1 /*
2 * Copyright (c) 1999, 2018, 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 "jvm.h"
27 #include "classfile/classLoader.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "code/icBuffer.hpp"
31 #include "code/vtableStubs.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/disassembler.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "logging/log.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "osContainer_linux.hpp"
42 #include "prims/jniFastGetField.hpp"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/threadSMR.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/align.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/vmError.hpp"
76
77 // put OS-includes here
78 # include <sys/types.h>
79 # include <sys/mman.h>
80 # include <sys/stat.h>
81 # include <sys/select.h>
82 # include <pthread.h>
83 # include <signal.h>
84 # include <errno.h>
85 # include <dlfcn.h>
86 # include <stdio.h>
87 # include <unistd.h>
88 # include <sys/resource.h>
89 # include <pthread.h>
90 # include <sys/stat.h>
91 # include <sys/time.h>
92 # include <sys/times.h>
93 # include <sys/utsname.h>
94 # include <sys/socket.h>
95 # include <sys/wait.h>
96 # include <pwd.h>
97 # include <poll.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 #define DAX_SHARED_BIT (1 << 8)
133 ////////////////////////////////////////////////////////////////////////////////
134 // global variables
135 julong os::Linux::_physical_memory = 0;
136
137 address os::Linux::_initial_thread_stack_bottom = NULL;
138 uintptr_t os::Linux::_initial_thread_stack_size = 0;
139
140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
143 Mutex* os::Linux::_createThread_lock = NULL;
144 pthread_t os::Linux::_main_thread;
145 int os::Linux::_page_size = -1;
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
151 static jlong initial_time_count=0;
152
153 static int clock_tics_per_sec = 100;
154
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
158
159 // Signal number used to suspend/resume a thread
160
161 // do not use any signal number less than SIGSEGV, see 4355769
162 static int SR_signum = SIGUSR2;
163 sigset_t SR_sigset;
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 julong avail_mem;
177
178 if (OSContainer::is_containerized()) {
179 jlong mem_limit, mem_usage;
180 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
181 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
182 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
183 }
184 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
185 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
186 }
187 if (mem_limit > 0 && mem_usage > 0 ) {
188 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
189 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
190 return avail_mem;
191 }
192 }
193
194 sysinfo(&si);
195 avail_mem = (julong)si.freeram * si.mem_unit;
196 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
197 return avail_mem;
198 }
199
200 julong os::physical_memory() {
201 jlong phys_mem = 0;
202 if (OSContainer::is_containerized()) {
203 jlong mem_limit;
204 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
205 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
206 return mem_limit;
207 }
208 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
209 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
210 }
211
212 phys_mem = Linux::physical_memory();
213 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
214 return phys_mem;
215 }
216
217 // Return true if user is running as root.
218
219 bool os::have_special_privileges() {
220 static bool init = false;
221 static bool privileges = false;
222 if (!init) {
223 privileges = (getuid() != geteuid()) || (getgid() != getegid());
224 init = true;
225 }
226 return privileges;
227 }
228
229
230 #ifndef SYS_gettid
231 // i386: 224, ia64: 1105, amd64: 186, sparc 143
232 #ifdef __ia64__
233 #define SYS_gettid 1105
234 #else
235 #ifdef __i386__
236 #define SYS_gettid 224
237 #else
238 #ifdef __amd64__
239 #define SYS_gettid 186
240 #else
241 #ifdef __sparc__
242 #define SYS_gettid 143
243 #else
244 #error define gettid for the arch
245 #endif
246 #endif
247 #endif
248 #endif
249 #endif
250
251
252 // pid_t gettid()
253 //
254 // Returns the kernel thread id of the currently running thread. Kernel
255 // thread id is used to access /proc.
256 pid_t os::Linux::gettid() {
257 int rslt = syscall(SYS_gettid);
258 assert(rslt != -1, "must be."); // old linuxthreads implementation?
259 return (pid_t)rslt;
260 }
261
262 // Most versions of linux have a bug where the number of processors are
263 // determined by looking at the /proc file system. In a chroot environment,
264 // the system call returns 1. This causes the VM to act as if it is
265 // a single processor and elide locking (see is_MP() call).
266 static bool unsafe_chroot_detected = false;
267 static const char *unstable_chroot_error = "/proc file system not found.\n"
268 "Java may be unstable running multithreaded in a chroot "
269 "environment on Linux when /proc filesystem is not mounted.";
270
271 void os::Linux::initialize_system_info() {
272 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
273 if (processor_count() == 1) {
274 pid_t pid = os::Linux::gettid();
275 char fname[32];
276 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
277 FILE *fp = fopen(fname, "r");
278 if (fp == NULL) {
279 unsafe_chroot_detected = true;
280 } else {
281 fclose(fp);
282 }
283 }
284 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
285 assert(processor_count() > 0, "linux error");
286 }
287
288 void os::init_system_properties_values() {
289 // The next steps are taken in the product version:
290 //
291 // Obtain the JAVA_HOME value from the location of libjvm.so.
292 // This library should be located at:
293 // <JAVA_HOME>/lib/{client|server}/libjvm.so.
294 //
295 // If "/jre/lib/" appears at the right place in the path, then we
296 // assume libjvm.so is installed in a JDK and we use this path.
297 //
298 // Otherwise exit with message: "Could not create the Java virtual machine."
299 //
300 // The following extra steps are taken in the debugging version:
301 //
302 // If "/jre/lib/" does NOT appear at the right place in the path
303 // instead of exit check for $JAVA_HOME environment variable.
304 //
305 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
306 // then we append a fake suffix "hotspot/libjvm.so" to this path so
307 // it looks like libjvm.so is installed there
308 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
309 //
310 // Otherwise exit.
311 //
312 // Important note: if the location of libjvm.so changes this
313 // code needs to be changed accordingly.
314
315 // See ld(1):
316 // The linker uses the following search paths to locate required
317 // shared libraries:
318 // 1: ...
319 // ...
320 // 7: The default directories, normally /lib and /usr/lib.
321 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
322 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
323 #else
324 #define DEFAULT_LIBPATH "/lib:/usr/lib"
325 #endif
326
327 // Base path of extensions installed on the system.
328 #define SYS_EXT_DIR "/usr/java/packages"
329 #define EXTENSIONS_DIR "/lib/ext"
330
331 // Buffer that fits several sprintfs.
332 // Note that the space for the colon and the trailing null are provided
333 // by the nulls included by the sizeof operator.
334 const size_t bufsize =
335 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
336 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
337 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
338
339 // sysclasspath, java_home, dll_dir
340 {
341 char *pslash;
342 os::jvm_path(buf, bufsize);
343
344 // Found the full path to libjvm.so.
345 // Now cut the path to <java_home>/jre if we can.
346 pslash = strrchr(buf, '/');
347 if (pslash != NULL) {
348 *pslash = '\0'; // Get rid of /libjvm.so.
349 }
350 pslash = strrchr(buf, '/');
351 if (pslash != NULL) {
352 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
353 }
354 Arguments::set_dll_dir(buf);
355
356 if (pslash != NULL) {
357 pslash = strrchr(buf, '/');
358 if (pslash != NULL) {
359 *pslash = '\0'; // Get rid of /lib.
360 }
361 }
362 Arguments::set_java_home(buf);
363 set_boot_path('/', ':');
364 }
365
366 // Where to look for native libraries.
367 //
368 // Note: Due to a legacy implementation, most of the library path
369 // is set in the launcher. This was to accomodate linking restrictions
370 // on legacy Linux implementations (which are no longer supported).
371 // Eventually, all the library path setting will be done here.
372 //
373 // However, to prevent the proliferation of improperly built native
374 // libraries, the new path component /usr/java/packages is added here.
375 // Eventually, all the library path setting will be done here.
376 {
377 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
378 // should always exist (until the legacy problem cited above is
379 // addressed).
380 const char *v = ::getenv("LD_LIBRARY_PATH");
381 const char *v_colon = ":";
382 if (v == NULL) { v = ""; v_colon = ""; }
383 // That's +1 for the colon and +1 for the trailing '\0'.
384 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
385 strlen(v) + 1 +
386 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
387 mtInternal);
388 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
389 Arguments::set_library_path(ld_library_path);
390 FREE_C_HEAP_ARRAY(char, ld_library_path);
391 }
392
393 // Extensions directories.
394 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
395 Arguments::set_ext_dirs(buf);
396
397 FREE_C_HEAP_ARRAY(char, buf);
398
399 #undef DEFAULT_LIBPATH
400 #undef SYS_EXT_DIR
401 #undef EXTENSIONS_DIR
402 }
403
404 ////////////////////////////////////////////////////////////////////////////////
405 // breakpoint support
406
407 void os::breakpoint() {
408 BREAKPOINT;
409 }
410
411 extern "C" void breakpoint() {
412 // use debugger to set breakpoint here
413 }
414
415 ////////////////////////////////////////////////////////////////////////////////
416 // signal support
417
418 debug_only(static bool signal_sets_initialized = false);
419 static sigset_t unblocked_sigs, vm_sigs;
420
421 bool os::Linux::is_sig_ignored(int sig) {
422 struct sigaction oact;
423 sigaction(sig, (struct sigaction*)NULL, &oact);
424 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
425 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
426 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
427 return true;
428 } else {
429 return false;
430 }
431 }
432
433 void os::Linux::signal_sets_init() {
434 // Should also have an assertion stating we are still single-threaded.
435 assert(!signal_sets_initialized, "Already initialized");
436 // Fill in signals that are necessarily unblocked for all threads in
437 // the VM. Currently, we unblock the following signals:
438 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
439 // by -Xrs (=ReduceSignalUsage));
440 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
441 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
442 // the dispositions or masks wrt these signals.
443 // Programs embedding the VM that want to use the above signals for their
444 // own purposes must, at this time, use the "-Xrs" option to prevent
445 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
446 // (See bug 4345157, and other related bugs).
447 // In reality, though, unblocking these signals is really a nop, since
448 // these signals are not blocked by default.
449 sigemptyset(&unblocked_sigs);
450 sigaddset(&unblocked_sigs, SIGILL);
451 sigaddset(&unblocked_sigs, SIGSEGV);
452 sigaddset(&unblocked_sigs, SIGBUS);
453 sigaddset(&unblocked_sigs, SIGFPE);
454 #if defined(PPC64)
455 sigaddset(&unblocked_sigs, SIGTRAP);
456 #endif
457 sigaddset(&unblocked_sigs, SR_signum);
458
459 if (!ReduceSignalUsage) {
460 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
461 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
462 }
463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 }
469 }
470 // Fill in signals that are blocked by all but the VM thread.
471 sigemptyset(&vm_sigs);
472 if (!ReduceSignalUsage) {
473 sigaddset(&vm_sigs, BREAK_SIGNAL);
474 }
475 debug_only(signal_sets_initialized = true);
476
477 }
478
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
485
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
492
493 void os::Linux::hotspot_sigmask(Thread* thread) {
494
495 //Save caller's signal mask before setting VM signal mask
496 sigset_t caller_sigmask;
497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
498
499 OSThread* osthread = thread->osthread();
500 osthread->set_caller_sigmask(caller_sigmask);
501
502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
503
504 if (!ReduceSignalUsage) {
505 if (thread->is_VM_thread()) {
506 // Only the VM thread handles BREAK_SIGNAL ...
507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
508 } else {
509 // ... all other threads block BREAK_SIGNAL
510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
511 }
512 }
513 }
514
515 //////////////////////////////////////////////////////////////////////////////
516 // detecting pthread library
517
518 void os::Linux::libpthread_init() {
519 // Save glibc and pthread version strings.
520 #if !defined(_CS_GNU_LIBC_VERSION) || \
521 !defined(_CS_GNU_LIBPTHREAD_VERSION)
522 #error "glibc too old (< 2.3.2)"
523 #endif
524
525 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
526 assert(n > 0, "cannot retrieve glibc version");
527 char *str = (char *)malloc(n, mtInternal);
528 confstr(_CS_GNU_LIBC_VERSION, str, n);
529 os::Linux::set_glibc_version(str);
530
531 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
532 assert(n > 0, "cannot retrieve pthread version");
533 str = (char *)malloc(n, mtInternal);
534 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
535 os::Linux::set_libpthread_version(str);
536 }
537
538 /////////////////////////////////////////////////////////////////////////////
539 // thread stack expansion
540
541 // os::Linux::manually_expand_stack() takes care of expanding the thread
542 // stack. Note that this is normally not needed: pthread stacks allocate
543 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
544 // committed. Therefore it is not necessary to expand the stack manually.
545 //
546 // Manually expanding the stack was historically needed on LinuxThreads
547 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
548 // it is kept to deal with very rare corner cases:
549 //
550 // For one, user may run the VM on an own implementation of threads
551 // whose stacks are - like the old LinuxThreads - implemented using
552 // mmap(MAP_GROWSDOWN).
553 //
554 // Also, this coding may be needed if the VM is running on the primordial
555 // thread. Normally we avoid running on the primordial thread; however,
556 // user may still invoke the VM on the primordial thread.
557 //
558 // The following historical comment describes the details about running
559 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
560
561
562 // Force Linux kernel to expand current thread stack. If "bottom" is close
563 // to the stack guard, caller should block all signals.
564 //
565 // MAP_GROWSDOWN:
566 // A special mmap() flag that is used to implement thread stacks. It tells
567 // kernel that the memory region should extend downwards when needed. This
568 // allows early versions of LinuxThreads to only mmap the first few pages
569 // when creating a new thread. Linux kernel will automatically expand thread
570 // stack as needed (on page faults).
571 //
572 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
573 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
574 // region, it's hard to tell if the fault is due to a legitimate stack
575 // access or because of reading/writing non-exist memory (e.g. buffer
576 // overrun). As a rule, if the fault happens below current stack pointer,
577 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
578 // application (see Linux kernel fault.c).
579 //
580 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
581 // stack overflow detection.
582 //
583 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
584 // not use MAP_GROWSDOWN.
585 //
586 // To get around the problem and allow stack banging on Linux, we need to
587 // manually expand thread stack after receiving the SIGSEGV.
588 //
589 // There are two ways to expand thread stack to address "bottom", we used
590 // both of them in JVM before 1.5:
591 // 1. adjust stack pointer first so that it is below "bottom", and then
592 // touch "bottom"
593 // 2. mmap() the page in question
594 //
595 // Now alternate signal stack is gone, it's harder to use 2. For instance,
596 // if current sp is already near the lower end of page 101, and we need to
597 // call mmap() to map page 100, it is possible that part of the mmap() frame
598 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
599 // That will destroy the mmap() frame and cause VM to crash.
600 //
601 // The following code works by adjusting sp first, then accessing the "bottom"
602 // page to force a page fault. Linux kernel will then automatically expand the
603 // stack mapping.
604 //
605 // _expand_stack_to() assumes its frame size is less than page size, which
606 // should always be true if the function is not inlined.
607
608 static void NOINLINE _expand_stack_to(address bottom) {
609 address sp;
610 size_t size;
611 volatile char *p;
612
613 // Adjust bottom to point to the largest address within the same page, it
614 // gives us a one-page buffer if alloca() allocates slightly more memory.
615 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
616 bottom += os::Linux::page_size() - 1;
617
618 // sp might be slightly above current stack pointer; if that's the case, we
619 // will alloca() a little more space than necessary, which is OK. Don't use
620 // os::current_stack_pointer(), as its result can be slightly below current
621 // stack pointer, causing us to not alloca enough to reach "bottom".
622 sp = (address)&sp;
623
624 if (sp > bottom) {
625 size = sp - bottom;
626 p = (volatile char *)alloca(size);
627 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
628 p[0] = '\0';
629 }
630 }
631
632 void os::Linux::expand_stack_to(address bottom) {
633 _expand_stack_to(bottom);
634 }
635
636 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
637 assert(t!=NULL, "just checking");
638 assert(t->osthread()->expanding_stack(), "expand should be set");
639 assert(t->stack_base() != NULL, "stack_base was not initialized");
640
641 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) {
642 sigset_t mask_all, old_sigset;
643 sigfillset(&mask_all);
644 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
645 _expand_stack_to(addr);
646 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
647 return true;
648 }
649 return false;
650 }
651
652 //////////////////////////////////////////////////////////////////////////////
653 // create new thread
654
655 // Thread start routine for all newly created threads
656 static void *thread_native_entry(Thread *thread) {
657 // Try to randomize the cache line index of hot stack frames.
658 // This helps when threads of the same stack traces evict each other's
659 // cache lines. The threads can be either from the same JVM instance, or
660 // from different JVM instances. The benefit is especially true for
661 // processors with hyperthreading technology.
662 static int counter = 0;
663 int pid = os::current_process_id();
664 alloca(((pid ^ counter++) & 7) * 128);
665
666 thread->initialize_thread_current();
667
668 OSThread* osthread = thread->osthread();
669 Monitor* sync = osthread->startThread_lock();
670
671 osthread->set_thread_id(os::current_thread_id());
672
673 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
674 os::current_thread_id(), (uintx) pthread_self());
675
676 if (UseNUMA) {
677 int lgrp_id = os::numa_get_group_id();
678 if (lgrp_id != -1) {
679 thread->set_lgrp_id(lgrp_id);
680 }
681 }
682 // initialize signal mask for this thread
683 os::Linux::hotspot_sigmask(thread);
684
685 // initialize floating point control register
686 os::Linux::init_thread_fpu_state();
687
688 // handshaking with parent thread
689 {
690 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
691
692 // notify parent thread
693 osthread->set_state(INITIALIZED);
694 sync->notify_all();
695
696 // wait until os::start_thread()
697 while (osthread->get_state() == INITIALIZED) {
698 sync->wait(Mutex::_no_safepoint_check_flag);
699 }
700 }
701
702 // call one more level start routine
703 thread->run();
704
705 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
706 os::current_thread_id(), (uintx) pthread_self());
707
708 // If a thread has not deleted itself ("delete this") as part of its
709 // termination sequence, we have to ensure thread-local-storage is
710 // cleared before we actually terminate. No threads should ever be
711 // deleted asynchronously with respect to their termination.
712 if (Thread::current_or_null_safe() != NULL) {
713 assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
714 thread->clear_thread_current();
715 }
716
717 return 0;
718 }
719
720 bool os::create_thread(Thread* thread, ThreadType thr_type,
721 size_t req_stack_size) {
722 assert(thread->osthread() == NULL, "caller responsible");
723
724 // Allocate the OSThread object
725 OSThread* osthread = new OSThread(NULL, NULL);
726 if (osthread == NULL) {
727 return false;
728 }
729
730 // set the correct thread state
731 osthread->set_thread_type(thr_type);
732
733 // Initial state is ALLOCATED but not INITIALIZED
734 osthread->set_state(ALLOCATED);
735
736 thread->set_osthread(osthread);
737
738 // init thread attributes
739 pthread_attr_t attr;
740 pthread_attr_init(&attr);
741 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
742
743 // Calculate stack size if it's not specified by caller.
744 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
745 // In the Linux NPTL pthread implementation the guard size mechanism
746 // is not implemented properly. The posix standard requires adding
747 // the size of the guard pages to the stack size, instead Linux
748 // takes the space out of 'stacksize'. Thus we adapt the requested
749 // stack_size by the size of the guard pages to mimick proper
750 // behaviour. However, be careful not to end up with a size
751 // of zero due to overflow. Don't add the guard page in that case.
752 size_t guard_size = os::Linux::default_guard_size(thr_type);
753 if (stack_size <= SIZE_MAX - guard_size) {
754 stack_size += guard_size;
755 }
756 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
757
758 int status = pthread_attr_setstacksize(&attr, stack_size);
759 assert_status(status == 0, status, "pthread_attr_setstacksize");
760
761 // Configure glibc guard page.
762 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
763
764 ThreadState state;
765
766 {
767 pthread_t tid;
768 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
769
770 char buf[64];
771 if (ret == 0) {
772 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
773 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
774 } else {
775 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
776 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
777 }
778
779 pthread_attr_destroy(&attr);
780
781 if (ret != 0) {
782 // Need to clean up stuff we've allocated so far
783 thread->set_osthread(NULL);
784 delete osthread;
785 return false;
786 }
787
788 // Store pthread info into the OSThread
789 osthread->set_pthread_id(tid);
790
791 // Wait until child thread is either initialized or aborted
792 {
793 Monitor* sync_with_child = osthread->startThread_lock();
794 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
795 while ((state = osthread->get_state()) == ALLOCATED) {
796 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
797 }
798 }
799 }
800
801 // Aborted due to thread limit being reached
802 if (state == ZOMBIE) {
803 thread->set_osthread(NULL);
804 delete osthread;
805 return false;
806 }
807
808 // The thread is returned suspended (in state INITIALIZED),
809 // and is started higher up in the call chain
810 assert(state == INITIALIZED, "race condition");
811 return true;
812 }
813
814 /////////////////////////////////////////////////////////////////////////////
815 // attach existing thread
816
817 // bootstrap the main thread
818 bool os::create_main_thread(JavaThread* thread) {
819 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
820 return create_attached_thread(thread);
821 }
822
823 bool os::create_attached_thread(JavaThread* thread) {
824 #ifdef ASSERT
825 thread->verify_not_published();
826 #endif
827
828 // Allocate the OSThread object
829 OSThread* osthread = new OSThread(NULL, NULL);
830
831 if (osthread == NULL) {
832 return false;
833 }
834
835 // Store pthread info into the OSThread
836 osthread->set_thread_id(os::Linux::gettid());
837 osthread->set_pthread_id(::pthread_self());
838
839 // initialize floating point control register
840 os::Linux::init_thread_fpu_state();
841
842 // Initial thread state is RUNNABLE
843 osthread->set_state(RUNNABLE);
844
845 thread->set_osthread(osthread);
846
847 if (UseNUMA) {
848 int lgrp_id = os::numa_get_group_id();
849 if (lgrp_id != -1) {
850 thread->set_lgrp_id(lgrp_id);
851 }
852 }
853
854 if (os::is_primordial_thread()) {
855 // If current thread is primordial thread, its stack is mapped on demand,
856 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
857 // the entire stack region to avoid SEGV in stack banging.
858 // It is also useful to get around the heap-stack-gap problem on SuSE
859 // kernel (see 4821821 for details). We first expand stack to the top
860 // of yellow zone, then enable stack yellow zone (order is significant,
861 // enabling yellow zone first will crash JVM on SuSE Linux), so there
862 // is no gap between the last two virtual memory regions.
863
864 JavaThread *jt = (JavaThread *)thread;
865 address addr = jt->stack_reserved_zone_base();
866 assert(addr != NULL, "initialization problem?");
867 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
868
869 osthread->set_expanding_stack();
870 os::Linux::manually_expand_stack(jt, addr);
871 osthread->clear_expanding_stack();
872 }
873
874 // initialize signal mask for this thread
875 // and save the caller's signal mask
876 os::Linux::hotspot_sigmask(thread);
877
878 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
879 os::current_thread_id(), (uintx) pthread_self());
880
881 return true;
882 }
883
884 void os::pd_start_thread(Thread* thread) {
885 OSThread * osthread = thread->osthread();
886 assert(osthread->get_state() != INITIALIZED, "just checking");
887 Monitor* sync_with_child = osthread->startThread_lock();
888 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
889 sync_with_child->notify();
890 }
891
892 // Free Linux resources related to the OSThread
893 void os::free_thread(OSThread* osthread) {
894 assert(osthread != NULL, "osthread not set");
895
896 // We are told to free resources of the argument thread,
897 // but we can only really operate on the current thread.
898 assert(Thread::current()->osthread() == osthread,
899 "os::free_thread but not current thread");
900
901 #ifdef ASSERT
902 sigset_t current;
903 sigemptyset(¤t);
904 pthread_sigmask(SIG_SETMASK, NULL, ¤t);
905 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!");
906 #endif
907
908 // Restore caller's signal mask
909 sigset_t sigmask = osthread->caller_sigmask();
910 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
911
912 delete osthread;
913 }
914
915 //////////////////////////////////////////////////////////////////////////////
916 // primordial thread
917
918 // Check if current thread is the primordial thread, similar to Solaris thr_main.
919 bool os::is_primordial_thread(void) {
920 char dummy;
921 // If called before init complete, thread stack bottom will be null.
922 // Can be called if fatal error occurs before initialization.
923 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
924 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
925 os::Linux::initial_thread_stack_size() != 0,
926 "os::init did not locate primordial thread's stack region");
927 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
928 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
929 os::Linux::initial_thread_stack_size()) {
930 return true;
931 } else {
932 return false;
933 }
934 }
935
936 // Find the virtual memory area that contains addr
937 static bool find_vma(address addr, address* vma_low, address* vma_high) {
938 FILE *fp = fopen("/proc/self/maps", "r");
939 if (fp) {
940 address low, high;
941 while (!feof(fp)) {
942 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
943 if (low <= addr && addr < high) {
944 if (vma_low) *vma_low = low;
945 if (vma_high) *vma_high = high;
946 fclose(fp);
947 return true;
948 }
949 }
950 for (;;) {
951 int ch = fgetc(fp);
952 if (ch == EOF || ch == (int)'\n') break;
953 }
954 }
955 fclose(fp);
956 }
957 return false;
958 }
959
960 // Locate primordial thread stack. This special handling of primordial thread stack
961 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
962 // bogus value for the primordial process thread. While the launcher has created
963 // the VM in a new thread since JDK 6, we still have to allow for the use of the
964 // JNI invocation API from a primordial thread.
965 void os::Linux::capture_initial_stack(size_t max_size) {
966
967 // max_size is either 0 (which means accept OS default for thread stacks) or
968 // a user-specified value known to be at least the minimum needed. If we
969 // are actually on the primordial thread we can make it appear that we have a
970 // smaller max_size stack by inserting the guard pages at that location. But we
971 // cannot do anything to emulate a larger stack than what has been provided by
972 // the OS or threading library. In fact if we try to use a stack greater than
973 // what is set by rlimit then we will crash the hosting process.
974
975 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
976 // If this is "unlimited" then it will be a huge value.
977 struct rlimit rlim;
978 getrlimit(RLIMIT_STACK, &rlim);
979 size_t stack_size = rlim.rlim_cur;
980
981 // 6308388: a bug in ld.so will relocate its own .data section to the
982 // lower end of primordial stack; reduce ulimit -s value a little bit
983 // so we won't install guard page on ld.so's data section.
984 // But ensure we don't underflow the stack size - allow 1 page spare
985 if (stack_size >= (size_t)(3 * page_size())) {
986 stack_size -= 2 * page_size();
987 }
988
989 // Try to figure out where the stack base (top) is. This is harder.
990 //
991 // When an application is started, glibc saves the initial stack pointer in
992 // a global variable "__libc_stack_end", which is then used by system
993 // libraries. __libc_stack_end should be pretty close to stack top. The
994 // variable is available since the very early days. However, because it is
995 // a private interface, it could disappear in the future.
996 //
997 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
998 // to __libc_stack_end, it is very close to stack top, but isn't the real
999 // stack top. Note that /proc may not exist if VM is running as a chroot
1000 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1001 // /proc/<pid>/stat could change in the future (though unlikely).
1002 //
1003 // We try __libc_stack_end first. If that doesn't work, look for
1004 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1005 // as a hint, which should work well in most cases.
1006
1007 uintptr_t stack_start;
1008
1009 // try __libc_stack_end first
1010 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1011 if (p && *p) {
1012 stack_start = *p;
1013 } else {
1014 // see if we can get the start_stack field from /proc/self/stat
1015 FILE *fp;
1016 int pid;
1017 char state;
1018 int ppid;
1019 int pgrp;
1020 int session;
1021 int nr;
1022 int tpgrp;
1023 unsigned long flags;
1024 unsigned long minflt;
1025 unsigned long cminflt;
1026 unsigned long majflt;
1027 unsigned long cmajflt;
1028 unsigned long utime;
1029 unsigned long stime;
1030 long cutime;
1031 long cstime;
1032 long prio;
1033 long nice;
1034 long junk;
1035 long it_real;
1036 uintptr_t start;
1037 uintptr_t vsize;
1038 intptr_t rss;
1039 uintptr_t rsslim;
1040 uintptr_t scodes;
1041 uintptr_t ecode;
1042 int i;
1043
1044 // Figure what the primordial thread stack base is. Code is inspired
1045 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1046 // followed by command name surrounded by parentheses, state, etc.
1047 char stat[2048];
1048 int statlen;
1049
1050 fp = fopen("/proc/self/stat", "r");
1051 if (fp) {
1052 statlen = fread(stat, 1, 2047, fp);
1053 stat[statlen] = '\0';
1054 fclose(fp);
1055
1056 // Skip pid and the command string. Note that we could be dealing with
1057 // weird command names, e.g. user could decide to rename java launcher
1058 // to "java 1.4.2 :)", then the stat file would look like
1059 // 1234 (java 1.4.2 :)) R ... ...
1060 // We don't really need to know the command string, just find the last
1061 // occurrence of ")" and then start parsing from there. See bug 4726580.
1062 char * s = strrchr(stat, ')');
1063
1064 i = 0;
1065 if (s) {
1066 // Skip blank chars
1067 do { s++; } while (s && isspace(*s));
1068
1069 #define _UFM UINTX_FORMAT
1070 #define _DFM INTX_FORMAT
1071
1072 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1073 // 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
1074 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,
1075 &state, // 3 %c
1076 &ppid, // 4 %d
1077 &pgrp, // 5 %d
1078 &session, // 6 %d
1079 &nr, // 7 %d
1080 &tpgrp, // 8 %d
1081 &flags, // 9 %lu
1082 &minflt, // 10 %lu
1083 &cminflt, // 11 %lu
1084 &majflt, // 12 %lu
1085 &cmajflt, // 13 %lu
1086 &utime, // 14 %lu
1087 &stime, // 15 %lu
1088 &cutime, // 16 %ld
1089 &cstime, // 17 %ld
1090 &prio, // 18 %ld
1091 &nice, // 19 %ld
1092 &junk, // 20 %ld
1093 &it_real, // 21 %ld
1094 &start, // 22 UINTX_FORMAT
1095 &vsize, // 23 UINTX_FORMAT
1096 &rss, // 24 INTX_FORMAT
1097 &rsslim, // 25 UINTX_FORMAT
1098 &scodes, // 26 UINTX_FORMAT
1099 &ecode, // 27 UINTX_FORMAT
1100 &stack_start); // 28 UINTX_FORMAT
1101 }
1102
1103 #undef _UFM
1104 #undef _DFM
1105
1106 if (i != 28 - 2) {
1107 assert(false, "Bad conversion from /proc/self/stat");
1108 // product mode - assume we are the primordial thread, good luck in the
1109 // embedded case.
1110 warning("Can't detect primordial thread stack location - bad conversion");
1111 stack_start = (uintptr_t) &rlim;
1112 }
1113 } else {
1114 // For some reason we can't open /proc/self/stat (for example, running on
1115 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1116 // most cases, so don't abort:
1117 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1118 stack_start = (uintptr_t) &rlim;
1119 }
1120 }
1121
1122 // Now we have a pointer (stack_start) very close to the stack top, the
1123 // next thing to do is to figure out the exact location of stack top. We
1124 // can find out the virtual memory area that contains stack_start by
1125 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1126 // and its upper limit is the real stack top. (again, this would fail if
1127 // running inside chroot, because /proc may not exist.)
1128
1129 uintptr_t stack_top;
1130 address low, high;
1131 if (find_vma((address)stack_start, &low, &high)) {
1132 // success, "high" is the true stack top. (ignore "low", because initial
1133 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1134 stack_top = (uintptr_t)high;
1135 } else {
1136 // failed, likely because /proc/self/maps does not exist
1137 warning("Can't detect primordial thread stack location - find_vma failed");
1138 // best effort: stack_start is normally within a few pages below the real
1139 // stack top, use it as stack top, and reduce stack size so we won't put
1140 // guard page outside stack.
1141 stack_top = stack_start;
1142 stack_size -= 16 * page_size();
1143 }
1144
1145 // stack_top could be partially down the page so align it
1146 stack_top = align_up(stack_top, page_size());
1147
1148 // Allowed stack value is minimum of max_size and what we derived from rlimit
1149 if (max_size > 0) {
1150 _initial_thread_stack_size = MIN2(max_size, stack_size);
1151 } else {
1152 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1153 // clamp it at 8MB as we do on Solaris
1154 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1155 }
1156 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1157 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1158
1159 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1160
1161 if (log_is_enabled(Info, os, thread)) {
1162 // See if we seem to be on primordial process thread
1163 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1164 uintptr_t(&rlim) < stack_top;
1165
1166 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1167 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1168 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K,
1169 stack_top, intptr_t(_initial_thread_stack_bottom));
1170 }
1171 }
1172
1173 ////////////////////////////////////////////////////////////////////////////////
1174 // time support
1175
1176 // Time since start-up in seconds to a fine granularity.
1177 // Used by VMSelfDestructTimer and the MemProfiler.
1178 double os::elapsedTime() {
1179
1180 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1181 }
1182
1183 jlong os::elapsed_counter() {
1184 return javaTimeNanos() - initial_time_count;
1185 }
1186
1187 jlong os::elapsed_frequency() {
1188 return NANOSECS_PER_SEC; // nanosecond resolution
1189 }
1190
1191 bool os::supports_vtime() { return true; }
1192 bool os::enable_vtime() { return false; }
1193 bool os::vtime_enabled() { return false; }
1194
1195 double os::elapsedVTime() {
1196 struct rusage usage;
1197 int retval = getrusage(RUSAGE_THREAD, &usage);
1198 if (retval == 0) {
1199 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);
1200 } else {
1201 // better than nothing, but not much
1202 return elapsedTime();
1203 }
1204 }
1205
1206 jlong os::javaTimeMillis() {
1207 timeval time;
1208 int status = gettimeofday(&time, NULL);
1209 assert(status != -1, "linux error");
1210 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1211 }
1212
1213 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1214 timeval time;
1215 int status = gettimeofday(&time, NULL);
1216 assert(status != -1, "linux error");
1217 seconds = jlong(time.tv_sec);
1218 nanos = jlong(time.tv_usec) * 1000;
1219 }
1220
1221
1222 #ifndef CLOCK_MONOTONIC
1223 #define CLOCK_MONOTONIC (1)
1224 #endif
1225
1226 void os::Linux::clock_init() {
1227 // we do dlopen's in this particular order due to bug in linux
1228 // dynamical loader (see 6348968) leading to crash on exit
1229 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1230 if (handle == NULL) {
1231 handle = dlopen("librt.so", RTLD_LAZY);
1232 }
1233
1234 if (handle) {
1235 int (*clock_getres_func)(clockid_t, struct timespec*) =
1236 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1237 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1238 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1239 if (clock_getres_func && clock_gettime_func) {
1240 // See if monotonic clock is supported by the kernel. Note that some
1241 // early implementations simply return kernel jiffies (updated every
1242 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1243 // for nano time (though the monotonic property is still nice to have).
1244 // It's fixed in newer kernels, however clock_getres() still returns
1245 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1246 // resolution for now. Hopefully as people move to new kernels, this
1247 // won't be a problem.
1248 struct timespec res;
1249 struct timespec tp;
1250 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1251 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1252 // yes, monotonic clock is supported
1253 _clock_gettime = clock_gettime_func;
1254 return;
1255 } else {
1256 // close librt if there is no monotonic clock
1257 dlclose(handle);
1258 }
1259 }
1260 }
1261 warning("No monotonic clock was available - timed services may " \
1262 "be adversely affected if the time-of-day clock changes");
1263 }
1264
1265 #ifndef SYS_clock_getres
1266 #if defined(X86) || defined(PPC64) || defined(S390)
1267 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1268 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1269 #else
1270 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1271 #define sys_clock_getres(x,y) -1
1272 #endif
1273 #else
1274 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1275 #endif
1276
1277 void os::Linux::fast_thread_clock_init() {
1278 if (!UseLinuxPosixThreadCPUClocks) {
1279 return;
1280 }
1281 clockid_t clockid;
1282 struct timespec tp;
1283 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1284 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1285
1286 // Switch to using fast clocks for thread cpu time if
1287 // the sys_clock_getres() returns 0 error code.
1288 // Note, that some kernels may support the current thread
1289 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1290 // returned by the pthread_getcpuclockid().
1291 // If the fast Posix clocks are supported then the sys_clock_getres()
1292 // must return at least tp.tv_sec == 0 which means a resolution
1293 // better than 1 sec. This is extra check for reliability.
1294
1295 if (pthread_getcpuclockid_func &&
1296 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1297 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1298 _supports_fast_thread_cpu_time = true;
1299 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1300 }
1301 }
1302
1303 jlong os::javaTimeNanos() {
1304 if (os::supports_monotonic_clock()) {
1305 struct timespec tp;
1306 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1307 assert(status == 0, "gettime error");
1308 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1309 return result;
1310 } else {
1311 timeval time;
1312 int status = gettimeofday(&time, NULL);
1313 assert(status != -1, "linux error");
1314 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1315 return 1000 * usecs;
1316 }
1317 }
1318
1319 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1320 if (os::supports_monotonic_clock()) {
1321 info_ptr->max_value = ALL_64_BITS;
1322
1323 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1324 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1325 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1326 } else {
1327 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1328 info_ptr->max_value = ALL_64_BITS;
1329
1330 // gettimeofday is a real time clock so it skips
1331 info_ptr->may_skip_backward = true;
1332 info_ptr->may_skip_forward = true;
1333 }
1334
1335 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1336 }
1337
1338 // Return the real, user, and system times in seconds from an
1339 // arbitrary fixed point in the past.
1340 bool os::getTimesSecs(double* process_real_time,
1341 double* process_user_time,
1342 double* process_system_time) {
1343 struct tms ticks;
1344 clock_t real_ticks = times(&ticks);
1345
1346 if (real_ticks == (clock_t) (-1)) {
1347 return false;
1348 } else {
1349 double ticks_per_second = (double) clock_tics_per_sec;
1350 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1351 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1352 *process_real_time = ((double) real_ticks) / ticks_per_second;
1353
1354 return true;
1355 }
1356 }
1357
1358
1359 char * os::local_time_string(char *buf, size_t buflen) {
1360 struct tm t;
1361 time_t long_time;
1362 time(&long_time);
1363 localtime_r(&long_time, &t);
1364 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1365 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1366 t.tm_hour, t.tm_min, t.tm_sec);
1367 return buf;
1368 }
1369
1370 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1371 return localtime_r(clock, res);
1372 }
1373
1374 ////////////////////////////////////////////////////////////////////////////////
1375 // runtime exit support
1376
1377 // Note: os::shutdown() might be called very early during initialization, or
1378 // called from signal handler. Before adding something to os::shutdown(), make
1379 // sure it is async-safe and can handle partially initialized VM.
1380 void os::shutdown() {
1381
1382 // allow PerfMemory to attempt cleanup of any persistent resources
1383 perfMemory_exit();
1384
1385 // needs to remove object in file system
1386 AttachListener::abort();
1387
1388 // flush buffered output, finish log files
1389 ostream_abort();
1390
1391 // Check for abort hook
1392 abort_hook_t abort_hook = Arguments::abort_hook();
1393 if (abort_hook != NULL) {
1394 abort_hook();
1395 }
1396
1397 }
1398
1399 // Note: os::abort() might be called very early during initialization, or
1400 // called from signal handler. Before adding something to os::abort(), make
1401 // sure it is async-safe and can handle partially initialized VM.
1402 void os::abort(bool dump_core, void* siginfo, const void* context) {
1403 os::shutdown();
1404 if (dump_core) {
1405 #ifndef PRODUCT
1406 fdStream out(defaultStream::output_fd());
1407 out.print_raw("Current thread is ");
1408 char buf[16];
1409 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1410 out.print_raw_cr(buf);
1411 out.print_raw_cr("Dumping core ...");
1412 #endif
1413 ::abort(); // dump core
1414 }
1415
1416 ::exit(1);
1417 }
1418
1419 // Die immediately, no exit hook, no abort hook, no cleanup.
1420 void os::die() {
1421 ::abort();
1422 }
1423
1424
1425 // This method is a copy of JDK's sysGetLastErrorString
1426 // from src/solaris/hpi/src/system_md.c
1427
1428 size_t os::lasterror(char *buf, size_t len) {
1429 if (errno == 0) return 0;
1430
1431 const char *s = os::strerror(errno);
1432 size_t n = ::strlen(s);
1433 if (n >= len) {
1434 n = len - 1;
1435 }
1436 ::strncpy(buf, s, n);
1437 buf[n] = '\0';
1438 return n;
1439 }
1440
1441 // thread_id is kernel thread id (similar to Solaris LWP id)
1442 intx os::current_thread_id() { return os::Linux::gettid(); }
1443 int os::current_process_id() {
1444 return ::getpid();
1445 }
1446
1447 // DLL functions
1448
1449 const char* os::dll_file_extension() { return ".so"; }
1450
1451 // This must be hard coded because it's the system's temporary
1452 // directory not the java application's temp directory, ala java.io.tmpdir.
1453 const char* os::get_temp_directory() { return "/tmp"; }
1454
1455 static bool file_exists(const char* filename) {
1456 struct stat statbuf;
1457 if (filename == NULL || strlen(filename) == 0) {
1458 return false;
1459 }
1460 return os::stat(filename, &statbuf) == 0;
1461 }
1462
1463 // check if addr is inside libjvm.so
1464 bool os::address_is_in_vm(address addr) {
1465 static address libjvm_base_addr;
1466 Dl_info dlinfo;
1467
1468 if (libjvm_base_addr == NULL) {
1469 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1470 libjvm_base_addr = (address)dlinfo.dli_fbase;
1471 }
1472 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1473 }
1474
1475 if (dladdr((void *)addr, &dlinfo) != 0) {
1476 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1477 }
1478
1479 return false;
1480 }
1481
1482 bool os::dll_address_to_function_name(address addr, char *buf,
1483 int buflen, int *offset,
1484 bool demangle) {
1485 // buf is not optional, but offset is optional
1486 assert(buf != NULL, "sanity check");
1487
1488 Dl_info dlinfo;
1489
1490 if (dladdr((void*)addr, &dlinfo) != 0) {
1491 // see if we have a matching symbol
1492 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1493 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1494 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1495 }
1496 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1497 return true;
1498 }
1499 // no matching symbol so try for just file info
1500 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1501 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1502 buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1503 return true;
1504 }
1505 }
1506 }
1507
1508 buf[0] = '\0';
1509 if (offset != NULL) *offset = -1;
1510 return false;
1511 }
1512
1513 struct _address_to_library_name {
1514 address addr; // input : memory address
1515 size_t buflen; // size of fname
1516 char* fname; // output: library name
1517 address base; // library base addr
1518 };
1519
1520 static int address_to_library_name_callback(struct dl_phdr_info *info,
1521 size_t size, void *data) {
1522 int i;
1523 bool found = false;
1524 address libbase = NULL;
1525 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1526
1527 // iterate through all loadable segments
1528 for (i = 0; i < info->dlpi_phnum; i++) {
1529 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1530 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1531 // base address of a library is the lowest address of its loaded
1532 // segments.
1533 if (libbase == NULL || libbase > segbase) {
1534 libbase = segbase;
1535 }
1536 // see if 'addr' is within current segment
1537 if (segbase <= d->addr &&
1538 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1539 found = true;
1540 }
1541 }
1542 }
1543
1544 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1545 // so dll_address_to_library_name() can fall through to use dladdr() which
1546 // can figure out executable name from argv[0].
1547 if (found && info->dlpi_name && info->dlpi_name[0]) {
1548 d->base = libbase;
1549 if (d->fname) {
1550 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1551 }
1552 return 1;
1553 }
1554 return 0;
1555 }
1556
1557 bool os::dll_address_to_library_name(address addr, char* buf,
1558 int buflen, int* offset) {
1559 // buf is not optional, but offset is optional
1560 assert(buf != NULL, "sanity check");
1561
1562 Dl_info dlinfo;
1563 struct _address_to_library_name data;
1564
1565 // There is a bug in old glibc dladdr() implementation that it could resolve
1566 // to wrong library name if the .so file has a base address != NULL. Here
1567 // we iterate through the program headers of all loaded libraries to find
1568 // out which library 'addr' really belongs to. This workaround can be
1569 // removed once the minimum requirement for glibc is moved to 2.3.x.
1570 data.addr = addr;
1571 data.fname = buf;
1572 data.buflen = buflen;
1573 data.base = NULL;
1574 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1575
1576 if (rslt) {
1577 // buf already contains library name
1578 if (offset) *offset = addr - data.base;
1579 return true;
1580 }
1581 if (dladdr((void*)addr, &dlinfo) != 0) {
1582 if (dlinfo.dli_fname != NULL) {
1583 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1584 }
1585 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1586 *offset = addr - (address)dlinfo.dli_fbase;
1587 }
1588 return true;
1589 }
1590
1591 buf[0] = '\0';
1592 if (offset) *offset = -1;
1593 return false;
1594 }
1595
1596 // Loads .dll/.so and
1597 // in case of error it checks if .dll/.so was built for the
1598 // same architecture as Hotspot is running on
1599
1600
1601 // Remember the stack's state. The Linux dynamic linker will change
1602 // the stack to 'executable' at most once, so we must safepoint only once.
1603 bool os::Linux::_stack_is_executable = false;
1604
1605 // VM operation that loads a library. This is necessary if stack protection
1606 // of the Java stacks can be lost during loading the library. If we
1607 // do not stop the Java threads, they can stack overflow before the stacks
1608 // are protected again.
1609 class VM_LinuxDllLoad: public VM_Operation {
1610 private:
1611 const char *_filename;
1612 char *_ebuf;
1613 int _ebuflen;
1614 void *_lib;
1615 public:
1616 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1617 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1618 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1619 void doit() {
1620 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1621 os::Linux::_stack_is_executable = true;
1622 }
1623 void* loaded_library() { return _lib; }
1624 };
1625
1626 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1627 void * result = NULL;
1628 bool load_attempted = false;
1629
1630 // Check whether the library to load might change execution rights
1631 // of the stack. If they are changed, the protection of the stack
1632 // guard pages will be lost. We need a safepoint to fix this.
1633 //
1634 // See Linux man page execstack(8) for more info.
1635 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1636 if (!ElfFile::specifies_noexecstack(filename)) {
1637 if (!is_init_completed()) {
1638 os::Linux::_stack_is_executable = true;
1639 // This is OK - No Java threads have been created yet, and hence no
1640 // stack guard pages to fix.
1641 //
1642 // This should happen only when you are building JDK7 using a very
1643 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1644 //
1645 // Dynamic loader will make all stacks executable after
1646 // this function returns, and will not do that again.
1647 #ifdef ASSERT
1648 ThreadsListHandle tlh;
1649 assert(tlh.length() == 0, "no Java threads should exist yet.");
1650 #endif
1651 } else {
1652 warning("You have loaded library %s which might have disabled stack guard. "
1653 "The VM will try to fix the stack guard now.\n"
1654 "It's highly recommended that you fix the library with "
1655 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1656 filename);
1657
1658 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1659 JavaThread *jt = JavaThread::current();
1660 if (jt->thread_state() != _thread_in_native) {
1661 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1662 // that requires ExecStack. Cannot enter safe point. Let's give up.
1663 warning("Unable to fix stack guard. Giving up.");
1664 } else {
1665 if (!LoadExecStackDllInVMThread) {
1666 // This is for the case where the DLL has an static
1667 // constructor function that executes JNI code. We cannot
1668 // load such DLLs in the VMThread.
1669 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1670 }
1671
1672 ThreadInVMfromNative tiv(jt);
1673 debug_only(VMNativeEntryWrapper vew;)
1674
1675 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1676 VMThread::execute(&op);
1677 if (LoadExecStackDllInVMThread) {
1678 result = op.loaded_library();
1679 }
1680 load_attempted = true;
1681 }
1682 }
1683 }
1684 }
1685
1686 if (!load_attempted) {
1687 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1688 }
1689
1690 if (result != NULL) {
1691 // Successful loading
1692 return result;
1693 }
1694
1695 Elf32_Ehdr elf_head;
1696 int diag_msg_max_length=ebuflen-strlen(ebuf);
1697 char* diag_msg_buf=ebuf+strlen(ebuf);
1698
1699 if (diag_msg_max_length==0) {
1700 // No more space in ebuf for additional diagnostics message
1701 return NULL;
1702 }
1703
1704
1705 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1706
1707 if (file_descriptor < 0) {
1708 // Can't open library, report dlerror() message
1709 return NULL;
1710 }
1711
1712 bool failed_to_read_elf_head=
1713 (sizeof(elf_head)!=
1714 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1715
1716 ::close(file_descriptor);
1717 if (failed_to_read_elf_head) {
1718 // file i/o error - report dlerror() msg
1719 return NULL;
1720 }
1721
1722 typedef struct {
1723 Elf32_Half code; // Actual value as defined in elf.h
1724 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1725 unsigned char elf_class; // 32 or 64 bit
1726 unsigned char endianess; // MSB or LSB
1727 char* name; // String representation
1728 } arch_t;
1729
1730 #ifndef EM_486
1731 #define EM_486 6 /* Intel 80486 */
1732 #endif
1733 #ifndef EM_AARCH64
1734 #define EM_AARCH64 183 /* ARM AARCH64 */
1735 #endif
1736
1737 static const arch_t arch_array[]={
1738 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1739 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1740 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1741 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1742 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1743 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1744 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1745 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1746 #if defined(VM_LITTLE_ENDIAN)
1747 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1748 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1749 #else
1750 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1751 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1752 #endif
1753 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1754 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1755 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1756 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1757 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1758 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1759 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1760 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1761 };
1762
1763 #if (defined IA32)
1764 static Elf32_Half running_arch_code=EM_386;
1765 #elif (defined AMD64)
1766 static Elf32_Half running_arch_code=EM_X86_64;
1767 #elif (defined IA64)
1768 static Elf32_Half running_arch_code=EM_IA_64;
1769 #elif (defined __sparc) && (defined _LP64)
1770 static Elf32_Half running_arch_code=EM_SPARCV9;
1771 #elif (defined __sparc) && (!defined _LP64)
1772 static Elf32_Half running_arch_code=EM_SPARC;
1773 #elif (defined __powerpc64__)
1774 static Elf32_Half running_arch_code=EM_PPC64;
1775 #elif (defined __powerpc__)
1776 static Elf32_Half running_arch_code=EM_PPC;
1777 #elif (defined AARCH64)
1778 static Elf32_Half running_arch_code=EM_AARCH64;
1779 #elif (defined ARM)
1780 static Elf32_Half running_arch_code=EM_ARM;
1781 #elif (defined S390)
1782 static Elf32_Half running_arch_code=EM_S390;
1783 #elif (defined ALPHA)
1784 static Elf32_Half running_arch_code=EM_ALPHA;
1785 #elif (defined MIPSEL)
1786 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1787 #elif (defined PARISC)
1788 static Elf32_Half running_arch_code=EM_PARISC;
1789 #elif (defined MIPS)
1790 static Elf32_Half running_arch_code=EM_MIPS;
1791 #elif (defined M68K)
1792 static Elf32_Half running_arch_code=EM_68K;
1793 #elif (defined SH)
1794 static Elf32_Half running_arch_code=EM_SH;
1795 #else
1796 #error Method os::dll_load requires that one of following is defined:\
1797 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1798 #endif
1799
1800 // Identify compatability class for VM's architecture and library's architecture
1801 // Obtain string descriptions for architectures
1802
1803 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1804 int running_arch_index=-1;
1805
1806 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1807 if (running_arch_code == arch_array[i].code) {
1808 running_arch_index = i;
1809 }
1810 if (lib_arch.code == arch_array[i].code) {
1811 lib_arch.compat_class = arch_array[i].compat_class;
1812 lib_arch.name = arch_array[i].name;
1813 }
1814 }
1815
1816 assert(running_arch_index != -1,
1817 "Didn't find running architecture code (running_arch_code) in arch_array");
1818 if (running_arch_index == -1) {
1819 // Even though running architecture detection failed
1820 // we may still continue with reporting dlerror() message
1821 return NULL;
1822 }
1823
1824 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1825 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1826 return NULL;
1827 }
1828
1829 #ifndef S390
1830 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1831 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1832 return NULL;
1833 }
1834 #endif // !S390
1835
1836 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1837 if (lib_arch.name!=NULL) {
1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1839 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1840 lib_arch.name, arch_array[running_arch_index].name);
1841 } else {
1842 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1843 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1844 lib_arch.code,
1845 arch_array[running_arch_index].name);
1846 }
1847 }
1848
1849 return NULL;
1850 }
1851
1852 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1853 int ebuflen) {
1854 void * result = ::dlopen(filename, RTLD_LAZY);
1855 if (result == NULL) {
1856 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1857 ebuf[ebuflen-1] = '\0';
1858 }
1859 return result;
1860 }
1861
1862 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1863 int ebuflen) {
1864 void * result = NULL;
1865 if (LoadExecStackDllInVMThread) {
1866 result = dlopen_helper(filename, ebuf, ebuflen);
1867 }
1868
1869 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1870 // library that requires an executable stack, or which does not have this
1871 // stack attribute set, dlopen changes the stack attribute to executable. The
1872 // read protection of the guard pages gets lost.
1873 //
1874 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1875 // may have been queued at the same time.
1876
1877 if (!_stack_is_executable) {
1878 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1879 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
1880 jt->stack_guards_enabled()) { // No pending stack overflow exceptions
1881 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1882 warning("Attempt to reguard stack yellow zone failed.");
1883 }
1884 }
1885 }
1886 }
1887
1888 return result;
1889 }
1890
1891 void* os::dll_lookup(void* handle, const char* name) {
1892 void* res = dlsym(handle, name);
1893 return res;
1894 }
1895
1896 void* os::get_default_process_handle() {
1897 return (void*)::dlopen(NULL, RTLD_LAZY);
1898 }
1899
1900 static bool _print_ascii_file(const char* filename, outputStream* st) {
1901 int fd = ::open(filename, O_RDONLY);
1902 if (fd == -1) {
1903 return false;
1904 }
1905
1906 char buf[33];
1907 int bytes;
1908 buf[32] = '\0';
1909 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1910 st->print_raw(buf, bytes);
1911 }
1912
1913 ::close(fd);
1914
1915 return true;
1916 }
1917
1918 void os::print_dll_info(outputStream *st) {
1919 st->print_cr("Dynamic libraries:");
1920
1921 char fname[32];
1922 pid_t pid = os::Linux::gettid();
1923
1924 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1925
1926 if (!_print_ascii_file(fname, st)) {
1927 st->print("Can not get library information for pid = %d\n", pid);
1928 }
1929 }
1930
1931 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1932 FILE *procmapsFile = NULL;
1933
1934 // Open the procfs maps file for the current process
1935 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1936 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1937 char line[PATH_MAX + 100];
1938
1939 // Read line by line from 'file'
1940 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1941 u8 base, top, offset, inode;
1942 char permissions[5];
1943 char device[6];
1944 char name[PATH_MAX + 1];
1945
1946 // Parse fields from line
1947 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1948 &base, &top, permissions, &offset, device, &inode, name);
1949
1950 // Filter by device id '00:00' so that we only get file system mapped files.
1951 if (strcmp(device, "00:00") != 0) {
1952
1953 // Call callback with the fields of interest
1954 if(callback(name, (address)base, (address)top, param)) {
1955 // Oops abort, callback aborted
1956 fclose(procmapsFile);
1957 return 1;
1958 }
1959 }
1960 }
1961 fclose(procmapsFile);
1962 }
1963 return 0;
1964 }
1965
1966 void os::print_os_info_brief(outputStream* st) {
1967 os::Linux::print_distro_info(st);
1968
1969 os::Posix::print_uname_info(st);
1970
1971 os::Linux::print_libversion_info(st);
1972
1973 }
1974
1975 void os::print_os_info(outputStream* st) {
1976 st->print("OS:");
1977
1978 os::Linux::print_distro_info(st);
1979
1980 os::Posix::print_uname_info(st);
1981
1982 // Print warning if unsafe chroot environment detected
1983 if (unsafe_chroot_detected) {
1984 st->print("WARNING!! ");
1985 st->print_cr("%s", unstable_chroot_error);
1986 }
1987
1988 os::Linux::print_libversion_info(st);
1989
1990 os::Posix::print_rlimit_info(st);
1991
1992 os::Posix::print_load_average(st);
1993
1994 os::Linux::print_full_memory_info(st);
1995
1996 os::Linux::print_container_info(st);
1997 }
1998
1999 // Try to identify popular distros.
2000 // Most Linux distributions have a /etc/XXX-release file, which contains
2001 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2002 // file that also contains the OS version string. Some have more than one
2003 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2004 // /etc/redhat-release.), so the order is important.
2005 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2006 // their own specific XXX-release file as well as a redhat-release file.
2007 // Because of this the XXX-release file needs to be searched for before the
2008 // redhat-release file.
2009 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2010 // search for redhat-release / SuSE-release needs to be before lsb-release.
2011 // Since the lsb-release file is the new standard it needs to be searched
2012 // before the older style release files.
2013 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2014 // next to last resort. The os-release file is a new standard that contains
2015 // distribution information and the system-release file seems to be an old
2016 // standard that has been replaced by the lsb-release and os-release files.
2017 // Searching for the debian_version file is the last resort. It contains
2018 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2019 // "Debian " is printed before the contents of the debian_version file.
2020
2021 const char* distro_files[] = {
2022 "/etc/oracle-release",
2023 "/etc/mandriva-release",
2024 "/etc/mandrake-release",
2025 "/etc/sun-release",
2026 "/etc/redhat-release",
2027 "/etc/SuSE-release",
2028 "/etc/lsb-release",
2029 "/etc/turbolinux-release",
2030 "/etc/gentoo-release",
2031 "/etc/ltib-release",
2032 "/etc/angstrom-version",
2033 "/etc/system-release",
2034 "/etc/os-release",
2035 NULL };
2036
2037 void os::Linux::print_distro_info(outputStream* st) {
2038 for (int i = 0;; i++) {
2039 const char* file = distro_files[i];
2040 if (file == NULL) {
2041 break; // done
2042 }
2043 // If file prints, we found it.
2044 if (_print_ascii_file(file, st)) {
2045 return;
2046 }
2047 }
2048
2049 if (file_exists("/etc/debian_version")) {
2050 st->print("Debian ");
2051 _print_ascii_file("/etc/debian_version", st);
2052 } else {
2053 st->print("Linux");
2054 }
2055 st->cr();
2056 }
2057
2058 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2059 char buf[256];
2060 while (fgets(buf, sizeof(buf), fp)) {
2061 // Edit out extra stuff in expected format
2062 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2063 char* ptr = strstr(buf, "\""); // the name is in quotes
2064 if (ptr != NULL) {
2065 ptr++; // go beyond first quote
2066 char* nl = strchr(ptr, '\"');
2067 if (nl != NULL) *nl = '\0';
2068 strncpy(distro, ptr, length);
2069 } else {
2070 ptr = strstr(buf, "=");
2071 ptr++; // go beyond equals then
2072 char* nl = strchr(ptr, '\n');
2073 if (nl != NULL) *nl = '\0';
2074 strncpy(distro, ptr, length);
2075 }
2076 return;
2077 } else if (get_first_line) {
2078 char* nl = strchr(buf, '\n');
2079 if (nl != NULL) *nl = '\0';
2080 strncpy(distro, buf, length);
2081 return;
2082 }
2083 }
2084 // print last line and close
2085 char* nl = strchr(buf, '\n');
2086 if (nl != NULL) *nl = '\0';
2087 strncpy(distro, buf, length);
2088 }
2089
2090 static void parse_os_info(char* distro, size_t length, const char* file) {
2091 FILE* fp = fopen(file, "r");
2092 if (fp != NULL) {
2093 // if suse format, print out first line
2094 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2095 parse_os_info_helper(fp, distro, length, get_first_line);
2096 fclose(fp);
2097 }
2098 }
2099
2100 void os::get_summary_os_info(char* buf, size_t buflen) {
2101 for (int i = 0;; i++) {
2102 const char* file = distro_files[i];
2103 if (file == NULL) {
2104 break; // ran out of distro_files
2105 }
2106 if (file_exists(file)) {
2107 parse_os_info(buf, buflen, file);
2108 return;
2109 }
2110 }
2111 // special case for debian
2112 if (file_exists("/etc/debian_version")) {
2113 strncpy(buf, "Debian ", buflen);
2114 parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2115 } else {
2116 strncpy(buf, "Linux", buflen);
2117 }
2118 }
2119
2120 void os::Linux::print_libversion_info(outputStream* st) {
2121 // libc, pthread
2122 st->print("libc:");
2123 st->print("%s ", os::Linux::glibc_version());
2124 st->print("%s ", os::Linux::libpthread_version());
2125 st->cr();
2126 }
2127
2128 void os::Linux::print_full_memory_info(outputStream* st) {
2129 st->print("\n/proc/meminfo:\n");
2130 _print_ascii_file("/proc/meminfo", st);
2131 st->cr();
2132 }
2133
2134 void os::Linux::print_container_info(outputStream* st) {
2135 if (!OSContainer::is_containerized()) {
2136 return;
2137 }
2138
2139 st->print("container (cgroup) information:\n");
2140
2141 const char *p_ct = OSContainer::container_type();
2142 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2143
2144 char *p = OSContainer::cpu_cpuset_cpus();
2145 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2146 free(p);
2147
2148 p = OSContainer::cpu_cpuset_memory_nodes();
2149 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2150 free(p);
2151
2152 int i = OSContainer::active_processor_count();
2153 if (i > 0) {
2154 st->print("active_processor_count: %d\n", i);
2155 } else {
2156 st->print("active_processor_count: failed\n");
2157 }
2158
2159 i = OSContainer::cpu_quota();
2160 st->print("cpu_quota: %d\n", i);
2161
2162 i = OSContainer::cpu_period();
2163 st->print("cpu_period: %d\n", i);
2164
2165 i = OSContainer::cpu_shares();
2166 st->print("cpu_shares: %d\n", i);
2167
2168 jlong j = OSContainer::memory_limit_in_bytes();
2169 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2170
2171 j = OSContainer::memory_and_swap_limit_in_bytes();
2172 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2173
2174 j = OSContainer::memory_soft_limit_in_bytes();
2175 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2176
2177 j = OSContainer::OSContainer::memory_usage_in_bytes();
2178 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2179
2180 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2181 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2182 st->cr();
2183 }
2184
2185 void os::print_memory_info(outputStream* st) {
2186
2187 st->print("Memory:");
2188 st->print(" %dk page", os::vm_page_size()>>10);
2189
2190 // values in struct sysinfo are "unsigned long"
2191 struct sysinfo si;
2192 sysinfo(&si);
2193
2194 st->print(", physical " UINT64_FORMAT "k",
2195 os::physical_memory() >> 10);
2196 st->print("(" UINT64_FORMAT "k free)",
2197 os::available_memory() >> 10);
2198 st->print(", swap " UINT64_FORMAT "k",
2199 ((jlong)si.totalswap * si.mem_unit) >> 10);
2200 st->print("(" UINT64_FORMAT "k free)",
2201 ((jlong)si.freeswap * si.mem_unit) >> 10);
2202 st->cr();
2203 }
2204
2205 // Print the first "model name" line and the first "flags" line
2206 // that we find and nothing more. We assume "model name" comes
2207 // before "flags" so if we find a second "model name", then the
2208 // "flags" field is considered missing.
2209 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2210 #if defined(IA32) || defined(AMD64)
2211 // Other platforms have less repetitive cpuinfo files
2212 FILE *fp = fopen("/proc/cpuinfo", "r");
2213 if (fp) {
2214 while (!feof(fp)) {
2215 if (fgets(buf, buflen, fp)) {
2216 // Assume model name comes before flags
2217 bool model_name_printed = false;
2218 if (strstr(buf, "model name") != NULL) {
2219 if (!model_name_printed) {
2220 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2221 st->print_raw(buf);
2222 model_name_printed = true;
2223 } else {
2224 // model name printed but not flags? Odd, just return
2225 fclose(fp);
2226 return true;
2227 }
2228 }
2229 // print the flags line too
2230 if (strstr(buf, "flags") != NULL) {
2231 st->print_raw(buf);
2232 fclose(fp);
2233 return true;
2234 }
2235 }
2236 }
2237 fclose(fp);
2238 }
2239 #endif // x86 platforms
2240 return false;
2241 }
2242
2243 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2244 // Only print the model name if the platform provides this as a summary
2245 if (!print_model_name_and_flags(st, buf, buflen)) {
2246 st->print("\n/proc/cpuinfo:\n");
2247 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2248 st->print_cr(" <Not Available>");
2249 }
2250 }
2251 }
2252
2253 #if defined(AMD64) || defined(IA32) || defined(X32)
2254 const char* search_string = "model name";
2255 #elif defined(M68K)
2256 const char* search_string = "CPU";
2257 #elif defined(PPC64)
2258 const char* search_string = "cpu";
2259 #elif defined(S390)
2260 const char* search_string = "processor";
2261 #elif defined(SPARC)
2262 const char* search_string = "cpu";
2263 #else
2264 const char* search_string = "Processor";
2265 #endif
2266
2267 // Parses the cpuinfo file for string representing the model name.
2268 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2269 FILE* fp = fopen("/proc/cpuinfo", "r");
2270 if (fp != NULL) {
2271 while (!feof(fp)) {
2272 char buf[256];
2273 if (fgets(buf, sizeof(buf), fp)) {
2274 char* start = strstr(buf, search_string);
2275 if (start != NULL) {
2276 char *ptr = start + strlen(search_string);
2277 char *end = buf + strlen(buf);
2278 while (ptr != end) {
2279 // skip whitespace and colon for the rest of the name.
2280 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2281 break;
2282 }
2283 ptr++;
2284 }
2285 if (ptr != end) {
2286 // reasonable string, get rid of newline and keep the rest
2287 char* nl = strchr(buf, '\n');
2288 if (nl != NULL) *nl = '\0';
2289 strncpy(cpuinfo, ptr, length);
2290 fclose(fp);
2291 return;
2292 }
2293 }
2294 }
2295 }
2296 fclose(fp);
2297 }
2298 // cpuinfo not found or parsing failed, just print generic string. The entire
2299 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2300 #if defined(AARCH64)
2301 strncpy(cpuinfo, "AArch64", length);
2302 #elif defined(AMD64)
2303 strncpy(cpuinfo, "x86_64", length);
2304 #elif defined(ARM) // Order wrt. AARCH64 is relevant!
2305 strncpy(cpuinfo, "ARM", length);
2306 #elif defined(IA32)
2307 strncpy(cpuinfo, "x86_32", length);
2308 #elif defined(IA64)
2309 strncpy(cpuinfo, "IA64", length);
2310 #elif defined(PPC)
2311 strncpy(cpuinfo, "PPC64", length);
2312 #elif defined(S390)
2313 strncpy(cpuinfo, "S390", length);
2314 #elif defined(SPARC)
2315 strncpy(cpuinfo, "sparcv9", length);
2316 #elif defined(ZERO_LIBARCH)
2317 strncpy(cpuinfo, ZERO_LIBARCH, length);
2318 #else
2319 strncpy(cpuinfo, "unknown", length);
2320 #endif
2321 }
2322
2323 static void print_signal_handler(outputStream* st, int sig,
2324 char* buf, size_t buflen);
2325
2326 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2327 st->print_cr("Signal Handlers:");
2328 print_signal_handler(st, SIGSEGV, buf, buflen);
2329 print_signal_handler(st, SIGBUS , buf, buflen);
2330 print_signal_handler(st, SIGFPE , buf, buflen);
2331 print_signal_handler(st, SIGPIPE, buf, buflen);
2332 print_signal_handler(st, SIGXFSZ, buf, buflen);
2333 print_signal_handler(st, SIGILL , buf, buflen);
2334 print_signal_handler(st, SR_signum, buf, buflen);
2335 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2336 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2337 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2338 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2339 #if defined(PPC64)
2340 print_signal_handler(st, SIGTRAP, buf, buflen);
2341 #endif
2342 }
2343
2344 static char saved_jvm_path[MAXPATHLEN] = {0};
2345
2346 // Find the full path to the current module, libjvm.so
2347 void os::jvm_path(char *buf, jint buflen) {
2348 // Error checking.
2349 if (buflen < MAXPATHLEN) {
2350 assert(false, "must use a large-enough buffer");
2351 buf[0] = '\0';
2352 return;
2353 }
2354 // Lazy resolve the path to current module.
2355 if (saved_jvm_path[0] != 0) {
2356 strcpy(buf, saved_jvm_path);
2357 return;
2358 }
2359
2360 char dli_fname[MAXPATHLEN];
2361 bool ret = dll_address_to_library_name(
2362 CAST_FROM_FN_PTR(address, os::jvm_path),
2363 dli_fname, sizeof(dli_fname), NULL);
2364 assert(ret, "cannot locate libjvm");
2365 char *rp = NULL;
2366 if (ret && dli_fname[0] != '\0') {
2367 rp = os::Posix::realpath(dli_fname, buf, buflen);
2368 }
2369 if (rp == NULL) {
2370 return;
2371 }
2372
2373 if (Arguments::sun_java_launcher_is_altjvm()) {
2374 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2375 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2376 // If "/jre/lib/" appears at the right place in the string, then
2377 // assume we are installed in a JDK and we're done. Otherwise, check
2378 // for a JAVA_HOME environment variable and fix up the path so it
2379 // looks like libjvm.so is installed there (append a fake suffix
2380 // hotspot/libjvm.so).
2381 const char *p = buf + strlen(buf) - 1;
2382 for (int count = 0; p > buf && count < 5; ++count) {
2383 for (--p; p > buf && *p != '/'; --p)
2384 /* empty */ ;
2385 }
2386
2387 if (strncmp(p, "/jre/lib/", 9) != 0) {
2388 // Look for JAVA_HOME in the environment.
2389 char* java_home_var = ::getenv("JAVA_HOME");
2390 if (java_home_var != NULL && java_home_var[0] != 0) {
2391 char* jrelib_p;
2392 int len;
2393
2394 // Check the current module name "libjvm.so".
2395 p = strrchr(buf, '/');
2396 if (p == NULL) {
2397 return;
2398 }
2399 assert(strstr(p, "/libjvm") == p, "invalid library name");
2400
2401 rp = os::Posix::realpath(java_home_var, buf, buflen);
2402 if (rp == NULL) {
2403 return;
2404 }
2405
2406 // determine if this is a legacy image or modules image
2407 // modules image doesn't have "jre" subdirectory
2408 len = strlen(buf);
2409 assert(len < buflen, "Ran out of buffer room");
2410 jrelib_p = buf + len;
2411 snprintf(jrelib_p, buflen-len, "/jre/lib");
2412 if (0 != access(buf, F_OK)) {
2413 snprintf(jrelib_p, buflen-len, "/lib");
2414 }
2415
2416 if (0 == access(buf, F_OK)) {
2417 // Use current module name "libjvm.so"
2418 len = strlen(buf);
2419 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2420 } else {
2421 // Go back to path of .so
2422 rp = os::Posix::realpath(dli_fname, buf, buflen);
2423 if (rp == NULL) {
2424 return;
2425 }
2426 }
2427 }
2428 }
2429 }
2430
2431 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2432 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2433 }
2434
2435 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2436 // no prefix required, not even "_"
2437 }
2438
2439 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2440 // no suffix required
2441 }
2442
2443 ////////////////////////////////////////////////////////////////////////////////
2444 // sun.misc.Signal support
2445
2446 static volatile jint sigint_count = 0;
2447
2448 static void UserHandler(int sig, void *siginfo, void *context) {
2449 // 4511530 - sem_post is serialized and handled by the manager thread. When
2450 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2451 // don't want to flood the manager thread with sem_post requests.
2452 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2453 return;
2454 }
2455
2456 // Ctrl-C is pressed during error reporting, likely because the error
2457 // handler fails to abort. Let VM die immediately.
2458 if (sig == SIGINT && VMError::is_error_reported()) {
2459 os::die();
2460 }
2461
2462 os::signal_notify(sig);
2463 }
2464
2465 void* os::user_handler() {
2466 return CAST_FROM_FN_PTR(void*, UserHandler);
2467 }
2468
2469 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2470 struct timespec ts;
2471 // Semaphore's are always associated with CLOCK_REALTIME
2472 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2473 // see os_posix.cpp for discussion on overflow checking
2474 if (sec >= MAX_SECS) {
2475 ts.tv_sec += MAX_SECS;
2476 ts.tv_nsec = 0;
2477 } else {
2478 ts.tv_sec += sec;
2479 ts.tv_nsec += nsec;
2480 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2481 ts.tv_nsec -= NANOSECS_PER_SEC;
2482 ++ts.tv_sec; // note: this must be <= max_secs
2483 }
2484 }
2485
2486 return ts;
2487 }
2488
2489 extern "C" {
2490 typedef void (*sa_handler_t)(int);
2491 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2492 }
2493
2494 void* os::signal(int signal_number, void* handler) {
2495 struct sigaction sigAct, oldSigAct;
2496
2497 sigfillset(&(sigAct.sa_mask));
2498 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2499 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2500
2501 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2502 // -1 means registration failed
2503 return (void *)-1;
2504 }
2505
2506 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2507 }
2508
2509 void os::signal_raise(int signal_number) {
2510 ::raise(signal_number);
2511 }
2512
2513 // The following code is moved from os.cpp for making this
2514 // code platform specific, which it is by its very nature.
2515
2516 // Will be modified when max signal is changed to be dynamic
2517 int os::sigexitnum_pd() {
2518 return NSIG;
2519 }
2520
2521 // a counter for each possible signal value
2522 static volatile jint pending_signals[NSIG+1] = { 0 };
2523
2524 // Linux(POSIX) specific hand shaking semaphore.
2525 static Semaphore* sig_sem = NULL;
2526 static PosixSemaphore sr_semaphore;
2527
2528 void os::signal_init_pd() {
2529 // Initialize signal structures
2530 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2531
2532 // Initialize signal semaphore
2533 sig_sem = new Semaphore();
2534 }
2535
2536 void os::signal_notify(int sig) {
2537 if (sig_sem != NULL) {
2538 Atomic::inc(&pending_signals[sig]);
2539 sig_sem->signal();
2540 } else {
2541 // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2542 // initialization isn't called.
2543 assert(ReduceSignalUsage, "signal semaphore should be created");
2544 }
2545 }
2546
2547 static int check_pending_signals() {
2548 Atomic::store(0, &sigint_count);
2549 for (;;) {
2550 for (int i = 0; i < NSIG + 1; i++) {
2551 jint n = pending_signals[i];
2552 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2553 return i;
2554 }
2555 }
2556 JavaThread *thread = JavaThread::current();
2557 ThreadBlockInVM tbivm(thread);
2558
2559 bool threadIsSuspended;
2560 do {
2561 thread->set_suspend_equivalent();
2562 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2563 sig_sem->wait();
2564
2565 // were we externally suspended while we were waiting?
2566 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2567 if (threadIsSuspended) {
2568 // The semaphore has been incremented, but while we were waiting
2569 // another thread suspended us. We don't want to continue running
2570 // while suspended because that would surprise the thread that
2571 // suspended us.
2572 sig_sem->signal();
2573
2574 thread->java_suspend_self();
2575 }
2576 } while (threadIsSuspended);
2577 }
2578 }
2579
2580 int os::signal_wait() {
2581 return check_pending_signals();
2582 }
2583
2584 ////////////////////////////////////////////////////////////////////////////////
2585 // Virtual Memory
2586
2587 int os::vm_page_size() {
2588 // Seems redundant as all get out
2589 assert(os::Linux::page_size() != -1, "must call os::init");
2590 return os::Linux::page_size();
2591 }
2592
2593 // Solaris allocates memory by pages.
2594 int os::vm_allocation_granularity() {
2595 assert(os::Linux::page_size() != -1, "must call os::init");
2596 return os::Linux::page_size();
2597 }
2598
2599 // Rationale behind this function:
2600 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2601 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2602 // samples for JITted code. Here we create private executable mapping over the code cache
2603 // and then we can use standard (well, almost, as mapping can change) way to provide
2604 // info for the reporting script by storing timestamp and location of symbol
2605 void linux_wrap_code(char* base, size_t size) {
2606 static volatile jint cnt = 0;
2607
2608 if (!UseOprofile) {
2609 return;
2610 }
2611
2612 char buf[PATH_MAX+1];
2613 int num = Atomic::add(1, &cnt);
2614
2615 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2616 os::get_temp_directory(), os::current_process_id(), num);
2617 unlink(buf);
2618
2619 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2620
2621 if (fd != -1) {
2622 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2623 if (rv != (off_t)-1) {
2624 if (::write(fd, "", 1) == 1) {
2625 mmap(base, size,
2626 PROT_READ|PROT_WRITE|PROT_EXEC,
2627 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2628 }
2629 }
2630 ::close(fd);
2631 unlink(buf);
2632 }
2633 }
2634
2635 static bool recoverable_mmap_error(int err) {
2636 // See if the error is one we can let the caller handle. This
2637 // list of errno values comes from JBS-6843484. I can't find a
2638 // Linux man page that documents this specific set of errno
2639 // values so while this list currently matches Solaris, it may
2640 // change as we gain experience with this failure mode.
2641 switch (err) {
2642 case EBADF:
2643 case EINVAL:
2644 case ENOTSUP:
2645 // let the caller deal with these errors
2646 return true;
2647
2648 default:
2649 // Any remaining errors on this OS can cause our reserved mapping
2650 // to be lost. That can cause confusion where different data
2651 // structures think they have the same memory mapped. The worst
2652 // scenario is if both the VM and a library think they have the
2653 // same memory mapped.
2654 return false;
2655 }
2656 }
2657
2658 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2659 int err) {
2660 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2661 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2662 os::strerror(err), err);
2663 }
2664
2665 static void warn_fail_commit_memory(char* addr, size_t size,
2666 size_t alignment_hint, bool exec,
2667 int err) {
2668 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2669 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2670 alignment_hint, exec, os::strerror(err), err);
2671 }
2672
2673 // NOTE: Linux kernel does not really reserve the pages for us.
2674 // All it does is to check if there are enough free pages
2675 // left at the time of mmap(). This could be a potential
2676 // problem.
2677 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2678 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2679 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2680 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2681 if (res != (uintptr_t) MAP_FAILED) {
2682 if (UseNUMAInterleaving) {
2683 numa_make_global(addr, size);
2684 }
2685 return 0;
2686 }
2687
2688 int err = errno; // save errno from mmap() call above
2689
2690 if (!recoverable_mmap_error(err)) {
2691 warn_fail_commit_memory(addr, size, exec, err);
2692 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2693 }
2694
2695 return err;
2696 }
2697
2698 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2699 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2700 }
2701
2702 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2703 const char* mesg) {
2704 assert(mesg != NULL, "mesg must be specified");
2705 int err = os::Linux::commit_memory_impl(addr, size, exec);
2706 if (err != 0) {
2707 // the caller wants all commit errors to exit with the specified mesg:
2708 warn_fail_commit_memory(addr, size, exec, err);
2709 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2710 }
2711 }
2712
2713 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2714 #ifndef MAP_HUGETLB
2715 #define MAP_HUGETLB 0x40000
2716 #endif
2717
2718 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2719 #ifndef MADV_HUGEPAGE
2720 #define MADV_HUGEPAGE 14
2721 #endif
2722
2723 int os::Linux::commit_memory_impl(char* addr, size_t size,
2724 size_t alignment_hint, bool exec) {
2725 int err = os::Linux::commit_memory_impl(addr, size, exec);
2726 if (err == 0) {
2727 realign_memory(addr, size, alignment_hint);
2728 }
2729 return err;
2730 }
2731
2732 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2733 bool exec) {
2734 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2735 }
2736
2737 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2738 size_t alignment_hint, bool exec,
2739 const char* mesg) {
2740 assert(mesg != NULL, "mesg must be specified");
2741 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2742 if (err != 0) {
2743 // the caller wants all commit errors to exit with the specified mesg:
2744 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2745 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2746 }
2747 }
2748
2749 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2750 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2751 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2752 // be supported or the memory may already be backed by huge pages.
2753 ::madvise(addr, bytes, MADV_HUGEPAGE);
2754 }
2755 }
2756
2757 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2758 // This method works by doing an mmap over an existing mmaping and effectively discarding
2759 // the existing pages. However it won't work for SHM-based large pages that cannot be
2760 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2761 // small pages on top of the SHM segment. This method always works for small pages, so we
2762 // allow that in any case.
2763 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2764 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2765 }
2766 }
2767
2768 void os::numa_make_global(char *addr, size_t bytes) {
2769 Linux::numa_interleave_memory(addr, bytes);
2770 }
2771
2772 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2773 // bind policy to MPOL_PREFERRED for the current thread.
2774 #define USE_MPOL_PREFERRED 0
2775
2776 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2777 // To make NUMA and large pages more robust when both enabled, we need to ease
2778 // the requirements on where the memory should be allocated. MPOL_BIND is the
2779 // default policy and it will force memory to be allocated on the specified
2780 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2781 // the specified node, but will not force it. Using this policy will prevent
2782 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2783 // free large pages.
2784 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2785 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2786 }
2787
2788 bool os::numa_topology_changed() { return false; }
2789
2790 size_t os::numa_get_groups_num() {
2791 // Return just the number of nodes in which it's possible to allocate memory
2792 // (in numa terminology, configured nodes).
2793 return Linux::numa_num_configured_nodes();
2794 }
2795
2796 int os::numa_get_group_id() {
2797 int cpu_id = Linux::sched_getcpu();
2798 if (cpu_id != -1) {
2799 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2800 if (lgrp_id != -1) {
2801 return lgrp_id;
2802 }
2803 }
2804 return 0;
2805 }
2806
2807 int os::Linux::get_existing_num_nodes() {
2808 size_t node;
2809 size_t highest_node_number = Linux::numa_max_node();
2810 int num_nodes = 0;
2811
2812 // Get the total number of nodes in the system including nodes without memory.
2813 for (node = 0; node <= highest_node_number; node++) {
2814 if (isnode_in_existing_nodes(node)) {
2815 num_nodes++;
2816 }
2817 }
2818 return num_nodes;
2819 }
2820
2821 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2822 size_t highest_node_number = Linux::numa_max_node();
2823 size_t i = 0;
2824
2825 // Map all node ids in which is possible to allocate memory. Also nodes are
2826 // not always consecutively available, i.e. available from 0 to the highest
2827 // node number.
2828 for (size_t node = 0; node <= highest_node_number; node++) {
2829 if (Linux::isnode_in_configured_nodes(node)) {
2830 ids[i++] = node;
2831 }
2832 }
2833 return i;
2834 }
2835
2836 bool os::get_page_info(char *start, page_info* info) {
2837 return false;
2838 }
2839
2840 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2841 page_info* page_found) {
2842 return end;
2843 }
2844
2845
2846 int os::Linux::sched_getcpu_syscall(void) {
2847 unsigned int cpu = 0;
2848 int retval = -1;
2849
2850 #if defined(IA32)
2851 #ifndef SYS_getcpu
2852 #define SYS_getcpu 318
2853 #endif
2854 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2855 #elif defined(AMD64)
2856 // Unfortunately we have to bring all these macros here from vsyscall.h
2857 // to be able to compile on old linuxes.
2858 #define __NR_vgetcpu 2
2859 #define VSYSCALL_START (-10UL << 20)
2860 #define VSYSCALL_SIZE 1024
2861 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2862 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2863 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2864 retval = vgetcpu(&cpu, NULL, NULL);
2865 #endif
2866
2867 return (retval == -1) ? retval : cpu;
2868 }
2869
2870 void os::Linux::sched_getcpu_init() {
2871 // sched_getcpu() should be in libc.
2872 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2873 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2874
2875 // If it's not, try a direct syscall.
2876 if (sched_getcpu() == -1) {
2877 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2878 (void*)&sched_getcpu_syscall));
2879 }
2880 }
2881
2882 // Something to do with the numa-aware allocator needs these symbols
2883 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2884 extern "C" JNIEXPORT void numa_error(char *where) { }
2885
2886 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2887 // load symbol from base version instead.
2888 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2889 void *f = dlvsym(handle, name, "libnuma_1.1");
2890 if (f == NULL) {
2891 f = dlsym(handle, name);
2892 }
2893 return f;
2894 }
2895
2896 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2897 // Return NULL if the symbol is not defined in this particular version.
2898 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2899 return dlvsym(handle, name, "libnuma_1.2");
2900 }
2901
2902 bool os::Linux::libnuma_init() {
2903 if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2904 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2905 if (handle != NULL) {
2906 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2907 libnuma_dlsym(handle, "numa_node_to_cpus")));
2908 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2909 libnuma_dlsym(handle, "numa_max_node")));
2910 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2911 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2912 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2913 libnuma_dlsym(handle, "numa_available")));
2914 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2915 libnuma_dlsym(handle, "numa_tonode_memory")));
2916 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2917 libnuma_dlsym(handle, "numa_interleave_memory")));
2918 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2919 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2920 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2921 libnuma_dlsym(handle, "numa_set_bind_policy")));
2922 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2923 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2924 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2925 libnuma_dlsym(handle, "numa_distance")));
2926
2927 if (numa_available() != -1) {
2928 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2929 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2930 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2931 // Create an index -> node mapping, since nodes are not always consecutive
2932 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2933 rebuild_nindex_to_node_map();
2934 // Create a cpu -> node mapping
2935 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2936 rebuild_cpu_to_node_map();
2937 return true;
2938 }
2939 }
2940 }
2941 return false;
2942 }
2943
2944 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2945 // Creating guard page is very expensive. Java thread has HotSpot
2946 // guard pages, only enable glibc guard page for non-Java threads.
2947 // (Remember: compiler thread is a Java thread, too!)
2948 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2949 }
2950
2951 void os::Linux::rebuild_nindex_to_node_map() {
2952 int highest_node_number = Linux::numa_max_node();
2953
2954 nindex_to_node()->clear();
2955 for (int node = 0; node <= highest_node_number; node++) {
2956 if (Linux::isnode_in_existing_nodes(node)) {
2957 nindex_to_node()->append(node);
2958 }
2959 }
2960 }
2961
2962 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2963 // The table is later used in get_node_by_cpu().
2964 void os::Linux::rebuild_cpu_to_node_map() {
2965 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2966 // in libnuma (possible values are starting from 16,
2967 // and continuing up with every other power of 2, but less
2968 // than the maximum number of CPUs supported by kernel), and
2969 // is a subject to change (in libnuma version 2 the requirements
2970 // are more reasonable) we'll just hardcode the number they use
2971 // in the library.
2972 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2973
2974 size_t cpu_num = processor_count();
2975 size_t cpu_map_size = NCPUS / BitsPerCLong;
2976 size_t cpu_map_valid_size =
2977 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2978
2979 cpu_to_node()->clear();
2980 cpu_to_node()->at_grow(cpu_num - 1);
2981
2982 size_t node_num = get_existing_num_nodes();
2983
2984 int distance = 0;
2985 int closest_distance = INT_MAX;
2986 int closest_node = 0;
2987 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2988 for (size_t i = 0; i < node_num; i++) {
2989 // Check if node is configured (not a memory-less node). If it is not, find
2990 // the closest configured node.
2991 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2992 closest_distance = INT_MAX;
2993 // Check distance from all remaining nodes in the system. Ignore distance
2994 // from itself and from another non-configured node.
2995 for (size_t m = 0; m < node_num; m++) {
2996 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2997 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2998 // If a closest node is found, update. There is always at least one
2999 // configured node in the system so there is always at least one node
3000 // close.
3001 if (distance != 0 && distance < closest_distance) {
3002 closest_distance = distance;
3003 closest_node = nindex_to_node()->at(m);
3004 }
3005 }
3006 }
3007 } else {
3008 // Current node is already a configured node.
3009 closest_node = nindex_to_node()->at(i);
3010 }
3011
3012 // Get cpus from the original node and map them to the closest node. If node
3013 // is a configured node (not a memory-less node), then original node and
3014 // closest node are the same.
3015 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3016 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3017 if (cpu_map[j] != 0) {
3018 for (size_t k = 0; k < BitsPerCLong; k++) {
3019 if (cpu_map[j] & (1UL << k)) {
3020 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3021 }
3022 }
3023 }
3024 }
3025 }
3026 }
3027 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3028 }
3029
3030 int os::Linux::get_node_by_cpu(int cpu_id) {
3031 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3032 return cpu_to_node()->at(cpu_id);
3033 }
3034 return -1;
3035 }
3036
3037 GrowableArray<int>* os::Linux::_cpu_to_node;
3038 GrowableArray<int>* os::Linux::_nindex_to_node;
3039 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3040 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3041 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3042 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3043 os::Linux::numa_available_func_t os::Linux::_numa_available;
3044 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3045 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3046 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3047 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3048 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3049 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3050 unsigned long* os::Linux::_numa_all_nodes;
3051 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3052 struct bitmask* os::Linux::_numa_nodes_ptr;
3053
3054 bool os::pd_uncommit_memory(char* addr, size_t size) {
3055 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3056 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3057 return res != (uintptr_t) MAP_FAILED;
3058 }
3059
3060 static address get_stack_commited_bottom(address bottom, size_t size) {
3061 address nbot = bottom;
3062 address ntop = bottom + size;
3063
3064 size_t page_sz = os::vm_page_size();
3065 unsigned pages = size / page_sz;
3066
3067 unsigned char vec[1];
3068 unsigned imin = 1, imax = pages + 1, imid;
3069 int mincore_return_value = 0;
3070
3071 assert(imin <= imax, "Unexpected page size");
3072
3073 while (imin < imax) {
3074 imid = (imax + imin) / 2;
3075 nbot = ntop - (imid * page_sz);
3076
3077 // Use a trick with mincore to check whether the page is mapped or not.
3078 // mincore sets vec to 1 if page resides in memory and to 0 if page
3079 // is swapped output but if page we are asking for is unmapped
3080 // it returns -1,ENOMEM
3081 mincore_return_value = mincore(nbot, page_sz, vec);
3082
3083 if (mincore_return_value == -1) {
3084 // Page is not mapped go up
3085 // to find first mapped page
3086 if (errno != EAGAIN) {
3087 assert(errno == ENOMEM, "Unexpected mincore errno");
3088 imax = imid;
3089 }
3090 } else {
3091 // Page is mapped go down
3092 // to find first not mapped page
3093 imin = imid + 1;
3094 }
3095 }
3096
3097 nbot = nbot + page_sz;
3098
3099 // Adjust stack bottom one page up if last checked page is not mapped
3100 if (mincore_return_value == -1) {
3101 nbot = nbot + page_sz;
3102 }
3103
3104 return nbot;
3105 }
3106
3107
3108 // Linux uses a growable mapping for the stack, and if the mapping for
3109 // the stack guard pages is not removed when we detach a thread the
3110 // stack cannot grow beyond the pages where the stack guard was
3111 // mapped. If at some point later in the process the stack expands to
3112 // that point, the Linux kernel cannot expand the stack any further
3113 // because the guard pages are in the way, and a segfault occurs.
3114 //
3115 // However, it's essential not to split the stack region by unmapping
3116 // a region (leaving a hole) that's already part of the stack mapping,
3117 // so if the stack mapping has already grown beyond the guard pages at
3118 // the time we create them, we have to truncate the stack mapping.
3119 // So, we need to know the extent of the stack mapping when
3120 // create_stack_guard_pages() is called.
3121
3122 // We only need this for stacks that are growable: at the time of
3123 // writing thread stacks don't use growable mappings (i.e. those
3124 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3125 // only applies to the main thread.
3126
3127 // If the (growable) stack mapping already extends beyond the point
3128 // where we're going to put our guard pages, truncate the mapping at
3129 // that point by munmap()ping it. This ensures that when we later
3130 // munmap() the guard pages we don't leave a hole in the stack
3131 // mapping. This only affects the main/primordial thread
3132
3133 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3134 if (os::is_primordial_thread()) {
3135 // As we manually grow stack up to bottom inside create_attached_thread(),
3136 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3137 // we don't need to do anything special.
3138 // Check it first, before calling heavy function.
3139 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3140 unsigned char vec[1];
3141
3142 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3143 // Fallback to slow path on all errors, including EAGAIN
3144 stack_extent = (uintptr_t) get_stack_commited_bottom(
3145 os::Linux::initial_thread_stack_bottom(),
3146 (size_t)addr - stack_extent);
3147 }
3148
3149 if (stack_extent < (uintptr_t)addr) {
3150 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3151 }
3152 }
3153
3154 return os::commit_memory(addr, size, !ExecMem);
3155 }
3156
3157 // If this is a growable mapping, remove the guard pages entirely by
3158 // munmap()ping them. If not, just call uncommit_memory(). This only
3159 // affects the main/primordial thread, but guard against future OS changes.
3160 // It's safe to always unmap guard pages for primordial thread because we
3161 // always place it right after end of the mapped region.
3162
3163 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3164 uintptr_t stack_extent, stack_base;
3165
3166 if (os::is_primordial_thread()) {
3167 return ::munmap(addr, size) == 0;
3168 }
3169
3170 return os::uncommit_memory(addr, size);
3171 }
3172
3173 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3174 // at 'requested_addr'. If there are existing memory mappings at the same
3175 // location, however, they will be overwritten. If 'fixed' is false,
3176 // 'requested_addr' is only treated as a hint, the return value may or
3177 // may not start from the requested address. Unlike Linux mmap(), this
3178 // function returns NULL to indicate failure.
3179 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3180 char * addr;
3181 int flags;
3182
3183 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3184 if (fixed) {
3185 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3186 flags |= MAP_FIXED;
3187 }
3188
3189 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3190 // touch an uncommitted page. Otherwise, the read/write might
3191 // succeed if we have enough swap space to back the physical page.
3192 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3193 flags, -1, 0);
3194
3195 return addr == MAP_FAILED ? NULL : addr;
3196 }
3197
3198 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3199 // (req_addr != NULL) or with a given alignment.
3200 // - bytes shall be a multiple of alignment.
3201 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3202 // - alignment sets the alignment at which memory shall be allocated.
3203 // It must be a multiple of allocation granularity.
3204 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3205 // req_addr or NULL.
3206 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3207
3208 size_t extra_size = bytes;
3209 if (req_addr == NULL && alignment > 0) {
3210 extra_size += alignment;
3211 }
3212
3213 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3214 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3215 -1, 0);
3216 if (start == MAP_FAILED) {
3217 start = NULL;
3218 } else {
3219 if (req_addr != NULL) {
3220 if (start != req_addr) {
3221 ::munmap(start, extra_size);
3222 start = NULL;
3223 }
3224 } else {
3225 char* const start_aligned = align_up(start, alignment);
3226 char* const end_aligned = start_aligned + bytes;
3227 char* const end = start + extra_size;
3228 if (start_aligned > start) {
3229 ::munmap(start, start_aligned - start);
3230 }
3231 if (end_aligned < end) {
3232 ::munmap(end_aligned, end - end_aligned);
3233 }
3234 start = start_aligned;
3235 }
3236 }
3237 return start;
3238 }
3239
3240 static int anon_munmap(char * addr, size_t size) {
3241 return ::munmap(addr, size) == 0;
3242 }
3243
3244 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3245 size_t alignment_hint) {
3246 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3247 }
3248
3249 bool os::pd_release_memory(char* addr, size_t size) {
3250 return anon_munmap(addr, size);
3251 }
3252
3253 static bool linux_mprotect(char* addr, size_t size, int prot) {
3254 // Linux wants the mprotect address argument to be page aligned.
3255 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3256
3257 // According to SUSv3, mprotect() should only be used with mappings
3258 // established by mmap(), and mmap() always maps whole pages. Unaligned
3259 // 'addr' likely indicates problem in the VM (e.g. trying to change
3260 // protection of malloc'ed or statically allocated memory). Check the
3261 // caller if you hit this assert.
3262 assert(addr == bottom, "sanity check");
3263
3264 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3265 return ::mprotect(bottom, size, prot) == 0;
3266 }
3267
3268 // Set protections specified
3269 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3270 bool is_committed) {
3271 unsigned int p = 0;
3272 switch (prot) {
3273 case MEM_PROT_NONE: p = PROT_NONE; break;
3274 case MEM_PROT_READ: p = PROT_READ; break;
3275 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3276 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3277 default:
3278 ShouldNotReachHere();
3279 }
3280 // is_committed is unused.
3281 return linux_mprotect(addr, bytes, p);
3282 }
3283
3284 bool os::guard_memory(char* addr, size_t size) {
3285 return linux_mprotect(addr, size, PROT_NONE);
3286 }
3287
3288 bool os::unguard_memory(char* addr, size_t size) {
3289 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3290 }
3291
3292 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3293 size_t page_size) {
3294 bool result = false;
3295 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3296 MAP_ANONYMOUS|MAP_PRIVATE,
3297 -1, 0);
3298 if (p != MAP_FAILED) {
3299 void *aligned_p = align_up(p, page_size);
3300
3301 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3302
3303 munmap(p, page_size * 2);
3304 }
3305
3306 if (warn && !result) {
3307 warning("TransparentHugePages is not supported by the operating system.");
3308 }
3309
3310 return result;
3311 }
3312
3313 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3314 bool result = false;
3315 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3316 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3317 -1, 0);
3318
3319 if (p != MAP_FAILED) {
3320 // We don't know if this really is a huge page or not.
3321 FILE *fp = fopen("/proc/self/maps", "r");
3322 if (fp) {
3323 while (!feof(fp)) {
3324 char chars[257];
3325 long x = 0;
3326 if (fgets(chars, sizeof(chars), fp)) {
3327 if (sscanf(chars, "%lx-%*x", &x) == 1
3328 && x == (long)p) {
3329 if (strstr (chars, "hugepage")) {
3330 result = true;
3331 break;
3332 }
3333 }
3334 }
3335 }
3336 fclose(fp);
3337 }
3338 munmap(p, page_size);
3339 }
3340
3341 if (warn && !result) {
3342 warning("HugeTLBFS is not supported by the operating system.");
3343 }
3344
3345 return result;
3346 }
3347
3348 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3349 //
3350 // From the coredump_filter documentation:
3351 //
3352 // - (bit 0) anonymous private memory
3353 // - (bit 1) anonymous shared memory
3354 // - (bit 2) file-backed private memory
3355 // - (bit 3) file-backed shared memory
3356 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3357 // effective only if the bit 2 is cleared)
3358 // - (bit 5) hugetlb private memory
3359 // - (bit 6) hugetlb shared memory
3360 // - (bit 7) dax private memory
3361 // - (bit 8) dax shared memory
3362 //
3363 static void set_coredump_filter(bool largepages, bool dax_shared) {
3364 FILE *f;
3365 long cdm;
3366 bool filter_changed = false;
3367
3368 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3369 return;
3370 }
3371
3372 if (fscanf(f, "%lx", &cdm) != 1) {
3373 fclose(f);
3374 return;
3375 }
3376
3377 rewind(f);
3378
3379 if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3380 cdm |= LARGEPAGES_BIT;
3381 filter_changed = true;
3382 }
3383 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3384 cdm |= DAX_SHARED_BIT;
3385 filter_changed = true;
3386 }
3387 if (filter_changed) {
3388 fprintf(f, "%#lx", cdm);
3389 }
3390
3391 fclose(f);
3392 }
3393
3394 // Large page support
3395
3396 static size_t _large_page_size = 0;
3397
3398 size_t os::Linux::find_large_page_size() {
3399 size_t large_page_size = 0;
3400
3401 // large_page_size on Linux is used to round up heap size. x86 uses either
3402 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3403 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3404 // page as large as 256M.
3405 //
3406 // Here we try to figure out page size by parsing /proc/meminfo and looking
3407 // for a line with the following format:
3408 // Hugepagesize: 2048 kB
3409 //
3410 // If we can't determine the value (e.g. /proc is not mounted, or the text
3411 // format has been changed), we'll use the largest page size supported by
3412 // the processor.
3413
3414 #ifndef ZERO
3415 large_page_size =
3416 AARCH64_ONLY(2 * M)
3417 AMD64_ONLY(2 * M)
3418 ARM32_ONLY(2 * M)
3419 IA32_ONLY(4 * M)
3420 IA64_ONLY(256 * M)
3421 PPC_ONLY(4 * M)
3422 S390_ONLY(1 * M)
3423 SPARC_ONLY(4 * M);
3424 #endif // ZERO
3425
3426 FILE *fp = fopen("/proc/meminfo", "r");
3427 if (fp) {
3428 while (!feof(fp)) {
3429 int x = 0;
3430 char buf[16];
3431 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3432 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3433 large_page_size = x * K;
3434 break;
3435 }
3436 } else {
3437 // skip to next line
3438 for (;;) {
3439 int ch = fgetc(fp);
3440 if (ch == EOF || ch == (int)'\n') break;
3441 }
3442 }
3443 }
3444 fclose(fp);
3445 }
3446
3447 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3448 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3449 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3450 proper_unit_for_byte_size(large_page_size));
3451 }
3452
3453 return large_page_size;
3454 }
3455
3456 size_t os::Linux::setup_large_page_size() {
3457 _large_page_size = Linux::find_large_page_size();
3458 const size_t default_page_size = (size_t)Linux::page_size();
3459 if (_large_page_size > default_page_size) {
3460 _page_sizes[0] = _large_page_size;
3461 _page_sizes[1] = default_page_size;
3462 _page_sizes[2] = 0;
3463 }
3464
3465 return _large_page_size;
3466 }
3467
3468 bool os::Linux::setup_large_page_type(size_t page_size) {
3469 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3470 FLAG_IS_DEFAULT(UseSHM) &&
3471 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3472
3473 // The type of large pages has not been specified by the user.
3474
3475 // Try UseHugeTLBFS and then UseSHM.
3476 UseHugeTLBFS = UseSHM = true;
3477
3478 // Don't try UseTransparentHugePages since there are known
3479 // performance issues with it turned on. This might change in the future.
3480 UseTransparentHugePages = false;
3481 }
3482
3483 if (UseTransparentHugePages) {
3484 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3485 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3486 UseHugeTLBFS = false;
3487 UseSHM = false;
3488 return true;
3489 }
3490 UseTransparentHugePages = false;
3491 }
3492
3493 if (UseHugeTLBFS) {
3494 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3495 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3496 UseSHM = false;
3497 return true;
3498 }
3499 UseHugeTLBFS = false;
3500 }
3501
3502 return UseSHM;
3503 }
3504
3505 void os::large_page_init() {
3506 if (!UseLargePages &&
3507 !UseTransparentHugePages &&
3508 !UseHugeTLBFS &&
3509 !UseSHM) {
3510 // Not using large pages.
3511 return;
3512 }
3513
3514 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3515 // The user explicitly turned off large pages.
3516 // Ignore the rest of the large pages flags.
3517 UseTransparentHugePages = false;
3518 UseHugeTLBFS = false;
3519 UseSHM = false;
3520 return;
3521 }
3522
3523 size_t large_page_size = Linux::setup_large_page_size();
3524 UseLargePages = Linux::setup_large_page_type(large_page_size);
3525
3526 set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3527 }
3528
3529 #ifndef SHM_HUGETLB
3530 #define SHM_HUGETLB 04000
3531 #endif
3532
3533 #define shm_warning_format(format, ...) \
3534 do { \
3535 if (UseLargePages && \
3536 (!FLAG_IS_DEFAULT(UseLargePages) || \
3537 !FLAG_IS_DEFAULT(UseSHM) || \
3538 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3539 warning(format, __VA_ARGS__); \
3540 } \
3541 } while (0)
3542
3543 #define shm_warning(str) shm_warning_format("%s", str)
3544
3545 #define shm_warning_with_errno(str) \
3546 do { \
3547 int err = errno; \
3548 shm_warning_format(str " (error = %d)", err); \
3549 } while (0)
3550
3551 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3552 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3553
3554 if (!is_aligned(alignment, SHMLBA)) {
3555 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3556 return NULL;
3557 }
3558
3559 // To ensure that we get 'alignment' aligned memory from shmat,
3560 // we pre-reserve aligned virtual memory and then attach to that.
3561
3562 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3563 if (pre_reserved_addr == NULL) {
3564 // Couldn't pre-reserve aligned memory.
3565 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3566 return NULL;
3567 }
3568
3569 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3570 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3571
3572 if ((intptr_t)addr == -1) {
3573 int err = errno;
3574 shm_warning_with_errno("Failed to attach shared memory.");
3575
3576 assert(err != EACCES, "Unexpected error");
3577 assert(err != EIDRM, "Unexpected error");
3578 assert(err != EINVAL, "Unexpected error");
3579
3580 // Since we don't know if the kernel unmapped the pre-reserved memory area
3581 // we can't unmap it, since that would potentially unmap memory that was
3582 // mapped from other threads.
3583 return NULL;
3584 }
3585
3586 return addr;
3587 }
3588
3589 static char* shmat_at_address(int shmid, char* req_addr) {
3590 if (!is_aligned(req_addr, SHMLBA)) {
3591 assert(false, "Requested address needs to be SHMLBA aligned");
3592 return NULL;
3593 }
3594
3595 char* addr = (char*)shmat(shmid, req_addr, 0);
3596
3597 if ((intptr_t)addr == -1) {
3598 shm_warning_with_errno("Failed to attach shared memory.");
3599 return NULL;
3600 }
3601
3602 return addr;
3603 }
3604
3605 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3606 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3607 if (req_addr != NULL) {
3608 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3609 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3610 return shmat_at_address(shmid, req_addr);
3611 }
3612
3613 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3614 // return large page size aligned memory addresses when req_addr == NULL.
3615 // However, if the alignment is larger than the large page size, we have
3616 // to manually ensure that the memory returned is 'alignment' aligned.
3617 if (alignment > os::large_page_size()) {
3618 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3619 return shmat_with_alignment(shmid, bytes, alignment);
3620 } else {
3621 return shmat_at_address(shmid, NULL);
3622 }
3623 }
3624
3625 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3626 char* req_addr, bool exec) {
3627 // "exec" is passed in but not used. Creating the shared image for
3628 // the code cache doesn't have an SHM_X executable permission to check.
3629 assert(UseLargePages && UseSHM, "only for SHM large pages");
3630 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3631 assert(is_aligned(req_addr, alignment), "Unaligned address");
3632
3633 if (!is_aligned(bytes, os::large_page_size())) {
3634 return NULL; // Fallback to small pages.
3635 }
3636
3637 // Create a large shared memory region to attach to based on size.
3638 // Currently, size is the total size of the heap.
3639 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3640 if (shmid == -1) {
3641 // Possible reasons for shmget failure:
3642 // 1. shmmax is too small for Java heap.
3643 // > check shmmax value: cat /proc/sys/kernel/shmmax
3644 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3645 // 2. not enough large page memory.
3646 // > check available large pages: cat /proc/meminfo
3647 // > increase amount of large pages:
3648 // echo new_value > /proc/sys/vm/nr_hugepages
3649 // Note 1: different Linux may use different name for this property,
3650 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3651 // Note 2: it's possible there's enough physical memory available but
3652 // they are so fragmented after a long run that they can't
3653 // coalesce into large pages. Try to reserve large pages when
3654 // the system is still "fresh".
3655 shm_warning_with_errno("Failed to reserve shared memory.");
3656 return NULL;
3657 }
3658
3659 // Attach to the region.
3660 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3661
3662 // Remove shmid. If shmat() is successful, the actual shared memory segment
3663 // will be deleted when it's detached by shmdt() or when the process
3664 // terminates. If shmat() is not successful this will remove the shared
3665 // segment immediately.
3666 shmctl(shmid, IPC_RMID, NULL);
3667
3668 return addr;
3669 }
3670
3671 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3672 int error) {
3673 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3674
3675 bool warn_on_failure = UseLargePages &&
3676 (!FLAG_IS_DEFAULT(UseLargePages) ||
3677 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3678 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3679
3680 if (warn_on_failure) {
3681 char msg[128];
3682 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3683 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3684 warning("%s", msg);
3685 }
3686 }
3687
3688 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3689 char* req_addr,
3690 bool exec) {
3691 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3692 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3693 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3694
3695 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3696 char* addr = (char*)::mmap(req_addr, bytes, prot,
3697 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3698 -1, 0);
3699
3700 if (addr == MAP_FAILED) {
3701 warn_on_large_pages_failure(req_addr, bytes, errno);
3702 return NULL;
3703 }
3704
3705 assert(is_aligned(addr, os::large_page_size()), "Must be");
3706
3707 return addr;
3708 }
3709
3710 // Reserve memory using mmap(MAP_HUGETLB).
3711 // - bytes shall be a multiple of alignment.
3712 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3713 // - alignment sets the alignment at which memory shall be allocated.
3714 // It must be a multiple of allocation granularity.
3715 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3716 // req_addr or NULL.
3717 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3718 size_t alignment,
3719 char* req_addr,
3720 bool exec) {
3721 size_t large_page_size = os::large_page_size();
3722 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3723
3724 assert(is_aligned(req_addr, alignment), "Must be");
3725 assert(is_aligned(bytes, alignment), "Must be");
3726
3727 // First reserve - but not commit - the address range in small pages.
3728 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3729
3730 if (start == NULL) {
3731 return NULL;
3732 }
3733
3734 assert(is_aligned(start, alignment), "Must be");
3735
3736 char* end = start + bytes;
3737
3738 // Find the regions of the allocated chunk that can be promoted to large pages.
3739 char* lp_start = align_up(start, large_page_size);
3740 char* lp_end = align_down(end, large_page_size);
3741
3742 size_t lp_bytes = lp_end - lp_start;
3743
3744 assert(is_aligned(lp_bytes, large_page_size), "Must be");
3745
3746 if (lp_bytes == 0) {
3747 // The mapped region doesn't even span the start and the end of a large page.
3748 // Fall back to allocate a non-special area.
3749 ::munmap(start, end - start);
3750 return NULL;
3751 }
3752
3753 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3754
3755 void* result;
3756
3757 // Commit small-paged leading area.
3758 if (start != lp_start) {
3759 result = ::mmap(start, lp_start - start, prot,
3760 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3761 -1, 0);
3762 if (result == MAP_FAILED) {
3763 ::munmap(lp_start, end - lp_start);
3764 return NULL;
3765 }
3766 }
3767
3768 // Commit large-paged area.
3769 result = ::mmap(lp_start, lp_bytes, prot,
3770 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3771 -1, 0);
3772 if (result == MAP_FAILED) {
3773 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3774 // If the mmap above fails, the large pages region will be unmapped and we
3775 // have regions before and after with small pages. Release these regions.
3776 //
3777 // | mapped | unmapped | mapped |
3778 // ^ ^ ^ ^
3779 // start lp_start lp_end end
3780 //
3781 ::munmap(start, lp_start - start);
3782 ::munmap(lp_end, end - lp_end);
3783 return NULL;
3784 }
3785
3786 // Commit small-paged trailing area.
3787 if (lp_end != end) {
3788 result = ::mmap(lp_end, end - lp_end, prot,
3789 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3790 -1, 0);
3791 if (result == MAP_FAILED) {
3792 ::munmap(start, lp_end - start);
3793 return NULL;
3794 }
3795 }
3796
3797 return start;
3798 }
3799
3800 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3801 size_t alignment,
3802 char* req_addr,
3803 bool exec) {
3804 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3805 assert(is_aligned(req_addr, alignment), "Must be");
3806 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3807 assert(is_power_of_2(os::large_page_size()), "Must be");
3808 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3809
3810 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3811 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3812 } else {
3813 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3814 }
3815 }
3816
3817 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3818 char* req_addr, bool exec) {
3819 assert(UseLargePages, "only for large pages");
3820
3821 char* addr;
3822 if (UseSHM) {
3823 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3824 } else {
3825 assert(UseHugeTLBFS, "must be");
3826 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3827 }
3828
3829 if (addr != NULL) {
3830 if (UseNUMAInterleaving) {
3831 numa_make_global(addr, bytes);
3832 }
3833
3834 // The memory is committed
3835 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3836 }
3837
3838 return addr;
3839 }
3840
3841 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3842 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3843 return shmdt(base) == 0;
3844 }
3845
3846 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3847 return pd_release_memory(base, bytes);
3848 }
3849
3850 bool os::release_memory_special(char* base, size_t bytes) {
3851 bool res;
3852 if (MemTracker::tracking_level() > NMT_minimal) {
3853 Tracker tkr(Tracker::release);
3854 res = os::Linux::release_memory_special_impl(base, bytes);
3855 if (res) {
3856 tkr.record((address)base, bytes);
3857 }
3858
3859 } else {
3860 res = os::Linux::release_memory_special_impl(base, bytes);
3861 }
3862 return res;
3863 }
3864
3865 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3866 assert(UseLargePages, "only for large pages");
3867 bool res;
3868
3869 if (UseSHM) {
3870 res = os::Linux::release_memory_special_shm(base, bytes);
3871 } else {
3872 assert(UseHugeTLBFS, "must be");
3873 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3874 }
3875 return res;
3876 }
3877
3878 size_t os::large_page_size() {
3879 return _large_page_size;
3880 }
3881
3882 // With SysV SHM the entire memory region must be allocated as shared
3883 // memory.
3884 // HugeTLBFS allows application to commit large page memory on demand.
3885 // However, when committing memory with HugeTLBFS fails, the region
3886 // that was supposed to be committed will lose the old reservation
3887 // and allow other threads to steal that memory region. Because of this
3888 // behavior we can't commit HugeTLBFS memory.
3889 bool os::can_commit_large_page_memory() {
3890 return UseTransparentHugePages;
3891 }
3892
3893 bool os::can_execute_large_page_memory() {
3894 return UseTransparentHugePages || UseHugeTLBFS;
3895 }
3896
3897 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3898 assert(file_desc >= 0, "file_desc is not valid");
3899 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3900 if (result != NULL) {
3901 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3902 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3903 }
3904 }
3905 return result;
3906 }
3907
3908 // Reserve memory at an arbitrary address, only if that area is
3909 // available (and not reserved for something else).
3910
3911 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3912 const int max_tries = 10;
3913 char* base[max_tries];
3914 size_t size[max_tries];
3915 const size_t gap = 0x000000;
3916
3917 // Assert only that the size is a multiple of the page size, since
3918 // that's all that mmap requires, and since that's all we really know
3919 // about at this low abstraction level. If we need higher alignment,
3920 // we can either pass an alignment to this method or verify alignment
3921 // in one of the methods further up the call chain. See bug 5044738.
3922 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3923
3924 // Repeatedly allocate blocks until the block is allocated at the
3925 // right spot.
3926
3927 // Linux mmap allows caller to pass an address as hint; give it a try first,
3928 // if kernel honors the hint then we can return immediately.
3929 char * addr = anon_mmap(requested_addr, bytes, false);
3930 if (addr == requested_addr) {
3931 return requested_addr;
3932 }
3933
3934 if (addr != NULL) {
3935 // mmap() is successful but it fails to reserve at the requested address
3936 anon_munmap(addr, bytes);
3937 }
3938
3939 int i;
3940 for (i = 0; i < max_tries; ++i) {
3941 base[i] = reserve_memory(bytes);
3942
3943 if (base[i] != NULL) {
3944 // Is this the block we wanted?
3945 if (base[i] == requested_addr) {
3946 size[i] = bytes;
3947 break;
3948 }
3949
3950 // Does this overlap the block we wanted? Give back the overlapped
3951 // parts and try again.
3952
3953 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3954 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3955 unmap_memory(base[i], top_overlap);
3956 base[i] += top_overlap;
3957 size[i] = bytes - top_overlap;
3958 } else {
3959 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3960 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3961 unmap_memory(requested_addr, bottom_overlap);
3962 size[i] = bytes - bottom_overlap;
3963 } else {
3964 size[i] = bytes;
3965 }
3966 }
3967 }
3968 }
3969
3970 // Give back the unused reserved pieces.
3971
3972 for (int j = 0; j < i; ++j) {
3973 if (base[j] != NULL) {
3974 unmap_memory(base[j], size[j]);
3975 }
3976 }
3977
3978 if (i < max_tries) {
3979 return requested_addr;
3980 } else {
3981 return NULL;
3982 }
3983 }
3984
3985 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3986 return ::read(fd, buf, nBytes);
3987 }
3988
3989 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3990 return ::pread(fd, buf, nBytes, offset);
3991 }
3992
3993 // Short sleep, direct OS call.
3994 //
3995 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3996 // sched_yield(2) will actually give up the CPU:
3997 //
3998 // * Alone on this pariticular CPU, keeps running.
3999 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4000 // (pre 2.6.39).
4001 //
4002 // So calling this with 0 is an alternative.
4003 //
4004 void os::naked_short_sleep(jlong ms) {
4005 struct timespec req;
4006
4007 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4008 req.tv_sec = 0;
4009 if (ms > 0) {
4010 req.tv_nsec = (ms % 1000) * 1000000;
4011 } else {
4012 req.tv_nsec = 1;
4013 }
4014
4015 nanosleep(&req, NULL);
4016
4017 return;
4018 }
4019
4020 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4021 void os::infinite_sleep() {
4022 while (true) { // sleep forever ...
4023 ::sleep(100); // ... 100 seconds at a time
4024 }
4025 }
4026
4027 // Used to convert frequent JVM_Yield() to nops
4028 bool os::dont_yield() {
4029 return DontYieldALot;
4030 }
4031
4032 void os::naked_yield() {
4033 sched_yield();
4034 }
4035
4036 ////////////////////////////////////////////////////////////////////////////////
4037 // thread priority support
4038
4039 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4040 // only supports dynamic priority, static priority must be zero. For real-time
4041 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4042 // However, for large multi-threaded applications, SCHED_RR is not only slower
4043 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4044 // of 5 runs - Sep 2005).
4045 //
4046 // The following code actually changes the niceness of kernel-thread/LWP. It
4047 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4048 // not the entire user process, and user level threads are 1:1 mapped to kernel
4049 // threads. It has always been the case, but could change in the future. For
4050 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4051 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4052
4053 int os::java_to_os_priority[CriticalPriority + 1] = {
4054 19, // 0 Entry should never be used
4055
4056 4, // 1 MinPriority
4057 3, // 2
4058 2, // 3
4059
4060 1, // 4
4061 0, // 5 NormPriority
4062 -1, // 6
4063
4064 -2, // 7
4065 -3, // 8
4066 -4, // 9 NearMaxPriority
4067
4068 -5, // 10 MaxPriority
4069
4070 -5 // 11 CriticalPriority
4071 };
4072
4073 static int prio_init() {
4074 if (ThreadPriorityPolicy == 1) {
4075 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4076 // if effective uid is not root. Perhaps, a more elegant way of doing
4077 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4078 if (geteuid() != 0) {
4079 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4080 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4081 }
4082 ThreadPriorityPolicy = 0;
4083 }
4084 }
4085 if (UseCriticalJavaThreadPriority) {
4086 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4087 }
4088 return 0;
4089 }
4090
4091 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4092 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4093
4094 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4095 return (ret == 0) ? OS_OK : OS_ERR;
4096 }
4097
4098 OSReturn os::get_native_priority(const Thread* const thread,
4099 int *priority_ptr) {
4100 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4101 *priority_ptr = java_to_os_priority[NormPriority];
4102 return OS_OK;
4103 }
4104
4105 errno = 0;
4106 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4107 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4108 }
4109
4110 // Hint to the underlying OS that a task switch would not be good.
4111 // Void return because it's a hint and can fail.
4112 void os::hint_no_preempt() {}
4113
4114 ////////////////////////////////////////////////////////////////////////////////
4115 // suspend/resume support
4116
4117 // The low-level signal-based suspend/resume support is a remnant from the
4118 // old VM-suspension that used to be for java-suspension, safepoints etc,
4119 // within hotspot. Currently used by JFR's OSThreadSampler
4120 //
4121 // The remaining code is greatly simplified from the more general suspension
4122 // code that used to be used.
4123 //
4124 // The protocol is quite simple:
4125 // - suspend:
4126 // - sends a signal to the target thread
4127 // - polls the suspend state of the osthread using a yield loop
4128 // - target thread signal handler (SR_handler) sets suspend state
4129 // and blocks in sigsuspend until continued
4130 // - resume:
4131 // - sets target osthread state to continue
4132 // - sends signal to end the sigsuspend loop in the SR_handler
4133 //
4134 // Note that the SR_lock plays no role in this suspend/resume protocol,
4135 // but is checked for NULL in SR_handler as a thread termination indicator.
4136 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4137 //
4138 // Note that resume_clear_context() and suspend_save_context() are needed
4139 // by SR_handler(), so that fetch_frame_from_ucontext() works,
4140 // which in part is used by:
4141 // - Forte Analyzer: AsyncGetCallTrace()
4142 // - StackBanging: get_frame_at_stack_banging_point()
4143
4144 static void resume_clear_context(OSThread *osthread) {
4145 osthread->set_ucontext(NULL);
4146 osthread->set_siginfo(NULL);
4147 }
4148
4149 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4150 ucontext_t* context) {
4151 osthread->set_ucontext(context);
4152 osthread->set_siginfo(siginfo);
4153 }
4154
4155 // Handler function invoked when a thread's execution is suspended or
4156 // resumed. We have to be careful that only async-safe functions are
4157 // called here (Note: most pthread functions are not async safe and
4158 // should be avoided.)
4159 //
4160 // Note: sigwait() is a more natural fit than sigsuspend() from an
4161 // interface point of view, but sigwait() prevents the signal hander
4162 // from being run. libpthread would get very confused by not having
4163 // its signal handlers run and prevents sigwait()'s use with the
4164 // mutex granting granting signal.
4165 //
4166 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4167 //
4168 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4169 // Save and restore errno to avoid confusing native code with EINTR
4170 // after sigsuspend.
4171 int old_errno = errno;
4172
4173 Thread* thread = Thread::current_or_null_safe();
4174 assert(thread != NULL, "Missing current thread in SR_handler");
4175
4176 // On some systems we have seen signal delivery get "stuck" until the signal
4177 // mask is changed as part of thread termination. Check that the current thread
4178 // has not already terminated (via SR_lock()) - else the following assertion
4179 // will fail because the thread is no longer a JavaThread as the ~JavaThread
4180 // destructor has completed.
4181
4182 if (thread->SR_lock() == NULL) {
4183 return;
4184 }
4185
4186 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4187
4188 OSThread* osthread = thread->osthread();
4189
4190 os::SuspendResume::State current = osthread->sr.state();
4191 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4192 suspend_save_context(osthread, siginfo, context);
4193
4194 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4195 os::SuspendResume::State state = osthread->sr.suspended();
4196 if (state == os::SuspendResume::SR_SUSPENDED) {
4197 sigset_t suspend_set; // signals for sigsuspend()
4198 sigemptyset(&suspend_set);
4199 // get current set of blocked signals and unblock resume signal
4200 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4201 sigdelset(&suspend_set, SR_signum);
4202
4203 sr_semaphore.signal();
4204 // wait here until we are resumed
4205 while (1) {
4206 sigsuspend(&suspend_set);
4207
4208 os::SuspendResume::State result = osthread->sr.running();
4209 if (result == os::SuspendResume::SR_RUNNING) {
4210 sr_semaphore.signal();
4211 break;
4212 }
4213 }
4214
4215 } else if (state == os::SuspendResume::SR_RUNNING) {
4216 // request was cancelled, continue
4217 } else {
4218 ShouldNotReachHere();
4219 }
4220
4221 resume_clear_context(osthread);
4222 } else if (current == os::SuspendResume::SR_RUNNING) {
4223 // request was cancelled, continue
4224 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4225 // ignore
4226 } else {
4227 // ignore
4228 }
4229
4230 errno = old_errno;
4231 }
4232
4233 static int SR_initialize() {
4234 struct sigaction act;
4235 char *s;
4236
4237 // Get signal number to use for suspend/resume
4238 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4239 int sig = ::strtol(s, 0, 10);
4240 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769.
4241 sig < NSIG) { // Must be legal signal and fit into sigflags[].
4242 SR_signum = sig;
4243 } else {
4244 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4245 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4246 }
4247 }
4248
4249 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4250 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4251
4252 sigemptyset(&SR_sigset);
4253 sigaddset(&SR_sigset, SR_signum);
4254
4255 // Set up signal handler for suspend/resume
4256 act.sa_flags = SA_RESTART|SA_SIGINFO;
4257 act.sa_handler = (void (*)(int)) SR_handler;
4258
4259 // SR_signum is blocked by default.
4260 // 4528190 - We also need to block pthread restart signal (32 on all
4261 // supported Linux platforms). Note that LinuxThreads need to block
4262 // this signal for all threads to work properly. So we don't have
4263 // to use hard-coded signal number when setting up the mask.
4264 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4265
4266 if (sigaction(SR_signum, &act, 0) == -1) {
4267 return -1;
4268 }
4269
4270 // Save signal flag
4271 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4272 return 0;
4273 }
4274
4275 static int sr_notify(OSThread* osthread) {
4276 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4277 assert_status(status == 0, status, "pthread_kill");
4278 return status;
4279 }
4280
4281 // "Randomly" selected value for how long we want to spin
4282 // before bailing out on suspending a thread, also how often
4283 // we send a signal to a thread we want to resume
4284 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4285 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4286
4287 // returns true on success and false on error - really an error is fatal
4288 // but this seems the normal response to library errors
4289 static bool do_suspend(OSThread* osthread) {
4290 assert(osthread->sr.is_running(), "thread should be running");
4291 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4292
4293 // mark as suspended and send signal
4294 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4295 // failed to switch, state wasn't running?
4296 ShouldNotReachHere();
4297 return false;
4298 }
4299
4300 if (sr_notify(osthread) != 0) {
4301 ShouldNotReachHere();
4302 }
4303
4304 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4305 while (true) {
4306 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4307 break;
4308 } else {
4309 // timeout
4310 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4311 if (cancelled == os::SuspendResume::SR_RUNNING) {
4312 return false;
4313 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4314 // make sure that we consume the signal on the semaphore as well
4315 sr_semaphore.wait();
4316 break;
4317 } else {
4318 ShouldNotReachHere();
4319 return false;
4320 }
4321 }
4322 }
4323
4324 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4325 return true;
4326 }
4327
4328 static void do_resume(OSThread* osthread) {
4329 assert(osthread->sr.is_suspended(), "thread should be suspended");
4330 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4331
4332 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4333 // failed to switch to WAKEUP_REQUEST
4334 ShouldNotReachHere();
4335 return;
4336 }
4337
4338 while (true) {
4339 if (sr_notify(osthread) == 0) {
4340 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4341 if (osthread->sr.is_running()) {
4342 return;
4343 }
4344 }
4345 } else {
4346 ShouldNotReachHere();
4347 }
4348 }
4349
4350 guarantee(osthread->sr.is_running(), "Must be running!");
4351 }
4352
4353 ///////////////////////////////////////////////////////////////////////////////////
4354 // signal handling (except suspend/resume)
4355
4356 // This routine may be used by user applications as a "hook" to catch signals.
4357 // The user-defined signal handler must pass unrecognized signals to this
4358 // routine, and if it returns true (non-zero), then the signal handler must
4359 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4360 // routine will never retun false (zero), but instead will execute a VM panic
4361 // routine kill the process.
4362 //
4363 // If this routine returns false, it is OK to call it again. This allows
4364 // the user-defined signal handler to perform checks either before or after
4365 // the VM performs its own checks. Naturally, the user code would be making
4366 // a serious error if it tried to handle an exception (such as a null check
4367 // or breakpoint) that the VM was generating for its own correct operation.
4368 //
4369 // This routine may recognize any of the following kinds of signals:
4370 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4371 // It should be consulted by handlers for any of those signals.
4372 //
4373 // The caller of this routine must pass in the three arguments supplied
4374 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4375 // field of the structure passed to sigaction(). This routine assumes that
4376 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4377 //
4378 // Note that the VM will print warnings if it detects conflicting signal
4379 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4380 //
4381 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4382 siginfo_t* siginfo,
4383 void* ucontext,
4384 int abort_if_unrecognized);
4385
4386 void signalHandler(int sig, siginfo_t* info, void* uc) {
4387 assert(info != NULL && uc != NULL, "it must be old kernel");
4388 int orig_errno = errno; // Preserve errno value over signal handler.
4389 JVM_handle_linux_signal(sig, info, uc, true);
4390 errno = orig_errno;
4391 }
4392
4393
4394 // This boolean allows users to forward their own non-matching signals
4395 // to JVM_handle_linux_signal, harmlessly.
4396 bool os::Linux::signal_handlers_are_installed = false;
4397
4398 // For signal-chaining
4399 struct sigaction sigact[NSIG];
4400 uint64_t sigs = 0;
4401 #if (64 < NSIG-1)
4402 #error "Not all signals can be encoded in sigs. Adapt its type!"
4403 #endif
4404 bool os::Linux::libjsig_is_loaded = false;
4405 typedef struct sigaction *(*get_signal_t)(int);
4406 get_signal_t os::Linux::get_signal_action = NULL;
4407
4408 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4409 struct sigaction *actp = NULL;
4410
4411 if (libjsig_is_loaded) {
4412 // Retrieve the old signal handler from libjsig
4413 actp = (*get_signal_action)(sig);
4414 }
4415 if (actp == NULL) {
4416 // Retrieve the preinstalled signal handler from jvm
4417 actp = get_preinstalled_handler(sig);
4418 }
4419
4420 return actp;
4421 }
4422
4423 static bool call_chained_handler(struct sigaction *actp, int sig,
4424 siginfo_t *siginfo, void *context) {
4425 // Call the old signal handler
4426 if (actp->sa_handler == SIG_DFL) {
4427 // It's more reasonable to let jvm treat it as an unexpected exception
4428 // instead of taking the default action.
4429 return false;
4430 } else if (actp->sa_handler != SIG_IGN) {
4431 if ((actp->sa_flags & SA_NODEFER) == 0) {
4432 // automaticlly block the signal
4433 sigaddset(&(actp->sa_mask), sig);
4434 }
4435
4436 sa_handler_t hand = NULL;
4437 sa_sigaction_t sa = NULL;
4438 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4439 // retrieve the chained handler
4440 if (siginfo_flag_set) {
4441 sa = actp->sa_sigaction;
4442 } else {
4443 hand = actp->sa_handler;
4444 }
4445
4446 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4447 actp->sa_handler = SIG_DFL;
4448 }
4449
4450 // try to honor the signal mask
4451 sigset_t oset;
4452 sigemptyset(&oset);
4453 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4454
4455 // call into the chained handler
4456 if (siginfo_flag_set) {
4457 (*sa)(sig, siginfo, context);
4458 } else {
4459 (*hand)(sig);
4460 }
4461
4462 // restore the signal mask
4463 pthread_sigmask(SIG_SETMASK, &oset, NULL);
4464 }
4465 // Tell jvm's signal handler the signal is taken care of.
4466 return true;
4467 }
4468
4469 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4470 bool chained = false;
4471 // signal-chaining
4472 if (UseSignalChaining) {
4473 struct sigaction *actp = get_chained_signal_action(sig);
4474 if (actp != NULL) {
4475 chained = call_chained_handler(actp, sig, siginfo, context);
4476 }
4477 }
4478 return chained;
4479 }
4480
4481 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4482 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4483 return &sigact[sig];
4484 }
4485 return NULL;
4486 }
4487
4488 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4489 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4490 sigact[sig] = oldAct;
4491 sigs |= (uint64_t)1 << (sig-1);
4492 }
4493
4494 // for diagnostic
4495 int sigflags[NSIG];
4496
4497 int os::Linux::get_our_sigflags(int sig) {
4498 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4499 return sigflags[sig];
4500 }
4501
4502 void os::Linux::set_our_sigflags(int sig, int flags) {
4503 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4504 if (sig > 0 && sig < NSIG) {
4505 sigflags[sig] = flags;
4506 }
4507 }
4508
4509 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4510 // Check for overwrite.
4511 struct sigaction oldAct;
4512 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4513
4514 void* oldhand = oldAct.sa_sigaction
4515 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4516 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4517 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4518 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4519 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4520 if (AllowUserSignalHandlers || !set_installed) {
4521 // Do not overwrite; user takes responsibility to forward to us.
4522 return;
4523 } else if (UseSignalChaining) {
4524 // save the old handler in jvm
4525 save_preinstalled_handler(sig, oldAct);
4526 // libjsig also interposes the sigaction() call below and saves the
4527 // old sigaction on it own.
4528 } else {
4529 fatal("Encountered unexpected pre-existing sigaction handler "
4530 "%#lx for signal %d.", (long)oldhand, sig);
4531 }
4532 }
4533
4534 struct sigaction sigAct;
4535 sigfillset(&(sigAct.sa_mask));
4536 sigAct.sa_handler = SIG_DFL;
4537 if (!set_installed) {
4538 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4539 } else {
4540 sigAct.sa_sigaction = signalHandler;
4541 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4542 }
4543 // Save flags, which are set by ours
4544 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4545 sigflags[sig] = sigAct.sa_flags;
4546
4547 int ret = sigaction(sig, &sigAct, &oldAct);
4548 assert(ret == 0, "check");
4549
4550 void* oldhand2 = oldAct.sa_sigaction
4551 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4552 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4553 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4554 }
4555
4556 // install signal handlers for signals that HotSpot needs to
4557 // handle in order to support Java-level exception handling.
4558
4559 void os::Linux::install_signal_handlers() {
4560 if (!signal_handlers_are_installed) {
4561 signal_handlers_are_installed = true;
4562
4563 // signal-chaining
4564 typedef void (*signal_setting_t)();
4565 signal_setting_t begin_signal_setting = NULL;
4566 signal_setting_t end_signal_setting = NULL;
4567 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4568 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4569 if (begin_signal_setting != NULL) {
4570 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4571 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4572 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4573 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4574 libjsig_is_loaded = true;
4575 assert(UseSignalChaining, "should enable signal-chaining");
4576 }
4577 if (libjsig_is_loaded) {
4578 // Tell libjsig jvm is setting signal handlers
4579 (*begin_signal_setting)();
4580 }
4581
4582 set_signal_handler(SIGSEGV, true);
4583 set_signal_handler(SIGPIPE, true);
4584 set_signal_handler(SIGBUS, true);
4585 set_signal_handler(SIGILL, true);
4586 set_signal_handler(SIGFPE, true);
4587 #if defined(PPC64)
4588 set_signal_handler(SIGTRAP, true);
4589 #endif
4590 set_signal_handler(SIGXFSZ, true);
4591
4592 if (libjsig_is_loaded) {
4593 // Tell libjsig jvm finishes setting signal handlers
4594 (*end_signal_setting)();
4595 }
4596
4597 // We don't activate signal checker if libjsig is in place, we trust ourselves
4598 // and if UserSignalHandler is installed all bets are off.
4599 // Log that signal checking is off only if -verbose:jni is specified.
4600 if (CheckJNICalls) {
4601 if (libjsig_is_loaded) {
4602 if (PrintJNIResolving) {
4603 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4604 }
4605 check_signals = false;
4606 }
4607 if (AllowUserSignalHandlers) {
4608 if (PrintJNIResolving) {
4609 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4610 }
4611 check_signals = false;
4612 }
4613 }
4614 }
4615 }
4616
4617 // This is the fastest way to get thread cpu time on Linux.
4618 // Returns cpu time (user+sys) for any thread, not only for current.
4619 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4620 // It might work on 2.6.10+ with a special kernel/glibc patch.
4621 // For reference, please, see IEEE Std 1003.1-2004:
4622 // http://www.unix.org/single_unix_specification
4623
4624 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4625 struct timespec tp;
4626 int rc = os::Linux::clock_gettime(clockid, &tp);
4627 assert(rc == 0, "clock_gettime is expected to return 0 code");
4628
4629 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4630 }
4631
4632 void os::Linux::initialize_os_info() {
4633 assert(_os_version == 0, "OS info already initialized");
4634
4635 struct utsname _uname;
4636
4637 uint32_t major;
4638 uint32_t minor;
4639 uint32_t fix;
4640
4641 int rc;
4642
4643 // Kernel version is unknown if
4644 // verification below fails.
4645 _os_version = 0x01000000;
4646
4647 rc = uname(&_uname);
4648 if (rc != -1) {
4649
4650 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4651 if (rc == 3) {
4652
4653 if (major < 256 && minor < 256 && fix < 256) {
4654 // Kernel version format is as expected,
4655 // set it overriding unknown state.
4656 _os_version = (major << 16) |
4657 (minor << 8 ) |
4658 (fix << 0 ) ;
4659 }
4660 }
4661 }
4662 }
4663
4664 uint32_t os::Linux::os_version() {
4665 assert(_os_version != 0, "not initialized");
4666 return _os_version & 0x00FFFFFF;
4667 }
4668
4669 bool os::Linux::os_version_is_known() {
4670 assert(_os_version != 0, "not initialized");
4671 return _os_version & 0x01000000 ? false : true;
4672 }
4673
4674 /////
4675 // glibc on Linux platform uses non-documented flag
4676 // to indicate, that some special sort of signal
4677 // trampoline is used.
4678 // We will never set this flag, and we should
4679 // ignore this flag in our diagnostic
4680 #ifdef SIGNIFICANT_SIGNAL_MASK
4681 #undef SIGNIFICANT_SIGNAL_MASK
4682 #endif
4683 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4684
4685 static const char* get_signal_handler_name(address handler,
4686 char* buf, int buflen) {
4687 int offset = 0;
4688 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4689 if (found) {
4690 // skip directory names
4691 const char *p1, *p2;
4692 p1 = buf;
4693 size_t len = strlen(os::file_separator());
4694 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4695 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4696 } else {
4697 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4698 }
4699 return buf;
4700 }
4701
4702 static void print_signal_handler(outputStream* st, int sig,
4703 char* buf, size_t buflen) {
4704 struct sigaction sa;
4705
4706 sigaction(sig, NULL, &sa);
4707
4708 // See comment for SIGNIFICANT_SIGNAL_MASK define
4709 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4710
4711 st->print("%s: ", os::exception_name(sig, buf, buflen));
4712
4713 address handler = (sa.sa_flags & SA_SIGINFO)
4714 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4715 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4716
4717 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4718 st->print("SIG_DFL");
4719 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4720 st->print("SIG_IGN");
4721 } else {
4722 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4723 }
4724
4725 st->print(", sa_mask[0]=");
4726 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4727
4728 address rh = VMError::get_resetted_sighandler(sig);
4729 // May be, handler was resetted by VMError?
4730 if (rh != NULL) {
4731 handler = rh;
4732 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4733 }
4734
4735 st->print(", sa_flags=");
4736 os::Posix::print_sa_flags(st, sa.sa_flags);
4737
4738 // Check: is it our handler?
4739 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4740 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4741 // It is our signal handler
4742 // check for flags, reset system-used one!
4743 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4744 st->print(
4745 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4746 os::Linux::get_our_sigflags(sig));
4747 }
4748 }
4749 st->cr();
4750 }
4751
4752
4753 #define DO_SIGNAL_CHECK(sig) \
4754 do { \
4755 if (!sigismember(&check_signal_done, sig)) { \
4756 os::Linux::check_signal_handler(sig); \
4757 } \
4758 } while (0)
4759
4760 // This method is a periodic task to check for misbehaving JNI applications
4761 // under CheckJNI, we can add any periodic checks here
4762
4763 void os::run_periodic_checks() {
4764 if (check_signals == false) return;
4765
4766 // SEGV and BUS if overridden could potentially prevent
4767 // generation of hs*.log in the event of a crash, debugging
4768 // such a case can be very challenging, so we absolutely
4769 // check the following for a good measure:
4770 DO_SIGNAL_CHECK(SIGSEGV);
4771 DO_SIGNAL_CHECK(SIGILL);
4772 DO_SIGNAL_CHECK(SIGFPE);
4773 DO_SIGNAL_CHECK(SIGBUS);
4774 DO_SIGNAL_CHECK(SIGPIPE);
4775 DO_SIGNAL_CHECK(SIGXFSZ);
4776 #if defined(PPC64)
4777 DO_SIGNAL_CHECK(SIGTRAP);
4778 #endif
4779
4780 // ReduceSignalUsage allows the user to override these handlers
4781 // see comments at the very top and jvm_md.h
4782 if (!ReduceSignalUsage) {
4783 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4784 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4785 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4786 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4787 }
4788
4789 DO_SIGNAL_CHECK(SR_signum);
4790 }
4791
4792 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4793
4794 static os_sigaction_t os_sigaction = NULL;
4795
4796 void os::Linux::check_signal_handler(int sig) {
4797 char buf[O_BUFLEN];
4798 address jvmHandler = NULL;
4799
4800
4801 struct sigaction act;
4802 if (os_sigaction == NULL) {
4803 // only trust the default sigaction, in case it has been interposed
4804 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4805 if (os_sigaction == NULL) return;
4806 }
4807
4808 os_sigaction(sig, (struct sigaction*)NULL, &act);
4809
4810
4811 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4812
4813 address thisHandler = (act.sa_flags & SA_SIGINFO)
4814 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4815 : CAST_FROM_FN_PTR(address, act.sa_handler);
4816
4817
4818 switch (sig) {
4819 case SIGSEGV:
4820 case SIGBUS:
4821 case SIGFPE:
4822 case SIGPIPE:
4823 case SIGILL:
4824 case SIGXFSZ:
4825 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4826 break;
4827
4828 case SHUTDOWN1_SIGNAL:
4829 case SHUTDOWN2_SIGNAL:
4830 case SHUTDOWN3_SIGNAL:
4831 case BREAK_SIGNAL:
4832 jvmHandler = (address)user_handler();
4833 break;
4834
4835 default:
4836 if (sig == SR_signum) {
4837 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4838 } else {
4839 return;
4840 }
4841 break;
4842 }
4843
4844 if (thisHandler != jvmHandler) {
4845 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4846 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4847 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4848 // No need to check this sig any longer
4849 sigaddset(&check_signal_done, sig);
4850 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4851 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4852 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4853 exception_name(sig, buf, O_BUFLEN));
4854 }
4855 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4856 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4857 tty->print("expected:");
4858 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4859 tty->cr();
4860 tty->print(" found:");
4861 os::Posix::print_sa_flags(tty, act.sa_flags);
4862 tty->cr();
4863 // No need to check this sig any longer
4864 sigaddset(&check_signal_done, sig);
4865 }
4866
4867 // Dump all the signal
4868 if (sigismember(&check_signal_done, sig)) {
4869 print_signal_handlers(tty, buf, O_BUFLEN);
4870 }
4871 }
4872
4873 extern void report_error(char* file_name, int line_no, char* title,
4874 char* format, ...);
4875
4876 // this is called _before_ most of the global arguments have been parsed
4877 void os::init(void) {
4878 char dummy; // used to get a guess on initial stack address
4879
4880 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4881
4882 init_random(1234567);
4883
4884 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4885 if (Linux::page_size() == -1) {
4886 fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4887 os::strerror(errno));
4888 }
4889 init_page_sizes((size_t) Linux::page_size());
4890
4891 Linux::initialize_system_info();
4892
4893 Linux::initialize_os_info();
4894
4895 // _main_thread points to the thread that created/loaded the JVM.
4896 Linux::_main_thread = pthread_self();
4897
4898 Linux::clock_init();
4899 initial_time_count = javaTimeNanos();
4900
4901 // retrieve entry point for pthread_setname_np
4902 Linux::_pthread_setname_np =
4903 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4904
4905 os::Posix::init();
4906 }
4907
4908 // To install functions for atexit system call
4909 extern "C" {
4910 static void perfMemory_exit_helper() {
4911 perfMemory_exit();
4912 }
4913 }
4914
4915 void os::pd_init_container_support() {
4916 OSContainer::init();
4917 }
4918
4919 // this is called _after_ the global arguments have been parsed
4920 jint os::init_2(void) {
4921
4922 os::Posix::init_2();
4923
4924 Linux::fast_thread_clock_init();
4925
4926 // initialize suspend/resume support - must do this before signal_sets_init()
4927 if (SR_initialize() != 0) {
4928 perror("SR_initialize failed");
4929 return JNI_ERR;
4930 }
4931
4932 Linux::signal_sets_init();
4933 Linux::install_signal_handlers();
4934
4935 // Check and sets minimum stack sizes against command line options
4936 if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4937 return JNI_ERR;
4938 }
4939 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4940
4941 #if defined(IA32)
4942 workaround_expand_exec_shield_cs_limit();
4943 #endif
4944
4945 Linux::libpthread_init();
4946 Linux::sched_getcpu_init();
4947 log_info(os)("HotSpot is running with %s, %s",
4948 Linux::glibc_version(), Linux::libpthread_version());
4949
4950 if (UseNUMA) {
4951 if (!Linux::libnuma_init()) {
4952 UseNUMA = false;
4953 } else {
4954 if ((Linux::numa_max_node() < 1)) {
4955 // There's only one node(they start from 0), disable NUMA.
4956 UseNUMA = false;
4957 }
4958 }
4959
4960 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4961 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4962 // we can make the adaptive lgrp chunk resizing work. If the user specified both
4963 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4964 // and disable adaptive resizing.
4965 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4966 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4967 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4968 UseAdaptiveSizePolicy = false;
4969 UseAdaptiveNUMAChunkSizing = false;
4970 }
4971 }
4972
4973 if (!UseNUMA && ForceNUMA) {
4974 UseNUMA = true;
4975 }
4976 }
4977
4978 if (MaxFDLimit) {
4979 // set the number of file descriptors to max. print out error
4980 // if getrlimit/setrlimit fails but continue regardless.
4981 struct rlimit nbr_files;
4982 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4983 if (status != 0) {
4984 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4985 } else {
4986 nbr_files.rlim_cur = nbr_files.rlim_max;
4987 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4988 if (status != 0) {
4989 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4990 }
4991 }
4992 }
4993
4994 // Initialize lock used to serialize thread creation (see os::create_thread)
4995 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4996
4997 // at-exit methods are called in the reverse order of their registration.
4998 // atexit functions are called on return from main or as a result of a
4999 // call to exit(3C). There can be only 32 of these functions registered
5000 // and atexit() does not set errno.
5001
5002 if (PerfAllowAtExitRegistration) {
5003 // only register atexit functions if PerfAllowAtExitRegistration is set.
5004 // atexit functions can be delayed until process exit time, which
5005 // can be problematic for embedded VM situations. Embedded VMs should
5006 // call DestroyJavaVM() to assure that VM resources are released.
5007
5008 // note: perfMemory_exit_helper atexit function may be removed in
5009 // the future if the appropriate cleanup code can be added to the
5010 // VM_Exit VMOperation's doit method.
5011 if (atexit(perfMemory_exit_helper) != 0) {
5012 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5013 }
5014 }
5015
5016 // initialize thread priority policy
5017 prio_init();
5018
5019 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5020 set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5021 }
5022 return JNI_OK;
5023 }
5024
5025 // Mark the polling page as unreadable
5026 void os::make_polling_page_unreadable(void) {
5027 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5028 fatal("Could not disable polling page");
5029 }
5030 }
5031
5032 // Mark the polling page as readable
5033 void os::make_polling_page_readable(void) {
5034 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5035 fatal("Could not enable polling page");
5036 }
5037 }
5038
5039 // older glibc versions don't have this macro (which expands to
5040 // an optimized bit-counting function) so we have to roll our own
5041 #ifndef CPU_COUNT
5042
5043 static int _cpu_count(const cpu_set_t* cpus) {
5044 int count = 0;
5045 // only look up to the number of configured processors
5046 for (int i = 0; i < os::processor_count(); i++) {
5047 if (CPU_ISSET(i, cpus)) {
5048 count++;
5049 }
5050 }
5051 return count;
5052 }
5053
5054 #define CPU_COUNT(cpus) _cpu_count(cpus)
5055
5056 #endif // CPU_COUNT
5057
5058 // Get the current number of available processors for this process.
5059 // This value can change at any time during a process's lifetime.
5060 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5061 // If it appears there may be more than 1024 processors then we do a
5062 // dynamic check - see 6515172 for details.
5063 // If anything goes wrong we fallback to returning the number of online
5064 // processors - which can be greater than the number available to the process.
5065 int os::Linux::active_processor_count() {
5066 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5067 cpu_set_t* cpus_p = &cpus;
5068 int cpus_size = sizeof(cpu_set_t);
5069
5070 int configured_cpus = os::processor_count(); // upper bound on available cpus
5071 int cpu_count = 0;
5072
5073 // old build platforms may not support dynamic cpu sets
5074 #ifdef CPU_ALLOC
5075
5076 // To enable easy testing of the dynamic path on different platforms we
5077 // introduce a diagnostic flag: UseCpuAllocPath
5078 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5079 // kernel may use a mask bigger than cpu_set_t
5080 log_trace(os)("active_processor_count: using dynamic path %s"
5081 "- configured processors: %d",
5082 UseCpuAllocPath ? "(forced) " : "",
5083 configured_cpus);
5084 cpus_p = CPU_ALLOC(configured_cpus);
5085 if (cpus_p != NULL) {
5086 cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5087 // zero it just to be safe
5088 CPU_ZERO_S(cpus_size, cpus_p);
5089 }
5090 else {
5091 // failed to allocate so fallback to online cpus
5092 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5093 log_trace(os)("active_processor_count: "
5094 "CPU_ALLOC failed (%s) - using "
5095 "online processor count: %d",
5096 os::strerror(errno), online_cpus);
5097 return online_cpus;
5098 }
5099 }
5100 else {
5101 log_trace(os)("active_processor_count: using static path - configured processors: %d",
5102 configured_cpus);
5103 }
5104 #else // CPU_ALLOC
5105 // these stubs won't be executed
5106 #define CPU_COUNT_S(size, cpus) -1
5107 #define CPU_FREE(cpus)
5108
5109 log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5110 configured_cpus);
5111 #endif // CPU_ALLOC
5112
5113 // pid 0 means the current thread - which we have to assume represents the process
5114 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5115 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5116 cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5117 }
5118 else {
5119 cpu_count = CPU_COUNT(cpus_p);
5120 }
5121 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5122 }
5123 else {
5124 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5125 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5126 "which may exceed available processors", os::strerror(errno), cpu_count);
5127 }
5128
5129 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5130 CPU_FREE(cpus_p);
5131 }
5132
5133 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5134 return cpu_count;
5135 }
5136
5137 // Determine the active processor count from one of
5138 // three different sources:
5139 //
5140 // 1. User option -XX:ActiveProcessorCount
5141 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5142 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5143 //
5144 // Option 1, if specified, will always override.
5145 // If the cgroup subsystem is active and configured, we
5146 // will return the min of the cgroup and option 2 results.
5147 // This is required since tools, such as numactl, that
5148 // alter cpu affinity do not update cgroup subsystem
5149 // cpuset configuration files.
5150 int os::active_processor_count() {
5151 // User has overridden the number of active processors
5152 if (ActiveProcessorCount > 0) {
5153 log_trace(os)("active_processor_count: "
5154 "active processor count set by user : %d",
5155 ActiveProcessorCount);
5156 return ActiveProcessorCount;
5157 }
5158
5159 int active_cpus;
5160 if (OSContainer::is_containerized()) {
5161 active_cpus = OSContainer::active_processor_count();
5162 log_trace(os)("active_processor_count: determined by OSContainer: %d",
5163 active_cpus);
5164 } else {
5165 active_cpus = os::Linux::active_processor_count();
5166 }
5167
5168 return active_cpus;
5169 }
5170
5171 void os::set_native_thread_name(const char *name) {
5172 if (Linux::_pthread_setname_np) {
5173 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5174 snprintf(buf, sizeof(buf), "%s", name);
5175 buf[sizeof(buf) - 1] = '\0';
5176 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5177 // ERANGE should not happen; all other errors should just be ignored.
5178 assert(rc != ERANGE, "pthread_setname_np failed");
5179 }
5180 }
5181
5182 bool os::distribute_processes(uint length, uint* distribution) {
5183 // Not yet implemented.
5184 return false;
5185 }
5186
5187 bool os::bind_to_processor(uint processor_id) {
5188 // Not yet implemented.
5189 return false;
5190 }
5191
5192 ///
5193
5194 void os::SuspendedThreadTask::internal_do_task() {
5195 if (do_suspend(_thread->osthread())) {
5196 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5197 do_task(context);
5198 do_resume(_thread->osthread());
5199 }
5200 }
5201
5202 ////////////////////////////////////////////////////////////////////////////////
5203 // debug support
5204
5205 bool os::find(address addr, outputStream* st) {
5206 Dl_info dlinfo;
5207 memset(&dlinfo, 0, sizeof(dlinfo));
5208 if (dladdr(addr, &dlinfo) != 0) {
5209 st->print(PTR_FORMAT ": ", p2i(addr));
5210 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5211 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5212 p2i(addr) - p2i(dlinfo.dli_saddr));
5213 } else if (dlinfo.dli_fbase != NULL) {
5214 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5215 } else {
5216 st->print("<absolute address>");
5217 }
5218 if (dlinfo.dli_fname != NULL) {
5219 st->print(" in %s", dlinfo.dli_fname);
5220 }
5221 if (dlinfo.dli_fbase != NULL) {
5222 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5223 }
5224 st->cr();
5225
5226 if (Verbose) {
5227 // decode some bytes around the PC
5228 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5229 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5230 address lowest = (address) dlinfo.dli_sname;
5231 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5232 if (begin < lowest) begin = lowest;
5233 Dl_info dlinfo2;
5234 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5235 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5236 end = (address) dlinfo2.dli_saddr;
5237 }
5238 Disassembler::decode(begin, end, st);
5239 }
5240 return true;
5241 }
5242 return false;
5243 }
5244
5245 ////////////////////////////////////////////////////////////////////////////////
5246 // misc
5247
5248 // This does not do anything on Linux. This is basically a hook for being
5249 // able to use structured exception handling (thread-local exception filters)
5250 // on, e.g., Win32.
5251 void
5252 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5253 JavaCallArguments* args, Thread* thread) {
5254 f(value, method, args, thread);
5255 }
5256
5257 void os::print_statistics() {
5258 }
5259
5260 bool os::message_box(const char* title, const char* message) {
5261 int i;
5262 fdStream err(defaultStream::error_fd());
5263 for (i = 0; i < 78; i++) err.print_raw("=");
5264 err.cr();
5265 err.print_raw_cr(title);
5266 for (i = 0; i < 78; i++) err.print_raw("-");
5267 err.cr();
5268 err.print_raw_cr(message);
5269 for (i = 0; i < 78; i++) err.print_raw("=");
5270 err.cr();
5271
5272 char buf[16];
5273 // Prevent process from exiting upon "read error" without consuming all CPU
5274 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5275
5276 return buf[0] == 'y' || buf[0] == 'Y';
5277 }
5278
5279 int os::stat(const char *path, struct stat *sbuf) {
5280 char pathbuf[MAX_PATH];
5281 if (strlen(path) > MAX_PATH - 1) {
5282 errno = ENAMETOOLONG;
5283 return -1;
5284 }
5285 os::native_path(strcpy(pathbuf, path));
5286 return ::stat(pathbuf, sbuf);
5287 }
5288
5289 // Is a (classpath) directory empty?
5290 bool os::dir_is_empty(const char* path) {
5291 DIR *dir = NULL;
5292 struct dirent *ptr;
5293
5294 dir = opendir(path);
5295 if (dir == NULL) return true;
5296
5297 // Scan the directory
5298 bool result = true;
5299 char buf[sizeof(struct dirent) + MAX_PATH];
5300 while (result && (ptr = ::readdir(dir)) != NULL) {
5301 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5302 result = false;
5303 }
5304 }
5305 closedir(dir);
5306 return result;
5307 }
5308
5309 // This code originates from JDK's sysOpen and open64_w
5310 // from src/solaris/hpi/src/system_md.c
5311
5312 int os::open(const char *path, int oflag, int mode) {
5313 if (strlen(path) > MAX_PATH - 1) {
5314 errno = ENAMETOOLONG;
5315 return -1;
5316 }
5317
5318 // All file descriptors that are opened in the Java process and not
5319 // specifically destined for a subprocess should have the close-on-exec
5320 // flag set. If we don't set it, then careless 3rd party native code
5321 // might fork and exec without closing all appropriate file descriptors
5322 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5323 // turn might:
5324 //
5325 // - cause end-of-file to fail to be detected on some file
5326 // descriptors, resulting in mysterious hangs, or
5327 //
5328 // - might cause an fopen in the subprocess to fail on a system
5329 // suffering from bug 1085341.
5330 //
5331 // (Yes, the default setting of the close-on-exec flag is a Unix
5332 // design flaw)
5333 //
5334 // See:
5335 // 1085341: 32-bit stdio routines should support file descriptors >255
5336 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5337 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5338 //
5339 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5340 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5341 // because it saves a system call and removes a small window where the flag
5342 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5343 // and we fall back to using FD_CLOEXEC (see below).
5344 #ifdef O_CLOEXEC
5345 oflag |= O_CLOEXEC;
5346 #endif
5347
5348 int fd = ::open64(path, oflag, mode);
5349 if (fd == -1) return -1;
5350
5351 //If the open succeeded, the file might still be a directory
5352 {
5353 struct stat64 buf64;
5354 int ret = ::fstat64(fd, &buf64);
5355 int st_mode = buf64.st_mode;
5356
5357 if (ret != -1) {
5358 if ((st_mode & S_IFMT) == S_IFDIR) {
5359 errno = EISDIR;
5360 ::close(fd);
5361 return -1;
5362 }
5363 } else {
5364 ::close(fd);
5365 return -1;
5366 }
5367 }
5368
5369 #ifdef FD_CLOEXEC
5370 // Validate that the use of the O_CLOEXEC flag on open above worked.
5371 // With recent kernels, we will perform this check exactly once.
5372 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5373 if (!O_CLOEXEC_is_known_to_work) {
5374 int flags = ::fcntl(fd, F_GETFD);
5375 if (flags != -1) {
5376 if ((flags & FD_CLOEXEC) != 0)
5377 O_CLOEXEC_is_known_to_work = 1;
5378 else
5379 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5380 }
5381 }
5382 #endif
5383
5384 return fd;
5385 }
5386
5387
5388 // create binary file, rewriting existing file if required
5389 int os::create_binary_file(const char* path, bool rewrite_existing) {
5390 int oflags = O_WRONLY | O_CREAT;
5391 if (!rewrite_existing) {
5392 oflags |= O_EXCL;
5393 }
5394 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5395 }
5396
5397 // return current position of file pointer
5398 jlong os::current_file_offset(int fd) {
5399 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5400 }
5401
5402 // move file pointer to the specified offset
5403 jlong os::seek_to_file_offset(int fd, jlong offset) {
5404 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5405 }
5406
5407 // This code originates from JDK's sysAvailable
5408 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5409
5410 int os::available(int fd, jlong *bytes) {
5411 jlong cur, end;
5412 int mode;
5413 struct stat64 buf64;
5414
5415 if (::fstat64(fd, &buf64) >= 0) {
5416 mode = buf64.st_mode;
5417 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5418 int n;
5419 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5420 *bytes = n;
5421 return 1;
5422 }
5423 }
5424 }
5425 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5426 return 0;
5427 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5428 return 0;
5429 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5430 return 0;
5431 }
5432 *bytes = end - cur;
5433 return 1;
5434 }
5435
5436 // Map a block of memory.
5437 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5438 char *addr, size_t bytes, bool read_only,
5439 bool allow_exec) {
5440 int prot;
5441 int flags = MAP_PRIVATE;
5442
5443 if (read_only) {
5444 prot = PROT_READ;
5445 } else {
5446 prot = PROT_READ | PROT_WRITE;
5447 }
5448
5449 if (allow_exec) {
5450 prot |= PROT_EXEC;
5451 }
5452
5453 if (addr != NULL) {
5454 flags |= MAP_FIXED;
5455 }
5456
5457 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5458 fd, file_offset);
5459 if (mapped_address == MAP_FAILED) {
5460 return NULL;
5461 }
5462 return mapped_address;
5463 }
5464
5465
5466 // Remap a block of memory.
5467 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5468 char *addr, size_t bytes, bool read_only,
5469 bool allow_exec) {
5470 // same as map_memory() on this OS
5471 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5472 allow_exec);
5473 }
5474
5475
5476 // Unmap a block of memory.
5477 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5478 return munmap(addr, bytes) == 0;
5479 }
5480
5481 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5482
5483 static clockid_t thread_cpu_clockid(Thread* thread) {
5484 pthread_t tid = thread->osthread()->pthread_id();
5485 clockid_t clockid;
5486
5487 // Get thread clockid
5488 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5489 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5490 return clockid;
5491 }
5492
5493 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5494 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5495 // of a thread.
5496 //
5497 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5498 // the fast estimate available on the platform.
5499
5500 jlong os::current_thread_cpu_time() {
5501 if (os::Linux::supports_fast_thread_cpu_time()) {
5502 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5503 } else {
5504 // return user + sys since the cost is the same
5505 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5506 }
5507 }
5508
5509 jlong os::thread_cpu_time(Thread* thread) {
5510 // consistent with what current_thread_cpu_time() returns
5511 if (os::Linux::supports_fast_thread_cpu_time()) {
5512 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5513 } else {
5514 return slow_thread_cpu_time(thread, true /* user + sys */);
5515 }
5516 }
5517
5518 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5519 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5520 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5521 } else {
5522 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5523 }
5524 }
5525
5526 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5527 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5528 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5529 } else {
5530 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5531 }
5532 }
5533
5534 // -1 on error.
5535 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5536 pid_t tid = thread->osthread()->thread_id();
5537 char *s;
5538 char stat[2048];
5539 int statlen;
5540 char proc_name[64];
5541 int count;
5542 long sys_time, user_time;
5543 char cdummy;
5544 int idummy;
5545 long ldummy;
5546 FILE *fp;
5547
5548 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5549 fp = fopen(proc_name, "r");
5550 if (fp == NULL) return -1;
5551 statlen = fread(stat, 1, 2047, fp);
5552 stat[statlen] = '\0';
5553 fclose(fp);
5554
5555 // Skip pid and the command string. Note that we could be dealing with
5556 // weird command names, e.g. user could decide to rename java launcher
5557 // to "java 1.4.2 :)", then the stat file would look like
5558 // 1234 (java 1.4.2 :)) R ... ...
5559 // We don't really need to know the command string, just find the last
5560 // occurrence of ")" and then start parsing from there. See bug 4726580.
5561 s = strrchr(stat, ')');
5562 if (s == NULL) return -1;
5563
5564 // Skip blank chars
5565 do { s++; } while (s && isspace(*s));
5566
5567 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5568 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5569 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5570 &user_time, &sys_time);
5571 if (count != 13) return -1;
5572 if (user_sys_cpu_time) {
5573 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5574 } else {
5575 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5576 }
5577 }
5578
5579 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5580 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5581 info_ptr->may_skip_backward = false; // elapsed time not wall time
5582 info_ptr->may_skip_forward = false; // elapsed time not wall time
5583 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5584 }
5585
5586 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5587 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5588 info_ptr->may_skip_backward = false; // elapsed time not wall time
5589 info_ptr->may_skip_forward = false; // elapsed time not wall time
5590 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5591 }
5592
5593 bool os::is_thread_cpu_time_supported() {
5594 return true;
5595 }
5596
5597 // System loadavg support. Returns -1 if load average cannot be obtained.
5598 // Linux doesn't yet have a (official) notion of processor sets,
5599 // so just return the system wide load average.
5600 int os::loadavg(double loadavg[], int nelem) {
5601 return ::getloadavg(loadavg, nelem);
5602 }
5603
5604 void os::pause() {
5605 char filename[MAX_PATH];
5606 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5607 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5608 } else {
5609 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5610 }
5611
5612 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5613 if (fd != -1) {
5614 struct stat buf;
5615 ::close(fd);
5616 while (::stat(filename, &buf) == 0) {
5617 (void)::poll(NULL, 0, 100);
5618 }
5619 } else {
5620 jio_fprintf(stderr,
5621 "Could not open pause file '%s', continuing immediately.\n", filename);
5622 }
5623 }
5624
5625 extern char** environ;
5626
5627 // Run the specified command in a separate process. Return its exit value,
5628 // or -1 on failure (e.g. can't fork a new process).
5629 // Unlike system(), this function can be called from signal handler. It
5630 // doesn't block SIGINT et al.
5631 int os::fork_and_exec(char* cmd) {
5632 const char * argv[4] = {"sh", "-c", cmd, NULL};
5633
5634 pid_t pid = fork();
5635
5636 if (pid < 0) {
5637 // fork failed
5638 return -1;
5639
5640 } else if (pid == 0) {
5641 // child process
5642
5643 execve("/bin/sh", (char* const*)argv, environ);
5644
5645 // execve failed
5646 _exit(-1);
5647
5648 } else {
5649 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5650 // care about the actual exit code, for now.
5651
5652 int status;
5653
5654 // Wait for the child process to exit. This returns immediately if
5655 // the child has already exited. */
5656 while (waitpid(pid, &status, 0) < 0) {
5657 switch (errno) {
5658 case ECHILD: return 0;
5659 case EINTR: break;
5660 default: return -1;
5661 }
5662 }
5663
5664 if (WIFEXITED(status)) {
5665 // The child exited normally; get its exit code.
5666 return WEXITSTATUS(status);
5667 } else if (WIFSIGNALED(status)) {
5668 // The child exited because of a signal
5669 // The best value to return is 0x80 + signal number,
5670 // because that is what all Unix shells do, and because
5671 // it allows callers to distinguish between process exit and
5672 // process death by signal.
5673 return 0x80 + WTERMSIG(status);
5674 } else {
5675 // Unknown exit code; pass it through
5676 return status;
5677 }
5678 }
5679 }
5680
5681 // Get the default path to the core file
5682 // Returns the length of the string
5683 int os::get_core_path(char* buffer, size_t bufferSize) {
5684 /*
5685 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5686 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5687 */
5688 const int core_pattern_len = 129;
5689 char core_pattern[core_pattern_len] = {0};
5690
5691 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5692 if (core_pattern_file == -1) {
5693 return -1;
5694 }
5695
5696 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5697 ::close(core_pattern_file);
5698 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5699 return -1;
5700 }
5701 if (core_pattern[ret-1] == '\n') {
5702 core_pattern[ret-1] = '\0';
5703 } else {
5704 core_pattern[ret] = '\0';
5705 }
5706
5707 char *pid_pos = strstr(core_pattern, "%p");
5708 int written;
5709
5710 if (core_pattern[0] == '/') {
5711 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5712 } else {
5713 char cwd[PATH_MAX];
5714
5715 const char* p = get_current_directory(cwd, PATH_MAX);
5716 if (p == NULL) {
5717 return -1;
5718 }
5719
5720 if (core_pattern[0] == '|') {
5721 written = jio_snprintf(buffer, bufferSize,
5722 "\"%s\" (or dumping to %s/core.%d)",
5723 &core_pattern[1], p, current_process_id());
5724 } else {
5725 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5726 }
5727 }
5728
5729 if (written < 0) {
5730 return -1;
5731 }
5732
5733 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5734 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5735
5736 if (core_uses_pid_file != -1) {
5737 char core_uses_pid = 0;
5738 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5739 ::close(core_uses_pid_file);
5740
5741 if (core_uses_pid == '1') {
5742 jio_snprintf(buffer + written, bufferSize - written,
5743 ".%d", current_process_id());
5744 }
5745 }
5746 }
5747
5748 return strlen(buffer);
5749 }
5750
5751 bool os::start_debugging(char *buf, int buflen) {
5752 int len = (int)strlen(buf);
5753 char *p = &buf[len];
5754
5755 jio_snprintf(p, buflen-len,
5756 "\n\n"
5757 "Do you want to debug the problem?\n\n"
5758 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5759 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5760 "Otherwise, press RETURN to abort...",
5761 os::current_process_id(), os::current_process_id(),
5762 os::current_thread_id(), os::current_thread_id());
5763
5764 bool yes = os::message_box("Unexpected Error", buf);
5765
5766 if (yes) {
5767 // yes, user asked VM to launch debugger
5768 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5769 os::current_process_id(), os::current_process_id());
5770
5771 os::fork_and_exec(buf);
5772 yes = false;
5773 }
5774 return yes;
5775 }
5776
5777
5778 // Java/Compiler thread:
5779 //
5780 // Low memory addresses
5781 // P0 +------------------------+
5782 // | |\ Java thread created by VM does not have glibc
5783 // | glibc guard page | - guard page, attached Java thread usually has
5784 // | |/ 1 glibc guard page.
5785 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5786 // | |\
5787 // | HotSpot Guard Pages | - red, yellow and reserved pages
5788 // | |/
5789 // +------------------------+ JavaThread::stack_reserved_zone_base()
5790 // | |\
5791 // | Normal Stack | -
5792 // | |/
5793 // P2 +------------------------+ Thread::stack_base()
5794 //
5795 // Non-Java thread:
5796 //
5797 // Low memory addresses
5798 // P0 +------------------------+
5799 // | |\
5800 // | glibc guard page | - usually 1 page
5801 // | |/
5802 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5803 // | |\
5804 // | Normal Stack | -
5805 // | |/
5806 // P2 +------------------------+ Thread::stack_base()
5807 //
5808 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5809 // returned from pthread_attr_getstack().
5810 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5811 // of the stack size given in pthread_attr. We work around this for
5812 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5813 //
5814 #ifndef ZERO
5815 static void current_stack_region(address * bottom, size_t * size) {
5816 if (os::is_primordial_thread()) {
5817 // primordial thread needs special handling because pthread_getattr_np()
5818 // may return bogus value.
5819 *bottom = os::Linux::initial_thread_stack_bottom();
5820 *size = os::Linux::initial_thread_stack_size();
5821 } else {
5822 pthread_attr_t attr;
5823
5824 int rslt = pthread_getattr_np(pthread_self(), &attr);
5825
5826 // JVM needs to know exact stack location, abort if it fails
5827 if (rslt != 0) {
5828 if (rslt == ENOMEM) {
5829 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5830 } else {
5831 fatal("pthread_getattr_np failed with error = %d", rslt);
5832 }
5833 }
5834
5835 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5836 fatal("Cannot locate current stack attributes!");
5837 }
5838
5839 // Work around NPTL stack guard error.
5840 size_t guard_size = 0;
5841 rslt = pthread_attr_getguardsize(&attr, &guard_size);
5842 if (rslt != 0) {
5843 fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5844 }
5845 *bottom += guard_size;
5846 *size -= guard_size;
5847
5848 pthread_attr_destroy(&attr);
5849
5850 }
5851 assert(os::current_stack_pointer() >= *bottom &&
5852 os::current_stack_pointer() < *bottom + *size, "just checking");
5853 }
5854
5855 address os::current_stack_base() {
5856 address bottom;
5857 size_t size;
5858 current_stack_region(&bottom, &size);
5859 return (bottom + size);
5860 }
5861
5862 size_t os::current_stack_size() {
5863 // This stack size includes the usable stack and HotSpot guard pages
5864 // (for the threads that have Hotspot guard pages).
5865 address bottom;
5866 size_t size;
5867 current_stack_region(&bottom, &size);
5868 return size;
5869 }
5870 #endif
5871
5872 static inline struct timespec get_mtime(const char* filename) {
5873 struct stat st;
5874 int ret = os::stat(filename, &st);
5875 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5876 return st.st_mtim;
5877 }
5878
5879 int os::compare_file_modified_times(const char* file1, const char* file2) {
5880 struct timespec filetime1 = get_mtime(file1);
5881 struct timespec filetime2 = get_mtime(file2);
5882 int diff = filetime1.tv_sec - filetime2.tv_sec;
5883 if (diff == 0) {
5884 return filetime1.tv_nsec - filetime2.tv_nsec;
5885 }
5886 return diff;
5887 }
5888
5889 /////////////// Unit tests ///////////////
5890
5891 #ifndef PRODUCT
5892
5893 #define test_log(...) \
5894 do { \
5895 if (VerboseInternalVMTests) { \
5896 tty->print_cr(__VA_ARGS__); \
5897 tty->flush(); \
5898 } \
5899 } while (false)
5900
5901 class TestReserveMemorySpecial : AllStatic {
5902 public:
5903 static void small_page_write(void* addr, size_t size) {
5904 size_t page_size = os::vm_page_size();
5905
5906 char* end = (char*)addr + size;
5907 for (char* p = (char*)addr; p < end; p += page_size) {
5908 *p = 1;
5909 }
5910 }
5911
5912 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5913 if (!UseHugeTLBFS) {
5914 return;
5915 }
5916
5917 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5918
5919 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5920
5921 if (addr != NULL) {
5922 small_page_write(addr, size);
5923
5924 os::Linux::release_memory_special_huge_tlbfs(addr, size);
5925 }
5926 }
5927
5928 static void test_reserve_memory_special_huge_tlbfs_only() {
5929 if (!UseHugeTLBFS) {
5930 return;
5931 }
5932
5933 size_t lp = os::large_page_size();
5934
5935 for (size_t size = lp; size <= lp * 10; size += lp) {
5936 test_reserve_memory_special_huge_tlbfs_only(size);
5937 }
5938 }
5939
5940 static void test_reserve_memory_special_huge_tlbfs_mixed() {
5941 size_t lp = os::large_page_size();
5942 size_t ag = os::vm_allocation_granularity();
5943
5944 // sizes to test
5945 const size_t sizes[] = {
5946 lp, lp + ag, lp + lp / 2, lp * 2,
5947 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5948 lp * 10, lp * 10 + lp / 2
5949 };
5950 const int num_sizes = sizeof(sizes) / sizeof(size_t);
5951
5952 // For each size/alignment combination, we test three scenarios:
5953 // 1) with req_addr == NULL
5954 // 2) with a non-null req_addr at which we expect to successfully allocate
5955 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5956 // expect the allocation to either fail or to ignore req_addr
5957
5958 // Pre-allocate two areas; they shall be as large as the largest allocation
5959 // and aligned to the largest alignment we will be testing.
5960 const size_t mapping_size = sizes[num_sizes - 1] * 2;
5961 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5962 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5963 -1, 0);
5964 assert(mapping1 != MAP_FAILED, "should work");
5965
5966 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5967 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5968 -1, 0);
5969 assert(mapping2 != MAP_FAILED, "should work");
5970
5971 // Unmap the first mapping, but leave the second mapping intact: the first
5972 // mapping will serve as a value for a "good" req_addr (case 2). The second
5973 // mapping, still intact, as "bad" req_addr (case 3).
5974 ::munmap(mapping1, mapping_size);
5975
5976 // Case 1
5977 test_log("%s, req_addr NULL:", __FUNCTION__);
5978 test_log("size align result");
5979
5980 for (int i = 0; i < num_sizes; i++) {
5981 const size_t size = sizes[i];
5982 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5983 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5984 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
5985 size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5986 if (p != NULL) {
5987 assert(is_aligned(p, alignment), "must be");
5988 small_page_write(p, size);
5989 os::Linux::release_memory_special_huge_tlbfs(p, size);
5990 }
5991 }
5992 }
5993
5994 // Case 2
5995 test_log("%s, req_addr non-NULL:", __FUNCTION__);
5996 test_log("size align req_addr result");
5997
5998 for (int i = 0; i < num_sizes; i++) {
5999 const size_t size = sizes[i];
6000 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6001 char* const req_addr = align_up(mapping1, alignment);
6002 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6003 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6004 size, alignment, p2i(req_addr), p2i(p),
6005 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6006 if (p != NULL) {
6007 assert(p == req_addr, "must be");
6008 small_page_write(p, size);
6009 os::Linux::release_memory_special_huge_tlbfs(p, size);
6010 }
6011 }
6012 }
6013
6014 // Case 3
6015 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6016 test_log("size align req_addr result");
6017
6018 for (int i = 0; i < num_sizes; i++) {
6019 const size_t size = sizes[i];
6020 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6021 char* const req_addr = align_up(mapping2, alignment);
6022 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6023 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6024 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6025 // as the area around req_addr contains already existing mappings, the API should always
6026 // return NULL (as per contract, it cannot return another address)
6027 assert(p == NULL, "must be");
6028 }
6029 }
6030
6031 ::munmap(mapping2, mapping_size);
6032
6033 }
6034
6035 static void test_reserve_memory_special_huge_tlbfs() {
6036 if (!UseHugeTLBFS) {
6037 return;
6038 }
6039
6040 test_reserve_memory_special_huge_tlbfs_only();
6041 test_reserve_memory_special_huge_tlbfs_mixed();
6042 }
6043
6044 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6045 if (!UseSHM) {
6046 return;
6047 }
6048
6049 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6050
6051 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6052
6053 if (addr != NULL) {
6054 assert(is_aligned(addr, alignment), "Check");
6055 assert(is_aligned(addr, os::large_page_size()), "Check");
6056
6057 small_page_write(addr, size);
6058
6059 os::Linux::release_memory_special_shm(addr, size);
6060 }
6061 }
6062
6063 static void test_reserve_memory_special_shm() {
6064 size_t lp = os::large_page_size();
6065 size_t ag = os::vm_allocation_granularity();
6066
6067 for (size_t size = ag; size < lp * 3; size += ag) {
6068 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6069 test_reserve_memory_special_shm(size, alignment);
6070 }
6071 }
6072 }
6073
6074 static void test() {
6075 test_reserve_memory_special_huge_tlbfs();
6076 test_reserve_memory_special_shm();
6077 }
6078 };
6079
6080 void TestReserveMemorySpecial_test() {
6081 TestReserveMemorySpecial::test();
6082 }
6083
6084 #endif