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 uint32_t os::Linux::_os_version = 0;
148 const char * os::Linux::_glibc_version = NULL;
149 const char * os::Linux::_libpthread_version = NULL;
150 pthread_condattr_t os::Linux::_condattr[1];
151
152 static jlong initial_time_count=0;
153
154 static int clock_tics_per_sec = 100;
155
156 // For diagnostics to print a message once. see run_periodic_checks
157 static sigset_t check_signal_done;
158 static bool check_signals = true;
159
160 // Signal number used to suspend/resume a thread
161
162 // do not use any signal number less than SIGSEGV, see 4355769
163 static int SR_signum = SIGUSR2;
164 sigset_t SR_sigset;
165
166 // Declarations
167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
168
169 // utility functions
170
171 static int SR_initialize();
172
173 julong os::available_memory() {
174 return Linux::available_memory();
175 }
176
177 julong os::Linux::available_memory() {
178 // values in struct sysinfo are "unsigned long"
179 struct sysinfo si;
180 sysinfo(&si);
181
182 return (julong)si.freeram * si.mem_unit;
183 }
184
185 julong os::physical_memory() {
186 return Linux::physical_memory();
187 }
188
189 // Return true if user is running as root.
190
191 bool os::have_special_privileges() {
192 static bool init = false;
193 static bool privileges = false;
194 if (!init) {
195 privileges = (getuid() != geteuid()) || (getgid() != getegid());
196 init = true;
197 }
198 return privileges;
199 }
200
201
202 #ifndef SYS_gettid
203 // i386: 224, ia64: 1105, amd64: 186, sparc 143
204 #ifdef __ia64__
205 #define SYS_gettid 1105
206 #else
207 #ifdef __i386__
208 #define SYS_gettid 224
209 #else
210 #ifdef __amd64__
211 #define SYS_gettid 186
212 #else
213 #ifdef __sparc__
214 #define SYS_gettid 143
215 #else
216 #error define gettid for the arch
217 #endif
218 #endif
219 #endif
220 #endif
221 #endif
222
223 // Cpu architecture string
224 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
225
226
227 // pid_t gettid()
228 //
229 // Returns the kernel thread id of the currently running thread. Kernel
230 // thread id is used to access /proc.
231 pid_t os::Linux::gettid() {
232 int rslt = syscall(SYS_gettid);
233 assert(rslt != -1, "must be."); // old linuxthreads implementation?
234 return (pid_t)rslt;
235 }
236
237 // Most versions of linux have a bug where the number of processors are
238 // determined by looking at the /proc file system. In a chroot environment,
239 // the system call returns 1. This causes the VM to act as if it is
240 // a single processor and elide locking (see is_MP() call).
241 static bool unsafe_chroot_detected = false;
242 static const char *unstable_chroot_error = "/proc file system not found.\n"
243 "Java may be unstable running multithreaded in a chroot "
244 "environment on Linux when /proc filesystem is not mounted.";
245
246 void os::Linux::initialize_system_info() {
247 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
248 if (processor_count() == 1) {
249 pid_t pid = os::Linux::gettid();
250 char fname[32];
251 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
252 FILE *fp = fopen(fname, "r");
253 if (fp == NULL) {
254 unsafe_chroot_detected = true;
255 } else {
256 fclose(fp);
257 }
258 }
259 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
260 assert(processor_count() > 0, "linux error");
261 }
262
263 void os::init_system_properties_values() {
264 // The next steps are taken in the product version:
265 //
266 // Obtain the JAVA_HOME value from the location of libjvm.so.
267 // This library should be located at:
268 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
269 //
270 // If "/jre/lib/" appears at the right place in the path, then we
271 // assume libjvm.so is installed in a JDK and we use this path.
272 //
273 // Otherwise exit with message: "Could not create the Java virtual machine."
274 //
275 // The following extra steps are taken in the debugging version:
276 //
277 // If "/jre/lib/" does NOT appear at the right place in the path
278 // instead of exit check for $JAVA_HOME environment variable.
279 //
280 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
281 // then we append a fake suffix "hotspot/libjvm.so" to this path so
282 // it looks like libjvm.so is installed there
283 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
284 //
285 // Otherwise exit.
286 //
287 // Important note: if the location of libjvm.so changes this
288 // code needs to be changed accordingly.
289
290 // See ld(1):
291 // The linker uses the following search paths to locate required
292 // shared libraries:
293 // 1: ...
294 // ...
295 // 7: The default directories, normally /lib and /usr/lib.
296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
297 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
298 #else
299 #define DEFAULT_LIBPATH "/lib:/usr/lib"
300 #endif
301
302 // Base path of extensions installed on the system.
303 #define SYS_EXT_DIR "/usr/java/packages"
304 #define EXTENSIONS_DIR "/lib/ext"
305
306 // Buffer that fits several sprintfs.
307 // Note that the space for the colon and the trailing null are provided
308 // by the nulls included by the sizeof operator.
309 const size_t bufsize =
310 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
311 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
312 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
313
314 // sysclasspath, java_home, dll_dir
315 {
316 char *pslash;
317 os::jvm_path(buf, bufsize);
318
319 // Found the full path to libjvm.so.
320 // Now cut the path to <java_home>/jre if we can.
321 pslash = strrchr(buf, '/');
322 if (pslash != NULL) {
323 *pslash = '\0'; // Get rid of /libjvm.so.
324 }
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
328 }
329 Arguments::set_dll_dir(buf);
330
331 if (pslash != NULL) {
332 pslash = strrchr(buf, '/');
333 if (pslash != NULL) {
334 *pslash = '\0'; // Get rid of /<arch>.
335 pslash = strrchr(buf, '/');
336 if (pslash != NULL) {
337 *pslash = '\0'; // Get rid of /lib.
338 }
339 }
340 }
341 Arguments::set_java_home(buf);
342 set_boot_path('/', ':');
343 }
344
345 // Where to look for native libraries.
346 //
347 // Note: Due to a legacy implementation, most of the library path
348 // is set in the launcher. This was to accomodate linking restrictions
349 // on legacy Linux implementations (which are no longer supported).
350 // Eventually, all the library path setting will be done here.
351 //
352 // However, to prevent the proliferation of improperly built native
353 // libraries, the new path component /usr/java/packages is added here.
354 // Eventually, all the library path setting will be done here.
355 {
356 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
357 // should always exist (until the legacy problem cited above is
358 // addressed).
359 const char *v = ::getenv("LD_LIBRARY_PATH");
360 const char *v_colon = ":";
361 if (v == NULL) { v = ""; v_colon = ""; }
362 // That's +1 for the colon and +1 for the trailing '\0'.
363 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
364 strlen(v) + 1 +
365 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
366 mtInternal);
367 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
368 Arguments::set_library_path(ld_library_path);
369 FREE_C_HEAP_ARRAY(char, ld_library_path);
370 }
371
372 // Extensions directories.
373 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
374 Arguments::set_ext_dirs(buf);
375
376 FREE_C_HEAP_ARRAY(char, buf);
377
378 #undef DEFAULT_LIBPATH
379 #undef SYS_EXT_DIR
380 #undef EXTENSIONS_DIR
381 }
382
383 ////////////////////////////////////////////////////////////////////////////////
384 // breakpoint support
385
386 void os::breakpoint() {
387 BREAKPOINT;
388 }
389
390 extern "C" void breakpoint() {
391 // use debugger to set breakpoint here
392 }
393
394 ////////////////////////////////////////////////////////////////////////////////
395 // signal support
396
397 debug_only(static bool signal_sets_initialized = false);
398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
399
400 bool os::Linux::is_sig_ignored(int sig) {
401 struct sigaction oact;
402 sigaction(sig, (struct sigaction*)NULL, &oact);
403 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
404 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
405 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
406 return true;
407 } else {
408 return false;
409 }
410 }
411
412 void os::Linux::signal_sets_init() {
413 // Should also have an assertion stating we are still single-threaded.
414 assert(!signal_sets_initialized, "Already initialized");
415 // Fill in signals that are necessarily unblocked for all threads in
416 // the VM. Currently, we unblock the following signals:
417 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
418 // by -Xrs (=ReduceSignalUsage));
419 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
420 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
421 // the dispositions or masks wrt these signals.
422 // Programs embedding the VM that want to use the above signals for their
423 // own purposes must, at this time, use the "-Xrs" option to prevent
424 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
425 // (See bug 4345157, and other related bugs).
426 // In reality, though, unblocking these signals is really a nop, since
427 // these signals are not blocked by default.
428 sigemptyset(&unblocked_sigs);
429 sigemptyset(&allowdebug_blocked_sigs);
430 sigaddset(&unblocked_sigs, SIGILL);
431 sigaddset(&unblocked_sigs, SIGSEGV);
432 sigaddset(&unblocked_sigs, SIGBUS);
433 sigaddset(&unblocked_sigs, SIGFPE);
434 #if defined(PPC64)
435 sigaddset(&unblocked_sigs, SIGTRAP);
436 #endif
437 sigaddset(&unblocked_sigs, SR_signum);
438
439 if (!ReduceSignalUsage) {
440 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
441 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
443 }
444 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
445 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
447 }
448 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
449 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
450 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
451 }
452 }
453 // Fill in signals that are blocked by all but the VM thread.
454 sigemptyset(&vm_sigs);
455 if (!ReduceSignalUsage) {
456 sigaddset(&vm_sigs, BREAK_SIGNAL);
457 }
458 debug_only(signal_sets_initialized = true);
459
460 }
461
462 // These are signals that are unblocked while a thread is running Java.
463 // (For some reason, they get blocked by default.)
464 sigset_t* os::Linux::unblocked_signals() {
465 assert(signal_sets_initialized, "Not initialized");
466 return &unblocked_sigs;
467 }
468
469 // These are the signals that are blocked while a (non-VM) thread is
470 // running Java. Only the VM thread handles these signals.
471 sigset_t* os::Linux::vm_signals() {
472 assert(signal_sets_initialized, "Not initialized");
473 return &vm_sigs;
474 }
475
476 // These are signals that are blocked during cond_wait to allow debugger in
477 sigset_t* os::Linux::allowdebug_blocked_signals() {
478 assert(signal_sets_initialized, "Not initialized");
479 return &allowdebug_blocked_sigs;
480 }
481
482 void os::Linux::hotspot_sigmask(Thread* thread) {
483
484 //Save caller's signal mask before setting VM signal mask
485 sigset_t caller_sigmask;
486 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
487
488 OSThread* osthread = thread->osthread();
489 osthread->set_caller_sigmask(caller_sigmask);
490
491 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
492
493 if (!ReduceSignalUsage) {
494 if (thread->is_VM_thread()) {
495 // Only the VM thread handles BREAK_SIGNAL ...
496 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
497 } else {
498 // ... all other threads block BREAK_SIGNAL
499 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
500 }
501 }
502 }
503
504 //////////////////////////////////////////////////////////////////////////////
505 // detecting pthread library
506
507 void os::Linux::libpthread_init() {
508 // Save glibc and pthread version strings.
509 #if !defined(_CS_GNU_LIBC_VERSION) || \
510 !defined(_CS_GNU_LIBPTHREAD_VERSION)
511 #error "glibc too old (< 2.3.2)"
512 #endif
513
514 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
515 assert(n > 0, "cannot retrieve glibc version");
516 char *str = (char *)malloc(n, mtInternal);
517 confstr(_CS_GNU_LIBC_VERSION, str, n);
518 os::Linux::set_glibc_version(str);
519
520 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
521 assert(n > 0, "cannot retrieve pthread version");
522 str = (char *)malloc(n, mtInternal);
523 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
524 os::Linux::set_libpthread_version(str);
525 }
526
527 /////////////////////////////////////////////////////////////////////////////
528 // thread stack expansion
529
530 // os::Linux::manually_expand_stack() takes care of expanding the thread
531 // stack. Note that this is normally not needed: pthread stacks allocate
532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
533 // committed. Therefore it is not necessary to expand the stack manually.
534 //
535 // Manually expanding the stack was historically needed on LinuxThreads
536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
537 // it is kept to deal with very rare corner cases:
538 //
539 // For one, user may run the VM on an own implementation of threads
540 // whose stacks are - like the old LinuxThreads - implemented using
541 // mmap(MAP_GROWSDOWN).
542 //
543 // Also, this coding may be needed if the VM is running on the primordial
544 // thread. Normally we avoid running on the primordial thread; however,
545 // user may still invoke the VM on the primordial thread.
546 //
547 // The following historical comment describes the details about running
548 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
549
550
551 // Force Linux kernel to expand current thread stack. If "bottom" is close
552 // to the stack guard, caller should block all signals.
553 //
554 // MAP_GROWSDOWN:
555 // A special mmap() flag that is used to implement thread stacks. It tells
556 // kernel that the memory region should extend downwards when needed. This
557 // allows early versions of LinuxThreads to only mmap the first few pages
558 // when creating a new thread. Linux kernel will automatically expand thread
559 // stack as needed (on page faults).
560 //
561 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
562 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
563 // region, it's hard to tell if the fault is due to a legitimate stack
564 // access or because of reading/writing non-exist memory (e.g. buffer
565 // overrun). As a rule, if the fault happens below current stack pointer,
566 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
567 // application (see Linux kernel fault.c).
568 //
569 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
570 // stack overflow detection.
571 //
572 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
573 // not use MAP_GROWSDOWN.
574 //
575 // To get around the problem and allow stack banging on Linux, we need to
576 // manually expand thread stack after receiving the SIGSEGV.
577 //
578 // There are two ways to expand thread stack to address "bottom", we used
579 // both of them in JVM before 1.5:
580 // 1. adjust stack pointer first so that it is below "bottom", and then
581 // touch "bottom"
582 // 2. mmap() the page in question
583 //
584 // Now alternate signal stack is gone, it's harder to use 2. For instance,
585 // if current sp is already near the lower end of page 101, and we need to
586 // call mmap() to map page 100, it is possible that part of the mmap() frame
587 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
588 // That will destroy the mmap() frame and cause VM to crash.
589 //
590 // The following code works by adjusting sp first, then accessing the "bottom"
591 // page to force a page fault. Linux kernel will then automatically expand the
592 // stack mapping.
593 //
594 // _expand_stack_to() assumes its frame size is less than page size, which
595 // should always be true if the function is not inlined.
596
597 static void NOINLINE _expand_stack_to(address bottom) {
598 address sp;
599 size_t size;
600 volatile char *p;
601
602 // Adjust bottom to point to the largest address within the same page, it
603 // gives us a one-page buffer if alloca() allocates slightly more memory.
604 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
605 bottom += os::Linux::page_size() - 1;
606
607 // sp might be slightly above current stack pointer; if that's the case, we
608 // will alloca() a little more space than necessary, which is OK. Don't use
609 // os::current_stack_pointer(), as its result can be slightly below current
610 // stack pointer, causing us to not alloca enough to reach "bottom".
611 sp = (address)&sp;
612
613 if (sp > bottom) {
614 size = sp - bottom;
615 p = (volatile char *)alloca(size);
616 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
617 p[0] = '\0';
618 }
619 }
620
621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
622 assert(t!=NULL, "just checking");
623 assert(t->osthread()->expanding_stack(), "expand should be set");
624 assert(t->stack_base() != NULL, "stack_base was not initialized");
625
626 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
627 sigset_t mask_all, old_sigset;
628 sigfillset(&mask_all);
629 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
630 _expand_stack_to(addr);
631 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
632 return true;
633 }
634 return false;
635 }
636
637 //////////////////////////////////////////////////////////////////////////////
638 // create new thread
639
640 // Thread start routine for all newly created threads
641 static void *java_start(Thread *thread) {
642 // Try to randomize the cache line index of hot stack frames.
643 // This helps when threads of the same stack traces evict each other's
644 // cache lines. The threads can be either from the same JVM instance, or
645 // from different JVM instances. The benefit is especially true for
646 // processors with hyperthreading technology.
647 static int counter = 0;
648 int pid = os::current_process_id();
649 alloca(((pid ^ counter++) & 7) * 128);
650
651 thread->initialize_thread_current();
652
653 OSThread* osthread = thread->osthread();
654 Monitor* sync = osthread->startThread_lock();
655
656 osthread->set_thread_id(os::current_thread_id());
657
658 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
659 os::current_thread_id(), (uintx) pthread_self());
660
661 if (UseNUMA) {
662 int lgrp_id = os::numa_get_group_id();
663 if (lgrp_id != -1) {
664 thread->set_lgrp_id(lgrp_id);
665 }
666 }
667 // initialize signal mask for this thread
668 os::Linux::hotspot_sigmask(thread);
669
670 // initialize floating point control register
671 os::Linux::init_thread_fpu_state();
672
673 // handshaking with parent thread
674 {
675 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
676
677 // notify parent thread
678 osthread->set_state(INITIALIZED);
679 sync->notify_all();
680
681 // wait until os::start_thread()
682 while (osthread->get_state() == INITIALIZED) {
683 sync->wait(Mutex::_no_safepoint_check_flag);
684 }
685 }
686
687 // call one more level start routine
688 thread->run();
689
690 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
691 os::current_thread_id(), (uintx) pthread_self());
692
693 return 0;
694 }
695
696 bool os::create_thread(Thread* thread, ThreadType thr_type,
697 size_t stack_size) {
698 assert(thread->osthread() == NULL, "caller responsible");
699
700 // Allocate the OSThread object
701 OSThread* osthread = new OSThread(NULL, NULL);
702 if (osthread == NULL) {
703 return false;
704 }
705
706 // set the correct thread state
707 osthread->set_thread_type(thr_type);
708
709 // Initial state is ALLOCATED but not INITIALIZED
710 osthread->set_state(ALLOCATED);
711
712 thread->set_osthread(osthread);
713
714 // init thread attributes
715 pthread_attr_t attr;
716 pthread_attr_init(&attr);
717 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
718
719 // stack size
720 // calculate stack size if it's not specified by caller
721 if (stack_size == 0) {
722 stack_size = os::Linux::default_stack_size(thr_type);
723
724 switch (thr_type) {
725 case os::java_thread:
726 // Java threads use ThreadStackSize which default value can be
727 // changed with the flag -Xss
728 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
729 stack_size = JavaThread::stack_size_at_create();
730 break;
731 case os::compiler_thread:
732 if (CompilerThreadStackSize > 0) {
733 stack_size = (size_t)(CompilerThreadStackSize * K);
734 break;
735 } // else fall through:
736 // use VMThreadStackSize if CompilerThreadStackSize is not defined
737 case os::vm_thread:
738 case os::pgc_thread:
739 case os::cgc_thread:
740 case os::watcher_thread:
741 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
742 break;
743 }
744 }
745
746 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
747 pthread_attr_setstacksize(&attr, stack_size);
748
749 // glibc guard page
750 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
751
752 ThreadState state;
753
754 {
755 pthread_t tid;
756 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
757
758 char buf[64];
759 if (ret == 0) {
760 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
761 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
762 } else {
763 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
764 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
765 }
766
767 pthread_attr_destroy(&attr);
768
769 if (ret != 0) {
770 // Need to clean up stuff we've allocated so far
771 thread->set_osthread(NULL);
772 delete osthread;
773 return false;
774 }
775
776 // Store pthread info into the OSThread
777 osthread->set_pthread_id(tid);
778
779 // Wait until child thread is either initialized or aborted
780 {
781 Monitor* sync_with_child = osthread->startThread_lock();
782 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
783 while ((state = osthread->get_state()) == ALLOCATED) {
784 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
785 }
786 }
787 }
788
789 // Aborted due to thread limit being reached
790 if (state == ZOMBIE) {
791 thread->set_osthread(NULL);
792 delete osthread;
793 return false;
794 }
795
796 // The thread is returned suspended (in state INITIALIZED),
797 // and is started higher up in the call chain
798 assert(state == INITIALIZED, "race condition");
799 return true;
800 }
801
802 /////////////////////////////////////////////////////////////////////////////
803 // attach existing thread
804
805 // bootstrap the main thread
806 bool os::create_main_thread(JavaThread* thread) {
807 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
808 return create_attached_thread(thread);
809 }
810
811 bool os::create_attached_thread(JavaThread* thread) {
812 #ifdef ASSERT
813 thread->verify_not_published();
814 #endif
815
816 // Allocate the OSThread object
817 OSThread* osthread = new OSThread(NULL, NULL);
818
819 if (osthread == NULL) {
820 return false;
821 }
822
823 // Store pthread info into the OSThread
824 osthread->set_thread_id(os::Linux::gettid());
825 osthread->set_pthread_id(::pthread_self());
826
827 // initialize floating point control register
828 os::Linux::init_thread_fpu_state();
829
830 // Initial thread state is RUNNABLE
831 osthread->set_state(RUNNABLE);
832
833 thread->set_osthread(osthread);
834
835 if (UseNUMA) {
836 int lgrp_id = os::numa_get_group_id();
837 if (lgrp_id != -1) {
838 thread->set_lgrp_id(lgrp_id);
839 }
840 }
841
842 if (os::Linux::is_initial_thread()) {
843 // If current thread is initial thread, its stack is mapped on demand,
844 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
845 // the entire stack region to avoid SEGV in stack banging.
846 // It is also useful to get around the heap-stack-gap problem on SuSE
847 // kernel (see 4821821 for details). We first expand stack to the top
848 // of yellow zone, then enable stack yellow zone (order is significant,
849 // enabling yellow zone first will crash JVM on SuSE Linux), so there
850 // is no gap between the last two virtual memory regions.
851
852 JavaThread *jt = (JavaThread *)thread;
853 address addr = jt->stack_reserved_zone_base();
854 assert(addr != NULL, "initialization problem?");
855 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
856
857 osthread->set_expanding_stack();
858 os::Linux::manually_expand_stack(jt, addr);
859 osthread->clear_expanding_stack();
860 }
861
862 // initialize signal mask for this thread
863 // and save the caller's signal mask
864 os::Linux::hotspot_sigmask(thread);
865
866 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
867 os::current_thread_id(), (uintx) pthread_self());
868
869 return true;
870 }
871
872 void os::pd_start_thread(Thread* thread) {
873 OSThread * osthread = thread->osthread();
874 assert(osthread->get_state() != INITIALIZED, "just checking");
875 Monitor* sync_with_child = osthread->startThread_lock();
876 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
877 sync_with_child->notify();
878 }
879
880 // Free Linux resources related to the OSThread
881 void os::free_thread(OSThread* osthread) {
882 assert(osthread != NULL, "osthread not set");
883
884 if (Thread::current()->osthread() == osthread) {
885 #ifdef ASSERT
886 sigset_t current;
887 sigemptyset(¤t);
888 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
889 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
890 #endif
891
892 // Restore caller's signal mask
893 sigset_t sigmask = osthread->caller_sigmask();
894 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
895 }
896
897 delete osthread;
898 }
899
900 //////////////////////////////////////////////////////////////////////////////
901 // initial thread
902
903 // Check if current thread is the initial thread, similar to Solaris thr_main.
904 bool os::Linux::is_initial_thread(void) {
905 char dummy;
906 // If called before init complete, thread stack bottom will be null.
907 // Can be called if fatal error occurs before initialization.
908 if (initial_thread_stack_bottom() == NULL) return false;
909 assert(initial_thread_stack_bottom() != NULL &&
910 initial_thread_stack_size() != 0,
911 "os::init did not locate initial thread's stack region");
912 if ((address)&dummy >= initial_thread_stack_bottom() &&
913 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
914 return true;
915 } else {
916 return false;
917 }
918 }
919
920 // Find the virtual memory area that contains addr
921 static bool find_vma(address addr, address* vma_low, address* vma_high) {
922 FILE *fp = fopen("/proc/self/maps", "r");
923 if (fp) {
924 address low, high;
925 while (!feof(fp)) {
926 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
927 if (low <= addr && addr < high) {
928 if (vma_low) *vma_low = low;
929 if (vma_high) *vma_high = high;
930 fclose(fp);
931 return true;
932 }
933 }
934 for (;;) {
935 int ch = fgetc(fp);
936 if (ch == EOF || ch == (int)'\n') break;
937 }
938 }
939 fclose(fp);
940 }
941 return false;
942 }
943
944 // Locate initial thread stack. This special handling of initial thread stack
945 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
946 // bogus value for initial thread.
947 void os::Linux::capture_initial_stack(size_t max_size) {
948 // stack size is the easy part, get it from RLIMIT_STACK
949 size_t stack_size;
950 struct rlimit rlim;
951 getrlimit(RLIMIT_STACK, &rlim);
952 stack_size = rlim.rlim_cur;
953
954 // 6308388: a bug in ld.so will relocate its own .data section to the
955 // lower end of primordial stack; reduce ulimit -s value a little bit
956 // so we won't install guard page on ld.so's data section.
957 stack_size -= 2 * page_size();
958
959 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
960 // 7.1, in both cases we will get 2G in return value.
961 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
962 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
963 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
964 // in case other parts in glibc still assumes 2M max stack size.
965 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
966 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
967 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
968 stack_size = 2 * K * K IA64_ONLY(*2);
969 }
970 // Try to figure out where the stack base (top) is. This is harder.
971 //
972 // When an application is started, glibc saves the initial stack pointer in
973 // a global variable "__libc_stack_end", which is then used by system
974 // libraries. __libc_stack_end should be pretty close to stack top. The
975 // variable is available since the very early days. However, because it is
976 // a private interface, it could disappear in the future.
977 //
978 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
979 // to __libc_stack_end, it is very close to stack top, but isn't the real
980 // stack top. Note that /proc may not exist if VM is running as a chroot
981 // program, so reading /proc/<pid>/stat could fail. Also the contents of
982 // /proc/<pid>/stat could change in the future (though unlikely).
983 //
984 // We try __libc_stack_end first. If that doesn't work, look for
985 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
986 // as a hint, which should work well in most cases.
987
988 uintptr_t stack_start;
989
990 // try __libc_stack_end first
991 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
992 if (p && *p) {
993 stack_start = *p;
994 } else {
995 // see if we can get the start_stack field from /proc/self/stat
996 FILE *fp;
997 int pid;
998 char state;
999 int ppid;
1000 int pgrp;
1001 int session;
1002 int nr;
1003 int tpgrp;
1004 unsigned long flags;
1005 unsigned long minflt;
1006 unsigned long cminflt;
1007 unsigned long majflt;
1008 unsigned long cmajflt;
1009 unsigned long utime;
1010 unsigned long stime;
1011 long cutime;
1012 long cstime;
1013 long prio;
1014 long nice;
1015 long junk;
1016 long it_real;
1017 uintptr_t start;
1018 uintptr_t vsize;
1019 intptr_t rss;
1020 uintptr_t rsslim;
1021 uintptr_t scodes;
1022 uintptr_t ecode;
1023 int i;
1024
1025 // Figure what the primordial thread stack base is. Code is inspired
1026 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1027 // followed by command name surrounded by parentheses, state, etc.
1028 char stat[2048];
1029 int statlen;
1030
1031 fp = fopen("/proc/self/stat", "r");
1032 if (fp) {
1033 statlen = fread(stat, 1, 2047, fp);
1034 stat[statlen] = '\0';
1035 fclose(fp);
1036
1037 // Skip pid and the command string. Note that we could be dealing with
1038 // weird command names, e.g. user could decide to rename java launcher
1039 // to "java 1.4.2 :)", then the stat file would look like
1040 // 1234 (java 1.4.2 :)) R ... ...
1041 // We don't really need to know the command string, just find the last
1042 // occurrence of ")" and then start parsing from there. See bug 4726580.
1043 char * s = strrchr(stat, ')');
1044
1045 i = 0;
1046 if (s) {
1047 // Skip blank chars
1048 do { s++; } while (s && isspace(*s));
1049
1050 #define _UFM UINTX_FORMAT
1051 #define _DFM INTX_FORMAT
1052
1053 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1054 // 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
1055 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,
1056 &state, // 3 %c
1057 &ppid, // 4 %d
1058 &pgrp, // 5 %d
1059 &session, // 6 %d
1060 &nr, // 7 %d
1061 &tpgrp, // 8 %d
1062 &flags, // 9 %lu
1063 &minflt, // 10 %lu
1064 &cminflt, // 11 %lu
1065 &majflt, // 12 %lu
1066 &cmajflt, // 13 %lu
1067 &utime, // 14 %lu
1068 &stime, // 15 %lu
1069 &cutime, // 16 %ld
1070 &cstime, // 17 %ld
1071 &prio, // 18 %ld
1072 &nice, // 19 %ld
1073 &junk, // 20 %ld
1074 &it_real, // 21 %ld
1075 &start, // 22 UINTX_FORMAT
1076 &vsize, // 23 UINTX_FORMAT
1077 &rss, // 24 INTX_FORMAT
1078 &rsslim, // 25 UINTX_FORMAT
1079 &scodes, // 26 UINTX_FORMAT
1080 &ecode, // 27 UINTX_FORMAT
1081 &stack_start); // 28 UINTX_FORMAT
1082 }
1083
1084 #undef _UFM
1085 #undef _DFM
1086
1087 if (i != 28 - 2) {
1088 assert(false, "Bad conversion from /proc/self/stat");
1089 // product mode - assume we are the initial thread, good luck in the
1090 // embedded case.
1091 warning("Can't detect initial thread stack location - bad conversion");
1092 stack_start = (uintptr_t) &rlim;
1093 }
1094 } else {
1095 // For some reason we can't open /proc/self/stat (for example, running on
1096 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1097 // most cases, so don't abort:
1098 warning("Can't detect initial thread stack location - no /proc/self/stat");
1099 stack_start = (uintptr_t) &rlim;
1100 }
1101 }
1102
1103 // Now we have a pointer (stack_start) very close to the stack top, the
1104 // next thing to do is to figure out the exact location of stack top. We
1105 // can find out the virtual memory area that contains stack_start by
1106 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1107 // and its upper limit is the real stack top. (again, this would fail if
1108 // running inside chroot, because /proc may not exist.)
1109
1110 uintptr_t stack_top;
1111 address low, high;
1112 if (find_vma((address)stack_start, &low, &high)) {
1113 // success, "high" is the true stack top. (ignore "low", because initial
1114 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1115 stack_top = (uintptr_t)high;
1116 } else {
1117 // failed, likely because /proc/self/maps does not exist
1118 warning("Can't detect initial thread stack location - find_vma failed");
1119 // best effort: stack_start is normally within a few pages below the real
1120 // stack top, use it as stack top, and reduce stack size so we won't put
1121 // guard page outside stack.
1122 stack_top = stack_start;
1123 stack_size -= 16 * page_size();
1124 }
1125
1126 // stack_top could be partially down the page so align it
1127 stack_top = align_size_up(stack_top, page_size());
1128
1129 if (max_size && stack_size > max_size) {
1130 _initial_thread_stack_size = max_size;
1131 } else {
1132 _initial_thread_stack_size = stack_size;
1133 }
1134
1135 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1136 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1137 }
1138
1139 ////////////////////////////////////////////////////////////////////////////////
1140 // time support
1141
1142 // Time since start-up in seconds to a fine granularity.
1143 // Used by VMSelfDestructTimer and the MemProfiler.
1144 double os::elapsedTime() {
1145
1146 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1147 }
1148
1149 jlong os::elapsed_counter() {
1150 return javaTimeNanos() - initial_time_count;
1151 }
1152
1153 jlong os::elapsed_frequency() {
1154 return NANOSECS_PER_SEC; // nanosecond resolution
1155 }
1156
1157 bool os::supports_vtime() { return true; }
1158 bool os::enable_vtime() { return false; }
1159 bool os::vtime_enabled() { return false; }
1160
1161 double os::elapsedVTime() {
1162 struct rusage usage;
1163 int retval = getrusage(RUSAGE_THREAD, &usage);
1164 if (retval == 0) {
1165 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);
1166 } else {
1167 // better than nothing, but not much
1168 return elapsedTime();
1169 }
1170 }
1171
1172 jlong os::javaTimeMillis() {
1173 timeval time;
1174 int status = gettimeofday(&time, NULL);
1175 assert(status != -1, "linux error");
1176 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1177 }
1178
1179 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1180 timeval time;
1181 int status = gettimeofday(&time, NULL);
1182 assert(status != -1, "linux error");
1183 seconds = jlong(time.tv_sec);
1184 nanos = jlong(time.tv_usec) * 1000;
1185 }
1186
1187
1188 #ifndef CLOCK_MONOTONIC
1189 #define CLOCK_MONOTONIC (1)
1190 #endif
1191
1192 void os::Linux::clock_init() {
1193 // we do dlopen's in this particular order due to bug in linux
1194 // dynamical loader (see 6348968) leading to crash on exit
1195 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1196 if (handle == NULL) {
1197 handle = dlopen("librt.so", RTLD_LAZY);
1198 }
1199
1200 if (handle) {
1201 int (*clock_getres_func)(clockid_t, struct timespec*) =
1202 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1203 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1204 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1205 if (clock_getres_func && clock_gettime_func) {
1206 // See if monotonic clock is supported by the kernel. Note that some
1207 // early implementations simply return kernel jiffies (updated every
1208 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1209 // for nano time (though the monotonic property is still nice to have).
1210 // It's fixed in newer kernels, however clock_getres() still returns
1211 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1212 // resolution for now. Hopefully as people move to new kernels, this
1213 // won't be a problem.
1214 struct timespec res;
1215 struct timespec tp;
1216 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1217 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1218 // yes, monotonic clock is supported
1219 _clock_gettime = clock_gettime_func;
1220 return;
1221 } else {
1222 // close librt if there is no monotonic clock
1223 dlclose(handle);
1224 }
1225 }
1226 }
1227 warning("No monotonic clock was available - timed services may " \
1228 "be adversely affected if the time-of-day clock changes");
1229 }
1230
1231 #ifndef SYS_clock_getres
1232 #if defined(IA32) || defined(AMD64)
1233 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1234 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1235 #else
1236 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1237 #define sys_clock_getres(x,y) -1
1238 #endif
1239 #else
1240 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1241 #endif
1242
1243 void os::Linux::fast_thread_clock_init() {
1244 if (!UseLinuxPosixThreadCPUClocks) {
1245 return;
1246 }
1247 clockid_t clockid;
1248 struct timespec tp;
1249 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1250 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1251
1252 // Switch to using fast clocks for thread cpu time if
1253 // the sys_clock_getres() returns 0 error code.
1254 // Note, that some kernels may support the current thread
1255 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1256 // returned by the pthread_getcpuclockid().
1257 // If the fast Posix clocks are supported then the sys_clock_getres()
1258 // must return at least tp.tv_sec == 0 which means a resolution
1259 // better than 1 sec. This is extra check for reliability.
1260
1261 if (pthread_getcpuclockid_func &&
1262 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1263 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1264 _supports_fast_thread_cpu_time = true;
1265 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1266 }
1267 }
1268
1269 jlong os::javaTimeNanos() {
1270 if (os::supports_monotonic_clock()) {
1271 struct timespec tp;
1272 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1273 assert(status == 0, "gettime error");
1274 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1275 return result;
1276 } else {
1277 timeval time;
1278 int status = gettimeofday(&time, NULL);
1279 assert(status != -1, "linux error");
1280 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1281 return 1000 * usecs;
1282 }
1283 }
1284
1285 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1286 if (os::supports_monotonic_clock()) {
1287 info_ptr->max_value = ALL_64_BITS;
1288
1289 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1290 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1291 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1292 } else {
1293 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1294 info_ptr->max_value = ALL_64_BITS;
1295
1296 // gettimeofday is a real time clock so it skips
1297 info_ptr->may_skip_backward = true;
1298 info_ptr->may_skip_forward = true;
1299 }
1300
1301 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1302 }
1303
1304 // Return the real, user, and system times in seconds from an
1305 // arbitrary fixed point in the past.
1306 bool os::getTimesSecs(double* process_real_time,
1307 double* process_user_time,
1308 double* process_system_time) {
1309 struct tms ticks;
1310 clock_t real_ticks = times(&ticks);
1311
1312 if (real_ticks == (clock_t) (-1)) {
1313 return false;
1314 } else {
1315 double ticks_per_second = (double) clock_tics_per_sec;
1316 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1317 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1318 *process_real_time = ((double) real_ticks) / ticks_per_second;
1319
1320 return true;
1321 }
1322 }
1323
1324
1325 char * os::local_time_string(char *buf, size_t buflen) {
1326 struct tm t;
1327 time_t long_time;
1328 time(&long_time);
1329 localtime_r(&long_time, &t);
1330 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1331 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1332 t.tm_hour, t.tm_min, t.tm_sec);
1333 return buf;
1334 }
1335
1336 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1337 return localtime_r(clock, res);
1338 }
1339
1340 ////////////////////////////////////////////////////////////////////////////////
1341 // runtime exit support
1342
1343 // Note: os::shutdown() might be called very early during initialization, or
1344 // called from signal handler. Before adding something to os::shutdown(), make
1345 // sure it is async-safe and can handle partially initialized VM.
1346 void os::shutdown() {
1347
1348 // allow PerfMemory to attempt cleanup of any persistent resources
1349 perfMemory_exit();
1350
1351 // needs to remove object in file system
1352 AttachListener::abort();
1353
1354 // flush buffered output, finish log files
1355 ostream_abort();
1356
1357 // Check for abort hook
1358 abort_hook_t abort_hook = Arguments::abort_hook();
1359 if (abort_hook != NULL) {
1360 abort_hook();
1361 }
1362
1363 }
1364
1365 // Note: os::abort() might be called very early during initialization, or
1366 // called from signal handler. Before adding something to os::abort(), make
1367 // sure it is async-safe and can handle partially initialized VM.
1368 void os::abort(bool dump_core, void* siginfo, const void* context) {
1369 os::shutdown();
1370 if (dump_core) {
1371 #ifndef PRODUCT
1372 fdStream out(defaultStream::output_fd());
1373 out.print_raw("Current thread is ");
1374 char buf[16];
1375 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1376 out.print_raw_cr(buf);
1377 out.print_raw_cr("Dumping core ...");
1378 #endif
1379 ::abort(); // dump core
1380 }
1381
1382 ::exit(1);
1383 }
1384
1385 // Die immediately, no exit hook, no abort hook, no cleanup.
1386 void os::die() {
1387 ::abort();
1388 }
1389
1390
1391 // This method is a copy of JDK's sysGetLastErrorString
1392 // from src/solaris/hpi/src/system_md.c
1393
1394 size_t os::lasterror(char *buf, size_t len) {
1395 if (errno == 0) return 0;
1396
1397 const char *s = os::strerror(errno);
1398 size_t n = ::strlen(s);
1399 if (n >= len) {
1400 n = len - 1;
1401 }
1402 ::strncpy(buf, s, n);
1403 buf[n] = '\0';
1404 return n;
1405 }
1406
1407 // thread_id is kernel thread id (similar to Solaris LWP id)
1408 intx os::current_thread_id() { return os::Linux::gettid(); }
1409 int os::current_process_id() {
1410 return ::getpid();
1411 }
1412
1413 // DLL functions
1414
1415 const char* os::dll_file_extension() { return ".so"; }
1416
1417 // This must be hard coded because it's the system's temporary
1418 // directory not the java application's temp directory, ala java.io.tmpdir.
1419 const char* os::get_temp_directory() { return "/tmp"; }
1420
1421 static bool file_exists(const char* filename) {
1422 struct stat statbuf;
1423 if (filename == NULL || strlen(filename) == 0) {
1424 return false;
1425 }
1426 return os::stat(filename, &statbuf) == 0;
1427 }
1428
1429 bool os::dll_build_name(char* buffer, size_t buflen,
1430 const char* pname, const char* fname) {
1431 bool retval = false;
1432 // Copied from libhpi
1433 const size_t pnamelen = pname ? strlen(pname) : 0;
1434
1435 // Return error on buffer overflow.
1436 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1437 return retval;
1438 }
1439
1440 if (pnamelen == 0) {
1441 snprintf(buffer, buflen, "lib%s.so", fname);
1442 retval = true;
1443 } else if (strchr(pname, *os::path_separator()) != NULL) {
1444 int n;
1445 char** pelements = split_path(pname, &n);
1446 if (pelements == NULL) {
1447 return false;
1448 }
1449 for (int i = 0; i < n; i++) {
1450 // Really shouldn't be NULL, but check can't hurt
1451 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1452 continue; // skip the empty path values
1453 }
1454 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1455 if (file_exists(buffer)) {
1456 retval = true;
1457 break;
1458 }
1459 }
1460 // release the storage
1461 for (int i = 0; i < n; i++) {
1462 if (pelements[i] != NULL) {
1463 FREE_C_HEAP_ARRAY(char, pelements[i]);
1464 }
1465 }
1466 if (pelements != NULL) {
1467 FREE_C_HEAP_ARRAY(char*, pelements);
1468 }
1469 } else {
1470 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1471 retval = true;
1472 }
1473 return retval;
1474 }
1475
1476 // check if addr is inside libjvm.so
1477 bool os::address_is_in_vm(address addr) {
1478 static address libjvm_base_addr;
1479 Dl_info dlinfo;
1480
1481 if (libjvm_base_addr == NULL) {
1482 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1483 libjvm_base_addr = (address)dlinfo.dli_fbase;
1484 }
1485 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1486 }
1487
1488 if (dladdr((void *)addr, &dlinfo) != 0) {
1489 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1490 }
1491
1492 return false;
1493 }
1494
1495 bool os::dll_address_to_function_name(address addr, char *buf,
1496 int buflen, int *offset,
1497 bool demangle) {
1498 // buf is not optional, but offset is optional
1499 assert(buf != NULL, "sanity check");
1500
1501 Dl_info dlinfo;
1502
1503 if (dladdr((void*)addr, &dlinfo) != 0) {
1504 // see if we have a matching symbol
1505 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1506 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1507 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1508 }
1509 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1510 return true;
1511 }
1512 // no matching symbol so try for just file info
1513 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1514 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1515 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1516 return true;
1517 }
1518 }
1519 }
1520
1521 buf[0] = '\0';
1522 if (offset != NULL) *offset = -1;
1523 return false;
1524 }
1525
1526 struct _address_to_library_name {
1527 address addr; // input : memory address
1528 size_t buflen; // size of fname
1529 char* fname; // output: library name
1530 address base; // library base addr
1531 };
1532
1533 static int address_to_library_name_callback(struct dl_phdr_info *info,
1534 size_t size, void *data) {
1535 int i;
1536 bool found = false;
1537 address libbase = NULL;
1538 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1539
1540 // iterate through all loadable segments
1541 for (i = 0; i < info->dlpi_phnum; i++) {
1542 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1543 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1544 // base address of a library is the lowest address of its loaded
1545 // segments.
1546 if (libbase == NULL || libbase > segbase) {
1547 libbase = segbase;
1548 }
1549 // see if 'addr' is within current segment
1550 if (segbase <= d->addr &&
1551 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1552 found = true;
1553 }
1554 }
1555 }
1556
1557 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1558 // so dll_address_to_library_name() can fall through to use dladdr() which
1559 // can figure out executable name from argv[0].
1560 if (found && info->dlpi_name && info->dlpi_name[0]) {
1561 d->base = libbase;
1562 if (d->fname) {
1563 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1564 }
1565 return 1;
1566 }
1567 return 0;
1568 }
1569
1570 bool os::dll_address_to_library_name(address addr, char* buf,
1571 int buflen, int* offset) {
1572 // buf is not optional, but offset is optional
1573 assert(buf != NULL, "sanity check");
1574
1575 Dl_info dlinfo;
1576 struct _address_to_library_name data;
1577
1578 // There is a bug in old glibc dladdr() implementation that it could resolve
1579 // to wrong library name if the .so file has a base address != NULL. Here
1580 // we iterate through the program headers of all loaded libraries to find
1581 // out which library 'addr' really belongs to. This workaround can be
1582 // removed once the minimum requirement for glibc is moved to 2.3.x.
1583 data.addr = addr;
1584 data.fname = buf;
1585 data.buflen = buflen;
1586 data.base = NULL;
1587 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1588
1589 if (rslt) {
1590 // buf already contains library name
1591 if (offset) *offset = addr - data.base;
1592 return true;
1593 }
1594 if (dladdr((void*)addr, &dlinfo) != 0) {
1595 if (dlinfo.dli_fname != NULL) {
1596 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1597 }
1598 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1599 *offset = addr - (address)dlinfo.dli_fbase;
1600 }
1601 return true;
1602 }
1603
1604 buf[0] = '\0';
1605 if (offset) *offset = -1;
1606 return false;
1607 }
1608
1609 // Loads .dll/.so and
1610 // in case of error it checks if .dll/.so was built for the
1611 // same architecture as Hotspot is running on
1612
1613
1614 // Remember the stack's state. The Linux dynamic linker will change
1615 // the stack to 'executable' at most once, so we must safepoint only once.
1616 bool os::Linux::_stack_is_executable = false;
1617
1618 // VM operation that loads a library. This is necessary if stack protection
1619 // of the Java stacks can be lost during loading the library. If we
1620 // do not stop the Java threads, they can stack overflow before the stacks
1621 // are protected again.
1622 class VM_LinuxDllLoad: public VM_Operation {
1623 private:
1624 const char *_filename;
1625 char *_ebuf;
1626 int _ebuflen;
1627 void *_lib;
1628 public:
1629 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1630 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1631 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1632 void doit() {
1633 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1634 os::Linux::_stack_is_executable = true;
1635 }
1636 void* loaded_library() { return _lib; }
1637 };
1638
1639 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1640 void * result = NULL;
1641 bool load_attempted = false;
1642
1643 // Check whether the library to load might change execution rights
1644 // of the stack. If they are changed, the protection of the stack
1645 // guard pages will be lost. We need a safepoint to fix this.
1646 //
1647 // See Linux man page execstack(8) for more info.
1648 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1649 ElfFile ef(filename);
1650 if (!ef.specifies_noexecstack()) {
1651 if (!is_init_completed()) {
1652 os::Linux::_stack_is_executable = true;
1653 // This is OK - No Java threads have been created yet, and hence no
1654 // stack guard pages to fix.
1655 //
1656 // This should happen only when you are building JDK7 using a very
1657 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1658 //
1659 // Dynamic loader will make all stacks executable after
1660 // this function returns, and will not do that again.
1661 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1662 } else {
1663 warning("You have loaded library %s which might have disabled stack guard. "
1664 "The VM will try to fix the stack guard now.\n"
1665 "It's highly recommended that you fix the library with "
1666 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1667 filename);
1668
1669 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1670 JavaThread *jt = JavaThread::current();
1671 if (jt->thread_state() != _thread_in_native) {
1672 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1673 // that requires ExecStack. Cannot enter safe point. Let's give up.
1674 warning("Unable to fix stack guard. Giving up.");
1675 } else {
1676 if (!LoadExecStackDllInVMThread) {
1677 // This is for the case where the DLL has an static
1678 // constructor function that executes JNI code. We cannot
1679 // load such DLLs in the VMThread.
1680 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1681 }
1682
1683 ThreadInVMfromNative tiv(jt);
1684 debug_only(VMNativeEntryWrapper vew;)
1685
1686 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1687 VMThread::execute(&op);
1688 if (LoadExecStackDllInVMThread) {
1689 result = op.loaded_library();
1690 }
1691 load_attempted = true;
1692 }
1693 }
1694 }
1695 }
1696
1697 if (!load_attempted) {
1698 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1699 }
1700
1701 if (result != NULL) {
1702 // Successful loading
1703 return result;
1704 }
1705
1706 Elf32_Ehdr elf_head;
1707 int diag_msg_max_length=ebuflen-strlen(ebuf);
1708 char* diag_msg_buf=ebuf+strlen(ebuf);
1709
1710 if (diag_msg_max_length==0) {
1711 // No more space in ebuf for additional diagnostics message
1712 return NULL;
1713 }
1714
1715
1716 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1717
1718 if (file_descriptor < 0) {
1719 // Can't open library, report dlerror() message
1720 return NULL;
1721 }
1722
1723 bool failed_to_read_elf_head=
1724 (sizeof(elf_head)!=
1725 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1726
1727 ::close(file_descriptor);
1728 if (failed_to_read_elf_head) {
1729 // file i/o error - report dlerror() msg
1730 return NULL;
1731 }
1732
1733 typedef struct {
1734 Elf32_Half code; // Actual value as defined in elf.h
1735 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1736 char elf_class; // 32 or 64 bit
1737 char endianess; // MSB or LSB
1738 char* name; // String representation
1739 } arch_t;
1740
1741 #ifndef EM_486
1742 #define EM_486 6 /* Intel 80486 */
1743 #endif
1744 #ifndef EM_AARCH64
1745 #define EM_AARCH64 183 /* ARM AARCH64 */
1746 #endif
1747
1748 static const arch_t arch_array[]={
1749 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1750 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1751 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1752 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1753 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1754 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1755 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1756 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1757 #if defined(VM_LITTLE_ENDIAN)
1758 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1759 #else
1760 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1761 #endif
1762 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1763 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1764 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1765 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1766 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1767 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1768 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1769 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1770 };
1771
1772 #if (defined IA32)
1773 static Elf32_Half running_arch_code=EM_386;
1774 #elif (defined AMD64)
1775 static Elf32_Half running_arch_code=EM_X86_64;
1776 #elif (defined IA64)
1777 static Elf32_Half running_arch_code=EM_IA_64;
1778 #elif (defined __sparc) && (defined _LP64)
1779 static Elf32_Half running_arch_code=EM_SPARCV9;
1780 #elif (defined __sparc) && (!defined _LP64)
1781 static Elf32_Half running_arch_code=EM_SPARC;
1782 #elif (defined __powerpc64__)
1783 static Elf32_Half running_arch_code=EM_PPC64;
1784 #elif (defined __powerpc__)
1785 static Elf32_Half running_arch_code=EM_PPC;
1786 #elif (defined ARM)
1787 static Elf32_Half running_arch_code=EM_ARM;
1788 #elif (defined S390)
1789 static Elf32_Half running_arch_code=EM_S390;
1790 #elif (defined ALPHA)
1791 static Elf32_Half running_arch_code=EM_ALPHA;
1792 #elif (defined MIPSEL)
1793 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1794 #elif (defined PARISC)
1795 static Elf32_Half running_arch_code=EM_PARISC;
1796 #elif (defined MIPS)
1797 static Elf32_Half running_arch_code=EM_MIPS;
1798 #elif (defined M68K)
1799 static Elf32_Half running_arch_code=EM_68K;
1800 #elif (defined AARCH64)
1801 static Elf32_Half running_arch_code=EM_AARCH64;
1802 #else
1803 #error Method os::dll_load requires that one of following is defined:\
1804 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1805 #endif
1806
1807 // Identify compatability class for VM's architecture and library's architecture
1808 // Obtain string descriptions for architectures
1809
1810 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1811 int running_arch_index=-1;
1812
1813 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1814 if (running_arch_code == arch_array[i].code) {
1815 running_arch_index = i;
1816 }
1817 if (lib_arch.code == arch_array[i].code) {
1818 lib_arch.compat_class = arch_array[i].compat_class;
1819 lib_arch.name = arch_array[i].name;
1820 }
1821 }
1822
1823 assert(running_arch_index != -1,
1824 "Didn't find running architecture code (running_arch_code) in arch_array");
1825 if (running_arch_index == -1) {
1826 // Even though running architecture detection failed
1827 // we may still continue with reporting dlerror() message
1828 return NULL;
1829 }
1830
1831 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1832 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1833 return NULL;
1834 }
1835
1836 #ifndef S390
1837 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1839 return NULL;
1840 }
1841 #endif // !S390
1842
1843 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1844 if (lib_arch.name!=NULL) {
1845 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1846 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1847 lib_arch.name, arch_array[running_arch_index].name);
1848 } else {
1849 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1850 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1851 lib_arch.code,
1852 arch_array[running_arch_index].name);
1853 }
1854 }
1855
1856 return NULL;
1857 }
1858
1859 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1860 int ebuflen) {
1861 void * result = ::dlopen(filename, RTLD_LAZY);
1862 if (result == NULL) {
1863 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1864 ebuf[ebuflen-1] = '\0';
1865 }
1866 return result;
1867 }
1868
1869 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1870 int ebuflen) {
1871 void * result = NULL;
1872 if (LoadExecStackDllInVMThread) {
1873 result = dlopen_helper(filename, ebuf, ebuflen);
1874 }
1875
1876 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1877 // library that requires an executable stack, or which does not have this
1878 // stack attribute set, dlopen changes the stack attribute to executable. The
1879 // read protection of the guard pages gets lost.
1880 //
1881 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1882 // may have been queued at the same time.
1883
1884 if (!_stack_is_executable) {
1885 JavaThread *jt = Threads::first();
1886
1887 while (jt) {
1888 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1889 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1890 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1891 warning("Attempt to reguard stack yellow zone failed.");
1892 }
1893 }
1894 jt = jt->next();
1895 }
1896 }
1897
1898 return result;
1899 }
1900
1901 void* os::dll_lookup(void* handle, const char* name) {
1902 void* res = dlsym(handle, name);
1903 return res;
1904 }
1905
1906 void* os::get_default_process_handle() {
1907 return (void*)::dlopen(NULL, RTLD_LAZY);
1908 }
1909
1910 static bool _print_ascii_file(const char* filename, outputStream* st) {
1911 int fd = ::open(filename, O_RDONLY);
1912 if (fd == -1) {
1913 return false;
1914 }
1915
1916 char buf[32];
1917 int bytes;
1918 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1919 st->print_raw(buf, bytes);
1920 }
1921
1922 ::close(fd);
1923
1924 return true;
1925 }
1926
1927 void os::print_dll_info(outputStream *st) {
1928 st->print_cr("Dynamic libraries:");
1929
1930 char fname[32];
1931 pid_t pid = os::Linux::gettid();
1932
1933 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1934
1935 if (!_print_ascii_file(fname, st)) {
1936 st->print("Can not get library information for pid = %d\n", pid);
1937 }
1938 }
1939
1940 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1941 FILE *procmapsFile = NULL;
1942
1943 // Open the procfs maps file for the current process
1944 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1945 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1946 char line[PATH_MAX + 100];
1947
1948 // Read line by line from 'file'
1949 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1950 u8 base, top, offset, inode;
1951 char permissions[5];
1952 char device[6];
1953 char name[PATH_MAX + 1];
1954
1955 // Parse fields from line
1956 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1957 &base, &top, permissions, &offset, device, &inode, name);
1958
1959 // Filter by device id '00:00' so that we only get file system mapped files.
1960 if (strcmp(device, "00:00") != 0) {
1961
1962 // Call callback with the fields of interest
1963 if(callback(name, (address)base, (address)top, param)) {
1964 // Oops abort, callback aborted
1965 fclose(procmapsFile);
1966 return 1;
1967 }
1968 }
1969 }
1970 fclose(procmapsFile);
1971 }
1972 return 0;
1973 }
1974
1975 void os::print_os_info_brief(outputStream* st) {
1976 os::Linux::print_distro_info(st);
1977
1978 os::Posix::print_uname_info(st);
1979
1980 os::Linux::print_libversion_info(st);
1981
1982 }
1983
1984 void os::print_os_info(outputStream* st) {
1985 st->print("OS:");
1986
1987 os::Linux::print_distro_info(st);
1988
1989 os::Posix::print_uname_info(st);
1990
1991 // Print warning if unsafe chroot environment detected
1992 if (unsafe_chroot_detected) {
1993 st->print("WARNING!! ");
1994 st->print_cr("%s", unstable_chroot_error);
1995 }
1996
1997 os::Linux::print_libversion_info(st);
1998
1999 os::Posix::print_rlimit_info(st);
2000
2001 os::Posix::print_load_average(st);
2002
2003 os::Linux::print_full_memory_info(st);
2004 }
2005
2006 // Try to identify popular distros.
2007 // Most Linux distributions have a /etc/XXX-release file, which contains
2008 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2009 // file that also contains the OS version string. Some have more than one
2010 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2011 // /etc/redhat-release.), so the order is important.
2012 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2013 // their own specific XXX-release file as well as a redhat-release file.
2014 // Because of this the XXX-release file needs to be searched for before the
2015 // redhat-release file.
2016 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2017 // search for redhat-release / SuSE-release needs to be before lsb-release.
2018 // Since the lsb-release file is the new standard it needs to be searched
2019 // before the older style release files.
2020 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2021 // next to last resort. The os-release file is a new standard that contains
2022 // distribution information and the system-release file seems to be an old
2023 // standard that has been replaced by the lsb-release and os-release files.
2024 // Searching for the debian_version file is the last resort. It contains
2025 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2026 // "Debian " is printed before the contents of the debian_version file.
2027
2028 const char* distro_files[] = {
2029 "/etc/oracle-release",
2030 "/etc/mandriva-release",
2031 "/etc/mandrake-release",
2032 "/etc/sun-release",
2033 "/etc/redhat-release",
2034 "/etc/SuSE-release",
2035 "/etc/lsb-release",
2036 "/etc/turbolinux-release",
2037 "/etc/gentoo-release",
2038 "/etc/ltib-release",
2039 "/etc/angstrom-version",
2040 "/etc/system-release",
2041 "/etc/os-release",
2042 NULL };
2043
2044 void os::Linux::print_distro_info(outputStream* st) {
2045 for (int i = 0;; i++) {
2046 const char* file = distro_files[i];
2047 if (file == NULL) {
2048 break; // done
2049 }
2050 // If file prints, we found it.
2051 if (_print_ascii_file(file, st)) {
2052 return;
2053 }
2054 }
2055
2056 if (file_exists("/etc/debian_version")) {
2057 st->print("Debian ");
2058 _print_ascii_file("/etc/debian_version", st);
2059 } else {
2060 st->print("Linux");
2061 }
2062 st->cr();
2063 }
2064
2065 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2066 char buf[256];
2067 while (fgets(buf, sizeof(buf), fp)) {
2068 // Edit out extra stuff in expected format
2069 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2070 char* ptr = strstr(buf, "\""); // the name is in quotes
2071 if (ptr != NULL) {
2072 ptr++; // go beyond first quote
2073 char* nl = strchr(ptr, '\"');
2074 if (nl != NULL) *nl = '\0';
2075 strncpy(distro, ptr, length);
2076 } else {
2077 ptr = strstr(buf, "=");
2078 ptr++; // go beyond equals then
2079 char* nl = strchr(ptr, '\n');
2080 if (nl != NULL) *nl = '\0';
2081 strncpy(distro, ptr, length);
2082 }
2083 return;
2084 } else if (get_first_line) {
2085 char* nl = strchr(buf, '\n');
2086 if (nl != NULL) *nl = '\0';
2087 strncpy(distro, buf, length);
2088 return;
2089 }
2090 }
2091 // print last line and close
2092 char* nl = strchr(buf, '\n');
2093 if (nl != NULL) *nl = '\0';
2094 strncpy(distro, buf, length);
2095 }
2096
2097 static void parse_os_info(char* distro, size_t length, const char* file) {
2098 FILE* fp = fopen(file, "r");
2099 if (fp != NULL) {
2100 // if suse format, print out first line
2101 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2102 parse_os_info_helper(fp, distro, length, get_first_line);
2103 fclose(fp);
2104 }
2105 }
2106
2107 void os::get_summary_os_info(char* buf, size_t buflen) {
2108 for (int i = 0;; i++) {
2109 const char* file = distro_files[i];
2110 if (file == NULL) {
2111 break; // ran out of distro_files
2112 }
2113 if (file_exists(file)) {
2114 parse_os_info(buf, buflen, file);
2115 return;
2116 }
2117 }
2118 // special case for debian
2119 if (file_exists("/etc/debian_version")) {
2120 strncpy(buf, "Debian ", buflen);
2121 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2122 } else {
2123 strncpy(buf, "Linux", buflen);
2124 }
2125 }
2126
2127 void os::Linux::print_libversion_info(outputStream* st) {
2128 // libc, pthread
2129 st->print("libc:");
2130 st->print("%s ", os::Linux::glibc_version());
2131 st->print("%s ", os::Linux::libpthread_version());
2132 st->cr();
2133 }
2134
2135 void os::Linux::print_full_memory_info(outputStream* st) {
2136 st->print("\n/proc/meminfo:\n");
2137 _print_ascii_file("/proc/meminfo", st);
2138 st->cr();
2139 }
2140
2141 void os::print_memory_info(outputStream* st) {
2142
2143 st->print("Memory:");
2144 st->print(" %dk page", os::vm_page_size()>>10);
2145
2146 // values in struct sysinfo are "unsigned long"
2147 struct sysinfo si;
2148 sysinfo(&si);
2149
2150 st->print(", physical " UINT64_FORMAT "k",
2151 os::physical_memory() >> 10);
2152 st->print("(" UINT64_FORMAT "k free)",
2153 os::available_memory() >> 10);
2154 st->print(", swap " UINT64_FORMAT "k",
2155 ((jlong)si.totalswap * si.mem_unit) >> 10);
2156 st->print("(" UINT64_FORMAT "k free)",
2157 ((jlong)si.freeswap * si.mem_unit) >> 10);
2158 st->cr();
2159 }
2160
2161 // Print the first "model name" line and the first "flags" line
2162 // that we find and nothing more. We assume "model name" comes
2163 // before "flags" so if we find a second "model name", then the
2164 // "flags" field is considered missing.
2165 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2166 #if defined(IA32) || defined(AMD64)
2167 // Other platforms have less repetitive cpuinfo files
2168 FILE *fp = fopen("/proc/cpuinfo", "r");
2169 if (fp) {
2170 while (!feof(fp)) {
2171 if (fgets(buf, buflen, fp)) {
2172 // Assume model name comes before flags
2173 bool model_name_printed = false;
2174 if (strstr(buf, "model name") != NULL) {
2175 if (!model_name_printed) {
2176 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2177 st->print_raw(buf);
2178 model_name_printed = true;
2179 } else {
2180 // model name printed but not flags? Odd, just return
2181 fclose(fp);
2182 return true;
2183 }
2184 }
2185 // print the flags line too
2186 if (strstr(buf, "flags") != NULL) {
2187 st->print_raw(buf);
2188 fclose(fp);
2189 return true;
2190 }
2191 }
2192 }
2193 fclose(fp);
2194 }
2195 #endif // x86 platforms
2196 return false;
2197 }
2198
2199 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2200 // Only print the model name if the platform provides this as a summary
2201 if (!print_model_name_and_flags(st, buf, buflen)) {
2202 st->print("\n/proc/cpuinfo:\n");
2203 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2204 st->print_cr(" <Not Available>");
2205 }
2206 }
2207 }
2208
2209 #if defined(AMD64) || defined(IA32) || defined(X32)
2210 const char* search_string = "model name";
2211 #elif defined(SPARC)
2212 const char* search_string = "cpu";
2213 #elif defined(PPC64)
2214 const char* search_string = "cpu";
2215 #else
2216 const char* search_string = "Processor";
2217 #endif
2218
2219 // Parses the cpuinfo file for string representing the model name.
2220 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2221 FILE* fp = fopen("/proc/cpuinfo", "r");
2222 if (fp != NULL) {
2223 while (!feof(fp)) {
2224 char buf[256];
2225 if (fgets(buf, sizeof(buf), fp)) {
2226 char* start = strstr(buf, search_string);
2227 if (start != NULL) {
2228 char *ptr = start + strlen(search_string);
2229 char *end = buf + strlen(buf);
2230 while (ptr != end) {
2231 // skip whitespace and colon for the rest of the name.
2232 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2233 break;
2234 }
2235 ptr++;
2236 }
2237 if (ptr != end) {
2238 // reasonable string, get rid of newline and keep the rest
2239 char* nl = strchr(buf, '\n');
2240 if (nl != NULL) *nl = '\0';
2241 strncpy(cpuinfo, ptr, length);
2242 fclose(fp);
2243 return;
2244 }
2245 }
2246 }
2247 }
2248 fclose(fp);
2249 }
2250 // cpuinfo not found or parsing failed, just print generic string. The entire
2251 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2252 #if defined(AMD64)
2253 strncpy(cpuinfo, "x86_64", length);
2254 #elif defined(IA32)
2255 strncpy(cpuinfo, "x86_32", length);
2256 #elif defined(IA64)
2257 strncpy(cpuinfo, "IA64", length);
2258 #elif defined(SPARC)
2259 strncpy(cpuinfo, "sparcv9", length);
2260 #elif defined(AARCH64)
2261 strncpy(cpuinfo, "AArch64", length);
2262 #elif defined(ARM)
2263 strncpy(cpuinfo, "ARM", length);
2264 #elif defined(PPC)
2265 strncpy(cpuinfo, "PPC64", length);
2266 #elif defined(ZERO_LIBARCH)
2267 strncpy(cpuinfo, ZERO_LIBARCH, length);
2268 #else
2269 strncpy(cpuinfo, "unknown", length);
2270 #endif
2271 }
2272
2273 static void print_signal_handler(outputStream* st, int sig,
2274 char* buf, size_t buflen);
2275
2276 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2277 st->print_cr("Signal Handlers:");
2278 print_signal_handler(st, SIGSEGV, buf, buflen);
2279 print_signal_handler(st, SIGBUS , buf, buflen);
2280 print_signal_handler(st, SIGFPE , buf, buflen);
2281 print_signal_handler(st, SIGPIPE, buf, buflen);
2282 print_signal_handler(st, SIGXFSZ, buf, buflen);
2283 print_signal_handler(st, SIGILL , buf, buflen);
2284 print_signal_handler(st, SR_signum, buf, buflen);
2285 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2286 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2287 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2288 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2289 #if defined(PPC64)
2290 print_signal_handler(st, SIGTRAP, buf, buflen);
2291 #endif
2292 }
2293
2294 static char saved_jvm_path[MAXPATHLEN] = {0};
2295
2296 // Find the full path to the current module, libjvm.so
2297 void os::jvm_path(char *buf, jint buflen) {
2298 // Error checking.
2299 if (buflen < MAXPATHLEN) {
2300 assert(false, "must use a large-enough buffer");
2301 buf[0] = '\0';
2302 return;
2303 }
2304 // Lazy resolve the path to current module.
2305 if (saved_jvm_path[0] != 0) {
2306 strcpy(buf, saved_jvm_path);
2307 return;
2308 }
2309
2310 char dli_fname[MAXPATHLEN];
2311 bool ret = dll_address_to_library_name(
2312 CAST_FROM_FN_PTR(address, os::jvm_path),
2313 dli_fname, sizeof(dli_fname), NULL);
2314 assert(ret, "cannot locate libjvm");
2315 char *rp = NULL;
2316 if (ret && dli_fname[0] != '\0') {
2317 rp = realpath(dli_fname, buf);
2318 }
2319 if (rp == NULL) {
2320 return;
2321 }
2322
2323 if (Arguments::sun_java_launcher_is_altjvm()) {
2324 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2325 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2326 // If "/jre/lib/" appears at the right place in the string, then
2327 // assume we are installed in a JDK and we're done. Otherwise, check
2328 // for a JAVA_HOME environment variable and fix up the path so it
2329 // looks like libjvm.so is installed there (append a fake suffix
2330 // hotspot/libjvm.so).
2331 const char *p = buf + strlen(buf) - 1;
2332 for (int count = 0; p > buf && count < 5; ++count) {
2333 for (--p; p > buf && *p != '/'; --p)
2334 /* empty */ ;
2335 }
2336
2337 if (strncmp(p, "/jre/lib/", 9) != 0) {
2338 // Look for JAVA_HOME in the environment.
2339 char* java_home_var = ::getenv("JAVA_HOME");
2340 if (java_home_var != NULL && java_home_var[0] != 0) {
2341 char* jrelib_p;
2342 int len;
2343
2344 // Check the current module name "libjvm.so".
2345 p = strrchr(buf, '/');
2346 if (p == NULL) {
2347 return;
2348 }
2349 assert(strstr(p, "/libjvm") == p, "invalid library name");
2350
2351 rp = realpath(java_home_var, buf);
2352 if (rp == NULL) {
2353 return;
2354 }
2355
2356 // determine if this is a legacy image or modules image
2357 // modules image doesn't have "jre" subdirectory
2358 len = strlen(buf);
2359 assert(len < buflen, "Ran out of buffer room");
2360 jrelib_p = buf + len;
2361 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2362 if (0 != access(buf, F_OK)) {
2363 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2364 }
2365
2366 if (0 == access(buf, F_OK)) {
2367 // Use current module name "libjvm.so"
2368 len = strlen(buf);
2369 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2370 } else {
2371 // Go back to path of .so
2372 rp = realpath(dli_fname, buf);
2373 if (rp == NULL) {
2374 return;
2375 }
2376 }
2377 }
2378 }
2379 }
2380
2381 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2382 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2383 }
2384
2385 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2386 // no prefix required, not even "_"
2387 }
2388
2389 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2390 // no suffix required
2391 }
2392
2393 ////////////////////////////////////////////////////////////////////////////////
2394 // sun.misc.Signal support
2395
2396 static volatile jint sigint_count = 0;
2397
2398 static void UserHandler(int sig, void *siginfo, void *context) {
2399 // 4511530 - sem_post is serialized and handled by the manager thread. When
2400 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2401 // don't want to flood the manager thread with sem_post requests.
2402 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2403 return;
2404 }
2405
2406 // Ctrl-C is pressed during error reporting, likely because the error
2407 // handler fails to abort. Let VM die immediately.
2408 if (sig == SIGINT && is_error_reported()) {
2409 os::die();
2410 }
2411
2412 os::signal_notify(sig);
2413 }
2414
2415 void* os::user_handler() {
2416 return CAST_FROM_FN_PTR(void*, UserHandler);
2417 }
2418
2419 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2420 struct timespec ts;
2421 // Semaphore's are always associated with CLOCK_REALTIME
2422 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2423 // see unpackTime for discussion on overflow checking
2424 if (sec >= MAX_SECS) {
2425 ts.tv_sec += MAX_SECS;
2426 ts.tv_nsec = 0;
2427 } else {
2428 ts.tv_sec += sec;
2429 ts.tv_nsec += nsec;
2430 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2431 ts.tv_nsec -= NANOSECS_PER_SEC;
2432 ++ts.tv_sec; // note: this must be <= max_secs
2433 }
2434 }
2435
2436 return ts;
2437 }
2438
2439 extern "C" {
2440 typedef void (*sa_handler_t)(int);
2441 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2442 }
2443
2444 void* os::signal(int signal_number, void* handler) {
2445 struct sigaction sigAct, oldSigAct;
2446
2447 sigfillset(&(sigAct.sa_mask));
2448 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2449 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2450
2451 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2452 // -1 means registration failed
2453 return (void *)-1;
2454 }
2455
2456 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2457 }
2458
2459 void os::signal_raise(int signal_number) {
2460 ::raise(signal_number);
2461 }
2462
2463 // The following code is moved from os.cpp for making this
2464 // code platform specific, which it is by its very nature.
2465
2466 // Will be modified when max signal is changed to be dynamic
2467 int os::sigexitnum_pd() {
2468 return NSIG;
2469 }
2470
2471 // a counter for each possible signal value
2472 static volatile jint pending_signals[NSIG+1] = { 0 };
2473
2474 // Linux(POSIX) specific hand shaking semaphore.
2475 static sem_t sig_sem;
2476 static PosixSemaphore sr_semaphore;
2477
2478 void os::signal_init_pd() {
2479 // Initialize signal structures
2480 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2481
2482 // Initialize signal semaphore
2483 ::sem_init(&sig_sem, 0, 0);
2484 }
2485
2486 void os::signal_notify(int sig) {
2487 Atomic::inc(&pending_signals[sig]);
2488 ::sem_post(&sig_sem);
2489 }
2490
2491 static int check_pending_signals(bool wait) {
2492 Atomic::store(0, &sigint_count);
2493 for (;;) {
2494 for (int i = 0; i < NSIG + 1; i++) {
2495 jint n = pending_signals[i];
2496 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2497 return i;
2498 }
2499 }
2500 if (!wait) {
2501 return -1;
2502 }
2503 JavaThread *thread = JavaThread::current();
2504 ThreadBlockInVM tbivm(thread);
2505
2506 bool threadIsSuspended;
2507 do {
2508 thread->set_suspend_equivalent();
2509 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2510 ::sem_wait(&sig_sem);
2511
2512 // were we externally suspended while we were waiting?
2513 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2514 if (threadIsSuspended) {
2515 // The semaphore has been incremented, but while we were waiting
2516 // another thread suspended us. We don't want to continue running
2517 // while suspended because that would surprise the thread that
2518 // suspended us.
2519 ::sem_post(&sig_sem);
2520
2521 thread->java_suspend_self();
2522 }
2523 } while (threadIsSuspended);
2524 }
2525 }
2526
2527 int os::signal_lookup() {
2528 return check_pending_signals(false);
2529 }
2530
2531 int os::signal_wait() {
2532 return check_pending_signals(true);
2533 }
2534
2535 ////////////////////////////////////////////////////////////////////////////////
2536 // Virtual Memory
2537
2538 int os::vm_page_size() {
2539 // Seems redundant as all get out
2540 assert(os::Linux::page_size() != -1, "must call os::init");
2541 return os::Linux::page_size();
2542 }
2543
2544 // Solaris allocates memory by pages.
2545 int os::vm_allocation_granularity() {
2546 assert(os::Linux::page_size() != -1, "must call os::init");
2547 return os::Linux::page_size();
2548 }
2549
2550 // Rationale behind this function:
2551 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2552 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2553 // samples for JITted code. Here we create private executable mapping over the code cache
2554 // and then we can use standard (well, almost, as mapping can change) way to provide
2555 // info for the reporting script by storing timestamp and location of symbol
2556 void linux_wrap_code(char* base, size_t size) {
2557 static volatile jint cnt = 0;
2558
2559 if (!UseOprofile) {
2560 return;
2561 }
2562
2563 char buf[PATH_MAX+1];
2564 int num = Atomic::add(1, &cnt);
2565
2566 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2567 os::get_temp_directory(), os::current_process_id(), num);
2568 unlink(buf);
2569
2570 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2571
2572 if (fd != -1) {
2573 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2574 if (rv != (off_t)-1) {
2575 if (::write(fd, "", 1) == 1) {
2576 mmap(base, size,
2577 PROT_READ|PROT_WRITE|PROT_EXEC,
2578 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2579 }
2580 }
2581 ::close(fd);
2582 unlink(buf);
2583 }
2584 }
2585
2586 static bool recoverable_mmap_error(int err) {
2587 // See if the error is one we can let the caller handle. This
2588 // list of errno values comes from JBS-6843484. I can't find a
2589 // Linux man page that documents this specific set of errno
2590 // values so while this list currently matches Solaris, it may
2591 // change as we gain experience with this failure mode.
2592 switch (err) {
2593 case EBADF:
2594 case EINVAL:
2595 case ENOTSUP:
2596 // let the caller deal with these errors
2597 return true;
2598
2599 default:
2600 // Any remaining errors on this OS can cause our reserved mapping
2601 // to be lost. That can cause confusion where different data
2602 // structures think they have the same memory mapped. The worst
2603 // scenario is if both the VM and a library think they have the
2604 // same memory mapped.
2605 return false;
2606 }
2607 }
2608
2609 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2610 int err) {
2611 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2612 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2613 os::strerror(err), err);
2614 }
2615
2616 static void warn_fail_commit_memory(char* addr, size_t size,
2617 size_t alignment_hint, bool exec,
2618 int err) {
2619 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2620 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2621 alignment_hint, exec, os::strerror(err), err);
2622 }
2623
2624 // NOTE: Linux kernel does not really reserve the pages for us.
2625 // All it does is to check if there are enough free pages
2626 // left at the time of mmap(). This could be a potential
2627 // problem.
2628 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2629 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2630 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2631 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2632 if (res != (uintptr_t) MAP_FAILED) {
2633 if (UseNUMAInterleaving) {
2634 numa_make_global(addr, size);
2635 }
2636 return 0;
2637 }
2638
2639 int err = errno; // save errno from mmap() call above
2640
2641 if (!recoverable_mmap_error(err)) {
2642 warn_fail_commit_memory(addr, size, exec, err);
2643 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2644 }
2645
2646 return err;
2647 }
2648
2649 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2650 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2651 }
2652
2653 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2654 const char* mesg) {
2655 assert(mesg != NULL, "mesg must be specified");
2656 int err = os::Linux::commit_memory_impl(addr, size, exec);
2657 if (err != 0) {
2658 // the caller wants all commit errors to exit with the specified mesg:
2659 warn_fail_commit_memory(addr, size, exec, err);
2660 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2661 }
2662 }
2663
2664 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2665 #ifndef MAP_HUGETLB
2666 #define MAP_HUGETLB 0x40000
2667 #endif
2668
2669 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2670 #ifndef MADV_HUGEPAGE
2671 #define MADV_HUGEPAGE 14
2672 #endif
2673
2674 int os::Linux::commit_memory_impl(char* addr, size_t size,
2675 size_t alignment_hint, bool exec) {
2676 int err = os::Linux::commit_memory_impl(addr, size, exec);
2677 if (err == 0) {
2678 realign_memory(addr, size, alignment_hint);
2679 }
2680 return err;
2681 }
2682
2683 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2684 bool exec) {
2685 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2686 }
2687
2688 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2689 size_t alignment_hint, bool exec,
2690 const char* mesg) {
2691 assert(mesg != NULL, "mesg must be specified");
2692 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2693 if (err != 0) {
2694 // the caller wants all commit errors to exit with the specified mesg:
2695 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2696 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2697 }
2698 }
2699
2700 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2701 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2702 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2703 // be supported or the memory may already be backed by huge pages.
2704 ::madvise(addr, bytes, MADV_HUGEPAGE);
2705 }
2706 }
2707
2708 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2709 // This method works by doing an mmap over an existing mmaping and effectively discarding
2710 // the existing pages. However it won't work for SHM-based large pages that cannot be
2711 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2712 // small pages on top of the SHM segment. This method always works for small pages, so we
2713 // allow that in any case.
2714 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2715 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2716 }
2717 }
2718
2719 void os::numa_make_global(char *addr, size_t bytes) {
2720 Linux::numa_interleave_memory(addr, bytes);
2721 }
2722
2723 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2724 // bind policy to MPOL_PREFERRED for the current thread.
2725 #define USE_MPOL_PREFERRED 0
2726
2727 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2728 // To make NUMA and large pages more robust when both enabled, we need to ease
2729 // the requirements on where the memory should be allocated. MPOL_BIND is the
2730 // default policy and it will force memory to be allocated on the specified
2731 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2732 // the specified node, but will not force it. Using this policy will prevent
2733 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2734 // free large pages.
2735 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2736 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2737 }
2738
2739 bool os::numa_topology_changed() { return false; }
2740
2741 size_t os::numa_get_groups_num() {
2742 int max_node = Linux::numa_max_node();
2743 return max_node > 0 ? max_node + 1 : 1;
2744 }
2745
2746 int os::numa_get_group_id() {
2747 int cpu_id = Linux::sched_getcpu();
2748 if (cpu_id != -1) {
2749 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2750 if (lgrp_id != -1) {
2751 return lgrp_id;
2752 }
2753 }
2754 return 0;
2755 }
2756
2757 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2758 for (size_t i = 0; i < size; i++) {
2759 ids[i] = i;
2760 }
2761 return size;
2762 }
2763
2764 bool os::get_page_info(char *start, page_info* info) {
2765 return false;
2766 }
2767
2768 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2769 page_info* page_found) {
2770 return end;
2771 }
2772
2773
2774 int os::Linux::sched_getcpu_syscall(void) {
2775 unsigned int cpu = 0;
2776 int retval = -1;
2777
2778 #if defined(IA32)
2779 #ifndef SYS_getcpu
2780 #define SYS_getcpu 318
2781 #endif
2782 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2783 #elif defined(AMD64)
2784 // Unfortunately we have to bring all these macros here from vsyscall.h
2785 // to be able to compile on old linuxes.
2786 #define __NR_vgetcpu 2
2787 #define VSYSCALL_START (-10UL << 20)
2788 #define VSYSCALL_SIZE 1024
2789 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2790 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2791 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2792 retval = vgetcpu(&cpu, NULL, NULL);
2793 #endif
2794
2795 return (retval == -1) ? retval : cpu;
2796 }
2797
2798 // Something to do with the numa-aware allocator needs these symbols
2799 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2800 extern "C" JNIEXPORT void numa_error(char *where) { }
2801
2802
2803 // If we are running with libnuma version > 2, then we should
2804 // be trying to use symbols with versions 1.1
2805 // If we are running with earlier version, which did not have symbol versions,
2806 // we should use the base version.
2807 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2808 void *f = dlvsym(handle, name, "libnuma_1.1");
2809 if (f == NULL) {
2810 f = dlsym(handle, name);
2811 }
2812 return f;
2813 }
2814
2815 bool os::Linux::libnuma_init() {
2816 // sched_getcpu() should be in libc.
2817 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2818 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2819
2820 // If it's not, try a direct syscall.
2821 if (sched_getcpu() == -1) {
2822 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2823 (void*)&sched_getcpu_syscall));
2824 }
2825
2826 if (sched_getcpu() != -1) { // Does it work?
2827 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2828 if (handle != NULL) {
2829 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2830 libnuma_dlsym(handle, "numa_node_to_cpus")));
2831 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2832 libnuma_dlsym(handle, "numa_max_node")));
2833 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2834 libnuma_dlsym(handle, "numa_available")));
2835 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2836 libnuma_dlsym(handle, "numa_tonode_memory")));
2837 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2838 libnuma_dlsym(handle, "numa_interleave_memory")));
2839 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2840 libnuma_dlsym(handle, "numa_set_bind_policy")));
2841
2842
2843 if (numa_available() != -1) {
2844 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2845 // Create a cpu -> node mapping
2846 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2847 rebuild_cpu_to_node_map();
2848 return true;
2849 }
2850 }
2851 }
2852 return false;
2853 }
2854
2855 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2856 // The table is later used in get_node_by_cpu().
2857 void os::Linux::rebuild_cpu_to_node_map() {
2858 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2859 // in libnuma (possible values are starting from 16,
2860 // and continuing up with every other power of 2, but less
2861 // than the maximum number of CPUs supported by kernel), and
2862 // is a subject to change (in libnuma version 2 the requirements
2863 // are more reasonable) we'll just hardcode the number they use
2864 // in the library.
2865 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2866
2867 size_t cpu_num = os::active_processor_count();
2868 size_t cpu_map_size = NCPUS / BitsPerCLong;
2869 size_t cpu_map_valid_size =
2870 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2871
2872 cpu_to_node()->clear();
2873 cpu_to_node()->at_grow(cpu_num - 1);
2874 size_t node_num = numa_get_groups_num();
2875
2876 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2877 for (size_t i = 0; i < node_num; i++) {
2878 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2879 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2880 if (cpu_map[j] != 0) {
2881 for (size_t k = 0; k < BitsPerCLong; k++) {
2882 if (cpu_map[j] & (1UL << k)) {
2883 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2884 }
2885 }
2886 }
2887 }
2888 }
2889 }
2890 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2891 }
2892
2893 int os::Linux::get_node_by_cpu(int cpu_id) {
2894 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2895 return cpu_to_node()->at(cpu_id);
2896 }
2897 return -1;
2898 }
2899
2900 GrowableArray<int>* os::Linux::_cpu_to_node;
2901 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2902 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2903 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2904 os::Linux::numa_available_func_t os::Linux::_numa_available;
2905 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2906 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2907 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2908 unsigned long* os::Linux::_numa_all_nodes;
2909
2910 bool os::pd_uncommit_memory(char* addr, size_t size) {
2911 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2912 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2913 return res != (uintptr_t) MAP_FAILED;
2914 }
2915
2916 static address get_stack_commited_bottom(address bottom, size_t size) {
2917 address nbot = bottom;
2918 address ntop = bottom + size;
2919
2920 size_t page_sz = os::vm_page_size();
2921 unsigned pages = size / page_sz;
2922
2923 unsigned char vec[1];
2924 unsigned imin = 1, imax = pages + 1, imid;
2925 int mincore_return_value = 0;
2926
2927 assert(imin <= imax, "Unexpected page size");
2928
2929 while (imin < imax) {
2930 imid = (imax + imin) / 2;
2931 nbot = ntop - (imid * page_sz);
2932
2933 // Use a trick with mincore to check whether the page is mapped or not.
2934 // mincore sets vec to 1 if page resides in memory and to 0 if page
2935 // is swapped output but if page we are asking for is unmapped
2936 // it returns -1,ENOMEM
2937 mincore_return_value = mincore(nbot, page_sz, vec);
2938
2939 if (mincore_return_value == -1) {
2940 // Page is not mapped go up
2941 // to find first mapped page
2942 if (errno != EAGAIN) {
2943 assert(errno == ENOMEM, "Unexpected mincore errno");
2944 imax = imid;
2945 }
2946 } else {
2947 // Page is mapped go down
2948 // to find first not mapped page
2949 imin = imid + 1;
2950 }
2951 }
2952
2953 nbot = nbot + page_sz;
2954
2955 // Adjust stack bottom one page up if last checked page is not mapped
2956 if (mincore_return_value == -1) {
2957 nbot = nbot + page_sz;
2958 }
2959
2960 return nbot;
2961 }
2962
2963
2964 // Linux uses a growable mapping for the stack, and if the mapping for
2965 // the stack guard pages is not removed when we detach a thread the
2966 // stack cannot grow beyond the pages where the stack guard was
2967 // mapped. If at some point later in the process the stack expands to
2968 // that point, the Linux kernel cannot expand the stack any further
2969 // because the guard pages are in the way, and a segfault occurs.
2970 //
2971 // However, it's essential not to split the stack region by unmapping
2972 // a region (leaving a hole) that's already part of the stack mapping,
2973 // so if the stack mapping has already grown beyond the guard pages at
2974 // the time we create them, we have to truncate the stack mapping.
2975 // So, we need to know the extent of the stack mapping when
2976 // create_stack_guard_pages() is called.
2977
2978 // We only need this for stacks that are growable: at the time of
2979 // writing thread stacks don't use growable mappings (i.e. those
2980 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2981 // only applies to the main thread.
2982
2983 // If the (growable) stack mapping already extends beyond the point
2984 // where we're going to put our guard pages, truncate the mapping at
2985 // that point by munmap()ping it. This ensures that when we later
2986 // munmap() the guard pages we don't leave a hole in the stack
2987 // mapping. This only affects the main/initial thread
2988
2989 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2990 if (os::Linux::is_initial_thread()) {
2991 // As we manually grow stack up to bottom inside create_attached_thread(),
2992 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2993 // we don't need to do anything special.
2994 // Check it first, before calling heavy function.
2995 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2996 unsigned char vec[1];
2997
2998 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2999 // Fallback to slow path on all errors, including EAGAIN
3000 stack_extent = (uintptr_t) get_stack_commited_bottom(
3001 os::Linux::initial_thread_stack_bottom(),
3002 (size_t)addr - stack_extent);
3003 }
3004
3005 if (stack_extent < (uintptr_t)addr) {
3006 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3007 }
3008 }
3009
3010 return os::commit_memory(addr, size, !ExecMem);
3011 }
3012
3013 // If this is a growable mapping, remove the guard pages entirely by
3014 // munmap()ping them. If not, just call uncommit_memory(). This only
3015 // affects the main/initial thread, but guard against future OS changes
3016 // It's safe to always unmap guard pages for initial thread because we
3017 // always place it right after end of the mapped region
3018
3019 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3020 uintptr_t stack_extent, stack_base;
3021
3022 if (os::Linux::is_initial_thread()) {
3023 return ::munmap(addr, size) == 0;
3024 }
3025
3026 return os::uncommit_memory(addr, size);
3027 }
3028
3029 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3030 // at 'requested_addr'. If there are existing memory mappings at the same
3031 // location, however, they will be overwritten. If 'fixed' is false,
3032 // 'requested_addr' is only treated as a hint, the return value may or
3033 // may not start from the requested address. Unlike Linux mmap(), this
3034 // function returns NULL to indicate failure.
3035 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3036 char * addr;
3037 int flags;
3038
3039 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3040 if (fixed) {
3041 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3042 flags |= MAP_FIXED;
3043 }
3044
3045 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3046 // touch an uncommitted page. Otherwise, the read/write might
3047 // succeed if we have enough swap space to back the physical page.
3048 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3049 flags, -1, 0);
3050
3051 return addr == MAP_FAILED ? NULL : addr;
3052 }
3053
3054 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3055 // (req_addr != NULL) or with a given alignment.
3056 // - bytes shall be a multiple of alignment.
3057 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3058 // - alignment sets the alignment at which memory shall be allocated.
3059 // It must be a multiple of allocation granularity.
3060 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3061 // req_addr or NULL.
3062 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3063
3064 size_t extra_size = bytes;
3065 if (req_addr == NULL && alignment > 0) {
3066 extra_size += alignment;
3067 }
3068
3069 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3070 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3071 -1, 0);
3072 if (start == MAP_FAILED) {
3073 start = NULL;
3074 } else {
3075 if (req_addr != NULL) {
3076 if (start != req_addr) {
3077 ::munmap(start, extra_size);
3078 start = NULL;
3079 }
3080 } else {
3081 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3082 char* const end_aligned = start_aligned + bytes;
3083 char* const end = start + extra_size;
3084 if (start_aligned > start) {
3085 ::munmap(start, start_aligned - start);
3086 }
3087 if (end_aligned < end) {
3088 ::munmap(end_aligned, end - end_aligned);
3089 }
3090 start = start_aligned;
3091 }
3092 }
3093 return start;
3094 }
3095
3096 static int anon_munmap(char * addr, size_t size) {
3097 return ::munmap(addr, size) == 0;
3098 }
3099
3100 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3101 size_t alignment_hint) {
3102 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3103 }
3104
3105 bool os::pd_release_memory(char* addr, size_t size) {
3106 return anon_munmap(addr, size);
3107 }
3108
3109 static bool linux_mprotect(char* addr, size_t size, int prot) {
3110 // Linux wants the mprotect address argument to be page aligned.
3111 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3112
3113 // According to SUSv3, mprotect() should only be used with mappings
3114 // established by mmap(), and mmap() always maps whole pages. Unaligned
3115 // 'addr' likely indicates problem in the VM (e.g. trying to change
3116 // protection of malloc'ed or statically allocated memory). Check the
3117 // caller if you hit this assert.
3118 assert(addr == bottom, "sanity check");
3119
3120 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3121 return ::mprotect(bottom, size, prot) == 0;
3122 }
3123
3124 // Set protections specified
3125 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3126 bool is_committed) {
3127 unsigned int p = 0;
3128 switch (prot) {
3129 case MEM_PROT_NONE: p = PROT_NONE; break;
3130 case MEM_PROT_READ: p = PROT_READ; break;
3131 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3132 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3133 default:
3134 ShouldNotReachHere();
3135 }
3136 // is_committed is unused.
3137 return linux_mprotect(addr, bytes, p);
3138 }
3139
3140 bool os::guard_memory(char* addr, size_t size) {
3141 return linux_mprotect(addr, size, PROT_NONE);
3142 }
3143
3144 bool os::unguard_memory(char* addr, size_t size) {
3145 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3146 }
3147
3148 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3149 size_t page_size) {
3150 bool result = false;
3151 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3152 MAP_ANONYMOUS|MAP_PRIVATE,
3153 -1, 0);
3154 if (p != MAP_FAILED) {
3155 void *aligned_p = align_ptr_up(p, page_size);
3156
3157 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3158
3159 munmap(p, page_size * 2);
3160 }
3161
3162 if (warn && !result) {
3163 warning("TransparentHugePages is not supported by the operating system.");
3164 }
3165
3166 return result;
3167 }
3168
3169 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3170 bool result = false;
3171 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3172 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3173 -1, 0);
3174
3175 if (p != MAP_FAILED) {
3176 // We don't know if this really is a huge page or not.
3177 FILE *fp = fopen("/proc/self/maps", "r");
3178 if (fp) {
3179 while (!feof(fp)) {
3180 char chars[257];
3181 long x = 0;
3182 if (fgets(chars, sizeof(chars), fp)) {
3183 if (sscanf(chars, "%lx-%*x", &x) == 1
3184 && x == (long)p) {
3185 if (strstr (chars, "hugepage")) {
3186 result = true;
3187 break;
3188 }
3189 }
3190 }
3191 }
3192 fclose(fp);
3193 }
3194 munmap(p, page_size);
3195 }
3196
3197 if (warn && !result) {
3198 warning("HugeTLBFS is not supported by the operating system.");
3199 }
3200
3201 return result;
3202 }
3203
3204 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3205 //
3206 // From the coredump_filter documentation:
3207 //
3208 // - (bit 0) anonymous private memory
3209 // - (bit 1) anonymous shared memory
3210 // - (bit 2) file-backed private memory
3211 // - (bit 3) file-backed shared memory
3212 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3213 // effective only if the bit 2 is cleared)
3214 // - (bit 5) hugetlb private memory
3215 // - (bit 6) hugetlb shared memory
3216 //
3217 static void set_coredump_filter(void) {
3218 FILE *f;
3219 long cdm;
3220
3221 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3222 return;
3223 }
3224
3225 if (fscanf(f, "%lx", &cdm) != 1) {
3226 fclose(f);
3227 return;
3228 }
3229
3230 rewind(f);
3231
3232 if ((cdm & LARGEPAGES_BIT) == 0) {
3233 cdm |= LARGEPAGES_BIT;
3234 fprintf(f, "%#lx", cdm);
3235 }
3236
3237 fclose(f);
3238 }
3239
3240 // Large page support
3241
3242 static size_t _large_page_size = 0;
3243
3244 size_t os::Linux::find_large_page_size() {
3245 size_t large_page_size = 0;
3246
3247 // large_page_size on Linux is used to round up heap size. x86 uses either
3248 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3249 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3250 // page as large as 256M.
3251 //
3252 // Here we try to figure out page size by parsing /proc/meminfo and looking
3253 // for a line with the following format:
3254 // Hugepagesize: 2048 kB
3255 //
3256 // If we can't determine the value (e.g. /proc is not mounted, or the text
3257 // format has been changed), we'll use the largest page size supported by
3258 // the processor.
3259
3260 #ifndef ZERO
3261 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3262 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3263 #endif // ZERO
3264
3265 FILE *fp = fopen("/proc/meminfo", "r");
3266 if (fp) {
3267 while (!feof(fp)) {
3268 int x = 0;
3269 char buf[16];
3270 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3271 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3272 large_page_size = x * K;
3273 break;
3274 }
3275 } else {
3276 // skip to next line
3277 for (;;) {
3278 int ch = fgetc(fp);
3279 if (ch == EOF || ch == (int)'\n') break;
3280 }
3281 }
3282 }
3283 fclose(fp);
3284 }
3285
3286 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3287 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3288 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3289 proper_unit_for_byte_size(large_page_size));
3290 }
3291
3292 return large_page_size;
3293 }
3294
3295 size_t os::Linux::setup_large_page_size() {
3296 _large_page_size = Linux::find_large_page_size();
3297 const size_t default_page_size = (size_t)Linux::page_size();
3298 if (_large_page_size > default_page_size) {
3299 _page_sizes[0] = _large_page_size;
3300 _page_sizes[1] = default_page_size;
3301 _page_sizes[2] = 0;
3302 }
3303
3304 return _large_page_size;
3305 }
3306
3307 bool os::Linux::setup_large_page_type(size_t page_size) {
3308 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3309 FLAG_IS_DEFAULT(UseSHM) &&
3310 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3311
3312 // The type of large pages has not been specified by the user.
3313
3314 // Try UseHugeTLBFS and then UseSHM.
3315 UseHugeTLBFS = UseSHM = true;
3316
3317 // Don't try UseTransparentHugePages since there are known
3318 // performance issues with it turned on. This might change in the future.
3319 UseTransparentHugePages = false;
3320 }
3321
3322 if (UseTransparentHugePages) {
3323 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3324 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3325 UseHugeTLBFS = false;
3326 UseSHM = false;
3327 return true;
3328 }
3329 UseTransparentHugePages = false;
3330 }
3331
3332 if (UseHugeTLBFS) {
3333 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3334 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3335 UseSHM = false;
3336 return true;
3337 }
3338 UseHugeTLBFS = false;
3339 }
3340
3341 return UseSHM;
3342 }
3343
3344 void os::large_page_init() {
3345 if (!UseLargePages &&
3346 !UseTransparentHugePages &&
3347 !UseHugeTLBFS &&
3348 !UseSHM) {
3349 // Not using large pages.
3350 return;
3351 }
3352
3353 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3354 // The user explicitly turned off large pages.
3355 // Ignore the rest of the large pages flags.
3356 UseTransparentHugePages = false;
3357 UseHugeTLBFS = false;
3358 UseSHM = false;
3359 return;
3360 }
3361
3362 size_t large_page_size = Linux::setup_large_page_size();
3363 UseLargePages = Linux::setup_large_page_type(large_page_size);
3364
3365 set_coredump_filter();
3366 }
3367
3368 #ifndef SHM_HUGETLB
3369 #define SHM_HUGETLB 04000
3370 #endif
3371
3372 #define shm_warning_format(format, ...) \
3373 do { \
3374 if (UseLargePages && \
3375 (!FLAG_IS_DEFAULT(UseLargePages) || \
3376 !FLAG_IS_DEFAULT(UseSHM) || \
3377 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3378 warning(format, __VA_ARGS__); \
3379 } \
3380 } while (0)
3381
3382 #define shm_warning(str) shm_warning_format("%s", str)
3383
3384 #define shm_warning_with_errno(str) \
3385 do { \
3386 int err = errno; \
3387 shm_warning_format(str " (error = %d)", err); \
3388 } while (0)
3389
3390 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3391 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3392
3393 if (!is_size_aligned(alignment, SHMLBA)) {
3394 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3395 return NULL;
3396 }
3397
3398 // To ensure that we get 'alignment' aligned memory from shmat,
3399 // we pre-reserve aligned virtual memory and then attach to that.
3400
3401 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3402 if (pre_reserved_addr == NULL) {
3403 // Couldn't pre-reserve aligned memory.
3404 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3405 return NULL;
3406 }
3407
3408 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3409 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3410
3411 if ((intptr_t)addr == -1) {
3412 int err = errno;
3413 shm_warning_with_errno("Failed to attach shared memory.");
3414
3415 assert(err != EACCES, "Unexpected error");
3416 assert(err != EIDRM, "Unexpected error");
3417 assert(err != EINVAL, "Unexpected error");
3418
3419 // Since we don't know if the kernel unmapped the pre-reserved memory area
3420 // we can't unmap it, since that would potentially unmap memory that was
3421 // mapped from other threads.
3422 return NULL;
3423 }
3424
3425 return addr;
3426 }
3427
3428 static char* shmat_at_address(int shmid, char* req_addr) {
3429 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3430 assert(false, "Requested address needs to be SHMLBA aligned");
3431 return NULL;
3432 }
3433
3434 char* addr = (char*)shmat(shmid, req_addr, 0);
3435
3436 if ((intptr_t)addr == -1) {
3437 shm_warning_with_errno("Failed to attach shared memory.");
3438 return NULL;
3439 }
3440
3441 return addr;
3442 }
3443
3444 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3445 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3446 if (req_addr != NULL) {
3447 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3448 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3449 return shmat_at_address(shmid, req_addr);
3450 }
3451
3452 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3453 // return large page size aligned memory addresses when req_addr == NULL.
3454 // However, if the alignment is larger than the large page size, we have
3455 // to manually ensure that the memory returned is 'alignment' aligned.
3456 if (alignment > os::large_page_size()) {
3457 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3458 return shmat_with_alignment(shmid, bytes, alignment);
3459 } else {
3460 return shmat_at_address(shmid, NULL);
3461 }
3462 }
3463
3464 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3465 char* req_addr, bool exec) {
3466 // "exec" is passed in but not used. Creating the shared image for
3467 // the code cache doesn't have an SHM_X executable permission to check.
3468 assert(UseLargePages && UseSHM, "only for SHM large pages");
3469 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3470 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3471
3472 if (!is_size_aligned(bytes, os::large_page_size())) {
3473 return NULL; // Fallback to small pages.
3474 }
3475
3476 // Create a large shared memory region to attach to based on size.
3477 // Currently, size is the total size of the heap.
3478 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3479 if (shmid == -1) {
3480 // Possible reasons for shmget failure:
3481 // 1. shmmax is too small for Java heap.
3482 // > check shmmax value: cat /proc/sys/kernel/shmmax
3483 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3484 // 2. not enough large page memory.
3485 // > check available large pages: cat /proc/meminfo
3486 // > increase amount of large pages:
3487 // echo new_value > /proc/sys/vm/nr_hugepages
3488 // Note 1: different Linux may use different name for this property,
3489 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3490 // Note 2: it's possible there's enough physical memory available but
3491 // they are so fragmented after a long run that they can't
3492 // coalesce into large pages. Try to reserve large pages when
3493 // the system is still "fresh".
3494 shm_warning_with_errno("Failed to reserve shared memory.");
3495 return NULL;
3496 }
3497
3498 // Attach to the region.
3499 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3500
3501 // Remove shmid. If shmat() is successful, the actual shared memory segment
3502 // will be deleted when it's detached by shmdt() or when the process
3503 // terminates. If shmat() is not successful this will remove the shared
3504 // segment immediately.
3505 shmctl(shmid, IPC_RMID, NULL);
3506
3507 return addr;
3508 }
3509
3510 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3511 int error) {
3512 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3513
3514 bool warn_on_failure = UseLargePages &&
3515 (!FLAG_IS_DEFAULT(UseLargePages) ||
3516 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3517 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3518
3519 if (warn_on_failure) {
3520 char msg[128];
3521 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3522 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3523 warning("%s", msg);
3524 }
3525 }
3526
3527 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3528 char* req_addr,
3529 bool exec) {
3530 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3531 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3532 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3533
3534 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3535 char* addr = (char*)::mmap(req_addr, bytes, prot,
3536 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3537 -1, 0);
3538
3539 if (addr == MAP_FAILED) {
3540 warn_on_large_pages_failure(req_addr, bytes, errno);
3541 return NULL;
3542 }
3543
3544 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3545
3546 return addr;
3547 }
3548
3549 // Reserve memory using mmap(MAP_HUGETLB).
3550 // - bytes shall be a multiple of alignment.
3551 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3552 // - alignment sets the alignment at which memory shall be allocated.
3553 // It must be a multiple of allocation granularity.
3554 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3555 // req_addr or NULL.
3556 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3557 size_t alignment,
3558 char* req_addr,
3559 bool exec) {
3560 size_t large_page_size = os::large_page_size();
3561 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3562
3563 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3564 assert(is_size_aligned(bytes, alignment), "Must be");
3565
3566 // First reserve - but not commit - the address range in small pages.
3567 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3568
3569 if (start == NULL) {
3570 return NULL;
3571 }
3572
3573 assert(is_ptr_aligned(start, alignment), "Must be");
3574
3575 char* end = start + bytes;
3576
3577 // Find the regions of the allocated chunk that can be promoted to large pages.
3578 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3579 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3580
3581 size_t lp_bytes = lp_end - lp_start;
3582
3583 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3584
3585 if (lp_bytes == 0) {
3586 // The mapped region doesn't even span the start and the end of a large page.
3587 // Fall back to allocate a non-special area.
3588 ::munmap(start, end - start);
3589 return NULL;
3590 }
3591
3592 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3593
3594 void* result;
3595
3596 // Commit small-paged leading area.
3597 if (start != lp_start) {
3598 result = ::mmap(start, lp_start - start, prot,
3599 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3600 -1, 0);
3601 if (result == MAP_FAILED) {
3602 ::munmap(lp_start, end - lp_start);
3603 return NULL;
3604 }
3605 }
3606
3607 // Commit large-paged area.
3608 result = ::mmap(lp_start, lp_bytes, prot,
3609 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3610 -1, 0);
3611 if (result == MAP_FAILED) {
3612 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3613 // If the mmap above fails, the large pages region will be unmapped and we
3614 // have regions before and after with small pages. Release these regions.
3615 //
3616 // | mapped | unmapped | mapped |
3617 // ^ ^ ^ ^
3618 // start lp_start lp_end end
3619 //
3620 ::munmap(start, lp_start - start);
3621 ::munmap(lp_end, end - lp_end);
3622 return NULL;
3623 }
3624
3625 // Commit small-paged trailing area.
3626 if (lp_end != end) {
3627 result = ::mmap(lp_end, end - lp_end, prot,
3628 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3629 -1, 0);
3630 if (result == MAP_FAILED) {
3631 ::munmap(start, lp_end - start);
3632 return NULL;
3633 }
3634 }
3635
3636 return start;
3637 }
3638
3639 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3640 size_t alignment,
3641 char* req_addr,
3642 bool exec) {
3643 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3644 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3645 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3646 assert(is_power_of_2(os::large_page_size()), "Must be");
3647 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3648
3649 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3650 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3651 } else {
3652 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3653 }
3654 }
3655
3656 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3657 char* req_addr, bool exec) {
3658 assert(UseLargePages, "only for large pages");
3659
3660 char* addr;
3661 if (UseSHM) {
3662 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3663 } else {
3664 assert(UseHugeTLBFS, "must be");
3665 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3666 }
3667
3668 if (addr != NULL) {
3669 if (UseNUMAInterleaving) {
3670 numa_make_global(addr, bytes);
3671 }
3672
3673 // The memory is committed
3674 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3675 }
3676
3677 return addr;
3678 }
3679
3680 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3681 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3682 return shmdt(base) == 0;
3683 }
3684
3685 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3686 return pd_release_memory(base, bytes);
3687 }
3688
3689 bool os::release_memory_special(char* base, size_t bytes) {
3690 bool res;
3691 if (MemTracker::tracking_level() > NMT_minimal) {
3692 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3693 res = os::Linux::release_memory_special_impl(base, bytes);
3694 if (res) {
3695 tkr.record((address)base, bytes);
3696 }
3697
3698 } else {
3699 res = os::Linux::release_memory_special_impl(base, bytes);
3700 }
3701 return res;
3702 }
3703
3704 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3705 assert(UseLargePages, "only for large pages");
3706 bool res;
3707
3708 if (UseSHM) {
3709 res = os::Linux::release_memory_special_shm(base, bytes);
3710 } else {
3711 assert(UseHugeTLBFS, "must be");
3712 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3713 }
3714 return res;
3715 }
3716
3717 size_t os::large_page_size() {
3718 return _large_page_size;
3719 }
3720
3721 // With SysV SHM the entire memory region must be allocated as shared
3722 // memory.
3723 // HugeTLBFS allows application to commit large page memory on demand.
3724 // However, when committing memory with HugeTLBFS fails, the region
3725 // that was supposed to be committed will lose the old reservation
3726 // and allow other threads to steal that memory region. Because of this
3727 // behavior we can't commit HugeTLBFS memory.
3728 bool os::can_commit_large_page_memory() {
3729 return UseTransparentHugePages;
3730 }
3731
3732 bool os::can_execute_large_page_memory() {
3733 return UseTransparentHugePages || UseHugeTLBFS;
3734 }
3735
3736 // Reserve memory at an arbitrary address, only if that area is
3737 // available (and not reserved for something else).
3738
3739 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3740 const int max_tries = 10;
3741 char* base[max_tries];
3742 size_t size[max_tries];
3743 const size_t gap = 0x000000;
3744
3745 // Assert only that the size is a multiple of the page size, since
3746 // that's all that mmap requires, and since that's all we really know
3747 // about at this low abstraction level. If we need higher alignment,
3748 // we can either pass an alignment to this method or verify alignment
3749 // in one of the methods further up the call chain. See bug 5044738.
3750 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3751
3752 // Repeatedly allocate blocks until the block is allocated at the
3753 // right spot.
3754
3755 // Linux mmap allows caller to pass an address as hint; give it a try first,
3756 // if kernel honors the hint then we can return immediately.
3757 char * addr = anon_mmap(requested_addr, bytes, false);
3758 if (addr == requested_addr) {
3759 return requested_addr;
3760 }
3761
3762 if (addr != NULL) {
3763 // mmap() is successful but it fails to reserve at the requested address
3764 anon_munmap(addr, bytes);
3765 }
3766
3767 int i;
3768 for (i = 0; i < max_tries; ++i) {
3769 base[i] = reserve_memory(bytes);
3770
3771 if (base[i] != NULL) {
3772 // Is this the block we wanted?
3773 if (base[i] == requested_addr) {
3774 size[i] = bytes;
3775 break;
3776 }
3777
3778 // Does this overlap the block we wanted? Give back the overlapped
3779 // parts and try again.
3780
3781 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3782 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3783 unmap_memory(base[i], top_overlap);
3784 base[i] += top_overlap;
3785 size[i] = bytes - top_overlap;
3786 } else {
3787 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3788 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3789 unmap_memory(requested_addr, bottom_overlap);
3790 size[i] = bytes - bottom_overlap;
3791 } else {
3792 size[i] = bytes;
3793 }
3794 }
3795 }
3796 }
3797
3798 // Give back the unused reserved pieces.
3799
3800 for (int j = 0; j < i; ++j) {
3801 if (base[j] != NULL) {
3802 unmap_memory(base[j], size[j]);
3803 }
3804 }
3805
3806 if (i < max_tries) {
3807 return requested_addr;
3808 } else {
3809 return NULL;
3810 }
3811 }
3812
3813 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3814 return ::read(fd, buf, nBytes);
3815 }
3816
3817 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3818 return ::pread(fd, buf, nBytes, offset);
3819 }
3820
3821 // Short sleep, direct OS call.
3822 //
3823 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3824 // sched_yield(2) will actually give up the CPU:
3825 //
3826 // * Alone on this pariticular CPU, keeps running.
3827 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3828 // (pre 2.6.39).
3829 //
3830 // So calling this with 0 is an alternative.
3831 //
3832 void os::naked_short_sleep(jlong ms) {
3833 struct timespec req;
3834
3835 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3836 req.tv_sec = 0;
3837 if (ms > 0) {
3838 req.tv_nsec = (ms % 1000) * 1000000;
3839 } else {
3840 req.tv_nsec = 1;
3841 }
3842
3843 nanosleep(&req, NULL);
3844
3845 return;
3846 }
3847
3848 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3849 void os::infinite_sleep() {
3850 while (true) { // sleep forever ...
3851 ::sleep(100); // ... 100 seconds at a time
3852 }
3853 }
3854
3855 // Used to convert frequent JVM_Yield() to nops
3856 bool os::dont_yield() {
3857 return DontYieldALot;
3858 }
3859
3860 void os::naked_yield() {
3861 sched_yield();
3862 }
3863
3864 ////////////////////////////////////////////////////////////////////////////////
3865 // thread priority support
3866
3867 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3868 // only supports dynamic priority, static priority must be zero. For real-time
3869 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3870 // However, for large multi-threaded applications, SCHED_RR is not only slower
3871 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3872 // of 5 runs - Sep 2005).
3873 //
3874 // The following code actually changes the niceness of kernel-thread/LWP. It
3875 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3876 // not the entire user process, and user level threads are 1:1 mapped to kernel
3877 // threads. It has always been the case, but could change in the future. For
3878 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3879 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3880
3881 int os::java_to_os_priority[CriticalPriority + 1] = {
3882 19, // 0 Entry should never be used
3883
3884 4, // 1 MinPriority
3885 3, // 2
3886 2, // 3
3887
3888 1, // 4
3889 0, // 5 NormPriority
3890 -1, // 6
3891
3892 -2, // 7
3893 -3, // 8
3894 -4, // 9 NearMaxPriority
3895
3896 -5, // 10 MaxPriority
3897
3898 -5 // 11 CriticalPriority
3899 };
3900
3901 static int prio_init() {
3902 if (ThreadPriorityPolicy == 1) {
3903 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3904 // if effective uid is not root. Perhaps, a more elegant way of doing
3905 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3906 if (geteuid() != 0) {
3907 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3908 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3909 }
3910 ThreadPriorityPolicy = 0;
3911 }
3912 }
3913 if (UseCriticalJavaThreadPriority) {
3914 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3915 }
3916 return 0;
3917 }
3918
3919 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3920 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3921
3922 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3923 return (ret == 0) ? OS_OK : OS_ERR;
3924 }
3925
3926 OSReturn os::get_native_priority(const Thread* const thread,
3927 int *priority_ptr) {
3928 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3929 *priority_ptr = java_to_os_priority[NormPriority];
3930 return OS_OK;
3931 }
3932
3933 errno = 0;
3934 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3935 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3936 }
3937
3938 // Hint to the underlying OS that a task switch would not be good.
3939 // Void return because it's a hint and can fail.
3940 void os::hint_no_preempt() {}
3941
3942 ////////////////////////////////////////////////////////////////////////////////
3943 // suspend/resume support
3944
3945 // the low-level signal-based suspend/resume support is a remnant from the
3946 // old VM-suspension that used to be for java-suspension, safepoints etc,
3947 // within hotspot. Now there is a single use-case for this:
3948 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3949 // that runs in the watcher thread.
3950 // The remaining code is greatly simplified from the more general suspension
3951 // code that used to be used.
3952 //
3953 // The protocol is quite simple:
3954 // - suspend:
3955 // - sends a signal to the target thread
3956 // - polls the suspend state of the osthread using a yield loop
3957 // - target thread signal handler (SR_handler) sets suspend state
3958 // and blocks in sigsuspend until continued
3959 // - resume:
3960 // - sets target osthread state to continue
3961 // - sends signal to end the sigsuspend loop in the SR_handler
3962 //
3963 // Note that the SR_lock plays no role in this suspend/resume protocol.
3964
3965 static void resume_clear_context(OSThread *osthread) {
3966 osthread->set_ucontext(NULL);
3967 osthread->set_siginfo(NULL);
3968 }
3969
3970 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3971 ucontext_t* context) {
3972 osthread->set_ucontext(context);
3973 osthread->set_siginfo(siginfo);
3974 }
3975
3976 // Handler function invoked when a thread's execution is suspended or
3977 // resumed. We have to be careful that only async-safe functions are
3978 // called here (Note: most pthread functions are not async safe and
3979 // should be avoided.)
3980 //
3981 // Note: sigwait() is a more natural fit than sigsuspend() from an
3982 // interface point of view, but sigwait() prevents the signal hander
3983 // from being run. libpthread would get very confused by not having
3984 // its signal handlers run and prevents sigwait()'s use with the
3985 // mutex granting granting signal.
3986 //
3987 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3988 //
3989 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3990 // Save and restore errno to avoid confusing native code with EINTR
3991 // after sigsuspend.
3992 int old_errno = errno;
3993
3994 Thread* thread = Thread::current_or_null_safe();
3995 assert(thread != NULL, "Missing current thread in SR_handler");
3996 OSThread* osthread = thread->osthread();
3997 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3998
3999 os::SuspendResume::State current = osthread->sr.state();
4000 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4001 suspend_save_context(osthread, siginfo, context);
4002
4003 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4004 os::SuspendResume::State state = osthread->sr.suspended();
4005 if (state == os::SuspendResume::SR_SUSPENDED) {
4006 sigset_t suspend_set; // signals for sigsuspend()
4007 sigemptyset(&suspend_set);
4008 // get current set of blocked signals and unblock resume signal
4009 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4010 sigdelset(&suspend_set, SR_signum);
4011
4012 sr_semaphore.signal();
4013 // wait here until we are resumed
4014 while (1) {
4015 sigsuspend(&suspend_set);
4016
4017 os::SuspendResume::State result = osthread->sr.running();
4018 if (result == os::SuspendResume::SR_RUNNING) {
4019 sr_semaphore.signal();
4020 break;
4021 }
4022 }
4023
4024 } else if (state == os::SuspendResume::SR_RUNNING) {
4025 // request was cancelled, continue
4026 } else {
4027 ShouldNotReachHere();
4028 }
4029
4030 resume_clear_context(osthread);
4031 } else if (current == os::SuspendResume::SR_RUNNING) {
4032 // request was cancelled, continue
4033 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4034 // ignore
4035 } else {
4036 // ignore
4037 }
4038
4039 errno = old_errno;
4040 }
4041
4042 static int SR_initialize() {
4043 struct sigaction act;
4044 char *s;
4045
4046 // Get signal number to use for suspend/resume
4047 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4048 int sig = ::strtol(s, 0, 10);
4049 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4050 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4051 SR_signum = sig;
4052 } else {
4053 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4054 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4055 }
4056 }
4057
4058 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4059 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4060
4061 sigemptyset(&SR_sigset);
4062 sigaddset(&SR_sigset, SR_signum);
4063
4064 // Set up signal handler for suspend/resume
4065 act.sa_flags = SA_RESTART|SA_SIGINFO;
4066 act.sa_handler = (void (*)(int)) SR_handler;
4067
4068 // SR_signum is blocked by default.
4069 // 4528190 - We also need to block pthread restart signal (32 on all
4070 // supported Linux platforms). Note that LinuxThreads need to block
4071 // this signal for all threads to work properly. So we don't have
4072 // to use hard-coded signal number when setting up the mask.
4073 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4074
4075 if (sigaction(SR_signum, &act, 0) == -1) {
4076 return -1;
4077 }
4078
4079 // Save signal flag
4080 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4081 return 0;
4082 }
4083
4084 static int sr_notify(OSThread* osthread) {
4085 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4086 assert_status(status == 0, status, "pthread_kill");
4087 return status;
4088 }
4089
4090 // "Randomly" selected value for how long we want to spin
4091 // before bailing out on suspending a thread, also how often
4092 // we send a signal to a thread we want to resume
4093 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4094 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4095
4096 // returns true on success and false on error - really an error is fatal
4097 // but this seems the normal response to library errors
4098 static bool do_suspend(OSThread* osthread) {
4099 assert(osthread->sr.is_running(), "thread should be running");
4100 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4101
4102 // mark as suspended and send signal
4103 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4104 // failed to switch, state wasn't running?
4105 ShouldNotReachHere();
4106 return false;
4107 }
4108
4109 if (sr_notify(osthread) != 0) {
4110 ShouldNotReachHere();
4111 }
4112
4113 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4114 while (true) {
4115 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4116 break;
4117 } else {
4118 // timeout
4119 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4120 if (cancelled == os::SuspendResume::SR_RUNNING) {
4121 return false;
4122 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4123 // make sure that we consume the signal on the semaphore as well
4124 sr_semaphore.wait();
4125 break;
4126 } else {
4127 ShouldNotReachHere();
4128 return false;
4129 }
4130 }
4131 }
4132
4133 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4134 return true;
4135 }
4136
4137 static void do_resume(OSThread* osthread) {
4138 assert(osthread->sr.is_suspended(), "thread should be suspended");
4139 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4140
4141 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4142 // failed to switch to WAKEUP_REQUEST
4143 ShouldNotReachHere();
4144 return;
4145 }
4146
4147 while (true) {
4148 if (sr_notify(osthread) == 0) {
4149 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4150 if (osthread->sr.is_running()) {
4151 return;
4152 }
4153 }
4154 } else {
4155 ShouldNotReachHere();
4156 }
4157 }
4158
4159 guarantee(osthread->sr.is_running(), "Must be running!");
4160 }
4161
4162 ///////////////////////////////////////////////////////////////////////////////////
4163 // signal handling (except suspend/resume)
4164
4165 // This routine may be used by user applications as a "hook" to catch signals.
4166 // The user-defined signal handler must pass unrecognized signals to this
4167 // routine, and if it returns true (non-zero), then the signal handler must
4168 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4169 // routine will never retun false (zero), but instead will execute a VM panic
4170 // routine kill the process.
4171 //
4172 // If this routine returns false, it is OK to call it again. This allows
4173 // the user-defined signal handler to perform checks either before or after
4174 // the VM performs its own checks. Naturally, the user code would be making
4175 // a serious error if it tried to handle an exception (such as a null check
4176 // or breakpoint) that the VM was generating for its own correct operation.
4177 //
4178 // This routine may recognize any of the following kinds of signals:
4179 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4180 // It should be consulted by handlers for any of those signals.
4181 //
4182 // The caller of this routine must pass in the three arguments supplied
4183 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4184 // field of the structure passed to sigaction(). This routine assumes that
4185 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4186 //
4187 // Note that the VM will print warnings if it detects conflicting signal
4188 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4189 //
4190 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4191 siginfo_t* siginfo,
4192 void* ucontext,
4193 int abort_if_unrecognized);
4194
4195 void signalHandler(int sig, siginfo_t* info, void* uc) {
4196 assert(info != NULL && uc != NULL, "it must be old kernel");
4197 int orig_errno = errno; // Preserve errno value over signal handler.
4198 JVM_handle_linux_signal(sig, info, uc, true);
4199 errno = orig_errno;
4200 }
4201
4202
4203 // This boolean allows users to forward their own non-matching signals
4204 // to JVM_handle_linux_signal, harmlessly.
4205 bool os::Linux::signal_handlers_are_installed = false;
4206
4207 // For signal-chaining
4208 struct sigaction sigact[NSIG];
4209 uint64_t sigs = 0;
4210 #if (64 < NSIG-1)
4211 #error "Not all signals can be encoded in sigs. Adapt its type!"
4212 #endif
4213 bool os::Linux::libjsig_is_loaded = false;
4214 typedef struct sigaction *(*get_signal_t)(int);
4215 get_signal_t os::Linux::get_signal_action = NULL;
4216
4217 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4218 struct sigaction *actp = NULL;
4219
4220 if (libjsig_is_loaded) {
4221 // Retrieve the old signal handler from libjsig
4222 actp = (*get_signal_action)(sig);
4223 }
4224 if (actp == NULL) {
4225 // Retrieve the preinstalled signal handler from jvm
4226 actp = get_preinstalled_handler(sig);
4227 }
4228
4229 return actp;
4230 }
4231
4232 static bool call_chained_handler(struct sigaction *actp, int sig,
4233 siginfo_t *siginfo, void *context) {
4234 // Call the old signal handler
4235 if (actp->sa_handler == SIG_DFL) {
4236 // It's more reasonable to let jvm treat it as an unexpected exception
4237 // instead of taking the default action.
4238 return false;
4239 } else if (actp->sa_handler != SIG_IGN) {
4240 if ((actp->sa_flags & SA_NODEFER) == 0) {
4241 // automaticlly block the signal
4242 sigaddset(&(actp->sa_mask), sig);
4243 }
4244
4245 sa_handler_t hand = NULL;
4246 sa_sigaction_t sa = NULL;
4247 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4248 // retrieve the chained handler
4249 if (siginfo_flag_set) {
4250 sa = actp->sa_sigaction;
4251 } else {
4252 hand = actp->sa_handler;
4253 }
4254
4255 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4256 actp->sa_handler = SIG_DFL;
4257 }
4258
4259 // try to honor the signal mask
4260 sigset_t oset;
4261 sigemptyset(&oset);
4262 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4263
4264 // call into the chained handler
4265 if (siginfo_flag_set) {
4266 (*sa)(sig, siginfo, context);
4267 } else {
4268 (*hand)(sig);
4269 }
4270
4271 // restore the signal mask
4272 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4273 }
4274 // Tell jvm's signal handler the signal is taken care of.
4275 return true;
4276 }
4277
4278 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4279 bool chained = false;
4280 // signal-chaining
4281 if (UseSignalChaining) {
4282 struct sigaction *actp = get_chained_signal_action(sig);
4283 if (actp != NULL) {
4284 chained = call_chained_handler(actp, sig, siginfo, context);
4285 }
4286 }
4287 return chained;
4288 }
4289
4290 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4291 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4292 return &sigact[sig];
4293 }
4294 return NULL;
4295 }
4296
4297 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4298 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4299 sigact[sig] = oldAct;
4300 sigs |= (uint64_t)1 << (sig-1);
4301 }
4302
4303 // for diagnostic
4304 int sigflags[NSIG];
4305
4306 int os::Linux::get_our_sigflags(int sig) {
4307 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4308 return sigflags[sig];
4309 }
4310
4311 void os::Linux::set_our_sigflags(int sig, int flags) {
4312 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4313 if (sig > 0 && sig < NSIG) {
4314 sigflags[sig] = flags;
4315 }
4316 }
4317
4318 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4319 // Check for overwrite.
4320 struct sigaction oldAct;
4321 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4322
4323 void* oldhand = oldAct.sa_sigaction
4324 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4325 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4326 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4327 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4328 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4329 if (AllowUserSignalHandlers || !set_installed) {
4330 // Do not overwrite; user takes responsibility to forward to us.
4331 return;
4332 } else if (UseSignalChaining) {
4333 // save the old handler in jvm
4334 save_preinstalled_handler(sig, oldAct);
4335 // libjsig also interposes the sigaction() call below and saves the
4336 // old sigaction on it own.
4337 } else {
4338 fatal("Encountered unexpected pre-existing sigaction handler "
4339 "%#lx for signal %d.", (long)oldhand, sig);
4340 }
4341 }
4342
4343 struct sigaction sigAct;
4344 sigfillset(&(sigAct.sa_mask));
4345 sigAct.sa_handler = SIG_DFL;
4346 if (!set_installed) {
4347 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4348 } else {
4349 sigAct.sa_sigaction = signalHandler;
4350 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4351 }
4352 // Save flags, which are set by ours
4353 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4354 sigflags[sig] = sigAct.sa_flags;
4355
4356 int ret = sigaction(sig, &sigAct, &oldAct);
4357 assert(ret == 0, "check");
4358
4359 void* oldhand2 = oldAct.sa_sigaction
4360 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4361 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4362 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4363 }
4364
4365 // install signal handlers for signals that HotSpot needs to
4366 // handle in order to support Java-level exception handling.
4367
4368 void os::Linux::install_signal_handlers() {
4369 if (!signal_handlers_are_installed) {
4370 signal_handlers_are_installed = true;
4371
4372 // signal-chaining
4373 typedef void (*signal_setting_t)();
4374 signal_setting_t begin_signal_setting = NULL;
4375 signal_setting_t end_signal_setting = NULL;
4376 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4377 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4378 if (begin_signal_setting != NULL) {
4379 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4380 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4381 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4382 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4383 libjsig_is_loaded = true;
4384 assert(UseSignalChaining, "should enable signal-chaining");
4385 }
4386 if (libjsig_is_loaded) {
4387 // Tell libjsig jvm is setting signal handlers
4388 (*begin_signal_setting)();
4389 }
4390
4391 set_signal_handler(SIGSEGV, true);
4392 set_signal_handler(SIGPIPE, true);
4393 set_signal_handler(SIGBUS, true);
4394 set_signal_handler(SIGILL, true);
4395 set_signal_handler(SIGFPE, true);
4396 #if defined(PPC64)
4397 set_signal_handler(SIGTRAP, true);
4398 #endif
4399 set_signal_handler(SIGXFSZ, true);
4400
4401 if (libjsig_is_loaded) {
4402 // Tell libjsig jvm finishes setting signal handlers
4403 (*end_signal_setting)();
4404 }
4405
4406 // We don't activate signal checker if libjsig is in place, we trust ourselves
4407 // and if UserSignalHandler is installed all bets are off.
4408 // Log that signal checking is off only if -verbose:jni is specified.
4409 if (CheckJNICalls) {
4410 if (libjsig_is_loaded) {
4411 if (PrintJNIResolving) {
4412 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4413 }
4414 check_signals = false;
4415 }
4416 if (AllowUserSignalHandlers) {
4417 if (PrintJNIResolving) {
4418 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4419 }
4420 check_signals = false;
4421 }
4422 }
4423 }
4424 }
4425
4426 // This is the fastest way to get thread cpu time on Linux.
4427 // Returns cpu time (user+sys) for any thread, not only for current.
4428 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4429 // It might work on 2.6.10+ with a special kernel/glibc patch.
4430 // For reference, please, see IEEE Std 1003.1-2004:
4431 // http://www.unix.org/single_unix_specification
4432
4433 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4434 struct timespec tp;
4435 int rc = os::Linux::clock_gettime(clockid, &tp);
4436 assert(rc == 0, "clock_gettime is expected to return 0 code");
4437
4438 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4439 }
4440
4441 void os::Linux::initialize_os_info() {
4442 assert(_os_version == 0, "OS info already initialized");
4443
4444 struct utsname _uname;
4445
4446 uint32_t major;
4447 uint32_t minor;
4448 uint32_t fix;
4449
4450 int rc;
4451
4452 // Kernel version is unknown if
4453 // verification below fails.
4454 _os_version = 0x01000000;
4455
4456 rc = uname(&_uname);
4457 if (rc != -1) {
4458
4459 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4460 if (rc == 3) {
4461
4462 if (major < 256 && minor < 256 && fix < 256) {
4463 // Kernel version format is as expected,
4464 // set it overriding unknown state.
4465 _os_version = (major << 16) |
4466 (minor << 8 ) |
4467 (fix << 0 ) ;
4468 }
4469 }
4470 }
4471 }
4472
4473 uint32_t os::Linux::os_version() {
4474 assert(_os_version != 0, "not initialized");
4475 return _os_version & 0x00FFFFFF;
4476 }
4477
4478 bool os::Linux::os_version_is_known() {
4479 assert(_os_version != 0, "not initialized");
4480 return _os_version & 0x01000000 ? false : true;
4481 }
4482
4483 /////
4484 // glibc on Linux platform uses non-documented flag
4485 // to indicate, that some special sort of signal
4486 // trampoline is used.
4487 // We will never set this flag, and we should
4488 // ignore this flag in our diagnostic
4489 #ifdef SIGNIFICANT_SIGNAL_MASK
4490 #undef SIGNIFICANT_SIGNAL_MASK
4491 #endif
4492 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4493
4494 static const char* get_signal_handler_name(address handler,
4495 char* buf, int buflen) {
4496 int offset = 0;
4497 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4498 if (found) {
4499 // skip directory names
4500 const char *p1, *p2;
4501 p1 = buf;
4502 size_t len = strlen(os::file_separator());
4503 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4504 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4505 } else {
4506 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4507 }
4508 return buf;
4509 }
4510
4511 static void print_signal_handler(outputStream* st, int sig,
4512 char* buf, size_t buflen) {
4513 struct sigaction sa;
4514
4515 sigaction(sig, NULL, &sa);
4516
4517 // See comment for SIGNIFICANT_SIGNAL_MASK define
4518 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4519
4520 st->print("%s: ", os::exception_name(sig, buf, buflen));
4521
4522 address handler = (sa.sa_flags & SA_SIGINFO)
4523 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4524 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4525
4526 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4527 st->print("SIG_DFL");
4528 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4529 st->print("SIG_IGN");
4530 } else {
4531 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4532 }
4533
4534 st->print(", sa_mask[0]=");
4535 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4536
4537 address rh = VMError::get_resetted_sighandler(sig);
4538 // May be, handler was resetted by VMError?
4539 if (rh != NULL) {
4540 handler = rh;
4541 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4542 }
4543
4544 st->print(", sa_flags=");
4545 os::Posix::print_sa_flags(st, sa.sa_flags);
4546
4547 // Check: is it our handler?
4548 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4549 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4550 // It is our signal handler
4551 // check for flags, reset system-used one!
4552 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4553 st->print(
4554 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4555 os::Linux::get_our_sigflags(sig));
4556 }
4557 }
4558 st->cr();
4559 }
4560
4561
4562 #define DO_SIGNAL_CHECK(sig) \
4563 do { \
4564 if (!sigismember(&check_signal_done, sig)) { \
4565 os::Linux::check_signal_handler(sig); \
4566 } \
4567 } while (0)
4568
4569 // This method is a periodic task to check for misbehaving JNI applications
4570 // under CheckJNI, we can add any periodic checks here
4571
4572 void os::run_periodic_checks() {
4573 if (check_signals == false) return;
4574
4575 // SEGV and BUS if overridden could potentially prevent
4576 // generation of hs*.log in the event of a crash, debugging
4577 // such a case can be very challenging, so we absolutely
4578 // check the following for a good measure:
4579 DO_SIGNAL_CHECK(SIGSEGV);
4580 DO_SIGNAL_CHECK(SIGILL);
4581 DO_SIGNAL_CHECK(SIGFPE);
4582 DO_SIGNAL_CHECK(SIGBUS);
4583 DO_SIGNAL_CHECK(SIGPIPE);
4584 DO_SIGNAL_CHECK(SIGXFSZ);
4585 #if defined(PPC64)
4586 DO_SIGNAL_CHECK(SIGTRAP);
4587 #endif
4588
4589 // ReduceSignalUsage allows the user to override these handlers
4590 // see comments at the very top and jvm_solaris.h
4591 if (!ReduceSignalUsage) {
4592 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4593 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4594 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4595 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4596 }
4597
4598 DO_SIGNAL_CHECK(SR_signum);
4599 }
4600
4601 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4602
4603 static os_sigaction_t os_sigaction = NULL;
4604
4605 void os::Linux::check_signal_handler(int sig) {
4606 char buf[O_BUFLEN];
4607 address jvmHandler = NULL;
4608
4609
4610 struct sigaction act;
4611 if (os_sigaction == NULL) {
4612 // only trust the default sigaction, in case it has been interposed
4613 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4614 if (os_sigaction == NULL) return;
4615 }
4616
4617 os_sigaction(sig, (struct sigaction*)NULL, &act);
4618
4619
4620 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4621
4622 address thisHandler = (act.sa_flags & SA_SIGINFO)
4623 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4624 : CAST_FROM_FN_PTR(address, act.sa_handler);
4625
4626
4627 switch (sig) {
4628 case SIGSEGV:
4629 case SIGBUS:
4630 case SIGFPE:
4631 case SIGPIPE:
4632 case SIGILL:
4633 case SIGXFSZ:
4634 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4635 break;
4636
4637 case SHUTDOWN1_SIGNAL:
4638 case SHUTDOWN2_SIGNAL:
4639 case SHUTDOWN3_SIGNAL:
4640 case BREAK_SIGNAL:
4641 jvmHandler = (address)user_handler();
4642 break;
4643
4644 default:
4645 if (sig == SR_signum) {
4646 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4647 } else {
4648 return;
4649 }
4650 break;
4651 }
4652
4653 if (thisHandler != jvmHandler) {
4654 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4655 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4656 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4657 // No need to check this sig any longer
4658 sigaddset(&check_signal_done, sig);
4659 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4660 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4661 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4662 exception_name(sig, buf, O_BUFLEN));
4663 }
4664 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4665 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4666 tty->print("expected:");
4667 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4668 tty->cr();
4669 tty->print(" found:");
4670 os::Posix::print_sa_flags(tty, act.sa_flags);
4671 tty->cr();
4672 // No need to check this sig any longer
4673 sigaddset(&check_signal_done, sig);
4674 }
4675
4676 // Dump all the signal
4677 if (sigismember(&check_signal_done, sig)) {
4678 print_signal_handlers(tty, buf, O_BUFLEN);
4679 }
4680 }
4681
4682 extern void report_error(char* file_name, int line_no, char* title,
4683 char* format, ...);
4684
4685 // this is called _before_ the most of global arguments have been parsed
4686 void os::init(void) {
4687 char dummy; // used to get a guess on initial stack address
4688 // first_hrtime = gethrtime();
4689
4690 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4691
4692 init_random(1234567);
4693
4694 ThreadCritical::initialize();
4695
4696 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4697 if (Linux::page_size() == -1) {
4698 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4699 os::strerror(errno));
4700 }
4701 init_page_sizes((size_t) Linux::page_size());
4702
4703 Linux::initialize_system_info();
4704
4705 Linux::initialize_os_info();
4706
4707 // main_thread points to the aboriginal thread
4708 Linux::_main_thread = pthread_self();
4709
4710 Linux::clock_init();
4711 initial_time_count = javaTimeNanos();
4712
4713 // pthread_condattr initialization for monotonic clock
4714 int status;
4715 pthread_condattr_t* _condattr = os::Linux::condAttr();
4716 if ((status = pthread_condattr_init(_condattr)) != 0) {
4717 fatal("pthread_condattr_init: %s", os::strerror(status));
4718 }
4719 // Only set the clock if CLOCK_MONOTONIC is available
4720 if (os::supports_monotonic_clock()) {
4721 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4722 if (status == EINVAL) {
4723 warning("Unable to use monotonic clock with relative timed-waits" \
4724 " - changes to the time-of-day clock may have adverse affects");
4725 } else {
4726 fatal("pthread_condattr_setclock: %s", os::strerror(status));
4727 }
4728 }
4729 }
4730 // else it defaults to CLOCK_REALTIME
4731
4732 // retrieve entry point for pthread_setname_np
4733 Linux::_pthread_setname_np =
4734 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4735
4736 }
4737
4738 // To install functions for atexit system call
4739 extern "C" {
4740 static void perfMemory_exit_helper() {
4741 perfMemory_exit();
4742 }
4743 }
4744
4745 // this is called _after_ the global arguments have been parsed
4746 jint os::init_2(void) {
4747 Linux::fast_thread_clock_init();
4748
4749 // Allocate a single page and mark it as readable for safepoint polling
4750 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4751 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4752
4753 os::set_polling_page(polling_page);
4754 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4755
4756 if (!UseMembar) {
4757 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4758 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4759 os::set_memory_serialize_page(mem_serialize_page);
4760 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4761 }
4762
4763 // initialize suspend/resume support - must do this before signal_sets_init()
4764 if (SR_initialize() != 0) {
4765 perror("SR_initialize failed");
4766 return JNI_ERR;
4767 }
4768
4769 Linux::signal_sets_init();
4770 Linux::install_signal_handlers();
4771
4772 // Check minimum allowable stack size for thread creation and to initialize
4773 // the java system classes, including StackOverflowError - depends on page
4774 // size. Add a page for compiler2 recursion in main thread.
4775 // Add in 2*BytesPerWord times page size to account for VM stack during
4776 // class initialization depending on 32 or 64 bit VM.
4777 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4778 JavaThread::stack_guard_zone_size() +
4779 JavaThread::stack_shadow_zone_size() +
4780 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4781
4782 size_t threadStackSizeInBytes = ThreadStackSize * K;
4783 if (threadStackSizeInBytes != 0 &&
4784 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4785 tty->print_cr("\nThe stack size specified is too small, "
4786 "Specify at least " SIZE_FORMAT "k",
4787 os::Linux::min_stack_allowed/ K);
4788 return JNI_ERR;
4789 }
4790
4791 // Make the stack size a multiple of the page size so that
4792 // the yellow/red zones can be guarded.
4793 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4794 vm_page_size()));
4795
4796 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4797
4798 #if defined(IA32)
4799 workaround_expand_exec_shield_cs_limit();
4800 #endif
4801
4802 Linux::libpthread_init();
4803 log_info(os)("HotSpot is running with %s, %s",
4804 Linux::glibc_version(), Linux::libpthread_version());
4805
4806 if (UseNUMA) {
4807 if (!Linux::libnuma_init()) {
4808 UseNUMA = false;
4809 } else {
4810 if ((Linux::numa_max_node() < 1)) {
4811 // There's only one node(they start from 0), disable NUMA.
4812 UseNUMA = false;
4813 }
4814 }
4815 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4816 // we can make the adaptive lgrp chunk resizing work. If the user specified
4817 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4818 // disable adaptive resizing.
4819 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4820 if (FLAG_IS_DEFAULT(UseNUMA)) {
4821 UseNUMA = false;
4822 } else {
4823 if (FLAG_IS_DEFAULT(UseLargePages) &&
4824 FLAG_IS_DEFAULT(UseSHM) &&
4825 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4826 UseLargePages = false;
4827 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4828 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4829 UseAdaptiveSizePolicy = false;
4830 UseAdaptiveNUMAChunkSizing = false;
4831 }
4832 }
4833 }
4834 if (!UseNUMA && ForceNUMA) {
4835 UseNUMA = true;
4836 }
4837 }
4838
4839 if (MaxFDLimit) {
4840 // set the number of file descriptors to max. print out error
4841 // if getrlimit/setrlimit fails but continue regardless.
4842 struct rlimit nbr_files;
4843 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4844 if (status != 0) {
4845 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4846 } else {
4847 nbr_files.rlim_cur = nbr_files.rlim_max;
4848 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4849 if (status != 0) {
4850 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4851 }
4852 }
4853 }
4854
4855 // Initialize lock used to serialize thread creation (see os::create_thread)
4856 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4857
4858 // at-exit methods are called in the reverse order of their registration.
4859 // atexit functions are called on return from main or as a result of a
4860 // call to exit(3C). There can be only 32 of these functions registered
4861 // and atexit() does not set errno.
4862
4863 if (PerfAllowAtExitRegistration) {
4864 // only register atexit functions if PerfAllowAtExitRegistration is set.
4865 // atexit functions can be delayed until process exit time, which
4866 // can be problematic for embedded VM situations. Embedded VMs should
4867 // call DestroyJavaVM() to assure that VM resources are released.
4868
4869 // note: perfMemory_exit_helper atexit function may be removed in
4870 // the future if the appropriate cleanup code can be added to the
4871 // VM_Exit VMOperation's doit method.
4872 if (atexit(perfMemory_exit_helper) != 0) {
4873 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4874 }
4875 }
4876
4877 // initialize thread priority policy
4878 prio_init();
4879
4880 return JNI_OK;
4881 }
4882
4883 // Mark the polling page as unreadable
4884 void os::make_polling_page_unreadable(void) {
4885 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4886 fatal("Could not disable polling page");
4887 }
4888 }
4889
4890 // Mark the polling page as readable
4891 void os::make_polling_page_readable(void) {
4892 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4893 fatal("Could not enable polling page");
4894 }
4895 }
4896
4897 // older glibc versions don't have this macro (which expands to
4898 // an optimized bit-counting function) so we have to roll our own
4899 #ifndef CPU_COUNT
4900
4901 static int _cpu_count(const cpu_set_t* cpus) {
4902 int count = 0;
4903 // only look up to the number of configured processors
4904 for (int i = 0; i < os::processor_count(); i++) {
4905 if (CPU_ISSET(i, cpus)) {
4906 count++;
4907 }
4908 }
4909 return count;
4910 }
4911
4912 #define CPU_COUNT(cpus) _cpu_count(cpus)
4913
4914 #endif // CPU_COUNT
4915
4916 // Get the current number of available processors for this process.
4917 // This value can change at any time during a process's lifetime.
4918 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4919 // If it appears there may be more than 1024 processors then we do a
4920 // dynamic check - see 6515172 for details.
4921 // If anything goes wrong we fallback to returning the number of online
4922 // processors - which can be greater than the number available to the process.
4923 int os::active_processor_count() {
4924 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
4925 cpu_set_t* cpus_p = &cpus;
4926 int cpus_size = sizeof(cpu_set_t);
4927
4928 int configured_cpus = processor_count(); // upper bound on available cpus
4929 int cpu_count = 0;
4930
4931 // old build platforms may not support dynamic cpu sets
4932 #ifdef CPU_ALLOC
4933
4934 // To enable easy testing of the dynamic path on different platforms we
4935 // introduce a diagnostic flag: UseCpuAllocPath
4936 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4937 // kernel may use a mask bigger than cpu_set_t
4938 log_trace(os)("active_processor_count: using dynamic path %s"
4939 "- configured processors: %d",
4940 UseCpuAllocPath ? "(forced) " : "",
4941 configured_cpus);
4942 cpus_p = CPU_ALLOC(configured_cpus);
4943 if (cpus_p != NULL) {
4944 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4945 // zero it just to be safe
4946 CPU_ZERO_S(cpus_size, cpus_p);
4947 }
4948 else {
4949 // failed to allocate so fallback to online cpus
4950 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4951 log_trace(os)("active_processor_count: "
4952 "CPU_ALLOC failed (%s) - using "
4953 "online processor count: %d",
4954 os::strerror(errno), online_cpus);
4955 return online_cpus;
4956 }
4957 }
4958 else {
4959 log_trace(os)("active_processor_count: using static path - configured processors: %d",
4960 configured_cpus);
4961 }
4962 #else // CPU_ALLOC
4963 // these stubs won't be executed
4964 #define CPU_COUNT_S(size, cpus) -1
4965 #define CPU_FREE(cpus)
4966
4967 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4968 configured_cpus);
4969 #endif // CPU_ALLOC
4970
4971 // pid 0 means the current thread - which we have to assume represents the process
4972 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4973 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4974 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4975 }
4976 else {
4977 cpu_count = CPU_COUNT(cpus_p);
4978 }
4979 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4980 }
4981 else {
4982 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4983 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4984 "which may exceed available processors", os::strerror(errno), cpu_count);
4985 }
4986
4987 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4988 CPU_FREE(cpus_p);
4989 }
4990
4991 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4992 return cpu_count;
4993 }
4994
4995 void os::set_native_thread_name(const char *name) {
4996 if (Linux::_pthread_setname_np) {
4997 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4998 snprintf(buf, sizeof(buf), "%s", name);
4999 buf[sizeof(buf) - 1] = '\0';
5000 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5001 // ERANGE should not happen; all other errors should just be ignored.
5002 assert(rc != ERANGE, "pthread_setname_np failed");
5003 }
5004 }
5005
5006 bool os::distribute_processes(uint length, uint* distribution) {
5007 // Not yet implemented.
5008 return false;
5009 }
5010
5011 bool os::bind_to_processor(uint processor_id) {
5012 // Not yet implemented.
5013 return false;
5014 }
5015
5016 ///
5017
5018 void os::SuspendedThreadTask::internal_do_task() {
5019 if (do_suspend(_thread->osthread())) {
5020 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5021 do_task(context);
5022 do_resume(_thread->osthread());
5023 }
5024 }
5025
5026 class PcFetcher : public os::SuspendedThreadTask {
5027 public:
5028 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5029 ExtendedPC result();
5030 protected:
5031 void do_task(const os::SuspendedThreadTaskContext& context);
5032 private:
5033 ExtendedPC _epc;
5034 };
5035
5036 ExtendedPC PcFetcher::result() {
5037 guarantee(is_done(), "task is not done yet.");
5038 return _epc;
5039 }
5040
5041 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5042 Thread* thread = context.thread();
5043 OSThread* osthread = thread->osthread();
5044 if (osthread->ucontext() != NULL) {
5045 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5046 } else {
5047 // NULL context is unexpected, double-check this is the VMThread
5048 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5049 }
5050 }
5051
5052 // Suspends the target using the signal mechanism and then grabs the PC before
5053 // resuming the target. Used by the flat-profiler only
5054 ExtendedPC os::get_thread_pc(Thread* thread) {
5055 // Make sure that it is called by the watcher for the VMThread
5056 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5057 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5058
5059 PcFetcher fetcher(thread);
5060 fetcher.run();
5061 return fetcher.result();
5062 }
5063
5064 ////////////////////////////////////////////////////////////////////////////////
5065 // debug support
5066
5067 bool os::find(address addr, outputStream* st) {
5068 Dl_info dlinfo;
5069 memset(&dlinfo, 0, sizeof(dlinfo));
5070 if (dladdr(addr, &dlinfo) != 0) {
5071 st->print(PTR_FORMAT ": ", p2i(addr));
5072 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5073 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5074 p2i(addr) - p2i(dlinfo.dli_saddr));
5075 } else if (dlinfo.dli_fbase != NULL) {
5076 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5077 } else {
5078 st->print("<absolute address>");
5079 }
5080 if (dlinfo.dli_fname != NULL) {
5081 st->print(" in %s", dlinfo.dli_fname);
5082 }
5083 if (dlinfo.dli_fbase != NULL) {
5084 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5085 }
5086 st->cr();
5087
5088 if (Verbose) {
5089 // decode some bytes around the PC
5090 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5091 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5092 address lowest = (address) dlinfo.dli_sname;
5093 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5094 if (begin < lowest) begin = lowest;
5095 Dl_info dlinfo2;
5096 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5097 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5098 end = (address) dlinfo2.dli_saddr;
5099 }
5100 Disassembler::decode(begin, end, st);
5101 }
5102 return true;
5103 }
5104 return false;
5105 }
5106
5107 ////////////////////////////////////////////////////////////////////////////////
5108 // misc
5109
5110 // This does not do anything on Linux. This is basically a hook for being
5111 // able to use structured exception handling (thread-local exception filters)
5112 // on, e.g., Win32.
5113 void
5114 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5115 JavaCallArguments* args, Thread* thread) {
5116 f(value, method, args, thread);
5117 }
5118
5119 void os::print_statistics() {
5120 }
5121
5122 bool os::message_box(const char* title, const char* message) {
5123 int i;
5124 fdStream err(defaultStream::error_fd());
5125 for (i = 0; i < 78; i++) err.print_raw("=");
5126 err.cr();
5127 err.print_raw_cr(title);
5128 for (i = 0; i < 78; i++) err.print_raw("-");
5129 err.cr();
5130 err.print_raw_cr(message);
5131 for (i = 0; i < 78; i++) err.print_raw("=");
5132 err.cr();
5133
5134 char buf[16];
5135 // Prevent process from exiting upon "read error" without consuming all CPU
5136 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5137
5138 return buf[0] == 'y' || buf[0] == 'Y';
5139 }
5140
5141 int os::stat(const char *path, struct stat *sbuf) {
5142 char pathbuf[MAX_PATH];
5143 if (strlen(path) > MAX_PATH - 1) {
5144 errno = ENAMETOOLONG;
5145 return -1;
5146 }
5147 os::native_path(strcpy(pathbuf, path));
5148 return ::stat(pathbuf, sbuf);
5149 }
5150
5151 bool os::check_heap(bool force) {
5152 return true;
5153 }
5154
5155 // Is a (classpath) directory empty?
5156 bool os::dir_is_empty(const char* path) {
5157 DIR *dir = NULL;
5158 struct dirent *ptr;
5159
5160 dir = opendir(path);
5161 if (dir == NULL) return true;
5162
5163 // Scan the directory
5164 bool result = true;
5165 char buf[sizeof(struct dirent) + MAX_PATH];
5166 while (result && (ptr = ::readdir(dir)) != NULL) {
5167 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5168 result = false;
5169 }
5170 }
5171 closedir(dir);
5172 return result;
5173 }
5174
5175 // This code originates from JDK's sysOpen and open64_w
5176 // from src/solaris/hpi/src/system_md.c
5177
5178 int os::open(const char *path, int oflag, int mode) {
5179 if (strlen(path) > MAX_PATH - 1) {
5180 errno = ENAMETOOLONG;
5181 return -1;
5182 }
5183
5184 // All file descriptors that are opened in the Java process and not
5185 // specifically destined for a subprocess should have the close-on-exec
5186 // flag set. If we don't set it, then careless 3rd party native code
5187 // might fork and exec without closing all appropriate file descriptors
5188 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5189 // turn might:
5190 //
5191 // - cause end-of-file to fail to be detected on some file
5192 // descriptors, resulting in mysterious hangs, or
5193 //
5194 // - might cause an fopen in the subprocess to fail on a system
5195 // suffering from bug 1085341.
5196 //
5197 // (Yes, the default setting of the close-on-exec flag is a Unix
5198 // design flaw)
5199 //
5200 // See:
5201 // 1085341: 32-bit stdio routines should support file descriptors >255
5202 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5203 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5204 //
5205 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5206 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5207 // because it saves a system call and removes a small window where the flag
5208 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5209 // and we fall back to using FD_CLOEXEC (see below).
5210 #ifdef O_CLOEXEC
5211 oflag |= O_CLOEXEC;
5212 #endif
5213
5214 int fd = ::open64(path, oflag, mode);
5215 if (fd == -1) return -1;
5216
5217 //If the open succeeded, the file might still be a directory
5218 {
5219 struct stat64 buf64;
5220 int ret = ::fstat64(fd, &buf64);
5221 int st_mode = buf64.st_mode;
5222
5223 if (ret != -1) {
5224 if ((st_mode & S_IFMT) == S_IFDIR) {
5225 errno = EISDIR;
5226 ::close(fd);
5227 return -1;
5228 }
5229 } else {
5230 ::close(fd);
5231 return -1;
5232 }
5233 }
5234
5235 #ifdef FD_CLOEXEC
5236 // Validate that the use of the O_CLOEXEC flag on open above worked.
5237 // With recent kernels, we will perform this check exactly once.
5238 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5239 if (!O_CLOEXEC_is_known_to_work) {
5240 int flags = ::fcntl(fd, F_GETFD);
5241 if (flags != -1) {
5242 if ((flags & FD_CLOEXEC) != 0)
5243 O_CLOEXEC_is_known_to_work = 1;
5244 else
5245 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5246 }
5247 }
5248 #endif
5249
5250 return fd;
5251 }
5252
5253
5254 // create binary file, rewriting existing file if required
5255 int os::create_binary_file(const char* path, bool rewrite_existing) {
5256 int oflags = O_WRONLY | O_CREAT;
5257 if (!rewrite_existing) {
5258 oflags |= O_EXCL;
5259 }
5260 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5261 }
5262
5263 // return current position of file pointer
5264 jlong os::current_file_offset(int fd) {
5265 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5266 }
5267
5268 // move file pointer to the specified offset
5269 jlong os::seek_to_file_offset(int fd, jlong offset) {
5270 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5271 }
5272
5273 // This code originates from JDK's sysAvailable
5274 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5275
5276 int os::available(int fd, jlong *bytes) {
5277 jlong cur, end;
5278 int mode;
5279 struct stat64 buf64;
5280
5281 if (::fstat64(fd, &buf64) >= 0) {
5282 mode = buf64.st_mode;
5283 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5284 int n;
5285 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5286 *bytes = n;
5287 return 1;
5288 }
5289 }
5290 }
5291 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5292 return 0;
5293 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5294 return 0;
5295 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5296 return 0;
5297 }
5298 *bytes = end - cur;
5299 return 1;
5300 }
5301
5302 // Map a block of memory.
5303 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5304 char *addr, size_t bytes, bool read_only,
5305 bool allow_exec) {
5306 int prot;
5307 int flags = MAP_PRIVATE;
5308
5309 if (read_only) {
5310 prot = PROT_READ;
5311 } else {
5312 prot = PROT_READ | PROT_WRITE;
5313 }
5314
5315 if (allow_exec) {
5316 prot |= PROT_EXEC;
5317 }
5318
5319 if (addr != NULL) {
5320 flags |= MAP_FIXED;
5321 }
5322
5323 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5324 fd, file_offset);
5325 if (mapped_address == MAP_FAILED) {
5326 return NULL;
5327 }
5328 return mapped_address;
5329 }
5330
5331
5332 // Remap a block of memory.
5333 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5334 char *addr, size_t bytes, bool read_only,
5335 bool allow_exec) {
5336 // same as map_memory() on this OS
5337 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5338 allow_exec);
5339 }
5340
5341
5342 // Unmap a block of memory.
5343 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5344 return munmap(addr, bytes) == 0;
5345 }
5346
5347 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5348
5349 static clockid_t thread_cpu_clockid(Thread* thread) {
5350 pthread_t tid = thread->osthread()->pthread_id();
5351 clockid_t clockid;
5352
5353 // Get thread clockid
5354 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5355 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5356 return clockid;
5357 }
5358
5359 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5360 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5361 // of a thread.
5362 //
5363 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5364 // the fast estimate available on the platform.
5365
5366 jlong os::current_thread_cpu_time() {
5367 if (os::Linux::supports_fast_thread_cpu_time()) {
5368 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5369 } else {
5370 // return user + sys since the cost is the same
5371 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5372 }
5373 }
5374
5375 jlong os::thread_cpu_time(Thread* thread) {
5376 // consistent with what current_thread_cpu_time() returns
5377 if (os::Linux::supports_fast_thread_cpu_time()) {
5378 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5379 } else {
5380 return slow_thread_cpu_time(thread, true /* user + sys */);
5381 }
5382 }
5383
5384 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5385 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5386 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5387 } else {
5388 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5389 }
5390 }
5391
5392 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5393 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5394 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5395 } else {
5396 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5397 }
5398 }
5399
5400 // -1 on error.
5401 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5402 pid_t tid = thread->osthread()->thread_id();
5403 char *s;
5404 char stat[2048];
5405 int statlen;
5406 char proc_name[64];
5407 int count;
5408 long sys_time, user_time;
5409 char cdummy;
5410 int idummy;
5411 long ldummy;
5412 FILE *fp;
5413
5414 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5415 fp = fopen(proc_name, "r");
5416 if (fp == NULL) return -1;
5417 statlen = fread(stat, 1, 2047, fp);
5418 stat[statlen] = '\0';
5419 fclose(fp);
5420
5421 // Skip pid and the command string. Note that we could be dealing with
5422 // weird command names, e.g. user could decide to rename java launcher
5423 // to "java 1.4.2 :)", then the stat file would look like
5424 // 1234 (java 1.4.2 :)) R ... ...
5425 // We don't really need to know the command string, just find the last
5426 // occurrence of ")" and then start parsing from there. See bug 4726580.
5427 s = strrchr(stat, ')');
5428 if (s == NULL) return -1;
5429
5430 // Skip blank chars
5431 do { s++; } while (s && isspace(*s));
5432
5433 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5434 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5435 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5436 &user_time, &sys_time);
5437 if (count != 13) return -1;
5438 if (user_sys_cpu_time) {
5439 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5440 } else {
5441 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5442 }
5443 }
5444
5445 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5446 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5447 info_ptr->may_skip_backward = false; // elapsed time not wall time
5448 info_ptr->may_skip_forward = false; // elapsed time not wall time
5449 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5450 }
5451
5452 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5453 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5454 info_ptr->may_skip_backward = false; // elapsed time not wall time
5455 info_ptr->may_skip_forward = false; // elapsed time not wall time
5456 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5457 }
5458
5459 bool os::is_thread_cpu_time_supported() {
5460 return true;
5461 }
5462
5463 // System loadavg support. Returns -1 if load average cannot be obtained.
5464 // Linux doesn't yet have a (official) notion of processor sets,
5465 // so just return the system wide load average.
5466 int os::loadavg(double loadavg[], int nelem) {
5467 return ::getloadavg(loadavg, nelem);
5468 }
5469
5470 void os::pause() {
5471 char filename[MAX_PATH];
5472 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5473 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5474 } else {
5475 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5476 }
5477
5478 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5479 if (fd != -1) {
5480 struct stat buf;
5481 ::close(fd);
5482 while (::stat(filename, &buf) == 0) {
5483 (void)::poll(NULL, 0, 100);
5484 }
5485 } else {
5486 jio_fprintf(stderr,
5487 "Could not open pause file '%s', continuing immediately.\n", filename);
5488 }
5489 }
5490
5491
5492 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5493 // comment paragraphs are worth repeating here:
5494 //
5495 // Assumption:
5496 // Only one parker can exist on an event, which is why we allocate
5497 // them per-thread. Multiple unparkers can coexist.
5498 //
5499 // _Event serves as a restricted-range semaphore.
5500 // -1 : thread is blocked, i.e. there is a waiter
5501 // 0 : neutral: thread is running or ready,
5502 // could have been signaled after a wait started
5503 // 1 : signaled - thread is running or ready
5504 //
5505
5506 // utility to compute the abstime argument to timedwait:
5507 // millis is the relative timeout time
5508 // abstime will be the absolute timeout time
5509 // TODO: replace compute_abstime() with unpackTime()
5510
5511 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5512 if (millis < 0) millis = 0;
5513
5514 jlong seconds = millis / 1000;
5515 millis %= 1000;
5516 if (seconds > 50000000) { // see man cond_timedwait(3T)
5517 seconds = 50000000;
5518 }
5519
5520 if (os::supports_monotonic_clock()) {
5521 struct timespec now;
5522 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5523 assert_status(status == 0, status, "clock_gettime");
5524 abstime->tv_sec = now.tv_sec + seconds;
5525 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5526 if (nanos >= NANOSECS_PER_SEC) {
5527 abstime->tv_sec += 1;
5528 nanos -= NANOSECS_PER_SEC;
5529 }
5530 abstime->tv_nsec = nanos;
5531 } else {
5532 struct timeval now;
5533 int status = gettimeofday(&now, NULL);
5534 assert(status == 0, "gettimeofday");
5535 abstime->tv_sec = now.tv_sec + seconds;
5536 long usec = now.tv_usec + millis * 1000;
5537 if (usec >= 1000000) {
5538 abstime->tv_sec += 1;
5539 usec -= 1000000;
5540 }
5541 abstime->tv_nsec = usec * 1000;
5542 }
5543 return abstime;
5544 }
5545
5546 void os::PlatformEvent::park() { // AKA "down()"
5547 // Transitions for _Event:
5548 // -1 => -1 : illegal
5549 // 1 => 0 : pass - return immediately
5550 // 0 => -1 : block; then set _Event to 0 before returning
5551
5552 // Invariant: Only the thread associated with the Event/PlatformEvent
5553 // may call park().
5554 // TODO: assert that _Assoc != NULL or _Assoc == Self
5555 assert(_nParked == 0, "invariant");
5556
5557 int v;
5558 for (;;) {
5559 v = _Event;
5560 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5561 }
5562 guarantee(v >= 0, "invariant");
5563 if (v == 0) {
5564 // Do this the hard way by blocking ...
5565 int status = pthread_mutex_lock(_mutex);
5566 assert_status(status == 0, status, "mutex_lock");
5567 guarantee(_nParked == 0, "invariant");
5568 ++_nParked;
5569 while (_Event < 0) {
5570 status = pthread_cond_wait(_cond, _mutex);
5571 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5572 // Treat this the same as if the wait was interrupted
5573 if (status == ETIME) { status = EINTR; }
5574 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5575 }
5576 --_nParked;
5577
5578 _Event = 0;
5579 status = pthread_mutex_unlock(_mutex);
5580 assert_status(status == 0, status, "mutex_unlock");
5581 // Paranoia to ensure our locked and lock-free paths interact
5582 // correctly with each other.
5583 OrderAccess::fence();
5584 }
5585 guarantee(_Event >= 0, "invariant");
5586 }
5587
5588 int os::PlatformEvent::park(jlong millis) {
5589 // Transitions for _Event:
5590 // -1 => -1 : illegal
5591 // 1 => 0 : pass - return immediately
5592 // 0 => -1 : block; then set _Event to 0 before returning
5593
5594 guarantee(_nParked == 0, "invariant");
5595
5596 int v;
5597 for (;;) {
5598 v = _Event;
5599 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5600 }
5601 guarantee(v >= 0, "invariant");
5602 if (v != 0) return OS_OK;
5603
5604 // We do this the hard way, by blocking the thread.
5605 // Consider enforcing a minimum timeout value.
5606 struct timespec abst;
5607 compute_abstime(&abst, millis);
5608
5609 int ret = OS_TIMEOUT;
5610 int status = pthread_mutex_lock(_mutex);
5611 assert_status(status == 0, status, "mutex_lock");
5612 guarantee(_nParked == 0, "invariant");
5613 ++_nParked;
5614
5615 // Object.wait(timo) will return because of
5616 // (a) notification
5617 // (b) timeout
5618 // (c) thread.interrupt
5619 //
5620 // Thread.interrupt and object.notify{All} both call Event::set.
5621 // That is, we treat thread.interrupt as a special case of notification.
5622 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5623 // We assume all ETIME returns are valid.
5624 //
5625 // TODO: properly differentiate simultaneous notify+interrupt.
5626 // In that case, we should propagate the notify to another waiter.
5627
5628 while (_Event < 0) {
5629 status = pthread_cond_timedwait(_cond, _mutex, &abst);
5630 assert_status(status == 0 || status == EINTR ||
5631 status == ETIME || status == ETIMEDOUT,
5632 status, "cond_timedwait");
5633 if (!FilterSpuriousWakeups) break; // previous semantics
5634 if (status == ETIME || status == ETIMEDOUT) break;
5635 // We consume and ignore EINTR and spurious wakeups.
5636 }
5637 --_nParked;
5638 if (_Event >= 0) {
5639 ret = OS_OK;
5640 }
5641 _Event = 0;
5642 status = pthread_mutex_unlock(_mutex);
5643 assert_status(status == 0, status, "mutex_unlock");
5644 assert(_nParked == 0, "invariant");
5645 // Paranoia to ensure our locked and lock-free paths interact
5646 // correctly with each other.
5647 OrderAccess::fence();
5648 return ret;
5649 }
5650
5651 void os::PlatformEvent::unpark() {
5652 // Transitions for _Event:
5653 // 0 => 1 : just return
5654 // 1 => 1 : just return
5655 // -1 => either 0 or 1; must signal target thread
5656 // That is, we can safely transition _Event from -1 to either
5657 // 0 or 1.
5658 // See also: "Semaphores in Plan 9" by Mullender & Cox
5659 //
5660 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5661 // that it will take two back-to-back park() calls for the owning
5662 // thread to block. This has the benefit of forcing a spurious return
5663 // from the first park() call after an unpark() call which will help
5664 // shake out uses of park() and unpark() without condition variables.
5665
5666 if (Atomic::xchg(1, &_Event) >= 0) return;
5667
5668 // Wait for the thread associated with the event to vacate
5669 int status = pthread_mutex_lock(_mutex);
5670 assert_status(status == 0, status, "mutex_lock");
5671 int AnyWaiters = _nParked;
5672 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5673 status = pthread_mutex_unlock(_mutex);
5674 assert_status(status == 0, status, "mutex_unlock");
5675 if (AnyWaiters != 0) {
5676 // Note that we signal() *after* dropping the lock for "immortal" Events.
5677 // This is safe and avoids a common class of futile wakeups. In rare
5678 // circumstances this can cause a thread to return prematurely from
5679 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5680 // will simply re-test the condition and re-park itself.
5681 // This provides particular benefit if the underlying platform does not
5682 // provide wait morphing.
5683 status = pthread_cond_signal(_cond);
5684 assert_status(status == 0, status, "cond_signal");
5685 }
5686 }
5687
5688
5689 // JSR166
5690 // -------------------------------------------------------
5691
5692 // The solaris and linux implementations of park/unpark are fairly
5693 // conservative for now, but can be improved. They currently use a
5694 // mutex/condvar pair, plus a a count.
5695 // Park decrements count if > 0, else does a condvar wait. Unpark
5696 // sets count to 1 and signals condvar. Only one thread ever waits
5697 // on the condvar. Contention seen when trying to park implies that someone
5698 // is unparking you, so don't wait. And spurious returns are fine, so there
5699 // is no need to track notifications.
5700
5701 // This code is common to linux and solaris and will be moved to a
5702 // common place in dolphin.
5703 //
5704 // The passed in time value is either a relative time in nanoseconds
5705 // or an absolute time in milliseconds. Either way it has to be unpacked
5706 // into suitable seconds and nanoseconds components and stored in the
5707 // given timespec structure.
5708 // Given time is a 64-bit value and the time_t used in the timespec is only
5709 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5710 // overflow if times way in the future are given. Further on Solaris versions
5711 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5712 // number of seconds, in abstime, is less than current_time + 100,000,000.
5713 // As it will be 28 years before "now + 100000000" will overflow we can
5714 // ignore overflow and just impose a hard-limit on seconds using the value
5715 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5716 // years from "now".
5717
5718 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5719 assert(time > 0, "convertTime");
5720 time_t max_secs = 0;
5721
5722 if (!os::supports_monotonic_clock() || isAbsolute) {
5723 struct timeval now;
5724 int status = gettimeofday(&now, NULL);
5725 assert(status == 0, "gettimeofday");
5726
5727 max_secs = now.tv_sec + MAX_SECS;
5728
5729 if (isAbsolute) {
5730 jlong secs = time / 1000;
5731 if (secs > max_secs) {
5732 absTime->tv_sec = max_secs;
5733 } else {
5734 absTime->tv_sec = secs;
5735 }
5736 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5737 } else {
5738 jlong secs = time / NANOSECS_PER_SEC;
5739 if (secs >= MAX_SECS) {
5740 absTime->tv_sec = max_secs;
5741 absTime->tv_nsec = 0;
5742 } else {
5743 absTime->tv_sec = now.tv_sec + secs;
5744 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5745 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5746 absTime->tv_nsec -= NANOSECS_PER_SEC;
5747 ++absTime->tv_sec; // note: this must be <= max_secs
5748 }
5749 }
5750 }
5751 } else {
5752 // must be relative using monotonic clock
5753 struct timespec now;
5754 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5755 assert_status(status == 0, status, "clock_gettime");
5756 max_secs = now.tv_sec + MAX_SECS;
5757 jlong secs = time / NANOSECS_PER_SEC;
5758 if (secs >= MAX_SECS) {
5759 absTime->tv_sec = max_secs;
5760 absTime->tv_nsec = 0;
5761 } else {
5762 absTime->tv_sec = now.tv_sec + secs;
5763 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5764 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5765 absTime->tv_nsec -= NANOSECS_PER_SEC;
5766 ++absTime->tv_sec; // note: this must be <= max_secs
5767 }
5768 }
5769 }
5770 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5771 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5772 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5773 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5774 }
5775
5776 void Parker::park(bool isAbsolute, jlong time) {
5777 // Ideally we'd do something useful while spinning, such
5778 // as calling unpackTime().
5779
5780 // Optional fast-path check:
5781 // Return immediately if a permit is available.
5782 // We depend on Atomic::xchg() having full barrier semantics
5783 // since we are doing a lock-free update to _counter.
5784 if (Atomic::xchg(0, &_counter) > 0) return;
5785
5786 Thread* thread = Thread::current();
5787 assert(thread->is_Java_thread(), "Must be JavaThread");
5788 JavaThread *jt = (JavaThread *)thread;
5789
5790 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5791 // Check interrupt before trying to wait
5792 if (Thread::is_interrupted(thread, false)) {
5793 return;
5794 }
5795
5796 // Next, demultiplex/decode time arguments
5797 timespec absTime;
5798 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5799 return;
5800 }
5801 if (time > 0) {
5802 unpackTime(&absTime, isAbsolute, time);
5803 }
5804
5805
5806 // Enter safepoint region
5807 // Beware of deadlocks such as 6317397.
5808 // The per-thread Parker:: mutex is a classic leaf-lock.
5809 // In particular a thread must never block on the Threads_lock while
5810 // holding the Parker:: mutex. If safepoints are pending both the
5811 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5812 ThreadBlockInVM tbivm(jt);
5813
5814 // Don't wait if cannot get lock since interference arises from
5815 // unblocking. Also. check interrupt before trying wait
5816 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5817 return;
5818 }
5819
5820 int status;
5821 if (_counter > 0) { // no wait needed
5822 _counter = 0;
5823 status = pthread_mutex_unlock(_mutex);
5824 assert_status(status == 0, status, "invariant");
5825 // Paranoia to ensure our locked and lock-free paths interact
5826 // correctly with each other and Java-level accesses.
5827 OrderAccess::fence();
5828 return;
5829 }
5830
5831 #ifdef ASSERT
5832 // Don't catch signals while blocked; let the running threads have the signals.
5833 // (This allows a debugger to break into the running thread.)
5834 sigset_t oldsigs;
5835 sigemptyset(&oldsigs);
5836 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5837 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5838 #endif
5839
5840 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5841 jt->set_suspend_equivalent();
5842 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5843
5844 assert(_cur_index == -1, "invariant");
5845 if (time == 0) {
5846 _cur_index = REL_INDEX; // arbitrary choice when not timed
5847 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5848 } else {
5849 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5850 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5851 }
5852 _cur_index = -1;
5853 assert_status(status == 0 || status == EINTR ||
5854 status == ETIME || status == ETIMEDOUT,
5855 status, "cond_timedwait");
5856
5857 #ifdef ASSERT
5858 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5859 #endif
5860
5861 _counter = 0;
5862 status = pthread_mutex_unlock(_mutex);
5863 assert_status(status == 0, status, "invariant");
5864 // Paranoia to ensure our locked and lock-free paths interact
5865 // correctly with each other and Java-level accesses.
5866 OrderAccess::fence();
5867
5868 // If externally suspended while waiting, re-suspend
5869 if (jt->handle_special_suspend_equivalent_condition()) {
5870 jt->java_suspend_self();
5871 }
5872 }
5873
5874 void Parker::unpark() {
5875 int status = pthread_mutex_lock(_mutex);
5876 assert_status(status == 0, status, "invariant");
5877 const int s = _counter;
5878 _counter = 1;
5879 // must capture correct index before unlocking
5880 int index = _cur_index;
5881 status = pthread_mutex_unlock(_mutex);
5882 assert_status(status == 0, status, "invariant");
5883 if (s < 1 && index != -1) {
5884 // thread is definitely parked
5885 status = pthread_cond_signal(&_cond[index]);
5886 assert_status(status == 0, status, "invariant");
5887 }
5888 }
5889
5890
5891 extern char** environ;
5892
5893 // Run the specified command in a separate process. Return its exit value,
5894 // or -1 on failure (e.g. can't fork a new process).
5895 // Unlike system(), this function can be called from signal handler. It
5896 // doesn't block SIGINT et al.
5897 int os::fork_and_exec(char* cmd) {
5898 const char * argv[4] = {"sh", "-c", cmd, NULL};
5899
5900 pid_t pid = fork();
5901
5902 if (pid < 0) {
5903 // fork failed
5904 return -1;
5905
5906 } else if (pid == 0) {
5907 // child process
5908
5909 execve("/bin/sh", (char* const*)argv, environ);
5910
5911 // execve failed
5912 _exit(-1);
5913
5914 } else {
5915 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5916 // care about the actual exit code, for now.
5917
5918 int status;
5919
5920 // Wait for the child process to exit. This returns immediately if
5921 // the child has already exited. */
5922 while (waitpid(pid, &status, 0) < 0) {
5923 switch (errno) {
5924 case ECHILD: return 0;
5925 case EINTR: break;
5926 default: return -1;
5927 }
5928 }
5929
5930 if (WIFEXITED(status)) {
5931 // The child exited normally; get its exit code.
5932 return WEXITSTATUS(status);
5933 } else if (WIFSIGNALED(status)) {
5934 // The child exited because of a signal
5935 // The best value to return is 0x80 + signal number,
5936 // because that is what all Unix shells do, and because
5937 // it allows callers to distinguish between process exit and
5938 // process death by signal.
5939 return 0x80 + WTERMSIG(status);
5940 } else {
5941 // Unknown exit code; pass it through
5942 return status;
5943 }
5944 }
5945 }
5946
5947 // is_headless_jre()
5948 //
5949 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5950 // in order to report if we are running in a headless jre
5951 //
5952 // Since JDK8 xawt/libmawt.so was moved into the same directory
5953 // as libawt.so, and renamed libawt_xawt.so
5954 //
5955 bool os::is_headless_jre() {
5956 struct stat statbuf;
5957 char buf[MAXPATHLEN];
5958 char libmawtpath[MAXPATHLEN];
5959 const char *xawtstr = "/xawt/libmawt.so";
5960 const char *new_xawtstr = "/libawt_xawt.so";
5961 char *p;
5962
5963 // Get path to libjvm.so
5964 os::jvm_path(buf, sizeof(buf));
5965
5966 // Get rid of libjvm.so
5967 p = strrchr(buf, '/');
5968 if (p == NULL) {
5969 return false;
5970 } else {
5971 *p = '\0';
5972 }
5973
5974 // Get rid of client or server
5975 p = strrchr(buf, '/');
5976 if (p == NULL) {
5977 return false;
5978 } else {
5979 *p = '\0';
5980 }
5981
5982 // check xawt/libmawt.so
5983 strcpy(libmawtpath, buf);
5984 strcat(libmawtpath, xawtstr);
5985 if (::stat(libmawtpath, &statbuf) == 0) return false;
5986
5987 // check libawt_xawt.so
5988 strcpy(libmawtpath, buf);
5989 strcat(libmawtpath, new_xawtstr);
5990 if (::stat(libmawtpath, &statbuf) == 0) return false;
5991
5992 return true;
5993 }
5994
5995 // Get the default path to the core file
5996 // Returns the length of the string
5997 int os::get_core_path(char* buffer, size_t bufferSize) {
5998 /*
5999 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6000 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6001 */
6002 const int core_pattern_len = 129;
6003 char core_pattern[core_pattern_len] = {0};
6004
6005 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6006 if (core_pattern_file == -1) {
6007 return -1;
6008 }
6009
6010 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6011 ::close(core_pattern_file);
6012 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6013 return -1;
6014 }
6015 if (core_pattern[ret-1] == '\n') {
6016 core_pattern[ret-1] = '\0';
6017 } else {
6018 core_pattern[ret] = '\0';
6019 }
6020
6021 char *pid_pos = strstr(core_pattern, "%p");
6022 int written;
6023
6024 if (core_pattern[0] == '/') {
6025 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6026 } else {
6027 char cwd[PATH_MAX];
6028
6029 const char* p = get_current_directory(cwd, PATH_MAX);
6030 if (p == NULL) {
6031 return -1;
6032 }
6033
6034 if (core_pattern[0] == '|') {
6035 written = jio_snprintf(buffer, bufferSize,
6036 "\"%s\" (or dumping to %s/core.%d)",
6037 &core_pattern[1], p, current_process_id());
6038 } else {
6039 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6040 }
6041 }
6042
6043 if (written < 0) {
6044 return -1;
6045 }
6046
6047 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6048 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6049
6050 if (core_uses_pid_file != -1) {
6051 char core_uses_pid = 0;
6052 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6053 ::close(core_uses_pid_file);
6054
6055 if (core_uses_pid == '1') {
6056 jio_snprintf(buffer + written, bufferSize - written,
6057 ".%d", current_process_id());
6058 }
6059 }
6060 }
6061
6062 return strlen(buffer);
6063 }
6064
6065 bool os::start_debugging(char *buf, int buflen) {
6066 int len = (int)strlen(buf);
6067 char *p = &buf[len];
6068
6069 jio_snprintf(p, buflen-len,
6070 "\n\n"
6071 "Do you want to debug the problem?\n\n"
6072 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6073 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6074 "Otherwise, press RETURN to abort...",
6075 os::current_process_id(), os::current_process_id(),
6076 os::current_thread_id(), os::current_thread_id());
6077
6078 bool yes = os::message_box("Unexpected Error", buf);
6079
6080 if (yes) {
6081 // yes, user asked VM to launch debugger
6082 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
6083 os::current_process_id(), os::current_process_id());
6084
6085 os::fork_and_exec(buf);
6086 yes = false;
6087 }
6088 return yes;
6089 }
6090
6091 static inline struct timespec get_mtime(const char* filename) {
6092 struct stat st;
6093 int ret = os::stat(filename, &st);
6094 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6095 return st.st_mtim;
6096 }
6097
6098 int os::compare_file_modified_times(const char* file1, const char* file2) {
6099 struct timespec filetime1 = get_mtime(file1);
6100 struct timespec filetime2 = get_mtime(file2);
6101 int diff = filetime1.tv_sec - filetime2.tv_sec;
6102 if (diff == 0) {
6103 return filetime1.tv_nsec - filetime2.tv_nsec;
6104 }
6105 return diff;
6106 }
6107
6108 /////////////// Unit tests ///////////////
6109
6110 #ifndef PRODUCT
6111
6112 #define test_log(...) \
6113 do { \
6114 if (VerboseInternalVMTests) { \
6115 tty->print_cr(__VA_ARGS__); \
6116 tty->flush(); \
6117 } \
6118 } while (false)
6119
6120 class TestReserveMemorySpecial : AllStatic {
6121 public:
6122 static void small_page_write(void* addr, size_t size) {
6123 size_t page_size = os::vm_page_size();
6124
6125 char* end = (char*)addr + size;
6126 for (char* p = (char*)addr; p < end; p += page_size) {
6127 *p = 1;
6128 }
6129 }
6130
6131 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6132 if (!UseHugeTLBFS) {
6133 return;
6134 }
6135
6136 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6137
6138 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6139
6140 if (addr != NULL) {
6141 small_page_write(addr, size);
6142
6143 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6144 }
6145 }
6146
6147 static void test_reserve_memory_special_huge_tlbfs_only() {
6148 if (!UseHugeTLBFS) {
6149 return;
6150 }
6151
6152 size_t lp = os::large_page_size();
6153
6154 for (size_t size = lp; size <= lp * 10; size += lp) {
6155 test_reserve_memory_special_huge_tlbfs_only(size);
6156 }
6157 }
6158
6159 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6160 size_t lp = os::large_page_size();
6161 size_t ag = os::vm_allocation_granularity();
6162
6163 // sizes to test
6164 const size_t sizes[] = {
6165 lp, lp + ag, lp + lp / 2, lp * 2,
6166 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6167 lp * 10, lp * 10 + lp / 2
6168 };
6169 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6170
6171 // For each size/alignment combination, we test three scenarios:
6172 // 1) with req_addr == NULL
6173 // 2) with a non-null req_addr at which we expect to successfully allocate
6174 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6175 // expect the allocation to either fail or to ignore req_addr
6176
6177 // Pre-allocate two areas; they shall be as large as the largest allocation
6178 // and aligned to the largest alignment we will be testing.
6179 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6180 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6181 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6182 -1, 0);
6183 assert(mapping1 != MAP_FAILED, "should work");
6184
6185 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6186 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6187 -1, 0);
6188 assert(mapping2 != MAP_FAILED, "should work");
6189
6190 // Unmap the first mapping, but leave the second mapping intact: the first
6191 // mapping will serve as a value for a "good" req_addr (case 2). The second
6192 // mapping, still intact, as "bad" req_addr (case 3).
6193 ::munmap(mapping1, mapping_size);
6194
6195 // Case 1
6196 test_log("%s, req_addr NULL:", __FUNCTION__);
6197 test_log("size align result");
6198
6199 for (int i = 0; i < num_sizes; i++) {
6200 const size_t size = sizes[i];
6201 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6202 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6203 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6204 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6205 if (p != NULL) {
6206 assert(is_ptr_aligned(p, alignment), "must be");
6207 small_page_write(p, size);
6208 os::Linux::release_memory_special_huge_tlbfs(p, size);
6209 }
6210 }
6211 }
6212
6213 // Case 2
6214 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6215 test_log("size align req_addr result");
6216
6217 for (int i = 0; i < num_sizes; i++) {
6218 const size_t size = sizes[i];
6219 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6220 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6221 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6222 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6223 size, alignment, p2i(req_addr), p2i(p),
6224 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6225 if (p != NULL) {
6226 assert(p == req_addr, "must be");
6227 small_page_write(p, size);
6228 os::Linux::release_memory_special_huge_tlbfs(p, size);
6229 }
6230 }
6231 }
6232
6233 // Case 3
6234 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6235 test_log("size align req_addr result");
6236
6237 for (int i = 0; i < num_sizes; i++) {
6238 const size_t size = sizes[i];
6239 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6240 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6241 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6242 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6243 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6244 // as the area around req_addr contains already existing mappings, the API should always
6245 // return NULL (as per contract, it cannot return another address)
6246 assert(p == NULL, "must be");
6247 }
6248 }
6249
6250 ::munmap(mapping2, mapping_size);
6251
6252 }
6253
6254 static void test_reserve_memory_special_huge_tlbfs() {
6255 if (!UseHugeTLBFS) {
6256 return;
6257 }
6258
6259 test_reserve_memory_special_huge_tlbfs_only();
6260 test_reserve_memory_special_huge_tlbfs_mixed();
6261 }
6262
6263 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6264 if (!UseSHM) {
6265 return;
6266 }
6267
6268 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6269
6270 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6271
6272 if (addr != NULL) {
6273 assert(is_ptr_aligned(addr, alignment), "Check");
6274 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6275
6276 small_page_write(addr, size);
6277
6278 os::Linux::release_memory_special_shm(addr, size);
6279 }
6280 }
6281
6282 static void test_reserve_memory_special_shm() {
6283 size_t lp = os::large_page_size();
6284 size_t ag = os::vm_allocation_granularity();
6285
6286 for (size_t size = ag; size < lp * 3; size += ag) {
6287 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6288 test_reserve_memory_special_shm(size, alignment);
6289 }
6290 }
6291 }
6292
6293 static void test() {
6294 test_reserve_memory_special_huge_tlbfs();
6295 test_reserve_memory_special_shm();
6296 }
6297 };
6298
6299 void TestReserveMemorySpecial_test() {
6300 TestReserveMemorySpecial::test();
6301 }
6302
6303 #endif