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 thread->initialize_thread_current();
650
651 OSThread* osthread = thread->osthread();
652 Monitor* sync = osthread->startThread_lock();
653
654 osthread->set_thread_id(os::current_thread_id());
655
656 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 // initial thread
878
879 // Check if current thread is the initial thread, similar to Solaris thr_main.
880 bool os::Linux::is_initial_thread(void) {
881 char dummy;
882 // If called before init complete, thread stack bottom will be null.
883 // Can be called if fatal error occurs before initialization.
884 if (initial_thread_stack_bottom() == NULL) return false;
885 assert(initial_thread_stack_bottom() != NULL &&
886 initial_thread_stack_size() != 0,
887 "os::init did not locate initial thread's stack region");
888 if ((address)&dummy >= initial_thread_stack_bottom() &&
889 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
890 return true;
891 } else {
892 return false;
893 }
894 }
895
896 // Find the virtual memory area that contains addr
897 static bool find_vma(address addr, address* vma_low, address* vma_high) {
898 FILE *fp = fopen("/proc/self/maps", "r");
899 if (fp) {
900 address low, high;
901 while (!feof(fp)) {
902 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
903 if (low <= addr && addr < high) {
904 if (vma_low) *vma_low = low;
905 if (vma_high) *vma_high = high;
906 fclose(fp);
907 return true;
908 }
909 }
910 for (;;) {
911 int ch = fgetc(fp);
912 if (ch == EOF || ch == (int)'\n') break;
913 }
914 }
915 fclose(fp);
916 }
917 return false;
918 }
919
920 // Locate initial thread stack. This special handling of initial thread stack
921 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
922 // bogus value for initial thread.
923 void os::Linux::capture_initial_stack(size_t max_size) {
924 // stack size is the easy part, get it from RLIMIT_STACK
925 size_t stack_size;
926 struct rlimit rlim;
927 getrlimit(RLIMIT_STACK, &rlim);
928 stack_size = rlim.rlim_cur;
929
930 // 6308388: a bug in ld.so will relocate its own .data section to the
931 // lower end of primordial stack; reduce ulimit -s value a little bit
932 // so we won't install guard page on ld.so's data section.
933 stack_size -= 2 * page_size();
934
935 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
936 // 7.1, in both cases we will get 2G in return value.
937 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
938 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
939 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
940 // in case other parts in glibc still assumes 2M max stack size.
941 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
942 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
943 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
944 stack_size = 2 * K * K IA64_ONLY(*2);
945 }
946 // Try to figure out where the stack base (top) is. This is harder.
947 //
948 // When an application is started, glibc saves the initial stack pointer in
949 // a global variable "__libc_stack_end", which is then used by system
950 // libraries. __libc_stack_end should be pretty close to stack top. The
951 // variable is available since the very early days. However, because it is
952 // a private interface, it could disappear in the future.
953 //
954 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
955 // to __libc_stack_end, it is very close to stack top, but isn't the real
956 // stack top. Note that /proc may not exist if VM is running as a chroot
957 // program, so reading /proc/<pid>/stat could fail. Also the contents of
958 // /proc/<pid>/stat could change in the future (though unlikely).
959 //
960 // We try __libc_stack_end first. If that doesn't work, look for
961 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
962 // as a hint, which should work well in most cases.
963
964 uintptr_t stack_start;
965
966 // try __libc_stack_end first
967 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
968 if (p && *p) {
969 stack_start = *p;
970 } else {
971 // see if we can get the start_stack field from /proc/self/stat
972 FILE *fp;
973 int pid;
974 char state;
975 int ppid;
976 int pgrp;
977 int session;
978 int nr;
979 int tpgrp;
980 unsigned long flags;
981 unsigned long minflt;
982 unsigned long cminflt;
983 unsigned long majflt;
984 unsigned long cmajflt;
985 unsigned long utime;
986 unsigned long stime;
987 long cutime;
988 long cstime;
989 long prio;
990 long nice;
991 long junk;
992 long it_real;
993 uintptr_t start;
994 uintptr_t vsize;
995 intptr_t rss;
996 uintptr_t rsslim;
997 uintptr_t scodes;
998 uintptr_t ecode;
999 int i;
1000
1001 // Figure what the primordial thread stack base is. Code is inspired
1002 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1003 // followed by command name surrounded by parentheses, state, etc.
1004 char stat[2048];
1005 int statlen;
1006
1007 fp = fopen("/proc/self/stat", "r");
1008 if (fp) {
1009 statlen = fread(stat, 1, 2047, fp);
1010 stat[statlen] = '\0';
1011 fclose(fp);
1012
1013 // Skip pid and the command string. Note that we could be dealing with
1014 // weird command names, e.g. user could decide to rename java launcher
1015 // to "java 1.4.2 :)", then the stat file would look like
1016 // 1234 (java 1.4.2 :)) R ... ...
1017 // We don't really need to know the command string, just find the last
1018 // occurrence of ")" and then start parsing from there. See bug 4726580.
1019 char * s = strrchr(stat, ')');
1020
1021 i = 0;
1022 if (s) {
1023 // Skip blank chars
1024 do { s++; } while (s && isspace(*s));
1025
1026 #define _UFM UINTX_FORMAT
1027 #define _DFM INTX_FORMAT
1028
1029 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1030 // 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
1031 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,
1032 &state, // 3 %c
1033 &ppid, // 4 %d
1034 &pgrp, // 5 %d
1035 &session, // 6 %d
1036 &nr, // 7 %d
1037 &tpgrp, // 8 %d
1038 &flags, // 9 %lu
1039 &minflt, // 10 %lu
1040 &cminflt, // 11 %lu
1041 &majflt, // 12 %lu
1042 &cmajflt, // 13 %lu
1043 &utime, // 14 %lu
1044 &stime, // 15 %lu
1045 &cutime, // 16 %ld
1046 &cstime, // 17 %ld
1047 &prio, // 18 %ld
1048 &nice, // 19 %ld
1049 &junk, // 20 %ld
1050 &it_real, // 21 %ld
1051 &start, // 22 UINTX_FORMAT
1052 &vsize, // 23 UINTX_FORMAT
1053 &rss, // 24 INTX_FORMAT
1054 &rsslim, // 25 UINTX_FORMAT
1055 &scodes, // 26 UINTX_FORMAT
1056 &ecode, // 27 UINTX_FORMAT
1057 &stack_start); // 28 UINTX_FORMAT
1058 }
1059
1060 #undef _UFM
1061 #undef _DFM
1062
1063 if (i != 28 - 2) {
1064 assert(false, "Bad conversion from /proc/self/stat");
1065 // product mode - assume we are the initial thread, good luck in the
1066 // embedded case.
1067 warning("Can't detect initial thread stack location - bad conversion");
1068 stack_start = (uintptr_t) &rlim;
1069 }
1070 } else {
1071 // For some reason we can't open /proc/self/stat (for example, running on
1072 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1073 // most cases, so don't abort:
1074 warning("Can't detect initial thread stack location - no /proc/self/stat");
1075 stack_start = (uintptr_t) &rlim;
1076 }
1077 }
1078
1079 // Now we have a pointer (stack_start) very close to the stack top, the
1080 // next thing to do is to figure out the exact location of stack top. We
1081 // can find out the virtual memory area that contains stack_start by
1082 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1083 // and its upper limit is the real stack top. (again, this would fail if
1084 // running inside chroot, because /proc may not exist.)
1085
1086 uintptr_t stack_top;
1087 address low, high;
1088 if (find_vma((address)stack_start, &low, &high)) {
1089 // success, "high" is the true stack top. (ignore "low", because initial
1090 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1091 stack_top = (uintptr_t)high;
1092 } else {
1093 // failed, likely because /proc/self/maps does not exist
1094 warning("Can't detect initial thread stack location - find_vma failed");
1095 // best effort: stack_start is normally within a few pages below the real
1096 // stack top, use it as stack top, and reduce stack size so we won't put
1097 // guard page outside stack.
1098 stack_top = stack_start;
1099 stack_size -= 16 * page_size();
1100 }
1101
1102 // stack_top could be partially down the page so align it
1103 stack_top = align_size_up(stack_top, page_size());
1104
1105 if (max_size && stack_size > max_size) {
1106 _initial_thread_stack_size = max_size;
1107 } else {
1108 _initial_thread_stack_size = stack_size;
1109 }
1110
1111 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1112 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1113 }
1114
1115 ////////////////////////////////////////////////////////////////////////////////
1116 // time support
1117
1118 // Time since start-up in seconds to a fine granularity.
1119 // Used by VMSelfDestructTimer and the MemProfiler.
1120 double os::elapsedTime() {
1121
1122 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1123 }
1124
1125 jlong os::elapsed_counter() {
1126 return javaTimeNanos() - initial_time_count;
1127 }
1128
1129 jlong os::elapsed_frequency() {
1130 return NANOSECS_PER_SEC; // nanosecond resolution
1131 }
1132
1133 bool os::supports_vtime() { return true; }
1134 bool os::enable_vtime() { return false; }
1135 bool os::vtime_enabled() { return false; }
1136
1137 double os::elapsedVTime() {
1138 struct rusage usage;
1139 int retval = getrusage(RUSAGE_THREAD, &usage);
1140 if (retval == 0) {
1141 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);
1142 } else {
1143 // better than nothing, but not much
1144 return elapsedTime();
1145 }
1146 }
1147
1148 jlong os::javaTimeMillis() {
1149 timeval time;
1150 int status = gettimeofday(&time, NULL);
1151 assert(status != -1, "linux error");
1152 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1153 }
1154
1155 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1156 timeval time;
1157 int status = gettimeofday(&time, NULL);
1158 assert(status != -1, "linux error");
1159 seconds = jlong(time.tv_sec);
1160 nanos = jlong(time.tv_usec) * 1000;
1161 }
1162
1163
1164 #ifndef CLOCK_MONOTONIC
1165 #define CLOCK_MONOTONIC (1)
1166 #endif
1167
1168 void os::Linux::clock_init() {
1169 // we do dlopen's in this particular order due to bug in linux
1170 // dynamical loader (see 6348968) leading to crash on exit
1171 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1172 if (handle == NULL) {
1173 handle = dlopen("librt.so", RTLD_LAZY);
1174 }
1175
1176 if (handle) {
1177 int (*clock_getres_func)(clockid_t, struct timespec*) =
1178 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1179 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1180 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1181 if (clock_getres_func && clock_gettime_func) {
1182 // See if monotonic clock is supported by the kernel. Note that some
1183 // early implementations simply return kernel jiffies (updated every
1184 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1185 // for nano time (though the monotonic property is still nice to have).
1186 // It's fixed in newer kernels, however clock_getres() still returns
1187 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1188 // resolution for now. Hopefully as people move to new kernels, this
1189 // won't be a problem.
1190 struct timespec res;
1191 struct timespec tp;
1192 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1193 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1194 // yes, monotonic clock is supported
1195 _clock_gettime = clock_gettime_func;
1196 return;
1197 } else {
1198 // close librt if there is no monotonic clock
1199 dlclose(handle);
1200 }
1201 }
1202 }
1203 warning("No monotonic clock was available - timed services may " \
1204 "be adversely affected if the time-of-day clock changes");
1205 }
1206
1207 #ifndef SYS_clock_getres
1208 #if defined(IA32) || defined(AMD64)
1209 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1210 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1211 #else
1212 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1213 #define sys_clock_getres(x,y) -1
1214 #endif
1215 #else
1216 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1217 #endif
1218
1219 void os::Linux::fast_thread_clock_init() {
1220 if (!UseLinuxPosixThreadCPUClocks) {
1221 return;
1222 }
1223 clockid_t clockid;
1224 struct timespec tp;
1225 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1226 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1227
1228 // Switch to using fast clocks for thread cpu time if
1229 // the sys_clock_getres() returns 0 error code.
1230 // Note, that some kernels may support the current thread
1231 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1232 // returned by the pthread_getcpuclockid().
1233 // If the fast Posix clocks are supported then the sys_clock_getres()
1234 // must return at least tp.tv_sec == 0 which means a resolution
1235 // better than 1 sec. This is extra check for reliability.
1236
1237 if (pthread_getcpuclockid_func &&
1238 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1239 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1240 _supports_fast_thread_cpu_time = true;
1241 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1242 }
1243 }
1244
1245 jlong os::javaTimeNanos() {
1246 if (os::supports_monotonic_clock()) {
1247 struct timespec tp;
1248 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1249 assert(status == 0, "gettime error");
1250 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1251 return result;
1252 } else {
1253 timeval time;
1254 int status = gettimeofday(&time, NULL);
1255 assert(status != -1, "linux error");
1256 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1257 return 1000 * usecs;
1258 }
1259 }
1260
1261 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1262 if (os::supports_monotonic_clock()) {
1263 info_ptr->max_value = ALL_64_BITS;
1264
1265 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1266 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1267 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1268 } else {
1269 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1270 info_ptr->max_value = ALL_64_BITS;
1271
1272 // gettimeofday is a real time clock so it skips
1273 info_ptr->may_skip_backward = true;
1274 info_ptr->may_skip_forward = true;
1275 }
1276
1277 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1278 }
1279
1280 // Return the real, user, and system times in seconds from an
1281 // arbitrary fixed point in the past.
1282 bool os::getTimesSecs(double* process_real_time,
1283 double* process_user_time,
1284 double* process_system_time) {
1285 struct tms ticks;
1286 clock_t real_ticks = times(&ticks);
1287
1288 if (real_ticks == (clock_t) (-1)) {
1289 return false;
1290 } else {
1291 double ticks_per_second = (double) clock_tics_per_sec;
1292 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1293 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1294 *process_real_time = ((double) real_ticks) / ticks_per_second;
1295
1296 return true;
1297 }
1298 }
1299
1300
1301 char * os::local_time_string(char *buf, size_t buflen) {
1302 struct tm t;
1303 time_t long_time;
1304 time(&long_time);
1305 localtime_r(&long_time, &t);
1306 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1307 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1308 t.tm_hour, t.tm_min, t.tm_sec);
1309 return buf;
1310 }
1311
1312 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1313 return localtime_r(clock, res);
1314 }
1315
1316 ////////////////////////////////////////////////////////////////////////////////
1317 // runtime exit support
1318
1319 // Note: os::shutdown() might be called very early during initialization, or
1320 // called from signal handler. Before adding something to os::shutdown(), make
1321 // sure it is async-safe and can handle partially initialized VM.
1322 void os::shutdown() {
1323
1324 // allow PerfMemory to attempt cleanup of any persistent resources
1325 perfMemory_exit();
1326
1327 // needs to remove object in file system
1328 AttachListener::abort();
1329
1330 // flush buffered output, finish log files
1331 ostream_abort();
1332
1333 // Check for abort hook
1334 abort_hook_t abort_hook = Arguments::abort_hook();
1335 if (abort_hook != NULL) {
1336 abort_hook();
1337 }
1338
1339 }
1340
1341 // Note: os::abort() might be called very early during initialization, or
1342 // called from signal handler. Before adding something to os::abort(), make
1343 // sure it is async-safe and can handle partially initialized VM.
1344 void os::abort(bool dump_core, void* siginfo, void* context) {
1345 os::shutdown();
1346 if (dump_core) {
1347 #ifndef PRODUCT
1348 fdStream out(defaultStream::output_fd());
1349 out.print_raw("Current thread is ");
1350 char buf[16];
1351 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1352 out.print_raw_cr(buf);
1353 out.print_raw_cr("Dumping core ...");
1354 #endif
1355 ::abort(); // dump core
1356 }
1357
1358 ::exit(1);
1359 }
1360
1361 // Die immediately, no exit hook, no abort hook, no cleanup.
1362 void os::die() {
1363 ::abort();
1364 }
1365
1366
1367 // This method is a copy of JDK's sysGetLastErrorString
1368 // from src/solaris/hpi/src/system_md.c
1369
1370 size_t os::lasterror(char *buf, size_t len) {
1371 if (errno == 0) return 0;
1372
1373 const char *s = ::strerror(errno);
1374 size_t n = ::strlen(s);
1375 if (n >= len) {
1376 n = len - 1;
1377 }
1378 ::strncpy(buf, s, n);
1379 buf[n] = '\0';
1380 return n;
1381 }
1382
1383 // thread_id is kernel thread id (similar to Solaris LWP id)
1384 intx os::current_thread_id() { return os::Linux::gettid(); }
1385 int os::current_process_id() {
1386 return ::getpid();
1387 }
1388
1389 // DLL functions
1390
1391 const char* os::dll_file_extension() { return ".so"; }
1392
1393 // This must be hard coded because it's the system's temporary
1394 // directory not the java application's temp directory, ala java.io.tmpdir.
1395 const char* os::get_temp_directory() { return "/tmp"; }
1396
1397 static bool file_exists(const char* filename) {
1398 struct stat statbuf;
1399 if (filename == NULL || strlen(filename) == 0) {
1400 return false;
1401 }
1402 return os::stat(filename, &statbuf) == 0;
1403 }
1404
1405 bool os::dll_build_name(char* buffer, size_t buflen,
1406 const char* pname, const char* fname) {
1407 bool retval = false;
1408 // Copied from libhpi
1409 const size_t pnamelen = pname ? strlen(pname) : 0;
1410
1411 // Return error on buffer overflow.
1412 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1413 return retval;
1414 }
1415
1416 if (pnamelen == 0) {
1417 snprintf(buffer, buflen, "lib%s.so", fname);
1418 retval = true;
1419 } else if (strchr(pname, *os::path_separator()) != NULL) {
1420 int n;
1421 char** pelements = split_path(pname, &n);
1422 if (pelements == NULL) {
1423 return false;
1424 }
1425 for (int i = 0; i < n; i++) {
1426 // Really shouldn't be NULL, but check can't hurt
1427 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1428 continue; // skip the empty path values
1429 }
1430 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1431 if (file_exists(buffer)) {
1432 retval = true;
1433 break;
1434 }
1435 }
1436 // release the storage
1437 for (int i = 0; i < n; i++) {
1438 if (pelements[i] != NULL) {
1439 FREE_C_HEAP_ARRAY(char, pelements[i]);
1440 }
1441 }
1442 if (pelements != NULL) {
1443 FREE_C_HEAP_ARRAY(char*, pelements);
1444 }
1445 } else {
1446 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1447 retval = true;
1448 }
1449 return retval;
1450 }
1451
1452 // check if addr is inside libjvm.so
1453 bool os::address_is_in_vm(address addr) {
1454 static address libjvm_base_addr;
1455 Dl_info dlinfo;
1456
1457 if (libjvm_base_addr == NULL) {
1458 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1459 libjvm_base_addr = (address)dlinfo.dli_fbase;
1460 }
1461 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1462 }
1463
1464 if (dladdr((void *)addr, &dlinfo) != 0) {
1465 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1466 }
1467
1468 return false;
1469 }
1470
1471 bool os::dll_address_to_function_name(address addr, char *buf,
1472 int buflen, int *offset,
1473 bool demangle) {
1474 // buf is not optional, but offset is optional
1475 assert(buf != NULL, "sanity check");
1476
1477 Dl_info dlinfo;
1478
1479 if (dladdr((void*)addr, &dlinfo) != 0) {
1480 // see if we have a matching symbol
1481 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1482 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1483 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1484 }
1485 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1486 return true;
1487 }
1488 // no matching symbol so try for just file info
1489 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1490 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1491 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1492 return true;
1493 }
1494 }
1495 }
1496
1497 buf[0] = '\0';
1498 if (offset != NULL) *offset = -1;
1499 return false;
1500 }
1501
1502 struct _address_to_library_name {
1503 address addr; // input : memory address
1504 size_t buflen; // size of fname
1505 char* fname; // output: library name
1506 address base; // library base addr
1507 };
1508
1509 static int address_to_library_name_callback(struct dl_phdr_info *info,
1510 size_t size, void *data) {
1511 int i;
1512 bool found = false;
1513 address libbase = NULL;
1514 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1515
1516 // iterate through all loadable segments
1517 for (i = 0; i < info->dlpi_phnum; i++) {
1518 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1519 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1520 // base address of a library is the lowest address of its loaded
1521 // segments.
1522 if (libbase == NULL || libbase > segbase) {
1523 libbase = segbase;
1524 }
1525 // see if 'addr' is within current segment
1526 if (segbase <= d->addr &&
1527 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1528 found = true;
1529 }
1530 }
1531 }
1532
1533 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1534 // so dll_address_to_library_name() can fall through to use dladdr() which
1535 // can figure out executable name from argv[0].
1536 if (found && info->dlpi_name && info->dlpi_name[0]) {
1537 d->base = libbase;
1538 if (d->fname) {
1539 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1540 }
1541 return 1;
1542 }
1543 return 0;
1544 }
1545
1546 bool os::dll_address_to_library_name(address addr, char* buf,
1547 int buflen, int* offset) {
1548 // buf is not optional, but offset is optional
1549 assert(buf != NULL, "sanity check");
1550
1551 Dl_info dlinfo;
1552 struct _address_to_library_name data;
1553
1554 // There is a bug in old glibc dladdr() implementation that it could resolve
1555 // to wrong library name if the .so file has a base address != NULL. Here
1556 // we iterate through the program headers of all loaded libraries to find
1557 // out which library 'addr' really belongs to. This workaround can be
1558 // removed once the minimum requirement for glibc is moved to 2.3.x.
1559 data.addr = addr;
1560 data.fname = buf;
1561 data.buflen = buflen;
1562 data.base = NULL;
1563 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1564
1565 if (rslt) {
1566 // buf already contains library name
1567 if (offset) *offset = addr - data.base;
1568 return true;
1569 }
1570 if (dladdr((void*)addr, &dlinfo) != 0) {
1571 if (dlinfo.dli_fname != NULL) {
1572 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1573 }
1574 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1575 *offset = addr - (address)dlinfo.dli_fbase;
1576 }
1577 return true;
1578 }
1579
1580 buf[0] = '\0';
1581 if (offset) *offset = -1;
1582 return false;
1583 }
1584
1585 // Loads .dll/.so and
1586 // in case of error it checks if .dll/.so was built for the
1587 // same architecture as Hotspot is running on
1588
1589
1590 // Remember the stack's state. The Linux dynamic linker will change
1591 // the stack to 'executable' at most once, so we must safepoint only once.
1592 bool os::Linux::_stack_is_executable = false;
1593
1594 // VM operation that loads a library. This is necessary if stack protection
1595 // of the Java stacks can be lost during loading the library. If we
1596 // do not stop the Java threads, they can stack overflow before the stacks
1597 // are protected again.
1598 class VM_LinuxDllLoad: public VM_Operation {
1599 private:
1600 const char *_filename;
1601 char *_ebuf;
1602 int _ebuflen;
1603 void *_lib;
1604 public:
1605 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1606 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1607 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1608 void doit() {
1609 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1610 os::Linux::_stack_is_executable = true;
1611 }
1612 void* loaded_library() { return _lib; }
1613 };
1614
1615 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1616 void * result = NULL;
1617 bool load_attempted = false;
1618
1619 // Check whether the library to load might change execution rights
1620 // of the stack. If they are changed, the protection of the stack
1621 // guard pages will be lost. We need a safepoint to fix this.
1622 //
1623 // See Linux man page execstack(8) for more info.
1624 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1625 ElfFile ef(filename);
1626 if (!ef.specifies_noexecstack()) {
1627 if (!is_init_completed()) {
1628 os::Linux::_stack_is_executable = true;
1629 // This is OK - No Java threads have been created yet, and hence no
1630 // stack guard pages to fix.
1631 //
1632 // This should happen only when you are building JDK7 using a very
1633 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1634 //
1635 // Dynamic loader will make all stacks executable after
1636 // this function returns, and will not do that again.
1637 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1638 } else {
1639 warning("You have loaded library %s which might have disabled stack guard. "
1640 "The VM will try to fix the stack guard now.\n"
1641 "It's highly recommended that you fix the library with "
1642 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1643 filename);
1644
1645 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1646 JavaThread *jt = JavaThread::current();
1647 if (jt->thread_state() != _thread_in_native) {
1648 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1649 // that requires ExecStack. Cannot enter safe point. Let's give up.
1650 warning("Unable to fix stack guard. Giving up.");
1651 } else {
1652 if (!LoadExecStackDllInVMThread) {
1653 // This is for the case where the DLL has an static
1654 // constructor function that executes JNI code. We cannot
1655 // load such DLLs in the VMThread.
1656 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1657 }
1658
1659 ThreadInVMfromNative tiv(jt);
1660 debug_only(VMNativeEntryWrapper vew;)
1661
1662 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1663 VMThread::execute(&op);
1664 if (LoadExecStackDllInVMThread) {
1665 result = op.loaded_library();
1666 }
1667 load_attempted = true;
1668 }
1669 }
1670 }
1671 }
1672
1673 if (!load_attempted) {
1674 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1675 }
1676
1677 if (result != NULL) {
1678 // Successful loading
1679 return result;
1680 }
1681
1682 Elf32_Ehdr elf_head;
1683 int diag_msg_max_length=ebuflen-strlen(ebuf);
1684 char* diag_msg_buf=ebuf+strlen(ebuf);
1685
1686 if (diag_msg_max_length==0) {
1687 // No more space in ebuf for additional diagnostics message
1688 return NULL;
1689 }
1690
1691
1692 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1693
1694 if (file_descriptor < 0) {
1695 // Can't open library, report dlerror() message
1696 return NULL;
1697 }
1698
1699 bool failed_to_read_elf_head=
1700 (sizeof(elf_head)!=
1701 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1702
1703 ::close(file_descriptor);
1704 if (failed_to_read_elf_head) {
1705 // file i/o error - report dlerror() msg
1706 return NULL;
1707 }
1708
1709 typedef struct {
1710 Elf32_Half code; // Actual value as defined in elf.h
1711 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1712 char elf_class; // 32 or 64 bit
1713 char endianess; // MSB or LSB
1714 char* name; // String representation
1715 } arch_t;
1716
1717 #ifndef EM_486
1718 #define EM_486 6 /* Intel 80486 */
1719 #endif
1720 #ifndef EM_AARCH64
1721 #define EM_AARCH64 183 /* ARM AARCH64 */
1722 #endif
1723
1724 static const arch_t arch_array[]={
1725 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1726 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1727 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1728 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1729 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1730 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1731 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1732 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1733 #if defined(VM_LITTLE_ENDIAN)
1734 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1735 #else
1736 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1737 #endif
1738 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1739 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1740 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1741 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1742 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1743 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1744 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1745 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1746 };
1747
1748 #if (defined IA32)
1749 static Elf32_Half running_arch_code=EM_386;
1750 #elif (defined AMD64)
1751 static Elf32_Half running_arch_code=EM_X86_64;
1752 #elif (defined IA64)
1753 static Elf32_Half running_arch_code=EM_IA_64;
1754 #elif (defined __sparc) && (defined _LP64)
1755 static Elf32_Half running_arch_code=EM_SPARCV9;
1756 #elif (defined __sparc) && (!defined _LP64)
1757 static Elf32_Half running_arch_code=EM_SPARC;
1758 #elif (defined __powerpc64__)
1759 static Elf32_Half running_arch_code=EM_PPC64;
1760 #elif (defined __powerpc__)
1761 static Elf32_Half running_arch_code=EM_PPC;
1762 #elif (defined ARM)
1763 static Elf32_Half running_arch_code=EM_ARM;
1764 #elif (defined S390)
1765 static Elf32_Half running_arch_code=EM_S390;
1766 #elif (defined ALPHA)
1767 static Elf32_Half running_arch_code=EM_ALPHA;
1768 #elif (defined MIPSEL)
1769 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1770 #elif (defined PARISC)
1771 static Elf32_Half running_arch_code=EM_PARISC;
1772 #elif (defined MIPS)
1773 static Elf32_Half running_arch_code=EM_MIPS;
1774 #elif (defined M68K)
1775 static Elf32_Half running_arch_code=EM_68K;
1776 #elif (defined AARCH64)
1777 static Elf32_Half running_arch_code=EM_AARCH64;
1778 #else
1779 #error Method os::dll_load requires that one of following is defined:\
1780 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1781 #endif
1782
1783 // Identify compatability class for VM's architecture and library's architecture
1784 // Obtain string descriptions for architectures
1785
1786 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1787 int running_arch_index=-1;
1788
1789 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1790 if (running_arch_code == arch_array[i].code) {
1791 running_arch_index = i;
1792 }
1793 if (lib_arch.code == arch_array[i].code) {
1794 lib_arch.compat_class = arch_array[i].compat_class;
1795 lib_arch.name = arch_array[i].name;
1796 }
1797 }
1798
1799 assert(running_arch_index != -1,
1800 "Didn't find running architecture code (running_arch_code) in arch_array");
1801 if (running_arch_index == -1) {
1802 // Even though running architecture detection failed
1803 // we may still continue with reporting dlerror() message
1804 return NULL;
1805 }
1806
1807 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1808 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1809 return NULL;
1810 }
1811
1812 #ifndef S390
1813 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1814 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1815 return NULL;
1816 }
1817 #endif // !S390
1818
1819 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1820 if (lib_arch.name!=NULL) {
1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1822 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1823 lib_arch.name, arch_array[running_arch_index].name);
1824 } else {
1825 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1826 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1827 lib_arch.code,
1828 arch_array[running_arch_index].name);
1829 }
1830 }
1831
1832 return NULL;
1833 }
1834
1835 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1836 int ebuflen) {
1837 void * result = ::dlopen(filename, RTLD_LAZY);
1838 if (result == NULL) {
1839 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1840 ebuf[ebuflen-1] = '\0';
1841 }
1842 return result;
1843 }
1844
1845 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1846 int ebuflen) {
1847 void * result = NULL;
1848 if (LoadExecStackDllInVMThread) {
1849 result = dlopen_helper(filename, ebuf, ebuflen);
1850 }
1851
1852 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1853 // library that requires an executable stack, or which does not have this
1854 // stack attribute set, dlopen changes the stack attribute to executable. The
1855 // read protection of the guard pages gets lost.
1856 //
1857 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1858 // may have been queued at the same time.
1859
1860 if (!_stack_is_executable) {
1861 JavaThread *jt = Threads::first();
1862
1863 while (jt) {
1864 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1865 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
1866 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1867 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1868 warning("Attempt to reguard stack yellow zone failed.");
1869 }
1870 }
1871 jt = jt->next();
1872 }
1873 }
1874
1875 return result;
1876 }
1877
1878 void* os::dll_lookup(void* handle, const char* name) {
1879 void* res = dlsym(handle, name);
1880 return res;
1881 }
1882
1883 void* os::get_default_process_handle() {
1884 return (void*)::dlopen(NULL, RTLD_LAZY);
1885 }
1886
1887 static bool _print_ascii_file(const char* filename, outputStream* st) {
1888 int fd = ::open(filename, O_RDONLY);
1889 if (fd == -1) {
1890 return false;
1891 }
1892
1893 char buf[32];
1894 int bytes;
1895 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1896 st->print_raw(buf, bytes);
1897 }
1898
1899 ::close(fd);
1900
1901 return true;
1902 }
1903
1904 void os::print_dll_info(outputStream *st) {
1905 st->print_cr("Dynamic libraries:");
1906
1907 char fname[32];
1908 pid_t pid = os::Linux::gettid();
1909
1910 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1911
1912 if (!_print_ascii_file(fname, st)) {
1913 st->print("Can not get library information for pid = %d\n", pid);
1914 }
1915 }
1916
1917 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1918 FILE *procmapsFile = NULL;
1919
1920 // Open the procfs maps file for the current process
1921 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1922 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1923 char line[PATH_MAX + 100];
1924
1925 // Read line by line from 'file'
1926 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1927 u8 base, top, offset, inode;
1928 char permissions[5];
1929 char device[6];
1930 char name[PATH_MAX + 1];
1931
1932 // Parse fields from line
1933 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1934 &base, &top, permissions, &offset, device, &inode, name);
1935
1936 // Filter by device id '00:00' so that we only get file system mapped files.
1937 if (strcmp(device, "00:00") != 0) {
1938
1939 // Call callback with the fields of interest
1940 if(callback(name, (address)base, (address)top, param)) {
1941 // Oops abort, callback aborted
1942 fclose(procmapsFile);
1943 return 1;
1944 }
1945 }
1946 }
1947 fclose(procmapsFile);
1948 }
1949 return 0;
1950 }
1951
1952 void os::print_os_info_brief(outputStream* st) {
1953 os::Linux::print_distro_info(st);
1954
1955 os::Posix::print_uname_info(st);
1956
1957 os::Linux::print_libversion_info(st);
1958
1959 }
1960
1961 void os::print_os_info(outputStream* st) {
1962 st->print("OS:");
1963
1964 os::Linux::print_distro_info(st);
1965
1966 os::Posix::print_uname_info(st);
1967
1968 // Print warning if unsafe chroot environment detected
1969 if (unsafe_chroot_detected) {
1970 st->print("WARNING!! ");
1971 st->print_cr("%s", unstable_chroot_error);
1972 }
1973
1974 os::Linux::print_libversion_info(st);
1975
1976 os::Posix::print_rlimit_info(st);
1977
1978 os::Posix::print_load_average(st);
1979
1980 os::Linux::print_full_memory_info(st);
1981 }
1982
1983 // Try to identify popular distros.
1984 // Most Linux distributions have a /etc/XXX-release file, which contains
1985 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1986 // file that also contains the OS version string. Some have more than one
1987 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1988 // /etc/redhat-release.), so the order is important.
1989 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1990 // their own specific XXX-release file as well as a redhat-release file.
1991 // Because of this the XXX-release file needs to be searched for before the
1992 // redhat-release file.
1993 // Since Red Hat has a lsb-release file that is not very descriptive the
1994 // search for redhat-release needs to be before lsb-release.
1995 // Since the lsb-release file is the new standard it needs to be searched
1996 // before the older style release files.
1997 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1998 // next to last resort. The os-release file is a new standard that contains
1999 // distribution information and the system-release file seems to be an old
2000 // standard that has been replaced by the lsb-release and os-release files.
2001 // Searching for the debian_version file is the last resort. It contains
2002 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2003 // "Debian " is printed before the contents of the debian_version file.
2004
2005 const char* distro_files[] = {
2006 "/etc/oracle-release",
2007 "/etc/mandriva-release",
2008 "/etc/mandrake-release",
2009 "/etc/sun-release",
2010 "/etc/redhat-release",
2011 "/etc/lsb-release",
2012 "/etc/SuSE-release",
2013 "/etc/turbolinux-release",
2014 "/etc/gentoo-release",
2015 "/etc/ltib-release",
2016 "/etc/angstrom-version",
2017 "/etc/system-release",
2018 "/etc/os-release",
2019 NULL };
2020
2021 void os::Linux::print_distro_info(outputStream* st) {
2022 for (int i = 0;; i++) {
2023 const char* file = distro_files[i];
2024 if (file == NULL) {
2025 break; // done
2026 }
2027 // If file prints, we found it.
2028 if (_print_ascii_file(file, st)) {
2029 return;
2030 }
2031 }
2032
2033 if (file_exists("/etc/debian_version")) {
2034 st->print("Debian ");
2035 _print_ascii_file("/etc/debian_version", st);
2036 } else {
2037 st->print("Linux");
2038 }
2039 st->cr();
2040 }
2041
2042 static void parse_os_info(char* distro, size_t length, const char* file) {
2043 FILE* fp = fopen(file, "r");
2044 if (fp != NULL) {
2045 char buf[256];
2046 // get last line of the file.
2047 while (fgets(buf, sizeof(buf), fp)) { }
2048 // Edit out extra stuff in expected ubuntu format
2049 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2050 char* ptr = strstr(buf, "\""); // the name is in quotes
2051 if (ptr != NULL) {
2052 ptr++; // go beyond first quote
2053 char* nl = strchr(ptr, '\"');
2054 if (nl != NULL) *nl = '\0';
2055 strncpy(distro, ptr, length);
2056 } else {
2057 ptr = strstr(buf, "=");
2058 ptr++; // go beyond equals then
2059 char* nl = strchr(ptr, '\n');
2060 if (nl != NULL) *nl = '\0';
2061 strncpy(distro, ptr, length);
2062 }
2063 } else {
2064 // if not in expected Ubuntu format, print out whole line minus \n
2065 char* nl = strchr(buf, '\n');
2066 if (nl != NULL) *nl = '\0';
2067 strncpy(distro, buf, length);
2068 }
2069 // close distro file
2070 fclose(fp);
2071 }
2072 }
2073
2074 void os::get_summary_os_info(char* buf, size_t buflen) {
2075 for (int i = 0;; i++) {
2076 const char* file = distro_files[i];
2077 if (file == NULL) {
2078 break; // ran out of distro_files
2079 }
2080 if (file_exists(file)) {
2081 parse_os_info(buf, buflen, file);
2082 return;
2083 }
2084 }
2085 // special case for debian
2086 if (file_exists("/etc/debian_version")) {
2087 strncpy(buf, "Debian ", buflen);
2088 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2089 } else {
2090 strncpy(buf, "Linux", buflen);
2091 }
2092 }
2093
2094 void os::Linux::print_libversion_info(outputStream* st) {
2095 // libc, pthread
2096 st->print("libc:");
2097 st->print("%s ", os::Linux::glibc_version());
2098 st->print("%s ", os::Linux::libpthread_version());
2099 st->cr();
2100 }
2101
2102 void os::Linux::print_full_memory_info(outputStream* st) {
2103 st->print("\n/proc/meminfo:\n");
2104 _print_ascii_file("/proc/meminfo", st);
2105 st->cr();
2106 }
2107
2108 void os::print_memory_info(outputStream* st) {
2109
2110 st->print("Memory:");
2111 st->print(" %dk page", os::vm_page_size()>>10);
2112
2113 // values in struct sysinfo are "unsigned long"
2114 struct sysinfo si;
2115 sysinfo(&si);
2116
2117 st->print(", physical " UINT64_FORMAT "k",
2118 os::physical_memory() >> 10);
2119 st->print("(" UINT64_FORMAT "k free)",
2120 os::available_memory() >> 10);
2121 st->print(", swap " UINT64_FORMAT "k",
2122 ((jlong)si.totalswap * si.mem_unit) >> 10);
2123 st->print("(" UINT64_FORMAT "k free)",
2124 ((jlong)si.freeswap * si.mem_unit) >> 10);
2125 st->cr();
2126 }
2127
2128 // Print the first "model name" line and the first "flags" line
2129 // that we find and nothing more. We assume "model name" comes
2130 // before "flags" so if we find a second "model name", then the
2131 // "flags" field is considered missing.
2132 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2133 #if defined(IA32) || defined(AMD64)
2134 // Other platforms have less repetitive cpuinfo files
2135 FILE *fp = fopen("/proc/cpuinfo", "r");
2136 if (fp) {
2137 while (!feof(fp)) {
2138 if (fgets(buf, buflen, fp)) {
2139 // Assume model name comes before flags
2140 bool model_name_printed = false;
2141 if (strstr(buf, "model name") != NULL) {
2142 if (!model_name_printed) {
2143 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2144 st->print_raw(buf);
2145 model_name_printed = true;
2146 } else {
2147 // model name printed but not flags? Odd, just return
2148 fclose(fp);
2149 return true;
2150 }
2151 }
2152 // print the flags line too
2153 if (strstr(buf, "flags") != NULL) {
2154 st->print_raw(buf);
2155 fclose(fp);
2156 return true;
2157 }
2158 }
2159 }
2160 fclose(fp);
2161 }
2162 #endif // x86 platforms
2163 return false;
2164 }
2165
2166 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2167 // Only print the model name if the platform provides this as a summary
2168 if (!print_model_name_and_flags(st, buf, buflen)) {
2169 st->print("\n/proc/cpuinfo:\n");
2170 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2171 st->print_cr(" <Not Available>");
2172 }
2173 }
2174 }
2175
2176 #if defined(AMD64) || defined(IA32) || defined(X32)
2177 const char* search_string = "model name";
2178 #elif defined(SPARC)
2179 const char* search_string = "cpu";
2180 #else
2181 const char* search_string = "Processor";
2182 #endif
2183
2184 // Parses the cpuinfo file for string representing the model name.
2185 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2186 FILE* fp = fopen("/proc/cpuinfo", "r");
2187 if (fp != NULL) {
2188 while (!feof(fp)) {
2189 char buf[256];
2190 if (fgets(buf, sizeof(buf), fp)) {
2191 char* start = strstr(buf, search_string);
2192 if (start != NULL) {
2193 char *ptr = start + strlen(search_string);
2194 char *end = buf + strlen(buf);
2195 while (ptr != end) {
2196 // skip whitespace and colon for the rest of the name.
2197 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2198 break;
2199 }
2200 ptr++;
2201 }
2202 if (ptr != end) {
2203 // reasonable string, get rid of newline and keep the rest
2204 char* nl = strchr(buf, '\n');
2205 if (nl != NULL) *nl = '\0';
2206 strncpy(cpuinfo, ptr, length);
2207 fclose(fp);
2208 return;
2209 }
2210 }
2211 }
2212 }
2213 fclose(fp);
2214 }
2215 // cpuinfo not found or parsing failed, just print generic string. The entire
2216 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2217 #if defined(AMD64)
2218 strncpy(cpuinfo, "x86_64", length);
2219 #elif defined(IA32)
2220 strncpy(cpuinfo, "x86_32", length);
2221 #elif defined(IA64)
2222 strncpy(cpuinfo, "IA64", length);
2223 #elif defined(SPARC)
2224 strncpy(cpuinfo, "sparcv9", length);
2225 #elif defined(AARCH64)
2226 strncpy(cpuinfo, "AArch64", length);
2227 #elif defined(ARM)
2228 strncpy(cpuinfo, "ARM", length);
2229 #elif defined(PPC)
2230 strncpy(cpuinfo, "PPC64", length);
2231 #elif defined(ZERO_LIBARCH)
2232 strncpy(cpuinfo, ZERO_LIBARCH, length);
2233 #else
2234 strncpy(cpuinfo, "unknown", length);
2235 #endif
2236 }
2237
2238 void os::print_siginfo(outputStream* st, void* siginfo) {
2239 const siginfo_t* si = (const siginfo_t*)siginfo;
2240
2241 os::Posix::print_siginfo_brief(st, si);
2242 #if INCLUDE_CDS
2243 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2244 UseSharedSpaces) {
2245 FileMapInfo* mapinfo = FileMapInfo::current_info();
2246 if (mapinfo->is_in_shared_space(si->si_addr)) {
2247 st->print("\n\nError accessing class data sharing archive." \
2248 " Mapped file inaccessible during execution, " \
2249 " possible disk/network problem.");
2250 }
2251 }
2252 #endif
2253 st->cr();
2254 }
2255
2256
2257 static void print_signal_handler(outputStream* st, int sig,
2258 char* buf, size_t buflen);
2259
2260 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2261 st->print_cr("Signal Handlers:");
2262 print_signal_handler(st, SIGSEGV, buf, buflen);
2263 print_signal_handler(st, SIGBUS , buf, buflen);
2264 print_signal_handler(st, SIGFPE , buf, buflen);
2265 print_signal_handler(st, SIGPIPE, buf, buflen);
2266 print_signal_handler(st, SIGXFSZ, buf, buflen);
2267 print_signal_handler(st, SIGILL , buf, buflen);
2268 print_signal_handler(st, SR_signum, buf, buflen);
2269 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2270 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2271 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2272 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2273 #if defined(PPC64)
2274 print_signal_handler(st, SIGTRAP, buf, buflen);
2275 #endif
2276 }
2277
2278 static char saved_jvm_path[MAXPATHLEN] = {0};
2279
2280 // Find the full path to the current module, libjvm.so
2281 void os::jvm_path(char *buf, jint buflen) {
2282 // Error checking.
2283 if (buflen < MAXPATHLEN) {
2284 assert(false, "must use a large-enough buffer");
2285 buf[0] = '\0';
2286 return;
2287 }
2288 // Lazy resolve the path to current module.
2289 if (saved_jvm_path[0] != 0) {
2290 strcpy(buf, saved_jvm_path);
2291 return;
2292 }
2293
2294 char dli_fname[MAXPATHLEN];
2295 bool ret = dll_address_to_library_name(
2296 CAST_FROM_FN_PTR(address, os::jvm_path),
2297 dli_fname, sizeof(dli_fname), NULL);
2298 assert(ret, "cannot locate libjvm");
2299 char *rp = NULL;
2300 if (ret && dli_fname[0] != '\0') {
2301 rp = realpath(dli_fname, buf);
2302 }
2303 if (rp == NULL) {
2304 return;
2305 }
2306
2307 if (Arguments::sun_java_launcher_is_altjvm()) {
2308 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2309 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2310 // If "/jre/lib/" appears at the right place in the string, then
2311 // assume we are installed in a JDK and we're done. Otherwise, check
2312 // for a JAVA_HOME environment variable and fix up the path so it
2313 // looks like libjvm.so is installed there (append a fake suffix
2314 // hotspot/libjvm.so).
2315 const char *p = buf + strlen(buf) - 1;
2316 for (int count = 0; p > buf && count < 5; ++count) {
2317 for (--p; p > buf && *p != '/'; --p)
2318 /* empty */ ;
2319 }
2320
2321 if (strncmp(p, "/jre/lib/", 9) != 0) {
2322 // Look for JAVA_HOME in the environment.
2323 char* java_home_var = ::getenv("JAVA_HOME");
2324 if (java_home_var != NULL && java_home_var[0] != 0) {
2325 char* jrelib_p;
2326 int len;
2327
2328 // Check the current module name "libjvm.so".
2329 p = strrchr(buf, '/');
2330 if (p == NULL) {
2331 return;
2332 }
2333 assert(strstr(p, "/libjvm") == p, "invalid library name");
2334
2335 rp = realpath(java_home_var, buf);
2336 if (rp == NULL) {
2337 return;
2338 }
2339
2340 // determine if this is a legacy image or modules image
2341 // modules image doesn't have "jre" subdirectory
2342 len = strlen(buf);
2343 assert(len < buflen, "Ran out of buffer room");
2344 jrelib_p = buf + len;
2345 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2346 if (0 != access(buf, F_OK)) {
2347 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2348 }
2349
2350 if (0 == access(buf, F_OK)) {
2351 // Use current module name "libjvm.so"
2352 len = strlen(buf);
2353 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2354 } else {
2355 // Go back to path of .so
2356 rp = realpath(dli_fname, buf);
2357 if (rp == NULL) {
2358 return;
2359 }
2360 }
2361 }
2362 }
2363 }
2364
2365 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2366 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2367 }
2368
2369 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2370 // no prefix required, not even "_"
2371 }
2372
2373 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2374 // no suffix required
2375 }
2376
2377 ////////////////////////////////////////////////////////////////////////////////
2378 // sun.misc.Signal support
2379
2380 static volatile jint sigint_count = 0;
2381
2382 static void UserHandler(int sig, void *siginfo, void *context) {
2383 // 4511530 - sem_post is serialized and handled by the manager thread. When
2384 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2385 // don't want to flood the manager thread with sem_post requests.
2386 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2387 return;
2388 }
2389
2390 // Ctrl-C is pressed during error reporting, likely because the error
2391 // handler fails to abort. Let VM die immediately.
2392 if (sig == SIGINT && is_error_reported()) {
2393 os::die();
2394 }
2395
2396 os::signal_notify(sig);
2397 }
2398
2399 void* os::user_handler() {
2400 return CAST_FROM_FN_PTR(void*, UserHandler);
2401 }
2402
2403 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2404 struct timespec ts;
2405 // Semaphore's are always associated with CLOCK_REALTIME
2406 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2407 // see unpackTime for discussion on overflow checking
2408 if (sec >= MAX_SECS) {
2409 ts.tv_sec += MAX_SECS;
2410 ts.tv_nsec = 0;
2411 } else {
2412 ts.tv_sec += sec;
2413 ts.tv_nsec += nsec;
2414 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2415 ts.tv_nsec -= NANOSECS_PER_SEC;
2416 ++ts.tv_sec; // note: this must be <= max_secs
2417 }
2418 }
2419
2420 return ts;
2421 }
2422
2423 extern "C" {
2424 typedef void (*sa_handler_t)(int);
2425 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2426 }
2427
2428 void* os::signal(int signal_number, void* handler) {
2429 struct sigaction sigAct, oldSigAct;
2430
2431 sigfillset(&(sigAct.sa_mask));
2432 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2433 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2434
2435 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2436 // -1 means registration failed
2437 return (void *)-1;
2438 }
2439
2440 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2441 }
2442
2443 void os::signal_raise(int signal_number) {
2444 ::raise(signal_number);
2445 }
2446
2447 // The following code is moved from os.cpp for making this
2448 // code platform specific, which it is by its very nature.
2449
2450 // Will be modified when max signal is changed to be dynamic
2451 int os::sigexitnum_pd() {
2452 return NSIG;
2453 }
2454
2455 // a counter for each possible signal value
2456 static volatile jint pending_signals[NSIG+1] = { 0 };
2457
2458 // Linux(POSIX) specific hand shaking semaphore.
2459 static sem_t sig_sem;
2460 static PosixSemaphore sr_semaphore;
2461
2462 void os::signal_init_pd() {
2463 // Initialize signal structures
2464 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2465
2466 // Initialize signal semaphore
2467 ::sem_init(&sig_sem, 0, 0);
2468 }
2469
2470 void os::signal_notify(int sig) {
2471 Atomic::inc(&pending_signals[sig]);
2472 ::sem_post(&sig_sem);
2473 }
2474
2475 static int check_pending_signals(bool wait) {
2476 Atomic::store(0, &sigint_count);
2477 for (;;) {
2478 for (int i = 0; i < NSIG + 1; i++) {
2479 jint n = pending_signals[i];
2480 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2481 return i;
2482 }
2483 }
2484 if (!wait) {
2485 return -1;
2486 }
2487 JavaThread *thread = JavaThread::current();
2488 ThreadBlockInVM tbivm(thread);
2489
2490 bool threadIsSuspended;
2491 do {
2492 thread->set_suspend_equivalent();
2493 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2494 ::sem_wait(&sig_sem);
2495
2496 // were we externally suspended while we were waiting?
2497 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2498 if (threadIsSuspended) {
2499 // The semaphore has been incremented, but while we were waiting
2500 // another thread suspended us. We don't want to continue running
2501 // while suspended because that would surprise the thread that
2502 // suspended us.
2503 ::sem_post(&sig_sem);
2504
2505 thread->java_suspend_self();
2506 }
2507 } while (threadIsSuspended);
2508 }
2509 }
2510
2511 int os::signal_lookup() {
2512 return check_pending_signals(false);
2513 }
2514
2515 int os::signal_wait() {
2516 return check_pending_signals(true);
2517 }
2518
2519 ////////////////////////////////////////////////////////////////////////////////
2520 // Virtual Memory
2521
2522 int os::vm_page_size() {
2523 // Seems redundant as all get out
2524 assert(os::Linux::page_size() != -1, "must call os::init");
2525 return os::Linux::page_size();
2526 }
2527
2528 // Solaris allocates memory by pages.
2529 int os::vm_allocation_granularity() {
2530 assert(os::Linux::page_size() != -1, "must call os::init");
2531 return os::Linux::page_size();
2532 }
2533
2534 // Rationale behind this function:
2535 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2536 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2537 // samples for JITted code. Here we create private executable mapping over the code cache
2538 // and then we can use standard (well, almost, as mapping can change) way to provide
2539 // info for the reporting script by storing timestamp and location of symbol
2540 void linux_wrap_code(char* base, size_t size) {
2541 static volatile jint cnt = 0;
2542
2543 if (!UseOprofile) {
2544 return;
2545 }
2546
2547 char buf[PATH_MAX+1];
2548 int num = Atomic::add(1, &cnt);
2549
2550 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2551 os::get_temp_directory(), os::current_process_id(), num);
2552 unlink(buf);
2553
2554 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2555
2556 if (fd != -1) {
2557 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2558 if (rv != (off_t)-1) {
2559 if (::write(fd, "", 1) == 1) {
2560 mmap(base, size,
2561 PROT_READ|PROT_WRITE|PROT_EXEC,
2562 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2563 }
2564 }
2565 ::close(fd);
2566 unlink(buf);
2567 }
2568 }
2569
2570 static bool recoverable_mmap_error(int err) {
2571 // See if the error is one we can let the caller handle. This
2572 // list of errno values comes from JBS-6843484. I can't find a
2573 // Linux man page that documents this specific set of errno
2574 // values so while this list currently matches Solaris, it may
2575 // change as we gain experience with this failure mode.
2576 switch (err) {
2577 case EBADF:
2578 case EINVAL:
2579 case ENOTSUP:
2580 // let the caller deal with these errors
2581 return true;
2582
2583 default:
2584 // Any remaining errors on this OS can cause our reserved mapping
2585 // to be lost. That can cause confusion where different data
2586 // structures think they have the same memory mapped. The worst
2587 // scenario is if both the VM and a library think they have the
2588 // same memory mapped.
2589 return false;
2590 }
2591 }
2592
2593 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2594 int err) {
2595 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2596 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2597 strerror(err), err);
2598 }
2599
2600 static void warn_fail_commit_memory(char* addr, size_t size,
2601 size_t alignment_hint, bool exec,
2602 int err) {
2603 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2604 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2605 alignment_hint, exec, strerror(err), err);
2606 }
2607
2608 // NOTE: Linux kernel does not really reserve the pages for us.
2609 // All it does is to check if there are enough free pages
2610 // left at the time of mmap(). This could be a potential
2611 // problem.
2612 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2613 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2614 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2615 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2616 if (res != (uintptr_t) MAP_FAILED) {
2617 if (UseNUMAInterleaving) {
2618 numa_make_global(addr, size);
2619 }
2620 return 0;
2621 }
2622
2623 int err = errno; // save errno from mmap() call above
2624
2625 if (!recoverable_mmap_error(err)) {
2626 warn_fail_commit_memory(addr, size, exec, err);
2627 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2628 }
2629
2630 return err;
2631 }
2632
2633 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2634 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2635 }
2636
2637 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2638 const char* mesg) {
2639 assert(mesg != NULL, "mesg must be specified");
2640 int err = os::Linux::commit_memory_impl(addr, size, exec);
2641 if (err != 0) {
2642 // the caller wants all commit errors to exit with the specified mesg:
2643 warn_fail_commit_memory(addr, size, exec, err);
2644 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2645 }
2646 }
2647
2648 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2649 #ifndef MAP_HUGETLB
2650 #define MAP_HUGETLB 0x40000
2651 #endif
2652
2653 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2654 #ifndef MADV_HUGEPAGE
2655 #define MADV_HUGEPAGE 14
2656 #endif
2657
2658 int os::Linux::commit_memory_impl(char* addr, size_t size,
2659 size_t alignment_hint, bool exec) {
2660 int err = os::Linux::commit_memory_impl(addr, size, exec);
2661 if (err == 0) {
2662 realign_memory(addr, size, alignment_hint);
2663 }
2664 return err;
2665 }
2666
2667 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2668 bool exec) {
2669 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2670 }
2671
2672 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2673 size_t alignment_hint, bool exec,
2674 const char* mesg) {
2675 assert(mesg != NULL, "mesg must be specified");
2676 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2677 if (err != 0) {
2678 // the caller wants all commit errors to exit with the specified mesg:
2679 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2680 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2681 }
2682 }
2683
2684 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2685 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2686 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2687 // be supported or the memory may already be backed by huge pages.
2688 ::madvise(addr, bytes, MADV_HUGEPAGE);
2689 }
2690 }
2691
2692 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2693 // This method works by doing an mmap over an existing mmaping and effectively discarding
2694 // the existing pages. However it won't work for SHM-based large pages that cannot be
2695 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2696 // small pages on top of the SHM segment. This method always works for small pages, so we
2697 // allow that in any case.
2698 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2699 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2700 }
2701 }
2702
2703 void os::numa_make_global(char *addr, size_t bytes) {
2704 Linux::numa_interleave_memory(addr, bytes);
2705 }
2706
2707 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2708 // bind policy to MPOL_PREFERRED for the current thread.
2709 #define USE_MPOL_PREFERRED 0
2710
2711 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2712 // To make NUMA and large pages more robust when both enabled, we need to ease
2713 // the requirements on where the memory should be allocated. MPOL_BIND is the
2714 // default policy and it will force memory to be allocated on the specified
2715 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2716 // the specified node, but will not force it. Using this policy will prevent
2717 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2718 // free large pages.
2719 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2720 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2721 }
2722
2723 bool os::numa_topology_changed() { return false; }
2724
2725 size_t os::numa_get_groups_num() {
2726 int max_node = Linux::numa_max_node();
2727 return max_node > 0 ? max_node + 1 : 1;
2728 }
2729
2730 int os::numa_get_group_id() {
2731 int cpu_id = Linux::sched_getcpu();
2732 if (cpu_id != -1) {
2733 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2734 if (lgrp_id != -1) {
2735 return lgrp_id;
2736 }
2737 }
2738 return 0;
2739 }
2740
2741 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2742 for (size_t i = 0; i < size; i++) {
2743 ids[i] = i;
2744 }
2745 return size;
2746 }
2747
2748 bool os::get_page_info(char *start, page_info* info) {
2749 return false;
2750 }
2751
2752 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2753 page_info* page_found) {
2754 return end;
2755 }
2756
2757
2758 int os::Linux::sched_getcpu_syscall(void) {
2759 unsigned int cpu = 0;
2760 int retval = -1;
2761
2762 #if defined(IA32)
2763 #ifndef SYS_getcpu
2764 #define SYS_getcpu 318
2765 #endif
2766 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2767 #elif defined(AMD64)
2768 // Unfortunately we have to bring all these macros here from vsyscall.h
2769 // to be able to compile on old linuxes.
2770 #define __NR_vgetcpu 2
2771 #define VSYSCALL_START (-10UL << 20)
2772 #define VSYSCALL_SIZE 1024
2773 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2774 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2775 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2776 retval = vgetcpu(&cpu, NULL, NULL);
2777 #endif
2778
2779 return (retval == -1) ? retval : cpu;
2780 }
2781
2782 // Something to do with the numa-aware allocator needs these symbols
2783 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2784 extern "C" JNIEXPORT void numa_error(char *where) { }
2785
2786
2787 // If we are running with libnuma version > 2, then we should
2788 // be trying to use symbols with versions 1.1
2789 // If we are running with earlier version, which did not have symbol versions,
2790 // we should use the base version.
2791 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2792 void *f = dlvsym(handle, name, "libnuma_1.1");
2793 if (f == NULL) {
2794 f = dlsym(handle, name);
2795 }
2796 return f;
2797 }
2798
2799 bool os::Linux::libnuma_init() {
2800 // sched_getcpu() should be in libc.
2801 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2802 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2803
2804 // If it's not, try a direct syscall.
2805 if (sched_getcpu() == -1) {
2806 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2807 (void*)&sched_getcpu_syscall));
2808 }
2809
2810 if (sched_getcpu() != -1) { // Does it work?
2811 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2812 if (handle != NULL) {
2813 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2814 libnuma_dlsym(handle, "numa_node_to_cpus")));
2815 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2816 libnuma_dlsym(handle, "numa_max_node")));
2817 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2818 libnuma_dlsym(handle, "numa_available")));
2819 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2820 libnuma_dlsym(handle, "numa_tonode_memory")));
2821 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2822 libnuma_dlsym(handle, "numa_interleave_memory")));
2823 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2824 libnuma_dlsym(handle, "numa_set_bind_policy")));
2825
2826
2827 if (numa_available() != -1) {
2828 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2829 // Create a cpu -> node mapping
2830 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2831 rebuild_cpu_to_node_map();
2832 return true;
2833 }
2834 }
2835 }
2836 return false;
2837 }
2838
2839 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2840 // The table is later used in get_node_by_cpu().
2841 void os::Linux::rebuild_cpu_to_node_map() {
2842 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2843 // in libnuma (possible values are starting from 16,
2844 // and continuing up with every other power of 2, but less
2845 // than the maximum number of CPUs supported by kernel), and
2846 // is a subject to change (in libnuma version 2 the requirements
2847 // are more reasonable) we'll just hardcode the number they use
2848 // in the library.
2849 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2850
2851 size_t cpu_num = os::active_processor_count();
2852 size_t cpu_map_size = NCPUS / BitsPerCLong;
2853 size_t cpu_map_valid_size =
2854 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2855
2856 cpu_to_node()->clear();
2857 cpu_to_node()->at_grow(cpu_num - 1);
2858 size_t node_num = numa_get_groups_num();
2859
2860 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2861 for (size_t i = 0; i < node_num; i++) {
2862 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2863 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2864 if (cpu_map[j] != 0) {
2865 for (size_t k = 0; k < BitsPerCLong; k++) {
2866 if (cpu_map[j] & (1UL << k)) {
2867 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2868 }
2869 }
2870 }
2871 }
2872 }
2873 }
2874 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2875 }
2876
2877 int os::Linux::get_node_by_cpu(int cpu_id) {
2878 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2879 return cpu_to_node()->at(cpu_id);
2880 }
2881 return -1;
2882 }
2883
2884 GrowableArray<int>* os::Linux::_cpu_to_node;
2885 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2886 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2887 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2888 os::Linux::numa_available_func_t os::Linux::_numa_available;
2889 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2890 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2891 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2892 unsigned long* os::Linux::_numa_all_nodes;
2893
2894 bool os::pd_uncommit_memory(char* addr, size_t size) {
2895 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2896 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2897 return res != (uintptr_t) MAP_FAILED;
2898 }
2899
2900 static address get_stack_commited_bottom(address bottom, size_t size) {
2901 address nbot = bottom;
2902 address ntop = bottom + size;
2903
2904 size_t page_sz = os::vm_page_size();
2905 unsigned pages = size / page_sz;
2906
2907 unsigned char vec[1];
2908 unsigned imin = 1, imax = pages + 1, imid;
2909 int mincore_return_value = 0;
2910
2911 assert(imin <= imax, "Unexpected page size");
2912
2913 while (imin < imax) {
2914 imid = (imax + imin) / 2;
2915 nbot = ntop - (imid * page_sz);
2916
2917 // Use a trick with mincore to check whether the page is mapped or not.
2918 // mincore sets vec to 1 if page resides in memory and to 0 if page
2919 // is swapped output but if page we are asking for is unmapped
2920 // it returns -1,ENOMEM
2921 mincore_return_value = mincore(nbot, page_sz, vec);
2922
2923 if (mincore_return_value == -1) {
2924 // Page is not mapped go up
2925 // to find first mapped page
2926 if (errno != EAGAIN) {
2927 assert(errno == ENOMEM, "Unexpected mincore errno");
2928 imax = imid;
2929 }
2930 } else {
2931 // Page is mapped go down
2932 // to find first not mapped page
2933 imin = imid + 1;
2934 }
2935 }
2936
2937 nbot = nbot + page_sz;
2938
2939 // Adjust stack bottom one page up if last checked page is not mapped
2940 if (mincore_return_value == -1) {
2941 nbot = nbot + page_sz;
2942 }
2943
2944 return nbot;
2945 }
2946
2947
2948 // Linux uses a growable mapping for the stack, and if the mapping for
2949 // the stack guard pages is not removed when we detach a thread the
2950 // stack cannot grow beyond the pages where the stack guard was
2951 // mapped. If at some point later in the process the stack expands to
2952 // that point, the Linux kernel cannot expand the stack any further
2953 // because the guard pages are in the way, and a segfault occurs.
2954 //
2955 // However, it's essential not to split the stack region by unmapping
2956 // a region (leaving a hole) that's already part of the stack mapping,
2957 // so if the stack mapping has already grown beyond the guard pages at
2958 // the time we create them, we have to truncate the stack mapping.
2959 // So, we need to know the extent of the stack mapping when
2960 // create_stack_guard_pages() is called.
2961
2962 // We only need this for stacks that are growable: at the time of
2963 // writing thread stacks don't use growable mappings (i.e. those
2964 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2965 // only applies to the main thread.
2966
2967 // If the (growable) stack mapping already extends beyond the point
2968 // where we're going to put our guard pages, truncate the mapping at
2969 // that point by munmap()ping it. This ensures that when we later
2970 // munmap() the guard pages we don't leave a hole in the stack
2971 // mapping. This only affects the main/initial thread
2972
2973 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2974 if (os::Linux::is_initial_thread()) {
2975 // As we manually grow stack up to bottom inside create_attached_thread(),
2976 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2977 // we don't need to do anything special.
2978 // Check it first, before calling heavy function.
2979 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2980 unsigned char vec[1];
2981
2982 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2983 // Fallback to slow path on all errors, including EAGAIN
2984 stack_extent = (uintptr_t) get_stack_commited_bottom(
2985 os::Linux::initial_thread_stack_bottom(),
2986 (size_t)addr - stack_extent);
2987 }
2988
2989 if (stack_extent < (uintptr_t)addr) {
2990 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2991 }
2992 }
2993
2994 return os::commit_memory(addr, size, !ExecMem);
2995 }
2996
2997 // If this is a growable mapping, remove the guard pages entirely by
2998 // munmap()ping them. If not, just call uncommit_memory(). This only
2999 // affects the main/initial thread, but guard against future OS changes
3000 // It's safe to always unmap guard pages for initial thread because we
3001 // always place it right after end of the mapped region
3002
3003 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3004 uintptr_t stack_extent, stack_base;
3005
3006 if (os::Linux::is_initial_thread()) {
3007 return ::munmap(addr, size) == 0;
3008 }
3009
3010 return os::uncommit_memory(addr, size);
3011 }
3012
3013 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3014 // at 'requested_addr'. If there are existing memory mappings at the same
3015 // location, however, they will be overwritten. If 'fixed' is false,
3016 // 'requested_addr' is only treated as a hint, the return value may or
3017 // may not start from the requested address. Unlike Linux mmap(), this
3018 // function returns NULL to indicate failure.
3019 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3020 char * addr;
3021 int flags;
3022
3023 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3024 if (fixed) {
3025 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3026 flags |= MAP_FIXED;
3027 }
3028
3029 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3030 // touch an uncommitted page. Otherwise, the read/write might
3031 // succeed if we have enough swap space to back the physical page.
3032 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3033 flags, -1, 0);
3034
3035 return addr == MAP_FAILED ? NULL : addr;
3036 }
3037
3038 static int anon_munmap(char * addr, size_t size) {
3039 return ::munmap(addr, size) == 0;
3040 }
3041
3042 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3043 size_t alignment_hint) {
3044 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3045 }
3046
3047 bool os::pd_release_memory(char* addr, size_t size) {
3048 return anon_munmap(addr, size);
3049 }
3050
3051 static bool linux_mprotect(char* addr, size_t size, int prot) {
3052 // Linux wants the mprotect address argument to be page aligned.
3053 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3054
3055 // According to SUSv3, mprotect() should only be used with mappings
3056 // established by mmap(), and mmap() always maps whole pages. Unaligned
3057 // 'addr' likely indicates problem in the VM (e.g. trying to change
3058 // protection of malloc'ed or statically allocated memory). Check the
3059 // caller if you hit this assert.
3060 assert(addr == bottom, "sanity check");
3061
3062 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3063 return ::mprotect(bottom, size, prot) == 0;
3064 }
3065
3066 // Set protections specified
3067 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3068 bool is_committed) {
3069 unsigned int p = 0;
3070 switch (prot) {
3071 case MEM_PROT_NONE: p = PROT_NONE; break;
3072 case MEM_PROT_READ: p = PROT_READ; break;
3073 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3074 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3075 default:
3076 ShouldNotReachHere();
3077 }
3078 // is_committed is unused.
3079 return linux_mprotect(addr, bytes, p);
3080 }
3081
3082 bool os::guard_memory(char* addr, size_t size) {
3083 return linux_mprotect(addr, size, PROT_NONE);
3084 }
3085
3086 bool os::unguard_memory(char* addr, size_t size) {
3087 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3088 }
3089
3090 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3091 size_t page_size) {
3092 bool result = false;
3093 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3094 MAP_ANONYMOUS|MAP_PRIVATE,
3095 -1, 0);
3096 if (p != MAP_FAILED) {
3097 void *aligned_p = align_ptr_up(p, page_size);
3098
3099 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3100
3101 munmap(p, page_size * 2);
3102 }
3103
3104 if (warn && !result) {
3105 warning("TransparentHugePages is not supported by the operating system.");
3106 }
3107
3108 return result;
3109 }
3110
3111 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3112 bool result = false;
3113 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3114 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3115 -1, 0);
3116
3117 if (p != MAP_FAILED) {
3118 // We don't know if this really is a huge page or not.
3119 FILE *fp = fopen("/proc/self/maps", "r");
3120 if (fp) {
3121 while (!feof(fp)) {
3122 char chars[257];
3123 long x = 0;
3124 if (fgets(chars, sizeof(chars), fp)) {
3125 if (sscanf(chars, "%lx-%*x", &x) == 1
3126 && x == (long)p) {
3127 if (strstr (chars, "hugepage")) {
3128 result = true;
3129 break;
3130 }
3131 }
3132 }
3133 }
3134 fclose(fp);
3135 }
3136 munmap(p, page_size);
3137 }
3138
3139 if (warn && !result) {
3140 warning("HugeTLBFS is not supported by the operating system.");
3141 }
3142
3143 return result;
3144 }
3145
3146 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3147 //
3148 // From the coredump_filter documentation:
3149 //
3150 // - (bit 0) anonymous private memory
3151 // - (bit 1) anonymous shared memory
3152 // - (bit 2) file-backed private memory
3153 // - (bit 3) file-backed shared memory
3154 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3155 // effective only if the bit 2 is cleared)
3156 // - (bit 5) hugetlb private memory
3157 // - (bit 6) hugetlb shared memory
3158 //
3159 static void set_coredump_filter(void) {
3160 FILE *f;
3161 long cdm;
3162
3163 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3164 return;
3165 }
3166
3167 if (fscanf(f, "%lx", &cdm) != 1) {
3168 fclose(f);
3169 return;
3170 }
3171
3172 rewind(f);
3173
3174 if ((cdm & LARGEPAGES_BIT) == 0) {
3175 cdm |= LARGEPAGES_BIT;
3176 fprintf(f, "%#lx", cdm);
3177 }
3178
3179 fclose(f);
3180 }
3181
3182 // Large page support
3183
3184 static size_t _large_page_size = 0;
3185
3186 size_t os::Linux::find_large_page_size() {
3187 size_t large_page_size = 0;
3188
3189 // large_page_size on Linux is used to round up heap size. x86 uses either
3190 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3191 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3192 // page as large as 256M.
3193 //
3194 // Here we try to figure out page size by parsing /proc/meminfo and looking
3195 // for a line with the following format:
3196 // Hugepagesize: 2048 kB
3197 //
3198 // If we can't determine the value (e.g. /proc is not mounted, or the text
3199 // format has been changed), we'll use the largest page size supported by
3200 // the processor.
3201
3202 #ifndef ZERO
3203 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3204 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3205 #endif // ZERO
3206
3207 FILE *fp = fopen("/proc/meminfo", "r");
3208 if (fp) {
3209 while (!feof(fp)) {
3210 int x = 0;
3211 char buf[16];
3212 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3213 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3214 large_page_size = x * K;
3215 break;
3216 }
3217 } else {
3218 // skip to next line
3219 for (;;) {
3220 int ch = fgetc(fp);
3221 if (ch == EOF || ch == (int)'\n') break;
3222 }
3223 }
3224 }
3225 fclose(fp);
3226 }
3227
3228 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3229 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3230 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3231 proper_unit_for_byte_size(large_page_size));
3232 }
3233
3234 return large_page_size;
3235 }
3236
3237 size_t os::Linux::setup_large_page_size() {
3238 _large_page_size = Linux::find_large_page_size();
3239 const size_t default_page_size = (size_t)Linux::page_size();
3240 if (_large_page_size > default_page_size) {
3241 _page_sizes[0] = _large_page_size;
3242 _page_sizes[1] = default_page_size;
3243 _page_sizes[2] = 0;
3244 }
3245
3246 return _large_page_size;
3247 }
3248
3249 bool os::Linux::setup_large_page_type(size_t page_size) {
3250 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3251 FLAG_IS_DEFAULT(UseSHM) &&
3252 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3253
3254 // The type of large pages has not been specified by the user.
3255
3256 // Try UseHugeTLBFS and then UseSHM.
3257 UseHugeTLBFS = UseSHM = true;
3258
3259 // Don't try UseTransparentHugePages since there are known
3260 // performance issues with it turned on. This might change in the future.
3261 UseTransparentHugePages = false;
3262 }
3263
3264 if (UseTransparentHugePages) {
3265 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3266 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3267 UseHugeTLBFS = false;
3268 UseSHM = false;
3269 return true;
3270 }
3271 UseTransparentHugePages = false;
3272 }
3273
3274 if (UseHugeTLBFS) {
3275 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3276 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3277 UseSHM = false;
3278 return true;
3279 }
3280 UseHugeTLBFS = false;
3281 }
3282
3283 return UseSHM;
3284 }
3285
3286 void os::large_page_init() {
3287 if (!UseLargePages &&
3288 !UseTransparentHugePages &&
3289 !UseHugeTLBFS &&
3290 !UseSHM) {
3291 // Not using large pages.
3292 return;
3293 }
3294
3295 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3296 // The user explicitly turned off large pages.
3297 // Ignore the rest of the large pages flags.
3298 UseTransparentHugePages = false;
3299 UseHugeTLBFS = false;
3300 UseSHM = false;
3301 return;
3302 }
3303
3304 size_t large_page_size = Linux::setup_large_page_size();
3305 UseLargePages = Linux::setup_large_page_type(large_page_size);
3306
3307 set_coredump_filter();
3308 }
3309
3310 #ifndef SHM_HUGETLB
3311 #define SHM_HUGETLB 04000
3312 #endif
3313
3314 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3315 char* req_addr, bool exec) {
3316 // "exec" is passed in but not used. Creating the shared image for
3317 // the code cache doesn't have an SHM_X executable permission to check.
3318 assert(UseLargePages && UseSHM, "only for SHM large pages");
3319 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3320
3321 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3322 return NULL; // Fallback to small pages.
3323 }
3324
3325 key_t key = IPC_PRIVATE;
3326 char *addr;
3327
3328 bool warn_on_failure = UseLargePages &&
3329 (!FLAG_IS_DEFAULT(UseLargePages) ||
3330 !FLAG_IS_DEFAULT(UseSHM) ||
3331 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3332 char msg[128];
3333
3334 // Create a large shared memory region to attach to based on size.
3335 // Currently, size is the total size of the heap
3336 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3337 if (shmid == -1) {
3338 // Possible reasons for shmget failure:
3339 // 1. shmmax is too small for Java heap.
3340 // > check shmmax value: cat /proc/sys/kernel/shmmax
3341 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3342 // 2. not enough large page memory.
3343 // > check available large pages: cat /proc/meminfo
3344 // > increase amount of large pages:
3345 // echo new_value > /proc/sys/vm/nr_hugepages
3346 // Note 1: different Linux may use different name for this property,
3347 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3348 // Note 2: it's possible there's enough physical memory available but
3349 // they are so fragmented after a long run that they can't
3350 // coalesce into large pages. Try to reserve large pages when
3351 // the system is still "fresh".
3352 if (warn_on_failure) {
3353 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3354 warning("%s", msg);
3355 }
3356 return NULL;
3357 }
3358
3359 // attach to the region
3360 addr = (char*)shmat(shmid, req_addr, 0);
3361 int err = errno;
3362
3363 // Remove shmid. If shmat() is successful, the actual shared memory segment
3364 // will be deleted when it's detached by shmdt() or when the process
3365 // terminates. If shmat() is not successful this will remove the shared
3366 // segment immediately.
3367 shmctl(shmid, IPC_RMID, NULL);
3368
3369 if ((intptr_t)addr == -1) {
3370 if (warn_on_failure) {
3371 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3372 warning("%s", msg);
3373 }
3374 return NULL;
3375 }
3376
3377 return addr;
3378 }
3379
3380 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3381 int error) {
3382 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3383
3384 bool warn_on_failure = UseLargePages &&
3385 (!FLAG_IS_DEFAULT(UseLargePages) ||
3386 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3387 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3388
3389 if (warn_on_failure) {
3390 char msg[128];
3391 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3392 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3393 warning("%s", msg);
3394 }
3395 }
3396
3397 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3398 char* req_addr,
3399 bool exec) {
3400 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3401 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3402 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3403
3404 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3405 char* addr = (char*)::mmap(req_addr, bytes, prot,
3406 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3407 -1, 0);
3408
3409 if (addr == MAP_FAILED) {
3410 warn_on_large_pages_failure(req_addr, bytes, errno);
3411 return NULL;
3412 }
3413
3414 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3415
3416 return addr;
3417 }
3418
3419 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3420 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3421 // (req_addr != NULL) or with a given alignment.
3422 // - bytes shall be a multiple of alignment.
3423 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3424 // - alignment sets the alignment at which memory shall be allocated.
3425 // It must be a multiple of allocation granularity.
3426 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3427 // req_addr or NULL.
3428 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3429
3430 size_t extra_size = bytes;
3431 if (req_addr == NULL && alignment > 0) {
3432 extra_size += alignment;
3433 }
3434
3435 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3436 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3437 -1, 0);
3438 if (start == MAP_FAILED) {
3439 start = NULL;
3440 } else {
3441 if (req_addr != NULL) {
3442 if (start != req_addr) {
3443 ::munmap(start, extra_size);
3444 start = NULL;
3445 }
3446 } else {
3447 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3448 char* const end_aligned = start_aligned + bytes;
3449 char* const end = start + extra_size;
3450 if (start_aligned > start) {
3451 ::munmap(start, start_aligned - start);
3452 }
3453 if (end_aligned < end) {
3454 ::munmap(end_aligned, end - end_aligned);
3455 }
3456 start = start_aligned;
3457 }
3458 }
3459 return start;
3460
3461 }
3462
3463 // Reserve memory using mmap(MAP_HUGETLB).
3464 // - bytes shall be a multiple of alignment.
3465 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3466 // - alignment sets the alignment at which memory shall be allocated.
3467 // It must be a multiple of allocation granularity.
3468 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3469 // req_addr or NULL.
3470 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3471 size_t alignment,
3472 char* req_addr,
3473 bool exec) {
3474 size_t large_page_size = os::large_page_size();
3475 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3476
3477 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3478 assert(is_size_aligned(bytes, alignment), "Must be");
3479
3480 // First reserve - but not commit - the address range in small pages.
3481 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3482
3483 if (start == NULL) {
3484 return NULL;
3485 }
3486
3487 assert(is_ptr_aligned(start, alignment), "Must be");
3488
3489 char* end = start + bytes;
3490
3491 // Find the regions of the allocated chunk that can be promoted to large pages.
3492 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3493 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3494
3495 size_t lp_bytes = lp_end - lp_start;
3496
3497 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3498
3499 if (lp_bytes == 0) {
3500 // The mapped region doesn't even span the start and the end of a large page.
3501 // Fall back to allocate a non-special area.
3502 ::munmap(start, end - start);
3503 return NULL;
3504 }
3505
3506 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3507
3508 void* result;
3509
3510 // Commit small-paged leading area.
3511 if (start != lp_start) {
3512 result = ::mmap(start, lp_start - start, prot,
3513 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3514 -1, 0);
3515 if (result == MAP_FAILED) {
3516 ::munmap(lp_start, end - lp_start);
3517 return NULL;
3518 }
3519 }
3520
3521 // Commit large-paged area.
3522 result = ::mmap(lp_start, lp_bytes, prot,
3523 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3524 -1, 0);
3525 if (result == MAP_FAILED) {
3526 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3527 // If the mmap above fails, the large pages region will be unmapped and we
3528 // have regions before and after with small pages. Release these regions.
3529 //
3530 // | mapped | unmapped | mapped |
3531 // ^ ^ ^ ^
3532 // start lp_start lp_end end
3533 //
3534 ::munmap(start, lp_start - start);
3535 ::munmap(lp_end, end - lp_end);
3536 return NULL;
3537 }
3538
3539 // Commit small-paged trailing area.
3540 if (lp_end != end) {
3541 result = ::mmap(lp_end, end - lp_end, prot,
3542 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3543 -1, 0);
3544 if (result == MAP_FAILED) {
3545 ::munmap(start, lp_end - start);
3546 return NULL;
3547 }
3548 }
3549
3550 return start;
3551 }
3552
3553 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3554 size_t alignment,
3555 char* req_addr,
3556 bool exec) {
3557 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3558 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3559 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3560 assert(is_power_of_2(os::large_page_size()), "Must be");
3561 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3562
3563 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3564 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3565 } else {
3566 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3567 }
3568 }
3569
3570 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3571 char* req_addr, bool exec) {
3572 assert(UseLargePages, "only for large pages");
3573
3574 char* addr;
3575 if (UseSHM) {
3576 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3577 } else {
3578 assert(UseHugeTLBFS, "must be");
3579 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3580 }
3581
3582 if (addr != NULL) {
3583 if (UseNUMAInterleaving) {
3584 numa_make_global(addr, bytes);
3585 }
3586
3587 // The memory is committed
3588 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3589 }
3590
3591 return addr;
3592 }
3593
3594 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3595 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3596 return shmdt(base) == 0;
3597 }
3598
3599 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3600 return pd_release_memory(base, bytes);
3601 }
3602
3603 bool os::release_memory_special(char* base, size_t bytes) {
3604 bool res;
3605 if (MemTracker::tracking_level() > NMT_minimal) {
3606 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3607 res = os::Linux::release_memory_special_impl(base, bytes);
3608 if (res) {
3609 tkr.record((address)base, bytes);
3610 }
3611
3612 } else {
3613 res = os::Linux::release_memory_special_impl(base, bytes);
3614 }
3615 return res;
3616 }
3617
3618 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3619 assert(UseLargePages, "only for large pages");
3620 bool res;
3621
3622 if (UseSHM) {
3623 res = os::Linux::release_memory_special_shm(base, bytes);
3624 } else {
3625 assert(UseHugeTLBFS, "must be");
3626 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3627 }
3628 return res;
3629 }
3630
3631 size_t os::large_page_size() {
3632 return _large_page_size;
3633 }
3634
3635 // With SysV SHM the entire memory region must be allocated as shared
3636 // memory.
3637 // HugeTLBFS allows application to commit large page memory on demand.
3638 // However, when committing memory with HugeTLBFS fails, the region
3639 // that was supposed to be committed will lose the old reservation
3640 // and allow other threads to steal that memory region. Because of this
3641 // behavior we can't commit HugeTLBFS memory.
3642 bool os::can_commit_large_page_memory() {
3643 return UseTransparentHugePages;
3644 }
3645
3646 bool os::can_execute_large_page_memory() {
3647 return UseTransparentHugePages || UseHugeTLBFS;
3648 }
3649
3650 // Reserve memory at an arbitrary address, only if that area is
3651 // available (and not reserved for something else).
3652
3653 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3654 const int max_tries = 10;
3655 char* base[max_tries];
3656 size_t size[max_tries];
3657 const size_t gap = 0x000000;
3658
3659 // Assert only that the size is a multiple of the page size, since
3660 // that's all that mmap requires, and since that's all we really know
3661 // about at this low abstraction level. If we need higher alignment,
3662 // we can either pass an alignment to this method or verify alignment
3663 // in one of the methods further up the call chain. See bug 5044738.
3664 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3665
3666 // Repeatedly allocate blocks until the block is allocated at the
3667 // right spot.
3668
3669 // Linux mmap allows caller to pass an address as hint; give it a try first,
3670 // if kernel honors the hint then we can return immediately.
3671 char * addr = anon_mmap(requested_addr, bytes, false);
3672 if (addr == requested_addr) {
3673 return requested_addr;
3674 }
3675
3676 if (addr != NULL) {
3677 // mmap() is successful but it fails to reserve at the requested address
3678 anon_munmap(addr, bytes);
3679 }
3680
3681 int i;
3682 for (i = 0; i < max_tries; ++i) {
3683 base[i] = reserve_memory(bytes);
3684
3685 if (base[i] != NULL) {
3686 // Is this the block we wanted?
3687 if (base[i] == requested_addr) {
3688 size[i] = bytes;
3689 break;
3690 }
3691
3692 // Does this overlap the block we wanted? Give back the overlapped
3693 // parts and try again.
3694
3695 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3696 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3697 unmap_memory(base[i], top_overlap);
3698 base[i] += top_overlap;
3699 size[i] = bytes - top_overlap;
3700 } else {
3701 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3702 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3703 unmap_memory(requested_addr, bottom_overlap);
3704 size[i] = bytes - bottom_overlap;
3705 } else {
3706 size[i] = bytes;
3707 }
3708 }
3709 }
3710 }
3711
3712 // Give back the unused reserved pieces.
3713
3714 for (int j = 0; j < i; ++j) {
3715 if (base[j] != NULL) {
3716 unmap_memory(base[j], size[j]);
3717 }
3718 }
3719
3720 if (i < max_tries) {
3721 return requested_addr;
3722 } else {
3723 return NULL;
3724 }
3725 }
3726
3727 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3728 return ::read(fd, buf, nBytes);
3729 }
3730
3731 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3732 return ::pread(fd, buf, nBytes, offset);
3733 }
3734
3735 // Short sleep, direct OS call.
3736 //
3737 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3738 // sched_yield(2) will actually give up the CPU:
3739 //
3740 // * Alone on this pariticular CPU, keeps running.
3741 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3742 // (pre 2.6.39).
3743 //
3744 // So calling this with 0 is an alternative.
3745 //
3746 void os::naked_short_sleep(jlong ms) {
3747 struct timespec req;
3748
3749 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3750 req.tv_sec = 0;
3751 if (ms > 0) {
3752 req.tv_nsec = (ms % 1000) * 1000000;
3753 } else {
3754 req.tv_nsec = 1;
3755 }
3756
3757 nanosleep(&req, NULL);
3758
3759 return;
3760 }
3761
3762 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3763 void os::infinite_sleep() {
3764 while (true) { // sleep forever ...
3765 ::sleep(100); // ... 100 seconds at a time
3766 }
3767 }
3768
3769 // Used to convert frequent JVM_Yield() to nops
3770 bool os::dont_yield() {
3771 return DontYieldALot;
3772 }
3773
3774 void os::naked_yield() {
3775 sched_yield();
3776 }
3777
3778 ////////////////////////////////////////////////////////////////////////////////
3779 // thread priority support
3780
3781 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3782 // only supports dynamic priority, static priority must be zero. For real-time
3783 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3784 // However, for large multi-threaded applications, SCHED_RR is not only slower
3785 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3786 // of 5 runs - Sep 2005).
3787 //
3788 // The following code actually changes the niceness of kernel-thread/LWP. It
3789 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3790 // not the entire user process, and user level threads are 1:1 mapped to kernel
3791 // threads. It has always been the case, but could change in the future. For
3792 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3793 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3794
3795 int os::java_to_os_priority[CriticalPriority + 1] = {
3796 19, // 0 Entry should never be used
3797
3798 4, // 1 MinPriority
3799 3, // 2
3800 2, // 3
3801
3802 1, // 4
3803 0, // 5 NormPriority
3804 -1, // 6
3805
3806 -2, // 7
3807 -3, // 8
3808 -4, // 9 NearMaxPriority
3809
3810 -5, // 10 MaxPriority
3811
3812 -5 // 11 CriticalPriority
3813 };
3814
3815 static int prio_init() {
3816 if (ThreadPriorityPolicy == 1) {
3817 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3818 // if effective uid is not root. Perhaps, a more elegant way of doing
3819 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3820 if (geteuid() != 0) {
3821 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3822 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3823 }
3824 ThreadPriorityPolicy = 0;
3825 }
3826 }
3827 if (UseCriticalJavaThreadPriority) {
3828 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3829 }
3830 return 0;
3831 }
3832
3833 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3834 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3835
3836 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3837 return (ret == 0) ? OS_OK : OS_ERR;
3838 }
3839
3840 OSReturn os::get_native_priority(const Thread* const thread,
3841 int *priority_ptr) {
3842 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3843 *priority_ptr = java_to_os_priority[NormPriority];
3844 return OS_OK;
3845 }
3846
3847 errno = 0;
3848 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3849 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3850 }
3851
3852 // Hint to the underlying OS that a task switch would not be good.
3853 // Void return because it's a hint and can fail.
3854 void os::hint_no_preempt() {}
3855
3856 ////////////////////////////////////////////////////////////////////////////////
3857 // suspend/resume support
3858
3859 // the low-level signal-based suspend/resume support is a remnant from the
3860 // old VM-suspension that used to be for java-suspension, safepoints etc,
3861 // within hotspot. Now there is a single use-case for this:
3862 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3863 // that runs in the watcher thread.
3864 // The remaining code is greatly simplified from the more general suspension
3865 // code that used to be used.
3866 //
3867 // The protocol is quite simple:
3868 // - suspend:
3869 // - sends a signal to the target thread
3870 // - polls the suspend state of the osthread using a yield loop
3871 // - target thread signal handler (SR_handler) sets suspend state
3872 // and blocks in sigsuspend until continued
3873 // - resume:
3874 // - sets target osthread state to continue
3875 // - sends signal to end the sigsuspend loop in the SR_handler
3876 //
3877 // Note that the SR_lock plays no role in this suspend/resume protocol.
3878
3879 static void resume_clear_context(OSThread *osthread) {
3880 osthread->set_ucontext(NULL);
3881 osthread->set_siginfo(NULL);
3882 }
3883
3884 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3885 ucontext_t* context) {
3886 osthread->set_ucontext(context);
3887 osthread->set_siginfo(siginfo);
3888 }
3889
3890 // Handler function invoked when a thread's execution is suspended or
3891 // resumed. We have to be careful that only async-safe functions are
3892 // called here (Note: most pthread functions are not async safe and
3893 // should be avoided.)
3894 //
3895 // Note: sigwait() is a more natural fit than sigsuspend() from an
3896 // interface point of view, but sigwait() prevents the signal hander
3897 // from being run. libpthread would get very confused by not having
3898 // its signal handlers run and prevents sigwait()'s use with the
3899 // mutex granting granting signal.
3900 //
3901 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3902 //
3903 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3904 // Save and restore errno to avoid confusing native code with EINTR
3905 // after sigsuspend.
3906 int old_errno = errno;
3907
3908 Thread* thread = Thread::current();
3909 OSThread* osthread = thread->osthread();
3910 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3911
3912 os::SuspendResume::State current = osthread->sr.state();
3913 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3914 suspend_save_context(osthread, siginfo, context);
3915
3916 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3917 os::SuspendResume::State state = osthread->sr.suspended();
3918 if (state == os::SuspendResume::SR_SUSPENDED) {
3919 sigset_t suspend_set; // signals for sigsuspend()
3920
3921 // get current set of blocked signals and unblock resume signal
3922 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3923 sigdelset(&suspend_set, SR_signum);
3924
3925 sr_semaphore.signal();
3926 // wait here until we are resumed
3927 while (1) {
3928 sigsuspend(&suspend_set);
3929
3930 os::SuspendResume::State result = osthread->sr.running();
3931 if (result == os::SuspendResume::SR_RUNNING) {
3932 sr_semaphore.signal();
3933 break;
3934 }
3935 }
3936
3937 } else if (state == os::SuspendResume::SR_RUNNING) {
3938 // request was cancelled, continue
3939 } else {
3940 ShouldNotReachHere();
3941 }
3942
3943 resume_clear_context(osthread);
3944 } else if (current == os::SuspendResume::SR_RUNNING) {
3945 // request was cancelled, continue
3946 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3947 // ignore
3948 } else {
3949 // ignore
3950 }
3951
3952 errno = old_errno;
3953 }
3954
3955 static int SR_initialize() {
3956 struct sigaction act;
3957 char *s;
3958
3959 // Get signal number to use for suspend/resume
3960 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3961 int sig = ::strtol(s, 0, 10);
3962 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
3963 sig < NSIG) { // Must be legal signal and fit into sigflags[].
3964 SR_signum = sig;
3965 } else {
3966 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3967 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3968 }
3969 }
3970
3971 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3972 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3973
3974 sigemptyset(&SR_sigset);
3975 sigaddset(&SR_sigset, SR_signum);
3976
3977 // Set up signal handler for suspend/resume
3978 act.sa_flags = SA_RESTART|SA_SIGINFO;
3979 act.sa_handler = (void (*)(int)) SR_handler;
3980
3981 // SR_signum is blocked by default.
3982 // 4528190 - We also need to block pthread restart signal (32 on all
3983 // supported Linux platforms). Note that LinuxThreads need to block
3984 // this signal for all threads to work properly. So we don't have
3985 // to use hard-coded signal number when setting up the mask.
3986 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3987
3988 if (sigaction(SR_signum, &act, 0) == -1) {
3989 return -1;
3990 }
3991
3992 // Save signal flag
3993 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3994 return 0;
3995 }
3996
3997 static int sr_notify(OSThread* osthread) {
3998 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3999 assert_status(status == 0, status, "pthread_kill");
4000 return status;
4001 }
4002
4003 // "Randomly" selected value for how long we want to spin
4004 // before bailing out on suspending a thread, also how often
4005 // we send a signal to a thread we want to resume
4006 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4007 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4008
4009 // returns true on success and false on error - really an error is fatal
4010 // but this seems the normal response to library errors
4011 static bool do_suspend(OSThread* osthread) {
4012 assert(osthread->sr.is_running(), "thread should be running");
4013 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4014
4015 // mark as suspended and send signal
4016 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4017 // failed to switch, state wasn't running?
4018 ShouldNotReachHere();
4019 return false;
4020 }
4021
4022 if (sr_notify(osthread) != 0) {
4023 ShouldNotReachHere();
4024 }
4025
4026 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4027 while (true) {
4028 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4029 break;
4030 } else {
4031 // timeout
4032 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4033 if (cancelled == os::SuspendResume::SR_RUNNING) {
4034 return false;
4035 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4036 // make sure that we consume the signal on the semaphore as well
4037 sr_semaphore.wait();
4038 break;
4039 } else {
4040 ShouldNotReachHere();
4041 return false;
4042 }
4043 }
4044 }
4045
4046 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4047 return true;
4048 }
4049
4050 static void do_resume(OSThread* osthread) {
4051 assert(osthread->sr.is_suspended(), "thread should be suspended");
4052 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4053
4054 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4055 // failed to switch to WAKEUP_REQUEST
4056 ShouldNotReachHere();
4057 return;
4058 }
4059
4060 while (true) {
4061 if (sr_notify(osthread) == 0) {
4062 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4063 if (osthread->sr.is_running()) {
4064 return;
4065 }
4066 }
4067 } else {
4068 ShouldNotReachHere();
4069 }
4070 }
4071
4072 guarantee(osthread->sr.is_running(), "Must be running!");
4073 }
4074
4075 ///////////////////////////////////////////////////////////////////////////////////
4076 // signal handling (except suspend/resume)
4077
4078 // This routine may be used by user applications as a "hook" to catch signals.
4079 // The user-defined signal handler must pass unrecognized signals to this
4080 // routine, and if it returns true (non-zero), then the signal handler must
4081 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4082 // routine will never retun false (zero), but instead will execute a VM panic
4083 // routine kill the process.
4084 //
4085 // If this routine returns false, it is OK to call it again. This allows
4086 // the user-defined signal handler to perform checks either before or after
4087 // the VM performs its own checks. Naturally, the user code would be making
4088 // a serious error if it tried to handle an exception (such as a null check
4089 // or breakpoint) that the VM was generating for its own correct operation.
4090 //
4091 // This routine may recognize any of the following kinds of signals:
4092 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4093 // It should be consulted by handlers for any of those signals.
4094 //
4095 // The caller of this routine must pass in the three arguments supplied
4096 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4097 // field of the structure passed to sigaction(). This routine assumes that
4098 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4099 //
4100 // Note that the VM will print warnings if it detects conflicting signal
4101 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4102 //
4103 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4104 siginfo_t* siginfo,
4105 void* ucontext,
4106 int abort_if_unrecognized);
4107
4108 void signalHandler(int sig, siginfo_t* info, void* uc) {
4109 assert(info != NULL && uc != NULL, "it must be old kernel");
4110 int orig_errno = errno; // Preserve errno value over signal handler.
4111 JVM_handle_linux_signal(sig, info, uc, true);
4112 errno = orig_errno;
4113 }
4114
4115
4116 // This boolean allows users to forward their own non-matching signals
4117 // to JVM_handle_linux_signal, harmlessly.
4118 bool os::Linux::signal_handlers_are_installed = false;
4119
4120 // For signal-chaining
4121 struct sigaction sigact[NSIG];
4122 uint64_t sigs = 0;
4123 #if (64 < NSIG-1)
4124 #error "Not all signals can be encoded in sigs. Adapt its type!"
4125 #endif
4126 bool os::Linux::libjsig_is_loaded = false;
4127 typedef struct sigaction *(*get_signal_t)(int);
4128 get_signal_t os::Linux::get_signal_action = NULL;
4129
4130 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4131 struct sigaction *actp = NULL;
4132
4133 if (libjsig_is_loaded) {
4134 // Retrieve the old signal handler from libjsig
4135 actp = (*get_signal_action)(sig);
4136 }
4137 if (actp == NULL) {
4138 // Retrieve the preinstalled signal handler from jvm
4139 actp = get_preinstalled_handler(sig);
4140 }
4141
4142 return actp;
4143 }
4144
4145 static bool call_chained_handler(struct sigaction *actp, int sig,
4146 siginfo_t *siginfo, void *context) {
4147 // Call the old signal handler
4148 if (actp->sa_handler == SIG_DFL) {
4149 // It's more reasonable to let jvm treat it as an unexpected exception
4150 // instead of taking the default action.
4151 return false;
4152 } else if (actp->sa_handler != SIG_IGN) {
4153 if ((actp->sa_flags & SA_NODEFER) == 0) {
4154 // automaticlly block the signal
4155 sigaddset(&(actp->sa_mask), sig);
4156 }
4157
4158 sa_handler_t hand = NULL;
4159 sa_sigaction_t sa = NULL;
4160 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4161 // retrieve the chained handler
4162 if (siginfo_flag_set) {
4163 sa = actp->sa_sigaction;
4164 } else {
4165 hand = actp->sa_handler;
4166 }
4167
4168 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4169 actp->sa_handler = SIG_DFL;
4170 }
4171
4172 // try to honor the signal mask
4173 sigset_t oset;
4174 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4175
4176 // call into the chained handler
4177 if (siginfo_flag_set) {
4178 (*sa)(sig, siginfo, context);
4179 } else {
4180 (*hand)(sig);
4181 }
4182
4183 // restore the signal mask
4184 pthread_sigmask(SIG_SETMASK, &oset, 0);
4185 }
4186 // Tell jvm's signal handler the signal is taken care of.
4187 return true;
4188 }
4189
4190 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4191 bool chained = false;
4192 // signal-chaining
4193 if (UseSignalChaining) {
4194 struct sigaction *actp = get_chained_signal_action(sig);
4195 if (actp != NULL) {
4196 chained = call_chained_handler(actp, sig, siginfo, context);
4197 }
4198 }
4199 return chained;
4200 }
4201
4202 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4203 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4204 return &sigact[sig];
4205 }
4206 return NULL;
4207 }
4208
4209 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4210 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4211 sigact[sig] = oldAct;
4212 sigs |= (uint64_t)1 << (sig-1);
4213 }
4214
4215 // for diagnostic
4216 int sigflags[NSIG];
4217
4218 int os::Linux::get_our_sigflags(int sig) {
4219 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4220 return sigflags[sig];
4221 }
4222
4223 void os::Linux::set_our_sigflags(int sig, int flags) {
4224 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4225 if (sig > 0 && sig < NSIG) {
4226 sigflags[sig] = flags;
4227 }
4228 }
4229
4230 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4231 // Check for overwrite.
4232 struct sigaction oldAct;
4233 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4234
4235 void* oldhand = oldAct.sa_sigaction
4236 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4237 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4238 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4239 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4240 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4241 if (AllowUserSignalHandlers || !set_installed) {
4242 // Do not overwrite; user takes responsibility to forward to us.
4243 return;
4244 } else if (UseSignalChaining) {
4245 // save the old handler in jvm
4246 save_preinstalled_handler(sig, oldAct);
4247 // libjsig also interposes the sigaction() call below and saves the
4248 // old sigaction on it own.
4249 } else {
4250 fatal("Encountered unexpected pre-existing sigaction handler "
4251 "%#lx for signal %d.", (long)oldhand, sig);
4252 }
4253 }
4254
4255 struct sigaction sigAct;
4256 sigfillset(&(sigAct.sa_mask));
4257 sigAct.sa_handler = SIG_DFL;
4258 if (!set_installed) {
4259 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4260 } else {
4261 sigAct.sa_sigaction = signalHandler;
4262 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4263 }
4264 // Save flags, which are set by ours
4265 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4266 sigflags[sig] = sigAct.sa_flags;
4267
4268 int ret = sigaction(sig, &sigAct, &oldAct);
4269 assert(ret == 0, "check");
4270
4271 void* oldhand2 = oldAct.sa_sigaction
4272 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4273 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4274 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4275 }
4276
4277 // install signal handlers for signals that HotSpot needs to
4278 // handle in order to support Java-level exception handling.
4279
4280 void os::Linux::install_signal_handlers() {
4281 if (!signal_handlers_are_installed) {
4282 signal_handlers_are_installed = true;
4283
4284 // signal-chaining
4285 typedef void (*signal_setting_t)();
4286 signal_setting_t begin_signal_setting = NULL;
4287 signal_setting_t end_signal_setting = NULL;
4288 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4289 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4290 if (begin_signal_setting != NULL) {
4291 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4292 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4293 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4294 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4295 libjsig_is_loaded = true;
4296 assert(UseSignalChaining, "should enable signal-chaining");
4297 }
4298 if (libjsig_is_loaded) {
4299 // Tell libjsig jvm is setting signal handlers
4300 (*begin_signal_setting)();
4301 }
4302
4303 set_signal_handler(SIGSEGV, true);
4304 set_signal_handler(SIGPIPE, true);
4305 set_signal_handler(SIGBUS, true);
4306 set_signal_handler(SIGILL, true);
4307 set_signal_handler(SIGFPE, true);
4308 #if defined(PPC64)
4309 set_signal_handler(SIGTRAP, true);
4310 #endif
4311 set_signal_handler(SIGXFSZ, true);
4312
4313 if (libjsig_is_loaded) {
4314 // Tell libjsig jvm finishes setting signal handlers
4315 (*end_signal_setting)();
4316 }
4317
4318 // We don't activate signal checker if libjsig is in place, we trust ourselves
4319 // and if UserSignalHandler is installed all bets are off.
4320 // Log that signal checking is off only if -verbose:jni is specified.
4321 if (CheckJNICalls) {
4322 if (libjsig_is_loaded) {
4323 if (PrintJNIResolving) {
4324 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4325 }
4326 check_signals = false;
4327 }
4328 if (AllowUserSignalHandlers) {
4329 if (PrintJNIResolving) {
4330 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4331 }
4332 check_signals = false;
4333 }
4334 }
4335 }
4336 }
4337
4338 // This is the fastest way to get thread cpu time on Linux.
4339 // Returns cpu time (user+sys) for any thread, not only for current.
4340 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4341 // It might work on 2.6.10+ with a special kernel/glibc patch.
4342 // For reference, please, see IEEE Std 1003.1-2004:
4343 // http://www.unix.org/single_unix_specification
4344
4345 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4346 struct timespec tp;
4347 int rc = os::Linux::clock_gettime(clockid, &tp);
4348 assert(rc == 0, "clock_gettime is expected to return 0 code");
4349
4350 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4351 }
4352
4353 /////
4354 // glibc on Linux platform uses non-documented flag
4355 // to indicate, that some special sort of signal
4356 // trampoline is used.
4357 // We will never set this flag, and we should
4358 // ignore this flag in our diagnostic
4359 #ifdef SIGNIFICANT_SIGNAL_MASK
4360 #undef SIGNIFICANT_SIGNAL_MASK
4361 #endif
4362 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4363
4364 static const char* get_signal_handler_name(address handler,
4365 char* buf, int buflen) {
4366 int offset = 0;
4367 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4368 if (found) {
4369 // skip directory names
4370 const char *p1, *p2;
4371 p1 = buf;
4372 size_t len = strlen(os::file_separator());
4373 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4374 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4375 } else {
4376 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4377 }
4378 return buf;
4379 }
4380
4381 static void print_signal_handler(outputStream* st, int sig,
4382 char* buf, size_t buflen) {
4383 struct sigaction sa;
4384
4385 sigaction(sig, NULL, &sa);
4386
4387 // See comment for SIGNIFICANT_SIGNAL_MASK define
4388 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4389
4390 st->print("%s: ", os::exception_name(sig, buf, buflen));
4391
4392 address handler = (sa.sa_flags & SA_SIGINFO)
4393 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4394 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4395
4396 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4397 st->print("SIG_DFL");
4398 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4399 st->print("SIG_IGN");
4400 } else {
4401 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4402 }
4403
4404 st->print(", sa_mask[0]=");
4405 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4406
4407 address rh = VMError::get_resetted_sighandler(sig);
4408 // May be, handler was resetted by VMError?
4409 if (rh != NULL) {
4410 handler = rh;
4411 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4412 }
4413
4414 st->print(", sa_flags=");
4415 os::Posix::print_sa_flags(st, sa.sa_flags);
4416
4417 // Check: is it our handler?
4418 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4419 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4420 // It is our signal handler
4421 // check for flags, reset system-used one!
4422 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4423 st->print(
4424 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4425 os::Linux::get_our_sigflags(sig));
4426 }
4427 }
4428 st->cr();
4429 }
4430
4431
4432 #define DO_SIGNAL_CHECK(sig) \
4433 do { \
4434 if (!sigismember(&check_signal_done, sig)) { \
4435 os::Linux::check_signal_handler(sig); \
4436 } \
4437 } while (0)
4438
4439 // This method is a periodic task to check for misbehaving JNI applications
4440 // under CheckJNI, we can add any periodic checks here
4441
4442 void os::run_periodic_checks() {
4443 if (check_signals == false) return;
4444
4445 // SEGV and BUS if overridden could potentially prevent
4446 // generation of hs*.log in the event of a crash, debugging
4447 // such a case can be very challenging, so we absolutely
4448 // check the following for a good measure:
4449 DO_SIGNAL_CHECK(SIGSEGV);
4450 DO_SIGNAL_CHECK(SIGILL);
4451 DO_SIGNAL_CHECK(SIGFPE);
4452 DO_SIGNAL_CHECK(SIGBUS);
4453 DO_SIGNAL_CHECK(SIGPIPE);
4454 DO_SIGNAL_CHECK(SIGXFSZ);
4455 #if defined(PPC64)
4456 DO_SIGNAL_CHECK(SIGTRAP);
4457 #endif
4458
4459 // ReduceSignalUsage allows the user to override these handlers
4460 // see comments at the very top and jvm_solaris.h
4461 if (!ReduceSignalUsage) {
4462 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4463 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4464 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4465 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4466 }
4467
4468 DO_SIGNAL_CHECK(SR_signum);
4469 }
4470
4471 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4472
4473 static os_sigaction_t os_sigaction = NULL;
4474
4475 void os::Linux::check_signal_handler(int sig) {
4476 char buf[O_BUFLEN];
4477 address jvmHandler = NULL;
4478
4479
4480 struct sigaction act;
4481 if (os_sigaction == NULL) {
4482 // only trust the default sigaction, in case it has been interposed
4483 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4484 if (os_sigaction == NULL) return;
4485 }
4486
4487 os_sigaction(sig, (struct sigaction*)NULL, &act);
4488
4489
4490 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4491
4492 address thisHandler = (act.sa_flags & SA_SIGINFO)
4493 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4494 : CAST_FROM_FN_PTR(address, act.sa_handler);
4495
4496
4497 switch (sig) {
4498 case SIGSEGV:
4499 case SIGBUS:
4500 case SIGFPE:
4501 case SIGPIPE:
4502 case SIGILL:
4503 case SIGXFSZ:
4504 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4505 break;
4506
4507 case SHUTDOWN1_SIGNAL:
4508 case SHUTDOWN2_SIGNAL:
4509 case SHUTDOWN3_SIGNAL:
4510 case BREAK_SIGNAL:
4511 jvmHandler = (address)user_handler();
4512 break;
4513
4514 default:
4515 if (sig == SR_signum) {
4516 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4517 } else {
4518 return;
4519 }
4520 break;
4521 }
4522
4523 if (thisHandler != jvmHandler) {
4524 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4525 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4526 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4527 // No need to check this sig any longer
4528 sigaddset(&check_signal_done, sig);
4529 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4530 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4531 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4532 exception_name(sig, buf, O_BUFLEN));
4533 }
4534 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4535 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4536 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4537 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4538 // No need to check this sig any longer
4539 sigaddset(&check_signal_done, sig);
4540 }
4541
4542 // Dump all the signal
4543 if (sigismember(&check_signal_done, sig)) {
4544 print_signal_handlers(tty, buf, O_BUFLEN);
4545 }
4546 }
4547
4548 extern void report_error(char* file_name, int line_no, char* title,
4549 char* format, ...);
4550
4551 extern bool signal_name(int signo, char* buf, size_t len);
4552
4553 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4554 if (0 < exception_code && exception_code <= SIGRTMAX) {
4555 // signal
4556 if (!signal_name(exception_code, buf, size)) {
4557 jio_snprintf(buf, size, "SIG%d", exception_code);
4558 }
4559 return buf;
4560 } else {
4561 return NULL;
4562 }
4563 }
4564
4565 // this is called _before_ the most of global arguments have been parsed
4566 void os::init(void) {
4567 char dummy; // used to get a guess on initial stack address
4568 // first_hrtime = gethrtime();
4569
4570 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4571
4572 init_random(1234567);
4573
4574 ThreadCritical::initialize();
4575
4576 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4577 if (Linux::page_size() == -1) {
4578 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4579 strerror(errno));
4580 }
4581 init_page_sizes((size_t) Linux::page_size());
4582
4583 Linux::initialize_system_info();
4584
4585 // main_thread points to the aboriginal thread
4586 Linux::_main_thread = pthread_self();
4587
4588 Linux::clock_init();
4589 initial_time_count = javaTimeNanos();
4590
4591 // pthread_condattr initialization for monotonic clock
4592 int status;
4593 pthread_condattr_t* _condattr = os::Linux::condAttr();
4594 if ((status = pthread_condattr_init(_condattr)) != 0) {
4595 fatal("pthread_condattr_init: %s", strerror(status));
4596 }
4597 // Only set the clock if CLOCK_MONOTONIC is available
4598 if (os::supports_monotonic_clock()) {
4599 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4600 if (status == EINVAL) {
4601 warning("Unable to use monotonic clock with relative timed-waits" \
4602 " - changes to the time-of-day clock may have adverse affects");
4603 } else {
4604 fatal("pthread_condattr_setclock: %s", strerror(status));
4605 }
4606 }
4607 }
4608 // else it defaults to CLOCK_REALTIME
4609
4610 // If the pagesize of the VM is greater than 8K determine the appropriate
4611 // number of initial guard pages. The user can change this with the
4612 // command line arguments, if needed.
4613 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4614 StackYellowPages = 1;
4615 StackRedPages = 1;
4616 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4617 }
4618
4619 // retrieve entry point for pthread_setname_np
4620 Linux::_pthread_setname_np =
4621 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4622
4623 }
4624
4625 // To install functions for atexit system call
4626 extern "C" {
4627 static void perfMemory_exit_helper() {
4628 perfMemory_exit();
4629 }
4630 }
4631
4632 // this is called _after_ the global arguments have been parsed
4633 jint os::init_2(void) {
4634 Linux::fast_thread_clock_init();
4635
4636 // Allocate a single page and mark it as readable for safepoint polling
4637 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4638 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4639
4640 os::set_polling_page(polling_page);
4641
4642 #ifndef PRODUCT
4643 if (Verbose && PrintMiscellaneous) {
4644 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4645 (intptr_t)polling_page);
4646 }
4647 #endif
4648
4649 if (!UseMembar) {
4650 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4651 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4652 os::set_memory_serialize_page(mem_serialize_page);
4653
4654 #ifndef PRODUCT
4655 if (Verbose && PrintMiscellaneous) {
4656 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4657 (intptr_t)mem_serialize_page);
4658 }
4659 #endif
4660 }
4661
4662 // initialize suspend/resume support - must do this before signal_sets_init()
4663 if (SR_initialize() != 0) {
4664 perror("SR_initialize failed");
4665 return JNI_ERR;
4666 }
4667
4668 Linux::signal_sets_init();
4669 Linux::install_signal_handlers();
4670
4671 // Check minimum allowable stack size for thread creation and to initialize
4672 // the java system classes, including StackOverflowError - depends on page
4673 // size. Add a page for compiler2 recursion in main thread.
4674 // Add in 2*BytesPerWord times page size to account for VM stack during
4675 // class initialization depending on 32 or 64 bit VM.
4676 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4677 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4678 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4679
4680 size_t threadStackSizeInBytes = ThreadStackSize * K;
4681 if (threadStackSizeInBytes != 0 &&
4682 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4683 tty->print_cr("\nThe stack size specified is too small, "
4684 "Specify at least " SIZE_FORMAT "k",
4685 os::Linux::min_stack_allowed/ K);
4686 return JNI_ERR;
4687 }
4688
4689 // Make the stack size a multiple of the page size so that
4690 // the yellow/red zones can be guarded.
4691 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4692 vm_page_size()));
4693
4694 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4695
4696 #if defined(IA32)
4697 workaround_expand_exec_shield_cs_limit();
4698 #endif
4699
4700 Linux::libpthread_init();
4701 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4702 tty->print_cr("[HotSpot is running with %s, %s]\n",
4703 Linux::glibc_version(), Linux::libpthread_version());
4704 }
4705
4706 if (UseNUMA) {
4707 if (!Linux::libnuma_init()) {
4708 UseNUMA = false;
4709 } else {
4710 if ((Linux::numa_max_node() < 1)) {
4711 // There's only one node(they start from 0), disable NUMA.
4712 UseNUMA = false;
4713 }
4714 }
4715 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4716 // we can make the adaptive lgrp chunk resizing work. If the user specified
4717 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4718 // disable adaptive resizing.
4719 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4720 if (FLAG_IS_DEFAULT(UseNUMA)) {
4721 UseNUMA = false;
4722 } else {
4723 if (FLAG_IS_DEFAULT(UseLargePages) &&
4724 FLAG_IS_DEFAULT(UseSHM) &&
4725 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4726 UseLargePages = false;
4727 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4728 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4729 UseAdaptiveSizePolicy = false;
4730 UseAdaptiveNUMAChunkSizing = false;
4731 }
4732 }
4733 }
4734 if (!UseNUMA && ForceNUMA) {
4735 UseNUMA = true;
4736 }
4737 }
4738
4739 if (MaxFDLimit) {
4740 // set the number of file descriptors to max. print out error
4741 // if getrlimit/setrlimit fails but continue regardless.
4742 struct rlimit nbr_files;
4743 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4744 if (status != 0) {
4745 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4746 perror("os::init_2 getrlimit failed");
4747 }
4748 } else {
4749 nbr_files.rlim_cur = nbr_files.rlim_max;
4750 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4751 if (status != 0) {
4752 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4753 perror("os::init_2 setrlimit failed");
4754 }
4755 }
4756 }
4757 }
4758
4759 // Initialize lock used to serialize thread creation (see os::create_thread)
4760 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4761
4762 // at-exit methods are called in the reverse order of their registration.
4763 // atexit functions are called on return from main or as a result of a
4764 // call to exit(3C). There can be only 32 of these functions registered
4765 // and atexit() does not set errno.
4766
4767 if (PerfAllowAtExitRegistration) {
4768 // only register atexit functions if PerfAllowAtExitRegistration is set.
4769 // atexit functions can be delayed until process exit time, which
4770 // can be problematic for embedded VM situations. Embedded VMs should
4771 // call DestroyJavaVM() to assure that VM resources are released.
4772
4773 // note: perfMemory_exit_helper atexit function may be removed in
4774 // the future if the appropriate cleanup code can be added to the
4775 // VM_Exit VMOperation's doit method.
4776 if (atexit(perfMemory_exit_helper) != 0) {
4777 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4778 }
4779 }
4780
4781 // initialize thread priority policy
4782 prio_init();
4783
4784 return JNI_OK;
4785 }
4786
4787 // Mark the polling page as unreadable
4788 void os::make_polling_page_unreadable(void) {
4789 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4790 fatal("Could not disable polling page");
4791 }
4792 }
4793
4794 // Mark the polling page as readable
4795 void os::make_polling_page_readable(void) {
4796 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4797 fatal("Could not enable polling page");
4798 }
4799 }
4800
4801 int os::active_processor_count() {
4802 // Linux doesn't yet have a (official) notion of processor sets,
4803 // so just return the number of online processors.
4804 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4805 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4806 return online_cpus;
4807 }
4808
4809 void os::set_native_thread_name(const char *name) {
4810 if (Linux::_pthread_setname_np) {
4811 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4812 snprintf(buf, sizeof(buf), "%s", name);
4813 buf[sizeof(buf) - 1] = '\0';
4814 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4815 // ERANGE should not happen; all other errors should just be ignored.
4816 assert(rc != ERANGE, "pthread_setname_np failed");
4817 }
4818 }
4819
4820 bool os::distribute_processes(uint length, uint* distribution) {
4821 // Not yet implemented.
4822 return false;
4823 }
4824
4825 bool os::bind_to_processor(uint processor_id) {
4826 // Not yet implemented.
4827 return false;
4828 }
4829
4830 ///
4831
4832 void os::SuspendedThreadTask::internal_do_task() {
4833 if (do_suspend(_thread->osthread())) {
4834 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4835 do_task(context);
4836 do_resume(_thread->osthread());
4837 }
4838 }
4839
4840 class PcFetcher : public os::SuspendedThreadTask {
4841 public:
4842 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4843 ExtendedPC result();
4844 protected:
4845 void do_task(const os::SuspendedThreadTaskContext& context);
4846 private:
4847 ExtendedPC _epc;
4848 };
4849
4850 ExtendedPC PcFetcher::result() {
4851 guarantee(is_done(), "task is not done yet.");
4852 return _epc;
4853 }
4854
4855 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4856 Thread* thread = context.thread();
4857 OSThread* osthread = thread->osthread();
4858 if (osthread->ucontext() != NULL) {
4859 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4860 } else {
4861 // NULL context is unexpected, double-check this is the VMThread
4862 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4863 }
4864 }
4865
4866 // Suspends the target using the signal mechanism and then grabs the PC before
4867 // resuming the target. Used by the flat-profiler only
4868 ExtendedPC os::get_thread_pc(Thread* thread) {
4869 // Make sure that it is called by the watcher for the VMThread
4870 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4871 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4872
4873 PcFetcher fetcher(thread);
4874 fetcher.run();
4875 return fetcher.result();
4876 }
4877
4878 ////////////////////////////////////////////////////////////////////////////////
4879 // debug support
4880
4881 bool os::find(address addr, outputStream* st) {
4882 Dl_info dlinfo;
4883 memset(&dlinfo, 0, sizeof(dlinfo));
4884 if (dladdr(addr, &dlinfo) != 0) {
4885 st->print(PTR_FORMAT ": ", p2i(addr));
4886 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4887 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4888 p2i(addr) - p2i(dlinfo.dli_saddr));
4889 } else if (dlinfo.dli_fbase != NULL) {
4890 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4891 } else {
4892 st->print("<absolute address>");
4893 }
4894 if (dlinfo.dli_fname != NULL) {
4895 st->print(" in %s", dlinfo.dli_fname);
4896 }
4897 if (dlinfo.dli_fbase != NULL) {
4898 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4899 }
4900 st->cr();
4901
4902 if (Verbose) {
4903 // decode some bytes around the PC
4904 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4905 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4906 address lowest = (address) dlinfo.dli_sname;
4907 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4908 if (begin < lowest) begin = lowest;
4909 Dl_info dlinfo2;
4910 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4911 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4912 end = (address) dlinfo2.dli_saddr;
4913 }
4914 Disassembler::decode(begin, end, st);
4915 }
4916 return true;
4917 }
4918 return false;
4919 }
4920
4921 ////////////////////////////////////////////////////////////////////////////////
4922 // misc
4923
4924 // This does not do anything on Linux. This is basically a hook for being
4925 // able to use structured exception handling (thread-local exception filters)
4926 // on, e.g., Win32.
4927 void
4928 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
4929 JavaCallArguments* args, Thread* thread) {
4930 f(value, method, args, thread);
4931 }
4932
4933 void os::print_statistics() {
4934 }
4935
4936 int os::message_box(const char* title, const char* message) {
4937 int i;
4938 fdStream err(defaultStream::error_fd());
4939 for (i = 0; i < 78; i++) err.print_raw("=");
4940 err.cr();
4941 err.print_raw_cr(title);
4942 for (i = 0; i < 78; i++) err.print_raw("-");
4943 err.cr();
4944 err.print_raw_cr(message);
4945 for (i = 0; i < 78; i++) err.print_raw("=");
4946 err.cr();
4947
4948 char buf[16];
4949 // Prevent process from exiting upon "read error" without consuming all CPU
4950 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4951
4952 return buf[0] == 'y' || buf[0] == 'Y';
4953 }
4954
4955 int os::stat(const char *path, struct stat *sbuf) {
4956 char pathbuf[MAX_PATH];
4957 if (strlen(path) > MAX_PATH - 1) {
4958 errno = ENAMETOOLONG;
4959 return -1;
4960 }
4961 os::native_path(strcpy(pathbuf, path));
4962 return ::stat(pathbuf, sbuf);
4963 }
4964
4965 bool os::check_heap(bool force) {
4966 return true;
4967 }
4968
4969 // Is a (classpath) directory empty?
4970 bool os::dir_is_empty(const char* path) {
4971 DIR *dir = NULL;
4972 struct dirent *ptr;
4973
4974 dir = opendir(path);
4975 if (dir == NULL) return true;
4976
4977 // Scan the directory
4978 bool result = true;
4979 char buf[sizeof(struct dirent) + MAX_PATH];
4980 while (result && (ptr = ::readdir(dir)) != NULL) {
4981 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4982 result = false;
4983 }
4984 }
4985 closedir(dir);
4986 return result;
4987 }
4988
4989 // This code originates from JDK's sysOpen and open64_w
4990 // from src/solaris/hpi/src/system_md.c
4991
4992 int os::open(const char *path, int oflag, int mode) {
4993 if (strlen(path) > MAX_PATH - 1) {
4994 errno = ENAMETOOLONG;
4995 return -1;
4996 }
4997
4998 // All file descriptors that are opened in the Java process and not
4999 // specifically destined for a subprocess should have the close-on-exec
5000 // flag set. If we don't set it, then careless 3rd party native code
5001 // might fork and exec without closing all appropriate file descriptors
5002 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5003 // turn might:
5004 //
5005 // - cause end-of-file to fail to be detected on some file
5006 // descriptors, resulting in mysterious hangs, or
5007 //
5008 // - might cause an fopen in the subprocess to fail on a system
5009 // suffering from bug 1085341.
5010 //
5011 // (Yes, the default setting of the close-on-exec flag is a Unix
5012 // design flaw)
5013 //
5014 // See:
5015 // 1085341: 32-bit stdio routines should support file descriptors >255
5016 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5017 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5018 //
5019 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5020 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5021 // because it saves a system call and removes a small window where the flag
5022 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5023 // and we fall back to using FD_CLOEXEC (see below).
5024 #ifdef O_CLOEXEC
5025 oflag |= O_CLOEXEC;
5026 #endif
5027
5028 int fd = ::open64(path, oflag, mode);
5029 if (fd == -1) return -1;
5030
5031 //If the open succeeded, the file might still be a directory
5032 {
5033 struct stat64 buf64;
5034 int ret = ::fstat64(fd, &buf64);
5035 int st_mode = buf64.st_mode;
5036
5037 if (ret != -1) {
5038 if ((st_mode & S_IFMT) == S_IFDIR) {
5039 errno = EISDIR;
5040 ::close(fd);
5041 return -1;
5042 }
5043 } else {
5044 ::close(fd);
5045 return -1;
5046 }
5047 }
5048
5049 #ifdef FD_CLOEXEC
5050 // Validate that the use of the O_CLOEXEC flag on open above worked.
5051 // With recent kernels, we will perform this check exactly once.
5052 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5053 if (!O_CLOEXEC_is_known_to_work) {
5054 int flags = ::fcntl(fd, F_GETFD);
5055 if (flags != -1) {
5056 if ((flags & FD_CLOEXEC) != 0)
5057 O_CLOEXEC_is_known_to_work = 1;
5058 else
5059 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5060 }
5061 }
5062 #endif
5063
5064 return fd;
5065 }
5066
5067
5068 // create binary file, rewriting existing file if required
5069 int os::create_binary_file(const char* path, bool rewrite_existing) {
5070 int oflags = O_WRONLY | O_CREAT;
5071 if (!rewrite_existing) {
5072 oflags |= O_EXCL;
5073 }
5074 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5075 }
5076
5077 // return current position of file pointer
5078 jlong os::current_file_offset(int fd) {
5079 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5080 }
5081
5082 // move file pointer to the specified offset
5083 jlong os::seek_to_file_offset(int fd, jlong offset) {
5084 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5085 }
5086
5087 // This code originates from JDK's sysAvailable
5088 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5089
5090 int os::available(int fd, jlong *bytes) {
5091 jlong cur, end;
5092 int mode;
5093 struct stat64 buf64;
5094
5095 if (::fstat64(fd, &buf64) >= 0) {
5096 mode = buf64.st_mode;
5097 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5098 int n;
5099 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5100 *bytes = n;
5101 return 1;
5102 }
5103 }
5104 }
5105 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5106 return 0;
5107 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5108 return 0;
5109 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5110 return 0;
5111 }
5112 *bytes = end - cur;
5113 return 1;
5114 }
5115
5116 // Map a block of memory.
5117 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5118 char *addr, size_t bytes, bool read_only,
5119 bool allow_exec) {
5120 int prot;
5121 int flags = MAP_PRIVATE;
5122
5123 if (read_only) {
5124 prot = PROT_READ;
5125 } else {
5126 prot = PROT_READ | PROT_WRITE;
5127 }
5128
5129 if (allow_exec) {
5130 prot |= PROT_EXEC;
5131 }
5132
5133 if (addr != NULL) {
5134 flags |= MAP_FIXED;
5135 }
5136
5137 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5138 fd, file_offset);
5139 if (mapped_address == MAP_FAILED) {
5140 return NULL;
5141 }
5142 return mapped_address;
5143 }
5144
5145
5146 // Remap a block of memory.
5147 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5148 char *addr, size_t bytes, bool read_only,
5149 bool allow_exec) {
5150 // same as map_memory() on this OS
5151 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5152 allow_exec);
5153 }
5154
5155
5156 // Unmap a block of memory.
5157 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5158 return munmap(addr, bytes) == 0;
5159 }
5160
5161 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5162
5163 static clockid_t thread_cpu_clockid(Thread* thread) {
5164 pthread_t tid = thread->osthread()->pthread_id();
5165 clockid_t clockid;
5166
5167 // Get thread clockid
5168 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5169 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5170 return clockid;
5171 }
5172
5173 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5174 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5175 // of a thread.
5176 //
5177 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5178 // the fast estimate available on the platform.
5179
5180 jlong os::current_thread_cpu_time() {
5181 if (os::Linux::supports_fast_thread_cpu_time()) {
5182 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5183 } else {
5184 // return user + sys since the cost is the same
5185 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5186 }
5187 }
5188
5189 jlong os::thread_cpu_time(Thread* thread) {
5190 // consistent with what current_thread_cpu_time() returns
5191 if (os::Linux::supports_fast_thread_cpu_time()) {
5192 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5193 } else {
5194 return slow_thread_cpu_time(thread, true /* user + sys */);
5195 }
5196 }
5197
5198 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5199 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5200 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5201 } else {
5202 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5203 }
5204 }
5205
5206 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5207 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5208 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5209 } else {
5210 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5211 }
5212 }
5213
5214 // -1 on error.
5215 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5216 pid_t tid = thread->osthread()->thread_id();
5217 char *s;
5218 char stat[2048];
5219 int statlen;
5220 char proc_name[64];
5221 int count;
5222 long sys_time, user_time;
5223 char cdummy;
5224 int idummy;
5225 long ldummy;
5226 FILE *fp;
5227
5228 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5229 fp = fopen(proc_name, "r");
5230 if (fp == NULL) return -1;
5231 statlen = fread(stat, 1, 2047, fp);
5232 stat[statlen] = '\0';
5233 fclose(fp);
5234
5235 // Skip pid and the command string. Note that we could be dealing with
5236 // weird command names, e.g. user could decide to rename java launcher
5237 // to "java 1.4.2 :)", then the stat file would look like
5238 // 1234 (java 1.4.2 :)) R ... ...
5239 // We don't really need to know the command string, just find the last
5240 // occurrence of ")" and then start parsing from there. See bug 4726580.
5241 s = strrchr(stat, ')');
5242 if (s == NULL) return -1;
5243
5244 // Skip blank chars
5245 do { s++; } while (s && isspace(*s));
5246
5247 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5248 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5249 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5250 &user_time, &sys_time);
5251 if (count != 13) return -1;
5252 if (user_sys_cpu_time) {
5253 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5254 } else {
5255 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5256 }
5257 }
5258
5259 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5260 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5261 info_ptr->may_skip_backward = false; // elapsed time not wall time
5262 info_ptr->may_skip_forward = false; // elapsed time not wall time
5263 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5264 }
5265
5266 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5267 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5268 info_ptr->may_skip_backward = false; // elapsed time not wall time
5269 info_ptr->may_skip_forward = false; // elapsed time not wall time
5270 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5271 }
5272
5273 bool os::is_thread_cpu_time_supported() {
5274 return true;
5275 }
5276
5277 // System loadavg support. Returns -1 if load average cannot be obtained.
5278 // Linux doesn't yet have a (official) notion of processor sets,
5279 // so just return the system wide load average.
5280 int os::loadavg(double loadavg[], int nelem) {
5281 return ::getloadavg(loadavg, nelem);
5282 }
5283
5284 void os::pause() {
5285 char filename[MAX_PATH];
5286 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5287 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5288 } else {
5289 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5290 }
5291
5292 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5293 if (fd != -1) {
5294 struct stat buf;
5295 ::close(fd);
5296 while (::stat(filename, &buf) == 0) {
5297 (void)::poll(NULL, 0, 100);
5298 }
5299 } else {
5300 jio_fprintf(stderr,
5301 "Could not open pause file '%s', continuing immediately.\n", filename);
5302 }
5303 }
5304
5305
5306 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5307 // comment paragraphs are worth repeating here:
5308 //
5309 // Assumption:
5310 // Only one parker can exist on an event, which is why we allocate
5311 // them per-thread. Multiple unparkers can coexist.
5312 //
5313 // _Event serves as a restricted-range semaphore.
5314 // -1 : thread is blocked, i.e. there is a waiter
5315 // 0 : neutral: thread is running or ready,
5316 // could have been signaled after a wait started
5317 // 1 : signaled - thread is running or ready
5318 //
5319 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5320 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5321 // For specifics regarding the bug see GLIBC BUGID 261237 :
5322 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5323 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5324 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5325 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5326 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5327 // and monitorenter when we're using 1-0 locking. All those operations may result in
5328 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5329 // of libpthread avoids the problem, but isn't practical.
5330 //
5331 // Possible remedies:
5332 //
5333 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5334 // This is palliative and probabilistic, however. If the thread is preempted
5335 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5336 // than the minimum period may have passed, and the abstime may be stale (in the
5337 // past) resultin in a hang. Using this technique reduces the odds of a hang
5338 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5339 //
5340 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5341 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5342 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5343 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5344 // thread.
5345 //
5346 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5347 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5348 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5349 // This also works well. In fact it avoids kernel-level scalability impediments
5350 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5351 // timers in a graceful fashion.
5352 //
5353 // 4. When the abstime value is in the past it appears that control returns
5354 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5355 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5356 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5357 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5358 // It may be possible to avoid reinitialization by checking the return
5359 // value from pthread_cond_timedwait(). In addition to reinitializing the
5360 // condvar we must establish the invariant that cond_signal() is only called
5361 // within critical sections protected by the adjunct mutex. This prevents
5362 // cond_signal() from "seeing" a condvar that's in the midst of being
5363 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5364 // desirable signal-after-unlock optimization that avoids futile context switching.
5365 //
5366 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5367 // structure when a condvar is used or initialized. cond_destroy() would
5368 // release the helper structure. Our reinitialize-after-timedwait fix
5369 // put excessive stress on malloc/free and locks protecting the c-heap.
5370 //
5371 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5372 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5373 // and only enabling the work-around for vulnerable environments.
5374
5375 // utility to compute the abstime argument to timedwait:
5376 // millis is the relative timeout time
5377 // abstime will be the absolute timeout time
5378 // TODO: replace compute_abstime() with unpackTime()
5379
5380 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5381 if (millis < 0) millis = 0;
5382
5383 jlong seconds = millis / 1000;
5384 millis %= 1000;
5385 if (seconds > 50000000) { // see man cond_timedwait(3T)
5386 seconds = 50000000;
5387 }
5388
5389 if (os::supports_monotonic_clock()) {
5390 struct timespec now;
5391 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5392 assert_status(status == 0, status, "clock_gettime");
5393 abstime->tv_sec = now.tv_sec + seconds;
5394 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5395 if (nanos >= NANOSECS_PER_SEC) {
5396 abstime->tv_sec += 1;
5397 nanos -= NANOSECS_PER_SEC;
5398 }
5399 abstime->tv_nsec = nanos;
5400 } else {
5401 struct timeval now;
5402 int status = gettimeofday(&now, NULL);
5403 assert(status == 0, "gettimeofday");
5404 abstime->tv_sec = now.tv_sec + seconds;
5405 long usec = now.tv_usec + millis * 1000;
5406 if (usec >= 1000000) {
5407 abstime->tv_sec += 1;
5408 usec -= 1000000;
5409 }
5410 abstime->tv_nsec = usec * 1000;
5411 }
5412 return abstime;
5413 }
5414
5415 void os::PlatformEvent::park() { // AKA "down()"
5416 // Transitions for _Event:
5417 // -1 => -1 : illegal
5418 // 1 => 0 : pass - return immediately
5419 // 0 => -1 : block; then set _Event to 0 before returning
5420
5421 // Invariant: Only the thread associated with the Event/PlatformEvent
5422 // may call park().
5423 // TODO: assert that _Assoc != NULL or _Assoc == Self
5424 assert(_nParked == 0, "invariant");
5425
5426 int v;
5427 for (;;) {
5428 v = _Event;
5429 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5430 }
5431 guarantee(v >= 0, "invariant");
5432 if (v == 0) {
5433 // Do this the hard way by blocking ...
5434 int status = pthread_mutex_lock(_mutex);
5435 assert_status(status == 0, status, "mutex_lock");
5436 guarantee(_nParked == 0, "invariant");
5437 ++_nParked;
5438 while (_Event < 0) {
5439 status = pthread_cond_wait(_cond, _mutex);
5440 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5441 // Treat this the same as if the wait was interrupted
5442 if (status == ETIME) { status = EINTR; }
5443 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5444 }
5445 --_nParked;
5446
5447 _Event = 0;
5448 status = pthread_mutex_unlock(_mutex);
5449 assert_status(status == 0, status, "mutex_unlock");
5450 // Paranoia to ensure our locked and lock-free paths interact
5451 // correctly with each other.
5452 OrderAccess::fence();
5453 }
5454 guarantee(_Event >= 0, "invariant");
5455 }
5456
5457 int os::PlatformEvent::park(jlong millis) {
5458 // Transitions for _Event:
5459 // -1 => -1 : illegal
5460 // 1 => 0 : pass - return immediately
5461 // 0 => -1 : block; then set _Event to 0 before returning
5462
5463 guarantee(_nParked == 0, "invariant");
5464
5465 int v;
5466 for (;;) {
5467 v = _Event;
5468 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5469 }
5470 guarantee(v >= 0, "invariant");
5471 if (v != 0) return OS_OK;
5472
5473 // We do this the hard way, by blocking the thread.
5474 // Consider enforcing a minimum timeout value.
5475 struct timespec abst;
5476 compute_abstime(&abst, millis);
5477
5478 int ret = OS_TIMEOUT;
5479 int status = pthread_mutex_lock(_mutex);
5480 assert_status(status == 0, status, "mutex_lock");
5481 guarantee(_nParked == 0, "invariant");
5482 ++_nParked;
5483
5484 // Object.wait(timo) will return because of
5485 // (a) notification
5486 // (b) timeout
5487 // (c) thread.interrupt
5488 //
5489 // Thread.interrupt and object.notify{All} both call Event::set.
5490 // That is, we treat thread.interrupt as a special case of notification.
5491 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5492 // We assume all ETIME returns are valid.
5493 //
5494 // TODO: properly differentiate simultaneous notify+interrupt.
5495 // In that case, we should propagate the notify to another waiter.
5496
5497 while (_Event < 0) {
5498 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5499 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5500 pthread_cond_destroy(_cond);
5501 pthread_cond_init(_cond, os::Linux::condAttr());
5502 }
5503 assert_status(status == 0 || status == EINTR ||
5504 status == ETIME || status == ETIMEDOUT,
5505 status, "cond_timedwait");
5506 if (!FilterSpuriousWakeups) break; // previous semantics
5507 if (status == ETIME || status == ETIMEDOUT) break;
5508 // We consume and ignore EINTR and spurious wakeups.
5509 }
5510 --_nParked;
5511 if (_Event >= 0) {
5512 ret = OS_OK;
5513 }
5514 _Event = 0;
5515 status = pthread_mutex_unlock(_mutex);
5516 assert_status(status == 0, status, "mutex_unlock");
5517 assert(_nParked == 0, "invariant");
5518 // Paranoia to ensure our locked and lock-free paths interact
5519 // correctly with each other.
5520 OrderAccess::fence();
5521 return ret;
5522 }
5523
5524 void os::PlatformEvent::unpark() {
5525 // Transitions for _Event:
5526 // 0 => 1 : just return
5527 // 1 => 1 : just return
5528 // -1 => either 0 or 1; must signal target thread
5529 // That is, we can safely transition _Event from -1 to either
5530 // 0 or 1.
5531 // See also: "Semaphores in Plan 9" by Mullender & Cox
5532 //
5533 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5534 // that it will take two back-to-back park() calls for the owning
5535 // thread to block. This has the benefit of forcing a spurious return
5536 // from the first park() call after an unpark() call which will help
5537 // shake out uses of park() and unpark() without condition variables.
5538
5539 if (Atomic::xchg(1, &_Event) >= 0) return;
5540
5541 // Wait for the thread associated with the event to vacate
5542 int status = pthread_mutex_lock(_mutex);
5543 assert_status(status == 0, status, "mutex_lock");
5544 int AnyWaiters = _nParked;
5545 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5546 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5547 AnyWaiters = 0;
5548 pthread_cond_signal(_cond);
5549 }
5550 status = pthread_mutex_unlock(_mutex);
5551 assert_status(status == 0, status, "mutex_unlock");
5552 if (AnyWaiters != 0) {
5553 // Note that we signal() *after* dropping the lock for "immortal" Events.
5554 // This is safe and avoids a common class of futile wakeups. In rare
5555 // circumstances this can cause a thread to return prematurely from
5556 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5557 // will simply re-test the condition and re-park itself.
5558 // This provides particular benefit if the underlying platform does not
5559 // provide wait morphing.
5560 status = pthread_cond_signal(_cond);
5561 assert_status(status == 0, status, "cond_signal");
5562 }
5563 }
5564
5565
5566 // JSR166
5567 // -------------------------------------------------------
5568
5569 // The solaris and linux implementations of park/unpark are fairly
5570 // conservative for now, but can be improved. They currently use a
5571 // mutex/condvar pair, plus a a count.
5572 // Park decrements count if > 0, else does a condvar wait. Unpark
5573 // sets count to 1 and signals condvar. Only one thread ever waits
5574 // on the condvar. Contention seen when trying to park implies that someone
5575 // is unparking you, so don't wait. And spurious returns are fine, so there
5576 // is no need to track notifications.
5577
5578 // This code is common to linux and solaris and will be moved to a
5579 // common place in dolphin.
5580 //
5581 // The passed in time value is either a relative time in nanoseconds
5582 // or an absolute time in milliseconds. Either way it has to be unpacked
5583 // into suitable seconds and nanoseconds components and stored in the
5584 // given timespec structure.
5585 // Given time is a 64-bit value and the time_t used in the timespec is only
5586 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5587 // overflow if times way in the future are given. Further on Solaris versions
5588 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5589 // number of seconds, in abstime, is less than current_time + 100,000,000.
5590 // As it will be 28 years before "now + 100000000" will overflow we can
5591 // ignore overflow and just impose a hard-limit on seconds using the value
5592 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5593 // years from "now".
5594
5595 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5596 assert(time > 0, "convertTime");
5597 time_t max_secs = 0;
5598
5599 if (!os::supports_monotonic_clock() || isAbsolute) {
5600 struct timeval now;
5601 int status = gettimeofday(&now, NULL);
5602 assert(status == 0, "gettimeofday");
5603
5604 max_secs = now.tv_sec + MAX_SECS;
5605
5606 if (isAbsolute) {
5607 jlong secs = time / 1000;
5608 if (secs > max_secs) {
5609 absTime->tv_sec = max_secs;
5610 } else {
5611 absTime->tv_sec = secs;
5612 }
5613 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5614 } else {
5615 jlong secs = time / NANOSECS_PER_SEC;
5616 if (secs >= MAX_SECS) {
5617 absTime->tv_sec = max_secs;
5618 absTime->tv_nsec = 0;
5619 } else {
5620 absTime->tv_sec = now.tv_sec + secs;
5621 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5622 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5623 absTime->tv_nsec -= NANOSECS_PER_SEC;
5624 ++absTime->tv_sec; // note: this must be <= max_secs
5625 }
5626 }
5627 }
5628 } else {
5629 // must be relative using monotonic clock
5630 struct timespec now;
5631 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5632 assert_status(status == 0, status, "clock_gettime");
5633 max_secs = now.tv_sec + MAX_SECS;
5634 jlong secs = time / NANOSECS_PER_SEC;
5635 if (secs >= MAX_SECS) {
5636 absTime->tv_sec = max_secs;
5637 absTime->tv_nsec = 0;
5638 } else {
5639 absTime->tv_sec = now.tv_sec + secs;
5640 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5641 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5642 absTime->tv_nsec -= NANOSECS_PER_SEC;
5643 ++absTime->tv_sec; // note: this must be <= max_secs
5644 }
5645 }
5646 }
5647 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5648 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5649 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5650 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5651 }
5652
5653 void Parker::park(bool isAbsolute, jlong time) {
5654 // Ideally we'd do something useful while spinning, such
5655 // as calling unpackTime().
5656
5657 // Optional fast-path check:
5658 // Return immediately if a permit is available.
5659 // We depend on Atomic::xchg() having full barrier semantics
5660 // since we are doing a lock-free update to _counter.
5661 if (Atomic::xchg(0, &_counter) > 0) return;
5662
5663 Thread* thread = Thread::current();
5664 assert(thread->is_Java_thread(), "Must be JavaThread");
5665 JavaThread *jt = (JavaThread *)thread;
5666
5667 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5668 // Check interrupt before trying to wait
5669 if (Thread::is_interrupted(thread, false)) {
5670 return;
5671 }
5672
5673 // Next, demultiplex/decode time arguments
5674 timespec absTime;
5675 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5676 return;
5677 }
5678 if (time > 0) {
5679 unpackTime(&absTime, isAbsolute, time);
5680 }
5681
5682
5683 // Enter safepoint region
5684 // Beware of deadlocks such as 6317397.
5685 // The per-thread Parker:: mutex is a classic leaf-lock.
5686 // In particular a thread must never block on the Threads_lock while
5687 // holding the Parker:: mutex. If safepoints are pending both the
5688 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5689 ThreadBlockInVM tbivm(jt);
5690
5691 // Don't wait if cannot get lock since interference arises from
5692 // unblocking. Also. check interrupt before trying wait
5693 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5694 return;
5695 }
5696
5697 int status;
5698 if (_counter > 0) { // no wait needed
5699 _counter = 0;
5700 status = pthread_mutex_unlock(_mutex);
5701 assert(status == 0, "invariant");
5702 // Paranoia to ensure our locked and lock-free paths interact
5703 // correctly with each other and Java-level accesses.
5704 OrderAccess::fence();
5705 return;
5706 }
5707
5708 #ifdef ASSERT
5709 // Don't catch signals while blocked; let the running threads have the signals.
5710 // (This allows a debugger to break into the running thread.)
5711 sigset_t oldsigs;
5712 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5713 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5714 #endif
5715
5716 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5717 jt->set_suspend_equivalent();
5718 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5719
5720 assert(_cur_index == -1, "invariant");
5721 if (time == 0) {
5722 _cur_index = REL_INDEX; // arbitrary choice when not timed
5723 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5724 } else {
5725 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5726 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5727 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5728 pthread_cond_destroy(&_cond[_cur_index]);
5729 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5730 }
5731 }
5732 _cur_index = -1;
5733 assert_status(status == 0 || status == EINTR ||
5734 status == ETIME || status == ETIMEDOUT,
5735 status, "cond_timedwait");
5736
5737 #ifdef ASSERT
5738 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5739 #endif
5740
5741 _counter = 0;
5742 status = pthread_mutex_unlock(_mutex);
5743 assert_status(status == 0, status, "invariant");
5744 // Paranoia to ensure our locked and lock-free paths interact
5745 // correctly with each other and Java-level accesses.
5746 OrderAccess::fence();
5747
5748 // If externally suspended while waiting, re-suspend
5749 if (jt->handle_special_suspend_equivalent_condition()) {
5750 jt->java_suspend_self();
5751 }
5752 }
5753
5754 void Parker::unpark() {
5755 int status = pthread_mutex_lock(_mutex);
5756 assert(status == 0, "invariant");
5757 const int s = _counter;
5758 _counter = 1;
5759 if (s < 1) {
5760 // thread might be parked
5761 if (_cur_index != -1) {
5762 // thread is definitely parked
5763 if (WorkAroundNPTLTimedWaitHang) {
5764 status = pthread_cond_signal(&_cond[_cur_index]);
5765 assert(status == 0, "invariant");
5766 status = pthread_mutex_unlock(_mutex);
5767 assert(status == 0, "invariant");
5768 } else {
5769 // must capture correct index before unlocking
5770 int index = _cur_index;
5771 status = pthread_mutex_unlock(_mutex);
5772 assert(status == 0, "invariant");
5773 status = pthread_cond_signal(&_cond[index]);
5774 assert(status == 0, "invariant");
5775 }
5776 } else {
5777 pthread_mutex_unlock(_mutex);
5778 assert(status == 0, "invariant");
5779 }
5780 } else {
5781 pthread_mutex_unlock(_mutex);
5782 assert(status == 0, "invariant");
5783 }
5784 }
5785
5786
5787 extern char** environ;
5788
5789 // Run the specified command in a separate process. Return its exit value,
5790 // or -1 on failure (e.g. can't fork a new process).
5791 // Unlike system(), this function can be called from signal handler. It
5792 // doesn't block SIGINT et al.
5793 int os::fork_and_exec(char* cmd) {
5794 const char * argv[4] = {"sh", "-c", cmd, NULL};
5795
5796 pid_t pid = fork();
5797
5798 if (pid < 0) {
5799 // fork failed
5800 return -1;
5801
5802 } else if (pid == 0) {
5803 // child process
5804
5805 execve("/bin/sh", (char* const*)argv, environ);
5806
5807 // execve failed
5808 _exit(-1);
5809
5810 } else {
5811 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5812 // care about the actual exit code, for now.
5813
5814 int status;
5815
5816 // Wait for the child process to exit. This returns immediately if
5817 // the child has already exited. */
5818 while (waitpid(pid, &status, 0) < 0) {
5819 switch (errno) {
5820 case ECHILD: return 0;
5821 case EINTR: break;
5822 default: return -1;
5823 }
5824 }
5825
5826 if (WIFEXITED(status)) {
5827 // The child exited normally; get its exit code.
5828 return WEXITSTATUS(status);
5829 } else if (WIFSIGNALED(status)) {
5830 // The child exited because of a signal
5831 // The best value to return is 0x80 + signal number,
5832 // because that is what all Unix shells do, and because
5833 // it allows callers to distinguish between process exit and
5834 // process death by signal.
5835 return 0x80 + WTERMSIG(status);
5836 } else {
5837 // Unknown exit code; pass it through
5838 return status;
5839 }
5840 }
5841 }
5842
5843 // is_headless_jre()
5844 //
5845 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5846 // in order to report if we are running in a headless jre
5847 //
5848 // Since JDK8 xawt/libmawt.so was moved into the same directory
5849 // as libawt.so, and renamed libawt_xawt.so
5850 //
5851 bool os::is_headless_jre() {
5852 struct stat statbuf;
5853 char buf[MAXPATHLEN];
5854 char libmawtpath[MAXPATHLEN];
5855 const char *xawtstr = "/xawt/libmawt.so";
5856 const char *new_xawtstr = "/libawt_xawt.so";
5857 char *p;
5858
5859 // Get path to libjvm.so
5860 os::jvm_path(buf, sizeof(buf));
5861
5862 // Get rid of libjvm.so
5863 p = strrchr(buf, '/');
5864 if (p == NULL) {
5865 return false;
5866 } else {
5867 *p = '\0';
5868 }
5869
5870 // Get rid of client or server
5871 p = strrchr(buf, '/');
5872 if (p == NULL) {
5873 return false;
5874 } else {
5875 *p = '\0';
5876 }
5877
5878 // check xawt/libmawt.so
5879 strcpy(libmawtpath, buf);
5880 strcat(libmawtpath, xawtstr);
5881 if (::stat(libmawtpath, &statbuf) == 0) return false;
5882
5883 // check libawt_xawt.so
5884 strcpy(libmawtpath, buf);
5885 strcat(libmawtpath, new_xawtstr);
5886 if (::stat(libmawtpath, &statbuf) == 0) return false;
5887
5888 return true;
5889 }
5890
5891 // Get the default path to the core file
5892 // Returns the length of the string
5893 int os::get_core_path(char* buffer, size_t bufferSize) {
5894 /*
5895 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5896 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5897 */
5898 const int core_pattern_len = 129;
5899 char core_pattern[core_pattern_len] = {0};
5900
5901 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5902 if (core_pattern_file == -1) {
5903 return -1;
5904 }
5905
5906 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5907 ::close(core_pattern_file);
5908 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5909 return -1;
5910 }
5911 if (core_pattern[ret-1] == '\n') {
5912 core_pattern[ret-1] = '\0';
5913 } else {
5914 core_pattern[ret] = '\0';
5915 }
5916
5917 char *pid_pos = strstr(core_pattern, "%p");
5918 int written;
5919
5920 if (core_pattern[0] == '/') {
5921 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5922 } else {
5923 char cwd[PATH_MAX];
5924
5925 const char* p = get_current_directory(cwd, PATH_MAX);
5926 if (p == NULL) {
5927 return -1;
5928 }
5929
5930 if (core_pattern[0] == '|') {
5931 written = jio_snprintf(buffer, bufferSize,
5932 "\"%s\" (or dumping to %s/core.%d)",
5933 &core_pattern[1], p, current_process_id());
5934 } else {
5935 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5936 }
5937 }
5938
5939 if (written < 0) {
5940 return -1;
5941 }
5942
5943 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5944 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5945
5946 if (core_uses_pid_file != -1) {
5947 char core_uses_pid = 0;
5948 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5949 ::close(core_uses_pid_file);
5950
5951 if (core_uses_pid == '1') {
5952 jio_snprintf(buffer + written, bufferSize - written,
5953 ".%d", current_process_id());
5954 }
5955 }
5956 }
5957
5958 return strlen(buffer);
5959 }
5960
5961 /////////////// Unit tests ///////////////
5962
5963 #ifndef PRODUCT
5964
5965 #define test_log(...) \
5966 do { \
5967 if (VerboseInternalVMTests) { \
5968 tty->print_cr(__VA_ARGS__); \
5969 tty->flush(); \
5970 } \
5971 } while (false)
5972
5973 class TestReserveMemorySpecial : AllStatic {
5974 public:
5975 static void small_page_write(void* addr, size_t size) {
5976 size_t page_size = os::vm_page_size();
5977
5978 char* end = (char*)addr + size;
5979 for (char* p = (char*)addr; p < end; p += page_size) {
5980 *p = 1;
5981 }
5982 }
5983
5984 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5985 if (!UseHugeTLBFS) {
5986 return;
5987 }
5988
5989 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5990
5991 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5992
5993 if (addr != NULL) {
5994 small_page_write(addr, size);
5995
5996 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5997 }
5998 }
5999
6000 static void test_reserve_memory_special_huge_tlbfs_only() {
6001 if (!UseHugeTLBFS) {
6002 return;
6003 }
6004
6005 size_t lp = os::large_page_size();
6006
6007 for (size_t size = lp; size <= lp * 10; size += lp) {
6008 test_reserve_memory_special_huge_tlbfs_only(size);
6009 }
6010 }
6011
6012 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6013 size_t lp = os::large_page_size();
6014 size_t ag = os::vm_allocation_granularity();
6015
6016 // sizes to test
6017 const size_t sizes[] = {
6018 lp, lp + ag, lp + lp / 2, lp * 2,
6019 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6020 lp * 10, lp * 10 + lp / 2
6021 };
6022 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6023
6024 // For each size/alignment combination, we test three scenarios:
6025 // 1) with req_addr == NULL
6026 // 2) with a non-null req_addr at which we expect to successfully allocate
6027 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6028 // expect the allocation to either fail or to ignore req_addr
6029
6030 // Pre-allocate two areas; they shall be as large as the largest allocation
6031 // and aligned to the largest alignment we will be testing.
6032 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6033 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6034 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6035 -1, 0);
6036 assert(mapping1 != MAP_FAILED, "should work");
6037
6038 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6039 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6040 -1, 0);
6041 assert(mapping2 != MAP_FAILED, "should work");
6042
6043 // Unmap the first mapping, but leave the second mapping intact: the first
6044 // mapping will serve as a value for a "good" req_addr (case 2). The second
6045 // mapping, still intact, as "bad" req_addr (case 3).
6046 ::munmap(mapping1, mapping_size);
6047
6048 // Case 1
6049 test_log("%s, req_addr NULL:", __FUNCTION__);
6050 test_log("size align result");
6051
6052 for (int i = 0; i < num_sizes; i++) {
6053 const size_t size = sizes[i];
6054 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6055 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6056 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6057 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6058 if (p != NULL) {
6059 assert(is_ptr_aligned(p, alignment), "must be");
6060 small_page_write(p, size);
6061 os::Linux::release_memory_special_huge_tlbfs(p, size);
6062 }
6063 }
6064 }
6065
6066 // Case 2
6067 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6068 test_log("size align req_addr result");
6069
6070 for (int i = 0; i < num_sizes; i++) {
6071 const size_t size = sizes[i];
6072 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6073 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6074 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6075 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6076 size, alignment, p2i(req_addr), p2i(p),
6077 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6078 if (p != NULL) {
6079 assert(p == req_addr, "must be");
6080 small_page_write(p, size);
6081 os::Linux::release_memory_special_huge_tlbfs(p, size);
6082 }
6083 }
6084 }
6085
6086 // Case 3
6087 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6088 test_log("size align req_addr result");
6089
6090 for (int i = 0; i < num_sizes; i++) {
6091 const size_t size = sizes[i];
6092 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6093 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6094 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6095 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6096 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6097 // as the area around req_addr contains already existing mappings, the API should always
6098 // return NULL (as per contract, it cannot return another address)
6099 assert(p == NULL, "must be");
6100 }
6101 }
6102
6103 ::munmap(mapping2, mapping_size);
6104
6105 }
6106
6107 static void test_reserve_memory_special_huge_tlbfs() {
6108 if (!UseHugeTLBFS) {
6109 return;
6110 }
6111
6112 test_reserve_memory_special_huge_tlbfs_only();
6113 test_reserve_memory_special_huge_tlbfs_mixed();
6114 }
6115
6116 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6117 if (!UseSHM) {
6118 return;
6119 }
6120
6121 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6122
6123 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6124
6125 if (addr != NULL) {
6126 assert(is_ptr_aligned(addr, alignment), "Check");
6127 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6128
6129 small_page_write(addr, size);
6130
6131 os::Linux::release_memory_special_shm(addr, size);
6132 }
6133 }
6134
6135 static void test_reserve_memory_special_shm() {
6136 size_t lp = os::large_page_size();
6137 size_t ag = os::vm_allocation_granularity();
6138
6139 for (size_t size = ag; size < lp * 3; size += ag) {
6140 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6141 test_reserve_memory_special_shm(size, alignment);
6142 }
6143 }
6144 }
6145
6146 static void test() {
6147 test_reserve_memory_special_huge_tlbfs();
6148 test_reserve_memory_special_shm();
6149 }
6150 };
6151
6152 void TestReserveMemorySpecial_test() {
6153 TestReserveMemorySpecial::test();
6154 }
6155
6156 #endif