() != _thread_in_native) {
1685 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1686 // that requires ExecStack. Cannot enter safe point. Let's give up.
1687 warning("Unable to fix stack guard. Giving up.");
1688 } else {
1689 if (!LoadExecStackDllInVMThread) {
1690 // This is for the case where the DLL has an static
1691 // constructor function that executes JNI code. We cannot
1692 // load such DLLs in the VMThread.
1693 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1694 }
1695
1696 ThreadInVMfromNative tiv(jt);
1697 debug_only(VMNativeEntryWrapper vew;)
1698
1699 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1700 VMThread::execute(&op);
1701 if (LoadExecStackDllInVMThread) {
1702 result = op.loaded_library();
1703 }
1704 load_attempted = true;
1705 }
1706 }
1707 }
1708 }
1709
1710 if (!load_attempted) {
1711 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1712 }
1713
1714 if (result != NULL) {
1715 // Successful loading
1716 return result;
1717 }
1718
1719 Elf32_Ehdr elf_head;
1720 int diag_msg_max_length=ebuflen-strlen(ebuf);
1721 char* diag_msg_buf=ebuf+strlen(ebuf);
1722
1723 if (diag_msg_max_length==0) {
1724 // No more space in ebuf for additional diagnostics message
1725 return NULL;
1726 }
1727
1728
1729 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1730
1731 if (file_descriptor < 0) {
1732 // Can't open library, report dlerror() message
1733 return NULL;
1734 }
1735
1736 bool failed_to_read_elf_head=
1737 (sizeof(elf_head)!=
1738 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1739
1740 ::close(file_descriptor);
1741 if (failed_to_read_elf_head) {
1742 // file i/o error - report dlerror() msg
1743 return NULL;
1744 }
1745
1746 typedef struct {
1747 Elf32_Half code; // Actual value as defined in elf.h
1748 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1749 char elf_class; // 32 or 64 bit
1750 char endianess; // MSB or LSB
1751 char* name; // String representation
1752 } arch_t;
1753
1754 #ifndef EM_486
1755 #define EM_486 6 /* Intel 80486 */
1756 #endif
1757 #ifndef EM_AARCH64
1758 #define EM_AARCH64 183 /* ARM AARCH64 */
1759 #endif
1760
1761 static const arch_t arch_array[]={
1762 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1763 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1764 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1765 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1766 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1767 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1768 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1769 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1770 #if defined(VM_LITTLE_ENDIAN)
1771 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1772 #else
1773 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1774 #endif
1775 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1776 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1777 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1778 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1779 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1780 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1781 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1782 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1783 };
1784
1785 #if (defined IA32)
1786 static Elf32_Half running_arch_code=EM_386;
1787 #elif (defined AMD64)
1788 static Elf32_Half running_arch_code=EM_X86_64;
1789 #elif (defined IA64)
1790 static Elf32_Half running_arch_code=EM_IA_64;
1791 #elif (defined __sparc) && (defined _LP64)
1792 static Elf32_Half running_arch_code=EM_SPARCV9;
1793 #elif (defined __sparc) && (!defined _LP64)
1794 static Elf32_Half running_arch_code=EM_SPARC;
1795 #elif (defined __powerpc64__)
1796 static Elf32_Half running_arch_code=EM_PPC64;
1797 #elif (defined __powerpc__)
1798 static Elf32_Half running_arch_code=EM_PPC;
1799 #elif (defined ARM)
1800 static Elf32_Half running_arch_code=EM_ARM;
1801 #elif (defined S390)
1802 static Elf32_Half running_arch_code=EM_S390;
1803 #elif (defined ALPHA)
1804 static Elf32_Half running_arch_code=EM_ALPHA;
1805 #elif (defined MIPSEL)
1806 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1807 #elif (defined PARISC)
1808 static Elf32_Half running_arch_code=EM_PARISC;
1809 #elif (defined MIPS)
1810 static Elf32_Half running_arch_code=EM_MIPS;
1811 #elif (defined M68K)
1812 static Elf32_Half running_arch_code=EM_68K;
1813 #elif (defined AARCH64)
1814 static Elf32_Half running_arch_code=EM_AARCH64;
1815 #else
1816 #error Method os::dll_load requires that one of following is defined:\
1817 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1818 #endif
1819
1820 // Identify compatability class for VM's architecture and library's architecture
1821 // Obtain string descriptions for architectures
1822
1823 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1824 int running_arch_index=-1;
1825
1826 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1827 if (running_arch_code == arch_array[i].code) {
1828 running_arch_index = i;
1829 }
1830 if (lib_arch.code == arch_array[i].code) {
1831 lib_arch.compat_class = arch_array[i].compat_class;
1832 lib_arch.name = arch_array[i].name;
1833 }
1834 }
1835
1836 assert(running_arch_index != -1,
1837 "Didn't find running architecture code (running_arch_code) in arch_array");
1838 if (running_arch_index == -1) {
1839 // Even though running architecture detection failed
1840 // we may still continue with reporting dlerror() message
1841 return NULL;
1842 }
1843
1844 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1845 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1846 return NULL;
1847 }
1848
1849 #ifndef S390
1850 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1851 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1852 return NULL;
1853 }
1854 #endif // !S390
1855
1856 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1857 if (lib_arch.name!=NULL) {
1858 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1859 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1860 lib_arch.name, arch_array[running_arch_index].name);
1861 } else {
1862 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1863 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1864 lib_arch.code,
1865 arch_array[running_arch_index].name);
1866 }
1867 }
1868
1869 return NULL;
1870 }
1871
1872 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1873 int ebuflen) {
1874 void * result = ::dlopen(filename, RTLD_LAZY);
1875 if (result == NULL) {
1876 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1877 ebuf[ebuflen-1] = '\0';
1878 }
1879 return result;
1880 }
1881
1882 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1883 int ebuflen) {
1884 void * result = NULL;
1885 if (LoadExecStackDllInVMThread) {
1886 result = dlopen_helper(filename, ebuf, ebuflen);
1887 }
1888
1889 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1890 // library that requires an executable stack, or which does not have this
1891 // stack attribute set, dlopen changes the stack attribute to executable. The
1892 // read protection of the guard pages gets lost.
1893 //
1894 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1895 // may have been queued at the same time.
1896
1897 if (!_stack_is_executable) {
1898 JavaThread *jt = Threads::first();
1899
1900 while (jt) {
1901 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1902 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
1903 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1904 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1905 warning("Attempt to reguard stack yellow zone failed.");
1906 }
1907 }
1908 jt = jt->next();
1909 }
1910 }
1911
1912 return result;
1913 }
1914
1915 void* os::dll_lookup(void* handle, const char* name) {
1916 void* res = dlsym(handle, name);
1917 return res;
1918 }
1919
1920 void* os::get_default_process_handle() {
1921 return (void*)::dlopen(NULL, RTLD_LAZY);
1922 }
1923
1924 static bool _print_ascii_file(const char* filename, outputStream* st) {
1925 int fd = ::open(filename, O_RDONLY);
1926 if (fd == -1) {
1927 return false;
1928 }
1929
1930 char buf[32];
1931 int bytes;
1932 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1933 st->print_raw(buf, bytes);
1934 }
1935
1936 ::close(fd);
1937
1938 return true;
1939 }
1940
1941 void os::print_dll_info(outputStream *st) {
1942 st->print_cr("Dynamic libraries:");
1943
1944 char fname[32];
1945 pid_t pid = os::Linux::gettid();
1946
1947 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1948
1949 if (!_print_ascii_file(fname, st)) {
1950 st->print("Can not get library information for pid = %d\n", pid);
1951 }
1952 }
1953
1954 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1955 FILE *procmapsFile = NULL;
1956
1957 // Open the procfs maps file for the current process
1958 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1959 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1960 char line[PATH_MAX + 100];
1961
1962 // Read line by line from 'file'
1963 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1964 u8 base, top, offset, inode;
1965 char permissions[5];
1966 char device[6];
1967 char name[PATH_MAX + 1];
1968
1969 // Parse fields from line
1970 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1971 &base, &top, permissions, &offset, device, &inode, name);
1972
1973 // Filter by device id '00:00' so that we only get file system mapped files.
1974 if (strcmp(device, "00:00") != 0) {
1975
1976 // Call callback with the fields of interest
1977 if(callback(name, (address)base, (address)top, param)) {
1978 // Oops abort, callback aborted
1979 fclose(procmapsFile);
1980 return 1;
1981 }
1982 }
1983 }
1984 fclose(procmapsFile);
1985 }
1986 return 0;
1987 }
1988
1989 void os::print_os_info_brief(outputStream* st) {
1990 os::Linux::print_distro_info(st);
1991
1992 os::Posix::print_uname_info(st);
1993
1994 os::Linux::print_libversion_info(st);
1995
1996 }
1997
1998 void os::print_os_info(outputStream* st) {
1999 st->print("OS:");
2000
2001 os::Linux::print_distro_info(st);
2002
2003 os::Posix::print_uname_info(st);
2004
2005 // Print warning if unsafe chroot environment detected
2006 if (unsafe_chroot_detected) {
2007 st->print("WARNING!! ");
2008 st->print_cr("%s", unstable_chroot_error);
2009 }
2010
2011 os::Linux::print_libversion_info(st);
2012
2013 os::Posix::print_rlimit_info(st);
2014
2015 os::Posix::print_load_average(st);
2016
2017 os::Linux::print_full_memory_info(st);
2018 }
2019
2020 // Try to identify popular distros.
2021 // Most Linux distributions have a /etc/XXX-release file, which contains
2022 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2023 // file that also contains the OS version string. Some have more than one
2024 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2025 // /etc/redhat-release.), so the order is important.
2026 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2027 // their own specific XXX-release file as well as a redhat-release file.
2028 // Because of this the XXX-release file needs to be searched for before the
2029 // redhat-release file.
2030 // Since Red Hat has a lsb-release file that is not very descriptive the
2031 // search for redhat-release needs to be before lsb-release.
2032 // Since the lsb-release file is the new standard it needs to be searched
2033 // before the older style release files.
2034 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2035 // next to last resort. The os-release file is a new standard that contains
2036 // distribution information and the system-release file seems to be an old
2037 // standard that has been replaced by the lsb-release and os-release files.
2038 // Searching for the debian_version file is the last resort. It contains
2039 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2040 // "Debian " is printed before the contents of the debian_version file.
2041
2042 const char* distro_files[] = {
2043 "/etc/oracle-release",
2044 "/etc/mandriva-release",
2045 "/etc/mandrake-release",
2046 "/etc/sun-release",
2047 "/etc/redhat-release",
2048 "/etc/lsb-release",
2049 "/etc/SuSE-release",
2050 "/etc/turbolinux-release",
2051 "/etc/gentoo-release",
2052 "/etc/ltib-release",
2053 "/etc/angstrom-version",
2054 "/etc/system-release",
2055 "/etc/os-release",
2056 NULL };
2057
2058 void os::Linux::print_distro_info(outputStream* st) {
2059 for (int i = 0;; i++) {
2060 const char* file = distro_files[i];
2061 if (file == NULL) {
2062 break; // done
2063 }
2064 // If file prints, we found it.
2065 if (_print_ascii_file(file, st)) {
2066 return;
2067 }
2068 }
2069
2070 if (file_exists("/etc/debian_version")) {
2071 st->print("Debian ");
2072 _print_ascii_file("/etc/debian_version", st);
2073 } else {
2074 st->print("Linux");
2075 }
2076 st->cr();
2077 }
2078
2079 static void parse_os_info(char* distro, size_t length, const char* file) {
2080 FILE* fp = fopen(file, "r");
2081 if (fp != NULL) {
2082 char buf[256];
2083 // get last line of the file.
2084 while (fgets(buf, sizeof(buf), fp)) { }
2085 // Edit out extra stuff in expected ubuntu format
2086 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2087 char* ptr = strstr(buf, "\""); // the name is in quotes
2088 if (ptr != NULL) {
2089 ptr++; // go beyond first quote
2090 char* nl = strchr(ptr, '\"');
2091 if (nl != NULL) *nl = '\0';
2092 strncpy(distro, ptr, length);
2093 } else {
2094 ptr = strstr(buf, "=");
2095 ptr++; // go beyond equals then
2096 char* nl = strchr(ptr, '\n');
2097 if (nl != NULL) *nl = '\0';
2098 strncpy(distro, ptr, length);
2099 }
2100 } else {
2101 // if not in expected Ubuntu format, print out whole line minus \n
2102 char* nl = strchr(buf, '\n');
2103 if (nl != NULL) *nl = '\0';
2104 strncpy(distro, buf, length);
2105 }
2106 // close distro file
2107 fclose(fp);
2108 }
2109 }
2110
2111 void os::get_summary_os_info(char* buf, size_t buflen) {
2112 for (int i = 0;; i++) {
2113 const char* file = distro_files[i];
2114 if (file == NULL) {
2115 break; // ran out of distro_files
2116 }
2117 if (file_exists(file)) {
2118 parse_os_info(buf, buflen, file);
2119 return;
2120 }
2121 }
2122 // special case for debian
2123 if (file_exists("/etc/debian_version")) {
2124 strncpy(buf, "Debian ", buflen);
2125 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2126 } else {
2127 strncpy(buf, "Linux", buflen);
2128 }
2129 }
2130
2131 void os::Linux::print_libversion_info(outputStream* st) {
2132 // libc, pthread
2133 st->print("libc:");
2134 st->print("%s ", os::Linux::glibc_version());
2135 st->print("%s ", os::Linux::libpthread_version());
2136 st->cr();
2137 }
2138
2139 void os::Linux::print_full_memory_info(outputStream* st) {
2140 st->print("\n/proc/meminfo:\n");
2141 _print_ascii_file("/proc/meminfo", st);
2142 st->cr();
2143 }
2144
2145 void os::print_memory_info(outputStream* st) {
2146
2147 st->print("Memory:");
2148 st->print(" %dk page", os::vm_page_size()>>10);
2149
2150 // values in struct sysinfo are "unsigned long"
2151 struct sysinfo si;
2152 sysinfo(&si);
2153
2154 st->print(", physical " UINT64_FORMAT "k",
2155 os::physical_memory() >> 10);
2156 st->print("(" UINT64_FORMAT "k free)",
2157 os::available_memory() >> 10);
2158 st->print(", swap " UINT64_FORMAT "k",
2159 ((jlong)si.totalswap * si.mem_unit) >> 10);
2160 st->print("(" UINT64_FORMAT "k free)",
2161 ((jlong)si.freeswap * si.mem_unit) >> 10);
2162 st->cr();
2163 }
2164
2165 // Print the first "model name" line and the first "flags" line
2166 // that we find and nothing more. We assume "model name" comes
2167 // before "flags" so if we find a second "model name", then the
2168 // "flags" field is considered missing.
2169 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2170 #if defined(IA32) || defined(AMD64)
2171 // Other platforms have less repetitive cpuinfo files
2172 FILE *fp = fopen("/proc/cpuinfo", "r");
2173 if (fp) {
2174 while (!feof(fp)) {
2175 if (fgets(buf, buflen, fp)) {
2176 // Assume model name comes before flags
2177 bool model_name_printed = false;
2178 if (strstr(buf, "model name") != NULL) {
2179 if (!model_name_printed) {
2180 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2181 st->print_raw(buf);
2182 model_name_printed = true;
2183 } else {
2184 // model name printed but not flags? Odd, just return
2185 fclose(fp);
2186 return true;
2187 }
2188 }
2189 // print the flags line too
2190 if (strstr(buf, "flags") != NULL) {
2191 st->print_raw(buf);
2192 fclose(fp);
2193 return true;
2194 }
2195 }
2196 }
2197 fclose(fp);
2198 }
2199 #endif // x86 platforms
2200 return false;
2201 }
2202
2203 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2204 // Only print the model name if the platform provides this as a summary
2205 if (!print_model_name_and_flags(st, buf, buflen)) {
2206 st->print("\n/proc/cpuinfo:\n");
2207 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2208 st->print_cr(" <Not Available>");
2209 }
2210 }
2211 }
2212
2213 #if defined(AMD64) || defined(IA32) || defined(X32)
2214 const char* search_string = "model name";
2215 #elif defined(SPARC)
2216 const char* search_string = "cpu";
2217 #else
2218 const char* search_string = "Processor";
2219 #endif
2220
2221 // Parses the cpuinfo file for string representing the model name.
2222 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2223 FILE* fp = fopen("/proc/cpuinfo", "r");
2224 if (fp != NULL) {
2225 while (!feof(fp)) {
2226 char buf[256];
2227 if (fgets(buf, sizeof(buf), fp)) {
2228 char* start = strstr(buf, search_string);
2229 if (start != NULL) {
2230 char *ptr = start + strlen(search_string);
2231 char *end = buf + strlen(buf);
2232 while (ptr != end) {
2233 // skip whitespace and colon for the rest of the name.
2234 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2235 break;
2236 }
2237 ptr++;
2238 }
2239 if (ptr != end) {
2240 // reasonable string, get rid of newline and keep the rest
2241 char* nl = strchr(buf, '\n');
2242 if (nl != NULL) *nl = '\0';
2243 strncpy(cpuinfo, ptr, length);
2244 fclose(fp);
2245 return;
2246 }
2247 }
2248 }
2249 }
2250 fclose(fp);
2251 }
2252 // cpuinfo not found or parsing failed, just print generic string. The entire
2253 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2254 #if defined(AMD64)
2255 strncpy(cpuinfo, "x86_64", length);
2256 #elif defined(IA32)
2257 strncpy(cpuinfo, "x86_32", length);
2258 #elif defined(IA64)
2259 strncpy(cpuinfo, "IA64", length);
2260 #elif defined(SPARC)
2261 strncpy(cpuinfo, "sparcv9", length);
2262 #elif defined(AARCH64)
2263 strncpy(cpuinfo, "AArch64", length);
2264 #elif defined(ARM)
2265 strncpy(cpuinfo, "ARM", length);
2266 #elif defined(PPC)
2267 strncpy(cpuinfo, "PPC64", length);
2268 #elif defined(ZERO_LIBARCH)
2269 strncpy(cpuinfo, ZERO_LIBARCH, length);
2270 #else
2271 strncpy(cpuinfo, "unknown", length);
2272 #endif
2273 }
2274
2275 void os::print_siginfo(outputStream* st, void* siginfo) {
2276 const siginfo_t* si = (const siginfo_t*)siginfo;
2277
2278 os::Posix::print_siginfo_brief(st, si);
2279 #if INCLUDE_CDS
2280 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2281 UseSharedSpaces) {
2282 FileMapInfo* mapinfo = FileMapInfo::current_info();
2283 if (mapinfo->is_in_shared_space(si->si_addr)) {
2284 st->print("\n\nError accessing class data sharing archive." \
2285 " Mapped file inaccessible during execution, " \
2286 " possible disk/network problem.");
2287 }
2288 }
2289 #endif
2290 st->cr();
2291 }
2292
2293
2294 static void print_signal_handler(outputStream* st, int sig,
2295 char* buf, size_t buflen);
2296
2297 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2298 st->print_cr("Signal Handlers:");
2299 print_signal_handler(st, SIGSEGV, buf, buflen);
2300 print_signal_handler(st, SIGBUS , buf, buflen);
2301 print_signal_handler(st, SIGFPE , buf, buflen);
2302 print_signal_handler(st, SIGPIPE, buf, buflen);
2303 print_signal_handler(st, SIGXFSZ, buf, buflen);
2304 print_signal_handler(st, SIGILL , buf, buflen);
2305 print_signal_handler(st, SR_signum, buf, buflen);
2306 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2307 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2308 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2309 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2310 #if defined(PPC64)
2311 print_signal_handler(st, SIGTRAP, buf, buflen);
2312 #endif
2313 }
2314
2315 static char saved_jvm_path[MAXPATHLEN] = {0};
2316
2317 // Find the full path to the current module, libjvm.so
2318 void os::jvm_path(char *buf, jint buflen) {
2319 // Error checking.
2320 if (buflen < MAXPATHLEN) {
2321 assert(false, "must use a large-enough buffer");
2322 buf[0] = '\0';
2323 return;
2324 }
2325 // Lazy resolve the path to current module.
2326 if (saved_jvm_path[0] != 0) {
2327 strcpy(buf, saved_jvm_path);
2328 return;
2329 }
2330
2331 char dli_fname[MAXPATHLEN];
2332 bool ret = dll_address_to_library_name(
2333 CAST_FROM_FN_PTR(address, os::jvm_path),
2334 dli_fname, sizeof(dli_fname), NULL);
2335 assert(ret, "cannot locate libjvm");
2336 char *rp = NULL;
2337 if (ret && dli_fname[0] != '\0') {
2338 rp = realpath(dli_fname, buf);
2339 }
2340 if (rp == NULL) {
2341 return;
2342 }
2343
2344 if (Arguments::sun_java_launcher_is_altjvm()) {
2345 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2346 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2347 // If "/jre/lib/" appears at the right place in the string, then
2348 // assume we are installed in a JDK and we're done. Otherwise, check
2349 // for a JAVA_HOME environment variable and fix up the path so it
2350 // looks like libjvm.so is installed there (append a fake suffix
2351 // hotspot/libjvm.so).
2352 const char *p = buf + strlen(buf) - 1;
2353 for (int count = 0; p > buf && count < 5; ++count) {
2354 for (--p; p > buf && *p != '/'; --p)
2355 /* empty */ ;
2356 }
2357
2358 if (strncmp(p, "/jre/lib/", 9) != 0) {
2359 // Look for JAVA_HOME in the environment.
2360 char* java_home_var = ::getenv("JAVA_HOME");
2361 if (java_home_var != NULL && java_home_var[0] != 0) {
2362 char* jrelib_p;
2363 int len;
2364
2365 // Check the current module name "libjvm.so".
2366 p = strrchr(buf, '/');
2367 if (p == NULL) {
2368 return;
2369 }
2370 assert(strstr(p, "/libjvm") == p, "invalid library name");
2371
2372 rp = realpath(java_home_var, buf);
2373 if (rp == NULL) {
2374 return;
2375 }
2376
2377 // determine if this is a legacy image or modules image
2378 // modules image doesn't have "jre" subdirectory
2379 len = strlen(buf);
2380 assert(len < buflen, "Ran out of buffer room");
2381 jrelib_p = buf + len;
2382 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2383 if (0 != access(buf, F_OK)) {
2384 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2385 }
2386
2387 if (0 == access(buf, F_OK)) {
2388 // Use current module name "libjvm.so"
2389 len = strlen(buf);
2390 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2391 } else {
2392 // Go back to path of .so
2393 rp = realpath(dli_fname, buf);
2394 if (rp == NULL) {
2395 return;
2396 }
2397 }
2398 }
2399 }
2400 }
2401
2402 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2403 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2404 }
2405
2406 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2407 // no prefix required, not even "_"
2408 }
2409
2410 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2411 // no suffix required
2412 }
2413
2414 ////////////////////////////////////////////////////////////////////////////////
2415 // sun.misc.Signal support
2416
2417 static volatile jint sigint_count = 0;
2418
2419 static void UserHandler(int sig, void *siginfo, void *context) {
2420 // 4511530 - sem_post is serialized and handled by the manager thread. When
2421 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2422 // don't want to flood the manager thread with sem_post requests.
2423 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2424 return;
2425 }
2426
2427 // Ctrl-C is pressed during error reporting, likely because the error
2428 // handler fails to abort. Let VM die immediately.
2429 if (sig == SIGINT && is_error_reported()) {
2430 os::die();
2431 }
2432
2433 os::signal_notify(sig);
2434 }
2435
2436 void* os::user_handler() {
2437 return CAST_FROM_FN_PTR(void*, UserHandler);
2438 }
2439
2440 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2441 struct timespec ts;
2442 // Semaphore's are always associated with CLOCK_REALTIME
2443 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2444 // see unpackTime for discussion on overflow checking
2445 if (sec >= MAX_SECS) {
2446 ts.tv_sec += MAX_SECS;
2447 ts.tv_nsec = 0;
2448 } else {
2449 ts.tv_sec += sec;
2450 ts.tv_nsec += nsec;
2451 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2452 ts.tv_nsec -= NANOSECS_PER_SEC;
2453 ++ts.tv_sec; // note: this must be <= max_secs
2454 }
2455 }
2456
2457 return ts;
2458 }
2459
2460 extern "C" {
2461 typedef void (*sa_handler_t)(int);
2462 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2463 }
2464
2465 void* os::signal(int signal_number, void* handler) {
2466 struct sigaction sigAct, oldSigAct;
2467
2468 sigfillset(&(sigAct.sa_mask));
2469 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2470 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2471
2472 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2473 // -1 means registration failed
2474 return (void *)-1;
2475 }
2476
2477 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2478 }
2479
2480 void os::signal_raise(int signal_number) {
2481 ::raise(signal_number);
2482 }
2483
2484 // The following code is moved from os.cpp for making this
2485 // code platform specific, which it is by its very nature.
2486
2487 // Will be modified when max signal is changed to be dynamic
2488 int os::sigexitnum_pd() {
2489 return NSIG;
2490 }
2491
2492 // a counter for each possible signal value
2493 static volatile jint pending_signals[NSIG+1] = { 0 };
2494
2495 // Linux(POSIX) specific hand shaking semaphore.
2496 static sem_t sig_sem;
2497 static PosixSemaphore sr_semaphore;
2498
2499 void os::signal_init_pd() {
2500 // Initialize signal structures
2501 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2502
2503 // Initialize signal semaphore
2504 ::sem_init(&sig_sem, 0, 0);
2505 }
2506
2507 void os::signal_notify(int sig) {
2508 Atomic::inc(&pending_signals[sig]);
2509 ::sem_post(&sig_sem);
2510 }
2511
2512 static int check_pending_signals(bool wait) {
2513 Atomic::store(0, &sigint_count);
2514 for (;;) {
2515 for (int i = 0; i < NSIG + 1; i++) {
2516 jint n = pending_signals[i];
2517 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2518 return i;
2519 }
2520 }
2521 if (!wait) {
2522 return -1;
2523 }
2524 JavaThread *thread = JavaThread::current();
2525 ThreadBlockInVM tbivm(thread);
2526
2527 bool threadIsSuspended;
2528 do {
2529 thread->set_suspend_equivalent();
2530 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2531 ::sem_wait(&sig_sem);
2532
2533 // were we externally suspended while we were waiting?
2534 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2535 if (threadIsSuspended) {
2536 // The semaphore has been incremented, but while we were waiting
2537 // another thread suspended us. We don't want to continue running
2538 // while suspended because that would surprise the thread that
2539 // suspended us.
2540 ::sem_post(&sig_sem);
2541
2542 thread->java_suspend_self();
2543 }
2544 } while (threadIsSuspended);
2545 }
2546 }
2547
2548 int os::signal_lookup() {
2549 return check_pending_signals(false);
2550 }
2551
2552 int os::signal_wait() {
2553 return check_pending_signals(true);
2554 }
2555
2556 ////////////////////////////////////////////////////////////////////////////////
2557 // Virtual Memory
2558
2559 int os::vm_page_size() {
2560 // Seems redundant as all get out
2561 assert(os::Linux::page_size() != -1, "must call os::init");
2562 return os::Linux::page_size();
2563 }
2564
2565 // Solaris allocates memory by pages.
2566 int os::vm_allocation_granularity() {
2567 assert(os::Linux::page_size() != -1, "must call os::init");
2568 return os::Linux::page_size();
2569 }
2570
2571 // Rationale behind this function:
2572 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2573 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2574 // samples for JITted code. Here we create private executable mapping over the code cache
2575 // and then we can use standard (well, almost, as mapping can change) way to provide
2576 // info for the reporting script by storing timestamp and location of symbol
2577 void linux_wrap_code(char* base, size_t size) {
2578 static volatile jint cnt = 0;
2579
2580 if (!UseOprofile) {
2581 return;
2582 }
2583
2584 char buf[PATH_MAX+1];
2585 int num = Atomic::add(1, &cnt);
2586
2587 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2588 os::get_temp_directory(), os::current_process_id(), num);
2589 unlink(buf);
2590
2591 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2592
2593 if (fd != -1) {
2594 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2595 if (rv != (off_t)-1) {
2596 if (::write(fd, "", 1) == 1) {
2597 mmap(base, size,
2598 PROT_READ|PROT_WRITE|PROT_EXEC,
2599 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2600 }
2601 }
2602 ::close(fd);
2603 unlink(buf);
2604 }
2605 }
2606
2607 static bool recoverable_mmap_error(int err) {
2608 // See if the error is one we can let the caller handle. This
2609 // list of errno values comes from JBS-6843484. I can't find a
2610 // Linux man page that documents this specific set of errno
2611 // values so while this list currently matches Solaris, it may
2612 // change as we gain experience with this failure mode.
2613 switch (err) {
2614 case EBADF:
2615 case EINVAL:
2616 case ENOTSUP:
2617 // let the caller deal with these errors
2618 return true;
2619
2620 default:
2621 // Any remaining errors on this OS can cause our reserved mapping
2622 // to be lost. That can cause confusion where different data
2623 // structures think they have the same memory mapped. The worst
2624 // scenario is if both the VM and a library think they have the
2625 // same memory mapped.
2626 return false;
2627 }
2628 }
2629
2630 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2631 int err) {
2632 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2633 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2634 strerror(err), err);
2635 }
2636
2637 static void warn_fail_commit_memory(char* addr, size_t size,
2638 size_t alignment_hint, bool exec,
2639 int err) {
2640 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2641 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2642 alignment_hint, exec, strerror(err), err);
2643 }
2644
2645 // NOTE: Linux kernel does not really reserve the pages for us.
2646 // All it does is to check if there are enough free pages
2647 // left at the time of mmap(). This could be a potential
2648 // problem.
2649 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2650 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2651 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2652 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2653 if (res != (uintptr_t) MAP_FAILED) {
2654 if (UseNUMAInterleaving) {
2655 numa_make_global(addr, size);
2656 }
2657 return 0;
2658 }
2659
2660 int err = errno; // save errno from mmap() call above
2661
2662 if (!recoverable_mmap_error(err)) {
2663 warn_fail_commit_memory(addr, size, exec, err);
2664 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2665 }
2666
2667 return err;
2668 }
2669
2670 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2671 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2672 }
2673
2674 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2675 const char* mesg) {
2676 assert(mesg != NULL, "mesg must be specified");
2677 int err = os::Linux::commit_memory_impl(addr, size, exec);
2678 if (err != 0) {
2679 // the caller wants all commit errors to exit with the specified mesg:
2680 warn_fail_commit_memory(addr, size, exec, err);
2681 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2682 }
2683 }
2684
2685 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2686 #ifndef MAP_HUGETLB
2687 #define MAP_HUGETLB 0x40000
2688 #endif
2689
2690 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2691 #ifndef MADV_HUGEPAGE
2692 #define MADV_HUGEPAGE 14
2693 #endif
2694
2695 int os::Linux::commit_memory_impl(char* addr, size_t size,
2696 size_t alignment_hint, bool exec) {
2697 int err = os::Linux::commit_memory_impl(addr, size, exec);
2698 if (err == 0) {
2699 realign_memory(addr, size, alignment_hint);
2700 }
2701 return err;
2702 }
2703
2704 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2705 bool exec) {
2706 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2707 }
2708
2709 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2710 size_t alignment_hint, bool exec,
2711 const char* mesg) {
2712 assert(mesg != NULL, "mesg must be specified");
2713 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2714 if (err != 0) {
2715 // the caller wants all commit errors to exit with the specified mesg:
2716 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2717 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2718 }
2719 }
2720
2721 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2722 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2723 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2724 // be supported or the memory may already be backed by huge pages.
2725 ::madvise(addr, bytes, MADV_HUGEPAGE);
2726 }
2727 }
2728
2729 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2730 // This method works by doing an mmap over an existing mmaping and effectively discarding
2731 // the existing pages. However it won't work for SHM-based large pages that cannot be
2732 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2733 // small pages on top of the SHM segment. This method always works for small pages, so we
2734 // allow that in any case.
2735 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2736 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2737 }
2738 }
2739
2740 void os::numa_make_global(char *addr, size_t bytes) {
2741 Linux::numa_interleave_memory(addr, bytes);
2742 }
2743
2744 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2745 // bind policy to MPOL_PREFERRED for the current thread.
2746 #define USE_MPOL_PREFERRED 0
2747
2748 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2749 // To make NUMA and large pages more robust when both enabled, we need to ease
2750 // the requirements on where the memory should be allocated. MPOL_BIND is the
2751 // default policy and it will force memory to be allocated on the specified
2752 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2753 // the specified node, but will not force it. Using this policy will prevent
2754 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2755 // free large pages.
2756 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2757 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2758 }
2759
2760 bool os::numa_topology_changed() { return false; }
2761
2762 size_t os::numa_get_groups_num() {
2763 int max_node = Linux::numa_max_node();
2764 return max_node > 0 ? max_node + 1 : 1;
2765 }
2766
2767 int os::numa_get_group_id() {
2768 int cpu_id = Linux::sched_getcpu();
2769 if (cpu_id != -1) {
2770 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2771 if (lgrp_id != -1) {
2772 return lgrp_id;
2773 }
2774 }
2775 return 0;
2776 }
2777
2778 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2779 for (size_t i = 0; i < size; i++) {
2780 ids[i] = i;
2781 }
2782 return size;
2783 }
2784
2785 bool os::get_page_info(char *start, page_info* info) {
2786 return false;
2787 }
2788
2789 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2790 page_info* page_found) {
2791 return end;
2792 }
2793
2794
2795 int os::Linux::sched_getcpu_syscall(void) {
2796 unsigned int cpu = 0;
2797 int retval = -1;
2798
2799 #if defined(IA32)
2800 #ifndef SYS_getcpu
2801 #define SYS_getcpu 318
2802 #endif
2803 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2804 #elif defined(AMD64)
2805 // Unfortunately we have to bring all these macros here from vsyscall.h
2806 // to be able to compile on old linuxes.
2807 #define __NR_vgetcpu 2
2808 #define VSYSCALL_START (-10UL << 20)
2809 #define VSYSCALL_SIZE 1024
2810 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2811 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2812 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2813 retval = vgetcpu(&cpu, NULL, NULL);
2814 #endif
2815
2816 return (retval == -1) ? retval : cpu;
2817 }
2818
2819 // Something to do with the numa-aware allocator needs these symbols
2820 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2821 extern "C" JNIEXPORT void numa_error(char *where) { }
2822 extern "C" JNIEXPORT int fork1() { return fork(); }
2823
2824
2825 // If we are running with libnuma version > 2, then we should
2826 // be trying to use symbols with versions 1.1
2827 // If we are running with earlier version, which did not have symbol versions,
2828 // we should use the base version.
2829 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2830 void *f = dlvsym(handle, name, "libnuma_1.1");
2831 if (f == NULL) {
2832 f = dlsym(handle, name);
2833 }
2834 return f;
2835 }
2836
2837 bool os::Linux::libnuma_init() {
2838 // sched_getcpu() should be in libc.
2839 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2840 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2841
2842 // If it's not, try a direct syscall.
2843 if (sched_getcpu() == -1) {
2844 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2845 (void*)&sched_getcpu_syscall));
2846 }
2847
2848 if (sched_getcpu() != -1) { // Does it work?
2849 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2850 if (handle != NULL) {
2851 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2852 libnuma_dlsym(handle, "numa_node_to_cpus")));
2853 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2854 libnuma_dlsym(handle, "numa_max_node")));
2855 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2856 libnuma_dlsym(handle, "numa_available")));
2857 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2858 libnuma_dlsym(handle, "numa_tonode_memory")));
2859 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2860 libnuma_dlsym(handle, "numa_interleave_memory")));
2861 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2862 libnuma_dlsym(handle, "numa_set_bind_policy")));
2863
2864
2865 if (numa_available() != -1) {
2866 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2867 // Create a cpu -> node mapping
2868 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2869 rebuild_cpu_to_node_map();
2870 return true;
2871 }
2872 }
2873 }
2874 return false;
2875 }
2876
2877 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2878 // The table is later used in get_node_by_cpu().
2879 void os::Linux::rebuild_cpu_to_node_map() {
2880 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2881 // in libnuma (possible values are starting from 16,
2882 // and continuing up with every other power of 2, but less
2883 // than the maximum number of CPUs supported by kernel), and
2884 // is a subject to change (in libnuma version 2 the requirements
2885 // are more reasonable) we'll just hardcode the number they use
2886 // in the library.
2887 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2888
2889 size_t cpu_num = os::active_processor_count();
2890 size_t cpu_map_size = NCPUS / BitsPerCLong;
2891 size_t cpu_map_valid_size =
2892 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2893
2894 cpu_to_node()->clear();
2895 cpu_to_node()->at_grow(cpu_num - 1);
2896 size_t node_num = numa_get_groups_num();
2897
2898 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2899 for (size_t i = 0; i < node_num; i++) {
2900 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2901 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2902 if (cpu_map[j] != 0) {
2903 for (size_t k = 0; k < BitsPerCLong; k++) {
2904 if (cpu_map[j] & (1UL << k)) {
2905 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2906 }
2907 }
2908 }
2909 }
2910 }
2911 }
2912 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2913 }
2914
2915 int os::Linux::get_node_by_cpu(int cpu_id) {
2916 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2917 return cpu_to_node()->at(cpu_id);
2918 }
2919 return -1;
2920 }
2921
2922 GrowableArray<int>* os::Linux::_cpu_to_node;
2923 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2924 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2925 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2926 os::Linux::numa_available_func_t os::Linux::_numa_available;
2927 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2928 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2929 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2930 unsigned long* os::Linux::_numa_all_nodes;
2931
2932 bool os::pd_uncommit_memory(char* addr, size_t size) {
2933 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2934 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2935 return res != (uintptr_t) MAP_FAILED;
2936 }
2937
2938 static address get_stack_commited_bottom(address bottom, size_t size) {
2939 address nbot = bottom;
2940 address ntop = bottom + size;
2941
2942 size_t page_sz = os::vm_page_size();
2943 unsigned pages = size / page_sz;
2944
2945 unsigned char vec[1];
2946 unsigned imin = 1, imax = pages + 1, imid;
2947 int mincore_return_value = 0;
2948
2949 assert(imin <= imax, "Unexpected page size");
2950
2951 while (imin < imax) {
2952 imid = (imax + imin) / 2;
2953 nbot = ntop - (imid * page_sz);
2954
2955 // Use a trick with mincore to check whether the page is mapped or not.
2956 // mincore sets vec to 1 if page resides in memory and to 0 if page
2957 // is swapped output but if page we are asking for is unmapped
2958 // it returns -1,ENOMEM
2959 mincore_return_value = mincore(nbot, page_sz, vec);
2960
2961 if (mincore_return_value == -1) {
2962 // Page is not mapped go up
2963 // to find first mapped page
2964 if (errno != EAGAIN) {
2965 assert(errno == ENOMEM, "Unexpected mincore errno");
2966 imax = imid;
2967 }
2968 } else {
2969 // Page is mapped go down
2970 // to find first not mapped page
2971 imin = imid + 1;
2972 }
2973 }
2974
2975 nbot = nbot + page_sz;
2976
2977 // Adjust stack bottom one page up if last checked page is not mapped
2978 if (mincore_return_value == -1) {
2979 nbot = nbot + page_sz;
2980 }
2981
2982 return nbot;
2983 }
2984
2985
2986 // Linux uses a growable mapping for the stack, and if the mapping for
2987 // the stack guard pages is not removed when we detach a thread the
2988 // stack cannot grow beyond the pages where the stack guard was
2989 // mapped. If at some point later in the process the stack expands to
2990 // that point, the Linux kernel cannot expand the stack any further
2991 // because the guard pages are in the way, and a segfault occurs.
2992 //
2993 // However, it's essential not to split the stack region by unmapping
2994 // a region (leaving a hole) that's already part of the stack mapping,
2995 // so if the stack mapping has already grown beyond the guard pages at
2996 // the time we create them, we have to truncate the stack mapping.
2997 // So, we need to know the extent of the stack mapping when
2998 // create_stack_guard_pages() is called.
2999
3000 // We only need this for stacks that are growable: at the time of
3001 // writing thread stacks don't use growable mappings (i.e. those
3002 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3003 // only applies to the main thread.
3004
3005 // If the (growable) stack mapping already extends beyond the point
3006 // where we're going to put our guard pages, truncate the mapping at
3007 // that point by munmap()ping it. This ensures that when we later
3008 // munmap() the guard pages we don't leave a hole in the stack
3009 // mapping. This only affects the main/initial thread
3010
3011 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3012 if (os::Linux::is_initial_thread()) {
3013 // As we manually grow stack up to bottom inside create_attached_thread(),
3014 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3015 // we don't need to do anything special.
3016 // Check it first, before calling heavy function.
3017 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3018 unsigned char vec[1];
3019
3020 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3021 // Fallback to slow path on all errors, including EAGAIN
3022 stack_extent = (uintptr_t) get_stack_commited_bottom(
3023 os::Linux::initial_thread_stack_bottom(),
3024 (size_t)addr - stack_extent);
3025 }
3026
3027 if (stack_extent < (uintptr_t)addr) {
3028 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3029 }
3030 }
3031
3032 return os::commit_memory(addr, size, !ExecMem);
3033 }
3034
3035 // If this is a growable mapping, remove the guard pages entirely by
3036 // munmap()ping them. If not, just call uncommit_memory(). This only
3037 // affects the main/initial thread, but guard against future OS changes
3038 // It's safe to always unmap guard pages for initial thread because we
3039 // always place it right after end of the mapped region
3040
3041 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3042 uintptr_t stack_extent, stack_base;
3043
3044 if (os::Linux::is_initial_thread()) {
3045 return ::munmap(addr, size) == 0;
3046 }
3047
3048 return os::uncommit_memory(addr, size);
3049 }
3050
3051 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3052 // at 'requested_addr'. If there are existing memory mappings at the same
3053 // location, however, they will be overwritten. If 'fixed' is false,
3054 // 'requested_addr' is only treated as a hint, the return value may or
3055 // may not start from the requested address. Unlike Linux mmap(), this
3056 // function returns NULL to indicate failure.
3057 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3058 char * addr;
3059 int flags;
3060
3061 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3062 if (fixed) {
3063 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3064 flags |= MAP_FIXED;
3065 }
3066
3067 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3068 // touch an uncommitted page. Otherwise, the read/write might
3069 // succeed if we have enough swap space to back the physical page.
3070 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3071 flags, -1, 0);
3072
3073 return addr == MAP_FAILED ? NULL : addr;
3074 }
3075
3076 static int anon_munmap(char * addr, size_t size) {
3077 return ::munmap(addr, size) == 0;
3078 }
3079
3080 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3081 size_t alignment_hint) {
3082 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3083 }
3084
3085 bool os::pd_release_memory(char* addr, size_t size) {
3086 return anon_munmap(addr, size);
3087 }
3088
3089 static bool linux_mprotect(char* addr, size_t size, int prot) {
3090 // Linux wants the mprotect address argument to be page aligned.
3091 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3092
3093 // According to SUSv3, mprotect() should only be used with mappings
3094 // established by mmap(), and mmap() always maps whole pages. Unaligned
3095 // 'addr' likely indicates problem in the VM (e.g. trying to change
3096 // protection of malloc'ed or statically allocated memory). Check the
3097 // caller if you hit this assert.
3098 assert(addr == bottom, "sanity check");
3099
3100 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3101 return ::mprotect(bottom, size, prot) == 0;
3102 }
3103
3104 // Set protections specified
3105 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3106 bool is_committed) {
3107 unsigned int p = 0;
3108 switch (prot) {
3109 case MEM_PROT_NONE: p = PROT_NONE; break;
3110 case MEM_PROT_READ: p = PROT_READ; break;
3111 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3112 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3113 default:
3114 ShouldNotReachHere();
3115 }
3116 // is_committed is unused.
3117 return linux_mprotect(addr, bytes, p);
3118 }
3119
3120 bool os::guard_memory(char* addr, size_t size) {
3121 return linux_mprotect(addr, size, PROT_NONE);
3122 }
3123
3124 bool os::unguard_memory(char* addr, size_t size) {
3125 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3126 }
3127
3128 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3129 size_t page_size) {
3130 bool result = false;
3131 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3132 MAP_ANONYMOUS|MAP_PRIVATE,
3133 -1, 0);
3134 if (p != MAP_FAILED) {
3135 void *aligned_p = align_ptr_up(p, page_size);
3136
3137 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3138
3139 munmap(p, page_size * 2);
3140 }
3141
3142 if (warn && !result) {
3143 warning("TransparentHugePages is not supported by the operating system.");
3144 }
3145
3146 return result;
3147 }
3148
3149 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3150 bool result = false;
3151 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3152 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3153 -1, 0);
3154
3155 if (p != MAP_FAILED) {
3156 // We don't know if this really is a huge page or not.
3157 FILE *fp = fopen("/proc/self/maps", "r");
3158 if (fp) {
3159 while (!feof(fp)) {
3160 char chars[257];
3161 long x = 0;
3162 if (fgets(chars, sizeof(chars), fp)) {
3163 if (sscanf(chars, "%lx-%*x", &x) == 1
3164 && x == (long)p) {
3165 if (strstr (chars, "hugepage")) {
3166 result = true;
3167 break;
3168 }
3169 }
3170 }
3171 }
3172 fclose(fp);
3173 }
3174 munmap(p, page_size);
3175 }
3176
3177 if (warn && !result) {
3178 warning("HugeTLBFS is not supported by the operating system.");
3179 }
3180
3181 return result;
3182 }
3183
3184 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3185 //
3186 // From the coredump_filter documentation:
3187 //
3188 // - (bit 0) anonymous private memory
3189 // - (bit 1) anonymous shared memory
3190 // - (bit 2) file-backed private memory
3191 // - (bit 3) file-backed shared memory
3192 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3193 // effective only if the bit 2 is cleared)
3194 // - (bit 5) hugetlb private memory
3195 // - (bit 6) hugetlb shared memory
3196 //
3197 static void set_coredump_filter(void) {
3198 FILE *f;
3199 long cdm;
3200
3201 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3202 return;
3203 }
3204
3205 if (fscanf(f, "%lx", &cdm) != 1) {
3206 fclose(f);
3207 return;
3208 }
3209
3210 rewind(f);
3211
3212 if ((cdm & LARGEPAGES_BIT) == 0) {
3213 cdm |= LARGEPAGES_BIT;
3214 fprintf(f, "%#lx", cdm);
3215 }
3216
3217 fclose(f);
3218 }
3219
3220 // Large page support
3221
3222 static size_t _large_page_size = 0;
3223
3224 size_t os::Linux::find_large_page_size() {
3225 size_t large_page_size = 0;
3226
3227 // large_page_size on Linux is used to round up heap size. x86 uses either
3228 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3229 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3230 // page as large as 256M.
3231 //
3232 // Here we try to figure out page size by parsing /proc/meminfo and looking
3233 // for a line with the following format:
3234 // Hugepagesize: 2048 kB
3235 //
3236 // If we can't determine the value (e.g. /proc is not mounted, or the text
3237 // format has been changed), we'll use the largest page size supported by
3238 // the processor.
3239
3240 #ifndef ZERO
3241 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3242 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3243 #endif // ZERO
3244
3245 FILE *fp = fopen("/proc/meminfo", "r");
3246 if (fp) {
3247 while (!feof(fp)) {
3248 int x = 0;
3249 char buf[16];
3250 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3251 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3252 large_page_size = x * K;
3253 break;
3254 }
3255 } else {
3256 // skip to next line
3257 for (;;) {
3258 int ch = fgetc(fp);
3259 if (ch == EOF || ch == (int)'\n') break;
3260 }
3261 }
3262 }
3263 fclose(fp);
3264 }
3265
3266 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3267 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3268 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3269 proper_unit_for_byte_size(large_page_size));
3270 }
3271
3272 return large_page_size;
3273 }
3274
3275 size_t os::Linux::setup_large_page_size() {
3276 _large_page_size = Linux::find_large_page_size();
3277 const size_t default_page_size = (size_t)Linux::page_size();
3278 if (_large_page_size > default_page_size) {
3279 _page_sizes[0] = _large_page_size;
3280 _page_sizes[1] = default_page_size;
3281 _page_sizes[2] = 0;
3282 }
3283
3284 return _large_page_size;
3285 }
3286
3287 bool os::Linux::setup_large_page_type(size_t page_size) {
3288 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3289 FLAG_IS_DEFAULT(UseSHM) &&
3290 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3291
3292 // The type of large pages has not been specified by the user.
3293
3294 // Try UseHugeTLBFS and then UseSHM.
3295 UseHugeTLBFS = UseSHM = true;
3296
3297 // Don't try UseTransparentHugePages since there are known
3298 // performance issues with it turned on. This might change in the future.
3299 UseTransparentHugePages = false;
3300 }
3301
3302 if (UseTransparentHugePages) {
3303 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3304 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3305 UseHugeTLBFS = false;
3306 UseSHM = false;
3307 return true;
3308 }
3309 UseTransparentHugePages = false;
3310 }
3311
3312 if (UseHugeTLBFS) {
3313 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3314 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3315 UseSHM = false;
3316 return true;
3317 }
3318 UseHugeTLBFS = false;
3319 }
3320
3321 return UseSHM;
3322 }
3323
3324 void os::large_page_init() {
3325 if (!UseLargePages &&
3326 !UseTransparentHugePages &&
3327 !UseHugeTLBFS &&
3328 !UseSHM) {
3329 // Not using large pages.
3330 return;
3331 }
3332
3333 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3334 // The user explicitly turned off large pages.
3335 // Ignore the rest of the large pages flags.
3336 UseTransparentHugePages = false;
3337 UseHugeTLBFS = false;
3338 UseSHM = false;
3339 return;
3340 }
3341
3342 size_t large_page_size = Linux::setup_large_page_size();
3343 UseLargePages = Linux::setup_large_page_type(large_page_size);
3344
3345 set_coredump_filter();
3346 }
3347
3348 #ifndef SHM_HUGETLB
3349 #define SHM_HUGETLB 04000
3350 #endif
3351
3352 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3353 char* req_addr, bool exec) {
3354 // "exec" is passed in but not used. Creating the shared image for
3355 // the code cache doesn't have an SHM_X executable permission to check.
3356 assert(UseLargePages && UseSHM, "only for SHM large pages");
3357 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3358
3359 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3360 return NULL; // Fallback to small pages.
3361 }
3362
3363 key_t key = IPC_PRIVATE;
3364 char *addr;
3365
3366 bool warn_on_failure = UseLargePages &&
3367 (!FLAG_IS_DEFAULT(UseLargePages) ||
3368 !FLAG_IS_DEFAULT(UseSHM) ||
3369 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3370 char msg[128];
3371
3372 // Create a large shared memory region to attach to based on size.
3373 // Currently, size is the total size of the heap
3374 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3375 if (shmid == -1) {
3376 // Possible reasons for shmget failure:
3377 // 1. shmmax is too small for Java heap.
3378 // > check shmmax value: cat /proc/sys/kernel/shmmax
3379 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3380 // 2. not enough large page memory.
3381 // > check available large pages: cat /proc/meminfo
3382 // > increase amount of large pages:
3383 // echo new_value > /proc/sys/vm/nr_hugepages
3384 // Note 1: different Linux may use different name for this property,
3385 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3386 // Note 2: it's possible there's enough physical memory available but
3387 // they are so fragmented after a long run that they can't
3388 // coalesce into large pages. Try to reserve large pages when
3389 // the system is still "fresh".
3390 if (warn_on_failure) {
3391 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3392 warning("%s", msg);
3393 }
3394 return NULL;
3395 }
3396
3397 // attach to the region
3398 addr = (char*)shmat(shmid, req_addr, 0);
3399 int err = errno;
3400
3401 // Remove shmid. If shmat() is successful, the actual shared memory segment
3402 // will be deleted when it's detached by shmdt() or when the process
3403 // terminates. If shmat() is not successful this will remove the shared
3404 // segment immediately.
3405 shmctl(shmid, IPC_RMID, NULL);
3406
3407 if ((intptr_t)addr == -1) {
3408 if (warn_on_failure) {
3409 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3410 warning("%s", msg);
3411 }
3412 return NULL;
3413 }
3414
3415 return addr;
3416 }
3417
3418 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3419 int error) {
3420 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3421
3422 bool warn_on_failure = UseLargePages &&
3423 (!FLAG_IS_DEFAULT(UseLargePages) ||
3424 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3425 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3426
3427 if (warn_on_failure) {
3428 char msg[128];
3429 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3430 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3431 warning("%s", msg);
3432 }
3433 }
3434
3435 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3436 char* req_addr,
3437 bool exec) {
3438 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3439 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3440 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3441
3442 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3443 char* addr = (char*)::mmap(req_addr, bytes, prot,
3444 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3445 -1, 0);
3446
3447 if (addr == MAP_FAILED) {
3448 warn_on_large_pages_failure(req_addr, bytes, errno);
3449 return NULL;
3450 }
3451
3452 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3453
3454 return addr;
3455 }
3456
3457 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3458 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3459 // (req_addr != NULL) or with a given alignment.
3460 // - bytes shall be a multiple of alignment.
3461 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3462 // - alignment sets the alignment at which memory shall be allocated.
3463 // It must be a multiple of allocation granularity.
3464 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3465 // req_addr or NULL.
3466 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3467
3468 size_t extra_size = bytes;
3469 if (req_addr == NULL && alignment > 0) {
3470 extra_size += alignment;
3471 }
3472
3473 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3474 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3475 -1, 0);
3476 if (start == MAP_FAILED) {
3477 start = NULL;
3478 } else {
3479 if (req_addr != NULL) {
3480 if (start != req_addr) {
3481 ::munmap(start, extra_size);
3482 start = NULL;
3483 }
3484 } else {
3485 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3486 char* const end_aligned = start_aligned + bytes;
3487 char* const end = start + extra_size;
3488 if (start_aligned > start) {
3489 ::munmap(start, start_aligned - start);
3490 }
3491 if (end_aligned < end) {
3492 ::munmap(end_aligned, end - end_aligned);
3493 }
3494 start = start_aligned;
3495 }
3496 }
3497 return start;
3498
3499 }
3500
3501 // Reserve memory using mmap(MAP_HUGETLB).
3502 // - bytes shall be a multiple of alignment.
3503 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3504 // - alignment sets the alignment at which memory shall be allocated.
3505 // It must be a multiple of allocation granularity.
3506 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3507 // req_addr or NULL.
3508 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3509 size_t alignment,
3510 char* req_addr,
3511 bool exec) {
3512 size_t large_page_size = os::large_page_size();
3513 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3514
3515 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3516 assert(is_size_aligned(bytes, alignment), "Must be");
3517
3518 // First reserve - but not commit - the address range in small pages.
3519 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3520
3521 if (start == NULL) {
3522 return NULL;
3523 }
3524
3525 assert(is_ptr_aligned(start, alignment), "Must be");
3526
3527 char* end = start + bytes;
3528
3529 // Find the regions of the allocated chunk that can be promoted to large pages.
3530 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3531 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3532
3533 size_t lp_bytes = lp_end - lp_start;
3534
3535 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3536
3537 if (lp_bytes == 0) {
3538 // The mapped region doesn't even span the start and the end of a large page.
3539 // Fall back to allocate a non-special area.
3540 ::munmap(start, end - start);
3541 return NULL;
3542 }
3543
3544 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3545
3546 void* result;
3547
3548 // Commit small-paged leading area.
3549 if (start != lp_start) {
3550 result = ::mmap(start, lp_start - start, prot,
3551 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3552 -1, 0);
3553 if (result == MAP_FAILED) {
3554 ::munmap(lp_start, end - lp_start);
3555 return NULL;
3556 }
3557 }
3558
3559 // Commit large-paged area.
3560 result = ::mmap(lp_start, lp_bytes, prot,
3561 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3562 -1, 0);
3563 if (result == MAP_FAILED) {
3564 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3565 // If the mmap above fails, the large pages region will be unmapped and we
3566 // have regions before and after with small pages. Release these regions.
3567 //
3568 // | mapped | unmapped | mapped |
3569 // ^ ^ ^ ^
3570 // start lp_start lp_end end
3571 //
3572 ::munmap(start, lp_start - start);
3573 ::munmap(lp_end, end - lp_end);
3574 return NULL;
3575 }
3576
3577 // Commit small-paged trailing area.
3578 if (lp_end != end) {
3579 result = ::mmap(lp_end, end - lp_end, prot,
3580 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3581 -1, 0);
3582 if (result == MAP_FAILED) {
3583 ::munmap(start, lp_end - start);
3584 return NULL;
3585 }
3586 }
3587
3588 return start;
3589 }
3590
3591 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3592 size_t alignment,
3593 char* req_addr,
3594 bool exec) {
3595 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3596 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3597 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3598 assert(is_power_of_2(os::large_page_size()), "Must be");
3599 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3600
3601 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3602 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3603 } else {
3604 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3605 }
3606 }
3607
3608 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3609 char* req_addr, bool exec) {
3610 assert(UseLargePages, "only for large pages");
3611
3612 char* addr;
3613 if (UseSHM) {
3614 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3615 } else {
3616 assert(UseHugeTLBFS, "must be");
3617 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3618 }
3619
3620 if (addr != NULL) {
3621 if (UseNUMAInterleaving) {
3622 numa_make_global(addr, bytes);
3623 }
3624
3625 // The memory is committed
3626 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3627 }
3628
3629 return addr;
3630 }
3631
3632 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3633 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3634 return shmdt(base) == 0;
3635 }
3636
3637 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3638 return pd_release_memory(base, bytes);
3639 }
3640
3641 bool os::release_memory_special(char* base, size_t bytes) {
3642 bool res;
3643 if (MemTracker::tracking_level() > NMT_minimal) {
3644 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3645 res = os::Linux::release_memory_special_impl(base, bytes);
3646 if (res) {
3647 tkr.record((address)base, bytes);
3648 }
3649
3650 } else {
3651 res = os::Linux::release_memory_special_impl(base, bytes);
3652 }
3653 return res;
3654 }
3655
3656 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3657 assert(UseLargePages, "only for large pages");
3658 bool res;
3659
3660 if (UseSHM) {
3661 res = os::Linux::release_memory_special_shm(base, bytes);
3662 } else {
3663 assert(UseHugeTLBFS, "must be");
3664 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3665 }
3666 return res;
3667 }
3668
3669 size_t os::large_page_size() {
3670 return _large_page_size;
3671 }
3672
3673 // With SysV SHM the entire memory region must be allocated as shared
3674 // memory.
3675 // HugeTLBFS allows application to commit large page memory on demand.
3676 // However, when committing memory with HugeTLBFS fails, the region
3677 // that was supposed to be committed will lose the old reservation
3678 // and allow other threads to steal that memory region. Because of this
3679 // behavior we can't commit HugeTLBFS memory.
3680 bool os::can_commit_large_page_memory() {
3681 return UseTransparentHugePages;
3682 }
3683
3684 bool os::can_execute_large_page_memory() {
3685 return UseTransparentHugePages || UseHugeTLBFS;
3686 }
3687
3688 // Reserve memory at an arbitrary address, only if that area is
3689 // available (and not reserved for something else).
3690
3691 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3692 const int max_tries = 10;
3693 char* base[max_tries];
3694 size_t size[max_tries];
3695 const size_t gap = 0x000000;
3696
3697 // Assert only that the size is a multiple of the page size, since
3698 // that's all that mmap requires, and since that's all we really know
3699 // about at this low abstraction level. If we need higher alignment,
3700 // we can either pass an alignment to this method or verify alignment
3701 // in one of the methods further up the call chain. See bug 5044738.
3702 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3703
3704 // Repeatedly allocate blocks until the block is allocated at the
3705 // right spot.
3706
3707 // Linux mmap allows caller to pass an address as hint; give it a try first,
3708 // if kernel honors the hint then we can return immediately.
3709 char * addr = anon_mmap(requested_addr, bytes, false);
3710 if (addr == requested_addr) {
3711 return requested_addr;
3712 }
3713
3714 if (addr != NULL) {
3715 // mmap() is successful but it fails to reserve at the requested address
3716 anon_munmap(addr, bytes);
3717 }
3718
3719 int i;
3720 for (i = 0; i < max_tries; ++i) {
3721 base[i] = reserve_memory(bytes);
3722
3723 if (base[i] != NULL) {
3724 // Is this the block we wanted?
3725 if (base[i] == requested_addr) {
3726 size[i] = bytes;
3727 break;
3728 }
3729
3730 // Does this overlap the block we wanted? Give back the overlapped
3731 // parts and try again.
3732
3733 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3734 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3735 unmap_memory(base[i], top_overlap);
3736 base[i] += top_overlap;
3737 size[i] = bytes - top_overlap;
3738 } else {
3739 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3740 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3741 unmap_memory(requested_addr, bottom_overlap);
3742 size[i] = bytes - bottom_overlap;
3743 } else {
3744 size[i] = bytes;
3745 }
3746 }
3747 }
3748 }
3749
3750 // Give back the unused reserved pieces.
3751
3752 for (int j = 0; j < i; ++j) {
3753 if (base[j] != NULL) {
3754 unmap_memory(base[j], size[j]);
3755 }
3756 }
3757
3758 if (i < max_tries) {
3759 return requested_addr;
3760 } else {
3761 return NULL;
3762 }
3763 }
3764
3765 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3766 return ::read(fd, buf, nBytes);
3767 }
3768
3769 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3770 return ::pread(fd, buf, nBytes, offset);
3771 }
3772
3773 // Short sleep, direct OS call.
3774 //
3775 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3776 // sched_yield(2) will actually give up the CPU:
3777 //
3778 // * Alone on this pariticular CPU, keeps running.
3779 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3780 // (pre 2.6.39).
3781 //
3782 // So calling this with 0 is an alternative.
3783 //
3784 void os::naked_short_sleep(jlong ms) {
3785 struct timespec req;
3786
3787 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3788 req.tv_sec = 0;
3789 if (ms > 0) {
3790 req.tv_nsec = (ms % 1000) * 1000000;
3791 } else {
3792 req.tv_nsec = 1;
3793 }
3794
3795 nanosleep(&req, NULL);
3796
3797 return;
3798 }
3799
3800 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3801 void os::infinite_sleep() {
3802 while (true) { // sleep forever ...
3803 ::sleep(100); // ... 100 seconds at a time
3804 }
3805 }
3806
3807 // Used to convert frequent JVM_Yield() to nops
3808 bool os::dont_yield() {
3809 return DontYieldALot;
3810 }
3811
3812 void os::naked_yield() {
3813 sched_yield();
3814 }
3815
3816 ////////////////////////////////////////////////////////////////////////////////
3817 // thread priority support
3818
3819 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3820 // only supports dynamic priority, static priority must be zero. For real-time
3821 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3822 // However, for large multi-threaded applications, SCHED_RR is not only slower
3823 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3824 // of 5 runs - Sep 2005).
3825 //
3826 // The following code actually changes the niceness of kernel-thread/LWP. It
3827 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3828 // not the entire user process, and user level threads are 1:1 mapped to kernel
3829 // threads. It has always been the case, but could change in the future. For
3830 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3831 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3832
3833 int os::java_to_os_priority[CriticalPriority + 1] = {
3834 19, // 0 Entry should never be used
3835
3836 4, // 1 MinPriority
3837 3, // 2
3838 2, // 3
3839
3840 1, // 4
3841 0, // 5 NormPriority
3842 -1, // 6
3843
3844 -2, // 7
3845 -3, // 8
3846 -4, // 9 NearMaxPriority
3847
3848 -5, // 10 MaxPriority
3849
3850 -5 // 11 CriticalPriority
3851 };
3852
3853 static int prio_init() {
3854 if (ThreadPriorityPolicy == 1) {
3855 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3856 // if effective uid is not root. Perhaps, a more elegant way of doing
3857 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3858 if (geteuid() != 0) {
3859 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3860 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3861 }
3862 ThreadPriorityPolicy = 0;
3863 }
3864 }
3865 if (UseCriticalJavaThreadPriority) {
3866 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3867 }
3868 return 0;
3869 }
3870
3871 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3872 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3873
3874 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3875 return (ret == 0) ? OS_OK : OS_ERR;
3876 }
3877
3878 OSReturn os::get_native_priority(const Thread* const thread,
3879 int *priority_ptr) {
3880 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3881 *priority_ptr = java_to_os_priority[NormPriority];
3882 return OS_OK;
3883 }
3884
3885 errno = 0;
3886 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3887 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3888 }
3889
3890 // Hint to the underlying OS that a task switch would not be good.
3891 // Void return because it's a hint and can fail.
3892 void os::hint_no_preempt() {}
3893
3894 ////////////////////////////////////////////////////////////////////////////////
3895 // suspend/resume support
3896
3897 // the low-level signal-based suspend/resume support is a remnant from the
3898 // old VM-suspension that used to be for java-suspension, safepoints etc,
3899 // within hotspot. Now there is a single use-case for this:
3900 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3901 // that runs in the watcher thread.
3902 // The remaining code is greatly simplified from the more general suspension
3903 // code that used to be used.
3904 //
3905 // The protocol is quite simple:
3906 // - suspend:
3907 // - sends a signal to the target thread
3908 // - polls the suspend state of the osthread using a yield loop
3909 // - target thread signal handler (SR_handler) sets suspend state
3910 // and blocks in sigsuspend until continued
3911 // - resume:
3912 // - sets target osthread state to continue
3913 // - sends signal to end the sigsuspend loop in the SR_handler
3914 //
3915 // Note that the SR_lock plays no role in this suspend/resume protocol.
3916
3917 static void resume_clear_context(OSThread *osthread) {
3918 osthread->set_ucontext(NULL);
3919 osthread->set_siginfo(NULL);
3920 }
3921
3922 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3923 ucontext_t* context) {
3924 osthread->set_ucontext(context);
3925 osthread->set_siginfo(siginfo);
3926 }
3927
3928 // Handler function invoked when a thread's execution is suspended or
3929 // resumed. We have to be careful that only async-safe functions are
3930 // called here (Note: most pthread functions are not async safe and
3931 // should be avoided.)
3932 //
3933 // Note: sigwait() is a more natural fit than sigsuspend() from an
3934 // interface point of view, but sigwait() prevents the signal hander
3935 // from being run. libpthread would get very confused by not having
3936 // its signal handlers run and prevents sigwait()'s use with the
3937 // mutex granting granting signal.
3938 //
3939 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3940 //
3941 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3942 // Save and restore errno to avoid confusing native code with EINTR
3943 // after sigsuspend.
3944 int old_errno = errno;
3945
3946 Thread* thread = Thread::current();
3947 OSThread* osthread = thread->osthread();
3948 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3949
3950 os::SuspendResume::State current = osthread->sr.state();
3951 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3952 suspend_save_context(osthread, siginfo, context);
3953
3954 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3955 os::SuspendResume::State state = osthread->sr.suspended();
3956 if (state == os::SuspendResume::SR_SUSPENDED) {
3957 sigset_t suspend_set; // signals for sigsuspend()
3958
3959 // get current set of blocked signals and unblock resume signal
3960 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3961 sigdelset(&suspend_set, SR_signum);
3962
3963 sr_semaphore.signal();
3964 // wait here until we are resumed
3965 while (1) {
3966 sigsuspend(&suspend_set);
3967
3968 os::SuspendResume::State result = osthread->sr.running();
3969 if (result == os::SuspendResume::SR_RUNNING) {
3970 sr_semaphore.signal();
3971 break;
3972 }
3973 }
3974
3975 } else if (state == os::SuspendResume::SR_RUNNING) {
3976 // request was cancelled, continue
3977 } else {
3978 ShouldNotReachHere();
3979 }
3980
3981 resume_clear_context(osthread);
3982 } else if (current == os::SuspendResume::SR_RUNNING) {
3983 // request was cancelled, continue
3984 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3985 // ignore
3986 } else {
3987 // ignore
3988 }
3989
3990 errno = old_errno;
3991 }
3992
3993
3994 static int SR_initialize() {
3995 struct sigaction act;
3996 char *s;
3997 // Get signal number to use for suspend/resume
3998 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3999 int sig = ::strtol(s, 0, 10);
4000 if (sig > 0 || sig < _NSIG) {
4001 SR_signum = sig;
4002 }
4003 }
4004
4005 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4006 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4007
4008 sigemptyset(&SR_sigset);
4009 sigaddset(&SR_sigset, SR_signum);
4010
4011 // Set up signal handler for suspend/resume
4012 act.sa_flags = SA_RESTART|SA_SIGINFO;
4013 act.sa_handler = (void (*)(int)) SR_handler;
4014
4015 // SR_signum is blocked by default.
4016 // 4528190 - We also need to block pthread restart signal (32 on all
4017 // supported Linux platforms). Note that LinuxThreads need to block
4018 // this signal for all threads to work properly. So we don't have
4019 // to use hard-coded signal number when setting up the mask.
4020 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4021
4022 if (sigaction(SR_signum, &act, 0) == -1) {
4023 return -1;
4024 }
4025
4026 // Save signal flag
4027 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4028 return 0;
4029 }
4030
4031 static int sr_notify(OSThread* osthread) {
4032 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4033 assert_status(status == 0, status, "pthread_kill");
4034 return status;
4035 }
4036
4037 // "Randomly" selected value for how long we want to spin
4038 // before bailing out on suspending a thread, also how often
4039 // we send a signal to a thread we want to resume
4040 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4041 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4042
4043 // returns true on success and false on error - really an error is fatal
4044 // but this seems the normal response to library errors
4045 static bool do_suspend(OSThread* osthread) {
4046 assert(osthread->sr.is_running(), "thread should be running");
4047 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4048
4049 // mark as suspended and send signal
4050 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4051 // failed to switch, state wasn't running?
4052 ShouldNotReachHere();
4053 return false;
4054 }
4055
4056 if (sr_notify(osthread) != 0) {
4057 ShouldNotReachHere();
4058 }
4059
4060 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4061 while (true) {
4062 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4063 break;
4064 } else {
4065 // timeout
4066 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4067 if (cancelled == os::SuspendResume::SR_RUNNING) {
4068 return false;
4069 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4070 // make sure that we consume the signal on the semaphore as well
4071 sr_semaphore.wait();
4072 break;
4073 } else {
4074 ShouldNotReachHere();
4075 return false;
4076 }
4077 }
4078 }
4079
4080 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4081 return true;
4082 }
4083
4084 static void do_resume(OSThread* osthread) {
4085 assert(osthread->sr.is_suspended(), "thread should be suspended");
4086 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4087
4088 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4089 // failed to switch to WAKEUP_REQUEST
4090 ShouldNotReachHere();
4091 return;
4092 }
4093
4094 while (true) {
4095 if (sr_notify(osthread) == 0) {
4096 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4097 if (osthread->sr.is_running()) {
4098 return;
4099 }
4100 }
4101 } else {
4102 ShouldNotReachHere();
4103 }
4104 }
4105
4106 guarantee(osthread->sr.is_running(), "Must be running!");
4107 }
4108
4109 ///////////////////////////////////////////////////////////////////////////////////
4110 // signal handling (except suspend/resume)
4111
4112 // This routine may be used by user applications as a "hook" to catch signals.
4113 // The user-defined signal handler must pass unrecognized signals to this
4114 // routine, and if it returns true (non-zero), then the signal handler must
4115 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4116 // routine will never retun false (zero), but instead will execute a VM panic
4117 // routine kill the process.
4118 //
4119 // If this routine returns false, it is OK to call it again. This allows
4120 // the user-defined signal handler to perform checks either before or after
4121 // the VM performs its own checks. Naturally, the user code would be making
4122 // a serious error if it tried to handle an exception (such as a null check
4123 // or breakpoint) that the VM was generating for its own correct operation.
4124 //
4125 // This routine may recognize any of the following kinds of signals:
4126 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4127 // It should be consulted by handlers for any of those signals.
4128 //
4129 // The caller of this routine must pass in the three arguments supplied
4130 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4131 // field of the structure passed to sigaction(). This routine assumes that
4132 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4133 //
4134 // Note that the VM will print warnings if it detects conflicting signal
4135 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4136 //
4137 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4138 siginfo_t* siginfo,
4139 void* ucontext,
4140 int abort_if_unrecognized);
4141
4142 void signalHandler(int sig, siginfo_t* info, void* uc) {
4143 assert(info != NULL && uc != NULL, "it must be old kernel");
4144 int orig_errno = errno; // Preserve errno value over signal handler.
4145 JVM_handle_linux_signal(sig, info, uc, true);
4146 errno = orig_errno;
4147 }
4148
4149
4150 // This boolean allows users to forward their own non-matching signals
4151 // to JVM_handle_linux_signal, harmlessly.
4152 bool os::Linux::signal_handlers_are_installed = false;
4153
4154 // For signal-chaining
4155 struct sigaction os::Linux::sigact[MAXSIGNUM];
4156 unsigned int os::Linux::sigs = 0;
4157 bool os::Linux::libjsig_is_loaded = false;
4158 typedef struct sigaction *(*get_signal_t)(int);
4159 get_signal_t os::Linux::get_signal_action = NULL;
4160
4161 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4162 struct sigaction *actp = NULL;
4163
4164 if (libjsig_is_loaded) {
4165 // Retrieve the old signal handler from libjsig
4166 actp = (*get_signal_action)(sig);
4167 }
4168 if (actp == NULL) {
4169 // Retrieve the preinstalled signal handler from jvm
4170 actp = get_preinstalled_handler(sig);
4171 }
4172
4173 return actp;
4174 }
4175
4176 static bool call_chained_handler(struct sigaction *actp, int sig,
4177 siginfo_t *siginfo, void *context) {
4178 // Call the old signal handler
4179 if (actp->sa_handler == SIG_DFL) {
4180 // It's more reasonable to let jvm treat it as an unexpected exception
4181 // instead of taking the default action.
4182 return false;
4183 } else if (actp->sa_handler != SIG_IGN) {
4184 if ((actp->sa_flags & SA_NODEFER) == 0) {
4185 // automaticlly block the signal
4186 sigaddset(&(actp->sa_mask), sig);
4187 }
4188
4189 sa_handler_t hand = NULL;
4190 sa_sigaction_t sa = NULL;
4191 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4192 // retrieve the chained handler
4193 if (siginfo_flag_set) {
4194 sa = actp->sa_sigaction;
4195 } else {
4196 hand = actp->sa_handler;
4197 }
4198
4199 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4200 actp->sa_handler = SIG_DFL;
4201 }
4202
4203 // try to honor the signal mask
4204 sigset_t oset;
4205 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4206
4207 // call into the chained handler
4208 if (siginfo_flag_set) {
4209 (*sa)(sig, siginfo, context);
4210 } else {
4211 (*hand)(sig);
4212 }
4213
4214 // restore the signal mask
4215 pthread_sigmask(SIG_SETMASK, &oset, 0);
4216 }
4217 // Tell jvm's signal handler the signal is taken care of.
4218 return true;
4219 }
4220
4221 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4222 bool chained = false;
4223 // signal-chaining
4224 if (UseSignalChaining) {
4225 struct sigaction *actp = get_chained_signal_action(sig);
4226 if (actp != NULL) {
4227 chained = call_chained_handler(actp, sig, siginfo, context);
4228 }
4229 }
4230 return chained;
4231 }
4232
4233 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4234 if ((((unsigned int)1 << sig) & sigs) != 0) {
4235 return &sigact[sig];
4236 }
4237 return NULL;
4238 }
4239
4240 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4241 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4242 sigact[sig] = oldAct;
4243 sigs |= (unsigned int)1 << sig;
4244 }
4245
4246 // for diagnostic
4247 int os::Linux::sigflags[MAXSIGNUM];
4248
4249 int os::Linux::get_our_sigflags(int sig) {
4250 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4251 return sigflags[sig];
4252 }
4253
4254 void os::Linux::set_our_sigflags(int sig, int flags) {
4255 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4256 sigflags[sig] = flags;
4257 }
4258
4259 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4260 // Check for overwrite.
4261 struct sigaction oldAct;
4262 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4263
4264 void* oldhand = oldAct.sa_sigaction
4265 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4266 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4267 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4268 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4269 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4270 if (AllowUserSignalHandlers || !set_installed) {
4271 // Do not overwrite; user takes responsibility to forward to us.
4272 return;
4273 } else if (UseSignalChaining) {
4274 // save the old handler in jvm
4275 save_preinstalled_handler(sig, oldAct);
4276 // libjsig also interposes the sigaction() call below and saves the
4277 // old sigaction on it own.
4278 } else {
4279 fatal("Encountered unexpected pre-existing sigaction handler "
4280 "%#lx for signal %d.", (long)oldhand, sig);
4281 }
4282 }
4283
4284 struct sigaction sigAct;
4285 sigfillset(&(sigAct.sa_mask));
4286 sigAct.sa_handler = SIG_DFL;
4287 if (!set_installed) {
4288 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4289 } else {
4290 sigAct.sa_sigaction = signalHandler;
4291 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4292 }
4293 // Save flags, which are set by ours
4294 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4295 sigflags[sig] = sigAct.sa_flags;
4296
4297 int ret = sigaction(sig, &sigAct, &oldAct);
4298 assert(ret == 0, "check");
4299
4300 void* oldhand2 = oldAct.sa_sigaction
4301 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4302 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4303 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4304 }
4305
4306 // install signal handlers for signals that HotSpot needs to
4307 // handle in order to support Java-level exception handling.
4308
4309 void os::Linux::install_signal_handlers() {
4310 if (!signal_handlers_are_installed) {
4311 signal_handlers_are_installed = true;
4312
4313 // signal-chaining
4314 typedef void (*signal_setting_t)();
4315 signal_setting_t begin_signal_setting = NULL;
4316 signal_setting_t end_signal_setting = NULL;
4317 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4318 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4319 if (begin_signal_setting != NULL) {
4320 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4321 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4322 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4323 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4324 libjsig_is_loaded = true;
4325 assert(UseSignalChaining, "should enable signal-chaining");
4326 }
4327 if (libjsig_is_loaded) {
4328 // Tell libjsig jvm is setting signal handlers
4329 (*begin_signal_setting)();
4330 }
4331
4332 set_signal_handler(SIGSEGV, true);
4333 set_signal_handler(SIGPIPE, true);
4334 set_signal_handler(SIGBUS, true);
4335 set_signal_handler(SIGILL, true);
4336 set_signal_handler(SIGFPE, true);
4337 #if defined(PPC64)
4338 set_signal_handler(SIGTRAP, true);
4339 #endif
4340 set_signal_handler(SIGXFSZ, true);
4341
4342 if (libjsig_is_loaded) {
4343 // Tell libjsig jvm finishes setting signal handlers
4344 (*end_signal_setting)();
4345 }
4346
4347 // We don't activate signal checker if libjsig is in place, we trust ourselves
4348 // and if UserSignalHandler is installed all bets are off.
4349 // Log that signal checking is off only if -verbose:jni is specified.
4350 if (CheckJNICalls) {
4351 if (libjsig_is_loaded) {
4352 if (PrintJNIResolving) {
4353 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4354 }
4355 check_signals = false;
4356 }
4357 if (AllowUserSignalHandlers) {
4358 if (PrintJNIResolving) {
4359 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4360 }
4361 check_signals = false;
4362 }
4363 }
4364 }
4365 }
4366
4367 // This is the fastest way to get thread cpu time on Linux.
4368 // Returns cpu time (user+sys) for any thread, not only for current.
4369 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4370 // It might work on 2.6.10+ with a special kernel/glibc patch.
4371 // For reference, please, see IEEE Std 1003.1-2004:
4372 // http://www.unix.org/single_unix_specification
4373
4374 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4375 struct timespec tp;
4376 int rc = os::Linux::clock_gettime(clockid, &tp);
4377 assert(rc == 0, "clock_gettime is expected to return 0 code");
4378
4379 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4380 }
4381
4382 /////
4383 // glibc on Linux platform uses non-documented flag
4384 // to indicate, that some special sort of signal
4385 // trampoline is used.
4386 // We will never set this flag, and we should
4387 // ignore this flag in our diagnostic
4388 #ifdef SIGNIFICANT_SIGNAL_MASK
4389 #undef SIGNIFICANT_SIGNAL_MASK
4390 #endif
4391 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4392
4393 static const char* get_signal_handler_name(address handler,
4394 char* buf, int buflen) {
4395 int offset = 0;
4396 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4397 if (found) {
4398 // skip directory names
4399 const char *p1, *p2;
4400 p1 = buf;
4401 size_t len = strlen(os::file_separator());
4402 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4403 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4404 } else {
4405 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4406 }
4407 return buf;
4408 }
4409
4410 static void print_signal_handler(outputStream* st, int sig,
4411 char* buf, size_t buflen) {
4412 struct sigaction sa;
4413
4414 sigaction(sig, NULL, &sa);
4415
4416 // See comment for SIGNIFICANT_SIGNAL_MASK define
4417 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4418
4419 st->print("%s: ", os::exception_name(sig, buf, buflen));
4420
4421 address handler = (sa.sa_flags & SA_SIGINFO)
4422 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4423 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4424
4425 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4426 st->print("SIG_DFL");
4427 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4428 st->print("SIG_IGN");
4429 } else {
4430 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4431 }
4432
4433 st->print(", sa_mask[0]=");
4434 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4435
4436 address rh = VMError::get_resetted_sighandler(sig);
4437 // May be, handler was resetted by VMError?
4438 if (rh != NULL) {
4439 handler = rh;
4440 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4441 }
4442
4443 st->print(", sa_flags=");
4444 os::Posix::print_sa_flags(st, sa.sa_flags);
4445
4446 // Check: is it our handler?
4447 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4448 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4449 // It is our signal handler
4450 // check for flags, reset system-used one!
4451 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4452 st->print(
4453 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4454 os::Linux::get_our_sigflags(sig));
4455 }
4456 }
4457 st->cr();
4458 }
4459
4460
4461 #define DO_SIGNAL_CHECK(sig) \
4462 do { \
4463 if (!sigismember(&check_signal_done, sig)) { \
4464 os::Linux::check_signal_handler(sig); \
4465 } \
4466 } while (0)
4467
4468 // This method is a periodic task to check for misbehaving JNI applications
4469 // under CheckJNI, we can add any periodic checks here
4470
4471 void os::run_periodic_checks() {
4472 if (check_signals == false) return;
4473
4474 // SEGV and BUS if overridden could potentially prevent
4475 // generation of hs*.log in the event of a crash, debugging
4476 // such a case can be very challenging, so we absolutely
4477 // check the following for a good measure:
4478 DO_SIGNAL_CHECK(SIGSEGV);
4479 DO_SIGNAL_CHECK(SIGILL);
4480 DO_SIGNAL_CHECK(SIGFPE);
4481 DO_SIGNAL_CHECK(SIGBUS);
4482 DO_SIGNAL_CHECK(SIGPIPE);
4483 DO_SIGNAL_CHECK(SIGXFSZ);
4484 #if defined(PPC64)
4485 DO_SIGNAL_CHECK(SIGTRAP);
4486 #endif
4487
4488 // ReduceSignalUsage allows the user to override these handlers
4489 // see comments at the very top and jvm_solaris.h
4490 if (!ReduceSignalUsage) {
4491 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4492 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4493 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4494 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4495 }
4496
4497 DO_SIGNAL_CHECK(SR_signum);
4498 }
4499
4500 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4501
4502 static os_sigaction_t os_sigaction = NULL;
4503
4504 void os::Linux::check_signal_handler(int sig) {
4505 char buf[O_BUFLEN];
4506 address jvmHandler = NULL;
4507
4508
4509 struct sigaction act;
4510 if (os_sigaction == NULL) {
4511 // only trust the default sigaction, in case it has been interposed
4512 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4513 if (os_sigaction == NULL) return;
4514 }
4515
4516 os_sigaction(sig, (struct sigaction*)NULL, &act);
4517
4518
4519 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4520
4521 address thisHandler = (act.sa_flags & SA_SIGINFO)
4522 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4523 : CAST_FROM_FN_PTR(address, act.sa_handler);
4524
4525
4526 switch (sig) {
4527 case SIGSEGV:
4528 case SIGBUS:
4529 case SIGFPE:
4530 case SIGPIPE:
4531 case SIGILL:
4532 case SIGXFSZ:
4533 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4534 break;
4535
4536 case SHUTDOWN1_SIGNAL:
4537 case SHUTDOWN2_SIGNAL:
4538 case SHUTDOWN3_SIGNAL:
4539 case BREAK_SIGNAL:
4540 jvmHandler = (address)user_handler();
4541 break;
4542
4543 default:
4544 if (sig == SR_signum) {
4545 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4546 } else {
4547 return;
4548 }
4549 break;
4550 }
4551
4552 if (thisHandler != jvmHandler) {
4553 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4554 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4555 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4556 // No need to check this sig any longer
4557 sigaddset(&check_signal_done, sig);
4558 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4559 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4560 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4561 exception_name(sig, buf, O_BUFLEN));
4562 }
4563 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4564 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4565 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4566 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4567 // No need to check this sig any longer
4568 sigaddset(&check_signal_done, sig);
4569 }
4570
4571 // Dump all the signal
4572 if (sigismember(&check_signal_done, sig)) {
4573 print_signal_handlers(tty, buf, O_BUFLEN);
4574 }
4575 }
4576
4577 extern void report_error(char* file_name, int line_no, char* title,
4578 char* format, ...);
4579
4580 extern bool signal_name(int signo, char* buf, size_t len);
4581
4582 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4583 if (0 < exception_code && exception_code <= SIGRTMAX) {
4584 // signal
4585 if (!signal_name(exception_code, buf, size)) {
4586 jio_snprintf(buf, size, "SIG%d", exception_code);
4587 }
4588 return buf;
4589 } else {
4590 return NULL;
4591 }
4592 }
4593
4594 // this is called _before_ the most of global arguments have been parsed
4595 void os::init(void) {
4596 char dummy; // used to get a guess on initial stack address
4597 // first_hrtime = gethrtime();
4598
4599 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4600
4601 init_random(1234567);
4602
4603 ThreadCritical::initialize();
4604
4605 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4606 if (Linux::page_size() == -1) {
4607 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4608 strerror(errno));
4609 }
4610 init_page_sizes((size_t) Linux::page_size());
4611
4612 Linux::initialize_system_info();
4613
4614 // main_thread points to the aboriginal thread
4615 Linux::_main_thread = pthread_self();
4616
4617 Linux::clock_init();
4618 initial_time_count = javaTimeNanos();
4619
4620 // pthread_condattr initialization for monotonic clock
4621 int status;
4622 pthread_condattr_t* _condattr = os::Linux::condAttr();
4623 if ((status = pthread_condattr_init(_condattr)) != 0) {
4624 fatal("pthread_condattr_init: %s", strerror(status));
4625 }
4626 // Only set the clock if CLOCK_MONOTONIC is available
4627 if (os::supports_monotonic_clock()) {
4628 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4629 if (status == EINVAL) {
4630 warning("Unable to use monotonic clock with relative timed-waits" \
4631 " - changes to the time-of-day clock may have adverse affects");
4632 } else {
4633 fatal("pthread_condattr_setclock: %s", strerror(status));
4634 }
4635 }
4636 }
4637 // else it defaults to CLOCK_REALTIME
4638
4639 // If the pagesize of the VM is greater than 8K determine the appropriate
4640 // number of initial guard pages. The user can change this with the
4641 // command line arguments, if needed.
4642 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4643 StackYellowPages = 1;
4644 StackRedPages = 1;
4645 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4646 }
4647
4648 // retrieve entry point for pthread_setname_np
4649 Linux::_pthread_setname_np =
4650 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4651
4652 }
4653
4654 // To install functions for atexit system call
4655 extern "C" {
4656 static void perfMemory_exit_helper() {
4657 perfMemory_exit();
4658 }
4659 }
4660
4661 // this is called _after_ the global arguments have been parsed
4662 jint os::init_2(void) {
4663 Linux::fast_thread_clock_init();
4664
4665 // Allocate a single page and mark it as readable for safepoint polling
4666 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4667 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4668
4669 os::set_polling_page(polling_page);
4670
4671 #ifndef PRODUCT
4672 if (Verbose && PrintMiscellaneous) {
4673 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4674 (intptr_t)polling_page);
4675 }
4676 #endif
4677
4678 if (!UseMembar) {
4679 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4680 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4681 os::set_memory_serialize_page(mem_serialize_page);
4682
4683 #ifndef PRODUCT
4684 if (Verbose && PrintMiscellaneous) {
4685 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4686 (intptr_t)mem_serialize_page);
4687 }
4688 #endif
4689 }
4690
4691 // initialize suspend/resume support - must do this before signal_sets_init()
4692 if (SR_initialize() != 0) {
4693 perror("SR_initialize failed");
4694 return JNI_ERR;
4695 }
4696
4697 Linux::signal_sets_init();
4698 Linux::install_signal_handlers();
4699
4700 // Check minimum allowable stack size for thread creation and to initialize
4701 // the java system classes, including StackOverflowError - depends on page
4702 // size. Add a page for compiler2 recursion in main thread.
4703 // Add in 2*BytesPerWord times page size to account for VM stack during
4704 // class initialization depending on 32 or 64 bit VM.
4705 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4706 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4707 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4708
4709 size_t threadStackSizeInBytes = ThreadStackSize * K;
4710 if (threadStackSizeInBytes != 0 &&
4711 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4712 tty->print_cr("\nThe stack size specified is too small, "
4713 "Specify at least " SIZE_FORMAT "k",
4714 os::Linux::min_stack_allowed/ K);
4715 return JNI_ERR;
4716 }
4717
4718 // Make the stack size a multiple of the page size so that
4719 // the yellow/red zones can be guarded.
4720 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4721 vm_page_size()));
4722
4723 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4724
4725 #if defined(IA32)
4726 workaround_expand_exec_shield_cs_limit();
4727 #endif
4728
4729 Linux::libpthread_init();
4730 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4731 tty->print_cr("[HotSpot is running with %s, %s]\n",
4732 Linux::glibc_version(), Linux::libpthread_version());
4733 }
4734
4735 if (UseNUMA) {
4736 if (!Linux::libnuma_init()) {
4737 UseNUMA = false;
4738 } else {
4739 if ((Linux::numa_max_node() < 1)) {
4740 // There's only one node(they start from 0), disable NUMA.
4741 UseNUMA = false;
4742 }
4743 }
4744 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4745 // we can make the adaptive lgrp chunk resizing work. If the user specified
4746 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4747 // disable adaptive resizing.
4748 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4749 if (FLAG_IS_DEFAULT(UseNUMA)) {
4750 UseNUMA = false;
4751 } else {
4752 if (FLAG_IS_DEFAULT(UseLargePages) &&
4753 FLAG_IS_DEFAULT(UseSHM) &&
4754 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4755 UseLargePages = false;
4756 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4757 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4758 UseAdaptiveSizePolicy = false;
4759 UseAdaptiveNUMAChunkSizing = false;
4760 }
4761 }
4762 }
4763 if (!UseNUMA && ForceNUMA) {
4764 UseNUMA = true;
4765 }
4766 }
4767
4768 if (MaxFDLimit) {
4769 // set the number of file descriptors to max. print out error
4770 // if getrlimit/setrlimit fails but continue regardless.
4771 struct rlimit nbr_files;
4772 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4773 if (status != 0) {
4774 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4775 perror("os::init_2 getrlimit failed");
4776 }
4777 } else {
4778 nbr_files.rlim_cur = nbr_files.rlim_max;
4779 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4780 if (status != 0) {
4781 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4782 perror("os::init_2 setrlimit failed");
4783 }
4784 }
4785 }
4786 }
4787
4788 // Initialize lock used to serialize thread creation (see os::create_thread)
4789 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4790
4791 // at-exit methods are called in the reverse order of their registration.
4792 // atexit functions are called on return from main or as a result of a
4793 // call to exit(3C). There can be only 32 of these functions registered
4794 // and atexit() does not set errno.
4795
4796 if (PerfAllowAtExitRegistration) {
4797 // only register atexit functions if PerfAllowAtExitRegistration is set.
4798 // atexit functions can be delayed until process exit time, which
4799 // can be problematic for embedded VM situations. Embedded VMs should
4800 // call DestroyJavaVM() to assure that VM resources are released.
4801
4802 // note: perfMemory_exit_helper atexit function may be removed in
4803 // the future if the appropriate cleanup code can be added to the
4804 // VM_Exit VMOperation's doit method.
4805 if (atexit(perfMemory_exit_helper) != 0) {
4806 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4807 }
4808 }
4809
4810 // initialize thread priority policy
4811 prio_init();
4812
4813 return JNI_OK;
4814 }
4815
4816 // Mark the polling page as unreadable
4817 void os::make_polling_page_unreadable(void) {
4818 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4819 fatal("Could not disable polling page");
4820 }
4821 }
4822
4823 // Mark the polling page as readable
4824 void os::make_polling_page_readable(void) {
4825 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4826 fatal("Could not enable polling page");
4827 }
4828 }
4829
4830 int os::active_processor_count() {
4831 // Linux doesn't yet have a (official) notion of processor sets,
4832 // so just return the number of online processors.
4833 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4834 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4835 return online_cpus;
4836 }
4837
4838 void os::set_native_thread_name(const char *name) {
4839 if (Linux::_pthread_setname_np) {
4840 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4841 snprintf(buf, sizeof(buf), "%s", name);
4842 buf[sizeof(buf) - 1] = '\0';
4843 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4844 // ERANGE should not happen; all other errors should just be ignored.
4845 assert(rc != ERANGE, "pthread_setname_np failed");
4846 }
4847 }
4848
4849 bool os::distribute_processes(uint length, uint* distribution) {
4850 // Not yet implemented.
4851 return false;
4852 }
4853
4854 bool os::bind_to_processor(uint processor_id) {
4855 // Not yet implemented.
4856 return false;
4857 }
4858
4859 ///
4860
4861 void os::SuspendedThreadTask::internal_do_task() {
4862 if (do_suspend(_thread->osthread())) {
4863 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4864 do_task(context);
4865 do_resume(_thread->osthread());
4866 }
4867 }
4868
4869 class PcFetcher : public os::SuspendedThreadTask {
4870 public:
4871 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4872 ExtendedPC result();
4873 protected:
4874 void do_task(const os::SuspendedThreadTaskContext& context);
4875 private:
4876 ExtendedPC _epc;
4877 };
4878
4879 ExtendedPC PcFetcher::result() {
4880 guarantee(is_done(), "task is not done yet.");
4881 return _epc;
4882 }
4883
4884 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4885 Thread* thread = context.thread();
4886 OSThread* osthread = thread->osthread();
4887 if (osthread->ucontext() != NULL) {
4888 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4889 } else {
4890 // NULL context is unexpected, double-check this is the VMThread
4891 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4892 }
4893 }
4894
4895 // Suspends the target using the signal mechanism and then grabs the PC before
4896 // resuming the target. Used by the flat-profiler only
4897 ExtendedPC os::get_thread_pc(Thread* thread) {
4898 // Make sure that it is called by the watcher for the VMThread
4899 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4900 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4901
4902 PcFetcher fetcher(thread);
4903 fetcher.run();
4904 return fetcher.result();
4905 }
4906
4907 ////////////////////////////////////////////////////////////////////////////////
4908 // debug support
4909
4910 bool os::find(address addr, outputStream* st) {
4911 Dl_info dlinfo;
4912 memset(&dlinfo, 0, sizeof(dlinfo));
4913 if (dladdr(addr, &dlinfo) != 0) {
4914 st->print(PTR_FORMAT ": ", p2i(addr));
4915 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4916 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4917 p2i(addr) - p2i(dlinfo.dli_saddr));
4918 } else if (dlinfo.dli_fbase != NULL) {
4919 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4920 } else {
4921 st->print("<absolute address>");
4922 }
4923 if (dlinfo.dli_fname != NULL) {
4924 st->print(" in %s", dlinfo.dli_fname);
4925 }
4926 if (dlinfo.dli_fbase != NULL) {
4927 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4928 }
4929 st->cr();
4930
4931 if (Verbose) {
4932 // decode some bytes around the PC
4933 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4934 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4935 address lowest = (address) dlinfo.dli_sname;
4936 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4937 if (begin < lowest) begin = lowest;
4938 Dl_info dlinfo2;
4939 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4940 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4941 end = (address) dlinfo2.dli_saddr;
4942 }
4943 Disassembler::decode(begin, end, st);
4944 }
4945 return true;
4946 }
4947 return false;
4948 }
4949
4950 ////////////////////////////////////////////////////////////////////////////////
4951 // misc
4952
4953 // This does not do anything on Linux. This is basically a hook for being
4954 // able to use structured exception handling (thread-local exception filters)
4955 // on, e.g., Win32.
4956 void
4957 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
4958 JavaCallArguments* args, Thread* thread) {
4959 f(value, method, args, thread);
4960 }
4961
4962 void os::print_statistics() {
4963 }
4964
4965 int os::message_box(const char* title, const char* message) {
4966 int i;
4967 fdStream err(defaultStream::error_fd());
4968 for (i = 0; i < 78; i++) err.print_raw("=");
4969 err.cr();
4970 err.print_raw_cr(title);
4971 for (i = 0; i < 78; i++) err.print_raw("-");
4972 err.cr();
4973 err.print_raw_cr(message);
4974 for (i = 0; i < 78; i++) err.print_raw("=");
4975 err.cr();
4976
4977 char buf[16];
4978 // Prevent process from exiting upon "read error" without consuming all CPU
4979 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4980
4981 return buf[0] == 'y' || buf[0] == 'Y';
4982 }
4983
4984 int os::stat(const char *path, struct stat *sbuf) {
4985 char pathbuf[MAX_PATH];
4986 if (strlen(path) > MAX_PATH - 1) {
4987 errno = ENAMETOOLONG;
4988 return -1;
4989 }
4990 os::native_path(strcpy(pathbuf, path));
4991 return ::stat(pathbuf, sbuf);
4992 }
4993
4994 bool os::check_heap(bool force) {
4995 return true;
4996 }
4997
4998 // Is a (classpath) directory empty?
4999 bool os::dir_is_empty(const char* path) {
5000 DIR *dir = NULL;
5001 struct dirent *ptr;
5002
5003 dir = opendir(path);
5004 if (dir == NULL) return true;
5005
5006 // Scan the directory
5007 bool result = true;
5008 char buf[sizeof(struct dirent) + MAX_PATH];
5009 while (result && (ptr = ::readdir(dir)) != NULL) {
5010 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5011 result = false;
5012 }
5013 }
5014 closedir(dir);
5015 return result;
5016 }
5017
5018 // This code originates from JDK's sysOpen and open64_w
5019 // from src/solaris/hpi/src/system_md.c
5020
5021 int os::open(const char *path, int oflag, int mode) {
5022 if (strlen(path) > MAX_PATH - 1) {
5023 errno = ENAMETOOLONG;
5024 return -1;
5025 }
5026
5027 // All file descriptors that are opened in the Java process and not
5028 // specifically destined for a subprocess should have the close-on-exec
5029 // flag set. If we don't set it, then careless 3rd party native code
5030 // might fork and exec without closing all appropriate file descriptors
5031 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5032 // turn might:
5033 //
5034 // - cause end-of-file to fail to be detected on some file
5035 // descriptors, resulting in mysterious hangs, or
5036 //
5037 // - might cause an fopen in the subprocess to fail on a system
5038 // suffering from bug 1085341.
5039 //
5040 // (Yes, the default setting of the close-on-exec flag is a Unix
5041 // design flaw)
5042 //
5043 // See:
5044 // 1085341: 32-bit stdio routines should support file descriptors >255
5045 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5046 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5047 //
5048 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5049 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5050 // because it saves a system call and removes a small window where the flag
5051 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5052 // and we fall back to using FD_CLOEXEC (see below).
5053 #ifdef O_CLOEXEC
5054 oflag |= O_CLOEXEC;
5055 #endif
5056
5057 int fd = ::open64(path, oflag, mode);
5058 if (fd == -1) return -1;
5059
5060 //If the open succeeded, the file might still be a directory
5061 {
5062 struct stat64 buf64;
5063 int ret = ::fstat64(fd, &buf64);
5064 int st_mode = buf64.st_mode;
5065
5066 if (ret != -1) {
5067 if ((st_mode & S_IFMT) == S_IFDIR) {
5068 errno = EISDIR;
5069 ::close(fd);
5070 return -1;
5071 }
5072 } else {
5073 ::close(fd);
5074 return -1;
5075 }
5076 }
5077
5078 #ifdef FD_CLOEXEC
5079 // Validate that the use of the O_CLOEXEC flag on open above worked.
5080 // With recent kernels, we will perform this check exactly once.
5081 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5082 if (!O_CLOEXEC_is_known_to_work) {
5083 int flags = ::fcntl(fd, F_GETFD);
5084 if (flags != -1) {
5085 if ((flags & FD_CLOEXEC) != 0)
5086 O_CLOEXEC_is_known_to_work = 1;
5087 else
5088 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5089 }
5090 }
5091 #endif
5092
5093 return fd;
5094 }
5095
5096
5097 // create binary file, rewriting existing file if required
5098 int os::create_binary_file(const char* path, bool rewrite_existing) {
5099 int oflags = O_WRONLY | O_CREAT;
5100 if (!rewrite_existing) {
5101 oflags |= O_EXCL;
5102 }
5103 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5104 }
5105
5106 // return current position of file pointer
5107 jlong os::current_file_offset(int fd) {
5108 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5109 }
5110
5111 // move file pointer to the specified offset
5112 jlong os::seek_to_file_offset(int fd, jlong offset) {
5113 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5114 }
5115
5116 // This code originates from JDK's sysAvailable
5117 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5118
5119 int os::available(int fd, jlong *bytes) {
5120 jlong cur, end;
5121 int mode;
5122 struct stat64 buf64;
5123
5124 if (::fstat64(fd, &buf64) >= 0) {
5125 mode = buf64.st_mode;
5126 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5127 int n;
5128 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5129 *bytes = n;
5130 return 1;
5131 }
5132 }
5133 }
5134 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5135 return 0;
5136 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5137 return 0;
5138 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5139 return 0;
5140 }
5141 *bytes = end - cur;
5142 return 1;
5143 }
5144
5145 // Map a block of memory.
5146 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5147 char *addr, size_t bytes, bool read_only,
5148 bool allow_exec) {
5149 int prot;
5150 int flags = MAP_PRIVATE;
5151
5152 if (read_only) {
5153 prot = PROT_READ;
5154 } else {
5155 prot = PROT_READ | PROT_WRITE;
5156 }
5157
5158 if (allow_exec) {
5159 prot |= PROT_EXEC;
5160 }
5161
5162 if (addr != NULL) {
5163 flags |= MAP_FIXED;
5164 }
5165
5166 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5167 fd, file_offset);
5168 if (mapped_address == MAP_FAILED) {
5169 return NULL;
5170 }
5171 return mapped_address;
5172 }
5173
5174
5175 // Remap a block of memory.
5176 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5177 char *addr, size_t bytes, bool read_only,
5178 bool allow_exec) {
5179 // same as map_memory() on this OS
5180 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5181 allow_exec);
5182 }
5183
5184
5185 // Unmap a block of memory.
5186 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5187 return munmap(addr, bytes) == 0;
5188 }
5189
5190 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5191
5192 static clockid_t thread_cpu_clockid(Thread* thread) {
5193 pthread_t tid = thread->osthread()->pthread_id();
5194 clockid_t clockid;
5195
5196 // Get thread clockid
5197 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5198 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5199 return clockid;
5200 }
5201
5202 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5203 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5204 // of a thread.
5205 //
5206 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5207 // the fast estimate available on the platform.
5208
5209 jlong os::current_thread_cpu_time() {
5210 if (os::Linux::supports_fast_thread_cpu_time()) {
5211 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5212 } else {
5213 // return user + sys since the cost is the same
5214 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5215 }
5216 }
5217
5218 jlong os::thread_cpu_time(Thread* thread) {
5219 // consistent with what current_thread_cpu_time() returns
5220 if (os::Linux::supports_fast_thread_cpu_time()) {
5221 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5222 } else {
5223 return slow_thread_cpu_time(thread, true /* user + sys */);
5224 }
5225 }
5226
5227 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5228 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5229 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5230 } else {
5231 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5232 }
5233 }
5234
5235 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5236 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5237 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5238 } else {
5239 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5240 }
5241 }
5242
5243 // -1 on error.
5244 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5245 pid_t tid = thread->osthread()->thread_id();
5246 char *s;
5247 char stat[2048];
5248 int statlen;
5249 char proc_name[64];
5250 int count;
5251 long sys_time, user_time;
5252 char cdummy;
5253 int idummy;
5254 long ldummy;
5255 FILE *fp;
5256
5257 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5258 fp = fopen(proc_name, "r");
5259 if (fp == NULL) return -1;
5260 statlen = fread(stat, 1, 2047, fp);
5261 stat[statlen] = '\0';
5262 fclose(fp);
5263
5264 // Skip pid and the command string. Note that we could be dealing with
5265 // weird command names, e.g. user could decide to rename java launcher
5266 // to "java 1.4.2 :)", then the stat file would look like
5267 // 1234 (java 1.4.2 :)) R ... ...
5268 // We don't really need to know the command string, just find the last
5269 // occurrence of ")" and then start parsing from there. See bug 4726580.
5270 s = strrchr(stat, ')');
5271 if (s == NULL) return -1;
5272
5273 // Skip blank chars
5274 do { s++; } while (s && isspace(*s));
5275
5276 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5277 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5278 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5279 &user_time, &sys_time);
5280 if (count != 13) return -1;
5281 if (user_sys_cpu_time) {
5282 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5283 } else {
5284 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5285 }
5286 }
5287
5288 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5289 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5290 info_ptr->may_skip_backward = false; // elapsed time not wall time
5291 info_ptr->may_skip_forward = false; // elapsed time not wall time
5292 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5293 }
5294
5295 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5296 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5297 info_ptr->may_skip_backward = false; // elapsed time not wall time
5298 info_ptr->may_skip_forward = false; // elapsed time not wall time
5299 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5300 }
5301
5302 bool os::is_thread_cpu_time_supported() {
5303 return true;
5304 }
5305
5306 // System loadavg support. Returns -1 if load average cannot be obtained.
5307 // Linux doesn't yet have a (official) notion of processor sets,
5308 // so just return the system wide load average.
5309 int os::loadavg(double loadavg[], int nelem) {
5310 return ::getloadavg(loadavg, nelem);
5311 }
5312
5313 void os::pause() {
5314 char filename[MAX_PATH];
5315 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5316 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5317 } else {
5318 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5319 }
5320
5321 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5322 if (fd != -1) {
5323 struct stat buf;
5324 ::close(fd);
5325 while (::stat(filename, &buf) == 0) {
5326 (void)::poll(NULL, 0, 100);
5327 }
5328 } else {
5329 jio_fprintf(stderr,
5330 "Could not open pause file '%s', continuing immediately.\n", filename);
5331 }
5332 }
5333
5334
5335 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5336 // comment paragraphs are worth repeating here:
5337 //
5338 // Assumption:
5339 // Only one parker can exist on an event, which is why we allocate
5340 // them per-thread. Multiple unparkers can coexist.
5341 //
5342 // _Event serves as a restricted-range semaphore.
5343 // -1 : thread is blocked, i.e. there is a waiter
5344 // 0 : neutral: thread is running or ready,
5345 // could have been signaled after a wait started
5346 // 1 : signaled - thread is running or ready
5347 //
5348 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5349 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5350 // For specifics regarding the bug see GLIBC BUGID 261237 :
5351 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5352 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5353 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5354 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5355 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5356 // and monitorenter when we're using 1-0 locking. All those operations may result in
5357 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5358 // of libpthread avoids the problem, but isn't practical.
5359 //
5360 // Possible remedies:
5361 //
5362 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5363 // This is palliative and probabilistic, however. If the thread is preempted
5364 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5365 // than the minimum period may have passed, and the abstime may be stale (in the
5366 // past) resultin in a hang. Using this technique reduces the odds of a hang
5367 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5368 //
5369 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5370 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5371 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5372 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5373 // thread.
5374 //
5375 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5376 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5377 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5378 // This also works well. In fact it avoids kernel-level scalability impediments
5379 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5380 // timers in a graceful fashion.
5381 //
5382 // 4. When the abstime value is in the past it appears that control returns
5383 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5384 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5385 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5386 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5387 // It may be possible to avoid reinitialization by checking the return
5388 // value from pthread_cond_timedwait(). In addition to reinitializing the
5389 // condvar we must establish the invariant that cond_signal() is only called
5390 // within critical sections protected by the adjunct mutex. This prevents
5391 // cond_signal() from "seeing" a condvar that's in the midst of being
5392 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5393 // desirable signal-after-unlock optimization that avoids futile context switching.
5394 //
5395 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5396 // structure when a condvar is used or initialized. cond_destroy() would
5397 // release the helper structure. Our reinitialize-after-timedwait fix
5398 // put excessive stress on malloc/free and locks protecting the c-heap.
5399 //
5400 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5401 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5402 // and only enabling the work-around for vulnerable environments.
5403
5404 // utility to compute the abstime argument to timedwait:
5405 // millis is the relative timeout time
5406 // abstime will be the absolute timeout time
5407 // TODO: replace compute_abstime() with unpackTime()
5408
5409 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5410 if (millis < 0) millis = 0;
5411
5412 jlong seconds = millis / 1000;
5413 millis %= 1000;
5414 if (seconds > 50000000) { // see man cond_timedwait(3T)
5415 seconds = 50000000;
5416 }
5417
5418 if (os::supports_monotonic_clock()) {
5419 struct timespec now;
5420 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5421 assert_status(status == 0, status, "clock_gettime");
5422 abstime->tv_sec = now.tv_sec + seconds;
5423 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5424 if (nanos >= NANOSECS_PER_SEC) {
5425 abstime->tv_sec += 1;
5426 nanos -= NANOSECS_PER_SEC;
5427 }
5428 abstime->tv_nsec = nanos;
5429 } else {
5430 struct timeval now;
5431 int status = gettimeofday(&now, NULL);
5432 assert(status == 0, "gettimeofday");
5433 abstime->tv_sec = now.tv_sec + seconds;
5434 long usec = now.tv_usec + millis * 1000;
5435 if (usec >= 1000000) {
5436 abstime->tv_sec += 1;
5437 usec -= 1000000;
5438 }
5439 abstime->tv_nsec = usec * 1000;
5440 }
5441 return abstime;
5442 }
5443
5444 void os::PlatformEvent::park() { // AKA "down()"
5445 // Transitions for _Event:
5446 // -1 => -1 : illegal
5447 // 1 => 0 : pass - return immediately
5448 // 0 => -1 : block; then set _Event to 0 before returning
5449
5450 // Invariant: Only the thread associated with the Event/PlatformEvent
5451 // may call park().
5452 // TODO: assert that _Assoc != NULL or _Assoc == Self
5453 assert(_nParked == 0, "invariant");
5454
5455 int v;
5456 for (;;) {
5457 v = _Event;
5458 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5459 }
5460 guarantee(v >= 0, "invariant");
5461 if (v == 0) {
5462 // Do this the hard way by blocking ...
5463 int status = pthread_mutex_lock(_mutex);
5464 assert_status(status == 0, status, "mutex_lock");
5465 guarantee(_nParked == 0, "invariant");
5466 ++_nParked;
5467 while (_Event < 0) {
5468 status = pthread_cond_wait(_cond, _mutex);
5469 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5470 // Treat this the same as if the wait was interrupted
5471 if (status == ETIME) { status = EINTR; }
5472 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5473 }
5474 --_nParked;
5475
5476 _Event = 0;
5477 status = pthread_mutex_unlock(_mutex);
5478 assert_status(status == 0, status, "mutex_unlock");
5479 // Paranoia to ensure our locked and lock-free paths interact
5480 // correctly with each other.
5481 OrderAccess::fence();
5482 }
5483 guarantee(_Event >= 0, "invariant");
5484 }
5485
5486 int os::PlatformEvent::park(jlong millis) {
5487 // Transitions for _Event:
5488 // -1 => -1 : illegal
5489 // 1 => 0 : pass - return immediately
5490 // 0 => -1 : block; then set _Event to 0 before returning
5491
5492 guarantee(_nParked == 0, "invariant");
5493
5494 int v;
5495 for (;;) {
5496 v = _Event;
5497 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5498 }
5499 guarantee(v >= 0, "invariant");
5500 if (v != 0) return OS_OK;
5501
5502 // We do this the hard way, by blocking the thread.
5503 // Consider enforcing a minimum timeout value.
5504 struct timespec abst;
5505 compute_abstime(&abst, millis);
5506
5507 int ret = OS_TIMEOUT;
5508 int status = pthread_mutex_lock(_mutex);
5509 assert_status(status == 0, status, "mutex_lock");
5510 guarantee(_nParked == 0, "invariant");
5511 ++_nParked;
5512
5513 // Object.wait(timo) will return because of
5514 // (a) notification
5515 // (b) timeout
5516 // (c) thread.interrupt
5517 //
5518 // Thread.interrupt and object.notify{All} both call Event::set.
5519 // That is, we treat thread.interrupt as a special case of notification.
5520 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5521 // We assume all ETIME returns are valid.
5522 //
5523 // TODO: properly differentiate simultaneous notify+interrupt.
5524 // In that case, we should propagate the notify to another waiter.
5525
5526 while (_Event < 0) {
5527 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5528 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5529 pthread_cond_destroy(_cond);
5530 pthread_cond_init(_cond, os::Linux::condAttr());
5531 }
5532 assert_status(status == 0 || status == EINTR ||
5533 status == ETIME || status == ETIMEDOUT,
5534 status, "cond_timedwait");
5535 if (!FilterSpuriousWakeups) break; // previous semantics
5536 if (status == ETIME || status == ETIMEDOUT) break;
5537 // We consume and ignore EINTR and spurious wakeups.
5538 }
5539 --_nParked;
5540 if (_Event >= 0) {
5541 ret = OS_OK;
5542 }
5543 _Event = 0;
5544 status = pthread_mutex_unlock(_mutex);
5545 assert_status(status == 0, status, "mutex_unlock");
5546 assert(_nParked == 0, "invariant");
5547 // Paranoia to ensure our locked and lock-free paths interact
5548 // correctly with each other.
5549 OrderAccess::fence();
5550 return ret;
5551 }
5552
5553 void os::PlatformEvent::unpark() {
5554 // Transitions for _Event:
5555 // 0 => 1 : just return
5556 // 1 => 1 : just return
5557 // -1 => either 0 or 1; must signal target thread
5558 // That is, we can safely transition _Event from -1 to either
5559 // 0 or 1.
5560 // See also: "Semaphores in Plan 9" by Mullender & Cox
5561 //
5562 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5563 // that it will take two back-to-back park() calls for the owning
5564 // thread to block. This has the benefit of forcing a spurious return
5565 // from the first park() call after an unpark() call which will help
5566 // shake out uses of park() and unpark() without condition variables.
5567
5568 if (Atomic::xchg(1, &_Event) >= 0) return;
5569
5570 // Wait for the thread associated with the event to vacate
5571 int status = pthread_mutex_lock(_mutex);
5572 assert_status(status == 0, status, "mutex_lock");
5573 int AnyWaiters = _nParked;
5574 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5575 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5576 AnyWaiters = 0;
5577 pthread_cond_signal(_cond);
5578 }
5579 status = pthread_mutex_unlock(_mutex);
5580 assert_status(status == 0, status, "mutex_unlock");
5581 if (AnyWaiters != 0) {
5582 // Note that we signal() *after* dropping the lock for "immortal" Events.
5583 // This is safe and avoids a common class of futile wakeups. In rare
5584 // circumstances this can cause a thread to return prematurely from
5585 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5586 // will simply re-test the condition and re-park itself.
5587 // This provides particular benefit if the underlying platform does not
5588 // provide wait morphing.
5589 status = pthread_cond_signal(_cond);
5590 assert_status(status == 0, status, "cond_signal");
5591 }
5592 }
5593
5594
5595 // JSR166
5596 // -------------------------------------------------------
5597
5598 // The solaris and linux implementations of park/unpark are fairly
5599 // conservative for now, but can be improved. They currently use a
5600 // mutex/condvar pair, plus a a count.
5601 // Park decrements count if > 0, else does a condvar wait. Unpark
5602 // sets count to 1 and signals condvar. Only one thread ever waits
5603 // on the condvar. Contention seen when trying to park implies that someone
5604 // is unparking you, so don't wait. And spurious returns are fine, so there
5605 // is no need to track notifications.
5606
5607 // This code is common to linux and solaris and will be moved to a
5608 // common place in dolphin.
5609 //
5610 // The passed in time value is either a relative time in nanoseconds
5611 // or an absolute time in milliseconds. Either way it has to be unpacked
5612 // into suitable seconds and nanoseconds components and stored in the
5613 // given timespec structure.
5614 // Given time is a 64-bit value and the time_t used in the timespec is only
5615 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5616 // overflow if times way in the future are given. Further on Solaris versions
5617 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5618 // number of seconds, in abstime, is less than current_time + 100,000,000.
5619 // As it will be 28 years before "now + 100000000" will overflow we can
5620 // ignore overflow and just impose a hard-limit on seconds using the value
5621 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5622 // years from "now".
5623
5624 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5625 assert(time > 0, "convertTime");
5626 time_t max_secs = 0;
5627
5628 if (!os::supports_monotonic_clock() || isAbsolute) {
5629 struct timeval now;
5630 int status = gettimeofday(&now, NULL);
5631 assert(status == 0, "gettimeofday");
5632
5633 max_secs = now.tv_sec + MAX_SECS;
5634
5635 if (isAbsolute) {
5636 jlong secs = time / 1000;
5637 if (secs > max_secs) {
5638 absTime->tv_sec = max_secs;
5639 } else {
5640 absTime->tv_sec = secs;
5641 }
5642 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5643 } else {
5644 jlong secs = time / NANOSECS_PER_SEC;
5645 if (secs >= MAX_SECS) {
5646 absTime->tv_sec = max_secs;
5647 absTime->tv_nsec = 0;
5648 } else {
5649 absTime->tv_sec = now.tv_sec + secs;
5650 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5651 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5652 absTime->tv_nsec -= NANOSECS_PER_SEC;
5653 ++absTime->tv_sec; // note: this must be <= max_secs
5654 }
5655 }
5656 }
5657 } else {
5658 // must be relative using monotonic clock
5659 struct timespec now;
5660 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5661 assert_status(status == 0, status, "clock_gettime");
5662 max_secs = now.tv_sec + MAX_SECS;
5663 jlong secs = time / NANOSECS_PER_SEC;
5664 if (secs >= MAX_SECS) {
5665 absTime->tv_sec = max_secs;
5666 absTime->tv_nsec = 0;
5667 } else {
5668 absTime->tv_sec = now.tv_sec + secs;
5669 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5670 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5671 absTime->tv_nsec -= NANOSECS_PER_SEC;
5672 ++absTime->tv_sec; // note: this must be <= max_secs
5673 }
5674 }
5675 }
5676 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5677 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5678 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5679 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5680 }
5681
5682 void Parker::park(bool isAbsolute, jlong time) {
5683 // Ideally we'd do something useful while spinning, such
5684 // as calling unpackTime().
5685
5686 // Optional fast-path check:
5687 // Return immediately if a permit is available.
5688 // We depend on Atomic::xchg() having full barrier semantics
5689 // since we are doing a lock-free update to _counter.
5690 if (Atomic::xchg(0, &_counter) > 0) return;
5691
5692 Thread* thread = Thread::current();
5693 assert(thread->is_Java_thread(), "Must be JavaThread");
5694 JavaThread *jt = (JavaThread *)thread;
5695
5696 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5697 // Check interrupt before trying to wait
5698 if (Thread::is_interrupted(thread, false)) {
5699 return;
5700 }
5701
5702 // Next, demultiplex/decode time arguments
5703 timespec absTime;
5704 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5705 return;
5706 }
5707 if (time > 0) {
5708 unpackTime(&absTime, isAbsolute, time);
5709 }
5710
5711
5712 // Enter safepoint region
5713 // Beware of deadlocks such as 6317397.
5714 // The per-thread Parker:: mutex is a classic leaf-lock.
5715 // In particular a thread must never block on the Threads_lock while
5716 // holding the Parker:: mutex. If safepoints are pending both the
5717 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5718 ThreadBlockInVM tbivm(jt);
5719
5720 // Don't wait if cannot get lock since interference arises from
5721 // unblocking. Also. check interrupt before trying wait
5722 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5723 return;
5724 }
5725
5726 int status;
5727 if (_counter > 0) { // no wait needed
5728 _counter = 0;
5729 status = pthread_mutex_unlock(_mutex);
5730 assert(status == 0, "invariant");
5731 // Paranoia to ensure our locked and lock-free paths interact
5732 // correctly with each other and Java-level accesses.
5733 OrderAccess::fence();
5734 return;
5735 }
5736
5737 #ifdef ASSERT
5738 // Don't catch signals while blocked; let the running threads have the signals.
5739 // (This allows a debugger to break into the running thread.)
5740 sigset_t oldsigs;
5741 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5742 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5743 #endif
5744
5745 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5746 jt->set_suspend_equivalent();
5747 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5748
5749 assert(_cur_index == -1, "invariant");
5750 if (time == 0) {
5751 _cur_index = REL_INDEX; // arbitrary choice when not timed
5752 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5753 } else {
5754 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5755 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5756 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5757 pthread_cond_destroy(&_cond[_cur_index]);
5758 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5759 }
5760 }
5761 _cur_index = -1;
5762 assert_status(status == 0 || status == EINTR ||
5763 status == ETIME || status == ETIMEDOUT,
5764 status, "cond_timedwait");
5765
5766 #ifdef ASSERT
5767 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5768 #endif
5769
5770 _counter = 0;
5771 status = pthread_mutex_unlock(_mutex);
5772 assert_status(status == 0, status, "invariant");
5773 // Paranoia to ensure our locked and lock-free paths interact
5774 // correctly with each other and Java-level accesses.
5775 OrderAccess::fence();
5776
5777 // If externally suspended while waiting, re-suspend
5778 if (jt->handle_special_suspend_equivalent_condition()) {
5779 jt->java_suspend_self();
5780 }
5781 }
5782
5783 void Parker::unpark() {
5784 int status = pthread_mutex_lock(_mutex);
5785 assert(status == 0, "invariant");
5786 const int s = _counter;
5787 _counter = 1;
5788 if (s < 1) {
5789 // thread might be parked
5790 if (_cur_index != -1) {
5791 // thread is definitely parked
5792 if (WorkAroundNPTLTimedWaitHang) {
5793 status = pthread_cond_signal(&_cond[_cur_index]);
5794 assert(status == 0, "invariant");
5795 status = pthread_mutex_unlock(_mutex);
5796 assert(status == 0, "invariant");
5797 } else {
5798 // must capture correct index before unlocking
5799 int index = _cur_index;
5800 status = pthread_mutex_unlock(_mutex);
5801 assert(status == 0, "invariant");
5802 status = pthread_cond_signal(&_cond[index]);
5803 assert(status == 0, "invariant");
5804 }
5805 } else {
5806 pthread_mutex_unlock(_mutex);
5807 assert(status == 0, "invariant");
5808 }
5809 } else {
5810 pthread_mutex_unlock(_mutex);
5811 assert(status == 0, "invariant");
5812 }
5813 }
5814
5815
5816 extern char** environ;
5817
5818 // Run the specified command in a separate process. Return its exit value,
5819 // or -1 on failure (e.g. can't fork a new process).
5820 // Unlike system(), this function can be called from signal handler. It
5821 // doesn't block SIGINT et al.
5822 int os::fork_and_exec(char* cmd) {
5823 const char * argv[4] = {"sh", "-c", cmd, NULL};
5824
5825 pid_t pid = fork();
5826
5827 if (pid < 0) {
5828 // fork failed
5829 return -1;
5830
5831 } else if (pid == 0) {
5832 // child process
5833
5834 execve("/bin/sh", (char* const*)argv, environ);
5835
5836 // execve failed
5837 _exit(-1);
5838
5839 } else {
5840 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5841 // care about the actual exit code, for now.
5842
5843 int status;
5844
5845 // Wait for the child process to exit. This returns immediately if
5846 // the child has already exited. */
5847 while (waitpid(pid, &status, 0) < 0) {
5848 switch (errno) {
5849 case ECHILD: return 0;
5850 case EINTR: break;
5851 default: return -1;
5852 }
5853 }
5854
5855 if (WIFEXITED(status)) {
5856 // The child exited normally; get its exit code.
5857 return WEXITSTATUS(status);
5858 } else if (WIFSIGNALED(status)) {
5859 // The child exited because of a signal
5860 // The best value to return is 0x80 + signal number,
5861 // because that is what all Unix shells do, and because
5862 // it allows callers to distinguish between process exit and
5863 // process death by signal.
5864 return 0x80 + WTERMSIG(status);
5865 } else {
5866 // Unknown exit code; pass it through
5867 return status;
5868 }
5869 }
5870 }
5871
5872 // is_headless_jre()
5873 //
5874 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5875 // in order to report if we are running in a headless jre
5876 //
5877 // Since JDK8 xawt/libmawt.so was moved into the same directory
5878 // as libawt.so, and renamed libawt_xawt.so
5879 //
5880 bool os::is_headless_jre() {
5881 struct stat statbuf;
5882 char buf[MAXPATHLEN];
5883 char libmawtpath[MAXPATHLEN];
5884 const char *xawtstr = "/xawt/libmawt.so";
5885 const char *new_xawtstr = "/libawt_xawt.so";
5886 char *p;
5887
5888 // Get path to libjvm.so
5889 os::jvm_path(buf, sizeof(buf));
5890
5891 // Get rid of libjvm.so
5892 p = strrchr(buf, '/');
5893 if (p == NULL) {
5894 return false;
5895 } else {
5896 *p = '\0';
5897 }
5898
5899 // Get rid of client or server
5900 p = strrchr(buf, '/');
5901 if (p == NULL) {
5902 return false;
5903 } else {
5904 *p = '\0';
5905 }
5906
5907 // check xawt/libmawt.so
5908 strcpy(libmawtpath, buf);
5909 strcat(libmawtpath, xawtstr);
5910 if (::stat(libmawtpath, &statbuf) == 0) return false;
5911
5912 // check libawt_xawt.so
5913 strcpy(libmawtpath, buf);
5914 strcat(libmawtpath, new_xawtstr);
5915 if (::stat(libmawtpath, &statbuf) == 0) return false;
5916
5917 return true;
5918 }
5919
5920 // Get the default path to the core file
5921 // Returns the length of the string
5922 int os::get_core_path(char* buffer, size_t bufferSize) {
5923 /*
5924 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5925 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5926 */
5927 const int core_pattern_len = 129;
5928 char core_pattern[core_pattern_len] = {0};
5929
5930 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5931 if (core_pattern_file != -1) {
5932 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5933 ::close(core_pattern_file);
5934
5935 if (ret > 0) {
5936 char *last_char = core_pattern + strlen(core_pattern) - 1;
5937
5938 if (*last_char == '\n') {
5939 *last_char = '\0';
5940 }
5941 }
5942 }
5943
5944 if (strlen(core_pattern) == 0) {
5945 return -1;
5946 }
5947
5948 char *pid_pos = strstr(core_pattern, "%p");
5949 int written;
5950
5951 if (core_pattern[0] == '/') {
5952 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5953 } else {
5954 char cwd[PATH_MAX];
5955
5956 const char* p = get_current_directory(cwd, PATH_MAX);
5957 if (p == NULL) {
5958 return -1;
5959 }
5960
5961 if (core_pattern[0] == '|') {
5962 written = jio_snprintf(buffer, bufferSize,
5963 "\"%s\" (or dumping to %s/core.%d)",
5964 &core_pattern[1], p, current_process_id());
5965 } else {
5966 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5967 }
5968 }
5969
5970 if (written < 0) {
5971 return -1;
5972 }
5973
5974 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5975 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5976
5977 if (core_uses_pid_file != -1) {
5978 char core_uses_pid = 0;
5979 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5980 ::close(core_uses_pid_file);
5981
5982 if (core_uses_pid == '1') {
5983 jio_snprintf(buffer + written, bufferSize - written,
5984 ".%d", current_process_id());
5985 }
5986 }
5987 }
5988
5989 return strlen(buffer);
5990 }
5991
5992 /////////////// Unit tests ///////////////
5993
5994 #ifndef PRODUCT
5995
5996 #define test_log(...) \
5997 do { \
5998 if (VerboseInternalVMTests) { \
5999 tty->print_cr(__VA_ARGS__); \
6000 tty->flush(); \
6001 } \
6002 } while (false)
6003
6004 class TestReserveMemorySpecial : AllStatic {
6005 public:
6006 static void small_page_write(void* addr, size_t size) {
6007 size_t page_size = os::vm_page_size();
6008
6009 char* end = (char*)addr + size;
6010 for (char* p = (char*)addr; p < end; p += page_size) {
6011 *p = 1;
6012 }
6013 }
6014
6015 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6016 if (!UseHugeTLBFS) {
6017 return;
6018 }
6019
6020 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6021
6022 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6023
6024 if (addr != NULL) {
6025 small_page_write(addr, size);
6026
6027 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6028 }
6029 }
6030
6031 static void test_reserve_memory_special_huge_tlbfs_only() {
6032 if (!UseHugeTLBFS) {
6033 return;
6034 }
6035
6036 size_t lp = os::large_page_size();
6037
6038 for (size_t size = lp; size <= lp * 10; size += lp) {
6039 test_reserve_memory_special_huge_tlbfs_only(size);
6040 }
6041 }
6042
6043 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6044 size_t lp = os::large_page_size();
6045 size_t ag = os::vm_allocation_granularity();
6046
6047 // sizes to test
6048 const size_t sizes[] = {
6049 lp, lp + ag, lp + lp / 2, lp * 2,
6050 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6051 lp * 10, lp * 10 + lp / 2
6052 };
6053 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6054
6055 // For each size/alignment combination, we test three scenarios:
6056 // 1) with req_addr == NULL
6057 // 2) with a non-null req_addr at which we expect to successfully allocate
6058 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6059 // expect the allocation to either fail or to ignore req_addr
6060
6061 // Pre-allocate two areas; they shall be as large as the largest allocation
6062 // and aligned to the largest alignment we will be testing.
6063 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6064 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6065 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6066 -1, 0);
6067 assert(mapping1 != MAP_FAILED, "should work");
6068
6069 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6070 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6071 -1, 0);
6072 assert(mapping2 != MAP_FAILED, "should work");
6073
6074 // Unmap the first mapping, but leave the second mapping intact: the first
6075 // mapping will serve as a value for a "good" req_addr (case 2). The second
6076 // mapping, still intact, as "bad" req_addr (case 3).
6077 ::munmap(mapping1, mapping_size);
6078
6079 // Case 1
6080 test_log("%s, req_addr NULL:", __FUNCTION__);
6081 test_log("size align result");
6082
6083 for (int i = 0; i < num_sizes; i++) {
6084 const size_t size = sizes[i];
6085 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6086 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6087 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6088 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6089 if (p != NULL) {
6090 assert(is_ptr_aligned(p, alignment), "must be");
6091 small_page_write(p, size);
6092 os::Linux::release_memory_special_huge_tlbfs(p, size);
6093 }
6094 }
6095 }
6096
6097 // Case 2
6098 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6099 test_log("size align req_addr result");
6100
6101 for (int i = 0; i < num_sizes; i++) {
6102 const size_t size = sizes[i];
6103 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6104 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6105 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6106 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6107 size, alignment, p2i(req_addr), p2i(p),
6108 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6109 if (p != NULL) {
6110 assert(p == req_addr, "must be");
6111 small_page_write(p, size);
6112 os::Linux::release_memory_special_huge_tlbfs(p, size);
6113 }
6114 }
6115 }
6116
6117 // Case 3
6118 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6119 test_log("size align req_addr result");
6120
6121 for (int i = 0; i < num_sizes; i++) {
6122 const size_t size = sizes[i];
6123 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6124 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6125 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6126 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6127 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6128 // as the area around req_addr contains already existing mappings, the API should always
6129 // return NULL (as per contract, it cannot return another address)
6130 assert(p == NULL, "must be");
6131 }
6132 }
6133
6134 ::munmap(mapping2, mapping_size);
6135
6136 }
6137
6138 static void test_reserve_memory_special_huge_tlbfs() {
6139 if (!UseHugeTLBFS) {
6140 return;
6141 }
6142
6143 test_reserve_memory_special_huge_tlbfs_only();
6144 test_reserve_memory_special_huge_tlbfs_mixed();
6145 }
6146
6147 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6148 if (!UseSHM) {
6149 return;
6150 }
6151
6152 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6153
6154 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6155
6156 if (addr != NULL) {
6157 assert(is_ptr_aligned(addr, alignment), "Check");
6158 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6159
6160 small_page_write(addr, size);
6161
6162 os::Linux::release_memory_special_shm(addr, size);
6163 }
6164 }
6165
6166 static void test_reserve_memory_special_shm() {
6167 size_t lp = os::large_page_size();
6168 size_t ag = os::vm_allocation_granularity();
6169
6170 for (size_t size = ag; size < lp * 3; size += ag) {
6171 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6172 test_reserve_memory_special_shm(size, alignment);
6173 }
6174 }
6175 }
6176
6177 static void test() {
6178 test_reserve_memory_special_huge_tlbfs();
6179 test_reserve_memory_special_shm();
6180 }
6181 };
6182
6183 void TestReserveMemorySpecial_test() {
6184 TestReserveMemorySpecial::test();
6185 }
6186
6187 #endif