1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_linux.inline.hpp"
41 #include "os_share_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm.h"
44 #include "prims/jvm_misc.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/atomic.inline.hpp"
47 #include "runtime/extendedPC.hpp"
48 #include "runtime/globals.hpp"
49 #include "runtime/interfaceSupport.hpp"
50 #include "runtime/init.hpp"
51 #include "runtime/java.hpp"
52 #include "runtime/javaCalls.hpp"
53 #include "runtime/mutexLocker.hpp"
54 #include "runtime/objectMonitor.hpp"
55 #include "runtime/orderAccess.inline.hpp"
56 #include "runtime/osThread.hpp"
57 #include "runtime/perfMemory.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/statSampler.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "runtime/thread.inline.hpp"
62 #include "runtime/threadCritical.hpp"
63 #include "runtime/timer.hpp"
64 #include "semaphore_posix.hpp"
65 #include "services/attachListener.hpp"
66 #include "services/memTracker.hpp"
67 #include "services/runtimeService.hpp"
68 #include "utilities/decoder.hpp"
69 #include "utilities/defaultStream.hpp"
70 #include "utilities/events.hpp"
71 #include "utilities/elfFile.hpp"
72 #include "utilities/growableArray.hpp"
73 #include "utilities/macros.hpp"
74 #include "utilities/vmError.hpp"
75
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
109
110 #ifndef _GNU_SOURCE
111 #define _GNU_SOURCE
112 #include <sched.h>
113 #undef _GNU_SOURCE
114 #else
115 #include <sched.h>
116 #endif
117
118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
119 // getrusage() is prepared to handle the associated failure.
120 #ifndef RUSAGE_THREAD
121 #define RUSAGE_THREAD (1) /* only the calling thread */
122 #endif
123
124 #define MAX_PATH (2 * K)
125
126 #define MAX_SECS 100000000
127
128 // for timer info max values which include all bits
129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
130
131 #define LARGEPAGES_BIT (1 << 6)
132 ////////////////////////////////////////////////////////////////////////////////
133 // global variables
134 julong os::Linux::_physical_memory = 0;
135
136 address os::Linux::_initial_thread_stack_bottom = NULL;
137 uintptr_t os::Linux::_initial_thread_stack_size = 0;
138
139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
142 Mutex* os::Linux::_createThread_lock = NULL;
143 pthread_t os::Linux::_main_thread;
144 int os::Linux::_page_size = -1;
145 const int os::Linux::_vm_default_page_size = (8 * K);
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149 pthread_condattr_t os::Linux::_condattr[1];
150
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // Declarations
166 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
167
168 // utility functions
169
170 static int SR_initialize();
171
172 julong os::available_memory() {
173 return Linux::available_memory();
174 }
175
176 julong os::Linux::available_memory() {
177 // values in struct sysinfo are "unsigned long"
178 struct sysinfo si;
179 sysinfo(&si);
180
181 return (julong)si.freeram * si.mem_unit;
182 }
183
184 julong os::physical_memory() {
185 return Linux::physical_memory();
186 }
187
188 // Return true if user is running as root.
189
190 bool os::have_special_privileges() {
191 static bool init = false;
192 static bool privileges = false;
193 if (!init) {
194 privileges = (getuid() != geteuid()) || (getgid() != getegid());
195 init = true;
196 }
197 return privileges;
198 }
199
200
201 #ifndef SYS_gettid
202 // i386: 224, ia64: 1105, amd64: 186, sparc 143
203 #ifdef __ia64__
204 #define SYS_gettid 1105
205 #else
206 #ifdef __i386__
207 #define SYS_gettid 224
208 #else
209 #ifdef __amd64__
210 #define SYS_gettid 186
211 #else
212 #ifdef __sparc__
213 #define SYS_gettid 143
214 #else
215 #error define gettid for the arch
216 #endif
217 #endif
218 #endif
219 #endif
220 #endif
221
222 // Cpu architecture string
223 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
224
225
226 // pid_t gettid()
227 //
228 // Returns the kernel thread id of the currently running thread. Kernel
229 // thread id is used to access /proc.
230 pid_t os::Linux::gettid() {
231 int rslt = syscall(SYS_gettid);
232 assert(rslt != -1, "must be."); // old linuxthreads implementation?
233 return (pid_t)rslt;
234 }
235
236 // Most versions of linux have a bug where the number of processors are
237 // determined by looking at the /proc file system. In a chroot environment,
238 // the system call returns 1. This causes the VM to act as if it is
239 // a single processor and elide locking (see is_MP() call).
240 static bool unsafe_chroot_detected = false;
241 static const char *unstable_chroot_error = "/proc file system not found.\n"
242 "Java may be unstable running multithreaded in a chroot "
243 "environment on Linux when /proc filesystem is not mounted.";
244
245 void os::Linux::initialize_system_info() {
246 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
247 if (processor_count() == 1) {
248 pid_t pid = os::Linux::gettid();
249 char fname[32];
250 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
251 FILE *fp = fopen(fname, "r");
252 if (fp == NULL) {
253 unsafe_chroot_detected = true;
254 } else {
255 fclose(fp);
256 }
257 }
258 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
259 assert(processor_count() > 0, "linux error");
260 }
261
262 void os::init_system_properties_values() {
263 // The next steps are taken in the product version:
264 //
265 // Obtain the JAVA_HOME value from the location of libjvm.so.
266 // This library should be located at:
267 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
268 //
269 // If "/jre/lib/" appears at the right place in the path, then we
270 // assume libjvm.so is installed in a JDK and we use this path.
271 //
272 // Otherwise exit with message: "Could not create the Java virtual machine."
273 //
274 // The following extra steps are taken in the debugging version:
275 //
276 // If "/jre/lib/" does NOT appear at the right place in the path
277 // instead of exit check for $JAVA_HOME environment variable.
278 //
279 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
280 // then we append a fake suffix "hotspot/libjvm.so" to this path so
281 // it looks like libjvm.so is installed there
282 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
283 //
284 // Otherwise exit.
285 //
286 // Important note: if the location of libjvm.so changes this
287 // code needs to be changed accordingly.
288
289 // See ld(1):
290 // The linker uses the following search paths to locate required
291 // shared libraries:
292 // 1: ...
293 // ...
294 // 7: The default directories, normally /lib and /usr/lib.
295 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
296 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
297 #else
298 #define DEFAULT_LIBPATH "/lib:/usr/lib"
299 #endif
300
301 // Base path of extensions installed on the system.
302 #define SYS_EXT_DIR "/usr/java/packages"
303 #define EXTENSIONS_DIR "/lib/ext"
304
305 // Buffer that fits several sprintfs.
306 // Note that the space for the colon and the trailing null are provided
307 // by the nulls included by the sizeof operator.
308 const size_t bufsize =
309 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
310 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
311 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
312
313 // sysclasspath, java_home, dll_dir
314 {
315 char *pslash;
316 os::jvm_path(buf, bufsize);
317
318 // Found the full path to libjvm.so.
319 // Now cut the path to <java_home>/jre if we can.
320 pslash = strrchr(buf, '/');
321 if (pslash != NULL) {
322 *pslash = '\0'; // Get rid of /libjvm.so.
323 }
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
327 }
328 Arguments::set_dll_dir(buf);
329
330 if (pslash != NULL) {
331 pslash = strrchr(buf, '/');
332 if (pslash != NULL) {
333 *pslash = '\0'; // Get rid of /<arch>.
334 pslash = strrchr(buf, '/');
335 if (pslash != NULL) {
336 *pslash = '\0'; // Get rid of /lib.
337 }
338 }
339 }
340 Arguments::set_java_home(buf);
341 set_boot_path('/', ':');
342 }
343
344 // Where to look for native libraries.
345 //
346 // Note: Due to a legacy implementation, most of the library path
347 // is set in the launcher. This was to accomodate linking restrictions
348 // on legacy Linux implementations (which are no longer supported).
349 // Eventually, all the library path setting will be done here.
350 //
351 // However, to prevent the proliferation of improperly built native
352 // libraries, the new path component /usr/java/packages is added here.
353 // Eventually, all the library path setting will be done here.
354 {
355 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
356 // should always exist (until the legacy problem cited above is
357 // addressed).
358 const char *v = ::getenv("LD_LIBRARY_PATH");
359 const char *v_colon = ":";
360 if (v == NULL) { v = ""; v_colon = ""; }
361 // That's +1 for the colon and +1 for the trailing '\0'.
362 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
363 strlen(v) + 1 +
364 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
365 mtInternal);
366 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
367 Arguments::set_library_path(ld_library_path);
368 FREE_C_HEAP_ARRAY(char, ld_library_path);
369 }
370
371 // Extensions directories.
372 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
373 Arguments::set_ext_dirs(buf);
374
375 FREE_C_HEAP_ARRAY(char, buf);
376
377 #undef DEFAULT_LIBPATH
378 #undef SYS_EXT_DIR
379 #undef EXTENSIONS_DIR
380 }
381
382 ////////////////////////////////////////////////////////////////////////////////
383 // breakpoint support
384
385 void os::breakpoint() {
386 BREAKPOINT;
387 }
388
389 extern "C" void breakpoint() {
390 // use debugger to set breakpoint here
391 }
392
393 ////////////////////////////////////////////////////////////////////////////////
394 // signal support
395
396 debug_only(static bool signal_sets_initialized = false);
397 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
398
399 bool os::Linux::is_sig_ignored(int sig) {
400 struct sigaction oact;
401 sigaction(sig, (struct sigaction*)NULL, &oact);
402 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
403 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
404 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
405 return true;
406 } else {
407 return false;
408 }
409 }
410
411 void os::Linux::signal_sets_init() {
412 // Should also have an assertion stating we are still single-threaded.
413 assert(!signal_sets_initialized, "Already initialized");
414 // Fill in signals that are necessarily unblocked for all threads in
415 // the VM. Currently, we unblock the following signals:
416 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
417 // by -Xrs (=ReduceSignalUsage));
418 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
419 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
420 // the dispositions or masks wrt these signals.
421 // Programs embedding the VM that want to use the above signals for their
422 // own purposes must, at this time, use the "-Xrs" option to prevent
423 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
424 // (See bug 4345157, and other related bugs).
425 // In reality, though, unblocking these signals is really a nop, since
426 // these signals are not blocked by default.
427 sigemptyset(&unblocked_sigs);
428 sigemptyset(&allowdebug_blocked_sigs);
429 sigaddset(&unblocked_sigs, SIGILL);
430 sigaddset(&unblocked_sigs, SIGSEGV);
431 sigaddset(&unblocked_sigs, SIGBUS);
432 sigaddset(&unblocked_sigs, SIGFPE);
433 #if defined(PPC64)
434 sigaddset(&unblocked_sigs, SIGTRAP);
435 #endif
436 sigaddset(&unblocked_sigs, SR_signum);
437
438 if (!ReduceSignalUsage) {
439 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
440 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
441 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
442 }
443 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
444 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
445 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
446 }
447 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
448 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
449 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
450 }
451 }
452 // Fill in signals that are blocked by all but the VM thread.
453 sigemptyset(&vm_sigs);
454 if (!ReduceSignalUsage) {
455 sigaddset(&vm_sigs, BREAK_SIGNAL);
456 }
457 debug_only(signal_sets_initialized = true);
458
459 }
460
461 // These are signals that are unblocked while a thread is running Java.
462 // (For some reason, they get blocked by default.)
463 sigset_t* os::Linux::unblocked_signals() {
464 assert(signal_sets_initialized, "Not initialized");
465 return &unblocked_sigs;
466 }
467
468 // These are the signals that are blocked while a (non-VM) thread is
469 // running Java. Only the VM thread handles these signals.
470 sigset_t* os::Linux::vm_signals() {
471 assert(signal_sets_initialized, "Not initialized");
472 return &vm_sigs;
473 }
474
475 // These are signals that are blocked during cond_wait to allow debugger in
476 sigset_t* os::Linux::allowdebug_blocked_signals() {
477 assert(signal_sets_initialized, "Not initialized");
478 return &allowdebug_blocked_sigs;
479 }
480
481 void os::Linux::hotspot_sigmask(Thread* thread) {
482
483 //Save caller's signal mask before setting VM signal mask
484 sigset_t caller_sigmask;
485 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
486
487 OSThread* osthread = thread->osthread();
488 osthread->set_caller_sigmask(caller_sigmask);
489
490 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
491
492 if (!ReduceSignalUsage) {
493 if (thread->is_VM_thread()) {
494 // Only the VM thread handles BREAK_SIGNAL ...
495 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
496 } else {
497 // ... all other threads block BREAK_SIGNAL
498 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
499 }
500 }
501 }
502
503 //////////////////////////////////////////////////////////////////////////////
504 // detecting pthread library
505
506 void os::Linux::libpthread_init() {
507 // Save glibc and pthread version strings.
508 #if !defined(_CS_GNU_LIBC_VERSION) || \
509 !defined(_CS_GNU_LIBPTHREAD_VERSION)
510 #error "glibc too old (< 2.3.2)"
511 #endif
512
513 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
514 assert(n > 0, "cannot retrieve glibc version");
515 char *str = (char *)malloc(n, mtInternal);
516 confstr(_CS_GNU_LIBC_VERSION, str, n);
517 os::Linux::set_glibc_version(str);
518
519 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
520 assert(n > 0, "cannot retrieve pthread version");
521 str = (char *)malloc(n, mtInternal);
522 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
523 os::Linux::set_libpthread_version(str);
524 }
525
526 /////////////////////////////////////////////////////////////////////////////
527 // thread stack expansion
528
529 // os::Linux::manually_expand_stack() takes care of expanding the thread
530 // stack. Note that this is normally not needed: pthread stacks allocate
531 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
532 // committed. Therefore it is not necessary to expand the stack manually.
533 //
534 // Manually expanding the stack was historically needed on LinuxThreads
535 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
536 // it is kept to deal with very rare corner cases:
537 //
538 // For one, user may run the VM on an own implementation of threads
539 // whose stacks are - like the old LinuxThreads - implemented using
540 // mmap(MAP_GROWSDOWN).
541 //
542 // Also, this coding may be needed if the VM is running on the primordial
543 // thread. Normally we avoid running on the primordial thread; however,
544 // user may still invoke the VM on the primordial thread.
545 //
546 // The following historical comment describes the details about running
547 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
548
549
550 // Force Linux kernel to expand current thread stack. If "bottom" is close
551 // to the stack guard, caller should block all signals.
552 //
553 // MAP_GROWSDOWN:
554 // A special mmap() flag that is used to implement thread stacks. It tells
555 // kernel that the memory region should extend downwards when needed. This
556 // allows early versions of LinuxThreads to only mmap the first few pages
557 // when creating a new thread. Linux kernel will automatically expand thread
558 // stack as needed (on page faults).
559 //
560 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
561 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
562 // region, it's hard to tell if the fault is due to a legitimate stack
563 // access or because of reading/writing non-exist memory (e.g. buffer
564 // overrun). As a rule, if the fault happens below current stack pointer,
565 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
566 // application (see Linux kernel fault.c).
567 //
568 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
569 // stack overflow detection.
570 //
571 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
572 // not use MAP_GROWSDOWN.
573 //
574 // To get around the problem and allow stack banging on Linux, we need to
575 // manually expand thread stack after receiving the SIGSEGV.
576 //
577 // There are two ways to expand thread stack to address "bottom", we used
578 // both of them in JVM before 1.5:
579 // 1. adjust stack pointer first so that it is below "bottom", and then
580 // touch "bottom"
581 // 2. mmap() the page in question
582 //
583 // Now alternate signal stack is gone, it's harder to use 2. For instance,
584 // if current sp is already near the lower end of page 101, and we need to
585 // call mmap() to map page 100, it is possible that part of the mmap() frame
586 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
587 // That will destroy the mmap() frame and cause VM to crash.
588 //
589 // The following code works by adjusting sp first, then accessing the "bottom"
590 // page to force a page fault. Linux kernel will then automatically expand the
591 // stack mapping.
592 //
593 // _expand_stack_to() assumes its frame size is less than page size, which
594 // should always be true if the function is not inlined.
595
596 static void NOINLINE _expand_stack_to(address bottom) {
597 address sp;
598 size_t size;
599 volatile char *p;
600
601 // Adjust bottom to point to the largest address within the same page, it
602 // gives us a one-page buffer if alloca() allocates slightly more memory.
603 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
604 bottom += os::Linux::page_size() - 1;
605
606 // sp might be slightly above current stack pointer; if that's the case, we
607 // will alloca() a little more space than necessary, which is OK. Don't use
608 // os::current_stack_pointer(), as its result can be slightly below current
609 // stack pointer, causing us to not alloca enough to reach "bottom".
610 sp = (address)&sp;
611
612 if (sp > bottom) {
613 size = sp - bottom;
614 p = (volatile char *)alloca(size);
615 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
616 p[0] = '\0';
617 }
618 }
619
620 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
621 assert(t!=NULL, "just checking");
622 assert(t->osthread()->expanding_stack(), "expand should be set");
623 assert(t->stack_base() != NULL, "stack_base was not initialized");
624
625 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
626 sigset_t mask_all, old_sigset;
627 sigfillset(&mask_all);
628 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
629 _expand_stack_to(addr);
630 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
631 return true;
632 }
633 return false;
634 }
635
636 //////////////////////////////////////////////////////////////////////////////
637 // create new thread
638
639 // Thread start routine for all newly created threads
640 static void *java_start(Thread *thread) {
641 // Try to randomize the cache line index of hot stack frames.
642 // This helps when threads of the same stack traces evict each other's
643 // cache lines. The threads can be either from the same JVM instance, or
644 // from different JVM instances. The benefit is especially true for
645 // processors with hyperthreading technology.
646 static int counter = 0;
647 int pid = os::current_process_id();
648 alloca(((pid ^ counter++) & 7) * 128);
649
650 thread->initialize_thread_current();
651
652 OSThread* osthread = thread->osthread();
653 Monitor* sync = osthread->startThread_lock();
654
655 osthread->set_thread_id(os::current_thread_id());
656
657 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
658 os::current_thread_id(), (uintx) pthread_self());
659
660 if (UseNUMA) {
661 int lgrp_id = os::numa_get_group_id();
662 if (lgrp_id != -1) {
663 thread->set_lgrp_id(lgrp_id);
664 }
665 }
666 // initialize signal mask for this thread
667 os::Linux::hotspot_sigmask(thread);
668
669 // initialize floating point control register
670 os::Linux::init_thread_fpu_state();
671
672 // handshaking with parent thread
673 {
674 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
675
676 // notify parent thread
677 osthread->set_state(INITIALIZED);
678 sync->notify_all();
679
680 // wait until os::start_thread()
681 while (osthread->get_state() == INITIALIZED) {
682 sync->wait(Mutex::_no_safepoint_check_flag);
683 }
684 }
685
686 // call one more level start routine
687 thread->run();
688
689 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
690 os::current_thread_id(), (uintx) pthread_self());
691
692 return 0;
693 }
694
695 bool os::create_thread(Thread* thread, ThreadType thr_type,
696 size_t stack_size) {
697 assert(thread->osthread() == NULL, "caller responsible");
698
699 // Allocate the OSThread object
700 OSThread* osthread = new OSThread(NULL, NULL);
701 if (osthread == NULL) {
702 return false;
703 }
704
705 // set the correct thread state
706 osthread->set_thread_type(thr_type);
707
708 // Initial state is ALLOCATED but not INITIALIZED
709 osthread->set_state(ALLOCATED);
710
711 thread->set_osthread(osthread);
712
713 // init thread attributes
714 pthread_attr_t attr;
715 pthread_attr_init(&attr);
716 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
717
718 // stack size
719 // calculate stack size if it's not specified by caller
720 if (stack_size == 0) {
721 stack_size = os::Linux::default_stack_size(thr_type);
722
723 switch (thr_type) {
724 case os::java_thread:
725 // Java threads use ThreadStackSize which default value can be
726 // changed with the flag -Xss
727 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
728 stack_size = JavaThread::stack_size_at_create();
729 break;
730 case os::compiler_thread:
731 if (CompilerThreadStackSize > 0) {
732 stack_size = (size_t)(CompilerThreadStackSize * K);
733 break;
734 } // else fall through:
735 // use VMThreadStackSize if CompilerThreadStackSize is not defined
736 case os::vm_thread:
737 case os::pgc_thread:
738 case os::cgc_thread:
739 case os::watcher_thread:
740 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
741 break;
742 }
743 }
744
745 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
746 pthread_attr_setstacksize(&attr, stack_size);
747
748 // glibc guard page
749 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
750
751 ThreadState state;
752
753 {
754 pthread_t tid;
755 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
756
757 char buf[64];
758 if (ret == 0) {
759 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
760 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
761 } else {
762 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
763 strerror(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
764 }
765
766 pthread_attr_destroy(&attr);
767
768 if (ret != 0) {
769 // Need to clean up stuff we've allocated so far
770 thread->set_osthread(NULL);
771 delete osthread;
772 return false;
773 }
774
775 // Store pthread info into the OSThread
776 osthread->set_pthread_id(tid);
777
778 // Wait until child thread is either initialized or aborted
779 {
780 Monitor* sync_with_child = osthread->startThread_lock();
781 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
782 while ((state = osthread->get_state()) == ALLOCATED) {
783 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
784 }
785 }
786 }
787
788 // Aborted due to thread limit being reached
789 if (state == ZOMBIE) {
790 thread->set_osthread(NULL);
791 delete osthread;
792 return false;
793 }
794
795 // The thread is returned suspended (in state INITIALIZED),
796 // and is started higher up in the call chain
797 assert(state == INITIALIZED, "race condition");
798 return true;
799 }
800
801 /////////////////////////////////////////////////////////////////////////////
802 // attach existing thread
803
804 // bootstrap the main thread
805 bool os::create_main_thread(JavaThread* thread) {
806 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
807 return create_attached_thread(thread);
808 }
809
810 bool os::create_attached_thread(JavaThread* thread) {
811 #ifdef ASSERT
812 thread->verify_not_published();
813 #endif
814
815 // Allocate the OSThread object
816 OSThread* osthread = new OSThread(NULL, NULL);
817
818 if (osthread == NULL) {
819 return false;
820 }
821
822 // Store pthread info into the OSThread
823 osthread->set_thread_id(os::Linux::gettid());
824 osthread->set_pthread_id(::pthread_self());
825
826 // initialize floating point control register
827 os::Linux::init_thread_fpu_state();
828
829 // Initial thread state is RUNNABLE
830 osthread->set_state(RUNNABLE);
831
832 thread->set_osthread(osthread);
833
834 if (UseNUMA) {
835 int lgrp_id = os::numa_get_group_id();
836 if (lgrp_id != -1) {
837 thread->set_lgrp_id(lgrp_id);
838 }
839 }
840
841 if (os::Linux::is_initial_thread()) {
842 // If current thread is initial thread, its stack is mapped on demand,
843 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
844 // the entire stack region to avoid SEGV in stack banging.
845 // It is also useful to get around the heap-stack-gap problem on SuSE
846 // kernel (see 4821821 for details). We first expand stack to the top
847 // of yellow zone, then enable stack yellow zone (order is significant,
848 // enabling yellow zone first will crash JVM on SuSE Linux), so there
849 // is no gap between the last two virtual memory regions.
850
851 JavaThread *jt = (JavaThread *)thread;
852 address addr = jt->stack_reserved_zone_base();
853 assert(addr != NULL, "initialization problem?");
854 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
855
856 osthread->set_expanding_stack();
857 os::Linux::manually_expand_stack(jt, addr);
858 osthread->clear_expanding_stack();
859 }
860
861 // initialize signal mask for this thread
862 // and save the caller's signal mask
863 os::Linux::hotspot_sigmask(thread);
864
865 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
866 os::current_thread_id(), (uintx) pthread_self());
867
868 return true;
869 }
870
871 void os::pd_start_thread(Thread* thread) {
872 OSThread * osthread = thread->osthread();
873 assert(osthread->get_state() != INITIALIZED, "just checking");
874 Monitor* sync_with_child = osthread->startThread_lock();
875 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
876 sync_with_child->notify();
877 }
878
879 // Free Linux resources related to the OSThread
880 void os::free_thread(OSThread* osthread) {
881 assert(osthread != NULL, "osthread not set");
882
883 if (Thread::current()->osthread() == osthread) {
884 // Restore caller's signal mask
885 sigset_t sigmask = osthread->caller_sigmask();
886 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
887 }
888
889 delete osthread;
890 }
891
892 //////////////////////////////////////////////////////////////////////////////
893 // initial thread
894
895 // Check if current thread is the initial thread, similar to Solaris thr_main.
896 bool os::Linux::is_initial_thread(void) {
897 char dummy;
898 // If called before init complete, thread stack bottom will be null.
899 // Can be called if fatal error occurs before initialization.
900 if (initial_thread_stack_bottom() == NULL) return false;
901 assert(initial_thread_stack_bottom() != NULL &&
902 initial_thread_stack_size() != 0,
903 "os::init did not locate initial thread's stack region");
904 if ((address)&dummy >= initial_thread_stack_bottom() &&
905 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
906 return true;
907 } else {
908 return false;
909 }
910 }
911
912 // Find the virtual memory area that contains addr
913 static bool find_vma(address addr, address* vma_low, address* vma_high) {
914 FILE *fp = fopen("/proc/self/maps", "r");
915 if (fp) {
916 address low, high;
917 while (!feof(fp)) {
918 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
919 if (low <= addr && addr < high) {
920 if (vma_low) *vma_low = low;
921 if (vma_high) *vma_high = high;
922 fclose(fp);
923 return true;
924 }
925 }
926 for (;;) {
927 int ch = fgetc(fp);
928 if (ch == EOF || ch == (int)'\n') break;
929 }
930 }
931 fclose(fp);
932 }
933 return false;
934 }
935
936 // Locate initial thread stack. This special handling of initial thread stack
937 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
938 // bogus value for initial thread.
939 void os::Linux::capture_initial_stack(size_t max_size) {
940 // stack size is the easy part, get it from RLIMIT_STACK
941 size_t stack_size;
942 struct rlimit rlim;
943 getrlimit(RLIMIT_STACK, &rlim);
944 stack_size = rlim.rlim_cur;
945
946 // 6308388: a bug in ld.so will relocate its own .data section to the
947 // lower end of primordial stack; reduce ulimit -s value a little bit
948 // so we won't install guard page on ld.so's data section.
949 stack_size -= 2 * page_size();
950
951 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
952 // 7.1, in both cases we will get 2G in return value.
953 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
954 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
955 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
956 // in case other parts in glibc still assumes 2M max stack size.
957 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
958 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
959 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
960 stack_size = 2 * K * K IA64_ONLY(*2);
961 }
962 // Try to figure out where the stack base (top) is. This is harder.
963 //
964 // When an application is started, glibc saves the initial stack pointer in
965 // a global variable "__libc_stack_end", which is then used by system
966 // libraries. __libc_stack_end should be pretty close to stack top. The
967 // variable is available since the very early days. However, because it is
968 // a private interface, it could disappear in the future.
969 //
970 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
971 // to __libc_stack_end, it is very close to stack top, but isn't the real
972 // stack top. Note that /proc may not exist if VM is running as a chroot
973 // program, so reading /proc/<pid>/stat could fail. Also the contents of
974 // /proc/<pid>/stat could change in the future (though unlikely).
975 //
976 // We try __libc_stack_end first. If that doesn't work, look for
977 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
978 // as a hint, which should work well in most cases.
979
980 uintptr_t stack_start;
981
982 // try __libc_stack_end first
983 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
984 if (p && *p) {
985 stack_start = *p;
986 } else {
987 // see if we can get the start_stack field from /proc/self/stat
988 FILE *fp;
989 int pid;
990 char state;
991 int ppid;
992 int pgrp;
993 int session;
994 int nr;
995 int tpgrp;
996 unsigned long flags;
997 unsigned long minflt;
998 unsigned long cminflt;
999 unsigned long majflt;
1000 unsigned long cmajflt;
1001 unsigned long utime;
1002 unsigned long stime;
1003 long cutime;
1004 long cstime;
1005 long prio;
1006 long nice;
1007 long junk;
1008 long it_real;
1009 uintptr_t start;
1010 uintptr_t vsize;
1011 intptr_t rss;
1012 uintptr_t rsslim;
1013 uintptr_t scodes;
1014 uintptr_t ecode;
1015 int i;
1016
1017 // Figure what the primordial thread stack base is. Code is inspired
1018 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1019 // followed by command name surrounded by parentheses, state, etc.
1020 char stat[2048];
1021 int statlen;
1022
1023 fp = fopen("/proc/self/stat", "r");
1024 if (fp) {
1025 statlen = fread(stat, 1, 2047, fp);
1026 stat[statlen] = '\0';
1027 fclose(fp);
1028
1029 // Skip pid and the command string. Note that we could be dealing with
1030 // weird command names, e.g. user could decide to rename java launcher
1031 // to "java 1.4.2 :)", then the stat file would look like
1032 // 1234 (java 1.4.2 :)) R ... ...
1033 // We don't really need to know the command string, just find the last
1034 // occurrence of ")" and then start parsing from there. See bug 4726580.
1035 char * s = strrchr(stat, ')');
1036
1037 i = 0;
1038 if (s) {
1039 // Skip blank chars
1040 do { s++; } while (s && isspace(*s));
1041
1042 #define _UFM UINTX_FORMAT
1043 #define _DFM INTX_FORMAT
1044
1045 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1046 // 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
1047 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,
1048 &state, // 3 %c
1049 &ppid, // 4 %d
1050 &pgrp, // 5 %d
1051 &session, // 6 %d
1052 &nr, // 7 %d
1053 &tpgrp, // 8 %d
1054 &flags, // 9 %lu
1055 &minflt, // 10 %lu
1056 &cminflt, // 11 %lu
1057 &majflt, // 12 %lu
1058 &cmajflt, // 13 %lu
1059 &utime, // 14 %lu
1060 &stime, // 15 %lu
1061 &cutime, // 16 %ld
1062 &cstime, // 17 %ld
1063 &prio, // 18 %ld
1064 &nice, // 19 %ld
1065 &junk, // 20 %ld
1066 &it_real, // 21 %ld
1067 &start, // 22 UINTX_FORMAT
1068 &vsize, // 23 UINTX_FORMAT
1069 &rss, // 24 INTX_FORMAT
1070 &rsslim, // 25 UINTX_FORMAT
1071 &scodes, // 26 UINTX_FORMAT
1072 &ecode, // 27 UINTX_FORMAT
1073 &stack_start); // 28 UINTX_FORMAT
1074 }
1075
1076 #undef _UFM
1077 #undef _DFM
1078
1079 if (i != 28 - 2) {
1080 assert(false, "Bad conversion from /proc/self/stat");
1081 // product mode - assume we are the initial thread, good luck in the
1082 // embedded case.
1083 warning("Can't detect initial thread stack location - bad conversion");
1084 stack_start = (uintptr_t) &rlim;
1085 }
1086 } else {
1087 // For some reason we can't open /proc/self/stat (for example, running on
1088 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1089 // most cases, so don't abort:
1090 warning("Can't detect initial thread stack location - no /proc/self/stat");
1091 stack_start = (uintptr_t) &rlim;
1092 }
1093 }
1094
1095 // Now we have a pointer (stack_start) very close to the stack top, the
1096 // next thing to do is to figure out the exact location of stack top. We
1097 // can find out the virtual memory area that contains stack_start by
1098 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1099 // and its upper limit is the real stack top. (again, this would fail if
1100 // running inside chroot, because /proc may not exist.)
1101
1102 uintptr_t stack_top;
1103 address low, high;
1104 if (find_vma((address)stack_start, &low, &high)) {
1105 // success, "high" is the true stack top. (ignore "low", because initial
1106 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1107 stack_top = (uintptr_t)high;
1108 } else {
1109 // failed, likely because /proc/self/maps does not exist
1110 warning("Can't detect initial thread stack location - find_vma failed");
1111 // best effort: stack_start is normally within a few pages below the real
1112 // stack top, use it as stack top, and reduce stack size so we won't put
1113 // guard page outside stack.
1114 stack_top = stack_start;
1115 stack_size -= 16 * page_size();
1116 }
1117
1118 // stack_top could be partially down the page so align it
1119 stack_top = align_size_up(stack_top, page_size());
1120
1121 if (max_size && stack_size > max_size) {
1122 _initial_thread_stack_size = max_size;
1123 } else {
1124 _initial_thread_stack_size = stack_size;
1125 }
1126
1127 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1128 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1129 }
1130
1131 ////////////////////////////////////////////////////////////////////////////////
1132 // time support
1133
1134 // Time since start-up in seconds to a fine granularity.
1135 // Used by VMSelfDestructTimer and the MemProfiler.
1136 double os::elapsedTime() {
1137
1138 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1139 }
1140
1141 jlong os::elapsed_counter() {
1142 return javaTimeNanos() - initial_time_count;
1143 }
1144
1145 jlong os::elapsed_frequency() {
1146 return NANOSECS_PER_SEC; // nanosecond resolution
1147 }
1148
1149 bool os::supports_vtime() { return true; }
1150 bool os::enable_vtime() { return false; }
1151 bool os::vtime_enabled() { return false; }
1152
1153 double os::elapsedVTime() {
1154 struct rusage usage;
1155 int retval = getrusage(RUSAGE_THREAD, &usage);
1156 if (retval == 0) {
1157 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);
1158 } else {
1159 // better than nothing, but not much
1160 return elapsedTime();
1161 }
1162 }
1163
1164 jlong os::javaTimeMillis() {
1165 timeval time;
1166 int status = gettimeofday(&time, NULL);
1167 assert(status != -1, "linux error");
1168 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1169 }
1170
1171 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1172 timeval time;
1173 int status = gettimeofday(&time, NULL);
1174 assert(status != -1, "linux error");
1175 seconds = jlong(time.tv_sec);
1176 nanos = jlong(time.tv_usec) * 1000;
1177 }
1178
1179
1180 #ifndef CLOCK_MONOTONIC
1181 #define CLOCK_MONOTONIC (1)
1182 #endif
1183
1184 void os::Linux::clock_init() {
1185 // we do dlopen's in this particular order due to bug in linux
1186 // dynamical loader (see 6348968) leading to crash on exit
1187 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1188 if (handle == NULL) {
1189 handle = dlopen("librt.so", RTLD_LAZY);
1190 }
1191
1192 if (handle) {
1193 int (*clock_getres_func)(clockid_t, struct timespec*) =
1194 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1195 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1196 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1197 if (clock_getres_func && clock_gettime_func) {
1198 // See if monotonic clock is supported by the kernel. Note that some
1199 // early implementations simply return kernel jiffies (updated every
1200 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1201 // for nano time (though the monotonic property is still nice to have).
1202 // It's fixed in newer kernels, however clock_getres() still returns
1203 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1204 // resolution for now. Hopefully as people move to new kernels, this
1205 // won't be a problem.
1206 struct timespec res;
1207 struct timespec tp;
1208 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1209 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1210 // yes, monotonic clock is supported
1211 _clock_gettime = clock_gettime_func;
1212 return;
1213 } else {
1214 // close librt if there is no monotonic clock
1215 dlclose(handle);
1216 }
1217 }
1218 }
1219 warning("No monotonic clock was available - timed services may " \
1220 "be adversely affected if the time-of-day clock changes");
1221 }
1222
1223 #ifndef SYS_clock_getres
1224 #if defined(IA32) || defined(AMD64)
1225 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1226 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1227 #else
1228 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1229 #define sys_clock_getres(x,y) -1
1230 #endif
1231 #else
1232 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1233 #endif
1234
1235 void os::Linux::fast_thread_clock_init() {
1236 if (!UseLinuxPosixThreadCPUClocks) {
1237 return;
1238 }
1239 clockid_t clockid;
1240 struct timespec tp;
1241 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1242 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1243
1244 // Switch to using fast clocks for thread cpu time if
1245 // the sys_clock_getres() returns 0 error code.
1246 // Note, that some kernels may support the current thread
1247 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1248 // returned by the pthread_getcpuclockid().
1249 // If the fast Posix clocks are supported then the sys_clock_getres()
1250 // must return at least tp.tv_sec == 0 which means a resolution
1251 // better than 1 sec. This is extra check for reliability.
1252
1253 if (pthread_getcpuclockid_func &&
1254 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1255 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1256 _supports_fast_thread_cpu_time = true;
1257 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1258 }
1259 }
1260
1261 jlong os::javaTimeNanos() {
1262 if (os::supports_monotonic_clock()) {
1263 struct timespec tp;
1264 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1265 assert(status == 0, "gettime error");
1266 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1267 return result;
1268 } else {
1269 timeval time;
1270 int status = gettimeofday(&time, NULL);
1271 assert(status != -1, "linux error");
1272 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1273 return 1000 * usecs;
1274 }
1275 }
1276
1277 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1278 if (os::supports_monotonic_clock()) {
1279 info_ptr->max_value = ALL_64_BITS;
1280
1281 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1282 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1283 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1284 } else {
1285 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1286 info_ptr->max_value = ALL_64_BITS;
1287
1288 // gettimeofday is a real time clock so it skips
1289 info_ptr->may_skip_backward = true;
1290 info_ptr->may_skip_forward = true;
1291 }
1292
1293 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1294 }
1295
1296 // Return the real, user, and system times in seconds from an
1297 // arbitrary fixed point in the past.
1298 bool os::getTimesSecs(double* process_real_time,
1299 double* process_user_time,
1300 double* process_system_time) {
1301 struct tms ticks;
1302 clock_t real_ticks = times(&ticks);
1303
1304 if (real_ticks == (clock_t) (-1)) {
1305 return false;
1306 } else {
1307 double ticks_per_second = (double) clock_tics_per_sec;
1308 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1309 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1310 *process_real_time = ((double) real_ticks) / ticks_per_second;
1311
1312 return true;
1313 }
1314 }
1315
1316
1317 char * os::local_time_string(char *buf, size_t buflen) {
1318 struct tm t;
1319 time_t long_time;
1320 time(&long_time);
1321 localtime_r(&long_time, &t);
1322 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1323 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1324 t.tm_hour, t.tm_min, t.tm_sec);
1325 return buf;
1326 }
1327
1328 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1329 return localtime_r(clock, res);
1330 }
1331
1332 ////////////////////////////////////////////////////////////////////////////////
1333 // runtime exit support
1334
1335 // Note: os::shutdown() might be called very early during initialization, or
1336 // called from signal handler. Before adding something to os::shutdown(), make
1337 // sure it is async-safe and can handle partially initialized VM.
1338 void os::shutdown() {
1339
1340 // allow PerfMemory to attempt cleanup of any persistent resources
1341 perfMemory_exit();
1342
1343 // needs to remove object in file system
1344 AttachListener::abort();
1345
1346 // flush buffered output, finish log files
1347 ostream_abort();
1348
1349 // Check for abort hook
1350 abort_hook_t abort_hook = Arguments::abort_hook();
1351 if (abort_hook != NULL) {
1352 abort_hook();
1353 }
1354
1355 }
1356
1357 // Note: os::abort() might be called very early during initialization, or
1358 // called from signal handler. Before adding something to os::abort(), make
1359 // sure it is async-safe and can handle partially initialized VM.
1360 void os::abort(bool dump_core, void* siginfo, const void* context) {
1361 os::shutdown();
1362 if (dump_core) {
1363 #ifndef PRODUCT
1364 fdStream out(defaultStream::output_fd());
1365 out.print_raw("Current thread is ");
1366 char buf[16];
1367 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1368 out.print_raw_cr(buf);
1369 out.print_raw_cr("Dumping core ...");
1370 #endif
1371 ::abort(); // dump core
1372 }
1373
1374 ::exit(1);
1375 }
1376
1377 // Die immediately, no exit hook, no abort hook, no cleanup.
1378 void os::die() {
1379 ::abort();
1380 }
1381
1382
1383 // This method is a copy of JDK's sysGetLastErrorString
1384 // from src/solaris/hpi/src/system_md.c
1385
1386 size_t os::lasterror(char *buf, size_t len) {
1387 if (errno == 0) return 0;
1388
1389 const char *s = ::strerror(errno);
1390 size_t n = ::strlen(s);
1391 if (n >= len) {
1392 n = len - 1;
1393 }
1394 ::strncpy(buf, s, n);
1395 buf[n] = '\0';
1396 return n;
1397 }
1398
1399 // thread_id is kernel thread id (similar to Solaris LWP id)
1400 intx os::current_thread_id() { return os::Linux::gettid(); }
1401 int os::current_process_id() {
1402 return ::getpid();
1403 }
1404
1405 // DLL functions
1406
1407 const char* os::dll_file_extension() { return ".so"; }
1408
1409 // This must be hard coded because it's the system's temporary
1410 // directory not the java application's temp directory, ala java.io.tmpdir.
1411 const char* os::get_temp_directory() { return "/tmp"; }
1412
1413 static bool file_exists(const char* filename) {
1414 struct stat statbuf;
1415 if (filename == NULL || strlen(filename) == 0) {
1416 return false;
1417 }
1418 return os::stat(filename, &statbuf) == 0;
1419 }
1420
1421 bool os::dll_build_name(char* buffer, size_t buflen,
1422 const char* pname, const char* fname) {
1423 bool retval = false;
1424 // Copied from libhpi
1425 const size_t pnamelen = pname ? strlen(pname) : 0;
1426
1427 // Return error on buffer overflow.
1428 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1429 return retval;
1430 }
1431
1432 if (pnamelen == 0) {
1433 snprintf(buffer, buflen, "lib%s.so", fname);
1434 retval = true;
1435 } else if (strchr(pname, *os::path_separator()) != NULL) {
1436 int n;
1437 char** pelements = split_path(pname, &n);
1438 if (pelements == NULL) {
1439 return false;
1440 }
1441 for (int i = 0; i < n; i++) {
1442 // Really shouldn't be NULL, but check can't hurt
1443 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1444 continue; // skip the empty path values
1445 }
1446 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1447 if (file_exists(buffer)) {
1448 retval = true;
1449 break;
1450 }
1451 }
1452 // release the storage
1453 for (int i = 0; i < n; i++) {
1454 if (pelements[i] != NULL) {
1455 FREE_C_HEAP_ARRAY(char, pelements[i]);
1456 }
1457 }
1458 if (pelements != NULL) {
1459 FREE_C_HEAP_ARRAY(char*, pelements);
1460 }
1461 } else {
1462 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1463 retval = true;
1464 }
1465 return retval;
1466 }
1467
1468 // check if addr is inside libjvm.so
1469 bool os::address_is_in_vm(address addr) {
1470 static address libjvm_base_addr;
1471 Dl_info dlinfo;
1472
1473 if (libjvm_base_addr == NULL) {
1474 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1475 libjvm_base_addr = (address)dlinfo.dli_fbase;
1476 }
1477 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1478 }
1479
1480 if (dladdr((void *)addr, &dlinfo) != 0) {
1481 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1482 }
1483
1484 return false;
1485 }
1486
1487 bool os::dll_address_to_function_name(address addr, char *buf,
1488 int buflen, int *offset,
1489 bool demangle) {
1490 // buf is not optional, but offset is optional
1491 assert(buf != NULL, "sanity check");
1492
1493 Dl_info dlinfo;
1494
1495 if (dladdr((void*)addr, &dlinfo) != 0) {
1496 // see if we have a matching symbol
1497 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1498 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1499 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1500 }
1501 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1502 return true;
1503 }
1504 // no matching symbol so try for just file info
1505 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1506 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1507 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1508 return true;
1509 }
1510 }
1511 }
1512
1513 buf[0] = '\0';
1514 if (offset != NULL) *offset = -1;
1515 return false;
1516 }
1517
1518 struct _address_to_library_name {
1519 address addr; // input : memory address
1520 size_t buflen; // size of fname
1521 char* fname; // output: library name
1522 address base; // library base addr
1523 };
1524
1525 static int address_to_library_name_callback(struct dl_phdr_info *info,
1526 size_t size, void *data) {
1527 int i;
1528 bool found = false;
1529 address libbase = NULL;
1530 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1531
1532 // iterate through all loadable segments
1533 for (i = 0; i < info->dlpi_phnum; i++) {
1534 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1535 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1536 // base address of a library is the lowest address of its loaded
1537 // segments.
1538 if (libbase == NULL || libbase > segbase) {
1539 libbase = segbase;
1540 }
1541 // see if 'addr' is within current segment
1542 if (segbase <= d->addr &&
1543 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1544 found = true;
1545 }
1546 }
1547 }
1548
1549 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1550 // so dll_address_to_library_name() can fall through to use dladdr() which
1551 // can figure out executable name from argv[0].
1552 if (found && info->dlpi_name && info->dlpi_name[0]) {
1553 d->base = libbase;
1554 if (d->fname) {
1555 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1556 }
1557 return 1;
1558 }
1559 return 0;
1560 }
1561
1562 bool os::dll_address_to_library_name(address addr, char* buf,
1563 int buflen, int* offset) {
1564 // buf is not optional, but offset is optional
1565 assert(buf != NULL, "sanity check");
1566
1567 Dl_info dlinfo;
1568 struct _address_to_library_name data;
1569
1570 // There is a bug in old glibc dladdr() implementation that it could resolve
1571 // to wrong library name if the .so file has a base address != NULL. Here
1572 // we iterate through the program headers of all loaded libraries to find
1573 // out which library 'addr' really belongs to. This workaround can be
1574 // removed once the minimum requirement for glibc is moved to 2.3.x.
1575 data.addr = addr;
1576 data.fname = buf;
1577 data.buflen = buflen;
1578 data.base = NULL;
1579 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1580
1581 if (rslt) {
1582 // buf already contains library name
1583 if (offset) *offset = addr - data.base;
1584 return true;
1585 }
1586 if (dladdr((void*)addr, &dlinfo) != 0) {
1587 if (dlinfo.dli_fname != NULL) {
1588 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1589 }
1590 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1591 *offset = addr - (address)dlinfo.dli_fbase;
1592 }
1593 return true;
1594 }
1595
1596 buf[0] = '\0';
1597 if (offset) *offset = -1;
1598 return false;
1599 }
1600
1601 // Loads .dll/.so and
1602 // in case of error it checks if .dll/.so was built for the
1603 // same architecture as Hotspot is running on
1604
1605
1606 // Remember the stack's state. The Linux dynamic linker will change
1607 // the stack to 'executable' at most once, so we must safepoint only once.
1608 bool os::Linux::_stack_is_executable = false;
1609
1610 // VM operation that loads a library. This is necessary if stack protection
1611 // of the Java stacks can be lost during loading the library. If we
1612 // do not stop the Java threads, they can stack overflow before the stacks
1613 // are protected again.
1614 class VM_LinuxDllLoad: public VM_Operation {
1615 private:
1616 const char *_filename;
1617 char *_ebuf;
1618 int _ebuflen;
1619 void *_lib;
1620 public:
1621 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1622 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1623 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1624 void doit() {
1625 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1626 os::Linux::_stack_is_executable = true;
1627 }
1628 void* loaded_library() { return _lib; }
1629 };
1630
1631 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1632 void * result = NULL;
1633 bool load_attempted = false;
1634
1635 // Check whether the library to load might change execution rights
1636 // of the stack. If they are changed, the protection of the stack
1637 // guard pages will be lost. We need a safepoint to fix this.
1638 //
1639 // See Linux man page execstack(8) for more info.
1640 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1641 ElfFile ef(filename);
1642 if (!ef.specifies_noexecstack()) {
1643 if (!is_init_completed()) {
1644 os::Linux::_stack_is_executable = true;
1645 // This is OK - No Java threads have been created yet, and hence no
1646 // stack guard pages to fix.
1647 //
1648 // This should happen only when you are building JDK7 using a very
1649 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1650 //
1651 // Dynamic loader will make all stacks executable after
1652 // this function returns, and will not do that again.
1653 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1654 } else {
1655 warning("You have loaded library %s which might have disabled stack guard. "
1656 "The VM will try to fix the stack guard now.\n"
1657 "It's highly recommended that you fix the library with "
1658 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1659 filename);
1660
1661 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1662 JavaThread *jt = JavaThread::current();
1663 if (jt->thread_state() != _thread_in_native) {
1664 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1665 // that requires ExecStack. Cannot enter safe point. Let's give up.
1666 warning("Unable to fix stack guard. Giving up.");
1667 } else {
1668 if (!LoadExecStackDllInVMThread) {
1669 // This is for the case where the DLL has an static
1670 // constructor function that executes JNI code. We cannot
1671 // load such DLLs in the VMThread.
1672 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1673 }
1674
1675 ThreadInVMfromNative tiv(jt);
1676 debug_only(VMNativeEntryWrapper vew;)
1677
1678 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1679 VMThread::execute(&op);
1680 if (LoadExecStackDllInVMThread) {
1681 result = op.loaded_library();
1682 }
1683 load_attempted = true;
1684 }
1685 }
1686 }
1687 }
1688
1689 if (!load_attempted) {
1690 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691 }
1692
1693 if (result != NULL) {
1694 // Successful loading
1695 return result;
1696 }
1697
1698 Elf32_Ehdr elf_head;
1699 int diag_msg_max_length=ebuflen-strlen(ebuf);
1700 char* diag_msg_buf=ebuf+strlen(ebuf);
1701
1702 if (diag_msg_max_length==0) {
1703 // No more space in ebuf for additional diagnostics message
1704 return NULL;
1705 }
1706
1707
1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1709
1710 if (file_descriptor < 0) {
1711 // Can't open library, report dlerror() message
1712 return NULL;
1713 }
1714
1715 bool failed_to_read_elf_head=
1716 (sizeof(elf_head)!=
1717 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1718
1719 ::close(file_descriptor);
1720 if (failed_to_read_elf_head) {
1721 // file i/o error - report dlerror() msg
1722 return NULL;
1723 }
1724
1725 typedef struct {
1726 Elf32_Half code; // Actual value as defined in elf.h
1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1728 char elf_class; // 32 or 64 bit
1729 char endianess; // MSB or LSB
1730 char* name; // String representation
1731 } arch_t;
1732
1733 #ifndef EM_486
1734 #define EM_486 6 /* Intel 80486 */
1735 #endif
1736 #ifndef EM_AARCH64
1737 #define EM_AARCH64 183 /* ARM AARCH64 */
1738 #endif
1739
1740 static const arch_t arch_array[]={
1741 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1742 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1743 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1744 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1745 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1746 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1747 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1748 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1749 #if defined(VM_LITTLE_ENDIAN)
1750 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1751 #else
1752 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1753 #endif
1754 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1755 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1756 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1757 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1758 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1759 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1760 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1761 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1762 };
1763
1764 #if (defined IA32)
1765 static Elf32_Half running_arch_code=EM_386;
1766 #elif (defined AMD64)
1767 static Elf32_Half running_arch_code=EM_X86_64;
1768 #elif (defined IA64)
1769 static Elf32_Half running_arch_code=EM_IA_64;
1770 #elif (defined __sparc) && (defined _LP64)
1771 static Elf32_Half running_arch_code=EM_SPARCV9;
1772 #elif (defined __sparc) && (!defined _LP64)
1773 static Elf32_Half running_arch_code=EM_SPARC;
1774 #elif (defined __powerpc64__)
1775 static Elf32_Half running_arch_code=EM_PPC64;
1776 #elif (defined __powerpc__)
1777 static Elf32_Half running_arch_code=EM_PPC;
1778 #elif (defined ARM)
1779 static Elf32_Half running_arch_code=EM_ARM;
1780 #elif (defined S390)
1781 static Elf32_Half running_arch_code=EM_S390;
1782 #elif (defined ALPHA)
1783 static Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif (defined MIPSEL)
1785 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif (defined PARISC)
1787 static Elf32_Half running_arch_code=EM_PARISC;
1788 #elif (defined MIPS)
1789 static Elf32_Half running_arch_code=EM_MIPS;
1790 #elif (defined M68K)
1791 static Elf32_Half running_arch_code=EM_68K;
1792 #elif (defined AARCH64)
1793 static Elf32_Half running_arch_code=EM_AARCH64;
1794 #else
1795 #error Method os::dll_load requires that one of following is defined:\
1796 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1797 #endif
1798
1799 // Identify compatability class for VM's architecture and library's architecture
1800 // Obtain string descriptions for architectures
1801
1802 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803 int running_arch_index=-1;
1804
1805 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806 if (running_arch_code == arch_array[i].code) {
1807 running_arch_index = i;
1808 }
1809 if (lib_arch.code == arch_array[i].code) {
1810 lib_arch.compat_class = arch_array[i].compat_class;
1811 lib_arch.name = arch_array[i].name;
1812 }
1813 }
1814
1815 assert(running_arch_index != -1,
1816 "Didn't find running architecture code (running_arch_code) in arch_array");
1817 if (running_arch_index == -1) {
1818 // Even though running architecture detection failed
1819 // we may still continue with reporting dlerror() message
1820 return NULL;
1821 }
1822
1823 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825 return NULL;
1826 }
1827
1828 #ifndef S390
1829 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831 return NULL;
1832 }
1833 #endif // !S390
1834
1835 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836 if (lib_arch.name!=NULL) {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839 lib_arch.name, arch_array[running_arch_index].name);
1840 } else {
1841 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843 lib_arch.code,
1844 arch_array[running_arch_index].name);
1845 }
1846 }
1847
1848 return NULL;
1849 }
1850
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852 int ebuflen) {
1853 void * result = ::dlopen(filename, RTLD_LAZY);
1854 if (result == NULL) {
1855 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856 ebuf[ebuflen-1] = '\0';
1857 }
1858 return result;
1859 }
1860
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862 int ebuflen) {
1863 void * result = NULL;
1864 if (LoadExecStackDllInVMThread) {
1865 result = dlopen_helper(filename, ebuf, ebuflen);
1866 }
1867
1868 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869 // library that requires an executable stack, or which does not have this
1870 // stack attribute set, dlopen changes the stack attribute to executable. The
1871 // read protection of the guard pages gets lost.
1872 //
1873 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874 // may have been queued at the same time.
1875
1876 if (!_stack_is_executable) {
1877 JavaThread *jt = Threads::first();
1878
1879 while (jt) {
1880 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1881 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1882 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1883 warning("Attempt to reguard stack yellow zone failed.");
1884 }
1885 }
1886 jt = jt->next();
1887 }
1888 }
1889
1890 return result;
1891 }
1892
1893 void* os::dll_lookup(void* handle, const char* name) {
1894 void* res = dlsym(handle, name);
1895 return res;
1896 }
1897
1898 void* os::get_default_process_handle() {
1899 return (void*)::dlopen(NULL, RTLD_LAZY);
1900 }
1901
1902 static bool _print_ascii_file(const char* filename, outputStream* st) {
1903 int fd = ::open(filename, O_RDONLY);
1904 if (fd == -1) {
1905 return false;
1906 }
1907
1908 char buf[32];
1909 int bytes;
1910 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1911 st->print_raw(buf, bytes);
1912 }
1913
1914 ::close(fd);
1915
1916 return true;
1917 }
1918
1919 void os::print_dll_info(outputStream *st) {
1920 st->print_cr("Dynamic libraries:");
1921
1922 char fname[32];
1923 pid_t pid = os::Linux::gettid();
1924
1925 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1926
1927 if (!_print_ascii_file(fname, st)) {
1928 st->print("Can not get library information for pid = %d\n", pid);
1929 }
1930 }
1931
1932 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1933 FILE *procmapsFile = NULL;
1934
1935 // Open the procfs maps file for the current process
1936 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1937 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1938 char line[PATH_MAX + 100];
1939
1940 // Read line by line from 'file'
1941 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1942 u8 base, top, offset, inode;
1943 char permissions[5];
1944 char device[6];
1945 char name[PATH_MAX + 1];
1946
1947 // Parse fields from line
1948 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1949 &base, &top, permissions, &offset, device, &inode, name);
1950
1951 // Filter by device id '00:00' so that we only get file system mapped files.
1952 if (strcmp(device, "00:00") != 0) {
1953
1954 // Call callback with the fields of interest
1955 if(callback(name, (address)base, (address)top, param)) {
1956 // Oops abort, callback aborted
1957 fclose(procmapsFile);
1958 return 1;
1959 }
1960 }
1961 }
1962 fclose(procmapsFile);
1963 }
1964 return 0;
1965 }
1966
1967 void os::print_os_info_brief(outputStream* st) {
1968 os::Linux::print_distro_info(st);
1969
1970 os::Posix::print_uname_info(st);
1971
1972 os::Linux::print_libversion_info(st);
1973
1974 }
1975
1976 void os::print_os_info(outputStream* st) {
1977 st->print("OS:");
1978
1979 os::Linux::print_distro_info(st);
1980
1981 os::Posix::print_uname_info(st);
1982
1983 // Print warning if unsafe chroot environment detected
1984 if (unsafe_chroot_detected) {
1985 st->print("WARNING!! ");
1986 st->print_cr("%s", unstable_chroot_error);
1987 }
1988
1989 os::Linux::print_libversion_info(st);
1990
1991 os::Posix::print_rlimit_info(st);
1992
1993 os::Posix::print_load_average(st);
1994
1995 os::Linux::print_full_memory_info(st);
1996 }
1997
1998 // Try to identify popular distros.
1999 // Most Linux distributions have a /etc/XXX-release file, which contains
2000 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2001 // file that also contains the OS version string. Some have more than one
2002 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2003 // /etc/redhat-release.), so the order is important.
2004 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2005 // their own specific XXX-release file as well as a redhat-release file.
2006 // Because of this the XXX-release file needs to be searched for before the
2007 // redhat-release file.
2008 // Since Red Hat has a lsb-release file that is not very descriptive the
2009 // search for redhat-release needs to be before lsb-release.
2010 // Since the lsb-release file is the new standard it needs to be searched
2011 // before the older style release files.
2012 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2013 // next to last resort. The os-release file is a new standard that contains
2014 // distribution information and the system-release file seems to be an old
2015 // standard that has been replaced by the lsb-release and os-release files.
2016 // Searching for the debian_version file is the last resort. It contains
2017 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2018 // "Debian " is printed before the contents of the debian_version file.
2019
2020 const char* distro_files[] = {
2021 "/etc/oracle-release",
2022 "/etc/mandriva-release",
2023 "/etc/mandrake-release",
2024 "/etc/sun-release",
2025 "/etc/redhat-release",
2026 "/etc/lsb-release",
2027 "/etc/SuSE-release",
2028 "/etc/turbolinux-release",
2029 "/etc/gentoo-release",
2030 "/etc/ltib-release",
2031 "/etc/angstrom-version",
2032 "/etc/system-release",
2033 "/etc/os-release",
2034 NULL };
2035
2036 void os::Linux::print_distro_info(outputStream* st) {
2037 for (int i = 0;; i++) {
2038 const char* file = distro_files[i];
2039 if (file == NULL) {
2040 break; // done
2041 }
2042 // If file prints, we found it.
2043 if (_print_ascii_file(file, st)) {
2044 return;
2045 }
2046 }
2047
2048 if (file_exists("/etc/debian_version")) {
2049 st->print("Debian ");
2050 _print_ascii_file("/etc/debian_version", st);
2051 } else {
2052 st->print("Linux");
2053 }
2054 st->cr();
2055 }
2056
2057 static void parse_os_info(char* distro, size_t length, const char* file) {
2058 FILE* fp = fopen(file, "r");
2059 if (fp != NULL) {
2060 char buf[256];
2061 // get last line of the file.
2062 while (fgets(buf, sizeof(buf), fp)) { }
2063 // Edit out extra stuff in expected ubuntu format
2064 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2065 char* ptr = strstr(buf, "\""); // the name is in quotes
2066 if (ptr != NULL) {
2067 ptr++; // go beyond first quote
2068 char* nl = strchr(ptr, '\"');
2069 if (nl != NULL) *nl = '\0';
2070 strncpy(distro, ptr, length);
2071 } else {
2072 ptr = strstr(buf, "=");
2073 ptr++; // go beyond equals then
2074 char* nl = strchr(ptr, '\n');
2075 if (nl != NULL) *nl = '\0';
2076 strncpy(distro, ptr, length);
2077 }
2078 } else {
2079 // if not in expected Ubuntu format, print out whole line minus \n
2080 char* nl = strchr(buf, '\n');
2081 if (nl != NULL) *nl = '\0';
2082 strncpy(distro, buf, length);
2083 }
2084 // close distro file
2085 fclose(fp);
2086 }
2087 }
2088
2089 void os::get_summary_os_info(char* buf, size_t buflen) {
2090 for (int i = 0;; i++) {
2091 const char* file = distro_files[i];
2092 if (file == NULL) {
2093 break; // ran out of distro_files
2094 }
2095 if (file_exists(file)) {
2096 parse_os_info(buf, buflen, file);
2097 return;
2098 }
2099 }
2100 // special case for debian
2101 if (file_exists("/etc/debian_version")) {
2102 strncpy(buf, "Debian ", buflen);
2103 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2104 } else {
2105 strncpy(buf, "Linux", buflen);
2106 }
2107 }
2108
2109 void os::Linux::print_libversion_info(outputStream* st) {
2110 // libc, pthread
2111 st->print("libc:");
2112 st->print("%s ", os::Linux::glibc_version());
2113 st->print("%s ", os::Linux::libpthread_version());
2114 st->cr();
2115 }
2116
2117 void os::Linux::print_full_memory_info(outputStream* st) {
2118 st->print("\n/proc/meminfo:\n");
2119 _print_ascii_file("/proc/meminfo", st);
2120 st->cr();
2121 }
2122
2123 void os::print_memory_info(outputStream* st) {
2124
2125 st->print("Memory:");
2126 st->print(" %dk page", os::vm_page_size()>>10);
2127
2128 // values in struct sysinfo are "unsigned long"
2129 struct sysinfo si;
2130 sysinfo(&si);
2131
2132 st->print(", physical " UINT64_FORMAT "k",
2133 os::physical_memory() >> 10);
2134 st->print("(" UINT64_FORMAT "k free)",
2135 os::available_memory() >> 10);
2136 st->print(", swap " UINT64_FORMAT "k",
2137 ((jlong)si.totalswap * si.mem_unit) >> 10);
2138 st->print("(" UINT64_FORMAT "k free)",
2139 ((jlong)si.freeswap * si.mem_unit) >> 10);
2140 st->cr();
2141 }
2142
2143 // Print the first "model name" line and the first "flags" line
2144 // that we find and nothing more. We assume "model name" comes
2145 // before "flags" so if we find a second "model name", then the
2146 // "flags" field is considered missing.
2147 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2148 #if defined(IA32) || defined(AMD64)
2149 // Other platforms have less repetitive cpuinfo files
2150 FILE *fp = fopen("/proc/cpuinfo", "r");
2151 if (fp) {
2152 while (!feof(fp)) {
2153 if (fgets(buf, buflen, fp)) {
2154 // Assume model name comes before flags
2155 bool model_name_printed = false;
2156 if (strstr(buf, "model name") != NULL) {
2157 if (!model_name_printed) {
2158 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2159 st->print_raw(buf);
2160 model_name_printed = true;
2161 } else {
2162 // model name printed but not flags? Odd, just return
2163 fclose(fp);
2164 return true;
2165 }
2166 }
2167 // print the flags line too
2168 if (strstr(buf, "flags") != NULL) {
2169 st->print_raw(buf);
2170 fclose(fp);
2171 return true;
2172 }
2173 }
2174 }
2175 fclose(fp);
2176 }
2177 #endif // x86 platforms
2178 return false;
2179 }
2180
2181 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2182 // Only print the model name if the platform provides this as a summary
2183 if (!print_model_name_and_flags(st, buf, buflen)) {
2184 st->print("\n/proc/cpuinfo:\n");
2185 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2186 st->print_cr(" <Not Available>");
2187 }
2188 }
2189 }
2190
2191 #if defined(AMD64) || defined(IA32) || defined(X32)
2192 const char* search_string = "model name";
2193 #elif defined(SPARC)
2194 const char* search_string = "cpu";
2195 #elif defined(PPC64)
2196 const char* search_string = "cpu";
2197 #else
2198 const char* search_string = "Processor";
2199 #endif
2200
2201 // Parses the cpuinfo file for string representing the model name.
2202 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2203 FILE* fp = fopen("/proc/cpuinfo", "r");
2204 if (fp != NULL) {
2205 while (!feof(fp)) {
2206 char buf[256];
2207 if (fgets(buf, sizeof(buf), fp)) {
2208 char* start = strstr(buf, search_string);
2209 if (start != NULL) {
2210 char *ptr = start + strlen(search_string);
2211 char *end = buf + strlen(buf);
2212 while (ptr != end) {
2213 // skip whitespace and colon for the rest of the name.
2214 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2215 break;
2216 }
2217 ptr++;
2218 }
2219 if (ptr != end) {
2220 // reasonable string, get rid of newline and keep the rest
2221 char* nl = strchr(buf, '\n');
2222 if (nl != NULL) *nl = '\0';
2223 strncpy(cpuinfo, ptr, length);
2224 fclose(fp);
2225 return;
2226 }
2227 }
2228 }
2229 }
2230 fclose(fp);
2231 }
2232 // cpuinfo not found or parsing failed, just print generic string. The entire
2233 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2234 #if defined(AMD64)
2235 strncpy(cpuinfo, "x86_64", length);
2236 #elif defined(IA32)
2237 strncpy(cpuinfo, "x86_32", length);
2238 #elif defined(IA64)
2239 strncpy(cpuinfo, "IA64", length);
2240 #elif defined(SPARC)
2241 strncpy(cpuinfo, "sparcv9", length);
2242 #elif defined(AARCH64)
2243 strncpy(cpuinfo, "AArch64", length);
2244 #elif defined(ARM)
2245 strncpy(cpuinfo, "ARM", length);
2246 #elif defined(PPC)
2247 strncpy(cpuinfo, "PPC64", length);
2248 #elif defined(ZERO_LIBARCH)
2249 strncpy(cpuinfo, ZERO_LIBARCH, length);
2250 #else
2251 strncpy(cpuinfo, "unknown", length);
2252 #endif
2253 }
2254
2255 static void print_signal_handler(outputStream* st, int sig,
2256 char* buf, size_t buflen);
2257
2258 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2259 st->print_cr("Signal Handlers:");
2260 print_signal_handler(st, SIGSEGV, buf, buflen);
2261 print_signal_handler(st, SIGBUS , buf, buflen);
2262 print_signal_handler(st, SIGFPE , buf, buflen);
2263 print_signal_handler(st, SIGPIPE, buf, buflen);
2264 print_signal_handler(st, SIGXFSZ, buf, buflen);
2265 print_signal_handler(st, SIGILL , buf, buflen);
2266 print_signal_handler(st, SR_signum, buf, buflen);
2267 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2268 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2269 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2270 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2271 #if defined(PPC64)
2272 print_signal_handler(st, SIGTRAP, buf, buflen);
2273 #endif
2274 }
2275
2276 static char saved_jvm_path[MAXPATHLEN] = {0};
2277
2278 // Find the full path to the current module, libjvm.so
2279 void os::jvm_path(char *buf, jint buflen) {
2280 // Error checking.
2281 if (buflen < MAXPATHLEN) {
2282 assert(false, "must use a large-enough buffer");
2283 buf[0] = '\0';
2284 return;
2285 }
2286 // Lazy resolve the path to current module.
2287 if (saved_jvm_path[0] != 0) {
2288 strcpy(buf, saved_jvm_path);
2289 return;
2290 }
2291
2292 char dli_fname[MAXPATHLEN];
2293 bool ret = dll_address_to_library_name(
2294 CAST_FROM_FN_PTR(address, os::jvm_path),
2295 dli_fname, sizeof(dli_fname), NULL);
2296 assert(ret, "cannot locate libjvm");
2297 char *rp = NULL;
2298 if (ret && dli_fname[0] != '\0') {
2299 rp = realpath(dli_fname, buf);
2300 }
2301 if (rp == NULL) {
2302 return;
2303 }
2304
2305 if (Arguments::sun_java_launcher_is_altjvm()) {
2306 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2307 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2308 // If "/jre/lib/" appears at the right place in the string, then
2309 // assume we are installed in a JDK and we're done. Otherwise, check
2310 // for a JAVA_HOME environment variable and fix up the path so it
2311 // looks like libjvm.so is installed there (append a fake suffix
2312 // hotspot/libjvm.so).
2313 const char *p = buf + strlen(buf) - 1;
2314 for (int count = 0; p > buf && count < 5; ++count) {
2315 for (--p; p > buf && *p != '/'; --p)
2316 /* empty */ ;
2317 }
2318
2319 if (strncmp(p, "/jre/lib/", 9) != 0) {
2320 // Look for JAVA_HOME in the environment.
2321 char* java_home_var = ::getenv("JAVA_HOME");
2322 if (java_home_var != NULL && java_home_var[0] != 0) {
2323 char* jrelib_p;
2324 int len;
2325
2326 // Check the current module name "libjvm.so".
2327 p = strrchr(buf, '/');
2328 if (p == NULL) {
2329 return;
2330 }
2331 assert(strstr(p, "/libjvm") == p, "invalid library name");
2332
2333 rp = realpath(java_home_var, buf);
2334 if (rp == NULL) {
2335 return;
2336 }
2337
2338 // determine if this is a legacy image or modules image
2339 // modules image doesn't have "jre" subdirectory
2340 len = strlen(buf);
2341 assert(len < buflen, "Ran out of buffer room");
2342 jrelib_p = buf + len;
2343 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2344 if (0 != access(buf, F_OK)) {
2345 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2346 }
2347
2348 if (0 == access(buf, F_OK)) {
2349 // Use current module name "libjvm.so"
2350 len = strlen(buf);
2351 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2352 } else {
2353 // Go back to path of .so
2354 rp = realpath(dli_fname, buf);
2355 if (rp == NULL) {
2356 return;
2357 }
2358 }
2359 }
2360 }
2361 }
2362
2363 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2364 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2365 }
2366
2367 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2368 // no prefix required, not even "_"
2369 }
2370
2371 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2372 // no suffix required
2373 }
2374
2375 ////////////////////////////////////////////////////////////////////////////////
2376 // sun.misc.Signal support
2377
2378 static volatile jint sigint_count = 0;
2379
2380 static void UserHandler(int sig, void *siginfo, void *context) {
2381 // 4511530 - sem_post is serialized and handled by the manager thread. When
2382 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2383 // don't want to flood the manager thread with sem_post requests.
2384 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2385 return;
2386 }
2387
2388 // Ctrl-C is pressed during error reporting, likely because the error
2389 // handler fails to abort. Let VM die immediately.
2390 if (sig == SIGINT && is_error_reported()) {
2391 os::die();
2392 }
2393
2394 os::signal_notify(sig);
2395 }
2396
2397 void* os::user_handler() {
2398 return CAST_FROM_FN_PTR(void*, UserHandler);
2399 }
2400
2401 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2402 struct timespec ts;
2403 // Semaphore's are always associated with CLOCK_REALTIME
2404 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2405 // see unpackTime for discussion on overflow checking
2406 if (sec >= MAX_SECS) {
2407 ts.tv_sec += MAX_SECS;
2408 ts.tv_nsec = 0;
2409 } else {
2410 ts.tv_sec += sec;
2411 ts.tv_nsec += nsec;
2412 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2413 ts.tv_nsec -= NANOSECS_PER_SEC;
2414 ++ts.tv_sec; // note: this must be <= max_secs
2415 }
2416 }
2417
2418 return ts;
2419 }
2420
2421 extern "C" {
2422 typedef void (*sa_handler_t)(int);
2423 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2424 }
2425
2426 void* os::signal(int signal_number, void* handler) {
2427 struct sigaction sigAct, oldSigAct;
2428
2429 sigfillset(&(sigAct.sa_mask));
2430 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2431 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2432
2433 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2434 // -1 means registration failed
2435 return (void *)-1;
2436 }
2437
2438 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2439 }
2440
2441 void os::signal_raise(int signal_number) {
2442 ::raise(signal_number);
2443 }
2444
2445 // The following code is moved from os.cpp for making this
2446 // code platform specific, which it is by its very nature.
2447
2448 // Will be modified when max signal is changed to be dynamic
2449 int os::sigexitnum_pd() {
2450 return NSIG;
2451 }
2452
2453 // a counter for each possible signal value
2454 static volatile jint pending_signals[NSIG+1] = { 0 };
2455
2456 // Linux(POSIX) specific hand shaking semaphore.
2457 static sem_t sig_sem;
2458 static PosixSemaphore sr_semaphore;
2459
2460 void os::signal_init_pd() {
2461 // Initialize signal structures
2462 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2463
2464 // Initialize signal semaphore
2465 ::sem_init(&sig_sem, 0, 0);
2466 }
2467
2468 void os::signal_notify(int sig) {
2469 Atomic::inc(&pending_signals[sig]);
2470 ::sem_post(&sig_sem);
2471 }
2472
2473 static int check_pending_signals(bool wait) {
2474 Atomic::store(0, &sigint_count);
2475 for (;;) {
2476 for (int i = 0; i < NSIG + 1; i++) {
2477 jint n = pending_signals[i];
2478 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2479 return i;
2480 }
2481 }
2482 if (!wait) {
2483 return -1;
2484 }
2485 JavaThread *thread = JavaThread::current();
2486 ThreadBlockInVM tbivm(thread);
2487
2488 bool threadIsSuspended;
2489 do {
2490 thread->set_suspend_equivalent();
2491 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2492 ::sem_wait(&sig_sem);
2493
2494 // were we externally suspended while we were waiting?
2495 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2496 if (threadIsSuspended) {
2497 // The semaphore has been incremented, but while we were waiting
2498 // another thread suspended us. We don't want to continue running
2499 // while suspended because that would surprise the thread that
2500 // suspended us.
2501 ::sem_post(&sig_sem);
2502
2503 thread->java_suspend_self();
2504 }
2505 } while (threadIsSuspended);
2506 }
2507 }
2508
2509 int os::signal_lookup() {
2510 return check_pending_signals(false);
2511 }
2512
2513 int os::signal_wait() {
2514 return check_pending_signals(true);
2515 }
2516
2517 ////////////////////////////////////////////////////////////////////////////////
2518 // Virtual Memory
2519
2520 int os::vm_page_size() {
2521 // Seems redundant as all get out
2522 assert(os::Linux::page_size() != -1, "must call os::init");
2523 return os::Linux::page_size();
2524 }
2525
2526 // Solaris allocates memory by pages.
2527 int os::vm_allocation_granularity() {
2528 assert(os::Linux::page_size() != -1, "must call os::init");
2529 return os::Linux::page_size();
2530 }
2531
2532 // Rationale behind this function:
2533 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2534 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2535 // samples for JITted code. Here we create private executable mapping over the code cache
2536 // and then we can use standard (well, almost, as mapping can change) way to provide
2537 // info for the reporting script by storing timestamp and location of symbol
2538 void linux_wrap_code(char* base, size_t size) {
2539 static volatile jint cnt = 0;
2540
2541 if (!UseOprofile) {
2542 return;
2543 }
2544
2545 char buf[PATH_MAX+1];
2546 int num = Atomic::add(1, &cnt);
2547
2548 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2549 os::get_temp_directory(), os::current_process_id(), num);
2550 unlink(buf);
2551
2552 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2553
2554 if (fd != -1) {
2555 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2556 if (rv != (off_t)-1) {
2557 if (::write(fd, "", 1) == 1) {
2558 mmap(base, size,
2559 PROT_READ|PROT_WRITE|PROT_EXEC,
2560 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2561 }
2562 }
2563 ::close(fd);
2564 unlink(buf);
2565 }
2566 }
2567
2568 static bool recoverable_mmap_error(int err) {
2569 // See if the error is one we can let the caller handle. This
2570 // list of errno values comes from JBS-6843484. I can't find a
2571 // Linux man page that documents this specific set of errno
2572 // values so while this list currently matches Solaris, it may
2573 // change as we gain experience with this failure mode.
2574 switch (err) {
2575 case EBADF:
2576 case EINVAL:
2577 case ENOTSUP:
2578 // let the caller deal with these errors
2579 return true;
2580
2581 default:
2582 // Any remaining errors on this OS can cause our reserved mapping
2583 // to be lost. That can cause confusion where different data
2584 // structures think they have the same memory mapped. The worst
2585 // scenario is if both the VM and a library think they have the
2586 // same memory mapped.
2587 return false;
2588 }
2589 }
2590
2591 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2592 int err) {
2593 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2594 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2595 strerror(err), err);
2596 }
2597
2598 static void warn_fail_commit_memory(char* addr, size_t size,
2599 size_t alignment_hint, bool exec,
2600 int err) {
2601 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2602 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2603 alignment_hint, exec, strerror(err), err);
2604 }
2605
2606 // NOTE: Linux kernel does not really reserve the pages for us.
2607 // All it does is to check if there are enough free pages
2608 // left at the time of mmap(). This could be a potential
2609 // problem.
2610 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2611 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2612 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2613 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2614 if (res != (uintptr_t) MAP_FAILED) {
2615 if (UseNUMAInterleaving) {
2616 numa_make_global(addr, size);
2617 }
2618 return 0;
2619 }
2620
2621 int err = errno; // save errno from mmap() call above
2622
2623 if (!recoverable_mmap_error(err)) {
2624 warn_fail_commit_memory(addr, size, exec, err);
2625 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2626 }
2627
2628 return err;
2629 }
2630
2631 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2632 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2633 }
2634
2635 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2636 const char* mesg) {
2637 assert(mesg != NULL, "mesg must be specified");
2638 int err = os::Linux::commit_memory_impl(addr, size, exec);
2639 if (err != 0) {
2640 // the caller wants all commit errors to exit with the specified mesg:
2641 warn_fail_commit_memory(addr, size, exec, err);
2642 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2643 }
2644 }
2645
2646 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2647 #ifndef MAP_HUGETLB
2648 #define MAP_HUGETLB 0x40000
2649 #endif
2650
2651 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2652 #ifndef MADV_HUGEPAGE
2653 #define MADV_HUGEPAGE 14
2654 #endif
2655
2656 int os::Linux::commit_memory_impl(char* addr, size_t size,
2657 size_t alignment_hint, bool exec) {
2658 int err = os::Linux::commit_memory_impl(addr, size, exec);
2659 if (err == 0) {
2660 realign_memory(addr, size, alignment_hint);
2661 }
2662 return err;
2663 }
2664
2665 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2666 bool exec) {
2667 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2668 }
2669
2670 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2671 size_t alignment_hint, bool exec,
2672 const char* mesg) {
2673 assert(mesg != NULL, "mesg must be specified");
2674 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2675 if (err != 0) {
2676 // the caller wants all commit errors to exit with the specified mesg:
2677 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2678 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2679 }
2680 }
2681
2682 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2683 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2684 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2685 // be supported or the memory may already be backed by huge pages.
2686 ::madvise(addr, bytes, MADV_HUGEPAGE);
2687 }
2688 }
2689
2690 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2691 // This method works by doing an mmap over an existing mmaping and effectively discarding
2692 // the existing pages. However it won't work for SHM-based large pages that cannot be
2693 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2694 // small pages on top of the SHM segment. This method always works for small pages, so we
2695 // allow that in any case.
2696 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2697 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2698 }
2699 }
2700
2701 void os::numa_make_global(char *addr, size_t bytes) {
2702 Linux::numa_interleave_memory(addr, bytes);
2703 }
2704
2705 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2706 // bind policy to MPOL_PREFERRED for the current thread.
2707 #define USE_MPOL_PREFERRED 0
2708
2709 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2710 // To make NUMA and large pages more robust when both enabled, we need to ease
2711 // the requirements on where the memory should be allocated. MPOL_BIND is the
2712 // default policy and it will force memory to be allocated on the specified
2713 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2714 // the specified node, but will not force it. Using this policy will prevent
2715 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2716 // free large pages.
2717 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2718 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2719 }
2720
2721 bool os::numa_topology_changed() { return false; }
2722
2723 size_t os::numa_get_groups_num() {
2724 int max_node = Linux::numa_max_node();
2725 return max_node > 0 ? max_node + 1 : 1;
2726 }
2727
2728 int os::numa_get_group_id() {
2729 int cpu_id = Linux::sched_getcpu();
2730 if (cpu_id != -1) {
2731 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2732 if (lgrp_id != -1) {
2733 return lgrp_id;
2734 }
2735 }
2736 return 0;
2737 }
2738
2739 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2740 for (size_t i = 0; i < size; i++) {
2741 ids[i] = i;
2742 }
2743 return size;
2744 }
2745
2746 bool os::get_page_info(char *start, page_info* info) {
2747 return false;
2748 }
2749
2750 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2751 page_info* page_found) {
2752 return end;
2753 }
2754
2755
2756 int os::Linux::sched_getcpu_syscall(void) {
2757 unsigned int cpu = 0;
2758 int retval = -1;
2759
2760 #if defined(IA32)
2761 #ifndef SYS_getcpu
2762 #define SYS_getcpu 318
2763 #endif
2764 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2765 #elif defined(AMD64)
2766 // Unfortunately we have to bring all these macros here from vsyscall.h
2767 // to be able to compile on old linuxes.
2768 #define __NR_vgetcpu 2
2769 #define VSYSCALL_START (-10UL << 20)
2770 #define VSYSCALL_SIZE 1024
2771 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2772 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2773 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2774 retval = vgetcpu(&cpu, NULL, NULL);
2775 #endif
2776
2777 return (retval == -1) ? retval : cpu;
2778 }
2779
2780 // Something to do with the numa-aware allocator needs these symbols
2781 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2782 extern "C" JNIEXPORT void numa_error(char *where) { }
2783
2784
2785 // If we are running with libnuma version > 2, then we should
2786 // be trying to use symbols with versions 1.1
2787 // If we are running with earlier version, which did not have symbol versions,
2788 // we should use the base version.
2789 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2790 void *f = dlvsym(handle, name, "libnuma_1.1");
2791 if (f == NULL) {
2792 f = dlsym(handle, name);
2793 }
2794 return f;
2795 }
2796
2797 bool os::Linux::libnuma_init() {
2798 // sched_getcpu() should be in libc.
2799 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2800 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2801
2802 // If it's not, try a direct syscall.
2803 if (sched_getcpu() == -1) {
2804 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2805 (void*)&sched_getcpu_syscall));
2806 }
2807
2808 if (sched_getcpu() != -1) { // Does it work?
2809 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2810 if (handle != NULL) {
2811 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2812 libnuma_dlsym(handle, "numa_node_to_cpus")));
2813 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2814 libnuma_dlsym(handle, "numa_max_node")));
2815 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2816 libnuma_dlsym(handle, "numa_available")));
2817 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2818 libnuma_dlsym(handle, "numa_tonode_memory")));
2819 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2820 libnuma_dlsym(handle, "numa_interleave_memory")));
2821 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2822 libnuma_dlsym(handle, "numa_set_bind_policy")));
2823
2824
2825 if (numa_available() != -1) {
2826 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2827 // Create a cpu -> node mapping
2828 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2829 rebuild_cpu_to_node_map();
2830 return true;
2831 }
2832 }
2833 }
2834 return false;
2835 }
2836
2837 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2838 // The table is later used in get_node_by_cpu().
2839 void os::Linux::rebuild_cpu_to_node_map() {
2840 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2841 // in libnuma (possible values are starting from 16,
2842 // and continuing up with every other power of 2, but less
2843 // than the maximum number of CPUs supported by kernel), and
2844 // is a subject to change (in libnuma version 2 the requirements
2845 // are more reasonable) we'll just hardcode the number they use
2846 // in the library.
2847 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2848
2849 size_t cpu_num = os::active_processor_count();
2850 size_t cpu_map_size = NCPUS / BitsPerCLong;
2851 size_t cpu_map_valid_size =
2852 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2853
2854 cpu_to_node()->clear();
2855 cpu_to_node()->at_grow(cpu_num - 1);
2856 size_t node_num = numa_get_groups_num();
2857
2858 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2859 for (size_t i = 0; i < node_num; i++) {
2860 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2861 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2862 if (cpu_map[j] != 0) {
2863 for (size_t k = 0; k < BitsPerCLong; k++) {
2864 if (cpu_map[j] & (1UL << k)) {
2865 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2866 }
2867 }
2868 }
2869 }
2870 }
2871 }
2872 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2873 }
2874
2875 int os::Linux::get_node_by_cpu(int cpu_id) {
2876 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2877 return cpu_to_node()->at(cpu_id);
2878 }
2879 return -1;
2880 }
2881
2882 GrowableArray<int>* os::Linux::_cpu_to_node;
2883 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2884 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2885 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2886 os::Linux::numa_available_func_t os::Linux::_numa_available;
2887 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2888 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2889 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2890 unsigned long* os::Linux::_numa_all_nodes;
2891
2892 bool os::pd_uncommit_memory(char* addr, size_t size) {
2893 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2894 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2895 return res != (uintptr_t) MAP_FAILED;
2896 }
2897
2898 static address get_stack_commited_bottom(address bottom, size_t size) {
2899 address nbot = bottom;
2900 address ntop = bottom + size;
2901
2902 size_t page_sz = os::vm_page_size();
2903 unsigned pages = size / page_sz;
2904
2905 unsigned char vec[1];
2906 unsigned imin = 1, imax = pages + 1, imid;
2907 int mincore_return_value = 0;
2908
2909 assert(imin <= imax, "Unexpected page size");
2910
2911 while (imin < imax) {
2912 imid = (imax + imin) / 2;
2913 nbot = ntop - (imid * page_sz);
2914
2915 // Use a trick with mincore to check whether the page is mapped or not.
2916 // mincore sets vec to 1 if page resides in memory and to 0 if page
2917 // is swapped output but if page we are asking for is unmapped
2918 // it returns -1,ENOMEM
2919 mincore_return_value = mincore(nbot, page_sz, vec);
2920
2921 if (mincore_return_value == -1) {
2922 // Page is not mapped go up
2923 // to find first mapped page
2924 if (errno != EAGAIN) {
2925 assert(errno == ENOMEM, "Unexpected mincore errno");
2926 imax = imid;
2927 }
2928 } else {
2929 // Page is mapped go down
2930 // to find first not mapped page
2931 imin = imid + 1;
2932 }
2933 }
2934
2935 nbot = nbot + page_sz;
2936
2937 // Adjust stack bottom one page up if last checked page is not mapped
2938 if (mincore_return_value == -1) {
2939 nbot = nbot + page_sz;
2940 }
2941
2942 return nbot;
2943 }
2944
2945
2946 // Linux uses a growable mapping for the stack, and if the mapping for
2947 // the stack guard pages is not removed when we detach a thread the
2948 // stack cannot grow beyond the pages where the stack guard was
2949 // mapped. If at some point later in the process the stack expands to
2950 // that point, the Linux kernel cannot expand the stack any further
2951 // because the guard pages are in the way, and a segfault occurs.
2952 //
2953 // However, it's essential not to split the stack region by unmapping
2954 // a region (leaving a hole) that's already part of the stack mapping,
2955 // so if the stack mapping has already grown beyond the guard pages at
2956 // the time we create them, we have to truncate the stack mapping.
2957 // So, we need to know the extent of the stack mapping when
2958 // create_stack_guard_pages() is called.
2959
2960 // We only need this for stacks that are growable: at the time of
2961 // writing thread stacks don't use growable mappings (i.e. those
2962 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2963 // only applies to the main thread.
2964
2965 // If the (growable) stack mapping already extends beyond the point
2966 // where we're going to put our guard pages, truncate the mapping at
2967 // that point by munmap()ping it. This ensures that when we later
2968 // munmap() the guard pages we don't leave a hole in the stack
2969 // mapping. This only affects the main/initial thread
2970
2971 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2972 if (os::Linux::is_initial_thread()) {
2973 // As we manually grow stack up to bottom inside create_attached_thread(),
2974 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2975 // we don't need to do anything special.
2976 // Check it first, before calling heavy function.
2977 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2978 unsigned char vec[1];
2979
2980 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2981 // Fallback to slow path on all errors, including EAGAIN
2982 stack_extent = (uintptr_t) get_stack_commited_bottom(
2983 os::Linux::initial_thread_stack_bottom(),
2984 (size_t)addr - stack_extent);
2985 }
2986
2987 if (stack_extent < (uintptr_t)addr) {
2988 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2989 }
2990 }
2991
2992 return os::commit_memory(addr, size, !ExecMem);
2993 }
2994
2995 // If this is a growable mapping, remove the guard pages entirely by
2996 // munmap()ping them. If not, just call uncommit_memory(). This only
2997 // affects the main/initial thread, but guard against future OS changes
2998 // It's safe to always unmap guard pages for initial thread because we
2999 // always place it right after end of the mapped region
3000
3001 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3002 uintptr_t stack_extent, stack_base;
3003
3004 if (os::Linux::is_initial_thread()) {
3005 return ::munmap(addr, size) == 0;
3006 }
3007
3008 return os::uncommit_memory(addr, size);
3009 }
3010
3011 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3012 // at 'requested_addr'. If there are existing memory mappings at the same
3013 // location, however, they will be overwritten. If 'fixed' is false,
3014 // 'requested_addr' is only treated as a hint, the return value may or
3015 // may not start from the requested address. Unlike Linux mmap(), this
3016 // function returns NULL to indicate failure.
3017 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3018 char * addr;
3019 int flags;
3020
3021 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3022 if (fixed) {
3023 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3024 flags |= MAP_FIXED;
3025 }
3026
3027 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3028 // touch an uncommitted page. Otherwise, the read/write might
3029 // succeed if we have enough swap space to back the physical page.
3030 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3031 flags, -1, 0);
3032
3033 return addr == MAP_FAILED ? NULL : addr;
3034 }
3035
3036 static int anon_munmap(char * addr, size_t size) {
3037 return ::munmap(addr, size) == 0;
3038 }
3039
3040 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3041 size_t alignment_hint) {
3042 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3043 }
3044
3045 bool os::pd_release_memory(char* addr, size_t size) {
3046 return anon_munmap(addr, size);
3047 }
3048
3049 static bool linux_mprotect(char* addr, size_t size, int prot) {
3050 // Linux wants the mprotect address argument to be page aligned.
3051 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3052
3053 // According to SUSv3, mprotect() should only be used with mappings
3054 // established by mmap(), and mmap() always maps whole pages. Unaligned
3055 // 'addr' likely indicates problem in the VM (e.g. trying to change
3056 // protection of malloc'ed or statically allocated memory). Check the
3057 // caller if you hit this assert.
3058 assert(addr == bottom, "sanity check");
3059
3060 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3061 return ::mprotect(bottom, size, prot) == 0;
3062 }
3063
3064 // Set protections specified
3065 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3066 bool is_committed) {
3067 unsigned int p = 0;
3068 switch (prot) {
3069 case MEM_PROT_NONE: p = PROT_NONE; break;
3070 case MEM_PROT_READ: p = PROT_READ; break;
3071 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3072 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3073 default:
3074 ShouldNotReachHere();
3075 }
3076 // is_committed is unused.
3077 return linux_mprotect(addr, bytes, p);
3078 }
3079
3080 bool os::guard_memory(char* addr, size_t size) {
3081 return linux_mprotect(addr, size, PROT_NONE);
3082 }
3083
3084 bool os::unguard_memory(char* addr, size_t size) {
3085 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3086 }
3087
3088 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3089 size_t page_size) {
3090 bool result = false;
3091 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3092 MAP_ANONYMOUS|MAP_PRIVATE,
3093 -1, 0);
3094 if (p != MAP_FAILED) {
3095 void *aligned_p = align_ptr_up(p, page_size);
3096
3097 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3098
3099 munmap(p, page_size * 2);
3100 }
3101
3102 if (warn && !result) {
3103 warning("TransparentHugePages is not supported by the operating system.");
3104 }
3105
3106 return result;
3107 }
3108
3109 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3110 bool result = false;
3111 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3112 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3113 -1, 0);
3114
3115 if (p != MAP_FAILED) {
3116 // We don't know if this really is a huge page or not.
3117 FILE *fp = fopen("/proc/self/maps", "r");
3118 if (fp) {
3119 while (!feof(fp)) {
3120 char chars[257];
3121 long x = 0;
3122 if (fgets(chars, sizeof(chars), fp)) {
3123 if (sscanf(chars, "%lx-%*x", &x) == 1
3124 && x == (long)p) {
3125 if (strstr (chars, "hugepage")) {
3126 result = true;
3127 break;
3128 }
3129 }
3130 }
3131 }
3132 fclose(fp);
3133 }
3134 munmap(p, page_size);
3135 }
3136
3137 if (warn && !result) {
3138 warning("HugeTLBFS is not supported by the operating system.");
3139 }
3140
3141 return result;
3142 }
3143
3144 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3145 //
3146 // From the coredump_filter documentation:
3147 //
3148 // - (bit 0) anonymous private memory
3149 // - (bit 1) anonymous shared memory
3150 // - (bit 2) file-backed private memory
3151 // - (bit 3) file-backed shared memory
3152 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3153 // effective only if the bit 2 is cleared)
3154 // - (bit 5) hugetlb private memory
3155 // - (bit 6) hugetlb shared memory
3156 //
3157 static void set_coredump_filter(void) {
3158 FILE *f;
3159 long cdm;
3160
3161 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3162 return;
3163 }
3164
3165 if (fscanf(f, "%lx", &cdm) != 1) {
3166 fclose(f);
3167 return;
3168 }
3169
3170 rewind(f);
3171
3172 if ((cdm & LARGEPAGES_BIT) == 0) {
3173 cdm |= LARGEPAGES_BIT;
3174 fprintf(f, "%#lx", cdm);
3175 }
3176
3177 fclose(f);
3178 }
3179
3180 // Large page support
3181
3182 static size_t _large_page_size = 0;
3183
3184 size_t os::Linux::find_large_page_size() {
3185 size_t large_page_size = 0;
3186
3187 // large_page_size on Linux is used to round up heap size. x86 uses either
3188 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3189 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3190 // page as large as 256M.
3191 //
3192 // Here we try to figure out page size by parsing /proc/meminfo and looking
3193 // for a line with the following format:
3194 // Hugepagesize: 2048 kB
3195 //
3196 // If we can't determine the value (e.g. /proc is not mounted, or the text
3197 // format has been changed), we'll use the largest page size supported by
3198 // the processor.
3199
3200 #ifndef ZERO
3201 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3202 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3203 #endif // ZERO
3204
3205 FILE *fp = fopen("/proc/meminfo", "r");
3206 if (fp) {
3207 while (!feof(fp)) {
3208 int x = 0;
3209 char buf[16];
3210 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3211 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3212 large_page_size = x * K;
3213 break;
3214 }
3215 } else {
3216 // skip to next line
3217 for (;;) {
3218 int ch = fgetc(fp);
3219 if (ch == EOF || ch == (int)'\n') break;
3220 }
3221 }
3222 }
3223 fclose(fp);
3224 }
3225
3226 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3227 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3228 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3229 proper_unit_for_byte_size(large_page_size));
3230 }
3231
3232 return large_page_size;
3233 }
3234
3235 size_t os::Linux::setup_large_page_size() {
3236 _large_page_size = Linux::find_large_page_size();
3237 const size_t default_page_size = (size_t)Linux::page_size();
3238 if (_large_page_size > default_page_size) {
3239 _page_sizes[0] = _large_page_size;
3240 _page_sizes[1] = default_page_size;
3241 _page_sizes[2] = 0;
3242 }
3243
3244 return _large_page_size;
3245 }
3246
3247 bool os::Linux::setup_large_page_type(size_t page_size) {
3248 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3249 FLAG_IS_DEFAULT(UseSHM) &&
3250 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3251
3252 // The type of large pages has not been specified by the user.
3253
3254 // Try UseHugeTLBFS and then UseSHM.
3255 UseHugeTLBFS = UseSHM = true;
3256
3257 // Don't try UseTransparentHugePages since there are known
3258 // performance issues with it turned on. This might change in the future.
3259 UseTransparentHugePages = false;
3260 }
3261
3262 if (UseTransparentHugePages) {
3263 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3264 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3265 UseHugeTLBFS = false;
3266 UseSHM = false;
3267 return true;
3268 }
3269 UseTransparentHugePages = false;
3270 }
3271
3272 if (UseHugeTLBFS) {
3273 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3274 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3275 UseSHM = false;
3276 return true;
3277 }
3278 UseHugeTLBFS = false;
3279 }
3280
3281 return UseSHM;
3282 }
3283
3284 void os::large_page_init() {
3285 if (!UseLargePages &&
3286 !UseTransparentHugePages &&
3287 !UseHugeTLBFS &&
3288 !UseSHM) {
3289 // Not using large pages.
3290 return;
3291 }
3292
3293 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3294 // The user explicitly turned off large pages.
3295 // Ignore the rest of the large pages flags.
3296 UseTransparentHugePages = false;
3297 UseHugeTLBFS = false;
3298 UseSHM = false;
3299 return;
3300 }
3301
3302 size_t large_page_size = Linux::setup_large_page_size();
3303 UseLargePages = Linux::setup_large_page_type(large_page_size);
3304
3305 set_coredump_filter();
3306 }
3307
3308 #ifndef SHM_HUGETLB
3309 #define SHM_HUGETLB 04000
3310 #endif
3311
3312 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3313 char* req_addr, bool exec) {
3314 // "exec" is passed in but not used. Creating the shared image for
3315 // the code cache doesn't have an SHM_X executable permission to check.
3316 assert(UseLargePages && UseSHM, "only for SHM large pages");
3317 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3318
3319 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3320 return NULL; // Fallback to small pages.
3321 }
3322
3323 key_t key = IPC_PRIVATE;
3324 char *addr;
3325
3326 bool warn_on_failure = UseLargePages &&
3327 (!FLAG_IS_DEFAULT(UseLargePages) ||
3328 !FLAG_IS_DEFAULT(UseSHM) ||
3329 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3330 char msg[128];
3331
3332 // Create a large shared memory region to attach to based on size.
3333 // Currently, size is the total size of the heap
3334 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3335 if (shmid == -1) {
3336 // Possible reasons for shmget failure:
3337 // 1. shmmax is too small for Java heap.
3338 // > check shmmax value: cat /proc/sys/kernel/shmmax
3339 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3340 // 2. not enough large page memory.
3341 // > check available large pages: cat /proc/meminfo
3342 // > increase amount of large pages:
3343 // echo new_value > /proc/sys/vm/nr_hugepages
3344 // Note 1: different Linux may use different name for this property,
3345 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3346 // Note 2: it's possible there's enough physical memory available but
3347 // they are so fragmented after a long run that they can't
3348 // coalesce into large pages. Try to reserve large pages when
3349 // the system is still "fresh".
3350 if (warn_on_failure) {
3351 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3352 warning("%s", msg);
3353 }
3354 return NULL;
3355 }
3356
3357 // attach to the region
3358 addr = (char*)shmat(shmid, req_addr, 0);
3359 int err = errno;
3360
3361 // Remove shmid. If shmat() is successful, the actual shared memory segment
3362 // will be deleted when it's detached by shmdt() or when the process
3363 // terminates. If shmat() is not successful this will remove the shared
3364 // segment immediately.
3365 shmctl(shmid, IPC_RMID, NULL);
3366
3367 if ((intptr_t)addr == -1) {
3368 if (warn_on_failure) {
3369 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3370 warning("%s", msg);
3371 }
3372 return NULL;
3373 }
3374
3375 return addr;
3376 }
3377
3378 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3379 int error) {
3380 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3381
3382 bool warn_on_failure = UseLargePages &&
3383 (!FLAG_IS_DEFAULT(UseLargePages) ||
3384 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3385 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3386
3387 if (warn_on_failure) {
3388 char msg[128];
3389 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3390 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3391 warning("%s", msg);
3392 }
3393 }
3394
3395 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3396 char* req_addr,
3397 bool exec) {
3398 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3399 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3400 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3401
3402 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3403 char* addr = (char*)::mmap(req_addr, bytes, prot,
3404 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3405 -1, 0);
3406
3407 if (addr == MAP_FAILED) {
3408 warn_on_large_pages_failure(req_addr, bytes, errno);
3409 return NULL;
3410 }
3411
3412 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3413
3414 return addr;
3415 }
3416
3417 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3418 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3419 // (req_addr != NULL) or with a given alignment.
3420 // - bytes shall be a multiple of alignment.
3421 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3422 // - alignment sets the alignment at which memory shall be allocated.
3423 // It must be a multiple of allocation granularity.
3424 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3425 // req_addr or NULL.
3426 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3427
3428 size_t extra_size = bytes;
3429 if (req_addr == NULL && alignment > 0) {
3430 extra_size += alignment;
3431 }
3432
3433 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3434 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3435 -1, 0);
3436 if (start == MAP_FAILED) {
3437 start = NULL;
3438 } else {
3439 if (req_addr != NULL) {
3440 if (start != req_addr) {
3441 ::munmap(start, extra_size);
3442 start = NULL;
3443 }
3444 } else {
3445 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3446 char* const end_aligned = start_aligned + bytes;
3447 char* const end = start + extra_size;
3448 if (start_aligned > start) {
3449 ::munmap(start, start_aligned - start);
3450 }
3451 if (end_aligned < end) {
3452 ::munmap(end_aligned, end - end_aligned);
3453 }
3454 start = start_aligned;
3455 }
3456 }
3457 return start;
3458
3459 }
3460
3461 // Reserve memory using mmap(MAP_HUGETLB).
3462 // - bytes shall be a multiple of alignment.
3463 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3464 // - alignment sets the alignment at which memory shall be allocated.
3465 // It must be a multiple of allocation granularity.
3466 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3467 // req_addr or NULL.
3468 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3469 size_t alignment,
3470 char* req_addr,
3471 bool exec) {
3472 size_t large_page_size = os::large_page_size();
3473 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3474
3475 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3476 assert(is_size_aligned(bytes, alignment), "Must be");
3477
3478 // First reserve - but not commit - the address range in small pages.
3479 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3480
3481 if (start == NULL) {
3482 return NULL;
3483 }
3484
3485 assert(is_ptr_aligned(start, alignment), "Must be");
3486
3487 char* end = start + bytes;
3488
3489 // Find the regions of the allocated chunk that can be promoted to large pages.
3490 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3491 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3492
3493 size_t lp_bytes = lp_end - lp_start;
3494
3495 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3496
3497 if (lp_bytes == 0) {
3498 // The mapped region doesn't even span the start and the end of a large page.
3499 // Fall back to allocate a non-special area.
3500 ::munmap(start, end - start);
3501 return NULL;
3502 }
3503
3504 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3505
3506 void* result;
3507
3508 // Commit small-paged leading area.
3509 if (start != lp_start) {
3510 result = ::mmap(start, lp_start - start, prot,
3511 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3512 -1, 0);
3513 if (result == MAP_FAILED) {
3514 ::munmap(lp_start, end - lp_start);
3515 return NULL;
3516 }
3517 }
3518
3519 // Commit large-paged area.
3520 result = ::mmap(lp_start, lp_bytes, prot,
3521 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3522 -1, 0);
3523 if (result == MAP_FAILED) {
3524 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3525 // If the mmap above fails, the large pages region will be unmapped and we
3526 // have regions before and after with small pages. Release these regions.
3527 //
3528 // | mapped | unmapped | mapped |
3529 // ^ ^ ^ ^
3530 // start lp_start lp_end end
3531 //
3532 ::munmap(start, lp_start - start);
3533 ::munmap(lp_end, end - lp_end);
3534 return NULL;
3535 }
3536
3537 // Commit small-paged trailing area.
3538 if (lp_end != end) {
3539 result = ::mmap(lp_end, end - lp_end, prot,
3540 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3541 -1, 0);
3542 if (result == MAP_FAILED) {
3543 ::munmap(start, lp_end - start);
3544 return NULL;
3545 }
3546 }
3547
3548 return start;
3549 }
3550
3551 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3552 size_t alignment,
3553 char* req_addr,
3554 bool exec) {
3555 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3556 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3557 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3558 assert(is_power_of_2(os::large_page_size()), "Must be");
3559 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3560
3561 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3562 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3563 } else {
3564 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3565 }
3566 }
3567
3568 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3569 char* req_addr, bool exec) {
3570 assert(UseLargePages, "only for large pages");
3571
3572 char* addr;
3573 if (UseSHM) {
3574 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3575 } else {
3576 assert(UseHugeTLBFS, "must be");
3577 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3578 }
3579
3580 if (addr != NULL) {
3581 if (UseNUMAInterleaving) {
3582 numa_make_global(addr, bytes);
3583 }
3584
3585 // The memory is committed
3586 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3587 }
3588
3589 return addr;
3590 }
3591
3592 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3593 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3594 return shmdt(base) == 0;
3595 }
3596
3597 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3598 return pd_release_memory(base, bytes);
3599 }
3600
3601 bool os::release_memory_special(char* base, size_t bytes) {
3602 bool res;
3603 if (MemTracker::tracking_level() > NMT_minimal) {
3604 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3605 res = os::Linux::release_memory_special_impl(base, bytes);
3606 if (res) {
3607 tkr.record((address)base, bytes);
3608 }
3609
3610 } else {
3611 res = os::Linux::release_memory_special_impl(base, bytes);
3612 }
3613 return res;
3614 }
3615
3616 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3617 assert(UseLargePages, "only for large pages");
3618 bool res;
3619
3620 if (UseSHM) {
3621 res = os::Linux::release_memory_special_shm(base, bytes);
3622 } else {
3623 assert(UseHugeTLBFS, "must be");
3624 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3625 }
3626 return res;
3627 }
3628
3629 size_t os::large_page_size() {
3630 return _large_page_size;
3631 }
3632
3633 // With SysV SHM the entire memory region must be allocated as shared
3634 // memory.
3635 // HugeTLBFS allows application to commit large page memory on demand.
3636 // However, when committing memory with HugeTLBFS fails, the region
3637 // that was supposed to be committed will lose the old reservation
3638 // and allow other threads to steal that memory region. Because of this
3639 // behavior we can't commit HugeTLBFS memory.
3640 bool os::can_commit_large_page_memory() {
3641 return UseTransparentHugePages;
3642 }
3643
3644 bool os::can_execute_large_page_memory() {
3645 return UseTransparentHugePages || UseHugeTLBFS;
3646 }
3647
3648 // Reserve memory at an arbitrary address, only if that area is
3649 // available (and not reserved for something else).
3650
3651 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3652 const int max_tries = 10;
3653 char* base[max_tries];
3654 size_t size[max_tries];
3655 const size_t gap = 0x000000;
3656
3657 // Assert only that the size is a multiple of the page size, since
3658 // that's all that mmap requires, and since that's all we really know
3659 // about at this low abstraction level. If we need higher alignment,
3660 // we can either pass an alignment to this method or verify alignment
3661 // in one of the methods further up the call chain. See bug 5044738.
3662 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3663
3664 // Repeatedly allocate blocks until the block is allocated at the
3665 // right spot.
3666
3667 // Linux mmap allows caller to pass an address as hint; give it a try first,
3668 // if kernel honors the hint then we can return immediately.
3669 char * addr = anon_mmap(requested_addr, bytes, false);
3670 if (addr == requested_addr) {
3671 return requested_addr;
3672 }
3673
3674 if (addr != NULL) {
3675 // mmap() is successful but it fails to reserve at the requested address
3676 anon_munmap(addr, bytes);
3677 }
3678
3679 int i;
3680 for (i = 0; i < max_tries; ++i) {
3681 base[i] = reserve_memory(bytes);
3682
3683 if (base[i] != NULL) {
3684 // Is this the block we wanted?
3685 if (base[i] == requested_addr) {
3686 size[i] = bytes;
3687 break;
3688 }
3689
3690 // Does this overlap the block we wanted? Give back the overlapped
3691 // parts and try again.
3692
3693 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3694 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3695 unmap_memory(base[i], top_overlap);
3696 base[i] += top_overlap;
3697 size[i] = bytes - top_overlap;
3698 } else {
3699 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3700 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3701 unmap_memory(requested_addr, bottom_overlap);
3702 size[i] = bytes - bottom_overlap;
3703 } else {
3704 size[i] = bytes;
3705 }
3706 }
3707 }
3708 }
3709
3710 // Give back the unused reserved pieces.
3711
3712 for (int j = 0; j < i; ++j) {
3713 if (base[j] != NULL) {
3714 unmap_memory(base[j], size[j]);
3715 }
3716 }
3717
3718 if (i < max_tries) {
3719 return requested_addr;
3720 } else {
3721 return NULL;
3722 }
3723 }
3724
3725 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3726 return ::read(fd, buf, nBytes);
3727 }
3728
3729 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3730 return ::pread(fd, buf, nBytes, offset);
3731 }
3732
3733 // Short sleep, direct OS call.
3734 //
3735 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3736 // sched_yield(2) will actually give up the CPU:
3737 //
3738 // * Alone on this pariticular CPU, keeps running.
3739 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3740 // (pre 2.6.39).
3741 //
3742 // So calling this with 0 is an alternative.
3743 //
3744 void os::naked_short_sleep(jlong ms) {
3745 struct timespec req;
3746
3747 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3748 req.tv_sec = 0;
3749 if (ms > 0) {
3750 req.tv_nsec = (ms % 1000) * 1000000;
3751 } else {
3752 req.tv_nsec = 1;
3753 }
3754
3755 nanosleep(&req, NULL);
3756
3757 return;
3758 }
3759
3760 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3761 void os::infinite_sleep() {
3762 while (true) { // sleep forever ...
3763 ::sleep(100); // ... 100 seconds at a time
3764 }
3765 }
3766
3767 // Used to convert frequent JVM_Yield() to nops
3768 bool os::dont_yield() {
3769 return DontYieldALot;
3770 }
3771
3772 void os::naked_yield() {
3773 sched_yield();
3774 }
3775
3776 ////////////////////////////////////////////////////////////////////////////////
3777 // thread priority support
3778
3779 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3780 // only supports dynamic priority, static priority must be zero. For real-time
3781 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3782 // However, for large multi-threaded applications, SCHED_RR is not only slower
3783 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3784 // of 5 runs - Sep 2005).
3785 //
3786 // The following code actually changes the niceness of kernel-thread/LWP. It
3787 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3788 // not the entire user process, and user level threads are 1:1 mapped to kernel
3789 // threads. It has always been the case, but could change in the future. For
3790 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3791 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3792
3793 int os::java_to_os_priority[CriticalPriority + 1] = {
3794 19, // 0 Entry should never be used
3795
3796 4, // 1 MinPriority
3797 3, // 2
3798 2, // 3
3799
3800 1, // 4
3801 0, // 5 NormPriority
3802 -1, // 6
3803
3804 -2, // 7
3805 -3, // 8
3806 -4, // 9 NearMaxPriority
3807
3808 -5, // 10 MaxPriority
3809
3810 -5 // 11 CriticalPriority
3811 };
3812
3813 static int prio_init() {
3814 if (ThreadPriorityPolicy == 1) {
3815 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3816 // if effective uid is not root. Perhaps, a more elegant way of doing
3817 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3818 if (geteuid() != 0) {
3819 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3820 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3821 }
3822 ThreadPriorityPolicy = 0;
3823 }
3824 }
3825 if (UseCriticalJavaThreadPriority) {
3826 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3827 }
3828 return 0;
3829 }
3830
3831 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3832 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3833
3834 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3835 return (ret == 0) ? OS_OK : OS_ERR;
3836 }
3837
3838 OSReturn os::get_native_priority(const Thread* const thread,
3839 int *priority_ptr) {
3840 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3841 *priority_ptr = java_to_os_priority[NormPriority];
3842 return OS_OK;
3843 }
3844
3845 errno = 0;
3846 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3847 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3848 }
3849
3850 // Hint to the underlying OS that a task switch would not be good.
3851 // Void return because it's a hint and can fail.
3852 void os::hint_no_preempt() {}
3853
3854 ////////////////////////////////////////////////////////////////////////////////
3855 // suspend/resume support
3856
3857 // the low-level signal-based suspend/resume support is a remnant from the
3858 // old VM-suspension that used to be for java-suspension, safepoints etc,
3859 // within hotspot. Now there is a single use-case for this:
3860 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3861 // that runs in the watcher thread.
3862 // The remaining code is greatly simplified from the more general suspension
3863 // code that used to be used.
3864 //
3865 // The protocol is quite simple:
3866 // - suspend:
3867 // - sends a signal to the target thread
3868 // - polls the suspend state of the osthread using a yield loop
3869 // - target thread signal handler (SR_handler) sets suspend state
3870 // and blocks in sigsuspend until continued
3871 // - resume:
3872 // - sets target osthread state to continue
3873 // - sends signal to end the sigsuspend loop in the SR_handler
3874 //
3875 // Note that the SR_lock plays no role in this suspend/resume protocol.
3876
3877 static void resume_clear_context(OSThread *osthread) {
3878 osthread->set_ucontext(NULL);
3879 osthread->set_siginfo(NULL);
3880 }
3881
3882 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3883 ucontext_t* context) {
3884 osthread->set_ucontext(context);
3885 osthread->set_siginfo(siginfo);
3886 }
3887
3888 // Handler function invoked when a thread's execution is suspended or
3889 // resumed. We have to be careful that only async-safe functions are
3890 // called here (Note: most pthread functions are not async safe and
3891 // should be avoided.)
3892 //
3893 // Note: sigwait() is a more natural fit than sigsuspend() from an
3894 // interface point of view, but sigwait() prevents the signal hander
3895 // from being run. libpthread would get very confused by not having
3896 // its signal handlers run and prevents sigwait()'s use with the
3897 // mutex granting granting signal.
3898 //
3899 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3900 //
3901 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3902 // Save and restore errno to avoid confusing native code with EINTR
3903 // after sigsuspend.
3904 int old_errno = errno;
3905
3906 Thread* thread = Thread::current();
3907 OSThread* osthread = thread->osthread();
3908 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3909
3910 os::SuspendResume::State current = osthread->sr.state();
3911 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3912 suspend_save_context(osthread, siginfo, context);
3913
3914 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3915 os::SuspendResume::State state = osthread->sr.suspended();
3916 if (state == os::SuspendResume::SR_SUSPENDED) {
3917 sigset_t suspend_set; // signals for sigsuspend()
3918
3919 // get current set of blocked signals and unblock resume signal
3920 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3921 sigdelset(&suspend_set, SR_signum);
3922
3923 sr_semaphore.signal();
3924 // wait here until we are resumed
3925 while (1) {
3926 sigsuspend(&suspend_set);
3927
3928 os::SuspendResume::State result = osthread->sr.running();
3929 if (result == os::SuspendResume::SR_RUNNING) {
3930 sr_semaphore.signal();
3931 break;
3932 }
3933 }
3934
3935 } else if (state == os::SuspendResume::SR_RUNNING) {
3936 // request was cancelled, continue
3937 } else {
3938 ShouldNotReachHere();
3939 }
3940
3941 resume_clear_context(osthread);
3942 } else if (current == os::SuspendResume::SR_RUNNING) {
3943 // request was cancelled, continue
3944 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3945 // ignore
3946 } else {
3947 // ignore
3948 }
3949
3950 errno = old_errno;
3951 }
3952
3953 static int SR_initialize() {
3954 struct sigaction act;
3955 char *s;
3956
3957 // Get signal number to use for suspend/resume
3958 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3959 int sig = ::strtol(s, 0, 10);
3960 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
3961 sig < NSIG) { // Must be legal signal and fit into sigflags[].
3962 SR_signum = sig;
3963 } else {
3964 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3965 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3966 }
3967 }
3968
3969 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3970 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3971
3972 sigemptyset(&SR_sigset);
3973 sigaddset(&SR_sigset, SR_signum);
3974
3975 // Set up signal handler for suspend/resume
3976 act.sa_flags = SA_RESTART|SA_SIGINFO;
3977 act.sa_handler = (void (*)(int)) SR_handler;
3978
3979 // SR_signum is blocked by default.
3980 // 4528190 - We also need to block pthread restart signal (32 on all
3981 // supported Linux platforms). Note that LinuxThreads need to block
3982 // this signal for all threads to work properly. So we don't have
3983 // to use hard-coded signal number when setting up the mask.
3984 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3985
3986 if (sigaction(SR_signum, &act, 0) == -1) {
3987 return -1;
3988 }
3989
3990 // Save signal flag
3991 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3992 return 0;
3993 }
3994
3995 static int sr_notify(OSThread* osthread) {
3996 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3997 assert_status(status == 0, status, "pthread_kill");
3998 return status;
3999 }
4000
4001 // "Randomly" selected value for how long we want to spin
4002 // before bailing out on suspending a thread, also how often
4003 // we send a signal to a thread we want to resume
4004 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4005 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4006
4007 // returns true on success and false on error - really an error is fatal
4008 // but this seems the normal response to library errors
4009 static bool do_suspend(OSThread* osthread) {
4010 assert(osthread->sr.is_running(), "thread should be running");
4011 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4012
4013 // mark as suspended and send signal
4014 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4015 // failed to switch, state wasn't running?
4016 ShouldNotReachHere();
4017 return false;
4018 }
4019
4020 if (sr_notify(osthread) != 0) {
4021 ShouldNotReachHere();
4022 }
4023
4024 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4025 while (true) {
4026 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4027 break;
4028 } else {
4029 // timeout
4030 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4031 if (cancelled == os::SuspendResume::SR_RUNNING) {
4032 return false;
4033 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4034 // make sure that we consume the signal on the semaphore as well
4035 sr_semaphore.wait();
4036 break;
4037 } else {
4038 ShouldNotReachHere();
4039 return false;
4040 }
4041 }
4042 }
4043
4044 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4045 return true;
4046 }
4047
4048 static void do_resume(OSThread* osthread) {
4049 assert(osthread->sr.is_suspended(), "thread should be suspended");
4050 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4051
4052 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4053 // failed to switch to WAKEUP_REQUEST
4054 ShouldNotReachHere();
4055 return;
4056 }
4057
4058 while (true) {
4059 if (sr_notify(osthread) == 0) {
4060 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4061 if (osthread->sr.is_running()) {
4062 return;
4063 }
4064 }
4065 } else {
4066 ShouldNotReachHere();
4067 }
4068 }
4069
4070 guarantee(osthread->sr.is_running(), "Must be running!");
4071 }
4072
4073 ///////////////////////////////////////////////////////////////////////////////////
4074 // signal handling (except suspend/resume)
4075
4076 // This routine may be used by user applications as a "hook" to catch signals.
4077 // The user-defined signal handler must pass unrecognized signals to this
4078 // routine, and if it returns true (non-zero), then the signal handler must
4079 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4080 // routine will never retun false (zero), but instead will execute a VM panic
4081 // routine kill the process.
4082 //
4083 // If this routine returns false, it is OK to call it again. This allows
4084 // the user-defined signal handler to perform checks either before or after
4085 // the VM performs its own checks. Naturally, the user code would be making
4086 // a serious error if it tried to handle an exception (such as a null check
4087 // or breakpoint) that the VM was generating for its own correct operation.
4088 //
4089 // This routine may recognize any of the following kinds of signals:
4090 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4091 // It should be consulted by handlers for any of those signals.
4092 //
4093 // The caller of this routine must pass in the three arguments supplied
4094 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4095 // field of the structure passed to sigaction(). This routine assumes that
4096 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4097 //
4098 // Note that the VM will print warnings if it detects conflicting signal
4099 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4100 //
4101 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4102 siginfo_t* siginfo,
4103 void* ucontext,
4104 int abort_if_unrecognized);
4105
4106 void signalHandler(int sig, siginfo_t* info, void* uc) {
4107 assert(info != NULL && uc != NULL, "it must be old kernel");
4108 int orig_errno = errno; // Preserve errno value over signal handler.
4109 JVM_handle_linux_signal(sig, info, uc, true);
4110 errno = orig_errno;
4111 }
4112
4113
4114 // This boolean allows users to forward their own non-matching signals
4115 // to JVM_handle_linux_signal, harmlessly.
4116 bool os::Linux::signal_handlers_are_installed = false;
4117
4118 // For signal-chaining
4119 struct sigaction sigact[NSIG];
4120 uint64_t sigs = 0;
4121 #if (64 < NSIG-1)
4122 #error "Not all signals can be encoded in sigs. Adapt its type!"
4123 #endif
4124 bool os::Linux::libjsig_is_loaded = false;
4125 typedef struct sigaction *(*get_signal_t)(int);
4126 get_signal_t os::Linux::get_signal_action = NULL;
4127
4128 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4129 struct sigaction *actp = NULL;
4130
4131 if (libjsig_is_loaded) {
4132 // Retrieve the old signal handler from libjsig
4133 actp = (*get_signal_action)(sig);
4134 }
4135 if (actp == NULL) {
4136 // Retrieve the preinstalled signal handler from jvm
4137 actp = get_preinstalled_handler(sig);
4138 }
4139
4140 return actp;
4141 }
4142
4143 static bool call_chained_handler(struct sigaction *actp, int sig,
4144 siginfo_t *siginfo, void *context) {
4145 // Call the old signal handler
4146 if (actp->sa_handler == SIG_DFL) {
4147 // It's more reasonable to let jvm treat it as an unexpected exception
4148 // instead of taking the default action.
4149 return false;
4150 } else if (actp->sa_handler != SIG_IGN) {
4151 if ((actp->sa_flags & SA_NODEFER) == 0) {
4152 // automaticlly block the signal
4153 sigaddset(&(actp->sa_mask), sig);
4154 }
4155
4156 sa_handler_t hand = NULL;
4157 sa_sigaction_t sa = NULL;
4158 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4159 // retrieve the chained handler
4160 if (siginfo_flag_set) {
4161 sa = actp->sa_sigaction;
4162 } else {
4163 hand = actp->sa_handler;
4164 }
4165
4166 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4167 actp->sa_handler = SIG_DFL;
4168 }
4169
4170 // try to honor the signal mask
4171 sigset_t oset;
4172 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4173
4174 // call into the chained handler
4175 if (siginfo_flag_set) {
4176 (*sa)(sig, siginfo, context);
4177 } else {
4178 (*hand)(sig);
4179 }
4180
4181 // restore the signal mask
4182 pthread_sigmask(SIG_SETMASK, &oset, 0);
4183 }
4184 // Tell jvm's signal handler the signal is taken care of.
4185 return true;
4186 }
4187
4188 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4189 bool chained = false;
4190 // signal-chaining
4191 if (UseSignalChaining) {
4192 struct sigaction *actp = get_chained_signal_action(sig);
4193 if (actp != NULL) {
4194 chained = call_chained_handler(actp, sig, siginfo, context);
4195 }
4196 }
4197 return chained;
4198 }
4199
4200 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4201 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4202 return &sigact[sig];
4203 }
4204 return NULL;
4205 }
4206
4207 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4208 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4209 sigact[sig] = oldAct;
4210 sigs |= (uint64_t)1 << (sig-1);
4211 }
4212
4213 // for diagnostic
4214 int sigflags[NSIG];
4215
4216 int os::Linux::get_our_sigflags(int sig) {
4217 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4218 return sigflags[sig];
4219 }
4220
4221 void os::Linux::set_our_sigflags(int sig, int flags) {
4222 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4223 if (sig > 0 && sig < NSIG) {
4224 sigflags[sig] = flags;
4225 }
4226 }
4227
4228 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4229 // Check for overwrite.
4230 struct sigaction oldAct;
4231 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4232
4233 void* oldhand = oldAct.sa_sigaction
4234 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4235 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4236 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4237 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4238 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4239 if (AllowUserSignalHandlers || !set_installed) {
4240 // Do not overwrite; user takes responsibility to forward to us.
4241 return;
4242 } else if (UseSignalChaining) {
4243 // save the old handler in jvm
4244 save_preinstalled_handler(sig, oldAct);
4245 // libjsig also interposes the sigaction() call below and saves the
4246 // old sigaction on it own.
4247 } else {
4248 fatal("Encountered unexpected pre-existing sigaction handler "
4249 "%#lx for signal %d.", (long)oldhand, sig);
4250 }
4251 }
4252
4253 struct sigaction sigAct;
4254 sigfillset(&(sigAct.sa_mask));
4255 sigAct.sa_handler = SIG_DFL;
4256 if (!set_installed) {
4257 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4258 } else {
4259 sigAct.sa_sigaction = signalHandler;
4260 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4261 }
4262 // Save flags, which are set by ours
4263 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4264 sigflags[sig] = sigAct.sa_flags;
4265
4266 int ret = sigaction(sig, &sigAct, &oldAct);
4267 assert(ret == 0, "check");
4268
4269 void* oldhand2 = oldAct.sa_sigaction
4270 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4271 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4272 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4273 }
4274
4275 // install signal handlers for signals that HotSpot needs to
4276 // handle in order to support Java-level exception handling.
4277
4278 void os::Linux::install_signal_handlers() {
4279 if (!signal_handlers_are_installed) {
4280 signal_handlers_are_installed = true;
4281
4282 // signal-chaining
4283 typedef void (*signal_setting_t)();
4284 signal_setting_t begin_signal_setting = NULL;
4285 signal_setting_t end_signal_setting = NULL;
4286 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4287 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4288 if (begin_signal_setting != NULL) {
4289 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4290 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4291 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4292 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4293 libjsig_is_loaded = true;
4294 assert(UseSignalChaining, "should enable signal-chaining");
4295 }
4296 if (libjsig_is_loaded) {
4297 // Tell libjsig jvm is setting signal handlers
4298 (*begin_signal_setting)();
4299 }
4300
4301 set_signal_handler(SIGSEGV, true);
4302 set_signal_handler(SIGPIPE, true);
4303 set_signal_handler(SIGBUS, true);
4304 set_signal_handler(SIGILL, true);
4305 set_signal_handler(SIGFPE, true);
4306 #if defined(PPC64)
4307 set_signal_handler(SIGTRAP, true);
4308 #endif
4309 set_signal_handler(SIGXFSZ, true);
4310
4311 if (libjsig_is_loaded) {
4312 // Tell libjsig jvm finishes setting signal handlers
4313 (*end_signal_setting)();
4314 }
4315
4316 // We don't activate signal checker if libjsig is in place, we trust ourselves
4317 // and if UserSignalHandler is installed all bets are off.
4318 // Log that signal checking is off only if -verbose:jni is specified.
4319 if (CheckJNICalls) {
4320 if (libjsig_is_loaded) {
4321 if (PrintJNIResolving) {
4322 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4323 }
4324 check_signals = false;
4325 }
4326 if (AllowUserSignalHandlers) {
4327 if (PrintJNIResolving) {
4328 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4329 }
4330 check_signals = false;
4331 }
4332 }
4333 }
4334 }
4335
4336 // This is the fastest way to get thread cpu time on Linux.
4337 // Returns cpu time (user+sys) for any thread, not only for current.
4338 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4339 // It might work on 2.6.10+ with a special kernel/glibc patch.
4340 // For reference, please, see IEEE Std 1003.1-2004:
4341 // http://www.unix.org/single_unix_specification
4342
4343 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4344 struct timespec tp;
4345 int rc = os::Linux::clock_gettime(clockid, &tp);
4346 assert(rc == 0, "clock_gettime is expected to return 0 code");
4347
4348 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4349 }
4350
4351 /////
4352 // glibc on Linux platform uses non-documented flag
4353 // to indicate, that some special sort of signal
4354 // trampoline is used.
4355 // We will never set this flag, and we should
4356 // ignore this flag in our diagnostic
4357 #ifdef SIGNIFICANT_SIGNAL_MASK
4358 #undef SIGNIFICANT_SIGNAL_MASK
4359 #endif
4360 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4361
4362 static const char* get_signal_handler_name(address handler,
4363 char* buf, int buflen) {
4364 int offset = 0;
4365 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4366 if (found) {
4367 // skip directory names
4368 const char *p1, *p2;
4369 p1 = buf;
4370 size_t len = strlen(os::file_separator());
4371 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4372 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4373 } else {
4374 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4375 }
4376 return buf;
4377 }
4378
4379 static void print_signal_handler(outputStream* st, int sig,
4380 char* buf, size_t buflen) {
4381 struct sigaction sa;
4382
4383 sigaction(sig, NULL, &sa);
4384
4385 // See comment for SIGNIFICANT_SIGNAL_MASK define
4386 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4387
4388 st->print("%s: ", os::exception_name(sig, buf, buflen));
4389
4390 address handler = (sa.sa_flags & SA_SIGINFO)
4391 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4392 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4393
4394 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4395 st->print("SIG_DFL");
4396 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4397 st->print("SIG_IGN");
4398 } else {
4399 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4400 }
4401
4402 st->print(", sa_mask[0]=");
4403 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4404
4405 address rh = VMError::get_resetted_sighandler(sig);
4406 // May be, handler was resetted by VMError?
4407 if (rh != NULL) {
4408 handler = rh;
4409 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4410 }
4411
4412 st->print(", sa_flags=");
4413 os::Posix::print_sa_flags(st, sa.sa_flags);
4414
4415 // Check: is it our handler?
4416 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4417 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4418 // It is our signal handler
4419 // check for flags, reset system-used one!
4420 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4421 st->print(
4422 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4423 os::Linux::get_our_sigflags(sig));
4424 }
4425 }
4426 st->cr();
4427 }
4428
4429
4430 #define DO_SIGNAL_CHECK(sig) \
4431 do { \
4432 if (!sigismember(&check_signal_done, sig)) { \
4433 os::Linux::check_signal_handler(sig); \
4434 } \
4435 } while (0)
4436
4437 // This method is a periodic task to check for misbehaving JNI applications
4438 // under CheckJNI, we can add any periodic checks here
4439
4440 void os::run_periodic_checks() {
4441 if (check_signals == false) return;
4442
4443 // SEGV and BUS if overridden could potentially prevent
4444 // generation of hs*.log in the event of a crash, debugging
4445 // such a case can be very challenging, so we absolutely
4446 // check the following for a good measure:
4447 DO_SIGNAL_CHECK(SIGSEGV);
4448 DO_SIGNAL_CHECK(SIGILL);
4449 DO_SIGNAL_CHECK(SIGFPE);
4450 DO_SIGNAL_CHECK(SIGBUS);
4451 DO_SIGNAL_CHECK(SIGPIPE);
4452 DO_SIGNAL_CHECK(SIGXFSZ);
4453 #if defined(PPC64)
4454 DO_SIGNAL_CHECK(SIGTRAP);
4455 #endif
4456
4457 // ReduceSignalUsage allows the user to override these handlers
4458 // see comments at the very top and jvm_solaris.h
4459 if (!ReduceSignalUsage) {
4460 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4461 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4462 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4463 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4464 }
4465
4466 DO_SIGNAL_CHECK(SR_signum);
4467 }
4468
4469 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4470
4471 static os_sigaction_t os_sigaction = NULL;
4472
4473 void os::Linux::check_signal_handler(int sig) {
4474 char buf[O_BUFLEN];
4475 address jvmHandler = NULL;
4476
4477
4478 struct sigaction act;
4479 if (os_sigaction == NULL) {
4480 // only trust the default sigaction, in case it has been interposed
4481 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4482 if (os_sigaction == NULL) return;
4483 }
4484
4485 os_sigaction(sig, (struct sigaction*)NULL, &act);
4486
4487
4488 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4489
4490 address thisHandler = (act.sa_flags & SA_SIGINFO)
4491 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4492 : CAST_FROM_FN_PTR(address, act.sa_handler);
4493
4494
4495 switch (sig) {
4496 case SIGSEGV:
4497 case SIGBUS:
4498 case SIGFPE:
4499 case SIGPIPE:
4500 case SIGILL:
4501 case SIGXFSZ:
4502 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4503 break;
4504
4505 case SHUTDOWN1_SIGNAL:
4506 case SHUTDOWN2_SIGNAL:
4507 case SHUTDOWN3_SIGNAL:
4508 case BREAK_SIGNAL:
4509 jvmHandler = (address)user_handler();
4510 break;
4511
4512 default:
4513 if (sig == SR_signum) {
4514 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4515 } else {
4516 return;
4517 }
4518 break;
4519 }
4520
4521 if (thisHandler != jvmHandler) {
4522 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4523 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4524 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4525 // No need to check this sig any longer
4526 sigaddset(&check_signal_done, sig);
4527 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4528 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4529 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4530 exception_name(sig, buf, O_BUFLEN));
4531 }
4532 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4533 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4534 tty->print("expected:");
4535 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4536 tty->cr();
4537 tty->print(" found:");
4538 os::Posix::print_sa_flags(tty, act.sa_flags);
4539 tty->cr();
4540 // No need to check this sig any longer
4541 sigaddset(&check_signal_done, sig);
4542 }
4543
4544 // Dump all the signal
4545 if (sigismember(&check_signal_done, sig)) {
4546 print_signal_handlers(tty, buf, O_BUFLEN);
4547 }
4548 }
4549
4550 extern void report_error(char* file_name, int line_no, char* title,
4551 char* format, ...);
4552
4553 // this is called _before_ the most of global arguments have been parsed
4554 void os::init(void) {
4555 char dummy; // used to get a guess on initial stack address
4556 // first_hrtime = gethrtime();
4557
4558 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4559
4560 init_random(1234567);
4561
4562 ThreadCritical::initialize();
4563
4564 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4565 if (Linux::page_size() == -1) {
4566 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4567 strerror(errno));
4568 }
4569 init_page_sizes((size_t) Linux::page_size());
4570
4571 Linux::initialize_system_info();
4572
4573 // main_thread points to the aboriginal thread
4574 Linux::_main_thread = pthread_self();
4575
4576 Linux::clock_init();
4577 initial_time_count = javaTimeNanos();
4578
4579 // pthread_condattr initialization for monotonic clock
4580 int status;
4581 pthread_condattr_t* _condattr = os::Linux::condAttr();
4582 if ((status = pthread_condattr_init(_condattr)) != 0) {
4583 fatal("pthread_condattr_init: %s", strerror(status));
4584 }
4585 // Only set the clock if CLOCK_MONOTONIC is available
4586 if (os::supports_monotonic_clock()) {
4587 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4588 if (status == EINVAL) {
4589 warning("Unable to use monotonic clock with relative timed-waits" \
4590 " - changes to the time-of-day clock may have adverse affects");
4591 } else {
4592 fatal("pthread_condattr_setclock: %s", strerror(status));
4593 }
4594 }
4595 }
4596 // else it defaults to CLOCK_REALTIME
4597
4598 // retrieve entry point for pthread_setname_np
4599 Linux::_pthread_setname_np =
4600 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4601
4602 }
4603
4604 // To install functions for atexit system call
4605 extern "C" {
4606 static void perfMemory_exit_helper() {
4607 perfMemory_exit();
4608 }
4609 }
4610
4611 // this is called _after_ the global arguments have been parsed
4612 jint os::init_2(void) {
4613 Linux::fast_thread_clock_init();
4614
4615 // Allocate a single page and mark it as readable for safepoint polling
4616 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4617 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4618
4619 os::set_polling_page(polling_page);
4620
4621 #ifndef PRODUCT
4622 if (Verbose && PrintMiscellaneous) {
4623 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4624 (intptr_t)polling_page);
4625 }
4626 #endif
4627
4628 if (!UseMembar) {
4629 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4630 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4631 os::set_memory_serialize_page(mem_serialize_page);
4632
4633 #ifndef PRODUCT
4634 if (Verbose && PrintMiscellaneous) {
4635 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4636 (intptr_t)mem_serialize_page);
4637 }
4638 #endif
4639 }
4640
4641 // initialize suspend/resume support - must do this before signal_sets_init()
4642 if (SR_initialize() != 0) {
4643 perror("SR_initialize failed");
4644 return JNI_ERR;
4645 }
4646
4647 Linux::signal_sets_init();
4648 Linux::install_signal_handlers();
4649
4650 // Check minimum allowable stack size for thread creation and to initialize
4651 // the java system classes, including StackOverflowError - depends on page
4652 // size. Add a page for compiler2 recursion in main thread.
4653 // Add in 2*BytesPerWord times page size to account for VM stack during
4654 // class initialization depending on 32 or 64 bit VM.
4655 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4656 JavaThread::stack_guard_zone_size() +
4657 JavaThread::stack_shadow_zone_size() +
4658 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4659
4660 size_t threadStackSizeInBytes = ThreadStackSize * K;
4661 if (threadStackSizeInBytes != 0 &&
4662 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4663 tty->print_cr("\nThe stack size specified is too small, "
4664 "Specify at least " SIZE_FORMAT "k",
4665 os::Linux::min_stack_allowed/ K);
4666 return JNI_ERR;
4667 }
4668
4669 // Make the stack size a multiple of the page size so that
4670 // the yellow/red zones can be guarded.
4671 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4672 vm_page_size()));
4673
4674 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4675
4676 #if defined(IA32)
4677 workaround_expand_exec_shield_cs_limit();
4678 #endif
4679
4680 Linux::libpthread_init();
4681 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4682 tty->print_cr("[HotSpot is running with %s, %s]\n",
4683 Linux::glibc_version(), Linux::libpthread_version());
4684 }
4685
4686 if (UseNUMA) {
4687 if (!Linux::libnuma_init()) {
4688 UseNUMA = false;
4689 } else {
4690 if ((Linux::numa_max_node() < 1)) {
4691 // There's only one node(they start from 0), disable NUMA.
4692 UseNUMA = false;
4693 }
4694 }
4695 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4696 // we can make the adaptive lgrp chunk resizing work. If the user specified
4697 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4698 // disable adaptive resizing.
4699 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4700 if (FLAG_IS_DEFAULT(UseNUMA)) {
4701 UseNUMA = false;
4702 } else {
4703 if (FLAG_IS_DEFAULT(UseLargePages) &&
4704 FLAG_IS_DEFAULT(UseSHM) &&
4705 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4706 UseLargePages = false;
4707 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4708 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4709 UseAdaptiveSizePolicy = false;
4710 UseAdaptiveNUMAChunkSizing = false;
4711 }
4712 }
4713 }
4714 if (!UseNUMA && ForceNUMA) {
4715 UseNUMA = true;
4716 }
4717 }
4718
4719 if (MaxFDLimit) {
4720 // set the number of file descriptors to max. print out error
4721 // if getrlimit/setrlimit fails but continue regardless.
4722 struct rlimit nbr_files;
4723 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4724 if (status != 0) {
4725 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4726 perror("os::init_2 getrlimit failed");
4727 }
4728 } else {
4729 nbr_files.rlim_cur = nbr_files.rlim_max;
4730 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4731 if (status != 0) {
4732 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4733 perror("os::init_2 setrlimit failed");
4734 }
4735 }
4736 }
4737 }
4738
4739 // Initialize lock used to serialize thread creation (see os::create_thread)
4740 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4741
4742 // at-exit methods are called in the reverse order of their registration.
4743 // atexit functions are called on return from main or as a result of a
4744 // call to exit(3C). There can be only 32 of these functions registered
4745 // and atexit() does not set errno.
4746
4747 if (PerfAllowAtExitRegistration) {
4748 // only register atexit functions if PerfAllowAtExitRegistration is set.
4749 // atexit functions can be delayed until process exit time, which
4750 // can be problematic for embedded VM situations. Embedded VMs should
4751 // call DestroyJavaVM() to assure that VM resources are released.
4752
4753 // note: perfMemory_exit_helper atexit function may be removed in
4754 // the future if the appropriate cleanup code can be added to the
4755 // VM_Exit VMOperation's doit method.
4756 if (atexit(perfMemory_exit_helper) != 0) {
4757 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4758 }
4759 }
4760
4761 // initialize thread priority policy
4762 prio_init();
4763
4764 return JNI_OK;
4765 }
4766
4767 // Mark the polling page as unreadable
4768 void os::make_polling_page_unreadable(void) {
4769 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4770 fatal("Could not disable polling page");
4771 }
4772 }
4773
4774 // Mark the polling page as readable
4775 void os::make_polling_page_readable(void) {
4776 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4777 fatal("Could not enable polling page");
4778 }
4779 }
4780
4781 // older glibc versions don't have this macro (which expands to
4782 // an optimized bit-counting function) so we have to roll our own
4783 #ifndef CPU_COUNT
4784
4785 static int _cpu_count(const cpu_set_t* cpus) {
4786 int count = 0;
4787 // only look up to the number of configured processors
4788 for (int i = 0; i < os::processor_count(); i++) {
4789 if (CPU_ISSET(i, cpus)) {
4790 count++;
4791 }
4792 }
4793 return count;
4794 }
4795
4796 #define CPU_COUNT(cpus) _cpu_count(cpus)
4797
4798 #endif // CPU_COUNT
4799
4800 // Get the current number of available processors for this process.
4801 // This value can change at any time during a process's lifetime.
4802 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4803 // If it appears there may be more than 1024 processors then we do a
4804 // dynamic check - see 6515172 for details.
4805 // If anything goes wrong we fallback to returning the number of online
4806 // processors - which can be greater than the number available to the process.
4807 int os::active_processor_count() {
4808 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4809 cpu_set_t* cpus_p = &cpus;
4810 int cpus_size = sizeof(cpu_set_t);
4811
4812 int configured_cpus = processor_count(); // upper bound on available cpus
4813 int cpu_count = 0;
4814
4815 // old build platforms may not support dynamic cpu sets
4816 #ifdef CPU_ALLOC
4817
4818 // To enable easy testing of the dynamic path on different platforms we
4819 // introduce a diagnostic flag: UseCpuAllocPath
4820 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4821 // kernel may use a mask bigger than cpu_set_t
4822 log_trace(os)("active_processor_count: using dynamic path %s"
4823 "- configured processors: %d",
4824 UseCpuAllocPath ? "(forced) " : "",
4825 configured_cpus);
4826 cpus_p = CPU_ALLOC(configured_cpus);
4827 if (cpus_p != NULL) {
4828 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4829 // zero it just to be safe
4830 CPU_ZERO_S(cpus_size, cpus_p);
4831 }
4832 else {
4833 // failed to allocate so fallback to online cpus
4834 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4835 log_trace(os)("active_processor_count: "
4836 "CPU_ALLOC failed (%s) - using "
4837 "online processor count: %d",
4838 strerror(errno), online_cpus);
4839 return online_cpus;
4840 }
4841 }
4842 else {
4843 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4844 configured_cpus);
4845 }
4846 #else // CPU_ALLOC
4847 // these stubs won't be executed
4848 #define CPU_COUNT_S(size, cpus) -1
4849 #define CPU_FREE(cpus)
4850
4851 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4852 configured_cpus);
4853 #endif // CPU_ALLOC
4854
4855 // pid 0 means the current thread - which we have to assume represents the process
4856 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4857 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4858 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4859 }
4860 else {
4861 cpu_count = CPU_COUNT(cpus_p);
4862 }
4863 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4864 }
4865 else {
4866 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4867 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4868 "which may exceed available processors", strerror(errno), cpu_count);
4869 }
4870
4871 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4872 CPU_FREE(cpus_p);
4873 }
4874
4875 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4876 return cpu_count;
4877 }
4878
4879 void os::set_native_thread_name(const char *name) {
4880 if (Linux::_pthread_setname_np) {
4881 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4882 snprintf(buf, sizeof(buf), "%s", name);
4883 buf[sizeof(buf) - 1] = '\0';
4884 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4885 // ERANGE should not happen; all other errors should just be ignored.
4886 assert(rc != ERANGE, "pthread_setname_np failed");
4887 }
4888 }
4889
4890 bool os::distribute_processes(uint length, uint* distribution) {
4891 // Not yet implemented.
4892 return false;
4893 }
4894
4895 bool os::bind_to_processor(uint processor_id) {
4896 // Not yet implemented.
4897 return false;
4898 }
4899
4900 ///
4901
4902 void os::SuspendedThreadTask::internal_do_task() {
4903 if (do_suspend(_thread->osthread())) {
4904 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4905 do_task(context);
4906 do_resume(_thread->osthread());
4907 }
4908 }
4909
4910 class PcFetcher : public os::SuspendedThreadTask {
4911 public:
4912 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4913 ExtendedPC result();
4914 protected:
4915 void do_task(const os::SuspendedThreadTaskContext& context);
4916 private:
4917 ExtendedPC _epc;
4918 };
4919
4920 ExtendedPC PcFetcher::result() {
4921 guarantee(is_done(), "task is not done yet.");
4922 return _epc;
4923 }
4924
4925 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4926 Thread* thread = context.thread();
4927 OSThread* osthread = thread->osthread();
4928 if (osthread->ucontext() != NULL) {
4929 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4930 } else {
4931 // NULL context is unexpected, double-check this is the VMThread
4932 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4933 }
4934 }
4935
4936 // Suspends the target using the signal mechanism and then grabs the PC before
4937 // resuming the target. Used by the flat-profiler only
4938 ExtendedPC os::get_thread_pc(Thread* thread) {
4939 // Make sure that it is called by the watcher for the VMThread
4940 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4941 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4942
4943 PcFetcher fetcher(thread);
4944 fetcher.run();
4945 return fetcher.result();
4946 }
4947
4948 ////////////////////////////////////////////////////////////////////////////////
4949 // debug support
4950
4951 bool os::find(address addr, outputStream* st) {
4952 Dl_info dlinfo;
4953 memset(&dlinfo, 0, sizeof(dlinfo));
4954 if (dladdr(addr, &dlinfo) != 0) {
4955 st->print(PTR_FORMAT ": ", p2i(addr));
4956 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4957 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4958 p2i(addr) - p2i(dlinfo.dli_saddr));
4959 } else if (dlinfo.dli_fbase != NULL) {
4960 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4961 } else {
4962 st->print("<absolute address>");
4963 }
4964 if (dlinfo.dli_fname != NULL) {
4965 st->print(" in %s", dlinfo.dli_fname);
4966 }
4967 if (dlinfo.dli_fbase != NULL) {
4968 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4969 }
4970 st->cr();
4971
4972 if (Verbose) {
4973 // decode some bytes around the PC
4974 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4975 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4976 address lowest = (address) dlinfo.dli_sname;
4977 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4978 if (begin < lowest) begin = lowest;
4979 Dl_info dlinfo2;
4980 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4981 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4982 end = (address) dlinfo2.dli_saddr;
4983 }
4984 Disassembler::decode(begin, end, st);
4985 }
4986 return true;
4987 }
4988 return false;
4989 }
4990
4991 ////////////////////////////////////////////////////////////////////////////////
4992 // misc
4993
4994 // This does not do anything on Linux. This is basically a hook for being
4995 // able to use structured exception handling (thread-local exception filters)
4996 // on, e.g., Win32.
4997 void
4998 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
4999 JavaCallArguments* args, Thread* thread) {
5000 f(value, method, args, thread);
5001 }
5002
5003 void os::print_statistics() {
5004 }
5005
5006 bool os::message_box(const char* title, const char* message) {
5007 int i;
5008 fdStream err(defaultStream::error_fd());
5009 for (i = 0; i < 78; i++) err.print_raw("=");
5010 err.cr();
5011 err.print_raw_cr(title);
5012 for (i = 0; i < 78; i++) err.print_raw("-");
5013 err.cr();
5014 err.print_raw_cr(message);
5015 for (i = 0; i < 78; i++) err.print_raw("=");
5016 err.cr();
5017
5018 char buf[16];
5019 // Prevent process from exiting upon "read error" without consuming all CPU
5020 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5021
5022 return buf[0] == 'y' || buf[0] == 'Y';
5023 }
5024
5025 int os::stat(const char *path, struct stat *sbuf) {
5026 char pathbuf[MAX_PATH];
5027 if (strlen(path) > MAX_PATH - 1) {
5028 errno = ENAMETOOLONG;
5029 return -1;
5030 }
5031 os::native_path(strcpy(pathbuf, path));
5032 return ::stat(pathbuf, sbuf);
5033 }
5034
5035 bool os::check_heap(bool force) {
5036 return true;
5037 }
5038
5039 // Is a (classpath) directory empty?
5040 bool os::dir_is_empty(const char* path) {
5041 DIR *dir = NULL;
5042 struct dirent *ptr;
5043
5044 dir = opendir(path);
5045 if (dir == NULL) return true;
5046
5047 // Scan the directory
5048 bool result = true;
5049 char buf[sizeof(struct dirent) + MAX_PATH];
5050 while (result && (ptr = ::readdir(dir)) != NULL) {
5051 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5052 result = false;
5053 }
5054 }
5055 closedir(dir);
5056 return result;
5057 }
5058
5059 // This code originates from JDK's sysOpen and open64_w
5060 // from src/solaris/hpi/src/system_md.c
5061
5062 int os::open(const char *path, int oflag, int mode) {
5063 if (strlen(path) > MAX_PATH - 1) {
5064 errno = ENAMETOOLONG;
5065 return -1;
5066 }
5067
5068 // All file descriptors that are opened in the Java process and not
5069 // specifically destined for a subprocess should have the close-on-exec
5070 // flag set. If we don't set it, then careless 3rd party native code
5071 // might fork and exec without closing all appropriate file descriptors
5072 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5073 // turn might:
5074 //
5075 // - cause end-of-file to fail to be detected on some file
5076 // descriptors, resulting in mysterious hangs, or
5077 //
5078 // - might cause an fopen in the subprocess to fail on a system
5079 // suffering from bug 1085341.
5080 //
5081 // (Yes, the default setting of the close-on-exec flag is a Unix
5082 // design flaw)
5083 //
5084 // See:
5085 // 1085341: 32-bit stdio routines should support file descriptors >255
5086 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5087 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5088 //
5089 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5090 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5091 // because it saves a system call and removes a small window where the flag
5092 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5093 // and we fall back to using FD_CLOEXEC (see below).
5094 #ifdef O_CLOEXEC
5095 oflag |= O_CLOEXEC;
5096 #endif
5097
5098 int fd = ::open64(path, oflag, mode);
5099 if (fd == -1) return -1;
5100
5101 //If the open succeeded, the file might still be a directory
5102 {
5103 struct stat64 buf64;
5104 int ret = ::fstat64(fd, &buf64);
5105 int st_mode = buf64.st_mode;
5106
5107 if (ret != -1) {
5108 if ((st_mode & S_IFMT) == S_IFDIR) {
5109 errno = EISDIR;
5110 ::close(fd);
5111 return -1;
5112 }
5113 } else {
5114 ::close(fd);
5115 return -1;
5116 }
5117 }
5118
5119 #ifdef FD_CLOEXEC
5120 // Validate that the use of the O_CLOEXEC flag on open above worked.
5121 // With recent kernels, we will perform this check exactly once.
5122 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5123 if (!O_CLOEXEC_is_known_to_work) {
5124 int flags = ::fcntl(fd, F_GETFD);
5125 if (flags != -1) {
5126 if ((flags & FD_CLOEXEC) != 0)
5127 O_CLOEXEC_is_known_to_work = 1;
5128 else
5129 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5130 }
5131 }
5132 #endif
5133
5134 return fd;
5135 }
5136
5137
5138 // create binary file, rewriting existing file if required
5139 int os::create_binary_file(const char* path, bool rewrite_existing) {
5140 int oflags = O_WRONLY | O_CREAT;
5141 if (!rewrite_existing) {
5142 oflags |= O_EXCL;
5143 }
5144 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5145 }
5146
5147 // return current position of file pointer
5148 jlong os::current_file_offset(int fd) {
5149 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5150 }
5151
5152 // move file pointer to the specified offset
5153 jlong os::seek_to_file_offset(int fd, jlong offset) {
5154 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5155 }
5156
5157 // This code originates from JDK's sysAvailable
5158 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5159
5160 int os::available(int fd, jlong *bytes) {
5161 jlong cur, end;
5162 int mode;
5163 struct stat64 buf64;
5164
5165 if (::fstat64(fd, &buf64) >= 0) {
5166 mode = buf64.st_mode;
5167 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5168 int n;
5169 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5170 *bytes = n;
5171 return 1;
5172 }
5173 }
5174 }
5175 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5176 return 0;
5177 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5178 return 0;
5179 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5180 return 0;
5181 }
5182 *bytes = end - cur;
5183 return 1;
5184 }
5185
5186 // Map a block of memory.
5187 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5188 char *addr, size_t bytes, bool read_only,
5189 bool allow_exec) {
5190 int prot;
5191 int flags = MAP_PRIVATE;
5192
5193 if (read_only) {
5194 prot = PROT_READ;
5195 } else {
5196 prot = PROT_READ | PROT_WRITE;
5197 }
5198
5199 if (allow_exec) {
5200 prot |= PROT_EXEC;
5201 }
5202
5203 if (addr != NULL) {
5204 flags |= MAP_FIXED;
5205 }
5206
5207 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5208 fd, file_offset);
5209 if (mapped_address == MAP_FAILED) {
5210 return NULL;
5211 }
5212 return mapped_address;
5213 }
5214
5215
5216 // Remap a block of memory.
5217 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5218 char *addr, size_t bytes, bool read_only,
5219 bool allow_exec) {
5220 // same as map_memory() on this OS
5221 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5222 allow_exec);
5223 }
5224
5225
5226 // Unmap a block of memory.
5227 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5228 return munmap(addr, bytes) == 0;
5229 }
5230
5231 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5232
5233 static clockid_t thread_cpu_clockid(Thread* thread) {
5234 pthread_t tid = thread->osthread()->pthread_id();
5235 clockid_t clockid;
5236
5237 // Get thread clockid
5238 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5239 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5240 return clockid;
5241 }
5242
5243 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5244 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5245 // of a thread.
5246 //
5247 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5248 // the fast estimate available on the platform.
5249
5250 jlong os::current_thread_cpu_time() {
5251 if (os::Linux::supports_fast_thread_cpu_time()) {
5252 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5253 } else {
5254 // return user + sys since the cost is the same
5255 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5256 }
5257 }
5258
5259 jlong os::thread_cpu_time(Thread* thread) {
5260 // consistent with what current_thread_cpu_time() returns
5261 if (os::Linux::supports_fast_thread_cpu_time()) {
5262 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5263 } else {
5264 return slow_thread_cpu_time(thread, true /* user + sys */);
5265 }
5266 }
5267
5268 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5269 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5270 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5271 } else {
5272 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5273 }
5274 }
5275
5276 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5277 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5278 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5279 } else {
5280 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5281 }
5282 }
5283
5284 // -1 on error.
5285 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5286 pid_t tid = thread->osthread()->thread_id();
5287 char *s;
5288 char stat[2048];
5289 int statlen;
5290 char proc_name[64];
5291 int count;
5292 long sys_time, user_time;
5293 char cdummy;
5294 int idummy;
5295 long ldummy;
5296 FILE *fp;
5297
5298 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5299 fp = fopen(proc_name, "r");
5300 if (fp == NULL) return -1;
5301 statlen = fread(stat, 1, 2047, fp);
5302 stat[statlen] = '\0';
5303 fclose(fp);
5304
5305 // Skip pid and the command string. Note that we could be dealing with
5306 // weird command names, e.g. user could decide to rename java launcher
5307 // to "java 1.4.2 :)", then the stat file would look like
5308 // 1234 (java 1.4.2 :)) R ... ...
5309 // We don't really need to know the command string, just find the last
5310 // occurrence of ")" and then start parsing from there. See bug 4726580.
5311 s = strrchr(stat, ')');
5312 if (s == NULL) return -1;
5313
5314 // Skip blank chars
5315 do { s++; } while (s && isspace(*s));
5316
5317 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5318 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5319 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5320 &user_time, &sys_time);
5321 if (count != 13) return -1;
5322 if (user_sys_cpu_time) {
5323 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5324 } else {
5325 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5326 }
5327 }
5328
5329 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5330 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5331 info_ptr->may_skip_backward = false; // elapsed time not wall time
5332 info_ptr->may_skip_forward = false; // elapsed time not wall time
5333 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5334 }
5335
5336 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5337 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5338 info_ptr->may_skip_backward = false; // elapsed time not wall time
5339 info_ptr->may_skip_forward = false; // elapsed time not wall time
5340 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5341 }
5342
5343 bool os::is_thread_cpu_time_supported() {
5344 return true;
5345 }
5346
5347 // System loadavg support. Returns -1 if load average cannot be obtained.
5348 // Linux doesn't yet have a (official) notion of processor sets,
5349 // so just return the system wide load average.
5350 int os::loadavg(double loadavg[], int nelem) {
5351 return ::getloadavg(loadavg, nelem);
5352 }
5353
5354 void os::pause() {
5355 char filename[MAX_PATH];
5356 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5357 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5358 } else {
5359 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5360 }
5361
5362 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5363 if (fd != -1) {
5364 struct stat buf;
5365 ::close(fd);
5366 while (::stat(filename, &buf) == 0) {
5367 (void)::poll(NULL, 0, 100);
5368 }
5369 } else {
5370 jio_fprintf(stderr,
5371 "Could not open pause file '%s', continuing immediately.\n", filename);
5372 }
5373 }
5374
5375
5376 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5377 // comment paragraphs are worth repeating here:
5378 //
5379 // Assumption:
5380 // Only one parker can exist on an event, which is why we allocate
5381 // them per-thread. Multiple unparkers can coexist.
5382 //
5383 // _Event serves as a restricted-range semaphore.
5384 // -1 : thread is blocked, i.e. there is a waiter
5385 // 0 : neutral: thread is running or ready,
5386 // could have been signaled after a wait started
5387 // 1 : signaled - thread is running or ready
5388 //
5389
5390 // utility to compute the abstime argument to timedwait:
5391 // millis is the relative timeout time
5392 // abstime will be the absolute timeout time
5393 // TODO: replace compute_abstime() with unpackTime()
5394
5395 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5396 if (millis < 0) millis = 0;
5397
5398 jlong seconds = millis / 1000;
5399 millis %= 1000;
5400 if (seconds > 50000000) { // see man cond_timedwait(3T)
5401 seconds = 50000000;
5402 }
5403
5404 if (os::supports_monotonic_clock()) {
5405 struct timespec now;
5406 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5407 assert_status(status == 0, status, "clock_gettime");
5408 abstime->tv_sec = now.tv_sec + seconds;
5409 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5410 if (nanos >= NANOSECS_PER_SEC) {
5411 abstime->tv_sec += 1;
5412 nanos -= NANOSECS_PER_SEC;
5413 }
5414 abstime->tv_nsec = nanos;
5415 } else {
5416 struct timeval now;
5417 int status = gettimeofday(&now, NULL);
5418 assert(status == 0, "gettimeofday");
5419 abstime->tv_sec = now.tv_sec + seconds;
5420 long usec = now.tv_usec + millis * 1000;
5421 if (usec >= 1000000) {
5422 abstime->tv_sec += 1;
5423 usec -= 1000000;
5424 }
5425 abstime->tv_nsec = usec * 1000;
5426 }
5427 return abstime;
5428 }
5429
5430 void os::PlatformEvent::park() { // AKA "down()"
5431 // Transitions for _Event:
5432 // -1 => -1 : illegal
5433 // 1 => 0 : pass - return immediately
5434 // 0 => -1 : block; then set _Event to 0 before returning
5435
5436 // Invariant: Only the thread associated with the Event/PlatformEvent
5437 // may call park().
5438 // TODO: assert that _Assoc != NULL or _Assoc == Self
5439 assert(_nParked == 0, "invariant");
5440
5441 int v;
5442 for (;;) {
5443 v = _Event;
5444 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5445 }
5446 guarantee(v >= 0, "invariant");
5447 if (v == 0) {
5448 // Do this the hard way by blocking ...
5449 int status = pthread_mutex_lock(_mutex);
5450 assert_status(status == 0, status, "mutex_lock");
5451 guarantee(_nParked == 0, "invariant");
5452 ++_nParked;
5453 while (_Event < 0) {
5454 status = pthread_cond_wait(_cond, _mutex);
5455 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5456 // Treat this the same as if the wait was interrupted
5457 if (status == ETIME) { status = EINTR; }
5458 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5459 }
5460 --_nParked;
5461
5462 _Event = 0;
5463 status = pthread_mutex_unlock(_mutex);
5464 assert_status(status == 0, status, "mutex_unlock");
5465 // Paranoia to ensure our locked and lock-free paths interact
5466 // correctly with each other.
5467 OrderAccess::fence();
5468 }
5469 guarantee(_Event >= 0, "invariant");
5470 }
5471
5472 int os::PlatformEvent::park(jlong millis) {
5473 // Transitions for _Event:
5474 // -1 => -1 : illegal
5475 // 1 => 0 : pass - return immediately
5476 // 0 => -1 : block; then set _Event to 0 before returning
5477
5478 guarantee(_nParked == 0, "invariant");
5479
5480 int v;
5481 for (;;) {
5482 v = _Event;
5483 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5484 }
5485 guarantee(v >= 0, "invariant");
5486 if (v != 0) return OS_OK;
5487
5488 // We do this the hard way, by blocking the thread.
5489 // Consider enforcing a minimum timeout value.
5490 struct timespec abst;
5491 compute_abstime(&abst, millis);
5492
5493 int ret = OS_TIMEOUT;
5494 int status = pthread_mutex_lock(_mutex);
5495 assert_status(status == 0, status, "mutex_lock");
5496 guarantee(_nParked == 0, "invariant");
5497 ++_nParked;
5498
5499 // Object.wait(timo) will return because of
5500 // (a) notification
5501 // (b) timeout
5502 // (c) thread.interrupt
5503 //
5504 // Thread.interrupt and object.notify{All} both call Event::set.
5505 // That is, we treat thread.interrupt as a special case of notification.
5506 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5507 // We assume all ETIME returns are valid.
5508 //
5509 // TODO: properly differentiate simultaneous notify+interrupt.
5510 // In that case, we should propagate the notify to another waiter.
5511
5512 while (_Event < 0) {
5513 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5514 assert_status(status == 0 || status == EINTR ||
5515 status == ETIME || status == ETIMEDOUT,
5516 status, "cond_timedwait");
5517 if (!FilterSpuriousWakeups) break; // previous semantics
5518 if (status == ETIME || status == ETIMEDOUT) break;
5519 // We consume and ignore EINTR and spurious wakeups.
5520 }
5521 --_nParked;
5522 if (_Event >= 0) {
5523 ret = OS_OK;
5524 }
5525 _Event = 0;
5526 status = pthread_mutex_unlock(_mutex);
5527 assert_status(status == 0, status, "mutex_unlock");
5528 assert(_nParked == 0, "invariant");
5529 // Paranoia to ensure our locked and lock-free paths interact
5530 // correctly with each other.
5531 OrderAccess::fence();
5532 return ret;
5533 }
5534
5535 void os::PlatformEvent::unpark() {
5536 // Transitions for _Event:
5537 // 0 => 1 : just return
5538 // 1 => 1 : just return
5539 // -1 => either 0 or 1; must signal target thread
5540 // That is, we can safely transition _Event from -1 to either
5541 // 0 or 1.
5542 // See also: "Semaphores in Plan 9" by Mullender & Cox
5543 //
5544 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5545 // that it will take two back-to-back park() calls for the owning
5546 // thread to block. This has the benefit of forcing a spurious return
5547 // from the first park() call after an unpark() call which will help
5548 // shake out uses of park() and unpark() without condition variables.
5549
5550 if (Atomic::xchg(1, &_Event) >= 0) return;
5551
5552 // Wait for the thread associated with the event to vacate
5553 int status = pthread_mutex_lock(_mutex);
5554 assert_status(status == 0, status, "mutex_lock");
5555 int AnyWaiters = _nParked;
5556 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5557 status = pthread_mutex_unlock(_mutex);
5558 assert_status(status == 0, status, "mutex_unlock");
5559 if (AnyWaiters != 0) {
5560 // Note that we signal() *after* dropping the lock for "immortal" Events.
5561 // This is safe and avoids a common class of futile wakeups. In rare
5562 // circumstances this can cause a thread to return prematurely from
5563 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5564 // will simply re-test the condition and re-park itself.
5565 // This provides particular benefit if the underlying platform does not
5566 // provide wait morphing.
5567 status = pthread_cond_signal(_cond);
5568 assert_status(status == 0, status, "cond_signal");
5569 }
5570 }
5571
5572
5573 // JSR166
5574 // -------------------------------------------------------
5575
5576 // The solaris and linux implementations of park/unpark are fairly
5577 // conservative for now, but can be improved. They currently use a
5578 // mutex/condvar pair, plus a a count.
5579 // Park decrements count if > 0, else does a condvar wait. Unpark
5580 // sets count to 1 and signals condvar. Only one thread ever waits
5581 // on the condvar. Contention seen when trying to park implies that someone
5582 // is unparking you, so don't wait. And spurious returns are fine, so there
5583 // is no need to track notifications.
5584
5585 // This code is common to linux and solaris and will be moved to a
5586 // common place in dolphin.
5587 //
5588 // The passed in time value is either a relative time in nanoseconds
5589 // or an absolute time in milliseconds. Either way it has to be unpacked
5590 // into suitable seconds and nanoseconds components and stored in the
5591 // given timespec structure.
5592 // Given time is a 64-bit value and the time_t used in the timespec is only
5593 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5594 // overflow if times way in the future are given. Further on Solaris versions
5595 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5596 // number of seconds, in abstime, is less than current_time + 100,000,000.
5597 // As it will be 28 years before "now + 100000000" will overflow we can
5598 // ignore overflow and just impose a hard-limit on seconds using the value
5599 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5600 // years from "now".
5601
5602 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5603 assert(time > 0, "convertTime");
5604 time_t max_secs = 0;
5605
5606 if (!os::supports_monotonic_clock() || isAbsolute) {
5607 struct timeval now;
5608 int status = gettimeofday(&now, NULL);
5609 assert(status == 0, "gettimeofday");
5610
5611 max_secs = now.tv_sec + MAX_SECS;
5612
5613 if (isAbsolute) {
5614 jlong secs = time / 1000;
5615 if (secs > max_secs) {
5616 absTime->tv_sec = max_secs;
5617 } else {
5618 absTime->tv_sec = secs;
5619 }
5620 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5621 } else {
5622 jlong secs = time / NANOSECS_PER_SEC;
5623 if (secs >= MAX_SECS) {
5624 absTime->tv_sec = max_secs;
5625 absTime->tv_nsec = 0;
5626 } else {
5627 absTime->tv_sec = now.tv_sec + secs;
5628 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5629 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5630 absTime->tv_nsec -= NANOSECS_PER_SEC;
5631 ++absTime->tv_sec; // note: this must be <= max_secs
5632 }
5633 }
5634 }
5635 } else {
5636 // must be relative using monotonic clock
5637 struct timespec now;
5638 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5639 assert_status(status == 0, status, "clock_gettime");
5640 max_secs = now.tv_sec + MAX_SECS;
5641 jlong secs = time / NANOSECS_PER_SEC;
5642 if (secs >= MAX_SECS) {
5643 absTime->tv_sec = max_secs;
5644 absTime->tv_nsec = 0;
5645 } else {
5646 absTime->tv_sec = now.tv_sec + secs;
5647 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5648 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5649 absTime->tv_nsec -= NANOSECS_PER_SEC;
5650 ++absTime->tv_sec; // note: this must be <= max_secs
5651 }
5652 }
5653 }
5654 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5655 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5656 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5657 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5658 }
5659
5660 void Parker::park(bool isAbsolute, jlong time) {
5661 // Ideally we'd do something useful while spinning, such
5662 // as calling unpackTime().
5663
5664 // Optional fast-path check:
5665 // Return immediately if a permit is available.
5666 // We depend on Atomic::xchg() having full barrier semantics
5667 // since we are doing a lock-free update to _counter.
5668 if (Atomic::xchg(0, &_counter) > 0) return;
5669
5670 Thread* thread = Thread::current();
5671 assert(thread->is_Java_thread(), "Must be JavaThread");
5672 JavaThread *jt = (JavaThread *)thread;
5673
5674 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5675 // Check interrupt before trying to wait
5676 if (Thread::is_interrupted(thread, false)) {
5677 return;
5678 }
5679
5680 // Next, demultiplex/decode time arguments
5681 timespec absTime;
5682 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5683 return;
5684 }
5685 if (time > 0) {
5686 unpackTime(&absTime, isAbsolute, time);
5687 }
5688
5689
5690 // Enter safepoint region
5691 // Beware of deadlocks such as 6317397.
5692 // The per-thread Parker:: mutex is a classic leaf-lock.
5693 // In particular a thread must never block on the Threads_lock while
5694 // holding the Parker:: mutex. If safepoints are pending both the
5695 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5696 ThreadBlockInVM tbivm(jt);
5697
5698 // Don't wait if cannot get lock since interference arises from
5699 // unblocking. Also. check interrupt before trying wait
5700 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5701 return;
5702 }
5703
5704 int status;
5705 if (_counter > 0) { // no wait needed
5706 _counter = 0;
5707 status = pthread_mutex_unlock(_mutex);
5708 assert_status(status == 0, status, "invariant");
5709 // Paranoia to ensure our locked and lock-free paths interact
5710 // correctly with each other and Java-level accesses.
5711 OrderAccess::fence();
5712 return;
5713 }
5714
5715 #ifdef ASSERT
5716 // Don't catch signals while blocked; let the running threads have the signals.
5717 // (This allows a debugger to break into the running thread.)
5718 sigset_t oldsigs;
5719 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5720 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5721 #endif
5722
5723 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5724 jt->set_suspend_equivalent();
5725 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5726
5727 assert(_cur_index == -1, "invariant");
5728 if (time == 0) {
5729 _cur_index = REL_INDEX; // arbitrary choice when not timed
5730 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5731 } else {
5732 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5733 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5734 }
5735 _cur_index = -1;
5736 assert_status(status == 0 || status == EINTR ||
5737 status == ETIME || status == ETIMEDOUT,
5738 status, "cond_timedwait");
5739
5740 #ifdef ASSERT
5741 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5742 #endif
5743
5744 _counter = 0;
5745 status = pthread_mutex_unlock(_mutex);
5746 assert_status(status == 0, status, "invariant");
5747 // Paranoia to ensure our locked and lock-free paths interact
5748 // correctly with each other and Java-level accesses.
5749 OrderAccess::fence();
5750
5751 // If externally suspended while waiting, re-suspend
5752 if (jt->handle_special_suspend_equivalent_condition()) {
5753 jt->java_suspend_self();
5754 }
5755 }
5756
5757 void Parker::unpark() {
5758 int status = pthread_mutex_lock(_mutex);
5759 assert_status(status == 0, status, "invariant");
5760 const int s = _counter;
5761 _counter = 1;
5762 // must capture correct index before unlocking
5763 int index = _cur_index;
5764 status = pthread_mutex_unlock(_mutex);
5765 assert_status(status == 0, status, "invariant");
5766 if (s < 1 && index != -1) {
5767 // thread is definitely parked
5768 status = pthread_cond_signal(&_cond[index]);
5769 assert_status(status == 0, status, "invariant");
5770 }
5771 }
5772
5773
5774 extern char** environ;
5775
5776 // Run the specified command in a separate process. Return its exit value,
5777 // or -1 on failure (e.g. can't fork a new process).
5778 // Unlike system(), this function can be called from signal handler. It
5779 // doesn't block SIGINT et al.
5780 int os::fork_and_exec(char* cmd) {
5781 const char * argv[4] = {"sh", "-c", cmd, NULL};
5782
5783 pid_t pid = fork();
5784
5785 if (pid < 0) {
5786 // fork failed
5787 return -1;
5788
5789 } else if (pid == 0) {
5790 // child process
5791
5792 execve("/bin/sh", (char* const*)argv, environ);
5793
5794 // execve failed
5795 _exit(-1);
5796
5797 } else {
5798 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5799 // care about the actual exit code, for now.
5800
5801 int status;
5802
5803 // Wait for the child process to exit. This returns immediately if
5804 // the child has already exited. */
5805 while (waitpid(pid, &status, 0) < 0) {
5806 switch (errno) {
5807 case ECHILD: return 0;
5808 case EINTR: break;
5809 default: return -1;
5810 }
5811 }
5812
5813 if (WIFEXITED(status)) {
5814 // The child exited normally; get its exit code.
5815 return WEXITSTATUS(status);
5816 } else if (WIFSIGNALED(status)) {
5817 // The child exited because of a signal
5818 // The best value to return is 0x80 + signal number,
5819 // because that is what all Unix shells do, and because
5820 // it allows callers to distinguish between process exit and
5821 // process death by signal.
5822 return 0x80 + WTERMSIG(status);
5823 } else {
5824 // Unknown exit code; pass it through
5825 return status;
5826 }
5827 }
5828 }
5829
5830 // is_headless_jre()
5831 //
5832 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5833 // in order to report if we are running in a headless jre
5834 //
5835 // Since JDK8 xawt/libmawt.so was moved into the same directory
5836 // as libawt.so, and renamed libawt_xawt.so
5837 //
5838 bool os::is_headless_jre() {
5839 struct stat statbuf;
5840 char buf[MAXPATHLEN];
5841 char libmawtpath[MAXPATHLEN];
5842 const char *xawtstr = "/xawt/libmawt.so";
5843 const char *new_xawtstr = "/libawt_xawt.so";
5844 char *p;
5845
5846 // Get path to libjvm.so
5847 os::jvm_path(buf, sizeof(buf));
5848
5849 // Get rid of libjvm.so
5850 p = strrchr(buf, '/');
5851 if (p == NULL) {
5852 return false;
5853 } else {
5854 *p = '\0';
5855 }
5856
5857 // Get rid of client or server
5858 p = strrchr(buf, '/');
5859 if (p == NULL) {
5860 return false;
5861 } else {
5862 *p = '\0';
5863 }
5864
5865 // check xawt/libmawt.so
5866 strcpy(libmawtpath, buf);
5867 strcat(libmawtpath, xawtstr);
5868 if (::stat(libmawtpath, &statbuf) == 0) return false;
5869
5870 // check libawt_xawt.so
5871 strcpy(libmawtpath, buf);
5872 strcat(libmawtpath, new_xawtstr);
5873 if (::stat(libmawtpath, &statbuf) == 0) return false;
5874
5875 return true;
5876 }
5877
5878 // Get the default path to the core file
5879 // Returns the length of the string
5880 int os::get_core_path(char* buffer, size_t bufferSize) {
5881 /*
5882 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5883 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5884 */
5885 const int core_pattern_len = 129;
5886 char core_pattern[core_pattern_len] = {0};
5887
5888 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5889 if (core_pattern_file == -1) {
5890 return -1;
5891 }
5892
5893 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5894 ::close(core_pattern_file);
5895 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5896 return -1;
5897 }
5898 if (core_pattern[ret-1] == '\n') {
5899 core_pattern[ret-1] = '\0';
5900 } else {
5901 core_pattern[ret] = '\0';
5902 }
5903
5904 char *pid_pos = strstr(core_pattern, "%p");
5905 int written;
5906
5907 if (core_pattern[0] == '/') {
5908 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5909 } else {
5910 char cwd[PATH_MAX];
5911
5912 const char* p = get_current_directory(cwd, PATH_MAX);
5913 if (p == NULL) {
5914 return -1;
5915 }
5916
5917 if (core_pattern[0] == '|') {
5918 written = jio_snprintf(buffer, bufferSize,
5919 "\"%s\" (or dumping to %s/core.%d)",
5920 &core_pattern[1], p, current_process_id());
5921 } else {
5922 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5923 }
5924 }
5925
5926 if (written < 0) {
5927 return -1;
5928 }
5929
5930 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5931 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5932
5933 if (core_uses_pid_file != -1) {
5934 char core_uses_pid = 0;
5935 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5936 ::close(core_uses_pid_file);
5937
5938 if (core_uses_pid == '1') {
5939 jio_snprintf(buffer + written, bufferSize - written,
5940 ".%d", current_process_id());
5941 }
5942 }
5943 }
5944
5945 return strlen(buffer);
5946 }
5947
5948 bool os::start_debugging(char *buf, int buflen) {
5949 int len = (int)strlen(buf);
5950 char *p = &buf[len];
5951
5952 jio_snprintf(p, buflen-len,
5953 "\n\n"
5954 "Do you want to debug the problem?\n\n"
5955 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5956 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5957 "Otherwise, press RETURN to abort...",
5958 os::current_process_id(), os::current_process_id(),
5959 os::current_thread_id(), os::current_thread_id());
5960
5961 bool yes = os::message_box("Unexpected Error", buf);
5962
5963 if (yes) {
5964 // yes, user asked VM to launch debugger
5965 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
5966 os::current_process_id(), os::current_process_id());
5967
5968 os::fork_and_exec(buf);
5969 yes = false;
5970 }
5971 return yes;
5972 }
5973
5974
5975
5976 /////////////// Unit tests ///////////////
5977
5978 #ifndef PRODUCT
5979
5980 #define test_log(...) \
5981 do { \
5982 if (VerboseInternalVMTests) { \
5983 tty->print_cr(__VA_ARGS__); \
5984 tty->flush(); \
5985 } \
5986 } while (false)
5987
5988 class TestReserveMemorySpecial : AllStatic {
5989 public:
5990 static void small_page_write(void* addr, size_t size) {
5991 size_t page_size = os::vm_page_size();
5992
5993 char* end = (char*)addr + size;
5994 for (char* p = (char*)addr; p < end; p += page_size) {
5995 *p = 1;
5996 }
5997 }
5998
5999 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6000 if (!UseHugeTLBFS) {
6001 return;
6002 }
6003
6004 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6005
6006 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6007
6008 if (addr != NULL) {
6009 small_page_write(addr, size);
6010
6011 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6012 }
6013 }
6014
6015 static void test_reserve_memory_special_huge_tlbfs_only() {
6016 if (!UseHugeTLBFS) {
6017 return;
6018 }
6019
6020 size_t lp = os::large_page_size();
6021
6022 for (size_t size = lp; size <= lp * 10; size += lp) {
6023 test_reserve_memory_special_huge_tlbfs_only(size);
6024 }
6025 }
6026
6027 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6028 size_t lp = os::large_page_size();
6029 size_t ag = os::vm_allocation_granularity();
6030
6031 // sizes to test
6032 const size_t sizes[] = {
6033 lp, lp + ag, lp + lp / 2, lp * 2,
6034 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6035 lp * 10, lp * 10 + lp / 2
6036 };
6037 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6038
6039 // For each size/alignment combination, we test three scenarios:
6040 // 1) with req_addr == NULL
6041 // 2) with a non-null req_addr at which we expect to successfully allocate
6042 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6043 // expect the allocation to either fail or to ignore req_addr
6044
6045 // Pre-allocate two areas; they shall be as large as the largest allocation
6046 // and aligned to the largest alignment we will be testing.
6047 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6048 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6049 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6050 -1, 0);
6051 assert(mapping1 != MAP_FAILED, "should work");
6052
6053 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6054 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6055 -1, 0);
6056 assert(mapping2 != MAP_FAILED, "should work");
6057
6058 // Unmap the first mapping, but leave the second mapping intact: the first
6059 // mapping will serve as a value for a "good" req_addr (case 2). The second
6060 // mapping, still intact, as "bad" req_addr (case 3).
6061 ::munmap(mapping1, mapping_size);
6062
6063 // Case 1
6064 test_log("%s, req_addr NULL:", __FUNCTION__);
6065 test_log("size align result");
6066
6067 for (int i = 0; i < num_sizes; i++) {
6068 const size_t size = sizes[i];
6069 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6070 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6071 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6072 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6073 if (p != NULL) {
6074 assert(is_ptr_aligned(p, alignment), "must be");
6075 small_page_write(p, size);
6076 os::Linux::release_memory_special_huge_tlbfs(p, size);
6077 }
6078 }
6079 }
6080
6081 // Case 2
6082 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6083 test_log("size align req_addr result");
6084
6085 for (int i = 0; i < num_sizes; i++) {
6086 const size_t size = sizes[i];
6087 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6088 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6089 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6090 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6091 size, alignment, p2i(req_addr), p2i(p),
6092 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6093 if (p != NULL) {
6094 assert(p == req_addr, "must be");
6095 small_page_write(p, size);
6096 os::Linux::release_memory_special_huge_tlbfs(p, size);
6097 }
6098 }
6099 }
6100
6101 // Case 3
6102 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6103 test_log("size align req_addr result");
6104
6105 for (int i = 0; i < num_sizes; i++) {
6106 const size_t size = sizes[i];
6107 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6108 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6109 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6110 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6111 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6112 // as the area around req_addr contains already existing mappings, the API should always
6113 // return NULL (as per contract, it cannot return another address)
6114 assert(p == NULL, "must be");
6115 }
6116 }
6117
6118 ::munmap(mapping2, mapping_size);
6119
6120 }
6121
6122 static void test_reserve_memory_special_huge_tlbfs() {
6123 if (!UseHugeTLBFS) {
6124 return;
6125 }
6126
6127 test_reserve_memory_special_huge_tlbfs_only();
6128 test_reserve_memory_special_huge_tlbfs_mixed();
6129 }
6130
6131 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6132 if (!UseSHM) {
6133 return;
6134 }
6135
6136 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6137
6138 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6139
6140 if (addr != NULL) {
6141 assert(is_ptr_aligned(addr, alignment), "Check");
6142 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6143
6144 small_page_write(addr, size);
6145
6146 os::Linux::release_memory_special_shm(addr, size);
6147 }
6148 }
6149
6150 static void test_reserve_memory_special_shm() {
6151 size_t lp = os::large_page_size();
6152 size_t ag = os::vm_allocation_granularity();
6153
6154 for (size_t size = ag; size < lp * 3; size += ag) {
6155 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6156 test_reserve_memory_special_shm(size, alignment);
6157 }
6158 }
6159 }
6160
6161 static void test() {
6162 test_reserve_memory_special_huge_tlbfs();
6163 test_reserve_memory_special_shm();
6164 }
6165 };
6166
6167 void TestReserveMemorySpecial_test() {
6168 TestReserveMemorySpecial::test();
6169 }
6170
6171 #endif